xref: /openbmc/linux/net/9p/trans_xen.c (revision 9aa2cba7)
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 		struct xen_9pfs_dataring *ring = &priv->rings[i];
284 
285 		cancel_work_sync(&ring->work);
286 
287 		if (!priv->rings[i].intf)
288 			break;
289 		if (priv->rings[i].irq > 0)
290 			unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
291 		if (priv->rings[i].data.in) {
292 			for (j = 0;
293 			     j < (1 << priv->rings[i].intf->ring_order);
294 			     j++) {
295 				grant_ref_t ref;
296 
297 				ref = priv->rings[i].intf->ref[j];
298 				gnttab_end_foreign_access(ref, NULL);
299 			}
300 			free_pages_exact(priv->rings[i].data.in,
301 				   1UL << (priv->rings[i].intf->ring_order +
302 					   XEN_PAGE_SHIFT));
303 		}
304 		gnttab_end_foreign_access(priv->rings[i].ref, NULL);
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 void 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 }
319 
320 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
321 					 struct xen_9pfs_dataring *ring,
322 					 unsigned int order)
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 = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
341 				  GFP_KERNEL | __GFP_ZERO);
342 	if (!bytes) {
343 		ret = -ENOMEM;
344 		goto out;
345 	}
346 	for (; i < (1 << 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 = order;
354 	ring->data.in = bytes;
355 	ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
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], NULL);
372 		free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
373 	}
374 	gnttab_end_foreign_access(ring->ref, NULL);
375 	free_page((unsigned long)ring->intf);
376 	return ret;
377 }
378 
379 static int xen_9pfs_front_init(struct xenbus_device *dev)
380 {
381 	int ret, i;
382 	struct xenbus_transaction xbt;
383 	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
384 	char *versions, *v;
385 	unsigned int max_rings, max_ring_order, len = 0;
386 
387 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
388 	if (IS_ERR(versions))
389 		return PTR_ERR(versions);
390 	for (v = versions; *v; v++) {
391 		if (simple_strtoul(v, &v, 10) == 1) {
392 			v = NULL;
393 			break;
394 		}
395 	}
396 	if (v) {
397 		kfree(versions);
398 		return -EINVAL;
399 	}
400 	kfree(versions);
401 	max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
402 	if (max_rings < XEN_9PFS_NUM_RINGS)
403 		return -EINVAL;
404 	max_ring_order = xenbus_read_unsigned(dev->otherend,
405 					      "max-ring-page-order", 0);
406 	if (max_ring_order > XEN_9PFS_RING_ORDER)
407 		max_ring_order = XEN_9PFS_RING_ORDER;
408 	if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
409 		p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
410 
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 						    max_ring_order);
423 		if (ret < 0)
424 			goto error;
425 	}
426 
427  again:
428 	ret = xenbus_transaction_start(&xbt);
429 	if (ret) {
430 		xenbus_dev_fatal(dev, ret, "starting transaction");
431 		goto error;
432 	}
433 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
434 	if (ret)
435 		goto error_xenbus;
436 	ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
437 			    priv->num_rings);
438 	if (ret)
439 		goto error_xenbus;
440 	for (i = 0; i < priv->num_rings; i++) {
441 		char str[16];
442 
443 		BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
444 		sprintf(str, "ring-ref%d", i);
445 		ret = xenbus_printf(xbt, dev->nodename, str, "%d",
446 				    priv->rings[i].ref);
447 		if (ret)
448 			goto error_xenbus;
449 
450 		sprintf(str, "event-channel-%d", i);
451 		ret = xenbus_printf(xbt, dev->nodename, str, "%u",
452 				    priv->rings[i].evtchn);
453 		if (ret)
454 			goto error_xenbus;
455 	}
456 	priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
457 	if (IS_ERR(priv->tag)) {
458 		ret = PTR_ERR(priv->tag);
459 		goto error_xenbus;
460 	}
461 	ret = xenbus_transaction_end(xbt, 0);
462 	if (ret) {
463 		if (ret == -EAGAIN)
464 			goto again;
465 		xenbus_dev_fatal(dev, ret, "completing transaction");
466 		goto error;
467 	}
468 
469 	return 0;
470 
471  error_xenbus:
472 	xenbus_transaction_end(xbt, 1);
473 	xenbus_dev_fatal(dev, ret, "writing xenstore");
474  error:
475 	xen_9pfs_front_free(priv);
476 	return ret;
477 }
478 
479 static int xen_9pfs_front_probe(struct xenbus_device *dev,
480 				const struct xenbus_device_id *id)
481 {
482 	struct xen_9pfs_front_priv *priv = NULL;
483 
484 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
485 	if (!priv)
486 		return -ENOMEM;
487 
488 	priv->dev = dev;
489 	dev_set_drvdata(&dev->dev, priv);
490 
491 	write_lock(&xen_9pfs_lock);
492 	list_add_tail(&priv->list, &xen_9pfs_devs);
493 	write_unlock(&xen_9pfs_lock);
494 
495 	return 0;
496 }
497 
498 static int xen_9pfs_front_resume(struct xenbus_device *dev)
499 {
500 	dev_warn(&dev->dev, "suspend/resume unsupported\n");
501 	return 0;
502 }
503 
504 static void xen_9pfs_front_changed(struct xenbus_device *dev,
505 				   enum xenbus_state backend_state)
506 {
507 	switch (backend_state) {
508 	case XenbusStateReconfiguring:
509 	case XenbusStateReconfigured:
510 	case XenbusStateInitialising:
511 	case XenbusStateInitialised:
512 	case XenbusStateUnknown:
513 		break;
514 
515 	case XenbusStateInitWait:
516 		if (!xen_9pfs_front_init(dev))
517 			xenbus_switch_state(dev, XenbusStateInitialised);
518 		break;
519 
520 	case XenbusStateConnected:
521 		xenbus_switch_state(dev, XenbusStateConnected);
522 		break;
523 
524 	case XenbusStateClosed:
525 		if (dev->state == XenbusStateClosed)
526 			break;
527 		fallthrough;	/* Missed the backend's CLOSING state */
528 	case XenbusStateClosing:
529 		xenbus_frontend_closed(dev);
530 		break;
531 	}
532 }
533 
534 static struct xenbus_driver xen_9pfs_front_driver = {
535 	.ids = xen_9pfs_front_ids,
536 	.probe = xen_9pfs_front_probe,
537 	.remove = xen_9pfs_front_remove,
538 	.resume = xen_9pfs_front_resume,
539 	.otherend_changed = xen_9pfs_front_changed,
540 };
541 
542 static int __init p9_trans_xen_init(void)
543 {
544 	int rc;
545 
546 	if (!xen_domain())
547 		return -ENODEV;
548 
549 	pr_info("Initialising Xen transport for 9pfs\n");
550 
551 	v9fs_register_trans(&p9_xen_trans);
552 	rc = xenbus_register_frontend(&xen_9pfs_front_driver);
553 	if (rc)
554 		v9fs_unregister_trans(&p9_xen_trans);
555 
556 	return rc;
557 }
558 module_init(p9_trans_xen_init);
559 MODULE_ALIAS_9P("xen");
560 
561 static void __exit p9_trans_xen_exit(void)
562 {
563 	v9fs_unregister_trans(&p9_xen_trans);
564 	return xenbus_unregister_driver(&xen_9pfs_front_driver);
565 }
566 module_exit(p9_trans_xen_exit);
567 
568 MODULE_ALIAS("xen:9pfs");
569 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
570 MODULE_DESCRIPTION("Xen Transport for 9P");
571 MODULE_LICENSE("GPL");
572