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; |
31a41898 IC |
37 | |
38 | const char *hotplug_script; | |
f942dc25 IC |
39 | }; |
40 | ||
e9ce7cb6 WL |
41 | static int connect_rings(struct backend_info *be, struct xenvif_queue *queue); |
42 | static void connect(struct backend_info *be); | |
43 | static int read_xenbus_vif_flags(struct backend_info *be); | |
2dd34339 | 44 | static int backend_create_xenvif(struct backend_info *be); |
f942dc25 | 45 | static void unregister_hotplug_status_watch(struct backend_info *be); |
edafc132 | 46 | static void xen_unregister_watchers(struct xenvif *vif); |
dc62ccac DV |
47 | static void set_backend_state(struct backend_info *be, |
48 | enum xenbus_state state); | |
f942dc25 | 49 | |
f51de243 ZK |
50 | #ifdef CONFIG_DEBUG_FS |
51 | struct dentry *xen_netback_dbg_root = NULL; | |
52 | ||
53 | static int xenvif_read_io_ring(struct seq_file *m, void *v) | |
54 | { | |
55 | struct xenvif_queue *queue = m->private; | |
56 | struct xen_netif_tx_back_ring *tx_ring = &queue->tx; | |
57 | struct xen_netif_rx_back_ring *rx_ring = &queue->rx; | |
f48da8b1 | 58 | struct netdev_queue *dev_queue; |
f51de243 ZK |
59 | |
60 | if (tx_ring->sring) { | |
61 | struct xen_netif_tx_sring *sring = tx_ring->sring; | |
62 | ||
63 | seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id, | |
64 | tx_ring->nr_ents); | |
65 | seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n", | |
66 | sring->req_prod, | |
67 | sring->req_prod - sring->rsp_prod, | |
68 | tx_ring->req_cons, | |
69 | tx_ring->req_cons - sring->rsp_prod, | |
70 | sring->req_event, | |
71 | sring->req_event - sring->rsp_prod); | |
72 | seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n", | |
73 | sring->rsp_prod, | |
74 | tx_ring->rsp_prod_pvt, | |
75 | tx_ring->rsp_prod_pvt - sring->rsp_prod, | |
76 | sring->rsp_event, | |
77 | sring->rsp_event - sring->rsp_prod); | |
78 | seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n", | |
79 | queue->pending_prod, | |
80 | queue->pending_cons, | |
81 | nr_pending_reqs(queue)); | |
82 | seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n", | |
83 | queue->dealloc_prod, | |
84 | queue->dealloc_cons, | |
85 | queue->dealloc_prod - queue->dealloc_cons); | |
86 | } | |
87 | ||
88 | if (rx_ring->sring) { | |
89 | struct xen_netif_rx_sring *sring = rx_ring->sring; | |
90 | ||
91 | seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents); | |
92 | seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n", | |
93 | sring->req_prod, | |
94 | sring->req_prod - sring->rsp_prod, | |
95 | rx_ring->req_cons, | |
96 | rx_ring->req_cons - sring->rsp_prod, | |
97 | sring->req_event, | |
98 | sring->req_event - sring->rsp_prod); | |
99 | seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n", | |
100 | sring->rsp_prod, | |
101 | rx_ring->rsp_prod_pvt, | |
102 | rx_ring->rsp_prod_pvt - sring->rsp_prod, | |
103 | sring->rsp_event, | |
104 | sring->rsp_event - sring->rsp_prod); | |
105 | } | |
106 | ||
107 | seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n" | |
108 | "Credit timer_pending: %d, credit: %lu, usec: %lu\n" | |
109 | "remaining: %lu, expires: %lu, now: %lu\n", | |
110 | queue->napi.state, queue->napi.weight, | |
111 | skb_queue_len(&queue->tx_queue), | |
112 | timer_pending(&queue->credit_timeout), | |
113 | queue->credit_bytes, | |
114 | queue->credit_usec, | |
115 | queue->remaining_credit, | |
116 | queue->credit_timeout.expires, | |
117 | jiffies); | |
118 | ||
f48da8b1 DV |
119 | dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id); |
120 | ||
121 | seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n", | |
122 | queue->rx_queue_len, queue->rx_queue_max, | |
123 | skb_queue_len(&queue->rx_queue), | |
124 | netif_tx_queue_stopped(dev_queue) ? "stopped" : "running"); | |
125 | ||
f51de243 ZK |
126 | return 0; |
127 | } | |
128 | ||
129 | #define XENVIF_KICK_STR "kick" | |
5c807005 | 130 | #define BUFFER_SIZE 32 |
f51de243 ZK |
131 | |
132 | static ssize_t | |
133 | xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count, | |
134 | loff_t *ppos) | |
135 | { | |
136 | struct xenvif_queue *queue = | |
137 | ((struct seq_file *)filp->private_data)->private; | |
138 | int len; | |
5c807005 | 139 | char write[BUFFER_SIZE]; |
f51de243 ZK |
140 | |
141 | /* don't allow partial writes and check the length */ | |
142 | if (*ppos != 0) | |
143 | return 0; | |
5c807005 | 144 | if (count >= sizeof(write)) |
f51de243 ZK |
145 | return -ENOSPC; |
146 | ||
147 | len = simple_write_to_buffer(write, | |
5c807005 | 148 | sizeof(write) - 1, |
f51de243 ZK |
149 | ppos, |
150 | buf, | |
151 | count); | |
152 | if (len < 0) | |
153 | return len; | |
154 | ||
5c807005 WL |
155 | write[len] = '\0'; |
156 | ||
f51de243 ZK |
157 | if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1)) |
158 | xenvif_interrupt(0, (void *)queue); | |
159 | else { | |
160 | pr_warn("Unknown command to io_ring_q%d. Available: kick\n", | |
161 | queue->id); | |
162 | count = -EINVAL; | |
163 | } | |
164 | return count; | |
165 | } | |
166 | ||
167 | static int xenvif_dump_open(struct inode *inode, struct file *filp) | |
168 | { | |
169 | int ret; | |
170 | void *queue = NULL; | |
171 | ||
172 | if (inode->i_private) | |
173 | queue = inode->i_private; | |
174 | ret = single_open(filp, xenvif_read_io_ring, queue); | |
175 | filp->f_mode |= FMODE_PWRITE; | |
176 | return ret; | |
177 | } | |
178 | ||
179 | static const struct file_operations xenvif_dbg_io_ring_ops_fops = { | |
180 | .owner = THIS_MODULE, | |
181 | .open = xenvif_dump_open, | |
182 | .read = seq_read, | |
183 | .llseek = seq_lseek, | |
184 | .release = single_release, | |
185 | .write = xenvif_write_io_ring, | |
186 | }; | |
187 | ||
628fa76b | 188 | static void xenvif_debugfs_addif(struct xenvif *vif) |
f51de243 ZK |
189 | { |
190 | struct dentry *pfile; | |
f51de243 ZK |
191 | int i; |
192 | ||
193 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
194 | return; | |
195 | ||
196 | vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name, | |
197 | xen_netback_dbg_root); | |
198 | if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) { | |
199 | for (i = 0; i < vif->num_queues; ++i) { | |
200 | char filename[sizeof("io_ring_q") + 4]; | |
201 | ||
202 | snprintf(filename, sizeof(filename), "io_ring_q%d", i); | |
203 | pfile = debugfs_create_file(filename, | |
204 | S_IRUSR | S_IWUSR, | |
205 | vif->xenvif_dbg_root, | |
206 | &vif->queues[i], | |
207 | &xenvif_dbg_io_ring_ops_fops); | |
208 | if (IS_ERR_OR_NULL(pfile)) | |
209 | pr_warn("Creation of io_ring file returned %ld!\n", | |
210 | PTR_ERR(pfile)); | |
211 | } | |
212 | } else | |
213 | netdev_warn(vif->dev, | |
214 | "Creation of vif debugfs dir returned %ld!\n", | |
215 | PTR_ERR(vif->xenvif_dbg_root)); | |
216 | } | |
217 | ||
218 | static void xenvif_debugfs_delif(struct xenvif *vif) | |
219 | { | |
220 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
221 | return; | |
222 | ||
223 | if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) | |
224 | debugfs_remove_recursive(vif->xenvif_dbg_root); | |
225 | vif->xenvif_dbg_root = NULL; | |
226 | } | |
227 | #endif /* CONFIG_DEBUG_FS */ | |
228 | ||
f942dc25 IC |
229 | static int netback_remove(struct xenbus_device *dev) |
230 | { | |
231 | struct backend_info *be = dev_get_drvdata(&dev->dev); | |
232 | ||
dc62ccac DV |
233 | set_backend_state(be, XenbusStateClosed); |
234 | ||
f942dc25 IC |
235 | unregister_hotplug_status_watch(be); |
236 | if (be->vif) { | |
237 | kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE); | |
edafc132 | 238 | xen_unregister_watchers(be->vif); |
f942dc25 | 239 | xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status"); |
279f438e | 240 | xenvif_free(be->vif); |
f942dc25 IC |
241 | be->vif = NULL; |
242 | } | |
31a41898 | 243 | kfree(be->hotplug_script); |
f942dc25 IC |
244 | kfree(be); |
245 | dev_set_drvdata(&dev->dev, NULL); | |
246 | return 0; | |
247 | } | |
248 | ||
249 | ||
250 | /** | |
251 | * Entry point to this code when a new device is created. Allocate the basic | |
252 | * structures and switch to InitWait. | |
253 | */ | |
254 | static int netback_probe(struct xenbus_device *dev, | |
255 | const struct xenbus_device_id *id) | |
256 | { | |
257 | const char *message; | |
258 | struct xenbus_transaction xbt; | |
259 | int err; | |
260 | int sg; | |
31a41898 | 261 | const char *script; |
f942dc25 IC |
262 | struct backend_info *be = kzalloc(sizeof(struct backend_info), |
263 | GFP_KERNEL); | |
264 | if (!be) { | |
265 | xenbus_dev_fatal(dev, -ENOMEM, | |
266 | "allocating backend structure"); | |
267 | return -ENOMEM; | |
268 | } | |
269 | ||
270 | be->dev = dev; | |
271 | dev_set_drvdata(&dev->dev, be); | |
272 | ||
273 | sg = 1; | |
274 | ||
275 | do { | |
276 | err = xenbus_transaction_start(&xbt); | |
277 | if (err) { | |
278 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
279 | goto fail; | |
280 | } | |
281 | ||
282 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg); | |
283 | if (err) { | |
284 | message = "writing feature-sg"; | |
285 | goto abort_transaction; | |
286 | } | |
287 | ||
288 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", | |
289 | "%d", sg); | |
290 | if (err) { | |
291 | message = "writing feature-gso-tcpv4"; | |
292 | goto abort_transaction; | |
293 | } | |
294 | ||
a9468587 PD |
295 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6", |
296 | "%d", sg); | |
297 | if (err) { | |
298 | message = "writing feature-gso-tcpv6"; | |
299 | goto abort_transaction; | |
300 | } | |
301 | ||
2eba61d5 PD |
302 | /* We support partial checksum setup for IPv6 packets */ |
303 | err = xenbus_printf(xbt, dev->nodename, | |
304 | "feature-ipv6-csum-offload", | |
305 | "%d", 1); | |
306 | if (err) { | |
307 | message = "writing feature-ipv6-csum-offload"; | |
308 | goto abort_transaction; | |
309 | } | |
310 | ||
f942dc25 IC |
311 | /* We support rx-copy path. */ |
312 | err = xenbus_printf(xbt, dev->nodename, | |
313 | "feature-rx-copy", "%d", 1); | |
314 | if (err) { | |
315 | message = "writing feature-rx-copy"; | |
316 | goto abort_transaction; | |
317 | } | |
318 | ||
319 | /* | |
320 | * We don't support rx-flip path (except old guests who don't | |
321 | * grok this feature flag). | |
322 | */ | |
323 | err = xenbus_printf(xbt, dev->nodename, | |
324 | "feature-rx-flip", "%d", 0); | |
325 | if (err) { | |
326 | message = "writing feature-rx-flip"; | |
327 | goto abort_transaction; | |
328 | } | |
329 | ||
330 | err = xenbus_transaction_end(xbt, 0); | |
331 | } while (err == -EAGAIN); | |
332 | ||
333 | if (err) { | |
334 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
335 | goto fail; | |
336 | } | |
337 | ||
e1f00a69 WL |
338 | /* |
339 | * Split event channels support, this is optional so it is not | |
340 | * put inside the above loop. | |
341 | */ | |
342 | err = xenbus_printf(XBT_NIL, dev->nodename, | |
343 | "feature-split-event-channels", | |
344 | "%u", separate_tx_rx_irq); | |
345 | if (err) | |
8ef2c3bc | 346 | pr_debug("Error writing feature-split-event-channels\n"); |
e1f00a69 | 347 | |
8d3d53b3 AB |
348 | /* Multi-queue support: This is an optional feature. */ |
349 | err = xenbus_printf(XBT_NIL, dev->nodename, | |
350 | "multi-queue-max-queues", "%u", xenvif_max_queues); | |
351 | if (err) | |
352 | pr_debug("Error writing multi-queue-max-queues\n"); | |
353 | ||
31a41898 IC |
354 | script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL); |
355 | if (IS_ERR(script)) { | |
356 | err = PTR_ERR(script); | |
357 | xenbus_dev_fatal(dev, err, "reading script"); | |
358 | goto fail; | |
359 | } | |
360 | ||
361 | be->hotplug_script = script; | |
362 | ||
f942dc25 IC |
363 | err = xenbus_switch_state(dev, XenbusStateInitWait); |
364 | if (err) | |
365 | goto fail; | |
366 | ||
ea732dff PD |
367 | be->state = XenbusStateInitWait; |
368 | ||
f942dc25 | 369 | /* This kicks hotplug scripts, so do it immediately. */ |
2dd34339 AK |
370 | err = backend_create_xenvif(be); |
371 | if (err) | |
372 | goto fail; | |
f942dc25 IC |
373 | |
374 | return 0; | |
375 | ||
376 | abort_transaction: | |
377 | xenbus_transaction_end(xbt, 1); | |
378 | xenbus_dev_fatal(dev, err, "%s", message); | |
379 | fail: | |
8ef2c3bc | 380 | pr_debug("failed\n"); |
f942dc25 IC |
381 | netback_remove(dev); |
382 | return err; | |
383 | } | |
384 | ||
385 | ||
386 | /* | |
387 | * Handle the creation of the hotplug script environment. We add the script | |
388 | * and vif variables to the environment, for the benefit of the vif-* hotplug | |
389 | * scripts. | |
390 | */ | |
391 | static int netback_uevent(struct xenbus_device *xdev, | |
392 | struct kobj_uevent_env *env) | |
393 | { | |
394 | struct backend_info *be = dev_get_drvdata(&xdev->dev); | |
f942dc25 | 395 | |
31a41898 IC |
396 | if (!be) |
397 | return 0; | |
398 | ||
399 | if (add_uevent_var(env, "script=%s", be->hotplug_script)) | |
400 | return -ENOMEM; | |
401 | ||
402 | if (!be->vif) | |
f942dc25 IC |
403 | return 0; |
404 | ||
405 | return add_uevent_var(env, "vif=%s", be->vif->dev->name); | |
406 | } | |
407 | ||
408 | ||
2dd34339 | 409 | static int backend_create_xenvif(struct backend_info *be) |
f942dc25 IC |
410 | { |
411 | int err; | |
412 | long handle; | |
413 | struct xenbus_device *dev = be->dev; | |
f15650b7 | 414 | struct xenvif *vif; |
f942dc25 IC |
415 | |
416 | if (be->vif != NULL) | |
2dd34339 | 417 | return 0; |
f942dc25 IC |
418 | |
419 | err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle); | |
420 | if (err != 1) { | |
421 | xenbus_dev_fatal(dev, err, "reading handle"); | |
2dd34339 | 422 | return (err < 0) ? err : -EINVAL; |
f942dc25 IC |
423 | } |
424 | ||
f15650b7 JB |
425 | vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle); |
426 | if (IS_ERR(vif)) { | |
427 | err = PTR_ERR(vif); | |
f942dc25 | 428 | xenbus_dev_fatal(dev, err, "creating interface"); |
2dd34339 | 429 | return err; |
f942dc25 | 430 | } |
f15650b7 | 431 | be->vif = vif; |
f942dc25 IC |
432 | |
433 | kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE); | |
2dd34339 | 434 | return 0; |
f942dc25 IC |
435 | } |
436 | ||
ea732dff | 437 | static void backend_disconnect(struct backend_info *be) |
f942dc25 | 438 | { |
f51de243 | 439 | if (be->vif) { |
edafc132 | 440 | xen_unregister_watchers(be->vif); |
f51de243 ZK |
441 | #ifdef CONFIG_DEBUG_FS |
442 | xenvif_debugfs_delif(be->vif); | |
443 | #endif /* CONFIG_DEBUG_FS */ | |
279f438e | 444 | xenvif_disconnect(be->vif); |
f51de243 | 445 | } |
279f438e PD |
446 | } |
447 | ||
ea732dff | 448 | static void backend_connect(struct backend_info *be) |
279f438e | 449 | { |
ea732dff PD |
450 | if (be->vif) |
451 | connect(be); | |
452 | } | |
279f438e | 453 | |
ea732dff PD |
454 | static inline void backend_switch_state(struct backend_info *be, |
455 | enum xenbus_state state) | |
456 | { | |
457 | struct xenbus_device *dev = be->dev; | |
458 | ||
459 | pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state)); | |
460 | be->state = state; | |
461 | ||
462 | /* If we are waiting for a hotplug script then defer the | |
463 | * actual xenbus state change. | |
464 | */ | |
465 | if (!be->have_hotplug_status_watch) | |
466 | xenbus_switch_state(dev, state); | |
467 | } | |
468 | ||
469 | /* Handle backend state transitions: | |
470 | * | |
471 | * The backend state starts in InitWait and the following transitions are | |
472 | * allowed. | |
473 | * | |
474 | * InitWait -> Connected | |
475 | * | |
476 | * ^ \ | | |
477 | * | \ | | |
478 | * | \ | | |
479 | * | \ | | |
480 | * | \ | | |
481 | * | \ | | |
482 | * | V V | |
483 | * | |
484 | * Closed <-> Closing | |
485 | * | |
486 | * The state argument specifies the eventual state of the backend and the | |
487 | * function transitions to that state via the shortest path. | |
488 | */ | |
489 | static void set_backend_state(struct backend_info *be, | |
490 | enum xenbus_state state) | |
491 | { | |
492 | while (be->state != state) { | |
493 | switch (be->state) { | |
494 | case XenbusStateClosed: | |
495 | switch (state) { | |
496 | case XenbusStateInitWait: | |
497 | case XenbusStateConnected: | |
498 | pr_info("%s: prepare for reconnect\n", | |
499 | be->dev->nodename); | |
500 | backend_switch_state(be, XenbusStateInitWait); | |
501 | break; | |
502 | case XenbusStateClosing: | |
503 | backend_switch_state(be, XenbusStateClosing); | |
504 | break; | |
505 | default: | |
506 | BUG(); | |
507 | } | |
508 | break; | |
509 | case XenbusStateInitWait: | |
510 | switch (state) { | |
511 | case XenbusStateConnected: | |
512 | backend_connect(be); | |
513 | backend_switch_state(be, XenbusStateConnected); | |
514 | break; | |
515 | case XenbusStateClosing: | |
516 | case XenbusStateClosed: | |
517 | backend_switch_state(be, XenbusStateClosing); | |
518 | break; | |
519 | default: | |
520 | BUG(); | |
521 | } | |
522 | break; | |
523 | case XenbusStateConnected: | |
524 | switch (state) { | |
525 | case XenbusStateInitWait: | |
526 | case XenbusStateClosing: | |
527 | case XenbusStateClosed: | |
528 | backend_disconnect(be); | |
529 | backend_switch_state(be, XenbusStateClosing); | |
530 | break; | |
531 | default: | |
532 | BUG(); | |
533 | } | |
534 | break; | |
535 | case XenbusStateClosing: | |
536 | switch (state) { | |
537 | case XenbusStateInitWait: | |
538 | case XenbusStateConnected: | |
539 | case XenbusStateClosed: | |
540 | backend_switch_state(be, XenbusStateClosed); | |
541 | break; | |
542 | default: | |
543 | BUG(); | |
544 | } | |
545 | break; | |
546 | default: | |
547 | BUG(); | |
548 | } | |
f942dc25 IC |
549 | } |
550 | } | |
551 | ||
552 | /** | |
553 | * Callback received when the frontend's state changes. | |
554 | */ | |
555 | static void frontend_changed(struct xenbus_device *dev, | |
556 | enum xenbus_state frontend_state) | |
557 | { | |
558 | struct backend_info *be = dev_get_drvdata(&dev->dev); | |
559 | ||
ea732dff | 560 | pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state)); |
f942dc25 IC |
561 | |
562 | be->frontend_state = frontend_state; | |
563 | ||
564 | switch (frontend_state) { | |
565 | case XenbusStateInitialising: | |
ea732dff | 566 | set_backend_state(be, XenbusStateInitWait); |
f942dc25 IC |
567 | break; |
568 | ||
569 | case XenbusStateInitialised: | |
570 | break; | |
571 | ||
572 | case XenbusStateConnected: | |
ea732dff | 573 | set_backend_state(be, XenbusStateConnected); |
f942dc25 IC |
574 | break; |
575 | ||
576 | case XenbusStateClosing: | |
ea732dff | 577 | set_backend_state(be, XenbusStateClosing); |
f942dc25 IC |
578 | break; |
579 | ||
580 | case XenbusStateClosed: | |
ea732dff | 581 | set_backend_state(be, XenbusStateClosed); |
f942dc25 IC |
582 | if (xenbus_dev_is_online(dev)) |
583 | break; | |
584 | /* fall through if not online */ | |
585 | case XenbusStateUnknown: | |
ea732dff | 586 | set_backend_state(be, XenbusStateClosed); |
f942dc25 IC |
587 | device_unregister(&dev->dev); |
588 | break; | |
589 | ||
590 | default: | |
591 | xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend", | |
592 | frontend_state); | |
593 | break; | |
594 | } | |
595 | } | |
596 | ||
597 | ||
598 | static void xen_net_read_rate(struct xenbus_device *dev, | |
599 | unsigned long *bytes, unsigned long *usec) | |
600 | { | |
601 | char *s, *e; | |
602 | unsigned long b, u; | |
603 | char *ratestr; | |
604 | ||
605 | /* Default to unlimited bandwidth. */ | |
606 | *bytes = ~0UL; | |
607 | *usec = 0; | |
608 | ||
609 | ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL); | |
610 | if (IS_ERR(ratestr)) | |
611 | return; | |
612 | ||
613 | s = ratestr; | |
614 | b = simple_strtoul(s, &e, 10); | |
615 | if ((s == e) || (*e != ',')) | |
616 | goto fail; | |
617 | ||
618 | s = e + 1; | |
619 | u = simple_strtoul(s, &e, 10); | |
620 | if ((s == e) || (*e != '\0')) | |
621 | goto fail; | |
622 | ||
623 | *bytes = b; | |
624 | *usec = u; | |
625 | ||
626 | kfree(ratestr); | |
627 | return; | |
628 | ||
629 | fail: | |
630 | pr_warn("Failed to parse network rate limit. Traffic unlimited.\n"); | |
631 | kfree(ratestr); | |
632 | } | |
633 | ||
634 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
635 | { | |
636 | char *s, *e, *macstr; | |
637 | int i; | |
638 | ||
639 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
640 | if (IS_ERR(macstr)) | |
641 | return PTR_ERR(macstr); | |
642 | ||
643 | for (i = 0; i < ETH_ALEN; i++) { | |
644 | mac[i] = simple_strtoul(s, &e, 16); | |
645 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
646 | kfree(macstr); | |
647 | return -ENOENT; | |
648 | } | |
649 | s = e+1; | |
650 | } | |
651 | ||
652 | kfree(macstr); | |
653 | return 0; | |
654 | } | |
655 | ||
edafc132 PI |
656 | static void xen_net_rate_changed(struct xenbus_watch *watch, |
657 | const char **vec, unsigned int len) | |
658 | { | |
659 | struct xenvif *vif = container_of(watch, struct xenvif, credit_watch); | |
660 | struct xenbus_device *dev = xenvif_to_xenbus_device(vif); | |
661 | unsigned long credit_bytes; | |
662 | unsigned long credit_usec; | |
663 | unsigned int queue_index; | |
664 | ||
665 | xen_net_read_rate(dev, &credit_bytes, &credit_usec); | |
666 | for (queue_index = 0; queue_index < vif->num_queues; queue_index++) { | |
667 | struct xenvif_queue *queue = &vif->queues[queue_index]; | |
668 | ||
669 | queue->credit_bytes = credit_bytes; | |
670 | queue->credit_usec = credit_usec; | |
671 | if (!mod_timer_pending(&queue->credit_timeout, jiffies) && | |
672 | queue->remaining_credit > queue->credit_bytes) { | |
673 | queue->remaining_credit = queue->credit_bytes; | |
674 | } | |
675 | } | |
676 | } | |
677 | ||
678 | static int xen_register_watchers(struct xenbus_device *dev, struct xenvif *vif) | |
679 | { | |
680 | int err = 0; | |
681 | char *node; | |
682 | unsigned maxlen = strlen(dev->nodename) + sizeof("/rate"); | |
683 | ||
684 | node = kmalloc(maxlen, GFP_KERNEL); | |
685 | if (!node) | |
686 | return -ENOMEM; | |
687 | snprintf(node, maxlen, "%s/rate", dev->nodename); | |
688 | vif->credit_watch.node = node; | |
689 | vif->credit_watch.callback = xen_net_rate_changed; | |
690 | err = register_xenbus_watch(&vif->credit_watch); | |
691 | if (err) { | |
692 | pr_err("Failed to set watcher %s\n", vif->credit_watch.node); | |
693 | kfree(node); | |
694 | vif->credit_watch.node = NULL; | |
695 | vif->credit_watch.callback = NULL; | |
696 | } | |
697 | return err; | |
698 | } | |
699 | ||
700 | static void xen_unregister_watchers(struct xenvif *vif) | |
701 | { | |
702 | if (vif->credit_watch.node) { | |
703 | unregister_xenbus_watch(&vif->credit_watch); | |
704 | kfree(vif->credit_watch.node); | |
705 | vif->credit_watch.node = NULL; | |
706 | } | |
707 | } | |
708 | ||
f942dc25 IC |
709 | static void unregister_hotplug_status_watch(struct backend_info *be) |
710 | { | |
711 | if (be->have_hotplug_status_watch) { | |
712 | unregister_xenbus_watch(&be->hotplug_status_watch); | |
713 | kfree(be->hotplug_status_watch.node); | |
714 | } | |
715 | be->have_hotplug_status_watch = 0; | |
716 | } | |
717 | ||
718 | static void hotplug_status_changed(struct xenbus_watch *watch, | |
719 | const char **vec, | |
720 | unsigned int vec_size) | |
721 | { | |
722 | struct backend_info *be = container_of(watch, | |
723 | struct backend_info, | |
724 | hotplug_status_watch); | |
725 | char *str; | |
726 | unsigned int len; | |
727 | ||
728 | str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len); | |
729 | if (IS_ERR(str)) | |
730 | return; | |
731 | if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) { | |
ea732dff PD |
732 | /* Complete any pending state change */ |
733 | xenbus_switch_state(be->dev, be->state); | |
734 | ||
f942dc25 IC |
735 | /* Not interested in this watch anymore. */ |
736 | unregister_hotplug_status_watch(be); | |
737 | } | |
738 | kfree(str); | |
739 | } | |
740 | ||
741 | static void connect(struct backend_info *be) | |
742 | { | |
743 | int err; | |
744 | struct xenbus_device *dev = be->dev; | |
e9ce7cb6 WL |
745 | unsigned long credit_bytes, credit_usec; |
746 | unsigned int queue_index; | |
8d3d53b3 | 747 | unsigned int requested_num_queues; |
e9ce7cb6 | 748 | struct xenvif_queue *queue; |
f942dc25 | 749 | |
8d3d53b3 AB |
750 | /* Check whether the frontend requested multiple queues |
751 | * and read the number requested. | |
752 | */ | |
753 | err = xenbus_scanf(XBT_NIL, dev->otherend, | |
754 | "multi-queue-num-queues", | |
755 | "%u", &requested_num_queues); | |
756 | if (err < 0) { | |
757 | requested_num_queues = 1; /* Fall back to single queue */ | |
758 | } else if (requested_num_queues > xenvif_max_queues) { | |
759 | /* buggy or malicious guest */ | |
760 | xenbus_dev_fatal(dev, err, | |
761 | "guest requested %u queues, exceeding the maximum of %u.", | |
762 | requested_num_queues, xenvif_max_queues); | |
763 | return; | |
764 | } | |
765 | ||
f942dc25 IC |
766 | err = xen_net_read_mac(dev, be->vif->fe_dev_addr); |
767 | if (err) { | |
768 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
769 | return; | |
770 | } | |
771 | ||
e9ce7cb6 | 772 | xen_net_read_rate(dev, &credit_bytes, &credit_usec); |
edafc132 | 773 | xen_register_watchers(dev, be->vif); |
e9ce7cb6 WL |
774 | read_xenbus_vif_flags(be); |
775 | ||
8d3d53b3 | 776 | /* Use the number of queues requested by the frontend */ |
e9ce7cb6 WL |
777 | be->vif->queues = vzalloc(requested_num_queues * |
778 | sizeof(struct xenvif_queue)); | |
f7b50c4e | 779 | be->vif->num_queues = requested_num_queues; |
ecf08d2d | 780 | be->vif->stalled_queues = requested_num_queues; |
e9ce7cb6 WL |
781 | |
782 | for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) { | |
783 | queue = &be->vif->queues[queue_index]; | |
784 | queue->vif = be->vif; | |
785 | queue->id = queue_index; | |
786 | snprintf(queue->name, sizeof(queue->name), "%s-q%u", | |
787 | be->vif->dev->name, queue->id); | |
788 | ||
789 | err = xenvif_init_queue(queue); | |
8d3d53b3 AB |
790 | if (err) { |
791 | /* xenvif_init_queue() cleans up after itself on | |
792 | * failure, but we need to clean up any previously | |
793 | * initialised queues. Set num_queues to i so that | |
794 | * earlier queues can be destroyed using the regular | |
795 | * disconnect logic. | |
796 | */ | |
f7b50c4e | 797 | be->vif->num_queues = queue_index; |
e9ce7cb6 | 798 | goto err; |
8d3d53b3 | 799 | } |
e9ce7cb6 | 800 | |
ce0e5c52 | 801 | queue->credit_bytes = credit_bytes; |
e9ce7cb6 | 802 | queue->remaining_credit = credit_bytes; |
07ff890d | 803 | queue->credit_usec = credit_usec; |
e9ce7cb6 WL |
804 | |
805 | err = connect_rings(be, queue); | |
8d3d53b3 AB |
806 | if (err) { |
807 | /* connect_rings() cleans up after itself on failure, | |
808 | * but we need to clean up after xenvif_init_queue() here, | |
809 | * and also clean up any previously initialised queues. | |
810 | */ | |
811 | xenvif_deinit_queue(queue); | |
f7b50c4e | 812 | be->vif->num_queues = queue_index; |
e9ce7cb6 | 813 | goto err; |
8d3d53b3 | 814 | } |
628fa76b WL |
815 | } |
816 | ||
f51de243 | 817 | #ifdef CONFIG_DEBUG_FS |
628fa76b | 818 | xenvif_debugfs_addif(be->vif); |
f51de243 | 819 | #endif /* CONFIG_DEBUG_FS */ |
e9ce7cb6 | 820 | |
f7b50c4e WL |
821 | /* Initialisation completed, tell core driver the number of |
822 | * active queues. | |
823 | */ | |
824 | rtnl_lock(); | |
825 | netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues); | |
826 | netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues); | |
827 | rtnl_unlock(); | |
828 | ||
e9ce7cb6 | 829 | xenvif_carrier_on(be->vif); |
f942dc25 IC |
830 | |
831 | unregister_hotplug_status_watch(be); | |
832 | err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch, | |
833 | hotplug_status_changed, | |
834 | "%s/%s", dev->nodename, "hotplug-status"); | |
ea732dff | 835 | if (!err) |
f942dc25 | 836 | be->have_hotplug_status_watch = 1; |
f942dc25 | 837 | |
e9ce7cb6 WL |
838 | netif_tx_wake_all_queues(be->vif->dev); |
839 | ||
840 | return; | |
841 | ||
842 | err: | |
f7b50c4e | 843 | if (be->vif->num_queues > 0) |
8d3d53b3 | 844 | xenvif_disconnect(be->vif); /* Clean up existing queues */ |
e9ce7cb6 WL |
845 | vfree(be->vif->queues); |
846 | be->vif->queues = NULL; | |
f7b50c4e | 847 | be->vif->num_queues = 0; |
e9ce7cb6 | 848 | return; |
f942dc25 IC |
849 | } |
850 | ||
851 | ||
e9ce7cb6 | 852 | static int connect_rings(struct backend_info *be, struct xenvif_queue *queue) |
f942dc25 | 853 | { |
f942dc25 | 854 | struct xenbus_device *dev = be->dev; |
f7b50c4e | 855 | unsigned int num_queues = queue->vif->num_queues; |
f942dc25 | 856 | unsigned long tx_ring_ref, rx_ring_ref; |
e9ce7cb6 | 857 | unsigned int tx_evtchn, rx_evtchn; |
f942dc25 | 858 | int err; |
8d3d53b3 AB |
859 | char *xspath; |
860 | size_t xspathsize; | |
861 | const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */ | |
862 | ||
863 | /* If the frontend requested 1 queue, or we have fallen back | |
864 | * to single queue due to lack of frontend support for multi- | |
865 | * queue, expect the remaining XenStore keys in the toplevel | |
866 | * directory. Otherwise, expect them in a subdirectory called | |
867 | * queue-N. | |
868 | */ | |
869 | if (num_queues == 1) { | |
870 | xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL); | |
871 | if (!xspath) { | |
872 | xenbus_dev_fatal(dev, -ENOMEM, | |
873 | "reading ring references"); | |
874 | return -ENOMEM; | |
875 | } | |
876 | strcpy(xspath, dev->otherend); | |
877 | } else { | |
878 | xspathsize = strlen(dev->otherend) + xenstore_path_ext_size; | |
879 | xspath = kzalloc(xspathsize, GFP_KERNEL); | |
880 | if (!xspath) { | |
881 | xenbus_dev_fatal(dev, -ENOMEM, | |
882 | "reading ring references"); | |
883 | return -ENOMEM; | |
884 | } | |
885 | snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, | |
886 | queue->id); | |
887 | } | |
f942dc25 | 888 | |
8d3d53b3 | 889 | err = xenbus_gather(XBT_NIL, xspath, |
f942dc25 | 890 | "tx-ring-ref", "%lu", &tx_ring_ref, |
e1f00a69 | 891 | "rx-ring-ref", "%lu", &rx_ring_ref, NULL); |
f942dc25 IC |
892 | if (err) { |
893 | xenbus_dev_fatal(dev, err, | |
e1f00a69 | 894 | "reading %s/ring-ref", |
8d3d53b3 AB |
895 | xspath); |
896 | goto err; | |
f942dc25 IC |
897 | } |
898 | ||
e1f00a69 | 899 | /* Try split event channels first, then single event channel. */ |
8d3d53b3 | 900 | err = xenbus_gather(XBT_NIL, xspath, |
e1f00a69 WL |
901 | "event-channel-tx", "%u", &tx_evtchn, |
902 | "event-channel-rx", "%u", &rx_evtchn, NULL); | |
903 | if (err < 0) { | |
8d3d53b3 | 904 | err = xenbus_scanf(XBT_NIL, xspath, |
e1f00a69 WL |
905 | "event-channel", "%u", &tx_evtchn); |
906 | if (err < 0) { | |
907 | xenbus_dev_fatal(dev, err, | |
908 | "reading %s/event-channel(-tx/rx)", | |
8d3d53b3 AB |
909 | xspath); |
910 | goto err; | |
e1f00a69 WL |
911 | } |
912 | rx_evtchn = tx_evtchn; | |
913 | } | |
914 | ||
e9ce7cb6 WL |
915 | /* Map the shared frame, irq etc. */ |
916 | err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref, | |
917 | tx_evtchn, rx_evtchn); | |
918 | if (err) { | |
919 | xenbus_dev_fatal(dev, err, | |
920 | "mapping shared-frames %lu/%lu port tx %u rx %u", | |
921 | tx_ring_ref, rx_ring_ref, | |
922 | tx_evtchn, rx_evtchn); | |
8d3d53b3 | 923 | goto err; |
e9ce7cb6 WL |
924 | } |
925 | ||
8d3d53b3 AB |
926 | err = 0; |
927 | err: /* Regular return falls through with err == 0 */ | |
928 | kfree(xspath); | |
929 | return err; | |
e9ce7cb6 WL |
930 | } |
931 | ||
932 | static int read_xenbus_vif_flags(struct backend_info *be) | |
933 | { | |
934 | struct xenvif *vif = be->vif; | |
935 | struct xenbus_device *dev = be->dev; | |
936 | unsigned int rx_copy; | |
937 | int err, val; | |
938 | ||
f942dc25 IC |
939 | err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u", |
940 | &rx_copy); | |
941 | if (err == -ENOENT) { | |
942 | err = 0; | |
943 | rx_copy = 0; | |
944 | } | |
945 | if (err < 0) { | |
946 | xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy", | |
947 | dev->otherend); | |
948 | return err; | |
949 | } | |
950 | if (!rx_copy) | |
951 | return -EOPNOTSUPP; | |
952 | ||
bc96f648 | 953 | if (xenbus_scanf(XBT_NIL, dev->otherend, |
26c0e102 DV |
954 | "feature-rx-notify", "%d", &val) < 0) |
955 | val = 0; | |
956 | if (!val) { | |
957 | /* - Reduce drain timeout to poll more frequently for | |
958 | * Rx requests. | |
959 | * - Disable Rx stall detection. | |
960 | */ | |
961 | be->vif->drain_timeout = msecs_to_jiffies(30); | |
962 | be->vif->stall_timeout = 0; | |
f942dc25 IC |
963 | } |
964 | ||
965 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg", | |
966 | "%d", &val) < 0) | |
967 | val = 0; | |
968 | vif->can_sg = !!val; | |
969 | ||
82cada22 PD |
970 | vif->gso_mask = 0; |
971 | vif->gso_prefix_mask = 0; | |
972 | ||
f942dc25 IC |
973 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4", |
974 | "%d", &val) < 0) | |
975 | val = 0; | |
82cada22 PD |
976 | if (val) |
977 | vif->gso_mask |= GSO_BIT(TCPV4); | |
f942dc25 IC |
978 | |
979 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix", | |
980 | "%d", &val) < 0) | |
981 | val = 0; | |
82cada22 PD |
982 | if (val) |
983 | vif->gso_prefix_mask |= GSO_BIT(TCPV4); | |
984 | ||
985 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6", | |
986 | "%d", &val) < 0) | |
987 | val = 0; | |
988 | if (val) | |
989 | vif->gso_mask |= GSO_BIT(TCPV6); | |
990 | ||
991 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix", | |
992 | "%d", &val) < 0) | |
993 | val = 0; | |
994 | if (val) | |
995 | vif->gso_prefix_mask |= GSO_BIT(TCPV6); | |
996 | ||
997 | if (vif->gso_mask & vif->gso_prefix_mask) { | |
998 | xenbus_dev_fatal(dev, err, | |
999 | "%s: gso and gso prefix flags are not " | |
1000 | "mutually exclusive", | |
1001 | dev->otherend); | |
1002 | return -EOPNOTSUPP; | |
1003 | } | |
f942dc25 IC |
1004 | |
1005 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload", | |
1006 | "%d", &val) < 0) | |
1007 | val = 0; | |
146c8a77 PD |
1008 | vif->ip_csum = !val; |
1009 | ||
1010 | if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload", | |
1011 | "%d", &val) < 0) | |
1012 | val = 0; | |
1013 | vif->ipv6_csum = !!val; | |
f942dc25 | 1014 | |
f942dc25 IC |
1015 | return 0; |
1016 | } | |
1017 | ||
f942dc25 IC |
1018 | static const struct xenbus_device_id netback_ids[] = { |
1019 | { "vif" }, | |
1020 | { "" } | |
1021 | }; | |
1022 | ||
95afae48 DV |
1023 | static struct xenbus_driver netback_driver = { |
1024 | .ids = netback_ids, | |
f942dc25 IC |
1025 | .probe = netback_probe, |
1026 | .remove = netback_remove, | |
1027 | .uevent = netback_uevent, | |
1028 | .otherend_changed = frontend_changed, | |
95afae48 | 1029 | }; |
f942dc25 IC |
1030 | |
1031 | int xenvif_xenbus_init(void) | |
1032 | { | |
73db144b | 1033 | return xenbus_register_backend(&netback_driver); |
f942dc25 | 1034 | } |
b103f358 WL |
1035 | |
1036 | void xenvif_xenbus_fini(void) | |
1037 | { | |
1038 | return xenbus_unregister_driver(&netback_driver); | |
1039 | } |