Commit | Line | Data |
---|---|---|
f942dc25 IC |
1 | /* |
2 | * Back-end of the driver for virtual network devices. This portion of the | |
3 | * driver exports a 'unified' network-device interface that can be accessed | |
4 | * by any operating system that implements a compatible front end. A | |
5 | * reference front-end implementation can be found in: | |
6 | * drivers/net/xen-netfront.c | |
7 | * | |
8 | * Copyright (c) 2002-2005, K A Fraser | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License version 2 | |
12 | * as published by the Free Software Foundation; or, when distributed | |
13 | * separately from the Linux kernel or incorporated into other | |
14 | * software packages, subject to the following license: | |
15 | * | |
16 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
17 | * of this source file (the "Software"), to deal in the Software without | |
18 | * restriction, including without limitation the rights to use, copy, modify, | |
19 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
20 | * and to permit persons to whom the Software is furnished to do so, subject to | |
21 | * the following conditions: | |
22 | * | |
23 | * The above copyright notice and this permission notice shall be included in | |
24 | * all copies or substantial portions of the Software. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
27 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
28 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
29 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
30 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
31 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
32 | * IN THE SOFTWARE. | |
33 | */ | |
34 | ||
35 | #include "common.h" | |
36 | ||
37 | #include <linux/kthread.h> | |
38 | #include <linux/if_vlan.h> | |
39 | #include <linux/udp.h> | |
e3377f36 | 40 | #include <linux/highmem.h> |
f942dc25 IC |
41 | |
42 | #include <net/tcp.h> | |
43 | ||
ca981633 | 44 | #include <xen/xen.h> |
f942dc25 IC |
45 | #include <xen/events.h> |
46 | #include <xen/interface/memory.h> | |
47 | ||
48 | #include <asm/xen/hypercall.h> | |
49 | #include <asm/xen/page.h> | |
50 | ||
e1f00a69 WL |
51 | /* Provide an option to disable split event channels at load time as |
52 | * event channels are limited resource. Split event channels are | |
53 | * enabled by default. | |
54 | */ | |
55 | bool separate_tx_rx_irq = 1; | |
56 | module_param(separate_tx_rx_irq, bool, 0644); | |
57 | ||
09350788 | 58 | /* When guest ring is filled up, qdisc queues the packets for us, but we have |
0e59a4a5 | 59 | * to timeout them, otherwise other guests' packets can get stuck there |
09350788 ZK |
60 | */ |
61 | unsigned int rx_drain_timeout_msecs = 10000; | |
62 | module_param(rx_drain_timeout_msecs, uint, 0444); | |
63 | unsigned int rx_drain_timeout_jiffies; | |
64 | ||
2810e5b9 WL |
65 | /* |
66 | * This is the maximum slots a skb can have. If a guest sends a skb | |
67 | * which exceeds this limit it is considered malicious. | |
68 | */ | |
37641494 WL |
69 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
70 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
71 | module_param(fatal_skb_slots, uint, 0444); | |
72 | ||
7376419a WL |
73 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
74 | u8 status); | |
75 | ||
f942dc25 IC |
76 | static void make_tx_response(struct xenvif *vif, |
77 | struct xen_netif_tx_request *txp, | |
78 | s8 st); | |
b3f980bd WL |
79 | |
80 | static inline int tx_work_todo(struct xenvif *vif); | |
81 | static inline int rx_work_todo(struct xenvif *vif); | |
82 | ||
f942dc25 IC |
83 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
84 | u16 id, | |
85 | s8 st, | |
86 | u16 offset, | |
87 | u16 size, | |
88 | u16 flags); | |
89 | ||
b3f980bd | 90 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 91 | u16 idx) |
f942dc25 | 92 | { |
b3f980bd | 93 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
94 | } |
95 | ||
b3f980bd | 96 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 97 | u16 idx) |
f942dc25 | 98 | { |
b3f980bd | 99 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
100 | } |
101 | ||
7aceb47a ZK |
102 | #define callback_param(vif, pending_idx) \ |
103 | (vif->pending_tx_info[pending_idx].callback_struct) | |
104 | ||
f53c3fe8 ZK |
105 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
106 | */ | |
3e2234b3 ZK |
107 | static inline struct xenvif* ubuf_to_vif(struct ubuf_info *ubuf) |
108 | { | |
f53c3fe8 ZK |
109 | u16 pending_idx = ubuf->desc; |
110 | struct pending_tx_info *temp = | |
111 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
112 | return container_of(temp - pending_idx, | |
113 | struct xenvif, | |
114 | pending_tx_info[0]); | |
3e2234b3 | 115 | } |
f53c3fe8 | 116 | |
2eba61d5 PD |
117 | /* This is a miniumum size for the linear area to avoid lots of |
118 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
119 | * value 128 was chosen as it covers all IPv4 and most likely | |
120 | * IPv6 headers. | |
f942dc25 | 121 | */ |
2eba61d5 | 122 | #define PKT_PROT_LEN 128 |
f942dc25 | 123 | |
ea066ad1 IC |
124 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
125 | { | |
126 | return (u16)frag->page_offset; | |
127 | } | |
128 | ||
129 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
130 | { | |
131 | frag->page_offset = pending_idx; | |
132 | } | |
133 | ||
f942dc25 IC |
134 | static inline pending_ring_idx_t pending_index(unsigned i) |
135 | { | |
136 | return i & (MAX_PENDING_REQS-1); | |
137 | } | |
138 | ||
ca2f09f2 | 139 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
f942dc25 | 140 | { |
ca2f09f2 | 141 | RING_IDX prod, cons; |
f942dc25 | 142 | |
ca2f09f2 PD |
143 | do { |
144 | prod = vif->rx.sring->req_prod; | |
145 | cons = vif->rx.req_cons; | |
f942dc25 | 146 | |
ca2f09f2 PD |
147 | if (prod - cons >= needed) |
148 | return true; | |
f942dc25 | 149 | |
ca2f09f2 | 150 | vif->rx.sring->req_event = prod + 1; |
f942dc25 | 151 | |
ca2f09f2 PD |
152 | /* Make sure event is visible before we check prod |
153 | * again. | |
154 | */ | |
155 | mb(); | |
156 | } while (vif->rx.sring->req_prod != prod); | |
f942dc25 | 157 | |
ca2f09f2 | 158 | return false; |
f942dc25 IC |
159 | } |
160 | ||
161 | /* | |
162 | * Returns true if we should start a new receive buffer instead of | |
163 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
164 | * bytes. | |
165 | */ | |
166 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
167 | { | |
168 | /* simple case: we have completely filled the current buffer. */ | |
169 | if (offset == MAX_BUFFER_OFFSET) | |
170 | return true; | |
171 | ||
172 | /* | |
173 | * complex case: start a fresh buffer if the current frag | |
174 | * would overflow the current buffer but only if: | |
175 | * (i) this frag would fit completely in the next buffer | |
176 | * and (ii) there is already some data in the current buffer | |
177 | * and (iii) this is not the head buffer. | |
178 | * | |
179 | * Where: | |
180 | * - (i) stops us splitting a frag into two copies | |
181 | * unless the frag is too large for a single buffer. | |
182 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
183 | * - (iii) stops us leaving the first buffer | |
184 | * empty. Strictly speaking this is already covered | |
185 | * by (ii) but is explicitly checked because | |
186 | * netfront relies on the first buffer being | |
187 | * non-empty and can crash otherwise. | |
188 | * | |
189 | * This means we will effectively linearise small | |
190 | * frags but do not needlessly split large buffers | |
191 | * into multiple copies tend to give large frags their | |
192 | * own buffers as before. | |
193 | */ | |
0576eddf PD |
194 | BUG_ON(size > MAX_BUFFER_OFFSET); |
195 | if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head) | |
f942dc25 IC |
196 | return true; |
197 | ||
198 | return false; | |
199 | } | |
200 | ||
f942dc25 IC |
201 | struct netrx_pending_operations { |
202 | unsigned copy_prod, copy_cons; | |
203 | unsigned meta_prod, meta_cons; | |
204 | struct gnttab_copy *copy; | |
b3f980bd | 205 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
206 | int copy_off; |
207 | grant_ref_t copy_gref; | |
208 | }; | |
209 | ||
b3f980bd WL |
210 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
211 | struct netrx_pending_operations *npo) | |
f942dc25 | 212 | { |
b3f980bd | 213 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
214 | struct xen_netif_rx_request *req; |
215 | ||
216 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
217 | ||
218 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 219 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
220 | meta->gso_size = 0; |
221 | meta->size = 0; | |
222 | meta->id = req->id; | |
223 | ||
224 | npo->copy_off = 0; | |
225 | npo->copy_gref = req->gref; | |
226 | ||
227 | return meta; | |
228 | } | |
229 | ||
33bc801d WL |
230 | /* |
231 | * Set up the grant operations for this fragment. If it's a flipping | |
232 | * interface, we also set up the unmap request from here. | |
233 | */ | |
7376419a WL |
234 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
235 | struct netrx_pending_operations *npo, | |
236 | struct page *page, unsigned long size, | |
3e2234b3 ZK |
237 | unsigned long offset, int *head, |
238 | struct xenvif *foreign_vif, | |
239 | grant_ref_t foreign_gref) | |
f942dc25 IC |
240 | { |
241 | struct gnttab_copy *copy_gop; | |
b3f980bd | 242 | struct xenvif_rx_meta *meta; |
f942dc25 | 243 | unsigned long bytes; |
5bd07670 | 244 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
245 | |
246 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 247 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
248 | |
249 | meta = npo->meta + npo->meta_prod - 1; | |
250 | ||
6a8ed462 IC |
251 | /* Skip unused frames from start of page */ |
252 | page += offset >> PAGE_SHIFT; | |
253 | offset &= ~PAGE_MASK; | |
254 | ||
f942dc25 | 255 | while (size > 0) { |
6a8ed462 | 256 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
257 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
258 | ||
6a8ed462 IC |
259 | bytes = PAGE_SIZE - offset; |
260 | ||
261 | if (bytes > size) | |
262 | bytes = size; | |
263 | ||
33bc801d | 264 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
265 | /* |
266 | * Netfront requires there to be some data in the head | |
267 | * buffer. | |
268 | */ | |
33bc801d | 269 | BUG_ON(*head); |
f942dc25 IC |
270 | |
271 | meta = get_next_rx_buffer(vif, npo); | |
272 | } | |
273 | ||
f942dc25 IC |
274 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
275 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
276 | ||
277 | copy_gop = npo->copy + npo->copy_prod++; | |
278 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
279 | copy_gop->len = bytes; |
280 | ||
3e2234b3 ZK |
281 | if (foreign_vif) { |
282 | copy_gop->source.domid = foreign_vif->domid; | |
283 | copy_gop->source.u.ref = foreign_gref; | |
284 | copy_gop->flags |= GNTCOPY_source_gref; | |
285 | } else { | |
286 | copy_gop->source.domid = DOMID_SELF; | |
287 | copy_gop->source.u.gmfn = | |
288 | virt_to_mfn(page_address(page)); | |
289 | } | |
f942dc25 | 290 | copy_gop->source.offset = offset; |
f942dc25 | 291 | |
b3f980bd | 292 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
293 | copy_gop->dest.offset = npo->copy_off; |
294 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
295 | |
296 | npo->copy_off += bytes; | |
297 | meta->size += bytes; | |
298 | ||
299 | offset += bytes; | |
300 | size -= bytes; | |
301 | ||
6a8ed462 IC |
302 | /* Next frame */ |
303 | if (offset == PAGE_SIZE && size) { | |
304 | BUG_ON(!PageCompound(page)); | |
305 | page++; | |
306 | offset = 0; | |
307 | } | |
308 | ||
f942dc25 | 309 | /* Leave a gap for the GSO descriptor. */ |
5bd07670 AL |
310 | if (skb_is_gso(skb)) { |
311 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
312 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
313 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
314 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
315 | } | |
82cada22 PD |
316 | |
317 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
318 | vif->rx.req_cons++; |
319 | ||
33bc801d | 320 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
321 | |
322 | } | |
323 | } | |
324 | ||
325 | /* | |
326 | * Prepare an SKB to be transmitted to the frontend. | |
327 | * | |
328 | * This function is responsible for allocating grant operations, meta | |
329 | * structures, etc. | |
330 | * | |
331 | * It returns the number of meta structures consumed. The number of | |
332 | * ring slots used is always equal to the number of meta slots used | |
333 | * plus the number of GSO descriptors used. Currently, we use either | |
334 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
335 | * frontend-side LRO). | |
336 | */ | |
7376419a WL |
337 | static int xenvif_gop_skb(struct sk_buff *skb, |
338 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
339 | { |
340 | struct xenvif *vif = netdev_priv(skb->dev); | |
341 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
342 | int i; | |
343 | struct xen_netif_rx_request *req; | |
b3f980bd | 344 | struct xenvif_rx_meta *meta; |
f942dc25 | 345 | unsigned char *data; |
33bc801d | 346 | int head = 1; |
f942dc25 | 347 | int old_meta_prod; |
82cada22 | 348 | int gso_type; |
3e2234b3 ZK |
349 | struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg; |
350 | grant_ref_t foreign_grefs[MAX_SKB_FRAGS]; | |
351 | struct xenvif *foreign_vif = NULL; | |
f942dc25 IC |
352 | |
353 | old_meta_prod = npo->meta_prod; | |
354 | ||
5bd07670 AL |
355 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
356 | if (skb_is_gso(skb)) { | |
357 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
358 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
359 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
360 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
361 | } |
362 | ||
f942dc25 | 363 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 364 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
365 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
366 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 367 | meta->gso_type = gso_type; |
5bd07670 | 368 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 IC |
369 | meta->size = 0; |
370 | meta->id = req->id; | |
371 | } | |
372 | ||
373 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
374 | meta = npo->meta + npo->meta_prod++; | |
375 | ||
82cada22 PD |
376 | if ((1 << gso_type) & vif->gso_mask) { |
377 | meta->gso_type = gso_type; | |
5bd07670 | 378 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
379 | } else { |
380 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 381 | meta->gso_size = 0; |
82cada22 | 382 | } |
f942dc25 IC |
383 | |
384 | meta->size = 0; | |
385 | meta->id = req->id; | |
386 | npo->copy_off = 0; | |
387 | npo->copy_gref = req->gref; | |
388 | ||
3e2234b3 ZK |
389 | if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) && |
390 | (ubuf->callback == &xenvif_zerocopy_callback)) { | |
391 | int i = 0; | |
392 | foreign_vif = ubuf_to_vif(ubuf); | |
393 | ||
394 | do { | |
395 | u16 pending_idx = ubuf->desc; | |
396 | foreign_grefs[i++] = | |
397 | foreign_vif->pending_tx_info[pending_idx].req.gref; | |
398 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
399 | } while (ubuf); | |
400 | } | |
401 | ||
f942dc25 IC |
402 | data = skb->data; |
403 | while (data < skb_tail_pointer(skb)) { | |
404 | unsigned int offset = offset_in_page(data); | |
405 | unsigned int len = PAGE_SIZE - offset; | |
406 | ||
407 | if (data + len > skb_tail_pointer(skb)) | |
408 | len = skb_tail_pointer(skb) - data; | |
409 | ||
7376419a | 410 | xenvif_gop_frag_copy(vif, skb, npo, |
3e2234b3 ZK |
411 | virt_to_page(data), len, offset, &head, |
412 | NULL, | |
413 | 0); | |
f942dc25 IC |
414 | data += len; |
415 | } | |
416 | ||
417 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
418 | xenvif_gop_frag_copy(vif, skb, npo, |
419 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
420 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
421 | skb_shinfo(skb)->frags[i].page_offset, | |
3e2234b3 ZK |
422 | &head, |
423 | foreign_vif, | |
424 | foreign_grefs[i]); | |
f942dc25 IC |
425 | } |
426 | ||
427 | return npo->meta_prod - old_meta_prod; | |
428 | } | |
429 | ||
430 | /* | |
7376419a | 431 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
432 | * used to set up the operations on the top of |
433 | * netrx_pending_operations, which have since been done. Check that | |
434 | * they didn't give any errors and advance over them. | |
435 | */ | |
7376419a WL |
436 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
437 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
438 | { |
439 | struct gnttab_copy *copy_op; | |
440 | int status = XEN_NETIF_RSP_OKAY; | |
441 | int i; | |
442 | ||
443 | for (i = 0; i < nr_meta_slots; i++) { | |
444 | copy_op = npo->copy + npo->copy_cons++; | |
445 | if (copy_op->status != GNTST_okay) { | |
446 | netdev_dbg(vif->dev, | |
447 | "Bad status %d from copy to DOM%d.\n", | |
448 | copy_op->status, vif->domid); | |
449 | status = XEN_NETIF_RSP_ERROR; | |
450 | } | |
451 | } | |
452 | ||
453 | return status; | |
454 | } | |
455 | ||
7376419a WL |
456 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
457 | struct xenvif_rx_meta *meta, | |
458 | int nr_meta_slots) | |
f942dc25 IC |
459 | { |
460 | int i; | |
461 | unsigned long offset; | |
462 | ||
463 | /* No fragments used */ | |
464 | if (nr_meta_slots <= 1) | |
465 | return; | |
466 | ||
467 | nr_meta_slots--; | |
468 | ||
469 | for (i = 0; i < nr_meta_slots; i++) { | |
470 | int flags; | |
471 | if (i == nr_meta_slots - 1) | |
472 | flags = 0; | |
473 | else | |
474 | flags = XEN_NETRXF_more_data; | |
475 | ||
476 | offset = 0; | |
477 | make_rx_response(vif, meta[i].id, status, offset, | |
478 | meta[i].size, flags); | |
479 | } | |
480 | } | |
481 | ||
8f13dd96 | 482 | struct xenvif_rx_cb { |
33bc801d WL |
483 | int meta_slots_used; |
484 | }; | |
485 | ||
8f13dd96 ZK |
486 | #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) |
487 | ||
ca2f09f2 | 488 | void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
489 | { |
490 | wake_up(&vif->wq); | |
491 | } | |
492 | ||
ca2f09f2 | 493 | static void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 494 | { |
f942dc25 | 495 | s8 status; |
e1f00a69 | 496 | u16 flags; |
f942dc25 IC |
497 | struct xen_netif_rx_response *resp; |
498 | struct sk_buff_head rxq; | |
499 | struct sk_buff *skb; | |
500 | LIST_HEAD(notify); | |
501 | int ret; | |
f942dc25 | 502 | unsigned long offset; |
11b57f90 | 503 | bool need_to_notify = false; |
f942dc25 IC |
504 | |
505 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
506 | .copy = vif->grant_copy_op, |
507 | .meta = vif->meta, | |
f942dc25 IC |
508 | }; |
509 | ||
510 | skb_queue_head_init(&rxq); | |
511 | ||
b3f980bd | 512 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
9ab9831b | 513 | RING_IDX max_slots_needed; |
1425c7a4 PD |
514 | RING_IDX old_req_cons; |
515 | RING_IDX ring_slots_used; | |
ca2f09f2 PD |
516 | int i; |
517 | ||
518 | /* We need a cheap worse case estimate for the number of | |
519 | * slots we'll use. | |
520 | */ | |
521 | ||
522 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
523 | skb_headlen(skb), | |
524 | PAGE_SIZE); | |
525 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
526 | unsigned int size; | |
a02eb473 PD |
527 | unsigned int offset; |
528 | ||
ca2f09f2 | 529 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
a02eb473 PD |
530 | offset = skb_shinfo(skb)->frags[i].page_offset; |
531 | ||
532 | /* For a worse-case estimate we need to factor in | |
533 | * the fragment page offset as this will affect the | |
534 | * number of times xenvif_gop_frag_copy() will | |
535 | * call start_new_rx_buffer(). | |
536 | */ | |
537 | max_slots_needed += DIV_ROUND_UP(offset + size, | |
538 | PAGE_SIZE); | |
ca2f09f2 | 539 | } |
a02eb473 PD |
540 | |
541 | /* To avoid the estimate becoming too pessimal for some | |
542 | * frontends that limit posted rx requests, cap the estimate | |
543 | * at MAX_SKB_FRAGS. | |
544 | */ | |
545 | if (max_slots_needed > MAX_SKB_FRAGS) | |
546 | max_slots_needed = MAX_SKB_FRAGS; | |
547 | ||
548 | /* We may need one more slot for GSO metadata */ | |
5bd07670 AL |
549 | if (skb_is_gso(skb) && |
550 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
551 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | |
ca2f09f2 PD |
552 | max_slots_needed++; |
553 | ||
554 | /* If the skb may not fit then bail out now */ | |
555 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | |
556 | skb_queue_head(&vif->rx_queue, skb); | |
11b57f90 | 557 | need_to_notify = true; |
9ab9831b | 558 | vif->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 559 | break; |
9ab9831b ZK |
560 | } else |
561 | vif->rx_last_skb_slots = 0; | |
f942dc25 | 562 | |
1425c7a4 | 563 | old_req_cons = vif->rx.req_cons; |
8f13dd96 | 564 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo); |
1425c7a4 PD |
565 | ring_slots_used = vif->rx.req_cons - old_req_cons; |
566 | ||
567 | BUG_ON(ring_slots_used > max_slots_needed); | |
f942dc25 IC |
568 | |
569 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
570 | } |
571 | ||
b3f980bd | 572 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
573 | |
574 | if (!npo.copy_prod) | |
ca2f09f2 | 575 | goto done; |
f942dc25 | 576 | |
ac3d5ac2 | 577 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
b3f980bd | 578 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
579 | |
580 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 581 | |
82cada22 PD |
582 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
583 | vif->gso_prefix_mask) { | |
f942dc25 | 584 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 585 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
586 | |
587 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
588 | ||
b3f980bd WL |
589 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
590 | resp->id = vif->meta[npo.meta_cons].id; | |
8f13dd96 | 591 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
592 | |
593 | npo.meta_cons++; | |
8f13dd96 | 594 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
595 | } |
596 | ||
597 | ||
598 | vif->dev->stats.tx_bytes += skb->len; | |
599 | vif->dev->stats.tx_packets++; | |
600 | ||
8f13dd96 ZK |
601 | status = xenvif_check_gop(vif, |
602 | XENVIF_RX_CB(skb)->meta_slots_used, | |
603 | &npo); | |
f942dc25 | 604 | |
8f13dd96 | 605 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
606 | flags = 0; |
607 | else | |
608 | flags = XEN_NETRXF_more_data; | |
609 | ||
610 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
611 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
612 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
613 | /* remote but checksummed. */ | |
614 | flags |= XEN_NETRXF_data_validated; | |
615 | ||
616 | offset = 0; | |
b3f980bd | 617 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 618 | status, offset, |
b3f980bd | 619 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
620 | flags); |
621 | ||
82cada22 PD |
622 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
623 | vif->gso_mask) { | |
f942dc25 IC |
624 | struct xen_netif_extra_info *gso = |
625 | (struct xen_netif_extra_info *) | |
626 | RING_GET_RESPONSE(&vif->rx, | |
627 | vif->rx.rsp_prod_pvt++); | |
628 | ||
629 | resp->flags |= XEN_NETRXF_extra_info; | |
630 | ||
82cada22 | 631 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 632 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
633 | gso->u.gso.pad = 0; |
634 | gso->u.gso.features = 0; | |
635 | ||
636 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
637 | gso->flags = 0; | |
638 | } | |
639 | ||
7376419a WL |
640 | xenvif_add_frag_responses(vif, status, |
641 | vif->meta + npo.meta_cons + 1, | |
8f13dd96 | 642 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 IC |
643 | |
644 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 645 | |
11b57f90 | 646 | need_to_notify |= !!ret; |
b3f980bd | 647 | |
8f13dd96 | 648 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
649 | dev_kfree_skb(skb); |
650 | } | |
651 | ||
ca2f09f2 | 652 | done: |
b3f980bd | 653 | if (need_to_notify) |
e1f00a69 | 654 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
655 | } |
656 | ||
7376419a | 657 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
658 | { |
659 | int more_to_do; | |
660 | ||
661 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
662 | ||
663 | if (more_to_do) | |
b3f980bd | 664 | napi_schedule(&vif->napi); |
f942dc25 IC |
665 | } |
666 | ||
667 | static void tx_add_credit(struct xenvif *vif) | |
668 | { | |
669 | unsigned long max_burst, max_credit; | |
670 | ||
671 | /* | |
672 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
673 | * Otherwise the interface can seize up due to insufficient credit. | |
674 | */ | |
675 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
676 | max_burst = min(max_burst, 131072UL); | |
677 | max_burst = max(max_burst, vif->credit_bytes); | |
678 | ||
679 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
680 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
681 | if (max_credit < vif->remaining_credit) | |
682 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
683 | ||
684 | vif->remaining_credit = min(max_credit, max_burst); | |
685 | } | |
686 | ||
687 | static void tx_credit_callback(unsigned long data) | |
688 | { | |
689 | struct xenvif *vif = (struct xenvif *)data; | |
690 | tx_add_credit(vif); | |
7376419a | 691 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
692 | } |
693 | ||
7376419a WL |
694 | static void xenvif_tx_err(struct xenvif *vif, |
695 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
696 | { |
697 | RING_IDX cons = vif->tx.req_cons; | |
f53c3fe8 | 698 | unsigned long flags; |
f942dc25 IC |
699 | |
700 | do { | |
f53c3fe8 | 701 | spin_lock_irqsave(&vif->response_lock, flags); |
f942dc25 | 702 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); |
f53c3fe8 | 703 | spin_unlock_irqrestore(&vif->response_lock, flags); |
b9149729 | 704 | if (cons == end) |
f942dc25 IC |
705 | break; |
706 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
707 | } while (1); | |
708 | vif->tx.req_cons = cons; | |
f942dc25 IC |
709 | } |
710 | ||
7376419a | 711 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
712 | { |
713 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 WL |
714 | vif->disabled = true; |
715 | xenvif_kick_thread(vif); | |
48856286 IC |
716 | } |
717 | ||
7376419a WL |
718 | static int xenvif_count_requests(struct xenvif *vif, |
719 | struct xen_netif_tx_request *first, | |
720 | struct xen_netif_tx_request *txp, | |
721 | int work_to_do) | |
f942dc25 IC |
722 | { |
723 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
724 | int slots = 0; |
725 | int drop_err = 0; | |
59ccb4eb | 726 | int more_data; |
f942dc25 IC |
727 | |
728 | if (!(first->flags & XEN_NETTXF_more_data)) | |
729 | return 0; | |
730 | ||
731 | do { | |
59ccb4eb WL |
732 | struct xen_netif_tx_request dropped_tx = { 0 }; |
733 | ||
2810e5b9 WL |
734 | if (slots >= work_to_do) { |
735 | netdev_err(vif->dev, | |
736 | "Asked for %d slots but exceeds this limit\n", | |
737 | work_to_do); | |
7376419a | 738 | xenvif_fatal_tx_err(vif); |
35876b5f | 739 | return -ENODATA; |
f942dc25 IC |
740 | } |
741 | ||
2810e5b9 WL |
742 | /* This guest is really using too many slots and |
743 | * considered malicious. | |
744 | */ | |
37641494 | 745 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
746 | netdev_err(vif->dev, |
747 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 748 | slots, fatal_skb_slots); |
7376419a | 749 | xenvif_fatal_tx_err(vif); |
35876b5f | 750 | return -E2BIG; |
f942dc25 IC |
751 | } |
752 | ||
2810e5b9 | 753 | /* Xen network protocol had implicit dependency on |
37641494 WL |
754 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
755 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
756 | * same behavior as before. Any packet using more than | |
757 | * 18 slots but less than fatal_skb_slots slots is | |
758 | * dropped | |
2810e5b9 | 759 | */ |
37641494 | 760 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
761 | if (net_ratelimit()) |
762 | netdev_dbg(vif->dev, | |
763 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 764 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
765 | drop_err = -E2BIG; |
766 | } | |
767 | ||
59ccb4eb WL |
768 | if (drop_err) |
769 | txp = &dropped_tx; | |
770 | ||
2810e5b9 | 771 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 772 | sizeof(*txp)); |
03393fd5 WL |
773 | |
774 | /* If the guest submitted a frame >= 64 KiB then | |
775 | * first->size overflowed and following slots will | |
776 | * appear to be larger than the frame. | |
777 | * | |
778 | * This cannot be fatal error as there are buggy | |
779 | * frontends that do this. | |
780 | * | |
781 | * Consume all slots and drop the packet. | |
782 | */ | |
783 | if (!drop_err && txp->size > first->size) { | |
784 | if (net_ratelimit()) | |
785 | netdev_dbg(vif->dev, | |
786 | "Invalid tx request, slot size %u > remaining size %u\n", | |
787 | txp->size, first->size); | |
788 | drop_err = -EIO; | |
f942dc25 IC |
789 | } |
790 | ||
791 | first->size -= txp->size; | |
2810e5b9 | 792 | slots++; |
f942dc25 IC |
793 | |
794 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 795 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 796 | txp->offset, txp->size); |
7376419a | 797 | xenvif_fatal_tx_err(vif); |
35876b5f | 798 | return -EINVAL; |
f942dc25 | 799 | } |
59ccb4eb WL |
800 | |
801 | more_data = txp->flags & XEN_NETTXF_more_data; | |
802 | ||
803 | if (!drop_err) | |
804 | txp++; | |
805 | ||
806 | } while (more_data); | |
2810e5b9 WL |
807 | |
808 | if (drop_err) { | |
7376419a | 809 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
810 | return drop_err; |
811 | } | |
812 | ||
813 | return slots; | |
f942dc25 IC |
814 | } |
815 | ||
8f13dd96 ZK |
816 | |
817 | struct xenvif_tx_cb { | |
818 | u16 pending_idx; | |
819 | }; | |
820 | ||
821 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
822 | ||
f53c3fe8 ZK |
823 | static inline void xenvif_tx_create_gop(struct xenvif *vif, |
824 | u16 pending_idx, | |
825 | struct xen_netif_tx_request *txp, | |
826 | struct gnttab_map_grant_ref *gop) | |
827 | { | |
828 | vif->pages_to_map[gop-vif->tx_map_ops] = vif->mmap_pages[pending_idx]; | |
829 | gnttab_set_map_op(gop, idx_to_kaddr(vif, pending_idx), | |
830 | GNTMAP_host_map | GNTMAP_readonly, | |
831 | txp->gref, vif->domid); | |
832 | ||
833 | memcpy(&vif->pending_tx_info[pending_idx].req, txp, | |
834 | sizeof(*txp)); | |
835 | } | |
836 | ||
e3377f36 ZK |
837 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
838 | { | |
839 | struct sk_buff *skb = | |
840 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
841 | GFP_ATOMIC | __GFP_NOWARN); | |
842 | if (unlikely(skb == NULL)) | |
843 | return NULL; | |
844 | ||
845 | /* Packets passed to netif_rx() must have some headroom. */ | |
846 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
847 | ||
848 | /* Initialize it here to avoid later surprises */ | |
849 | skb_shinfo(skb)->destructor_arg = NULL; | |
850 | ||
851 | return skb; | |
852 | } | |
853 | ||
f53c3fe8 ZK |
854 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif *vif, |
855 | struct sk_buff *skb, | |
856 | struct xen_netif_tx_request *txp, | |
857 | struct gnttab_map_grant_ref *gop) | |
f942dc25 IC |
858 | { |
859 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
860 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 861 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
862 | int start; |
863 | pending_ring_idx_t index; | |
e3377f36 | 864 | unsigned int nr_slots, frag_overflow = 0; |
2810e5b9 WL |
865 | |
866 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 867 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 | 868 | */ |
e3377f36 ZK |
869 | if (shinfo->nr_frags > MAX_SKB_FRAGS) { |
870 | frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS; | |
871 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
872 | shinfo->nr_frags = MAX_SKB_FRAGS; | |
873 | } | |
2810e5b9 | 874 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
875 | |
876 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 877 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 878 | |
f53c3fe8 ZK |
879 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
880 | shinfo->nr_frags++, txp++, gop++) { | |
62bad319 ZK |
881 | index = pending_index(vif->pending_cons++); |
882 | pending_idx = vif->pending_ring[index]; | |
f53c3fe8 ZK |
883 | xenvif_tx_create_gop(vif, pending_idx, txp, gop); |
884 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); | |
f942dc25 IC |
885 | } |
886 | ||
e3377f36 ZK |
887 | if (frag_overflow) { |
888 | struct sk_buff *nskb = xenvif_alloc_skb(0); | |
889 | if (unlikely(nskb == NULL)) { | |
890 | if (net_ratelimit()) | |
891 | netdev_err(vif->dev, | |
892 | "Can't allocate the frag_list skb.\n"); | |
893 | return NULL; | |
894 | } | |
895 | ||
896 | shinfo = skb_shinfo(nskb); | |
897 | frags = shinfo->frags; | |
898 | ||
899 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
900 | shinfo->nr_frags++, txp++, gop++) { | |
901 | index = pending_index(vif->pending_cons++); | |
902 | pending_idx = vif->pending_ring[index]; | |
903 | xenvif_tx_create_gop(vif, pending_idx, txp, gop); | |
904 | frag_set_pending_idx(&frags[shinfo->nr_frags], | |
905 | pending_idx); | |
906 | } | |
907 | ||
908 | skb_shinfo(skb)->frag_list = nskb; | |
909 | } | |
2810e5b9 | 910 | |
f942dc25 IC |
911 | return gop; |
912 | } | |
913 | ||
f53c3fe8 ZK |
914 | static inline void xenvif_grant_handle_set(struct xenvif *vif, |
915 | u16 pending_idx, | |
916 | grant_handle_t handle) | |
917 | { | |
918 | if (unlikely(vif->grant_tx_handle[pending_idx] != | |
919 | NETBACK_INVALID_HANDLE)) { | |
920 | netdev_err(vif->dev, | |
921 | "Trying to overwrite active handle! pending_idx: %x\n", | |
922 | pending_idx); | |
923 | BUG(); | |
924 | } | |
925 | vif->grant_tx_handle[pending_idx] = handle; | |
926 | } | |
927 | ||
928 | static inline void xenvif_grant_handle_reset(struct xenvif *vif, | |
929 | u16 pending_idx) | |
930 | { | |
931 | if (unlikely(vif->grant_tx_handle[pending_idx] == | |
932 | NETBACK_INVALID_HANDLE)) { | |
933 | netdev_err(vif->dev, | |
934 | "Trying to unmap invalid handle! pending_idx: %x\n", | |
935 | pending_idx); | |
936 | BUG(); | |
937 | } | |
938 | vif->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; | |
939 | } | |
940 | ||
7376419a WL |
941 | static int xenvif_tx_check_gop(struct xenvif *vif, |
942 | struct sk_buff *skb, | |
f53c3fe8 | 943 | struct gnttab_map_grant_ref **gopp) |
f942dc25 | 944 | { |
f53c3fe8 | 945 | struct gnttab_map_grant_ref *gop = *gopp; |
8f13dd96 | 946 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
f942dc25 | 947 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 948 | struct pending_tx_info *tx_info; |
f942dc25 IC |
949 | int nr_frags = shinfo->nr_frags; |
950 | int i, err, start; | |
e3377f36 | 951 | struct sk_buff *first_skb = NULL; |
f942dc25 IC |
952 | |
953 | /* Check status of header. */ | |
954 | err = gop->status; | |
7d5145d8 | 955 | if (unlikely(err)) |
7376419a | 956 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f53c3fe8 ZK |
957 | else |
958 | xenvif_grant_handle_set(vif, pending_idx , gop->handle); | |
f942dc25 IC |
959 | |
960 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 961 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 962 | |
e3377f36 | 963 | check_frags: |
f942dc25 IC |
964 | for (i = start; i < nr_frags; i++) { |
965 | int j, newerr; | |
f942dc25 | 966 | |
ea066ad1 | 967 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 968 | tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 IC |
969 | |
970 | /* Check error status: if okay then remember grant handle. */ | |
62bad319 | 971 | newerr = (++gop)->status; |
2810e5b9 | 972 | |
f942dc25 | 973 | if (likely(!newerr)) { |
f53c3fe8 | 974 | xenvif_grant_handle_set(vif, pending_idx , gop->handle); |
f942dc25 IC |
975 | /* Had a previous error? Invalidate this fragment. */ |
976 | if (unlikely(err)) | |
f53c3fe8 | 977 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
978 | continue; |
979 | } | |
980 | ||
981 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 982 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
983 | |
984 | /* Not the first error? Preceding frags already invalidated. */ | |
985 | if (err) | |
986 | continue; | |
f942dc25 | 987 | /* First error: invalidate header and preceding fragments. */ |
e3377f36 ZK |
988 | if (!first_skb) |
989 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
990 | else | |
991 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
f53c3fe8 | 992 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 | 993 | for (j = start; j < i; j++) { |
5ccb3ea7 | 994 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
f53c3fe8 | 995 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
996 | } |
997 | ||
998 | /* Remember the error: invalidate all subsequent fragments. */ | |
999 | err = newerr; | |
1000 | } | |
1001 | ||
e3377f36 ZK |
1002 | if (skb_has_frag_list(skb)) { |
1003 | first_skb = skb; | |
1004 | skb = shinfo->frag_list; | |
1005 | shinfo = skb_shinfo(skb); | |
1006 | nr_frags = shinfo->nr_frags; | |
1007 | start = 0; | |
1008 | ||
1009 | goto check_frags; | |
1010 | } | |
1011 | ||
1012 | /* There was a mapping error in the frag_list skb. We have to unmap | |
1013 | * the first skb's frags | |
1014 | */ | |
1015 | if (first_skb && err) { | |
1016 | int j; | |
1017 | shinfo = skb_shinfo(first_skb); | |
1018 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
1019 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); | |
1020 | for (j = start; j < shinfo->nr_frags; j++) { | |
1021 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); | |
1022 | xenvif_idx_unmap(vif, pending_idx); | |
1023 | } | |
1024 | } | |
1025 | ||
f942dc25 IC |
1026 | *gopp = gop + 1; |
1027 | return err; | |
1028 | } | |
1029 | ||
7376419a | 1030 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
1031 | { |
1032 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1033 | int nr_frags = shinfo->nr_frags; | |
1034 | int i; | |
f53c3fe8 ZK |
1035 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1036 | ||
1037 | if (skb_shinfo(skb)->destructor_arg) | |
1038 | prev_pending_idx = XENVIF_TX_CB(skb)->pending_idx; | |
f942dc25 IC |
1039 | |
1040 | for (i = 0; i < nr_frags; i++) { | |
1041 | skb_frag_t *frag = shinfo->frags + i; | |
1042 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1043 | struct page *page; |
1044 | u16 pending_idx; | |
f942dc25 | 1045 | |
ea066ad1 | 1046 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1047 | |
f53c3fe8 ZK |
1048 | /* If this is not the first frag, chain it to the previous*/ |
1049 | if (unlikely(prev_pending_idx == INVALID_PENDING_IDX)) | |
1050 | skb_shinfo(skb)->destructor_arg = | |
7aceb47a | 1051 | &callback_param(vif, pending_idx); |
f53c3fe8 | 1052 | else if (likely(pending_idx != prev_pending_idx)) |
7aceb47a ZK |
1053 | callback_param(vif, prev_pending_idx).ctx = |
1054 | &callback_param(vif, pending_idx); | |
f53c3fe8 | 1055 | |
7aceb47a | 1056 | callback_param(vif, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1057 | prev_pending_idx = pending_idx; |
1058 | ||
b3f980bd WL |
1059 | txp = &vif->pending_tx_info[pending_idx].req; |
1060 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 1061 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1062 | skb->len += txp->size; |
1063 | skb->data_len += txp->size; | |
1064 | skb->truesize += txp->size; | |
1065 | ||
f53c3fe8 | 1066 | /* Take an extra reference to offset network stack's put_page */ |
b3f980bd | 1067 | get_page(vif->mmap_pages[pending_idx]); |
f942dc25 | 1068 | } |
f53c3fe8 ZK |
1069 | /* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc |
1070 | * overlaps with "index", and "mapping" is not set. I think mapping | |
1071 | * should be set. If delivered to local stack, it would drop this | |
1072 | * skb in sk_filter unless the socket has the right to use it. | |
1073 | */ | |
1074 | skb->pfmemalloc = false; | |
f942dc25 IC |
1075 | } |
1076 | ||
7376419a | 1077 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
1078 | struct xen_netif_extra_info *extras, |
1079 | int work_to_do) | |
1080 | { | |
1081 | struct xen_netif_extra_info extra; | |
1082 | RING_IDX cons = vif->tx.req_cons; | |
1083 | ||
1084 | do { | |
1085 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 1086 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 1087 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1088 | return -EBADR; |
1089 | } | |
1090 | ||
1091 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1092 | sizeof(extra)); | |
1093 | if (unlikely(!extra.type || | |
1094 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1095 | vif->tx.req_cons = ++cons; | |
48856286 | 1096 | netdev_err(vif->dev, |
f942dc25 | 1097 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1098 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1099 | return -EINVAL; |
1100 | } | |
1101 | ||
1102 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1103 | vif->tx.req_cons = ++cons; | |
1104 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1105 | ||
1106 | return work_to_do; | |
1107 | } | |
1108 | ||
7376419a WL |
1109 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1110 | struct sk_buff *skb, | |
1111 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1112 | { |
1113 | if (!gso->u.gso.size) { | |
48856286 | 1114 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1115 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1116 | return -EINVAL; |
1117 | } | |
1118 | ||
a9468587 PD |
1119 | switch (gso->u.gso.type) { |
1120 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1121 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1122 | break; | |
1123 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1124 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1125 | break; | |
1126 | default: | |
48856286 | 1127 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1128 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1129 | return -EINVAL; |
1130 | } | |
1131 | ||
1132 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1133 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1134 | |
1135 | return 0; | |
1136 | } | |
1137 | ||
2eba61d5 PD |
1138 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) |
1139 | { | |
2721637c | 1140 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1141 | |
1142 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1143 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1144 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1145 | * recalculate the partial checksum. | |
1146 | */ | |
1147 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1148 | vif->rx_gso_checksum_fixup++; | |
1149 | skb->ip_summed = CHECKSUM_PARTIAL; | |
2721637c | 1150 | recalculate_partial_csum = true; |
2eba61d5 PD |
1151 | } |
1152 | ||
1153 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1154 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1155 | return 0; | |
1156 | ||
2721637c | 1157 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1158 | } |
1159 | ||
f942dc25 IC |
1160 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1161 | { | |
059dfa6a WL |
1162 | u64 now = get_jiffies_64(); |
1163 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1164 | msecs_to_jiffies(vif->credit_usec / 1000); |
1165 | ||
1166 | /* Timer could already be pending in rare cases. */ | |
1167 | if (timer_pending(&vif->credit_timeout)) | |
1168 | return true; | |
1169 | ||
1170 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1171 | if (time_after_eq64(now, next_credit)) { |
1172 | vif->credit_window_start = now; | |
f942dc25 IC |
1173 | tx_add_credit(vif); |
1174 | } | |
1175 | ||
1176 | /* Still too big to send right now? Set a callback. */ | |
1177 | if (size > vif->remaining_credit) { | |
1178 | vif->credit_timeout.data = | |
1179 | (unsigned long)vif; | |
1180 | vif->credit_timeout.function = | |
1181 | tx_credit_callback; | |
1182 | mod_timer(&vif->credit_timeout, | |
1183 | next_credit); | |
059dfa6a | 1184 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1185 | |
1186 | return true; | |
1187 | } | |
1188 | ||
1189 | return false; | |
1190 | } | |
1191 | ||
10574059 | 1192 | static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget) |
f942dc25 | 1193 | { |
f53c3fe8 | 1194 | struct gnttab_map_grant_ref *gop = vif->tx_map_ops, *request_gop; |
f942dc25 IC |
1195 | struct sk_buff *skb; |
1196 | int ret; | |
1197 | ||
869b9b19 | 1198 | while (skb_queue_len(&vif->tx_queue) < budget) { |
f942dc25 | 1199 | struct xen_netif_tx_request txreq; |
37641494 | 1200 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1201 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1202 | u16 pending_idx; | |
1203 | RING_IDX idx; | |
1204 | int work_to_do; | |
1205 | unsigned int data_len; | |
1206 | pending_ring_idx_t index; | |
1207 | ||
48856286 IC |
1208 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1209 | XEN_NETIF_TX_RING_SIZE) { | |
1210 | netdev_err(vif->dev, | |
1211 | "Impossible number of requests. " | |
1212 | "req_prod %d, req_cons %d, size %ld\n", | |
1213 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1214 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1215 | xenvif_fatal_tx_err(vif); |
e9d8b2c2 | 1216 | break; |
48856286 IC |
1217 | } |
1218 | ||
d9601a36 | 1219 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
b3f980bd WL |
1220 | if (!work_to_do) |
1221 | break; | |
f942dc25 IC |
1222 | |
1223 | idx = vif->tx.req_cons; | |
1224 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1225 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1226 | ||
1227 | /* Credit-based scheduling. */ | |
1228 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1229 | tx_credit_exceeded(vif, txreq.size)) |
1230 | break; | |
f942dc25 IC |
1231 | |
1232 | vif->remaining_credit -= txreq.size; | |
1233 | ||
1234 | work_to_do--; | |
1235 | vif->tx.req_cons = ++idx; | |
1236 | ||
1237 | memset(extras, 0, sizeof(extras)); | |
1238 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1239 | work_to_do = xenvif_get_extras(vif, extras, |
1240 | work_to_do); | |
f942dc25 | 1241 | idx = vif->tx.req_cons; |
48856286 | 1242 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1243 | break; |
f942dc25 IC |
1244 | } |
1245 | ||
7376419a | 1246 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1247 | if (unlikely(ret < 0)) |
b3f980bd | 1248 | break; |
48856286 | 1249 | |
f942dc25 IC |
1250 | idx += ret; |
1251 | ||
1252 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1253 | netdev_dbg(vif->dev, | |
1254 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1255 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1256 | break; |
f942dc25 IC |
1257 | } |
1258 | ||
1259 | /* No crossing a page as the payload mustn't fragment. */ | |
1260 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1261 | netdev_err(vif->dev, |
f942dc25 IC |
1262 | "txreq.offset: %x, size: %u, end: %lu\n", |
1263 | txreq.offset, txreq.size, | |
1264 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1265 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1266 | break; |
f942dc25 IC |
1267 | } |
1268 | ||
b3f980bd WL |
1269 | index = pending_index(vif->pending_cons); |
1270 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1271 | |
1272 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1273 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1274 | PKT_PROT_LEN : txreq.size; |
1275 | ||
e3377f36 | 1276 | skb = xenvif_alloc_skb(data_len); |
f942dc25 IC |
1277 | if (unlikely(skb == NULL)) { |
1278 | netdev_dbg(vif->dev, | |
1279 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1280 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1281 | break; |
1282 | } | |
1283 | ||
f942dc25 IC |
1284 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1285 | struct xen_netif_extra_info *gso; | |
1286 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1287 | ||
7376419a WL |
1288 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1289 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1290 | kfree_skb(skb); |
b3f980bd | 1291 | break; |
f942dc25 IC |
1292 | } |
1293 | } | |
1294 | ||
f53c3fe8 | 1295 | xenvif_tx_create_gop(vif, pending_idx, &txreq, gop); |
f942dc25 IC |
1296 | |
1297 | gop++; | |
1298 | ||
8f13dd96 | 1299 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1300 | |
1301 | __skb_put(skb, data_len); | |
1302 | ||
1303 | skb_shinfo(skb)->nr_frags = ret; | |
1304 | if (data_len < txreq.size) { | |
1305 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1306 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1307 | pending_idx); | |
f942dc25 | 1308 | } else { |
ea066ad1 IC |
1309 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1310 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1311 | } |
1312 | ||
b3f980bd | 1313 | vif->pending_cons++; |
f942dc25 | 1314 | |
7376419a | 1315 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1316 | if (request_gop == NULL) { |
1317 | kfree_skb(skb); | |
7376419a | 1318 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1319 | break; |
f942dc25 IC |
1320 | } |
1321 | gop = request_gop; | |
1322 | ||
b3f980bd | 1323 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1324 | |
f942dc25 | 1325 | vif->tx.req_cons = idx; |
f942dc25 | 1326 | |
f53c3fe8 | 1327 | if ((gop-vif->tx_map_ops) >= ARRAY_SIZE(vif->tx_map_ops)) |
f942dc25 IC |
1328 | break; |
1329 | } | |
1330 | ||
f53c3fe8 | 1331 | return gop - vif->tx_map_ops; |
f942dc25 IC |
1332 | } |
1333 | ||
e3377f36 ZK |
1334 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1335 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1336 | */ | |
1337 | static int xenvif_handle_frag_list(struct xenvif *vif, struct sk_buff *skb) | |
1338 | { | |
1339 | unsigned int offset = skb_headlen(skb); | |
1340 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
1341 | int i; | |
1342 | struct ubuf_info *uarg; | |
1343 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1344 | ||
1345 | vif->tx_zerocopy_sent += 2; | |
1346 | vif->tx_frag_overflow++; | |
1347 | ||
1348 | xenvif_fill_frags(vif, nskb); | |
1349 | /* Subtract frags size, we will correct it later */ | |
1350 | skb->truesize -= skb->data_len; | |
1351 | skb->len += nskb->len; | |
1352 | skb->data_len += nskb->len; | |
1353 | ||
1354 | /* create a brand new frags array and coalesce there */ | |
1355 | for (i = 0; offset < skb->len; i++) { | |
1356 | struct page *page; | |
1357 | unsigned int len; | |
1358 | ||
1359 | BUG_ON(i >= MAX_SKB_FRAGS); | |
1360 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
1361 | if (!page) { | |
1362 | int j; | |
1363 | skb->truesize += skb->data_len; | |
1364 | for (j = 0; j < i; j++) | |
1365 | put_page(frags[j].page.p); | |
1366 | return -ENOMEM; | |
1367 | } | |
1368 | ||
1369 | if (offset + PAGE_SIZE < skb->len) | |
1370 | len = PAGE_SIZE; | |
1371 | else | |
1372 | len = skb->len - offset; | |
1373 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1374 | BUG(); | |
1375 | ||
1376 | offset += len; | |
1377 | frags[i].page.p = page; | |
1378 | frags[i].page_offset = 0; | |
1379 | skb_frag_size_set(&frags[i], len); | |
1380 | } | |
1381 | /* swap out with old one */ | |
1382 | memcpy(skb_shinfo(skb)->frags, | |
1383 | frags, | |
1384 | i * sizeof(skb_frag_t)); | |
1385 | skb_shinfo(skb)->nr_frags = i; | |
1386 | skb->truesize += i * PAGE_SIZE; | |
1387 | ||
1388 | /* remove traces of mapped pages and frag_list */ | |
1389 | skb_frag_list_init(skb); | |
1390 | uarg = skb_shinfo(skb)->destructor_arg; | |
1391 | uarg->callback(uarg, true); | |
1392 | skb_shinfo(skb)->destructor_arg = NULL; | |
1393 | ||
1394 | skb_shinfo(nskb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1395 | kfree_skb(nskb); | |
1396 | ||
1397 | return 0; | |
1398 | } | |
b3f980bd | 1399 | |
10574059 | 1400 | static int xenvif_tx_submit(struct xenvif *vif) |
f942dc25 | 1401 | { |
f53c3fe8 | 1402 | struct gnttab_map_grant_ref *gop = vif->tx_map_ops; |
f942dc25 | 1403 | struct sk_buff *skb; |
b3f980bd | 1404 | int work_done = 0; |
f942dc25 | 1405 | |
10574059 | 1406 | while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) { |
f942dc25 | 1407 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1408 | u16 pending_idx; |
1409 | unsigned data_len; | |
1410 | ||
8f13dd96 | 1411 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
b3f980bd | 1412 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1413 | |
1414 | /* Check the remap error code. */ | |
7376419a | 1415 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1416 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1417 | skb_shinfo(skb)->nr_frags = 0; | |
1418 | kfree_skb(skb); | |
1419 | continue; | |
1420 | } | |
1421 | ||
1422 | data_len = skb->len; | |
1423 | memcpy(skb->data, | |
b3f980bd | 1424 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 | 1425 | data_len); |
7aceb47a | 1426 | callback_param(vif, pending_idx).ctx = NULL; |
f942dc25 IC |
1427 | if (data_len < txp->size) { |
1428 | /* Append the packet payload as a fragment. */ | |
1429 | txp->offset += data_len; | |
1430 | txp->size -= data_len; | |
f53c3fe8 | 1431 | skb_shinfo(skb)->destructor_arg = |
7aceb47a | 1432 | &callback_param(vif, pending_idx); |
f942dc25 IC |
1433 | } else { |
1434 | /* Schedule a response immediately. */ | |
f53c3fe8 | 1435 | xenvif_idx_unmap(vif, pending_idx); |
f942dc25 IC |
1436 | } |
1437 | ||
1438 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1439 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1440 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1441 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1442 | ||
7376419a | 1443 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1444 | |
e3377f36 ZK |
1445 | if (unlikely(skb_has_frag_list(skb))) { |
1446 | if (xenvif_handle_frag_list(vif, skb)) { | |
1447 | if (net_ratelimit()) | |
1448 | netdev_err(vif->dev, | |
1449 | "Not enough memory to consolidate frag_list!\n"); | |
1450 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1451 | kfree_skb(skb); | |
1452 | continue; | |
1453 | } | |
1454 | } | |
1455 | ||
2eba61d5 | 1456 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1457 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1458 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1459 | } | |
1460 | ||
1461 | skb->dev = vif->dev; | |
1462 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1463 | skb_reset_network_header(skb); |
f942dc25 IC |
1464 | |
1465 | if (checksum_setup(vif, skb)) { | |
1466 | netdev_dbg(vif->dev, | |
1467 | "Can't setup checksum in net_tx_action\n"); | |
f53c3fe8 ZK |
1468 | /* We have to set this flag to trigger the callback */ |
1469 | if (skb_shinfo(skb)->destructor_arg) | |
1470 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
f942dc25 IC |
1471 | kfree_skb(skb); |
1472 | continue; | |
1473 | } | |
1474 | ||
40893fd0 | 1475 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1476 | |
b89587a7 PD |
1477 | /* If the packet is GSO then we will have just set up the |
1478 | * transport header offset in checksum_setup so it's now | |
1479 | * straightforward to calculate gso_segs. | |
1480 | */ | |
1481 | if (skb_is_gso(skb)) { | |
1482 | int mss = skb_shinfo(skb)->gso_size; | |
1483 | int hdrlen = skb_transport_header(skb) - | |
1484 | skb_mac_header(skb) + | |
1485 | tcp_hdrlen(skb); | |
1486 | ||
1487 | skb_shinfo(skb)->gso_segs = | |
1488 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1489 | } | |
1490 | ||
f942dc25 IC |
1491 | vif->dev->stats.rx_bytes += skb->len; |
1492 | vif->dev->stats.rx_packets++; | |
1493 | ||
b3f980bd WL |
1494 | work_done++; |
1495 | ||
f53c3fe8 ZK |
1496 | /* Set this flag right before netif_receive_skb, otherwise |
1497 | * someone might think this packet already left netback, and | |
1498 | * do a skb_copy_ubufs while we are still in control of the | |
1499 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1500 | */ | |
1bb332af | 1501 | if (skb_shinfo(skb)->destructor_arg) { |
f53c3fe8 | 1502 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
1bb332af ZK |
1503 | vif->tx_zerocopy_sent++; |
1504 | } | |
f53c3fe8 | 1505 | |
b3f980bd | 1506 | netif_receive_skb(skb); |
f942dc25 | 1507 | } |
b3f980bd WL |
1508 | |
1509 | return work_done; | |
f942dc25 IC |
1510 | } |
1511 | ||
3e2234b3 ZK |
1512 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1513 | { | |
f53c3fe8 ZK |
1514 | unsigned long flags; |
1515 | pending_ring_idx_t index; | |
1516 | struct xenvif *vif = ubuf_to_vif(ubuf); | |
1517 | ||
1518 | /* This is the only place where we grab this lock, to protect callbacks | |
1519 | * from each other. | |
1520 | */ | |
1521 | spin_lock_irqsave(&vif->callback_lock, flags); | |
1522 | do { | |
1523 | u16 pending_idx = ubuf->desc; | |
1524 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
1525 | BUG_ON(vif->dealloc_prod - vif->dealloc_cons >= | |
1526 | MAX_PENDING_REQS); | |
1527 | index = pending_index(vif->dealloc_prod); | |
1528 | vif->dealloc_ring[index] = pending_idx; | |
1529 | /* Sync with xenvif_tx_dealloc_action: | |
1530 | * insert idx then incr producer. | |
1531 | */ | |
1532 | smp_wmb(); | |
1533 | vif->dealloc_prod++; | |
1534 | } while (ubuf); | |
1535 | wake_up(&vif->dealloc_wq); | |
1536 | spin_unlock_irqrestore(&vif->callback_lock, flags); | |
1537 | ||
1bb332af ZK |
1538 | if (likely(zerocopy_success)) |
1539 | vif->tx_zerocopy_success++; | |
1540 | else | |
1541 | vif->tx_zerocopy_fail++; | |
f53c3fe8 ZK |
1542 | } |
1543 | ||
1544 | static inline void xenvif_tx_dealloc_action(struct xenvif *vif) | |
1545 | { | |
1546 | struct gnttab_unmap_grant_ref *gop; | |
1547 | pending_ring_idx_t dc, dp; | |
1548 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1549 | unsigned int i = 0; | |
1550 | ||
1551 | dc = vif->dealloc_cons; | |
1552 | gop = vif->tx_unmap_ops; | |
1553 | ||
1554 | /* Free up any grants we have finished using */ | |
1555 | do { | |
1556 | dp = vif->dealloc_prod; | |
1557 | ||
1558 | /* Ensure we see all indices enqueued by all | |
1559 | * xenvif_zerocopy_callback(). | |
1560 | */ | |
1561 | smp_rmb(); | |
1562 | ||
1563 | while (dc != dp) { | |
1564 | BUG_ON(gop - vif->tx_unmap_ops > MAX_PENDING_REQS); | |
1565 | pending_idx = | |
1566 | vif->dealloc_ring[pending_index(dc++)]; | |
1567 | ||
1568 | pending_idx_release[gop-vif->tx_unmap_ops] = | |
1569 | pending_idx; | |
1570 | vif->pages_to_unmap[gop-vif->tx_unmap_ops] = | |
1571 | vif->mmap_pages[pending_idx]; | |
1572 | gnttab_set_unmap_op(gop, | |
1573 | idx_to_kaddr(vif, pending_idx), | |
1574 | GNTMAP_host_map, | |
1575 | vif->grant_tx_handle[pending_idx]); | |
f53c3fe8 ZK |
1576 | xenvif_grant_handle_reset(vif, pending_idx); |
1577 | ++gop; | |
1578 | } | |
1579 | ||
1580 | } while (dp != vif->dealloc_prod); | |
1581 | ||
1582 | vif->dealloc_cons = dc; | |
1583 | ||
1584 | if (gop - vif->tx_unmap_ops > 0) { | |
1585 | int ret; | |
1586 | ret = gnttab_unmap_refs(vif->tx_unmap_ops, | |
1587 | NULL, | |
1588 | vif->pages_to_unmap, | |
1589 | gop - vif->tx_unmap_ops); | |
1590 | if (ret) { | |
2c5f4f84 | 1591 | netdev_err(vif->dev, "Unmap fail: nr_ops %tx ret %d\n", |
f53c3fe8 ZK |
1592 | gop - vif->tx_unmap_ops, ret); |
1593 | for (i = 0; i < gop - vif->tx_unmap_ops; ++i) { | |
1594 | if (gop[i].status != GNTST_okay) | |
1595 | netdev_err(vif->dev, | |
1596 | " host_addr: %llx handle: %x status: %d\n", | |
1597 | gop[i].host_addr, | |
1598 | gop[i].handle, | |
1599 | gop[i].status); | |
1600 | } | |
1601 | BUG(); | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | for (i = 0; i < gop - vif->tx_unmap_ops; ++i) | |
1606 | xenvif_idx_release(vif, pending_idx_release[i], | |
1607 | XEN_NETIF_RSP_OKAY); | |
3e2234b3 ZK |
1608 | } |
1609 | ||
f53c3fe8 | 1610 | |
f942dc25 | 1611 | /* Called after netfront has transmitted */ |
7376419a | 1612 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1613 | { |
1614 | unsigned nr_gops; | |
f53c3fe8 | 1615 | int work_done, ret; |
f942dc25 | 1616 | |
b3f980bd WL |
1617 | if (unlikely(!tx_work_todo(vif))) |
1618 | return 0; | |
1619 | ||
10574059 | 1620 | nr_gops = xenvif_tx_build_gops(vif, budget); |
f942dc25 IC |
1621 | |
1622 | if (nr_gops == 0) | |
b3f980bd WL |
1623 | return 0; |
1624 | ||
f53c3fe8 ZK |
1625 | ret = gnttab_map_refs(vif->tx_map_ops, |
1626 | NULL, | |
1627 | vif->pages_to_map, | |
1628 | nr_gops); | |
1629 | BUG_ON(ret); | |
f942dc25 | 1630 | |
10574059 | 1631 | work_done = xenvif_tx_submit(vif); |
f942dc25 | 1632 | |
b3f980bd | 1633 | return work_done; |
f942dc25 IC |
1634 | } |
1635 | ||
7376419a WL |
1636 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1637 | u8 status) | |
f942dc25 | 1638 | { |
f942dc25 | 1639 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1640 | pending_ring_idx_t index; |
f53c3fe8 | 1641 | unsigned long flags; |
2810e5b9 | 1642 | |
62bad319 ZK |
1643 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
1644 | spin_lock_irqsave(&vif->response_lock, flags); | |
1645 | make_tx_response(vif, &pending_tx_info->req, status); | |
1646 | index = pending_index(vif->pending_prod); | |
1647 | vif->pending_ring[index] = pending_idx; | |
1648 | /* TX shouldn't use the index before we give it back here */ | |
1649 | mb(); | |
1650 | vif->pending_prod++; | |
1651 | spin_unlock_irqrestore(&vif->response_lock, flags); | |
f942dc25 IC |
1652 | } |
1653 | ||
2810e5b9 | 1654 | |
f942dc25 IC |
1655 | static void make_tx_response(struct xenvif *vif, |
1656 | struct xen_netif_tx_request *txp, | |
1657 | s8 st) | |
1658 | { | |
1659 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1660 | struct xen_netif_tx_response *resp; | |
1661 | int notify; | |
1662 | ||
1663 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1664 | resp->id = txp->id; | |
1665 | resp->status = st; | |
1666 | ||
1667 | if (txp->flags & XEN_NETTXF_extra_info) | |
1668 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1669 | ||
1670 | vif->tx.rsp_prod_pvt = ++i; | |
1671 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1672 | if (notify) | |
e1f00a69 | 1673 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1674 | } |
1675 | ||
1676 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1677 | u16 id, | |
1678 | s8 st, | |
1679 | u16 offset, | |
1680 | u16 size, | |
1681 | u16 flags) | |
1682 | { | |
1683 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1684 | struct xen_netif_rx_response *resp; | |
1685 | ||
1686 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1687 | resp->offset = offset; | |
1688 | resp->flags = flags; | |
1689 | resp->id = id; | |
1690 | resp->status = (s16)size; | |
1691 | if (st < 0) | |
1692 | resp->status = (s16)st; | |
1693 | ||
1694 | vif->rx.rsp_prod_pvt = ++i; | |
1695 | ||
1696 | return resp; | |
1697 | } | |
1698 | ||
f53c3fe8 ZK |
1699 | void xenvif_idx_unmap(struct xenvif *vif, u16 pending_idx) |
1700 | { | |
1701 | int ret; | |
1702 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1703 | ||
1704 | gnttab_set_unmap_op(&tx_unmap_op, | |
1705 | idx_to_kaddr(vif, pending_idx), | |
1706 | GNTMAP_host_map, | |
1707 | vif->grant_tx_handle[pending_idx]); | |
f53c3fe8 ZK |
1708 | xenvif_grant_handle_reset(vif, pending_idx); |
1709 | ||
1710 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
1711 | &vif->mmap_pages[pending_idx], 1); | |
7aceb47a ZK |
1712 | if (ret) { |
1713 | netdev_err(vif->dev, | |
1714 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: %x status: %d\n", | |
1715 | ret, | |
1716 | pending_idx, | |
1717 | tx_unmap_op.host_addr, | |
1718 | tx_unmap_op.handle, | |
1719 | tx_unmap_op.status); | |
1720 | BUG(); | |
1721 | } | |
f53c3fe8 ZK |
1722 | |
1723 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); | |
1724 | } | |
1725 | ||
b3f980bd | 1726 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1727 | { |
09350788 ZK |
1728 | return (!skb_queue_empty(&vif->rx_queue) && |
1729 | xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots)) || | |
1730 | vif->rx_queue_purge; | |
f942dc25 IC |
1731 | } |
1732 | ||
b3f980bd | 1733 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1734 | { |
1735 | ||
869b9b19 | 1736 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))) |
f942dc25 IC |
1737 | return 1; |
1738 | ||
1739 | return 0; | |
1740 | } | |
1741 | ||
f53c3fe8 ZK |
1742 | static inline bool tx_dealloc_work_todo(struct xenvif *vif) |
1743 | { | |
397dfd9f | 1744 | return vif->dealloc_cons != vif->dealloc_prod; |
f53c3fe8 ZK |
1745 | } |
1746 | ||
7376419a | 1747 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1748 | { |
c9d63699 DV |
1749 | if (vif->tx.sring) |
1750 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1751 | vif->tx.sring); | |
1752 | if (vif->rx.sring) | |
1753 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1754 | vif->rx.sring); | |
f942dc25 IC |
1755 | } |
1756 | ||
7376419a WL |
1757 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1758 | grant_ref_t tx_ring_ref, | |
1759 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1760 | { |
c9d63699 | 1761 | void *addr; |
f942dc25 IC |
1762 | struct xen_netif_tx_sring *txs; |
1763 | struct xen_netif_rx_sring *rxs; | |
1764 | ||
1765 | int err = -ENOMEM; | |
1766 | ||
c9d63699 DV |
1767 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1768 | tx_ring_ref, &addr); | |
1769 | if (err) | |
f942dc25 IC |
1770 | goto err; |
1771 | ||
c9d63699 | 1772 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1773 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1774 | ||
c9d63699 DV |
1775 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1776 | rx_ring_ref, &addr); | |
1777 | if (err) | |
f942dc25 | 1778 | goto err; |
f942dc25 | 1779 | |
c9d63699 | 1780 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1781 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1782 | ||
1783 | return 0; | |
1784 | ||
1785 | err: | |
7376419a | 1786 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1787 | return err; |
1788 | } | |
1789 | ||
ca2f09f2 PD |
1790 | void xenvif_stop_queue(struct xenvif *vif) |
1791 | { | |
1792 | if (!vif->can_queue) | |
1793 | return; | |
1794 | ||
1795 | netif_stop_queue(vif->dev); | |
1796 | } | |
1797 | ||
1798 | static void xenvif_start_queue(struct xenvif *vif) | |
1799 | { | |
1800 | if (xenvif_schedulable(vif)) | |
1801 | netif_wake_queue(vif->dev); | |
1802 | } | |
1803 | ||
121fa4b7 | 1804 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd WL |
1805 | { |
1806 | struct xenvif *vif = data; | |
ca2f09f2 | 1807 | struct sk_buff *skb; |
b3f980bd WL |
1808 | |
1809 | while (!kthread_should_stop()) { | |
1810 | wait_event_interruptible(vif->wq, | |
1811 | rx_work_todo(vif) || | |
e9d8b2c2 | 1812 | vif->disabled || |
b3f980bd | 1813 | kthread_should_stop()); |
e9d8b2c2 WL |
1814 | |
1815 | /* This frontend is found to be rogue, disable it in | |
1816 | * kthread context. Currently this is only set when | |
1817 | * netback finds out frontend sends malformed packet, | |
1818 | * but we cannot disable the interface in softirq | |
1819 | * context so we defer it here. | |
1820 | */ | |
1821 | if (unlikely(vif->disabled && netif_carrier_ok(vif->dev))) | |
1822 | xenvif_carrier_off(vif); | |
1823 | ||
b3f980bd WL |
1824 | if (kthread_should_stop()) |
1825 | break; | |
1826 | ||
09350788 ZK |
1827 | if (vif->rx_queue_purge) { |
1828 | skb_queue_purge(&vif->rx_queue); | |
1829 | vif->rx_queue_purge = false; | |
1830 | } | |
1831 | ||
ca2f09f2 | 1832 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 1833 | xenvif_rx_action(vif); |
b3f980bd | 1834 | |
ca2f09f2 | 1835 | if (skb_queue_empty(&vif->rx_queue) && |
09350788 ZK |
1836 | netif_queue_stopped(vif->dev)) { |
1837 | del_timer_sync(&vif->wake_queue); | |
ca2f09f2 | 1838 | xenvif_start_queue(vif); |
09350788 | 1839 | } |
ca2f09f2 | 1840 | |
b3f980bd WL |
1841 | cond_resched(); |
1842 | } | |
1843 | ||
ca2f09f2 PD |
1844 | /* Bin any remaining skbs */ |
1845 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | |
1846 | dev_kfree_skb(skb); | |
1847 | ||
b3f980bd WL |
1848 | return 0; |
1849 | } | |
1850 | ||
f53c3fe8 ZK |
1851 | int xenvif_dealloc_kthread(void *data) |
1852 | { | |
1853 | struct xenvif *vif = data; | |
1854 | ||
1855 | while (!kthread_should_stop()) { | |
1856 | wait_event_interruptible(vif->dealloc_wq, | |
1857 | tx_dealloc_work_todo(vif) || | |
1858 | kthread_should_stop()); | |
1859 | if (kthread_should_stop()) | |
1860 | break; | |
1861 | ||
1862 | xenvif_tx_dealloc_action(vif); | |
1863 | cond_resched(); | |
1864 | } | |
1865 | ||
1866 | /* Unmap anything remaining*/ | |
1867 | if (tx_dealloc_work_todo(vif)) | |
1868 | xenvif_tx_dealloc_action(vif); | |
1869 | ||
1870 | return 0; | |
1871 | } | |
1872 | ||
f942dc25 IC |
1873 | static int __init netback_init(void) |
1874 | { | |
f942dc25 | 1875 | int rc = 0; |
f942dc25 | 1876 | |
2a14b244 | 1877 | if (!xen_domain()) |
f942dc25 IC |
1878 | return -ENODEV; |
1879 | ||
37641494 | 1880 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1881 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1882 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1883 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1884 | } |
1885 | ||
f942dc25 IC |
1886 | rc = xenvif_xenbus_init(); |
1887 | if (rc) | |
1888 | goto failed_init; | |
1889 | ||
09350788 ZK |
1890 | rx_drain_timeout_jiffies = msecs_to_jiffies(rx_drain_timeout_msecs); |
1891 | ||
f942dc25 IC |
1892 | return 0; |
1893 | ||
1894 | failed_init: | |
f942dc25 | 1895 | return rc; |
f942dc25 IC |
1896 | } |
1897 | ||
1898 | module_init(netback_init); | |
1899 | ||
b103f358 WL |
1900 | static void __exit netback_fini(void) |
1901 | { | |
b103f358 | 1902 | xenvif_xenbus_fini(); |
b103f358 WL |
1903 | } |
1904 | module_exit(netback_fini); | |
1905 | ||
f942dc25 | 1906 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1907 | MODULE_ALIAS("xen-backend:vif"); |