xref: /linux/net/9p/trans_xen.c (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
1 /*
2  * linux/fs/9p/trans_xen
3  *
4  * Xen transport layer.
5  *
6  * Copyright (C) 2017 by Stefano Stabellini <stefano@aporeto.com>
7  *
8  * This program is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU General Public License version 2
10  * as published by the Free Software Foundation; or, when distributed
11  * separately from the Linux kernel or incorporated into other
12  * software packages, subject to the following license:
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this source file (the "Software"), to deal in the Software without
16  * restriction, including without limitation the rights to use, copy, modify,
17  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
18  * and to permit persons to whom the Software is furnished to do so, subject to
19  * the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
27  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
29  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
30  * IN THE SOFTWARE.
31  */
32 
33 #include <xen/events.h>
34 #include <xen/grant_table.h>
35 #include <xen/xen.h>
36 #include <xen/xenbus.h>
37 #include <xen/interface/io/9pfs.h>
38 
39 #include <linux/module.h>
40 #include <linux/spinlock.h>
41 #include <net/9p/9p.h>
42 #include <net/9p/client.h>
43 #include <net/9p/transport.h>
44 
45 #define XEN_9PFS_NUM_RINGS 2
46 #define XEN_9PFS_RING_ORDER 6
47 #define XEN_9PFS_RING_SIZE  XEN_FLEX_RING_SIZE(XEN_9PFS_RING_ORDER)
48 
49 struct xen_9pfs_header {
50 	uint32_t size;
51 	uint8_t id;
52 	uint16_t tag;
53 
54 	/* uint8_t sdata[]; */
55 } __attribute__((packed));
56 
57 /* One per ring, more than one per 9pfs share */
58 struct xen_9pfs_dataring {
59 	struct xen_9pfs_front_priv *priv;
60 
61 	struct xen_9pfs_data_intf *intf;
62 	grant_ref_t ref;
63 	int evtchn;
64 	int irq;
65 	/* protect a ring from concurrent accesses */
66 	spinlock_t lock;
67 
68 	struct xen_9pfs_data data;
69 	wait_queue_head_t wq;
70 	struct work_struct work;
71 };
72 
73 /* One per 9pfs share */
74 struct xen_9pfs_front_priv {
75 	struct list_head list;
76 	struct xenbus_device *dev;
77 	char *tag;
78 	struct p9_client *client;
79 
80 	int num_rings;
81 	struct xen_9pfs_dataring *rings;
82 };
83 
84 static LIST_HEAD(xen_9pfs_devs);
85 static DEFINE_RWLOCK(xen_9pfs_lock);
86 
87 /* We don't currently allow canceling of requests */
88 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
89 {
90 	return 1;
91 }
92 
93 static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
94 {
95 	struct xen_9pfs_front_priv *priv;
96 
97 	read_lock(&xen_9pfs_lock);
98 	list_for_each_entry(priv, &xen_9pfs_devs, list) {
99 		if (!strcmp(priv->tag, addr)) {
100 			priv->client = client;
101 			read_unlock(&xen_9pfs_lock);
102 			return 0;
103 		}
104 	}
105 	read_unlock(&xen_9pfs_lock);
106 	return -EINVAL;
107 }
108 
109 static void p9_xen_close(struct p9_client *client)
110 {
111 	struct xen_9pfs_front_priv *priv;
112 
113 	read_lock(&xen_9pfs_lock);
114 	list_for_each_entry(priv, &xen_9pfs_devs, list) {
115 		if (priv->client == client) {
116 			priv->client = NULL;
117 			read_unlock(&xen_9pfs_lock);
118 			return;
119 		}
120 	}
121 	read_unlock(&xen_9pfs_lock);
122 }
123 
124 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
125 {
126 	RING_IDX cons, prod;
127 
128 	cons = ring->intf->out_cons;
129 	prod = ring->intf->out_prod;
130 	virt_mb();
131 
132 	return XEN_9PFS_RING_SIZE -
133 		xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) >= size;
134 }
135 
136 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
137 {
138 	struct xen_9pfs_front_priv *priv = NULL;
139 	RING_IDX cons, prod, masked_cons, masked_prod;
140 	unsigned long flags;
141 	u32 size = p9_req->tc->size;
142 	struct xen_9pfs_dataring *ring;
143 	int num;
144 
145 	read_lock(&xen_9pfs_lock);
146 	list_for_each_entry(priv, &xen_9pfs_devs, list) {
147 		if (priv->client == client)
148 			break;
149 	}
150 	read_unlock(&xen_9pfs_lock);
151 	if (!priv || priv->client != client)
152 		return -EINVAL;
153 
154 	num = p9_req->tc->tag % priv->num_rings;
155 	ring = &priv->rings[num];
156 
157 again:
158 	while (wait_event_killable(ring->wq,
159 				   p9_xen_write_todo(ring, size)) != 0)
160 		;
161 
162 	spin_lock_irqsave(&ring->lock, flags);
163 	cons = ring->intf->out_cons;
164 	prod = ring->intf->out_prod;
165 	virt_mb();
166 
167 	if (XEN_9PFS_RING_SIZE - xen_9pfs_queued(prod, cons,
168 						 XEN_9PFS_RING_SIZE) < size) {
169 		spin_unlock_irqrestore(&ring->lock, flags);
170 		goto again;
171 	}
172 
173 	masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
174 	masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
175 
176 	xen_9pfs_write_packet(ring->data.out, p9_req->tc->sdata, size,
177 			      &masked_prod, masked_cons, XEN_9PFS_RING_SIZE);
178 
179 	p9_req->status = REQ_STATUS_SENT;
180 	virt_wmb();			/* write ring before updating pointer */
181 	prod += size;
182 	ring->intf->out_prod = prod;
183 	spin_unlock_irqrestore(&ring->lock, flags);
184 	notify_remote_via_irq(ring->irq);
185 
186 	return 0;
187 }
188 
189 static void p9_xen_response(struct work_struct *work)
190 {
191 	struct xen_9pfs_front_priv *priv;
192 	struct xen_9pfs_dataring *ring;
193 	RING_IDX cons, prod, masked_cons, masked_prod;
194 	struct xen_9pfs_header h;
195 	struct p9_req_t *req;
196 	int status;
197 
198 	ring = container_of(work, struct xen_9pfs_dataring, work);
199 	priv = ring->priv;
200 
201 	while (1) {
202 		cons = ring->intf->in_cons;
203 		prod = ring->intf->in_prod;
204 		virt_rmb();
205 
206 		if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) <
207 		    sizeof(h)) {
208 			notify_remote_via_irq(ring->irq);
209 			return;
210 		}
211 
212 		masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
213 		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
214 
215 		/* First, read just the header */
216 		xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
217 				     masked_prod, &masked_cons,
218 				     XEN_9PFS_RING_SIZE);
219 
220 		req = p9_tag_lookup(priv->client, h.tag);
221 		if (!req || req->status != REQ_STATUS_SENT) {
222 			dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
223 			cons += h.size;
224 			virt_mb();
225 			ring->intf->in_cons = cons;
226 			continue;
227 		}
228 
229 		memcpy(req->rc, &h, sizeof(h));
230 		req->rc->offset = 0;
231 
232 		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
233 		/* Then, read the whole packet (including the header) */
234 		xen_9pfs_read_packet(req->rc->sdata, ring->data.in, h.size,
235 				     masked_prod, &masked_cons,
236 				     XEN_9PFS_RING_SIZE);
237 
238 		virt_mb();
239 		cons += h.size;
240 		ring->intf->in_cons = cons;
241 
242 		status = (req->status != REQ_STATUS_ERROR) ?
243 			REQ_STATUS_RCVD : REQ_STATUS_ERROR;
244 
245 		p9_client_cb(priv->client, req, status);
246 	}
247 }
248 
249 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
250 {
251 	struct xen_9pfs_dataring *ring = r;
252 
253 	if (!ring || !ring->priv->client) {
254 		/* ignore spurious interrupt */
255 		return IRQ_HANDLED;
256 	}
257 
258 	wake_up_interruptible(&ring->wq);
259 	schedule_work(&ring->work);
260 
261 	return IRQ_HANDLED;
262 }
263 
264 static struct p9_trans_module p9_xen_trans = {
265 	.name = "xen",
266 	.maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
267 	.def = 1,
268 	.create = p9_xen_create,
269 	.close = p9_xen_close,
270 	.request = p9_xen_request,
271 	.cancel = p9_xen_cancel,
272 	.owner = THIS_MODULE,
273 };
274 
275 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
276 	{ "9pfs" },
277 	{ "" }
278 };
279 
280 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
281 {
282 	int i, j;
283 
284 	write_lock(&xen_9pfs_lock);
285 	list_del(&priv->list);
286 	write_unlock(&xen_9pfs_lock);
287 
288 	for (i = 0; i < priv->num_rings; i++) {
289 		if (!priv->rings[i].intf)
290 			break;
291 		if (priv->rings[i].irq > 0)
292 			unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
293 		if (priv->rings[i].data.in) {
294 			for (j = 0; j < (1 << XEN_9PFS_RING_ORDER); j++) {
295 				grant_ref_t ref;
296 
297 				ref = priv->rings[i].intf->ref[j];
298 				gnttab_end_foreign_access(ref, 0, 0);
299 			}
300 			free_pages((unsigned long)priv->rings[i].data.in,
301 				   XEN_9PFS_RING_ORDER -
302 				   (PAGE_SHIFT - XEN_PAGE_SHIFT));
303 		}
304 		gnttab_end_foreign_access(priv->rings[i].ref, 0, 0);
305 		free_page((unsigned long)priv->rings[i].intf);
306 	}
307 	kfree(priv->rings);
308 	kfree(priv->tag);
309 	kfree(priv);
310 }
311 
312 static int xen_9pfs_front_remove(struct xenbus_device *dev)
313 {
314 	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
315 
316 	dev_set_drvdata(&dev->dev, NULL);
317 	xen_9pfs_front_free(priv);
318 	return 0;
319 }
320 
321 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
322 					 struct xen_9pfs_dataring *ring)
323 {
324 	int i = 0;
325 	int ret = -ENOMEM;
326 	void *bytes = NULL;
327 
328 	init_waitqueue_head(&ring->wq);
329 	spin_lock_init(&ring->lock);
330 	INIT_WORK(&ring->work, p9_xen_response);
331 
332 	ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
333 	if (!ring->intf)
334 		return ret;
335 	ret = gnttab_grant_foreign_access(dev->otherend_id,
336 					  virt_to_gfn(ring->intf), 0);
337 	if (ret < 0)
338 		goto out;
339 	ring->ref = ret;
340 	bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
341 			XEN_9PFS_RING_ORDER - (PAGE_SHIFT - XEN_PAGE_SHIFT));
342 	if (!bytes) {
343 		ret = -ENOMEM;
344 		goto out;
345 	}
346 	for (; i < (1 << XEN_9PFS_RING_ORDER); i++) {
347 		ret = gnttab_grant_foreign_access(
348 				dev->otherend_id, virt_to_gfn(bytes) + i, 0);
349 		if (ret < 0)
350 			goto out;
351 		ring->intf->ref[i] = ret;
352 	}
353 	ring->intf->ring_order = XEN_9PFS_RING_ORDER;
354 	ring->data.in = bytes;
355 	ring->data.out = bytes + XEN_9PFS_RING_SIZE;
356 
357 	ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
358 	if (ret)
359 		goto out;
360 	ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
361 					      xen_9pfs_front_event_handler,
362 					      0, "xen_9pfs-frontend", ring);
363 	if (ring->irq >= 0)
364 		return 0;
365 
366 	xenbus_free_evtchn(dev, ring->evtchn);
367 	ret = ring->irq;
368 out:
369 	if (bytes) {
370 		for (i--; i >= 0; i--)
371 			gnttab_end_foreign_access(ring->intf->ref[i], 0, 0);
372 		free_pages((unsigned long)bytes,
373 			   XEN_9PFS_RING_ORDER -
374 			   (PAGE_SHIFT - XEN_PAGE_SHIFT));
375 	}
376 	gnttab_end_foreign_access(ring->ref, 0, 0);
377 	free_page((unsigned long)ring->intf);
378 	return ret;
379 }
380 
381 static int xen_9pfs_front_probe(struct xenbus_device *dev,
382 				const struct xenbus_device_id *id)
383 {
384 	int ret, i;
385 	struct xenbus_transaction xbt;
386 	struct xen_9pfs_front_priv *priv = NULL;
387 	char *versions;
388 	unsigned int max_rings, max_ring_order, len = 0;
389 
390 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
391 	if (!len)
392 		return -EINVAL;
393 	if (strcmp(versions, "1")) {
394 		kfree(versions);
395 		return -EINVAL;
396 	}
397 	kfree(versions);
398 	max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
399 	if (max_rings < XEN_9PFS_NUM_RINGS)
400 		return -EINVAL;
401 	max_ring_order = xenbus_read_unsigned(dev->otherend,
402 					      "max-ring-page-order", 0);
403 	if (max_ring_order < XEN_9PFS_RING_ORDER)
404 		return -EINVAL;
405 
406 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
407 	if (!priv)
408 		return -ENOMEM;
409 
410 	priv->dev = dev;
411 	priv->num_rings = XEN_9PFS_NUM_RINGS;
412 	priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
413 			      GFP_KERNEL);
414 	if (!priv->rings) {
415 		kfree(priv);
416 		return -ENOMEM;
417 	}
418 
419 	for (i = 0; i < priv->num_rings; i++) {
420 		priv->rings[i].priv = priv;
421 		ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i]);
422 		if (ret < 0)
423 			goto error;
424 	}
425 
426  again:
427 	ret = xenbus_transaction_start(&xbt);
428 	if (ret) {
429 		xenbus_dev_fatal(dev, ret, "starting transaction");
430 		goto error;
431 	}
432 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
433 	if (ret)
434 		goto error_xenbus;
435 	ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
436 			    priv->num_rings);
437 	if (ret)
438 		goto error_xenbus;
439 	for (i = 0; i < priv->num_rings; i++) {
440 		char str[16];
441 
442 		BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
443 		sprintf(str, "ring-ref%u", i);
444 		ret = xenbus_printf(xbt, dev->nodename, str, "%d",
445 				    priv->rings[i].ref);
446 		if (ret)
447 			goto error_xenbus;
448 
449 		sprintf(str, "event-channel-%u", i);
450 		ret = xenbus_printf(xbt, dev->nodename, str, "%u",
451 				    priv->rings[i].evtchn);
452 		if (ret)
453 			goto error_xenbus;
454 	}
455 	priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
456 	if (IS_ERR(priv->tag)) {
457 		ret = PTR_ERR(priv->tag);
458 		goto error_xenbus;
459 	}
460 	ret = xenbus_transaction_end(xbt, 0);
461 	if (ret) {
462 		if (ret == -EAGAIN)
463 			goto again;
464 		xenbus_dev_fatal(dev, ret, "completing transaction");
465 		goto error;
466 	}
467 
468 	write_lock(&xen_9pfs_lock);
469 	list_add_tail(&priv->list, &xen_9pfs_devs);
470 	write_unlock(&xen_9pfs_lock);
471 	dev_set_drvdata(&dev->dev, priv);
472 	xenbus_switch_state(dev, XenbusStateInitialised);
473 
474 	return 0;
475 
476  error_xenbus:
477 	xenbus_transaction_end(xbt, 1);
478 	xenbus_dev_fatal(dev, ret, "writing xenstore");
479  error:
480 	dev_set_drvdata(&dev->dev, NULL);
481 	xen_9pfs_front_free(priv);
482 	return ret;
483 }
484 
485 static int xen_9pfs_front_resume(struct xenbus_device *dev)
486 {
487 	dev_warn(&dev->dev, "suspend/resume unsupported\n");
488 	return 0;
489 }
490 
491 static void xen_9pfs_front_changed(struct xenbus_device *dev,
492 				   enum xenbus_state backend_state)
493 {
494 	switch (backend_state) {
495 	case XenbusStateReconfiguring:
496 	case XenbusStateReconfigured:
497 	case XenbusStateInitialising:
498 	case XenbusStateInitialised:
499 	case XenbusStateUnknown:
500 		break;
501 
502 	case XenbusStateInitWait:
503 		break;
504 
505 	case XenbusStateConnected:
506 		xenbus_switch_state(dev, XenbusStateConnected);
507 		break;
508 
509 	case XenbusStateClosed:
510 		if (dev->state == XenbusStateClosed)
511 			break;
512 		/* Missed the backend's CLOSING state -- fallthrough */
513 	case XenbusStateClosing:
514 		xenbus_frontend_closed(dev);
515 		break;
516 	}
517 }
518 
519 static struct xenbus_driver xen_9pfs_front_driver = {
520 	.ids = xen_9pfs_front_ids,
521 	.probe = xen_9pfs_front_probe,
522 	.remove = xen_9pfs_front_remove,
523 	.resume = xen_9pfs_front_resume,
524 	.otherend_changed = xen_9pfs_front_changed,
525 };
526 
527 static int p9_trans_xen_init(void)
528 {
529 	if (!xen_domain())
530 		return -ENODEV;
531 
532 	pr_info("Initialising Xen transport for 9pfs\n");
533 
534 	v9fs_register_trans(&p9_xen_trans);
535 	return xenbus_register_frontend(&xen_9pfs_front_driver);
536 }
537 module_init(p9_trans_xen_init);
538 
539 static void p9_trans_xen_exit(void)
540 {
541 	v9fs_unregister_trans(&p9_xen_trans);
542 	return xenbus_unregister_driver(&xen_9pfs_front_driver);
543 }
544 module_exit(p9_trans_xen_exit);
545 
546 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
547 MODULE_DESCRIPTION("Xen Transport for 9P");
548 MODULE_LICENSE("GPL");
549