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 <net/9p/9p.h>
  42 #include <net/9p/client.h>
  43 #include <net/9p/transport.h>
  44
  45 #define XEN_9PFS_NUM_RINGS 2
  46 #define XEN_9PFS_RING_ORDER 9
  47 #define XEN_9PFS_RING_SIZE(ring)  XEN_FLEX_RING_SIZE(ring->intf->ring_order)
  48
  49 struct xen_9pfs_header {
  50         uint32_t size;
  51         uint8_t id;
  52         uint16_t tag;
  53
  54         /* uint8_t sdata[]; */
  55 } __attribute__((packed));
  56
  57 /* One per ring, more than one per 9pfs share */
  58 struct xen_9pfs_dataring {
  59         struct xen_9pfs_front_priv *priv;
  60
  61         struct xen_9pfs_data_intf *intf;
  62         grant_ref_t ref;
  63         int evtchn;
  64         int irq;
  65         /* protect a ring from concurrent accesses */
  66         spinlock_t lock;
  67
  68         struct xen_9pfs_data data;
  69         wait_queue_head_t wq;
  70         struct work_struct work;
  71 };
  72
  73 /* One per 9pfs share */
  74 struct xen_9pfs_front_priv {
  75         struct list_head list;
  76         struct xenbus_device *dev;
  77         char *tag;
  78         struct p9_client *client;
  79
  80         int num_rings;
  81         struct xen_9pfs_dataring *rings;
  82 };
  83
  84 static LIST_HEAD(xen_9pfs_devs);
  85 static DEFINE_RWLOCK(xen_9pfs_lock);
  86
  87 /* We don't currently allow canceling of requests */
  88 static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req)
  89 {
  90         return 1;
  91 }
  92
  93 static int p9_xen_create(struct p9_client *client, const char *addr, char *args)
  94 {
  95         struct xen_9pfs_front_priv *priv;
  96
  97         if (addr == NULL)
  98                 return -EINVAL;
  99
 100         read_lock(&xen_9pfs_lock);
 101         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 102                 if (!strcmp(priv->tag, addr)) {
 103                         priv->client = client;
 104                         read_unlock(&xen_9pfs_lock);
 105                         return 0;
 106                 }
 107         }
 108         read_unlock(&xen_9pfs_lock);
 109         return -EINVAL;
 110 }
 111
 112 static void p9_xen_close(struct p9_client *client)
 113 {
 114         struct xen_9pfs_front_priv *priv;
 115
 116         read_lock(&xen_9pfs_lock);
 117         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 118                 if (priv->client == client) {
 119                         priv->client = NULL;
 120                         read_unlock(&xen_9pfs_lock);
 121                         return;
 122                 }
 123         }
 124         read_unlock(&xen_9pfs_lock);
 125 }
 126
 127 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
 128 {
 129         RING_IDX cons, prod;
 130
 131         cons = ring->intf->out_cons;
 132         prod = ring->intf->out_prod;
 133         virt_mb();
 134
 135         return XEN_9PFS_RING_SIZE(ring) -
 136                 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size;
 137 }
 138
 139 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
 140 {
 141         struct xen_9pfs_front_priv *priv;
 142         RING_IDX cons, prod, masked_cons, masked_prod;
 143         unsigned long flags;
 144         u32 size = p9_req->tc.size;
 145         struct xen_9pfs_dataring *ring;
 146         int num;
 147
 148         read_lock(&xen_9pfs_lock);
 149         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 150                 if (priv->client == client)
 151                         break;
 152         }
 153         read_unlock(&xen_9pfs_lock);
 154         if (list_entry_is_head(priv, &xen_9pfs_devs, list))
 155                 return -EINVAL;
 156
 157         num = p9_req->tc.tag % priv->num_rings;
 158         ring = &priv->rings[num];
 159
 160 again:
 161         while (wait_event_killable(ring->wq,
 162                                    p9_xen_write_todo(ring, size)) != 0)
 163                 ;
 164
 165         spin_lock_irqsave(&ring->lock, flags);
 166         cons = ring->intf->out_cons;
 167         prod = ring->intf->out_prod;
 168         virt_mb();
 169
 170         if (XEN_9PFS_RING_SIZE(ring) -
 171             xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) < size) {
 172                 spin_unlock_irqrestore(&ring->lock, flags);
 173                 goto again;
 174         }
 175
 176         masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
 177         masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
 178
 179         xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
 180                               &masked_prod, masked_cons,
 181                               XEN_9PFS_RING_SIZE(ring));
 182
 183         p9_req->status = REQ_STATUS_SENT;
 184         virt_wmb();                     /* write ring before updating pointer */
 185         prod += size;
 186         ring->intf->out_prod = prod;
 187         spin_unlock_irqrestore(&ring->lock, flags);
 188         notify_remote_via_irq(ring->irq);
 189         p9_req_put(client, p9_req);
 190
 191         return 0;
 192 }
 193
 194 static void p9_xen_response(struct work_struct *work)
 195 {
 196         struct xen_9pfs_front_priv *priv;
 197         struct xen_9pfs_dataring *ring;
 198         RING_IDX cons, prod, masked_cons, masked_prod;
 199         struct xen_9pfs_header h;
 200         struct p9_req_t *req;
 201         int status;
 202
 203         ring = container_of(work, struct xen_9pfs_dataring, work);
 204         priv = ring->priv;
 205
 206         while (1) {
 207                 cons = ring->intf->in_cons;
 208                 prod = ring->intf->in_prod;
 209                 virt_rmb();
 210
 211                 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) <
 212                     sizeof(h)) {
 213                         notify_remote_via_irq(ring->irq);
 214                         return;
 215                 }
 216
 217                 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring));
 218                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
 219
 220                 /* First, read just the header */
 221                 xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
 222                                      masked_prod, &masked_cons,
 223                                      XEN_9PFS_RING_SIZE(ring));
 224
 225                 req = p9_tag_lookup(priv->client, h.tag);
 226                 if (!req || req->status != REQ_STATUS_SENT) {
 227                         dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
 228                         cons += h.size;
 229                         virt_mb();
 230                         ring->intf->in_cons = cons;
 231                         continue;
 232                 }
 233
 234                 if (h.size > req->rc.capacity) {
 235                         dev_warn(&priv->dev->dev,
 236                                  "requested packet size too big: %d for tag %d with capacity %zd\n",
 237                                  h.size, h.tag, req->rc.capacity);
 238                         req->status = REQ_STATUS_ERROR;
 239                         goto recv_error;
 240                 }
 241
 242                 memcpy(&req->rc, &h, sizeof(h));
 243                 req->rc.offset = 0;
 244
 245                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring));
 246                 /* Then, read the whole packet (including the header) */
 247                 xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
 248                                      masked_prod, &masked_cons,
 249                                      XEN_9PFS_RING_SIZE(ring));
 250
 251 recv_error:
 252                 virt_mb();
 253                 cons += h.size;
 254                 ring->intf->in_cons = cons;
 255
 256                 status = (req->status != REQ_STATUS_ERROR) ?
 257                         REQ_STATUS_RCVD : REQ_STATUS_ERROR;
 258
 259                 p9_client_cb(priv->client, req, status);
 260         }
 261 }
 262
 263 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
 264 {
 265         struct xen_9pfs_dataring *ring = r;
 266
 267         if (!ring || !ring->priv->client) {
 268                 /* ignore spurious interrupt */
 269                 return IRQ_HANDLED;
 270         }
 271
 272         wake_up_interruptible(&ring->wq);
 273         schedule_work(&ring->work);
 274
 275         return IRQ_HANDLED;
 276 }
 277
 278 static struct p9_trans_module p9_xen_trans = {
 279         .name = "xen",
 280         .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
 281         .def = 1,
 282         .create = p9_xen_create,
 283         .close = p9_xen_close,
 284         .request = p9_xen_request,
 285         .cancel = p9_xen_cancel,
 286         .owner = THIS_MODULE,
 287 };
 288
 289 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
 290         { "9pfs" },
 291         { "" }
 292 };
 293
 294 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
 295 {
 296         int i, j;
 297
 298         write_lock(&xen_9pfs_lock);
 299         list_del(&priv->list);
 300         write_unlock(&xen_9pfs_lock);
 301
 302         for (i = 0; i < priv->num_rings; i++) {
 303                 if (!priv->rings[i].intf)
 304                         break;
 305                 if (priv->rings[i].irq > 0)
 306                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
 307                 if (priv->rings[i].data.in) {
 308                         for (j = 0;
 309                              j < (1 << priv->rings[i].intf->ring_order);
 310                              j++) {
 311                                 grant_ref_t ref;
 312
 313                                 ref = priv->rings[i].intf->ref[j];
 314                                 gnttab_end_foreign_access(ref, 0, 0);
 315                         }
 316                         free_pages_exact(priv->rings[i].data.in,
 317                                    1UL << (priv->rings[i].intf->ring_order +
 318                                            XEN_PAGE_SHIFT));
 319                 }
 320                 gnttab_end_foreign_access(priv->rings[i].ref, 0, 0);
 321                 free_page((unsigned long)priv->rings[i].intf);
 322         }
 323         kfree(priv->rings);
 324         kfree(priv->tag);
 325         kfree(priv);
 326 }
 327
 328 static int xen_9pfs_front_remove(struct xenbus_device *dev)
 329 {
 330         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 331
 332         dev_set_drvdata(&dev->dev, NULL);
 333         xen_9pfs_front_free(priv);
 334         return 0;
 335 }
 336
 337 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
 338                                          struct xen_9pfs_dataring *ring,
 339                                          unsigned int order)
 340 {
 341         int i = 0;
 342         int ret = -ENOMEM;
 343         void *bytes = NULL;
 344
 345         init_waitqueue_head(&ring->wq);
 346         spin_lock_init(&ring->lock);
 347         INIT_WORK(&ring->work, p9_xen_response);
 348
 349         ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
 350         if (!ring->intf)
 351                 return ret;
 352         ret = gnttab_grant_foreign_access(dev->otherend_id,
 353                                           virt_to_gfn(ring->intf), 0);
 354         if (ret < 0)
 355                 goto out;
 356         ring->ref = ret;
 357         bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT),
 358                                   GFP_KERNEL | __GFP_ZERO);
 359         if (!bytes) {
 360                 ret = -ENOMEM;
 361                 goto out;
 362         }
 363         for (; i < (1 << order); i++) {
 364                 ret = gnttab_grant_foreign_access(
 365                                 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
 366                 if (ret < 0)
 367                         goto out;
 368                 ring->intf->ref[i] = ret;
 369         }
 370         ring->intf->ring_order = order;
 371         ring->data.in = bytes;
 372         ring->data.out = bytes + XEN_FLEX_RING_SIZE(order);
 373
 374         ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
 375         if (ret)
 376                 goto out;
 377         ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
 378                                               xen_9pfs_front_event_handler,
 379                                               0, "xen_9pfs-frontend", ring);
 380         if (ring->irq >= 0)
 381                 return 0;
 382
 383         xenbus_free_evtchn(dev, ring->evtchn);
 384         ret = ring->irq;
 385 out:
 386         if (bytes) {
 387                 for (i--; i >= 0; i--)
 388                         gnttab_end_foreign_access(ring->intf->ref[i], 0, 0);
 389                 free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT));
 390         }
 391         gnttab_end_foreign_access(ring->ref, 0, 0);
 392         free_page((unsigned long)ring->intf);
 393         return ret;
 394 }
 395
 396 static int xen_9pfs_front_probe(struct xenbus_device *dev,
 397                                 const struct xenbus_device_id *id)
 398 {
 399         int ret, i;
 400         struct xenbus_transaction xbt;
 401         struct xen_9pfs_front_priv *priv = NULL;
 402         char *versions;
 403         unsigned int max_rings, max_ring_order, len = 0;
 404
 405         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
 406         if (IS_ERR(versions))
 407                 return PTR_ERR(versions);
 408         if (strcmp(versions, "1")) {
 409                 kfree(versions);
 410                 return -EINVAL;
 411         }
 412         kfree(versions);
 413         max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
 414         if (max_rings < XEN_9PFS_NUM_RINGS)
 415                 return -EINVAL;
 416         max_ring_order = xenbus_read_unsigned(dev->otherend,
 417                                               "max-ring-page-order", 0);
 418         if (max_ring_order > XEN_9PFS_RING_ORDER)
 419                 max_ring_order = XEN_9PFS_RING_ORDER;
 420         if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order))
 421                 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2;
 422
 423         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 424         if (!priv)
 425                 return -ENOMEM;
 426
 427         priv->dev = dev;
 428         priv->num_rings = XEN_9PFS_NUM_RINGS;
 429         priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
 430                               GFP_KERNEL);
 431         if (!priv->rings) {
 432                 kfree(priv);
 433                 return -ENOMEM;
 434         }
 435
 436         for (i = 0; i < priv->num_rings; i++) {
 437                 priv->rings[i].priv = priv;
 438                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i],
 439                                                     max_ring_order);
 440                 if (ret < 0)
 441                         goto error;
 442         }
 443
 444  again:
 445         ret = xenbus_transaction_start(&xbt);
 446         if (ret) {
 447                 xenbus_dev_fatal(dev, ret, "starting transaction");
 448                 goto error;
 449         }
 450         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
 451         if (ret)
 452                 goto error_xenbus;
 453         ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
 454                             priv->num_rings);
 455         if (ret)
 456                 goto error_xenbus;
 457         for (i = 0; i < priv->num_rings; i++) {
 458                 char str[16];
 459
 460                 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
 461                 sprintf(str, "ring-ref%d", i);
 462                 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
 463                                     priv->rings[i].ref);
 464                 if (ret)
 465                         goto error_xenbus;
 466
 467                 sprintf(str, "event-channel-%d", i);
 468                 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
 469                                     priv->rings[i].evtchn);
 470                 if (ret)
 471                         goto error_xenbus;
 472         }
 473         priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
 474         if (IS_ERR(priv->tag)) {
 475                 ret = PTR_ERR(priv->tag);
 476                 goto error_xenbus;
 477         }
 478         ret = xenbus_transaction_end(xbt, 0);
 479         if (ret) {
 480                 if (ret == -EAGAIN)
 481                         goto again;
 482                 xenbus_dev_fatal(dev, ret, "completing transaction");
 483                 goto error;
 484         }
 485
 486         write_lock(&xen_9pfs_lock);
 487         list_add_tail(&priv->list, &xen_9pfs_devs);
 488         write_unlock(&xen_9pfs_lock);
 489         dev_set_drvdata(&dev->dev, priv);
 490         xenbus_switch_state(dev, XenbusStateInitialised);
 491
 492         return 0;
 493
 494  error_xenbus:
 495         xenbus_transaction_end(xbt, 1);
 496         xenbus_dev_fatal(dev, ret, "writing xenstore");
 497  error:
 498         dev_set_drvdata(&dev->dev, NULL);
 499         xen_9pfs_front_free(priv);
 500         return ret;
 501 }
 502
 503 static int xen_9pfs_front_resume(struct xenbus_device *dev)
 504 {
 505         dev_warn(&dev->dev, "suspend/resume unsupported\n");
 506         return 0;
 507 }
 508
 509 static void xen_9pfs_front_changed(struct xenbus_device *dev,
 510                                    enum xenbus_state backend_state)
 511 {
 512         switch (backend_state) {
 513         case XenbusStateReconfiguring:
 514         case XenbusStateReconfigured:
 515         case XenbusStateInitialising:
 516         case XenbusStateInitialised:
 517         case XenbusStateUnknown:
 518                 break;
 519
 520         case XenbusStateInitWait:
 521                 break;
 522
 523         case XenbusStateConnected:
 524                 xenbus_switch_state(dev, XenbusStateConnected);
 525                 break;
 526
 527         case XenbusStateClosed:
 528                 if (dev->state == XenbusStateClosed)
 529                         break;
 530                 fallthrough;    /* Missed the backend's CLOSING state */
 531         case XenbusStateClosing:
 532                 xenbus_frontend_closed(dev);
 533                 break;
 534         }
 535 }
 536
 537 static struct xenbus_driver xen_9pfs_front_driver = {
 538         .ids = xen_9pfs_front_ids,
 539         .probe = xen_9pfs_front_probe,
 540         .remove = xen_9pfs_front_remove,
 541         .resume = xen_9pfs_front_resume,
 542         .otherend_changed = xen_9pfs_front_changed,
 543 };
 544
 545 static int p9_trans_xen_init(void)
 546 {
 547         int rc;
 548
 549         if (!xen_domain())
 550                 return -ENODEV;
 551
 552         pr_info("Initialising Xen transport for 9pfs\n");
 553
 554         v9fs_register_trans(&p9_xen_trans);
 555         rc = xenbus_register_frontend(&xen_9pfs_front_driver);
 556         if (rc)
 557                 v9fs_unregister_trans(&p9_xen_trans);
 558
 559         return rc;
 560 }
 561 module_init(p9_trans_xen_init);
 562
 563 static void p9_trans_xen_exit(void)
 564 {
 565         v9fs_unregister_trans(&p9_xen_trans);
 566         return xenbus_unregister_driver(&xen_9pfs_front_driver);
 567 }
 568 module_exit(p9_trans_xen_exit);
 569
 570 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
 571 MODULE_DESCRIPTION("Xen Transport for 9P");
 572 MODULE_LICENSE("GPL");