Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[muen/linux.git] / net / 9p / trans_xen.c
1 /*
2  * linux/fs/9p/trans_xen
3  *
4  * Xen transport layer.
5  *
6  * Copyright (C) 2017 by Stefano Stabellini <stefano@aporeto.com>
7  *
8  * This program is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU General Public License version 2
10  * as published by the Free Software Foundation; or, when distributed
11  * separately from the Linux kernel or incorporated into other
12  * software packages, subject to the following license:
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this source file (the "Software"), to deal in the Software without
16  * restriction, including without limitation the rights to use, copy, modify,
17  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
18  * and to permit persons to whom the Software is furnished to do so, subject to
19  * the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
27  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
29  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
30  * IN THE SOFTWARE.
31  */
32
33 #include <xen/events.h>
34 #include <xen/grant_table.h>
35 #include <xen/xen.h>
36 #include <xen/xenbus.h>
37 #include <xen/interface/io/9pfs.h>
38
39 #include <linux/module.h>
40 #include <linux/spinlock.h>
41 #include <linux/rwlock.h>
42 #include <net/9p/9p.h>
43 #include <net/9p/client.h>
44 #include <net/9p/transport.h>
45
46 #define XEN_9PFS_NUM_RINGS 2
47 #define XEN_9PFS_RING_ORDER 6
48 #define XEN_9PFS_RING_SIZE  XEN_FLEX_RING_SIZE(XEN_9PFS_RING_ORDER)
49
50 struct xen_9pfs_header {
51         uint32_t size;
52         uint8_t id;
53         uint16_t tag;
54
55         /* uint8_t sdata[]; */
56 } __attribute__((packed));
57
58 /* One per ring, more than one per 9pfs share */
59 struct xen_9pfs_dataring {
60         struct xen_9pfs_front_priv *priv;
61
62         struct xen_9pfs_data_intf *intf;
63         grant_ref_t ref;
64         int evtchn;
65         int irq;
66         /* protect a ring from concurrent accesses */
67         spinlock_t lock;
68
69         struct xen_9pfs_data data;
70         wait_queue_head_t wq;
71         struct work_struct work;
72 };
73
74 /* One per 9pfs share */
75 struct xen_9pfs_front_priv {
76         struct list_head list;
77         struct xenbus_device *dev;
78         char *tag;
79         struct p9_client *client;
80
81         int num_rings;
82         struct xen_9pfs_dataring *rings;
83 };
84
85 static LIST_HEAD(xen_9pfs_devs);
86 static DEFINE_RWLOCK(xen_9pfs_lock);
87
88 /* We don't currently allow canceling of requests */
89 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
90 {
91         return 1;
92 }
93
94 static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
95 {
96         struct xen_9pfs_front_priv *priv;
97
98         read_lock(&xen_9pfs_lock);
99         list_for_each_entry(priv, &xen_9pfs_devs, list) {
100                 if (!strcmp(priv->tag, addr)) {
101                         priv->client = client;
102                         read_unlock(&xen_9pfs_lock);
103                         return 0;
104                 }
105         }
106         read_unlock(&xen_9pfs_lock);
107         return -EINVAL;
108 }
109
110 static void p9_xen_close(struct p9_client *client)
111 {
112         struct xen_9pfs_front_priv *priv;
113
114         read_lock(&xen_9pfs_lock);
115         list_for_each_entry(priv, &xen_9pfs_devs, list) {
116                 if (priv->client == client) {
117                         priv->client = NULL;
118                         read_unlock(&xen_9pfs_lock);
119                         return;
120                 }
121         }
122         read_unlock(&xen_9pfs_lock);
123 }
124
125 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
126 {
127         RING_IDX cons, prod;
128
129         cons = ring->intf->out_cons;
130         prod = ring->intf->out_prod;
131         virt_mb();
132
133         return XEN_9PFS_RING_SIZE -
134                 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) >= size;
135 }
136
137 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
138 {
139         struct xen_9pfs_front_priv *priv = NULL;
140         RING_IDX cons, prod, masked_cons, masked_prod;
141         unsigned long flags;
142         u32 size = p9_req->tc->size;
143         struct xen_9pfs_dataring *ring;
144         int num;
145
146         read_lock(&xen_9pfs_lock);
147         list_for_each_entry(priv, &xen_9pfs_devs, list) {
148                 if (priv->client == client)
149                         break;
150         }
151         read_unlock(&xen_9pfs_lock);
152         if (!priv || priv->client != client)
153                 return -EINVAL;
154
155         num = p9_req->tc->tag % priv->num_rings;
156         ring = &priv->rings[num];
157
158 again:
159         while (wait_event_interruptible(ring->wq,
160                                         p9_xen_write_todo(ring, size)) != 0)
161                 ;
162
163         spin_lock_irqsave(&ring->lock, flags);
164         cons = ring->intf->out_cons;
165         prod = ring->intf->out_prod;
166         virt_mb();
167
168         if (XEN_9PFS_RING_SIZE - xen_9pfs_queued(prod, cons,
169                                                  XEN_9PFS_RING_SIZE) < size) {
170                 spin_unlock_irqrestore(&ring->lock, flags);
171                 goto again;
172         }
173
174         masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
175         masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
176
177         xen_9pfs_write_packet(ring->data.out, p9_req->tc->sdata, size,
178                               &masked_prod, masked_cons, XEN_9PFS_RING_SIZE);
179
180         p9_req->status = REQ_STATUS_SENT;
181         virt_wmb();                     /* write ring before updating pointer */
182         prod += size;
183         ring->intf->out_prod = prod;
184         spin_unlock_irqrestore(&ring->lock, flags);
185         notify_remote_via_irq(ring->irq);
186
187         return 0;
188 }
189
190 static void p9_xen_response(struct work_struct *work)
191 {
192         struct xen_9pfs_front_priv *priv;
193         struct xen_9pfs_dataring *ring;
194         RING_IDX cons, prod, masked_cons, masked_prod;
195         struct xen_9pfs_header h;
196         struct p9_req_t *req;
197         int status;
198
199         ring = container_of(work, struct xen_9pfs_dataring, work);
200         priv = ring->priv;
201
202         while (1) {
203                 cons = ring->intf->in_cons;
204                 prod = ring->intf->in_prod;
205                 virt_rmb();
206
207                 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) <
208                     sizeof(h)) {
209                         notify_remote_via_irq(ring->irq);
210                         return;
211                 }
212
213                 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
214                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
215
216                 /* First, read just the header */
217                 xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
218                                      masked_prod, &masked_cons,
219                                      XEN_9PFS_RING_SIZE);
220
221                 req = p9_tag_lookup(priv->client, h.tag);
222                 if (!req || req->status != REQ_STATUS_SENT) {
223                         dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
224                         cons += h.size;
225                         virt_mb();
226                         ring->intf->in_cons = cons;
227                         continue;
228                 }
229
230                 memcpy(req->rc, &h, sizeof(h));
231                 req->rc->offset = 0;
232
233                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
234                 /* Then, read the whole packet (including the header) */
235                 xen_9pfs_read_packet(req->rc->sdata, ring->data.in, h.size,
236                                      masked_prod, &masked_cons,
237                                      XEN_9PFS_RING_SIZE);
238
239                 virt_mb();
240                 cons += h.size;
241                 ring->intf->in_cons = cons;
242
243                 status = (req->status != REQ_STATUS_ERROR) ?
244                         REQ_STATUS_RCVD : REQ_STATUS_ERROR;
245
246                 p9_client_cb(priv->client, req, status);
247         }
248 }
249
250 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
251 {
252         struct xen_9pfs_dataring *ring = r;
253
254         if (!ring || !ring->priv->client) {
255                 /* ignore spurious interrupt */
256                 return IRQ_HANDLED;
257         }
258
259         wake_up_interruptible(&ring->wq);
260         schedule_work(&ring->work);
261
262         return IRQ_HANDLED;
263 }
264
265 static struct p9_trans_module p9_xen_trans = {
266         .name = "xen",
267         .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
268         .def = 1,
269         .create = p9_xen_create,
270         .close = p9_xen_close,
271         .request = p9_xen_request,
272         .cancel = p9_xen_cancel,
273         .owner = THIS_MODULE,
274 };
275
276 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
277         { "9pfs" },
278         { "" }
279 };
280
281 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
282 {
283         int i, j;
284
285         write_lock(&xen_9pfs_lock);
286         list_del(&priv->list);
287         write_unlock(&xen_9pfs_lock);
288
289         for (i = 0; i < priv->num_rings; i++) {
290                 if (!priv->rings[i].intf)
291                         break;
292                 if (priv->rings[i].irq > 0)
293                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
294                 if (priv->rings[i].data.in) {
295                         for (j = 0; j < (1 << XEN_9PFS_RING_ORDER); j++) {
296                                 grant_ref_t ref;
297
298                                 ref = priv->rings[i].intf->ref[j];
299                                 gnttab_end_foreign_access(ref, 0, 0);
300                         }
301                         free_pages((unsigned long)priv->rings[i].data.in,
302                                    XEN_9PFS_RING_ORDER -
303                                    (PAGE_SHIFT - XEN_PAGE_SHIFT));
304                 }
305                 gnttab_end_foreign_access(priv->rings[i].ref, 0, 0);
306                 free_page((unsigned long)priv->rings[i].intf);
307         }
308         kfree(priv->rings);
309         kfree(priv->tag);
310         kfree(priv);
311 }
312
313 static int xen_9pfs_front_remove(struct xenbus_device *dev)
314 {
315         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
316
317         dev_set_drvdata(&dev->dev, NULL);
318         xen_9pfs_front_free(priv);
319         return 0;
320 }
321
322 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
323                                          struct xen_9pfs_dataring *ring)
324 {
325         int i = 0;
326         int ret = -ENOMEM;
327         void *bytes = NULL;
328
329         init_waitqueue_head(&ring->wq);
330         spin_lock_init(&ring->lock);
331         INIT_WORK(&ring->work, p9_xen_response);
332
333         ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
334         if (!ring->intf)
335                 return ret;
336         ret = gnttab_grant_foreign_access(dev->otherend_id,
337                                           virt_to_gfn(ring->intf), 0);
338         if (ret < 0)
339                 goto out;
340         ring->ref = ret;
341         bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
342                         XEN_9PFS_RING_ORDER - (PAGE_SHIFT - XEN_PAGE_SHIFT));
343         if (!bytes) {
344                 ret = -ENOMEM;
345                 goto out;
346         }
347         for (; i < (1 << XEN_9PFS_RING_ORDER); i++) {
348                 ret = gnttab_grant_foreign_access(
349                                 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
350                 if (ret < 0)
351                         goto out;
352                 ring->intf->ref[i] = ret;
353         }
354         ring->intf->ring_order = XEN_9PFS_RING_ORDER;
355         ring->data.in = bytes;
356         ring->data.out = bytes + XEN_9PFS_RING_SIZE;
357
358         ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
359         if (ret)
360                 goto out;
361         ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
362                                               xen_9pfs_front_event_handler,
363                                               0, "xen_9pfs-frontend", ring);
364         if (ring->irq >= 0)
365                 return 0;
366
367         xenbus_free_evtchn(dev, ring->evtchn);
368         ret = ring->irq;
369 out:
370         if (bytes) {
371                 for (i--; i >= 0; i--)
372                         gnttab_end_foreign_access(ring->intf->ref[i], 0, 0);
373                 free_pages((unsigned long)bytes,
374                            XEN_9PFS_RING_ORDER -
375                            (PAGE_SHIFT - XEN_PAGE_SHIFT));
376         }
377         gnttab_end_foreign_access(ring->ref, 0, 0);
378         free_page((unsigned long)ring->intf);
379         return ret;
380 }
381
382 static int xen_9pfs_front_probe(struct xenbus_device *dev,
383                                 const struct xenbus_device_id *id)
384 {
385         int ret, i;
386         struct xenbus_transaction xbt;
387         struct xen_9pfs_front_priv *priv = NULL;
388         char *versions;
389         unsigned int max_rings, max_ring_order, len = 0;
390
391         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
392         if (!len)
393                 return -EINVAL;
394         if (strcmp(versions, "1")) {
395                 kfree(versions);
396                 return -EINVAL;
397         }
398         kfree(versions);
399         max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
400         if (max_rings < XEN_9PFS_NUM_RINGS)
401                 return -EINVAL;
402         max_ring_order = xenbus_read_unsigned(dev->otherend,
403                                               "max-ring-page-order", 0);
404         if (max_ring_order < XEN_9PFS_RING_ORDER)
405                 return -EINVAL;
406
407         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
408         if (!priv)
409                 return -ENOMEM;
410
411         priv->dev = dev;
412         priv->num_rings = XEN_9PFS_NUM_RINGS;
413         priv->rings = kcalloc(priv->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 < priv->num_rings; i++) {
421                 priv->rings[i].priv = priv;
422                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i]);
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%u", 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-%u", 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         write_lock(&xen_9pfs_lock);
470         list_add_tail(&priv->list, &xen_9pfs_devs);
471         write_unlock(&xen_9pfs_lock);
472         dev_set_drvdata(&dev->dev, priv);
473         xenbus_switch_state(dev, XenbusStateInitialised);
474
475         return 0;
476
477  error_xenbus:
478         xenbus_transaction_end(xbt, 1);
479         xenbus_dev_fatal(dev, ret, "writing xenstore");
480  error:
481         dev_set_drvdata(&dev->dev, NULL);
482         xen_9pfs_front_free(priv);
483         return ret;
484 }
485
486 static int xen_9pfs_front_resume(struct xenbus_device *dev)
487 {
488         dev_warn(&dev->dev, "suspsend/resume unsupported\n");
489         return 0;
490 }
491
492 static void xen_9pfs_front_changed(struct xenbus_device *dev,
493                                    enum xenbus_state backend_state)
494 {
495         switch (backend_state) {
496         case XenbusStateReconfiguring:
497         case XenbusStateReconfigured:
498         case XenbusStateInitialising:
499         case XenbusStateInitialised:
500         case XenbusStateUnknown:
501                 break;
502
503         case XenbusStateInitWait:
504                 break;
505
506         case XenbusStateConnected:
507                 xenbus_switch_state(dev, XenbusStateConnected);
508                 break;
509
510         case XenbusStateClosed:
511                 if (dev->state == XenbusStateClosed)
512                         break;
513                 /* Missed the backend's CLOSING state -- fallthrough */
514         case XenbusStateClosing:
515                 xenbus_frontend_closed(dev);
516                 break;
517         }
518 }
519
520 static struct xenbus_driver xen_9pfs_front_driver = {
521         .ids = xen_9pfs_front_ids,
522         .probe = xen_9pfs_front_probe,
523         .remove = xen_9pfs_front_remove,
524         .resume = xen_9pfs_front_resume,
525         .otherend_changed = xen_9pfs_front_changed,
526 };
527
528 static int p9_trans_xen_init(void)
529 {
530         if (!xen_domain())
531                 return -ENODEV;
532
533         pr_info("Initialising Xen transport for 9pfs\n");
534
535         v9fs_register_trans(&p9_xen_trans);
536         return xenbus_register_frontend(&xen_9pfs_front_driver);
537 }
538 module_init(p9_trans_xen_init);
539
540 static void p9_trans_xen_exit(void)
541 {
542         v9fs_unregister_trans(&p9_xen_trans);
543         return xenbus_unregister_driver(&xen_9pfs_front_driver);
544 }
545 module_exit(p9_trans_xen_exit);