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> | |
e3377f36 | 40 | #include <linux/highmem.h> |
f942dc25 IC |
41 | |
42 | #include <net/tcp.h> | |
43 | ||
ca981633 | 44 | #include <xen/xen.h> |
f942dc25 IC |
45 | #include <xen/events.h> |
46 | #include <xen/interface/memory.h> | |
47 | ||
48 | #include <asm/xen/hypercall.h> | |
49 | #include <asm/xen/page.h> | |
50 | ||
e1f00a69 WL |
51 | /* Provide an option to disable split event channels at load time as |
52 | * event channels are limited resource. Split event channels are | |
53 | * enabled by default. | |
54 | */ | |
55 | bool separate_tx_rx_irq = 1; | |
56 | module_param(separate_tx_rx_irq, bool, 0644); | |
57 | ||
09350788 | 58 | /* When guest ring is filled up, qdisc queues the packets for us, but we have |
0e59a4a5 | 59 | * to timeout them, otherwise other guests' packets can get stuck there |
09350788 ZK |
60 | */ |
61 | unsigned int rx_drain_timeout_msecs = 10000; | |
62 | module_param(rx_drain_timeout_msecs, uint, 0444); | |
63 | unsigned int rx_drain_timeout_jiffies; | |
64 | ||
2810e5b9 WL |
65 | /* |
66 | * This is the maximum slots a skb can have. If a guest sends a skb | |
67 | * which exceeds this limit it is considered malicious. | |
68 | */ | |
37641494 WL |
69 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
70 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
71 | module_param(fatal_skb_slots, uint, 0444); | |
72 | ||
7376419a WL |
73 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
74 | u8 status); | |
75 | ||
f942dc25 IC |
76 | static void make_tx_response(struct xenvif *vif, |
77 | struct xen_netif_tx_request *txp, | |
78 | s8 st); | |
b3f980bd WL |
79 | |
80 | static inline int tx_work_todo(struct xenvif *vif); | |
81 | static inline int rx_work_todo(struct xenvif *vif); | |
82 | ||
f942dc25 IC |
83 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
84 | u16 id, | |
85 | s8 st, | |
86 | u16 offset, | |
87 | u16 size, | |
88 | u16 flags); | |
89 | ||
b3f980bd | 90 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 91 | u16 idx) |
f942dc25 | 92 | { |
b3f980bd | 93 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
94 | } |
95 | ||
b3f980bd | 96 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 97 | u16 idx) |
f942dc25 | 98 | { |
b3f980bd | 99 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
100 | } |
101 | ||
7aceb47a ZK |
102 | #define callback_param(vif, pending_idx) \ |
103 | (vif->pending_tx_info[pending_idx].callback_struct) | |
104 | ||
f53c3fe8 ZK |
105 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
106 | */ | |
58375744 | 107 | static inline struct xenvif *ubuf_to_vif(const struct ubuf_info *ubuf) |
3e2234b3 | 108 | { |
f53c3fe8 ZK |
109 | u16 pending_idx = ubuf->desc; |
110 | struct pending_tx_info *temp = | |
111 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
112 | return container_of(temp - pending_idx, | |
113 | struct xenvif, | |
114 | pending_tx_info[0]); | |
3e2234b3 | 115 | } |
f53c3fe8 | 116 | |
2eba61d5 PD |
117 | /* This is a miniumum size for the linear area to avoid lots of |
118 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
119 | * value 128 was chosen as it covers all IPv4 and most likely | |
120 | * IPv6 headers. | |
f942dc25 | 121 | */ |
2eba61d5 | 122 | #define PKT_PROT_LEN 128 |
f942dc25 | 123 | |
ea066ad1 IC |
124 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
125 | { | |
126 | return (u16)frag->page_offset; | |
127 | } | |
128 | ||
129 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
130 | { | |
131 | frag->page_offset = pending_idx; | |
132 | } | |
133 | ||
f942dc25 IC |
134 | static inline pending_ring_idx_t pending_index(unsigned i) |
135 | { | |
136 | return i & (MAX_PENDING_REQS-1); | |
137 | } | |
138 | ||
ca2f09f2 | 139 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
f942dc25 | 140 | { |
ca2f09f2 | 141 | RING_IDX prod, cons; |
f942dc25 | 142 | |
ca2f09f2 PD |
143 | do { |
144 | prod = vif->rx.sring->req_prod; | |
145 | cons = vif->rx.req_cons; | |
f942dc25 | 146 | |
ca2f09f2 PD |
147 | if (prod - cons >= needed) |
148 | return true; | |
f942dc25 | 149 | |
ca2f09f2 | 150 | vif->rx.sring->req_event = prod + 1; |
f942dc25 | 151 | |
ca2f09f2 PD |
152 | /* Make sure event is visible before we check prod |
153 | * again. | |
154 | */ | |
155 | mb(); | |
156 | } while (vif->rx.sring->req_prod != prod); | |
f942dc25 | 157 | |
ca2f09f2 | 158 | return false; |
f942dc25 IC |
159 | } |
160 | ||
161 | /* | |
162 | * Returns true if we should start a new receive buffer instead of | |
163 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
164 | * bytes. | |
165 | */ | |
166 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
167 | { | |
168 | /* simple case: we have completely filled the current buffer. */ | |
169 | if (offset == MAX_BUFFER_OFFSET) | |
170 | return true; | |
171 | ||
172 | /* | |
173 | * complex case: start a fresh buffer if the current frag | |
174 | * would overflow the current buffer but only if: | |
175 | * (i) this frag would fit completely in the next buffer | |
176 | * and (ii) there is already some data in the current buffer | |
177 | * and (iii) this is not the head buffer. | |
178 | * | |
179 | * Where: | |
180 | * - (i) stops us splitting a frag into two copies | |
181 | * unless the frag is too large for a single buffer. | |
182 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
183 | * - (iii) stops us leaving the first buffer | |
184 | * empty. Strictly speaking this is already covered | |
185 | * by (ii) but is explicitly checked because | |
186 | * netfront relies on the first buffer being | |
187 | * non-empty and can crash otherwise. | |
188 | * | |
189 | * This means we will effectively linearise small | |
190 | * frags but do not needlessly split large buffers | |
191 | * into multiple copies tend to give large frags their | |
192 | * own buffers as before. | |
193 | */ | |
0576eddf PD |
194 | BUG_ON(size > MAX_BUFFER_OFFSET); |
195 | if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head) | |
f942dc25 IC |
196 | return true; |
197 | ||
198 | return false; | |
199 | } | |
200 | ||
f942dc25 IC |
201 | struct netrx_pending_operations { |
202 | unsigned copy_prod, copy_cons; | |
203 | unsigned meta_prod, meta_cons; | |
204 | struct gnttab_copy *copy; | |
b3f980bd | 205 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
206 | int copy_off; |
207 | grant_ref_t copy_gref; | |
208 | }; | |
209 | ||
b3f980bd WL |
210 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
211 | struct netrx_pending_operations *npo) | |
f942dc25 | 212 | { |
b3f980bd | 213 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
214 | struct xen_netif_rx_request *req; |
215 | ||
216 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
217 | ||
218 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 219 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
220 | meta->gso_size = 0; |
221 | meta->size = 0; | |
222 | meta->id = req->id; | |
223 | ||
224 | npo->copy_off = 0; | |
225 | npo->copy_gref = req->gref; | |
226 | ||
227 | return meta; | |
228 | } | |
229 | ||
33bc801d WL |
230 | /* |
231 | * Set up the grant operations for this fragment. If it's a flipping | |
232 | * interface, we also set up the unmap request from here. | |
233 | */ | |
7376419a WL |
234 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
235 | struct netrx_pending_operations *npo, | |
236 | struct page *page, unsigned long size, | |
3e2234b3 ZK |
237 | unsigned long offset, int *head, |
238 | struct xenvif *foreign_vif, | |
239 | grant_ref_t foreign_gref) | |
f942dc25 IC |
240 | { |
241 | struct gnttab_copy *copy_gop; | |
b3f980bd | 242 | struct xenvif_rx_meta *meta; |
f942dc25 | 243 | unsigned long bytes; |
5bd07670 | 244 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
245 | |
246 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 247 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
248 | |
249 | meta = npo->meta + npo->meta_prod - 1; | |
250 | ||
6a8ed462 IC |
251 | /* Skip unused frames from start of page */ |
252 | page += offset >> PAGE_SHIFT; | |
253 | offset &= ~PAGE_MASK; | |
254 | ||
f942dc25 | 255 | while (size > 0) { |
6a8ed462 | 256 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
257 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
258 | ||
6a8ed462 IC |
259 | bytes = PAGE_SIZE - offset; |
260 | ||
261 | if (bytes > size) | |
262 | bytes = size; | |
263 | ||
33bc801d | 264 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
265 | /* |
266 | * Netfront requires there to be some data in the head | |
267 | * buffer. | |
268 | */ | |
33bc801d | 269 | BUG_ON(*head); |
f942dc25 IC |
270 | |
271 | meta = get_next_rx_buffer(vif, npo); | |
272 | } | |
273 | ||
f942dc25 IC |
274 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
275 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
276 | ||
277 | copy_gop = npo->copy + npo->copy_prod++; | |
278 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
279 | copy_gop->len = bytes; |
280 | ||
3e2234b3 ZK |
281 | if (foreign_vif) { |
282 | copy_gop->source.domid = foreign_vif->domid; | |
283 | copy_gop->source.u.ref = foreign_gref; | |
284 | copy_gop->flags |= GNTCOPY_source_gref; | |
285 | } else { | |
286 | copy_gop->source.domid = DOMID_SELF; | |
287 | copy_gop->source.u.gmfn = | |
288 | virt_to_mfn(page_address(page)); | |
289 | } | |
f942dc25 | 290 | copy_gop->source.offset = offset; |
f942dc25 | 291 | |
b3f980bd | 292 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
293 | copy_gop->dest.offset = npo->copy_off; |
294 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
295 | |
296 | npo->copy_off += bytes; | |
297 | meta->size += bytes; | |
298 | ||
299 | offset += bytes; | |
300 | size -= bytes; | |
301 | ||
6a8ed462 IC |
302 | /* Next frame */ |
303 | if (offset == PAGE_SIZE && size) { | |
304 | BUG_ON(!PageCompound(page)); | |
305 | page++; | |
306 | offset = 0; | |
307 | } | |
308 | ||
f942dc25 | 309 | /* Leave a gap for the GSO descriptor. */ |
5bd07670 AL |
310 | if (skb_is_gso(skb)) { |
311 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
312 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
313 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
314 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
315 | } | |
82cada22 PD |
316 | |
317 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
318 | vif->rx.req_cons++; |
319 | ||
33bc801d | 320 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
321 | |
322 | } | |
323 | } | |
324 | ||
58375744 ZK |
325 | /* |
326 | * Find the grant ref for a given frag in a chain of struct ubuf_info's | |
327 | * skb: the skb itself | |
328 | * i: the frag's number | |
329 | * ubuf: a pointer to an element in the chain. It should not be NULL | |
330 | * | |
331 | * Returns a pointer to the element in the chain where the page were found. If | |
332 | * not found, returns NULL. | |
333 | * See the definition of callback_struct in common.h for more details about | |
334 | * the chain. | |
335 | */ | |
336 | static const struct ubuf_info *xenvif_find_gref(const struct sk_buff *const skb, | |
337 | const int i, | |
338 | const struct ubuf_info *ubuf) | |
339 | { | |
340 | struct xenvif *foreign_vif = ubuf_to_vif(ubuf); | |
341 | ||
342 | do { | |
343 | u16 pending_idx = ubuf->desc; | |
344 | ||
345 | if (skb_shinfo(skb)->frags[i].page.p == | |
346 | foreign_vif->mmap_pages[pending_idx]) | |
347 | break; | |
348 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
349 | } while (ubuf); | |
350 | ||
351 | return ubuf; | |
352 | } | |
353 | ||
f942dc25 IC |
354 | /* |
355 | * Prepare an SKB to be transmitted to the frontend. | |
356 | * | |
357 | * This function is responsible for allocating grant operations, meta | |
358 | * structures, etc. | |
359 | * | |
360 | * It returns the number of meta structures consumed. The number of | |
361 | * ring slots used is always equal to the number of meta slots used | |
362 | * plus the number of GSO descriptors used. Currently, we use either | |
363 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
364 | * frontend-side LRO). | |
365 | */ | |
7376419a WL |
366 | static int xenvif_gop_skb(struct sk_buff *skb, |
367 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
368 | { |
369 | struct xenvif *vif = netdev_priv(skb->dev); | |
370 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
371 | int i; | |
372 | struct xen_netif_rx_request *req; | |
b3f980bd | 373 | struct xenvif_rx_meta *meta; |
f942dc25 | 374 | unsigned char *data; |
33bc801d | 375 | int head = 1; |
f942dc25 | 376 | int old_meta_prod; |
82cada22 | 377 | int gso_type; |
58375744 ZK |
378 | const struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg; |
379 | const struct ubuf_info *const head_ubuf = ubuf; | |
f942dc25 IC |
380 | |
381 | old_meta_prod = npo->meta_prod; | |
382 | ||
5bd07670 AL |
383 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
384 | if (skb_is_gso(skb)) { | |
385 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
386 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
387 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
388 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
389 | } |
390 | ||
f942dc25 | 391 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 392 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
393 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
394 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 395 | meta->gso_type = gso_type; |
5bd07670 | 396 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 IC |
397 | meta->size = 0; |
398 | meta->id = req->id; | |
399 | } | |
400 | ||
401 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
402 | meta = npo->meta + npo->meta_prod++; | |
403 | ||
82cada22 PD |
404 | if ((1 << gso_type) & vif->gso_mask) { |
405 | meta->gso_type = gso_type; | |
5bd07670 | 406 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
407 | } else { |
408 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 409 | meta->gso_size = 0; |
82cada22 | 410 | } |
f942dc25 IC |
411 | |
412 | meta->size = 0; | |
413 | meta->id = req->id; | |
414 | npo->copy_off = 0; | |
415 | npo->copy_gref = req->gref; | |
416 | ||
417 | data = skb->data; | |
418 | while (data < skb_tail_pointer(skb)) { | |
419 | unsigned int offset = offset_in_page(data); | |
420 | unsigned int len = PAGE_SIZE - offset; | |
421 | ||
422 | if (data + len > skb_tail_pointer(skb)) | |
423 | len = skb_tail_pointer(skb) - data; | |
424 | ||
7376419a | 425 | xenvif_gop_frag_copy(vif, skb, npo, |
3e2234b3 ZK |
426 | virt_to_page(data), len, offset, &head, |
427 | NULL, | |
428 | 0); | |
f942dc25 IC |
429 | data += len; |
430 | } | |
431 | ||
432 | for (i = 0; i < nr_frags; i++) { | |
58375744 ZK |
433 | /* This variable also signals whether foreign_gref has a real |
434 | * value or not. | |
435 | */ | |
436 | struct xenvif *foreign_vif = NULL; | |
437 | grant_ref_t foreign_gref; | |
438 | ||
439 | if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) && | |
440 | (ubuf->callback == &xenvif_zerocopy_callback)) { | |
441 | const struct ubuf_info *const startpoint = ubuf; | |
442 | ||
443 | /* Ideally ubuf points to the chain element which | |
444 | * belongs to this frag. Or if frags were removed from | |
445 | * the beginning, then shortly before it. | |
446 | */ | |
447 | ubuf = xenvif_find_gref(skb, i, ubuf); | |
448 | ||
449 | /* Try again from the beginning of the list, if we | |
450 | * haven't tried from there. This only makes sense in | |
451 | * the unlikely event of reordering the original frags. | |
452 | * For injected local pages it's an unnecessary second | |
453 | * run. | |
454 | */ | |
455 | if (unlikely(!ubuf) && startpoint != head_ubuf) | |
456 | ubuf = xenvif_find_gref(skb, i, head_ubuf); | |
457 | ||
458 | if (likely(ubuf)) { | |
459 | u16 pending_idx = ubuf->desc; | |
460 | ||
461 | foreign_vif = ubuf_to_vif(ubuf); | |
462 | foreign_gref = foreign_vif->pending_tx_info[pending_idx].req.gref; | |
463 | /* Just a safety measure. If this was the last | |
464 | * element on the list, the for loop will | |
465 | * iterate again if a local page were added to | |
466 | * the end. Using head_ubuf here prevents the | |
467 | * second search on the chain. Or the original | |
468 | * frags changed order, but that's less likely. | |
469 | * In any way, ubuf shouldn't be NULL. | |
470 | */ | |
471 | ubuf = ubuf->ctx ? | |
472 | (struct ubuf_info *) ubuf->ctx : | |
473 | head_ubuf; | |
474 | } else | |
475 | /* This frag was a local page, added to the | |
476 | * array after the skb left netback. | |
477 | */ | |
478 | ubuf = head_ubuf; | |
479 | } | |
7376419a WL |
480 | xenvif_gop_frag_copy(vif, skb, npo, |
481 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
482 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
483 | skb_shinfo(skb)->frags[i].page_offset, | |
3e2234b3 ZK |
484 | &head, |
485 | foreign_vif, | |
58375744 | 486 | foreign_vif ? foreign_gref : UINT_MAX); |
f942dc25 IC |
487 | } |
488 | ||
489 | return npo->meta_prod - old_meta_prod; | |
490 | } | |
491 | ||
492 | /* | |
7376419a | 493 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
494 | * used to set up the operations on the top of |
495 | * netrx_pending_operations, which have since been done. Check that | |
496 | * they didn't give any errors and advance over them. | |
497 | */ | |
7376419a WL |
498 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
499 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
500 | { |
501 | struct gnttab_copy *copy_op; | |
502 | int status = XEN_NETIF_RSP_OKAY; | |
503 | int i; | |
504 | ||
505 | for (i = 0; i < nr_meta_slots; i++) { | |
506 | copy_op = npo->copy + npo->copy_cons++; | |
507 | if (copy_op->status != GNTST_okay) { | |
508 | netdev_dbg(vif->dev, | |
509 | "Bad status %d from copy to DOM%d.\n", | |
510 | copy_op->status, vif->domid); | |
511 | status = XEN_NETIF_RSP_ERROR; | |
512 | } | |
513 | } | |
514 | ||
515 | return status; | |
516 | } | |
517 | ||
7376419a WL |
518 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
519 | struct xenvif_rx_meta *meta, | |
520 | int nr_meta_slots) | |
f942dc25 IC |
521 | { |
522 | int i; | |
523 | unsigned long offset; | |
524 | ||
525 | /* No fragments used */ | |
526 | if (nr_meta_slots <= 1) | |
527 | return; | |
528 | ||
529 | nr_meta_slots--; | |
530 | ||
531 | for (i = 0; i < nr_meta_slots; i++) { | |
532 | int flags; | |
533 | if (i == nr_meta_slots - 1) | |
534 | flags = 0; | |
535 | else | |
536 | flags = XEN_NETRXF_more_data; | |
537 | ||
538 | offset = 0; | |
539 | make_rx_response(vif, meta[i].id, status, offset, | |
540 | meta[i].size, flags); | |
541 | } | |
542 | } | |
543 | ||
8f13dd96 | 544 | struct xenvif_rx_cb { |
33bc801d WL |
545 | int meta_slots_used; |
546 | }; | |
547 | ||
8f13dd96 ZK |
548 | #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) |
549 | ||
ca2f09f2 | 550 | void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
551 | { |
552 | wake_up(&vif->wq); | |
553 | } | |
554 | ||
ca2f09f2 | 555 | static void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 556 | { |
f942dc25 | 557 | s8 status; |
e1f00a69 | 558 | u16 flags; |
f942dc25 IC |
559 | struct xen_netif_rx_response *resp; |
560 | struct sk_buff_head rxq; | |
561 | struct sk_buff *skb; | |
562 | LIST_HEAD(notify); | |
563 | int ret; | |
f942dc25 | 564 | unsigned long offset; |
11b57f90 | 565 | bool need_to_notify = false; |
f942dc25 IC |
566 | |
567 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
568 | .copy = vif->grant_copy_op, |
569 | .meta = vif->meta, | |
f942dc25 IC |
570 | }; |
571 | ||
572 | skb_queue_head_init(&rxq); | |
573 | ||
b3f980bd | 574 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
9ab9831b | 575 | RING_IDX max_slots_needed; |
1425c7a4 PD |
576 | RING_IDX old_req_cons; |
577 | RING_IDX ring_slots_used; | |
ca2f09f2 PD |
578 | int i; |
579 | ||
580 | /* We need a cheap worse case estimate for the number of | |
581 | * slots we'll use. | |
582 | */ | |
583 | ||
584 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
585 | skb_headlen(skb), | |
586 | PAGE_SIZE); | |
587 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
588 | unsigned int size; | |
a02eb473 PD |
589 | unsigned int offset; |
590 | ||
ca2f09f2 | 591 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
a02eb473 PD |
592 | offset = skb_shinfo(skb)->frags[i].page_offset; |
593 | ||
594 | /* For a worse-case estimate we need to factor in | |
595 | * the fragment page offset as this will affect the | |
596 | * number of times xenvif_gop_frag_copy() will | |
597 | * call start_new_rx_buffer(). | |
598 | */ | |
599 | max_slots_needed += DIV_ROUND_UP(offset + size, | |
600 | PAGE_SIZE); | |
ca2f09f2 | 601 | } |
a02eb473 PD |
602 | |
603 | /* To avoid the estimate becoming too pessimal for some | |
604 | * frontends that limit posted rx requests, cap the estimate | |
605 | * at MAX_SKB_FRAGS. | |
606 | */ | |
607 | if (max_slots_needed > MAX_SKB_FRAGS) | |
608 | max_slots_needed = MAX_SKB_FRAGS; | |
609 | ||
610 | /* We may need one more slot for GSO metadata */ | |
5bd07670 AL |
611 | if (skb_is_gso(skb) && |
612 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
613 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | |
ca2f09f2 PD |
614 | max_slots_needed++; |
615 | ||
616 | /* If the skb may not fit then bail out now */ | |
617 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | |
618 | skb_queue_head(&vif->rx_queue, skb); | |
11b57f90 | 619 | need_to_notify = true; |
9ab9831b | 620 | vif->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 621 | break; |
9ab9831b ZK |
622 | } else |
623 | vif->rx_last_skb_slots = 0; | |
f942dc25 | 624 | |
1425c7a4 | 625 | old_req_cons = vif->rx.req_cons; |
8f13dd96 | 626 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo); |
1425c7a4 PD |
627 | ring_slots_used = vif->rx.req_cons - old_req_cons; |
628 | ||
629 | BUG_ON(ring_slots_used > max_slots_needed); | |
f942dc25 IC |
630 | |
631 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
632 | } |
633 | ||
b3f980bd | 634 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
635 | |
636 | if (!npo.copy_prod) | |
ca2f09f2 | 637 | goto done; |
f942dc25 | 638 | |
ac3d5ac2 | 639 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
b3f980bd | 640 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
641 | |
642 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 643 | |
82cada22 PD |
644 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
645 | vif->gso_prefix_mask) { | |
f942dc25 | 646 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 647 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
648 | |
649 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
650 | ||
b3f980bd WL |
651 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
652 | resp->id = vif->meta[npo.meta_cons].id; | |
8f13dd96 | 653 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
654 | |
655 | npo.meta_cons++; | |
8f13dd96 | 656 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
657 | } |
658 | ||
659 | ||
660 | vif->dev->stats.tx_bytes += skb->len; | |
661 | vif->dev->stats.tx_packets++; | |
662 | ||
8f13dd96 ZK |
663 | status = xenvif_check_gop(vif, |
664 | XENVIF_RX_CB(skb)->meta_slots_used, | |
665 | &npo); | |
f942dc25 | 666 | |
8f13dd96 | 667 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
668 | flags = 0; |
669 | else | |
670 | flags = XEN_NETRXF_more_data; | |
671 | ||
672 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
673 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
674 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
675 | /* remote but checksummed. */ | |
676 | flags |= XEN_NETRXF_data_validated; | |
677 | ||
678 | offset = 0; | |
b3f980bd | 679 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 680 | status, offset, |
b3f980bd | 681 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
682 | flags); |
683 | ||
82cada22 PD |
684 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
685 | vif->gso_mask) { | |
f942dc25 IC |
686 | struct xen_netif_extra_info *gso = |
687 | (struct xen_netif_extra_info *) | |
688 | RING_GET_RESPONSE(&vif->rx, | |
689 | vif->rx.rsp_prod_pvt++); | |
690 | ||
691 | resp->flags |= XEN_NETRXF_extra_info; | |
692 | ||
82cada22 | 693 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 694 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
695 | gso->u.gso.pad = 0; |
696 | gso->u.gso.features = 0; | |
697 | ||
698 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
699 | gso->flags = 0; | |
700 | } | |
701 | ||
7376419a WL |
702 | xenvif_add_frag_responses(vif, status, |
703 | vif->meta + npo.meta_cons + 1, | |
8f13dd96 | 704 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 IC |
705 | |
706 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 707 | |
11b57f90 | 708 | need_to_notify |= !!ret; |
b3f980bd | 709 | |
8f13dd96 | 710 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
711 | dev_kfree_skb(skb); |
712 | } | |
713 | ||
ca2f09f2 | 714 | done: |
b3f980bd | 715 | if (need_to_notify) |
e1f00a69 | 716 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
717 | } |
718 | ||
0d08fceb | 719 | void xenvif_napi_schedule_or_enable_events(struct xenvif *vif) |
f942dc25 IC |
720 | { |
721 | int more_to_do; | |
722 | ||
723 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
724 | ||
725 | if (more_to_do) | |
b3f980bd | 726 | napi_schedule(&vif->napi); |
f942dc25 IC |
727 | } |
728 | ||
729 | static void tx_add_credit(struct xenvif *vif) | |
730 | { | |
731 | unsigned long max_burst, max_credit; | |
732 | ||
733 | /* | |
734 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
735 | * Otherwise the interface can seize up due to insufficient credit. | |
736 | */ | |
737 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
738 | max_burst = min(max_burst, 131072UL); | |
739 | max_burst = max(max_burst, vif->credit_bytes); | |
740 | ||
741 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
742 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
743 | if (max_credit < vif->remaining_credit) | |
744 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
745 | ||
746 | vif->remaining_credit = min(max_credit, max_burst); | |
747 | } | |
748 | ||
749 | static void tx_credit_callback(unsigned long data) | |
750 | { | |
751 | struct xenvif *vif = (struct xenvif *)data; | |
752 | tx_add_credit(vif); | |
0d08fceb | 753 | xenvif_napi_schedule_or_enable_events(vif); |
f942dc25 IC |
754 | } |
755 | ||
7376419a WL |
756 | static void xenvif_tx_err(struct xenvif *vif, |
757 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
758 | { |
759 | RING_IDX cons = vif->tx.req_cons; | |
f53c3fe8 | 760 | unsigned long flags; |
f942dc25 IC |
761 | |
762 | do { | |
f53c3fe8 | 763 | spin_lock_irqsave(&vif->response_lock, flags); |
f942dc25 | 764 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); |
f53c3fe8 | 765 | spin_unlock_irqrestore(&vif->response_lock, flags); |
b9149729 | 766 | if (cons == end) |
f942dc25 IC |
767 | break; |
768 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
769 | } while (1); | |
770 | vif->tx.req_cons = cons; | |
f942dc25 IC |
771 | } |
772 | ||
7376419a | 773 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
774 | { |
775 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 WL |
776 | vif->disabled = true; |
777 | xenvif_kick_thread(vif); | |
48856286 IC |
778 | } |
779 | ||
7376419a WL |
780 | static int xenvif_count_requests(struct xenvif *vif, |
781 | struct xen_netif_tx_request *first, | |
782 | struct xen_netif_tx_request *txp, | |
783 | int work_to_do) | |
f942dc25 IC |
784 | { |
785 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
786 | int slots = 0; |
787 | int drop_err = 0; | |
59ccb4eb | 788 | int more_data; |
f942dc25 IC |
789 | |
790 | if (!(first->flags & XEN_NETTXF_more_data)) | |
791 | return 0; | |
792 | ||
793 | do { | |
59ccb4eb WL |
794 | struct xen_netif_tx_request dropped_tx = { 0 }; |
795 | ||
2810e5b9 WL |
796 | if (slots >= work_to_do) { |
797 | netdev_err(vif->dev, | |
798 | "Asked for %d slots but exceeds this limit\n", | |
799 | work_to_do); | |
7376419a | 800 | xenvif_fatal_tx_err(vif); |
35876b5f | 801 | return -ENODATA; |
f942dc25 IC |
802 | } |
803 | ||
2810e5b9 WL |
804 | /* This guest is really using too many slots and |
805 | * considered malicious. | |
806 | */ | |
37641494 | 807 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
808 | netdev_err(vif->dev, |
809 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 810 | slots, fatal_skb_slots); |
7376419a | 811 | xenvif_fatal_tx_err(vif); |
35876b5f | 812 | return -E2BIG; |
f942dc25 IC |
813 | } |
814 | ||
2810e5b9 | 815 | /* Xen network protocol had implicit dependency on |
37641494 WL |
816 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
817 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
818 | * same behavior as before. Any packet using more than | |
819 | * 18 slots but less than fatal_skb_slots slots is | |
820 | * dropped | |
2810e5b9 | 821 | */ |
37641494 | 822 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
823 | if (net_ratelimit()) |
824 | netdev_dbg(vif->dev, | |
825 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 826 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
827 | drop_err = -E2BIG; |
828 | } | |
829 | ||
59ccb4eb WL |
830 | if (drop_err) |
831 | txp = &dropped_tx; | |
832 | ||
2810e5b9 | 833 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 834 | sizeof(*txp)); |
03393fd5 WL |
835 | |
836 | /* If the guest submitted a frame >= 64 KiB then | |
837 | * first->size overflowed and following slots will | |
838 | * appear to be larger than the frame. | |
839 | * | |
840 | * This cannot be fatal error as there are buggy | |
841 | * frontends that do this. | |
842 | * | |
843 | * Consume all slots and drop the packet. | |
844 | */ | |
845 | if (!drop_err && txp->size > first->size) { | |
846 | if (net_ratelimit()) | |
847 | netdev_dbg(vif->dev, | |
848 | "Invalid tx request, slot size %u > remaining size %u\n", | |
849 | txp->size, first->size); | |
850 | drop_err = -EIO; | |
f942dc25 IC |
851 | } |
852 | ||
853 | first->size -= txp->size; | |
2810e5b9 | 854 | slots++; |
f942dc25 IC |
855 | |
856 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 857 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 858 | txp->offset, txp->size); |
7376419a | 859 | xenvif_fatal_tx_err(vif); |
35876b5f | 860 | return -EINVAL; |
f942dc25 | 861 | } |
59ccb4eb WL |
862 | |
863 | more_data = txp->flags & XEN_NETTXF_more_data; | |
864 | ||
865 | if (!drop_err) | |
866 | txp++; | |
867 | ||
868 | } while (more_data); | |
2810e5b9 WL |
869 | |
870 | if (drop_err) { | |
7376419a | 871 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
872 | return drop_err; |
873 | } | |
874 | ||
875 | return slots; | |
f942dc25 IC |
876 | } |
877 | ||
8f13dd96 ZK |
878 | |
879 | struct xenvif_tx_cb { | |
880 | u16 pending_idx; | |
881 | }; | |
882 | ||
883 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
884 | ||
9074ce24 ZK |
885 | static inline void xenvif_tx_create_map_op(struct xenvif *vif, |
886 | u16 pending_idx, | |
887 | struct xen_netif_tx_request *txp, | |
888 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 889 | { |
9074ce24 ZK |
890 | vif->pages_to_map[mop-vif->tx_map_ops] = vif->mmap_pages[pending_idx]; |
891 | gnttab_set_map_op(mop, idx_to_kaddr(vif, pending_idx), | |
f53c3fe8 ZK |
892 | GNTMAP_host_map | GNTMAP_readonly, |
893 | txp->gref, vif->domid); | |
894 | ||
895 | memcpy(&vif->pending_tx_info[pending_idx].req, txp, | |
896 | sizeof(*txp)); | |
897 | } | |
898 | ||
e3377f36 ZK |
899 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
900 | { | |
901 | struct sk_buff *skb = | |
902 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
903 | GFP_ATOMIC | __GFP_NOWARN); | |
904 | if (unlikely(skb == NULL)) | |
905 | return NULL; | |
906 | ||
907 | /* Packets passed to netif_rx() must have some headroom. */ | |
908 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
909 | ||
910 | /* Initialize it here to avoid later surprises */ | |
911 | skb_shinfo(skb)->destructor_arg = NULL; | |
912 | ||
913 | return skb; | |
914 | } | |
915 | ||
f53c3fe8 ZK |
916 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif *vif, |
917 | struct sk_buff *skb, | |
918 | struct xen_netif_tx_request *txp, | |
919 | struct gnttab_map_grant_ref *gop) | |
f942dc25 IC |
920 | { |
921 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
922 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 923 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
924 | int start; |
925 | pending_ring_idx_t index; | |
e3377f36 | 926 | unsigned int nr_slots, frag_overflow = 0; |
2810e5b9 WL |
927 | |
928 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 929 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 | 930 | */ |
e3377f36 ZK |
931 | if (shinfo->nr_frags > MAX_SKB_FRAGS) { |
932 | frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS; | |
933 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
934 | shinfo->nr_frags = MAX_SKB_FRAGS; | |
935 | } | |
2810e5b9 | 936 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
937 | |
938 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 939 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 940 | |
f53c3fe8 ZK |
941 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
942 | shinfo->nr_frags++, txp++, gop++) { | |
62bad319 ZK |
943 | index = pending_index(vif->pending_cons++); |
944 | pending_idx = vif->pending_ring[index]; | |
9074ce24 | 945 | xenvif_tx_create_map_op(vif, pending_idx, txp, gop); |
f53c3fe8 | 946 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
947 | } |
948 | ||
e3377f36 ZK |
949 | if (frag_overflow) { |
950 | struct sk_buff *nskb = xenvif_alloc_skb(0); | |
951 | if (unlikely(nskb == NULL)) { | |
952 | if (net_ratelimit()) | |
953 | netdev_err(vif->dev, | |
954 | "Can't allocate the frag_list skb.\n"); | |
955 | return NULL; | |
956 | } | |
957 | ||
958 | shinfo = skb_shinfo(nskb); | |
959 | frags = shinfo->frags; | |
960 | ||
961 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
962 | shinfo->nr_frags++, txp++, gop++) { | |
963 | index = pending_index(vif->pending_cons++); | |
964 | pending_idx = vif->pending_ring[index]; | |
9074ce24 | 965 | xenvif_tx_create_map_op(vif, pending_idx, txp, gop); |
e3377f36 ZK |
966 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
967 | pending_idx); | |
968 | } | |
969 | ||
970 | skb_shinfo(skb)->frag_list = nskb; | |
971 | } | |
2810e5b9 | 972 | |
f942dc25 IC |
973 | return gop; |
974 | } | |
975 | ||
f53c3fe8 ZK |
976 | static inline void xenvif_grant_handle_set(struct xenvif *vif, |
977 | u16 pending_idx, | |
978 | grant_handle_t handle) | |
979 | { | |
980 | if (unlikely(vif->grant_tx_handle[pending_idx] != | |
981 | NETBACK_INVALID_HANDLE)) { | |
982 | netdev_err(vif->dev, | |
983 | "Trying to overwrite active handle! pending_idx: %x\n", | |
984 | pending_idx); | |
985 | BUG(); | |
986 | } | |
987 | vif->grant_tx_handle[pending_idx] = handle; | |
988 | } | |
989 | ||
990 | static inline void xenvif_grant_handle_reset(struct xenvif *vif, | |
991 | u16 pending_idx) | |
992 | { | |
993 | if (unlikely(vif->grant_tx_handle[pending_idx] == | |
994 | NETBACK_INVALID_HANDLE)) { | |
995 | netdev_err(vif->dev, | |
996 | "Trying to unmap invalid handle! pending_idx: %x\n", | |
997 | pending_idx); | |
998 | BUG(); | |
999 | } | |
1000 | vif->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; | |
1001 | } | |
1002 | ||
7376419a WL |
1003 | static int xenvif_tx_check_gop(struct xenvif *vif, |
1004 | struct sk_buff *skb, | |
bdab8275 ZK |
1005 | struct gnttab_map_grant_ref **gopp_map, |
1006 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 1007 | { |
9074ce24 | 1008 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 1009 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
f942dc25 IC |
1010 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1011 | int nr_frags = shinfo->nr_frags; | |
bdab8275 | 1012 | int i, err; |
e3377f36 | 1013 | struct sk_buff *first_skb = NULL; |
f942dc25 IC |
1014 | |
1015 | /* Check status of header. */ | |
bdab8275 ZK |
1016 | err = (*gopp_copy)->status; |
1017 | (*gopp_copy)++; | |
1018 | if (unlikely(err)) { | |
1019 | if (net_ratelimit()) | |
1020 | netdev_dbg(vif->dev, | |
00aefceb | 1021 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1022 | (*gopp_copy)->status, |
1023 | pending_idx, | |
1024 | (*gopp_copy)->source.u.ref); | |
7376419a | 1025 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
bdab8275 | 1026 | } |
f942dc25 | 1027 | |
e3377f36 | 1028 | check_frags: |
bdab8275 | 1029 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 1030 | int j, newerr; |
f942dc25 | 1031 | |
ea066ad1 | 1032 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
1033 | |
1034 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 1035 | newerr = gop_map->status; |
2810e5b9 | 1036 | |
f942dc25 | 1037 | if (likely(!newerr)) { |
9074ce24 ZK |
1038 | xenvif_grant_handle_set(vif, |
1039 | pending_idx, | |
1040 | gop_map->handle); | |
f942dc25 IC |
1041 | /* Had a previous error? Invalidate this fragment. */ |
1042 | if (unlikely(err)) | |
f53c3fe8 | 1043 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
1044 | continue; |
1045 | } | |
1046 | ||
1047 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 ZK |
1048 | if (net_ratelimit()) |
1049 | netdev_dbg(vif->dev, | |
00aefceb | 1050 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1051 | i, |
1052 | gop_map->status, | |
1053 | pending_idx, | |
1054 | gop_map->ref); | |
7376419a | 1055 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1056 | |
1057 | /* Not the first error? Preceding frags already invalidated. */ | |
1058 | if (err) | |
1059 | continue; | |
bdab8275 ZK |
1060 | /* First error: invalidate preceding fragments. */ |
1061 | for (j = 0; j < i; j++) { | |
5ccb3ea7 | 1062 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
f53c3fe8 | 1063 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
1064 | } |
1065 | ||
1066 | /* Remember the error: invalidate all subsequent fragments. */ | |
1067 | err = newerr; | |
1068 | } | |
1069 | ||
e3377f36 ZK |
1070 | if (skb_has_frag_list(skb)) { |
1071 | first_skb = skb; | |
1072 | skb = shinfo->frag_list; | |
1073 | shinfo = skb_shinfo(skb); | |
1074 | nr_frags = shinfo->nr_frags; | |
e3377f36 ZK |
1075 | |
1076 | goto check_frags; | |
1077 | } | |
1078 | ||
1079 | /* There was a mapping error in the frag_list skb. We have to unmap | |
1080 | * the first skb's frags | |
1081 | */ | |
1082 | if (first_skb && err) { | |
1083 | int j; | |
1084 | shinfo = skb_shinfo(first_skb); | |
bdab8275 | 1085 | for (j = 0; j < shinfo->nr_frags; j++) { |
e3377f36 ZK |
1086 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
1087 | xenvif_idx_unmap(vif, pending_idx); | |
1088 | } | |
1089 | } | |
1090 | ||
bdab8275 | 1091 | *gopp_map = gop_map; |
f942dc25 IC |
1092 | return err; |
1093 | } | |
1094 | ||
7376419a | 1095 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
1096 | { |
1097 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1098 | int nr_frags = shinfo->nr_frags; | |
1099 | int i; | |
f53c3fe8 ZK |
1100 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1101 | ||
f942dc25 IC |
1102 | for (i = 0; i < nr_frags; i++) { |
1103 | skb_frag_t *frag = shinfo->frags + i; | |
1104 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1105 | struct page *page; |
1106 | u16 pending_idx; | |
f942dc25 | 1107 | |
ea066ad1 | 1108 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1109 | |
f53c3fe8 | 1110 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 1111 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 1112 | skb_shinfo(skb)->destructor_arg = |
7aceb47a | 1113 | &callback_param(vif, pending_idx); |
bdab8275 | 1114 | else |
7aceb47a ZK |
1115 | callback_param(vif, prev_pending_idx).ctx = |
1116 | &callback_param(vif, pending_idx); | |
f53c3fe8 | 1117 | |
7aceb47a | 1118 | callback_param(vif, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1119 | prev_pending_idx = pending_idx; |
1120 | ||
b3f980bd WL |
1121 | txp = &vif->pending_tx_info[pending_idx].req; |
1122 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 1123 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1124 | skb->len += txp->size; |
1125 | skb->data_len += txp->size; | |
1126 | skb->truesize += txp->size; | |
1127 | ||
f53c3fe8 | 1128 | /* Take an extra reference to offset network stack's put_page */ |
b3f980bd | 1129 | get_page(vif->mmap_pages[pending_idx]); |
f942dc25 | 1130 | } |
f53c3fe8 ZK |
1131 | /* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc |
1132 | * overlaps with "index", and "mapping" is not set. I think mapping | |
1133 | * should be set. If delivered to local stack, it would drop this | |
1134 | * skb in sk_filter unless the socket has the right to use it. | |
1135 | */ | |
1136 | skb->pfmemalloc = false; | |
f942dc25 IC |
1137 | } |
1138 | ||
7376419a | 1139 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
1140 | struct xen_netif_extra_info *extras, |
1141 | int work_to_do) | |
1142 | { | |
1143 | struct xen_netif_extra_info extra; | |
1144 | RING_IDX cons = vif->tx.req_cons; | |
1145 | ||
1146 | do { | |
1147 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 1148 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 1149 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1150 | return -EBADR; |
1151 | } | |
1152 | ||
1153 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1154 | sizeof(extra)); | |
1155 | if (unlikely(!extra.type || | |
1156 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1157 | vif->tx.req_cons = ++cons; | |
48856286 | 1158 | netdev_err(vif->dev, |
f942dc25 | 1159 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1160 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1161 | return -EINVAL; |
1162 | } | |
1163 | ||
1164 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1165 | vif->tx.req_cons = ++cons; | |
1166 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1167 | ||
1168 | return work_to_do; | |
1169 | } | |
1170 | ||
7376419a WL |
1171 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1172 | struct sk_buff *skb, | |
1173 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1174 | { |
1175 | if (!gso->u.gso.size) { | |
48856286 | 1176 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1177 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1178 | return -EINVAL; |
1179 | } | |
1180 | ||
a9468587 PD |
1181 | switch (gso->u.gso.type) { |
1182 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1183 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1184 | break; | |
1185 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1186 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1187 | break; | |
1188 | default: | |
48856286 | 1189 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1190 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1191 | return -EINVAL; |
1192 | } | |
1193 | ||
1194 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1195 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1196 | |
1197 | return 0; | |
1198 | } | |
1199 | ||
2eba61d5 PD |
1200 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) |
1201 | { | |
2721637c | 1202 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1203 | |
1204 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1205 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1206 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1207 | * recalculate the partial checksum. | |
1208 | */ | |
1209 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1210 | vif->rx_gso_checksum_fixup++; | |
1211 | skb->ip_summed = CHECKSUM_PARTIAL; | |
2721637c | 1212 | recalculate_partial_csum = true; |
2eba61d5 PD |
1213 | } |
1214 | ||
1215 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1216 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1217 | return 0; | |
1218 | ||
2721637c | 1219 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1220 | } |
1221 | ||
f942dc25 IC |
1222 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1223 | { | |
059dfa6a WL |
1224 | u64 now = get_jiffies_64(); |
1225 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1226 | msecs_to_jiffies(vif->credit_usec / 1000); |
1227 | ||
1228 | /* Timer could already be pending in rare cases. */ | |
1229 | if (timer_pending(&vif->credit_timeout)) | |
1230 | return true; | |
1231 | ||
1232 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1233 | if (time_after_eq64(now, next_credit)) { |
1234 | vif->credit_window_start = now; | |
f942dc25 IC |
1235 | tx_add_credit(vif); |
1236 | } | |
1237 | ||
1238 | /* Still too big to send right now? Set a callback. */ | |
1239 | if (size > vif->remaining_credit) { | |
1240 | vif->credit_timeout.data = | |
1241 | (unsigned long)vif; | |
1242 | vif->credit_timeout.function = | |
1243 | tx_credit_callback; | |
1244 | mod_timer(&vif->credit_timeout, | |
1245 | next_credit); | |
059dfa6a | 1246 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1247 | |
1248 | return true; | |
1249 | } | |
1250 | ||
1251 | return false; | |
1252 | } | |
1253 | ||
bdab8275 ZK |
1254 | static void xenvif_tx_build_gops(struct xenvif *vif, |
1255 | int budget, | |
1256 | unsigned *copy_ops, | |
1257 | unsigned *map_ops) | |
f942dc25 | 1258 | { |
f53c3fe8 | 1259 | struct gnttab_map_grant_ref *gop = vif->tx_map_ops, *request_gop; |
f942dc25 IC |
1260 | struct sk_buff *skb; |
1261 | int ret; | |
1262 | ||
869b9b19 | 1263 | while (skb_queue_len(&vif->tx_queue) < budget) { |
f942dc25 | 1264 | struct xen_netif_tx_request txreq; |
37641494 | 1265 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1266 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1267 | u16 pending_idx; | |
1268 | RING_IDX idx; | |
1269 | int work_to_do; | |
1270 | unsigned int data_len; | |
1271 | pending_ring_idx_t index; | |
1272 | ||
48856286 IC |
1273 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1274 | XEN_NETIF_TX_RING_SIZE) { | |
1275 | netdev_err(vif->dev, | |
1276 | "Impossible number of requests. " | |
1277 | "req_prod %d, req_cons %d, size %ld\n", | |
1278 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1279 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1280 | xenvif_fatal_tx_err(vif); |
e9d8b2c2 | 1281 | break; |
48856286 IC |
1282 | } |
1283 | ||
d9601a36 | 1284 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
b3f980bd WL |
1285 | if (!work_to_do) |
1286 | break; | |
f942dc25 IC |
1287 | |
1288 | idx = vif->tx.req_cons; | |
1289 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1290 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1291 | ||
1292 | /* Credit-based scheduling. */ | |
1293 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1294 | tx_credit_exceeded(vif, txreq.size)) |
1295 | break; | |
f942dc25 IC |
1296 | |
1297 | vif->remaining_credit -= txreq.size; | |
1298 | ||
1299 | work_to_do--; | |
1300 | vif->tx.req_cons = ++idx; | |
1301 | ||
1302 | memset(extras, 0, sizeof(extras)); | |
1303 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1304 | work_to_do = xenvif_get_extras(vif, extras, |
1305 | work_to_do); | |
f942dc25 | 1306 | idx = vif->tx.req_cons; |
48856286 | 1307 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1308 | break; |
f942dc25 IC |
1309 | } |
1310 | ||
7376419a | 1311 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1312 | if (unlikely(ret < 0)) |
b3f980bd | 1313 | break; |
48856286 | 1314 | |
f942dc25 IC |
1315 | idx += ret; |
1316 | ||
1317 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1318 | netdev_dbg(vif->dev, | |
1319 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1320 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1321 | break; |
f942dc25 IC |
1322 | } |
1323 | ||
1324 | /* No crossing a page as the payload mustn't fragment. */ | |
1325 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1326 | netdev_err(vif->dev, |
f942dc25 IC |
1327 | "txreq.offset: %x, size: %u, end: %lu\n", |
1328 | txreq.offset, txreq.size, | |
1329 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1330 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1331 | break; |
f942dc25 IC |
1332 | } |
1333 | ||
b3f980bd WL |
1334 | index = pending_index(vif->pending_cons); |
1335 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1336 | |
1337 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1338 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1339 | PKT_PROT_LEN : txreq.size; |
1340 | ||
e3377f36 | 1341 | skb = xenvif_alloc_skb(data_len); |
f942dc25 IC |
1342 | if (unlikely(skb == NULL)) { |
1343 | netdev_dbg(vif->dev, | |
1344 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1345 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1346 | break; |
1347 | } | |
1348 | ||
f942dc25 IC |
1349 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1350 | struct xen_netif_extra_info *gso; | |
1351 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1352 | ||
7376419a WL |
1353 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1354 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1355 | kfree_skb(skb); |
b3f980bd | 1356 | break; |
f942dc25 IC |
1357 | } |
1358 | } | |
1359 | ||
8f13dd96 | 1360 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1361 | |
1362 | __skb_put(skb, data_len); | |
bdab8275 ZK |
1363 | vif->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
1364 | vif->tx_copy_ops[*copy_ops].source.domid = vif->domid; | |
1365 | vif->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
1366 | ||
1367 | vif->tx_copy_ops[*copy_ops].dest.u.gmfn = | |
1368 | virt_to_mfn(skb->data); | |
1369 | vif->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; | |
1370 | vif->tx_copy_ops[*copy_ops].dest.offset = | |
1371 | offset_in_page(skb->data); | |
1372 | ||
1373 | vif->tx_copy_ops[*copy_ops].len = data_len; | |
1374 | vif->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
1375 | ||
1376 | (*copy_ops)++; | |
f942dc25 IC |
1377 | |
1378 | skb_shinfo(skb)->nr_frags = ret; | |
1379 | if (data_len < txreq.size) { | |
1380 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1381 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1382 | pending_idx); | |
bdab8275 ZK |
1383 | xenvif_tx_create_map_op(vif, pending_idx, &txreq, gop); |
1384 | gop++; | |
f942dc25 | 1385 | } else { |
ea066ad1 IC |
1386 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1387 | INVALID_PENDING_IDX); | |
bdab8275 ZK |
1388 | memcpy(&vif->pending_tx_info[pending_idx].req, &txreq, |
1389 | sizeof(txreq)); | |
f942dc25 IC |
1390 | } |
1391 | ||
b3f980bd | 1392 | vif->pending_cons++; |
f942dc25 | 1393 | |
7376419a | 1394 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1395 | if (request_gop == NULL) { |
1396 | kfree_skb(skb); | |
7376419a | 1397 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1398 | break; |
f942dc25 IC |
1399 | } |
1400 | gop = request_gop; | |
1401 | ||
b3f980bd | 1402 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1403 | |
f942dc25 | 1404 | vif->tx.req_cons = idx; |
f942dc25 | 1405 | |
bdab8275 ZK |
1406 | if (((gop-vif->tx_map_ops) >= ARRAY_SIZE(vif->tx_map_ops)) || |
1407 | (*copy_ops >= ARRAY_SIZE(vif->tx_copy_ops))) | |
f942dc25 IC |
1408 | break; |
1409 | } | |
1410 | ||
bdab8275 ZK |
1411 | (*map_ops) = gop - vif->tx_map_ops; |
1412 | return; | |
f942dc25 IC |
1413 | } |
1414 | ||
e3377f36 ZK |
1415 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1416 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1417 | */ | |
1418 | static int xenvif_handle_frag_list(struct xenvif *vif, struct sk_buff *skb) | |
1419 | { | |
1420 | unsigned int offset = skb_headlen(skb); | |
1421 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
1422 | int i; | |
1423 | struct ubuf_info *uarg; | |
1424 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1425 | ||
1426 | vif->tx_zerocopy_sent += 2; | |
1427 | vif->tx_frag_overflow++; | |
1428 | ||
1429 | xenvif_fill_frags(vif, nskb); | |
1430 | /* Subtract frags size, we will correct it later */ | |
1431 | skb->truesize -= skb->data_len; | |
1432 | skb->len += nskb->len; | |
1433 | skb->data_len += nskb->len; | |
1434 | ||
1435 | /* create a brand new frags array and coalesce there */ | |
1436 | for (i = 0; offset < skb->len; i++) { | |
1437 | struct page *page; | |
1438 | unsigned int len; | |
1439 | ||
1440 | BUG_ON(i >= MAX_SKB_FRAGS); | |
1441 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
1442 | if (!page) { | |
1443 | int j; | |
1444 | skb->truesize += skb->data_len; | |
1445 | for (j = 0; j < i; j++) | |
1446 | put_page(frags[j].page.p); | |
1447 | return -ENOMEM; | |
1448 | } | |
1449 | ||
1450 | if (offset + PAGE_SIZE < skb->len) | |
1451 | len = PAGE_SIZE; | |
1452 | else | |
1453 | len = skb->len - offset; | |
1454 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1455 | BUG(); | |
1456 | ||
1457 | offset += len; | |
1458 | frags[i].page.p = page; | |
1459 | frags[i].page_offset = 0; | |
1460 | skb_frag_size_set(&frags[i], len); | |
1461 | } | |
1462 | /* swap out with old one */ | |
1463 | memcpy(skb_shinfo(skb)->frags, | |
1464 | frags, | |
1465 | i * sizeof(skb_frag_t)); | |
1466 | skb_shinfo(skb)->nr_frags = i; | |
1467 | skb->truesize += i * PAGE_SIZE; | |
1468 | ||
1469 | /* remove traces of mapped pages and frag_list */ | |
1470 | skb_frag_list_init(skb); | |
1471 | uarg = skb_shinfo(skb)->destructor_arg; | |
1472 | uarg->callback(uarg, true); | |
1473 | skb_shinfo(skb)->destructor_arg = NULL; | |
1474 | ||
1475 | skb_shinfo(nskb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1476 | kfree_skb(nskb); | |
1477 | ||
1478 | return 0; | |
1479 | } | |
b3f980bd | 1480 | |
10574059 | 1481 | static int xenvif_tx_submit(struct xenvif *vif) |
f942dc25 | 1482 | { |
9074ce24 | 1483 | struct gnttab_map_grant_ref *gop_map = vif->tx_map_ops; |
bdab8275 | 1484 | struct gnttab_copy *gop_copy = vif->tx_copy_ops; |
f942dc25 | 1485 | struct sk_buff *skb; |
b3f980bd | 1486 | int work_done = 0; |
f942dc25 | 1487 | |
10574059 | 1488 | while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) { |
f942dc25 | 1489 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1490 | u16 pending_idx; |
1491 | unsigned data_len; | |
1492 | ||
8f13dd96 | 1493 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
b3f980bd | 1494 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1495 | |
1496 | /* Check the remap error code. */ | |
bdab8275 | 1497 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop_map, &gop_copy))) { |
f942dc25 IC |
1498 | skb_shinfo(skb)->nr_frags = 0; |
1499 | kfree_skb(skb); | |
1500 | continue; | |
1501 | } | |
1502 | ||
1503 | data_len = skb->len; | |
7aceb47a | 1504 | callback_param(vif, pending_idx).ctx = NULL; |
f942dc25 IC |
1505 | if (data_len < txp->size) { |
1506 | /* Append the packet payload as a fragment. */ | |
1507 | txp->offset += data_len; | |
1508 | txp->size -= data_len; | |
1509 | } else { | |
1510 | /* Schedule a response immediately. */ | |
bdab8275 ZK |
1511 | xenvif_idx_release(vif, pending_idx, |
1512 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1513 | } |
1514 | ||
1515 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1516 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1517 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1518 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1519 | ||
7376419a | 1520 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1521 | |
e3377f36 ZK |
1522 | if (unlikely(skb_has_frag_list(skb))) { |
1523 | if (xenvif_handle_frag_list(vif, skb)) { | |
1524 | if (net_ratelimit()) | |
1525 | netdev_err(vif->dev, | |
1526 | "Not enough memory to consolidate frag_list!\n"); | |
1527 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1528 | kfree_skb(skb); | |
1529 | continue; | |
1530 | } | |
1531 | } | |
1532 | ||
2eba61d5 | 1533 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1534 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1535 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1536 | } | |
1537 | ||
1538 | skb->dev = vif->dev; | |
1539 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1540 | skb_reset_network_header(skb); |
f942dc25 IC |
1541 | |
1542 | if (checksum_setup(vif, skb)) { | |
1543 | netdev_dbg(vif->dev, | |
1544 | "Can't setup checksum in net_tx_action\n"); | |
f53c3fe8 ZK |
1545 | /* We have to set this flag to trigger the callback */ |
1546 | if (skb_shinfo(skb)->destructor_arg) | |
1547 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
f942dc25 IC |
1548 | kfree_skb(skb); |
1549 | continue; | |
1550 | } | |
1551 | ||
40893fd0 | 1552 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1553 | |
b89587a7 PD |
1554 | /* If the packet is GSO then we will have just set up the |
1555 | * transport header offset in checksum_setup so it's now | |
1556 | * straightforward to calculate gso_segs. | |
1557 | */ | |
1558 | if (skb_is_gso(skb)) { | |
1559 | int mss = skb_shinfo(skb)->gso_size; | |
1560 | int hdrlen = skb_transport_header(skb) - | |
1561 | skb_mac_header(skb) + | |
1562 | tcp_hdrlen(skb); | |
1563 | ||
1564 | skb_shinfo(skb)->gso_segs = | |
1565 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1566 | } | |
1567 | ||
f942dc25 IC |
1568 | vif->dev->stats.rx_bytes += skb->len; |
1569 | vif->dev->stats.rx_packets++; | |
1570 | ||
b3f980bd WL |
1571 | work_done++; |
1572 | ||
f53c3fe8 ZK |
1573 | /* Set this flag right before netif_receive_skb, otherwise |
1574 | * someone might think this packet already left netback, and | |
1575 | * do a skb_copy_ubufs while we are still in control of the | |
1576 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1577 | */ | |
1bb332af | 1578 | if (skb_shinfo(skb)->destructor_arg) { |
f53c3fe8 | 1579 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
1bb332af ZK |
1580 | vif->tx_zerocopy_sent++; |
1581 | } | |
f53c3fe8 | 1582 | |
b3f980bd | 1583 | netif_receive_skb(skb); |
f942dc25 | 1584 | } |
b3f980bd WL |
1585 | |
1586 | return work_done; | |
f942dc25 IC |
1587 | } |
1588 | ||
3e2234b3 ZK |
1589 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1590 | { | |
f53c3fe8 ZK |
1591 | unsigned long flags; |
1592 | pending_ring_idx_t index; | |
1593 | struct xenvif *vif = ubuf_to_vif(ubuf); | |
1594 | ||
1595 | /* This is the only place where we grab this lock, to protect callbacks | |
1596 | * from each other. | |
1597 | */ | |
1598 | spin_lock_irqsave(&vif->callback_lock, flags); | |
1599 | do { | |
1600 | u16 pending_idx = ubuf->desc; | |
1601 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
1602 | BUG_ON(vif->dealloc_prod - vif->dealloc_cons >= | |
1603 | MAX_PENDING_REQS); | |
1604 | index = pending_index(vif->dealloc_prod); | |
1605 | vif->dealloc_ring[index] = pending_idx; | |
1606 | /* Sync with xenvif_tx_dealloc_action: | |
1607 | * insert idx then incr producer. | |
1608 | */ | |
1609 | smp_wmb(); | |
1610 | vif->dealloc_prod++; | |
1611 | } while (ubuf); | |
1612 | wake_up(&vif->dealloc_wq); | |
1613 | spin_unlock_irqrestore(&vif->callback_lock, flags); | |
1614 | ||
1bb332af ZK |
1615 | if (likely(zerocopy_success)) |
1616 | vif->tx_zerocopy_success++; | |
1617 | else | |
1618 | vif->tx_zerocopy_fail++; | |
f53c3fe8 ZK |
1619 | } |
1620 | ||
1621 | static inline void xenvif_tx_dealloc_action(struct xenvif *vif) | |
1622 | { | |
1623 | struct gnttab_unmap_grant_ref *gop; | |
1624 | pending_ring_idx_t dc, dp; | |
1625 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1626 | unsigned int i = 0; | |
1627 | ||
1628 | dc = vif->dealloc_cons; | |
1629 | gop = vif->tx_unmap_ops; | |
1630 | ||
1631 | /* Free up any grants we have finished using */ | |
1632 | do { | |
1633 | dp = vif->dealloc_prod; | |
1634 | ||
1635 | /* Ensure we see all indices enqueued by all | |
1636 | * xenvif_zerocopy_callback(). | |
1637 | */ | |
1638 | smp_rmb(); | |
1639 | ||
1640 | while (dc != dp) { | |
1641 | BUG_ON(gop - vif->tx_unmap_ops > MAX_PENDING_REQS); | |
1642 | pending_idx = | |
1643 | vif->dealloc_ring[pending_index(dc++)]; | |
1644 | ||
1645 | pending_idx_release[gop-vif->tx_unmap_ops] = | |
1646 | pending_idx; | |
1647 | vif->pages_to_unmap[gop-vif->tx_unmap_ops] = | |
1648 | vif->mmap_pages[pending_idx]; | |
1649 | gnttab_set_unmap_op(gop, | |
1650 | idx_to_kaddr(vif, pending_idx), | |
1651 | GNTMAP_host_map, | |
1652 | vif->grant_tx_handle[pending_idx]); | |
f53c3fe8 ZK |
1653 | xenvif_grant_handle_reset(vif, pending_idx); |
1654 | ++gop; | |
1655 | } | |
1656 | ||
1657 | } while (dp != vif->dealloc_prod); | |
1658 | ||
1659 | vif->dealloc_cons = dc; | |
1660 | ||
1661 | if (gop - vif->tx_unmap_ops > 0) { | |
1662 | int ret; | |
1663 | ret = gnttab_unmap_refs(vif->tx_unmap_ops, | |
1664 | NULL, | |
1665 | vif->pages_to_unmap, | |
1666 | gop - vif->tx_unmap_ops); | |
1667 | if (ret) { | |
2c5f4f84 | 1668 | netdev_err(vif->dev, "Unmap fail: nr_ops %tx ret %d\n", |
f53c3fe8 ZK |
1669 | gop - vif->tx_unmap_ops, ret); |
1670 | for (i = 0; i < gop - vif->tx_unmap_ops; ++i) { | |
1671 | if (gop[i].status != GNTST_okay) | |
1672 | netdev_err(vif->dev, | |
1673 | " host_addr: %llx handle: %x status: %d\n", | |
1674 | gop[i].host_addr, | |
1675 | gop[i].handle, | |
1676 | gop[i].status); | |
1677 | } | |
1678 | BUG(); | |
1679 | } | |
1680 | } | |
1681 | ||
1682 | for (i = 0; i < gop - vif->tx_unmap_ops; ++i) | |
1683 | xenvif_idx_release(vif, pending_idx_release[i], | |
1684 | XEN_NETIF_RSP_OKAY); | |
3e2234b3 ZK |
1685 | } |
1686 | ||
f53c3fe8 | 1687 | |
f942dc25 | 1688 | /* Called after netfront has transmitted */ |
7376419a | 1689 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 | 1690 | { |
bdab8275 | 1691 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1692 | int work_done, ret; |
f942dc25 | 1693 | |
b3f980bd WL |
1694 | if (unlikely(!tx_work_todo(vif))) |
1695 | return 0; | |
1696 | ||
bdab8275 | 1697 | xenvif_tx_build_gops(vif, budget, &nr_cops, &nr_mops); |
f942dc25 | 1698 | |
bdab8275 | 1699 | if (nr_cops == 0) |
b3f980bd WL |
1700 | return 0; |
1701 | ||
bdab8275 ZK |
1702 | gnttab_batch_copy(vif->tx_copy_ops, nr_cops); |
1703 | if (nr_mops != 0) { | |
1704 | ret = gnttab_map_refs(vif->tx_map_ops, | |
1705 | NULL, | |
1706 | vif->pages_to_map, | |
1707 | nr_mops); | |
1708 | BUG_ON(ret); | |
1709 | } | |
f942dc25 | 1710 | |
10574059 | 1711 | work_done = xenvif_tx_submit(vif); |
f942dc25 | 1712 | |
b3f980bd | 1713 | return work_done; |
f942dc25 IC |
1714 | } |
1715 | ||
7376419a WL |
1716 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1717 | u8 status) | |
f942dc25 | 1718 | { |
f942dc25 | 1719 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1720 | pending_ring_idx_t index; |
f53c3fe8 | 1721 | unsigned long flags; |
2810e5b9 | 1722 | |
62bad319 ZK |
1723 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
1724 | spin_lock_irqsave(&vif->response_lock, flags); | |
1725 | make_tx_response(vif, &pending_tx_info->req, status); | |
1726 | index = pending_index(vif->pending_prod); | |
1727 | vif->pending_ring[index] = pending_idx; | |
1728 | /* TX shouldn't use the index before we give it back here */ | |
1729 | mb(); | |
1730 | vif->pending_prod++; | |
1731 | spin_unlock_irqrestore(&vif->response_lock, flags); | |
f942dc25 IC |
1732 | } |
1733 | ||
2810e5b9 | 1734 | |
f942dc25 IC |
1735 | static void make_tx_response(struct xenvif *vif, |
1736 | struct xen_netif_tx_request *txp, | |
1737 | s8 st) | |
1738 | { | |
1739 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1740 | struct xen_netif_tx_response *resp; | |
1741 | int notify; | |
1742 | ||
1743 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1744 | resp->id = txp->id; | |
1745 | resp->status = st; | |
1746 | ||
1747 | if (txp->flags & XEN_NETTXF_extra_info) | |
1748 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1749 | ||
1750 | vif->tx.rsp_prod_pvt = ++i; | |
1751 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1752 | if (notify) | |
e1f00a69 | 1753 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1754 | } |
1755 | ||
1756 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1757 | u16 id, | |
1758 | s8 st, | |
1759 | u16 offset, | |
1760 | u16 size, | |
1761 | u16 flags) | |
1762 | { | |
1763 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1764 | struct xen_netif_rx_response *resp; | |
1765 | ||
1766 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1767 | resp->offset = offset; | |
1768 | resp->flags = flags; | |
1769 | resp->id = id; | |
1770 | resp->status = (s16)size; | |
1771 | if (st < 0) | |
1772 | resp->status = (s16)st; | |
1773 | ||
1774 | vif->rx.rsp_prod_pvt = ++i; | |
1775 | ||
1776 | return resp; | |
1777 | } | |
1778 | ||
f53c3fe8 ZK |
1779 | void xenvif_idx_unmap(struct xenvif *vif, u16 pending_idx) |
1780 | { | |
1781 | int ret; | |
1782 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1783 | ||
1784 | gnttab_set_unmap_op(&tx_unmap_op, | |
1785 | idx_to_kaddr(vif, pending_idx), | |
1786 | GNTMAP_host_map, | |
1787 | vif->grant_tx_handle[pending_idx]); | |
f53c3fe8 ZK |
1788 | xenvif_grant_handle_reset(vif, pending_idx); |
1789 | ||
1790 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
1791 | &vif->mmap_pages[pending_idx], 1); | |
7aceb47a ZK |
1792 | if (ret) { |
1793 | netdev_err(vif->dev, | |
1794 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: %x status: %d\n", | |
1795 | ret, | |
1796 | pending_idx, | |
1797 | tx_unmap_op.host_addr, | |
1798 | tx_unmap_op.handle, | |
1799 | tx_unmap_op.status); | |
1800 | BUG(); | |
1801 | } | |
f53c3fe8 ZK |
1802 | |
1803 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); | |
1804 | } | |
1805 | ||
b3f980bd | 1806 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1807 | { |
09350788 ZK |
1808 | return (!skb_queue_empty(&vif->rx_queue) && |
1809 | xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots)) || | |
1810 | vif->rx_queue_purge; | |
f942dc25 IC |
1811 | } |
1812 | ||
b3f980bd | 1813 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1814 | { |
1815 | ||
869b9b19 | 1816 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))) |
f942dc25 IC |
1817 | return 1; |
1818 | ||
1819 | return 0; | |
1820 | } | |
1821 | ||
f53c3fe8 ZK |
1822 | static inline bool tx_dealloc_work_todo(struct xenvif *vif) |
1823 | { | |
397dfd9f | 1824 | return vif->dealloc_cons != vif->dealloc_prod; |
f53c3fe8 ZK |
1825 | } |
1826 | ||
7376419a | 1827 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1828 | { |
c9d63699 DV |
1829 | if (vif->tx.sring) |
1830 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1831 | vif->tx.sring); | |
1832 | if (vif->rx.sring) | |
1833 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1834 | vif->rx.sring); | |
f942dc25 IC |
1835 | } |
1836 | ||
7376419a WL |
1837 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1838 | grant_ref_t tx_ring_ref, | |
1839 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1840 | { |
c9d63699 | 1841 | void *addr; |
f942dc25 IC |
1842 | struct xen_netif_tx_sring *txs; |
1843 | struct xen_netif_rx_sring *rxs; | |
1844 | ||
1845 | int err = -ENOMEM; | |
1846 | ||
c9d63699 DV |
1847 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1848 | tx_ring_ref, &addr); | |
1849 | if (err) | |
f942dc25 IC |
1850 | goto err; |
1851 | ||
c9d63699 | 1852 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1853 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1854 | ||
c9d63699 DV |
1855 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1856 | rx_ring_ref, &addr); | |
1857 | if (err) | |
f942dc25 | 1858 | goto err; |
f942dc25 | 1859 | |
c9d63699 | 1860 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1861 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1862 | ||
1863 | return 0; | |
1864 | ||
1865 | err: | |
7376419a | 1866 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1867 | return err; |
1868 | } | |
1869 | ||
ca2f09f2 PD |
1870 | void xenvif_stop_queue(struct xenvif *vif) |
1871 | { | |
1872 | if (!vif->can_queue) | |
1873 | return; | |
1874 | ||
1875 | netif_stop_queue(vif->dev); | |
1876 | } | |
1877 | ||
1878 | static void xenvif_start_queue(struct xenvif *vif) | |
1879 | { | |
1880 | if (xenvif_schedulable(vif)) | |
1881 | netif_wake_queue(vif->dev); | |
1882 | } | |
1883 | ||
121fa4b7 | 1884 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd WL |
1885 | { |
1886 | struct xenvif *vif = data; | |
ca2f09f2 | 1887 | struct sk_buff *skb; |
b3f980bd WL |
1888 | |
1889 | while (!kthread_should_stop()) { | |
1890 | wait_event_interruptible(vif->wq, | |
1891 | rx_work_todo(vif) || | |
e9d8b2c2 | 1892 | vif->disabled || |
b3f980bd | 1893 | kthread_should_stop()); |
e9d8b2c2 WL |
1894 | |
1895 | /* This frontend is found to be rogue, disable it in | |
1896 | * kthread context. Currently this is only set when | |
1897 | * netback finds out frontend sends malformed packet, | |
1898 | * but we cannot disable the interface in softirq | |
1899 | * context so we defer it here. | |
1900 | */ | |
1901 | if (unlikely(vif->disabled && netif_carrier_ok(vif->dev))) | |
1902 | xenvif_carrier_off(vif); | |
1903 | ||
b3f980bd WL |
1904 | if (kthread_should_stop()) |
1905 | break; | |
1906 | ||
09350788 ZK |
1907 | if (vif->rx_queue_purge) { |
1908 | skb_queue_purge(&vif->rx_queue); | |
1909 | vif->rx_queue_purge = false; | |
1910 | } | |
1911 | ||
ca2f09f2 | 1912 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 1913 | xenvif_rx_action(vif); |
b3f980bd | 1914 | |
ca2f09f2 | 1915 | if (skb_queue_empty(&vif->rx_queue) && |
09350788 ZK |
1916 | netif_queue_stopped(vif->dev)) { |
1917 | del_timer_sync(&vif->wake_queue); | |
ca2f09f2 | 1918 | xenvif_start_queue(vif); |
09350788 | 1919 | } |
ca2f09f2 | 1920 | |
b3f980bd WL |
1921 | cond_resched(); |
1922 | } | |
1923 | ||
ca2f09f2 PD |
1924 | /* Bin any remaining skbs */ |
1925 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | |
1926 | dev_kfree_skb(skb); | |
1927 | ||
b3f980bd WL |
1928 | return 0; |
1929 | } | |
1930 | ||
f53c3fe8 ZK |
1931 | int xenvif_dealloc_kthread(void *data) |
1932 | { | |
1933 | struct xenvif *vif = data; | |
1934 | ||
1935 | while (!kthread_should_stop()) { | |
1936 | wait_event_interruptible(vif->dealloc_wq, | |
1937 | tx_dealloc_work_todo(vif) || | |
1938 | kthread_should_stop()); | |
1939 | if (kthread_should_stop()) | |
1940 | break; | |
1941 | ||
1942 | xenvif_tx_dealloc_action(vif); | |
1943 | cond_resched(); | |
1944 | } | |
1945 | ||
1946 | /* Unmap anything remaining*/ | |
1947 | if (tx_dealloc_work_todo(vif)) | |
1948 | xenvif_tx_dealloc_action(vif); | |
1949 | ||
1950 | return 0; | |
1951 | } | |
1952 | ||
f942dc25 IC |
1953 | static int __init netback_init(void) |
1954 | { | |
f942dc25 | 1955 | int rc = 0; |
f942dc25 | 1956 | |
2a14b244 | 1957 | if (!xen_domain()) |
f942dc25 IC |
1958 | return -ENODEV; |
1959 | ||
37641494 | 1960 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1961 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1962 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1963 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1964 | } |
1965 | ||
f942dc25 IC |
1966 | rc = xenvif_xenbus_init(); |
1967 | if (rc) | |
1968 | goto failed_init; | |
1969 | ||
09350788 ZK |
1970 | rx_drain_timeout_jiffies = msecs_to_jiffies(rx_drain_timeout_msecs); |
1971 | ||
f942dc25 IC |
1972 | return 0; |
1973 | ||
1974 | failed_init: | |
f942dc25 | 1975 | return rc; |
f942dc25 IC |
1976 | } |
1977 | ||
1978 | module_init(netback_init); | |
1979 | ||
b103f358 WL |
1980 | static void __exit netback_fini(void) |
1981 | { | |
b103f358 | 1982 | xenvif_xenbus_fini(); |
b103f358 WL |
1983 | } |
1984 | module_exit(netback_fini); | |
1985 | ||
f942dc25 | 1986 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1987 | MODULE_ALIAS("xen-backend:vif"); |