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 | ||
2810e5b9 WL |
58 | /* |
59 | * This is the maximum slots a skb can have. If a guest sends a skb | |
60 | * which exceeds this limit it is considered malicious. | |
61 | */ | |
37641494 WL |
62 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
63 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
64 | module_param(fatal_skb_slots, uint, 0444); | |
65 | ||
7376419a WL |
66 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
67 | u8 status); | |
68 | ||
f942dc25 IC |
69 | static void make_tx_response(struct xenvif *vif, |
70 | struct xen_netif_tx_request *txp, | |
71 | s8 st); | |
b3f980bd WL |
72 | |
73 | static inline int tx_work_todo(struct xenvif *vif); | |
74 | static inline int rx_work_todo(struct xenvif *vif); | |
75 | ||
f942dc25 IC |
76 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
77 | u16 id, | |
78 | s8 st, | |
79 | u16 offset, | |
80 | u16 size, | |
81 | u16 flags); | |
82 | ||
b3f980bd | 83 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 84 | u16 idx) |
f942dc25 | 85 | { |
b3f980bd | 86 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
87 | } |
88 | ||
b3f980bd | 89 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 90 | u16 idx) |
f942dc25 | 91 | { |
b3f980bd | 92 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
93 | } |
94 | ||
f53c3fe8 ZK |
95 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
96 | */ | |
3e2234b3 ZK |
97 | static inline struct xenvif* ubuf_to_vif(struct ubuf_info *ubuf) |
98 | { | |
f53c3fe8 ZK |
99 | u16 pending_idx = ubuf->desc; |
100 | struct pending_tx_info *temp = | |
101 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
102 | return container_of(temp - pending_idx, | |
103 | struct xenvif, | |
104 | pending_tx_info[0]); | |
3e2234b3 | 105 | } |
f53c3fe8 | 106 | |
2eba61d5 PD |
107 | /* This is a miniumum size for the linear area to avoid lots of |
108 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
109 | * value 128 was chosen as it covers all IPv4 and most likely | |
110 | * IPv6 headers. | |
f942dc25 | 111 | */ |
2eba61d5 | 112 | #define PKT_PROT_LEN 128 |
f942dc25 | 113 | |
ea066ad1 IC |
114 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
115 | { | |
116 | return (u16)frag->page_offset; | |
117 | } | |
118 | ||
119 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
120 | { | |
121 | frag->page_offset = pending_idx; | |
122 | } | |
123 | ||
f942dc25 IC |
124 | static inline pending_ring_idx_t pending_index(unsigned i) |
125 | { | |
126 | return i & (MAX_PENDING_REQS-1); | |
127 | } | |
128 | ||
ca2f09f2 | 129 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
f942dc25 | 130 | { |
ca2f09f2 | 131 | RING_IDX prod, cons; |
f942dc25 | 132 | |
ca2f09f2 PD |
133 | do { |
134 | prod = vif->rx.sring->req_prod; | |
135 | cons = vif->rx.req_cons; | |
f942dc25 | 136 | |
ca2f09f2 PD |
137 | if (prod - cons >= needed) |
138 | return true; | |
f942dc25 | 139 | |
ca2f09f2 | 140 | vif->rx.sring->req_event = prod + 1; |
f942dc25 | 141 | |
ca2f09f2 PD |
142 | /* Make sure event is visible before we check prod |
143 | * again. | |
144 | */ | |
145 | mb(); | |
146 | } while (vif->rx.sring->req_prod != prod); | |
f942dc25 | 147 | |
ca2f09f2 | 148 | return false; |
f942dc25 IC |
149 | } |
150 | ||
151 | /* | |
152 | * Returns true if we should start a new receive buffer instead of | |
153 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
154 | * bytes. | |
155 | */ | |
156 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
157 | { | |
158 | /* simple case: we have completely filled the current buffer. */ | |
159 | if (offset == MAX_BUFFER_OFFSET) | |
160 | return true; | |
161 | ||
162 | /* | |
163 | * complex case: start a fresh buffer if the current frag | |
164 | * would overflow the current buffer but only if: | |
165 | * (i) this frag would fit completely in the next buffer | |
166 | * and (ii) there is already some data in the current buffer | |
167 | * and (iii) this is not the head buffer. | |
168 | * | |
169 | * Where: | |
170 | * - (i) stops us splitting a frag into two copies | |
171 | * unless the frag is too large for a single buffer. | |
172 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
173 | * - (iii) stops us leaving the first buffer | |
174 | * empty. Strictly speaking this is already covered | |
175 | * by (ii) but is explicitly checked because | |
176 | * netfront relies on the first buffer being | |
177 | * non-empty and can crash otherwise. | |
178 | * | |
179 | * This means we will effectively linearise small | |
180 | * frags but do not needlessly split large buffers | |
181 | * into multiple copies tend to give large frags their | |
182 | * own buffers as before. | |
183 | */ | |
184 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
185 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
186 | return true; | |
187 | ||
188 | return false; | |
189 | } | |
190 | ||
f942dc25 IC |
191 | struct netrx_pending_operations { |
192 | unsigned copy_prod, copy_cons; | |
193 | unsigned meta_prod, meta_cons; | |
194 | struct gnttab_copy *copy; | |
b3f980bd | 195 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
196 | int copy_off; |
197 | grant_ref_t copy_gref; | |
198 | }; | |
199 | ||
b3f980bd WL |
200 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
201 | struct netrx_pending_operations *npo) | |
f942dc25 | 202 | { |
b3f980bd | 203 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
204 | struct xen_netif_rx_request *req; |
205 | ||
206 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
207 | ||
208 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 209 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
210 | meta->gso_size = 0; |
211 | meta->size = 0; | |
212 | meta->id = req->id; | |
213 | ||
214 | npo->copy_off = 0; | |
215 | npo->copy_gref = req->gref; | |
216 | ||
217 | return meta; | |
218 | } | |
219 | ||
33bc801d WL |
220 | /* |
221 | * Set up the grant operations for this fragment. If it's a flipping | |
222 | * interface, we also set up the unmap request from here. | |
223 | */ | |
7376419a WL |
224 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
225 | struct netrx_pending_operations *npo, | |
226 | struct page *page, unsigned long size, | |
3e2234b3 ZK |
227 | unsigned long offset, int *head, |
228 | struct xenvif *foreign_vif, | |
229 | grant_ref_t foreign_gref) | |
f942dc25 IC |
230 | { |
231 | struct gnttab_copy *copy_gop; | |
b3f980bd | 232 | struct xenvif_rx_meta *meta; |
f942dc25 | 233 | unsigned long bytes; |
82cada22 | 234 | int gso_type; |
f942dc25 IC |
235 | |
236 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 237 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
238 | |
239 | meta = npo->meta + npo->meta_prod - 1; | |
240 | ||
6a8ed462 IC |
241 | /* Skip unused frames from start of page */ |
242 | page += offset >> PAGE_SHIFT; | |
243 | offset &= ~PAGE_MASK; | |
244 | ||
f942dc25 | 245 | while (size > 0) { |
6a8ed462 | 246 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
247 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
248 | ||
6a8ed462 IC |
249 | bytes = PAGE_SIZE - offset; |
250 | ||
251 | if (bytes > size) | |
252 | bytes = size; | |
253 | ||
33bc801d | 254 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
255 | /* |
256 | * Netfront requires there to be some data in the head | |
257 | * buffer. | |
258 | */ | |
33bc801d | 259 | BUG_ON(*head); |
f942dc25 IC |
260 | |
261 | meta = get_next_rx_buffer(vif, npo); | |
262 | } | |
263 | ||
f942dc25 IC |
264 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
265 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
266 | ||
267 | copy_gop = npo->copy + npo->copy_prod++; | |
268 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
269 | copy_gop->len = bytes; |
270 | ||
3e2234b3 ZK |
271 | if (foreign_vif) { |
272 | copy_gop->source.domid = foreign_vif->domid; | |
273 | copy_gop->source.u.ref = foreign_gref; | |
274 | copy_gop->flags |= GNTCOPY_source_gref; | |
275 | } else { | |
276 | copy_gop->source.domid = DOMID_SELF; | |
277 | copy_gop->source.u.gmfn = | |
278 | virt_to_mfn(page_address(page)); | |
279 | } | |
f942dc25 | 280 | copy_gop->source.offset = offset; |
f942dc25 | 281 | |
b3f980bd | 282 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
283 | copy_gop->dest.offset = npo->copy_off; |
284 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
285 | |
286 | npo->copy_off += bytes; | |
287 | meta->size += bytes; | |
288 | ||
289 | offset += bytes; | |
290 | size -= bytes; | |
291 | ||
6a8ed462 IC |
292 | /* Next frame */ |
293 | if (offset == PAGE_SIZE && size) { | |
294 | BUG_ON(!PageCompound(page)); | |
295 | page++; | |
296 | offset = 0; | |
297 | } | |
298 | ||
f942dc25 | 299 | /* Leave a gap for the GSO descriptor. */ |
82cada22 PD |
300 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
301 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
302 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
303 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
304 | else | |
305 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
306 | ||
307 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
308 | vif->rx.req_cons++; |
309 | ||
33bc801d | 310 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
311 | |
312 | } | |
313 | } | |
314 | ||
315 | /* | |
316 | * Prepare an SKB to be transmitted to the frontend. | |
317 | * | |
318 | * This function is responsible for allocating grant operations, meta | |
319 | * structures, etc. | |
320 | * | |
321 | * It returns the number of meta structures consumed. The number of | |
322 | * ring slots used is always equal to the number of meta slots used | |
323 | * plus the number of GSO descriptors used. Currently, we use either | |
324 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
325 | * frontend-side LRO). | |
326 | */ | |
7376419a WL |
327 | static int xenvif_gop_skb(struct sk_buff *skb, |
328 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
329 | { |
330 | struct xenvif *vif = netdev_priv(skb->dev); | |
331 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
332 | int i; | |
333 | struct xen_netif_rx_request *req; | |
b3f980bd | 334 | struct xenvif_rx_meta *meta; |
f942dc25 | 335 | unsigned char *data; |
33bc801d | 336 | int head = 1; |
f942dc25 | 337 | int old_meta_prod; |
82cada22 PD |
338 | int gso_type; |
339 | int gso_size; | |
3e2234b3 ZK |
340 | struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg; |
341 | grant_ref_t foreign_grefs[MAX_SKB_FRAGS]; | |
342 | struct xenvif *foreign_vif = NULL; | |
f942dc25 IC |
343 | |
344 | old_meta_prod = npo->meta_prod; | |
345 | ||
82cada22 PD |
346 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { |
347 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
348 | gso_size = skb_shinfo(skb)->gso_size; | |
349 | } else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { | |
350 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
351 | gso_size = skb_shinfo(skb)->gso_size; | |
352 | } else { | |
353 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
354 | gso_size = 0; | |
355 | } | |
356 | ||
f942dc25 | 357 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 358 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
359 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
360 | meta = npo->meta + npo->meta_prod++; | |
82cada22 PD |
361 | meta->gso_type = gso_type; |
362 | meta->gso_size = gso_size; | |
f942dc25 IC |
363 | meta->size = 0; |
364 | meta->id = req->id; | |
365 | } | |
366 | ||
367 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
368 | meta = npo->meta + npo->meta_prod++; | |
369 | ||
82cada22 PD |
370 | if ((1 << gso_type) & vif->gso_mask) { |
371 | meta->gso_type = gso_type; | |
372 | meta->gso_size = gso_size; | |
373 | } else { | |
374 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 375 | meta->gso_size = 0; |
82cada22 | 376 | } |
f942dc25 IC |
377 | |
378 | meta->size = 0; | |
379 | meta->id = req->id; | |
380 | npo->copy_off = 0; | |
381 | npo->copy_gref = req->gref; | |
382 | ||
3e2234b3 ZK |
383 | if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) && |
384 | (ubuf->callback == &xenvif_zerocopy_callback)) { | |
385 | int i = 0; | |
386 | foreign_vif = ubuf_to_vif(ubuf); | |
387 | ||
388 | do { | |
389 | u16 pending_idx = ubuf->desc; | |
390 | foreign_grefs[i++] = | |
391 | foreign_vif->pending_tx_info[pending_idx].req.gref; | |
392 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
393 | } while (ubuf); | |
394 | } | |
395 | ||
f942dc25 IC |
396 | data = skb->data; |
397 | while (data < skb_tail_pointer(skb)) { | |
398 | unsigned int offset = offset_in_page(data); | |
399 | unsigned int len = PAGE_SIZE - offset; | |
400 | ||
401 | if (data + len > skb_tail_pointer(skb)) | |
402 | len = skb_tail_pointer(skb) - data; | |
403 | ||
7376419a | 404 | xenvif_gop_frag_copy(vif, skb, npo, |
3e2234b3 ZK |
405 | virt_to_page(data), len, offset, &head, |
406 | NULL, | |
407 | 0); | |
f942dc25 IC |
408 | data += len; |
409 | } | |
410 | ||
411 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
412 | xenvif_gop_frag_copy(vif, skb, npo, |
413 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
414 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
415 | skb_shinfo(skb)->frags[i].page_offset, | |
3e2234b3 ZK |
416 | &head, |
417 | foreign_vif, | |
418 | foreign_grefs[i]); | |
f942dc25 IC |
419 | } |
420 | ||
421 | return npo->meta_prod - old_meta_prod; | |
422 | } | |
423 | ||
424 | /* | |
7376419a | 425 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
426 | * used to set up the operations on the top of |
427 | * netrx_pending_operations, which have since been done. Check that | |
428 | * they didn't give any errors and advance over them. | |
429 | */ | |
7376419a WL |
430 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
431 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
432 | { |
433 | struct gnttab_copy *copy_op; | |
434 | int status = XEN_NETIF_RSP_OKAY; | |
435 | int i; | |
436 | ||
437 | for (i = 0; i < nr_meta_slots; i++) { | |
438 | copy_op = npo->copy + npo->copy_cons++; | |
439 | if (copy_op->status != GNTST_okay) { | |
440 | netdev_dbg(vif->dev, | |
441 | "Bad status %d from copy to DOM%d.\n", | |
442 | copy_op->status, vif->domid); | |
443 | status = XEN_NETIF_RSP_ERROR; | |
444 | } | |
445 | } | |
446 | ||
447 | return status; | |
448 | } | |
449 | ||
7376419a WL |
450 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
451 | struct xenvif_rx_meta *meta, | |
452 | int nr_meta_slots) | |
f942dc25 IC |
453 | { |
454 | int i; | |
455 | unsigned long offset; | |
456 | ||
457 | /* No fragments used */ | |
458 | if (nr_meta_slots <= 1) | |
459 | return; | |
460 | ||
461 | nr_meta_slots--; | |
462 | ||
463 | for (i = 0; i < nr_meta_slots; i++) { | |
464 | int flags; | |
465 | if (i == nr_meta_slots - 1) | |
466 | flags = 0; | |
467 | else | |
468 | flags = XEN_NETRXF_more_data; | |
469 | ||
470 | offset = 0; | |
471 | make_rx_response(vif, meta[i].id, status, offset, | |
472 | meta[i].size, flags); | |
473 | } | |
474 | } | |
475 | ||
8f13dd96 | 476 | struct xenvif_rx_cb { |
33bc801d WL |
477 | int meta_slots_used; |
478 | }; | |
479 | ||
8f13dd96 ZK |
480 | #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) |
481 | ||
ca2f09f2 | 482 | void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
483 | { |
484 | wake_up(&vif->wq); | |
485 | } | |
486 | ||
ca2f09f2 | 487 | static void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 488 | { |
f942dc25 | 489 | s8 status; |
e1f00a69 | 490 | u16 flags; |
f942dc25 IC |
491 | struct xen_netif_rx_response *resp; |
492 | struct sk_buff_head rxq; | |
493 | struct sk_buff *skb; | |
494 | LIST_HEAD(notify); | |
495 | int ret; | |
f942dc25 | 496 | unsigned long offset; |
11b57f90 | 497 | bool need_to_notify = false; |
f942dc25 IC |
498 | |
499 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
500 | .copy = vif->grant_copy_op, |
501 | .meta = vif->meta, | |
f942dc25 IC |
502 | }; |
503 | ||
504 | skb_queue_head_init(&rxq); | |
505 | ||
b3f980bd | 506 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
9ab9831b | 507 | RING_IDX max_slots_needed; |
ca2f09f2 PD |
508 | int i; |
509 | ||
510 | /* We need a cheap worse case estimate for the number of | |
511 | * slots we'll use. | |
512 | */ | |
513 | ||
514 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
515 | skb_headlen(skb), | |
516 | PAGE_SIZE); | |
517 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
518 | unsigned int size; | |
519 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | |
520 | max_slots_needed += DIV_ROUND_UP(size, PAGE_SIZE); | |
521 | } | |
522 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
523 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
524 | max_slots_needed++; | |
525 | ||
526 | /* If the skb may not fit then bail out now */ | |
527 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | |
528 | skb_queue_head(&vif->rx_queue, skb); | |
11b57f90 | 529 | need_to_notify = true; |
9ab9831b | 530 | vif->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 531 | break; |
9ab9831b ZK |
532 | } else |
533 | vif->rx_last_skb_slots = 0; | |
f942dc25 | 534 | |
8f13dd96 ZK |
535 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo); |
536 | BUG_ON(XENVIF_RX_CB(skb)->meta_slots_used > max_slots_needed); | |
f942dc25 IC |
537 | |
538 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
539 | } |
540 | ||
b3f980bd | 541 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
542 | |
543 | if (!npo.copy_prod) | |
ca2f09f2 | 544 | goto done; |
f942dc25 | 545 | |
ac3d5ac2 | 546 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
b3f980bd | 547 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
548 | |
549 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 550 | |
82cada22 PD |
551 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
552 | vif->gso_prefix_mask) { | |
f942dc25 | 553 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 554 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
555 | |
556 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
557 | ||
b3f980bd WL |
558 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
559 | resp->id = vif->meta[npo.meta_cons].id; | |
8f13dd96 | 560 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
561 | |
562 | npo.meta_cons++; | |
8f13dd96 | 563 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
564 | } |
565 | ||
566 | ||
567 | vif->dev->stats.tx_bytes += skb->len; | |
568 | vif->dev->stats.tx_packets++; | |
569 | ||
8f13dd96 ZK |
570 | status = xenvif_check_gop(vif, |
571 | XENVIF_RX_CB(skb)->meta_slots_used, | |
572 | &npo); | |
f942dc25 | 573 | |
8f13dd96 | 574 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
575 | flags = 0; |
576 | else | |
577 | flags = XEN_NETRXF_more_data; | |
578 | ||
579 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
580 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
581 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
582 | /* remote but checksummed. */ | |
583 | flags |= XEN_NETRXF_data_validated; | |
584 | ||
585 | offset = 0; | |
b3f980bd | 586 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 587 | status, offset, |
b3f980bd | 588 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
589 | flags); |
590 | ||
82cada22 PD |
591 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
592 | vif->gso_mask) { | |
f942dc25 IC |
593 | struct xen_netif_extra_info *gso = |
594 | (struct xen_netif_extra_info *) | |
595 | RING_GET_RESPONSE(&vif->rx, | |
596 | vif->rx.rsp_prod_pvt++); | |
597 | ||
598 | resp->flags |= XEN_NETRXF_extra_info; | |
599 | ||
82cada22 | 600 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 601 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
602 | gso->u.gso.pad = 0; |
603 | gso->u.gso.features = 0; | |
604 | ||
605 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
606 | gso->flags = 0; | |
607 | } | |
608 | ||
7376419a WL |
609 | xenvif_add_frag_responses(vif, status, |
610 | vif->meta + npo.meta_cons + 1, | |
8f13dd96 | 611 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 IC |
612 | |
613 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 614 | |
11b57f90 | 615 | need_to_notify |= !!ret; |
b3f980bd | 616 | |
8f13dd96 | 617 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
618 | dev_kfree_skb(skb); |
619 | } | |
620 | ||
ca2f09f2 | 621 | done: |
b3f980bd | 622 | if (need_to_notify) |
e1f00a69 | 623 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
624 | } |
625 | ||
7376419a | 626 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
627 | { |
628 | int more_to_do; | |
629 | ||
630 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
631 | ||
632 | if (more_to_do) | |
b3f980bd | 633 | napi_schedule(&vif->napi); |
f942dc25 IC |
634 | } |
635 | ||
636 | static void tx_add_credit(struct xenvif *vif) | |
637 | { | |
638 | unsigned long max_burst, max_credit; | |
639 | ||
640 | /* | |
641 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
642 | * Otherwise the interface can seize up due to insufficient credit. | |
643 | */ | |
644 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
645 | max_burst = min(max_burst, 131072UL); | |
646 | max_burst = max(max_burst, vif->credit_bytes); | |
647 | ||
648 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
649 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
650 | if (max_credit < vif->remaining_credit) | |
651 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
652 | ||
653 | vif->remaining_credit = min(max_credit, max_burst); | |
654 | } | |
655 | ||
656 | static void tx_credit_callback(unsigned long data) | |
657 | { | |
658 | struct xenvif *vif = (struct xenvif *)data; | |
659 | tx_add_credit(vif); | |
7376419a | 660 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
661 | } |
662 | ||
7376419a WL |
663 | static void xenvif_tx_err(struct xenvif *vif, |
664 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
665 | { |
666 | RING_IDX cons = vif->tx.req_cons; | |
f53c3fe8 | 667 | unsigned long flags; |
f942dc25 IC |
668 | |
669 | do { | |
f53c3fe8 | 670 | spin_lock_irqsave(&vif->response_lock, flags); |
f942dc25 | 671 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); |
f53c3fe8 | 672 | spin_unlock_irqrestore(&vif->response_lock, flags); |
b9149729 | 673 | if (cons == end) |
f942dc25 IC |
674 | break; |
675 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
676 | } while (1); | |
677 | vif->tx.req_cons = cons; | |
f942dc25 IC |
678 | } |
679 | ||
7376419a | 680 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
681 | { |
682 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
683 | xenvif_carrier_off(vif); | |
48856286 IC |
684 | } |
685 | ||
7376419a WL |
686 | static int xenvif_count_requests(struct xenvif *vif, |
687 | struct xen_netif_tx_request *first, | |
688 | struct xen_netif_tx_request *txp, | |
689 | int work_to_do) | |
f942dc25 IC |
690 | { |
691 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
692 | int slots = 0; |
693 | int drop_err = 0; | |
59ccb4eb | 694 | int more_data; |
f942dc25 IC |
695 | |
696 | if (!(first->flags & XEN_NETTXF_more_data)) | |
697 | return 0; | |
698 | ||
699 | do { | |
59ccb4eb WL |
700 | struct xen_netif_tx_request dropped_tx = { 0 }; |
701 | ||
2810e5b9 WL |
702 | if (slots >= work_to_do) { |
703 | netdev_err(vif->dev, | |
704 | "Asked for %d slots but exceeds this limit\n", | |
705 | work_to_do); | |
7376419a | 706 | xenvif_fatal_tx_err(vif); |
35876b5f | 707 | return -ENODATA; |
f942dc25 IC |
708 | } |
709 | ||
2810e5b9 WL |
710 | /* This guest is really using too many slots and |
711 | * considered malicious. | |
712 | */ | |
37641494 | 713 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
714 | netdev_err(vif->dev, |
715 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 716 | slots, fatal_skb_slots); |
7376419a | 717 | xenvif_fatal_tx_err(vif); |
35876b5f | 718 | return -E2BIG; |
f942dc25 IC |
719 | } |
720 | ||
2810e5b9 | 721 | /* Xen network protocol had implicit dependency on |
37641494 WL |
722 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
723 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
724 | * same behavior as before. Any packet using more than | |
725 | * 18 slots but less than fatal_skb_slots slots is | |
726 | * dropped | |
2810e5b9 | 727 | */ |
37641494 | 728 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
729 | if (net_ratelimit()) |
730 | netdev_dbg(vif->dev, | |
731 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 732 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
733 | drop_err = -E2BIG; |
734 | } | |
735 | ||
59ccb4eb WL |
736 | if (drop_err) |
737 | txp = &dropped_tx; | |
738 | ||
2810e5b9 | 739 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 740 | sizeof(*txp)); |
03393fd5 WL |
741 | |
742 | /* If the guest submitted a frame >= 64 KiB then | |
743 | * first->size overflowed and following slots will | |
744 | * appear to be larger than the frame. | |
745 | * | |
746 | * This cannot be fatal error as there are buggy | |
747 | * frontends that do this. | |
748 | * | |
749 | * Consume all slots and drop the packet. | |
750 | */ | |
751 | if (!drop_err && txp->size > first->size) { | |
752 | if (net_ratelimit()) | |
753 | netdev_dbg(vif->dev, | |
754 | "Invalid tx request, slot size %u > remaining size %u\n", | |
755 | txp->size, first->size); | |
756 | drop_err = -EIO; | |
f942dc25 IC |
757 | } |
758 | ||
759 | first->size -= txp->size; | |
2810e5b9 | 760 | slots++; |
f942dc25 IC |
761 | |
762 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 763 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 764 | txp->offset, txp->size); |
7376419a | 765 | xenvif_fatal_tx_err(vif); |
35876b5f | 766 | return -EINVAL; |
f942dc25 | 767 | } |
59ccb4eb WL |
768 | |
769 | more_data = txp->flags & XEN_NETTXF_more_data; | |
770 | ||
771 | if (!drop_err) | |
772 | txp++; | |
773 | ||
774 | } while (more_data); | |
2810e5b9 WL |
775 | |
776 | if (drop_err) { | |
7376419a | 777 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
778 | return drop_err; |
779 | } | |
780 | ||
781 | return slots; | |
f942dc25 IC |
782 | } |
783 | ||
8f13dd96 ZK |
784 | |
785 | struct xenvif_tx_cb { | |
786 | u16 pending_idx; | |
787 | }; | |
788 | ||
789 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
790 | ||
f53c3fe8 ZK |
791 | static inline void xenvif_tx_create_gop(struct xenvif *vif, |
792 | u16 pending_idx, | |
793 | struct xen_netif_tx_request *txp, | |
794 | struct gnttab_map_grant_ref *gop) | |
795 | { | |
796 | vif->pages_to_map[gop-vif->tx_map_ops] = vif->mmap_pages[pending_idx]; | |
797 | gnttab_set_map_op(gop, idx_to_kaddr(vif, pending_idx), | |
798 | GNTMAP_host_map | GNTMAP_readonly, | |
799 | txp->gref, vif->domid); | |
800 | ||
801 | memcpy(&vif->pending_tx_info[pending_idx].req, txp, | |
802 | sizeof(*txp)); | |
803 | } | |
804 | ||
e3377f36 ZK |
805 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
806 | { | |
807 | struct sk_buff *skb = | |
808 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
809 | GFP_ATOMIC | __GFP_NOWARN); | |
810 | if (unlikely(skb == NULL)) | |
811 | return NULL; | |
812 | ||
813 | /* Packets passed to netif_rx() must have some headroom. */ | |
814 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
815 | ||
816 | /* Initialize it here to avoid later surprises */ | |
817 | skb_shinfo(skb)->destructor_arg = NULL; | |
818 | ||
819 | return skb; | |
820 | } | |
821 | ||
f53c3fe8 ZK |
822 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif *vif, |
823 | struct sk_buff *skb, | |
824 | struct xen_netif_tx_request *txp, | |
825 | struct gnttab_map_grant_ref *gop) | |
f942dc25 IC |
826 | { |
827 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
828 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 829 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
830 | int start; |
831 | pending_ring_idx_t index; | |
e3377f36 | 832 | unsigned int nr_slots, frag_overflow = 0; |
2810e5b9 WL |
833 | |
834 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 835 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 | 836 | */ |
e3377f36 ZK |
837 | if (shinfo->nr_frags > MAX_SKB_FRAGS) { |
838 | frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS; | |
839 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
840 | shinfo->nr_frags = MAX_SKB_FRAGS; | |
841 | } | |
2810e5b9 | 842 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
843 | |
844 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 845 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 846 | |
f53c3fe8 ZK |
847 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
848 | shinfo->nr_frags++, txp++, gop++) { | |
62bad319 ZK |
849 | index = pending_index(vif->pending_cons++); |
850 | pending_idx = vif->pending_ring[index]; | |
f53c3fe8 ZK |
851 | xenvif_tx_create_gop(vif, pending_idx, txp, gop); |
852 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); | |
f942dc25 IC |
853 | } |
854 | ||
e3377f36 ZK |
855 | if (frag_overflow) { |
856 | struct sk_buff *nskb = xenvif_alloc_skb(0); | |
857 | if (unlikely(nskb == NULL)) { | |
858 | if (net_ratelimit()) | |
859 | netdev_err(vif->dev, | |
860 | "Can't allocate the frag_list skb.\n"); | |
861 | return NULL; | |
862 | } | |
863 | ||
864 | shinfo = skb_shinfo(nskb); | |
865 | frags = shinfo->frags; | |
866 | ||
867 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
868 | shinfo->nr_frags++, txp++, gop++) { | |
869 | index = pending_index(vif->pending_cons++); | |
870 | pending_idx = vif->pending_ring[index]; | |
871 | xenvif_tx_create_gop(vif, pending_idx, txp, gop); | |
872 | frag_set_pending_idx(&frags[shinfo->nr_frags], | |
873 | pending_idx); | |
874 | } | |
875 | ||
876 | skb_shinfo(skb)->frag_list = nskb; | |
877 | } | |
2810e5b9 | 878 | |
f942dc25 IC |
879 | return gop; |
880 | } | |
881 | ||
f53c3fe8 ZK |
882 | static inline void xenvif_grant_handle_set(struct xenvif *vif, |
883 | u16 pending_idx, | |
884 | grant_handle_t handle) | |
885 | { | |
886 | if (unlikely(vif->grant_tx_handle[pending_idx] != | |
887 | NETBACK_INVALID_HANDLE)) { | |
888 | netdev_err(vif->dev, | |
889 | "Trying to overwrite active handle! pending_idx: %x\n", | |
890 | pending_idx); | |
891 | BUG(); | |
892 | } | |
893 | vif->grant_tx_handle[pending_idx] = handle; | |
894 | } | |
895 | ||
896 | static inline void xenvif_grant_handle_reset(struct xenvif *vif, | |
897 | u16 pending_idx) | |
898 | { | |
899 | if (unlikely(vif->grant_tx_handle[pending_idx] == | |
900 | NETBACK_INVALID_HANDLE)) { | |
901 | netdev_err(vif->dev, | |
902 | "Trying to unmap invalid handle! pending_idx: %x\n", | |
903 | pending_idx); | |
904 | BUG(); | |
905 | } | |
906 | vif->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; | |
907 | } | |
908 | ||
7376419a WL |
909 | static int xenvif_tx_check_gop(struct xenvif *vif, |
910 | struct sk_buff *skb, | |
f53c3fe8 | 911 | struct gnttab_map_grant_ref **gopp) |
f942dc25 | 912 | { |
f53c3fe8 | 913 | struct gnttab_map_grant_ref *gop = *gopp; |
8f13dd96 | 914 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
f942dc25 | 915 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 916 | struct pending_tx_info *tx_info; |
f942dc25 IC |
917 | int nr_frags = shinfo->nr_frags; |
918 | int i, err, start; | |
e3377f36 | 919 | struct sk_buff *first_skb = NULL; |
f942dc25 IC |
920 | |
921 | /* Check status of header. */ | |
922 | err = gop->status; | |
7d5145d8 | 923 | if (unlikely(err)) |
7376419a | 924 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f53c3fe8 ZK |
925 | else |
926 | xenvif_grant_handle_set(vif, pending_idx , gop->handle); | |
f942dc25 IC |
927 | |
928 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 929 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 930 | |
e3377f36 | 931 | check_frags: |
f942dc25 IC |
932 | for (i = start; i < nr_frags; i++) { |
933 | int j, newerr; | |
f942dc25 | 934 | |
ea066ad1 | 935 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 936 | tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 IC |
937 | |
938 | /* Check error status: if okay then remember grant handle. */ | |
62bad319 | 939 | newerr = (++gop)->status; |
2810e5b9 | 940 | |
f942dc25 | 941 | if (likely(!newerr)) { |
f53c3fe8 | 942 | xenvif_grant_handle_set(vif, pending_idx , gop->handle); |
f942dc25 IC |
943 | /* Had a previous error? Invalidate this fragment. */ |
944 | if (unlikely(err)) | |
f53c3fe8 | 945 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
946 | continue; |
947 | } | |
948 | ||
949 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 950 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
951 | |
952 | /* Not the first error? Preceding frags already invalidated. */ | |
953 | if (err) | |
954 | continue; | |
f942dc25 | 955 | /* First error: invalidate header and preceding fragments. */ |
e3377f36 ZK |
956 | if (!first_skb) |
957 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
958 | else | |
959 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
f53c3fe8 | 960 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 | 961 | for (j = start; j < i; j++) { |
5ccb3ea7 | 962 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
f53c3fe8 | 963 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
964 | } |
965 | ||
966 | /* Remember the error: invalidate all subsequent fragments. */ | |
967 | err = newerr; | |
968 | } | |
969 | ||
e3377f36 ZK |
970 | if (skb_has_frag_list(skb)) { |
971 | first_skb = skb; | |
972 | skb = shinfo->frag_list; | |
973 | shinfo = skb_shinfo(skb); | |
974 | nr_frags = shinfo->nr_frags; | |
975 | start = 0; | |
976 | ||
977 | goto check_frags; | |
978 | } | |
979 | ||
980 | /* There was a mapping error in the frag_list skb. We have to unmap | |
981 | * the first skb's frags | |
982 | */ | |
983 | if (first_skb && err) { | |
984 | int j; | |
985 | shinfo = skb_shinfo(first_skb); | |
986 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
987 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); | |
988 | for (j = start; j < shinfo->nr_frags; j++) { | |
989 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); | |
990 | xenvif_idx_unmap(vif, pending_idx); | |
991 | } | |
992 | } | |
993 | ||
f942dc25 IC |
994 | *gopp = gop + 1; |
995 | return err; | |
996 | } | |
997 | ||
7376419a | 998 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
999 | { |
1000 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1001 | int nr_frags = shinfo->nr_frags; | |
1002 | int i; | |
f53c3fe8 ZK |
1003 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1004 | ||
1005 | if (skb_shinfo(skb)->destructor_arg) | |
1006 | prev_pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
f942dc25 IC |
1007 | |
1008 | for (i = 0; i < nr_frags; i++) { | |
1009 | skb_frag_t *frag = shinfo->frags + i; | |
1010 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1011 | struct page *page; |
1012 | u16 pending_idx; | |
f942dc25 | 1013 | |
ea066ad1 | 1014 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1015 | |
f53c3fe8 ZK |
1016 | /* If this is not the first frag, chain it to the previous*/ |
1017 | if (unlikely(prev_pending_idx == INVALID_PENDING_IDX)) | |
1018 | skb_shinfo(skb)->destructor_arg = | |
1019 | &vif->pending_tx_info[pending_idx].callback_struct; | |
1020 | else if (likely(pending_idx != prev_pending_idx)) | |
1021 | vif->pending_tx_info[prev_pending_idx].callback_struct.ctx = | |
1022 | &(vif->pending_tx_info[pending_idx].callback_struct); | |
1023 | ||
1024 | vif->pending_tx_info[pending_idx].callback_struct.ctx = NULL; | |
1025 | prev_pending_idx = pending_idx; | |
1026 | ||
b3f980bd WL |
1027 | txp = &vif->pending_tx_info[pending_idx].req; |
1028 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 1029 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1030 | skb->len += txp->size; |
1031 | skb->data_len += txp->size; | |
1032 | skb->truesize += txp->size; | |
1033 | ||
f53c3fe8 | 1034 | /* Take an extra reference to offset network stack's put_page */ |
b3f980bd | 1035 | get_page(vif->mmap_pages[pending_idx]); |
f942dc25 | 1036 | } |
f53c3fe8 ZK |
1037 | /* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc |
1038 | * overlaps with "index", and "mapping" is not set. I think mapping | |
1039 | * should be set. If delivered to local stack, it would drop this | |
1040 | * skb in sk_filter unless the socket has the right to use it. | |
1041 | */ | |
1042 | skb->pfmemalloc = false; | |
f942dc25 IC |
1043 | } |
1044 | ||
7376419a | 1045 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
1046 | struct xen_netif_extra_info *extras, |
1047 | int work_to_do) | |
1048 | { | |
1049 | struct xen_netif_extra_info extra; | |
1050 | RING_IDX cons = vif->tx.req_cons; | |
1051 | ||
1052 | do { | |
1053 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 1054 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 1055 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1056 | return -EBADR; |
1057 | } | |
1058 | ||
1059 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1060 | sizeof(extra)); | |
1061 | if (unlikely(!extra.type || | |
1062 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1063 | vif->tx.req_cons = ++cons; | |
48856286 | 1064 | netdev_err(vif->dev, |
f942dc25 | 1065 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1066 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1067 | return -EINVAL; |
1068 | } | |
1069 | ||
1070 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1071 | vif->tx.req_cons = ++cons; | |
1072 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1073 | ||
1074 | return work_to_do; | |
1075 | } | |
1076 | ||
7376419a WL |
1077 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1078 | struct sk_buff *skb, | |
1079 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1080 | { |
1081 | if (!gso->u.gso.size) { | |
48856286 | 1082 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1083 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1084 | return -EINVAL; |
1085 | } | |
1086 | ||
a9468587 PD |
1087 | switch (gso->u.gso.type) { |
1088 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1089 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1090 | break; | |
1091 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1092 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1093 | break; | |
1094 | default: | |
48856286 | 1095 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1096 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1097 | return -EINVAL; |
1098 | } | |
1099 | ||
1100 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1101 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1102 | |
1103 | return 0; | |
1104 | } | |
1105 | ||
2eba61d5 PD |
1106 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) |
1107 | { | |
2721637c | 1108 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1109 | |
1110 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1111 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1112 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1113 | * recalculate the partial checksum. | |
1114 | */ | |
1115 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1116 | vif->rx_gso_checksum_fixup++; | |
1117 | skb->ip_summed = CHECKSUM_PARTIAL; | |
2721637c | 1118 | recalculate_partial_csum = true; |
2eba61d5 PD |
1119 | } |
1120 | ||
1121 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1122 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1123 | return 0; | |
1124 | ||
2721637c | 1125 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1126 | } |
1127 | ||
f942dc25 IC |
1128 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1129 | { | |
059dfa6a WL |
1130 | u64 now = get_jiffies_64(); |
1131 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1132 | msecs_to_jiffies(vif->credit_usec / 1000); |
1133 | ||
1134 | /* Timer could already be pending in rare cases. */ | |
1135 | if (timer_pending(&vif->credit_timeout)) | |
1136 | return true; | |
1137 | ||
1138 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1139 | if (time_after_eq64(now, next_credit)) { |
1140 | vif->credit_window_start = now; | |
f942dc25 IC |
1141 | tx_add_credit(vif); |
1142 | } | |
1143 | ||
1144 | /* Still too big to send right now? Set a callback. */ | |
1145 | if (size > vif->remaining_credit) { | |
1146 | vif->credit_timeout.data = | |
1147 | (unsigned long)vif; | |
1148 | vif->credit_timeout.function = | |
1149 | tx_credit_callback; | |
1150 | mod_timer(&vif->credit_timeout, | |
1151 | next_credit); | |
059dfa6a | 1152 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1153 | |
1154 | return true; | |
1155 | } | |
1156 | ||
1157 | return false; | |
1158 | } | |
1159 | ||
10574059 | 1160 | static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget) |
f942dc25 | 1161 | { |
f53c3fe8 | 1162 | struct gnttab_map_grant_ref *gop = vif->tx_map_ops, *request_gop; |
f942dc25 IC |
1163 | struct sk_buff *skb; |
1164 | int ret; | |
1165 | ||
121fa4b7 | 1166 | while (xenvif_tx_pending_slots_available(vif) && |
10574059 | 1167 | (skb_queue_len(&vif->tx_queue) < budget)) { |
f942dc25 | 1168 | struct xen_netif_tx_request txreq; |
37641494 | 1169 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1170 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1171 | u16 pending_idx; | |
1172 | RING_IDX idx; | |
1173 | int work_to_do; | |
1174 | unsigned int data_len; | |
1175 | pending_ring_idx_t index; | |
1176 | ||
48856286 IC |
1177 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1178 | XEN_NETIF_TX_RING_SIZE) { | |
1179 | netdev_err(vif->dev, | |
1180 | "Impossible number of requests. " | |
1181 | "req_prod %d, req_cons %d, size %ld\n", | |
1182 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1183 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1184 | xenvif_fatal_tx_err(vif); |
48856286 IC |
1185 | continue; |
1186 | } | |
1187 | ||
d9601a36 | 1188 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
b3f980bd WL |
1189 | if (!work_to_do) |
1190 | break; | |
f942dc25 IC |
1191 | |
1192 | idx = vif->tx.req_cons; | |
1193 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1194 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1195 | ||
1196 | /* Credit-based scheduling. */ | |
1197 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1198 | tx_credit_exceeded(vif, txreq.size)) |
1199 | break; | |
f942dc25 IC |
1200 | |
1201 | vif->remaining_credit -= txreq.size; | |
1202 | ||
1203 | work_to_do--; | |
1204 | vif->tx.req_cons = ++idx; | |
1205 | ||
1206 | memset(extras, 0, sizeof(extras)); | |
1207 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1208 | work_to_do = xenvif_get_extras(vif, extras, |
1209 | work_to_do); | |
f942dc25 | 1210 | idx = vif->tx.req_cons; |
48856286 | 1211 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1212 | break; |
f942dc25 IC |
1213 | } |
1214 | ||
7376419a | 1215 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1216 | if (unlikely(ret < 0)) |
b3f980bd | 1217 | break; |
48856286 | 1218 | |
f942dc25 IC |
1219 | idx += ret; |
1220 | ||
1221 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1222 | netdev_dbg(vif->dev, | |
1223 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1224 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1225 | break; |
f942dc25 IC |
1226 | } |
1227 | ||
1228 | /* No crossing a page as the payload mustn't fragment. */ | |
1229 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1230 | netdev_err(vif->dev, |
f942dc25 IC |
1231 | "txreq.offset: %x, size: %u, end: %lu\n", |
1232 | txreq.offset, txreq.size, | |
1233 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1234 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1235 | break; |
f942dc25 IC |
1236 | } |
1237 | ||
b3f980bd WL |
1238 | index = pending_index(vif->pending_cons); |
1239 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1240 | |
1241 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1242 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1243 | PKT_PROT_LEN : txreq.size; |
1244 | ||
e3377f36 | 1245 | skb = xenvif_alloc_skb(data_len); |
f942dc25 IC |
1246 | if (unlikely(skb == NULL)) { |
1247 | netdev_dbg(vif->dev, | |
1248 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1249 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1250 | break; |
1251 | } | |
1252 | ||
f942dc25 IC |
1253 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1254 | struct xen_netif_extra_info *gso; | |
1255 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1256 | ||
7376419a WL |
1257 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1258 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1259 | kfree_skb(skb); |
b3f980bd | 1260 | break; |
f942dc25 IC |
1261 | } |
1262 | } | |
1263 | ||
f53c3fe8 | 1264 | xenvif_tx_create_gop(vif, pending_idx, &txreq, gop); |
f942dc25 IC |
1265 | |
1266 | gop++; | |
1267 | ||
8f13dd96 | 1268 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1269 | |
1270 | __skb_put(skb, data_len); | |
1271 | ||
1272 | skb_shinfo(skb)->nr_frags = ret; | |
1273 | if (data_len < txreq.size) { | |
1274 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1275 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1276 | pending_idx); | |
f942dc25 | 1277 | } else { |
ea066ad1 IC |
1278 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1279 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1280 | } |
1281 | ||
b3f980bd | 1282 | vif->pending_cons++; |
f942dc25 | 1283 | |
7376419a | 1284 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1285 | if (request_gop == NULL) { |
1286 | kfree_skb(skb); | |
7376419a | 1287 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1288 | break; |
f942dc25 IC |
1289 | } |
1290 | gop = request_gop; | |
1291 | ||
b3f980bd | 1292 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1293 | |
f942dc25 | 1294 | vif->tx.req_cons = idx; |
f942dc25 | 1295 | |
f53c3fe8 | 1296 | if ((gop-vif->tx_map_ops) >= ARRAY_SIZE(vif->tx_map_ops)) |
f942dc25 IC |
1297 | break; |
1298 | } | |
1299 | ||
f53c3fe8 | 1300 | return gop - vif->tx_map_ops; |
f942dc25 IC |
1301 | } |
1302 | ||
e3377f36 ZK |
1303 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1304 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1305 | */ | |
1306 | static int xenvif_handle_frag_list(struct xenvif *vif, struct sk_buff *skb) | |
1307 | { | |
1308 | unsigned int offset = skb_headlen(skb); | |
1309 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
1310 | int i; | |
1311 | struct ubuf_info *uarg; | |
1312 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1313 | ||
1314 | vif->tx_zerocopy_sent += 2; | |
1315 | vif->tx_frag_overflow++; | |
1316 | ||
1317 | xenvif_fill_frags(vif, nskb); | |
1318 | /* Subtract frags size, we will correct it later */ | |
1319 | skb->truesize -= skb->data_len; | |
1320 | skb->len += nskb->len; | |
1321 | skb->data_len += nskb->len; | |
1322 | ||
1323 | /* create a brand new frags array and coalesce there */ | |
1324 | for (i = 0; offset < skb->len; i++) { | |
1325 | struct page *page; | |
1326 | unsigned int len; | |
1327 | ||
1328 | BUG_ON(i >= MAX_SKB_FRAGS); | |
1329 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
1330 | if (!page) { | |
1331 | int j; | |
1332 | skb->truesize += skb->data_len; | |
1333 | for (j = 0; j < i; j++) | |
1334 | put_page(frags[j].page.p); | |
1335 | return -ENOMEM; | |
1336 | } | |
1337 | ||
1338 | if (offset + PAGE_SIZE < skb->len) | |
1339 | len = PAGE_SIZE; | |
1340 | else | |
1341 | len = skb->len - offset; | |
1342 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1343 | BUG(); | |
1344 | ||
1345 | offset += len; | |
1346 | frags[i].page.p = page; | |
1347 | frags[i].page_offset = 0; | |
1348 | skb_frag_size_set(&frags[i], len); | |
1349 | } | |
1350 | /* swap out with old one */ | |
1351 | memcpy(skb_shinfo(skb)->frags, | |
1352 | frags, | |
1353 | i * sizeof(skb_frag_t)); | |
1354 | skb_shinfo(skb)->nr_frags = i; | |
1355 | skb->truesize += i * PAGE_SIZE; | |
1356 | ||
1357 | /* remove traces of mapped pages and frag_list */ | |
1358 | skb_frag_list_init(skb); | |
1359 | uarg = skb_shinfo(skb)->destructor_arg; | |
1360 | uarg->callback(uarg, true); | |
1361 | skb_shinfo(skb)->destructor_arg = NULL; | |
1362 | ||
1363 | skb_shinfo(nskb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1364 | kfree_skb(nskb); | |
1365 | ||
1366 | return 0; | |
1367 | } | |
b3f980bd | 1368 | |
10574059 | 1369 | static int xenvif_tx_submit(struct xenvif *vif) |
f942dc25 | 1370 | { |
f53c3fe8 | 1371 | struct gnttab_map_grant_ref *gop = vif->tx_map_ops; |
f942dc25 | 1372 | struct sk_buff *skb; |
b3f980bd | 1373 | int work_done = 0; |
f942dc25 | 1374 | |
10574059 | 1375 | while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) { |
f942dc25 | 1376 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1377 | u16 pending_idx; |
1378 | unsigned data_len; | |
1379 | ||
8f13dd96 | 1380 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
b3f980bd | 1381 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1382 | |
1383 | /* Check the remap error code. */ | |
7376419a | 1384 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1385 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1386 | skb_shinfo(skb)->nr_frags = 0; | |
1387 | kfree_skb(skb); | |
1388 | continue; | |
1389 | } | |
1390 | ||
1391 | data_len = skb->len; | |
1392 | memcpy(skb->data, | |
b3f980bd | 1393 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 | 1394 | data_len); |
f53c3fe8 | 1395 | vif->pending_tx_info[pending_idx].callback_struct.ctx = NULL; |
f942dc25 IC |
1396 | if (data_len < txp->size) { |
1397 | /* Append the packet payload as a fragment. */ | |
1398 | txp->offset += data_len; | |
1399 | txp->size -= data_len; | |
f53c3fe8 ZK |
1400 | skb_shinfo(skb)->destructor_arg = |
1401 | &vif->pending_tx_info[pending_idx].callback_struct; | |
f942dc25 IC |
1402 | } else { |
1403 | /* Schedule a response immediately. */ | |
f53c3fe8 | 1404 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
1405 | } |
1406 | ||
1407 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1408 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1409 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1410 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1411 | ||
7376419a | 1412 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1413 | |
e3377f36 ZK |
1414 | if (unlikely(skb_has_frag_list(skb))) { |
1415 | if (xenvif_handle_frag_list(vif, skb)) { | |
1416 | if (net_ratelimit()) | |
1417 | netdev_err(vif->dev, | |
1418 | "Not enough memory to consolidate frag_list!\n"); | |
1419 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1420 | kfree_skb(skb); | |
1421 | continue; | |
1422 | } | |
1423 | } | |
1424 | ||
2eba61d5 | 1425 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1426 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1427 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1428 | } | |
1429 | ||
1430 | skb->dev = vif->dev; | |
1431 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1432 | skb_reset_network_header(skb); |
f942dc25 IC |
1433 | |
1434 | if (checksum_setup(vif, skb)) { | |
1435 | netdev_dbg(vif->dev, | |
1436 | "Can't setup checksum in net_tx_action\n"); | |
f53c3fe8 ZK |
1437 | /* We have to set this flag to trigger the callback */ |
1438 | if (skb_shinfo(skb)->destructor_arg) | |
1439 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
f942dc25 IC |
1440 | kfree_skb(skb); |
1441 | continue; | |
1442 | } | |
1443 | ||
40893fd0 | 1444 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1445 | |
b89587a7 PD |
1446 | /* If the packet is GSO then we will have just set up the |
1447 | * transport header offset in checksum_setup so it's now | |
1448 | * straightforward to calculate gso_segs. | |
1449 | */ | |
1450 | if (skb_is_gso(skb)) { | |
1451 | int mss = skb_shinfo(skb)->gso_size; | |
1452 | int hdrlen = skb_transport_header(skb) - | |
1453 | skb_mac_header(skb) + | |
1454 | tcp_hdrlen(skb); | |
1455 | ||
1456 | skb_shinfo(skb)->gso_segs = | |
1457 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1458 | } | |
1459 | ||
f942dc25 IC |
1460 | vif->dev->stats.rx_bytes += skb->len; |
1461 | vif->dev->stats.rx_packets++; | |
1462 | ||
b3f980bd WL |
1463 | work_done++; |
1464 | ||
f53c3fe8 ZK |
1465 | /* Set this flag right before netif_receive_skb, otherwise |
1466 | * someone might think this packet already left netback, and | |
1467 | * do a skb_copy_ubufs while we are still in control of the | |
1468 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1469 | */ | |
1bb332af | 1470 | if (skb_shinfo(skb)->destructor_arg) { |
f53c3fe8 | 1471 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
1bb332af ZK |
1472 | vif->tx_zerocopy_sent++; |
1473 | } | |
f53c3fe8 | 1474 | |
b3f980bd | 1475 | netif_receive_skb(skb); |
f942dc25 | 1476 | } |
b3f980bd WL |
1477 | |
1478 | return work_done; | |
f942dc25 IC |
1479 | } |
1480 | ||
3e2234b3 ZK |
1481 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1482 | { | |
f53c3fe8 ZK |
1483 | unsigned long flags; |
1484 | pending_ring_idx_t index; | |
1485 | struct xenvif *vif = ubuf_to_vif(ubuf); | |
1486 | ||
1487 | /* This is the only place where we grab this lock, to protect callbacks | |
1488 | * from each other. | |
1489 | */ | |
1490 | spin_lock_irqsave(&vif->callback_lock, flags); | |
1491 | do { | |
1492 | u16 pending_idx = ubuf->desc; | |
1493 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
1494 | BUG_ON(vif->dealloc_prod - vif->dealloc_cons >= | |
1495 | MAX_PENDING_REQS); | |
1496 | index = pending_index(vif->dealloc_prod); | |
1497 | vif->dealloc_ring[index] = pending_idx; | |
1498 | /* Sync with xenvif_tx_dealloc_action: | |
1499 | * insert idx then incr producer. | |
1500 | */ | |
1501 | smp_wmb(); | |
1502 | vif->dealloc_prod++; | |
1503 | } while (ubuf); | |
1504 | wake_up(&vif->dealloc_wq); | |
1505 | spin_unlock_irqrestore(&vif->callback_lock, flags); | |
1506 | ||
1507 | if (RING_HAS_UNCONSUMED_REQUESTS(&vif->tx) && | |
1508 | xenvif_tx_pending_slots_available(vif)) { | |
1509 | local_bh_disable(); | |
1510 | napi_schedule(&vif->napi); | |
1511 | local_bh_enable(); | |
1512 | } | |
1bb332af ZK |
1513 | |
1514 | if (likely(zerocopy_success)) | |
1515 | vif->tx_zerocopy_success++; | |
1516 | else | |
1517 | vif->tx_zerocopy_fail++; | |
f53c3fe8 ZK |
1518 | } |
1519 | ||
1520 | static inline void xenvif_tx_dealloc_action(struct xenvif *vif) | |
1521 | { | |
1522 | struct gnttab_unmap_grant_ref *gop; | |
1523 | pending_ring_idx_t dc, dp; | |
1524 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1525 | unsigned int i = 0; | |
1526 | ||
1527 | dc = vif->dealloc_cons; | |
1528 | gop = vif->tx_unmap_ops; | |
1529 | ||
1530 | /* Free up any grants we have finished using */ | |
1531 | do { | |
1532 | dp = vif->dealloc_prod; | |
1533 | ||
1534 | /* Ensure we see all indices enqueued by all | |
1535 | * xenvif_zerocopy_callback(). | |
1536 | */ | |
1537 | smp_rmb(); | |
1538 | ||
1539 | while (dc != dp) { | |
1540 | BUG_ON(gop - vif->tx_unmap_ops > MAX_PENDING_REQS); | |
1541 | pending_idx = | |
1542 | vif->dealloc_ring[pending_index(dc++)]; | |
1543 | ||
1544 | pending_idx_release[gop-vif->tx_unmap_ops] = | |
1545 | pending_idx; | |
1546 | vif->pages_to_unmap[gop-vif->tx_unmap_ops] = | |
1547 | vif->mmap_pages[pending_idx]; | |
1548 | gnttab_set_unmap_op(gop, | |
1549 | idx_to_kaddr(vif, pending_idx), | |
1550 | GNTMAP_host_map, | |
1551 | vif->grant_tx_handle[pending_idx]); | |
1552 | /* Btw. already unmapped? */ | |
1553 | xenvif_grant_handle_reset(vif, pending_idx); | |
1554 | ++gop; | |
1555 | } | |
1556 | ||
1557 | } while (dp != vif->dealloc_prod); | |
1558 | ||
1559 | vif->dealloc_cons = dc; | |
1560 | ||
1561 | if (gop - vif->tx_unmap_ops > 0) { | |
1562 | int ret; | |
1563 | ret = gnttab_unmap_refs(vif->tx_unmap_ops, | |
1564 | NULL, | |
1565 | vif->pages_to_unmap, | |
1566 | gop - vif->tx_unmap_ops); | |
1567 | if (ret) { | |
1568 | netdev_err(vif->dev, "Unmap fail: nr_ops %x ret %d\n", | |
1569 | gop - vif->tx_unmap_ops, ret); | |
1570 | for (i = 0; i < gop - vif->tx_unmap_ops; ++i) { | |
1571 | if (gop[i].status != GNTST_okay) | |
1572 | netdev_err(vif->dev, | |
1573 | " host_addr: %llx handle: %x status: %d\n", | |
1574 | gop[i].host_addr, | |
1575 | gop[i].handle, | |
1576 | gop[i].status); | |
1577 | } | |
1578 | BUG(); | |
1579 | } | |
1580 | } | |
1581 | ||
1582 | for (i = 0; i < gop - vif->tx_unmap_ops; ++i) | |
1583 | xenvif_idx_release(vif, pending_idx_release[i], | |
1584 | XEN_NETIF_RSP_OKAY); | |
3e2234b3 ZK |
1585 | } |
1586 | ||
f53c3fe8 | 1587 | |
f942dc25 | 1588 | /* Called after netfront has transmitted */ |
7376419a | 1589 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1590 | { |
1591 | unsigned nr_gops; | |
f53c3fe8 | 1592 | int work_done, ret; |
f942dc25 | 1593 | |
b3f980bd WL |
1594 | if (unlikely(!tx_work_todo(vif))) |
1595 | return 0; | |
1596 | ||
10574059 | 1597 | nr_gops = xenvif_tx_build_gops(vif, budget); |
f942dc25 IC |
1598 | |
1599 | if (nr_gops == 0) | |
b3f980bd WL |
1600 | return 0; |
1601 | ||
f53c3fe8 ZK |
1602 | ret = gnttab_map_refs(vif->tx_map_ops, |
1603 | NULL, | |
1604 | vif->pages_to_map, | |
1605 | nr_gops); | |
1606 | BUG_ON(ret); | |
f942dc25 | 1607 | |
10574059 | 1608 | work_done = xenvif_tx_submit(vif); |
f942dc25 | 1609 | |
b3f980bd | 1610 | return work_done; |
f942dc25 IC |
1611 | } |
1612 | ||
7376419a WL |
1613 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1614 | u8 status) | |
f942dc25 | 1615 | { |
f942dc25 | 1616 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1617 | pending_ring_idx_t index; |
f53c3fe8 | 1618 | unsigned long flags; |
2810e5b9 | 1619 | |
62bad319 ZK |
1620 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
1621 | spin_lock_irqsave(&vif->response_lock, flags); | |
1622 | make_tx_response(vif, &pending_tx_info->req, status); | |
1623 | index = pending_index(vif->pending_prod); | |
1624 | vif->pending_ring[index] = pending_idx; | |
1625 | /* TX shouldn't use the index before we give it back here */ | |
1626 | mb(); | |
1627 | vif->pending_prod++; | |
1628 | spin_unlock_irqrestore(&vif->response_lock, flags); | |
f942dc25 IC |
1629 | } |
1630 | ||
2810e5b9 | 1631 | |
f942dc25 IC |
1632 | static void make_tx_response(struct xenvif *vif, |
1633 | struct xen_netif_tx_request *txp, | |
1634 | s8 st) | |
1635 | { | |
1636 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1637 | struct xen_netif_tx_response *resp; | |
1638 | int notify; | |
1639 | ||
1640 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1641 | resp->id = txp->id; | |
1642 | resp->status = st; | |
1643 | ||
1644 | if (txp->flags & XEN_NETTXF_extra_info) | |
1645 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1646 | ||
1647 | vif->tx.rsp_prod_pvt = ++i; | |
1648 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1649 | if (notify) | |
e1f00a69 | 1650 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1651 | } |
1652 | ||
1653 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1654 | u16 id, | |
1655 | s8 st, | |
1656 | u16 offset, | |
1657 | u16 size, | |
1658 | u16 flags) | |
1659 | { | |
1660 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1661 | struct xen_netif_rx_response *resp; | |
1662 | ||
1663 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1664 | resp->offset = offset; | |
1665 | resp->flags = flags; | |
1666 | resp->id = id; | |
1667 | resp->status = (s16)size; | |
1668 | if (st < 0) | |
1669 | resp->status = (s16)st; | |
1670 | ||
1671 | vif->rx.rsp_prod_pvt = ++i; | |
1672 | ||
1673 | return resp; | |
1674 | } | |
1675 | ||
f53c3fe8 ZK |
1676 | void xenvif_idx_unmap(struct xenvif *vif, u16 pending_idx) |
1677 | { | |
1678 | int ret; | |
1679 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1680 | ||
1681 | gnttab_set_unmap_op(&tx_unmap_op, | |
1682 | idx_to_kaddr(vif, pending_idx), | |
1683 | GNTMAP_host_map, | |
1684 | vif->grant_tx_handle[pending_idx]); | |
1685 | /* Btw. already unmapped? */ | |
1686 | xenvif_grant_handle_reset(vif, pending_idx); | |
1687 | ||
1688 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
1689 | &vif->mmap_pages[pending_idx], 1); | |
1690 | BUG_ON(ret); | |
1691 | ||
1692 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); | |
1693 | } | |
1694 | ||
b3f980bd | 1695 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1696 | { |
9ab9831b ZK |
1697 | return !skb_queue_empty(&vif->rx_queue) && |
1698 | xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots); | |
f942dc25 IC |
1699 | } |
1700 | ||
b3f980bd | 1701 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1702 | { |
1703 | ||
b3f980bd | 1704 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) && |
121fa4b7 | 1705 | xenvif_tx_pending_slots_available(vif)) |
f942dc25 IC |
1706 | return 1; |
1707 | ||
1708 | return 0; | |
1709 | } | |
1710 | ||
f53c3fe8 ZK |
1711 | static inline bool tx_dealloc_work_todo(struct xenvif *vif) |
1712 | { | |
1713 | return vif->dealloc_cons != vif->dealloc_prod; | |
1714 | } | |
1715 | ||
7376419a | 1716 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1717 | { |
c9d63699 DV |
1718 | if (vif->tx.sring) |
1719 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1720 | vif->tx.sring); | |
1721 | if (vif->rx.sring) | |
1722 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1723 | vif->rx.sring); | |
f942dc25 IC |
1724 | } |
1725 | ||
7376419a WL |
1726 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1727 | grant_ref_t tx_ring_ref, | |
1728 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1729 | { |
c9d63699 | 1730 | void *addr; |
f942dc25 IC |
1731 | struct xen_netif_tx_sring *txs; |
1732 | struct xen_netif_rx_sring *rxs; | |
1733 | ||
1734 | int err = -ENOMEM; | |
1735 | ||
c9d63699 DV |
1736 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1737 | tx_ring_ref, &addr); | |
1738 | if (err) | |
f942dc25 IC |
1739 | goto err; |
1740 | ||
c9d63699 | 1741 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1742 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1743 | ||
c9d63699 DV |
1744 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1745 | rx_ring_ref, &addr); | |
1746 | if (err) | |
f942dc25 | 1747 | goto err; |
f942dc25 | 1748 | |
c9d63699 | 1749 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1750 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1751 | ||
1752 | return 0; | |
1753 | ||
1754 | err: | |
7376419a | 1755 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1756 | return err; |
1757 | } | |
1758 | ||
ca2f09f2 PD |
1759 | void xenvif_stop_queue(struct xenvif *vif) |
1760 | { | |
1761 | if (!vif->can_queue) | |
1762 | return; | |
1763 | ||
1764 | netif_stop_queue(vif->dev); | |
1765 | } | |
1766 | ||
1767 | static void xenvif_start_queue(struct xenvif *vif) | |
1768 | { | |
1769 | if (xenvif_schedulable(vif)) | |
1770 | netif_wake_queue(vif->dev); | |
1771 | } | |
1772 | ||
121fa4b7 | 1773 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd WL |
1774 | { |
1775 | struct xenvif *vif = data; | |
ca2f09f2 | 1776 | struct sk_buff *skb; |
b3f980bd WL |
1777 | |
1778 | while (!kthread_should_stop()) { | |
1779 | wait_event_interruptible(vif->wq, | |
1780 | rx_work_todo(vif) || | |
1781 | kthread_should_stop()); | |
1782 | if (kthread_should_stop()) | |
1783 | break; | |
1784 | ||
ca2f09f2 | 1785 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 1786 | xenvif_rx_action(vif); |
b3f980bd | 1787 | |
ca2f09f2 PD |
1788 | if (skb_queue_empty(&vif->rx_queue) && |
1789 | netif_queue_stopped(vif->dev)) | |
1790 | xenvif_start_queue(vif); | |
1791 | ||
b3f980bd WL |
1792 | cond_resched(); |
1793 | } | |
1794 | ||
ca2f09f2 PD |
1795 | /* Bin any remaining skbs */ |
1796 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | |
1797 | dev_kfree_skb(skb); | |
1798 | ||
b3f980bd WL |
1799 | return 0; |
1800 | } | |
1801 | ||
f53c3fe8 ZK |
1802 | int xenvif_dealloc_kthread(void *data) |
1803 | { | |
1804 | struct xenvif *vif = data; | |
1805 | ||
1806 | while (!kthread_should_stop()) { | |
1807 | wait_event_interruptible(vif->dealloc_wq, | |
1808 | tx_dealloc_work_todo(vif) || | |
1809 | kthread_should_stop()); | |
1810 | if (kthread_should_stop()) | |
1811 | break; | |
1812 | ||
1813 | xenvif_tx_dealloc_action(vif); | |
1814 | cond_resched(); | |
1815 | } | |
1816 | ||
1817 | /* Unmap anything remaining*/ | |
1818 | if (tx_dealloc_work_todo(vif)) | |
1819 | xenvif_tx_dealloc_action(vif); | |
1820 | ||
1821 | return 0; | |
1822 | } | |
1823 | ||
f942dc25 IC |
1824 | static int __init netback_init(void) |
1825 | { | |
f942dc25 | 1826 | int rc = 0; |
f942dc25 | 1827 | |
2a14b244 | 1828 | if (!xen_domain()) |
f942dc25 IC |
1829 | return -ENODEV; |
1830 | ||
37641494 | 1831 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1832 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1833 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1834 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1835 | } |
1836 | ||
f942dc25 IC |
1837 | rc = xenvif_xenbus_init(); |
1838 | if (rc) | |
1839 | goto failed_init; | |
1840 | ||
1841 | return 0; | |
1842 | ||
1843 | failed_init: | |
f942dc25 | 1844 | return rc; |
f942dc25 IC |
1845 | } |
1846 | ||
1847 | module_init(netback_init); | |
1848 | ||
b103f358 WL |
1849 | static void __exit netback_fini(void) |
1850 | { | |
b103f358 | 1851 | xenvif_xenbus_fini(); |
b103f358 WL |
1852 | } |
1853 | module_exit(netback_fini); | |
1854 | ||
f942dc25 | 1855 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1856 | MODULE_ALIAS("xen-backend:vif"); |