xref: /linux/net/9p/trans_xen.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/9p/trans_xen
4  *
5  * Xen transport layer.
6  *
7  * Copyright (C) 2017 by Stefano Stabellini <stefano@aporeto.com>
8  */
9 
10 #include <xen/events.h>
11 #include <xen/grant_table.h>
12 #include <xen/xen.h>
13 #include <xen/xenbus.h>
14 #include <xen/interface/io/9pfs.h>
15 
16 #include <linux/module.h>
17 #include <linux/spinlock.h>
18 #include <net/9p/9p.h>
19 #include <net/9p/client.h>
20 #include <net/9p/transport.h>
21 
22 #define XEN_9PFS_NUM_RINGS 2
23 #define XEN_9PFS_RING_ORDER 9
24 #define XEN_9PFS_RING_SIZE(ring)  XEN_FLEX_RING_SIZE(ring->intf->ring_order)
25 
26 struct xen_9pfs_header {
27 	uint32_t size;
28 	uint8_t id;
29 	uint16_t tag;
30 
31 	/* uint8_t sdata[]; */
32 } __attribute__((packed));
33 
34 /* One per ring, more than one per 9pfs share */
35 struct xen_9pfs_dataring {
36 	struct xen_9pfs_front_priv *priv;
37 
38 	struct xen_9pfs_data_intf *intf;
39 	grant_ref_t ref;
40 	int evtchn;
41 	int irq;
42 	/* protect a ring from concurrent accesses */
43 	spinlock_t lock;
44 
45 	struct xen_9pfs_data data;
46 	wait_queue_head_t wq;
47 	struct work_struct work;
48 };
49 
50 /* One per 9pfs share */
51 struct xen_9pfs_front_priv {
52 	struct list_head list;
53 	struct xenbus_device *dev;
54 	char *tag;
55 	struct p9_client *client;
56 
57 	int num_rings;
58 	struct xen_9pfs_dataring *rings;
59 };
60 
61 static LIST_HEAD(xen_9pfs_devs);
62 static DEFINE_RWLOCK(xen_9pfs_lock);
63 
64 /* We don't currently allow canceling of requests */
65 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
66 {
67 	return 1;
68 }
69 
70 static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
71 {
72 	struct xen_9pfs_front_priv *priv;
73 
74 	if (addr == NULL)
75 		return -EINVAL;
76 
77 	read_lock(&xen_9pfs_lock);
78 	list_for_each_entry(priv, &xen_9pfs_devs, list) {
79 		if (!strcmp(priv->tag, addr)) {
80 			priv->client = client;
81 			read_unlock(&xen_9pfs_lock);
82 			return 0;
83 		}
84 	}
85 	read_unlock(&xen_9pfs_lock);
86 	return -EINVAL;
87 }
88 
89 static void p9_xen_close(struct p9_client *client)
90 {
91 	struct xen_9pfs_front_priv *priv;
92 
93 	read_lock(&xen_9pfs_lock);
94 	list_for_each_entry(priv, &xen_9pfs_devs, list) {
95 		if (priv->client == client) {
96 			priv->client = NULL;
97 			read_unlock(&xen_9pfs_lock);
98 			return;
99 		}
100 	}
101 	read_unlock(&xen_9pfs_lock);
102 }
103 
104 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
105 {
106 	RING_IDX cons, prod;
107 
108 	cons = ring->intf->out_cons;
109 	prod = ring->intf->out_prod;
110 	virt_mb();
111 
112 	return XEN_9PFS_RING_SIZE(ring) -
113 		xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size;
114 }
115 
116 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
117 {
118 	struct xen_9pfs_front_priv *priv;
119 	RING_IDX cons, prod, masked_cons, masked_prod;
120 	unsigned long flags;
121 	u32 size = p9_req->tc.size;
122 	struct xen_9pfs_dataring *ring;
123 	int num;
124 
125 	read_lock(&xen_9pfs_lock);
126 	list_for_each_entry(priv, &xen_9pfs_devs, list) {
127 		if (priv->client == client)
128 			break;
129 	}
130 	read_unlock(&xen_9pfs_lock);
131 	if (list_entry_is_head(priv, &xen_9pfs_devs, list))
132 		return -EINVAL;
133 
134 	num = p9_req->tc.tag % priv->num_rings;
135 	ring = &priv->rings[num];
136 
137 again:
138 	while (wait_event_killable(ring->wq,
139 				   p9_xen_write_todo(ring, size)) != 0)
140 		;
141 
142 	spin_lock_irqsave(&ring->lock, flags);
143 	cons = ring->intf->out_cons;
144 	prod = ring->intf->out_prod;
145 	virt_mb();
146 
147 	if (XEN_9PFS_RING_SIZE(ring) -
148 	    xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) < size) {
149 		spin_unlock_irqrestore(&ring->lock, flags);
150 		goto again;
151 	}
152 
153 	masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
154 	masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
155 
156 	xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
157 			      &masked_prod, masked_cons,
158 			      XEN_9PFS_RING_SIZE(ring));
159 
160 	p9_req->status = REQ_STATUS_SENT;
161 	virt_wmb();			/* write ring before updating pointer */
162 	prod += size;
163 	ring->intf->out_prod = prod;
164 	spin_unlock_irqrestore(&ring->lock, flags);
165 	notify_remote_via_irq(ring->irq);
166 	p9_req_put(p9_req);
167 
168 	return 0;
169 }
170 
171 static void p9_xen_response(struct work_struct *work)
172 {
173 	struct xen_9pfs_front_priv *priv;
174 	struct xen_9pfs_dataring *ring;
175 	RING_IDX cons, prod, masked_cons, masked_prod;
176 	struct xen_9pfs_header h;
177 	struct p9_req_t *req;
178 	int status;
179 
180 	ring = container_of(work, struct xen_9pfs_dataring, work);
181 	priv = ring->priv;
182 
183 	while (1) {
184 		cons = ring->intf->in_cons;
185 		prod = ring->intf->in_prod;
186 		virt_rmb();
187 
188 		if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) <
189 		    sizeof(h)) {
190 			notify_remote_via_irq(ring->irq);
191 			return;
192 		}
193 
194 		masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
195 		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
196 
197 		/* First, read just the header */
198 		xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
199 				     masked_prod, &masked_cons,
200 				     XEN_9PFS_RING_SIZE(ring));
201 
202 		req = p9_tag_lookup(priv->client, h.tag);
203 		if (!req || req->status != REQ_STATUS_SENT) {
204 			dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
205 			cons += h.size;
206 			virt_mb();
207 			ring->intf->in_cons = cons;
208 			continue;
209 		}
210 
211 		memcpy(&req->rc, &h, sizeof(h));
212 		req->rc.offset = 0;
213 
214 		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
215 		/* Then, read the whole packet (including the header) */
216 		xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
217 				     masked_prod, &masked_cons,
218 				     XEN_9PFS_RING_SIZE(ring));
219 
220 		virt_mb();
221 		cons += h.size;
222 		ring->intf->in_cons = cons;
223 
224 		status = (req->status != REQ_STATUS_ERROR) ?
225 			REQ_STATUS_RCVD : REQ_STATUS_ERROR;
226 
227 		p9_client_cb(priv->client, req, status);
228 	}
229 }
230 
231 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
232 {
233 	struct xen_9pfs_dataring *ring = r;
234 
235 	if (!ring || !ring->priv->client) {
236 		/* ignore spurious interrupt */
237 		return IRQ_HANDLED;
238 	}
239 
240 	wake_up_interruptible(&ring->wq);
241 	schedule_work(&ring->work);
242 
243 	return IRQ_HANDLED;
244 }
245 
246 static struct p9_trans_module p9_xen_trans = {
247 	.name = "xen",
248 	.maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
249 	.def = 1,
250 	.create = p9_xen_create,
251 	.close = p9_xen_close,
252 	.request = p9_xen_request,
253 	.cancel = p9_xen_cancel,
254 	.owner = THIS_MODULE,
255 };
256 
257 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
258 	{ "9pfs" },
259 	{ "" }
260 };
261 
262 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
263 {
264 	int i, j;
265 
266 	write_lock(&xen_9pfs_lock);
267 	list_del(&priv->list);
268 	write_unlock(&xen_9pfs_lock);
269 
270 	for (i = 0; i < priv->num_rings; i++) {
271 		if (!priv->rings[i].intf)
272 			break;
273 		if (priv->rings[i].irq > 0)
274 			unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
275 		if (priv->rings[i].data.in) {
276 			for (j = 0;
277 			     j < (1 << priv->rings[i].intf->ring_order);
278 			     j++) {
279 				grant_ref_t ref;
280 
281 				ref = priv->rings[i].intf->ref[j];
282 				gnttab_end_foreign_access(ref, 0);
283 			}
284 			free_pages_exact(priv->rings[i].data.in,
285 				   1UL << (priv->rings[i].intf->ring_order +
286 					   XEN_PAGE_SHIFT));
287 		}
288 		gnttab_end_foreign_access(priv->rings[i].ref, 0);
289 		free_page((unsigned long)priv->rings[i].intf);
290 	}
291 	kfree(priv->rings);
292 	kfree(priv->tag);
293 	kfree(priv);
294 }
295 
296 static int xen_9pfs_front_remove(struct xenbus_device *dev)
297 {
298 	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
299 
300 	dev_set_drvdata(&dev->dev, NULL);
301 	xen_9pfs_front_free(priv);
302 	return 0;
303 }
304 
305 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
306 					 struct xen_9pfs_dataring *ring,
307 					 unsigned int order)
308 {
309 	int i = 0;
310 	int ret = -ENOMEM;
311 	void *bytes = NULL;
312 
313 	init_waitqueue_head(&ring->wq);
314 	spin_lock_init(&ring->lock);
315 	INIT_WORK(&ring->work, p9_xen_response);
316 
317 	ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
318 	if (!ring->intf)
319 		return ret;
320 	ret = gnttab_grant_foreign_access(dev->otherend_id,
321 					  virt_to_gfn(ring->intf), 0);
322 	if (ret < 0)
323 		goto out;
324 	ring->ref = ret;
325 	bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
326 				  GFP_KERNEL | __GFP_ZERO);
327 	if (!bytes) {
328 		ret = -ENOMEM;
329 		goto out;
330 	}
331 	for (; i < (1 << order); i++) {
332 		ret = gnttab_grant_foreign_access(
333 				dev->otherend_id, virt_to_gfn(bytes) + i, 0);
334 		if (ret < 0)
335 			goto out;
336 		ring->intf->ref[i] = ret;
337 	}
338 	ring->intf->ring_order = order;
339 	ring->data.in = bytes;
340 	ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
341 
342 	ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
343 	if (ret)
344 		goto out;
345 	ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
346 					      xen_9pfs_front_event_handler,
347 					      0, "xen_9pfs-frontend", ring);
348 	if (ring->irq >= 0)
349 		return 0;
350 
351 	xenbus_free_evtchn(dev, ring->evtchn);
352 	ret = ring->irq;
353 out:
354 	if (bytes) {
355 		for (i--; i >= 0; i--)
356 			gnttab_end_foreign_access(ring->intf->ref[i], 0);
357 		free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
358 	}
359 	gnttab_end_foreign_access(ring->ref, 0);
360 	free_page((unsigned long)ring->intf);
361 	return ret;
362 }
363 
364 static int xen_9pfs_front_probe(struct xenbus_device *dev,
365 				const struct xenbus_device_id *id)
366 {
367 	int ret, i;
368 	struct xenbus_transaction xbt;
369 	struct xen_9pfs_front_priv *priv = NULL;
370 	char *versions;
371 	unsigned int max_rings, max_ring_order, len = 0;
372 
373 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
374 	if (IS_ERR(versions))
375 		return PTR_ERR(versions);
376 	if (strcmp(versions, "1")) {
377 		kfree(versions);
378 		return -EINVAL;
379 	}
380 	kfree(versions);
381 	max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
382 	if (max_rings < XEN_9PFS_NUM_RINGS)
383 		return -EINVAL;
384 	max_ring_order = xenbus_read_unsigned(dev->otherend,
385 					      "max-ring-page-order", 0);
386 	if (max_ring_order > XEN_9PFS_RING_ORDER)
387 		max_ring_order = XEN_9PFS_RING_ORDER;
388 	if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
389 		p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
390 
391 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
392 	if (!priv)
393 		return -ENOMEM;
394 
395 	priv->dev = dev;
396 	priv->num_rings = XEN_9PFS_NUM_RINGS;
397 	priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
398 			      GFP_KERNEL);
399 	if (!priv->rings) {
400 		kfree(priv);
401 		return -ENOMEM;
402 	}
403 
404 	for (i = 0; i < priv->num_rings; i++) {
405 		priv->rings[i].priv = priv;
406 		ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
407 						    max_ring_order);
408 		if (ret < 0)
409 			goto error;
410 	}
411 
412  again:
413 	ret = xenbus_transaction_start(&xbt);
414 	if (ret) {
415 		xenbus_dev_fatal(dev, ret, "starting transaction");
416 		goto error;
417 	}
418 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
419 	if (ret)
420 		goto error_xenbus;
421 	ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
422 			    priv->num_rings);
423 	if (ret)
424 		goto error_xenbus;
425 	for (i = 0; i < priv->num_rings; i++) {
426 		char str[16];
427 
428 		BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
429 		sprintf(str, "ring-ref%d", i);
430 		ret = xenbus_printf(xbt, dev->nodename, str, "%d",
431 				    priv->rings[i].ref);
432 		if (ret)
433 			goto error_xenbus;
434 
435 		sprintf(str, "event-channel-%d", i);
436 		ret = xenbus_printf(xbt, dev->nodename, str, "%u",
437 				    priv->rings[i].evtchn);
438 		if (ret)
439 			goto error_xenbus;
440 	}
441 	priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
442 	if (IS_ERR(priv->tag)) {
443 		ret = PTR_ERR(priv->tag);
444 		goto error_xenbus;
445 	}
446 	ret = xenbus_transaction_end(xbt, 0);
447 	if (ret) {
448 		if (ret == -EAGAIN)
449 			goto again;
450 		xenbus_dev_fatal(dev, ret, "completing transaction");
451 		goto error;
452 	}
453 
454 	write_lock(&xen_9pfs_lock);
455 	list_add_tail(&priv->list, &xen_9pfs_devs);
456 	write_unlock(&xen_9pfs_lock);
457 	dev_set_drvdata(&dev->dev, priv);
458 	xenbus_switch_state(dev, XenbusStateInitialised);
459 
460 	return 0;
461 
462  error_xenbus:
463 	xenbus_transaction_end(xbt, 1);
464 	xenbus_dev_fatal(dev, ret, "writing xenstore");
465  error:
466 	dev_set_drvdata(&dev->dev, NULL);
467 	xen_9pfs_front_free(priv);
468 	return ret;
469 }
470 
471 static int xen_9pfs_front_resume(struct xenbus_device *dev)
472 {
473 	dev_warn(&dev->dev, "suspend/resume unsupported\n");
474 	return 0;
475 }
476 
477 static void xen_9pfs_front_changed(struct xenbus_device *dev,
478 				   enum xenbus_state backend_state)
479 {
480 	switch (backend_state) {
481 	case XenbusStateReconfiguring:
482 	case XenbusStateReconfigured:
483 	case XenbusStateInitialising:
484 	case XenbusStateInitialised:
485 	case XenbusStateUnknown:
486 		break;
487 
488 	case XenbusStateInitWait:
489 		break;
490 
491 	case XenbusStateConnected:
492 		xenbus_switch_state(dev, XenbusStateConnected);
493 		break;
494 
495 	case XenbusStateClosed:
496 		if (dev->state == XenbusStateClosed)
497 			break;
498 		fallthrough;	/* Missed the backend's CLOSING state */
499 	case XenbusStateClosing:
500 		xenbus_frontend_closed(dev);
501 		break;
502 	}
503 }
504 
505 static struct xenbus_driver xen_9pfs_front_driver = {
506 	.ids = xen_9pfs_front_ids,
507 	.probe = xen_9pfs_front_probe,
508 	.remove = xen_9pfs_front_remove,
509 	.resume = xen_9pfs_front_resume,
510 	.otherend_changed = xen_9pfs_front_changed,
511 };
512 
513 static int p9_trans_xen_init(void)
514 {
515 	int rc;
516 
517 	if (!xen_domain())
518 		return -ENODEV;
519 
520 	pr_info("Initialising Xen transport for 9pfs\n");
521 
522 	v9fs_register_trans(&p9_xen_trans);
523 	rc = xenbus_register_frontend(&xen_9pfs_front_driver);
524 	if (rc)
525 		v9fs_unregister_trans(&p9_xen_trans);
526 
527 	return rc;
528 }
529 module_init(p9_trans_xen_init);
530 MODULE_ALIAS_9P("xen");
531 
532 static void p9_trans_xen_exit(void)
533 {
534 	v9fs_unregister_trans(&p9_xen_trans);
535 	return xenbus_unregister_driver(&xen_9pfs_front_driver);
536 }
537 module_exit(p9_trans_xen_exit);
538 
539 MODULE_ALIAS("xen:9pfs");
540 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
541 MODULE_DESCRIPTION("Xen Transport for 9P");
542 MODULE_LICENSE("GPL");
543