xref: /openbmc/linux/net/9p/trans_xen.c (revision 1a0aae88)
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 	WRITE_ONCE(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(client, 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 		if (h.size > req->rc.capacity) {
212 			dev_warn(&priv->dev->dev,
213 				 "requested packet size too big: %d for tag %d with capacity %zd\n",
214 				 h.size, h.tag, req->rc.capacity);
215 			WRITE_ONCE(req->status, REQ_STATUS_ERROR);
216 			goto recv_error;
217 		}
218 
219 		req->rc.size = h.size;
220 		req->rc.id = h.id;
221 		req->rc.tag = h.tag;
222 		req->rc.offset = 0;
223 
224 		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
225 		/* Then, read the whole packet (including the header) */
226 		xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
227 				     masked_prod, &masked_cons,
228 				     XEN_9PFS_RING_SIZE(ring));
229 
230 recv_error:
231 		virt_mb();
232 		cons += h.size;
233 		ring->intf->in_cons = cons;
234 
235 		status = (req->status != REQ_STATUS_ERROR) ?
236 			REQ_STATUS_RCVD : REQ_STATUS_ERROR;
237 
238 		p9_client_cb(priv->client, req, status);
239 	}
240 }
241 
242 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
243 {
244 	struct xen_9pfs_dataring *ring = r;
245 
246 	if (!ring || !ring->priv->client) {
247 		/* ignore spurious interrupt */
248 		return IRQ_HANDLED;
249 	}
250 
251 	wake_up_interruptible(&ring->wq);
252 	schedule_work(&ring->work);
253 
254 	return IRQ_HANDLED;
255 }
256 
257 static struct p9_trans_module p9_xen_trans = {
258 	.name = "xen",
259 	.maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
260 	.pooled_rbuffers = false,
261 	.def = 1,
262 	.create = p9_xen_create,
263 	.close = p9_xen_close,
264 	.request = p9_xen_request,
265 	.cancel = p9_xen_cancel,
266 	.owner = THIS_MODULE,
267 };
268 
269 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
270 	{ "9pfs" },
271 	{ "" }
272 };
273 
274 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
275 {
276 	int i, j;
277 
278 	write_lock(&xen_9pfs_lock);
279 	list_del(&priv->list);
280 	write_unlock(&xen_9pfs_lock);
281 
282 	for (i = 0; i < priv->num_rings; i++) {
283 		if (!priv->rings[i].intf)
284 			break;
285 		if (priv->rings[i].irq > 0)
286 			unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
287 		if (priv->rings[i].data.in) {
288 			for (j = 0;
289 			     j < (1 << priv->rings[i].intf->ring_order);
290 			     j++) {
291 				grant_ref_t ref;
292 
293 				ref = priv->rings[i].intf->ref[j];
294 				gnttab_end_foreign_access(ref, NULL);
295 			}
296 			free_pages_exact(priv->rings[i].data.in,
297 				   1UL << (priv->rings[i].intf->ring_order +
298 					   XEN_PAGE_SHIFT));
299 		}
300 		gnttab_end_foreign_access(priv->rings[i].ref, NULL);
301 		free_page((unsigned long)priv->rings[i].intf);
302 	}
303 	kfree(priv->rings);
304 	kfree(priv->tag);
305 	kfree(priv);
306 }
307 
308 static void xen_9pfs_front_remove(struct xenbus_device *dev)
309 {
310 	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
311 
312 	dev_set_drvdata(&dev->dev, NULL);
313 	xen_9pfs_front_free(priv);
314 }
315 
316 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
317 					 struct xen_9pfs_dataring *ring,
318 					 unsigned int order)
319 {
320 	int i = 0;
321 	int ret = -ENOMEM;
322 	void *bytes = NULL;
323 
324 	init_waitqueue_head(&ring->wq);
325 	spin_lock_init(&ring->lock);
326 	INIT_WORK(&ring->work, p9_xen_response);
327 
328 	ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
329 	if (!ring->intf)
330 		return ret;
331 	ret = gnttab_grant_foreign_access(dev->otherend_id,
332 					  virt_to_gfn(ring->intf), 0);
333 	if (ret < 0)
334 		goto out;
335 	ring->ref = ret;
336 	bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
337 				  GFP_KERNEL | __GFP_ZERO);
338 	if (!bytes) {
339 		ret = -ENOMEM;
340 		goto out;
341 	}
342 	for (; i < (1 << order); i++) {
343 		ret = gnttab_grant_foreign_access(
344 				dev->otherend_id, virt_to_gfn(bytes) + i, 0);
345 		if (ret < 0)
346 			goto out;
347 		ring->intf->ref[i] = ret;
348 	}
349 	ring->intf->ring_order = order;
350 	ring->data.in = bytes;
351 	ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
352 
353 	ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
354 	if (ret)
355 		goto out;
356 	ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
357 					      xen_9pfs_front_event_handler,
358 					      0, "xen_9pfs-frontend", ring);
359 	if (ring->irq >= 0)
360 		return 0;
361 
362 	xenbus_free_evtchn(dev, ring->evtchn);
363 	ret = ring->irq;
364 out:
365 	if (bytes) {
366 		for (i--; i >= 0; i--)
367 			gnttab_end_foreign_access(ring->intf->ref[i], NULL);
368 		free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
369 	}
370 	gnttab_end_foreign_access(ring->ref, NULL);
371 	free_page((unsigned long)ring->intf);
372 	return ret;
373 }
374 
375 static int xen_9pfs_front_init(struct xenbus_device *dev)
376 {
377 	int ret, i;
378 	struct xenbus_transaction xbt;
379 	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
380 	char *versions, *v;
381 	unsigned int max_rings, max_ring_order, len = 0;
382 
383 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
384 	if (IS_ERR(versions))
385 		return PTR_ERR(versions);
386 	for (v = versions; *v; v++) {
387 		if (simple_strtoul(v, &v, 10) == 1) {
388 			v = NULL;
389 			break;
390 		}
391 	}
392 	if (v) {
393 		kfree(versions);
394 		return -EINVAL;
395 	}
396 	kfree(versions);
397 	max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
398 	if (max_rings < XEN_9PFS_NUM_RINGS)
399 		return -EINVAL;
400 	max_ring_order = xenbus_read_unsigned(dev->otherend,
401 					      "max-ring-page-order", 0);
402 	if (max_ring_order > XEN_9PFS_RING_ORDER)
403 		max_ring_order = XEN_9PFS_RING_ORDER;
404 	if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
405 		p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
406 
407 	priv->num_rings = XEN_9PFS_NUM_RINGS;
408 	priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
409 			      GFP_KERNEL);
410 	if (!priv->rings) {
411 		kfree(priv);
412 		return -ENOMEM;
413 	}
414 
415 	for (i = 0; i < priv->num_rings; i++) {
416 		priv->rings[i].priv = priv;
417 		ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
418 						    max_ring_order);
419 		if (ret < 0)
420 			goto error;
421 	}
422 
423  again:
424 	ret = xenbus_transaction_start(&xbt);
425 	if (ret) {
426 		xenbus_dev_fatal(dev, ret, "starting transaction");
427 		goto error;
428 	}
429 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
430 	if (ret)
431 		goto error_xenbus;
432 	ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
433 			    priv->num_rings);
434 	if (ret)
435 		goto error_xenbus;
436 	for (i = 0; i < priv->num_rings; i++) {
437 		char str[16];
438 
439 		BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
440 		sprintf(str, "ring-ref%d", i);
441 		ret = xenbus_printf(xbt, dev->nodename, str, "%d",
442 				    priv->rings[i].ref);
443 		if (ret)
444 			goto error_xenbus;
445 
446 		sprintf(str, "event-channel-%d", i);
447 		ret = xenbus_printf(xbt, dev->nodename, str, "%u",
448 				    priv->rings[i].evtchn);
449 		if (ret)
450 			goto error_xenbus;
451 	}
452 	priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
453 	if (IS_ERR(priv->tag)) {
454 		ret = PTR_ERR(priv->tag);
455 		goto error_xenbus;
456 	}
457 	ret = xenbus_transaction_end(xbt, 0);
458 	if (ret) {
459 		if (ret == -EAGAIN)
460 			goto again;
461 		xenbus_dev_fatal(dev, ret, "completing transaction");
462 		goto error;
463 	}
464 
465 	return 0;
466 
467  error_xenbus:
468 	xenbus_transaction_end(xbt, 1);
469 	xenbus_dev_fatal(dev, ret, "writing xenstore");
470  error:
471 	xen_9pfs_front_free(priv);
472 	return ret;
473 }
474 
475 static int xen_9pfs_front_probe(struct xenbus_device *dev,
476 				const struct xenbus_device_id *id)
477 {
478 	struct xen_9pfs_front_priv *priv = NULL;
479 
480 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
481 	if (!priv)
482 		return -ENOMEM;
483 
484 	priv->dev = dev;
485 	dev_set_drvdata(&dev->dev, priv);
486 
487 	write_lock(&xen_9pfs_lock);
488 	list_add_tail(&priv->list, &xen_9pfs_devs);
489 	write_unlock(&xen_9pfs_lock);
490 
491 	return 0;
492 }
493 
494 static int xen_9pfs_front_resume(struct xenbus_device *dev)
495 {
496 	dev_warn(&dev->dev, "suspend/resume unsupported\n");
497 	return 0;
498 }
499 
500 static void xen_9pfs_front_changed(struct xenbus_device *dev,
501 				   enum xenbus_state backend_state)
502 {
503 	switch (backend_state) {
504 	case XenbusStateReconfiguring:
505 	case XenbusStateReconfigured:
506 	case XenbusStateInitialising:
507 	case XenbusStateInitialised:
508 	case XenbusStateUnknown:
509 		break;
510 
511 	case XenbusStateInitWait:
512 		if (!xen_9pfs_front_init(dev))
513 			xenbus_switch_state(dev, XenbusStateInitialised);
514 		break;
515 
516 	case XenbusStateConnected:
517 		xenbus_switch_state(dev, XenbusStateConnected);
518 		break;
519 
520 	case XenbusStateClosed:
521 		if (dev->state == XenbusStateClosed)
522 			break;
523 		fallthrough;	/* Missed the backend's CLOSING state */
524 	case XenbusStateClosing:
525 		xenbus_frontend_closed(dev);
526 		break;
527 	}
528 }
529 
530 static struct xenbus_driver xen_9pfs_front_driver = {
531 	.ids = xen_9pfs_front_ids,
532 	.probe = xen_9pfs_front_probe,
533 	.remove = xen_9pfs_front_remove,
534 	.resume = xen_9pfs_front_resume,
535 	.otherend_changed = xen_9pfs_front_changed,
536 };
537 
538 static int __init p9_trans_xen_init(void)
539 {
540 	int rc;
541 
542 	if (!xen_domain())
543 		return -ENODEV;
544 
545 	pr_info("Initialising Xen transport for 9pfs\n");
546 
547 	v9fs_register_trans(&p9_xen_trans);
548 	rc = xenbus_register_frontend(&xen_9pfs_front_driver);
549 	if (rc)
550 		v9fs_unregister_trans(&p9_xen_trans);
551 
552 	return rc;
553 }
554 module_init(p9_trans_xen_init);
555 MODULE_ALIAS_9P("xen");
556 
557 static void __exit p9_trans_xen_exit(void)
558 {
559 	v9fs_unregister_trans(&p9_xen_trans);
560 	return xenbus_unregister_driver(&xen_9pfs_front_driver);
561 }
562 module_exit(p9_trans_xen_exit);
563 
564 MODULE_ALIAS("xen:9pfs");
565 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
566 MODULE_DESCRIPTION("Xen Transport for 9P");
567 MODULE_LICENSE("GPL");
568