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> | |
a9fd60e2 | 47 | #include <xen/page.h> |
f942dc25 IC |
48 | |
49 | #include <asm/xen/hypercall.h> | |
f942dc25 | 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 | */ | |
c489dbb1 | 55 | bool separate_tx_rx_irq = true; |
e1f00a69 WL |
56 | module_param(separate_tx_rx_irq, bool, 0644); |
57 | ||
f48da8b1 DV |
58 | /* The time that packets can stay on the guest Rx internal queue |
59 | * before they are dropped. | |
09350788 ZK |
60 | */ |
61 | unsigned int rx_drain_timeout_msecs = 10000; | |
62 | module_param(rx_drain_timeout_msecs, uint, 0444); | |
09350788 | 63 | |
ecf08d2d DV |
64 | /* The length of time before the frontend is considered unresponsive |
65 | * because it isn't providing Rx slots. | |
66 | */ | |
26c0e102 | 67 | unsigned int rx_stall_timeout_msecs = 60000; |
ecf08d2d | 68 | module_param(rx_stall_timeout_msecs, uint, 0444); |
ecf08d2d | 69 | |
8d3d53b3 AB |
70 | unsigned int xenvif_max_queues; |
71 | module_param_named(max_queues, xenvif_max_queues, uint, 0644); | |
72 | MODULE_PARM_DESC(max_queues, | |
73 | "Maximum number of queues per virtual interface"); | |
74 | ||
2810e5b9 WL |
75 | /* |
76 | * This is the maximum slots a skb can have. If a guest sends a skb | |
77 | * which exceeds this limit it is considered malicious. | |
78 | */ | |
37641494 WL |
79 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
80 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
81 | module_param(fatal_skb_slots, uint, 0444); | |
82 | ||
7e5d7753 MC |
83 | /* The amount to copy out of the first guest Tx slot into the skb's |
84 | * linear area. If the first slot has more data, it will be mapped | |
85 | * and put into the first frag. | |
86 | * | |
87 | * This is sized to avoid pulling headers from the frags for most | |
88 | * TCP/IP packets. | |
89 | */ | |
90 | #define XEN_NETBACK_TX_COPY_LEN 128 | |
91 | ||
92 | ||
e9ce7cb6 | 93 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a WL |
94 | u8 status); |
95 | ||
e9ce7cb6 | 96 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 IC |
97 | struct xen_netif_tx_request *txp, |
98 | s8 st); | |
c8a4d299 | 99 | static void push_tx_responses(struct xenvif_queue *queue); |
b3f980bd | 100 | |
e9ce7cb6 | 101 | static inline int tx_work_todo(struct xenvif_queue *queue); |
b3f980bd | 102 | |
e9ce7cb6 | 103 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
104 | u16 id, |
105 | s8 st, | |
106 | u16 offset, | |
107 | u16 size, | |
108 | u16 flags); | |
109 | ||
e9ce7cb6 | 110 | static inline unsigned long idx_to_pfn(struct xenvif_queue *queue, |
ea066ad1 | 111 | u16 idx) |
f942dc25 | 112 | { |
e9ce7cb6 | 113 | return page_to_pfn(queue->mmap_pages[idx]); |
f942dc25 IC |
114 | } |
115 | ||
e9ce7cb6 | 116 | static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue, |
ea066ad1 | 117 | u16 idx) |
f942dc25 | 118 | { |
e9ce7cb6 | 119 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx)); |
f942dc25 IC |
120 | } |
121 | ||
7aceb47a ZK |
122 | #define callback_param(vif, pending_idx) \ |
123 | (vif->pending_tx_info[pending_idx].callback_struct) | |
124 | ||
f53c3fe8 ZK |
125 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
126 | */ | |
e9ce7cb6 | 127 | static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf) |
3e2234b3 | 128 | { |
f53c3fe8 ZK |
129 | u16 pending_idx = ubuf->desc; |
130 | struct pending_tx_info *temp = | |
131 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
132 | return container_of(temp - pending_idx, | |
e9ce7cb6 | 133 | struct xenvif_queue, |
f53c3fe8 | 134 | pending_tx_info[0]); |
3e2234b3 | 135 | } |
f53c3fe8 | 136 | |
ea066ad1 IC |
137 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
138 | { | |
139 | return (u16)frag->page_offset; | |
140 | } | |
141 | ||
142 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
143 | { | |
144 | frag->page_offset = pending_idx; | |
145 | } | |
146 | ||
f942dc25 IC |
147 | static inline pending_ring_idx_t pending_index(unsigned i) |
148 | { | |
149 | return i & (MAX_PENDING_REQS-1); | |
150 | } | |
151 | ||
1d5d4852 | 152 | static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue) |
f942dc25 | 153 | { |
ca2f09f2 | 154 | RING_IDX prod, cons; |
99a2dea5 | 155 | struct sk_buff *skb; |
1d5d4852 DV |
156 | int needed; |
157 | ||
99a2dea5 DV |
158 | skb = skb_peek(&queue->rx_queue); |
159 | if (!skb) | |
160 | return false; | |
161 | ||
162 | needed = DIV_ROUND_UP(skb->len, XEN_PAGE_SIZE); | |
163 | if (skb_is_gso(skb)) | |
164 | needed++; | |
f942dc25 | 165 | |
ca2f09f2 | 166 | do { |
e9ce7cb6 WL |
167 | prod = queue->rx.sring->req_prod; |
168 | cons = queue->rx.req_cons; | |
f942dc25 | 169 | |
ca2f09f2 PD |
170 | if (prod - cons >= needed) |
171 | return true; | |
f942dc25 | 172 | |
e9ce7cb6 | 173 | queue->rx.sring->req_event = prod + 1; |
f942dc25 | 174 | |
ca2f09f2 PD |
175 | /* Make sure event is visible before we check prod |
176 | * again. | |
177 | */ | |
178 | mb(); | |
e9ce7cb6 | 179 | } while (queue->rx.sring->req_prod != prod); |
f942dc25 | 180 | |
ca2f09f2 | 181 | return false; |
f942dc25 IC |
182 | } |
183 | ||
f48da8b1 DV |
184 | void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb) |
185 | { | |
186 | unsigned long flags; | |
187 | ||
188 | spin_lock_irqsave(&queue->rx_queue.lock, flags); | |
189 | ||
190 | __skb_queue_tail(&queue->rx_queue, skb); | |
191 | ||
192 | queue->rx_queue_len += skb->len; | |
193 | if (queue->rx_queue_len > queue->rx_queue_max) | |
194 | netif_tx_stop_queue(netdev_get_tx_queue(queue->vif->dev, queue->id)); | |
195 | ||
196 | spin_unlock_irqrestore(&queue->rx_queue.lock, flags); | |
197 | } | |
198 | ||
199 | static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue) | |
200 | { | |
201 | struct sk_buff *skb; | |
202 | ||
203 | spin_lock_irq(&queue->rx_queue.lock); | |
204 | ||
205 | skb = __skb_dequeue(&queue->rx_queue); | |
206 | if (skb) | |
207 | queue->rx_queue_len -= skb->len; | |
208 | ||
209 | spin_unlock_irq(&queue->rx_queue.lock); | |
210 | ||
211 | return skb; | |
212 | } | |
213 | ||
214 | static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue) | |
215 | { | |
216 | spin_lock_irq(&queue->rx_queue.lock); | |
217 | ||
218 | if (queue->rx_queue_len < queue->rx_queue_max) | |
219 | netif_tx_wake_queue(netdev_get_tx_queue(queue->vif->dev, queue->id)); | |
220 | ||
221 | spin_unlock_irq(&queue->rx_queue.lock); | |
222 | } | |
223 | ||
224 | ||
225 | static void xenvif_rx_queue_purge(struct xenvif_queue *queue) | |
226 | { | |
227 | struct sk_buff *skb; | |
228 | while ((skb = xenvif_rx_dequeue(queue)) != NULL) | |
229 | kfree_skb(skb); | |
230 | } | |
231 | ||
232 | static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue) | |
233 | { | |
234 | struct sk_buff *skb; | |
235 | ||
236 | for(;;) { | |
237 | skb = skb_peek(&queue->rx_queue); | |
238 | if (!skb) | |
239 | break; | |
240 | if (time_before(jiffies, XENVIF_RX_CB(skb)->expires)) | |
241 | break; | |
242 | xenvif_rx_dequeue(queue); | |
243 | kfree_skb(skb); | |
244 | } | |
245 | } | |
246 | ||
f942dc25 IC |
247 | struct netrx_pending_operations { |
248 | unsigned copy_prod, copy_cons; | |
249 | unsigned meta_prod, meta_cons; | |
250 | struct gnttab_copy *copy; | |
b3f980bd | 251 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
252 | int copy_off; |
253 | grant_ref_t copy_gref; | |
254 | }; | |
255 | ||
e9ce7cb6 | 256 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue, |
b3f980bd | 257 | struct netrx_pending_operations *npo) |
f942dc25 | 258 | { |
b3f980bd | 259 | struct xenvif_rx_meta *meta; |
68a33bfd | 260 | struct xen_netif_rx_request req; |
f942dc25 | 261 | |
68a33bfd | 262 | RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req); |
f942dc25 IC |
263 | |
264 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 265 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
266 | meta->gso_size = 0; |
267 | meta->size = 0; | |
68a33bfd | 268 | meta->id = req.id; |
f942dc25 IC |
269 | |
270 | npo->copy_off = 0; | |
68a33bfd | 271 | npo->copy_gref = req.gref; |
f942dc25 IC |
272 | |
273 | return meta; | |
274 | } | |
275 | ||
d0089e8a JG |
276 | struct gop_frag_copy { |
277 | struct xenvif_queue *queue; | |
278 | struct netrx_pending_operations *npo; | |
279 | struct xenvif_rx_meta *meta; | |
280 | int head; | |
281 | int gso_type; | |
282 | ||
283 | struct page *page; | |
284 | }; | |
285 | ||
286 | static void xenvif_setup_copy_gop(unsigned long gfn, | |
287 | unsigned int offset, | |
288 | unsigned int *len, | |
289 | struct gop_frag_copy *info) | |
290 | { | |
291 | struct gnttab_copy *copy_gop; | |
292 | struct xen_page_foreign *foreign; | |
293 | /* Convenient aliases */ | |
294 | struct xenvif_queue *queue = info->queue; | |
295 | struct netrx_pending_operations *npo = info->npo; | |
296 | struct page *page = info->page; | |
297 | ||
298 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); | |
299 | ||
300 | if (npo->copy_off == MAX_BUFFER_OFFSET) | |
301 | info->meta = get_next_rx_buffer(queue, npo); | |
302 | ||
303 | if (npo->copy_off + *len > MAX_BUFFER_OFFSET) | |
304 | *len = MAX_BUFFER_OFFSET - npo->copy_off; | |
305 | ||
306 | copy_gop = npo->copy + npo->copy_prod++; | |
307 | copy_gop->flags = GNTCOPY_dest_gref; | |
308 | copy_gop->len = *len; | |
309 | ||
310 | foreign = xen_page_foreign(page); | |
311 | if (foreign) { | |
312 | copy_gop->source.domid = foreign->domid; | |
313 | copy_gop->source.u.ref = foreign->gref; | |
314 | copy_gop->flags |= GNTCOPY_source_gref; | |
315 | } else { | |
316 | copy_gop->source.domid = DOMID_SELF; | |
317 | copy_gop->source.u.gmfn = gfn; | |
318 | } | |
319 | copy_gop->source.offset = offset; | |
320 | ||
321 | copy_gop->dest.domid = queue->vif->domid; | |
322 | copy_gop->dest.offset = npo->copy_off; | |
323 | copy_gop->dest.u.ref = npo->copy_gref; | |
324 | ||
325 | npo->copy_off += *len; | |
326 | info->meta->size += *len; | |
327 | ||
328 | /* Leave a gap for the GSO descriptor. */ | |
329 | if (info->head && ((1 << info->gso_type) & queue->vif->gso_mask)) | |
330 | queue->rx.req_cons++; | |
331 | ||
332 | info->head = 0; /* There must be something in this buffer now */ | |
333 | } | |
334 | ||
335 | static void xenvif_gop_frag_copy_grant(unsigned long gfn, | |
336 | unsigned offset, | |
337 | unsigned int len, | |
338 | void *data) | |
339 | { | |
340 | unsigned int bytes; | |
341 | ||
342 | while (len) { | |
343 | bytes = len; | |
344 | xenvif_setup_copy_gop(gfn, offset, &bytes, data); | |
345 | offset += bytes; | |
346 | len -= bytes; | |
347 | } | |
348 | } | |
349 | ||
33bc801d WL |
350 | /* |
351 | * Set up the grant operations for this fragment. If it's a flipping | |
352 | * interface, we also set up the unmap request from here. | |
353 | */ | |
e9ce7cb6 | 354 | static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb, |
7376419a WL |
355 | struct netrx_pending_operations *npo, |
356 | struct page *page, unsigned long size, | |
c2677a6f | 357 | unsigned long offset, int *head) |
f942dc25 | 358 | { |
d0089e8a JG |
359 | struct gop_frag_copy info = { |
360 | .queue = queue, | |
361 | .npo = npo, | |
362 | .head = *head, | |
363 | .gso_type = XEN_NETIF_GSO_TYPE_NONE, | |
364 | }; | |
f942dc25 IC |
365 | unsigned long bytes; |
366 | ||
a0f2e80f JG |
367 | if (skb_is_gso(skb)) { |
368 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
d0089e8a | 369 | info.gso_type = XEN_NETIF_GSO_TYPE_TCPV4; |
a0f2e80f | 370 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
d0089e8a | 371 | info.gso_type = XEN_NETIF_GSO_TYPE_TCPV6; |
a0f2e80f JG |
372 | } |
373 | ||
f942dc25 | 374 | /* Data must not cross a page boundary. */ |
6a8ed462 | 375 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 | 376 | |
d0089e8a | 377 | info.meta = npo->meta + npo->meta_prod - 1; |
f942dc25 | 378 | |
6a8ed462 IC |
379 | /* Skip unused frames from start of page */ |
380 | page += offset >> PAGE_SHIFT; | |
381 | offset &= ~PAGE_MASK; | |
382 | ||
f942dc25 | 383 | while (size > 0) { |
6a8ed462 | 384 | BUG_ON(offset >= PAGE_SIZE); |
6a8ed462 | 385 | |
1650d545 | 386 | bytes = PAGE_SIZE - offset; |
6a8ed462 IC |
387 | if (bytes > size) |
388 | bytes = size; | |
389 | ||
d0089e8a JG |
390 | info.page = page; |
391 | gnttab_foreach_grant_in_range(page, offset, bytes, | |
392 | xenvif_gop_frag_copy_grant, | |
393 | &info); | |
f942dc25 | 394 | size -= bytes; |
d0089e8a | 395 | offset = 0; |
f942dc25 | 396 | |
d0089e8a JG |
397 | /* Next page */ |
398 | if (size) { | |
6a8ed462 IC |
399 | BUG_ON(!PageCompound(page)); |
400 | page++; | |
6a8ed462 | 401 | } |
f942dc25 | 402 | } |
d0089e8a JG |
403 | |
404 | *head = info.head; | |
f942dc25 IC |
405 | } |
406 | ||
407 | /* | |
408 | * Prepare an SKB to be transmitted to the frontend. | |
409 | * | |
410 | * This function is responsible for allocating grant operations, meta | |
411 | * structures, etc. | |
412 | * | |
413 | * It returns the number of meta structures consumed. The number of | |
414 | * ring slots used is always equal to the number of meta slots used | |
415 | * plus the number of GSO descriptors used. Currently, we use either | |
416 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
417 | * frontend-side LRO). | |
418 | */ | |
7376419a | 419 | static int xenvif_gop_skb(struct sk_buff *skb, |
e9ce7cb6 WL |
420 | struct netrx_pending_operations *npo, |
421 | struct xenvif_queue *queue) | |
f942dc25 IC |
422 | { |
423 | struct xenvif *vif = netdev_priv(skb->dev); | |
424 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
425 | int i; | |
68a33bfd | 426 | struct xen_netif_rx_request req; |
b3f980bd | 427 | struct xenvif_rx_meta *meta; |
f942dc25 | 428 | unsigned char *data; |
33bc801d | 429 | int head = 1; |
f942dc25 | 430 | int old_meta_prod; |
82cada22 | 431 | int gso_type; |
f942dc25 IC |
432 | |
433 | old_meta_prod = npo->meta_prod; | |
434 | ||
5bd07670 AL |
435 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
436 | if (skb_is_gso(skb)) { | |
437 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
438 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
439 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
440 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
441 | } |
442 | ||
f942dc25 | 443 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 444 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
68a33bfd | 445 | RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req); |
f942dc25 | 446 | meta = npo->meta + npo->meta_prod++; |
82cada22 | 447 | meta->gso_type = gso_type; |
5bd07670 | 448 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 | 449 | meta->size = 0; |
68a33bfd | 450 | meta->id = req.id; |
f942dc25 IC |
451 | } |
452 | ||
68a33bfd | 453 | RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req); |
f942dc25 IC |
454 | meta = npo->meta + npo->meta_prod++; |
455 | ||
82cada22 PD |
456 | if ((1 << gso_type) & vif->gso_mask) { |
457 | meta->gso_type = gso_type; | |
5bd07670 | 458 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
459 | } else { |
460 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 461 | meta->gso_size = 0; |
82cada22 | 462 | } |
f942dc25 IC |
463 | |
464 | meta->size = 0; | |
68a33bfd | 465 | meta->id = req.id; |
f942dc25 | 466 | npo->copy_off = 0; |
68a33bfd | 467 | npo->copy_gref = req.gref; |
f942dc25 IC |
468 | |
469 | data = skb->data; | |
470 | while (data < skb_tail_pointer(skb)) { | |
471 | unsigned int offset = offset_in_page(data); | |
472 | unsigned int len = PAGE_SIZE - offset; | |
473 | ||
474 | if (data + len > skb_tail_pointer(skb)) | |
475 | len = skb_tail_pointer(skb) - data; | |
476 | ||
e9ce7cb6 | 477 | xenvif_gop_frag_copy(queue, skb, npo, |
c2677a6f | 478 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
479 | data += len; |
480 | } | |
481 | ||
482 | for (i = 0; i < nr_frags; i++) { | |
e9ce7cb6 | 483 | xenvif_gop_frag_copy(queue, skb, npo, |
7376419a WL |
484 | skb_frag_page(&skb_shinfo(skb)->frags[i]), |
485 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
486 | skb_shinfo(skb)->frags[i].page_offset, | |
c2677a6f | 487 | &head); |
f942dc25 IC |
488 | } |
489 | ||
490 | return npo->meta_prod - old_meta_prod; | |
491 | } | |
492 | ||
493 | /* | |
7376419a | 494 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
495 | * used to set up the operations on the top of |
496 | * netrx_pending_operations, which have since been done. Check that | |
497 | * they didn't give any errors and advance over them. | |
498 | */ | |
7376419a WL |
499 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
500 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
501 | { |
502 | struct gnttab_copy *copy_op; | |
503 | int status = XEN_NETIF_RSP_OKAY; | |
504 | int i; | |
505 | ||
506 | for (i = 0; i < nr_meta_slots; i++) { | |
507 | copy_op = npo->copy + npo->copy_cons++; | |
508 | if (copy_op->status != GNTST_okay) { | |
509 | netdev_dbg(vif->dev, | |
510 | "Bad status %d from copy to DOM%d.\n", | |
511 | copy_op->status, vif->domid); | |
512 | status = XEN_NETIF_RSP_ERROR; | |
513 | } | |
514 | } | |
515 | ||
516 | return status; | |
517 | } | |
518 | ||
e9ce7cb6 | 519 | static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status, |
7376419a WL |
520 | struct xenvif_rx_meta *meta, |
521 | int nr_meta_slots) | |
f942dc25 IC |
522 | { |
523 | int i; | |
524 | unsigned long offset; | |
525 | ||
526 | /* No fragments used */ | |
527 | if (nr_meta_slots <= 1) | |
528 | return; | |
529 | ||
530 | nr_meta_slots--; | |
531 | ||
532 | for (i = 0; i < nr_meta_slots; i++) { | |
533 | int flags; | |
534 | if (i == nr_meta_slots - 1) | |
535 | flags = 0; | |
536 | else | |
537 | flags = XEN_NETRXF_more_data; | |
538 | ||
539 | offset = 0; | |
e9ce7cb6 | 540 | make_rx_response(queue, meta[i].id, status, offset, |
f942dc25 IC |
541 | meta[i].size, flags); |
542 | } | |
543 | } | |
544 | ||
e9ce7cb6 | 545 | void xenvif_kick_thread(struct xenvif_queue *queue) |
b3f980bd | 546 | { |
e9ce7cb6 | 547 | wake_up(&queue->wq); |
b3f980bd WL |
548 | } |
549 | ||
e9ce7cb6 | 550 | static void xenvif_rx_action(struct xenvif_queue *queue) |
f942dc25 | 551 | { |
f942dc25 | 552 | s8 status; |
e1f00a69 | 553 | u16 flags; |
f942dc25 IC |
554 | struct xen_netif_rx_response *resp; |
555 | struct sk_buff_head rxq; | |
556 | struct sk_buff *skb; | |
557 | LIST_HEAD(notify); | |
558 | int ret; | |
f942dc25 | 559 | unsigned long offset; |
11b57f90 | 560 | bool need_to_notify = false; |
f942dc25 IC |
561 | |
562 | struct netrx_pending_operations npo = { | |
e9ce7cb6 WL |
563 | .copy = queue->grant_copy_op, |
564 | .meta = queue->meta, | |
f942dc25 IC |
565 | }; |
566 | ||
567 | skb_queue_head_init(&rxq); | |
568 | ||
1d5d4852 | 569 | while (xenvif_rx_ring_slots_available(queue) |
f48da8b1 | 570 | && (skb = xenvif_rx_dequeue(queue)) != NULL) { |
ecf08d2d DV |
571 | queue->last_rx_time = jiffies; |
572 | ||
e9ce7cb6 | 573 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue); |
1425c7a4 | 574 | |
f942dc25 | 575 | __skb_queue_tail(&rxq, skb); |
f942dc25 IC |
576 | } |
577 | ||
e9ce7cb6 | 578 | BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta)); |
f942dc25 IC |
579 | |
580 | if (!npo.copy_prod) | |
ca2f09f2 | 581 | goto done; |
f942dc25 | 582 | |
ac3d5ac2 | 583 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
e9ce7cb6 | 584 | gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
585 | |
586 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 587 | |
e9ce7cb6 WL |
588 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
589 | queue->vif->gso_prefix_mask) { | |
590 | resp = RING_GET_RESPONSE(&queue->rx, | |
591 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
592 | |
593 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
594 | ||
e9ce7cb6 WL |
595 | resp->offset = queue->meta[npo.meta_cons].gso_size; |
596 | resp->id = queue->meta[npo.meta_cons].id; | |
8f13dd96 | 597 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
598 | |
599 | npo.meta_cons++; | |
8f13dd96 | 600 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
601 | } |
602 | ||
603 | ||
e9ce7cb6 WL |
604 | queue->stats.tx_bytes += skb->len; |
605 | queue->stats.tx_packets++; | |
f942dc25 | 606 | |
e9ce7cb6 | 607 | status = xenvif_check_gop(queue->vif, |
8f13dd96 ZK |
608 | XENVIF_RX_CB(skb)->meta_slots_used, |
609 | &npo); | |
f942dc25 | 610 | |
8f13dd96 | 611 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
612 | flags = 0; |
613 | else | |
614 | flags = XEN_NETRXF_more_data; | |
615 | ||
616 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
617 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
618 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
619 | /* remote but checksummed. */ | |
620 | flags |= XEN_NETRXF_data_validated; | |
621 | ||
622 | offset = 0; | |
e9ce7cb6 | 623 | resp = make_rx_response(queue, queue->meta[npo.meta_cons].id, |
f942dc25 | 624 | status, offset, |
e9ce7cb6 | 625 | queue->meta[npo.meta_cons].size, |
f942dc25 IC |
626 | flags); |
627 | ||
e9ce7cb6 WL |
628 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
629 | queue->vif->gso_mask) { | |
f942dc25 IC |
630 | struct xen_netif_extra_info *gso = |
631 | (struct xen_netif_extra_info *) | |
e9ce7cb6 WL |
632 | RING_GET_RESPONSE(&queue->rx, |
633 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
634 | |
635 | resp->flags |= XEN_NETRXF_extra_info; | |
636 | ||
e9ce7cb6 WL |
637 | gso->u.gso.type = queue->meta[npo.meta_cons].gso_type; |
638 | gso->u.gso.size = queue->meta[npo.meta_cons].gso_size; | |
f942dc25 IC |
639 | gso->u.gso.pad = 0; |
640 | gso->u.gso.features = 0; | |
641 | ||
642 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
643 | gso->flags = 0; | |
644 | } | |
645 | ||
e9ce7cb6 WL |
646 | xenvif_add_frag_responses(queue, status, |
647 | queue->meta + npo.meta_cons + 1, | |
8f13dd96 | 648 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 | 649 | |
e9ce7cb6 | 650 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret); |
f942dc25 | 651 | |
11b57f90 | 652 | need_to_notify |= !!ret; |
b3f980bd | 653 | |
8f13dd96 | 654 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
655 | dev_kfree_skb(skb); |
656 | } | |
657 | ||
ca2f09f2 | 658 | done: |
b3f980bd | 659 | if (need_to_notify) |
e9ce7cb6 | 660 | notify_remote_via_irq(queue->rx_irq); |
f942dc25 IC |
661 | } |
662 | ||
e9ce7cb6 | 663 | void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue) |
f942dc25 IC |
664 | { |
665 | int more_to_do; | |
666 | ||
e9ce7cb6 | 667 | RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do); |
f942dc25 IC |
668 | |
669 | if (more_to_do) | |
e9ce7cb6 | 670 | napi_schedule(&queue->napi); |
f942dc25 IC |
671 | } |
672 | ||
e9ce7cb6 | 673 | static void tx_add_credit(struct xenvif_queue *queue) |
f942dc25 IC |
674 | { |
675 | unsigned long max_burst, max_credit; | |
676 | ||
677 | /* | |
678 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
679 | * Otherwise the interface can seize up due to insufficient credit. | |
680 | */ | |
0f589967 | 681 | max_burst = max(131072UL, queue->credit_bytes); |
f942dc25 IC |
682 | |
683 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
e9ce7cb6 WL |
684 | max_credit = queue->remaining_credit + queue->credit_bytes; |
685 | if (max_credit < queue->remaining_credit) | |
f942dc25 IC |
686 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ |
687 | ||
e9ce7cb6 | 688 | queue->remaining_credit = min(max_credit, max_burst); |
f942dc25 IC |
689 | } |
690 | ||
edafc132 | 691 | void xenvif_tx_credit_callback(unsigned long data) |
f942dc25 | 692 | { |
e9ce7cb6 WL |
693 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
694 | tx_add_credit(queue); | |
695 | xenvif_napi_schedule_or_enable_events(queue); | |
f942dc25 IC |
696 | } |
697 | ||
e9ce7cb6 | 698 | static void xenvif_tx_err(struct xenvif_queue *queue, |
7376419a | 699 | struct xen_netif_tx_request *txp, RING_IDX end) |
f942dc25 | 700 | { |
e9ce7cb6 | 701 | RING_IDX cons = queue->tx.req_cons; |
f53c3fe8 | 702 | unsigned long flags; |
f942dc25 IC |
703 | |
704 | do { | |
e9ce7cb6 WL |
705 | spin_lock_irqsave(&queue->response_lock, flags); |
706 | make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR); | |
c8a4d299 | 707 | push_tx_responses(queue); |
e9ce7cb6 | 708 | spin_unlock_irqrestore(&queue->response_lock, flags); |
b9149729 | 709 | if (cons == end) |
f942dc25 | 710 | break; |
68a33bfd | 711 | RING_COPY_REQUEST(&queue->tx, cons++, txp); |
f942dc25 | 712 | } while (1); |
e9ce7cb6 | 713 | queue->tx.req_cons = cons; |
f942dc25 IC |
714 | } |
715 | ||
7376419a | 716 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
717 | { |
718 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 | 719 | vif->disabled = true; |
e9ce7cb6 WL |
720 | /* Disable the vif from queue 0's kthread */ |
721 | if (vif->queues) | |
722 | xenvif_kick_thread(&vif->queues[0]); | |
48856286 IC |
723 | } |
724 | ||
e9ce7cb6 | 725 | static int xenvif_count_requests(struct xenvif_queue *queue, |
7376419a WL |
726 | struct xen_netif_tx_request *first, |
727 | struct xen_netif_tx_request *txp, | |
728 | int work_to_do) | |
f942dc25 | 729 | { |
e9ce7cb6 | 730 | RING_IDX cons = queue->tx.req_cons; |
2810e5b9 WL |
731 | int slots = 0; |
732 | int drop_err = 0; | |
59ccb4eb | 733 | int more_data; |
f942dc25 IC |
734 | |
735 | if (!(first->flags & XEN_NETTXF_more_data)) | |
736 | return 0; | |
737 | ||
738 | do { | |
59ccb4eb WL |
739 | struct xen_netif_tx_request dropped_tx = { 0 }; |
740 | ||
2810e5b9 | 741 | if (slots >= work_to_do) { |
e9ce7cb6 | 742 | netdev_err(queue->vif->dev, |
2810e5b9 WL |
743 | "Asked for %d slots but exceeds this limit\n", |
744 | work_to_do); | |
e9ce7cb6 | 745 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 746 | return -ENODATA; |
f942dc25 IC |
747 | } |
748 | ||
2810e5b9 WL |
749 | /* This guest is really using too many slots and |
750 | * considered malicious. | |
751 | */ | |
37641494 | 752 | if (unlikely(slots >= fatal_skb_slots)) { |
e9ce7cb6 | 753 | netdev_err(queue->vif->dev, |
2810e5b9 | 754 | "Malicious frontend using %d slots, threshold %u\n", |
37641494 | 755 | slots, fatal_skb_slots); |
e9ce7cb6 | 756 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 757 | return -E2BIG; |
f942dc25 IC |
758 | } |
759 | ||
2810e5b9 | 760 | /* Xen network protocol had implicit dependency on |
37641494 WL |
761 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
762 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
763 | * same behavior as before. Any packet using more than | |
764 | * 18 slots but less than fatal_skb_slots slots is | |
765 | * dropped | |
2810e5b9 | 766 | */ |
37641494 | 767 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 | 768 | if (net_ratelimit()) |
e9ce7cb6 | 769 | netdev_dbg(queue->vif->dev, |
2810e5b9 | 770 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", |
37641494 | 771 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
772 | drop_err = -E2BIG; |
773 | } | |
774 | ||
59ccb4eb WL |
775 | if (drop_err) |
776 | txp = &dropped_tx; | |
777 | ||
68a33bfd | 778 | RING_COPY_REQUEST(&queue->tx, cons + slots, txp); |
03393fd5 WL |
779 | |
780 | /* If the guest submitted a frame >= 64 KiB then | |
781 | * first->size overflowed and following slots will | |
782 | * appear to be larger than the frame. | |
783 | * | |
784 | * This cannot be fatal error as there are buggy | |
785 | * frontends that do this. | |
786 | * | |
787 | * Consume all slots and drop the packet. | |
788 | */ | |
789 | if (!drop_err && txp->size > first->size) { | |
790 | if (net_ratelimit()) | |
e9ce7cb6 | 791 | netdev_dbg(queue->vif->dev, |
03393fd5 WL |
792 | "Invalid tx request, slot size %u > remaining size %u\n", |
793 | txp->size, first->size); | |
794 | drop_err = -EIO; | |
f942dc25 IC |
795 | } |
796 | ||
797 | first->size -= txp->size; | |
2810e5b9 | 798 | slots++; |
f942dc25 | 799 | |
d0089e8a | 800 | if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) { |
68946159 | 801 | netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n", |
f942dc25 | 802 | txp->offset, txp->size); |
e9ce7cb6 | 803 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 804 | return -EINVAL; |
f942dc25 | 805 | } |
59ccb4eb WL |
806 | |
807 | more_data = txp->flags & XEN_NETTXF_more_data; | |
808 | ||
809 | if (!drop_err) | |
810 | txp++; | |
811 | ||
812 | } while (more_data); | |
2810e5b9 WL |
813 | |
814 | if (drop_err) { | |
e9ce7cb6 | 815 | xenvif_tx_err(queue, first, cons + slots); |
2810e5b9 WL |
816 | return drop_err; |
817 | } | |
818 | ||
819 | return slots; | |
f942dc25 IC |
820 | } |
821 | ||
8f13dd96 ZK |
822 | |
823 | struct xenvif_tx_cb { | |
824 | u16 pending_idx; | |
825 | }; | |
826 | ||
827 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
828 | ||
e9ce7cb6 | 829 | static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue, |
9074ce24 ZK |
830 | u16 pending_idx, |
831 | struct xen_netif_tx_request *txp, | |
832 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 833 | { |
e9ce7cb6 WL |
834 | queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx]; |
835 | gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx), | |
f53c3fe8 | 836 | GNTMAP_host_map | GNTMAP_readonly, |
e9ce7cb6 | 837 | txp->gref, queue->vif->domid); |
f53c3fe8 | 838 | |
e9ce7cb6 | 839 | memcpy(&queue->pending_tx_info[pending_idx].req, txp, |
f53c3fe8 ZK |
840 | sizeof(*txp)); |
841 | } | |
842 | ||
e3377f36 ZK |
843 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
844 | { | |
845 | struct sk_buff *skb = | |
846 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
847 | GFP_ATOMIC | __GFP_NOWARN); | |
848 | if (unlikely(skb == NULL)) | |
849 | return NULL; | |
850 | ||
851 | /* Packets passed to netif_rx() must have some headroom. */ | |
852 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
853 | ||
854 | /* Initialize it here to avoid later surprises */ | |
855 | skb_shinfo(skb)->destructor_arg = NULL; | |
856 | ||
857 | return skb; | |
858 | } | |
859 | ||
e9ce7cb6 | 860 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue, |
f53c3fe8 ZK |
861 | struct sk_buff *skb, |
862 | struct xen_netif_tx_request *txp, | |
2475b225 RL |
863 | struct gnttab_map_grant_ref *gop, |
864 | unsigned int frag_overflow, | |
865 | struct sk_buff *nskb) | |
f942dc25 IC |
866 | { |
867 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
868 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 869 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
870 | int start; |
871 | pending_ring_idx_t index; | |
2475b225 | 872 | unsigned int nr_slots; |
2810e5b9 | 873 | |
2810e5b9 | 874 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
875 | |
876 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 877 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 878 | |
f53c3fe8 ZK |
879 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
880 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
881 | index = pending_index(queue->pending_cons++); |
882 | pending_idx = queue->pending_ring[index]; | |
883 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
f53c3fe8 | 884 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
885 | } |
886 | ||
e3377f36 | 887 | if (frag_overflow) { |
e3377f36 ZK |
888 | |
889 | shinfo = skb_shinfo(nskb); | |
890 | frags = shinfo->frags; | |
891 | ||
892 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
893 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
894 | index = pending_index(queue->pending_cons++); |
895 | pending_idx = queue->pending_ring[index]; | |
896 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
e3377f36 ZK |
897 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
898 | pending_idx); | |
899 | } | |
900 | ||
901 | skb_shinfo(skb)->frag_list = nskb; | |
902 | } | |
2810e5b9 | 903 | |
f942dc25 IC |
904 | return gop; |
905 | } | |
906 | ||
e9ce7cb6 | 907 | static inline void xenvif_grant_handle_set(struct xenvif_queue *queue, |
f53c3fe8 ZK |
908 | u16 pending_idx, |
909 | grant_handle_t handle) | |
910 | { | |
e9ce7cb6 | 911 | if (unlikely(queue->grant_tx_handle[pending_idx] != |
f53c3fe8 | 912 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 913 | netdev_err(queue->vif->dev, |
68946159 | 914 | "Trying to overwrite active handle! pending_idx: 0x%x\n", |
f53c3fe8 ZK |
915 | pending_idx); |
916 | BUG(); | |
917 | } | |
e9ce7cb6 | 918 | queue->grant_tx_handle[pending_idx] = handle; |
f53c3fe8 ZK |
919 | } |
920 | ||
e9ce7cb6 | 921 | static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue, |
f53c3fe8 ZK |
922 | u16 pending_idx) |
923 | { | |
e9ce7cb6 | 924 | if (unlikely(queue->grant_tx_handle[pending_idx] == |
f53c3fe8 | 925 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 926 | netdev_err(queue->vif->dev, |
68946159 | 927 | "Trying to unmap invalid handle! pending_idx: 0x%x\n", |
f53c3fe8 ZK |
928 | pending_idx); |
929 | BUG(); | |
930 | } | |
e9ce7cb6 | 931 | queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; |
f53c3fe8 ZK |
932 | } |
933 | ||
e9ce7cb6 | 934 | static int xenvif_tx_check_gop(struct xenvif_queue *queue, |
7376419a | 935 | struct sk_buff *skb, |
bdab8275 ZK |
936 | struct gnttab_map_grant_ref **gopp_map, |
937 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 938 | { |
9074ce24 | 939 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 940 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
1a998d3e ZK |
941 | /* This always points to the shinfo of the skb being checked, which |
942 | * could be either the first or the one on the frag_list | |
943 | */ | |
f942dc25 | 944 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1a998d3e ZK |
945 | /* If this is non-NULL, we are currently checking the frag_list skb, and |
946 | * this points to the shinfo of the first one | |
947 | */ | |
948 | struct skb_shared_info *first_shinfo = NULL; | |
f942dc25 | 949 | int nr_frags = shinfo->nr_frags; |
1b860da0 ZK |
950 | const bool sharedslot = nr_frags && |
951 | frag_get_pending_idx(&shinfo->frags[0]) == pending_idx; | |
bdab8275 | 952 | int i, err; |
f942dc25 IC |
953 | |
954 | /* Check status of header. */ | |
bdab8275 | 955 | err = (*gopp_copy)->status; |
bdab8275 ZK |
956 | if (unlikely(err)) { |
957 | if (net_ratelimit()) | |
e9ce7cb6 | 958 | netdev_dbg(queue->vif->dev, |
00aefceb | 959 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
960 | (*gopp_copy)->status, |
961 | pending_idx, | |
962 | (*gopp_copy)->source.u.ref); | |
1b860da0 ZK |
963 | /* The first frag might still have this slot mapped */ |
964 | if (!sharedslot) | |
965 | xenvif_idx_release(queue, pending_idx, | |
966 | XEN_NETIF_RSP_ERROR); | |
bdab8275 | 967 | } |
d8cfbfc4 | 968 | (*gopp_copy)++; |
f942dc25 | 969 | |
e3377f36 | 970 | check_frags: |
bdab8275 | 971 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 972 | int j, newerr; |
f942dc25 | 973 | |
ea066ad1 | 974 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
975 | |
976 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 977 | newerr = gop_map->status; |
2810e5b9 | 978 | |
f942dc25 | 979 | if (likely(!newerr)) { |
e9ce7cb6 | 980 | xenvif_grant_handle_set(queue, |
9074ce24 ZK |
981 | pending_idx, |
982 | gop_map->handle); | |
f942dc25 | 983 | /* Had a previous error? Invalidate this fragment. */ |
1b860da0 | 984 | if (unlikely(err)) { |
e9ce7cb6 | 985 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
986 | /* If the mapping of the first frag was OK, but |
987 | * the header's copy failed, and they are | |
988 | * sharing a slot, send an error | |
989 | */ | |
990 | if (i == 0 && sharedslot) | |
991 | xenvif_idx_release(queue, pending_idx, | |
992 | XEN_NETIF_RSP_ERROR); | |
993 | else | |
994 | xenvif_idx_release(queue, pending_idx, | |
995 | XEN_NETIF_RSP_OKAY); | |
996 | } | |
f942dc25 IC |
997 | continue; |
998 | } | |
999 | ||
1000 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 | 1001 | if (net_ratelimit()) |
e9ce7cb6 | 1002 | netdev_dbg(queue->vif->dev, |
00aefceb | 1003 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1004 | i, |
1005 | gop_map->status, | |
1006 | pending_idx, | |
1007 | gop_map->ref); | |
1b860da0 | 1008 | |
e9ce7cb6 | 1009 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1010 | |
1011 | /* Not the first error? Preceding frags already invalidated. */ | |
1012 | if (err) | |
1013 | continue; | |
1b860da0 ZK |
1014 | |
1015 | /* First error: if the header haven't shared a slot with the | |
1016 | * first frag, release it as well. | |
1017 | */ | |
1018 | if (!sharedslot) | |
1019 | xenvif_idx_release(queue, | |
1020 | XENVIF_TX_CB(skb)->pending_idx, | |
1021 | XEN_NETIF_RSP_OKAY); | |
1022 | ||
1023 | /* Invalidate preceding fragments of this skb. */ | |
bdab8275 | 1024 | for (j = 0; j < i; j++) { |
5ccb3ea7 | 1025 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 1026 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
1027 | xenvif_idx_release(queue, pending_idx, |
1028 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1029 | } |
1030 | ||
1a998d3e ZK |
1031 | /* And if we found the error while checking the frag_list, unmap |
1032 | * the first skb's frags | |
1033 | */ | |
1034 | if (first_shinfo) { | |
1035 | for (j = 0; j < first_shinfo->nr_frags; j++) { | |
1036 | pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]); | |
1037 | xenvif_idx_unmap(queue, pending_idx); | |
1b860da0 ZK |
1038 | xenvif_idx_release(queue, pending_idx, |
1039 | XEN_NETIF_RSP_OKAY); | |
1a998d3e | 1040 | } |
f942dc25 IC |
1041 | } |
1042 | ||
1043 | /* Remember the error: invalidate all subsequent fragments. */ | |
1044 | err = newerr; | |
1045 | } | |
1046 | ||
1a998d3e ZK |
1047 | if (skb_has_frag_list(skb) && !first_shinfo) { |
1048 | first_shinfo = skb_shinfo(skb); | |
1049 | shinfo = skb_shinfo(skb_shinfo(skb)->frag_list); | |
e3377f36 | 1050 | nr_frags = shinfo->nr_frags; |
e3377f36 ZK |
1051 | |
1052 | goto check_frags; | |
1053 | } | |
1054 | ||
bdab8275 | 1055 | *gopp_map = gop_map; |
f942dc25 IC |
1056 | return err; |
1057 | } | |
1058 | ||
e9ce7cb6 | 1059 | static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb) |
f942dc25 IC |
1060 | { |
1061 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1062 | int nr_frags = shinfo->nr_frags; | |
1063 | int i; | |
f53c3fe8 ZK |
1064 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1065 | ||
f942dc25 IC |
1066 | for (i = 0; i < nr_frags; i++) { |
1067 | skb_frag_t *frag = shinfo->frags + i; | |
1068 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1069 | struct page *page; |
1070 | u16 pending_idx; | |
f942dc25 | 1071 | |
ea066ad1 | 1072 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1073 | |
f53c3fe8 | 1074 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 1075 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 1076 | skb_shinfo(skb)->destructor_arg = |
e9ce7cb6 | 1077 | &callback_param(queue, pending_idx); |
bdab8275 | 1078 | else |
e9ce7cb6 WL |
1079 | callback_param(queue, prev_pending_idx).ctx = |
1080 | &callback_param(queue, pending_idx); | |
f53c3fe8 | 1081 | |
e9ce7cb6 | 1082 | callback_param(queue, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1083 | prev_pending_idx = pending_idx; |
1084 | ||
e9ce7cb6 WL |
1085 | txp = &queue->pending_tx_info[pending_idx].req; |
1086 | page = virt_to_page(idx_to_kaddr(queue, pending_idx)); | |
ea066ad1 | 1087 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1088 | skb->len += txp->size; |
1089 | skb->data_len += txp->size; | |
1090 | skb->truesize += txp->size; | |
1091 | ||
f53c3fe8 | 1092 | /* Take an extra reference to offset network stack's put_page */ |
e9ce7cb6 | 1093 | get_page(queue->mmap_pages[pending_idx]); |
f942dc25 IC |
1094 | } |
1095 | } | |
1096 | ||
e9ce7cb6 | 1097 | static int xenvif_get_extras(struct xenvif_queue *queue, |
f942dc25 IC |
1098 | struct xen_netif_extra_info *extras, |
1099 | int work_to_do) | |
1100 | { | |
1101 | struct xen_netif_extra_info extra; | |
e9ce7cb6 | 1102 | RING_IDX cons = queue->tx.req_cons; |
f942dc25 IC |
1103 | |
1104 | do { | |
1105 | if (unlikely(work_to_do-- <= 0)) { | |
e9ce7cb6 WL |
1106 | netdev_err(queue->vif->dev, "Missing extra info\n"); |
1107 | xenvif_fatal_tx_err(queue->vif); | |
f942dc25 IC |
1108 | return -EBADR; |
1109 | } | |
1110 | ||
68a33bfd | 1111 | RING_COPY_REQUEST(&queue->tx, cons, &extra); |
f942dc25 IC |
1112 | if (unlikely(!extra.type || |
1113 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
e9ce7cb6 WL |
1114 | queue->tx.req_cons = ++cons; |
1115 | netdev_err(queue->vif->dev, | |
f942dc25 | 1116 | "Invalid extra type: %d\n", extra.type); |
e9ce7cb6 | 1117 | xenvif_fatal_tx_err(queue->vif); |
f942dc25 IC |
1118 | return -EINVAL; |
1119 | } | |
1120 | ||
1121 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
e9ce7cb6 | 1122 | queue->tx.req_cons = ++cons; |
f942dc25 IC |
1123 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
1124 | ||
1125 | return work_to_do; | |
1126 | } | |
1127 | ||
7376419a WL |
1128 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1129 | struct sk_buff *skb, | |
1130 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1131 | { |
1132 | if (!gso->u.gso.size) { | |
48856286 | 1133 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1134 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1135 | return -EINVAL; |
1136 | } | |
1137 | ||
a9468587 PD |
1138 | switch (gso->u.gso.type) { |
1139 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1140 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1141 | break; | |
1142 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1143 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1144 | break; | |
1145 | default: | |
48856286 | 1146 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1147 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1148 | return -EINVAL; |
1149 | } | |
1150 | ||
1151 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1152 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1153 | |
1154 | return 0; | |
1155 | } | |
1156 | ||
e9ce7cb6 | 1157 | static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb) |
2eba61d5 | 1158 | { |
2721637c | 1159 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1160 | |
1161 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1162 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1163 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1164 | * recalculate the partial checksum. | |
1165 | */ | |
1166 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
e9ce7cb6 | 1167 | queue->stats.rx_gso_checksum_fixup++; |
2eba61d5 | 1168 | skb->ip_summed = CHECKSUM_PARTIAL; |
2721637c | 1169 | recalculate_partial_csum = true; |
2eba61d5 PD |
1170 | } |
1171 | ||
1172 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1173 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1174 | return 0; | |
1175 | ||
2721637c | 1176 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1177 | } |
1178 | ||
e9ce7cb6 | 1179 | static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size) |
f942dc25 | 1180 | { |
059dfa6a | 1181 | u64 now = get_jiffies_64(); |
e9ce7cb6 WL |
1182 | u64 next_credit = queue->credit_window_start + |
1183 | msecs_to_jiffies(queue->credit_usec / 1000); | |
f942dc25 IC |
1184 | |
1185 | /* Timer could already be pending in rare cases. */ | |
e9ce7cb6 | 1186 | if (timer_pending(&queue->credit_timeout)) |
f942dc25 IC |
1187 | return true; |
1188 | ||
1189 | /* Passed the point where we can replenish credit? */ | |
059dfa6a | 1190 | if (time_after_eq64(now, next_credit)) { |
e9ce7cb6 WL |
1191 | queue->credit_window_start = now; |
1192 | tx_add_credit(queue); | |
f942dc25 IC |
1193 | } |
1194 | ||
1195 | /* Still too big to send right now? Set a callback. */ | |
e9ce7cb6 WL |
1196 | if (size > queue->remaining_credit) { |
1197 | queue->credit_timeout.data = | |
1198 | (unsigned long)queue; | |
e9ce7cb6 | 1199 | mod_timer(&queue->credit_timeout, |
f942dc25 | 1200 | next_credit); |
e9ce7cb6 | 1201 | queue->credit_window_start = next_credit; |
f942dc25 IC |
1202 | |
1203 | return true; | |
1204 | } | |
1205 | ||
1206 | return false; | |
1207 | } | |
1208 | ||
210c34dc PD |
1209 | /* No locking is required in xenvif_mcast_add/del() as they are |
1210 | * only ever invoked from NAPI poll. An RCU list is used because | |
1211 | * xenvif_mcast_match() is called asynchronously, during start_xmit. | |
1212 | */ | |
1213 | ||
1214 | static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr) | |
1215 | { | |
1216 | struct xenvif_mcast_addr *mcast; | |
1217 | ||
1218 | if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) { | |
1219 | if (net_ratelimit()) | |
1220 | netdev_err(vif->dev, | |
1221 | "Too many multicast addresses\n"); | |
1222 | return -ENOSPC; | |
1223 | } | |
1224 | ||
1225 | mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC); | |
1226 | if (!mcast) | |
1227 | return -ENOMEM; | |
1228 | ||
1229 | ether_addr_copy(mcast->addr, addr); | |
1230 | list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr); | |
1231 | vif->fe_mcast_count++; | |
1232 | ||
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr) | |
1237 | { | |
1238 | struct xenvif_mcast_addr *mcast; | |
1239 | ||
1240 | list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) { | |
1241 | if (ether_addr_equal(addr, mcast->addr)) { | |
1242 | --vif->fe_mcast_count; | |
1243 | list_del_rcu(&mcast->entry); | |
1244 | kfree_rcu(mcast, rcu); | |
1245 | break; | |
1246 | } | |
1247 | } | |
1248 | } | |
1249 | ||
1250 | bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr) | |
1251 | { | |
1252 | struct xenvif_mcast_addr *mcast; | |
1253 | ||
1254 | rcu_read_lock(); | |
1255 | list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) { | |
1256 | if (ether_addr_equal(addr, mcast->addr)) { | |
1257 | rcu_read_unlock(); | |
1258 | return true; | |
1259 | } | |
1260 | } | |
1261 | rcu_read_unlock(); | |
1262 | ||
1263 | return false; | |
1264 | } | |
1265 | ||
1266 | void xenvif_mcast_addr_list_free(struct xenvif *vif) | |
1267 | { | |
1268 | /* No need for locking or RCU here. NAPI poll and TX queue | |
1269 | * are stopped. | |
1270 | */ | |
1271 | while (!list_empty(&vif->fe_mcast_addr)) { | |
1272 | struct xenvif_mcast_addr *mcast; | |
1273 | ||
1274 | mcast = list_first_entry(&vif->fe_mcast_addr, | |
1275 | struct xenvif_mcast_addr, | |
1276 | entry); | |
1277 | --vif->fe_mcast_count; | |
1278 | list_del(&mcast->entry); | |
1279 | kfree(mcast); | |
1280 | } | |
1281 | } | |
1282 | ||
e9ce7cb6 | 1283 | static void xenvif_tx_build_gops(struct xenvif_queue *queue, |
bdab8275 ZK |
1284 | int budget, |
1285 | unsigned *copy_ops, | |
1286 | unsigned *map_ops) | |
f942dc25 | 1287 | { |
2475b225 RL |
1288 | struct gnttab_map_grant_ref *gop = queue->tx_map_ops; |
1289 | struct sk_buff *skb, *nskb; | |
f942dc25 | 1290 | int ret; |
2475b225 | 1291 | unsigned int frag_overflow; |
f942dc25 | 1292 | |
e9ce7cb6 | 1293 | while (skb_queue_len(&queue->tx_queue) < budget) { |
f942dc25 | 1294 | struct xen_netif_tx_request txreq; |
37641494 | 1295 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1296 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1297 | u16 pending_idx; | |
1298 | RING_IDX idx; | |
1299 | int work_to_do; | |
1300 | unsigned int data_len; | |
1301 | pending_ring_idx_t index; | |
1302 | ||
e9ce7cb6 | 1303 | if (queue->tx.sring->req_prod - queue->tx.req_cons > |
48856286 | 1304 | XEN_NETIF_TX_RING_SIZE) { |
e9ce7cb6 | 1305 | netdev_err(queue->vif->dev, |
48856286 IC |
1306 | "Impossible number of requests. " |
1307 | "req_prod %d, req_cons %d, size %ld\n", | |
e9ce7cb6 | 1308 | queue->tx.sring->req_prod, queue->tx.req_cons, |
48856286 | 1309 | XEN_NETIF_TX_RING_SIZE); |
e9ce7cb6 | 1310 | xenvif_fatal_tx_err(queue->vif); |
e9d8b2c2 | 1311 | break; |
48856286 IC |
1312 | } |
1313 | ||
e9ce7cb6 | 1314 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx); |
b3f980bd WL |
1315 | if (!work_to_do) |
1316 | break; | |
f942dc25 | 1317 | |
e9ce7cb6 | 1318 | idx = queue->tx.req_cons; |
f942dc25 | 1319 | rmb(); /* Ensure that we see the request before we copy it. */ |
68a33bfd | 1320 | RING_COPY_REQUEST(&queue->tx, idx, &txreq); |
f942dc25 IC |
1321 | |
1322 | /* Credit-based scheduling. */ | |
e9ce7cb6 WL |
1323 | if (txreq.size > queue->remaining_credit && |
1324 | tx_credit_exceeded(queue, txreq.size)) | |
b3f980bd | 1325 | break; |
f942dc25 | 1326 | |
e9ce7cb6 | 1327 | queue->remaining_credit -= txreq.size; |
f942dc25 IC |
1328 | |
1329 | work_to_do--; | |
e9ce7cb6 | 1330 | queue->tx.req_cons = ++idx; |
f942dc25 IC |
1331 | |
1332 | memset(extras, 0, sizeof(extras)); | |
1333 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
e9ce7cb6 | 1334 | work_to_do = xenvif_get_extras(queue, extras, |
7376419a | 1335 | work_to_do); |
e9ce7cb6 | 1336 | idx = queue->tx.req_cons; |
48856286 | 1337 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1338 | break; |
f942dc25 IC |
1339 | } |
1340 | ||
210c34dc PD |
1341 | if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) { |
1342 | struct xen_netif_extra_info *extra; | |
1343 | ||
1344 | extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1]; | |
1345 | ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr); | |
1346 | ||
1347 | make_tx_response(queue, &txreq, | |
1348 | (ret == 0) ? | |
1349 | XEN_NETIF_RSP_OKAY : | |
1350 | XEN_NETIF_RSP_ERROR); | |
1351 | push_tx_responses(queue); | |
1352 | continue; | |
1353 | } | |
1354 | ||
1355 | if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) { | |
1356 | struct xen_netif_extra_info *extra; | |
1357 | ||
1358 | extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1]; | |
1359 | xenvif_mcast_del(queue->vif, extra->u.mcast.addr); | |
1360 | ||
1361 | make_tx_response(queue, &txreq, XEN_NETIF_RSP_OKAY); | |
1362 | push_tx_responses(queue); | |
1363 | continue; | |
1364 | } | |
1365 | ||
e9ce7cb6 | 1366 | ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do); |
48856286 | 1367 | if (unlikely(ret < 0)) |
b3f980bd | 1368 | break; |
48856286 | 1369 | |
f942dc25 IC |
1370 | idx += ret; |
1371 | ||
1372 | if (unlikely(txreq.size < ETH_HLEN)) { | |
e9ce7cb6 | 1373 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1374 | "Bad packet size: %d\n", txreq.size); |
e9ce7cb6 | 1375 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1376 | break; |
f942dc25 IC |
1377 | } |
1378 | ||
1379 | /* No crossing a page as the payload mustn't fragment. */ | |
d0089e8a | 1380 | if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) { |
e9ce7cb6 | 1381 | netdev_err(queue->vif->dev, |
68946159 | 1382 | "txreq.offset: %u, size: %u, end: %lu\n", |
f942dc25 | 1383 | txreq.offset, txreq.size, |
d0089e8a | 1384 | (unsigned long)(txreq.offset&~XEN_PAGE_MASK) + txreq.size); |
e9ce7cb6 | 1385 | xenvif_fatal_tx_err(queue->vif); |
b3f980bd | 1386 | break; |
f942dc25 IC |
1387 | } |
1388 | ||
e9ce7cb6 WL |
1389 | index = pending_index(queue->pending_cons); |
1390 | pending_idx = queue->pending_ring[index]; | |
f942dc25 | 1391 | |
7e5d7753 | 1392 | data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN && |
37641494 | 1393 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
7e5d7753 | 1394 | XEN_NETBACK_TX_COPY_LEN : txreq.size; |
f942dc25 | 1395 | |
e3377f36 | 1396 | skb = xenvif_alloc_skb(data_len); |
f942dc25 | 1397 | if (unlikely(skb == NULL)) { |
e9ce7cb6 | 1398 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1399 | "Can't allocate a skb in start_xmit.\n"); |
e9ce7cb6 | 1400 | xenvif_tx_err(queue, &txreq, idx); |
f942dc25 IC |
1401 | break; |
1402 | } | |
1403 | ||
2475b225 RL |
1404 | skb_shinfo(skb)->nr_frags = ret; |
1405 | if (data_len < txreq.size) | |
1406 | skb_shinfo(skb)->nr_frags++; | |
1407 | /* At this point shinfo->nr_frags is in fact the number of | |
1408 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. | |
1409 | */ | |
1410 | frag_overflow = 0; | |
1411 | nskb = NULL; | |
1412 | if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) { | |
1413 | frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS; | |
1414 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
1415 | skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS; | |
1416 | nskb = xenvif_alloc_skb(0); | |
1417 | if (unlikely(nskb == NULL)) { | |
1418 | kfree_skb(skb); | |
1419 | xenvif_tx_err(queue, &txreq, idx); | |
1420 | if (net_ratelimit()) | |
1421 | netdev_err(queue->vif->dev, | |
1422 | "Can't allocate the frag_list skb.\n"); | |
1423 | break; | |
1424 | } | |
1425 | } | |
1426 | ||
f942dc25 IC |
1427 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1428 | struct xen_netif_extra_info *gso; | |
1429 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1430 | ||
e9ce7cb6 | 1431 | if (xenvif_set_skb_gso(queue->vif, skb, gso)) { |
7376419a | 1432 | /* Failure in xenvif_set_skb_gso is fatal. */ |
f942dc25 | 1433 | kfree_skb(skb); |
2475b225 | 1434 | kfree_skb(nskb); |
b3f980bd | 1435 | break; |
f942dc25 IC |
1436 | } |
1437 | } | |
1438 | ||
8f13dd96 | 1439 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1440 | |
1441 | __skb_put(skb, data_len); | |
e9ce7cb6 WL |
1442 | queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
1443 | queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid; | |
1444 | queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
bdab8275 | 1445 | |
e9ce7cb6 | 1446 | queue->tx_copy_ops[*copy_ops].dest.u.gmfn = |
0df4f266 | 1447 | virt_to_gfn(skb->data); |
e9ce7cb6 WL |
1448 | queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; |
1449 | queue->tx_copy_ops[*copy_ops].dest.offset = | |
d0089e8a | 1450 | offset_in_page(skb->data) & ~XEN_PAGE_MASK; |
bdab8275 | 1451 | |
e9ce7cb6 WL |
1452 | queue->tx_copy_ops[*copy_ops].len = data_len; |
1453 | queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
bdab8275 ZK |
1454 | |
1455 | (*copy_ops)++; | |
f942dc25 | 1456 | |
f942dc25 | 1457 | if (data_len < txreq.size) { |
ea066ad1 IC |
1458 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1459 | pending_idx); | |
e9ce7cb6 | 1460 | xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop); |
bdab8275 | 1461 | gop++; |
f942dc25 | 1462 | } else { |
ea066ad1 IC |
1463 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1464 | INVALID_PENDING_IDX); | |
e9ce7cb6 | 1465 | memcpy(&queue->pending_tx_info[pending_idx].req, &txreq, |
bdab8275 | 1466 | sizeof(txreq)); |
f942dc25 IC |
1467 | } |
1468 | ||
e9ce7cb6 | 1469 | queue->pending_cons++; |
f942dc25 | 1470 | |
2475b225 RL |
1471 | gop = xenvif_get_requests(queue, skb, txfrags, gop, |
1472 | frag_overflow, nskb); | |
f942dc25 | 1473 | |
e9ce7cb6 | 1474 | __skb_queue_tail(&queue->tx_queue, skb); |
1e0b6eac | 1475 | |
e9ce7cb6 | 1476 | queue->tx.req_cons = idx; |
f942dc25 | 1477 | |
e9ce7cb6 WL |
1478 | if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) || |
1479 | (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops))) | |
f942dc25 IC |
1480 | break; |
1481 | } | |
1482 | ||
e9ce7cb6 | 1483 | (*map_ops) = gop - queue->tx_map_ops; |
bdab8275 | 1484 | return; |
f942dc25 IC |
1485 | } |
1486 | ||
e3377f36 ZK |
1487 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1488 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1489 | */ | |
e9ce7cb6 | 1490 | static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb) |
e3377f36 ZK |
1491 | { |
1492 | unsigned int offset = skb_headlen(skb); | |
1493 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
49d9991a | 1494 | int i, f; |
e3377f36 ZK |
1495 | struct ubuf_info *uarg; |
1496 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1497 | ||
e9ce7cb6 WL |
1498 | queue->stats.tx_zerocopy_sent += 2; |
1499 | queue->stats.tx_frag_overflow++; | |
e3377f36 | 1500 | |
e9ce7cb6 | 1501 | xenvif_fill_frags(queue, nskb); |
e3377f36 ZK |
1502 | /* Subtract frags size, we will correct it later */ |
1503 | skb->truesize -= skb->data_len; | |
1504 | skb->len += nskb->len; | |
1505 | skb->data_len += nskb->len; | |
1506 | ||
1507 | /* create a brand new frags array and coalesce there */ | |
1508 | for (i = 0; offset < skb->len; i++) { | |
1509 | struct page *page; | |
1510 | unsigned int len; | |
1511 | ||
1512 | BUG_ON(i >= MAX_SKB_FRAGS); | |
44cc8ed1 | 1513 | page = alloc_page(GFP_ATOMIC); |
e3377f36 ZK |
1514 | if (!page) { |
1515 | int j; | |
1516 | skb->truesize += skb->data_len; | |
1517 | for (j = 0; j < i; j++) | |
1518 | put_page(frags[j].page.p); | |
1519 | return -ENOMEM; | |
1520 | } | |
1521 | ||
1522 | if (offset + PAGE_SIZE < skb->len) | |
1523 | len = PAGE_SIZE; | |
1524 | else | |
1525 | len = skb->len - offset; | |
1526 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1527 | BUG(); | |
1528 | ||
1529 | offset += len; | |
1530 | frags[i].page.p = page; | |
1531 | frags[i].page_offset = 0; | |
1532 | skb_frag_size_set(&frags[i], len); | |
1533 | } | |
49d9991a | 1534 | |
b0c21bad DV |
1535 | /* Copied all the bits from the frag list -- free it. */ |
1536 | skb_frag_list_init(skb); | |
1537 | xenvif_skb_zerocopy_prepare(queue, nskb); | |
1538 | kfree_skb(nskb); | |
1539 | ||
49d9991a DV |
1540 | /* Release all the original (foreign) frags. */ |
1541 | for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) | |
1542 | skb_frag_unref(skb, f); | |
e3377f36 | 1543 | uarg = skb_shinfo(skb)->destructor_arg; |
a64bd934 WL |
1544 | /* increase inflight counter to offset decrement in callback */ |
1545 | atomic_inc(&queue->inflight_packets); | |
e3377f36 ZK |
1546 | uarg->callback(uarg, true); |
1547 | skb_shinfo(skb)->destructor_arg = NULL; | |
1548 | ||
b0c21bad DV |
1549 | /* Fill the skb with the new (local) frags. */ |
1550 | memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t)); | |
1551 | skb_shinfo(skb)->nr_frags = i; | |
1552 | skb->truesize += i * PAGE_SIZE; | |
e3377f36 ZK |
1553 | |
1554 | return 0; | |
1555 | } | |
b3f980bd | 1556 | |
e9ce7cb6 | 1557 | static int xenvif_tx_submit(struct xenvif_queue *queue) |
f942dc25 | 1558 | { |
e9ce7cb6 WL |
1559 | struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops; |
1560 | struct gnttab_copy *gop_copy = queue->tx_copy_ops; | |
f942dc25 | 1561 | struct sk_buff *skb; |
b3f980bd | 1562 | int work_done = 0; |
f942dc25 | 1563 | |
e9ce7cb6 | 1564 | while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) { |
f942dc25 | 1565 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1566 | u16 pending_idx; |
1567 | unsigned data_len; | |
1568 | ||
8f13dd96 | 1569 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
e9ce7cb6 | 1570 | txp = &queue->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1571 | |
1572 | /* Check the remap error code. */ | |
e9ce7cb6 | 1573 | if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) { |
b42cc6e4 ZK |
1574 | /* If there was an error, xenvif_tx_check_gop is |
1575 | * expected to release all the frags which were mapped, | |
1576 | * so kfree_skb shouldn't do it again | |
1577 | */ | |
f942dc25 | 1578 | skb_shinfo(skb)->nr_frags = 0; |
b42cc6e4 ZK |
1579 | if (skb_has_frag_list(skb)) { |
1580 | struct sk_buff *nskb = | |
1581 | skb_shinfo(skb)->frag_list; | |
1582 | skb_shinfo(nskb)->nr_frags = 0; | |
1583 | } | |
f942dc25 IC |
1584 | kfree_skb(skb); |
1585 | continue; | |
1586 | } | |
1587 | ||
1588 | data_len = skb->len; | |
e9ce7cb6 | 1589 | callback_param(queue, pending_idx).ctx = NULL; |
f942dc25 IC |
1590 | if (data_len < txp->size) { |
1591 | /* Append the packet payload as a fragment. */ | |
1592 | txp->offset += data_len; | |
1593 | txp->size -= data_len; | |
1594 | } else { | |
1595 | /* Schedule a response immediately. */ | |
e9ce7cb6 | 1596 | xenvif_idx_release(queue, pending_idx, |
bdab8275 | 1597 | XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1598 | } |
1599 | ||
1600 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1601 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1602 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1603 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1604 | ||
e9ce7cb6 | 1605 | xenvif_fill_frags(queue, skb); |
f942dc25 | 1606 | |
e3377f36 | 1607 | if (unlikely(skb_has_frag_list(skb))) { |
e9ce7cb6 | 1608 | if (xenvif_handle_frag_list(queue, skb)) { |
e3377f36 | 1609 | if (net_ratelimit()) |
e9ce7cb6 | 1610 | netdev_err(queue->vif->dev, |
e3377f36 | 1611 | "Not enough memory to consolidate frag_list!\n"); |
a64bd934 | 1612 | xenvif_skb_zerocopy_prepare(queue, skb); |
e3377f36 ZK |
1613 | kfree_skb(skb); |
1614 | continue; | |
1615 | } | |
1616 | } | |
1617 | ||
e9ce7cb6 | 1618 | skb->dev = queue->vif->dev; |
f942dc25 | 1619 | skb->protocol = eth_type_trans(skb, skb->dev); |
f9ca8f74 | 1620 | skb_reset_network_header(skb); |
f942dc25 | 1621 | |
e9ce7cb6 WL |
1622 | if (checksum_setup(queue, skb)) { |
1623 | netdev_dbg(queue->vif->dev, | |
f942dc25 | 1624 | "Can't setup checksum in net_tx_action\n"); |
f53c3fe8 ZK |
1625 | /* We have to set this flag to trigger the callback */ |
1626 | if (skb_shinfo(skb)->destructor_arg) | |
a64bd934 | 1627 | xenvif_skb_zerocopy_prepare(queue, skb); |
f942dc25 IC |
1628 | kfree_skb(skb); |
1629 | continue; | |
1630 | } | |
1631 | ||
40893fd0 | 1632 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1633 | |
b89587a7 PD |
1634 | /* If the packet is GSO then we will have just set up the |
1635 | * transport header offset in checksum_setup so it's now | |
1636 | * straightforward to calculate gso_segs. | |
1637 | */ | |
1638 | if (skb_is_gso(skb)) { | |
1639 | int mss = skb_shinfo(skb)->gso_size; | |
1640 | int hdrlen = skb_transport_header(skb) - | |
1641 | skb_mac_header(skb) + | |
1642 | tcp_hdrlen(skb); | |
1643 | ||
1644 | skb_shinfo(skb)->gso_segs = | |
1645 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1646 | } | |
1647 | ||
e9ce7cb6 WL |
1648 | queue->stats.rx_bytes += skb->len; |
1649 | queue->stats.rx_packets++; | |
f942dc25 | 1650 | |
b3f980bd WL |
1651 | work_done++; |
1652 | ||
f53c3fe8 ZK |
1653 | /* Set this flag right before netif_receive_skb, otherwise |
1654 | * someone might think this packet already left netback, and | |
1655 | * do a skb_copy_ubufs while we are still in control of the | |
1656 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1657 | */ | |
1bb332af | 1658 | if (skb_shinfo(skb)->destructor_arg) { |
a64bd934 | 1659 | xenvif_skb_zerocopy_prepare(queue, skb); |
e9ce7cb6 | 1660 | queue->stats.tx_zerocopy_sent++; |
1bb332af | 1661 | } |
f53c3fe8 | 1662 | |
b3f980bd | 1663 | netif_receive_skb(skb); |
f942dc25 | 1664 | } |
b3f980bd WL |
1665 | |
1666 | return work_done; | |
f942dc25 IC |
1667 | } |
1668 | ||
3e2234b3 ZK |
1669 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1670 | { | |
f53c3fe8 ZK |
1671 | unsigned long flags; |
1672 | pending_ring_idx_t index; | |
e9ce7cb6 | 1673 | struct xenvif_queue *queue = ubuf_to_queue(ubuf); |
f53c3fe8 ZK |
1674 | |
1675 | /* This is the only place where we grab this lock, to protect callbacks | |
1676 | * from each other. | |
1677 | */ | |
e9ce7cb6 | 1678 | spin_lock_irqsave(&queue->callback_lock, flags); |
f53c3fe8 ZK |
1679 | do { |
1680 | u16 pending_idx = ubuf->desc; | |
1681 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
e9ce7cb6 | 1682 | BUG_ON(queue->dealloc_prod - queue->dealloc_cons >= |
f53c3fe8 | 1683 | MAX_PENDING_REQS); |
e9ce7cb6 WL |
1684 | index = pending_index(queue->dealloc_prod); |
1685 | queue->dealloc_ring[index] = pending_idx; | |
f53c3fe8 ZK |
1686 | /* Sync with xenvif_tx_dealloc_action: |
1687 | * insert idx then incr producer. | |
1688 | */ | |
1689 | smp_wmb(); | |
e9ce7cb6 | 1690 | queue->dealloc_prod++; |
f53c3fe8 | 1691 | } while (ubuf); |
e9ce7cb6 | 1692 | spin_unlock_irqrestore(&queue->callback_lock, flags); |
f53c3fe8 | 1693 | |
1bb332af | 1694 | if (likely(zerocopy_success)) |
e9ce7cb6 | 1695 | queue->stats.tx_zerocopy_success++; |
1bb332af | 1696 | else |
e9ce7cb6 | 1697 | queue->stats.tx_zerocopy_fail++; |
a64bd934 | 1698 | xenvif_skb_zerocopy_complete(queue); |
f53c3fe8 ZK |
1699 | } |
1700 | ||
e9ce7cb6 | 1701 | static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue) |
f53c3fe8 ZK |
1702 | { |
1703 | struct gnttab_unmap_grant_ref *gop; | |
1704 | pending_ring_idx_t dc, dp; | |
1705 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1706 | unsigned int i = 0; | |
1707 | ||
e9ce7cb6 WL |
1708 | dc = queue->dealloc_cons; |
1709 | gop = queue->tx_unmap_ops; | |
f53c3fe8 ZK |
1710 | |
1711 | /* Free up any grants we have finished using */ | |
1712 | do { | |
e9ce7cb6 | 1713 | dp = queue->dealloc_prod; |
f53c3fe8 ZK |
1714 | |
1715 | /* Ensure we see all indices enqueued by all | |
1716 | * xenvif_zerocopy_callback(). | |
1717 | */ | |
1718 | smp_rmb(); | |
1719 | ||
1720 | while (dc != dp) { | |
50c2e4dd | 1721 | BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS); |
f53c3fe8 | 1722 | pending_idx = |
e9ce7cb6 | 1723 | queue->dealloc_ring[pending_index(dc++)]; |
f53c3fe8 | 1724 | |
50c2e4dd | 1725 | pending_idx_release[gop - queue->tx_unmap_ops] = |
f53c3fe8 | 1726 | pending_idx; |
50c2e4dd | 1727 | queue->pages_to_unmap[gop - queue->tx_unmap_ops] = |
e9ce7cb6 | 1728 | queue->mmap_pages[pending_idx]; |
f53c3fe8 | 1729 | gnttab_set_unmap_op(gop, |
e9ce7cb6 | 1730 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1731 | GNTMAP_host_map, |
e9ce7cb6 WL |
1732 | queue->grant_tx_handle[pending_idx]); |
1733 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1734 | ++gop; |
1735 | } | |
1736 | ||
e9ce7cb6 | 1737 | } while (dp != queue->dealloc_prod); |
f53c3fe8 | 1738 | |
e9ce7cb6 | 1739 | queue->dealloc_cons = dc; |
f53c3fe8 | 1740 | |
e9ce7cb6 | 1741 | if (gop - queue->tx_unmap_ops > 0) { |
f53c3fe8 | 1742 | int ret; |
e9ce7cb6 | 1743 | ret = gnttab_unmap_refs(queue->tx_unmap_ops, |
f53c3fe8 | 1744 | NULL, |
e9ce7cb6 WL |
1745 | queue->pages_to_unmap, |
1746 | gop - queue->tx_unmap_ops); | |
f53c3fe8 | 1747 | if (ret) { |
68946159 | 1748 | netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n", |
e9ce7cb6 WL |
1749 | gop - queue->tx_unmap_ops, ret); |
1750 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) { | |
f53c3fe8 | 1751 | if (gop[i].status != GNTST_okay) |
e9ce7cb6 | 1752 | netdev_err(queue->vif->dev, |
68946159 | 1753 | " host_addr: 0x%llx handle: 0x%x status: %d\n", |
f53c3fe8 ZK |
1754 | gop[i].host_addr, |
1755 | gop[i].handle, | |
1756 | gop[i].status); | |
1757 | } | |
1758 | BUG(); | |
1759 | } | |
1760 | } | |
1761 | ||
e9ce7cb6 WL |
1762 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) |
1763 | xenvif_idx_release(queue, pending_idx_release[i], | |
f53c3fe8 | 1764 | XEN_NETIF_RSP_OKAY); |
3e2234b3 ZK |
1765 | } |
1766 | ||
f53c3fe8 | 1767 | |
f942dc25 | 1768 | /* Called after netfront has transmitted */ |
e9ce7cb6 | 1769 | int xenvif_tx_action(struct xenvif_queue *queue, int budget) |
f942dc25 | 1770 | { |
bdab8275 | 1771 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1772 | int work_done, ret; |
f942dc25 | 1773 | |
e9ce7cb6 | 1774 | if (unlikely(!tx_work_todo(queue))) |
b3f980bd WL |
1775 | return 0; |
1776 | ||
e9ce7cb6 | 1777 | xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops); |
f942dc25 | 1778 | |
bdab8275 | 1779 | if (nr_cops == 0) |
b3f980bd WL |
1780 | return 0; |
1781 | ||
e9ce7cb6 | 1782 | gnttab_batch_copy(queue->tx_copy_ops, nr_cops); |
bdab8275 | 1783 | if (nr_mops != 0) { |
e9ce7cb6 | 1784 | ret = gnttab_map_refs(queue->tx_map_ops, |
bdab8275 | 1785 | NULL, |
e9ce7cb6 | 1786 | queue->pages_to_map, |
bdab8275 ZK |
1787 | nr_mops); |
1788 | BUG_ON(ret); | |
1789 | } | |
f942dc25 | 1790 | |
e9ce7cb6 | 1791 | work_done = xenvif_tx_submit(queue); |
f942dc25 | 1792 | |
b3f980bd | 1793 | return work_done; |
f942dc25 IC |
1794 | } |
1795 | ||
e9ce7cb6 | 1796 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a | 1797 | u8 status) |
f942dc25 | 1798 | { |
f942dc25 | 1799 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1800 | pending_ring_idx_t index; |
f53c3fe8 | 1801 | unsigned long flags; |
2810e5b9 | 1802 | |
e9ce7cb6 | 1803 | pending_tx_info = &queue->pending_tx_info[pending_idx]; |
7fbb9d84 | 1804 | |
e9ce7cb6 | 1805 | spin_lock_irqsave(&queue->response_lock, flags); |
7fbb9d84 | 1806 | |
e9ce7cb6 | 1807 | make_tx_response(queue, &pending_tx_info->req, status); |
7fbb9d84 DV |
1808 | |
1809 | /* Release the pending index before pusing the Tx response so | |
1810 | * its available before a new Tx request is pushed by the | |
1811 | * frontend. | |
1812 | */ | |
1813 | index = pending_index(queue->pending_prod++); | |
e9ce7cb6 | 1814 | queue->pending_ring[index] = pending_idx; |
7fbb9d84 | 1815 | |
c8a4d299 | 1816 | push_tx_responses(queue); |
7fbb9d84 | 1817 | |
e9ce7cb6 | 1818 | spin_unlock_irqrestore(&queue->response_lock, flags); |
f942dc25 IC |
1819 | } |
1820 | ||
2810e5b9 | 1821 | |
e9ce7cb6 | 1822 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1823 | struct xen_netif_tx_request *txp, |
1824 | s8 st) | |
1825 | { | |
e9ce7cb6 | 1826 | RING_IDX i = queue->tx.rsp_prod_pvt; |
f942dc25 | 1827 | struct xen_netif_tx_response *resp; |
f942dc25 | 1828 | |
e9ce7cb6 | 1829 | resp = RING_GET_RESPONSE(&queue->tx, i); |
f942dc25 IC |
1830 | resp->id = txp->id; |
1831 | resp->status = st; | |
1832 | ||
1833 | if (txp->flags & XEN_NETTXF_extra_info) | |
e9ce7cb6 | 1834 | RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL; |
f942dc25 | 1835 | |
e9ce7cb6 | 1836 | queue->tx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1837 | } |
1838 | ||
c8a4d299 DV |
1839 | static void push_tx_responses(struct xenvif_queue *queue) |
1840 | { | |
1841 | int notify; | |
1842 | ||
1843 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify); | |
1844 | if (notify) | |
1845 | notify_remote_via_irq(queue->tx_irq); | |
1846 | } | |
1847 | ||
e9ce7cb6 | 1848 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1849 | u16 id, |
1850 | s8 st, | |
1851 | u16 offset, | |
1852 | u16 size, | |
1853 | u16 flags) | |
1854 | { | |
e9ce7cb6 | 1855 | RING_IDX i = queue->rx.rsp_prod_pvt; |
f942dc25 IC |
1856 | struct xen_netif_rx_response *resp; |
1857 | ||
e9ce7cb6 | 1858 | resp = RING_GET_RESPONSE(&queue->rx, i); |
f942dc25 IC |
1859 | resp->offset = offset; |
1860 | resp->flags = flags; | |
1861 | resp->id = id; | |
1862 | resp->status = (s16)size; | |
1863 | if (st < 0) | |
1864 | resp->status = (s16)st; | |
1865 | ||
e9ce7cb6 | 1866 | queue->rx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1867 | |
1868 | return resp; | |
1869 | } | |
1870 | ||
e9ce7cb6 | 1871 | void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx) |
f53c3fe8 ZK |
1872 | { |
1873 | int ret; | |
1874 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1875 | ||
1876 | gnttab_set_unmap_op(&tx_unmap_op, | |
e9ce7cb6 | 1877 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1878 | GNTMAP_host_map, |
e9ce7cb6 WL |
1879 | queue->grant_tx_handle[pending_idx]); |
1880 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1881 | |
1882 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
e9ce7cb6 | 1883 | &queue->mmap_pages[pending_idx], 1); |
7aceb47a | 1884 | if (ret) { |
e9ce7cb6 | 1885 | netdev_err(queue->vif->dev, |
68946159 | 1886 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n", |
7aceb47a ZK |
1887 | ret, |
1888 | pending_idx, | |
1889 | tx_unmap_op.host_addr, | |
1890 | tx_unmap_op.handle, | |
1891 | tx_unmap_op.status); | |
1892 | BUG(); | |
1893 | } | |
f53c3fe8 ZK |
1894 | } |
1895 | ||
e9ce7cb6 | 1896 | static inline int tx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1897 | { |
e9ce7cb6 | 1898 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))) |
f942dc25 IC |
1899 | return 1; |
1900 | ||
1901 | return 0; | |
1902 | } | |
1903 | ||
e9ce7cb6 | 1904 | static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue) |
f53c3fe8 | 1905 | { |
e9ce7cb6 | 1906 | return queue->dealloc_cons != queue->dealloc_prod; |
f53c3fe8 ZK |
1907 | } |
1908 | ||
e9ce7cb6 | 1909 | void xenvif_unmap_frontend_rings(struct xenvif_queue *queue) |
f942dc25 | 1910 | { |
e9ce7cb6 WL |
1911 | if (queue->tx.sring) |
1912 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1913 | queue->tx.sring); | |
1914 | if (queue->rx.sring) | |
1915 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1916 | queue->rx.sring); | |
f942dc25 IC |
1917 | } |
1918 | ||
e9ce7cb6 | 1919 | int xenvif_map_frontend_rings(struct xenvif_queue *queue, |
7376419a WL |
1920 | grant_ref_t tx_ring_ref, |
1921 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1922 | { |
c9d63699 | 1923 | void *addr; |
f942dc25 IC |
1924 | struct xen_netif_tx_sring *txs; |
1925 | struct xen_netif_rx_sring *rxs; | |
1926 | ||
1927 | int err = -ENOMEM; | |
1928 | ||
e9ce7cb6 | 1929 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
ccc9d90a | 1930 | &tx_ring_ref, 1, &addr); |
c9d63699 | 1931 | if (err) |
f942dc25 IC |
1932 | goto err; |
1933 | ||
c9d63699 | 1934 | txs = (struct xen_netif_tx_sring *)addr; |
d0089e8a | 1935 | BACK_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
f942dc25 | 1936 | |
e9ce7cb6 | 1937 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
ccc9d90a | 1938 | &rx_ring_ref, 1, &addr); |
c9d63699 | 1939 | if (err) |
f942dc25 | 1940 | goto err; |
f942dc25 | 1941 | |
c9d63699 | 1942 | rxs = (struct xen_netif_rx_sring *)addr; |
d0089e8a | 1943 | BACK_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
f942dc25 IC |
1944 | |
1945 | return 0; | |
1946 | ||
1947 | err: | |
e9ce7cb6 | 1948 | xenvif_unmap_frontend_rings(queue); |
f942dc25 IC |
1949 | return err; |
1950 | } | |
1951 | ||
ecf08d2d | 1952 | static void xenvif_queue_carrier_off(struct xenvif_queue *queue) |
ca2f09f2 | 1953 | { |
ecf08d2d DV |
1954 | struct xenvif *vif = queue->vif; |
1955 | ||
1956 | queue->stalled = true; | |
1957 | ||
1958 | /* At least one queue has stalled? Disable the carrier. */ | |
1959 | spin_lock(&vif->lock); | |
1960 | if (vif->stalled_queues++ == 0) { | |
1961 | netdev_info(vif->dev, "Guest Rx stalled"); | |
1962 | netif_carrier_off(vif->dev); | |
1963 | } | |
1964 | spin_unlock(&vif->lock); | |
ca2f09f2 PD |
1965 | } |
1966 | ||
ecf08d2d | 1967 | static void xenvif_queue_carrier_on(struct xenvif_queue *queue) |
f34a4cf9 | 1968 | { |
ecf08d2d | 1969 | struct xenvif *vif = queue->vif; |
f34a4cf9 | 1970 | |
ecf08d2d DV |
1971 | queue->last_rx_time = jiffies; /* Reset Rx stall detection. */ |
1972 | queue->stalled = false; | |
f34a4cf9 | 1973 | |
ecf08d2d DV |
1974 | /* All queues are ready? Enable the carrier. */ |
1975 | spin_lock(&vif->lock); | |
1976 | if (--vif->stalled_queues == 0) { | |
1977 | netdev_info(vif->dev, "Guest Rx ready"); | |
1978 | netif_carrier_on(vif->dev); | |
1979 | } | |
1980 | spin_unlock(&vif->lock); | |
1981 | } | |
f34a4cf9 | 1982 | |
ecf08d2d DV |
1983 | static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue) |
1984 | { | |
1985 | RING_IDX prod, cons; | |
1986 | ||
1987 | prod = queue->rx.sring->req_prod; | |
1988 | cons = queue->rx.req_cons; | |
1989 | ||
1d5d4852 | 1990 | return !queue->stalled && prod - cons < 1 |
ecf08d2d | 1991 | && time_after(jiffies, |
26c0e102 | 1992 | queue->last_rx_time + queue->vif->stall_timeout); |
ecf08d2d DV |
1993 | } |
1994 | ||
1995 | static bool xenvif_rx_queue_ready(struct xenvif_queue *queue) | |
1996 | { | |
1997 | RING_IDX prod, cons; | |
1998 | ||
1999 | prod = queue->rx.sring->req_prod; | |
2000 | cons = queue->rx.req_cons; | |
2001 | ||
1d5d4852 | 2002 | return queue->stalled && prod - cons >= 1; |
ecf08d2d DV |
2003 | } |
2004 | ||
f48da8b1 | 2005 | static bool xenvif_have_rx_work(struct xenvif_queue *queue) |
ca2f09f2 | 2006 | { |
99a2dea5 | 2007 | return xenvif_rx_ring_slots_available(queue) |
26c0e102 DV |
2008 | || (queue->vif->stall_timeout && |
2009 | (xenvif_rx_queue_stalled(queue) | |
2010 | || xenvif_rx_queue_ready(queue))) | |
f48da8b1 DV |
2011 | || kthread_should_stop() |
2012 | || queue->vif->disabled; | |
ca2f09f2 PD |
2013 | } |
2014 | ||
f48da8b1 | 2015 | static long xenvif_rx_queue_timeout(struct xenvif_queue *queue) |
f34a4cf9 | 2016 | { |
f48da8b1 DV |
2017 | struct sk_buff *skb; |
2018 | long timeout; | |
f34a4cf9 | 2019 | |
f48da8b1 DV |
2020 | skb = skb_peek(&queue->rx_queue); |
2021 | if (!skb) | |
2022 | return MAX_SCHEDULE_TIMEOUT; | |
f34a4cf9 | 2023 | |
f48da8b1 DV |
2024 | timeout = XENVIF_RX_CB(skb)->expires - jiffies; |
2025 | return timeout < 0 ? 0 : timeout; | |
2026 | } | |
f34a4cf9 | 2027 | |
f48da8b1 DV |
2028 | /* Wait until the guest Rx thread has work. |
2029 | * | |
2030 | * The timeout needs to be adjusted based on the current head of the | |
2031 | * queue (and not just the head at the beginning). In particular, if | |
2032 | * the queue is initially empty an infinite timeout is used and this | |
2033 | * needs to be reduced when a skb is queued. | |
2034 | * | |
2035 | * This cannot be done with wait_event_timeout() because it only | |
2036 | * calculates the timeout once. | |
2037 | */ | |
2038 | static void xenvif_wait_for_rx_work(struct xenvif_queue *queue) | |
2039 | { | |
2040 | DEFINE_WAIT(wait); | |
2041 | ||
2042 | if (xenvif_have_rx_work(queue)) | |
2043 | return; | |
2044 | ||
2045 | for (;;) { | |
2046 | long ret; | |
2047 | ||
2048 | prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE); | |
2049 | if (xenvif_have_rx_work(queue)) | |
2050 | break; | |
2051 | ret = schedule_timeout(xenvif_rx_queue_timeout(queue)); | |
2052 | if (!ret) | |
2053 | break; | |
f34a4cf9 | 2054 | } |
f48da8b1 | 2055 | finish_wait(&queue->wq, &wait); |
f34a4cf9 ZK |
2056 | } |
2057 | ||
121fa4b7 | 2058 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd | 2059 | { |
e9ce7cb6 | 2060 | struct xenvif_queue *queue = data; |
f48da8b1 | 2061 | struct xenvif *vif = queue->vif; |
b3f980bd | 2062 | |
26c0e102 DV |
2063 | if (!vif->stall_timeout) |
2064 | xenvif_queue_carrier_on(queue); | |
2065 | ||
f48da8b1 DV |
2066 | for (;;) { |
2067 | xenvif_wait_for_rx_work(queue); | |
e9d8b2c2 | 2068 | |
f34a4cf9 ZK |
2069 | if (kthread_should_stop()) |
2070 | break; | |
2071 | ||
e9d8b2c2 WL |
2072 | /* This frontend is found to be rogue, disable it in |
2073 | * kthread context. Currently this is only set when | |
2074 | * netback finds out frontend sends malformed packet, | |
2075 | * but we cannot disable the interface in softirq | |
e9ce7cb6 WL |
2076 | * context so we defer it here, if this thread is |
2077 | * associated with queue 0. | |
e9d8b2c2 | 2078 | */ |
f48da8b1 DV |
2079 | if (unlikely(vif->disabled && queue->id == 0)) { |
2080 | xenvif_carrier_off(vif); | |
42b5212f | 2081 | break; |
09350788 ZK |
2082 | } |
2083 | ||
e9ce7cb6 WL |
2084 | if (!skb_queue_empty(&queue->rx_queue)) |
2085 | xenvif_rx_action(queue); | |
b3f980bd | 2086 | |
ecf08d2d DV |
2087 | /* If the guest hasn't provided any Rx slots for a |
2088 | * while it's probably not responsive, drop the | |
2089 | * carrier so packets are dropped earlier. | |
2090 | */ | |
26c0e102 DV |
2091 | if (vif->stall_timeout) { |
2092 | if (xenvif_rx_queue_stalled(queue)) | |
2093 | xenvif_queue_carrier_off(queue); | |
2094 | else if (xenvif_rx_queue_ready(queue)) | |
2095 | xenvif_queue_carrier_on(queue); | |
2096 | } | |
ecf08d2d | 2097 | |
f48da8b1 DV |
2098 | /* Queued packets may have foreign pages from other |
2099 | * domains. These cannot be queued indefinitely as | |
2100 | * this would starve guests of grant refs and transmit | |
2101 | * slots. | |
2102 | */ | |
2103 | xenvif_rx_queue_drop_expired(queue); | |
2104 | ||
2105 | xenvif_rx_queue_maybe_wake(queue); | |
2106 | ||
b3f980bd WL |
2107 | cond_resched(); |
2108 | } | |
2109 | ||
ca2f09f2 | 2110 | /* Bin any remaining skbs */ |
f48da8b1 | 2111 | xenvif_rx_queue_purge(queue); |
ca2f09f2 | 2112 | |
b3f980bd WL |
2113 | return 0; |
2114 | } | |
2115 | ||
a64bd934 WL |
2116 | static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue) |
2117 | { | |
2118 | /* Dealloc thread must remain running until all inflight | |
2119 | * packets complete. | |
2120 | */ | |
2121 | return kthread_should_stop() && | |
2122 | !atomic_read(&queue->inflight_packets); | |
2123 | } | |
2124 | ||
f53c3fe8 ZK |
2125 | int xenvif_dealloc_kthread(void *data) |
2126 | { | |
e9ce7cb6 | 2127 | struct xenvif_queue *queue = data; |
f53c3fe8 | 2128 | |
a64bd934 | 2129 | for (;;) { |
e9ce7cb6 WL |
2130 | wait_event_interruptible(queue->dealloc_wq, |
2131 | tx_dealloc_work_todo(queue) || | |
a64bd934 WL |
2132 | xenvif_dealloc_kthread_should_stop(queue)); |
2133 | if (xenvif_dealloc_kthread_should_stop(queue)) | |
f53c3fe8 ZK |
2134 | break; |
2135 | ||
e9ce7cb6 | 2136 | xenvif_tx_dealloc_action(queue); |
f53c3fe8 ZK |
2137 | cond_resched(); |
2138 | } | |
2139 | ||
2140 | /* Unmap anything remaining*/ | |
e9ce7cb6 WL |
2141 | if (tx_dealloc_work_todo(queue)) |
2142 | xenvif_tx_dealloc_action(queue); | |
f53c3fe8 ZK |
2143 | |
2144 | return 0; | |
2145 | } | |
2146 | ||
f942dc25 IC |
2147 | static int __init netback_init(void) |
2148 | { | |
f942dc25 | 2149 | int rc = 0; |
f942dc25 | 2150 | |
2a14b244 | 2151 | if (!xen_domain()) |
f942dc25 IC |
2152 | return -ENODEV; |
2153 | ||
4c82ac3c WL |
2154 | /* Allow as many queues as there are CPUs if user has not |
2155 | * specified a value. | |
2156 | */ | |
2157 | if (xenvif_max_queues == 0) | |
2158 | xenvif_max_queues = num_online_cpus(); | |
8d3d53b3 | 2159 | |
37641494 | 2160 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
2161 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
2162 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 2163 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
2164 | } |
2165 | ||
f942dc25 IC |
2166 | rc = xenvif_xenbus_init(); |
2167 | if (rc) | |
2168 | goto failed_init; | |
2169 | ||
f51de243 ZK |
2170 | #ifdef CONFIG_DEBUG_FS |
2171 | xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL); | |
2172 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
2173 | pr_warn("Init of debugfs returned %ld!\n", | |
2174 | PTR_ERR(xen_netback_dbg_root)); | |
2175 | #endif /* CONFIG_DEBUG_FS */ | |
2176 | ||
f942dc25 IC |
2177 | return 0; |
2178 | ||
2179 | failed_init: | |
f942dc25 | 2180 | return rc; |
f942dc25 IC |
2181 | } |
2182 | ||
2183 | module_init(netback_init); | |
2184 | ||
b103f358 WL |
2185 | static void __exit netback_fini(void) |
2186 | { | |
f51de243 ZK |
2187 | #ifdef CONFIG_DEBUG_FS |
2188 | if (!IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
2189 | debugfs_remove_recursive(xen_netback_dbg_root); | |
2190 | #endif /* CONFIG_DEBUG_FS */ | |
b103f358 | 2191 | xenvif_xenbus_fini(); |
b103f358 WL |
2192 | } |
2193 | module_exit(netback_fini); | |
2194 | ||
f942dc25 | 2195 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 2196 | MODULE_ALIAS("xen-backend:vif"); |