Commit | Line | Data |
---|---|---|
f942dc25 IC |
1 | /* |
2 | * Xenbus code for netif backend | |
3 | * | |
4 | * Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au> | |
5 | * Copyright (C) 2005 XenSource Ltd | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
adf8d3ff | 18 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
f942dc25 IC |
19 | */ |
20 | ||
21 | #include "common.h" | |
e9ce7cb6 WL |
22 | #include <linux/vmalloc.h> |
23 | #include <linux/rtnetlink.h> | |
f942dc25 IC |
24 | |
25 | struct backend_info { | |
26 | struct xenbus_device *dev; | |
27 | struct xenvif *vif; | |
ea732dff PD |
28 | |
29 | /* This is the state that will be reflected in xenstore when any | |
30 | * active hotplug script completes. | |
31 | */ | |
32 | enum xenbus_state state; | |
33 | ||
f942dc25 IC |
34 | enum xenbus_state frontend_state; |
35 | struct xenbus_watch hotplug_status_watch; | |
17938a69 | 36 | u8 have_hotplug_status_watch:1; |
f942dc25 IC |
37 | }; |
38 | ||
e9ce7cb6 WL |
39 | static int connect_rings(struct backend_info *be, struct xenvif_queue *queue); |
40 | static void connect(struct backend_info *be); | |
41 | static int read_xenbus_vif_flags(struct backend_info *be); | |
f942dc25 IC |
42 | static void backend_create_xenvif(struct backend_info *be); |
43 | static void unregister_hotplug_status_watch(struct backend_info *be); | |
dc62ccac DV |
44 | static void set_backend_state(struct backend_info *be, |
45 | enum xenbus_state state); | |
f942dc25 | 46 | |
f51de243 ZK |
47 | #ifdef CONFIG_DEBUG_FS |
48 | struct dentry *xen_netback_dbg_root = NULL; | |
49 | ||
50 | static int xenvif_read_io_ring(struct seq_file *m, void *v) | |
51 | { | |
52 | struct xenvif_queue *queue = m->private; | |
53 | struct xen_netif_tx_back_ring *tx_ring = &queue->tx; | |
54 | struct xen_netif_rx_back_ring *rx_ring = &queue->rx; | |
55 | ||
56 | if (tx_ring->sring) { | |
57 | struct xen_netif_tx_sring *sring = tx_ring->sring; | |
58 | ||
59 | seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id, | |
60 | tx_ring->nr_ents); | |
61 | seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n", | |
62 | sring->req_prod, | |
63 | sring->req_prod - sring->rsp_prod, | |
64 | tx_ring->req_cons, | |
65 | tx_ring->req_cons - sring->rsp_prod, | |
66 | sring->req_event, | |
67 | sring->req_event - sring->rsp_prod); | |
68 | seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n", | |
69 | sring->rsp_prod, | |
70 | tx_ring->rsp_prod_pvt, | |
71 | tx_ring->rsp_prod_pvt - sring->rsp_prod, | |
72 | sring->rsp_event, | |
73 | sring->rsp_event - sring->rsp_prod); | |
74 | seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n", | |
75 | queue->pending_prod, | |
76 | queue->pending_cons, | |
77 | nr_pending_reqs(queue)); | |
78 | seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n", | |
79 | queue->dealloc_prod, | |
80 | queue->dealloc_cons, | |
81 | queue->dealloc_prod - queue->dealloc_cons); | |
82 | } | |
83 | ||
84 | if (rx_ring->sring) { | |
85 | struct xen_netif_rx_sring *sring = rx_ring->sring; | |
86 | ||
87 | seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents); | |
88 | seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n", | |
89 | sring->req_prod, | |
90 | sring->req_prod - sring->rsp_prod, | |
91 | rx_ring->req_cons, | |
92 | rx_ring->req_cons - sring->rsp_prod, | |
93 | sring->req_event, | |
94 | sring->req_event - sring->rsp_prod); | |
95 | seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n", | |
96 | sring->rsp_prod, | |
97 | rx_ring->rsp_prod_pvt, | |
98 | rx_ring->rsp_prod_pvt - sring->rsp_prod, | |
99 | sring->rsp_event, | |
100 | sring->rsp_event - sring->rsp_prod); | |
101 | } | |
102 | ||
103 | seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n" | |
104 | "Credit timer_pending: %d, credit: %lu, usec: %lu\n" | |
105 | "remaining: %lu, expires: %lu, now: %lu\n", | |
106 | queue->napi.state, queue->napi.weight, | |
107 | skb_queue_len(&queue->tx_queue), | |
108 | timer_pending(&queue->credit_timeout), | |
109 | queue->credit_bytes, | |
110 | queue->credit_usec, | |
111 | queue->remaining_credit, | |
112 | queue->credit_timeout.expires, | |
113 | jiffies); | |
114 | ||
115 | return 0; | |
116 | } | |
117 | ||
118 | #define XENVIF_KICK_STR "kick" | |
119 | ||
120 | static ssize_t | |
121 | xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count, | |
122 | loff_t *ppos) | |
123 | { | |
124 | struct xenvif_queue *queue = | |
125 | ((struct seq_file *)filp->private_data)->private; | |
126 | int len; | |
127 | char write[sizeof(XENVIF_KICK_STR)]; | |
128 | ||
129 | /* don't allow partial writes and check the length */ | |
130 | if (*ppos != 0) | |
131 | return 0; | |
132 | if (count < sizeof(XENVIF_KICK_STR) - 1) | |
133 | return -ENOSPC; | |
134 | ||
135 | len = simple_write_to_buffer(write, | |
136 | sizeof(write), | |
137 | ppos, | |
138 | buf, | |
139 | count); | |
140 | if (len < 0) | |
141 | return len; | |
142 | ||
143 | if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1)) | |
144 | xenvif_interrupt(0, (void *)queue); | |
145 | else { | |
146 | pr_warn("Unknown command to io_ring_q%d. Available: kick\n", | |
147 | queue->id); | |
148 | count = -EINVAL; | |
149 | } | |
150 | return count; | |
151 | } | |
152 | ||
153 | static int xenvif_dump_open(struct inode *inode, struct file *filp) | |
154 | { | |
155 | int ret; | |
156 | void *queue = NULL; | |
157 | ||
158 | if (inode->i_private) | |
159 | queue = inode->i_private; | |
160 | ret = single_open(filp, xenvif_read_io_ring, queue); | |
161 | filp->f_mode |= FMODE_PWRITE; | |
162 | return ret; | |
163 | } | |
164 | ||
165 | static const struct file_operations xenvif_dbg_io_ring_ops_fops = { | |
166 | .owner = THIS_MODULE, | |
167 | .open = xenvif_dump_open, | |
168 | .read = seq_read, | |
169 | .llseek = seq_lseek, | |
170 | .release = single_release, | |
171 | .write = xenvif_write_io_ring, | |
172 | }; | |
173 | ||
174 | static void xenvif_debugfs_addif(struct xenvif_queue *queue) | |
175 | { | |
176 | struct dentry *pfile; | |
177 | struct xenvif *vif = queue->vif; | |
178 | int i; | |
179 | ||
180 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
181 | return; | |
182 | ||
183 | vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name, | |
184 | xen_netback_dbg_root); | |
185 | if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) { | |
186 | for (i = 0; i < vif->num_queues; ++i) { | |
187 | char filename[sizeof("io_ring_q") + 4]; | |
188 | ||
189 | snprintf(filename, sizeof(filename), "io_ring_q%d", i); | |
190 | pfile = debugfs_create_file(filename, | |
191 | S_IRUSR | S_IWUSR, | |
192 | vif->xenvif_dbg_root, | |
193 | &vif->queues[i], | |
194 | &xenvif_dbg_io_ring_ops_fops); | |
195 | if (IS_ERR_OR_NULL(pfile)) | |
196 | pr_warn("Creation of io_ring file returned %ld!\n", | |
197 | PTR_ERR(pfile)); | |
198 | } | |
199 | } else | |
200 | netdev_warn(vif->dev, | |
201 | "Creation of vif debugfs dir returned %ld!\n", | |
202 | PTR_ERR(vif->xenvif_dbg_root)); | |
203 | } | |
204 | ||
205 | static void xenvif_debugfs_delif(struct xenvif *vif) | |
206 | { | |
207 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
208 | return; | |
209 | ||
210 | if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) | |
211 | debugfs_remove_recursive(vif->xenvif_dbg_root); | |
212 | vif->xenvif_dbg_root = NULL; | |
213 | } | |
214 | #endif /* CONFIG_DEBUG_FS */ | |
215 | ||
f942dc25 IC |
216 | static int netback_remove(struct xenbus_device *dev) |
217 | { | |
218 | struct backend_info *be = dev_get_drvdata(&dev->dev); | |
219 | ||
dc62ccac DV |
220 | set_backend_state(be, XenbusStateClosed); |
221 | ||
f942dc25 IC |
222 | unregister_hotplug_status_watch(be); |
223 | if (be->vif) { | |
224 | kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE); | |
225 | xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status"); | |
279f438e | 226 | xenvif_free(be->vif); |
f942dc25 IC |
227 | be->vif = NULL; |
228 | } | |
229 | kfree(be); | |
230 | dev_set_drvdata(&dev->dev, NULL); | |
231 | return 0; | |
232 | } | |
233 | ||
234 | ||
235 | /** | |
236 | * Entry point to this code when a new device is created. Allocate the basic | |
237 | * structures and switch to InitWait. | |
238 | */ | |
239 | static int netback_probe(struct xenbus_device *dev, | |
240 | const struct xenbus_device_id *id) | |
241 | { | |
242 | const char *message; | |
243 | struct xenbus_transaction xbt; | |
244 | int err; | |
245 | int sg; | |
246 | struct backend_info *be = kzalloc(sizeof(struct backend_info), | |
247 | GFP_KERNEL); | |
248 | if (!be) { | |
249 | xenbus_dev_fatal(dev, -ENOMEM, | |
250 | "allocating backend structure"); | |
251 | return -ENOMEM; | |
252 | } | |
253 | ||
254 | be->dev = dev; | |
255 | dev_set_drvdata(&dev->dev, be); | |
256 | ||
257 | sg = 1; | |
258 | ||
259 | do { | |
260 | err = xenbus_transaction_start(&xbt); | |
261 | if (err) { | |
262 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
263 | goto fail; | |
264 | } | |
265 | ||
266 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg); | |
267 | if (err) { | |
268 | message = "writing feature-sg"; | |
269 | goto abort_transaction; | |
270 | } | |
271 | ||
272 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", | |
273 | "%d", sg); | |
274 | if (err) { | |
275 | message = "writing feature-gso-tcpv4"; | |
276 | goto abort_transaction; | |
277 | } | |
278 | ||
a9468587 PD |
279 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6", |
280 | "%d", sg); | |
281 | if (err) { | |
282 | message = "writing feature-gso-tcpv6"; | |
283 | goto abort_transaction; | |
284 | } | |
285 | ||
2eba61d5 PD |
286 | /* We support partial checksum setup for IPv6 packets */ |
287 | err = xenbus_printf(xbt, dev->nodename, | |
288 | "feature-ipv6-csum-offload", | |
289 | "%d", 1); | |
290 | if (err) { | |
291 | message = "writing feature-ipv6-csum-offload"; | |
292 | goto abort_transaction; | |
293 | } | |
294 | ||
f942dc25 IC |
295 | /* We support rx-copy path. */ |
296 | err = xenbus_printf(xbt, dev->nodename, | |
297 | "feature-rx-copy", "%d", 1); | |
298 | if (err) { | |
299 | message = "writing feature-rx-copy"; | |
300 | goto abort_transaction; | |
301 | } | |
302 | ||
303 | /* | |
304 | * We don't support rx-flip path (except old guests who don't | |
305 | * grok this feature flag). | |
306 | */ | |
307 | err = xenbus_printf(xbt, dev->nodename, | |
308 | "feature-rx-flip", "%d", 0); | |
309 | if (err) { | |
310 | message = "writing feature-rx-flip"; | |
311 | goto abort_transaction; | |
312 | } | |
313 | ||
314 | err = xenbus_transaction_end(xbt, 0); | |
315 | } while (err == -EAGAIN); | |
316 | ||
317 | if (err) { | |
318 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
319 | goto fail; | |
320 | } | |
321 | ||
e1f00a69 WL |
322 | /* |
323 | * Split event channels support, this is optional so it is not | |
324 | * put inside the above loop. | |
325 | */ | |
326 | err = xenbus_printf(XBT_NIL, dev->nodename, | |
327 | "feature-split-event-channels", | |
328 | "%u", separate_tx_rx_irq); | |
329 | if (err) | |
8ef2c3bc | 330 | pr_debug("Error writing feature-split-event-channels\n"); |
e1f00a69 | 331 | |
8d3d53b3 AB |
332 | /* Multi-queue support: This is an optional feature. */ |
333 | err = xenbus_printf(XBT_NIL, dev->nodename, | |
334 | "multi-queue-max-queues", "%u", xenvif_max_queues); | |
335 | if (err) | |
336 | pr_debug("Error writing multi-queue-max-queues\n"); | |
337 | ||
f942dc25 IC |
338 | err = xenbus_switch_state(dev, XenbusStateInitWait); |
339 | if (err) | |
340 | goto fail; | |
341 | ||
ea732dff PD |
342 | be->state = XenbusStateInitWait; |
343 | ||
f942dc25 IC |
344 | /* This kicks hotplug scripts, so do it immediately. */ |
345 | backend_create_xenvif(be); | |
346 | ||
347 | return 0; | |
348 | ||
349 | abort_transaction: | |
350 | xenbus_transaction_end(xbt, 1); | |
351 | xenbus_dev_fatal(dev, err, "%s", message); | |
352 | fail: | |
8ef2c3bc | 353 | pr_debug("failed\n"); |
f942dc25 IC |
354 | netback_remove(dev); |
355 | return err; | |
356 | } | |
357 | ||
358 | ||
359 | /* | |
360 | * Handle the creation of the hotplug script environment. We add the script | |
361 | * and vif variables to the environment, for the benefit of the vif-* hotplug | |
362 | * scripts. | |
363 | */ | |
364 | static int netback_uevent(struct xenbus_device *xdev, | |
365 | struct kobj_uevent_env *env) | |
366 | { | |
367 | struct backend_info *be = dev_get_drvdata(&xdev->dev); | |
368 | char *val; | |
369 | ||
370 | val = xenbus_read(XBT_NIL, xdev->nodename, "script", NULL); | |
371 | if (IS_ERR(val)) { | |
372 | int err = PTR_ERR(val); | |
373 | xenbus_dev_fatal(xdev, err, "reading script"); | |
374 | return err; | |
375 | } else { | |
376 | if (add_uevent_var(env, "script=%s", val)) { | |
377 | kfree(val); | |
378 | return -ENOMEM; | |
379 | } | |
380 | kfree(val); | |
381 | } | |
382 | ||
383 | if (!be || !be->vif) | |
384 | return 0; | |
385 | ||
386 | return add_uevent_var(env, "vif=%s", be->vif->dev->name); | |
387 | } | |
388 | ||
389 | ||
390 | static void backend_create_xenvif(struct backend_info *be) | |
391 | { | |
392 | int err; | |
393 | long handle; | |
394 | struct xenbus_device *dev = be->dev; | |
395 | ||
396 | if (be->vif != NULL) | |
397 | return; | |
398 | ||
399 | err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle); | |
400 | if (err != 1) { | |
401 | xenbus_dev_fatal(dev, err, "reading handle"); | |
402 | return; | |
403 | } | |
404 | ||
405 | be->vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle); | |
406 | if (IS_ERR(be->vif)) { | |
407 | err = PTR_ERR(be->vif); | |
408 | be->vif = NULL; | |
409 | xenbus_dev_fatal(dev, err, "creating interface"); | |
410 | return; | |
411 | } | |
412 | ||
413 | kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE); | |
414 | } | |
415 | ||
ea732dff | 416 | static void backend_disconnect(struct backend_info *be) |
f942dc25 | 417 | { |
f51de243 ZK |
418 | if (be->vif) { |
419 | #ifdef CONFIG_DEBUG_FS | |
420 | xenvif_debugfs_delif(be->vif); | |
421 | #endif /* CONFIG_DEBUG_FS */ | |
279f438e | 422 | xenvif_disconnect(be->vif); |
f51de243 | 423 | } |
279f438e PD |
424 | } |
425 | ||
ea732dff | 426 | static void backend_connect(struct backend_info *be) |
279f438e | 427 | { |
ea732dff PD |
428 | if (be->vif) |
429 | connect(be); | |
430 | } | |
279f438e | 431 | |
ea732dff PD |
432 | static inline void backend_switch_state(struct backend_info *be, |
433 | enum xenbus_state state) | |
434 | { | |
435 | struct xenbus_device *dev = be->dev; | |
436 | ||
437 | pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state)); | |
438 | be->state = state; | |
439 | ||
440 | /* If we are waiting for a hotplug script then defer the | |
441 | * actual xenbus state change. | |
442 | */ | |
443 | if (!be->have_hotplug_status_watch) | |
444 | xenbus_switch_state(dev, state); | |
445 | } | |
446 | ||
447 | /* Handle backend state transitions: | |
448 | * | |
449 | * The backend state starts in InitWait and the following transitions are | |
450 | * allowed. | |
451 | * | |
452 | * InitWait -> Connected | |
453 | * | |
454 | * ^ \ | | |
455 | * | \ | | |
456 | * | \ | | |
457 | * | \ | | |
458 | * | \ | | |
459 | * | \ | | |
460 | * | V V | |
461 | * | |
462 | * Closed <-> Closing | |
463 | * | |
464 | * The state argument specifies the eventual state of the backend and the | |
465 | * function transitions to that state via the shortest path. | |
466 | */ | |
467 | static void set_backend_state(struct backend_info *be, | |
468 | enum xenbus_state state) | |
469 | { | |
470 | while (be->state != state) { | |
471 | switch (be->state) { | |
472 | case XenbusStateClosed: | |
473 | switch (state) { | |
474 | case XenbusStateInitWait: | |
475 | case XenbusStateConnected: | |
476 | pr_info("%s: prepare for reconnect\n", | |
477 | be->dev->nodename); | |
478 | backend_switch_state(be, XenbusStateInitWait); | |
479 | break; | |
480 | case XenbusStateClosing: | |
481 | backend_switch_state(be, XenbusStateClosing); | |
482 | break; | |
483 | default: | |
484 | BUG(); | |
485 | } | |
486 | break; | |
487 | case XenbusStateInitWait: | |
488 | switch (state) { | |
489 | case XenbusStateConnected: | |
490 | backend_connect(be); | |
491 | backend_switch_state(be, XenbusStateConnected); | |
492 | break; | |
493 | case XenbusStateClosing: | |
494 | case XenbusStateClosed: | |
495 | backend_switch_state(be, XenbusStateClosing); | |
496 | break; | |
497 | default: | |
498 | BUG(); | |
499 | } | |
500 | break; | |
501 | case XenbusStateConnected: | |
502 | switch (state) { | |
503 | case XenbusStateInitWait: | |
504 | case XenbusStateClosing: | |
505 | case XenbusStateClosed: | |
506 | backend_disconnect(be); | |
507 | backend_switch_state(be, XenbusStateClosing); | |
508 | break; | |
509 | default: | |
510 | BUG(); | |
511 | } | |
512 | break; | |
513 | case XenbusStateClosing: | |
514 | switch (state) { | |
515 | case XenbusStateInitWait: | |
516 | case XenbusStateConnected: | |
517 | case XenbusStateClosed: | |
518 | backend_switch_state(be, XenbusStateClosed); | |
519 | break; | |
520 | default: | |
521 | BUG(); | |
522 | } | |
523 | break; | |
524 | default: | |
525 | BUG(); | |
526 | } | |
f942dc25 IC |
527 | } |
528 | } | |
529 | ||
530 | /** | |
531 | * Callback received when the frontend's state changes. | |
532 | */ | |
533 | static void frontend_changed(struct xenbus_device *dev, | |
534 | enum xenbus_state frontend_state) | |
535 | { | |
536 | struct backend_info *be = dev_get_drvdata(&dev->dev); | |
537 | ||
ea732dff | 538 | pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state)); |
f942dc25 IC |
539 | |
540 | be->frontend_state = frontend_state; | |
541 | ||
542 | switch (frontend_state) { | |
543 | case XenbusStateInitialising: | |
ea732dff | 544 | set_backend_state(be, XenbusStateInitWait); |
f942dc25 IC |
545 | break; |
546 | ||
547 | case XenbusStateInitialised: | |
548 | break; | |
549 | ||
550 | case XenbusStateConnected: | |
ea732dff | 551 | set_backend_state(be, XenbusStateConnected); |
f942dc25 IC |
552 | break; |
553 | ||
554 | case XenbusStateClosing: | |
ea732dff | 555 | set_backend_state(be, XenbusStateClosing); |
f942dc25 IC |
556 | break; |
557 | ||
558 | case XenbusStateClosed: | |
ea732dff | 559 | set_backend_state(be, XenbusStateClosed); |
f942dc25 IC |
560 | if (xenbus_dev_is_online(dev)) |
561 | break; | |
562 | /* fall through if not online */ | |
563 | case XenbusStateUnknown: | |
ea732dff | 564 | set_backend_state(be, XenbusStateClosed); |
f942dc25 IC |
565 | device_unregister(&dev->dev); |
566 | break; | |
567 | ||
568 | default: | |
569 | xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend", | |
570 | frontend_state); | |
571 | break; | |
572 | } | |
573 | } | |
574 | ||
575 | ||
576 | static void xen_net_read_rate(struct xenbus_device *dev, | |
577 | unsigned long *bytes, unsigned long *usec) | |
578 | { | |
579 | char *s, *e; | |
580 | unsigned long b, u; | |
581 | char *ratestr; | |
582 | ||
583 | /* Default to unlimited bandwidth. */ | |
584 | *bytes = ~0UL; | |
585 | *usec = 0; | |
586 | ||
587 | ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL); | |
588 | if (IS_ERR(ratestr)) | |
589 | return; | |
590 | ||
591 | s = ratestr; | |
592 | b = simple_strtoul(s, &e, 10); | |
593 | if ((s == e) || (*e != ',')) | |
594 | goto fail; | |
595 | ||
596 | s = e + 1; | |
597 | u = simple_strtoul(s, &e, 10); | |
598 | if ((s == e) || (*e != '\0')) | |
599 | goto fail; | |
600 | ||
601 | *bytes = b; | |
602 | *usec = u; | |
603 | ||
604 | kfree(ratestr); | |
605 | return; | |
606 | ||
607 | fail: | |
608 | pr_warn("Failed to parse network rate limit. Traffic unlimited.\n"); | |
609 | kfree(ratestr); | |
610 | } | |
611 | ||
612 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
613 | { | |
614 | char *s, *e, *macstr; | |
615 | int i; | |
616 | ||
617 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
618 | if (IS_ERR(macstr)) | |
619 | return PTR_ERR(macstr); | |
620 | ||
621 | for (i = 0; i < ETH_ALEN; i++) { | |
622 | mac[i] = simple_strtoul(s, &e, 16); | |
623 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
624 | kfree(macstr); | |
625 | return -ENOENT; | |
626 | } | |
627 | s = e+1; | |
628 | } | |
629 | ||
630 | kfree(macstr); | |
631 | return 0; | |
632 | } | |
633 | ||
634 | static void unregister_hotplug_status_watch(struct backend_info *be) | |
635 | { | |
636 | if (be->have_hotplug_status_watch) { | |
637 | unregister_xenbus_watch(&be->hotplug_status_watch); | |
638 | kfree(be->hotplug_status_watch.node); | |
639 | } | |
640 | be->have_hotplug_status_watch = 0; | |
641 | } | |
642 | ||
643 | static void hotplug_status_changed(struct xenbus_watch *watch, | |
644 | const char **vec, | |
645 | unsigned int vec_size) | |
646 | { | |
647 | struct backend_info *be = container_of(watch, | |
648 | struct backend_info, | |
649 | hotplug_status_watch); | |
650 | char *str; | |
651 | unsigned int len; | |
652 | ||
653 | str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len); | |
654 | if (IS_ERR(str)) | |
655 | return; | |
656 | if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) { | |
ea732dff PD |
657 | /* Complete any pending state change */ |
658 | xenbus_switch_state(be->dev, be->state); | |
659 | ||
f942dc25 IC |
660 | /* Not interested in this watch anymore. */ |
661 | unregister_hotplug_status_watch(be); | |
662 | } | |
663 | kfree(str); | |
664 | } | |
665 | ||
666 | static void connect(struct backend_info *be) | |
667 | { | |
668 | int err; | |
669 | struct xenbus_device *dev = be->dev; | |
e9ce7cb6 WL |
670 | unsigned long credit_bytes, credit_usec; |
671 | unsigned int queue_index; | |
8d3d53b3 | 672 | unsigned int requested_num_queues; |
e9ce7cb6 | 673 | struct xenvif_queue *queue; |
f942dc25 | 674 | |
8d3d53b3 AB |
675 | /* Check whether the frontend requested multiple queues |
676 | * and read the number requested. | |
677 | */ | |
678 | err = xenbus_scanf(XBT_NIL, dev->otherend, | |
679 | "multi-queue-num-queues", | |
680 | "%u", &requested_num_queues); | |
681 | if (err < 0) { | |
682 | requested_num_queues = 1; /* Fall back to single queue */ | |
683 | } else if (requested_num_queues > xenvif_max_queues) { | |
684 | /* buggy or malicious guest */ | |
685 | xenbus_dev_fatal(dev, err, | |
686 | "guest requested %u queues, exceeding the maximum of %u.", | |
687 | requested_num_queues, xenvif_max_queues); | |
688 | return; | |
689 | } | |
690 | ||
f942dc25 IC |
691 | err = xen_net_read_mac(dev, be->vif->fe_dev_addr); |
692 | if (err) { | |
693 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
694 | return; | |
695 | } | |
696 | ||
e9ce7cb6 WL |
697 | xen_net_read_rate(dev, &credit_bytes, &credit_usec); |
698 | read_xenbus_vif_flags(be); | |
699 | ||
8d3d53b3 | 700 | /* Use the number of queues requested by the frontend */ |
e9ce7cb6 WL |
701 | be->vif->queues = vzalloc(requested_num_queues * |
702 | sizeof(struct xenvif_queue)); | |
f7b50c4e | 703 | be->vif->num_queues = requested_num_queues; |
e9ce7cb6 WL |
704 | |
705 | for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) { | |
706 | queue = &be->vif->queues[queue_index]; | |
707 | queue->vif = be->vif; | |
708 | queue->id = queue_index; | |
709 | snprintf(queue->name, sizeof(queue->name), "%s-q%u", | |
710 | be->vif->dev->name, queue->id); | |
711 | ||
712 | err = xenvif_init_queue(queue); | |
8d3d53b3 AB |
713 | if (err) { |
714 | /* xenvif_init_queue() cleans up after itself on | |
715 | * failure, but we need to clean up any previously | |
716 | * initialised queues. Set num_queues to i so that | |
717 | * earlier queues can be destroyed using the regular | |
718 | * disconnect logic. | |
719 | */ | |
f7b50c4e | 720 | be->vif->num_queues = queue_index; |
e9ce7cb6 | 721 | goto err; |
8d3d53b3 | 722 | } |
e9ce7cb6 WL |
723 | |
724 | queue->remaining_credit = credit_bytes; | |
725 | ||
726 | err = connect_rings(be, queue); | |
8d3d53b3 AB |
727 | if (err) { |
728 | /* connect_rings() cleans up after itself on failure, | |
729 | * but we need to clean up after xenvif_init_queue() here, | |
730 | * and also clean up any previously initialised queues. | |
731 | */ | |
732 | xenvif_deinit_queue(queue); | |
f7b50c4e | 733 | be->vif->num_queues = queue_index; |
e9ce7cb6 | 734 | goto err; |
8d3d53b3 | 735 | } |
f51de243 ZK |
736 | #ifdef CONFIG_DEBUG_FS |
737 | xenvif_debugfs_addif(queue); | |
738 | #endif /* CONFIG_DEBUG_FS */ | |
e9ce7cb6 WL |
739 | } |
740 | ||
f7b50c4e WL |
741 | /* Initialisation completed, tell core driver the number of |
742 | * active queues. | |
743 | */ | |
744 | rtnl_lock(); | |
745 | netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues); | |
746 | netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues); | |
747 | rtnl_unlock(); | |
748 | ||
e9ce7cb6 | 749 | xenvif_carrier_on(be->vif); |
f942dc25 IC |
750 | |
751 | unregister_hotplug_status_watch(be); | |
752 | err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch, | |
753 | hotplug_status_changed, | |
754 | "%s/%s", dev->nodename, "hotplug-status"); | |
ea732dff | 755 | if (!err) |
f942dc25 | 756 | be->have_hotplug_status_watch = 1; |
f942dc25 | 757 | |
e9ce7cb6 WL |
758 | netif_tx_wake_all_queues(be->vif->dev); |
759 | ||
760 | return; | |
761 | ||
762 | err: | |
f7b50c4e | 763 | if (be->vif->num_queues > 0) |
8d3d53b3 | 764 | xenvif_disconnect(be->vif); /* Clean up existing queues */ |
e9ce7cb6 WL |
765 | vfree(be->vif->queues); |
766 | be->vif->queues = NULL; | |
f7b50c4e | 767 | be->vif->num_queues = 0; |
e9ce7cb6 | 768 | return; |
f942dc25 IC |
769 | } |
770 | ||
771 | ||
e9ce7cb6 | 772 | static int connect_rings(struct backend_info *be, struct xenvif_queue *queue) |
f942dc25 | 773 | { |
f942dc25 | 774 | struct xenbus_device *dev = be->dev; |
f7b50c4e | 775 | unsigned int num_queues = queue->vif->num_queues; |
f942dc25 | 776 | unsigned long tx_ring_ref, rx_ring_ref; |
e9ce7cb6 | 777 | unsigned int tx_evtchn, rx_evtchn; |
f942dc25 | 778 | int err; |
8d3d53b3 AB |
779 | char *xspath; |
780 | size_t xspathsize; | |
781 | const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */ | |
782 | ||
783 | /* If the frontend requested 1 queue, or we have fallen back | |
784 | * to single queue due to lack of frontend support for multi- | |
785 | * queue, expect the remaining XenStore keys in the toplevel | |
786 | * directory. Otherwise, expect them in a subdirectory called | |
787 | * queue-N. | |
788 | */ | |
789 | if (num_queues == 1) { | |
790 | xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL); | |
791 | if (!xspath) { | |
792 | xenbus_dev_fatal(dev, -ENOMEM, | |
793 | "reading ring references"); | |
794 | return -ENOMEM; | |
795 | } | |
796 | strcpy(xspath, dev->otherend); | |
797 | } else { | |
798 | xspathsize = strlen(dev->otherend) + xenstore_path_ext_size; | |
799 | xspath = kzalloc(xspathsize, GFP_KERNEL); | |
800 | if (!xspath) { | |
801 | xenbus_dev_fatal(dev, -ENOMEM, | |
802 | "reading ring references"); | |
803 | return -ENOMEM; | |
804 | } | |
805 | snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, | |
806 | queue->id); | |
807 | } | |
f942dc25 | 808 | |
8d3d53b3 | 809 | err = xenbus_gather(XBT_NIL, xspath, |
f942dc25 | 810 | "tx-ring-ref", "%lu", &tx_ring_ref, |
e1f00a69 | 811 | "rx-ring-ref", "%lu", &rx_ring_ref, NULL); |
f942dc25 IC |
812 | if (err) { |
813 | xenbus_dev_fatal(dev, err, | |
e1f00a69 | 814 | "reading %s/ring-ref", |
8d3d53b3 AB |
815 | xspath); |
816 | goto err; | |
f942dc25 IC |
817 | } |
818 | ||
e1f00a69 | 819 | /* Try split event channels first, then single event channel. */ |
8d3d53b3 | 820 | err = xenbus_gather(XBT_NIL, xspath, |
e1f00a69 WL |
821 | "event-channel-tx", "%u", &tx_evtchn, |
822 | "event-channel-rx", "%u", &rx_evtchn, NULL); | |
823 | if (err < 0) { | |
8d3d53b3 | 824 | err = xenbus_scanf(XBT_NIL, xspath, |
e1f00a69 WL |
825 | "event-channel", "%u", &tx_evtchn); |
826 | if (err < 0) { | |
827 | xenbus_dev_fatal(dev, err, | |
828 | "reading %s/event-channel(-tx/rx)", | |
8d3d53b3 AB |
829 | xspath); |
830 | goto err; | |
e1f00a69 WL |
831 | } |
832 | rx_evtchn = tx_evtchn; | |
833 | } | |
834 | ||
e9ce7cb6 WL |
835 | /* Map the shared frame, irq etc. */ |
836 | err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref, | |
837 | tx_evtchn, rx_evtchn); | |
838 | if (err) { | |
839 | xenbus_dev_fatal(dev, err, | |
840 | "mapping shared-frames %lu/%lu port tx %u rx %u", | |
841 | tx_ring_ref, rx_ring_ref, | |
842 | tx_evtchn, rx_evtchn); | |
8d3d53b3 | 843 | goto err; |
e9ce7cb6 WL |
844 | } |
845 | ||
8d3d53b3 AB |
846 | err = 0; |
847 | err: /* Regular return falls through with err == 0 */ | |
848 | kfree(xspath); | |
849 | return err; | |
e9ce7cb6 WL |
850 | } |
851 | ||
852 | static int read_xenbus_vif_flags(struct backend_info *be) | |
853 | { | |
854 | struct xenvif *vif = be->vif; | |
855 | struct xenbus_device *dev = be->dev; | |
856 | unsigned int rx_copy; | |
857 | int err, val; | |
858 | ||
f942dc25 IC |
859 | err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u", |
860 | &rx_copy); | |
861 | if (err == -ENOENT) { | |
862 | err = 0; | |
863 | rx_copy = 0; | |
864 | } | |
865 | if (err < 0) { | |
866 | xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy", | |
867 | dev->otherend); | |
868 | return err; | |
869 | } | |
870 | if (!rx_copy) | |
871 | return -EOPNOTSUPP; | |
872 | ||
873 | if (vif->dev->tx_queue_len != 0) { | |
874 | if (xenbus_scanf(XBT_NIL, dev->otherend, | |
875 | "feature-rx-notify", "%d", &val) < 0) | |
876 | val = 0; | |
877 | if (val) | |
878 | vif->can_queue = 1; | |
879 | else | |
880 | /* Must be non-zero for pfifo_fast to work. */ | |
881 | vif->dev->tx_queue_len = 1; | |
882 | } | |
883 | ||
884 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg", | |
885 | "%d", &val) < 0) | |
886 | val = 0; | |
887 | vif->can_sg = !!val; | |
888 | ||
82cada22 PD |
889 | vif->gso_mask = 0; |
890 | vif->gso_prefix_mask = 0; | |
891 | ||
f942dc25 IC |
892 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4", |
893 | "%d", &val) < 0) | |
894 | val = 0; | |
82cada22 PD |
895 | if (val) |
896 | vif->gso_mask |= GSO_BIT(TCPV4); | |
f942dc25 IC |
897 | |
898 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix", | |
899 | "%d", &val) < 0) | |
900 | val = 0; | |
82cada22 PD |
901 | if (val) |
902 | vif->gso_prefix_mask |= GSO_BIT(TCPV4); | |
903 | ||
904 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6", | |
905 | "%d", &val) < 0) | |
906 | val = 0; | |
907 | if (val) | |
908 | vif->gso_mask |= GSO_BIT(TCPV6); | |
909 | ||
910 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix", | |
911 | "%d", &val) < 0) | |
912 | val = 0; | |
913 | if (val) | |
914 | vif->gso_prefix_mask |= GSO_BIT(TCPV6); | |
915 | ||
916 | if (vif->gso_mask & vif->gso_prefix_mask) { | |
917 | xenbus_dev_fatal(dev, err, | |
918 | "%s: gso and gso prefix flags are not " | |
919 | "mutually exclusive", | |
920 | dev->otherend); | |
921 | return -EOPNOTSUPP; | |
922 | } | |
f942dc25 IC |
923 | |
924 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload", | |
925 | "%d", &val) < 0) | |
926 | val = 0; | |
146c8a77 PD |
927 | vif->ip_csum = !val; |
928 | ||
929 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload", | |
930 | "%d", &val) < 0) | |
931 | val = 0; | |
932 | vif->ipv6_csum = !!val; | |
f942dc25 | 933 | |
f942dc25 IC |
934 | return 0; |
935 | } | |
936 | ||
937 | ||
938 | /* ** Driver Registration ** */ | |
939 | ||
940 | ||
941 | static const struct xenbus_device_id netback_ids[] = { | |
942 | { "vif" }, | |
943 | { "" } | |
944 | }; | |
945 | ||
946 | ||
73db144b | 947 | static DEFINE_XENBUS_DRIVER(netback, , |
f942dc25 IC |
948 | .probe = netback_probe, |
949 | .remove = netback_remove, | |
950 | .uevent = netback_uevent, | |
951 | .otherend_changed = frontend_changed, | |
73db144b | 952 | ); |
f942dc25 IC |
953 | |
954 | int xenvif_xenbus_init(void) | |
955 | { | |
73db144b | 956 | return xenbus_register_backend(&netback_driver); |
f942dc25 | 957 | } |
b103f358 WL |
958 | |
959 | void xenvif_xenbus_fini(void) | |
960 | { | |
961 | return xenbus_unregister_driver(&netback_driver); | |
962 | } |