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                 if (!priv->rings[i].intf)
 294                         break;
 295                 if (priv->rings[i].irq > 0)
 296                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
 297                 if (priv->rings[i].data.in) {
 298                         for (j = 0; j < (1 << XEN_9PFS_RING_ORDER); j++) {
 299                                 grant_ref_t ref;
 300
 301                                 ref = priv->rings[i].intf->ref[j];
 302                                 gnttab_end_foreign_access(ref, 0, 0);
 303                         }
 304                         free_pages_exact(priv->rings[i].data.in,
 305                                    1UL << (XEN_9PFS_RING_ORDER +
 306                                            XEN_PAGE_SHIFT));
 307                 }
 308                 gnttab_end_foreign_access(priv->rings[i].ref, 0, 0);
 309                 free_page((unsigned long)priv->rings[i].intf);
 310         }
 311         kfree(priv->rings);
 312         kfree(priv->tag);
 313         kfree(priv);
 314 }
 315
 316 static int xen_9pfs_front_remove(struct xenbus_device *dev)
 317 {
 318         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 319
 320         dev_set_drvdata(&dev->dev, NULL);
 321         xen_9pfs_front_free(priv);
 322         return 0;
 323 }
 324
 325 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
 326                                          struct xen_9pfs_dataring *ring)
 327 {
 328         int i = 0;
 329         int ret = -ENOMEM;
 330         void *bytes = NULL;
 331
 332         init_waitqueue_head(&ring->wq);
 333         spin_lock_init(&ring->lock);
 334         INIT_WORK(&ring->work, p9_xen_response);
 335
 336         ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL);
 337         if (!ring->intf)
 338                 return ret;
 339         ret = gnttab_grant_foreign_access(dev->otherend_id,
 340                                           virt_to_gfn(ring->intf), 0);
 341         if (ret < 0)
 342                 goto out;
 343         ring->ref = ret;
 344         bytes = alloc_pages_exact(1UL << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
 345                                   GFP_KERNEL | __GFP_ZERO);
 346         if (!bytes) {
 347                 ret = -ENOMEM;
 348                 goto out;
 349         }
 350         for (; i < (1 << XEN_9PFS_RING_ORDER); i++) {
 351                 ret = gnttab_grant_foreign_access(
 352                                 dev->otherend_id, virt_to_gfn(bytes) + i, 0);
 353                 if (ret < 0)
 354                         goto out;
 355                 ring->intf->ref[i] = ret;
 356         }
 357         ring->intf->ring_order = XEN_9PFS_RING_ORDER;
 358         ring->data.in = bytes;
 359         ring->data.out = bytes + XEN_9PFS_RING_SIZE;
 360
 361         ret = xenbus_alloc_evtchn(dev, &ring->evtchn);
 362         if (ret)
 363                 goto out;
 364         ring->irq = bind_evtchn_to_irqhandler(ring->evtchn,
 365                                               xen_9pfs_front_event_handler,
 366                                               0, "xen_9pfs-frontend", ring);
 367         if (ring->irq >= 0)
 368                 return 0;
 369
 370         xenbus_free_evtchn(dev, ring->evtchn);
 371         ret = ring->irq;
 372 out:
 373         if (bytes) {
 374                 for (i--; i >= 0; i--)
 375                         gnttab_end_foreign_access(ring->intf->ref[i], 0, 0);
 376                 free_pages_exact(bytes, 1UL << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT));
 377         }
 378         gnttab_end_foreign_access(ring->ref, 0, 0);
 379         free_page((unsigned long)ring->intf);
 380         return ret;
 381 }
 382
 383 static int xen_9pfs_front_init(struct xenbus_device *dev)
 384 {
 385         int ret, i;
 386         struct xenbus_transaction xbt;
 387         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 388         char *versions, *v;
 389         unsigned int max_rings, max_ring_order, len = 0;
 390
 391         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
 392         if (IS_ERR(versions))
 393                 return PTR_ERR(versions);
 394         for (v = versions; *v; v++) {
 395                 if (simple_strtoul(v, &v, 10) == 1) {
 396                         v = NULL;
 397                         break;
 398                 }
 399         }
 400         if (v) {
 401                 kfree(versions);
 402                 return -EINVAL;
 403         }
 404         kfree(versions);
 405         max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
 406         if (max_rings < XEN_9PFS_NUM_RINGS)
 407                 return -EINVAL;
 408         max_ring_order = xenbus_read_unsigned(dev->otherend,
 409                                               "max-ring-page-order", 0);
 410         if (max_ring_order < XEN_9PFS_RING_ORDER)
 411                 return -EINVAL;
 412
 413         priv->num_rings = XEN_9PFS_NUM_RINGS;
 414         priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
 415                               GFP_KERNEL);
 416         if (!priv->rings) {
 417                 kfree(priv);
 418                 return -ENOMEM;
 419         }
 420
 421         for (i = 0; i < priv->num_rings; i++) {
 422                 priv->rings[i].priv = priv;
 423                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i]);
 424                 if (ret < 0)
 425                         goto error;
 426         }
 427
 428  again:
 429         ret = xenbus_transaction_start(&xbt);
 430         if (ret) {
 431                 xenbus_dev_fatal(dev, ret, "starting transaction");
 432                 goto error;
 433         }
 434         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
 435         if (ret)
 436                 goto error_xenbus;
 437         ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
 438                             priv->num_rings);
 439         if (ret)
 440                 goto error_xenbus;
 441         for (i = 0; i < priv->num_rings; i++) {
 442                 char str[16];
 443
 444                 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
 445                 sprintf(str, "ring-ref%u", i);
 446                 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
 447                                     priv->rings[i].ref);
 448                 if (ret)
 449                         goto error_xenbus;
 450
 451                 sprintf(str, "event-channel-%u", i);
 452                 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
 453                                     priv->rings[i].evtchn);
 454                 if (ret)
 455                         goto error_xenbus;
 456         }
 457         priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
 458         if (IS_ERR(priv->tag)) {
 459                 ret = PTR_ERR(priv->tag);
 460                 goto error_xenbus;
 461         }
 462         ret = xenbus_transaction_end(xbt, 0);
 463         if (ret) {
 464                 if (ret == -EAGAIN)
 465                         goto again;
 466                 xenbus_dev_fatal(dev, ret, "completing transaction");
 467                 goto error;
 468         }
 469
 470         return 0;
 471
 472  error_xenbus:
 473         xenbus_transaction_end(xbt, 1);
 474         xenbus_dev_fatal(dev, ret, "writing xenstore");
 475  error:
 476         xen_9pfs_front_free(priv);
 477         return ret;
 478 }
 479
 480 static int xen_9pfs_front_probe(struct xenbus_device *dev,
 481                                 const struct xenbus_device_id *id)
 482 {
 483         struct xen_9pfs_front_priv *priv = NULL;
 484
 485         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 486         if (!priv)
 487                 return -ENOMEM;
 488
 489         priv->dev = dev;
 490         dev_set_drvdata(&dev->dev, priv);
 491
 492         write_lock(&xen_9pfs_lock);
 493         list_add_tail(&priv->list, &xen_9pfs_devs);
 494         write_unlock(&xen_9pfs_lock);
 495
 496         return 0;
 497 }
 498
 499 static int xen_9pfs_front_resume(struct xenbus_device *dev)
 500 {
 501         dev_warn(&dev->dev, "suspsend/resume unsupported\n");
 502         return 0;
 503 }
 504
 505 static void xen_9pfs_front_changed(struct xenbus_device *dev,
 506                                    enum xenbus_state backend_state)
 507 {
 508         switch (backend_state) {
 509         case XenbusStateReconfiguring:
 510         case XenbusStateReconfigured:
 511         case XenbusStateInitialising:
 512         case XenbusStateInitialised:
 513         case XenbusStateUnknown:
 514                 break;
 515
 516         case XenbusStateInitWait:
 517                 if (!xen_9pfs_front_init(dev))
 518                         xenbus_switch_state(dev, XenbusStateInitialised);
 519                 break;
 520
 521         case XenbusStateConnected:
 522                 xenbus_switch_state(dev, XenbusStateConnected);
 523                 break;
 524
 525         case XenbusStateClosed:
 526                 if (dev->state == XenbusStateClosed)
 527                         break;
 528                 /* Missed the backend's CLOSING state -- fallthrough */
 529         case XenbusStateClosing:
 530                 xenbus_frontend_closed(dev);
 531                 break;
 532         }
 533 }
 534
 535 static struct xenbus_driver xen_9pfs_front_driver = {
 536         .ids = xen_9pfs_front_ids,
 537         .probe = xen_9pfs_front_probe,
 538         .remove = xen_9pfs_front_remove,
 539         .resume = xen_9pfs_front_resume,
 540         .otherend_changed = xen_9pfs_front_changed,
 541 };
 542
 543 static int p9_trans_xen_init(void)
 544 {
 545         int rc;
 546
 547         if (!xen_domain())
 548                 return -ENODEV;
 549
 550         pr_info("Initialising Xen transport for 9pfs\n");
 551
 552         v9fs_register_trans(&p9_xen_trans);
 553         rc = xenbus_register_frontend(&xen_9pfs_front_driver);
 554         if (rc)
 555                 v9fs_unregister_trans(&p9_xen_trans);
 556
 557         return rc;
 558 }
 559 module_init(p9_trans_xen_init);
 560
 561 static void p9_trans_xen_exit(void)
 562 {
 563         v9fs_unregister_trans(&p9_xen_trans);
 564         return xenbus_unregister_driver(&xen_9pfs_front_driver);
 565 }
 566 module_exit(p9_trans_xen_exit);