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 <linux/fs_context.h>
19 #include <net/9p/9p.h>
20 #include <net/9p/client.h>
21 #include <net/9p/transport.h>
22
23 #define XEN_9PFS_NUM_RINGS 2
24 #define XEN_9PFS_RING_ORDER 9
25 #define XEN_9PFS_RING_SIZE(ring) XEN_FLEX_RING_SIZE(ring->intf->ring_order)
26
27 struct xen_9pfs_header {
28 uint32_t size;
29 uint8_t id;
30 uint16_t tag;
31
32 /* uint8_t sdata[]; */
33 } __attribute__((packed));
34
35 /* One per ring, more than one per 9pfs share */
36 struct xen_9pfs_dataring {
37 struct xen_9pfs_front_priv *priv;
38
39 struct xen_9pfs_data_intf *intf;
40 grant_ref_t ref;
41 int evtchn;
42 int irq;
43 /* protect a ring from concurrent accesses */
44 spinlock_t lock;
45
46 struct xen_9pfs_data data;
47 wait_queue_head_t wq;
48 struct work_struct work;
49 };
50
51 /* One per 9pfs share */
52 struct xen_9pfs_front_priv {
53 struct list_head list;
54 struct xenbus_device *dev;
55 char *tag;
56 struct p9_client *client;
57
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 */
p9_xen_cancel(struct p9_client * client,struct p9_req_t * req)65 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
66 {
67 return 1;
68 }
69
p9_xen_create(struct p9_client * client,struct fs_context * fc)70 static int p9_xen_create(struct p9_client *client, struct fs_context *fc)
71 {
72 const char *addr = fc->source;
73 struct xen_9pfs_front_priv *priv;
74
75 if (addr == NULL)
76 return -EINVAL;
77
78 read_lock(&xen_9pfs_lock);
79 list_for_each_entry(priv, &xen_9pfs_devs, list) {
80 if (!strcmp(priv->tag, addr)) {
81 priv->client = client;
82 read_unlock(&xen_9pfs_lock);
83 return 0;
84 }
85 }
86 read_unlock(&xen_9pfs_lock);
87 return -EINVAL;
88 }
89
p9_xen_close(struct p9_client * client)90 static void p9_xen_close(struct p9_client *client)
91 {
92 struct xen_9pfs_front_priv *priv;
93
94 read_lock(&xen_9pfs_lock);
95 list_for_each_entry(priv, &xen_9pfs_devs, list) {
96 if (priv->client == client) {
97 priv->client = NULL;
98 read_unlock(&xen_9pfs_lock);
99 return;
100 }
101 }
102 read_unlock(&xen_9pfs_lock);
103 }
104
p9_xen_write_todo(struct xen_9pfs_dataring * ring,RING_IDX size)105 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
106 {
107 RING_IDX cons, prod;
108
109 cons = ring->intf->out_cons;
110 prod = ring->intf->out_prod;
111 virt_mb();
112
113 return XEN_9PFS_RING_SIZE(ring) -
114 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size;
115 }
116
p9_xen_request(struct p9_client * client,struct p9_req_t * p9_req)117 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
118 {
119 struct xen_9pfs_front_priv *priv;
120 RING_IDX cons, prod, masked_cons, masked_prod;
121 unsigned long flags;
122 u32 size = p9_req->tc.size;
123 struct xen_9pfs_dataring *ring;
124 int num;
125
126 read_lock(&xen_9pfs_lock);
127 list_for_each_entry(priv, &xen_9pfs_devs, list) {
128 if (priv->client == client)
129 break;
130 }
131 read_unlock(&xen_9pfs_lock);
132 if (list_entry_is_head(priv, &xen_9pfs_devs, list))
133 return -EINVAL;
134
135 num = p9_req->tc.tag % XEN_9PFS_NUM_RINGS;
136 ring = &priv->rings[num];
137
138 again:
139 while (io_wait_event_killable(ring->wq,
140 p9_xen_write_todo(ring, size)) != 0)
141 ;
142
143 spin_lock_irqsave(&ring->lock, flags);
144 cons = ring->intf->out_cons;
145 prod = ring->intf->out_prod;
146 virt_mb();
147
148 if (XEN_9PFS_RING_SIZE(ring) -
149 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) < size) {
150 spin_unlock_irqrestore(&ring->lock, flags);
151 goto again;
152 }
153
154 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
155 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
156
157 xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
158 &masked_prod, masked_cons,
159 XEN_9PFS_RING_SIZE(ring));
160
161 WRITE_ONCE(p9_req->status, REQ_STATUS_SENT);
162 virt_wmb(); /* write ring before updating pointer */
163 prod += size;
164 ring->intf->out_prod = prod;
165 spin_unlock_irqrestore(&ring->lock, flags);
166 notify_remote_via_irq(ring->irq);
167 p9_req_put(client, p9_req);
168
169 return 0;
170 }
171
p9_xen_response(struct work_struct * work)172 static void p9_xen_response(struct work_struct *work)
173 {
174 struct xen_9pfs_front_priv *priv;
175 struct xen_9pfs_dataring *ring;
176 RING_IDX cons, prod, masked_cons, masked_prod;
177 struct xen_9pfs_header h;
178 struct p9_req_t *req;
179 int status;
180
181 ring = container_of(work, struct xen_9pfs_dataring, work);
182 priv = ring->priv;
183
184 while (1) {
185 cons = ring->intf->in_cons;
186 prod = ring->intf->in_prod;
187 virt_rmb();
188
189 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) <
190 sizeof(h)) {
191 notify_remote_via_irq(ring->irq);
192 return;
193 }
194
195 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
196 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
197
198 /* First, read just the header */
199 xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
200 masked_prod, &masked_cons,
201 XEN_9PFS_RING_SIZE(ring));
202
203 req = p9_tag_lookup(priv->client, h.tag);
204 if (!req || req->status != REQ_STATUS_SENT) {
205 dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
206 cons += h.size;
207 virt_mb();
208 ring->intf->in_cons = cons;
209 continue;
210 }
211
212 if (h.size > req->rc.capacity) {
213 dev_warn(&priv->dev->dev,
214 "requested packet size too big: %d for tag %d with capacity %zd\n",
215 h.size, h.tag, req->rc.capacity);
216 WRITE_ONCE(req->status, REQ_STATUS_ERROR);
217 goto recv_error;
218 }
219
220 req->rc.size = h.size;
221 req->rc.id = h.id;
222 req->rc.tag = h.tag;
223 req->rc.offset = 0;
224
225 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
226 /* Then, read the whole packet (including the header) */
227 xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
228 masked_prod, &masked_cons,
229 XEN_9PFS_RING_SIZE(ring));
230
231 recv_error:
232 virt_mb();
233 cons += h.size;
234 ring->intf->in_cons = cons;
235
236 status = (req->status != REQ_STATUS_ERROR) ?
237 REQ_STATUS_RCVD : REQ_STATUS_ERROR;
238
239 p9_client_cb(priv->client, req, status);
240 }
241 }
242
xen_9pfs_front_event_handler(int irq,void * r)243 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
244 {
245 struct xen_9pfs_dataring *ring = r;
246
247 if (!ring || !ring->priv->client) {
248 /* ignore spurious interrupt */
249 return IRQ_HANDLED;
250 }
251
252 wake_up_interruptible(&ring->wq);
253 schedule_work(&ring->work);
254
255 return IRQ_HANDLED;
256 }
257
258 static struct p9_trans_module p9_xen_trans = {
259 .name = "xen",
260 .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
261 .pooled_rbuffers = false,
262 .def = true,
263 .supports_vmalloc = false,
264 .create = p9_xen_create,
265 .close = p9_xen_close,
266 .request = p9_xen_request,
267 .cancel = p9_xen_cancel,
268 .owner = THIS_MODULE,
269 };
270
271 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
272 { "9pfs" },
273 { "" }
274 };
275
xen_9pfs_front_free(struct xen_9pfs_front_priv * priv)276 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
277 {
278 int i, j;
279
280 if (priv->rings) {
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 (!ring->intf)
287 break;
288 if (ring->irq >= 0) {
289 unbind_from_irqhandler(ring->irq, ring);
290 ring->irq = -1;
291 }
292 if (ring->data.in) {
293 for (j = 0; j < (1 << ring->intf->ring_order);
294 j++) {
295 grant_ref_t ref;
296
297 ref = ring->intf->ref[j];
298 gnttab_end_foreign_access(ref, NULL);
299 ring->intf->ref[j] = INVALID_GRANT_REF;
300 }
301 free_pages_exact(ring->data.in,
302 1UL << (ring->intf->ring_order +
303 XEN_PAGE_SHIFT));
304 ring->data.in = NULL;
305 ring->data.out = NULL;
306 }
307 if (ring->ref != INVALID_GRANT_REF) {
308 gnttab_end_foreign_access(ring->ref, NULL);
309 ring->ref = INVALID_GRANT_REF;
310 }
311 free_page((unsigned long)ring->intf);
312 ring->intf = NULL;
313 }
314 kfree(priv->rings);
315 }
316 kfree(priv->tag);
317 kfree(priv);
318 }
319
xen_9pfs_front_remove(struct xenbus_device * dev)320 static void xen_9pfs_front_remove(struct xenbus_device *dev)
321 {
322 struct xen_9pfs_front_priv *priv;
323
324 write_lock(&xen_9pfs_lock);
325 priv = dev_get_drvdata(&dev->dev);
326 if (priv == NULL) {
327 write_unlock(&xen_9pfs_lock);
328 return;
329 }
330 dev_set_drvdata(&dev->dev, NULL);
331 list_del(&priv->list);
332 write_unlock(&xen_9pfs_lock);
333
334 xen_9pfs_front_free(priv);
335 }
336
xen_9pfs_front_alloc_dataring(struct xenbus_device * dev,struct xen_9pfs_dataring * ring,unsigned int order)337 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
338 struct xen_9pfs_dataring *ring,
339 unsigned int order)
340 {
341 int i = 0;
342 int ret = -ENOMEM;
343 void *bytes = NULL;
344
345 ring->intf = NULL;
346 ring->data.in = NULL;
347 ring->data.out = NULL;
348 ring->ref = INVALID_GRANT_REF;
349 ring->irq = -1;
350
351 init_waitqueue_head(&ring->wq);
352 spin_lock_init(&ring->lock);
353 INIT_WORK(&ring->work, p9_xen_response);
354
355 ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
356 if (!ring->intf)
357 return ret;
358 ret = gnttab_grant_foreign_access(dev->otherend_id,
359 virt_to_gfn(ring->intf), 0);
360 if (ret < 0)
361 goto out;
362 ring->ref = ret;
363 bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
364 GFP_KERNEL | __GFP_ZERO);
365 if (!bytes) {
366 ret = -ENOMEM;
367 goto out;
368 }
369 for (; i < (1 << order); i++) {
370 ret = gnttab_grant_foreign_access(
371 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
372 if (ret < 0)
373 goto out;
374 ring->intf->ref[i] = ret;
375 }
376 ring->intf->ring_order = order;
377 ring->data.in = bytes;
378 ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
379
380 ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
381 if (ret)
382 goto out;
383 ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
384 xen_9pfs_front_event_handler,
385 0, "xen_9pfs-frontend", ring);
386 if (ring->irq >= 0)
387 return 0;
388
389 xenbus_free_evtchn(dev, ring->evtchn);
390 ret = ring->irq;
391 out:
392 if (bytes) {
393 for (i--; i >= 0; i--)
394 gnttab_end_foreign_access(ring->intf->ref[i], NULL);
395 free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
396 ring->data.in = NULL;
397 ring->data.out = NULL;
398 }
399 if (ring->ref != INVALID_GRANT_REF) {
400 gnttab_end_foreign_access(ring->ref, NULL);
401 ring->ref = INVALID_GRANT_REF;
402 }
403 if (ring->intf) {
404 free_page((unsigned long)ring->intf);
405 ring->intf = NULL;
406 }
407 ring->irq = -1;
408 return ret;
409 }
410
xen_9pfs_front_init(struct xenbus_device * dev)411 static int xen_9pfs_front_init(struct xenbus_device *dev)
412 {
413 int ret, i;
414 struct xenbus_transaction xbt;
415 struct xen_9pfs_front_priv *priv;
416 char *versions, *v, *token;
417 bool version_1 = false;
418 unsigned int max_rings, max_ring_order, len = 0, version;
419
420 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
421 if (IS_ERR(versions))
422 return PTR_ERR(versions);
423 for (v = versions; (token = strsep(&v, ",")); ) {
424 if (!*token)
425 continue;
426
427 ret = kstrtouint(token, 10, &version);
428 if (ret) {
429 kfree(versions);
430 return ret;
431 }
432 if (version == 1)
433 version_1 = true;
434 }
435 kfree(versions);
436 if (!version_1)
437 return -EINVAL;
438
439 max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
440 if (max_rings < XEN_9PFS_NUM_RINGS)
441 return -EINVAL;
442 max_ring_order = xenbus_read_unsigned(dev->otherend,
443 "max-ring-page-order", 0);
444 if (max_ring_order > XEN_9PFS_RING_ORDER)
445 max_ring_order = XEN_9PFS_RING_ORDER;
446 if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
447 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
448
449 priv = kzalloc_obj(*priv);
450 if (!priv)
451 return -ENOMEM;
452 priv->dev = dev;
453 priv->rings = kzalloc_objs(*priv->rings, XEN_9PFS_NUM_RINGS);
454 if (!priv->rings) {
455 kfree(priv);
456 return -ENOMEM;
457 }
458
459 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
460 priv->rings[i].priv = priv;
461 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
462 max_ring_order);
463 if (ret < 0)
464 goto error;
465 }
466
467 again:
468 ret = xenbus_transaction_start(&xbt);
469 if (ret) {
470 xenbus_dev_fatal(dev, ret, "starting transaction");
471 goto error;
472 }
473 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
474 if (ret)
475 goto error_xenbus;
476 ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
477 XEN_9PFS_NUM_RINGS);
478 if (ret)
479 goto error_xenbus;
480
481 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
482 char str[16];
483
484 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
485 sprintf(str, "ring-ref%d", i);
486 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
487 priv->rings[i].ref);
488 if (ret)
489 goto error_xenbus;
490
491 sprintf(str, "event-channel-%d", i);
492 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
493 priv->rings[i].evtchn);
494 if (ret)
495 goto error_xenbus;
496 }
497 priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
498 if (IS_ERR(priv->tag)) {
499 ret = PTR_ERR(priv->tag);
500 goto error_xenbus;
501 }
502 ret = xenbus_transaction_end(xbt, 0);
503 if (ret) {
504 if (ret == -EAGAIN)
505 goto again;
506 xenbus_dev_fatal(dev, ret, "completing transaction");
507 goto error;
508 }
509
510 write_lock(&xen_9pfs_lock);
511 dev_set_drvdata(&dev->dev, priv);
512 list_add_tail(&priv->list, &xen_9pfs_devs);
513 write_unlock(&xen_9pfs_lock);
514
515 xenbus_switch_state(dev, XenbusStateInitialised);
516 return 0;
517
518 error_xenbus:
519 xenbus_transaction_end(xbt, 1);
520 xenbus_dev_fatal(dev, ret, "writing xenstore");
521 error:
522 xen_9pfs_front_free(priv);
523 return ret;
524 }
525
xen_9pfs_front_probe(struct xenbus_device * dev,const struct xenbus_device_id * id)526 static int xen_9pfs_front_probe(struct xenbus_device *dev,
527 const struct xenbus_device_id *id)
528 {
529 return 0;
530 }
531
xen_9pfs_front_resume(struct xenbus_device * dev)532 static int xen_9pfs_front_resume(struct xenbus_device *dev)
533 {
534 dev_warn(&dev->dev, "suspend/resume unsupported\n");
535 return 0;
536 }
537
xen_9pfs_front_changed(struct xenbus_device * dev,enum xenbus_state backend_state)538 static void xen_9pfs_front_changed(struct xenbus_device *dev,
539 enum xenbus_state backend_state)
540 {
541 switch (backend_state) {
542 case XenbusStateReconfiguring:
543 case XenbusStateReconfigured:
544 case XenbusStateInitialising:
545 case XenbusStateInitialised:
546 case XenbusStateUnknown:
547 break;
548
549 case XenbusStateInitWait:
550 if (dev->state != XenbusStateInitialising)
551 break;
552
553 xen_9pfs_front_init(dev);
554 break;
555
556 case XenbusStateConnected:
557 xenbus_switch_state(dev, XenbusStateConnected);
558 break;
559
560 case XenbusStateClosed:
561 if (dev->state == XenbusStateClosed)
562 break;
563 fallthrough; /* Missed the backend's CLOSING state */
564 case XenbusStateClosing:
565 xenbus_frontend_closed(dev);
566 break;
567 }
568 }
569
570 static struct xenbus_driver xen_9pfs_front_driver = {
571 .ids = xen_9pfs_front_ids,
572 .probe = xen_9pfs_front_probe,
573 .remove = xen_9pfs_front_remove,
574 .resume = xen_9pfs_front_resume,
575 .otherend_changed = xen_9pfs_front_changed,
576 };
577
p9_trans_xen_init(void)578 static int __init p9_trans_xen_init(void)
579 {
580 int rc;
581
582 if (!xen_domain())
583 return -ENODEV;
584
585 pr_info("Initialising Xen transport for 9pfs\n");
586
587 v9fs_register_trans(&p9_xen_trans);
588 rc = xenbus_register_frontend(&xen_9pfs_front_driver);
589 if (rc)
590 v9fs_unregister_trans(&p9_xen_trans);
591
592 return rc;
593 }
594 module_init(p9_trans_xen_init);
595 MODULE_ALIAS_9P("xen");
596
p9_trans_xen_exit(void)597 static void __exit p9_trans_xen_exit(void)
598 {
599 v9fs_unregister_trans(&p9_xen_trans);
600 return xenbus_unregister_driver(&xen_9pfs_front_driver);
601 }
602 module_exit(p9_trans_xen_exit);
603
604 MODULE_ALIAS("xen:9pfs");
605 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
606 MODULE_DESCRIPTION("Xen Transport for 9P");
607 MODULE_LICENSE("GPL");
608