net/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                 req->rc.size = h.size;
 220                 req->rc.id = h.id;
 221                 req->rc.tag = h.tag;
 222                 req->rc.offset = 0;
 223
 224                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
 225                 /* Then, read the whole packet (including the header) */
 226                 xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
 227                                      masked_prod, &masked_cons,
 228                                      XEN_9PFS_RING_SIZE(ring));
 229
 230 recv_error:
 231                 virt_mb();
 232                 cons += h.size;
 233                 ring->intf->in_cons = cons;
 234
 235                 status = (req->status != REQ_STATUS_ERROR) ?
 236                         REQ_STATUS_RCVD : REQ_STATUS_ERROR;
 237
 238                 p9_client_cb(priv->client, req, status);
 239         }
 240 }
 241
 242 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
 243 {
 244         struct xen_9pfs_dataring *ring = r;
 245
 246         if (!ring || !ring->priv->client) {
 247                 /* ignore spurious interrupt */
 248                 return IRQ_HANDLED;
 249         }
 250
 251         wake_up_interruptible(&ring->wq);
 252         schedule_work(&ring->work);
 253
 254         return IRQ_HANDLED;
 255 }
 256
 257 static struct p9_trans_module p9_xen_trans = {
 258         .name = "xen",
 259         .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
 260         .pooled_rbuffers = false,
 261         .def = 1,
 262         .create = p9_xen_create,
 263         .close = p9_xen_close,
 264         .request = p9_xen_request,
 265         .cancel = p9_xen_cancel,
 266         .owner = THIS_MODULE,
 267 };
 268
 269 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
 270         { "9pfs" },
 271         { "" }
 272 };
 273
 274 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
 275 {
 276         int i, j;
 277
 278         write_lock(&xen_9pfs_lock);
 279         list_del(&priv->list);
 280         write_unlock(&xen_9pfs_lock);
 281
 282         for (i = 0; i < priv->num_rings; i++) {
 283                 if (!priv->rings[i].intf)
 284                         break;
 285                 if (priv->rings[i].irq > 0)
 286                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
 287                 if (priv->rings[i].data.in) {
 288                         for (j = 0;
 289                              j < (1 << priv->rings[i].intf->ring_order);
 290                              j++) {
 291                                 grant_ref_t ref;
 292
 293                                 ref = priv->rings[i].intf->ref[j];
 294                                 gnttab_end_foreign_access(ref, NULL);
 295                         }
 296                         free_pages_exact(priv->rings[i].data.in,
 297                                    1UL << (priv->rings[i].intf->ring_order +
 298                                            XEN_PAGE_SHIFT));
 299                 }
 300                 gnttab_end_foreign_access(priv->rings[i].ref, NULL);
 301                 free_page((unsigned long)priv->rings[i].intf);
 302         }
 303         kfree(priv->rings);
 304         kfree(priv->tag);
 305         kfree(priv);
 306 }
 307
 308 static int xen_9pfs_front_remove(struct xenbus_device *dev)
 309 {
 310         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 311
 312         dev_set_drvdata(&dev->dev, NULL);
 313         xen_9pfs_front_free(priv);
 314         return 0;
 315 }
 316
 317 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
 318                                          struct xen_9pfs_dataring *ring,
 319                                          unsigned int order)
 320 {
 321         int i = 0;
 322         int ret = -ENOMEM;
 323         void *bytes = NULL;
 324
 325         init_waitqueue_head(&ring->wq);
 326         spin_lock_init(&ring->lock);
 327         INIT_WORK(&ring->work, p9_xen_response);
 328
 329         ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
 330         if (!ring->intf)
 331                 return ret;
 332         ret = gnttab_grant_foreign_access(dev->otherend_id,
 333                                           virt_to_gfn(ring->intf), 0);
 334         if (ret < 0)
 335                 goto out;
 336         ring->ref = ret;
 337         bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
 338                                   GFP_KERNEL | __GFP_ZERO);
 339         if (!bytes) {
 340                 ret = -ENOMEM;
 341                 goto out;
 342         }
 343         for (; i < (1 << order); i++) {
 344                 ret = gnttab_grant_foreign_access(
 345                                 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
 346                 if (ret < 0)
 347                         goto out;
 348                 ring->intf->ref[i] = ret;
 349         }
 350         ring->intf->ring_order = order;
 351         ring->data.in = bytes;
 352         ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
 353
 354         ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
 355         if (ret)
 356                 goto out;
 357         ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
 358                                               xen_9pfs_front_event_handler,
 359                                               0, "xen_9pfs-frontend", ring);
 360         if (ring->irq >= 0)
 361                 return 0;
 362
 363         xenbus_free_evtchn(dev, ring->evtchn);
 364         ret = ring->irq;
 365 out:
 366         if (bytes) {
 367                 for (i--; i >= 0; i--)
 368                         gnttab_end_foreign_access(ring->intf->ref[i], NULL);
 369                 free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
 370         }
 371         gnttab_end_foreign_access(ring->ref, NULL);
 372         free_page((unsigned long)ring->intf);
 373         return ret;
 374 }
 375
 376 static int xen_9pfs_front_init(struct xenbus_device *dev)
 377 {
 378         int ret, i;
 379         struct xenbus_transaction xbt;
 380         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 381         char *versions, *v;
 382         unsigned int max_rings, max_ring_order, len = 0;
 383
 384         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
 385         if (IS_ERR(versions))
 386                 return PTR_ERR(versions);
 387         for (v = versions; *v; v++) {
 388                 if (simple_strtoul(v, &v, 10) == 1) {
 389                         v = NULL;
 390                         break;
 391                 }
 392         }
 393         if (v) {
 394                 kfree(versions);
 395                 return -EINVAL;
 396         }
 397         kfree(versions);
 398         max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
 399         if (max_rings < XEN_9PFS_NUM_RINGS)
 400                 return -EINVAL;
 401         max_ring_order = xenbus_read_unsigned(dev->otherend,
 402                                               "max-ring-page-order", 0);
 403         if (max_ring_order > XEN_9PFS_RING_ORDER)
 404                 max_ring_order = XEN_9PFS_RING_ORDER;
 405         if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
 406                 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
 407
 408         priv->num_rings = XEN_9PFS_NUM_RINGS;
 409         priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
 410                               GFP_KERNEL);
 411         if (!priv->rings) {
 412                 kfree(priv);
 413                 return -ENOMEM;
 414         }
 415
 416         for (i = 0; i < priv->num_rings; i++) {
 417                 priv->rings[i].priv = priv;
 418                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
 419                                                     max_ring_order);
 420                 if (ret < 0)
 421                         goto error;
 422         }
 423
 424  again:
 425         ret = xenbus_transaction_start(&xbt);
 426         if (ret) {
 427                 xenbus_dev_fatal(dev, ret, "starting transaction");
 428                 goto error;
 429         }
 430         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
 431         if (ret)
 432                 goto error_xenbus;
 433         ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
 434                             priv->num_rings);
 435         if (ret)
 436                 goto error_xenbus;
 437         for (i = 0; i < priv->num_rings; i++) {
 438                 char str[16];
 439
 440                 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
 441                 sprintf(str, "ring-ref%d", i);
 442                 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
 443                                     priv->rings[i].ref);
 444                 if (ret)
 445                         goto error_xenbus;
 446
 447                 sprintf(str, "event-channel-%d", i);
 448                 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
 449                                     priv->rings[i].evtchn);
 450                 if (ret)
 451                         goto error_xenbus;
 452         }
 453         priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
 454         if (IS_ERR(priv->tag)) {
 455                 ret = PTR_ERR(priv->tag);
 456                 goto error_xenbus;
 457         }
 458         ret = xenbus_transaction_end(xbt, 0);
 459         if (ret) {
 460                 if (ret == -EAGAIN)
 461                         goto again;
 462                 xenbus_dev_fatal(dev, ret, "completing transaction");
 463                 goto error;
 464         }
 465
 466         return 0;
 467
 468  error_xenbus:
 469         xenbus_transaction_end(xbt, 1);
 470         xenbus_dev_fatal(dev, ret, "writing xenstore");
 471  error:
 472         xen_9pfs_front_free(priv);
 473         return ret;
 474 }
 475
 476 static int xen_9pfs_front_probe(struct xenbus_device *dev,
 477                                 const struct xenbus_device_id *id)
 478 {
 479         struct xen_9pfs_front_priv *priv = NULL;
 480
 481         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 482         if (!priv)
 483                 return -ENOMEM;
 484
 485         priv->dev = dev;
 486         dev_set_drvdata(&dev->dev, priv);
 487
 488         write_lock(&xen_9pfs_lock);
 489         list_add_tail(&priv->list, &xen_9pfs_devs);
 490         write_unlock(&xen_9pfs_lock);
 491
 492         return 0;
 493 }
 494
 495 static int xen_9pfs_front_resume(struct xenbus_device *dev)
 496 {
 497         dev_warn(&dev->dev, "suspend/resume unsupported\n");
 498         return 0;
 499 }
 500
 501 static void xen_9pfs_front_changed(struct xenbus_device *dev,
 502                                    enum xenbus_state backend_state)
 503 {
 504         switch (backend_state) {
 505         case XenbusStateReconfiguring:
 506         case XenbusStateReconfigured:
 507         case XenbusStateInitialising:
 508         case XenbusStateInitialised:
 509         case XenbusStateUnknown:
 510                 break;
 511
 512         case XenbusStateInitWait:
 513                 if (!xen_9pfs_front_init(dev))
 514                         xenbus_switch_state(dev, XenbusStateInitialised);
 515                 break;
 516
 517         case XenbusStateConnected:
 518                 xenbus_switch_state(dev, XenbusStateConnected);
 519                 break;
 520
 521         case XenbusStateClosed:
 522                 if (dev->state == XenbusStateClosed)
 523                         break;
 524                 fallthrough;    /* Missed the backend's CLOSING state */
 525         case XenbusStateClosing:
 526                 xenbus_frontend_closed(dev);
 527                 break;
 528         }
 529 }
 530
 531 static struct xenbus_driver xen_9pfs_front_driver = {
 532         .ids = xen_9pfs_front_ids,
 533         .probe = xen_9pfs_front_probe,
 534         .remove = xen_9pfs_front_remove,
 535         .resume = xen_9pfs_front_resume,
 536         .otherend_changed = xen_9pfs_front_changed,
 537 };
 538
 539 static int __init p9_trans_xen_init(void)
 540 {
 541         int rc;
 542
 543         if (!xen_domain())
 544                 return -ENODEV;
 545
 546         pr_info("Initialising Xen transport for 9pfs\n");
 547
 548         v9fs_register_trans(&p9_xen_trans);
 549         rc = xenbus_register_frontend(&xen_9pfs_front_driver);
 550         if (rc)
 551                 v9fs_unregister_trans(&p9_xen_trans);
 552
 553         return rc;
 554 }
 555 module_init(p9_trans_xen_init);
 556 MODULE_ALIAS_9P("xen");
 557
 558 static void __exit p9_trans_xen_exit(void)
 559 {
 560         v9fs_unregister_trans(&p9_xen_trans);
 561         return xenbus_unregister_driver(&xen_9pfs_front_driver);
 562 }
 563 module_exit(p9_trans_xen_exit);
 564
 565 MODULE_ALIAS("xen:9pfs");
 566 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
 567 MODULE_DESCRIPTION("Xen Transport for 9P");
 568 MODULE_LICENSE("GPL");