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 | 97 | struct xen_netif_tx_request *txp, |
562abd39 | 98 | unsigned int extra_count, |
f942dc25 | 99 | s8 st); |
c8a4d299 | 100 | static void push_tx_responses(struct xenvif_queue *queue); |
b3f980bd | 101 | |
e9ce7cb6 | 102 | static inline int tx_work_todo(struct xenvif_queue *queue); |
b3f980bd | 103 | |
e9ce7cb6 | 104 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
105 | u16 id, |
106 | s8 st, | |
107 | u16 offset, | |
108 | u16 size, | |
109 | u16 flags); | |
110 | ||
e9ce7cb6 | 111 | static inline unsigned long idx_to_pfn(struct xenvif_queue *queue, |
ea066ad1 | 112 | u16 idx) |
f942dc25 | 113 | { |
e9ce7cb6 | 114 | return page_to_pfn(queue->mmap_pages[idx]); |
f942dc25 IC |
115 | } |
116 | ||
e9ce7cb6 | 117 | static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue, |
ea066ad1 | 118 | u16 idx) |
f942dc25 | 119 | { |
e9ce7cb6 | 120 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx)); |
f942dc25 IC |
121 | } |
122 | ||
7aceb47a ZK |
123 | #define callback_param(vif, pending_idx) \ |
124 | (vif->pending_tx_info[pending_idx].callback_struct) | |
125 | ||
f53c3fe8 ZK |
126 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
127 | */ | |
e9ce7cb6 | 128 | static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf) |
3e2234b3 | 129 | { |
f53c3fe8 ZK |
130 | u16 pending_idx = ubuf->desc; |
131 | struct pending_tx_info *temp = | |
132 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
133 | return container_of(temp - pending_idx, | |
e9ce7cb6 | 134 | struct xenvif_queue, |
f53c3fe8 | 135 | pending_tx_info[0]); |
3e2234b3 | 136 | } |
f53c3fe8 | 137 | |
ea066ad1 IC |
138 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
139 | { | |
140 | return (u16)frag->page_offset; | |
141 | } | |
142 | ||
143 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
144 | { | |
145 | frag->page_offset = pending_idx; | |
146 | } | |
147 | ||
f942dc25 IC |
148 | static inline pending_ring_idx_t pending_index(unsigned i) |
149 | { | |
150 | return i & (MAX_PENDING_REQS-1); | |
151 | } | |
152 | ||
1d5d4852 | 153 | static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue) |
f942dc25 | 154 | { |
ca2f09f2 | 155 | RING_IDX prod, cons; |
99a2dea5 | 156 | struct sk_buff *skb; |
1d5d4852 DV |
157 | int needed; |
158 | ||
99a2dea5 DV |
159 | skb = skb_peek(&queue->rx_queue); |
160 | if (!skb) | |
161 | return false; | |
162 | ||
163 | needed = DIV_ROUND_UP(skb->len, XEN_PAGE_SIZE); | |
164 | if (skb_is_gso(skb)) | |
165 | needed++; | |
f942dc25 | 166 | |
ca2f09f2 | 167 | do { |
e9ce7cb6 WL |
168 | prod = queue->rx.sring->req_prod; |
169 | cons = queue->rx.req_cons; | |
f942dc25 | 170 | |
ca2f09f2 PD |
171 | if (prod - cons >= needed) |
172 | return true; | |
f942dc25 | 173 | |
e9ce7cb6 | 174 | queue->rx.sring->req_event = prod + 1; |
f942dc25 | 175 | |
ca2f09f2 PD |
176 | /* Make sure event is visible before we check prod |
177 | * again. | |
178 | */ | |
179 | mb(); | |
e9ce7cb6 | 180 | } while (queue->rx.sring->req_prod != prod); |
f942dc25 | 181 | |
ca2f09f2 | 182 | return false; |
f942dc25 IC |
183 | } |
184 | ||
f48da8b1 DV |
185 | void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb) |
186 | { | |
187 | unsigned long flags; | |
188 | ||
189 | spin_lock_irqsave(&queue->rx_queue.lock, flags); | |
190 | ||
191 | __skb_queue_tail(&queue->rx_queue, skb); | |
192 | ||
193 | queue->rx_queue_len += skb->len; | |
194 | if (queue->rx_queue_len > queue->rx_queue_max) | |
195 | netif_tx_stop_queue(netdev_get_tx_queue(queue->vif->dev, queue->id)); | |
196 | ||
197 | spin_unlock_irqrestore(&queue->rx_queue.lock, flags); | |
198 | } | |
199 | ||
200 | static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue) | |
201 | { | |
202 | struct sk_buff *skb; | |
203 | ||
204 | spin_lock_irq(&queue->rx_queue.lock); | |
205 | ||
206 | skb = __skb_dequeue(&queue->rx_queue); | |
207 | if (skb) | |
208 | queue->rx_queue_len -= skb->len; | |
209 | ||
210 | spin_unlock_irq(&queue->rx_queue.lock); | |
211 | ||
212 | return skb; | |
213 | } | |
214 | ||
215 | static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue) | |
216 | { | |
217 | spin_lock_irq(&queue->rx_queue.lock); | |
218 | ||
219 | if (queue->rx_queue_len < queue->rx_queue_max) | |
220 | netif_tx_wake_queue(netdev_get_tx_queue(queue->vif->dev, queue->id)); | |
221 | ||
222 | spin_unlock_irq(&queue->rx_queue.lock); | |
223 | } | |
224 | ||
225 | ||
226 | static void xenvif_rx_queue_purge(struct xenvif_queue *queue) | |
227 | { | |
228 | struct sk_buff *skb; | |
229 | while ((skb = xenvif_rx_dequeue(queue)) != NULL) | |
230 | kfree_skb(skb); | |
231 | } | |
232 | ||
233 | static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue) | |
234 | { | |
235 | struct sk_buff *skb; | |
236 | ||
237 | for(;;) { | |
238 | skb = skb_peek(&queue->rx_queue); | |
239 | if (!skb) | |
240 | break; | |
241 | if (time_before(jiffies, XENVIF_RX_CB(skb)->expires)) | |
242 | break; | |
243 | xenvif_rx_dequeue(queue); | |
244 | kfree_skb(skb); | |
245 | } | |
246 | } | |
247 | ||
f942dc25 IC |
248 | struct netrx_pending_operations { |
249 | unsigned copy_prod, copy_cons; | |
250 | unsigned meta_prod, meta_cons; | |
251 | struct gnttab_copy *copy; | |
b3f980bd | 252 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
253 | int copy_off; |
254 | grant_ref_t copy_gref; | |
255 | }; | |
256 | ||
e9ce7cb6 | 257 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue, |
b3f980bd | 258 | struct netrx_pending_operations *npo) |
f942dc25 | 259 | { |
b3f980bd | 260 | struct xenvif_rx_meta *meta; |
68a33bfd | 261 | struct xen_netif_rx_request req; |
f942dc25 | 262 | |
68a33bfd | 263 | RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req); |
f942dc25 IC |
264 | |
265 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 266 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
267 | meta->gso_size = 0; |
268 | meta->size = 0; | |
68a33bfd | 269 | meta->id = req.id; |
f942dc25 IC |
270 | |
271 | npo->copy_off = 0; | |
68a33bfd | 272 | npo->copy_gref = req.gref; |
f942dc25 IC |
273 | |
274 | return meta; | |
275 | } | |
276 | ||
d0089e8a JG |
277 | struct gop_frag_copy { |
278 | struct xenvif_queue *queue; | |
279 | struct netrx_pending_operations *npo; | |
280 | struct xenvif_rx_meta *meta; | |
281 | int head; | |
282 | int gso_type; | |
283 | ||
284 | struct page *page; | |
285 | }; | |
286 | ||
287 | static void xenvif_setup_copy_gop(unsigned long gfn, | |
288 | unsigned int offset, | |
289 | unsigned int *len, | |
290 | struct gop_frag_copy *info) | |
291 | { | |
292 | struct gnttab_copy *copy_gop; | |
293 | struct xen_page_foreign *foreign; | |
294 | /* Convenient aliases */ | |
295 | struct xenvif_queue *queue = info->queue; | |
296 | struct netrx_pending_operations *npo = info->npo; | |
297 | struct page *page = info->page; | |
298 | ||
299 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); | |
300 | ||
301 | if (npo->copy_off == MAX_BUFFER_OFFSET) | |
302 | info->meta = get_next_rx_buffer(queue, npo); | |
303 | ||
304 | if (npo->copy_off + *len > MAX_BUFFER_OFFSET) | |
305 | *len = MAX_BUFFER_OFFSET - npo->copy_off; | |
306 | ||
307 | copy_gop = npo->copy + npo->copy_prod++; | |
308 | copy_gop->flags = GNTCOPY_dest_gref; | |
309 | copy_gop->len = *len; | |
310 | ||
311 | foreign = xen_page_foreign(page); | |
312 | if (foreign) { | |
313 | copy_gop->source.domid = foreign->domid; | |
314 | copy_gop->source.u.ref = foreign->gref; | |
315 | copy_gop->flags |= GNTCOPY_source_gref; | |
316 | } else { | |
317 | copy_gop->source.domid = DOMID_SELF; | |
318 | copy_gop->source.u.gmfn = gfn; | |
319 | } | |
320 | copy_gop->source.offset = offset; | |
321 | ||
322 | copy_gop->dest.domid = queue->vif->domid; | |
323 | copy_gop->dest.offset = npo->copy_off; | |
324 | copy_gop->dest.u.ref = npo->copy_gref; | |
325 | ||
326 | npo->copy_off += *len; | |
327 | info->meta->size += *len; | |
328 | ||
329 | /* Leave a gap for the GSO descriptor. */ | |
330 | if (info->head && ((1 << info->gso_type) & queue->vif->gso_mask)) | |
331 | queue->rx.req_cons++; | |
332 | ||
333 | info->head = 0; /* There must be something in this buffer now */ | |
334 | } | |
335 | ||
336 | static void xenvif_gop_frag_copy_grant(unsigned long gfn, | |
337 | unsigned offset, | |
338 | unsigned int len, | |
339 | void *data) | |
340 | { | |
341 | unsigned int bytes; | |
342 | ||
343 | while (len) { | |
344 | bytes = len; | |
345 | xenvif_setup_copy_gop(gfn, offset, &bytes, data); | |
346 | offset += bytes; | |
347 | len -= bytes; | |
348 | } | |
349 | } | |
350 | ||
33bc801d WL |
351 | /* |
352 | * Set up the grant operations for this fragment. If it's a flipping | |
353 | * interface, we also set up the unmap request from here. | |
354 | */ | |
e9ce7cb6 | 355 | static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb, |
7376419a WL |
356 | struct netrx_pending_operations *npo, |
357 | struct page *page, unsigned long size, | |
c2677a6f | 358 | unsigned long offset, int *head) |
f942dc25 | 359 | { |
d0089e8a JG |
360 | struct gop_frag_copy info = { |
361 | .queue = queue, | |
362 | .npo = npo, | |
363 | .head = *head, | |
364 | .gso_type = XEN_NETIF_GSO_TYPE_NONE, | |
365 | }; | |
f942dc25 IC |
366 | unsigned long bytes; |
367 | ||
a0f2e80f JG |
368 | if (skb_is_gso(skb)) { |
369 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
d0089e8a | 370 | info.gso_type = XEN_NETIF_GSO_TYPE_TCPV4; |
a0f2e80f | 371 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
d0089e8a | 372 | info.gso_type = XEN_NETIF_GSO_TYPE_TCPV6; |
a0f2e80f JG |
373 | } |
374 | ||
f942dc25 | 375 | /* Data must not cross a page boundary. */ |
6a8ed462 | 376 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 | 377 | |
d0089e8a | 378 | info.meta = npo->meta + npo->meta_prod - 1; |
f942dc25 | 379 | |
6a8ed462 IC |
380 | /* Skip unused frames from start of page */ |
381 | page += offset >> PAGE_SHIFT; | |
382 | offset &= ~PAGE_MASK; | |
383 | ||
f942dc25 | 384 | while (size > 0) { |
6a8ed462 | 385 | BUG_ON(offset >= PAGE_SIZE); |
6a8ed462 | 386 | |
1650d545 | 387 | bytes = PAGE_SIZE - offset; |
6a8ed462 IC |
388 | if (bytes > size) |
389 | bytes = size; | |
390 | ||
d0089e8a JG |
391 | info.page = page; |
392 | gnttab_foreach_grant_in_range(page, offset, bytes, | |
393 | xenvif_gop_frag_copy_grant, | |
394 | &info); | |
f942dc25 | 395 | size -= bytes; |
d0089e8a | 396 | offset = 0; |
f942dc25 | 397 | |
d0089e8a JG |
398 | /* Next page */ |
399 | if (size) { | |
6a8ed462 IC |
400 | BUG_ON(!PageCompound(page)); |
401 | page++; | |
6a8ed462 | 402 | } |
f942dc25 | 403 | } |
d0089e8a JG |
404 | |
405 | *head = info.head; | |
f942dc25 IC |
406 | } |
407 | ||
408 | /* | |
409 | * Prepare an SKB to be transmitted to the frontend. | |
410 | * | |
411 | * This function is responsible for allocating grant operations, meta | |
412 | * structures, etc. | |
413 | * | |
414 | * It returns the number of meta structures consumed. The number of | |
415 | * ring slots used is always equal to the number of meta slots used | |
416 | * plus the number of GSO descriptors used. Currently, we use either | |
417 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
418 | * frontend-side LRO). | |
419 | */ | |
7376419a | 420 | static int xenvif_gop_skb(struct sk_buff *skb, |
e9ce7cb6 WL |
421 | struct netrx_pending_operations *npo, |
422 | struct xenvif_queue *queue) | |
f942dc25 IC |
423 | { |
424 | struct xenvif *vif = netdev_priv(skb->dev); | |
425 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
426 | int i; | |
68a33bfd | 427 | struct xen_netif_rx_request req; |
b3f980bd | 428 | struct xenvif_rx_meta *meta; |
f942dc25 | 429 | unsigned char *data; |
33bc801d | 430 | int head = 1; |
f942dc25 | 431 | int old_meta_prod; |
82cada22 | 432 | int gso_type; |
f942dc25 IC |
433 | |
434 | old_meta_prod = npo->meta_prod; | |
435 | ||
5bd07670 AL |
436 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
437 | if (skb_is_gso(skb)) { | |
438 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
439 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
440 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
441 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
442 | } |
443 | ||
f942dc25 | 444 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 445 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
68a33bfd | 446 | RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req); |
f942dc25 | 447 | meta = npo->meta + npo->meta_prod++; |
82cada22 | 448 | meta->gso_type = gso_type; |
5bd07670 | 449 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 | 450 | meta->size = 0; |
68a33bfd | 451 | meta->id = req.id; |
f942dc25 IC |
452 | } |
453 | ||
68a33bfd | 454 | RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req); |
f942dc25 IC |
455 | meta = npo->meta + npo->meta_prod++; |
456 | ||
82cada22 PD |
457 | if ((1 << gso_type) & vif->gso_mask) { |
458 | meta->gso_type = gso_type; | |
5bd07670 | 459 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
460 | } else { |
461 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 462 | meta->gso_size = 0; |
82cada22 | 463 | } |
f942dc25 IC |
464 | |
465 | meta->size = 0; | |
68a33bfd | 466 | meta->id = req.id; |
f942dc25 | 467 | npo->copy_off = 0; |
68a33bfd | 468 | npo->copy_gref = req.gref; |
f942dc25 IC |
469 | |
470 | data = skb->data; | |
471 | while (data < skb_tail_pointer(skb)) { | |
472 | unsigned int offset = offset_in_page(data); | |
473 | unsigned int len = PAGE_SIZE - offset; | |
474 | ||
475 | if (data + len > skb_tail_pointer(skb)) | |
476 | len = skb_tail_pointer(skb) - data; | |
477 | ||
e9ce7cb6 | 478 | xenvif_gop_frag_copy(queue, skb, npo, |
c2677a6f | 479 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
480 | data += len; |
481 | } | |
482 | ||
483 | for (i = 0; i < nr_frags; i++) { | |
e9ce7cb6 | 484 | xenvif_gop_frag_copy(queue, skb, npo, |
7376419a WL |
485 | skb_frag_page(&skb_shinfo(skb)->frags[i]), |
486 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
487 | skb_shinfo(skb)->frags[i].page_offset, | |
c2677a6f | 488 | &head); |
f942dc25 IC |
489 | } |
490 | ||
491 | return npo->meta_prod - old_meta_prod; | |
492 | } | |
493 | ||
494 | /* | |
7376419a | 495 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
496 | * used to set up the operations on the top of |
497 | * netrx_pending_operations, which have since been done. Check that | |
498 | * they didn't give any errors and advance over them. | |
499 | */ | |
7376419a WL |
500 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
501 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
502 | { |
503 | struct gnttab_copy *copy_op; | |
504 | int status = XEN_NETIF_RSP_OKAY; | |
505 | int i; | |
506 | ||
507 | for (i = 0; i < nr_meta_slots; i++) { | |
508 | copy_op = npo->copy + npo->copy_cons++; | |
509 | if (copy_op->status != GNTST_okay) { | |
510 | netdev_dbg(vif->dev, | |
511 | "Bad status %d from copy to DOM%d.\n", | |
512 | copy_op->status, vif->domid); | |
513 | status = XEN_NETIF_RSP_ERROR; | |
514 | } | |
515 | } | |
516 | ||
517 | return status; | |
518 | } | |
519 | ||
e9ce7cb6 | 520 | static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status, |
7376419a WL |
521 | struct xenvif_rx_meta *meta, |
522 | int nr_meta_slots) | |
f942dc25 IC |
523 | { |
524 | int i; | |
525 | unsigned long offset; | |
526 | ||
527 | /* No fragments used */ | |
528 | if (nr_meta_slots <= 1) | |
529 | return; | |
530 | ||
531 | nr_meta_slots--; | |
532 | ||
533 | for (i = 0; i < nr_meta_slots; i++) { | |
534 | int flags; | |
535 | if (i == nr_meta_slots - 1) | |
536 | flags = 0; | |
537 | else | |
538 | flags = XEN_NETRXF_more_data; | |
539 | ||
540 | offset = 0; | |
e9ce7cb6 | 541 | make_rx_response(queue, meta[i].id, status, offset, |
f942dc25 IC |
542 | meta[i].size, flags); |
543 | } | |
544 | } | |
545 | ||
e9ce7cb6 | 546 | void xenvif_kick_thread(struct xenvif_queue *queue) |
b3f980bd | 547 | { |
e9ce7cb6 | 548 | wake_up(&queue->wq); |
b3f980bd WL |
549 | } |
550 | ||
e9ce7cb6 | 551 | static void xenvif_rx_action(struct xenvif_queue *queue) |
f942dc25 | 552 | { |
f942dc25 | 553 | s8 status; |
e1f00a69 | 554 | u16 flags; |
f942dc25 IC |
555 | struct xen_netif_rx_response *resp; |
556 | struct sk_buff_head rxq; | |
557 | struct sk_buff *skb; | |
558 | LIST_HEAD(notify); | |
559 | int ret; | |
f942dc25 | 560 | unsigned long offset; |
11b57f90 | 561 | bool need_to_notify = false; |
f942dc25 IC |
562 | |
563 | struct netrx_pending_operations npo = { | |
e9ce7cb6 WL |
564 | .copy = queue->grant_copy_op, |
565 | .meta = queue->meta, | |
f942dc25 IC |
566 | }; |
567 | ||
568 | skb_queue_head_init(&rxq); | |
569 | ||
1d5d4852 | 570 | while (xenvif_rx_ring_slots_available(queue) |
f48da8b1 | 571 | && (skb = xenvif_rx_dequeue(queue)) != NULL) { |
ecf08d2d DV |
572 | queue->last_rx_time = jiffies; |
573 | ||
e9ce7cb6 | 574 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue); |
1425c7a4 | 575 | |
f942dc25 | 576 | __skb_queue_tail(&rxq, skb); |
f942dc25 IC |
577 | } |
578 | ||
e9ce7cb6 | 579 | BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta)); |
f942dc25 IC |
580 | |
581 | if (!npo.copy_prod) | |
ca2f09f2 | 582 | goto done; |
f942dc25 | 583 | |
ac3d5ac2 | 584 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
e9ce7cb6 | 585 | gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
586 | |
587 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 588 | |
e9ce7cb6 WL |
589 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
590 | queue->vif->gso_prefix_mask) { | |
591 | resp = RING_GET_RESPONSE(&queue->rx, | |
592 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
593 | |
594 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
595 | ||
e9ce7cb6 WL |
596 | resp->offset = queue->meta[npo.meta_cons].gso_size; |
597 | resp->id = queue->meta[npo.meta_cons].id; | |
8f13dd96 | 598 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
599 | |
600 | npo.meta_cons++; | |
8f13dd96 | 601 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
602 | } |
603 | ||
604 | ||
e9ce7cb6 WL |
605 | queue->stats.tx_bytes += skb->len; |
606 | queue->stats.tx_packets++; | |
f942dc25 | 607 | |
e9ce7cb6 | 608 | status = xenvif_check_gop(queue->vif, |
8f13dd96 ZK |
609 | XENVIF_RX_CB(skb)->meta_slots_used, |
610 | &npo); | |
f942dc25 | 611 | |
8f13dd96 | 612 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
613 | flags = 0; |
614 | else | |
615 | flags = XEN_NETRXF_more_data; | |
616 | ||
617 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
618 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
619 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
620 | /* remote but checksummed. */ | |
621 | flags |= XEN_NETRXF_data_validated; | |
622 | ||
623 | offset = 0; | |
e9ce7cb6 | 624 | resp = make_rx_response(queue, queue->meta[npo.meta_cons].id, |
f942dc25 | 625 | status, offset, |
e9ce7cb6 | 626 | queue->meta[npo.meta_cons].size, |
f942dc25 IC |
627 | flags); |
628 | ||
e9ce7cb6 WL |
629 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
630 | queue->vif->gso_mask) { | |
f942dc25 IC |
631 | struct xen_netif_extra_info *gso = |
632 | (struct xen_netif_extra_info *) | |
e9ce7cb6 WL |
633 | RING_GET_RESPONSE(&queue->rx, |
634 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
635 | |
636 | resp->flags |= XEN_NETRXF_extra_info; | |
637 | ||
e9ce7cb6 WL |
638 | gso->u.gso.type = queue->meta[npo.meta_cons].gso_type; |
639 | gso->u.gso.size = queue->meta[npo.meta_cons].gso_size; | |
f942dc25 IC |
640 | gso->u.gso.pad = 0; |
641 | gso->u.gso.features = 0; | |
642 | ||
643 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
644 | gso->flags = 0; | |
645 | } | |
646 | ||
e9ce7cb6 WL |
647 | xenvif_add_frag_responses(queue, status, |
648 | queue->meta + npo.meta_cons + 1, | |
8f13dd96 | 649 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 | 650 | |
e9ce7cb6 | 651 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret); |
f942dc25 | 652 | |
11b57f90 | 653 | need_to_notify |= !!ret; |
b3f980bd | 654 | |
8f13dd96 | 655 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
656 | dev_kfree_skb(skb); |
657 | } | |
658 | ||
ca2f09f2 | 659 | done: |
b3f980bd | 660 | if (need_to_notify) |
e9ce7cb6 | 661 | notify_remote_via_irq(queue->rx_irq); |
f942dc25 IC |
662 | } |
663 | ||
e9ce7cb6 | 664 | void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue) |
f942dc25 IC |
665 | { |
666 | int more_to_do; | |
667 | ||
e9ce7cb6 | 668 | RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do); |
f942dc25 IC |
669 | |
670 | if (more_to_do) | |
e9ce7cb6 | 671 | napi_schedule(&queue->napi); |
f942dc25 IC |
672 | } |
673 | ||
e9ce7cb6 | 674 | static void tx_add_credit(struct xenvif_queue *queue) |
f942dc25 IC |
675 | { |
676 | unsigned long max_burst, max_credit; | |
677 | ||
678 | /* | |
679 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
680 | * Otherwise the interface can seize up due to insufficient credit. | |
681 | */ | |
0f589967 | 682 | max_burst = max(131072UL, queue->credit_bytes); |
f942dc25 IC |
683 | |
684 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
e9ce7cb6 WL |
685 | max_credit = queue->remaining_credit + queue->credit_bytes; |
686 | if (max_credit < queue->remaining_credit) | |
f942dc25 IC |
687 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ |
688 | ||
e9ce7cb6 | 689 | queue->remaining_credit = min(max_credit, max_burst); |
f942dc25 IC |
690 | } |
691 | ||
edafc132 | 692 | void xenvif_tx_credit_callback(unsigned long data) |
f942dc25 | 693 | { |
e9ce7cb6 WL |
694 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
695 | tx_add_credit(queue); | |
696 | xenvif_napi_schedule_or_enable_events(queue); | |
f942dc25 IC |
697 | } |
698 | ||
e9ce7cb6 | 699 | static void xenvif_tx_err(struct xenvif_queue *queue, |
562abd39 PD |
700 | struct xen_netif_tx_request *txp, |
701 | unsigned int extra_count, RING_IDX end) | |
f942dc25 | 702 | { |
e9ce7cb6 | 703 | RING_IDX cons = queue->tx.req_cons; |
f53c3fe8 | 704 | unsigned long flags; |
f942dc25 IC |
705 | |
706 | do { | |
e9ce7cb6 | 707 | spin_lock_irqsave(&queue->response_lock, flags); |
562abd39 | 708 | make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR); |
c8a4d299 | 709 | push_tx_responses(queue); |
e9ce7cb6 | 710 | spin_unlock_irqrestore(&queue->response_lock, flags); |
b9149729 | 711 | if (cons == end) |
f942dc25 | 712 | break; |
68a33bfd | 713 | RING_COPY_REQUEST(&queue->tx, cons++, txp); |
72eec92a | 714 | extra_count = 0; /* only the first frag can have extras */ |
f942dc25 | 715 | } while (1); |
e9ce7cb6 | 716 | queue->tx.req_cons = cons; |
f942dc25 IC |
717 | } |
718 | ||
7376419a | 719 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
720 | { |
721 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 | 722 | vif->disabled = true; |
e9ce7cb6 WL |
723 | /* Disable the vif from queue 0's kthread */ |
724 | if (vif->queues) | |
725 | xenvif_kick_thread(&vif->queues[0]); | |
48856286 IC |
726 | } |
727 | ||
e9ce7cb6 | 728 | static int xenvif_count_requests(struct xenvif_queue *queue, |
7376419a | 729 | struct xen_netif_tx_request *first, |
562abd39 | 730 | unsigned int extra_count, |
7376419a WL |
731 | struct xen_netif_tx_request *txp, |
732 | int work_to_do) | |
f942dc25 | 733 | { |
e9ce7cb6 | 734 | RING_IDX cons = queue->tx.req_cons; |
2810e5b9 WL |
735 | int slots = 0; |
736 | int drop_err = 0; | |
59ccb4eb | 737 | int more_data; |
f942dc25 IC |
738 | |
739 | if (!(first->flags & XEN_NETTXF_more_data)) | |
740 | return 0; | |
741 | ||
742 | do { | |
59ccb4eb WL |
743 | struct xen_netif_tx_request dropped_tx = { 0 }; |
744 | ||
2810e5b9 | 745 | if (slots >= work_to_do) { |
e9ce7cb6 | 746 | netdev_err(queue->vif->dev, |
2810e5b9 WL |
747 | "Asked for %d slots but exceeds this limit\n", |
748 | work_to_do); | |
e9ce7cb6 | 749 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 750 | return -ENODATA; |
f942dc25 IC |
751 | } |
752 | ||
2810e5b9 WL |
753 | /* This guest is really using too many slots and |
754 | * considered malicious. | |
755 | */ | |
37641494 | 756 | if (unlikely(slots >= fatal_skb_slots)) { |
e9ce7cb6 | 757 | netdev_err(queue->vif->dev, |
2810e5b9 | 758 | "Malicious frontend using %d slots, threshold %u\n", |
37641494 | 759 | slots, fatal_skb_slots); |
e9ce7cb6 | 760 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 761 | return -E2BIG; |
f942dc25 IC |
762 | } |
763 | ||
2810e5b9 | 764 | /* Xen network protocol had implicit dependency on |
37641494 WL |
765 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
766 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
767 | * same behavior as before. Any packet using more than | |
768 | * 18 slots but less than fatal_skb_slots slots is | |
769 | * dropped | |
2810e5b9 | 770 | */ |
37641494 | 771 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 | 772 | if (net_ratelimit()) |
e9ce7cb6 | 773 | netdev_dbg(queue->vif->dev, |
2810e5b9 | 774 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", |
37641494 | 775 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
776 | drop_err = -E2BIG; |
777 | } | |
778 | ||
59ccb4eb WL |
779 | if (drop_err) |
780 | txp = &dropped_tx; | |
781 | ||
68a33bfd | 782 | RING_COPY_REQUEST(&queue->tx, cons + slots, txp); |
03393fd5 WL |
783 | |
784 | /* If the guest submitted a frame >= 64 KiB then | |
785 | * first->size overflowed and following slots will | |
786 | * appear to be larger than the frame. | |
787 | * | |
788 | * This cannot be fatal error as there are buggy | |
789 | * frontends that do this. | |
790 | * | |
791 | * Consume all slots and drop the packet. | |
792 | */ | |
793 | if (!drop_err && txp->size > first->size) { | |
794 | if (net_ratelimit()) | |
e9ce7cb6 | 795 | netdev_dbg(queue->vif->dev, |
03393fd5 WL |
796 | "Invalid tx request, slot size %u > remaining size %u\n", |
797 | txp->size, first->size); | |
798 | drop_err = -EIO; | |
f942dc25 IC |
799 | } |
800 | ||
801 | first->size -= txp->size; | |
2810e5b9 | 802 | slots++; |
f942dc25 | 803 | |
d0089e8a | 804 | if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) { |
68946159 | 805 | netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n", |
f942dc25 | 806 | txp->offset, txp->size); |
e9ce7cb6 | 807 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 808 | return -EINVAL; |
f942dc25 | 809 | } |
59ccb4eb WL |
810 | |
811 | more_data = txp->flags & XEN_NETTXF_more_data; | |
812 | ||
813 | if (!drop_err) | |
814 | txp++; | |
815 | ||
816 | } while (more_data); | |
2810e5b9 WL |
817 | |
818 | if (drop_err) { | |
562abd39 | 819 | xenvif_tx_err(queue, first, extra_count, cons + slots); |
2810e5b9 WL |
820 | return drop_err; |
821 | } | |
822 | ||
823 | return slots; | |
f942dc25 IC |
824 | } |
825 | ||
8f13dd96 ZK |
826 | |
827 | struct xenvif_tx_cb { | |
828 | u16 pending_idx; | |
829 | }; | |
830 | ||
831 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
832 | ||
e9ce7cb6 | 833 | static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue, |
562abd39 PD |
834 | u16 pending_idx, |
835 | struct xen_netif_tx_request *txp, | |
836 | unsigned int extra_count, | |
837 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 838 | { |
e9ce7cb6 WL |
839 | queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx]; |
840 | gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx), | |
f53c3fe8 | 841 | GNTMAP_host_map | GNTMAP_readonly, |
e9ce7cb6 | 842 | txp->gref, queue->vif->domid); |
f53c3fe8 | 843 | |
e9ce7cb6 | 844 | memcpy(&queue->pending_tx_info[pending_idx].req, txp, |
f53c3fe8 | 845 | sizeof(*txp)); |
562abd39 | 846 | queue->pending_tx_info[pending_idx].extra_count = extra_count; |
f53c3fe8 ZK |
847 | } |
848 | ||
e3377f36 ZK |
849 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
850 | { | |
851 | struct sk_buff *skb = | |
852 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
853 | GFP_ATOMIC | __GFP_NOWARN); | |
854 | if (unlikely(skb == NULL)) | |
855 | return NULL; | |
856 | ||
857 | /* Packets passed to netif_rx() must have some headroom. */ | |
858 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
859 | ||
860 | /* Initialize it here to avoid later surprises */ | |
861 | skb_shinfo(skb)->destructor_arg = NULL; | |
862 | ||
863 | return skb; | |
864 | } | |
865 | ||
e9ce7cb6 | 866 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue, |
f53c3fe8 ZK |
867 | struct sk_buff *skb, |
868 | struct xen_netif_tx_request *txp, | |
2475b225 RL |
869 | struct gnttab_map_grant_ref *gop, |
870 | unsigned int frag_overflow, | |
871 | struct sk_buff *nskb) | |
f942dc25 IC |
872 | { |
873 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
874 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 875 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
876 | int start; |
877 | pending_ring_idx_t index; | |
2475b225 | 878 | unsigned int nr_slots; |
2810e5b9 | 879 | |
2810e5b9 | 880 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
881 | |
882 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 883 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 884 | |
f53c3fe8 ZK |
885 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
886 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
887 | index = pending_index(queue->pending_cons++); |
888 | pending_idx = queue->pending_ring[index]; | |
562abd39 | 889 | xenvif_tx_create_map_op(queue, pending_idx, txp, 0, gop); |
f53c3fe8 | 890 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
891 | } |
892 | ||
e3377f36 | 893 | if (frag_overflow) { |
e3377f36 ZK |
894 | |
895 | shinfo = skb_shinfo(nskb); | |
896 | frags = shinfo->frags; | |
897 | ||
898 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
899 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
900 | index = pending_index(queue->pending_cons++); |
901 | pending_idx = queue->pending_ring[index]; | |
562abd39 PD |
902 | xenvif_tx_create_map_op(queue, pending_idx, txp, 0, |
903 | gop); | |
e3377f36 ZK |
904 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
905 | pending_idx); | |
906 | } | |
907 | ||
908 | skb_shinfo(skb)->frag_list = nskb; | |
909 | } | |
2810e5b9 | 910 | |
f942dc25 IC |
911 | return gop; |
912 | } | |
913 | ||
e9ce7cb6 | 914 | static inline void xenvif_grant_handle_set(struct xenvif_queue *queue, |
f53c3fe8 ZK |
915 | u16 pending_idx, |
916 | grant_handle_t handle) | |
917 | { | |
e9ce7cb6 | 918 | if (unlikely(queue->grant_tx_handle[pending_idx] != |
f53c3fe8 | 919 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 920 | netdev_err(queue->vif->dev, |
68946159 | 921 | "Trying to overwrite active handle! pending_idx: 0x%x\n", |
f53c3fe8 ZK |
922 | pending_idx); |
923 | BUG(); | |
924 | } | |
e9ce7cb6 | 925 | queue->grant_tx_handle[pending_idx] = handle; |
f53c3fe8 ZK |
926 | } |
927 | ||
e9ce7cb6 | 928 | static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue, |
f53c3fe8 ZK |
929 | u16 pending_idx) |
930 | { | |
e9ce7cb6 | 931 | if (unlikely(queue->grant_tx_handle[pending_idx] == |
f53c3fe8 | 932 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 933 | netdev_err(queue->vif->dev, |
68946159 | 934 | "Trying to unmap invalid handle! pending_idx: 0x%x\n", |
f53c3fe8 ZK |
935 | pending_idx); |
936 | BUG(); | |
937 | } | |
e9ce7cb6 | 938 | queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; |
f53c3fe8 ZK |
939 | } |
940 | ||
e9ce7cb6 | 941 | static int xenvif_tx_check_gop(struct xenvif_queue *queue, |
7376419a | 942 | struct sk_buff *skb, |
bdab8275 ZK |
943 | struct gnttab_map_grant_ref **gopp_map, |
944 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 945 | { |
9074ce24 | 946 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 947 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
1a998d3e ZK |
948 | /* This always points to the shinfo of the skb being checked, which |
949 | * could be either the first or the one on the frag_list | |
950 | */ | |
f942dc25 | 951 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1a998d3e ZK |
952 | /* If this is non-NULL, we are currently checking the frag_list skb, and |
953 | * this points to the shinfo of the first one | |
954 | */ | |
955 | struct skb_shared_info *first_shinfo = NULL; | |
f942dc25 | 956 | int nr_frags = shinfo->nr_frags; |
1b860da0 ZK |
957 | const bool sharedslot = nr_frags && |
958 | frag_get_pending_idx(&shinfo->frags[0]) == pending_idx; | |
bdab8275 | 959 | int i, err; |
f942dc25 IC |
960 | |
961 | /* Check status of header. */ | |
bdab8275 | 962 | err = (*gopp_copy)->status; |
bdab8275 ZK |
963 | if (unlikely(err)) { |
964 | if (net_ratelimit()) | |
e9ce7cb6 | 965 | netdev_dbg(queue->vif->dev, |
00aefceb | 966 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
967 | (*gopp_copy)->status, |
968 | pending_idx, | |
969 | (*gopp_copy)->source.u.ref); | |
1b860da0 ZK |
970 | /* The first frag might still have this slot mapped */ |
971 | if (!sharedslot) | |
972 | xenvif_idx_release(queue, pending_idx, | |
973 | XEN_NETIF_RSP_ERROR); | |
bdab8275 | 974 | } |
d8cfbfc4 | 975 | (*gopp_copy)++; |
f942dc25 | 976 | |
e3377f36 | 977 | check_frags: |
bdab8275 | 978 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 979 | int j, newerr; |
f942dc25 | 980 | |
ea066ad1 | 981 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
982 | |
983 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 984 | newerr = gop_map->status; |
2810e5b9 | 985 | |
f942dc25 | 986 | if (likely(!newerr)) { |
e9ce7cb6 | 987 | xenvif_grant_handle_set(queue, |
9074ce24 ZK |
988 | pending_idx, |
989 | gop_map->handle); | |
f942dc25 | 990 | /* Had a previous error? Invalidate this fragment. */ |
1b860da0 | 991 | if (unlikely(err)) { |
e9ce7cb6 | 992 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
993 | /* If the mapping of the first frag was OK, but |
994 | * the header's copy failed, and they are | |
995 | * sharing a slot, send an error | |
996 | */ | |
997 | if (i == 0 && sharedslot) | |
998 | xenvif_idx_release(queue, pending_idx, | |
999 | XEN_NETIF_RSP_ERROR); | |
1000 | else | |
1001 | xenvif_idx_release(queue, pending_idx, | |
1002 | XEN_NETIF_RSP_OKAY); | |
1003 | } | |
f942dc25 IC |
1004 | continue; |
1005 | } | |
1006 | ||
1007 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 | 1008 | if (net_ratelimit()) |
e9ce7cb6 | 1009 | netdev_dbg(queue->vif->dev, |
00aefceb | 1010 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1011 | i, |
1012 | gop_map->status, | |
1013 | pending_idx, | |
1014 | gop_map->ref); | |
1b860da0 | 1015 | |
e9ce7cb6 | 1016 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1017 | |
1018 | /* Not the first error? Preceding frags already invalidated. */ | |
1019 | if (err) | |
1020 | continue; | |
1b860da0 ZK |
1021 | |
1022 | /* First error: if the header haven't shared a slot with the | |
1023 | * first frag, release it as well. | |
1024 | */ | |
1025 | if (!sharedslot) | |
1026 | xenvif_idx_release(queue, | |
1027 | XENVIF_TX_CB(skb)->pending_idx, | |
1028 | XEN_NETIF_RSP_OKAY); | |
1029 | ||
1030 | /* Invalidate preceding fragments of this skb. */ | |
bdab8275 | 1031 | for (j = 0; j < i; j++) { |
5ccb3ea7 | 1032 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 1033 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
1034 | xenvif_idx_release(queue, pending_idx, |
1035 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1036 | } |
1037 | ||
1a998d3e ZK |
1038 | /* And if we found the error while checking the frag_list, unmap |
1039 | * the first skb's frags | |
1040 | */ | |
1041 | if (first_shinfo) { | |
1042 | for (j = 0; j < first_shinfo->nr_frags; j++) { | |
1043 | pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]); | |
1044 | xenvif_idx_unmap(queue, pending_idx); | |
1b860da0 ZK |
1045 | xenvif_idx_release(queue, pending_idx, |
1046 | XEN_NETIF_RSP_OKAY); | |
1a998d3e | 1047 | } |
f942dc25 IC |
1048 | } |
1049 | ||
1050 | /* Remember the error: invalidate all subsequent fragments. */ | |
1051 | err = newerr; | |
1052 | } | |
1053 | ||
1a998d3e ZK |
1054 | if (skb_has_frag_list(skb) && !first_shinfo) { |
1055 | first_shinfo = skb_shinfo(skb); | |
1056 | shinfo = skb_shinfo(skb_shinfo(skb)->frag_list); | |
e3377f36 | 1057 | nr_frags = shinfo->nr_frags; |
e3377f36 ZK |
1058 | |
1059 | goto check_frags; | |
1060 | } | |
1061 | ||
bdab8275 | 1062 | *gopp_map = gop_map; |
f942dc25 IC |
1063 | return err; |
1064 | } | |
1065 | ||
e9ce7cb6 | 1066 | static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb) |
f942dc25 IC |
1067 | { |
1068 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1069 | int nr_frags = shinfo->nr_frags; | |
1070 | int i; | |
f53c3fe8 ZK |
1071 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1072 | ||
f942dc25 IC |
1073 | for (i = 0; i < nr_frags; i++) { |
1074 | skb_frag_t *frag = shinfo->frags + i; | |
1075 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1076 | struct page *page; |
1077 | u16 pending_idx; | |
f942dc25 | 1078 | |
ea066ad1 | 1079 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1080 | |
f53c3fe8 | 1081 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 1082 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 1083 | skb_shinfo(skb)->destructor_arg = |
e9ce7cb6 | 1084 | &callback_param(queue, pending_idx); |
bdab8275 | 1085 | else |
e9ce7cb6 WL |
1086 | callback_param(queue, prev_pending_idx).ctx = |
1087 | &callback_param(queue, pending_idx); | |
f53c3fe8 | 1088 | |
e9ce7cb6 | 1089 | callback_param(queue, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1090 | prev_pending_idx = pending_idx; |
1091 | ||
e9ce7cb6 WL |
1092 | txp = &queue->pending_tx_info[pending_idx].req; |
1093 | page = virt_to_page(idx_to_kaddr(queue, pending_idx)); | |
ea066ad1 | 1094 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1095 | skb->len += txp->size; |
1096 | skb->data_len += txp->size; | |
1097 | skb->truesize += txp->size; | |
1098 | ||
f53c3fe8 | 1099 | /* Take an extra reference to offset network stack's put_page */ |
e9ce7cb6 | 1100 | get_page(queue->mmap_pages[pending_idx]); |
f942dc25 IC |
1101 | } |
1102 | } | |
1103 | ||
e9ce7cb6 | 1104 | static int xenvif_get_extras(struct xenvif_queue *queue, |
562abd39 PD |
1105 | struct xen_netif_extra_info *extras, |
1106 | unsigned int *extra_count, | |
1107 | int work_to_do) | |
f942dc25 IC |
1108 | { |
1109 | struct xen_netif_extra_info extra; | |
e9ce7cb6 | 1110 | RING_IDX cons = queue->tx.req_cons; |
f942dc25 IC |
1111 | |
1112 | do { | |
1113 | if (unlikely(work_to_do-- <= 0)) { | |
e9ce7cb6 WL |
1114 | netdev_err(queue->vif->dev, "Missing extra info\n"); |
1115 | xenvif_fatal_tx_err(queue->vif); | |
f942dc25 IC |
1116 | return -EBADR; |
1117 | } | |
1118 | ||
68a33bfd | 1119 | RING_COPY_REQUEST(&queue->tx, cons, &extra); |
562abd39 PD |
1120 | |
1121 | queue->tx.req_cons = ++cons; | |
1122 | (*extra_count)++; | |
1123 | ||
f942dc25 IC |
1124 | if (unlikely(!extra.type || |
1125 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
e9ce7cb6 | 1126 | netdev_err(queue->vif->dev, |
f942dc25 | 1127 | "Invalid extra type: %d\n", extra.type); |
e9ce7cb6 | 1128 | xenvif_fatal_tx_err(queue->vif); |
f942dc25 IC |
1129 | return -EINVAL; |
1130 | } | |
1131 | ||
1132 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
f942dc25 IC |
1133 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
1134 | ||
1135 | return work_to_do; | |
1136 | } | |
1137 | ||
7376419a WL |
1138 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1139 | struct sk_buff *skb, | |
1140 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1141 | { |
1142 | if (!gso->u.gso.size) { | |
48856286 | 1143 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1144 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1145 | return -EINVAL; |
1146 | } | |
1147 | ||
a9468587 PD |
1148 | switch (gso->u.gso.type) { |
1149 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1150 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1151 | break; | |
1152 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1153 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1154 | break; | |
1155 | default: | |
48856286 | 1156 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1157 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1158 | return -EINVAL; |
1159 | } | |
1160 | ||
1161 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1162 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1163 | |
1164 | return 0; | |
1165 | } | |
1166 | ||
e9ce7cb6 | 1167 | static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb) |
2eba61d5 | 1168 | { |
2721637c | 1169 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1170 | |
1171 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1172 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1173 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1174 | * recalculate the partial checksum. | |
1175 | */ | |
1176 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
e9ce7cb6 | 1177 | queue->stats.rx_gso_checksum_fixup++; |
2eba61d5 | 1178 | skb->ip_summed = CHECKSUM_PARTIAL; |
2721637c | 1179 | recalculate_partial_csum = true; |
2eba61d5 PD |
1180 | } |
1181 | ||
1182 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1183 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1184 | return 0; | |
1185 | ||
2721637c | 1186 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1187 | } |
1188 | ||
e9ce7cb6 | 1189 | static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size) |
f942dc25 | 1190 | { |
059dfa6a | 1191 | u64 now = get_jiffies_64(); |
e9ce7cb6 WL |
1192 | u64 next_credit = queue->credit_window_start + |
1193 | msecs_to_jiffies(queue->credit_usec / 1000); | |
f942dc25 IC |
1194 | |
1195 | /* Timer could already be pending in rare cases. */ | |
e9ce7cb6 | 1196 | if (timer_pending(&queue->credit_timeout)) |
f942dc25 IC |
1197 | return true; |
1198 | ||
1199 | /* Passed the point where we can replenish credit? */ | |
059dfa6a | 1200 | if (time_after_eq64(now, next_credit)) { |
e9ce7cb6 WL |
1201 | queue->credit_window_start = now; |
1202 | tx_add_credit(queue); | |
f942dc25 IC |
1203 | } |
1204 | ||
1205 | /* Still too big to send right now? Set a callback. */ | |
e9ce7cb6 WL |
1206 | if (size > queue->remaining_credit) { |
1207 | queue->credit_timeout.data = | |
1208 | (unsigned long)queue; | |
e9ce7cb6 | 1209 | mod_timer(&queue->credit_timeout, |
f942dc25 | 1210 | next_credit); |
e9ce7cb6 | 1211 | queue->credit_window_start = next_credit; |
f942dc25 IC |
1212 | |
1213 | return true; | |
1214 | } | |
1215 | ||
1216 | return false; | |
1217 | } | |
1218 | ||
210c34dc PD |
1219 | /* No locking is required in xenvif_mcast_add/del() as they are |
1220 | * only ever invoked from NAPI poll. An RCU list is used because | |
1221 | * xenvif_mcast_match() is called asynchronously, during start_xmit. | |
1222 | */ | |
1223 | ||
1224 | static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr) | |
1225 | { | |
1226 | struct xenvif_mcast_addr *mcast; | |
1227 | ||
1228 | if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) { | |
1229 | if (net_ratelimit()) | |
1230 | netdev_err(vif->dev, | |
1231 | "Too many multicast addresses\n"); | |
1232 | return -ENOSPC; | |
1233 | } | |
1234 | ||
1235 | mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC); | |
1236 | if (!mcast) | |
1237 | return -ENOMEM; | |
1238 | ||
1239 | ether_addr_copy(mcast->addr, addr); | |
1240 | list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr); | |
1241 | vif->fe_mcast_count++; | |
1242 | ||
1243 | return 0; | |
1244 | } | |
1245 | ||
1246 | static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr) | |
1247 | { | |
1248 | struct xenvif_mcast_addr *mcast; | |
1249 | ||
1250 | list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) { | |
1251 | if (ether_addr_equal(addr, mcast->addr)) { | |
1252 | --vif->fe_mcast_count; | |
1253 | list_del_rcu(&mcast->entry); | |
1254 | kfree_rcu(mcast, rcu); | |
1255 | break; | |
1256 | } | |
1257 | } | |
1258 | } | |
1259 | ||
1260 | bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr) | |
1261 | { | |
1262 | struct xenvif_mcast_addr *mcast; | |
1263 | ||
1264 | rcu_read_lock(); | |
1265 | list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) { | |
1266 | if (ether_addr_equal(addr, mcast->addr)) { | |
1267 | rcu_read_unlock(); | |
1268 | return true; | |
1269 | } | |
1270 | } | |
1271 | rcu_read_unlock(); | |
1272 | ||
1273 | return false; | |
1274 | } | |
1275 | ||
1276 | void xenvif_mcast_addr_list_free(struct xenvif *vif) | |
1277 | { | |
1278 | /* No need for locking or RCU here. NAPI poll and TX queue | |
1279 | * are stopped. | |
1280 | */ | |
1281 | while (!list_empty(&vif->fe_mcast_addr)) { | |
1282 | struct xenvif_mcast_addr *mcast; | |
1283 | ||
1284 | mcast = list_first_entry(&vif->fe_mcast_addr, | |
1285 | struct xenvif_mcast_addr, | |
1286 | entry); | |
1287 | --vif->fe_mcast_count; | |
1288 | list_del(&mcast->entry); | |
1289 | kfree(mcast); | |
1290 | } | |
1291 | } | |
1292 | ||
e9ce7cb6 | 1293 | static void xenvif_tx_build_gops(struct xenvif_queue *queue, |
bdab8275 ZK |
1294 | int budget, |
1295 | unsigned *copy_ops, | |
1296 | unsigned *map_ops) | |
f942dc25 | 1297 | { |
2475b225 RL |
1298 | struct gnttab_map_grant_ref *gop = queue->tx_map_ops; |
1299 | struct sk_buff *skb, *nskb; | |
f942dc25 | 1300 | int ret; |
2475b225 | 1301 | unsigned int frag_overflow; |
f942dc25 | 1302 | |
e9ce7cb6 | 1303 | while (skb_queue_len(&queue->tx_queue) < budget) { |
f942dc25 | 1304 | struct xen_netif_tx_request txreq; |
37641494 | 1305 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 | 1306 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
562abd39 | 1307 | unsigned int extra_count; |
f942dc25 IC |
1308 | u16 pending_idx; |
1309 | RING_IDX idx; | |
1310 | int work_to_do; | |
1311 | unsigned int data_len; | |
1312 | pending_ring_idx_t index; | |
1313 | ||
e9ce7cb6 | 1314 | if (queue->tx.sring->req_prod - queue->tx.req_cons > |
48856286 | 1315 | XEN_NETIF_TX_RING_SIZE) { |
e9ce7cb6 | 1316 | netdev_err(queue->vif->dev, |
48856286 IC |
1317 | "Impossible number of requests. " |
1318 | "req_prod %d, req_cons %d, size %ld\n", | |
e9ce7cb6 | 1319 | queue->tx.sring->req_prod, queue->tx.req_cons, |
48856286 | 1320 | XEN_NETIF_TX_RING_SIZE); |
e9ce7cb6 | 1321 | xenvif_fatal_tx_err(queue->vif); |
e9d8b2c2 | 1322 | break; |
48856286 IC |
1323 | } |
1324 | ||
e9ce7cb6 | 1325 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx); |
b3f980bd WL |
1326 | if (!work_to_do) |
1327 | break; | |
f942dc25 | 1328 | |
e9ce7cb6 | 1329 | idx = queue->tx.req_cons; |
f942dc25 | 1330 | rmb(); /* Ensure that we see the request before we copy it. */ |
68a33bfd | 1331 | RING_COPY_REQUEST(&queue->tx, idx, &txreq); |
f942dc25 IC |
1332 | |
1333 | /* Credit-based scheduling. */ | |
e9ce7cb6 WL |
1334 | if (txreq.size > queue->remaining_credit && |
1335 | tx_credit_exceeded(queue, txreq.size)) | |
b3f980bd | 1336 | break; |
f942dc25 | 1337 | |
e9ce7cb6 | 1338 | queue->remaining_credit -= txreq.size; |
f942dc25 IC |
1339 | |
1340 | work_to_do--; | |
e9ce7cb6 | 1341 | queue->tx.req_cons = ++idx; |
f942dc25 IC |
1342 | |
1343 | memset(extras, 0, sizeof(extras)); | |
562abd39 | 1344 | extra_count = 0; |
f942dc25 | 1345 | if (txreq.flags & XEN_NETTXF_extra_info) { |
e9ce7cb6 | 1346 | work_to_do = xenvif_get_extras(queue, extras, |
562abd39 | 1347 | &extra_count, |
7376419a | 1348 | work_to_do); |
e9ce7cb6 | 1349 | idx = queue->tx.req_cons; |
48856286 | 1350 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1351 | break; |
f942dc25 IC |
1352 | } |
1353 | ||
210c34dc PD |
1354 | if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) { |
1355 | struct xen_netif_extra_info *extra; | |
1356 | ||
1357 | extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1]; | |
1358 | ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr); | |
1359 | ||
562abd39 | 1360 | make_tx_response(queue, &txreq, extra_count, |
210c34dc PD |
1361 | (ret == 0) ? |
1362 | XEN_NETIF_RSP_OKAY : | |
1363 | XEN_NETIF_RSP_ERROR); | |
1364 | push_tx_responses(queue); | |
1365 | continue; | |
1366 | } | |
1367 | ||
1368 | if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) { | |
1369 | struct xen_netif_extra_info *extra; | |
1370 | ||
1371 | extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1]; | |
1372 | xenvif_mcast_del(queue->vif, extra->u.mcast.addr); | |
1373 | ||
562abd39 PD |
1374 | make_tx_response(queue, &txreq, extra_count, |
1375 | XEN_NETIF_RSP_OKAY); | |
210c34dc PD |
1376 | push_tx_responses(queue); |
1377 | continue; | |
1378 | } | |
1379 | ||
562abd39 PD |
1380 | ret = xenvif_count_requests(queue, &txreq, extra_count, |
1381 | txfrags, work_to_do); | |
48856286 | 1382 | if (unlikely(ret < 0)) |
b3f980bd | 1383 | break; |
48856286 | 1384 | |
f942dc25 IC |
1385 | idx += ret; |
1386 | ||
1387 | if (unlikely(txreq.size < ETH_HLEN)) { | |
e9ce7cb6 | 1388 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1389 | "Bad packet size: %d\n", txreq.size); |
562abd39 | 1390 | xenvif_tx_err(queue, &txreq, extra_count, idx); |
b3f980bd | 1391 | break; |
f942dc25 IC |
1392 | } |
1393 | ||
1394 | /* No crossing a page as the payload mustn't fragment. */ | |
d0089e8a | 1395 | if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) { |
e9ce7cb6 | 1396 | netdev_err(queue->vif->dev, |
68946159 | 1397 | "txreq.offset: %u, size: %u, end: %lu\n", |
f942dc25 | 1398 | txreq.offset, txreq.size, |
d0089e8a | 1399 | (unsigned long)(txreq.offset&~XEN_PAGE_MASK) + txreq.size); |
e9ce7cb6 | 1400 | xenvif_fatal_tx_err(queue->vif); |
b3f980bd | 1401 | break; |
f942dc25 IC |
1402 | } |
1403 | ||
e9ce7cb6 WL |
1404 | index = pending_index(queue->pending_cons); |
1405 | pending_idx = queue->pending_ring[index]; | |
f942dc25 | 1406 | |
7e5d7753 | 1407 | data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN && |
37641494 | 1408 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
7e5d7753 | 1409 | XEN_NETBACK_TX_COPY_LEN : txreq.size; |
f942dc25 | 1410 | |
e3377f36 | 1411 | skb = xenvif_alloc_skb(data_len); |
f942dc25 | 1412 | if (unlikely(skb == NULL)) { |
e9ce7cb6 | 1413 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1414 | "Can't allocate a skb in start_xmit.\n"); |
562abd39 | 1415 | xenvif_tx_err(queue, &txreq, extra_count, idx); |
f942dc25 IC |
1416 | break; |
1417 | } | |
1418 | ||
2475b225 RL |
1419 | skb_shinfo(skb)->nr_frags = ret; |
1420 | if (data_len < txreq.size) | |
1421 | skb_shinfo(skb)->nr_frags++; | |
1422 | /* At this point shinfo->nr_frags is in fact the number of | |
1423 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. | |
1424 | */ | |
1425 | frag_overflow = 0; | |
1426 | nskb = NULL; | |
1427 | if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) { | |
1428 | frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS; | |
1429 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
1430 | skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS; | |
1431 | nskb = xenvif_alloc_skb(0); | |
1432 | if (unlikely(nskb == NULL)) { | |
1433 | kfree_skb(skb); | |
562abd39 | 1434 | xenvif_tx_err(queue, &txreq, extra_count, idx); |
2475b225 RL |
1435 | if (net_ratelimit()) |
1436 | netdev_err(queue->vif->dev, | |
1437 | "Can't allocate the frag_list skb.\n"); | |
1438 | break; | |
1439 | } | |
1440 | } | |
1441 | ||
f942dc25 IC |
1442 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1443 | struct xen_netif_extra_info *gso; | |
1444 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1445 | ||
e9ce7cb6 | 1446 | if (xenvif_set_skb_gso(queue->vif, skb, gso)) { |
7376419a | 1447 | /* Failure in xenvif_set_skb_gso is fatal. */ |
f942dc25 | 1448 | kfree_skb(skb); |
2475b225 | 1449 | kfree_skb(nskb); |
b3f980bd | 1450 | break; |
f942dc25 IC |
1451 | } |
1452 | } | |
1453 | ||
8f13dd96 | 1454 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1455 | |
1456 | __skb_put(skb, data_len); | |
e9ce7cb6 WL |
1457 | queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
1458 | queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid; | |
1459 | queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
bdab8275 | 1460 | |
e9ce7cb6 | 1461 | queue->tx_copy_ops[*copy_ops].dest.u.gmfn = |
0df4f266 | 1462 | virt_to_gfn(skb->data); |
e9ce7cb6 WL |
1463 | queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; |
1464 | queue->tx_copy_ops[*copy_ops].dest.offset = | |
d0089e8a | 1465 | offset_in_page(skb->data) & ~XEN_PAGE_MASK; |
bdab8275 | 1466 | |
e9ce7cb6 WL |
1467 | queue->tx_copy_ops[*copy_ops].len = data_len; |
1468 | queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
bdab8275 ZK |
1469 | |
1470 | (*copy_ops)++; | |
f942dc25 | 1471 | |
f942dc25 | 1472 | if (data_len < txreq.size) { |
ea066ad1 IC |
1473 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1474 | pending_idx); | |
562abd39 PD |
1475 | xenvif_tx_create_map_op(queue, pending_idx, &txreq, |
1476 | extra_count, gop); | |
bdab8275 | 1477 | gop++; |
f942dc25 | 1478 | } else { |
ea066ad1 IC |
1479 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1480 | INVALID_PENDING_IDX); | |
562abd39 PD |
1481 | memcpy(&queue->pending_tx_info[pending_idx].req, |
1482 | &txreq, sizeof(txreq)); | |
1483 | queue->pending_tx_info[pending_idx].extra_count = | |
1484 | extra_count; | |
f942dc25 IC |
1485 | } |
1486 | ||
e9ce7cb6 | 1487 | queue->pending_cons++; |
f942dc25 | 1488 | |
2475b225 RL |
1489 | gop = xenvif_get_requests(queue, skb, txfrags, gop, |
1490 | frag_overflow, nskb); | |
f942dc25 | 1491 | |
e9ce7cb6 | 1492 | __skb_queue_tail(&queue->tx_queue, skb); |
1e0b6eac | 1493 | |
e9ce7cb6 | 1494 | queue->tx.req_cons = idx; |
f942dc25 | 1495 | |
e9ce7cb6 WL |
1496 | if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) || |
1497 | (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops))) | |
f942dc25 IC |
1498 | break; |
1499 | } | |
1500 | ||
e9ce7cb6 | 1501 | (*map_ops) = gop - queue->tx_map_ops; |
bdab8275 | 1502 | return; |
f942dc25 IC |
1503 | } |
1504 | ||
e3377f36 ZK |
1505 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1506 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1507 | */ | |
e9ce7cb6 | 1508 | static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb) |
e3377f36 ZK |
1509 | { |
1510 | unsigned int offset = skb_headlen(skb); | |
1511 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
49d9991a | 1512 | int i, f; |
e3377f36 ZK |
1513 | struct ubuf_info *uarg; |
1514 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1515 | ||
e9ce7cb6 WL |
1516 | queue->stats.tx_zerocopy_sent += 2; |
1517 | queue->stats.tx_frag_overflow++; | |
e3377f36 | 1518 | |
e9ce7cb6 | 1519 | xenvif_fill_frags(queue, nskb); |
e3377f36 ZK |
1520 | /* Subtract frags size, we will correct it later */ |
1521 | skb->truesize -= skb->data_len; | |
1522 | skb->len += nskb->len; | |
1523 | skb->data_len += nskb->len; | |
1524 | ||
1525 | /* create a brand new frags array and coalesce there */ | |
1526 | for (i = 0; offset < skb->len; i++) { | |
1527 | struct page *page; | |
1528 | unsigned int len; | |
1529 | ||
1530 | BUG_ON(i >= MAX_SKB_FRAGS); | |
44cc8ed1 | 1531 | page = alloc_page(GFP_ATOMIC); |
e3377f36 ZK |
1532 | if (!page) { |
1533 | int j; | |
1534 | skb->truesize += skb->data_len; | |
1535 | for (j = 0; j < i; j++) | |
1536 | put_page(frags[j].page.p); | |
1537 | return -ENOMEM; | |
1538 | } | |
1539 | ||
1540 | if (offset + PAGE_SIZE < skb->len) | |
1541 | len = PAGE_SIZE; | |
1542 | else | |
1543 | len = skb->len - offset; | |
1544 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1545 | BUG(); | |
1546 | ||
1547 | offset += len; | |
1548 | frags[i].page.p = page; | |
1549 | frags[i].page_offset = 0; | |
1550 | skb_frag_size_set(&frags[i], len); | |
1551 | } | |
49d9991a | 1552 | |
b0c21bad DV |
1553 | /* Copied all the bits from the frag list -- free it. */ |
1554 | skb_frag_list_init(skb); | |
1555 | xenvif_skb_zerocopy_prepare(queue, nskb); | |
1556 | kfree_skb(nskb); | |
1557 | ||
49d9991a DV |
1558 | /* Release all the original (foreign) frags. */ |
1559 | for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) | |
1560 | skb_frag_unref(skb, f); | |
e3377f36 | 1561 | uarg = skb_shinfo(skb)->destructor_arg; |
a64bd934 WL |
1562 | /* increase inflight counter to offset decrement in callback */ |
1563 | atomic_inc(&queue->inflight_packets); | |
e3377f36 ZK |
1564 | uarg->callback(uarg, true); |
1565 | skb_shinfo(skb)->destructor_arg = NULL; | |
1566 | ||
b0c21bad DV |
1567 | /* Fill the skb with the new (local) frags. */ |
1568 | memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t)); | |
1569 | skb_shinfo(skb)->nr_frags = i; | |
1570 | skb->truesize += i * PAGE_SIZE; | |
e3377f36 ZK |
1571 | |
1572 | return 0; | |
1573 | } | |
b3f980bd | 1574 | |
e9ce7cb6 | 1575 | static int xenvif_tx_submit(struct xenvif_queue *queue) |
f942dc25 | 1576 | { |
e9ce7cb6 WL |
1577 | struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops; |
1578 | struct gnttab_copy *gop_copy = queue->tx_copy_ops; | |
f942dc25 | 1579 | struct sk_buff *skb; |
b3f980bd | 1580 | int work_done = 0; |
f942dc25 | 1581 | |
e9ce7cb6 | 1582 | while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) { |
f942dc25 | 1583 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1584 | u16 pending_idx; |
1585 | unsigned data_len; | |
1586 | ||
8f13dd96 | 1587 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
e9ce7cb6 | 1588 | txp = &queue->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1589 | |
1590 | /* Check the remap error code. */ | |
e9ce7cb6 | 1591 | if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) { |
b42cc6e4 ZK |
1592 | /* If there was an error, xenvif_tx_check_gop is |
1593 | * expected to release all the frags which were mapped, | |
1594 | * so kfree_skb shouldn't do it again | |
1595 | */ | |
f942dc25 | 1596 | skb_shinfo(skb)->nr_frags = 0; |
b42cc6e4 ZK |
1597 | if (skb_has_frag_list(skb)) { |
1598 | struct sk_buff *nskb = | |
1599 | skb_shinfo(skb)->frag_list; | |
1600 | skb_shinfo(nskb)->nr_frags = 0; | |
1601 | } | |
f942dc25 IC |
1602 | kfree_skb(skb); |
1603 | continue; | |
1604 | } | |
1605 | ||
1606 | data_len = skb->len; | |
e9ce7cb6 | 1607 | callback_param(queue, pending_idx).ctx = NULL; |
f942dc25 IC |
1608 | if (data_len < txp->size) { |
1609 | /* Append the packet payload as a fragment. */ | |
1610 | txp->offset += data_len; | |
1611 | txp->size -= data_len; | |
1612 | } else { | |
1613 | /* Schedule a response immediately. */ | |
e9ce7cb6 | 1614 | xenvif_idx_release(queue, pending_idx, |
bdab8275 | 1615 | XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1616 | } |
1617 | ||
1618 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1619 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1620 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1621 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1622 | ||
e9ce7cb6 | 1623 | xenvif_fill_frags(queue, skb); |
f942dc25 | 1624 | |
e3377f36 | 1625 | if (unlikely(skb_has_frag_list(skb))) { |
e9ce7cb6 | 1626 | if (xenvif_handle_frag_list(queue, skb)) { |
e3377f36 | 1627 | if (net_ratelimit()) |
e9ce7cb6 | 1628 | netdev_err(queue->vif->dev, |
e3377f36 | 1629 | "Not enough memory to consolidate frag_list!\n"); |
a64bd934 | 1630 | xenvif_skb_zerocopy_prepare(queue, skb); |
e3377f36 ZK |
1631 | kfree_skb(skb); |
1632 | continue; | |
1633 | } | |
1634 | } | |
1635 | ||
e9ce7cb6 | 1636 | skb->dev = queue->vif->dev; |
f942dc25 | 1637 | skb->protocol = eth_type_trans(skb, skb->dev); |
f9ca8f74 | 1638 | skb_reset_network_header(skb); |
f942dc25 | 1639 | |
e9ce7cb6 WL |
1640 | if (checksum_setup(queue, skb)) { |
1641 | netdev_dbg(queue->vif->dev, | |
f942dc25 | 1642 | "Can't setup checksum in net_tx_action\n"); |
f53c3fe8 ZK |
1643 | /* We have to set this flag to trigger the callback */ |
1644 | if (skb_shinfo(skb)->destructor_arg) | |
a64bd934 | 1645 | xenvif_skb_zerocopy_prepare(queue, skb); |
f942dc25 IC |
1646 | kfree_skb(skb); |
1647 | continue; | |
1648 | } | |
1649 | ||
40893fd0 | 1650 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1651 | |
b89587a7 PD |
1652 | /* If the packet is GSO then we will have just set up the |
1653 | * transport header offset in checksum_setup so it's now | |
1654 | * straightforward to calculate gso_segs. | |
1655 | */ | |
1656 | if (skb_is_gso(skb)) { | |
1657 | int mss = skb_shinfo(skb)->gso_size; | |
1658 | int hdrlen = skb_transport_header(skb) - | |
1659 | skb_mac_header(skb) + | |
1660 | tcp_hdrlen(skb); | |
1661 | ||
1662 | skb_shinfo(skb)->gso_segs = | |
1663 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1664 | } | |
1665 | ||
e9ce7cb6 WL |
1666 | queue->stats.rx_bytes += skb->len; |
1667 | queue->stats.rx_packets++; | |
f942dc25 | 1668 | |
b3f980bd WL |
1669 | work_done++; |
1670 | ||
f53c3fe8 ZK |
1671 | /* Set this flag right before netif_receive_skb, otherwise |
1672 | * someone might think this packet already left netback, and | |
1673 | * do a skb_copy_ubufs while we are still in control of the | |
1674 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1675 | */ | |
1bb332af | 1676 | if (skb_shinfo(skb)->destructor_arg) { |
a64bd934 | 1677 | xenvif_skb_zerocopy_prepare(queue, skb); |
e9ce7cb6 | 1678 | queue->stats.tx_zerocopy_sent++; |
1bb332af | 1679 | } |
f53c3fe8 | 1680 | |
b3f980bd | 1681 | netif_receive_skb(skb); |
f942dc25 | 1682 | } |
b3f980bd WL |
1683 | |
1684 | return work_done; | |
f942dc25 IC |
1685 | } |
1686 | ||
3e2234b3 ZK |
1687 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1688 | { | |
f53c3fe8 ZK |
1689 | unsigned long flags; |
1690 | pending_ring_idx_t index; | |
e9ce7cb6 | 1691 | struct xenvif_queue *queue = ubuf_to_queue(ubuf); |
f53c3fe8 ZK |
1692 | |
1693 | /* This is the only place where we grab this lock, to protect callbacks | |
1694 | * from each other. | |
1695 | */ | |
e9ce7cb6 | 1696 | spin_lock_irqsave(&queue->callback_lock, flags); |
f53c3fe8 ZK |
1697 | do { |
1698 | u16 pending_idx = ubuf->desc; | |
1699 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
e9ce7cb6 | 1700 | BUG_ON(queue->dealloc_prod - queue->dealloc_cons >= |
f53c3fe8 | 1701 | MAX_PENDING_REQS); |
e9ce7cb6 WL |
1702 | index = pending_index(queue->dealloc_prod); |
1703 | queue->dealloc_ring[index] = pending_idx; | |
f53c3fe8 ZK |
1704 | /* Sync with xenvif_tx_dealloc_action: |
1705 | * insert idx then incr producer. | |
1706 | */ | |
1707 | smp_wmb(); | |
e9ce7cb6 | 1708 | queue->dealloc_prod++; |
f53c3fe8 | 1709 | } while (ubuf); |
e9ce7cb6 | 1710 | spin_unlock_irqrestore(&queue->callback_lock, flags); |
f53c3fe8 | 1711 | |
1bb332af | 1712 | if (likely(zerocopy_success)) |
e9ce7cb6 | 1713 | queue->stats.tx_zerocopy_success++; |
1bb332af | 1714 | else |
e9ce7cb6 | 1715 | queue->stats.tx_zerocopy_fail++; |
a64bd934 | 1716 | xenvif_skb_zerocopy_complete(queue); |
f53c3fe8 ZK |
1717 | } |
1718 | ||
e9ce7cb6 | 1719 | static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue) |
f53c3fe8 ZK |
1720 | { |
1721 | struct gnttab_unmap_grant_ref *gop; | |
1722 | pending_ring_idx_t dc, dp; | |
1723 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1724 | unsigned int i = 0; | |
1725 | ||
e9ce7cb6 WL |
1726 | dc = queue->dealloc_cons; |
1727 | gop = queue->tx_unmap_ops; | |
f53c3fe8 ZK |
1728 | |
1729 | /* Free up any grants we have finished using */ | |
1730 | do { | |
e9ce7cb6 | 1731 | dp = queue->dealloc_prod; |
f53c3fe8 ZK |
1732 | |
1733 | /* Ensure we see all indices enqueued by all | |
1734 | * xenvif_zerocopy_callback(). | |
1735 | */ | |
1736 | smp_rmb(); | |
1737 | ||
1738 | while (dc != dp) { | |
50c2e4dd | 1739 | BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS); |
f53c3fe8 | 1740 | pending_idx = |
e9ce7cb6 | 1741 | queue->dealloc_ring[pending_index(dc++)]; |
f53c3fe8 | 1742 | |
50c2e4dd | 1743 | pending_idx_release[gop - queue->tx_unmap_ops] = |
f53c3fe8 | 1744 | pending_idx; |
50c2e4dd | 1745 | queue->pages_to_unmap[gop - queue->tx_unmap_ops] = |
e9ce7cb6 | 1746 | queue->mmap_pages[pending_idx]; |
f53c3fe8 | 1747 | gnttab_set_unmap_op(gop, |
e9ce7cb6 | 1748 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1749 | GNTMAP_host_map, |
e9ce7cb6 WL |
1750 | queue->grant_tx_handle[pending_idx]); |
1751 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1752 | ++gop; |
1753 | } | |
1754 | ||
e9ce7cb6 | 1755 | } while (dp != queue->dealloc_prod); |
f53c3fe8 | 1756 | |
e9ce7cb6 | 1757 | queue->dealloc_cons = dc; |
f53c3fe8 | 1758 | |
e9ce7cb6 | 1759 | if (gop - queue->tx_unmap_ops > 0) { |
f53c3fe8 | 1760 | int ret; |
e9ce7cb6 | 1761 | ret = gnttab_unmap_refs(queue->tx_unmap_ops, |
f53c3fe8 | 1762 | NULL, |
e9ce7cb6 WL |
1763 | queue->pages_to_unmap, |
1764 | gop - queue->tx_unmap_ops); | |
f53c3fe8 | 1765 | if (ret) { |
68946159 | 1766 | netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n", |
e9ce7cb6 WL |
1767 | gop - queue->tx_unmap_ops, ret); |
1768 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) { | |
f53c3fe8 | 1769 | if (gop[i].status != GNTST_okay) |
e9ce7cb6 | 1770 | netdev_err(queue->vif->dev, |
68946159 | 1771 | " host_addr: 0x%llx handle: 0x%x status: %d\n", |
f53c3fe8 ZK |
1772 | gop[i].host_addr, |
1773 | gop[i].handle, | |
1774 | gop[i].status); | |
1775 | } | |
1776 | BUG(); | |
1777 | } | |
1778 | } | |
1779 | ||
e9ce7cb6 WL |
1780 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) |
1781 | xenvif_idx_release(queue, pending_idx_release[i], | |
f53c3fe8 | 1782 | XEN_NETIF_RSP_OKAY); |
3e2234b3 ZK |
1783 | } |
1784 | ||
f53c3fe8 | 1785 | |
f942dc25 | 1786 | /* Called after netfront has transmitted */ |
e9ce7cb6 | 1787 | int xenvif_tx_action(struct xenvif_queue *queue, int budget) |
f942dc25 | 1788 | { |
bdab8275 | 1789 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1790 | int work_done, ret; |
f942dc25 | 1791 | |
e9ce7cb6 | 1792 | if (unlikely(!tx_work_todo(queue))) |
b3f980bd WL |
1793 | return 0; |
1794 | ||
e9ce7cb6 | 1795 | xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops); |
f942dc25 | 1796 | |
bdab8275 | 1797 | if (nr_cops == 0) |
b3f980bd WL |
1798 | return 0; |
1799 | ||
e9ce7cb6 | 1800 | gnttab_batch_copy(queue->tx_copy_ops, nr_cops); |
bdab8275 | 1801 | if (nr_mops != 0) { |
e9ce7cb6 | 1802 | ret = gnttab_map_refs(queue->tx_map_ops, |
bdab8275 | 1803 | NULL, |
e9ce7cb6 | 1804 | queue->pages_to_map, |
bdab8275 ZK |
1805 | nr_mops); |
1806 | BUG_ON(ret); | |
1807 | } | |
f942dc25 | 1808 | |
e9ce7cb6 | 1809 | work_done = xenvif_tx_submit(queue); |
f942dc25 | 1810 | |
b3f980bd | 1811 | return work_done; |
f942dc25 IC |
1812 | } |
1813 | ||
e9ce7cb6 | 1814 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a | 1815 | u8 status) |
f942dc25 | 1816 | { |
f942dc25 | 1817 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1818 | pending_ring_idx_t index; |
f53c3fe8 | 1819 | unsigned long flags; |
2810e5b9 | 1820 | |
e9ce7cb6 | 1821 | pending_tx_info = &queue->pending_tx_info[pending_idx]; |
7fbb9d84 | 1822 | |
e9ce7cb6 | 1823 | spin_lock_irqsave(&queue->response_lock, flags); |
7fbb9d84 | 1824 | |
562abd39 PD |
1825 | make_tx_response(queue, &pending_tx_info->req, |
1826 | pending_tx_info->extra_count, status); | |
7fbb9d84 DV |
1827 | |
1828 | /* Release the pending index before pusing the Tx response so | |
1829 | * its available before a new Tx request is pushed by the | |
1830 | * frontend. | |
1831 | */ | |
1832 | index = pending_index(queue->pending_prod++); | |
e9ce7cb6 | 1833 | queue->pending_ring[index] = pending_idx; |
7fbb9d84 | 1834 | |
c8a4d299 | 1835 | push_tx_responses(queue); |
7fbb9d84 | 1836 | |
e9ce7cb6 | 1837 | spin_unlock_irqrestore(&queue->response_lock, flags); |
f942dc25 IC |
1838 | } |
1839 | ||
2810e5b9 | 1840 | |
e9ce7cb6 | 1841 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 | 1842 | struct xen_netif_tx_request *txp, |
562abd39 | 1843 | unsigned int extra_count, |
f942dc25 IC |
1844 | s8 st) |
1845 | { | |
e9ce7cb6 | 1846 | RING_IDX i = queue->tx.rsp_prod_pvt; |
f942dc25 | 1847 | struct xen_netif_tx_response *resp; |
f942dc25 | 1848 | |
e9ce7cb6 | 1849 | resp = RING_GET_RESPONSE(&queue->tx, i); |
f942dc25 IC |
1850 | resp->id = txp->id; |
1851 | resp->status = st; | |
1852 | ||
562abd39 | 1853 | while (extra_count-- != 0) |
e9ce7cb6 | 1854 | RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL; |
f942dc25 | 1855 | |
e9ce7cb6 | 1856 | queue->tx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1857 | } |
1858 | ||
c8a4d299 DV |
1859 | static void push_tx_responses(struct xenvif_queue *queue) |
1860 | { | |
1861 | int notify; | |
1862 | ||
1863 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify); | |
1864 | if (notify) | |
1865 | notify_remote_via_irq(queue->tx_irq); | |
1866 | } | |
1867 | ||
e9ce7cb6 | 1868 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1869 | u16 id, |
1870 | s8 st, | |
1871 | u16 offset, | |
1872 | u16 size, | |
1873 | u16 flags) | |
1874 | { | |
e9ce7cb6 | 1875 | RING_IDX i = queue->rx.rsp_prod_pvt; |
f942dc25 IC |
1876 | struct xen_netif_rx_response *resp; |
1877 | ||
e9ce7cb6 | 1878 | resp = RING_GET_RESPONSE(&queue->rx, i); |
f942dc25 IC |
1879 | resp->offset = offset; |
1880 | resp->flags = flags; | |
1881 | resp->id = id; | |
1882 | resp->status = (s16)size; | |
1883 | if (st < 0) | |
1884 | resp->status = (s16)st; | |
1885 | ||
e9ce7cb6 | 1886 | queue->rx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1887 | |
1888 | return resp; | |
1889 | } | |
1890 | ||
e9ce7cb6 | 1891 | void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx) |
f53c3fe8 ZK |
1892 | { |
1893 | int ret; | |
1894 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1895 | ||
1896 | gnttab_set_unmap_op(&tx_unmap_op, | |
e9ce7cb6 | 1897 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1898 | GNTMAP_host_map, |
e9ce7cb6 WL |
1899 | queue->grant_tx_handle[pending_idx]); |
1900 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1901 | |
1902 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
e9ce7cb6 | 1903 | &queue->mmap_pages[pending_idx], 1); |
7aceb47a | 1904 | if (ret) { |
e9ce7cb6 | 1905 | netdev_err(queue->vif->dev, |
68946159 | 1906 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n", |
7aceb47a ZK |
1907 | ret, |
1908 | pending_idx, | |
1909 | tx_unmap_op.host_addr, | |
1910 | tx_unmap_op.handle, | |
1911 | tx_unmap_op.status); | |
1912 | BUG(); | |
1913 | } | |
f53c3fe8 ZK |
1914 | } |
1915 | ||
e9ce7cb6 | 1916 | static inline int tx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1917 | { |
e9ce7cb6 | 1918 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))) |
f942dc25 IC |
1919 | return 1; |
1920 | ||
1921 | return 0; | |
1922 | } | |
1923 | ||
e9ce7cb6 | 1924 | static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue) |
f53c3fe8 | 1925 | { |
e9ce7cb6 | 1926 | return queue->dealloc_cons != queue->dealloc_prod; |
f53c3fe8 ZK |
1927 | } |
1928 | ||
e9ce7cb6 | 1929 | void xenvif_unmap_frontend_rings(struct xenvif_queue *queue) |
f942dc25 | 1930 | { |
e9ce7cb6 WL |
1931 | if (queue->tx.sring) |
1932 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1933 | queue->tx.sring); | |
1934 | if (queue->rx.sring) | |
1935 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1936 | queue->rx.sring); | |
f942dc25 IC |
1937 | } |
1938 | ||
e9ce7cb6 | 1939 | int xenvif_map_frontend_rings(struct xenvif_queue *queue, |
7376419a WL |
1940 | grant_ref_t tx_ring_ref, |
1941 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1942 | { |
c9d63699 | 1943 | void *addr; |
f942dc25 IC |
1944 | struct xen_netif_tx_sring *txs; |
1945 | struct xen_netif_rx_sring *rxs; | |
1946 | ||
1947 | int err = -ENOMEM; | |
1948 | ||
e9ce7cb6 | 1949 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
ccc9d90a | 1950 | &tx_ring_ref, 1, &addr); |
c9d63699 | 1951 | if (err) |
f942dc25 IC |
1952 | goto err; |
1953 | ||
c9d63699 | 1954 | txs = (struct xen_netif_tx_sring *)addr; |
d0089e8a | 1955 | BACK_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
f942dc25 | 1956 | |
e9ce7cb6 | 1957 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
ccc9d90a | 1958 | &rx_ring_ref, 1, &addr); |
c9d63699 | 1959 | if (err) |
f942dc25 | 1960 | goto err; |
f942dc25 | 1961 | |
c9d63699 | 1962 | rxs = (struct xen_netif_rx_sring *)addr; |
d0089e8a | 1963 | BACK_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
f942dc25 IC |
1964 | |
1965 | return 0; | |
1966 | ||
1967 | err: | |
e9ce7cb6 | 1968 | xenvif_unmap_frontend_rings(queue); |
f942dc25 IC |
1969 | return err; |
1970 | } | |
1971 | ||
ecf08d2d | 1972 | static void xenvif_queue_carrier_off(struct xenvif_queue *queue) |
ca2f09f2 | 1973 | { |
ecf08d2d DV |
1974 | struct xenvif *vif = queue->vif; |
1975 | ||
1976 | queue->stalled = true; | |
1977 | ||
1978 | /* At least one queue has stalled? Disable the carrier. */ | |
1979 | spin_lock(&vif->lock); | |
1980 | if (vif->stalled_queues++ == 0) { | |
1981 | netdev_info(vif->dev, "Guest Rx stalled"); | |
1982 | netif_carrier_off(vif->dev); | |
1983 | } | |
1984 | spin_unlock(&vif->lock); | |
ca2f09f2 PD |
1985 | } |
1986 | ||
ecf08d2d | 1987 | static void xenvif_queue_carrier_on(struct xenvif_queue *queue) |
f34a4cf9 | 1988 | { |
ecf08d2d | 1989 | struct xenvif *vif = queue->vif; |
f34a4cf9 | 1990 | |
ecf08d2d DV |
1991 | queue->last_rx_time = jiffies; /* Reset Rx stall detection. */ |
1992 | queue->stalled = false; | |
f34a4cf9 | 1993 | |
ecf08d2d DV |
1994 | /* All queues are ready? Enable the carrier. */ |
1995 | spin_lock(&vif->lock); | |
1996 | if (--vif->stalled_queues == 0) { | |
1997 | netdev_info(vif->dev, "Guest Rx ready"); | |
1998 | netif_carrier_on(vif->dev); | |
1999 | } | |
2000 | spin_unlock(&vif->lock); | |
2001 | } | |
f34a4cf9 | 2002 | |
ecf08d2d DV |
2003 | static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue) |
2004 | { | |
2005 | RING_IDX prod, cons; | |
2006 | ||
2007 | prod = queue->rx.sring->req_prod; | |
2008 | cons = queue->rx.req_cons; | |
2009 | ||
1d5d4852 | 2010 | return !queue->stalled && prod - cons < 1 |
ecf08d2d | 2011 | && time_after(jiffies, |
26c0e102 | 2012 | queue->last_rx_time + queue->vif->stall_timeout); |
ecf08d2d DV |
2013 | } |
2014 | ||
2015 | static bool xenvif_rx_queue_ready(struct xenvif_queue *queue) | |
2016 | { | |
2017 | RING_IDX prod, cons; | |
2018 | ||
2019 | prod = queue->rx.sring->req_prod; | |
2020 | cons = queue->rx.req_cons; | |
2021 | ||
1d5d4852 | 2022 | return queue->stalled && prod - cons >= 1; |
ecf08d2d DV |
2023 | } |
2024 | ||
f48da8b1 | 2025 | static bool xenvif_have_rx_work(struct xenvif_queue *queue) |
ca2f09f2 | 2026 | { |
99a2dea5 | 2027 | return xenvif_rx_ring_slots_available(queue) |
26c0e102 DV |
2028 | || (queue->vif->stall_timeout && |
2029 | (xenvif_rx_queue_stalled(queue) | |
2030 | || xenvif_rx_queue_ready(queue))) | |
f48da8b1 DV |
2031 | || kthread_should_stop() |
2032 | || queue->vif->disabled; | |
ca2f09f2 PD |
2033 | } |
2034 | ||
f48da8b1 | 2035 | static long xenvif_rx_queue_timeout(struct xenvif_queue *queue) |
f34a4cf9 | 2036 | { |
f48da8b1 DV |
2037 | struct sk_buff *skb; |
2038 | long timeout; | |
f34a4cf9 | 2039 | |
f48da8b1 DV |
2040 | skb = skb_peek(&queue->rx_queue); |
2041 | if (!skb) | |
2042 | return MAX_SCHEDULE_TIMEOUT; | |
f34a4cf9 | 2043 | |
f48da8b1 DV |
2044 | timeout = XENVIF_RX_CB(skb)->expires - jiffies; |
2045 | return timeout < 0 ? 0 : timeout; | |
2046 | } | |
f34a4cf9 | 2047 | |
f48da8b1 DV |
2048 | /* Wait until the guest Rx thread has work. |
2049 | * | |
2050 | * The timeout needs to be adjusted based on the current head of the | |
2051 | * queue (and not just the head at the beginning). In particular, if | |
2052 | * the queue is initially empty an infinite timeout is used and this | |
2053 | * needs to be reduced when a skb is queued. | |
2054 | * | |
2055 | * This cannot be done with wait_event_timeout() because it only | |
2056 | * calculates the timeout once. | |
2057 | */ | |
2058 | static void xenvif_wait_for_rx_work(struct xenvif_queue *queue) | |
2059 | { | |
2060 | DEFINE_WAIT(wait); | |
2061 | ||
2062 | if (xenvif_have_rx_work(queue)) | |
2063 | return; | |
2064 | ||
2065 | for (;;) { | |
2066 | long ret; | |
2067 | ||
2068 | prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE); | |
2069 | if (xenvif_have_rx_work(queue)) | |
2070 | break; | |
2071 | ret = schedule_timeout(xenvif_rx_queue_timeout(queue)); | |
2072 | if (!ret) | |
2073 | break; | |
f34a4cf9 | 2074 | } |
f48da8b1 | 2075 | finish_wait(&queue->wq, &wait); |
f34a4cf9 ZK |
2076 | } |
2077 | ||
121fa4b7 | 2078 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd | 2079 | { |
e9ce7cb6 | 2080 | struct xenvif_queue *queue = data; |
f48da8b1 | 2081 | struct xenvif *vif = queue->vif; |
b3f980bd | 2082 | |
26c0e102 DV |
2083 | if (!vif->stall_timeout) |
2084 | xenvif_queue_carrier_on(queue); | |
2085 | ||
f48da8b1 DV |
2086 | for (;;) { |
2087 | xenvif_wait_for_rx_work(queue); | |
e9d8b2c2 | 2088 | |
f34a4cf9 ZK |
2089 | if (kthread_should_stop()) |
2090 | break; | |
2091 | ||
e9d8b2c2 WL |
2092 | /* This frontend is found to be rogue, disable it in |
2093 | * kthread context. Currently this is only set when | |
2094 | * netback finds out frontend sends malformed packet, | |
2095 | * but we cannot disable the interface in softirq | |
e9ce7cb6 WL |
2096 | * context so we defer it here, if this thread is |
2097 | * associated with queue 0. | |
e9d8b2c2 | 2098 | */ |
f48da8b1 DV |
2099 | if (unlikely(vif->disabled && queue->id == 0)) { |
2100 | xenvif_carrier_off(vif); | |
42b5212f | 2101 | break; |
09350788 ZK |
2102 | } |
2103 | ||
e9ce7cb6 WL |
2104 | if (!skb_queue_empty(&queue->rx_queue)) |
2105 | xenvif_rx_action(queue); | |
b3f980bd | 2106 | |
ecf08d2d DV |
2107 | /* If the guest hasn't provided any Rx slots for a |
2108 | * while it's probably not responsive, drop the | |
2109 | * carrier so packets are dropped earlier. | |
2110 | */ | |
26c0e102 DV |
2111 | if (vif->stall_timeout) { |
2112 | if (xenvif_rx_queue_stalled(queue)) | |
2113 | xenvif_queue_carrier_off(queue); | |
2114 | else if (xenvif_rx_queue_ready(queue)) | |
2115 | xenvif_queue_carrier_on(queue); | |
2116 | } | |
ecf08d2d | 2117 | |
f48da8b1 DV |
2118 | /* Queued packets may have foreign pages from other |
2119 | * domains. These cannot be queued indefinitely as | |
2120 | * this would starve guests of grant refs and transmit | |
2121 | * slots. | |
2122 | */ | |
2123 | xenvif_rx_queue_drop_expired(queue); | |
2124 | ||
2125 | xenvif_rx_queue_maybe_wake(queue); | |
2126 | ||
b3f980bd WL |
2127 | cond_resched(); |
2128 | } | |
2129 | ||
ca2f09f2 | 2130 | /* Bin any remaining skbs */ |
f48da8b1 | 2131 | xenvif_rx_queue_purge(queue); |
ca2f09f2 | 2132 | |
b3f980bd WL |
2133 | return 0; |
2134 | } | |
2135 | ||
a64bd934 WL |
2136 | static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue) |
2137 | { | |
2138 | /* Dealloc thread must remain running until all inflight | |
2139 | * packets complete. | |
2140 | */ | |
2141 | return kthread_should_stop() && | |
2142 | !atomic_read(&queue->inflight_packets); | |
2143 | } | |
2144 | ||
f53c3fe8 ZK |
2145 | int xenvif_dealloc_kthread(void *data) |
2146 | { | |
e9ce7cb6 | 2147 | struct xenvif_queue *queue = data; |
f53c3fe8 | 2148 | |
a64bd934 | 2149 | for (;;) { |
e9ce7cb6 WL |
2150 | wait_event_interruptible(queue->dealloc_wq, |
2151 | tx_dealloc_work_todo(queue) || | |
a64bd934 WL |
2152 | xenvif_dealloc_kthread_should_stop(queue)); |
2153 | if (xenvif_dealloc_kthread_should_stop(queue)) | |
f53c3fe8 ZK |
2154 | break; |
2155 | ||
e9ce7cb6 | 2156 | xenvif_tx_dealloc_action(queue); |
f53c3fe8 ZK |
2157 | cond_resched(); |
2158 | } | |
2159 | ||
2160 | /* Unmap anything remaining*/ | |
e9ce7cb6 WL |
2161 | if (tx_dealloc_work_todo(queue)) |
2162 | xenvif_tx_dealloc_action(queue); | |
f53c3fe8 ZK |
2163 | |
2164 | return 0; | |
2165 | } | |
2166 | ||
f942dc25 IC |
2167 | static int __init netback_init(void) |
2168 | { | |
f942dc25 | 2169 | int rc = 0; |
f942dc25 | 2170 | |
2a14b244 | 2171 | if (!xen_domain()) |
f942dc25 IC |
2172 | return -ENODEV; |
2173 | ||
4c82ac3c WL |
2174 | /* Allow as many queues as there are CPUs if user has not |
2175 | * specified a value. | |
2176 | */ | |
2177 | if (xenvif_max_queues == 0) | |
2178 | xenvif_max_queues = num_online_cpus(); | |
8d3d53b3 | 2179 | |
37641494 | 2180 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
2181 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
2182 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 2183 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
2184 | } |
2185 | ||
f942dc25 IC |
2186 | rc = xenvif_xenbus_init(); |
2187 | if (rc) | |
2188 | goto failed_init; | |
2189 | ||
f51de243 ZK |
2190 | #ifdef CONFIG_DEBUG_FS |
2191 | xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL); | |
2192 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
2193 | pr_warn("Init of debugfs returned %ld!\n", | |
2194 | PTR_ERR(xen_netback_dbg_root)); | |
2195 | #endif /* CONFIG_DEBUG_FS */ | |
2196 | ||
f942dc25 IC |
2197 | return 0; |
2198 | ||
2199 | failed_init: | |
f942dc25 | 2200 | return rc; |
f942dc25 IC |
2201 | } |
2202 | ||
2203 | module_init(netback_init); | |
2204 | ||
b103f358 WL |
2205 | static void __exit netback_fini(void) |
2206 | { | |
f51de243 ZK |
2207 | #ifdef CONFIG_DEBUG_FS |
2208 | if (!IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
2209 | debugfs_remove_recursive(xen_netback_dbg_root); | |
2210 | #endif /* CONFIG_DEBUG_FS */ | |
b103f358 | 2211 | xenvif_xenbus_fini(); |
b103f358 WL |
2212 | } |
2213 | module_exit(netback_fini); | |
2214 | ||
f942dc25 | 2215 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 2216 | MODULE_ALIAS("xen-backend:vif"); |