xref: /linux/net/9p/trans_xen.c (revision cc8c418b4fc09ed58ddd27b8e90ec797e9ca1e67)
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(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 		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 	.pooled_rbuffers = false,
250 	.def = 1,
251 	.create = p9_xen_create,
252 	.close = p9_xen_close,
253 	.request = p9_xen_request,
254 	.cancel = p9_xen_cancel,
255 	.owner = THIS_MODULE,
256 };
257 
258 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
259 	{ "9pfs" },
260 	{ "" }
261 };
262 
263 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
264 {
265 	int i, j;
266 
267 	write_lock(&xen_9pfs_lock);
268 	list_del(&priv->list);
269 	write_unlock(&xen_9pfs_lock);
270 
271 	for (i = 0; i < priv->num_rings; i++) {
272 		if (!priv->rings[i].intf)
273 			break;
274 		if (priv->rings[i].irq > 0)
275 			unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
276 		if (priv->rings[i].data.in) {
277 			for (j = 0;
278 			     j < (1 << priv->rings[i].intf->ring_order);
279 			     j++) {
280 				grant_ref_t ref;
281 
282 				ref = priv->rings[i].intf->ref[j];
283 				gnttab_end_foreign_access(ref, NULL);
284 			}
285 			free_pages_exact(priv->rings[i].data.in,
286 				   1UL << (priv->rings[i].intf->ring_order +
287 					   XEN_PAGE_SHIFT));
288 		}
289 		gnttab_end_foreign_access(priv->rings[i].ref, NULL);
290 		free_page((unsigned long)priv->rings[i].intf);
291 	}
292 	kfree(priv->rings);
293 	kfree(priv->tag);
294 	kfree(priv);
295 }
296 
297 static int xen_9pfs_front_remove(struct xenbus_device *dev)
298 {
299 	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
300 
301 	dev_set_drvdata(&dev->dev, NULL);
302 	xen_9pfs_front_free(priv);
303 	return 0;
304 }
305 
306 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
307 					 struct xen_9pfs_dataring *ring,
308 					 unsigned int order)
309 {
310 	int i = 0;
311 	int ret = -ENOMEM;
312 	void *bytes = NULL;
313 
314 	init_waitqueue_head(&ring->wq);
315 	spin_lock_init(&ring->lock);
316 	INIT_WORK(&ring->work, p9_xen_response);
317 
318 	ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
319 	if (!ring->intf)
320 		return ret;
321 	ret = gnttab_grant_foreign_access(dev->otherend_id,
322 					  virt_to_gfn(ring->intf), 0);
323 	if (ret < 0)
324 		goto out;
325 	ring->ref = ret;
326 	bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
327 				  GFP_KERNEL | __GFP_ZERO);
328 	if (!bytes) {
329 		ret = -ENOMEM;
330 		goto out;
331 	}
332 	for (; i < (1 << order); i++) {
333 		ret = gnttab_grant_foreign_access(
334 				dev->otherend_id, virt_to_gfn(bytes) + i, 0);
335 		if (ret < 0)
336 			goto out;
337 		ring->intf->ref[i] = ret;
338 	}
339 	ring->intf->ring_order = order;
340 	ring->data.in = bytes;
341 	ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
342 
343 	ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
344 	if (ret)
345 		goto out;
346 	ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
347 					      xen_9pfs_front_event_handler,
348 					      0, "xen_9pfs-frontend", ring);
349 	if (ring->irq >= 0)
350 		return 0;
351 
352 	xenbus_free_evtchn(dev, ring->evtchn);
353 	ret = ring->irq;
354 out:
355 	if (bytes) {
356 		for (i--; i >= 0; i--)
357 			gnttab_end_foreign_access(ring->intf->ref[i], NULL);
358 		free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
359 	}
360 	gnttab_end_foreign_access(ring->ref, NULL);
361 	free_page((unsigned long)ring->intf);
362 	return ret;
363 }
364 
365 static int xen_9pfs_front_probe(struct xenbus_device *dev,
366 				const struct xenbus_device_id *id)
367 {
368 	int ret, i;
369 	struct xenbus_transaction xbt;
370 	struct xen_9pfs_front_priv *priv = NULL;
371 	char *versions;
372 	unsigned int max_rings, max_ring_order, len = 0;
373 
374 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
375 	if (IS_ERR(versions))
376 		return PTR_ERR(versions);
377 	if (strcmp(versions, "1")) {
378 		kfree(versions);
379 		return -EINVAL;
380 	}
381 	kfree(versions);
382 	max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
383 	if (max_rings < XEN_9PFS_NUM_RINGS)
384 		return -EINVAL;
385 	max_ring_order = xenbus_read_unsigned(dev->otherend,
386 					      "max-ring-page-order", 0);
387 	if (max_ring_order > XEN_9PFS_RING_ORDER)
388 		max_ring_order = XEN_9PFS_RING_ORDER;
389 	if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
390 		p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
391 
392 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
393 	if (!priv)
394 		return -ENOMEM;
395 
396 	priv->dev = dev;
397 	priv->num_rings = XEN_9PFS_NUM_RINGS;
398 	priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
399 			      GFP_KERNEL);
400 	if (!priv->rings) {
401 		kfree(priv);
402 		return -ENOMEM;
403 	}
404 
405 	for (i = 0; i < priv->num_rings; i++) {
406 		priv->rings[i].priv = priv;
407 		ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
408 						    max_ring_order);
409 		if (ret < 0)
410 			goto error;
411 	}
412 
413  again:
414 	ret = xenbus_transaction_start(&xbt);
415 	if (ret) {
416 		xenbus_dev_fatal(dev, ret, "starting transaction");
417 		goto error;
418 	}
419 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
420 	if (ret)
421 		goto error_xenbus;
422 	ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
423 			    priv->num_rings);
424 	if (ret)
425 		goto error_xenbus;
426 	for (i = 0; i < priv->num_rings; i++) {
427 		char str[16];
428 
429 		BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
430 		sprintf(str, "ring-ref%d", i);
431 		ret = xenbus_printf(xbt, dev->nodename, str, "%d",
432 				    priv->rings[i].ref);
433 		if (ret)
434 			goto error_xenbus;
435 
436 		sprintf(str, "event-channel-%d", i);
437 		ret = xenbus_printf(xbt, dev->nodename, str, "%u",
438 				    priv->rings[i].evtchn);
439 		if (ret)
440 			goto error_xenbus;
441 	}
442 	priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
443 	if (IS_ERR(priv->tag)) {
444 		ret = PTR_ERR(priv->tag);
445 		goto error_xenbus;
446 	}
447 	ret = xenbus_transaction_end(xbt, 0);
448 	if (ret) {
449 		if (ret == -EAGAIN)
450 			goto again;
451 		xenbus_dev_fatal(dev, ret, "completing transaction");
452 		goto error;
453 	}
454 
455 	write_lock(&xen_9pfs_lock);
456 	list_add_tail(&priv->list, &xen_9pfs_devs);
457 	write_unlock(&xen_9pfs_lock);
458 	dev_set_drvdata(&dev->dev, priv);
459 	xenbus_switch_state(dev, XenbusStateInitialised);
460 
461 	return 0;
462 
463  error_xenbus:
464 	xenbus_transaction_end(xbt, 1);
465 	xenbus_dev_fatal(dev, ret, "writing xenstore");
466  error:
467 	dev_set_drvdata(&dev->dev, NULL);
468 	xen_9pfs_front_free(priv);
469 	return ret;
470 }
471 
472 static int xen_9pfs_front_resume(struct xenbus_device *dev)
473 {
474 	dev_warn(&dev->dev, "suspend/resume unsupported\n");
475 	return 0;
476 }
477 
478 static void xen_9pfs_front_changed(struct xenbus_device *dev,
479 				   enum xenbus_state backend_state)
480 {
481 	switch (backend_state) {
482 	case XenbusStateReconfiguring:
483 	case XenbusStateReconfigured:
484 	case XenbusStateInitialising:
485 	case XenbusStateInitialised:
486 	case XenbusStateUnknown:
487 		break;
488 
489 	case XenbusStateInitWait:
490 		break;
491 
492 	case XenbusStateConnected:
493 		xenbus_switch_state(dev, XenbusStateConnected);
494 		break;
495 
496 	case XenbusStateClosed:
497 		if (dev->state == XenbusStateClosed)
498 			break;
499 		fallthrough;	/* Missed the backend's CLOSING state */
500 	case XenbusStateClosing:
501 		xenbus_frontend_closed(dev);
502 		break;
503 	}
504 }
505 
506 static struct xenbus_driver xen_9pfs_front_driver = {
507 	.ids = xen_9pfs_front_ids,
508 	.probe = xen_9pfs_front_probe,
509 	.remove = xen_9pfs_front_remove,
510 	.resume = xen_9pfs_front_resume,
511 	.otherend_changed = xen_9pfs_front_changed,
512 };
513 
514 static int __init p9_trans_xen_init(void)
515 {
516 	int rc;
517 
518 	if (!xen_domain())
519 		return -ENODEV;
520 
521 	pr_info("Initialising Xen transport for 9pfs\n");
522 
523 	v9fs_register_trans(&p9_xen_trans);
524 	rc = xenbus_register_frontend(&xen_9pfs_front_driver);
525 	if (rc)
526 		v9fs_unregister_trans(&p9_xen_trans);
527 
528 	return rc;
529 }
530 module_init(p9_trans_xen_init);
531 MODULE_ALIAS_9P("xen");
532 
533 static void __exit p9_trans_xen_exit(void)
534 {
535 	v9fs_unregister_trans(&p9_xen_trans);
536 	return xenbus_unregister_driver(&xen_9pfs_front_driver);
537 }
538 module_exit(p9_trans_xen_exit);
539 
540 MODULE_ALIAS("xen:9pfs");
541 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
542 MODULE_DESCRIPTION("Xen Transport for 9P");
543 MODULE_LICENSE("GPL");
544