Commit | Line | Data |
---|---|---|
f942dc25 IC |
1 | /* |
2 | * Back-end of the driver for virtual network devices. This portion of the | |
3 | * driver exports a 'unified' network-device interface that can be accessed | |
4 | * by any operating system that implements a compatible front end. A | |
5 | * reference front-end implementation can be found in: | |
6 | * drivers/net/xen-netfront.c | |
7 | * | |
8 | * Copyright (c) 2002-2005, K A Fraser | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License version 2 | |
12 | * as published by the Free Software Foundation; or, when distributed | |
13 | * separately from the Linux kernel or incorporated into other | |
14 | * software packages, subject to the following license: | |
15 | * | |
16 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
17 | * of this source file (the "Software"), to deal in the Software without | |
18 | * restriction, including without limitation the rights to use, copy, modify, | |
19 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
20 | * and to permit persons to whom the Software is furnished to do so, subject to | |
21 | * the following conditions: | |
22 | * | |
23 | * The above copyright notice and this permission notice shall be included in | |
24 | * all copies or substantial portions of the Software. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
27 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
28 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
29 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
30 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
31 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
32 | * IN THE SOFTWARE. | |
33 | */ | |
34 | ||
35 | #include "common.h" | |
36 | ||
37 | #include <linux/kthread.h> | |
38 | #include <linux/if_vlan.h> | |
39 | #include <linux/udp.h> | |
e3377f36 | 40 | #include <linux/highmem.h> |
f942dc25 IC |
41 | |
42 | #include <net/tcp.h> | |
43 | ||
ca981633 | 44 | #include <xen/xen.h> |
f942dc25 IC |
45 | #include <xen/events.h> |
46 | #include <xen/interface/memory.h> | |
47 | ||
48 | #include <asm/xen/hypercall.h> | |
49 | #include <asm/xen/page.h> | |
50 | ||
e1f00a69 WL |
51 | /* Provide an option to disable split event channels at load time as |
52 | * event channels are limited resource. Split event channels are | |
53 | * enabled by default. | |
54 | */ | |
55 | bool separate_tx_rx_irq = 1; | |
56 | module_param(separate_tx_rx_irq, bool, 0644); | |
57 | ||
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); | |
b3f980bd | 99 | |
e9ce7cb6 | 100 | static inline int tx_work_todo(struct xenvif_queue *queue); |
b3f980bd | 101 | |
e9ce7cb6 | 102 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
103 | u16 id, |
104 | s8 st, | |
105 | u16 offset, | |
106 | u16 size, | |
107 | u16 flags); | |
108 | ||
e9ce7cb6 | 109 | static inline unsigned long idx_to_pfn(struct xenvif_queue *queue, |
ea066ad1 | 110 | u16 idx) |
f942dc25 | 111 | { |
e9ce7cb6 | 112 | return page_to_pfn(queue->mmap_pages[idx]); |
f942dc25 IC |
113 | } |
114 | ||
e9ce7cb6 | 115 | static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue, |
ea066ad1 | 116 | u16 idx) |
f942dc25 | 117 | { |
e9ce7cb6 | 118 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx)); |
f942dc25 IC |
119 | } |
120 | ||
7aceb47a ZK |
121 | #define callback_param(vif, pending_idx) \ |
122 | (vif->pending_tx_info[pending_idx].callback_struct) | |
123 | ||
f53c3fe8 ZK |
124 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
125 | */ | |
e9ce7cb6 | 126 | static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf) |
3e2234b3 | 127 | { |
f53c3fe8 ZK |
128 | u16 pending_idx = ubuf->desc; |
129 | struct pending_tx_info *temp = | |
130 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
131 | return container_of(temp - pending_idx, | |
e9ce7cb6 | 132 | struct xenvif_queue, |
f53c3fe8 | 133 | pending_tx_info[0]); |
3e2234b3 | 134 | } |
f53c3fe8 | 135 | |
ea066ad1 IC |
136 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
137 | { | |
138 | return (u16)frag->page_offset; | |
139 | } | |
140 | ||
141 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
142 | { | |
143 | frag->page_offset = pending_idx; | |
144 | } | |
145 | ||
f942dc25 IC |
146 | static inline pending_ring_idx_t pending_index(unsigned i) |
147 | { | |
148 | return i & (MAX_PENDING_REQS-1); | |
149 | } | |
150 | ||
e9ce7cb6 | 151 | bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue, int needed) |
f942dc25 | 152 | { |
ca2f09f2 | 153 | RING_IDX prod, cons; |
f942dc25 | 154 | |
ca2f09f2 | 155 | do { |
e9ce7cb6 WL |
156 | prod = queue->rx.sring->req_prod; |
157 | cons = queue->rx.req_cons; | |
f942dc25 | 158 | |
ca2f09f2 PD |
159 | if (prod - cons >= needed) |
160 | return true; | |
f942dc25 | 161 | |
e9ce7cb6 | 162 | queue->rx.sring->req_event = prod + 1; |
f942dc25 | 163 | |
ca2f09f2 PD |
164 | /* Make sure event is visible before we check prod |
165 | * again. | |
166 | */ | |
167 | mb(); | |
e9ce7cb6 | 168 | } while (queue->rx.sring->req_prod != prod); |
f942dc25 | 169 | |
ca2f09f2 | 170 | return false; |
f942dc25 IC |
171 | } |
172 | ||
f48da8b1 DV |
173 | void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb) |
174 | { | |
175 | unsigned long flags; | |
176 | ||
177 | spin_lock_irqsave(&queue->rx_queue.lock, flags); | |
178 | ||
179 | __skb_queue_tail(&queue->rx_queue, skb); | |
180 | ||
181 | queue->rx_queue_len += skb->len; | |
182 | if (queue->rx_queue_len > queue->rx_queue_max) | |
183 | netif_tx_stop_queue(netdev_get_tx_queue(queue->vif->dev, queue->id)); | |
184 | ||
185 | spin_unlock_irqrestore(&queue->rx_queue.lock, flags); | |
186 | } | |
187 | ||
188 | static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue) | |
189 | { | |
190 | struct sk_buff *skb; | |
191 | ||
192 | spin_lock_irq(&queue->rx_queue.lock); | |
193 | ||
194 | skb = __skb_dequeue(&queue->rx_queue); | |
195 | if (skb) | |
196 | queue->rx_queue_len -= skb->len; | |
197 | ||
198 | spin_unlock_irq(&queue->rx_queue.lock); | |
199 | ||
200 | return skb; | |
201 | } | |
202 | ||
203 | static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue) | |
204 | { | |
205 | spin_lock_irq(&queue->rx_queue.lock); | |
206 | ||
207 | if (queue->rx_queue_len < queue->rx_queue_max) | |
208 | netif_tx_wake_queue(netdev_get_tx_queue(queue->vif->dev, queue->id)); | |
209 | ||
210 | spin_unlock_irq(&queue->rx_queue.lock); | |
211 | } | |
212 | ||
213 | ||
214 | static void xenvif_rx_queue_purge(struct xenvif_queue *queue) | |
215 | { | |
216 | struct sk_buff *skb; | |
217 | while ((skb = xenvif_rx_dequeue(queue)) != NULL) | |
218 | kfree_skb(skb); | |
219 | } | |
220 | ||
221 | static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue) | |
222 | { | |
223 | struct sk_buff *skb; | |
224 | ||
225 | for(;;) { | |
226 | skb = skb_peek(&queue->rx_queue); | |
227 | if (!skb) | |
228 | break; | |
229 | if (time_before(jiffies, XENVIF_RX_CB(skb)->expires)) | |
230 | break; | |
231 | xenvif_rx_dequeue(queue); | |
232 | kfree_skb(skb); | |
233 | } | |
234 | } | |
235 | ||
f942dc25 IC |
236 | /* |
237 | * Returns true if we should start a new receive buffer instead of | |
238 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
239 | * bytes. | |
240 | */ | |
59ae9fc6 ZK |
241 | static bool start_new_rx_buffer(int offset, unsigned long size, int head, |
242 | bool full_coalesce) | |
f942dc25 IC |
243 | { |
244 | /* simple case: we have completely filled the current buffer. */ | |
245 | if (offset == MAX_BUFFER_OFFSET) | |
246 | return true; | |
247 | ||
248 | /* | |
249 | * complex case: start a fresh buffer if the current frag | |
250 | * would overflow the current buffer but only if: | |
251 | * (i) this frag would fit completely in the next buffer | |
252 | * and (ii) there is already some data in the current buffer | |
253 | * and (iii) this is not the head buffer. | |
59ae9fc6 | 254 | * and (iv) there is no need to fully utilize the buffers |
f942dc25 IC |
255 | * |
256 | * Where: | |
257 | * - (i) stops us splitting a frag into two copies | |
258 | * unless the frag is too large for a single buffer. | |
259 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
260 | * - (iii) stops us leaving the first buffer | |
261 | * empty. Strictly speaking this is already covered | |
262 | * by (ii) but is explicitly checked because | |
263 | * netfront relies on the first buffer being | |
264 | * non-empty and can crash otherwise. | |
59ae9fc6 ZK |
265 | * - (iv) is needed for skbs which can use up more than MAX_SKB_FRAGS |
266 | * slot | |
f942dc25 IC |
267 | * |
268 | * This means we will effectively linearise small | |
269 | * frags but do not needlessly split large buffers | |
270 | * into multiple copies tend to give large frags their | |
271 | * own buffers as before. | |
272 | */ | |
0576eddf | 273 | BUG_ON(size > MAX_BUFFER_OFFSET); |
59ae9fc6 ZK |
274 | if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head && |
275 | !full_coalesce) | |
f942dc25 IC |
276 | return true; |
277 | ||
278 | return false; | |
279 | } | |
280 | ||
f942dc25 IC |
281 | struct netrx_pending_operations { |
282 | unsigned copy_prod, copy_cons; | |
283 | unsigned meta_prod, meta_cons; | |
284 | struct gnttab_copy *copy; | |
b3f980bd | 285 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
286 | int copy_off; |
287 | grant_ref_t copy_gref; | |
288 | }; | |
289 | ||
e9ce7cb6 | 290 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue, |
b3f980bd | 291 | struct netrx_pending_operations *npo) |
f942dc25 | 292 | { |
b3f980bd | 293 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
294 | struct xen_netif_rx_request *req; |
295 | ||
e9ce7cb6 | 296 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 IC |
297 | |
298 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 299 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
300 | meta->gso_size = 0; |
301 | meta->size = 0; | |
302 | meta->id = req->id; | |
303 | ||
304 | npo->copy_off = 0; | |
305 | npo->copy_gref = req->gref; | |
306 | ||
307 | return meta; | |
308 | } | |
309 | ||
33bc801d WL |
310 | /* |
311 | * Set up the grant operations for this fragment. If it's a flipping | |
312 | * interface, we also set up the unmap request from here. | |
313 | */ | |
e9ce7cb6 | 314 | static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb, |
7376419a WL |
315 | struct netrx_pending_operations *npo, |
316 | struct page *page, unsigned long size, | |
3e2234b3 | 317 | unsigned long offset, int *head, |
e9ce7cb6 | 318 | struct xenvif_queue *foreign_queue, |
3e2234b3 | 319 | grant_ref_t foreign_gref) |
f942dc25 IC |
320 | { |
321 | struct gnttab_copy *copy_gop; | |
b3f980bd | 322 | struct xenvif_rx_meta *meta; |
f942dc25 | 323 | unsigned long bytes; |
5bd07670 | 324 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
325 | |
326 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 327 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
328 | |
329 | meta = npo->meta + npo->meta_prod - 1; | |
330 | ||
6a8ed462 IC |
331 | /* Skip unused frames from start of page */ |
332 | page += offset >> PAGE_SHIFT; | |
333 | offset &= ~PAGE_MASK; | |
334 | ||
f942dc25 | 335 | while (size > 0) { |
6a8ed462 | 336 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
337 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
338 | ||
6a8ed462 IC |
339 | bytes = PAGE_SIZE - offset; |
340 | ||
341 | if (bytes > size) | |
342 | bytes = size; | |
343 | ||
59ae9fc6 ZK |
344 | if (start_new_rx_buffer(npo->copy_off, |
345 | bytes, | |
346 | *head, | |
347 | XENVIF_RX_CB(skb)->full_coalesce)) { | |
f942dc25 IC |
348 | /* |
349 | * Netfront requires there to be some data in the head | |
350 | * buffer. | |
351 | */ | |
33bc801d | 352 | BUG_ON(*head); |
f942dc25 | 353 | |
e9ce7cb6 | 354 | meta = get_next_rx_buffer(queue, npo); |
f942dc25 IC |
355 | } |
356 | ||
f942dc25 IC |
357 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
358 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
359 | ||
360 | copy_gop = npo->copy + npo->copy_prod++; | |
361 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
362 | copy_gop->len = bytes; |
363 | ||
e9ce7cb6 WL |
364 | if (foreign_queue) { |
365 | copy_gop->source.domid = foreign_queue->vif->domid; | |
3e2234b3 ZK |
366 | copy_gop->source.u.ref = foreign_gref; |
367 | copy_gop->flags |= GNTCOPY_source_gref; | |
368 | } else { | |
369 | copy_gop->source.domid = DOMID_SELF; | |
370 | copy_gop->source.u.gmfn = | |
371 | virt_to_mfn(page_address(page)); | |
372 | } | |
f942dc25 | 373 | copy_gop->source.offset = offset; |
f942dc25 | 374 | |
e9ce7cb6 | 375 | copy_gop->dest.domid = queue->vif->domid; |
f942dc25 IC |
376 | copy_gop->dest.offset = npo->copy_off; |
377 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
378 | |
379 | npo->copy_off += bytes; | |
380 | meta->size += bytes; | |
381 | ||
382 | offset += bytes; | |
383 | size -= bytes; | |
384 | ||
6a8ed462 IC |
385 | /* Next frame */ |
386 | if (offset == PAGE_SIZE && size) { | |
387 | BUG_ON(!PageCompound(page)); | |
388 | page++; | |
389 | offset = 0; | |
390 | } | |
391 | ||
f942dc25 | 392 | /* Leave a gap for the GSO descriptor. */ |
5bd07670 AL |
393 | if (skb_is_gso(skb)) { |
394 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
395 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
396 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
397 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
398 | } | |
82cada22 | 399 | |
e9ce7cb6 WL |
400 | if (*head && ((1 << gso_type) & queue->vif->gso_mask)) |
401 | queue->rx.req_cons++; | |
f942dc25 | 402 | |
33bc801d | 403 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
404 | |
405 | } | |
406 | } | |
407 | ||
58375744 ZK |
408 | /* |
409 | * Find the grant ref for a given frag in a chain of struct ubuf_info's | |
410 | * skb: the skb itself | |
411 | * i: the frag's number | |
412 | * ubuf: a pointer to an element in the chain. It should not be NULL | |
413 | * | |
414 | * Returns a pointer to the element in the chain where the page were found. If | |
415 | * not found, returns NULL. | |
416 | * See the definition of callback_struct in common.h for more details about | |
417 | * the chain. | |
418 | */ | |
419 | static const struct ubuf_info *xenvif_find_gref(const struct sk_buff *const skb, | |
420 | const int i, | |
421 | const struct ubuf_info *ubuf) | |
422 | { | |
e9ce7cb6 | 423 | struct xenvif_queue *foreign_queue = ubuf_to_queue(ubuf); |
58375744 ZK |
424 | |
425 | do { | |
426 | u16 pending_idx = ubuf->desc; | |
427 | ||
428 | if (skb_shinfo(skb)->frags[i].page.p == | |
e9ce7cb6 | 429 | foreign_queue->mmap_pages[pending_idx]) |
58375744 ZK |
430 | break; |
431 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
432 | } while (ubuf); | |
433 | ||
434 | return ubuf; | |
435 | } | |
436 | ||
f942dc25 IC |
437 | /* |
438 | * Prepare an SKB to be transmitted to the frontend. | |
439 | * | |
440 | * This function is responsible for allocating grant operations, meta | |
441 | * structures, etc. | |
442 | * | |
443 | * It returns the number of meta structures consumed. The number of | |
444 | * ring slots used is always equal to the number of meta slots used | |
445 | * plus the number of GSO descriptors used. Currently, we use either | |
446 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
447 | * frontend-side LRO). | |
448 | */ | |
7376419a | 449 | static int xenvif_gop_skb(struct sk_buff *skb, |
e9ce7cb6 WL |
450 | struct netrx_pending_operations *npo, |
451 | struct xenvif_queue *queue) | |
f942dc25 IC |
452 | { |
453 | struct xenvif *vif = netdev_priv(skb->dev); | |
454 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
455 | int i; | |
456 | struct xen_netif_rx_request *req; | |
b3f980bd | 457 | struct xenvif_rx_meta *meta; |
f942dc25 | 458 | unsigned char *data; |
33bc801d | 459 | int head = 1; |
f942dc25 | 460 | int old_meta_prod; |
82cada22 | 461 | int gso_type; |
58375744 ZK |
462 | const struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg; |
463 | const struct ubuf_info *const head_ubuf = ubuf; | |
f942dc25 IC |
464 | |
465 | old_meta_prod = npo->meta_prod; | |
466 | ||
5bd07670 AL |
467 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
468 | if (skb_is_gso(skb)) { | |
469 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
470 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
471 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
472 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
473 | } |
474 | ||
f942dc25 | 475 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 476 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
e9ce7cb6 | 477 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 | 478 | meta = npo->meta + npo->meta_prod++; |
82cada22 | 479 | meta->gso_type = gso_type; |
5bd07670 | 480 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 IC |
481 | meta->size = 0; |
482 | meta->id = req->id; | |
483 | } | |
484 | ||
e9ce7cb6 | 485 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 IC |
486 | meta = npo->meta + npo->meta_prod++; |
487 | ||
82cada22 PD |
488 | if ((1 << gso_type) & vif->gso_mask) { |
489 | meta->gso_type = gso_type; | |
5bd07670 | 490 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
491 | } else { |
492 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 493 | meta->gso_size = 0; |
82cada22 | 494 | } |
f942dc25 IC |
495 | |
496 | meta->size = 0; | |
497 | meta->id = req->id; | |
498 | npo->copy_off = 0; | |
499 | npo->copy_gref = req->gref; | |
500 | ||
501 | data = skb->data; | |
502 | while (data < skb_tail_pointer(skb)) { | |
503 | unsigned int offset = offset_in_page(data); | |
504 | unsigned int len = PAGE_SIZE - offset; | |
505 | ||
506 | if (data + len > skb_tail_pointer(skb)) | |
507 | len = skb_tail_pointer(skb) - data; | |
508 | ||
e9ce7cb6 | 509 | xenvif_gop_frag_copy(queue, skb, npo, |
3e2234b3 ZK |
510 | virt_to_page(data), len, offset, &head, |
511 | NULL, | |
512 | 0); | |
f942dc25 IC |
513 | data += len; |
514 | } | |
515 | ||
516 | for (i = 0; i < nr_frags; i++) { | |
58375744 ZK |
517 | /* This variable also signals whether foreign_gref has a real |
518 | * value or not. | |
519 | */ | |
e9ce7cb6 | 520 | struct xenvif_queue *foreign_queue = NULL; |
58375744 ZK |
521 | grant_ref_t foreign_gref; |
522 | ||
523 | if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) && | |
524 | (ubuf->callback == &xenvif_zerocopy_callback)) { | |
525 | const struct ubuf_info *const startpoint = ubuf; | |
526 | ||
527 | /* Ideally ubuf points to the chain element which | |
528 | * belongs to this frag. Or if frags were removed from | |
529 | * the beginning, then shortly before it. | |
530 | */ | |
531 | ubuf = xenvif_find_gref(skb, i, ubuf); | |
532 | ||
533 | /* Try again from the beginning of the list, if we | |
534 | * haven't tried from there. This only makes sense in | |
535 | * the unlikely event of reordering the original frags. | |
536 | * For injected local pages it's an unnecessary second | |
537 | * run. | |
538 | */ | |
539 | if (unlikely(!ubuf) && startpoint != head_ubuf) | |
540 | ubuf = xenvif_find_gref(skb, i, head_ubuf); | |
541 | ||
542 | if (likely(ubuf)) { | |
543 | u16 pending_idx = ubuf->desc; | |
544 | ||
e9ce7cb6 WL |
545 | foreign_queue = ubuf_to_queue(ubuf); |
546 | foreign_gref = | |
547 | foreign_queue->pending_tx_info[pending_idx].req.gref; | |
58375744 ZK |
548 | /* Just a safety measure. If this was the last |
549 | * element on the list, the for loop will | |
550 | * iterate again if a local page were added to | |
551 | * the end. Using head_ubuf here prevents the | |
552 | * second search on the chain. Or the original | |
553 | * frags changed order, but that's less likely. | |
554 | * In any way, ubuf shouldn't be NULL. | |
555 | */ | |
556 | ubuf = ubuf->ctx ? | |
557 | (struct ubuf_info *) ubuf->ctx : | |
558 | head_ubuf; | |
559 | } else | |
560 | /* This frag was a local page, added to the | |
561 | * array after the skb left netback. | |
562 | */ | |
563 | ubuf = head_ubuf; | |
564 | } | |
e9ce7cb6 | 565 | xenvif_gop_frag_copy(queue, skb, npo, |
7376419a WL |
566 | skb_frag_page(&skb_shinfo(skb)->frags[i]), |
567 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
568 | skb_shinfo(skb)->frags[i].page_offset, | |
3e2234b3 | 569 | &head, |
e9ce7cb6 WL |
570 | foreign_queue, |
571 | foreign_queue ? foreign_gref : UINT_MAX); | |
f942dc25 IC |
572 | } |
573 | ||
574 | return npo->meta_prod - old_meta_prod; | |
575 | } | |
576 | ||
577 | /* | |
7376419a | 578 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
579 | * used to set up the operations on the top of |
580 | * netrx_pending_operations, which have since been done. Check that | |
581 | * they didn't give any errors and advance over them. | |
582 | */ | |
7376419a WL |
583 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
584 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
585 | { |
586 | struct gnttab_copy *copy_op; | |
587 | int status = XEN_NETIF_RSP_OKAY; | |
588 | int i; | |
589 | ||
590 | for (i = 0; i < nr_meta_slots; i++) { | |
591 | copy_op = npo->copy + npo->copy_cons++; | |
592 | if (copy_op->status != GNTST_okay) { | |
593 | netdev_dbg(vif->dev, | |
594 | "Bad status %d from copy to DOM%d.\n", | |
595 | copy_op->status, vif->domid); | |
596 | status = XEN_NETIF_RSP_ERROR; | |
597 | } | |
598 | } | |
599 | ||
600 | return status; | |
601 | } | |
602 | ||
e9ce7cb6 | 603 | static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status, |
7376419a WL |
604 | struct xenvif_rx_meta *meta, |
605 | int nr_meta_slots) | |
f942dc25 IC |
606 | { |
607 | int i; | |
608 | unsigned long offset; | |
609 | ||
610 | /* No fragments used */ | |
611 | if (nr_meta_slots <= 1) | |
612 | return; | |
613 | ||
614 | nr_meta_slots--; | |
615 | ||
616 | for (i = 0; i < nr_meta_slots; i++) { | |
617 | int flags; | |
618 | if (i == nr_meta_slots - 1) | |
619 | flags = 0; | |
620 | else | |
621 | flags = XEN_NETRXF_more_data; | |
622 | ||
623 | offset = 0; | |
e9ce7cb6 | 624 | make_rx_response(queue, meta[i].id, status, offset, |
f942dc25 IC |
625 | meta[i].size, flags); |
626 | } | |
627 | } | |
628 | ||
e9ce7cb6 | 629 | void xenvif_kick_thread(struct xenvif_queue *queue) |
b3f980bd | 630 | { |
e9ce7cb6 | 631 | wake_up(&queue->wq); |
b3f980bd WL |
632 | } |
633 | ||
e9ce7cb6 | 634 | static void xenvif_rx_action(struct xenvif_queue *queue) |
f942dc25 | 635 | { |
f942dc25 | 636 | s8 status; |
e1f00a69 | 637 | u16 flags; |
f942dc25 IC |
638 | struct xen_netif_rx_response *resp; |
639 | struct sk_buff_head rxq; | |
640 | struct sk_buff *skb; | |
641 | LIST_HEAD(notify); | |
642 | int ret; | |
f942dc25 | 643 | unsigned long offset; |
11b57f90 | 644 | bool need_to_notify = false; |
f942dc25 IC |
645 | |
646 | struct netrx_pending_operations npo = { | |
e9ce7cb6 WL |
647 | .copy = queue->grant_copy_op, |
648 | .meta = queue->meta, | |
f942dc25 IC |
649 | }; |
650 | ||
651 | skb_queue_head_init(&rxq); | |
652 | ||
f48da8b1 DV |
653 | while (xenvif_rx_ring_slots_available(queue, XEN_NETBK_RX_SLOTS_MAX) |
654 | && (skb = xenvif_rx_dequeue(queue)) != NULL) { | |
9ab9831b | 655 | RING_IDX max_slots_needed; |
1425c7a4 PD |
656 | RING_IDX old_req_cons; |
657 | RING_IDX ring_slots_used; | |
ca2f09f2 PD |
658 | int i; |
659 | ||
ecf08d2d DV |
660 | queue->last_rx_time = jiffies; |
661 | ||
ca2f09f2 PD |
662 | /* We need a cheap worse case estimate for the number of |
663 | * slots we'll use. | |
664 | */ | |
665 | ||
666 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
667 | skb_headlen(skb), | |
668 | PAGE_SIZE); | |
669 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
670 | unsigned int size; | |
a02eb473 PD |
671 | unsigned int offset; |
672 | ||
ca2f09f2 | 673 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
a02eb473 PD |
674 | offset = skb_shinfo(skb)->frags[i].page_offset; |
675 | ||
676 | /* For a worse-case estimate we need to factor in | |
677 | * the fragment page offset as this will affect the | |
678 | * number of times xenvif_gop_frag_copy() will | |
679 | * call start_new_rx_buffer(). | |
680 | */ | |
681 | max_slots_needed += DIV_ROUND_UP(offset + size, | |
682 | PAGE_SIZE); | |
ca2f09f2 | 683 | } |
a02eb473 PD |
684 | |
685 | /* To avoid the estimate becoming too pessimal for some | |
686 | * frontends that limit posted rx requests, cap the estimate | |
59ae9fc6 ZK |
687 | * at MAX_SKB_FRAGS. In this case netback will fully coalesce |
688 | * the skb into the provided slots. | |
a02eb473 | 689 | */ |
59ae9fc6 | 690 | if (max_slots_needed > MAX_SKB_FRAGS) { |
a02eb473 | 691 | max_slots_needed = MAX_SKB_FRAGS; |
59ae9fc6 ZK |
692 | XENVIF_RX_CB(skb)->full_coalesce = true; |
693 | } else { | |
694 | XENVIF_RX_CB(skb)->full_coalesce = false; | |
695 | } | |
a02eb473 PD |
696 | |
697 | /* We may need one more slot for GSO metadata */ | |
5bd07670 AL |
698 | if (skb_is_gso(skb) && |
699 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
700 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | |
ca2f09f2 PD |
701 | max_slots_needed++; |
702 | ||
e9ce7cb6 WL |
703 | old_req_cons = queue->rx.req_cons; |
704 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue); | |
705 | ring_slots_used = queue->rx.req_cons - old_req_cons; | |
1425c7a4 PD |
706 | |
707 | BUG_ON(ring_slots_used > max_slots_needed); | |
f942dc25 IC |
708 | |
709 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
710 | } |
711 | ||
e9ce7cb6 | 712 | BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta)); |
f942dc25 IC |
713 | |
714 | if (!npo.copy_prod) | |
ca2f09f2 | 715 | goto done; |
f942dc25 | 716 | |
ac3d5ac2 | 717 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
e9ce7cb6 | 718 | gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
719 | |
720 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 721 | |
e9ce7cb6 WL |
722 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
723 | queue->vif->gso_prefix_mask) { | |
724 | resp = RING_GET_RESPONSE(&queue->rx, | |
725 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
726 | |
727 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
728 | ||
e9ce7cb6 WL |
729 | resp->offset = queue->meta[npo.meta_cons].gso_size; |
730 | resp->id = queue->meta[npo.meta_cons].id; | |
8f13dd96 | 731 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
732 | |
733 | npo.meta_cons++; | |
8f13dd96 | 734 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
735 | } |
736 | ||
737 | ||
e9ce7cb6 WL |
738 | queue->stats.tx_bytes += skb->len; |
739 | queue->stats.tx_packets++; | |
f942dc25 | 740 | |
e9ce7cb6 | 741 | status = xenvif_check_gop(queue->vif, |
8f13dd96 ZK |
742 | XENVIF_RX_CB(skb)->meta_slots_used, |
743 | &npo); | |
f942dc25 | 744 | |
8f13dd96 | 745 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
746 | flags = 0; |
747 | else | |
748 | flags = XEN_NETRXF_more_data; | |
749 | ||
750 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
751 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
752 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
753 | /* remote but checksummed. */ | |
754 | flags |= XEN_NETRXF_data_validated; | |
755 | ||
756 | offset = 0; | |
e9ce7cb6 | 757 | resp = make_rx_response(queue, queue->meta[npo.meta_cons].id, |
f942dc25 | 758 | status, offset, |
e9ce7cb6 | 759 | queue->meta[npo.meta_cons].size, |
f942dc25 IC |
760 | flags); |
761 | ||
e9ce7cb6 WL |
762 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
763 | queue->vif->gso_mask) { | |
f942dc25 IC |
764 | struct xen_netif_extra_info *gso = |
765 | (struct xen_netif_extra_info *) | |
e9ce7cb6 WL |
766 | RING_GET_RESPONSE(&queue->rx, |
767 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
768 | |
769 | resp->flags |= XEN_NETRXF_extra_info; | |
770 | ||
e9ce7cb6 WL |
771 | gso->u.gso.type = queue->meta[npo.meta_cons].gso_type; |
772 | gso->u.gso.size = queue->meta[npo.meta_cons].gso_size; | |
f942dc25 IC |
773 | gso->u.gso.pad = 0; |
774 | gso->u.gso.features = 0; | |
775 | ||
776 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
777 | gso->flags = 0; | |
778 | } | |
779 | ||
e9ce7cb6 WL |
780 | xenvif_add_frag_responses(queue, status, |
781 | queue->meta + npo.meta_cons + 1, | |
8f13dd96 | 782 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 | 783 | |
e9ce7cb6 | 784 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret); |
f942dc25 | 785 | |
11b57f90 | 786 | need_to_notify |= !!ret; |
b3f980bd | 787 | |
8f13dd96 | 788 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
789 | dev_kfree_skb(skb); |
790 | } | |
791 | ||
ca2f09f2 | 792 | done: |
b3f980bd | 793 | if (need_to_notify) |
e9ce7cb6 | 794 | notify_remote_via_irq(queue->rx_irq); |
f942dc25 IC |
795 | } |
796 | ||
e9ce7cb6 | 797 | void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue) |
f942dc25 IC |
798 | { |
799 | int more_to_do; | |
800 | ||
e9ce7cb6 | 801 | RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do); |
f942dc25 IC |
802 | |
803 | if (more_to_do) | |
e9ce7cb6 | 804 | napi_schedule(&queue->napi); |
f942dc25 IC |
805 | } |
806 | ||
e9ce7cb6 | 807 | static void tx_add_credit(struct xenvif_queue *queue) |
f942dc25 IC |
808 | { |
809 | unsigned long max_burst, max_credit; | |
810 | ||
811 | /* | |
812 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
813 | * Otherwise the interface can seize up due to insufficient credit. | |
814 | */ | |
e9ce7cb6 | 815 | max_burst = RING_GET_REQUEST(&queue->tx, queue->tx.req_cons)->size; |
f942dc25 | 816 | max_burst = min(max_burst, 131072UL); |
e9ce7cb6 | 817 | max_burst = max(max_burst, queue->credit_bytes); |
f942dc25 IC |
818 | |
819 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
e9ce7cb6 WL |
820 | max_credit = queue->remaining_credit + queue->credit_bytes; |
821 | if (max_credit < queue->remaining_credit) | |
f942dc25 IC |
822 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ |
823 | ||
e9ce7cb6 | 824 | queue->remaining_credit = min(max_credit, max_burst); |
f942dc25 IC |
825 | } |
826 | ||
827 | static void tx_credit_callback(unsigned long data) | |
828 | { | |
e9ce7cb6 WL |
829 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
830 | tx_add_credit(queue); | |
831 | xenvif_napi_schedule_or_enable_events(queue); | |
f942dc25 IC |
832 | } |
833 | ||
e9ce7cb6 | 834 | static void xenvif_tx_err(struct xenvif_queue *queue, |
7376419a | 835 | struct xen_netif_tx_request *txp, RING_IDX end) |
f942dc25 | 836 | { |
e9ce7cb6 | 837 | RING_IDX cons = queue->tx.req_cons; |
f53c3fe8 | 838 | unsigned long flags; |
f942dc25 IC |
839 | |
840 | do { | |
e9ce7cb6 WL |
841 | spin_lock_irqsave(&queue->response_lock, flags); |
842 | make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR); | |
843 | spin_unlock_irqrestore(&queue->response_lock, flags); | |
b9149729 | 844 | if (cons == end) |
f942dc25 | 845 | break; |
e9ce7cb6 | 846 | txp = RING_GET_REQUEST(&queue->tx, cons++); |
f942dc25 | 847 | } while (1); |
e9ce7cb6 | 848 | queue->tx.req_cons = cons; |
f942dc25 IC |
849 | } |
850 | ||
7376419a | 851 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
852 | { |
853 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 | 854 | vif->disabled = true; |
e9ce7cb6 WL |
855 | /* Disable the vif from queue 0's kthread */ |
856 | if (vif->queues) | |
857 | xenvif_kick_thread(&vif->queues[0]); | |
48856286 IC |
858 | } |
859 | ||
e9ce7cb6 | 860 | static int xenvif_count_requests(struct xenvif_queue *queue, |
7376419a WL |
861 | struct xen_netif_tx_request *first, |
862 | struct xen_netif_tx_request *txp, | |
863 | int work_to_do) | |
f942dc25 | 864 | { |
e9ce7cb6 | 865 | RING_IDX cons = queue->tx.req_cons; |
2810e5b9 WL |
866 | int slots = 0; |
867 | int drop_err = 0; | |
59ccb4eb | 868 | int more_data; |
f942dc25 IC |
869 | |
870 | if (!(first->flags & XEN_NETTXF_more_data)) | |
871 | return 0; | |
872 | ||
873 | do { | |
59ccb4eb WL |
874 | struct xen_netif_tx_request dropped_tx = { 0 }; |
875 | ||
2810e5b9 | 876 | if (slots >= work_to_do) { |
e9ce7cb6 | 877 | netdev_err(queue->vif->dev, |
2810e5b9 WL |
878 | "Asked for %d slots but exceeds this limit\n", |
879 | work_to_do); | |
e9ce7cb6 | 880 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 881 | return -ENODATA; |
f942dc25 IC |
882 | } |
883 | ||
2810e5b9 WL |
884 | /* This guest is really using too many slots and |
885 | * considered malicious. | |
886 | */ | |
37641494 | 887 | if (unlikely(slots >= fatal_skb_slots)) { |
e9ce7cb6 | 888 | netdev_err(queue->vif->dev, |
2810e5b9 | 889 | "Malicious frontend using %d slots, threshold %u\n", |
37641494 | 890 | slots, fatal_skb_slots); |
e9ce7cb6 | 891 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 892 | return -E2BIG; |
f942dc25 IC |
893 | } |
894 | ||
2810e5b9 | 895 | /* Xen network protocol had implicit dependency on |
37641494 WL |
896 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
897 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
898 | * same behavior as before. Any packet using more than | |
899 | * 18 slots but less than fatal_skb_slots slots is | |
900 | * dropped | |
2810e5b9 | 901 | */ |
37641494 | 902 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 | 903 | if (net_ratelimit()) |
e9ce7cb6 | 904 | netdev_dbg(queue->vif->dev, |
2810e5b9 | 905 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", |
37641494 | 906 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
907 | drop_err = -E2BIG; |
908 | } | |
909 | ||
59ccb4eb WL |
910 | if (drop_err) |
911 | txp = &dropped_tx; | |
912 | ||
e9ce7cb6 | 913 | memcpy(txp, RING_GET_REQUEST(&queue->tx, cons + slots), |
f942dc25 | 914 | sizeof(*txp)); |
03393fd5 WL |
915 | |
916 | /* If the guest submitted a frame >= 64 KiB then | |
917 | * first->size overflowed and following slots will | |
918 | * appear to be larger than the frame. | |
919 | * | |
920 | * This cannot be fatal error as there are buggy | |
921 | * frontends that do this. | |
922 | * | |
923 | * Consume all slots and drop the packet. | |
924 | */ | |
925 | if (!drop_err && txp->size > first->size) { | |
926 | if (net_ratelimit()) | |
e9ce7cb6 | 927 | netdev_dbg(queue->vif->dev, |
03393fd5 WL |
928 | "Invalid tx request, slot size %u > remaining size %u\n", |
929 | txp->size, first->size); | |
930 | drop_err = -EIO; | |
f942dc25 IC |
931 | } |
932 | ||
933 | first->size -= txp->size; | |
2810e5b9 | 934 | slots++; |
f942dc25 IC |
935 | |
936 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
e9ce7cb6 | 937 | netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 938 | txp->offset, txp->size); |
e9ce7cb6 | 939 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 940 | return -EINVAL; |
f942dc25 | 941 | } |
59ccb4eb WL |
942 | |
943 | more_data = txp->flags & XEN_NETTXF_more_data; | |
944 | ||
945 | if (!drop_err) | |
946 | txp++; | |
947 | ||
948 | } while (more_data); | |
2810e5b9 WL |
949 | |
950 | if (drop_err) { | |
e9ce7cb6 | 951 | xenvif_tx_err(queue, first, cons + slots); |
2810e5b9 WL |
952 | return drop_err; |
953 | } | |
954 | ||
955 | return slots; | |
f942dc25 IC |
956 | } |
957 | ||
8f13dd96 ZK |
958 | |
959 | struct xenvif_tx_cb { | |
960 | u16 pending_idx; | |
961 | }; | |
962 | ||
963 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
964 | ||
e9ce7cb6 | 965 | static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue, |
9074ce24 ZK |
966 | u16 pending_idx, |
967 | struct xen_netif_tx_request *txp, | |
968 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 969 | { |
e9ce7cb6 WL |
970 | queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx]; |
971 | gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx), | |
f53c3fe8 | 972 | GNTMAP_host_map | GNTMAP_readonly, |
e9ce7cb6 | 973 | txp->gref, queue->vif->domid); |
f53c3fe8 | 974 | |
e9ce7cb6 | 975 | memcpy(&queue->pending_tx_info[pending_idx].req, txp, |
f53c3fe8 ZK |
976 | sizeof(*txp)); |
977 | } | |
978 | ||
e3377f36 ZK |
979 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
980 | { | |
981 | struct sk_buff *skb = | |
982 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
983 | GFP_ATOMIC | __GFP_NOWARN); | |
984 | if (unlikely(skb == NULL)) | |
985 | return NULL; | |
986 | ||
987 | /* Packets passed to netif_rx() must have some headroom. */ | |
988 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
989 | ||
990 | /* Initialize it here to avoid later surprises */ | |
991 | skb_shinfo(skb)->destructor_arg = NULL; | |
992 | ||
993 | return skb; | |
994 | } | |
995 | ||
e9ce7cb6 | 996 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue, |
f53c3fe8 ZK |
997 | struct sk_buff *skb, |
998 | struct xen_netif_tx_request *txp, | |
999 | struct gnttab_map_grant_ref *gop) | |
f942dc25 IC |
1000 | { |
1001 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1002 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 1003 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
1004 | int start; |
1005 | pending_ring_idx_t index; | |
e3377f36 | 1006 | unsigned int nr_slots, frag_overflow = 0; |
2810e5b9 WL |
1007 | |
1008 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 1009 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 | 1010 | */ |
e3377f36 ZK |
1011 | if (shinfo->nr_frags > MAX_SKB_FRAGS) { |
1012 | frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS; | |
1013 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
1014 | shinfo->nr_frags = MAX_SKB_FRAGS; | |
1015 | } | |
2810e5b9 | 1016 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
1017 | |
1018 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 1019 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 1020 | |
f53c3fe8 ZK |
1021 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
1022 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
1023 | index = pending_index(queue->pending_cons++); |
1024 | pending_idx = queue->pending_ring[index]; | |
1025 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
f53c3fe8 | 1026 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
1027 | } |
1028 | ||
e3377f36 ZK |
1029 | if (frag_overflow) { |
1030 | struct sk_buff *nskb = xenvif_alloc_skb(0); | |
1031 | if (unlikely(nskb == NULL)) { | |
1032 | if (net_ratelimit()) | |
e9ce7cb6 | 1033 | netdev_err(queue->vif->dev, |
e3377f36 ZK |
1034 | "Can't allocate the frag_list skb.\n"); |
1035 | return NULL; | |
1036 | } | |
1037 | ||
1038 | shinfo = skb_shinfo(nskb); | |
1039 | frags = shinfo->frags; | |
1040 | ||
1041 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
1042 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
1043 | index = pending_index(queue->pending_cons++); |
1044 | pending_idx = queue->pending_ring[index]; | |
1045 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
e3377f36 ZK |
1046 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
1047 | pending_idx); | |
1048 | } | |
1049 | ||
1050 | skb_shinfo(skb)->frag_list = nskb; | |
1051 | } | |
2810e5b9 | 1052 | |
f942dc25 IC |
1053 | return gop; |
1054 | } | |
1055 | ||
e9ce7cb6 | 1056 | static inline void xenvif_grant_handle_set(struct xenvif_queue *queue, |
f53c3fe8 ZK |
1057 | u16 pending_idx, |
1058 | grant_handle_t handle) | |
1059 | { | |
e9ce7cb6 | 1060 | if (unlikely(queue->grant_tx_handle[pending_idx] != |
f53c3fe8 | 1061 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 1062 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1063 | "Trying to overwrite active handle! pending_idx: %x\n", |
1064 | pending_idx); | |
1065 | BUG(); | |
1066 | } | |
e9ce7cb6 | 1067 | queue->grant_tx_handle[pending_idx] = handle; |
f53c3fe8 ZK |
1068 | } |
1069 | ||
e9ce7cb6 | 1070 | static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue, |
f53c3fe8 ZK |
1071 | u16 pending_idx) |
1072 | { | |
e9ce7cb6 | 1073 | if (unlikely(queue->grant_tx_handle[pending_idx] == |
f53c3fe8 | 1074 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 1075 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1076 | "Trying to unmap invalid handle! pending_idx: %x\n", |
1077 | pending_idx); | |
1078 | BUG(); | |
1079 | } | |
e9ce7cb6 | 1080 | queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; |
f53c3fe8 ZK |
1081 | } |
1082 | ||
e9ce7cb6 | 1083 | static int xenvif_tx_check_gop(struct xenvif_queue *queue, |
7376419a | 1084 | struct sk_buff *skb, |
bdab8275 ZK |
1085 | struct gnttab_map_grant_ref **gopp_map, |
1086 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 1087 | { |
9074ce24 | 1088 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 1089 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
1a998d3e ZK |
1090 | /* This always points to the shinfo of the skb being checked, which |
1091 | * could be either the first or the one on the frag_list | |
1092 | */ | |
f942dc25 | 1093 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1a998d3e ZK |
1094 | /* If this is non-NULL, we are currently checking the frag_list skb, and |
1095 | * this points to the shinfo of the first one | |
1096 | */ | |
1097 | struct skb_shared_info *first_shinfo = NULL; | |
f942dc25 | 1098 | int nr_frags = shinfo->nr_frags; |
1b860da0 ZK |
1099 | const bool sharedslot = nr_frags && |
1100 | frag_get_pending_idx(&shinfo->frags[0]) == pending_idx; | |
bdab8275 | 1101 | int i, err; |
f942dc25 IC |
1102 | |
1103 | /* Check status of header. */ | |
bdab8275 | 1104 | err = (*gopp_copy)->status; |
bdab8275 ZK |
1105 | if (unlikely(err)) { |
1106 | if (net_ratelimit()) | |
e9ce7cb6 | 1107 | netdev_dbg(queue->vif->dev, |
00aefceb | 1108 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1109 | (*gopp_copy)->status, |
1110 | pending_idx, | |
1111 | (*gopp_copy)->source.u.ref); | |
1b860da0 ZK |
1112 | /* The first frag might still have this slot mapped */ |
1113 | if (!sharedslot) | |
1114 | xenvif_idx_release(queue, pending_idx, | |
1115 | XEN_NETIF_RSP_ERROR); | |
bdab8275 | 1116 | } |
d8cfbfc4 | 1117 | (*gopp_copy)++; |
f942dc25 | 1118 | |
e3377f36 | 1119 | check_frags: |
bdab8275 | 1120 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 1121 | int j, newerr; |
f942dc25 | 1122 | |
ea066ad1 | 1123 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
1124 | |
1125 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 1126 | newerr = gop_map->status; |
2810e5b9 | 1127 | |
f942dc25 | 1128 | if (likely(!newerr)) { |
e9ce7cb6 | 1129 | xenvif_grant_handle_set(queue, |
9074ce24 ZK |
1130 | pending_idx, |
1131 | gop_map->handle); | |
f942dc25 | 1132 | /* Had a previous error? Invalidate this fragment. */ |
1b860da0 | 1133 | if (unlikely(err)) { |
e9ce7cb6 | 1134 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
1135 | /* If the mapping of the first frag was OK, but |
1136 | * the header's copy failed, and they are | |
1137 | * sharing a slot, send an error | |
1138 | */ | |
1139 | if (i == 0 && sharedslot) | |
1140 | xenvif_idx_release(queue, pending_idx, | |
1141 | XEN_NETIF_RSP_ERROR); | |
1142 | else | |
1143 | xenvif_idx_release(queue, pending_idx, | |
1144 | XEN_NETIF_RSP_OKAY); | |
1145 | } | |
f942dc25 IC |
1146 | continue; |
1147 | } | |
1148 | ||
1149 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 | 1150 | if (net_ratelimit()) |
e9ce7cb6 | 1151 | netdev_dbg(queue->vif->dev, |
00aefceb | 1152 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1153 | i, |
1154 | gop_map->status, | |
1155 | pending_idx, | |
1156 | gop_map->ref); | |
1b860da0 | 1157 | |
e9ce7cb6 | 1158 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1159 | |
1160 | /* Not the first error? Preceding frags already invalidated. */ | |
1161 | if (err) | |
1162 | continue; | |
1b860da0 ZK |
1163 | |
1164 | /* First error: if the header haven't shared a slot with the | |
1165 | * first frag, release it as well. | |
1166 | */ | |
1167 | if (!sharedslot) | |
1168 | xenvif_idx_release(queue, | |
1169 | XENVIF_TX_CB(skb)->pending_idx, | |
1170 | XEN_NETIF_RSP_OKAY); | |
1171 | ||
1172 | /* Invalidate preceding fragments of this skb. */ | |
bdab8275 | 1173 | for (j = 0; j < i; j++) { |
5ccb3ea7 | 1174 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 1175 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
1176 | xenvif_idx_release(queue, pending_idx, |
1177 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1178 | } |
1179 | ||
1a998d3e ZK |
1180 | /* And if we found the error while checking the frag_list, unmap |
1181 | * the first skb's frags | |
1182 | */ | |
1183 | if (first_shinfo) { | |
1184 | for (j = 0; j < first_shinfo->nr_frags; j++) { | |
1185 | pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]); | |
1186 | xenvif_idx_unmap(queue, pending_idx); | |
1b860da0 ZK |
1187 | xenvif_idx_release(queue, pending_idx, |
1188 | XEN_NETIF_RSP_OKAY); | |
1a998d3e | 1189 | } |
f942dc25 IC |
1190 | } |
1191 | ||
1192 | /* Remember the error: invalidate all subsequent fragments. */ | |
1193 | err = newerr; | |
1194 | } | |
1195 | ||
1a998d3e ZK |
1196 | if (skb_has_frag_list(skb) && !first_shinfo) { |
1197 | first_shinfo = skb_shinfo(skb); | |
1198 | shinfo = skb_shinfo(skb_shinfo(skb)->frag_list); | |
e3377f36 | 1199 | nr_frags = shinfo->nr_frags; |
e3377f36 ZK |
1200 | |
1201 | goto check_frags; | |
1202 | } | |
1203 | ||
bdab8275 | 1204 | *gopp_map = gop_map; |
f942dc25 IC |
1205 | return err; |
1206 | } | |
1207 | ||
e9ce7cb6 | 1208 | static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb) |
f942dc25 IC |
1209 | { |
1210 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1211 | int nr_frags = shinfo->nr_frags; | |
1212 | int i; | |
f53c3fe8 ZK |
1213 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1214 | ||
f942dc25 IC |
1215 | for (i = 0; i < nr_frags; i++) { |
1216 | skb_frag_t *frag = shinfo->frags + i; | |
1217 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1218 | struct page *page; |
1219 | u16 pending_idx; | |
f942dc25 | 1220 | |
ea066ad1 | 1221 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1222 | |
f53c3fe8 | 1223 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 1224 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 1225 | skb_shinfo(skb)->destructor_arg = |
e9ce7cb6 | 1226 | &callback_param(queue, pending_idx); |
bdab8275 | 1227 | else |
e9ce7cb6 WL |
1228 | callback_param(queue, prev_pending_idx).ctx = |
1229 | &callback_param(queue, pending_idx); | |
f53c3fe8 | 1230 | |
e9ce7cb6 | 1231 | callback_param(queue, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1232 | prev_pending_idx = pending_idx; |
1233 | ||
e9ce7cb6 WL |
1234 | txp = &queue->pending_tx_info[pending_idx].req; |
1235 | page = virt_to_page(idx_to_kaddr(queue, pending_idx)); | |
ea066ad1 | 1236 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1237 | skb->len += txp->size; |
1238 | skb->data_len += txp->size; | |
1239 | skb->truesize += txp->size; | |
1240 | ||
f53c3fe8 | 1241 | /* Take an extra reference to offset network stack's put_page */ |
e9ce7cb6 | 1242 | get_page(queue->mmap_pages[pending_idx]); |
f942dc25 | 1243 | } |
f53c3fe8 ZK |
1244 | /* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc |
1245 | * overlaps with "index", and "mapping" is not set. I think mapping | |
1246 | * should be set. If delivered to local stack, it would drop this | |
1247 | * skb in sk_filter unless the socket has the right to use it. | |
1248 | */ | |
1249 | skb->pfmemalloc = false; | |
f942dc25 IC |
1250 | } |
1251 | ||
e9ce7cb6 | 1252 | static int xenvif_get_extras(struct xenvif_queue *queue, |
f942dc25 IC |
1253 | struct xen_netif_extra_info *extras, |
1254 | int work_to_do) | |
1255 | { | |
1256 | struct xen_netif_extra_info extra; | |
e9ce7cb6 | 1257 | RING_IDX cons = queue->tx.req_cons; |
f942dc25 IC |
1258 | |
1259 | do { | |
1260 | if (unlikely(work_to_do-- <= 0)) { | |
e9ce7cb6 WL |
1261 | netdev_err(queue->vif->dev, "Missing extra info\n"); |
1262 | xenvif_fatal_tx_err(queue->vif); | |
f942dc25 IC |
1263 | return -EBADR; |
1264 | } | |
1265 | ||
e9ce7cb6 | 1266 | memcpy(&extra, RING_GET_REQUEST(&queue->tx, cons), |
f942dc25 IC |
1267 | sizeof(extra)); |
1268 | if (unlikely(!extra.type || | |
1269 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
e9ce7cb6 WL |
1270 | queue->tx.req_cons = ++cons; |
1271 | netdev_err(queue->vif->dev, | |
f942dc25 | 1272 | "Invalid extra type: %d\n", extra.type); |
e9ce7cb6 | 1273 | xenvif_fatal_tx_err(queue->vif); |
f942dc25 IC |
1274 | return -EINVAL; |
1275 | } | |
1276 | ||
1277 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
e9ce7cb6 | 1278 | queue->tx.req_cons = ++cons; |
f942dc25 IC |
1279 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
1280 | ||
1281 | return work_to_do; | |
1282 | } | |
1283 | ||
7376419a WL |
1284 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1285 | struct sk_buff *skb, | |
1286 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1287 | { |
1288 | if (!gso->u.gso.size) { | |
48856286 | 1289 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1290 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1291 | return -EINVAL; |
1292 | } | |
1293 | ||
a9468587 PD |
1294 | switch (gso->u.gso.type) { |
1295 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1296 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1297 | break; | |
1298 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1299 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1300 | break; | |
1301 | default: | |
48856286 | 1302 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1303 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1304 | return -EINVAL; |
1305 | } | |
1306 | ||
1307 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1308 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1309 | |
1310 | return 0; | |
1311 | } | |
1312 | ||
e9ce7cb6 | 1313 | static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb) |
2eba61d5 | 1314 | { |
2721637c | 1315 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1316 | |
1317 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1318 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1319 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1320 | * recalculate the partial checksum. | |
1321 | */ | |
1322 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
e9ce7cb6 | 1323 | queue->stats.rx_gso_checksum_fixup++; |
2eba61d5 | 1324 | skb->ip_summed = CHECKSUM_PARTIAL; |
2721637c | 1325 | recalculate_partial_csum = true; |
2eba61d5 PD |
1326 | } |
1327 | ||
1328 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1329 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1330 | return 0; | |
1331 | ||
2721637c | 1332 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1333 | } |
1334 | ||
e9ce7cb6 | 1335 | static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size) |
f942dc25 | 1336 | { |
059dfa6a | 1337 | u64 now = get_jiffies_64(); |
e9ce7cb6 WL |
1338 | u64 next_credit = queue->credit_window_start + |
1339 | msecs_to_jiffies(queue->credit_usec / 1000); | |
f942dc25 IC |
1340 | |
1341 | /* Timer could already be pending in rare cases. */ | |
e9ce7cb6 | 1342 | if (timer_pending(&queue->credit_timeout)) |
f942dc25 IC |
1343 | return true; |
1344 | ||
1345 | /* Passed the point where we can replenish credit? */ | |
059dfa6a | 1346 | if (time_after_eq64(now, next_credit)) { |
e9ce7cb6 WL |
1347 | queue->credit_window_start = now; |
1348 | tx_add_credit(queue); | |
f942dc25 IC |
1349 | } |
1350 | ||
1351 | /* Still too big to send right now? Set a callback. */ | |
e9ce7cb6 WL |
1352 | if (size > queue->remaining_credit) { |
1353 | queue->credit_timeout.data = | |
1354 | (unsigned long)queue; | |
1355 | queue->credit_timeout.function = | |
f942dc25 | 1356 | tx_credit_callback; |
e9ce7cb6 | 1357 | mod_timer(&queue->credit_timeout, |
f942dc25 | 1358 | next_credit); |
e9ce7cb6 | 1359 | queue->credit_window_start = next_credit; |
f942dc25 IC |
1360 | |
1361 | return true; | |
1362 | } | |
1363 | ||
1364 | return false; | |
1365 | } | |
1366 | ||
e9ce7cb6 | 1367 | static void xenvif_tx_build_gops(struct xenvif_queue *queue, |
bdab8275 ZK |
1368 | int budget, |
1369 | unsigned *copy_ops, | |
1370 | unsigned *map_ops) | |
f942dc25 | 1371 | { |
e9ce7cb6 | 1372 | struct gnttab_map_grant_ref *gop = queue->tx_map_ops, *request_gop; |
f942dc25 IC |
1373 | struct sk_buff *skb; |
1374 | int ret; | |
1375 | ||
e9ce7cb6 | 1376 | while (skb_queue_len(&queue->tx_queue) < budget) { |
f942dc25 | 1377 | struct xen_netif_tx_request txreq; |
37641494 | 1378 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1379 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1380 | u16 pending_idx; | |
1381 | RING_IDX idx; | |
1382 | int work_to_do; | |
1383 | unsigned int data_len; | |
1384 | pending_ring_idx_t index; | |
1385 | ||
e9ce7cb6 | 1386 | if (queue->tx.sring->req_prod - queue->tx.req_cons > |
48856286 | 1387 | XEN_NETIF_TX_RING_SIZE) { |
e9ce7cb6 | 1388 | netdev_err(queue->vif->dev, |
48856286 IC |
1389 | "Impossible number of requests. " |
1390 | "req_prod %d, req_cons %d, size %ld\n", | |
e9ce7cb6 | 1391 | queue->tx.sring->req_prod, queue->tx.req_cons, |
48856286 | 1392 | XEN_NETIF_TX_RING_SIZE); |
e9ce7cb6 | 1393 | xenvif_fatal_tx_err(queue->vif); |
e9d8b2c2 | 1394 | break; |
48856286 IC |
1395 | } |
1396 | ||
e9ce7cb6 | 1397 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx); |
b3f980bd WL |
1398 | if (!work_to_do) |
1399 | break; | |
f942dc25 | 1400 | |
e9ce7cb6 | 1401 | idx = queue->tx.req_cons; |
f942dc25 | 1402 | rmb(); /* Ensure that we see the request before we copy it. */ |
e9ce7cb6 | 1403 | memcpy(&txreq, RING_GET_REQUEST(&queue->tx, idx), sizeof(txreq)); |
f942dc25 IC |
1404 | |
1405 | /* Credit-based scheduling. */ | |
e9ce7cb6 WL |
1406 | if (txreq.size > queue->remaining_credit && |
1407 | tx_credit_exceeded(queue, txreq.size)) | |
b3f980bd | 1408 | break; |
f942dc25 | 1409 | |
e9ce7cb6 | 1410 | queue->remaining_credit -= txreq.size; |
f942dc25 IC |
1411 | |
1412 | work_to_do--; | |
e9ce7cb6 | 1413 | queue->tx.req_cons = ++idx; |
f942dc25 IC |
1414 | |
1415 | memset(extras, 0, sizeof(extras)); | |
1416 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
e9ce7cb6 | 1417 | work_to_do = xenvif_get_extras(queue, extras, |
7376419a | 1418 | work_to_do); |
e9ce7cb6 | 1419 | idx = queue->tx.req_cons; |
48856286 | 1420 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1421 | break; |
f942dc25 IC |
1422 | } |
1423 | ||
e9ce7cb6 | 1424 | ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do); |
48856286 | 1425 | if (unlikely(ret < 0)) |
b3f980bd | 1426 | break; |
48856286 | 1427 | |
f942dc25 IC |
1428 | idx += ret; |
1429 | ||
1430 | if (unlikely(txreq.size < ETH_HLEN)) { | |
e9ce7cb6 | 1431 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1432 | "Bad packet size: %d\n", txreq.size); |
e9ce7cb6 | 1433 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1434 | break; |
f942dc25 IC |
1435 | } |
1436 | ||
1437 | /* No crossing a page as the payload mustn't fragment. */ | |
1438 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
e9ce7cb6 | 1439 | netdev_err(queue->vif->dev, |
f942dc25 IC |
1440 | "txreq.offset: %x, size: %u, end: %lu\n", |
1441 | txreq.offset, txreq.size, | |
1442 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
e9ce7cb6 | 1443 | xenvif_fatal_tx_err(queue->vif); |
b3f980bd | 1444 | break; |
f942dc25 IC |
1445 | } |
1446 | ||
e9ce7cb6 WL |
1447 | index = pending_index(queue->pending_cons); |
1448 | pending_idx = queue->pending_ring[index]; | |
f942dc25 | 1449 | |
7e5d7753 | 1450 | data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN && |
37641494 | 1451 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
7e5d7753 | 1452 | XEN_NETBACK_TX_COPY_LEN : txreq.size; |
f942dc25 | 1453 | |
e3377f36 | 1454 | skb = xenvif_alloc_skb(data_len); |
f942dc25 | 1455 | if (unlikely(skb == NULL)) { |
e9ce7cb6 | 1456 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1457 | "Can't allocate a skb in start_xmit.\n"); |
e9ce7cb6 | 1458 | xenvif_tx_err(queue, &txreq, idx); |
f942dc25 IC |
1459 | break; |
1460 | } | |
1461 | ||
f942dc25 IC |
1462 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1463 | struct xen_netif_extra_info *gso; | |
1464 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1465 | ||
e9ce7cb6 | 1466 | if (xenvif_set_skb_gso(queue->vif, skb, gso)) { |
7376419a | 1467 | /* Failure in xenvif_set_skb_gso is fatal. */ |
f942dc25 | 1468 | kfree_skb(skb); |
b3f980bd | 1469 | break; |
f942dc25 IC |
1470 | } |
1471 | } | |
1472 | ||
8f13dd96 | 1473 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1474 | |
1475 | __skb_put(skb, data_len); | |
e9ce7cb6 WL |
1476 | queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
1477 | queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid; | |
1478 | queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
bdab8275 | 1479 | |
e9ce7cb6 | 1480 | queue->tx_copy_ops[*copy_ops].dest.u.gmfn = |
bdab8275 | 1481 | virt_to_mfn(skb->data); |
e9ce7cb6 WL |
1482 | queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; |
1483 | queue->tx_copy_ops[*copy_ops].dest.offset = | |
bdab8275 ZK |
1484 | offset_in_page(skb->data); |
1485 | ||
e9ce7cb6 WL |
1486 | queue->tx_copy_ops[*copy_ops].len = data_len; |
1487 | queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
bdab8275 ZK |
1488 | |
1489 | (*copy_ops)++; | |
f942dc25 IC |
1490 | |
1491 | skb_shinfo(skb)->nr_frags = ret; | |
1492 | if (data_len < txreq.size) { | |
1493 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1494 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1495 | pending_idx); | |
e9ce7cb6 | 1496 | xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop); |
bdab8275 | 1497 | gop++; |
f942dc25 | 1498 | } else { |
ea066ad1 IC |
1499 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1500 | INVALID_PENDING_IDX); | |
e9ce7cb6 | 1501 | memcpy(&queue->pending_tx_info[pending_idx].req, &txreq, |
bdab8275 | 1502 | sizeof(txreq)); |
f942dc25 IC |
1503 | } |
1504 | ||
e9ce7cb6 | 1505 | queue->pending_cons++; |
f942dc25 | 1506 | |
e9ce7cb6 | 1507 | request_gop = xenvif_get_requests(queue, skb, txfrags, gop); |
f942dc25 IC |
1508 | if (request_gop == NULL) { |
1509 | kfree_skb(skb); | |
e9ce7cb6 | 1510 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1511 | break; |
f942dc25 IC |
1512 | } |
1513 | gop = request_gop; | |
1514 | ||
e9ce7cb6 | 1515 | __skb_queue_tail(&queue->tx_queue, skb); |
1e0b6eac | 1516 | |
e9ce7cb6 | 1517 | queue->tx.req_cons = idx; |
f942dc25 | 1518 | |
e9ce7cb6 WL |
1519 | if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) || |
1520 | (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops))) | |
f942dc25 IC |
1521 | break; |
1522 | } | |
1523 | ||
e9ce7cb6 | 1524 | (*map_ops) = gop - queue->tx_map_ops; |
bdab8275 | 1525 | return; |
f942dc25 IC |
1526 | } |
1527 | ||
e3377f36 ZK |
1528 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1529 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1530 | */ | |
e9ce7cb6 | 1531 | static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb) |
e3377f36 ZK |
1532 | { |
1533 | unsigned int offset = skb_headlen(skb); | |
1534 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
1535 | int i; | |
1536 | struct ubuf_info *uarg; | |
1537 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1538 | ||
e9ce7cb6 WL |
1539 | queue->stats.tx_zerocopy_sent += 2; |
1540 | queue->stats.tx_frag_overflow++; | |
e3377f36 | 1541 | |
e9ce7cb6 | 1542 | xenvif_fill_frags(queue, nskb); |
e3377f36 ZK |
1543 | /* Subtract frags size, we will correct it later */ |
1544 | skb->truesize -= skb->data_len; | |
1545 | skb->len += nskb->len; | |
1546 | skb->data_len += nskb->len; | |
1547 | ||
1548 | /* create a brand new frags array and coalesce there */ | |
1549 | for (i = 0; offset < skb->len; i++) { | |
1550 | struct page *page; | |
1551 | unsigned int len; | |
1552 | ||
1553 | BUG_ON(i >= MAX_SKB_FRAGS); | |
44cc8ed1 | 1554 | page = alloc_page(GFP_ATOMIC); |
e3377f36 ZK |
1555 | if (!page) { |
1556 | int j; | |
1557 | skb->truesize += skb->data_len; | |
1558 | for (j = 0; j < i; j++) | |
1559 | put_page(frags[j].page.p); | |
1560 | return -ENOMEM; | |
1561 | } | |
1562 | ||
1563 | if (offset + PAGE_SIZE < skb->len) | |
1564 | len = PAGE_SIZE; | |
1565 | else | |
1566 | len = skb->len - offset; | |
1567 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1568 | BUG(); | |
1569 | ||
1570 | offset += len; | |
1571 | frags[i].page.p = page; | |
1572 | frags[i].page_offset = 0; | |
1573 | skb_frag_size_set(&frags[i], len); | |
1574 | } | |
1575 | /* swap out with old one */ | |
1576 | memcpy(skb_shinfo(skb)->frags, | |
1577 | frags, | |
1578 | i * sizeof(skb_frag_t)); | |
1579 | skb_shinfo(skb)->nr_frags = i; | |
1580 | skb->truesize += i * PAGE_SIZE; | |
1581 | ||
1582 | /* remove traces of mapped pages and frag_list */ | |
1583 | skb_frag_list_init(skb); | |
1584 | uarg = skb_shinfo(skb)->destructor_arg; | |
a64bd934 WL |
1585 | /* increase inflight counter to offset decrement in callback */ |
1586 | atomic_inc(&queue->inflight_packets); | |
e3377f36 ZK |
1587 | uarg->callback(uarg, true); |
1588 | skb_shinfo(skb)->destructor_arg = NULL; | |
1589 | ||
a64bd934 | 1590 | xenvif_skb_zerocopy_prepare(queue, nskb); |
e3377f36 ZK |
1591 | kfree_skb(nskb); |
1592 | ||
1593 | return 0; | |
1594 | } | |
b3f980bd | 1595 | |
e9ce7cb6 | 1596 | static int xenvif_tx_submit(struct xenvif_queue *queue) |
f942dc25 | 1597 | { |
e9ce7cb6 WL |
1598 | struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops; |
1599 | struct gnttab_copy *gop_copy = queue->tx_copy_ops; | |
f942dc25 | 1600 | struct sk_buff *skb; |
b3f980bd | 1601 | int work_done = 0; |
f942dc25 | 1602 | |
e9ce7cb6 | 1603 | while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) { |
f942dc25 | 1604 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1605 | u16 pending_idx; |
1606 | unsigned data_len; | |
1607 | ||
8f13dd96 | 1608 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
e9ce7cb6 | 1609 | txp = &queue->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1610 | |
1611 | /* Check the remap error code. */ | |
e9ce7cb6 | 1612 | if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) { |
b42cc6e4 ZK |
1613 | /* If there was an error, xenvif_tx_check_gop is |
1614 | * expected to release all the frags which were mapped, | |
1615 | * so kfree_skb shouldn't do it again | |
1616 | */ | |
f942dc25 | 1617 | skb_shinfo(skb)->nr_frags = 0; |
b42cc6e4 ZK |
1618 | if (skb_has_frag_list(skb)) { |
1619 | struct sk_buff *nskb = | |
1620 | skb_shinfo(skb)->frag_list; | |
1621 | skb_shinfo(nskb)->nr_frags = 0; | |
1622 | } | |
f942dc25 IC |
1623 | kfree_skb(skb); |
1624 | continue; | |
1625 | } | |
1626 | ||
1627 | data_len = skb->len; | |
e9ce7cb6 | 1628 | callback_param(queue, pending_idx).ctx = NULL; |
f942dc25 IC |
1629 | if (data_len < txp->size) { |
1630 | /* Append the packet payload as a fragment. */ | |
1631 | txp->offset += data_len; | |
1632 | txp->size -= data_len; | |
1633 | } else { | |
1634 | /* Schedule a response immediately. */ | |
e9ce7cb6 | 1635 | xenvif_idx_release(queue, pending_idx, |
bdab8275 | 1636 | XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1637 | } |
1638 | ||
1639 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1640 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1641 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1642 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1643 | ||
e9ce7cb6 | 1644 | xenvif_fill_frags(queue, skb); |
f942dc25 | 1645 | |
e3377f36 | 1646 | if (unlikely(skb_has_frag_list(skb))) { |
e9ce7cb6 | 1647 | if (xenvif_handle_frag_list(queue, skb)) { |
e3377f36 | 1648 | if (net_ratelimit()) |
e9ce7cb6 | 1649 | netdev_err(queue->vif->dev, |
e3377f36 | 1650 | "Not enough memory to consolidate frag_list!\n"); |
a64bd934 | 1651 | xenvif_skb_zerocopy_prepare(queue, skb); |
e3377f36 ZK |
1652 | kfree_skb(skb); |
1653 | continue; | |
1654 | } | |
1655 | } | |
1656 | ||
e9ce7cb6 | 1657 | skb->dev = queue->vif->dev; |
f942dc25 | 1658 | skb->protocol = eth_type_trans(skb, skb->dev); |
f9ca8f74 | 1659 | skb_reset_network_header(skb); |
f942dc25 | 1660 | |
e9ce7cb6 WL |
1661 | if (checksum_setup(queue, skb)) { |
1662 | netdev_dbg(queue->vif->dev, | |
f942dc25 | 1663 | "Can't setup checksum in net_tx_action\n"); |
f53c3fe8 ZK |
1664 | /* We have to set this flag to trigger the callback */ |
1665 | if (skb_shinfo(skb)->destructor_arg) | |
a64bd934 | 1666 | xenvif_skb_zerocopy_prepare(queue, skb); |
f942dc25 IC |
1667 | kfree_skb(skb); |
1668 | continue; | |
1669 | } | |
1670 | ||
40893fd0 | 1671 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1672 | |
b89587a7 PD |
1673 | /* If the packet is GSO then we will have just set up the |
1674 | * transport header offset in checksum_setup so it's now | |
1675 | * straightforward to calculate gso_segs. | |
1676 | */ | |
1677 | if (skb_is_gso(skb)) { | |
1678 | int mss = skb_shinfo(skb)->gso_size; | |
1679 | int hdrlen = skb_transport_header(skb) - | |
1680 | skb_mac_header(skb) + | |
1681 | tcp_hdrlen(skb); | |
1682 | ||
1683 | skb_shinfo(skb)->gso_segs = | |
1684 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1685 | } | |
1686 | ||
e9ce7cb6 WL |
1687 | queue->stats.rx_bytes += skb->len; |
1688 | queue->stats.rx_packets++; | |
f942dc25 | 1689 | |
b3f980bd WL |
1690 | work_done++; |
1691 | ||
f53c3fe8 ZK |
1692 | /* Set this flag right before netif_receive_skb, otherwise |
1693 | * someone might think this packet already left netback, and | |
1694 | * do a skb_copy_ubufs while we are still in control of the | |
1695 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1696 | */ | |
1bb332af | 1697 | if (skb_shinfo(skb)->destructor_arg) { |
a64bd934 | 1698 | xenvif_skb_zerocopy_prepare(queue, skb); |
e9ce7cb6 | 1699 | queue->stats.tx_zerocopy_sent++; |
1bb332af | 1700 | } |
f53c3fe8 | 1701 | |
b3f980bd | 1702 | netif_receive_skb(skb); |
f942dc25 | 1703 | } |
b3f980bd WL |
1704 | |
1705 | return work_done; | |
f942dc25 IC |
1706 | } |
1707 | ||
3e2234b3 ZK |
1708 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1709 | { | |
f53c3fe8 ZK |
1710 | unsigned long flags; |
1711 | pending_ring_idx_t index; | |
e9ce7cb6 | 1712 | struct xenvif_queue *queue = ubuf_to_queue(ubuf); |
f53c3fe8 ZK |
1713 | |
1714 | /* This is the only place where we grab this lock, to protect callbacks | |
1715 | * from each other. | |
1716 | */ | |
e9ce7cb6 | 1717 | spin_lock_irqsave(&queue->callback_lock, flags); |
f53c3fe8 ZK |
1718 | do { |
1719 | u16 pending_idx = ubuf->desc; | |
1720 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
e9ce7cb6 | 1721 | BUG_ON(queue->dealloc_prod - queue->dealloc_cons >= |
f53c3fe8 | 1722 | MAX_PENDING_REQS); |
e9ce7cb6 WL |
1723 | index = pending_index(queue->dealloc_prod); |
1724 | queue->dealloc_ring[index] = pending_idx; | |
f53c3fe8 ZK |
1725 | /* Sync with xenvif_tx_dealloc_action: |
1726 | * insert idx then incr producer. | |
1727 | */ | |
1728 | smp_wmb(); | |
e9ce7cb6 | 1729 | queue->dealloc_prod++; |
f53c3fe8 | 1730 | } while (ubuf); |
e9ce7cb6 WL |
1731 | wake_up(&queue->dealloc_wq); |
1732 | spin_unlock_irqrestore(&queue->callback_lock, flags); | |
f53c3fe8 | 1733 | |
1bb332af | 1734 | if (likely(zerocopy_success)) |
e9ce7cb6 | 1735 | queue->stats.tx_zerocopy_success++; |
1bb332af | 1736 | else |
e9ce7cb6 | 1737 | queue->stats.tx_zerocopy_fail++; |
a64bd934 | 1738 | xenvif_skb_zerocopy_complete(queue); |
f53c3fe8 ZK |
1739 | } |
1740 | ||
e9ce7cb6 | 1741 | static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue) |
f53c3fe8 ZK |
1742 | { |
1743 | struct gnttab_unmap_grant_ref *gop; | |
1744 | pending_ring_idx_t dc, dp; | |
1745 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1746 | unsigned int i = 0; | |
1747 | ||
e9ce7cb6 WL |
1748 | dc = queue->dealloc_cons; |
1749 | gop = queue->tx_unmap_ops; | |
f53c3fe8 ZK |
1750 | |
1751 | /* Free up any grants we have finished using */ | |
1752 | do { | |
e9ce7cb6 | 1753 | dp = queue->dealloc_prod; |
f53c3fe8 ZK |
1754 | |
1755 | /* Ensure we see all indices enqueued by all | |
1756 | * xenvif_zerocopy_callback(). | |
1757 | */ | |
1758 | smp_rmb(); | |
1759 | ||
1760 | while (dc != dp) { | |
e9ce7cb6 | 1761 | BUG_ON(gop - queue->tx_unmap_ops > MAX_PENDING_REQS); |
f53c3fe8 | 1762 | pending_idx = |
e9ce7cb6 | 1763 | queue->dealloc_ring[pending_index(dc++)]; |
f53c3fe8 | 1764 | |
e9ce7cb6 | 1765 | pending_idx_release[gop-queue->tx_unmap_ops] = |
f53c3fe8 | 1766 | pending_idx; |
e9ce7cb6 WL |
1767 | queue->pages_to_unmap[gop-queue->tx_unmap_ops] = |
1768 | queue->mmap_pages[pending_idx]; | |
f53c3fe8 | 1769 | gnttab_set_unmap_op(gop, |
e9ce7cb6 | 1770 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1771 | GNTMAP_host_map, |
e9ce7cb6 WL |
1772 | queue->grant_tx_handle[pending_idx]); |
1773 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1774 | ++gop; |
1775 | } | |
1776 | ||
e9ce7cb6 | 1777 | } while (dp != queue->dealloc_prod); |
f53c3fe8 | 1778 | |
e9ce7cb6 | 1779 | queue->dealloc_cons = dc; |
f53c3fe8 | 1780 | |
e9ce7cb6 | 1781 | if (gop - queue->tx_unmap_ops > 0) { |
f53c3fe8 | 1782 | int ret; |
e9ce7cb6 | 1783 | ret = gnttab_unmap_refs(queue->tx_unmap_ops, |
f53c3fe8 | 1784 | NULL, |
e9ce7cb6 WL |
1785 | queue->pages_to_unmap, |
1786 | gop - queue->tx_unmap_ops); | |
f53c3fe8 | 1787 | if (ret) { |
e9ce7cb6 WL |
1788 | netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tx ret %d\n", |
1789 | gop - queue->tx_unmap_ops, ret); | |
1790 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) { | |
f53c3fe8 | 1791 | if (gop[i].status != GNTST_okay) |
e9ce7cb6 | 1792 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1793 | " host_addr: %llx handle: %x status: %d\n", |
1794 | gop[i].host_addr, | |
1795 | gop[i].handle, | |
1796 | gop[i].status); | |
1797 | } | |
1798 | BUG(); | |
1799 | } | |
1800 | } | |
1801 | ||
e9ce7cb6 WL |
1802 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) |
1803 | xenvif_idx_release(queue, pending_idx_release[i], | |
f53c3fe8 | 1804 | XEN_NETIF_RSP_OKAY); |
3e2234b3 ZK |
1805 | } |
1806 | ||
f53c3fe8 | 1807 | |
f942dc25 | 1808 | /* Called after netfront has transmitted */ |
e9ce7cb6 | 1809 | int xenvif_tx_action(struct xenvif_queue *queue, int budget) |
f942dc25 | 1810 | { |
bdab8275 | 1811 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1812 | int work_done, ret; |
f942dc25 | 1813 | |
e9ce7cb6 | 1814 | if (unlikely(!tx_work_todo(queue))) |
b3f980bd WL |
1815 | return 0; |
1816 | ||
e9ce7cb6 | 1817 | xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops); |
f942dc25 | 1818 | |
bdab8275 | 1819 | if (nr_cops == 0) |
b3f980bd WL |
1820 | return 0; |
1821 | ||
e9ce7cb6 | 1822 | gnttab_batch_copy(queue->tx_copy_ops, nr_cops); |
bdab8275 | 1823 | if (nr_mops != 0) { |
e9ce7cb6 | 1824 | ret = gnttab_map_refs(queue->tx_map_ops, |
bdab8275 | 1825 | NULL, |
e9ce7cb6 | 1826 | queue->pages_to_map, |
bdab8275 ZK |
1827 | nr_mops); |
1828 | BUG_ON(ret); | |
1829 | } | |
f942dc25 | 1830 | |
e9ce7cb6 | 1831 | work_done = xenvif_tx_submit(queue); |
f942dc25 | 1832 | |
b3f980bd | 1833 | return work_done; |
f942dc25 IC |
1834 | } |
1835 | ||
e9ce7cb6 | 1836 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a | 1837 | u8 status) |
f942dc25 | 1838 | { |
f942dc25 | 1839 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1840 | pending_ring_idx_t index; |
f53c3fe8 | 1841 | unsigned long flags; |
2810e5b9 | 1842 | |
e9ce7cb6 WL |
1843 | pending_tx_info = &queue->pending_tx_info[pending_idx]; |
1844 | spin_lock_irqsave(&queue->response_lock, flags); | |
1845 | make_tx_response(queue, &pending_tx_info->req, status); | |
1846 | index = pending_index(queue->pending_prod); | |
1847 | queue->pending_ring[index] = pending_idx; | |
62bad319 ZK |
1848 | /* TX shouldn't use the index before we give it back here */ |
1849 | mb(); | |
e9ce7cb6 WL |
1850 | queue->pending_prod++; |
1851 | spin_unlock_irqrestore(&queue->response_lock, flags); | |
f942dc25 IC |
1852 | } |
1853 | ||
2810e5b9 | 1854 | |
e9ce7cb6 | 1855 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1856 | struct xen_netif_tx_request *txp, |
1857 | s8 st) | |
1858 | { | |
e9ce7cb6 | 1859 | RING_IDX i = queue->tx.rsp_prod_pvt; |
f942dc25 IC |
1860 | struct xen_netif_tx_response *resp; |
1861 | int notify; | |
1862 | ||
e9ce7cb6 | 1863 | resp = RING_GET_RESPONSE(&queue->tx, i); |
f942dc25 IC |
1864 | resp->id = txp->id; |
1865 | resp->status = st; | |
1866 | ||
1867 | if (txp->flags & XEN_NETTXF_extra_info) | |
e9ce7cb6 | 1868 | RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL; |
f942dc25 | 1869 | |
e9ce7cb6 WL |
1870 | queue->tx.rsp_prod_pvt = ++i; |
1871 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify); | |
f942dc25 | 1872 | if (notify) |
e9ce7cb6 | 1873 | notify_remote_via_irq(queue->tx_irq); |
f942dc25 IC |
1874 | } |
1875 | ||
e9ce7cb6 | 1876 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1877 | u16 id, |
1878 | s8 st, | |
1879 | u16 offset, | |
1880 | u16 size, | |
1881 | u16 flags) | |
1882 | { | |
e9ce7cb6 | 1883 | RING_IDX i = queue->rx.rsp_prod_pvt; |
f942dc25 IC |
1884 | struct xen_netif_rx_response *resp; |
1885 | ||
e9ce7cb6 | 1886 | resp = RING_GET_RESPONSE(&queue->rx, i); |
f942dc25 IC |
1887 | resp->offset = offset; |
1888 | resp->flags = flags; | |
1889 | resp->id = id; | |
1890 | resp->status = (s16)size; | |
1891 | if (st < 0) | |
1892 | resp->status = (s16)st; | |
1893 | ||
e9ce7cb6 | 1894 | queue->rx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1895 | |
1896 | return resp; | |
1897 | } | |
1898 | ||
e9ce7cb6 | 1899 | void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx) |
f53c3fe8 ZK |
1900 | { |
1901 | int ret; | |
1902 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1903 | ||
1904 | gnttab_set_unmap_op(&tx_unmap_op, | |
e9ce7cb6 | 1905 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1906 | GNTMAP_host_map, |
e9ce7cb6 WL |
1907 | queue->grant_tx_handle[pending_idx]); |
1908 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1909 | |
1910 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
e9ce7cb6 | 1911 | &queue->mmap_pages[pending_idx], 1); |
7aceb47a | 1912 | if (ret) { |
e9ce7cb6 | 1913 | netdev_err(queue->vif->dev, |
7aceb47a ZK |
1914 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: %x status: %d\n", |
1915 | ret, | |
1916 | pending_idx, | |
1917 | tx_unmap_op.host_addr, | |
1918 | tx_unmap_op.handle, | |
1919 | tx_unmap_op.status); | |
1920 | BUG(); | |
1921 | } | |
f53c3fe8 ZK |
1922 | } |
1923 | ||
e9ce7cb6 | 1924 | static inline int tx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1925 | { |
e9ce7cb6 | 1926 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))) |
f942dc25 IC |
1927 | return 1; |
1928 | ||
1929 | return 0; | |
1930 | } | |
1931 | ||
e9ce7cb6 | 1932 | static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue) |
f53c3fe8 | 1933 | { |
e9ce7cb6 | 1934 | return queue->dealloc_cons != queue->dealloc_prod; |
f53c3fe8 ZK |
1935 | } |
1936 | ||
e9ce7cb6 | 1937 | void xenvif_unmap_frontend_rings(struct xenvif_queue *queue) |
f942dc25 | 1938 | { |
e9ce7cb6 WL |
1939 | if (queue->tx.sring) |
1940 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1941 | queue->tx.sring); | |
1942 | if (queue->rx.sring) | |
1943 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1944 | queue->rx.sring); | |
f942dc25 IC |
1945 | } |
1946 | ||
e9ce7cb6 | 1947 | int xenvif_map_frontend_rings(struct xenvif_queue *queue, |
7376419a WL |
1948 | grant_ref_t tx_ring_ref, |
1949 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1950 | { |
c9d63699 | 1951 | void *addr; |
f942dc25 IC |
1952 | struct xen_netif_tx_sring *txs; |
1953 | struct xen_netif_rx_sring *rxs; | |
1954 | ||
1955 | int err = -ENOMEM; | |
1956 | ||
e9ce7cb6 | 1957 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
c9d63699 DV |
1958 | tx_ring_ref, &addr); |
1959 | if (err) | |
f942dc25 IC |
1960 | goto err; |
1961 | ||
c9d63699 | 1962 | txs = (struct xen_netif_tx_sring *)addr; |
e9ce7cb6 | 1963 | BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE); |
f942dc25 | 1964 | |
e9ce7cb6 | 1965 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
c9d63699 DV |
1966 | rx_ring_ref, &addr); |
1967 | if (err) | |
f942dc25 | 1968 | goto err; |
f942dc25 | 1969 | |
c9d63699 | 1970 | rxs = (struct xen_netif_rx_sring *)addr; |
e9ce7cb6 | 1971 | BACK_RING_INIT(&queue->rx, rxs, PAGE_SIZE); |
f942dc25 IC |
1972 | |
1973 | return 0; | |
1974 | ||
1975 | err: | |
e9ce7cb6 | 1976 | xenvif_unmap_frontend_rings(queue); |
f942dc25 IC |
1977 | return err; |
1978 | } | |
1979 | ||
ecf08d2d | 1980 | static void xenvif_queue_carrier_off(struct xenvif_queue *queue) |
ca2f09f2 | 1981 | { |
ecf08d2d DV |
1982 | struct xenvif *vif = queue->vif; |
1983 | ||
1984 | queue->stalled = true; | |
1985 | ||
1986 | /* At least one queue has stalled? Disable the carrier. */ | |
1987 | spin_lock(&vif->lock); | |
1988 | if (vif->stalled_queues++ == 0) { | |
1989 | netdev_info(vif->dev, "Guest Rx stalled"); | |
1990 | netif_carrier_off(vif->dev); | |
1991 | } | |
1992 | spin_unlock(&vif->lock); | |
ca2f09f2 PD |
1993 | } |
1994 | ||
ecf08d2d | 1995 | static void xenvif_queue_carrier_on(struct xenvif_queue *queue) |
f34a4cf9 | 1996 | { |
ecf08d2d | 1997 | struct xenvif *vif = queue->vif; |
f34a4cf9 | 1998 | |
ecf08d2d DV |
1999 | queue->last_rx_time = jiffies; /* Reset Rx stall detection. */ |
2000 | queue->stalled = false; | |
f34a4cf9 | 2001 | |
ecf08d2d DV |
2002 | /* All queues are ready? Enable the carrier. */ |
2003 | spin_lock(&vif->lock); | |
2004 | if (--vif->stalled_queues == 0) { | |
2005 | netdev_info(vif->dev, "Guest Rx ready"); | |
2006 | netif_carrier_on(vif->dev); | |
2007 | } | |
2008 | spin_unlock(&vif->lock); | |
2009 | } | |
f34a4cf9 | 2010 | |
ecf08d2d DV |
2011 | static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue) |
2012 | { | |
2013 | RING_IDX prod, cons; | |
2014 | ||
2015 | prod = queue->rx.sring->req_prod; | |
2016 | cons = queue->rx.req_cons; | |
2017 | ||
2018 | return !queue->stalled | |
2019 | && prod - cons < XEN_NETBK_RX_SLOTS_MAX | |
2020 | && time_after(jiffies, | |
26c0e102 | 2021 | queue->last_rx_time + queue->vif->stall_timeout); |
ecf08d2d DV |
2022 | } |
2023 | ||
2024 | static bool xenvif_rx_queue_ready(struct xenvif_queue *queue) | |
2025 | { | |
2026 | RING_IDX prod, cons; | |
2027 | ||
2028 | prod = queue->rx.sring->req_prod; | |
2029 | cons = queue->rx.req_cons; | |
2030 | ||
2031 | return queue->stalled | |
2032 | && prod - cons >= XEN_NETBK_RX_SLOTS_MAX; | |
2033 | } | |
2034 | ||
f48da8b1 | 2035 | static bool xenvif_have_rx_work(struct xenvif_queue *queue) |
ca2f09f2 | 2036 | { |
f48da8b1 DV |
2037 | return (!skb_queue_empty(&queue->rx_queue) |
2038 | && xenvif_rx_ring_slots_available(queue, XEN_NETBK_RX_SLOTS_MAX)) | |
26c0e102 DV |
2039 | || (queue->vif->stall_timeout && |
2040 | (xenvif_rx_queue_stalled(queue) | |
2041 | || xenvif_rx_queue_ready(queue))) | |
f48da8b1 DV |
2042 | || kthread_should_stop() |
2043 | || queue->vif->disabled; | |
ca2f09f2 PD |
2044 | } |
2045 | ||
f48da8b1 | 2046 | static long xenvif_rx_queue_timeout(struct xenvif_queue *queue) |
f34a4cf9 | 2047 | { |
f48da8b1 DV |
2048 | struct sk_buff *skb; |
2049 | long timeout; | |
f34a4cf9 | 2050 | |
f48da8b1 DV |
2051 | skb = skb_peek(&queue->rx_queue); |
2052 | if (!skb) | |
2053 | return MAX_SCHEDULE_TIMEOUT; | |
f34a4cf9 | 2054 | |
f48da8b1 DV |
2055 | timeout = XENVIF_RX_CB(skb)->expires - jiffies; |
2056 | return timeout < 0 ? 0 : timeout; | |
2057 | } | |
f34a4cf9 | 2058 | |
f48da8b1 DV |
2059 | /* Wait until the guest Rx thread has work. |
2060 | * | |
2061 | * The timeout needs to be adjusted based on the current head of the | |
2062 | * queue (and not just the head at the beginning). In particular, if | |
2063 | * the queue is initially empty an infinite timeout is used and this | |
2064 | * needs to be reduced when a skb is queued. | |
2065 | * | |
2066 | * This cannot be done with wait_event_timeout() because it only | |
2067 | * calculates the timeout once. | |
2068 | */ | |
2069 | static void xenvif_wait_for_rx_work(struct xenvif_queue *queue) | |
2070 | { | |
2071 | DEFINE_WAIT(wait); | |
2072 | ||
2073 | if (xenvif_have_rx_work(queue)) | |
2074 | return; | |
2075 | ||
2076 | for (;;) { | |
2077 | long ret; | |
2078 | ||
2079 | prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE); | |
2080 | if (xenvif_have_rx_work(queue)) | |
2081 | break; | |
2082 | ret = schedule_timeout(xenvif_rx_queue_timeout(queue)); | |
2083 | if (!ret) | |
2084 | break; | |
f34a4cf9 | 2085 | } |
f48da8b1 | 2086 | finish_wait(&queue->wq, &wait); |
f34a4cf9 ZK |
2087 | } |
2088 | ||
121fa4b7 | 2089 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd | 2090 | { |
e9ce7cb6 | 2091 | struct xenvif_queue *queue = data; |
f48da8b1 | 2092 | struct xenvif *vif = queue->vif; |
b3f980bd | 2093 | |
26c0e102 DV |
2094 | if (!vif->stall_timeout) |
2095 | xenvif_queue_carrier_on(queue); | |
2096 | ||
f48da8b1 DV |
2097 | for (;;) { |
2098 | xenvif_wait_for_rx_work(queue); | |
e9d8b2c2 | 2099 | |
f34a4cf9 ZK |
2100 | if (kthread_should_stop()) |
2101 | break; | |
2102 | ||
e9d8b2c2 WL |
2103 | /* This frontend is found to be rogue, disable it in |
2104 | * kthread context. Currently this is only set when | |
2105 | * netback finds out frontend sends malformed packet, | |
2106 | * but we cannot disable the interface in softirq | |
e9ce7cb6 WL |
2107 | * context so we defer it here, if this thread is |
2108 | * associated with queue 0. | |
e9d8b2c2 | 2109 | */ |
f48da8b1 DV |
2110 | if (unlikely(vif->disabled && queue->id == 0)) { |
2111 | xenvif_carrier_off(vif); | |
2112 | xenvif_rx_queue_purge(queue); | |
2113 | continue; | |
09350788 ZK |
2114 | } |
2115 | ||
e9ce7cb6 WL |
2116 | if (!skb_queue_empty(&queue->rx_queue)) |
2117 | xenvif_rx_action(queue); | |
b3f980bd | 2118 | |
ecf08d2d DV |
2119 | /* If the guest hasn't provided any Rx slots for a |
2120 | * while it's probably not responsive, drop the | |
2121 | * carrier so packets are dropped earlier. | |
2122 | */ | |
26c0e102 DV |
2123 | if (vif->stall_timeout) { |
2124 | if (xenvif_rx_queue_stalled(queue)) | |
2125 | xenvif_queue_carrier_off(queue); | |
2126 | else if (xenvif_rx_queue_ready(queue)) | |
2127 | xenvif_queue_carrier_on(queue); | |
2128 | } | |
ecf08d2d | 2129 | |
f48da8b1 DV |
2130 | /* Queued packets may have foreign pages from other |
2131 | * domains. These cannot be queued indefinitely as | |
2132 | * this would starve guests of grant refs and transmit | |
2133 | * slots. | |
2134 | */ | |
2135 | xenvif_rx_queue_drop_expired(queue); | |
2136 | ||
2137 | xenvif_rx_queue_maybe_wake(queue); | |
2138 | ||
b3f980bd WL |
2139 | cond_resched(); |
2140 | } | |
2141 | ||
ca2f09f2 | 2142 | /* Bin any remaining skbs */ |
f48da8b1 | 2143 | xenvif_rx_queue_purge(queue); |
ca2f09f2 | 2144 | |
b3f980bd WL |
2145 | return 0; |
2146 | } | |
2147 | ||
a64bd934 WL |
2148 | static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue) |
2149 | { | |
2150 | /* Dealloc thread must remain running until all inflight | |
2151 | * packets complete. | |
2152 | */ | |
2153 | return kthread_should_stop() && | |
2154 | !atomic_read(&queue->inflight_packets); | |
2155 | } | |
2156 | ||
f53c3fe8 ZK |
2157 | int xenvif_dealloc_kthread(void *data) |
2158 | { | |
e9ce7cb6 | 2159 | struct xenvif_queue *queue = data; |
f53c3fe8 | 2160 | |
a64bd934 | 2161 | for (;;) { |
e9ce7cb6 WL |
2162 | wait_event_interruptible(queue->dealloc_wq, |
2163 | tx_dealloc_work_todo(queue) || | |
a64bd934 WL |
2164 | xenvif_dealloc_kthread_should_stop(queue)); |
2165 | if (xenvif_dealloc_kthread_should_stop(queue)) | |
f53c3fe8 ZK |
2166 | break; |
2167 | ||
e9ce7cb6 | 2168 | xenvif_tx_dealloc_action(queue); |
f53c3fe8 ZK |
2169 | cond_resched(); |
2170 | } | |
2171 | ||
2172 | /* Unmap anything remaining*/ | |
e9ce7cb6 WL |
2173 | if (tx_dealloc_work_todo(queue)) |
2174 | xenvif_tx_dealloc_action(queue); | |
f53c3fe8 ZK |
2175 | |
2176 | return 0; | |
2177 | } | |
2178 | ||
f942dc25 IC |
2179 | static int __init netback_init(void) |
2180 | { | |
f942dc25 | 2181 | int rc = 0; |
f942dc25 | 2182 | |
2a14b244 | 2183 | if (!xen_domain()) |
f942dc25 IC |
2184 | return -ENODEV; |
2185 | ||
8d3d53b3 AB |
2186 | /* Allow as many queues as there are CPUs, by default */ |
2187 | xenvif_max_queues = num_online_cpus(); | |
2188 | ||
37641494 | 2189 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
2190 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
2191 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 2192 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
2193 | } |
2194 | ||
f942dc25 IC |
2195 | rc = xenvif_xenbus_init(); |
2196 | if (rc) | |
2197 | goto failed_init; | |
2198 | ||
f51de243 ZK |
2199 | #ifdef CONFIG_DEBUG_FS |
2200 | xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL); | |
2201 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
2202 | pr_warn("Init of debugfs returned %ld!\n", | |
2203 | PTR_ERR(xen_netback_dbg_root)); | |
2204 | #endif /* CONFIG_DEBUG_FS */ | |
2205 | ||
f942dc25 IC |
2206 | return 0; |
2207 | ||
2208 | failed_init: | |
f942dc25 | 2209 | return rc; |
f942dc25 IC |
2210 | } |
2211 | ||
2212 | module_init(netback_init); | |
2213 | ||
b103f358 WL |
2214 | static void __exit netback_fini(void) |
2215 | { | |
f51de243 ZK |
2216 | #ifdef CONFIG_DEBUG_FS |
2217 | if (!IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
2218 | debugfs_remove_recursive(xen_netback_dbg_root); | |
2219 | #endif /* CONFIG_DEBUG_FS */ | |
b103f358 | 2220 | xenvif_xenbus_fini(); |
b103f358 WL |
2221 | } |
2222 | module_exit(netback_fini); | |
2223 | ||
f942dc25 | 2224 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 2225 | MODULE_ALIAS("xen-backend:vif"); |