1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
2 
3 #define DPRINTK(fmt, ...)				\
4 	pr_debug("(%s:%d) " fmt "\n",			\
5 		 __func__, __LINE__, ##__VA_ARGS__)
6 
7 #include <linux/kernel.h>
8 #include <linux/err.h>
9 #include <linux/string.h>
10 #include <linux/ctype.h>
11 #include <linux/fcntl.h>
12 #include <linux/mm.h>
13 #include <linux/proc_fs.h>
14 #include <linux/notifier.h>
15 #include <linux/kthread.h>
16 #include <linux/mutex.h>
17 #include <linux/io.h>
18 #include <linux/module.h>
19 
20 #include <asm/page.h>
21 #include <asm/pgtable.h>
22 #include <asm/xen/hypervisor.h>
23 #include <xen/xenbus.h>
24 #include <xen/events.h>
25 #include <xen/page.h>
26 #include <xen/xen.h>
27 
28 #include <xen/platform_pci.h>
29 
30 #include "xenbus_comms.h"
31 #include "xenbus_probe.h"
32 
33 
34 static struct workqueue_struct *xenbus_frontend_wq;
35 
36 /* device/<type>/<id> => <type>-<id> */
37 static int frontend_bus_id(char bus_id[XEN_BUS_ID_SIZE], const char *nodename)
38 {
39 	nodename = strchr(nodename, '/');
40 	if (!nodename || strlen(nodename + 1) >= XEN_BUS_ID_SIZE) {
41 		pr_warn("bad frontend %s\n", nodename);
42 		return -EINVAL;
43 	}
44 
45 	strlcpy(bus_id, nodename + 1, XEN_BUS_ID_SIZE);
46 	if (!strchr(bus_id, '/')) {
47 		pr_warn("bus_id %s no slash\n", bus_id);
48 		return -EINVAL;
49 	}
50 	*strchr(bus_id, '/') = '-';
51 	return 0;
52 }
53 
54 /* device/<typename>/<name> */
55 static int xenbus_probe_frontend(struct xen_bus_type *bus, const char *type,
56 				 const char *name)
57 {
58 	char *nodename;
59 	int err;
60 
61 	/* ignore console/0 */
62 	if (!strncmp(type, "console", 7) && !strncmp(name, "0", 1)) {
63 		DPRINTK("Ignoring buggy device entry console/0");
64 		return 0;
65 	}
66 
67 	nodename = kasprintf(GFP_KERNEL, "%s/%s/%s", bus->root, type, name);
68 	if (!nodename)
69 		return -ENOMEM;
70 
71 	DPRINTK("%s", nodename);
72 
73 	err = xenbus_probe_node(bus, type, nodename);
74 	kfree(nodename);
75 	return err;
76 }
77 
78 static int xenbus_uevent_frontend(struct device *_dev,
79 				  struct kobj_uevent_env *env)
80 {
81 	struct xenbus_device *dev = to_xenbus_device(_dev);
82 
83 	if (add_uevent_var(env, "MODALIAS=xen:%s", dev->devicetype))
84 		return -ENOMEM;
85 
86 	return 0;
87 }
88 
89 
90 static void backend_changed(struct xenbus_watch *watch,
91 			    const char **vec, unsigned int len)
92 {
93 	xenbus_otherend_changed(watch, vec, len, 1);
94 }
95 
96 static void xenbus_frontend_delayed_resume(struct work_struct *w)
97 {
98 	struct xenbus_device *xdev = container_of(w, struct xenbus_device, work);
99 
100 	xenbus_dev_resume(&xdev->dev);
101 }
102 
103 static int xenbus_frontend_dev_resume(struct device *dev)
104 {
105 	/*
106 	 * If xenstored is running in this domain, we cannot access the backend
107 	 * state at the moment, so we need to defer xenbus_dev_resume
108 	 */
109 	if (xen_store_domain_type == XS_LOCAL) {
110 		struct xenbus_device *xdev = to_xenbus_device(dev);
111 
112 		if (!xenbus_frontend_wq) {
113 			pr_err("%s: no workqueue to process delayed resume\n",
114 			       xdev->nodename);
115 			return -EFAULT;
116 		}
117 
118 		queue_work(xenbus_frontend_wq, &xdev->work);
119 
120 		return 0;
121 	}
122 
123 	return xenbus_dev_resume(dev);
124 }
125 
126 static int xenbus_frontend_dev_probe(struct device *dev)
127 {
128 	if (xen_store_domain_type == XS_LOCAL) {
129 		struct xenbus_device *xdev = to_xenbus_device(dev);
130 		INIT_WORK(&xdev->work, xenbus_frontend_delayed_resume);
131 	}
132 
133 	return xenbus_dev_probe(dev);
134 }
135 
136 static const struct dev_pm_ops xenbus_pm_ops = {
137 	.suspend	= xenbus_dev_suspend,
138 	.resume		= xenbus_frontend_dev_resume,
139 	.freeze		= xenbus_dev_suspend,
140 	.thaw		= xenbus_dev_cancel,
141 	.restore	= xenbus_dev_resume,
142 };
143 
144 static struct xen_bus_type xenbus_frontend = {
145 	.root = "device",
146 	.levels = 2,		/* device/type/<id> */
147 	.get_bus_id = frontend_bus_id,
148 	.probe = xenbus_probe_frontend,
149 	.otherend_changed = backend_changed,
150 	.bus = {
151 		.name		= "xen",
152 		.match		= xenbus_match,
153 		.uevent		= xenbus_uevent_frontend,
154 		.probe		= xenbus_frontend_dev_probe,
155 		.remove		= xenbus_dev_remove,
156 		.shutdown	= xenbus_dev_shutdown,
157 		.dev_attrs	= xenbus_dev_attrs,
158 
159 		.pm		= &xenbus_pm_ops,
160 	},
161 };
162 
163 static void frontend_changed(struct xenbus_watch *watch,
164 			     const char **vec, unsigned int len)
165 {
166 	DPRINTK("");
167 
168 	xenbus_dev_changed(vec[XS_WATCH_PATH], &xenbus_frontend);
169 }
170 
171 
172 /* We watch for devices appearing and vanishing. */
173 static struct xenbus_watch fe_watch = {
174 	.node = "device",
175 	.callback = frontend_changed,
176 };
177 
178 static int read_backend_details(struct xenbus_device *xendev)
179 {
180 	return xenbus_read_otherend_details(xendev, "backend-id", "backend");
181 }
182 
183 static int is_device_connecting(struct device *dev, void *data, bool ignore_nonessential)
184 {
185 	struct xenbus_device *xendev = to_xenbus_device(dev);
186 	struct device_driver *drv = data;
187 	struct xenbus_driver *xendrv;
188 
189 	/*
190 	 * A device with no driver will never connect. We care only about
191 	 * devices which should currently be in the process of connecting.
192 	 */
193 	if (!dev->driver)
194 		return 0;
195 
196 	/* Is this search limited to a particular driver? */
197 	if (drv && (dev->driver != drv))
198 		return 0;
199 
200 	if (ignore_nonessential) {
201 		/* With older QEMU, for PVonHVM guests the guest config files
202 		 * could contain: vfb = [ 'vnc=1, vnclisten=0.0.0.0']
203 		 * which is nonsensical as there is no PV FB (there can be
204 		 * a PVKB) running as HVM guest. */
205 
206 		if ((strncmp(xendev->nodename, "device/vkbd", 11) == 0))
207 			return 0;
208 
209 		if ((strncmp(xendev->nodename, "device/vfb", 10) == 0))
210 			return 0;
211 	}
212 	xendrv = to_xenbus_driver(dev->driver);
213 	return (xendev->state < XenbusStateConnected ||
214 		(xendev->state == XenbusStateConnected &&
215 		 xendrv->is_ready && !xendrv->is_ready(xendev)));
216 }
217 static int essential_device_connecting(struct device *dev, void *data)
218 {
219 	return is_device_connecting(dev, data, true /* ignore PV[KBB+FB] */);
220 }
221 static int non_essential_device_connecting(struct device *dev, void *data)
222 {
223 	return is_device_connecting(dev, data, false);
224 }
225 
226 static int exists_essential_connecting_device(struct device_driver *drv)
227 {
228 	return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv,
229 				essential_device_connecting);
230 }
231 static int exists_non_essential_connecting_device(struct device_driver *drv)
232 {
233 	return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv,
234 				non_essential_device_connecting);
235 }
236 
237 static int print_device_status(struct device *dev, void *data)
238 {
239 	struct xenbus_device *xendev = to_xenbus_device(dev);
240 	struct device_driver *drv = data;
241 
242 	/* Is this operation limited to a particular driver? */
243 	if (drv && (dev->driver != drv))
244 		return 0;
245 
246 	if (!dev->driver) {
247 		/* Information only: is this too noisy? */
248 		pr_info("Device with no driver: %s\n", xendev->nodename);
249 	} else if (xendev->state < XenbusStateConnected) {
250 		enum xenbus_state rstate = XenbusStateUnknown;
251 		if (xendev->otherend)
252 			rstate = xenbus_read_driver_state(xendev->otherend);
253 		pr_warn("Timeout connecting to device: %s (local state %d, remote state %d)\n",
254 			xendev->nodename, xendev->state, rstate);
255 	}
256 
257 	return 0;
258 }
259 
260 /* We only wait for device setup after most initcalls have run. */
261 static int ready_to_wait_for_devices;
262 
263 static bool wait_loop(unsigned long start, unsigned int max_delay,
264 		     unsigned int *seconds_waited)
265 {
266 	if (time_after(jiffies, start + (*seconds_waited+5)*HZ)) {
267 		if (!*seconds_waited)
268 			pr_warn("Waiting for devices to initialise: ");
269 		*seconds_waited += 5;
270 		pr_cont("%us...", max_delay - *seconds_waited);
271 		if (*seconds_waited == max_delay) {
272 			pr_cont("\n");
273 			return true;
274 		}
275 	}
276 
277 	schedule_timeout_interruptible(HZ/10);
278 
279 	return false;
280 }
281 /*
282  * On a 5-minute timeout, wait for all devices currently configured.  We need
283  * to do this to guarantee that the filesystems and / or network devices
284  * needed for boot are available, before we can allow the boot to proceed.
285  *
286  * This needs to be on a late_initcall, to happen after the frontend device
287  * drivers have been initialised, but before the root fs is mounted.
288  *
289  * A possible improvement here would be to have the tools add a per-device
290  * flag to the store entry, indicating whether it is needed at boot time.
291  * This would allow people who knew what they were doing to accelerate their
292  * boot slightly, but of course needs tools or manual intervention to set up
293  * those flags correctly.
294  */
295 static void wait_for_devices(struct xenbus_driver *xendrv)
296 {
297 	unsigned long start = jiffies;
298 	struct device_driver *drv = xendrv ? &xendrv->driver : NULL;
299 	unsigned int seconds_waited = 0;
300 
301 	if (!ready_to_wait_for_devices || !xen_domain())
302 		return;
303 
304 	while (exists_non_essential_connecting_device(drv))
305 		if (wait_loop(start, 30, &seconds_waited))
306 			break;
307 
308 	/* Skips PVKB and PVFB check.*/
309 	while (exists_essential_connecting_device(drv))
310 		if (wait_loop(start, 270, &seconds_waited))
311 			break;
312 
313 	if (seconds_waited)
314 		printk("\n");
315 
316 	bus_for_each_dev(&xenbus_frontend.bus, NULL, drv,
317 			 print_device_status);
318 }
319 
320 int xenbus_register_frontend(struct xenbus_driver *drv)
321 {
322 	int ret;
323 
324 	drv->read_otherend_details = read_backend_details;
325 
326 	ret = xenbus_register_driver_common(drv, &xenbus_frontend);
327 	if (ret)
328 		return ret;
329 
330 	/* If this driver is loaded as a module wait for devices to attach. */
331 	wait_for_devices(drv);
332 
333 	return 0;
334 }
335 EXPORT_SYMBOL_GPL(xenbus_register_frontend);
336 
337 static DECLARE_WAIT_QUEUE_HEAD(backend_state_wq);
338 static int backend_state;
339 
340 static void xenbus_reset_backend_state_changed(struct xenbus_watch *w,
341 					const char **v, unsigned int l)
342 {
343 	xenbus_scanf(XBT_NIL, v[XS_WATCH_PATH], "", "%i", &backend_state);
344 	printk(KERN_DEBUG "XENBUS: backend %s %s\n",
345 			v[XS_WATCH_PATH], xenbus_strstate(backend_state));
346 	wake_up(&backend_state_wq);
347 }
348 
349 static void xenbus_reset_wait_for_backend(char *be, int expected)
350 {
351 	long timeout;
352 	timeout = wait_event_interruptible_timeout(backend_state_wq,
353 			backend_state == expected, 5 * HZ);
354 	if (timeout <= 0)
355 		pr_info("backend %s timed out\n", be);
356 }
357 
358 /*
359  * Reset frontend if it is in Connected or Closed state.
360  * Wait for backend to catch up.
361  * State Connected happens during kdump, Closed after kexec.
362  */
363 static void xenbus_reset_frontend(char *fe, char *be, int be_state)
364 {
365 	struct xenbus_watch be_watch;
366 
367 	printk(KERN_DEBUG "XENBUS: backend %s %s\n",
368 			be, xenbus_strstate(be_state));
369 
370 	memset(&be_watch, 0, sizeof(be_watch));
371 	be_watch.node = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/state", be);
372 	if (!be_watch.node)
373 		return;
374 
375 	be_watch.callback = xenbus_reset_backend_state_changed;
376 	backend_state = XenbusStateUnknown;
377 
378 	pr_info("triggering reconnect on %s\n", be);
379 	register_xenbus_watch(&be_watch);
380 
381 	/* fall through to forward backend to state XenbusStateInitialising */
382 	switch (be_state) {
383 	case XenbusStateConnected:
384 		xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosing);
385 		xenbus_reset_wait_for_backend(be, XenbusStateClosing);
386 
387 	case XenbusStateClosing:
388 		xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosed);
389 		xenbus_reset_wait_for_backend(be, XenbusStateClosed);
390 
391 	case XenbusStateClosed:
392 		xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateInitialising);
393 		xenbus_reset_wait_for_backend(be, XenbusStateInitWait);
394 	}
395 
396 	unregister_xenbus_watch(&be_watch);
397 	pr_info("reconnect done on %s\n", be);
398 	kfree(be_watch.node);
399 }
400 
401 static void xenbus_check_frontend(char *class, char *dev)
402 {
403 	int be_state, fe_state, err;
404 	char *backend, *frontend;
405 
406 	frontend = kasprintf(GFP_NOIO | __GFP_HIGH, "device/%s/%s", class, dev);
407 	if (!frontend)
408 		return;
409 
410 	err = xenbus_scanf(XBT_NIL, frontend, "state", "%i", &fe_state);
411 	if (err != 1)
412 		goto out;
413 
414 	switch (fe_state) {
415 	case XenbusStateConnected:
416 	case XenbusStateClosed:
417 		printk(KERN_DEBUG "XENBUS: frontend %s %s\n",
418 				frontend, xenbus_strstate(fe_state));
419 		backend = xenbus_read(XBT_NIL, frontend, "backend", NULL);
420 		if (!backend || IS_ERR(backend))
421 			goto out;
422 		err = xenbus_scanf(XBT_NIL, backend, "state", "%i", &be_state);
423 		if (err == 1)
424 			xenbus_reset_frontend(frontend, backend, be_state);
425 		kfree(backend);
426 		break;
427 	default:
428 		break;
429 	}
430 out:
431 	kfree(frontend);
432 }
433 
434 static void xenbus_reset_state(void)
435 {
436 	char **devclass, **dev;
437 	int devclass_n, dev_n;
438 	int i, j;
439 
440 	devclass = xenbus_directory(XBT_NIL, "device", "", &devclass_n);
441 	if (IS_ERR(devclass))
442 		return;
443 
444 	for (i = 0; i < devclass_n; i++) {
445 		dev = xenbus_directory(XBT_NIL, "device", devclass[i], &dev_n);
446 		if (IS_ERR(dev))
447 			continue;
448 		for (j = 0; j < dev_n; j++)
449 			xenbus_check_frontend(devclass[i], dev[j]);
450 		kfree(dev);
451 	}
452 	kfree(devclass);
453 }
454 
455 static int frontend_probe_and_watch(struct notifier_block *notifier,
456 				   unsigned long event,
457 				   void *data)
458 {
459 	/* reset devices in Connected or Closed state */
460 	if (xen_hvm_domain())
461 		xenbus_reset_state();
462 	/* Enumerate devices in xenstore and watch for changes. */
463 	xenbus_probe_devices(&xenbus_frontend);
464 	register_xenbus_watch(&fe_watch);
465 
466 	return NOTIFY_DONE;
467 }
468 
469 
470 static int __init xenbus_probe_frontend_init(void)
471 {
472 	static struct notifier_block xenstore_notifier = {
473 		.notifier_call = frontend_probe_and_watch
474 	};
475 	int err;
476 
477 	DPRINTK("");
478 
479 	/* Register ourselves with the kernel bus subsystem */
480 	err = bus_register(&xenbus_frontend.bus);
481 	if (err)
482 		return err;
483 
484 	register_xenstore_notifier(&xenstore_notifier);
485 
486 	if (xen_store_domain_type == XS_LOCAL) {
487 		xenbus_frontend_wq = create_workqueue("xenbus_frontend");
488 		if (!xenbus_frontend_wq)
489 			pr_warn("create xenbus frontend workqueue failed, S3 resume is likely to fail\n");
490 	}
491 
492 	return 0;
493 }
494 subsys_initcall(xenbus_probe_frontend_init);
495 
496 #ifndef MODULE
497 static int __init boot_wait_for_devices(void)
498 {
499 	if (xen_hvm_domain() && !xen_platform_pci_unplug)
500 		return -ENODEV;
501 
502 	ready_to_wait_for_devices = 1;
503 	wait_for_devices(NULL);
504 	return 0;
505 }
506 
507 late_initcall(boot_wait_for_devices);
508 #endif
509 
510 MODULE_LICENSE("GPL");
511