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         if (addr == NULL)
  99                 return -EINVAL;
 100
 101         read_lock(&xen_9pfs_lock);
 102         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 103                 if (!strcmp(priv->tag, addr)) {
 104                         priv->client = client;
 105                         read_unlock(&xen_9pfs_lock);
 106                         return 0;
 107                 }
 108         }
 109         read_unlock(&xen_9pfs_lock);
 110         return -EINVAL;
 111 }
 112
 113 static void p9_xen_close(struct p9_client *client)
 114 {
 115         struct xen_9pfs_front_priv *priv;
 116
 117         read_lock(&xen_9pfs_lock);
 118         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 119                 if (priv->client == client) {
 120                         priv->client = NULL;
 121                         read_unlock(&xen_9pfs_lock);
 122                         return;
 123                 }
 124         }
 125         read_unlock(&xen_9pfs_lock);
 126 }
 127
 128 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size)
 129 {
 130         RING_IDX cons, prod;
 131
 132         cons = ring->intf->out_cons;
 133         prod = ring->intf->out_prod;
 134         virt_mb();
 135
 136         return XEN_9PFS_RING_SIZE -
 137                 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) >= size;
 138 }
 139
 140 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req)
 141 {
 142         struct xen_9pfs_front_priv *priv;
 143         RING_IDX cons, prod, masked_cons, masked_prod;
 144         unsigned long flags;
 145         u32 size = p9_req->tc->size;
 146         struct xen_9pfs_dataring *ring;
 147         int num;
 148
 149         read_lock(&xen_9pfs_lock);
 150         list_for_each_entry(priv, &xen_9pfs_devs, list) {
 151                 if (priv->client == client)
 152                         break;
 153         }
 154         read_unlock(&xen_9pfs_lock);
 155         if (list_entry_is_head(priv, &xen_9pfs_devs, list))
 156                 return -EINVAL;
 157
 158         num = p9_req->tc->tag % priv->num_rings;
 159         ring = &priv->rings[num];
 160
 161 again:
 162         while (wait_event_killable(ring->wq,
 163                                    p9_xen_write_todo(ring, size)) != 0)
 164                 ;
 165
 166         spin_lock_irqsave(&ring->lock, flags);
 167         cons = ring->intf->out_cons;
 168         prod = ring->intf->out_prod;
 169         virt_mb();
 170
 171         if (XEN_9PFS_RING_SIZE - xen_9pfs_queued(prod, cons,
 172                                                  XEN_9PFS_RING_SIZE) < size) {
 173                 spin_unlock_irqrestore(&ring->lock, flags);
 174                 goto again;
 175         }
 176
 177         masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
 178         masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
 179
 180         xen_9pfs_write_packet(ring->data.out, p9_req->tc->sdata, size,
 181                               &masked_prod, masked_cons, XEN_9PFS_RING_SIZE);
 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
 190         return 0;
 191 }
 192
 193 static void p9_xen_response(struct work_struct *work)
 194 {
 195         struct xen_9pfs_front_priv *priv;
 196         struct xen_9pfs_dataring *ring;
 197         RING_IDX cons, prod, masked_cons, masked_prod;
 198         struct xen_9pfs_header h;
 199         struct p9_req_t *req;
 200         int status;
 201
 202         ring = container_of(work, struct xen_9pfs_dataring, work);
 203         priv = ring->priv;
 204
 205         while (1) {
 206                 cons = ring->intf->in_cons;
 207                 prod = ring->intf->in_prod;
 208                 virt_rmb();
 209
 210                 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) <
 211                     sizeof(h)) {
 212                         notify_remote_via_irq(ring->irq);
 213                         return;
 214                 }
 215
 216                 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
 217                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
 218
 219                 /* First, read just the header */
 220                 xen_9pfs_read_packet(&h, ring->data.in, sizeof(h),
 221                                      masked_prod, &masked_cons,
 222                                      XEN_9PFS_RING_SIZE);
 223
 224                 req = p9_tag_lookup(priv->client, h.tag);
 225                 if (!req || req->status != REQ_STATUS_SENT) {
 226                         dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag);
 227                         cons += h.size;
 228                         virt_mb();
 229                         ring->intf->in_cons = cons;
 230                         continue;
 231                 }
 232
 233                 memcpy(req->rc, &h, sizeof(h));
 234                 req->rc->offset = 0;
 235
 236                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
 237                 /* Then, read the whole packet (including the header) */
 238                 xen_9pfs_read_packet(req->rc->sdata, ring->data.in, h.size,
 239                                      masked_prod, &masked_cons,
 240                                      XEN_9PFS_RING_SIZE);
 241
 242                 virt_mb();
 243                 cons += h.size;
 244                 ring->intf->in_cons = cons;
 245
 246                 status = (req->status != REQ_STATUS_ERROR) ?
 247                         REQ_STATUS_RCVD : REQ_STATUS_ERROR;
 248
 249                 p9_client_cb(priv->client, req, status);
 250         }
 251 }
 252
 253 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
 254 {
 255         struct xen_9pfs_dataring *ring = r;
 256
 257         if (!ring || !ring->priv->client) {
 258                 /* ignore spurious interrupt */
 259                 return IRQ_HANDLED;
 260         }
 261
 262         wake_up_interruptible(&ring->wq);
 263         schedule_work(&ring->work);
 264
 265         return IRQ_HANDLED;
 266 }
 267
 268 static struct p9_trans_module p9_xen_trans = {
 269         .name = "xen",
 270         .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
 271         .def = 1,
 272         .create = p9_xen_create,
 273         .close = p9_xen_close,
 274         .request = p9_xen_request,
 275         .cancel = p9_xen_cancel,
 276         .owner = THIS_MODULE,
 277 };
 278
 279 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
 280         { "9pfs" },
 281         { "" }
 282 };
 283
 284 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
 285 {
 286         int i, j;
 287
 288         write_lock(&xen_9pfs_lock);
 289         list_del(&priv->list);
 290         write_unlock(&xen_9pfs_lock);
 291
 292         for (i = 0; i < priv->num_rings; i++) {
 293                 struct xen_9pfs_dataring *ring = &priv->rings[i];
 294
 295                 cancel_work_sync(&ring->work);
 296
 297                 if (!priv->rings[i].intf)
 298                         break;
 299                 if (priv->rings[i].irq > 0)
 300                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
 301                 if (priv->rings[i].data.in) {
 302                         for (j = 0; j < (1 << XEN_9PFS_RING_ORDER); j++) {
 303                                 grant_ref_t ref;
 304
 305                                 ref = priv->rings[i].intf->ref[j];
 306                                 gnttab_end_foreign_access(ref, 0, 0);
 307                         }
 308                         free_pages_exact(priv->rings[i].data.in,
 309                                    1UL << (XEN_9PFS_RING_ORDER +
 310                                            XEN_PAGE_SHIFT));
 311                 }
 312                 gnttab_end_foreign_access(priv->rings[i].ref, 0, 0);
 313                 free_page((unsigned long)priv->rings[i].intf);
 314         }
 315         kfree(priv->rings);
 316         kfree(priv->tag);
 317         kfree(priv);
 318 }
 319
 320 static int xen_9pfs_front_remove(struct xenbus_device *dev)
 321 {
 322         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 323
 324         dev_set_drvdata(&dev->dev, NULL);
 325         xen_9pfs_front_free(priv);
 326         return 0;
 327 }
 328
 329 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
 330                                          struct xen_9pfs_dataring *ring)
 331 {
 332         int i = 0;
 333         int ret = -ENOMEM;
 334         void *bytes = NULL;
 335
 336         init_waitqueue_head(&ring->wq);
 337         spin_lock_init(&ring->lock);
 338         INIT_WORK(&ring->work, p9_xen_response);
 339
 340         ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
 341         if (!ring->intf)
 342                 return ret;
 343         ret = gnttab_grant_foreign_access(dev->otherend_id,
 344                                           virt_to_gfn(ring->intf), 0);
 345         if (ret < 0)
 346                 goto out;
 347         ring->ref = ret;
 348         bytes = alloc_pages_exact(1UL << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
 349                                   GFP_KERNEL | __GFP_ZERO);
 350         if (!bytes) {
 351                 ret = -ENOMEM;
 352                 goto out;
 353         }
 354         for (; i < (1 << XEN_9PFS_RING_ORDER); i++) {
 355                 ret = gnttab_grant_foreign_access(
 356                                 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
 357                 if (ret < 0)
 358                         goto out;
 359                 ring->intf->ref[i] = ret;
 360         }
 361         ring->intf->ring_order = XEN_9PFS_RING_ORDER;
 362         ring->data.in = bytes;
 363         ring->data.out = bytes + XEN_9PFS_RING_SIZE;
 364
 365         ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
 366         if (ret)
 367                 goto out;
 368         ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
 369                                               xen_9pfs_front_event_handler,
 370                                               0, "xen_9pfs-frontend", ring);
 371         if (ring->irq >= 0)
 372                 return 0;
 373
 374         xenbus_free_evtchn(dev, ring->evtchn);
 375         ret = ring->irq;
 376 out:
 377         if (bytes) {
 378                 for (i--; i >= 0; i--)
 379                         gnttab_end_foreign_access(ring->intf->ref[i], 0, 0);
 380                 free_pages_exact(bytes, 1UL << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT));
 381         }
 382         gnttab_end_foreign_access(ring->ref, 0, 0);
 383         free_page((unsigned long)ring->intf);
 384         return ret;
 385 }
 386
 387 static int xen_9pfs_front_init(struct xenbus_device *dev)
 388 {
 389         int ret, i;
 390         struct xenbus_transaction xbt;
 391         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 392         char *versions, *v;
 393         unsigned int max_rings, max_ring_order, len = 0;
 394
 395         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
 396         if (IS_ERR(versions))
 397                 return PTR_ERR(versions);
 398         for (v = versions; *v; v++) {
 399                 if (simple_strtoul(v, &v, 10) == 1) {
 400                         v = NULL;
 401                         break;
 402                 }
 403         }
 404         if (v) {
 405                 kfree(versions);
 406                 return -EINVAL;
 407         }
 408         kfree(versions);
 409         max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
 410         if (max_rings < XEN_9PFS_NUM_RINGS)
 411                 return -EINVAL;
 412         max_ring_order = xenbus_read_unsigned(dev->otherend,
 413                                               "max-ring-page-order", 0);
 414         if (max_ring_order < XEN_9PFS_RING_ORDER)
 415                 return -EINVAL;
 416
 417         priv->num_rings = XEN_9PFS_NUM_RINGS;
 418         priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
 419                               GFP_KERNEL);
 420         if (!priv->rings) {
 421                 kfree(priv);
 422                 return -ENOMEM;
 423         }
 424
 425         for (i = 0; i < priv->num_rings; i++) {
 426                 priv->rings[i].priv = priv;
 427                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i]);
 428                 if (ret < 0)
 429                         goto error;
 430         }
 431
 432  again:
 433         ret = xenbus_transaction_start(&xbt);
 434         if (ret) {
 435                 xenbus_dev_fatal(dev, ret, "starting transaction");
 436                 goto error;
 437         }
 438         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
 439         if (ret)
 440                 goto error_xenbus;
 441         ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
 442                             priv->num_rings);
 443         if (ret)
 444                 goto error_xenbus;
 445         for (i = 0; i < priv->num_rings; i++) {
 446                 char str[16];
 447
 448                 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
 449                 sprintf(str, "ring-ref%u", i);
 450                 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
 451                                     priv->rings[i].ref);
 452                 if (ret)
 453                         goto error_xenbus;
 454
 455                 sprintf(str, "event-channel-%u", i);
 456                 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
 457                                     priv->rings[i].evtchn);
 458                 if (ret)
 459                         goto error_xenbus;
 460         }
 461         priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
 462         if (IS_ERR(priv->tag)) {
 463                 ret = PTR_ERR(priv->tag);
 464                 goto error_xenbus;
 465         }
 466         ret = xenbus_transaction_end(xbt, 0);
 467         if (ret) {
 468                 if (ret == -EAGAIN)
 469                         goto again;
 470                 xenbus_dev_fatal(dev, ret, "completing transaction");
 471                 goto error;
 472         }
 473
 474         return 0;
 475
 476  error_xenbus:
 477         xenbus_transaction_end(xbt, 1);
 478         xenbus_dev_fatal(dev, ret, "writing xenstore");
 479  error:
 480         xen_9pfs_front_free(priv);
 481         return ret;
 482 }
 483
 484 static int xen_9pfs_front_probe(struct xenbus_device *dev,
 485                                 const struct xenbus_device_id *id)
 486 {
 487         struct xen_9pfs_front_priv *priv = NULL;
 488
 489         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 490         if (!priv)
 491                 return -ENOMEM;
 492
 493         priv->dev = dev;
 494         dev_set_drvdata(&dev->dev, priv);
 495
 496         write_lock(&xen_9pfs_lock);
 497         list_add_tail(&priv->list, &xen_9pfs_devs);
 498         write_unlock(&xen_9pfs_lock);
 499
 500         return 0;
 501 }
 502
 503 static int xen_9pfs_front_resume(struct xenbus_device *dev)
 504 {
 505         dev_warn(&dev->dev, "suspsend/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                 if (!xen_9pfs_front_init(dev))
 522                         xenbus_switch_state(dev, XenbusStateInitialised);
 523                 break;
 524
 525         case XenbusStateConnected:
 526                 xenbus_switch_state(dev, XenbusStateConnected);
 527                 break;
 528
 529         case XenbusStateClosed:
 530                 if (dev->state == XenbusStateClosed)
 531                         break;
 532                 /* Missed the backend's CLOSING state -- fallthrough */
 533         case XenbusStateClosing:
 534                 xenbus_frontend_closed(dev);
 535                 break;
 536         }
 537 }
 538
 539 static struct xenbus_driver xen_9pfs_front_driver = {
 540         .ids = xen_9pfs_front_ids,
 541         .probe = xen_9pfs_front_probe,
 542         .remove = xen_9pfs_front_remove,
 543         .resume = xen_9pfs_front_resume,
 544         .otherend_changed = xen_9pfs_front_changed,
 545 };
 546
 547 static int p9_trans_xen_init(void)
 548 {
 549         int rc;
 550
 551         if (!xen_domain())
 552                 return -ENODEV;
 553
 554         pr_info("Initialising Xen transport for 9pfs\n");
 555
 556         v9fs_register_trans(&p9_xen_trans);
 557         rc = xenbus_register_frontend(&xen_9pfs_front_driver);
 558         if (rc)
 559                 v9fs_unregister_trans(&p9_xen_trans);
 560
 561         return rc;
 562 }
 563 module_init(p9_trans_xen_init);
 564
 565 static void p9_trans_xen_exit(void)
 566 {
 567         v9fs_unregister_trans(&p9_xen_trans);
 568         return xenbus_unregister_driver(&xen_9pfs_front_driver);
 569 }
 570 module_exit(p9_trans_xen_exit);