1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/fs/9p/trans_xen
7 * Copyright (C) 2017 by Stefano Stabellini <stefano@aporeto.com>
10 #include <xen/events.h>
11 #include <xen/grant_table.h>
13 #include <xen/xenbus.h>
14 #include <xen/interface/io/9pfs.h>
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>
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)
26 struct xen_9pfs_header {
31 /* uint8_t sdata[]; */
32 } __attribute__((packed));
34 /* One per ring, more than one per 9pfs share */
35 struct xen_9pfs_dataring {
36 struct xen_9pfs_front_priv *priv;
38 struct xen_9pfs_data_intf *intf;
42 /* protect a ring from concurrent accesses */
45 struct xen_9pfs_data data;
47 struct work_struct work;
50 /* One per 9pfs share */
51 struct xen_9pfs_front_priv {
52 struct list_head list;
53 struct xenbus_device *dev;
55 struct p9_client *client;
58 struct xen_9pfs_dataring *rings;
61 static LIST_HEAD(xen_9pfs_devs);
62 static DEFINE_RWLOCK(xen_9pfs_lock);
64 /* We don't currently allow canceling of requests */
65 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
70 static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
72 struct xen_9pfs_front_priv *priv;
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);
85 read_unlock(&xen_9pfs_lock);
89 static void p9_xen_close(struct p9_client *client)
91 struct xen_9pfs_front_priv *priv;
93 read_lock(&xen_9pfs_lock);
94 list_for_each_entry(priv, &xen_9pfs_devs, list) {
95 if (priv->client == client) {
97 read_unlock(&xen_9pfs_lock);
101 read_unlock(&xen_9pfs_lock);
104 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
108 cons = ring->intf->out_cons;
109 prod = ring->intf->out_prod;
112 return XEN_9PFS_RING_SIZE(ring) -
113 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size;
116 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
118 struct xen_9pfs_front_priv *priv;
119 RING_IDX cons, prod, masked_cons, masked_prod;
121 u32 size = p9_req->tc.size;
122 struct xen_9pfs_dataring *ring;
125 read_lock(&xen_9pfs_lock);
126 list_for_each_entry(priv, &xen_9pfs_devs, list) {
127 if (priv->client == client)
130 read_unlock(&xen_9pfs_lock);
131 if (list_entry_is_head(priv, &xen_9pfs_devs, list))
134 num = p9_req->tc.tag % priv->num_rings;
135 ring = &priv->rings[num];
138 while (wait_event_killable(ring->wq,
139 p9_xen_write_todo(ring, size)) != 0)
142 spin_lock_irqsave(&ring->lock, flags);
143 cons = ring->intf->out_cons;
144 prod = ring->intf->out_prod;
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);
153 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
154 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
156 xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
157 &masked_prod, masked_cons,
158 XEN_9PFS_RING_SIZE(ring));
160 WRITE_ONCE(p9_req->status, REQ_STATUS_SENT);
161 virt_wmb(); /* write ring before updating pointer */
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);
171 static void p9_xen_response(struct work_struct *work)
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;
180 ring = container_of(work, struct xen_9pfs_dataring, work);
184 cons = ring->intf->in_cons;
185 prod = ring->intf->in_prod;
188 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) <
190 notify_remote_via_irq(ring->irq);
194 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
195 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
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));
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);
207 ring->intf->in_cons = cons;
211 if (h.size > req->rc.capacity) {
212 dev_warn(&priv->dev->dev,
213 "requested packet size too big: %d for tag %d with capacity %zd\n",
214 h.size, h.tag, req->rc.capacity);
215 WRITE_ONCE(req->status, REQ_STATUS_ERROR);
219 memcpy(&req->rc, &h, sizeof(h));
222 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
223 /* Then, read the whole packet (including the header) */
224 xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
225 masked_prod, &masked_cons,
226 XEN_9PFS_RING_SIZE(ring));
231 ring->intf->in_cons = cons;
233 status = (req->status != REQ_STATUS_ERROR) ?
234 REQ_STATUS_RCVD : REQ_STATUS_ERROR;
236 p9_client_cb(priv->client, req, status);
240 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
242 struct xen_9pfs_dataring *ring = r;
244 if (!ring || !ring->priv->client) {
245 /* ignore spurious interrupt */
249 wake_up_interruptible(&ring->wq);
250 schedule_work(&ring->work);
255 static struct p9_trans_module p9_xen_trans = {
257 .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
258 .pooled_rbuffers = false,
260 .create = p9_xen_create,
261 .close = p9_xen_close,
262 .request = p9_xen_request,
263 .cancel = p9_xen_cancel,
264 .owner = THIS_MODULE,
267 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
272 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
276 write_lock(&xen_9pfs_lock);
277 list_del(&priv->list);
278 write_unlock(&xen_9pfs_lock);
280 for (i = 0; i < priv->num_rings; i++) {
281 if (!priv->rings[i].intf)
283 if (priv->rings[i].irq > 0)
284 unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
285 if (priv->rings[i].data.in) {
287 j < (1 << priv->rings[i].intf->ring_order);
291 ref = priv->rings[i].intf->ref[j];
292 gnttab_end_foreign_access(ref, NULL);
294 free_pages_exact(priv->rings[i].data.in,
295 1UL << (priv->rings[i].intf->ring_order +
298 gnttab_end_foreign_access(priv->rings[i].ref, NULL);
299 free_page((unsigned long)priv->rings[i].intf);
306 static int xen_9pfs_front_remove(struct xenbus_device *dev)
308 struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
310 dev_set_drvdata(&dev->dev, NULL);
311 xen_9pfs_front_free(priv);
315 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
316 struct xen_9pfs_dataring *ring,
323 init_waitqueue_head(&ring->wq);
324 spin_lock_init(&ring->lock);
325 INIT_WORK(&ring->work, p9_xen_response);
327 ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
330 ret = gnttab_grant_foreign_access(dev->otherend_id,
331 virt_to_gfn(ring->intf), 0);
335 bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
336 GFP_KERNEL | __GFP_ZERO);
341 for (; i < (1 << order); i++) {
342 ret = gnttab_grant_foreign_access(
343 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
346 ring->intf->ref[i] = ret;
348 ring->intf->ring_order = order;
349 ring->data.in = bytes;
350 ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
352 ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
355 ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
356 xen_9pfs_front_event_handler,
357 0, "xen_9pfs-frontend", ring);
361 xenbus_free_evtchn(dev, ring->evtchn);
365 for (i--; i >= 0; i--)
366 gnttab_end_foreign_access(ring->intf->ref[i], NULL);
367 free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
369 gnttab_end_foreign_access(ring->ref, NULL);
370 free_page((unsigned long)ring->intf);
374 static int xen_9pfs_front_init(struct xenbus_device *dev)
377 struct xenbus_transaction xbt;
378 struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
380 unsigned int max_rings, max_ring_order, len = 0;
382 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
383 if (IS_ERR(versions))
384 return PTR_ERR(versions);
385 for (v = versions; *v; v++) {
386 if (simple_strtoul(v, &v, 10) == 1) {
396 max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
397 if (max_rings < XEN_9PFS_NUM_RINGS)
399 max_ring_order = xenbus_read_unsigned(dev->otherend,
400 "max-ring-page-order", 0);
401 if (max_ring_order > XEN_9PFS_RING_ORDER)
402 max_ring_order = XEN_9PFS_RING_ORDER;
403 if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
404 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
406 priv->num_rings = XEN_9PFS_NUM_RINGS;
407 priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
414 for (i = 0; i < priv->num_rings; i++) {
415 priv->rings[i].priv = priv;
416 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
423 ret = xenbus_transaction_start(&xbt);
425 xenbus_dev_fatal(dev, ret, "starting transaction");
428 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
431 ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
435 for (i = 0; i < priv->num_rings; i++) {
438 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
439 sprintf(str, "ring-ref%d", i);
440 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
445 sprintf(str, "event-channel-%d", i);
446 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
447 priv->rings[i].evtchn);
451 priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
452 if (IS_ERR(priv->tag)) {
453 ret = PTR_ERR(priv->tag);
456 ret = xenbus_transaction_end(xbt, 0);
460 xenbus_dev_fatal(dev, ret, "completing transaction");
467 xenbus_transaction_end(xbt, 1);
468 xenbus_dev_fatal(dev, ret, "writing xenstore");
470 xen_9pfs_front_free(priv);
474 static int xen_9pfs_front_probe(struct xenbus_device *dev,
475 const struct xenbus_device_id *id)
477 struct xen_9pfs_front_priv *priv = NULL;
479 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
484 dev_set_drvdata(&dev->dev, priv);
486 write_lock(&xen_9pfs_lock);
487 list_add_tail(&priv->list, &xen_9pfs_devs);
488 write_unlock(&xen_9pfs_lock);
493 static int xen_9pfs_front_resume(struct xenbus_device *dev)
495 dev_warn(&dev->dev, "suspend/resume unsupported\n");
499 static void xen_9pfs_front_changed(struct xenbus_device *dev,
500 enum xenbus_state backend_state)
502 switch (backend_state) {
503 case XenbusStateReconfiguring:
504 case XenbusStateReconfigured:
505 case XenbusStateInitialising:
506 case XenbusStateInitialised:
507 case XenbusStateUnknown:
510 case XenbusStateInitWait:
511 if (!xen_9pfs_front_init(dev))
512 xenbus_switch_state(dev, XenbusStateInitialised);
515 case XenbusStateConnected:
516 xenbus_switch_state(dev, XenbusStateConnected);
519 case XenbusStateClosed:
520 if (dev->state == XenbusStateClosed)
522 fallthrough; /* Missed the backend's CLOSING state */
523 case XenbusStateClosing:
524 xenbus_frontend_closed(dev);
529 static struct xenbus_driver xen_9pfs_front_driver = {
530 .ids = xen_9pfs_front_ids,
531 .probe = xen_9pfs_front_probe,
532 .remove = xen_9pfs_front_remove,
533 .resume = xen_9pfs_front_resume,
534 .otherend_changed = xen_9pfs_front_changed,
537 static int __init p9_trans_xen_init(void)
544 pr_info("Initialising Xen transport for 9pfs\n");
546 v9fs_register_trans(&p9_xen_trans);
547 rc = xenbus_register_frontend(&xen_9pfs_front_driver);
549 v9fs_unregister_trans(&p9_xen_trans);
553 module_init(p9_trans_xen_init);
554 MODULE_ALIAS_9P("xen");
556 static void __exit p9_trans_xen_exit(void)
558 v9fs_unregister_trans(&p9_xen_trans);
559 return xenbus_unregister_driver(&xen_9pfs_front_driver);
561 module_exit(p9_trans_xen_exit);
563 MODULE_ALIAS("xen:9pfs");
564 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
565 MODULE_DESCRIPTION("Xen Transport for 9P");
566 MODULE_LICENSE("GPL");