9p/trans_xen.c

   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
  26 struct 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 */
  35 struct 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 */
  51 struct xen_9pfs_front_priv {
  52         struct list_head list;
  53         struct xenbus_device *dev;
  54         char *tag;
  55         struct p9_client *client;
  56
  57         int num_rings;
  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 */
  65 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
  66 {
  67         return 1;
  68 }
  69
  70 static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
  71 {
  72         struct xen_9pfs_front_priv *priv;
  73
  74         if (addr == NULL)
  75                 return -EINVAL;
  76
  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);
  82                         return 0;
  83                 }
  84         }
  85         read_unlock(&xen_9pfs_lock);
  86         return -EINVAL;
  87 }
  88
  89 static void p9_xen_close(struct p9_client *client)
  90 {
  91         struct xen_9pfs_front_priv *priv;
  92
  93         read_lock(&xen_9pfs_lock);
  94         list_for_each_entry(priv, &xen_9pfs_devs, list) {
  95                 if (priv->client == client) {
  96                         priv->client = NULL;
  97                         read_unlock(&xen_9pfs_lock);
  98                         return;
  99                 }
 100         }
 101         read_unlock(&xen_9pfs_lock);
 102 }
 103
 104 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
 105 {
 106         RING_IDX cons, prod;
 107
 108         cons = ring->intf->out_cons;
 109         prod = ring->intf->out_prod;
 110         virt_mb();
 111
 112         return XEN_9PFS_RING_SIZE(ring) -
 113                 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size;
 114 }
 115
 116 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
 117 {
 118         struct xen_9pfs_front_priv *priv;
 119         RING_IDX cons, prod, masked_cons, masked_prod;
 120         unsigned long flags;
 121         u32 size = p9_req->tc.size;
 122         struct xen_9pfs_dataring *ring;
 123         int num;
 124
 125         read_lock(&xen_9pfs_lock);
 126         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 127                 if (priv->client == client)
 128                         break;
 129         }
 130         read_unlock(&xen_9pfs_lock);
 131         if (list_entry_is_head(priv, &xen_9pfs_devs, list))
 132                 return -EINVAL;
 133
 134         num = p9_req->tc.tag % priv->num_rings;
 135         ring = &priv->rings[num];
 136
 137 again:
 138         while (wait_event_killable(ring->wq,
 139                                    p9_xen_write_todo(ring, size)) != 0)
 140                 ;
 141
 142         spin_lock_irqsave(&ring->lock, flags);
 143         cons = ring->intf->out_cons;
 144         prod = ring->intf->out_prod;
 145         virt_mb();
 146
 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);
 150                 goto again;
 151         }
 152
 153         masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
 154         masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
 155
 156         xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
 157                               &masked_prod, masked_cons,
 158                               XEN_9PFS_RING_SIZE(ring));
 159
 160         WRITE_ONCE(p9_req->status, REQ_STATUS_SENT);
 161         virt_wmb();                     /* write ring before updating pointer */
 162         prod += size;
 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);
 167
 168         return 0;
 169 }
 170
 171 static void p9_xen_response(struct work_struct *work)
 172 {
 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;
 178         int status;
 179
 180         ring = container_of(work, struct xen_9pfs_dataring, work);
 181         priv = ring->priv;
 182
 183         while (1) {
 184                 cons = ring->intf->in_cons;
 185                 prod = ring->intf->in_prod;
 186                 virt_rmb();
 187
 188                 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) <
 189                     sizeof(h)) {
 190                         notify_remote_via_irq(ring->irq);
 191                         return;
 192                 }
 193
 194                 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
 195                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
 196
 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));
 201
 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);
 205                         cons += h.size;
 206                         virt_mb();
 207                         ring->intf->in_cons = cons;
 208                         continue;
 209                 }
 210
 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);
 216                         goto recv_error;
 217                 }
 218
 219                 memcpy(&req->rc, &h, sizeof(h));
 220                 req->rc.offset = 0;
 221
 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));
 227
 228 recv_error:
 229                 virt_mb();
 230                 cons += h.size;
 231                 ring->intf->in_cons = cons;
 232
 233                 status = (req->status != REQ_STATUS_ERROR) ?
 234                         REQ_STATUS_RCVD : REQ_STATUS_ERROR;
 235
 236                 p9_client_cb(priv->client, req, status);
 237         }
 238 }
 239
 240 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
 241 {
 242         struct xen_9pfs_dataring *ring = r;
 243
 244         if (!ring || !ring->priv->client) {
 245                 /* ignore spurious interrupt */
 246                 return IRQ_HANDLED;
 247         }
 248
 249         wake_up_interruptible(&ring->wq);
 250         schedule_work(&ring->work);
 251
 252         return IRQ_HANDLED;
 253 }
 254
 255 static struct p9_trans_module p9_xen_trans = {
 256         .name = "xen",
 257         .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
 258         .pooled_rbuffers = false,
 259         .def = 1,
 260         .create = p9_xen_create,
 261         .close = p9_xen_close,
 262         .request = p9_xen_request,
 263         .cancel = p9_xen_cancel,
 264         .owner = THIS_MODULE,
 265 };
 266
 267 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
 268         { "9pfs" },
 269         { "" }
 270 };
 271
 272 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
 273 {
 274         int i, j;
 275
 276         write_lock(&xen_9pfs_lock);
 277         list_del(&priv->list);
 278         write_unlock(&xen_9pfs_lock);
 279
 280         for (i = 0; i < priv->num_rings; i++) {
 281                 struct xen_9pfs_dataring *ring = &priv->rings[i];
 282
 283                 cancel_work_sync(&ring->work);
 284
 285                 if (!priv->rings[i].intf)
 286                         break;
 287                 if (priv->rings[i].irq > 0)
 288                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
 289                 if (priv->rings[i].data.in) {
 290                         for (j = 0;
 291                              j < (1 << priv->rings[i].intf->ring_order);
 292                              j++) {
 293                                 grant_ref_t ref;
 294
 295                                 ref = priv->rings[i].intf->ref[j];
 296                                 gnttab_end_foreign_access(ref, NULL);
 297                         }
 298                         free_pages_exact(priv->rings[i].data.in,
 299                                    1UL << (priv->rings[i].intf->ring_order +
 300                                            XEN_PAGE_SHIFT));
 301                 }
 302                 gnttab_end_foreign_access(priv->rings[i].ref, NULL);
 303                 free_page((unsigned long)priv->rings[i].intf);
 304         }
 305         kfree(priv->rings);
 306         kfree(priv->tag);
 307         kfree(priv);
 308 }
 309
 310 static int xen_9pfs_front_remove(struct xenbus_device *dev)
 311 {
 312         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 313
 314         dev_set_drvdata(&dev->dev, NULL);
 315         xen_9pfs_front_free(priv);
 316         return 0;
 317 }
 318
 319 static 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;
 367 out:
 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
 378 static 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->num_rings = XEN_9PFS_NUM_RINGS;
 411         priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
 412                               GFP_KERNEL);
 413         if (!priv->rings) {
 414                 kfree(priv);
 415                 return -ENOMEM;
 416         }
 417
 418         for (i = 0; i < priv->num_rings; i++) {
 419                 priv->rings[i].priv = priv;
 420                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
 421                                                     max_ring_order);
 422                 if (ret < 0)
 423                         goto error;
 424         }
 425
 426  again:
 427         ret = xenbus_transaction_start(&xbt);
 428         if (ret) {
 429                 xenbus_dev_fatal(dev, ret, "starting transaction");
 430                 goto error;
 431         }
 432         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
 433         if (ret)
 434                 goto error_xenbus;
 435         ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
 436                             priv->num_rings);
 437         if (ret)
 438                 goto error_xenbus;
 439         for (i = 0; i < priv->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
 478 static 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
 497 static int xen_9pfs_front_resume(struct xenbus_device *dev)
 498 {
 499         dev_warn(&dev->dev, "suspend/resume unsupported\n");
 500         return 0;
 501 }
 502
 503 static 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
 533 static 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
 541 static 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 }
 557 module_init(p9_trans_xen_init);
 558 MODULE_ALIAS_9P("xen");
 559
 560 static 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 }
 565 module_exit(p9_trans_xen_exit);
 566
 567 MODULE_ALIAS("xen:9pfs");
 568 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
 569 MODULE_DESCRIPTION("Xen Transport for 9P");
 570 MODULE_LICENSE("GPL");