Commit | Line | Data |
---|---|---|
f942dc25 IC |
1 | /* |
2 | * Back-end of the driver for virtual network devices. This portion of the | |
3 | * driver exports a 'unified' network-device interface that can be accessed | |
4 | * by any operating system that implements a compatible front end. A | |
5 | * reference front-end implementation can be found in: | |
6 | * drivers/net/xen-netfront.c | |
7 | * | |
8 | * Copyright (c) 2002-2005, K A Fraser | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License version 2 | |
12 | * as published by the Free Software Foundation; or, when distributed | |
13 | * separately from the Linux kernel or incorporated into other | |
14 | * software packages, subject to the following license: | |
15 | * | |
16 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
17 | * of this source file (the "Software"), to deal in the Software without | |
18 | * restriction, including without limitation the rights to use, copy, modify, | |
19 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
20 | * and to permit persons to whom the Software is furnished to do so, subject to | |
21 | * the following conditions: | |
22 | * | |
23 | * The above copyright notice and this permission notice shall be included in | |
24 | * all copies or substantial portions of the Software. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
27 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
28 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
29 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
30 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
31 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
32 | * IN THE SOFTWARE. | |
33 | */ | |
34 | ||
35 | #include "common.h" | |
36 | ||
37 | #include <linux/kthread.h> | |
38 | #include <linux/if_vlan.h> | |
39 | #include <linux/udp.h> | |
40 | ||
41 | #include <net/tcp.h> | |
42 | ||
ca981633 | 43 | #include <xen/xen.h> |
f942dc25 IC |
44 | #include <xen/events.h> |
45 | #include <xen/interface/memory.h> | |
46 | ||
47 | #include <asm/xen/hypercall.h> | |
48 | #include <asm/xen/page.h> | |
49 | ||
e1f00a69 WL |
50 | /* Provide an option to disable split event channels at load time as |
51 | * event channels are limited resource. Split event channels are | |
52 | * enabled by default. | |
53 | */ | |
54 | bool separate_tx_rx_irq = 1; | |
55 | module_param(separate_tx_rx_irq, bool, 0644); | |
56 | ||
2810e5b9 WL |
57 | /* |
58 | * This is the maximum slots a skb can have. If a guest sends a skb | |
59 | * which exceeds this limit it is considered malicious. | |
60 | */ | |
37641494 WL |
61 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
62 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
63 | module_param(fatal_skb_slots, uint, 0444); | |
64 | ||
65 | /* | |
66 | * To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating | |
67 | * the maximum slots a valid packet can use. Now this value is defined | |
68 | * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by | |
69 | * all backend. | |
70 | */ | |
71 | #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN | |
2810e5b9 | 72 | |
2810e5b9 WL |
73 | /* |
74 | * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of | |
75 | * one or more merged tx requests, otherwise it is the continuation of | |
76 | * previous tx request. | |
77 | */ | |
b3f980bd | 78 | static inline int pending_tx_is_head(struct xenvif *vif, RING_IDX idx) |
f942dc25 | 79 | { |
b3f980bd | 80 | return vif->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX; |
f942dc25 IC |
81 | } |
82 | ||
7376419a WL |
83 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
84 | u8 status); | |
85 | ||
f942dc25 IC |
86 | static void make_tx_response(struct xenvif *vif, |
87 | struct xen_netif_tx_request *txp, | |
88 | s8 st); | |
b3f980bd WL |
89 | |
90 | static inline int tx_work_todo(struct xenvif *vif); | |
91 | static inline int rx_work_todo(struct xenvif *vif); | |
92 | ||
f942dc25 IC |
93 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
94 | u16 id, | |
95 | s8 st, | |
96 | u16 offset, | |
97 | u16 size, | |
98 | u16 flags); | |
99 | ||
b3f980bd | 100 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 101 | u16 idx) |
f942dc25 | 102 | { |
b3f980bd | 103 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
104 | } |
105 | ||
b3f980bd | 106 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 107 | u16 idx) |
f942dc25 | 108 | { |
b3f980bd | 109 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
110 | } |
111 | ||
2eba61d5 PD |
112 | /* This is a miniumum size for the linear area to avoid lots of |
113 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
114 | * value 128 was chosen as it covers all IPv4 and most likely | |
115 | * IPv6 headers. | |
f942dc25 | 116 | */ |
2eba61d5 | 117 | #define PKT_PROT_LEN 128 |
f942dc25 | 118 | |
ea066ad1 IC |
119 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
120 | { | |
121 | return (u16)frag->page_offset; | |
122 | } | |
123 | ||
124 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
125 | { | |
126 | frag->page_offset = pending_idx; | |
127 | } | |
128 | ||
f942dc25 IC |
129 | static inline pending_ring_idx_t pending_index(unsigned i) |
130 | { | |
131 | return i & (MAX_PENDING_REQS-1); | |
132 | } | |
133 | ||
b3f980bd | 134 | static inline pending_ring_idx_t nr_pending_reqs(struct xenvif *vif) |
f942dc25 IC |
135 | { |
136 | return MAX_PENDING_REQS - | |
b3f980bd | 137 | vif->pending_prod + vif->pending_cons; |
f942dc25 IC |
138 | } |
139 | ||
ca2f09f2 | 140 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
f942dc25 | 141 | { |
ca2f09f2 | 142 | RING_IDX prod, cons; |
f942dc25 | 143 | |
ca2f09f2 PD |
144 | do { |
145 | prod = vif->rx.sring->req_prod; | |
146 | cons = vif->rx.req_cons; | |
f942dc25 | 147 | |
ca2f09f2 PD |
148 | if (prod - cons >= needed) |
149 | return true; | |
f942dc25 | 150 | |
ca2f09f2 | 151 | vif->rx.sring->req_event = prod + 1; |
f942dc25 | 152 | |
ca2f09f2 PD |
153 | /* Make sure event is visible before we check prod |
154 | * again. | |
155 | */ | |
156 | mb(); | |
157 | } while (vif->rx.sring->req_prod != prod); | |
f942dc25 | 158 | |
ca2f09f2 | 159 | return false; |
f942dc25 IC |
160 | } |
161 | ||
162 | /* | |
163 | * Returns true if we should start a new receive buffer instead of | |
164 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
165 | * bytes. | |
166 | */ | |
167 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
168 | { | |
169 | /* simple case: we have completely filled the current buffer. */ | |
170 | if (offset == MAX_BUFFER_OFFSET) | |
171 | return true; | |
172 | ||
173 | /* | |
174 | * complex case: start a fresh buffer if the current frag | |
175 | * would overflow the current buffer but only if: | |
176 | * (i) this frag would fit completely in the next buffer | |
177 | * and (ii) there is already some data in the current buffer | |
178 | * and (iii) this is not the head buffer. | |
179 | * | |
180 | * Where: | |
181 | * - (i) stops us splitting a frag into two copies | |
182 | * unless the frag is too large for a single buffer. | |
183 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
184 | * - (iii) stops us leaving the first buffer | |
185 | * empty. Strictly speaking this is already covered | |
186 | * by (ii) but is explicitly checked because | |
187 | * netfront relies on the first buffer being | |
188 | * non-empty and can crash otherwise. | |
189 | * | |
190 | * This means we will effectively linearise small | |
191 | * frags but do not needlessly split large buffers | |
192 | * into multiple copies tend to give large frags their | |
193 | * own buffers as before. | |
194 | */ | |
195 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
196 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
197 | return true; | |
198 | ||
199 | return false; | |
200 | } | |
201 | ||
f942dc25 IC |
202 | struct netrx_pending_operations { |
203 | unsigned copy_prod, copy_cons; | |
204 | unsigned meta_prod, meta_cons; | |
205 | struct gnttab_copy *copy; | |
b3f980bd | 206 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
207 | int copy_off; |
208 | grant_ref_t copy_gref; | |
209 | }; | |
210 | ||
b3f980bd WL |
211 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
212 | struct netrx_pending_operations *npo) | |
f942dc25 | 213 | { |
b3f980bd | 214 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
215 | struct xen_netif_rx_request *req; |
216 | ||
217 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
218 | ||
219 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 220 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
221 | meta->gso_size = 0; |
222 | meta->size = 0; | |
223 | meta->id = req->id; | |
224 | ||
225 | npo->copy_off = 0; | |
226 | npo->copy_gref = req->gref; | |
227 | ||
228 | return meta; | |
229 | } | |
230 | ||
33bc801d WL |
231 | /* |
232 | * Set up the grant operations for this fragment. If it's a flipping | |
233 | * interface, we also set up the unmap request from here. | |
234 | */ | |
7376419a WL |
235 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
236 | struct netrx_pending_operations *npo, | |
237 | struct page *page, unsigned long size, | |
33bc801d | 238 | unsigned long offset, int *head) |
f942dc25 IC |
239 | { |
240 | struct gnttab_copy *copy_gop; | |
b3f980bd | 241 | struct xenvif_rx_meta *meta; |
f942dc25 | 242 | unsigned long bytes; |
5bd07670 | 243 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
244 | |
245 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 246 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
247 | |
248 | meta = npo->meta + npo->meta_prod - 1; | |
249 | ||
6a8ed462 IC |
250 | /* Skip unused frames from start of page */ |
251 | page += offset >> PAGE_SHIFT; | |
252 | offset &= ~PAGE_MASK; | |
253 | ||
f942dc25 | 254 | while (size > 0) { |
6a8ed462 | 255 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
256 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
257 | ||
6a8ed462 IC |
258 | bytes = PAGE_SIZE - offset; |
259 | ||
260 | if (bytes > size) | |
261 | bytes = size; | |
262 | ||
33bc801d | 263 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
264 | /* |
265 | * Netfront requires there to be some data in the head | |
266 | * buffer. | |
267 | */ | |
33bc801d | 268 | BUG_ON(*head); |
f942dc25 IC |
269 | |
270 | meta = get_next_rx_buffer(vif, npo); | |
271 | } | |
272 | ||
f942dc25 IC |
273 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
274 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
275 | ||
276 | copy_gop = npo->copy + npo->copy_prod++; | |
277 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
278 | copy_gop->len = bytes; |
279 | ||
43e9d194 WL |
280 | copy_gop->source.domid = DOMID_SELF; |
281 | copy_gop->source.u.gmfn = virt_to_mfn(page_address(page)); | |
f942dc25 | 282 | copy_gop->source.offset = offset; |
f942dc25 | 283 | |
b3f980bd | 284 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
285 | copy_gop->dest.offset = npo->copy_off; |
286 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
287 | |
288 | npo->copy_off += bytes; | |
289 | meta->size += bytes; | |
290 | ||
291 | offset += bytes; | |
292 | size -= bytes; | |
293 | ||
6a8ed462 IC |
294 | /* Next frame */ |
295 | if (offset == PAGE_SIZE && size) { | |
296 | BUG_ON(!PageCompound(page)); | |
297 | page++; | |
298 | offset = 0; | |
299 | } | |
300 | ||
f942dc25 | 301 | /* Leave a gap for the GSO descriptor. */ |
5bd07670 AL |
302 | if (skb_is_gso(skb)) { |
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 | } | |
82cada22 PD |
308 | |
309 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
310 | vif->rx.req_cons++; |
311 | ||
33bc801d | 312 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
313 | |
314 | } | |
315 | } | |
316 | ||
317 | /* | |
318 | * Prepare an SKB to be transmitted to the frontend. | |
319 | * | |
320 | * This function is responsible for allocating grant operations, meta | |
321 | * structures, etc. | |
322 | * | |
323 | * It returns the number of meta structures consumed. The number of | |
324 | * ring slots used is always equal to the number of meta slots used | |
325 | * plus the number of GSO descriptors used. Currently, we use either | |
326 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
327 | * frontend-side LRO). | |
328 | */ | |
7376419a WL |
329 | static int xenvif_gop_skb(struct sk_buff *skb, |
330 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
331 | { |
332 | struct xenvif *vif = netdev_priv(skb->dev); | |
333 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
334 | int i; | |
335 | struct xen_netif_rx_request *req; | |
b3f980bd | 336 | struct xenvif_rx_meta *meta; |
f942dc25 | 337 | unsigned char *data; |
33bc801d | 338 | int head = 1; |
f942dc25 | 339 | int old_meta_prod; |
82cada22 | 340 | int gso_type; |
f942dc25 IC |
341 | |
342 | old_meta_prod = npo->meta_prod; | |
343 | ||
5bd07670 AL |
344 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
345 | if (skb_is_gso(skb)) { | |
346 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
347 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
348 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
349 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
350 | } |
351 | ||
f942dc25 | 352 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 353 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
354 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
355 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 356 | meta->gso_type = gso_type; |
5bd07670 | 357 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 IC |
358 | meta->size = 0; |
359 | meta->id = req->id; | |
360 | } | |
361 | ||
362 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
363 | meta = npo->meta + npo->meta_prod++; | |
364 | ||
82cada22 PD |
365 | if ((1 << gso_type) & vif->gso_mask) { |
366 | meta->gso_type = gso_type; | |
5bd07670 | 367 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
368 | } else { |
369 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 370 | meta->gso_size = 0; |
82cada22 | 371 | } |
f942dc25 IC |
372 | |
373 | meta->size = 0; | |
374 | meta->id = req->id; | |
375 | npo->copy_off = 0; | |
376 | npo->copy_gref = req->gref; | |
377 | ||
378 | data = skb->data; | |
379 | while (data < skb_tail_pointer(skb)) { | |
380 | unsigned int offset = offset_in_page(data); | |
381 | unsigned int len = PAGE_SIZE - offset; | |
382 | ||
383 | if (data + len > skb_tail_pointer(skb)) | |
384 | len = skb_tail_pointer(skb) - data; | |
385 | ||
7376419a | 386 | xenvif_gop_frag_copy(vif, skb, npo, |
33bc801d | 387 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
388 | data += len; |
389 | } | |
390 | ||
391 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
392 | xenvif_gop_frag_copy(vif, skb, npo, |
393 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
394 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
395 | skb_shinfo(skb)->frags[i].page_offset, | |
33bc801d | 396 | &head); |
f942dc25 IC |
397 | } |
398 | ||
399 | return npo->meta_prod - old_meta_prod; | |
400 | } | |
401 | ||
402 | /* | |
7376419a | 403 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
404 | * used to set up the operations on the top of |
405 | * netrx_pending_operations, which have since been done. Check that | |
406 | * they didn't give any errors and advance over them. | |
407 | */ | |
7376419a WL |
408 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
409 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
410 | { |
411 | struct gnttab_copy *copy_op; | |
412 | int status = XEN_NETIF_RSP_OKAY; | |
413 | int i; | |
414 | ||
415 | for (i = 0; i < nr_meta_slots; i++) { | |
416 | copy_op = npo->copy + npo->copy_cons++; | |
417 | if (copy_op->status != GNTST_okay) { | |
418 | netdev_dbg(vif->dev, | |
419 | "Bad status %d from copy to DOM%d.\n", | |
420 | copy_op->status, vif->domid); | |
421 | status = XEN_NETIF_RSP_ERROR; | |
422 | } | |
423 | } | |
424 | ||
425 | return status; | |
426 | } | |
427 | ||
7376419a WL |
428 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
429 | struct xenvif_rx_meta *meta, | |
430 | int nr_meta_slots) | |
f942dc25 IC |
431 | { |
432 | int i; | |
433 | unsigned long offset; | |
434 | ||
435 | /* No fragments used */ | |
436 | if (nr_meta_slots <= 1) | |
437 | return; | |
438 | ||
439 | nr_meta_slots--; | |
440 | ||
441 | for (i = 0; i < nr_meta_slots; i++) { | |
442 | int flags; | |
443 | if (i == nr_meta_slots - 1) | |
444 | flags = 0; | |
445 | else | |
446 | flags = XEN_NETRXF_more_data; | |
447 | ||
448 | offset = 0; | |
449 | make_rx_response(vif, meta[i].id, status, offset, | |
450 | meta[i].size, flags); | |
451 | } | |
452 | } | |
453 | ||
33bc801d WL |
454 | struct skb_cb_overlay { |
455 | int meta_slots_used; | |
456 | }; | |
457 | ||
ca2f09f2 | 458 | void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
459 | { |
460 | wake_up(&vif->wq); | |
461 | } | |
462 | ||
ca2f09f2 | 463 | static void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 464 | { |
f942dc25 | 465 | s8 status; |
e1f00a69 | 466 | u16 flags; |
f942dc25 IC |
467 | struct xen_netif_rx_response *resp; |
468 | struct sk_buff_head rxq; | |
469 | struct sk_buff *skb; | |
470 | LIST_HEAD(notify); | |
471 | int ret; | |
f942dc25 IC |
472 | unsigned long offset; |
473 | struct skb_cb_overlay *sco; | |
11b57f90 | 474 | bool need_to_notify = false; |
f942dc25 IC |
475 | |
476 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
477 | .copy = vif->grant_copy_op, |
478 | .meta = vif->meta, | |
f942dc25 IC |
479 | }; |
480 | ||
481 | skb_queue_head_init(&rxq); | |
482 | ||
b3f980bd | 483 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
9ab9831b | 484 | RING_IDX max_slots_needed; |
ca2f09f2 PD |
485 | int i; |
486 | ||
487 | /* We need a cheap worse case estimate for the number of | |
488 | * slots we'll use. | |
489 | */ | |
490 | ||
491 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
492 | skb_headlen(skb), | |
493 | PAGE_SIZE); | |
494 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
495 | unsigned int size; | |
496 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | |
497 | max_slots_needed += DIV_ROUND_UP(size, PAGE_SIZE); | |
498 | } | |
5bd07670 AL |
499 | if (skb_is_gso(skb) && |
500 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
501 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | |
ca2f09f2 PD |
502 | max_slots_needed++; |
503 | ||
504 | /* If the skb may not fit then bail out now */ | |
505 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | |
506 | skb_queue_head(&vif->rx_queue, skb); | |
11b57f90 | 507 | need_to_notify = true; |
9ab9831b | 508 | vif->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 509 | break; |
9ab9831b ZK |
510 | } else |
511 | vif->rx_last_skb_slots = 0; | |
f942dc25 IC |
512 | |
513 | sco = (struct skb_cb_overlay *)skb->cb; | |
7376419a | 514 | sco->meta_slots_used = xenvif_gop_skb(skb, &npo); |
ca2f09f2 | 515 | BUG_ON(sco->meta_slots_used > max_slots_needed); |
f942dc25 IC |
516 | |
517 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
518 | } |
519 | ||
b3f980bd | 520 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
521 | |
522 | if (!npo.copy_prod) | |
ca2f09f2 | 523 | goto done; |
f942dc25 | 524 | |
ac3d5ac2 | 525 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
b3f980bd | 526 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
527 | |
528 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
529 | sco = (struct skb_cb_overlay *)skb->cb; | |
530 | ||
82cada22 PD |
531 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
532 | vif->gso_prefix_mask) { | |
f942dc25 | 533 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 534 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
535 | |
536 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
537 | ||
b3f980bd WL |
538 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
539 | resp->id = vif->meta[npo.meta_cons].id; | |
f942dc25 IC |
540 | resp->status = sco->meta_slots_used; |
541 | ||
542 | npo.meta_cons++; | |
543 | sco->meta_slots_used--; | |
544 | } | |
545 | ||
546 | ||
547 | vif->dev->stats.tx_bytes += skb->len; | |
548 | vif->dev->stats.tx_packets++; | |
549 | ||
7376419a | 550 | status = xenvif_check_gop(vif, sco->meta_slots_used, &npo); |
f942dc25 IC |
551 | |
552 | if (sco->meta_slots_used == 1) | |
553 | flags = 0; | |
554 | else | |
555 | flags = XEN_NETRXF_more_data; | |
556 | ||
557 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
558 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
559 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
560 | /* remote but checksummed. */ | |
561 | flags |= XEN_NETRXF_data_validated; | |
562 | ||
563 | offset = 0; | |
b3f980bd | 564 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 565 | status, offset, |
b3f980bd | 566 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
567 | flags); |
568 | ||
82cada22 PD |
569 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
570 | vif->gso_mask) { | |
f942dc25 IC |
571 | struct xen_netif_extra_info *gso = |
572 | (struct xen_netif_extra_info *) | |
573 | RING_GET_RESPONSE(&vif->rx, | |
574 | vif->rx.rsp_prod_pvt++); | |
575 | ||
576 | resp->flags |= XEN_NETRXF_extra_info; | |
577 | ||
82cada22 | 578 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 579 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
580 | gso->u.gso.pad = 0; |
581 | gso->u.gso.features = 0; | |
582 | ||
583 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
584 | gso->flags = 0; | |
585 | } | |
586 | ||
7376419a WL |
587 | xenvif_add_frag_responses(vif, status, |
588 | vif->meta + npo.meta_cons + 1, | |
589 | sco->meta_slots_used); | |
f942dc25 IC |
590 | |
591 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 592 | |
11b57f90 | 593 | need_to_notify |= !!ret; |
b3f980bd | 594 | |
f942dc25 IC |
595 | npo.meta_cons += sco->meta_slots_used; |
596 | dev_kfree_skb(skb); | |
597 | } | |
598 | ||
ca2f09f2 | 599 | done: |
b3f980bd | 600 | if (need_to_notify) |
e1f00a69 | 601 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
602 | } |
603 | ||
7376419a | 604 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
605 | { |
606 | int more_to_do; | |
607 | ||
608 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
609 | ||
610 | if (more_to_do) | |
b3f980bd | 611 | napi_schedule(&vif->napi); |
f942dc25 IC |
612 | } |
613 | ||
614 | static void tx_add_credit(struct xenvif *vif) | |
615 | { | |
616 | unsigned long max_burst, max_credit; | |
617 | ||
618 | /* | |
619 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
620 | * Otherwise the interface can seize up due to insufficient credit. | |
621 | */ | |
622 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
623 | max_burst = min(max_burst, 131072UL); | |
624 | max_burst = max(max_burst, vif->credit_bytes); | |
625 | ||
626 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
627 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
628 | if (max_credit < vif->remaining_credit) | |
629 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
630 | ||
631 | vif->remaining_credit = min(max_credit, max_burst); | |
632 | } | |
633 | ||
634 | static void tx_credit_callback(unsigned long data) | |
635 | { | |
636 | struct xenvif *vif = (struct xenvif *)data; | |
637 | tx_add_credit(vif); | |
7376419a | 638 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
639 | } |
640 | ||
7376419a WL |
641 | static void xenvif_tx_err(struct xenvif *vif, |
642 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
643 | { |
644 | RING_IDX cons = vif->tx.req_cons; | |
645 | ||
646 | do { | |
647 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); | |
b9149729 | 648 | if (cons == end) |
f942dc25 IC |
649 | break; |
650 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
651 | } while (1); | |
652 | vif->tx.req_cons = cons; | |
f942dc25 IC |
653 | } |
654 | ||
7376419a | 655 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
656 | { |
657 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
658 | xenvif_carrier_off(vif); | |
48856286 IC |
659 | } |
660 | ||
7376419a WL |
661 | static int xenvif_count_requests(struct xenvif *vif, |
662 | struct xen_netif_tx_request *first, | |
663 | struct xen_netif_tx_request *txp, | |
664 | int work_to_do) | |
f942dc25 IC |
665 | { |
666 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
667 | int slots = 0; |
668 | int drop_err = 0; | |
59ccb4eb | 669 | int more_data; |
f942dc25 IC |
670 | |
671 | if (!(first->flags & XEN_NETTXF_more_data)) | |
672 | return 0; | |
673 | ||
674 | do { | |
59ccb4eb WL |
675 | struct xen_netif_tx_request dropped_tx = { 0 }; |
676 | ||
2810e5b9 WL |
677 | if (slots >= work_to_do) { |
678 | netdev_err(vif->dev, | |
679 | "Asked for %d slots but exceeds this limit\n", | |
680 | work_to_do); | |
7376419a | 681 | xenvif_fatal_tx_err(vif); |
35876b5f | 682 | return -ENODATA; |
f942dc25 IC |
683 | } |
684 | ||
2810e5b9 WL |
685 | /* This guest is really using too many slots and |
686 | * considered malicious. | |
687 | */ | |
37641494 | 688 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
689 | netdev_err(vif->dev, |
690 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 691 | slots, fatal_skb_slots); |
7376419a | 692 | xenvif_fatal_tx_err(vif); |
35876b5f | 693 | return -E2BIG; |
f942dc25 IC |
694 | } |
695 | ||
2810e5b9 | 696 | /* Xen network protocol had implicit dependency on |
37641494 WL |
697 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
698 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
699 | * same behavior as before. Any packet using more than | |
700 | * 18 slots but less than fatal_skb_slots slots is | |
701 | * dropped | |
2810e5b9 | 702 | */ |
37641494 | 703 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
704 | if (net_ratelimit()) |
705 | netdev_dbg(vif->dev, | |
706 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 707 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
708 | drop_err = -E2BIG; |
709 | } | |
710 | ||
59ccb4eb WL |
711 | if (drop_err) |
712 | txp = &dropped_tx; | |
713 | ||
2810e5b9 | 714 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 715 | sizeof(*txp)); |
03393fd5 WL |
716 | |
717 | /* If the guest submitted a frame >= 64 KiB then | |
718 | * first->size overflowed and following slots will | |
719 | * appear to be larger than the frame. | |
720 | * | |
721 | * This cannot be fatal error as there are buggy | |
722 | * frontends that do this. | |
723 | * | |
724 | * Consume all slots and drop the packet. | |
725 | */ | |
726 | if (!drop_err && txp->size > first->size) { | |
727 | if (net_ratelimit()) | |
728 | netdev_dbg(vif->dev, | |
729 | "Invalid tx request, slot size %u > remaining size %u\n", | |
730 | txp->size, first->size); | |
731 | drop_err = -EIO; | |
f942dc25 IC |
732 | } |
733 | ||
734 | first->size -= txp->size; | |
2810e5b9 | 735 | slots++; |
f942dc25 IC |
736 | |
737 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 738 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 739 | txp->offset, txp->size); |
7376419a | 740 | xenvif_fatal_tx_err(vif); |
35876b5f | 741 | return -EINVAL; |
f942dc25 | 742 | } |
59ccb4eb WL |
743 | |
744 | more_data = txp->flags & XEN_NETTXF_more_data; | |
745 | ||
746 | if (!drop_err) | |
747 | txp++; | |
748 | ||
749 | } while (more_data); | |
2810e5b9 WL |
750 | |
751 | if (drop_err) { | |
7376419a | 752 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
753 | return drop_err; |
754 | } | |
755 | ||
756 | return slots; | |
f942dc25 IC |
757 | } |
758 | ||
7376419a WL |
759 | static struct page *xenvif_alloc_page(struct xenvif *vif, |
760 | u16 pending_idx) | |
f942dc25 IC |
761 | { |
762 | struct page *page; | |
b3f980bd WL |
763 | |
764 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
f942dc25 IC |
765 | if (!page) |
766 | return NULL; | |
b3f980bd WL |
767 | vif->mmap_pages[pending_idx] = page; |
768 | ||
f942dc25 IC |
769 | return page; |
770 | } | |
771 | ||
7376419a WL |
772 | static struct gnttab_copy *xenvif_get_requests(struct xenvif *vif, |
773 | struct sk_buff *skb, | |
774 | struct xen_netif_tx_request *txp, | |
775 | struct gnttab_copy *gop) | |
f942dc25 IC |
776 | { |
777 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
778 | skb_frag_t *frags = shinfo->frags; | |
ea066ad1 | 779 | u16 pending_idx = *((u16 *)skb->data); |
2810e5b9 WL |
780 | u16 head_idx = 0; |
781 | int slot, start; | |
782 | struct page *page; | |
783 | pending_ring_idx_t index, start_idx = 0; | |
784 | uint16_t dst_offset; | |
785 | unsigned int nr_slots; | |
786 | struct pending_tx_info *first = NULL; | |
787 | ||
788 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 789 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 WL |
790 | */ |
791 | nr_slots = shinfo->nr_frags; | |
f942dc25 IC |
792 | |
793 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 794 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 795 | |
2810e5b9 WL |
796 | /* Coalesce tx requests, at this point the packet passed in |
797 | * should be <= 64K. Any packets larger than 64K have been | |
7376419a | 798 | * handled in xenvif_count_requests(). |
2810e5b9 WL |
799 | */ |
800 | for (shinfo->nr_frags = slot = start; slot < nr_slots; | |
801 | shinfo->nr_frags++) { | |
f942dc25 | 802 | struct pending_tx_info *pending_tx_info = |
b3f980bd | 803 | vif->pending_tx_info; |
f942dc25 | 804 | |
b3f980bd | 805 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); |
f942dc25 | 806 | if (!page) |
4cc7c1cb | 807 | goto err; |
f942dc25 | 808 | |
2810e5b9 WL |
809 | dst_offset = 0; |
810 | first = NULL; | |
811 | while (dst_offset < PAGE_SIZE && slot < nr_slots) { | |
812 | gop->flags = GNTCOPY_source_gref; | |
813 | ||
814 | gop->source.u.ref = txp->gref; | |
815 | gop->source.domid = vif->domid; | |
816 | gop->source.offset = txp->offset; | |
817 | ||
818 | gop->dest.domid = DOMID_SELF; | |
819 | ||
820 | gop->dest.offset = dst_offset; | |
821 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
822 | ||
823 | if (dst_offset + txp->size > PAGE_SIZE) { | |
824 | /* This page can only merge a portion | |
825 | * of tx request. Do not increment any | |
826 | * pointer / counter here. The txp | |
827 | * will be dealt with in future | |
828 | * rounds, eventually hitting the | |
829 | * `else` branch. | |
830 | */ | |
831 | gop->len = PAGE_SIZE - dst_offset; | |
832 | txp->offset += gop->len; | |
833 | txp->size -= gop->len; | |
834 | dst_offset += gop->len; /* quit loop */ | |
835 | } else { | |
836 | /* This tx request can be merged in the page */ | |
837 | gop->len = txp->size; | |
838 | dst_offset += gop->len; | |
839 | ||
b3f980bd | 840 | index = pending_index(vif->pending_cons++); |
2810e5b9 | 841 | |
b3f980bd | 842 | pending_idx = vif->pending_ring[index]; |
2810e5b9 WL |
843 | |
844 | memcpy(&pending_tx_info[pending_idx].req, txp, | |
845 | sizeof(*txp)); | |
2810e5b9 WL |
846 | |
847 | /* Poison these fields, corresponding | |
848 | * fields for head tx req will be set | |
849 | * to correct values after the loop. | |
850 | */ | |
b3f980bd | 851 | vif->mmap_pages[pending_idx] = (void *)(~0UL); |
2810e5b9 WL |
852 | pending_tx_info[pending_idx].head = |
853 | INVALID_PENDING_RING_IDX; | |
854 | ||
855 | if (!first) { | |
856 | first = &pending_tx_info[pending_idx]; | |
857 | start_idx = index; | |
858 | head_idx = pending_idx; | |
859 | } | |
860 | ||
861 | txp++; | |
862 | slot++; | |
863 | } | |
f942dc25 | 864 | |
2810e5b9 WL |
865 | gop++; |
866 | } | |
f942dc25 | 867 | |
2810e5b9 WL |
868 | first->req.offset = 0; |
869 | first->req.size = dst_offset; | |
870 | first->head = start_idx; | |
b3f980bd | 871 | vif->mmap_pages[head_idx] = page; |
2810e5b9 | 872 | frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx); |
f942dc25 IC |
873 | } |
874 | ||
2810e5b9 WL |
875 | BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS); |
876 | ||
f942dc25 | 877 | return gop; |
4cc7c1cb IC |
878 | err: |
879 | /* Unwind, freeing all pages and sending error responses. */ | |
2810e5b9 | 880 | while (shinfo->nr_frags-- > start) { |
7376419a | 881 | xenvif_idx_release(vif, |
2810e5b9 WL |
882 | frag_get_pending_idx(&frags[shinfo->nr_frags]), |
883 | XEN_NETIF_RSP_ERROR); | |
4cc7c1cb IC |
884 | } |
885 | /* The head too, if necessary. */ | |
886 | if (start) | |
7376419a | 887 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
4cc7c1cb IC |
888 | |
889 | return NULL; | |
f942dc25 IC |
890 | } |
891 | ||
7376419a WL |
892 | static int xenvif_tx_check_gop(struct xenvif *vif, |
893 | struct sk_buff *skb, | |
894 | struct gnttab_copy **gopp) | |
f942dc25 IC |
895 | { |
896 | struct gnttab_copy *gop = *gopp; | |
ea066ad1 | 897 | u16 pending_idx = *((u16 *)skb->data); |
f942dc25 | 898 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 899 | struct pending_tx_info *tx_info; |
f942dc25 IC |
900 | int nr_frags = shinfo->nr_frags; |
901 | int i, err, start; | |
2810e5b9 | 902 | u16 peek; /* peek into next tx request */ |
f942dc25 IC |
903 | |
904 | /* Check status of header. */ | |
905 | err = gop->status; | |
7d5145d8 | 906 | if (unlikely(err)) |
7376419a | 907 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
908 | |
909 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 910 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 IC |
911 | |
912 | for (i = start; i < nr_frags; i++) { | |
913 | int j, newerr; | |
2810e5b9 | 914 | pending_ring_idx_t head; |
f942dc25 | 915 | |
ea066ad1 | 916 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 917 | tx_info = &vif->pending_tx_info[pending_idx]; |
2810e5b9 | 918 | head = tx_info->head; |
f942dc25 IC |
919 | |
920 | /* Check error status: if okay then remember grant handle. */ | |
2810e5b9 WL |
921 | do { |
922 | newerr = (++gop)->status; | |
923 | if (newerr) | |
924 | break; | |
b3f980bd WL |
925 | peek = vif->pending_ring[pending_index(++head)]; |
926 | } while (!pending_tx_is_head(vif, peek)); | |
2810e5b9 | 927 | |
f942dc25 IC |
928 | if (likely(!newerr)) { |
929 | /* Had a previous error? Invalidate this fragment. */ | |
930 | if (unlikely(err)) | |
7376419a WL |
931 | xenvif_idx_release(vif, pending_idx, |
932 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
933 | continue; |
934 | } | |
935 | ||
936 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 937 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
938 | |
939 | /* Not the first error? Preceding frags already invalidated. */ | |
940 | if (err) | |
941 | continue; | |
942 | ||
943 | /* First error: invalidate header and preceding fragments. */ | |
944 | pending_idx = *((u16 *)skb->data); | |
7376419a | 945 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 | 946 | for (j = start; j < i; j++) { |
5ccb3ea7 | 947 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
7376419a WL |
948 | xenvif_idx_release(vif, pending_idx, |
949 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
950 | } |
951 | ||
952 | /* Remember the error: invalidate all subsequent fragments. */ | |
953 | err = newerr; | |
954 | } | |
955 | ||
956 | *gopp = gop + 1; | |
957 | return err; | |
958 | } | |
959 | ||
7376419a | 960 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
961 | { |
962 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
963 | int nr_frags = shinfo->nr_frags; | |
964 | int i; | |
965 | ||
966 | for (i = 0; i < nr_frags; i++) { | |
967 | skb_frag_t *frag = shinfo->frags + i; | |
968 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
969 | struct page *page; |
970 | u16 pending_idx; | |
f942dc25 | 971 | |
ea066ad1 | 972 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 973 | |
b3f980bd WL |
974 | txp = &vif->pending_tx_info[pending_idx].req; |
975 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 976 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
977 | skb->len += txp->size; |
978 | skb->data_len += txp->size; | |
979 | skb->truesize += txp->size; | |
980 | ||
7376419a | 981 | /* Take an extra reference to offset xenvif_idx_release */ |
b3f980bd | 982 | get_page(vif->mmap_pages[pending_idx]); |
7376419a | 983 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
984 | } |
985 | } | |
986 | ||
7376419a | 987 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
988 | struct xen_netif_extra_info *extras, |
989 | int work_to_do) | |
990 | { | |
991 | struct xen_netif_extra_info extra; | |
992 | RING_IDX cons = vif->tx.req_cons; | |
993 | ||
994 | do { | |
995 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 996 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 997 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
998 | return -EBADR; |
999 | } | |
1000 | ||
1001 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1002 | sizeof(extra)); | |
1003 | if (unlikely(!extra.type || | |
1004 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1005 | vif->tx.req_cons = ++cons; | |
48856286 | 1006 | netdev_err(vif->dev, |
f942dc25 | 1007 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1008 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1009 | return -EINVAL; |
1010 | } | |
1011 | ||
1012 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1013 | vif->tx.req_cons = ++cons; | |
1014 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1015 | ||
1016 | return work_to_do; | |
1017 | } | |
1018 | ||
7376419a WL |
1019 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1020 | struct sk_buff *skb, | |
1021 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1022 | { |
1023 | if (!gso->u.gso.size) { | |
48856286 | 1024 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1025 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1026 | return -EINVAL; |
1027 | } | |
1028 | ||
a9468587 PD |
1029 | switch (gso->u.gso.type) { |
1030 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1031 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1032 | break; | |
1033 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1034 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1035 | break; | |
1036 | default: | |
48856286 | 1037 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1038 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1039 | return -EINVAL; |
1040 | } | |
1041 | ||
1042 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1043 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1044 | |
1045 | return 0; | |
1046 | } | |
1047 | ||
2eba61d5 PD |
1048 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) |
1049 | { | |
2721637c | 1050 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1051 | |
1052 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1053 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1054 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1055 | * recalculate the partial checksum. | |
1056 | */ | |
1057 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1058 | vif->rx_gso_checksum_fixup++; | |
1059 | skb->ip_summed = CHECKSUM_PARTIAL; | |
2721637c | 1060 | recalculate_partial_csum = true; |
2eba61d5 PD |
1061 | } |
1062 | ||
1063 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1064 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1065 | return 0; | |
1066 | ||
2721637c | 1067 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1068 | } |
1069 | ||
f942dc25 IC |
1070 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1071 | { | |
059dfa6a WL |
1072 | u64 now = get_jiffies_64(); |
1073 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1074 | msecs_to_jiffies(vif->credit_usec / 1000); |
1075 | ||
1076 | /* Timer could already be pending in rare cases. */ | |
1077 | if (timer_pending(&vif->credit_timeout)) | |
1078 | return true; | |
1079 | ||
1080 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1081 | if (time_after_eq64(now, next_credit)) { |
1082 | vif->credit_window_start = now; | |
f942dc25 IC |
1083 | tx_add_credit(vif); |
1084 | } | |
1085 | ||
1086 | /* Still too big to send right now? Set a callback. */ | |
1087 | if (size > vif->remaining_credit) { | |
1088 | vif->credit_timeout.data = | |
1089 | (unsigned long)vif; | |
1090 | vif->credit_timeout.function = | |
1091 | tx_credit_callback; | |
1092 | mod_timer(&vif->credit_timeout, | |
1093 | next_credit); | |
059dfa6a | 1094 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1095 | |
1096 | return true; | |
1097 | } | |
1098 | ||
1099 | return false; | |
1100 | } | |
1101 | ||
10574059 | 1102 | static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget) |
f942dc25 | 1103 | { |
b3f980bd | 1104 | struct gnttab_copy *gop = vif->tx_copy_ops, *request_gop; |
f942dc25 IC |
1105 | struct sk_buff *skb; |
1106 | int ret; | |
1107 | ||
b3f980bd | 1108 | while ((nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX |
10574059 PD |
1109 | < MAX_PENDING_REQS) && |
1110 | (skb_queue_len(&vif->tx_queue) < budget)) { | |
f942dc25 | 1111 | struct xen_netif_tx_request txreq; |
37641494 | 1112 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1113 | struct page *page; |
1114 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; | |
1115 | u16 pending_idx; | |
1116 | RING_IDX idx; | |
1117 | int work_to_do; | |
1118 | unsigned int data_len; | |
1119 | pending_ring_idx_t index; | |
1120 | ||
48856286 IC |
1121 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1122 | XEN_NETIF_TX_RING_SIZE) { | |
1123 | netdev_err(vif->dev, | |
1124 | "Impossible number of requests. " | |
1125 | "req_prod %d, req_cons %d, size %ld\n", | |
1126 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1127 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1128 | xenvif_fatal_tx_err(vif); |
48856286 IC |
1129 | continue; |
1130 | } | |
1131 | ||
d9601a36 | 1132 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
b3f980bd WL |
1133 | if (!work_to_do) |
1134 | break; | |
f942dc25 IC |
1135 | |
1136 | idx = vif->tx.req_cons; | |
1137 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1138 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1139 | ||
1140 | /* Credit-based scheduling. */ | |
1141 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1142 | tx_credit_exceeded(vif, txreq.size)) |
1143 | break; | |
f942dc25 IC |
1144 | |
1145 | vif->remaining_credit -= txreq.size; | |
1146 | ||
1147 | work_to_do--; | |
1148 | vif->tx.req_cons = ++idx; | |
1149 | ||
1150 | memset(extras, 0, sizeof(extras)); | |
1151 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1152 | work_to_do = xenvif_get_extras(vif, extras, |
1153 | work_to_do); | |
f942dc25 | 1154 | idx = vif->tx.req_cons; |
48856286 | 1155 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1156 | break; |
f942dc25 IC |
1157 | } |
1158 | ||
7376419a | 1159 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1160 | if (unlikely(ret < 0)) |
b3f980bd | 1161 | break; |
48856286 | 1162 | |
f942dc25 IC |
1163 | idx += ret; |
1164 | ||
1165 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1166 | netdev_dbg(vif->dev, | |
1167 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1168 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1169 | break; |
f942dc25 IC |
1170 | } |
1171 | ||
1172 | /* No crossing a page as the payload mustn't fragment. */ | |
1173 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1174 | netdev_err(vif->dev, |
f942dc25 IC |
1175 | "txreq.offset: %x, size: %u, end: %lu\n", |
1176 | txreq.offset, txreq.size, | |
1177 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1178 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1179 | break; |
f942dc25 IC |
1180 | } |
1181 | ||
b3f980bd WL |
1182 | index = pending_index(vif->pending_cons); |
1183 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1184 | |
1185 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1186 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1187 | PKT_PROT_LEN : txreq.size; |
1188 | ||
1189 | skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN, | |
1190 | GFP_ATOMIC | __GFP_NOWARN); | |
1191 | if (unlikely(skb == NULL)) { | |
1192 | netdev_dbg(vif->dev, | |
1193 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1194 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1195 | break; |
1196 | } | |
1197 | ||
1198 | /* Packets passed to netif_rx() must have some headroom. */ | |
1199 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
1200 | ||
1201 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1202 | struct xen_netif_extra_info *gso; | |
1203 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1204 | ||
7376419a WL |
1205 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1206 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1207 | kfree_skb(skb); |
b3f980bd | 1208 | break; |
f942dc25 IC |
1209 | } |
1210 | } | |
1211 | ||
1212 | /* XXX could copy straight to head */ | |
7376419a | 1213 | page = xenvif_alloc_page(vif, pending_idx); |
f942dc25 IC |
1214 | if (!page) { |
1215 | kfree_skb(skb); | |
7376419a | 1216 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1217 | break; |
f942dc25 IC |
1218 | } |
1219 | ||
f942dc25 IC |
1220 | gop->source.u.ref = txreq.gref; |
1221 | gop->source.domid = vif->domid; | |
1222 | gop->source.offset = txreq.offset; | |
1223 | ||
1224 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1225 | gop->dest.domid = DOMID_SELF; | |
1226 | gop->dest.offset = txreq.offset; | |
1227 | ||
1228 | gop->len = txreq.size; | |
1229 | gop->flags = GNTCOPY_source_gref; | |
1230 | ||
1231 | gop++; | |
1232 | ||
b3f980bd | 1233 | memcpy(&vif->pending_tx_info[pending_idx].req, |
f942dc25 | 1234 | &txreq, sizeof(txreq)); |
b3f980bd | 1235 | vif->pending_tx_info[pending_idx].head = index; |
f942dc25 IC |
1236 | *((u16 *)skb->data) = pending_idx; |
1237 | ||
1238 | __skb_put(skb, data_len); | |
1239 | ||
1240 | skb_shinfo(skb)->nr_frags = ret; | |
1241 | if (data_len < txreq.size) { | |
1242 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1243 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1244 | pending_idx); | |
f942dc25 | 1245 | } else { |
ea066ad1 IC |
1246 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1247 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1248 | } |
1249 | ||
b3f980bd | 1250 | vif->pending_cons++; |
f942dc25 | 1251 | |
7376419a | 1252 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1253 | if (request_gop == NULL) { |
1254 | kfree_skb(skb); | |
7376419a | 1255 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1256 | break; |
f942dc25 IC |
1257 | } |
1258 | gop = request_gop; | |
1259 | ||
b3f980bd | 1260 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1261 | |
f942dc25 | 1262 | vif->tx.req_cons = idx; |
f942dc25 | 1263 | |
b3f980bd | 1264 | if ((gop-vif->tx_copy_ops) >= ARRAY_SIZE(vif->tx_copy_ops)) |
f942dc25 IC |
1265 | break; |
1266 | } | |
1267 | ||
b3f980bd | 1268 | return gop - vif->tx_copy_ops; |
f942dc25 IC |
1269 | } |
1270 | ||
b3f980bd | 1271 | |
10574059 | 1272 | static int xenvif_tx_submit(struct xenvif *vif) |
f942dc25 | 1273 | { |
b3f980bd | 1274 | struct gnttab_copy *gop = vif->tx_copy_ops; |
f942dc25 | 1275 | struct sk_buff *skb; |
b3f980bd | 1276 | int work_done = 0; |
f942dc25 | 1277 | |
10574059 | 1278 | while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) { |
f942dc25 | 1279 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1280 | u16 pending_idx; |
1281 | unsigned data_len; | |
1282 | ||
1283 | pending_idx = *((u16 *)skb->data); | |
b3f980bd | 1284 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1285 | |
1286 | /* Check the remap error code. */ | |
7376419a | 1287 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1288 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1289 | skb_shinfo(skb)->nr_frags = 0; | |
1290 | kfree_skb(skb); | |
1291 | continue; | |
1292 | } | |
1293 | ||
1294 | data_len = skb->len; | |
1295 | memcpy(skb->data, | |
b3f980bd | 1296 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 IC |
1297 | data_len); |
1298 | if (data_len < txp->size) { | |
1299 | /* Append the packet payload as a fragment. */ | |
1300 | txp->offset += data_len; | |
1301 | txp->size -= data_len; | |
1302 | } else { | |
1303 | /* Schedule a response immediately. */ | |
7376419a WL |
1304 | xenvif_idx_release(vif, pending_idx, |
1305 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1306 | } |
1307 | ||
1308 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1309 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1310 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1311 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1312 | ||
7376419a | 1313 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1314 | |
2eba61d5 | 1315 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1316 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1317 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1318 | } | |
1319 | ||
1320 | skb->dev = vif->dev; | |
1321 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1322 | skb_reset_network_header(skb); |
f942dc25 IC |
1323 | |
1324 | if (checksum_setup(vif, skb)) { | |
1325 | netdev_dbg(vif->dev, | |
1326 | "Can't setup checksum in net_tx_action\n"); | |
1327 | kfree_skb(skb); | |
1328 | continue; | |
1329 | } | |
1330 | ||
40893fd0 | 1331 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1332 | |
b89587a7 PD |
1333 | /* If the packet is GSO then we will have just set up the |
1334 | * transport header offset in checksum_setup so it's now | |
1335 | * straightforward to calculate gso_segs. | |
1336 | */ | |
1337 | if (skb_is_gso(skb)) { | |
1338 | int mss = skb_shinfo(skb)->gso_size; | |
1339 | int hdrlen = skb_transport_header(skb) - | |
1340 | skb_mac_header(skb) + | |
1341 | tcp_hdrlen(skb); | |
1342 | ||
1343 | skb_shinfo(skb)->gso_segs = | |
1344 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1345 | } | |
1346 | ||
f942dc25 IC |
1347 | vif->dev->stats.rx_bytes += skb->len; |
1348 | vif->dev->stats.rx_packets++; | |
1349 | ||
b3f980bd WL |
1350 | work_done++; |
1351 | ||
1352 | netif_receive_skb(skb); | |
f942dc25 | 1353 | } |
b3f980bd WL |
1354 | |
1355 | return work_done; | |
f942dc25 IC |
1356 | } |
1357 | ||
1358 | /* Called after netfront has transmitted */ | |
7376419a | 1359 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1360 | { |
1361 | unsigned nr_gops; | |
b3f980bd | 1362 | int work_done; |
f942dc25 | 1363 | |
b3f980bd WL |
1364 | if (unlikely(!tx_work_todo(vif))) |
1365 | return 0; | |
1366 | ||
10574059 | 1367 | nr_gops = xenvif_tx_build_gops(vif, budget); |
f942dc25 IC |
1368 | |
1369 | if (nr_gops == 0) | |
b3f980bd WL |
1370 | return 0; |
1371 | ||
1372 | gnttab_batch_copy(vif->tx_copy_ops, nr_gops); | |
f942dc25 | 1373 | |
10574059 | 1374 | work_done = xenvif_tx_submit(vif); |
f942dc25 | 1375 | |
b3f980bd | 1376 | return work_done; |
f942dc25 IC |
1377 | } |
1378 | ||
7376419a WL |
1379 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1380 | u8 status) | |
f942dc25 | 1381 | { |
f942dc25 | 1382 | struct pending_tx_info *pending_tx_info; |
2810e5b9 WL |
1383 | pending_ring_idx_t head; |
1384 | u16 peek; /* peek into next tx request */ | |
1385 | ||
b3f980bd | 1386 | BUG_ON(vif->mmap_pages[pending_idx] == (void *)(~0UL)); |
f942dc25 IC |
1387 | |
1388 | /* Already complete? */ | |
b3f980bd | 1389 | if (vif->mmap_pages[pending_idx] == NULL) |
f942dc25 IC |
1390 | return; |
1391 | ||
b3f980bd | 1392 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 | 1393 | |
2810e5b9 | 1394 | head = pending_tx_info->head; |
f942dc25 | 1395 | |
b3f980bd WL |
1396 | BUG_ON(!pending_tx_is_head(vif, head)); |
1397 | BUG_ON(vif->pending_ring[pending_index(head)] != pending_idx); | |
f942dc25 | 1398 | |
2810e5b9 WL |
1399 | do { |
1400 | pending_ring_idx_t index; | |
1401 | pending_ring_idx_t idx = pending_index(head); | |
b3f980bd | 1402 | u16 info_idx = vif->pending_ring[idx]; |
f942dc25 | 1403 | |
b3f980bd | 1404 | pending_tx_info = &vif->pending_tx_info[info_idx]; |
2810e5b9 WL |
1405 | make_tx_response(vif, &pending_tx_info->req, status); |
1406 | ||
1407 | /* Setting any number other than | |
1408 | * INVALID_PENDING_RING_IDX indicates this slot is | |
1409 | * starting a new packet / ending a previous packet. | |
1410 | */ | |
1411 | pending_tx_info->head = 0; | |
1412 | ||
b3f980bd WL |
1413 | index = pending_index(vif->pending_prod++); |
1414 | vif->pending_ring[index] = vif->pending_ring[info_idx]; | |
f942dc25 | 1415 | |
b3f980bd | 1416 | peek = vif->pending_ring[pending_index(++head)]; |
2810e5b9 | 1417 | |
b3f980bd | 1418 | } while (!pending_tx_is_head(vif, peek)); |
2810e5b9 | 1419 | |
b3f980bd WL |
1420 | put_page(vif->mmap_pages[pending_idx]); |
1421 | vif->mmap_pages[pending_idx] = NULL; | |
f942dc25 IC |
1422 | } |
1423 | ||
2810e5b9 | 1424 | |
f942dc25 IC |
1425 | static void make_tx_response(struct xenvif *vif, |
1426 | struct xen_netif_tx_request *txp, | |
1427 | s8 st) | |
1428 | { | |
1429 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1430 | struct xen_netif_tx_response *resp; | |
1431 | int notify; | |
1432 | ||
1433 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1434 | resp->id = txp->id; | |
1435 | resp->status = st; | |
1436 | ||
1437 | if (txp->flags & XEN_NETTXF_extra_info) | |
1438 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1439 | ||
1440 | vif->tx.rsp_prod_pvt = ++i; | |
1441 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1442 | if (notify) | |
e1f00a69 | 1443 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1444 | } |
1445 | ||
1446 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1447 | u16 id, | |
1448 | s8 st, | |
1449 | u16 offset, | |
1450 | u16 size, | |
1451 | u16 flags) | |
1452 | { | |
1453 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1454 | struct xen_netif_rx_response *resp; | |
1455 | ||
1456 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1457 | resp->offset = offset; | |
1458 | resp->flags = flags; | |
1459 | resp->id = id; | |
1460 | resp->status = (s16)size; | |
1461 | if (st < 0) | |
1462 | resp->status = (s16)st; | |
1463 | ||
1464 | vif->rx.rsp_prod_pvt = ++i; | |
1465 | ||
1466 | return resp; | |
1467 | } | |
1468 | ||
b3f980bd | 1469 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1470 | { |
9ab9831b ZK |
1471 | return !skb_queue_empty(&vif->rx_queue) && |
1472 | xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots); | |
f942dc25 IC |
1473 | } |
1474 | ||
b3f980bd | 1475 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1476 | { |
1477 | ||
b3f980bd WL |
1478 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) && |
1479 | (nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX | |
1480 | < MAX_PENDING_REQS)) | |
f942dc25 IC |
1481 | return 1; |
1482 | ||
1483 | return 0; | |
1484 | } | |
1485 | ||
7376419a | 1486 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1487 | { |
c9d63699 DV |
1488 | if (vif->tx.sring) |
1489 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1490 | vif->tx.sring); | |
1491 | if (vif->rx.sring) | |
1492 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1493 | vif->rx.sring); | |
f942dc25 IC |
1494 | } |
1495 | ||
7376419a WL |
1496 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1497 | grant_ref_t tx_ring_ref, | |
1498 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1499 | { |
c9d63699 | 1500 | void *addr; |
f942dc25 IC |
1501 | struct xen_netif_tx_sring *txs; |
1502 | struct xen_netif_rx_sring *rxs; | |
1503 | ||
1504 | int err = -ENOMEM; | |
1505 | ||
c9d63699 DV |
1506 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1507 | tx_ring_ref, &addr); | |
1508 | if (err) | |
f942dc25 IC |
1509 | goto err; |
1510 | ||
c9d63699 | 1511 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1512 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1513 | ||
c9d63699 DV |
1514 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1515 | rx_ring_ref, &addr); | |
1516 | if (err) | |
f942dc25 | 1517 | goto err; |
f942dc25 | 1518 | |
c9d63699 | 1519 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1520 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1521 | ||
1522 | return 0; | |
1523 | ||
1524 | err: | |
7376419a | 1525 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1526 | return err; |
1527 | } | |
1528 | ||
ca2f09f2 PD |
1529 | void xenvif_stop_queue(struct xenvif *vif) |
1530 | { | |
1531 | if (!vif->can_queue) | |
1532 | return; | |
1533 | ||
1534 | netif_stop_queue(vif->dev); | |
1535 | } | |
1536 | ||
1537 | static void xenvif_start_queue(struct xenvif *vif) | |
1538 | { | |
1539 | if (xenvif_schedulable(vif)) | |
1540 | netif_wake_queue(vif->dev); | |
1541 | } | |
1542 | ||
7376419a | 1543 | int xenvif_kthread(void *data) |
b3f980bd WL |
1544 | { |
1545 | struct xenvif *vif = data; | |
ca2f09f2 | 1546 | struct sk_buff *skb; |
b3f980bd WL |
1547 | |
1548 | while (!kthread_should_stop()) { | |
1549 | wait_event_interruptible(vif->wq, | |
1550 | rx_work_todo(vif) || | |
1551 | kthread_should_stop()); | |
1552 | if (kthread_should_stop()) | |
1553 | break; | |
1554 | ||
ca2f09f2 | 1555 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 1556 | xenvif_rx_action(vif); |
b3f980bd | 1557 | |
ca2f09f2 PD |
1558 | if (skb_queue_empty(&vif->rx_queue) && |
1559 | netif_queue_stopped(vif->dev)) | |
1560 | xenvif_start_queue(vif); | |
1561 | ||
b3f980bd WL |
1562 | cond_resched(); |
1563 | } | |
1564 | ||
ca2f09f2 PD |
1565 | /* Bin any remaining skbs */ |
1566 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | |
1567 | dev_kfree_skb(skb); | |
1568 | ||
b3f980bd WL |
1569 | return 0; |
1570 | } | |
1571 | ||
f942dc25 IC |
1572 | static int __init netback_init(void) |
1573 | { | |
f942dc25 | 1574 | int rc = 0; |
f942dc25 | 1575 | |
2a14b244 | 1576 | if (!xen_domain()) |
f942dc25 IC |
1577 | return -ENODEV; |
1578 | ||
37641494 | 1579 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1580 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1581 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1582 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1583 | } |
1584 | ||
f942dc25 IC |
1585 | rc = xenvif_xenbus_init(); |
1586 | if (rc) | |
1587 | goto failed_init; | |
1588 | ||
1589 | return 0; | |
1590 | ||
1591 | failed_init: | |
f942dc25 | 1592 | return rc; |
f942dc25 IC |
1593 | } |
1594 | ||
1595 | module_init(netback_init); | |
1596 | ||
b103f358 WL |
1597 | static void __exit netback_fini(void) |
1598 | { | |
b103f358 | 1599 | xenvif_xenbus_fini(); |
b103f358 WL |
1600 | } |
1601 | module_exit(netback_fini); | |
1602 | ||
f942dc25 | 1603 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1604 | MODULE_ALIAS("xen-backend:vif"); |