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
27struct 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 */
36struct 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 */
52struct 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
61static LIST_HEAD(xen_9pfs_devs);
62static DEFINE_RWLOCK(xen_9pfs_lock);
63
64/* We don't currently allow canceling of requests */
65static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
66{
67 return 1;
68}
69
70static 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
90static 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
105static 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
117static 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
138again:
139 while (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(lock: &ring->lock, flags);
151 goto again;
152 }
153
154 masked_prod = xen_9pfs_mask(idx: prod, XEN_9PFS_RING_SIZE(ring));
155 masked_cons = xen_9pfs_mask(idx: cons, XEN_9PFS_RING_SIZE(ring));
156
157 xen_9pfs_write_packet(buf: ring->data.out, opaque: p9_req->tc.sdata, size,
158 masked_prod: &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(lock: &ring->lock, flags);
166 notify_remote_via_irq(irq: ring->irq);
167 p9_req_put(c: client, r: p9_req);
168
169 return 0;
170}
171
172static 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(irq: ring->irq);
192 return;
193 }
194
195 masked_prod = xen_9pfs_mask(idx: prod, XEN_9PFS_RING_SIZE(ring));
196 masked_cons = xen_9pfs_mask(idx: cons, XEN_9PFS_RING_SIZE(ring));
197
198 /* First, read just the header */
199 xen_9pfs_read_packet(opaque: &h, buf: ring->data.in, size: sizeof(h),
200 masked_prod, masked_cons: &masked_cons,
201 XEN_9PFS_RING_SIZE(ring));
202
203 req = p9_tag_lookup(c: priv->client, tag: 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(idx: cons, XEN_9PFS_RING_SIZE(ring));
226 /* Then, read the whole packet (including the header) */
227 xen_9pfs_read_packet(opaque: req->rc.sdata, buf: ring->data.in, size: h.size,
228 masked_prod, masked_cons: &masked_cons,
229 XEN_9PFS_RING_SIZE(ring));
230
231recv_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(c: priv->client, req, status);
240 }
241}
242
243static 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(work: &ring->work);
254
255 return IRQ_HANDLED;
256}
257
258static 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
271static const struct xenbus_device_id xen_9pfs_front_ids[] = {
272 { "9pfs" },
273 { "" }
274};
275
276static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
277{
278 int i, j;
279
280 write_lock(&xen_9pfs_lock);
281 list_del(entry: &priv->list);
282 write_unlock(&xen_9pfs_lock);
283
284 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
285 struct xen_9pfs_dataring *ring = &priv->rings[i];
286
287 cancel_work_sync(work: &ring->work);
288
289 if (!priv->rings[i].intf)
290 break;
291 if (priv->rings[i].irq > 0)
292 unbind_from_irqhandler(irq: priv->rings[i].irq, dev_id: ring);
293 if (priv->rings[i].data.in) {
294 for (j = 0;
295 j < (1 << priv->rings[i].intf->ring_order);
296 j++) {
297 grant_ref_t ref;
298
299 ref = priv->rings[i].intf->ref[j];
300 gnttab_end_foreign_access(ref, NULL);
301 }
302 free_pages_exact(virt: priv->rings[i].data.in,
303 size: 1UL << (priv->rings[i].intf->ring_order +
304 XEN_PAGE_SHIFT));
305 }
306 gnttab_end_foreign_access(ref: priv->rings[i].ref, NULL);
307 free_page((unsigned long)priv->rings[i].intf);
308 }
309 kfree(objp: priv->rings);
310 kfree(objp: priv->tag);
311 kfree(objp: priv);
312}
313
314static void xen_9pfs_front_remove(struct xenbus_device *dev)
315{
316 struct xen_9pfs_front_priv *priv = dev_get_drvdata(dev: &dev->dev);
317
318 dev_set_drvdata(dev: &dev->dev, NULL);
319 xen_9pfs_front_free(priv);
320}
321
322static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
323 struct xen_9pfs_dataring *ring,
324 unsigned int order)
325{
326 int i = 0;
327 int ret = -ENOMEM;
328 void *bytes = NULL;
329
330 init_waitqueue_head(&ring->wq);
331 spin_lock_init(&ring->lock);
332 INIT_WORK(&ring->work, p9_xen_response);
333
334 ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
335 if (!ring->intf)
336 return ret;
337 ret = gnttab_grant_foreign_access(domid: dev->otherend_id,
338 virt_to_gfn(ring->intf), readonly: 0);
339 if (ret < 0)
340 goto out;
341 ring->ref = ret;
342 bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
343 GFP_KERNEL | __GFP_ZERO);
344 if (!bytes) {
345 ret = -ENOMEM;
346 goto out;
347 }
348 for (; i < (1 << order); i++) {
349 ret = gnttab_grant_foreign_access(
350 domid: dev->otherend_id, virt_to_gfn(bytes) + i, readonly: 0);
351 if (ret < 0)
352 goto out;
353 ring->intf->ref[i] = ret;
354 }
355 ring->intf->ring_order = order;
356 ring->data.in = bytes;
357 ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
358
359 ret = xenbus_alloc_evtchn(dev, port: &ring->evtchn);
360 if (ret)
361 goto out;
362 ring->irq = bind_evtchn_to_irqhandler(evtchn: ring->evtchn,
363 handler: xen_9pfs_front_event_handler,
364 irqflags: 0, devname: "xen_9pfs-frontend", dev_id: ring);
365 if (ring->irq >= 0)
366 return 0;
367
368 xenbus_free_evtchn(dev, port: ring->evtchn);
369 ret = ring->irq;
370out:
371 if (bytes) {
372 for (i--; i >= 0; i--)
373 gnttab_end_foreign_access(ref: ring->intf->ref[i], NULL);
374 free_pages_exact(virt: bytes, size: 1UL << (order + XEN_PAGE_SHIFT));
375 }
376 gnttab_end_foreign_access(ref: ring->ref, NULL);
377 free_page((unsigned long)ring->intf);
378 return ret;
379}
380
381static int xen_9pfs_front_init(struct xenbus_device *dev)
382{
383 int ret, i;
384 struct xenbus_transaction xbt;
385 struct xen_9pfs_front_priv *priv = dev_get_drvdata(dev: &dev->dev);
386 char *versions, *v;
387 unsigned int max_rings, max_ring_order, len = 0;
388
389 versions = xenbus_read(XBT_NIL, dir: dev->otherend, node: "versions", len: &len);
390 if (IS_ERR(versions))
391 return PTR_ERR(versions);
392 for (v = versions; *v; v++) {
393 if (simple_strtoul(v, &v, 10) == 1) {
394 v = NULL;
395 break;
396 }
397 }
398 if (v) {
399 kfree(versions);
400 return -EINVAL;
401 }
402 kfree(versions);
403 max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
404 if (max_rings < XEN_9PFS_NUM_RINGS)
405 return -EINVAL;
406 max_ring_order = xenbus_read_unsigned(dev->otherend,
407 "max-ring-page-order", 0);
408 if (max_ring_order > XEN_9PFS_RING_ORDER)
409 max_ring_order = XEN_9PFS_RING_ORDER;
410 if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
411 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
412
413 priv->rings = kcalloc(XEN_9PFS_NUM_RINGS, sizeof(*priv->rings),
414 GFP_KERNEL);
415 if (!priv->rings) {
416 kfree(priv);
417 return -ENOMEM;
418 }
419
420 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
421 priv->rings[i].priv = priv;
422 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
423 max_ring_order);
424 if (ret < 0)
425 goto error;
426 }
427
428 again:
429 ret = xenbus_transaction_start(&xbt);
430 if (ret) {
431 xenbus_dev_fatal(dev, ret, "starting transaction");
432 goto error;
433 }
434 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
435 if (ret)
436 goto error_xenbus;
437 ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
438 XEN_9PFS_NUM_RINGS);
439 if (ret)
440 goto error_xenbus;
441
442 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) {
443 char str[16];
444
445 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
446 sprintf(str, "ring-ref%d", i);
447 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
448 priv->rings[i].ref);
449 if (ret)
450 goto error_xenbus;
451
452 sprintf(str, "event-channel-%d", i);
453 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
454 priv->rings[i].evtchn);
455 if (ret)
456 goto error_xenbus;
457 }
458 priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
459 if (IS_ERR(priv->tag)) {
460 ret = PTR_ERR(priv->tag);
461 goto error_xenbus;
462 }
463 ret = xenbus_transaction_end(xbt, 0);
464 if (ret) {
465 if (ret == -EAGAIN)
466 goto again;
467 xenbus_dev_fatal(dev, ret, "completing transaction");
468 goto error;
469 }
470
471 xenbus_switch_state(dev, XenbusStateInitialised);
472 return 0;
473
474 error_xenbus:
475 xenbus_transaction_end(xbt, 1);
476 xenbus_dev_fatal(dev, ret, "writing xenstore");
477 error:
478 xen_9pfs_front_free(priv);
479 return ret;
480}
481
482static int xen_9pfs_front_probe(struct xenbus_device *dev,
483 const struct xenbus_device_id *id)
484{
485 struct xen_9pfs_front_priv *priv = NULL;
486
487 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
488 if (!priv)
489 return -ENOMEM;
490
491 priv->dev = dev;
492 dev_set_drvdata(dev: &dev->dev, data: priv);
493
494 write_lock(&xen_9pfs_lock);
495 list_add_tail(new: &priv->list, head: &xen_9pfs_devs);
496 write_unlock(&xen_9pfs_lock);
497
498 return 0;
499}
500
501static int xen_9pfs_front_resume(struct xenbus_device *dev)
502{
503 dev_warn(&dev->dev, "suspend/resume unsupported\n");
504 return 0;
505}
506
507static void xen_9pfs_front_changed(struct xenbus_device *dev,
508 enum xenbus_state backend_state)
509{
510 switch (backend_state) {
511 case XenbusStateReconfiguring:
512 case XenbusStateReconfigured:
513 case XenbusStateInitialising:
514 case XenbusStateInitialised:
515 case XenbusStateUnknown:
516 break;
517
518 case XenbusStateInitWait:
519 if (dev->state != XenbusStateInitialising)
520 break;
521
522 xen_9pfs_front_init(dev);
523 break;
524
525 case XenbusStateConnected:
526 xenbus_switch_state(dev, new_state: XenbusStateConnected);
527 break;
528
529 case XenbusStateClosed:
530 if (dev->state == XenbusStateClosed)
531 break;
532 fallthrough; /* Missed the backend's CLOSING state */
533 case XenbusStateClosing:
534 xenbus_frontend_closed(dev);
535 break;
536 }
537}
538
539static struct xenbus_driver xen_9pfs_front_driver = {
540 .ids = xen_9pfs_front_ids,
541 .probe = xen_9pfs_front_probe,
542 .remove = xen_9pfs_front_remove,
543 .resume = xen_9pfs_front_resume,
544 .otherend_changed = xen_9pfs_front_changed,
545};
546
547static int __init p9_trans_xen_init(void)
548{
549 int rc;
550
551 if (!xen_domain())
552 return -ENODEV;
553
554 pr_info("Initialising Xen transport for 9pfs\n");
555
556 v9fs_register_trans(m: &p9_xen_trans);
557 rc = xenbus_register_frontend(&xen_9pfs_front_driver);
558 if (rc)
559 v9fs_unregister_trans(m: &p9_xen_trans);
560
561 return rc;
562}
563module_init(p9_trans_xen_init);
564MODULE_ALIAS_9P("xen");
565
566static void __exit p9_trans_xen_exit(void)
567{
568 v9fs_unregister_trans(m: &p9_xen_trans);
569 return xenbus_unregister_driver(drv: &xen_9pfs_front_driver);
570}
571module_exit(p9_trans_xen_exit);
572
573MODULE_ALIAS("xen:9pfs");
574MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
575MODULE_DESCRIPTION("Xen Transport for 9P");
576MODULE_LICENSE("GPL");
577

source code of linux/net/9p/trans_xen.c