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
26struct 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 */
35struct 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 */
51struct xen_9pfs_front_priv {
52	struct list_head list;
53	struct xenbus_device *dev;
54	char *tag;
55	struct p9_client *client;
56
57	struct xen_9pfs_dataring *rings;
58};
59
60static LIST_HEAD(xen_9pfs_devs);
61static DEFINE_RWLOCK(xen_9pfs_lock);
62
63/* We don't currently allow canceling of requests */
64static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
65{
66	return 1;
67}
68
69static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
70{
71	struct xen_9pfs_front_priv *priv;
72
73	if (addr == NULL)
74		return -EINVAL;
75
76	read_lock(&xen_9pfs_lock);
77	list_for_each_entry(priv, &xen_9pfs_devs, list) {
78		if (!strcmp(priv->tag, addr)) {
79			priv->client = client;
80			read_unlock(&xen_9pfs_lock);
81			return 0;
82		}
83	}
84	read_unlock(&xen_9pfs_lock);
85	return -EINVAL;
86}
87
88static void p9_xen_close(struct p9_client *client)
89{
90	struct xen_9pfs_front_priv *priv;
91
92	read_lock(&xen_9pfs_lock);
93	list_for_each_entry(priv, &xen_9pfs_devs, list) {
94		if (priv->client == client) {
95			priv->client = NULL;
96			read_unlock(&xen_9pfs_lock);
97			return;
98		}
99	}
100	read_unlock(&xen_9pfs_lock);
101}
102
103static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
104{
105	RING_IDX cons, prod;
106
107	cons = ring->intf->out_cons;
108	prod = ring->intf->out_prod;
109	virt_mb();
110
111	return XEN_9PFS_RING_SIZE(ring) -
112		xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size;
113}
114
115static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
116{
117	struct xen_9pfs_front_priv *priv;
118	RING_IDX cons, prod, masked_cons, masked_prod;
119	unsigned long flags;
120	u32 size = p9_req->tc.size;
121	struct xen_9pfs_dataring *ring;
122	int num;
123
124	read_lock(&xen_9pfs_lock);
125	list_for_each_entry(priv, &xen_9pfs_devs, list) {
126		if (priv->client == client)
127			break;
128	}
129	read_unlock(&xen_9pfs_lock);
130	if (list_entry_is_head(priv, &xen_9pfs_devs, list))
131		return -EINVAL;
132
133	num = p9_req->tc.tag % XEN_9PFS_NUM_RINGS;
134	ring = &priv->rings[num];
135
136again:
137	while (wait_event_killable(ring->wq,
138				   p9_xen_write_todo(ring, size)) != 0)
139		;
140
141	spin_lock_irqsave(&ring->lock, flags);
142	cons = ring->intf->out_cons;
143	prod = ring->intf->out_prod;
144	virt_mb();
145
146	if (XEN_9PFS_RING_SIZE(ring) -
147	    xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) < size) {
148		spin_unlock_irqrestore(&ring->lock, flags);
149		goto again;
150	}
151
152	masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
153	masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
154
155	xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
156			      &masked_prod, masked_cons,
157			      XEN_9PFS_RING_SIZE(ring));
158
159	WRITE_ONCE(p9_req->status, REQ_STATUS_SENT);
160	virt_wmb();			/* write ring before updating pointer */
161	prod += size;
162	ring->intf->out_prod = prod;
163	spin_unlock_irqrestore(&ring->lock, flags);
164	notify_remote_via_irq(ring->irq);
165	p9_req_put(client, p9_req);
166
167	return 0;
168}
169
170static void p9_xen_response(struct work_struct *work)
171{
172	struct xen_9pfs_front_priv *priv;
173	struct xen_9pfs_dataring *ring;
174	RING_IDX cons, prod, masked_cons, masked_prod;
175	struct xen_9pfs_header h;
176	struct p9_req_t *req;
177	int status;
178
179	ring = container_of(work, struct xen_9pfs_dataring, work);
180	priv = ring->priv;
181
182	while (1) {
183		cons = ring->intf->in_cons;
184		prod = ring->intf->in_prod;
185		virt_rmb();
186
187		if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) <
188		    sizeof(h)) {
189			notify_remote_via_irq(ring->irq);
190			return;
191		}
192
193		masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
194		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
195
196		/* First, read just the header */
197		xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
198				     masked_prod, &masked_cons,
199				     XEN_9PFS_RING_SIZE(ring));
200
201		req = p9_tag_lookup(priv->client, h.tag);
202		if (!req || req->status != REQ_STATUS_SENT) {
203			dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
204			cons += h.size;
205			virt_mb();
206			ring->intf->in_cons = cons;
207			continue;
208		}
209
210		if (h.size > req->rc.capacity) {
211			dev_warn(&priv->dev->dev,
212				 "requested packet size too big: %d for tag %d with capacity %zd\n",
213				 h.size, h.tag, req->rc.capacity);
214			WRITE_ONCE(req->status, REQ_STATUS_ERROR);
215			goto recv_error;
216		}
217
218		req->rc.size = h.size;
219		req->rc.id = h.id;
220		req->rc.tag = h.tag;
221		req->rc.offset = 0;
222
223		masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
224		/* Then, read the whole packet (including the header) */
225		xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
226				     masked_prod, &masked_cons,
227				     XEN_9PFS_RING_SIZE(ring));
228
229recv_error:
230		virt_mb();
231		cons += h.size;
232		ring->intf->in_cons = cons;
233
234		status = (req->status != REQ_STATUS_ERROR) ?
235			REQ_STATUS_RCVD : REQ_STATUS_ERROR;
236
237		p9_client_cb(priv->client, req, status);
238	}
239}
240
241static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
242{
243	struct xen_9pfs_dataring *ring = r;
244
245	if (!ring || !ring->priv->client) {
246		/* ignore spurious interrupt */
247		return IRQ_HANDLED;
248	}
249
250	wake_up_interruptible(&ring->wq);
251	schedule_work(&ring->work);
252
253	return IRQ_HANDLED;
254}
255
256static struct p9_trans_module p9_xen_trans = {
257	.name = "xen",
258	.maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
259	.pooled_rbuffers = false,
260	.def = 1,
261	.create = p9_xen_create,
262	.close = p9_xen_close,
263	.request = p9_xen_request,
264	.cancel = p9_xen_cancel,
265	.owner = THIS_MODULE,
266};
267
268static const struct xenbus_device_id xen_9pfs_front_ids[] = {
269	{ "9pfs" },
270	{ "" }
271};
272
273static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
274{
275	int i, j;
276
277	write_lock(&xen_9pfs_lock);
278	list_del(&priv->list);
279	write_unlock(&xen_9pfs_lock);
280
281	for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
282		struct xen_9pfs_dataring *ring = &priv->rings[i];
283
284		cancel_work_sync(&ring->work);
285
286		if (!priv->rings[i].intf)
287			break;
288		if (priv->rings[i].irq > 0)
289			unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
290		if (priv->rings[i].data.in) {
291			for (j = 0;
292			     j < (1 << priv->rings[i].intf->ring_order);
293			     j++) {
294				grant_ref_t ref;
295
296				ref = priv->rings[i].intf->ref[j];
297				gnttab_end_foreign_access(ref, NULL);
298			}
299			free_pages_exact(priv->rings[i].data.in,
300				   1UL << (priv->rings[i].intf->ring_order +
301					   XEN_PAGE_SHIFT));
302		}
303		gnttab_end_foreign_access(priv->rings[i].ref, NULL);
304		free_page((unsigned long)priv->rings[i].intf);
305	}
306	kfree(priv->rings);
307	kfree(priv->tag);
308	kfree(priv);
309}
310
311static void xen_9pfs_front_remove(struct xenbus_device *dev)
312{
313	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
314
315	dev_set_drvdata(&dev->dev, NULL);
316	xen_9pfs_front_free(priv);
317}
318
319static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
320					 struct xen_9pfs_dataring *ring,
321					 unsigned int order)
322{
323	int i = 0;
324	int ret = -ENOMEM;
325	void *bytes = NULL;
326
327	init_waitqueue_head(&ring->wq);
328	spin_lock_init(&ring->lock);
329	INIT_WORK(&ring->work, p9_xen_response);
330
331	ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
332	if (!ring->intf)
333		return ret;
334	ret = gnttab_grant_foreign_access(dev->otherend_id,
335					  virt_to_gfn(ring->intf), 0);
336	if (ret < 0)
337		goto out;
338	ring->ref = ret;
339	bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
340				  GFP_KERNEL | __GFP_ZERO);
341	if (!bytes) {
342		ret = -ENOMEM;
343		goto out;
344	}
345	for (; i < (1 << order); i++) {
346		ret = gnttab_grant_foreign_access(
347				dev->otherend_id, virt_to_gfn(bytes) + i, 0);
348		if (ret < 0)
349			goto out;
350		ring->intf->ref[i] = ret;
351	}
352	ring->intf->ring_order = order;
353	ring->data.in = bytes;
354	ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
355
356	ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
357	if (ret)
358		goto out;
359	ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
360					      xen_9pfs_front_event_handler,
361					      0, "xen_9pfs-frontend", ring);
362	if (ring->irq >= 0)
363		return 0;
364
365	xenbus_free_evtchn(dev, ring->evtchn);
366	ret = ring->irq;
367out:
368	if (bytes) {
369		for (i--; i >= 0; i--)
370			gnttab_end_foreign_access(ring->intf->ref[i], NULL);
371		free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
372	}
373	gnttab_end_foreign_access(ring->ref, NULL);
374	free_page((unsigned long)ring->intf);
375	return ret;
376}
377
378static int xen_9pfs_front_init(struct xenbus_device *dev)
379{
380	int ret, i;
381	struct xenbus_transaction xbt;
382	struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
383	char *versions, *v;
384	unsigned int max_rings, max_ring_order, len = 0;
385
386	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
387	if (IS_ERR(versions))
388		return PTR_ERR(versions);
389	for (v = versions; *v; v++) {
390		if (simple_strtoul(v, &v, 10) == 1) {
391			v = NULL;
392			break;
393		}
394	}
395	if (v) {
396		kfree(versions);
397		return -EINVAL;
398	}
399	kfree(versions);
400	max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
401	if (max_rings < XEN_9PFS_NUM_RINGS)
402		return -EINVAL;
403	max_ring_order = xenbus_read_unsigned(dev->otherend,
404					      "max-ring-page-order", 0);
405	if (max_ring_order > XEN_9PFS_RING_ORDER)
406		max_ring_order = XEN_9PFS_RING_ORDER;
407	if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
408		p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
409
410	priv->rings = kcalloc(XEN_9PFS_NUM_RINGS, sizeof(*priv->rings),
411			      GFP_KERNEL);
412	if (!priv->rings) {
413		kfree(priv);
414		return -ENOMEM;
415	}
416
417	for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
418		priv->rings[i].priv = priv;
419		ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
420						    max_ring_order);
421		if (ret < 0)
422			goto error;
423	}
424
425 again:
426	ret = xenbus_transaction_start(&xbt);
427	if (ret) {
428		xenbus_dev_fatal(dev, ret, "starting transaction");
429		goto error;
430	}
431	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
432	if (ret)
433		goto error_xenbus;
434	ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
435			    XEN_9PFS_NUM_RINGS);
436	if (ret)
437		goto error_xenbus;
438
439	for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
440		char str[16];
441
442		BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
443		sprintf(str, "ring-ref%d", 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-%d", 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	return 0;
469
470 error_xenbus:
471	xenbus_transaction_end(xbt, 1);
472	xenbus_dev_fatal(dev, ret, "writing xenstore");
473 error:
474	xen_9pfs_front_free(priv);
475	return ret;
476}
477
478static int xen_9pfs_front_probe(struct xenbus_device *dev,
479				const struct xenbus_device_id *id)
480{
481	struct xen_9pfs_front_priv *priv = NULL;
482
483	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
484	if (!priv)
485		return -ENOMEM;
486
487	priv->dev = dev;
488	dev_set_drvdata(&dev->dev, priv);
489
490	write_lock(&xen_9pfs_lock);
491	list_add_tail(&priv->list, &xen_9pfs_devs);
492	write_unlock(&xen_9pfs_lock);
493
494	return 0;
495}
496
497static int xen_9pfs_front_resume(struct xenbus_device *dev)
498{
499	dev_warn(&dev->dev, "suspend/resume unsupported\n");
500	return 0;
501}
502
503static void xen_9pfs_front_changed(struct xenbus_device *dev,
504				   enum xenbus_state backend_state)
505{
506	switch (backend_state) {
507	case XenbusStateReconfiguring:
508	case XenbusStateReconfigured:
509	case XenbusStateInitialising:
510	case XenbusStateInitialised:
511	case XenbusStateUnknown:
512		break;
513
514	case XenbusStateInitWait:
515		if (!xen_9pfs_front_init(dev))
516			xenbus_switch_state(dev, XenbusStateInitialised);
517		break;
518
519	case XenbusStateConnected:
520		xenbus_switch_state(dev, XenbusStateConnected);
521		break;
522
523	case XenbusStateClosed:
524		if (dev->state == XenbusStateClosed)
525			break;
526		fallthrough;	/* Missed the backend's CLOSING state */
527	case XenbusStateClosing:
528		xenbus_frontend_closed(dev);
529		break;
530	}
531}
532
533static struct xenbus_driver xen_9pfs_front_driver = {
534	.ids = xen_9pfs_front_ids,
535	.probe = xen_9pfs_front_probe,
536	.remove = xen_9pfs_front_remove,
537	.resume = xen_9pfs_front_resume,
538	.otherend_changed = xen_9pfs_front_changed,
539};
540
541static int __init p9_trans_xen_init(void)
542{
543	int rc;
544
545	if (!xen_domain())
546		return -ENODEV;
547
548	pr_info("Initialising Xen transport for 9pfs\n");
549
550	v9fs_register_trans(&p9_xen_trans);
551	rc = xenbus_register_frontend(&xen_9pfs_front_driver);
552	if (rc)
553		v9fs_unregister_trans(&p9_xen_trans);
554
555	return rc;
556}
557module_init(p9_trans_xen_init);
558MODULE_ALIAS_9P("xen");
559
560static void __exit p9_trans_xen_exit(void)
561{
562	v9fs_unregister_trans(&p9_xen_trans);
563	return xenbus_unregister_driver(&xen_9pfs_front_driver);
564}
565module_exit(p9_trans_xen_exit);
566
567MODULE_ALIAS("xen:9pfs");
568MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
569MODULE_DESCRIPTION("Xen Transport for 9P");
570MODULE_LICENSE("GPL");
571