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); | |
2dd34339 | 42 | static int backend_create_xenvif(struct backend_info *be); |
f942dc25 | 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; | |
f48da8b1 | 55 | struct netdev_queue *dev_queue; |
f51de243 ZK |
56 | |
57 | if (tx_ring->sring) { | |
58 | struct xen_netif_tx_sring *sring = tx_ring->sring; | |
59 | ||
60 | seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id, | |
61 | tx_ring->nr_ents); | |
62 | seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n", | |
63 | sring->req_prod, | |
64 | sring->req_prod - sring->rsp_prod, | |
65 | tx_ring->req_cons, | |
66 | tx_ring->req_cons - sring->rsp_prod, | |
67 | sring->req_event, | |
68 | sring->req_event - sring->rsp_prod); | |
69 | seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n", | |
70 | sring->rsp_prod, | |
71 | tx_ring->rsp_prod_pvt, | |
72 | tx_ring->rsp_prod_pvt - sring->rsp_prod, | |
73 | sring->rsp_event, | |
74 | sring->rsp_event - sring->rsp_prod); | |
75 | seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n", | |
76 | queue->pending_prod, | |
77 | queue->pending_cons, | |
78 | nr_pending_reqs(queue)); | |
79 | seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n", | |
80 | queue->dealloc_prod, | |
81 | queue->dealloc_cons, | |
82 | queue->dealloc_prod - queue->dealloc_cons); | |
83 | } | |
84 | ||
85 | if (rx_ring->sring) { | |
86 | struct xen_netif_rx_sring *sring = rx_ring->sring; | |
87 | ||
88 | seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents); | |
89 | seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n", | |
90 | sring->req_prod, | |
91 | sring->req_prod - sring->rsp_prod, | |
92 | rx_ring->req_cons, | |
93 | rx_ring->req_cons - sring->rsp_prod, | |
94 | sring->req_event, | |
95 | sring->req_event - sring->rsp_prod); | |
96 | seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n", | |
97 | sring->rsp_prod, | |
98 | rx_ring->rsp_prod_pvt, | |
99 | rx_ring->rsp_prod_pvt - sring->rsp_prod, | |
100 | sring->rsp_event, | |
101 | sring->rsp_event - sring->rsp_prod); | |
102 | } | |
103 | ||
104 | seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n" | |
105 | "Credit timer_pending: %d, credit: %lu, usec: %lu\n" | |
106 | "remaining: %lu, expires: %lu, now: %lu\n", | |
107 | queue->napi.state, queue->napi.weight, | |
108 | skb_queue_len(&queue->tx_queue), | |
109 | timer_pending(&queue->credit_timeout), | |
110 | queue->credit_bytes, | |
111 | queue->credit_usec, | |
112 | queue->remaining_credit, | |
113 | queue->credit_timeout.expires, | |
114 | jiffies); | |
115 | ||
f48da8b1 DV |
116 | dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id); |
117 | ||
118 | seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n", | |
119 | queue->rx_queue_len, queue->rx_queue_max, | |
120 | skb_queue_len(&queue->rx_queue), | |
121 | netif_tx_queue_stopped(dev_queue) ? "stopped" : "running"); | |
122 | ||
f51de243 ZK |
123 | return 0; |
124 | } | |
125 | ||
126 | #define XENVIF_KICK_STR "kick" | |
5c807005 | 127 | #define BUFFER_SIZE 32 |
f51de243 ZK |
128 | |
129 | static ssize_t | |
130 | xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count, | |
131 | loff_t *ppos) | |
132 | { | |
133 | struct xenvif_queue *queue = | |
134 | ((struct seq_file *)filp->private_data)->private; | |
135 | int len; | |
5c807005 | 136 | char write[BUFFER_SIZE]; |
f51de243 ZK |
137 | |
138 | /* don't allow partial writes and check the length */ | |
139 | if (*ppos != 0) | |
140 | return 0; | |
5c807005 | 141 | if (count >= sizeof(write)) |
f51de243 ZK |
142 | return -ENOSPC; |
143 | ||
144 | len = simple_write_to_buffer(write, | |
5c807005 | 145 | sizeof(write) - 1, |
f51de243 ZK |
146 | ppos, |
147 | buf, | |
148 | count); | |
149 | if (len < 0) | |
150 | return len; | |
151 | ||
5c807005 WL |
152 | write[len] = '\0'; |
153 | ||
f51de243 ZK |
154 | if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1)) |
155 | xenvif_interrupt(0, (void *)queue); | |
156 | else { | |
157 | pr_warn("Unknown command to io_ring_q%d. Available: kick\n", | |
158 | queue->id); | |
159 | count = -EINVAL; | |
160 | } | |
161 | return count; | |
162 | } | |
163 | ||
164 | static int xenvif_dump_open(struct inode *inode, struct file *filp) | |
165 | { | |
166 | int ret; | |
167 | void *queue = NULL; | |
168 | ||
169 | if (inode->i_private) | |
170 | queue = inode->i_private; | |
171 | ret = single_open(filp, xenvif_read_io_ring, queue); | |
172 | filp->f_mode |= FMODE_PWRITE; | |
173 | return ret; | |
174 | } | |
175 | ||
176 | static const struct file_operations xenvif_dbg_io_ring_ops_fops = { | |
177 | .owner = THIS_MODULE, | |
178 | .open = xenvif_dump_open, | |
179 | .read = seq_read, | |
180 | .llseek = seq_lseek, | |
181 | .release = single_release, | |
182 | .write = xenvif_write_io_ring, | |
183 | }; | |
184 | ||
628fa76b | 185 | static void xenvif_debugfs_addif(struct xenvif *vif) |
f51de243 ZK |
186 | { |
187 | struct dentry *pfile; | |
f51de243 ZK |
188 | int i; |
189 | ||
190 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
191 | return; | |
192 | ||
193 | vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name, | |
194 | xen_netback_dbg_root); | |
195 | if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) { | |
196 | for (i = 0; i < vif->num_queues; ++i) { | |
197 | char filename[sizeof("io_ring_q") + 4]; | |
198 | ||
199 | snprintf(filename, sizeof(filename), "io_ring_q%d", i); | |
200 | pfile = debugfs_create_file(filename, | |
201 | S_IRUSR | S_IWUSR, | |
202 | vif->xenvif_dbg_root, | |
203 | &vif->queues[i], | |
204 | &xenvif_dbg_io_ring_ops_fops); | |
205 | if (IS_ERR_OR_NULL(pfile)) | |
206 | pr_warn("Creation of io_ring file returned %ld!\n", | |
207 | PTR_ERR(pfile)); | |
208 | } | |
209 | } else | |
210 | netdev_warn(vif->dev, | |
211 | "Creation of vif debugfs dir returned %ld!\n", | |
212 | PTR_ERR(vif->xenvif_dbg_root)); | |
213 | } | |
214 | ||
215 | static void xenvif_debugfs_delif(struct xenvif *vif) | |
216 | { | |
217 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
218 | return; | |
219 | ||
220 | if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) | |
221 | debugfs_remove_recursive(vif->xenvif_dbg_root); | |
222 | vif->xenvif_dbg_root = NULL; | |
223 | } | |
224 | #endif /* CONFIG_DEBUG_FS */ | |
225 | ||
f942dc25 IC |
226 | static int netback_remove(struct xenbus_device *dev) |
227 | { | |
228 | struct backend_info *be = dev_get_drvdata(&dev->dev); | |
229 | ||
dc62ccac DV |
230 | set_backend_state(be, XenbusStateClosed); |
231 | ||
f942dc25 IC |
232 | unregister_hotplug_status_watch(be); |
233 | if (be->vif) { | |
234 | kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE); | |
235 | xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status"); | |
279f438e | 236 | xenvif_free(be->vif); |
f942dc25 IC |
237 | be->vif = NULL; |
238 | } | |
239 | kfree(be); | |
240 | dev_set_drvdata(&dev->dev, NULL); | |
241 | return 0; | |
242 | } | |
243 | ||
244 | ||
245 | /** | |
246 | * Entry point to this code when a new device is created. Allocate the basic | |
247 | * structures and switch to InitWait. | |
248 | */ | |
249 | static int netback_probe(struct xenbus_device *dev, | |
250 | const struct xenbus_device_id *id) | |
251 | { | |
252 | const char *message; | |
253 | struct xenbus_transaction xbt; | |
254 | int err; | |
255 | int sg; | |
256 | struct backend_info *be = kzalloc(sizeof(struct backend_info), | |
257 | GFP_KERNEL); | |
258 | if (!be) { | |
259 | xenbus_dev_fatal(dev, -ENOMEM, | |
260 | "allocating backend structure"); | |
261 | return -ENOMEM; | |
262 | } | |
263 | ||
264 | be->dev = dev; | |
265 | dev_set_drvdata(&dev->dev, be); | |
266 | ||
267 | sg = 1; | |
268 | ||
269 | do { | |
270 | err = xenbus_transaction_start(&xbt); | |
271 | if (err) { | |
272 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
273 | goto fail; | |
274 | } | |
275 | ||
276 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg); | |
277 | if (err) { | |
278 | message = "writing feature-sg"; | |
279 | goto abort_transaction; | |
280 | } | |
281 | ||
282 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", | |
283 | "%d", sg); | |
284 | if (err) { | |
285 | message = "writing feature-gso-tcpv4"; | |
286 | goto abort_transaction; | |
287 | } | |
288 | ||
a9468587 PD |
289 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6", |
290 | "%d", sg); | |
291 | if (err) { | |
292 | message = "writing feature-gso-tcpv6"; | |
293 | goto abort_transaction; | |
294 | } | |
295 | ||
2eba61d5 PD |
296 | /* We support partial checksum setup for IPv6 packets */ |
297 | err = xenbus_printf(xbt, dev->nodename, | |
298 | "feature-ipv6-csum-offload", | |
299 | "%d", 1); | |
300 | if (err) { | |
301 | message = "writing feature-ipv6-csum-offload"; | |
302 | goto abort_transaction; | |
303 | } | |
304 | ||
f942dc25 IC |
305 | /* We support rx-copy path. */ |
306 | err = xenbus_printf(xbt, dev->nodename, | |
307 | "feature-rx-copy", "%d", 1); | |
308 | if (err) { | |
309 | message = "writing feature-rx-copy"; | |
310 | goto abort_transaction; | |
311 | } | |
312 | ||
313 | /* | |
314 | * We don't support rx-flip path (except old guests who don't | |
315 | * grok this feature flag). | |
316 | */ | |
317 | err = xenbus_printf(xbt, dev->nodename, | |
318 | "feature-rx-flip", "%d", 0); | |
319 | if (err) { | |
320 | message = "writing feature-rx-flip"; | |
321 | goto abort_transaction; | |
322 | } | |
323 | ||
324 | err = xenbus_transaction_end(xbt, 0); | |
325 | } while (err == -EAGAIN); | |
326 | ||
327 | if (err) { | |
328 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
329 | goto fail; | |
330 | } | |
331 | ||
e1f00a69 WL |
332 | /* |
333 | * Split event channels support, this is optional so it is not | |
334 | * put inside the above loop. | |
335 | */ | |
336 | err = xenbus_printf(XBT_NIL, dev->nodename, | |
337 | "feature-split-event-channels", | |
338 | "%u", separate_tx_rx_irq); | |
339 | if (err) | |
8ef2c3bc | 340 | pr_debug("Error writing feature-split-event-channels\n"); |
e1f00a69 | 341 | |
8d3d53b3 AB |
342 | /* Multi-queue support: This is an optional feature. */ |
343 | err = xenbus_printf(XBT_NIL, dev->nodename, | |
344 | "multi-queue-max-queues", "%u", xenvif_max_queues); | |
345 | if (err) | |
346 | pr_debug("Error writing multi-queue-max-queues\n"); | |
347 | ||
f942dc25 IC |
348 | err = xenbus_switch_state(dev, XenbusStateInitWait); |
349 | if (err) | |
350 | goto fail; | |
351 | ||
ea732dff PD |
352 | be->state = XenbusStateInitWait; |
353 | ||
f942dc25 | 354 | /* This kicks hotplug scripts, so do it immediately. */ |
2dd34339 AK |
355 | err = backend_create_xenvif(be); |
356 | if (err) | |
357 | goto fail; | |
f942dc25 IC |
358 | |
359 | return 0; | |
360 | ||
361 | abort_transaction: | |
362 | xenbus_transaction_end(xbt, 1); | |
363 | xenbus_dev_fatal(dev, err, "%s", message); | |
364 | fail: | |
8ef2c3bc | 365 | pr_debug("failed\n"); |
f942dc25 IC |
366 | netback_remove(dev); |
367 | return err; | |
368 | } | |
369 | ||
370 | ||
371 | /* | |
372 | * Handle the creation of the hotplug script environment. We add the script | |
373 | * and vif variables to the environment, for the benefit of the vif-* hotplug | |
374 | * scripts. | |
375 | */ | |
376 | static int netback_uevent(struct xenbus_device *xdev, | |
377 | struct kobj_uevent_env *env) | |
378 | { | |
379 | struct backend_info *be = dev_get_drvdata(&xdev->dev); | |
380 | char *val; | |
381 | ||
382 | val = xenbus_read(XBT_NIL, xdev->nodename, "script", NULL); | |
383 | if (IS_ERR(val)) { | |
384 | int err = PTR_ERR(val); | |
385 | xenbus_dev_fatal(xdev, err, "reading script"); | |
386 | return err; | |
387 | } else { | |
388 | if (add_uevent_var(env, "script=%s", val)) { | |
389 | kfree(val); | |
390 | return -ENOMEM; | |
391 | } | |
392 | kfree(val); | |
393 | } | |
394 | ||
395 | if (!be || !be->vif) | |
396 | return 0; | |
397 | ||
398 | return add_uevent_var(env, "vif=%s", be->vif->dev->name); | |
399 | } | |
400 | ||
401 | ||
2dd34339 | 402 | static int backend_create_xenvif(struct backend_info *be) |
f942dc25 IC |
403 | { |
404 | int err; | |
405 | long handle; | |
406 | struct xenbus_device *dev = be->dev; | |
f15650b7 | 407 | struct xenvif *vif; |
f942dc25 IC |
408 | |
409 | if (be->vif != NULL) | |
2dd34339 | 410 | return 0; |
f942dc25 IC |
411 | |
412 | err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle); | |
413 | if (err != 1) { | |
414 | xenbus_dev_fatal(dev, err, "reading handle"); | |
2dd34339 | 415 | return (err < 0) ? err : -EINVAL; |
f942dc25 IC |
416 | } |
417 | ||
f15650b7 JB |
418 | vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle); |
419 | if (IS_ERR(vif)) { | |
420 | err = PTR_ERR(vif); | |
f942dc25 | 421 | xenbus_dev_fatal(dev, err, "creating interface"); |
2dd34339 | 422 | return err; |
f942dc25 | 423 | } |
f15650b7 | 424 | be->vif = vif; |
f942dc25 IC |
425 | |
426 | kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE); | |
2dd34339 | 427 | return 0; |
f942dc25 IC |
428 | } |
429 | ||
ea732dff | 430 | static void backend_disconnect(struct backend_info *be) |
f942dc25 | 431 | { |
f51de243 ZK |
432 | if (be->vif) { |
433 | #ifdef CONFIG_DEBUG_FS | |
434 | xenvif_debugfs_delif(be->vif); | |
435 | #endif /* CONFIG_DEBUG_FS */ | |
279f438e | 436 | xenvif_disconnect(be->vif); |
f51de243 | 437 | } |
279f438e PD |
438 | } |
439 | ||
ea732dff | 440 | static void backend_connect(struct backend_info *be) |
279f438e | 441 | { |
ea732dff PD |
442 | if (be->vif) |
443 | connect(be); | |
444 | } | |
279f438e | 445 | |
ea732dff PD |
446 | static inline void backend_switch_state(struct backend_info *be, |
447 | enum xenbus_state state) | |
448 | { | |
449 | struct xenbus_device *dev = be->dev; | |
450 | ||
451 | pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state)); | |
452 | be->state = state; | |
453 | ||
454 | /* If we are waiting for a hotplug script then defer the | |
455 | * actual xenbus state change. | |
456 | */ | |
457 | if (!be->have_hotplug_status_watch) | |
458 | xenbus_switch_state(dev, state); | |
459 | } | |
460 | ||
461 | /* Handle backend state transitions: | |
462 | * | |
463 | * The backend state starts in InitWait and the following transitions are | |
464 | * allowed. | |
465 | * | |
466 | * InitWait -> Connected | |
467 | * | |
468 | * ^ \ | | |
469 | * | \ | | |
470 | * | \ | | |
471 | * | \ | | |
472 | * | \ | | |
473 | * | \ | | |
474 | * | V V | |
475 | * | |
476 | * Closed <-> Closing | |
477 | * | |
478 | * The state argument specifies the eventual state of the backend and the | |
479 | * function transitions to that state via the shortest path. | |
480 | */ | |
481 | static void set_backend_state(struct backend_info *be, | |
482 | enum xenbus_state state) | |
483 | { | |
484 | while (be->state != state) { | |
485 | switch (be->state) { | |
486 | case XenbusStateClosed: | |
487 | switch (state) { | |
488 | case XenbusStateInitWait: | |
489 | case XenbusStateConnected: | |
490 | pr_info("%s: prepare for reconnect\n", | |
491 | be->dev->nodename); | |
492 | backend_switch_state(be, XenbusStateInitWait); | |
493 | break; | |
494 | case XenbusStateClosing: | |
495 | backend_switch_state(be, XenbusStateClosing); | |
496 | break; | |
497 | default: | |
498 | BUG(); | |
499 | } | |
500 | break; | |
501 | case XenbusStateInitWait: | |
502 | switch (state) { | |
503 | case XenbusStateConnected: | |
504 | backend_connect(be); | |
505 | backend_switch_state(be, XenbusStateConnected); | |
506 | break; | |
507 | case XenbusStateClosing: | |
508 | case XenbusStateClosed: | |
509 | backend_switch_state(be, XenbusStateClosing); | |
510 | break; | |
511 | default: | |
512 | BUG(); | |
513 | } | |
514 | break; | |
515 | case XenbusStateConnected: | |
516 | switch (state) { | |
517 | case XenbusStateInitWait: | |
518 | case XenbusStateClosing: | |
519 | case XenbusStateClosed: | |
520 | backend_disconnect(be); | |
521 | backend_switch_state(be, XenbusStateClosing); | |
522 | break; | |
523 | default: | |
524 | BUG(); | |
525 | } | |
526 | break; | |
527 | case XenbusStateClosing: | |
528 | switch (state) { | |
529 | case XenbusStateInitWait: | |
530 | case XenbusStateConnected: | |
531 | case XenbusStateClosed: | |
532 | backend_switch_state(be, XenbusStateClosed); | |
533 | break; | |
534 | default: | |
535 | BUG(); | |
536 | } | |
537 | break; | |
538 | default: | |
539 | BUG(); | |
540 | } | |
f942dc25 IC |
541 | } |
542 | } | |
543 | ||
544 | /** | |
545 | * Callback received when the frontend's state changes. | |
546 | */ | |
547 | static void frontend_changed(struct xenbus_device *dev, | |
548 | enum xenbus_state frontend_state) | |
549 | { | |
550 | struct backend_info *be = dev_get_drvdata(&dev->dev); | |
551 | ||
ea732dff | 552 | pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state)); |
f942dc25 IC |
553 | |
554 | be->frontend_state = frontend_state; | |
555 | ||
556 | switch (frontend_state) { | |
557 | case XenbusStateInitialising: | |
ea732dff | 558 | set_backend_state(be, XenbusStateInitWait); |
f942dc25 IC |
559 | break; |
560 | ||
561 | case XenbusStateInitialised: | |
562 | break; | |
563 | ||
564 | case XenbusStateConnected: | |
ea732dff | 565 | set_backend_state(be, XenbusStateConnected); |
f942dc25 IC |
566 | break; |
567 | ||
568 | case XenbusStateClosing: | |
ea732dff | 569 | set_backend_state(be, XenbusStateClosing); |
f942dc25 IC |
570 | break; |
571 | ||
572 | case XenbusStateClosed: | |
ea732dff | 573 | set_backend_state(be, XenbusStateClosed); |
f942dc25 IC |
574 | if (xenbus_dev_is_online(dev)) |
575 | break; | |
576 | /* fall through if not online */ | |
577 | case XenbusStateUnknown: | |
ea732dff | 578 | set_backend_state(be, XenbusStateClosed); |
f942dc25 IC |
579 | device_unregister(&dev->dev); |
580 | break; | |
581 | ||
582 | default: | |
583 | xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend", | |
584 | frontend_state); | |
585 | break; | |
586 | } | |
587 | } | |
588 | ||
589 | ||
590 | static void xen_net_read_rate(struct xenbus_device *dev, | |
591 | unsigned long *bytes, unsigned long *usec) | |
592 | { | |
593 | char *s, *e; | |
594 | unsigned long b, u; | |
595 | char *ratestr; | |
596 | ||
597 | /* Default to unlimited bandwidth. */ | |
598 | *bytes = ~0UL; | |
599 | *usec = 0; | |
600 | ||
601 | ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL); | |
602 | if (IS_ERR(ratestr)) | |
603 | return; | |
604 | ||
605 | s = ratestr; | |
606 | b = simple_strtoul(s, &e, 10); | |
607 | if ((s == e) || (*e != ',')) | |
608 | goto fail; | |
609 | ||
610 | s = e + 1; | |
611 | u = simple_strtoul(s, &e, 10); | |
612 | if ((s == e) || (*e != '\0')) | |
613 | goto fail; | |
614 | ||
615 | *bytes = b; | |
616 | *usec = u; | |
617 | ||
618 | kfree(ratestr); | |
619 | return; | |
620 | ||
621 | fail: | |
622 | pr_warn("Failed to parse network rate limit. Traffic unlimited.\n"); | |
623 | kfree(ratestr); | |
624 | } | |
625 | ||
626 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
627 | { | |
628 | char *s, *e, *macstr; | |
629 | int i; | |
630 | ||
631 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
632 | if (IS_ERR(macstr)) | |
633 | return PTR_ERR(macstr); | |
634 | ||
635 | for (i = 0; i < ETH_ALEN; i++) { | |
636 | mac[i] = simple_strtoul(s, &e, 16); | |
637 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
638 | kfree(macstr); | |
639 | return -ENOENT; | |
640 | } | |
641 | s = e+1; | |
642 | } | |
643 | ||
644 | kfree(macstr); | |
645 | return 0; | |
646 | } | |
647 | ||
648 | static void unregister_hotplug_status_watch(struct backend_info *be) | |
649 | { | |
650 | if (be->have_hotplug_status_watch) { | |
651 | unregister_xenbus_watch(&be->hotplug_status_watch); | |
652 | kfree(be->hotplug_status_watch.node); | |
653 | } | |
654 | be->have_hotplug_status_watch = 0; | |
655 | } | |
656 | ||
657 | static void hotplug_status_changed(struct xenbus_watch *watch, | |
658 | const char **vec, | |
659 | unsigned int vec_size) | |
660 | { | |
661 | struct backend_info *be = container_of(watch, | |
662 | struct backend_info, | |
663 | hotplug_status_watch); | |
664 | char *str; | |
665 | unsigned int len; | |
666 | ||
667 | str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len); | |
668 | if (IS_ERR(str)) | |
669 | return; | |
670 | if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) { | |
ea732dff PD |
671 | /* Complete any pending state change */ |
672 | xenbus_switch_state(be->dev, be->state); | |
673 | ||
f942dc25 IC |
674 | /* Not interested in this watch anymore. */ |
675 | unregister_hotplug_status_watch(be); | |
676 | } | |
677 | kfree(str); | |
678 | } | |
679 | ||
680 | static void connect(struct backend_info *be) | |
681 | { | |
682 | int err; | |
683 | struct xenbus_device *dev = be->dev; | |
e9ce7cb6 WL |
684 | unsigned long credit_bytes, credit_usec; |
685 | unsigned int queue_index; | |
8d3d53b3 | 686 | unsigned int requested_num_queues; |
e9ce7cb6 | 687 | struct xenvif_queue *queue; |
f942dc25 | 688 | |
8d3d53b3 AB |
689 | /* Check whether the frontend requested multiple queues |
690 | * and read the number requested. | |
691 | */ | |
692 | err = xenbus_scanf(XBT_NIL, dev->otherend, | |
693 | "multi-queue-num-queues", | |
694 | "%u", &requested_num_queues); | |
695 | if (err < 0) { | |
696 | requested_num_queues = 1; /* Fall back to single queue */ | |
697 | } else if (requested_num_queues > xenvif_max_queues) { | |
698 | /* buggy or malicious guest */ | |
699 | xenbus_dev_fatal(dev, err, | |
700 | "guest requested %u queues, exceeding the maximum of %u.", | |
701 | requested_num_queues, xenvif_max_queues); | |
702 | return; | |
703 | } | |
704 | ||
f942dc25 IC |
705 | err = xen_net_read_mac(dev, be->vif->fe_dev_addr); |
706 | if (err) { | |
707 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
708 | return; | |
709 | } | |
710 | ||
e9ce7cb6 WL |
711 | xen_net_read_rate(dev, &credit_bytes, &credit_usec); |
712 | read_xenbus_vif_flags(be); | |
713 | ||
8d3d53b3 | 714 | /* Use the number of queues requested by the frontend */ |
e9ce7cb6 WL |
715 | be->vif->queues = vzalloc(requested_num_queues * |
716 | sizeof(struct xenvif_queue)); | |
f7b50c4e | 717 | be->vif->num_queues = requested_num_queues; |
ecf08d2d | 718 | be->vif->stalled_queues = requested_num_queues; |
e9ce7cb6 WL |
719 | |
720 | for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) { | |
721 | queue = &be->vif->queues[queue_index]; | |
722 | queue->vif = be->vif; | |
723 | queue->id = queue_index; | |
724 | snprintf(queue->name, sizeof(queue->name), "%s-q%u", | |
725 | be->vif->dev->name, queue->id); | |
726 | ||
727 | err = xenvif_init_queue(queue); | |
8d3d53b3 AB |
728 | if (err) { |
729 | /* xenvif_init_queue() cleans up after itself on | |
730 | * failure, but we need to clean up any previously | |
731 | * initialised queues. Set num_queues to i so that | |
732 | * earlier queues can be destroyed using the regular | |
733 | * disconnect logic. | |
734 | */ | |
f7b50c4e | 735 | be->vif->num_queues = queue_index; |
e9ce7cb6 | 736 | goto err; |
8d3d53b3 | 737 | } |
e9ce7cb6 WL |
738 | |
739 | queue->remaining_credit = credit_bytes; | |
07ff890d | 740 | queue->credit_usec = credit_usec; |
e9ce7cb6 WL |
741 | |
742 | err = connect_rings(be, queue); | |
8d3d53b3 AB |
743 | if (err) { |
744 | /* connect_rings() cleans up after itself on failure, | |
745 | * but we need to clean up after xenvif_init_queue() here, | |
746 | * and also clean up any previously initialised queues. | |
747 | */ | |
748 | xenvif_deinit_queue(queue); | |
f7b50c4e | 749 | be->vif->num_queues = queue_index; |
e9ce7cb6 | 750 | goto err; |
8d3d53b3 | 751 | } |
628fa76b WL |
752 | } |
753 | ||
f51de243 | 754 | #ifdef CONFIG_DEBUG_FS |
628fa76b | 755 | xenvif_debugfs_addif(be->vif); |
f51de243 | 756 | #endif /* CONFIG_DEBUG_FS */ |
e9ce7cb6 | 757 | |
f7b50c4e WL |
758 | /* Initialisation completed, tell core driver the number of |
759 | * active queues. | |
760 | */ | |
761 | rtnl_lock(); | |
762 | netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues); | |
763 | netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues); | |
764 | rtnl_unlock(); | |
765 | ||
e9ce7cb6 | 766 | xenvif_carrier_on(be->vif); |
f942dc25 IC |
767 | |
768 | unregister_hotplug_status_watch(be); | |
769 | err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch, | |
770 | hotplug_status_changed, | |
771 | "%s/%s", dev->nodename, "hotplug-status"); | |
ea732dff | 772 | if (!err) |
f942dc25 | 773 | be->have_hotplug_status_watch = 1; |
f942dc25 | 774 | |
e9ce7cb6 WL |
775 | netif_tx_wake_all_queues(be->vif->dev); |
776 | ||
777 | return; | |
778 | ||
779 | err: | |
f7b50c4e | 780 | if (be->vif->num_queues > 0) |
8d3d53b3 | 781 | xenvif_disconnect(be->vif); /* Clean up existing queues */ |
e9ce7cb6 WL |
782 | vfree(be->vif->queues); |
783 | be->vif->queues = NULL; | |
f7b50c4e | 784 | be->vif->num_queues = 0; |
e9ce7cb6 | 785 | return; |
f942dc25 IC |
786 | } |
787 | ||
788 | ||
e9ce7cb6 | 789 | static int connect_rings(struct backend_info *be, struct xenvif_queue *queue) |
f942dc25 | 790 | { |
f942dc25 | 791 | struct xenbus_device *dev = be->dev; |
f7b50c4e | 792 | unsigned int num_queues = queue->vif->num_queues; |
f942dc25 | 793 | unsigned long tx_ring_ref, rx_ring_ref; |
e9ce7cb6 | 794 | unsigned int tx_evtchn, rx_evtchn; |
f942dc25 | 795 | int err; |
8d3d53b3 AB |
796 | char *xspath; |
797 | size_t xspathsize; | |
798 | const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */ | |
799 | ||
800 | /* If the frontend requested 1 queue, or we have fallen back | |
801 | * to single queue due to lack of frontend support for multi- | |
802 | * queue, expect the remaining XenStore keys in the toplevel | |
803 | * directory. Otherwise, expect them in a subdirectory called | |
804 | * queue-N. | |
805 | */ | |
806 | if (num_queues == 1) { | |
807 | xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL); | |
808 | if (!xspath) { | |
809 | xenbus_dev_fatal(dev, -ENOMEM, | |
810 | "reading ring references"); | |
811 | return -ENOMEM; | |
812 | } | |
813 | strcpy(xspath, dev->otherend); | |
814 | } else { | |
815 | xspathsize = strlen(dev->otherend) + xenstore_path_ext_size; | |
816 | xspath = kzalloc(xspathsize, GFP_KERNEL); | |
817 | if (!xspath) { | |
818 | xenbus_dev_fatal(dev, -ENOMEM, | |
819 | "reading ring references"); | |
820 | return -ENOMEM; | |
821 | } | |
822 | snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, | |
823 | queue->id); | |
824 | } | |
f942dc25 | 825 | |
8d3d53b3 | 826 | err = xenbus_gather(XBT_NIL, xspath, |
f942dc25 | 827 | "tx-ring-ref", "%lu", &tx_ring_ref, |
e1f00a69 | 828 | "rx-ring-ref", "%lu", &rx_ring_ref, NULL); |
f942dc25 IC |
829 | if (err) { |
830 | xenbus_dev_fatal(dev, err, | |
e1f00a69 | 831 | "reading %s/ring-ref", |
8d3d53b3 AB |
832 | xspath); |
833 | goto err; | |
f942dc25 IC |
834 | } |
835 | ||
e1f00a69 | 836 | /* Try split event channels first, then single event channel. */ |
8d3d53b3 | 837 | err = xenbus_gather(XBT_NIL, xspath, |
e1f00a69 WL |
838 | "event-channel-tx", "%u", &tx_evtchn, |
839 | "event-channel-rx", "%u", &rx_evtchn, NULL); | |
840 | if (err < 0) { | |
8d3d53b3 | 841 | err = xenbus_scanf(XBT_NIL, xspath, |
e1f00a69 WL |
842 | "event-channel", "%u", &tx_evtchn); |
843 | if (err < 0) { | |
844 | xenbus_dev_fatal(dev, err, | |
845 | "reading %s/event-channel(-tx/rx)", | |
8d3d53b3 AB |
846 | xspath); |
847 | goto err; | |
e1f00a69 WL |
848 | } |
849 | rx_evtchn = tx_evtchn; | |
850 | } | |
851 | ||
e9ce7cb6 WL |
852 | /* Map the shared frame, irq etc. */ |
853 | err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref, | |
854 | tx_evtchn, rx_evtchn); | |
855 | if (err) { | |
856 | xenbus_dev_fatal(dev, err, | |
857 | "mapping shared-frames %lu/%lu port tx %u rx %u", | |
858 | tx_ring_ref, rx_ring_ref, | |
859 | tx_evtchn, rx_evtchn); | |
8d3d53b3 | 860 | goto err; |
e9ce7cb6 WL |
861 | } |
862 | ||
8d3d53b3 AB |
863 | err = 0; |
864 | err: /* Regular return falls through with err == 0 */ | |
865 | kfree(xspath); | |
866 | return err; | |
e9ce7cb6 WL |
867 | } |
868 | ||
869 | static int read_xenbus_vif_flags(struct backend_info *be) | |
870 | { | |
871 | struct xenvif *vif = be->vif; | |
872 | struct xenbus_device *dev = be->dev; | |
873 | unsigned int rx_copy; | |
874 | int err, val; | |
875 | ||
f942dc25 IC |
876 | err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u", |
877 | &rx_copy); | |
878 | if (err == -ENOENT) { | |
879 | err = 0; | |
880 | rx_copy = 0; | |
881 | } | |
882 | if (err < 0) { | |
883 | xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy", | |
884 | dev->otherend); | |
885 | return err; | |
886 | } | |
887 | if (!rx_copy) | |
888 | return -EOPNOTSUPP; | |
889 | ||
bc96f648 | 890 | if (xenbus_scanf(XBT_NIL, dev->otherend, |
26c0e102 DV |
891 | "feature-rx-notify", "%d", &val) < 0) |
892 | val = 0; | |
893 | if (!val) { | |
894 | /* - Reduce drain timeout to poll more frequently for | |
895 | * Rx requests. | |
896 | * - Disable Rx stall detection. | |
897 | */ | |
898 | be->vif->drain_timeout = msecs_to_jiffies(30); | |
899 | be->vif->stall_timeout = 0; | |
f942dc25 IC |
900 | } |
901 | ||
902 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg", | |
903 | "%d", &val) < 0) | |
904 | val = 0; | |
905 | vif->can_sg = !!val; | |
906 | ||
82cada22 PD |
907 | vif->gso_mask = 0; |
908 | vif->gso_prefix_mask = 0; | |
909 | ||
f942dc25 IC |
910 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4", |
911 | "%d", &val) < 0) | |
912 | val = 0; | |
82cada22 PD |
913 | if (val) |
914 | vif->gso_mask |= GSO_BIT(TCPV4); | |
f942dc25 IC |
915 | |
916 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix", | |
917 | "%d", &val) < 0) | |
918 | val = 0; | |
82cada22 PD |
919 | if (val) |
920 | vif->gso_prefix_mask |= GSO_BIT(TCPV4); | |
921 | ||
922 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6", | |
923 | "%d", &val) < 0) | |
924 | val = 0; | |
925 | if (val) | |
926 | vif->gso_mask |= GSO_BIT(TCPV6); | |
927 | ||
928 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix", | |
929 | "%d", &val) < 0) | |
930 | val = 0; | |
931 | if (val) | |
932 | vif->gso_prefix_mask |= GSO_BIT(TCPV6); | |
933 | ||
934 | if (vif->gso_mask & vif->gso_prefix_mask) { | |
935 | xenbus_dev_fatal(dev, err, | |
936 | "%s: gso and gso prefix flags are not " | |
937 | "mutually exclusive", | |
938 | dev->otherend); | |
939 | return -EOPNOTSUPP; | |
940 | } | |
f942dc25 IC |
941 | |
942 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload", | |
943 | "%d", &val) < 0) | |
944 | val = 0; | |
146c8a77 PD |
945 | vif->ip_csum = !val; |
946 | ||
947 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload", | |
948 | "%d", &val) < 0) | |
949 | val = 0; | |
950 | vif->ipv6_csum = !!val; | |
f942dc25 | 951 | |
f942dc25 IC |
952 | return 0; |
953 | } | |
954 | ||
f942dc25 IC |
955 | static const struct xenbus_device_id netback_ids[] = { |
956 | { "vif" }, | |
957 | { "" } | |
958 | }; | |
959 | ||
95afae48 DV |
960 | static struct xenbus_driver netback_driver = { |
961 | .ids = netback_ids, | |
f942dc25 IC |
962 | .probe = netback_probe, |
963 | .remove = netback_remove, | |
964 | .uevent = netback_uevent, | |
965 | .otherend_changed = frontend_changed, | |
95afae48 | 966 | }; |
f942dc25 IC |
967 | |
968 | int xenvif_xenbus_init(void) | |
969 | { | |
73db144b | 970 | return xenbus_register_backend(&netback_driver); |
f942dc25 | 971 | } |
b103f358 WL |
972 | |
973 | void xenvif_xenbus_fini(void) | |
974 | { | |
975 | return xenbus_unregister_driver(&netback_driver); | |
976 | } |