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 6
  47 #define XEN_9PFS_RING_SIZE  XEN_FLEX_RING_SIZE(XEN_9PFS_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 -
 136                 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE) >= 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 - xen_9pfs_queued(prod, cons,
 171                                                  XEN_9PFS_RING_SIZE) < size) {
 172                 spin_unlock_irqrestore(&ring->lock, flags);
 173                 goto again;
 174         }
 175
 176         masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE);
 177         masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
 178
 179         xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size,
 180                               &masked_prod, masked_cons, XEN_9PFS_RING_SIZE);
 181
 182         p9_req->status = REQ_STATUS_SENT;
 183         virt_wmb();                     /* write ring before updating pointer */
 184         prod += size;
 185         ring->intf->out_prod = prod;
 186         spin_unlock_irqrestore(&ring->lock, flags);
 187         notify_remote_via_irq(ring->irq);
 188         p9_req_put(p9_req);
 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                 if (h.size > req->rc.capacity) {
 234                         dev_warn(&priv->dev->dev,
 235                                  "requested packet size too big: %d for tag %d with capacity %zd\n",
 236                                  h.size, h.tag, req->rc.capacity);
 237                         req->status = REQ_STATUS_ERROR;
 238                         goto recv_error;
 239                 }
 240
 241                 memcpy(&req->rc, &h, sizeof(h));
 242                 req->rc.offset = 0;
 243
 244                 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE);
 245                 /* Then, read the whole packet (including the header) */
 246                 xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size,
 247                                      masked_prod, &masked_cons,
 248                                      XEN_9PFS_RING_SIZE);
 249
 250 recv_error:
 251                 virt_mb();
 252                 cons += h.size;
 253                 ring->intf->in_cons = cons;
 254
 255                 status = (req->status != REQ_STATUS_ERROR) ?
 256                         REQ_STATUS_RCVD : REQ_STATUS_ERROR;
 257
 258                 p9_client_cb(priv->client, req, status);
 259         }
 260 }
 261
 262 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
 263 {
 264         struct xen_9pfs_dataring *ring = r;
 265
 266         if (!ring || !ring->priv->client) {
 267                 /* ignore spurious interrupt */
 268                 return IRQ_HANDLED;
 269         }
 270
 271         wake_up_interruptible(&ring->wq);
 272         schedule_work(&ring->work);
 273
 274         return IRQ_HANDLED;
 275 }
 276
 277 static struct p9_trans_module p9_xen_trans = {
 278         .name = "xen",
 279         .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
 280         .def = 1,
 281         .create = p9_xen_create,
 282         .close = p9_xen_close,
 283         .request = p9_xen_request,
 284         .cancel = p9_xen_cancel,
 285         .owner = THIS_MODULE,
 286 };
 287
 288 static const struct xenbus_device_id xen_9pfs_front_ids[] = {
 289         { "9pfs" },
 290         { "" }
 291 };
 292
 293 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
 294 {
 295         int i, j;
 296
 297         write_lock(&xen_9pfs_lock);
 298         list_del(&priv->list);
 299         write_unlock(&xen_9pfs_lock);
 300
 301         for (i = 0; i < priv->num_rings; i++) {
 302                 struct xen_9pfs_dataring *ring = &priv->rings[i];
 303
 304                 cancel_work_sync(&ring->work);
 305
 306                 if (!priv->rings[i].intf)
 307                         break;
 308                 if (priv->rings[i].irq > 0)
 309                         unbind_from_irqhandler(priv->rings[i].irq, priv->dev);
 310                 if (priv->rings[i].data.in) {
 311                         for (j = 0; j < (1 << XEN_9PFS_RING_ORDER); j++) {
 312                                 grant_ref_t ref;
 313
 314                                 ref = priv->rings[i].intf->ref[j];
 315                                 gnttab_end_foreign_access(ref, 0, 0);
 316                         }
 317                         free_pages_exact(priv->rings[i].data.in,
 318                                    1UL << (XEN_9PFS_RING_ORDER +
 319                                            XEN_PAGE_SHIFT));
 320                 }
 321                 gnttab_end_foreign_access(priv->rings[i].ref, 0, 0);
 322                 free_page((unsigned long)priv->rings[i].intf);
 323         }
 324         kfree(priv->rings);
 325         kfree(priv->tag);
 326         kfree(priv);
 327 }
 328
 329 static int xen_9pfs_front_remove(struct xenbus_device *dev)
 330 {
 331         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 332
 333         dev_set_drvdata(&dev->dev, NULL);
 334         xen_9pfs_front_free(priv);
 335         return 0;
 336 }
 337
 338 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev,
 339                                          struct xen_9pfs_dataring *ring)
 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 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT),
 358                                   GFP_KERNEL | __GFP_ZERO);
 359         if (!bytes) {
 360                 ret = -ENOMEM;
 361                 goto out;
 362         }
 363         for (; i < (1 << XEN_9PFS_RING_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 = XEN_9PFS_RING_ORDER;
 371         ring->data.in = bytes;
 372         ring->data.out = bytes + XEN_9PFS_RING_SIZE;
 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 << (XEN_9PFS_RING_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_init(struct xenbus_device *dev)
 397 {
 398         int ret, i;
 399         struct xenbus_transaction xbt;
 400         struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev);
 401         char *versions, *v;
 402         unsigned int max_rings, max_ring_order, len = 0;
 403
 404         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
 405         if (IS_ERR(versions))
 406                 return PTR_ERR(versions);
 407         for (v = versions; *v; v++) {
 408                 if (simple_strtoul(v, &v, 10) == 1) {
 409                         v = NULL;
 410                         break;
 411                 }
 412         }
 413         if (v) {
 414                 kfree(versions);
 415                 return -EINVAL;
 416         }
 417         kfree(versions);
 418         max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0);
 419         if (max_rings < XEN_9PFS_NUM_RINGS)
 420                 return -EINVAL;
 421         max_ring_order = xenbus_read_unsigned(dev->otherend,
 422                                               "max-ring-page-order", 0);
 423         if (max_ring_order < XEN_9PFS_RING_ORDER)
 424                 return -EINVAL;
 425
 426         priv->num_rings = XEN_9PFS_NUM_RINGS;
 427         priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings),
 428                               GFP_KERNEL);
 429         if (!priv->rings) {
 430                 kfree(priv);
 431                 return -ENOMEM;
 432         }
 433
 434         for (i = 0; i < priv->num_rings; i++) {
 435                 priv->rings[i].priv = priv;
 436                 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i]);
 437                 if (ret < 0)
 438                         goto error;
 439         }
 440
 441  again:
 442         ret = xenbus_transaction_start(&xbt);
 443         if (ret) {
 444                 xenbus_dev_fatal(dev, ret, "starting transaction");
 445                 goto error;
 446         }
 447         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
 448         if (ret)
 449                 goto error_xenbus;
 450         ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u",
 451                             priv->num_rings);
 452         if (ret)
 453                 goto error_xenbus;
 454         for (i = 0; i < priv->num_rings; i++) {
 455                 char str[16];
 456
 457                 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9);
 458                 sprintf(str, "ring-ref%u", i);
 459                 ret = xenbus_printf(xbt, dev->nodename, str, "%d",
 460                                     priv->rings[i].ref);
 461                 if (ret)
 462                         goto error_xenbus;
 463
 464                 sprintf(str, "event-channel-%u", i);
 465                 ret = xenbus_printf(xbt, dev->nodename, str, "%u",
 466                                     priv->rings[i].evtchn);
 467                 if (ret)
 468                         goto error_xenbus;
 469         }
 470         priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL);
 471         if (IS_ERR(priv->tag)) {
 472                 ret = PTR_ERR(priv->tag);
 473                 goto error_xenbus;
 474         }
 475         ret = xenbus_transaction_end(xbt, 0);
 476         if (ret) {
 477                 if (ret == -EAGAIN)
 478                         goto again;
 479                 xenbus_dev_fatal(dev, ret, "completing transaction");
 480                 goto error;
 481         }
 482
 483         return 0;
 484
 485  error_xenbus:
 486         xenbus_transaction_end(xbt, 1);
 487         xenbus_dev_fatal(dev, ret, "writing xenstore");
 488  error:
 489         xen_9pfs_front_free(priv);
 490         return ret;
 491 }
 492
 493 static int xen_9pfs_front_probe(struct xenbus_device *dev,
 494                                 const struct xenbus_device_id *id)
 495 {
 496         struct xen_9pfs_front_priv *priv = NULL;
 497
 498         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 499         if (!priv)
 500                 return -ENOMEM;
 501
 502         priv->dev = dev;
 503         dev_set_drvdata(&dev->dev, priv);
 504
 505         write_lock(&xen_9pfs_lock);
 506         list_add_tail(&priv->list, &xen_9pfs_devs);
 507         write_unlock(&xen_9pfs_lock);
 508
 509         return 0;
 510 }
 511
 512 static int xen_9pfs_front_resume(struct xenbus_device *dev)
 513 {
 514         dev_warn(&dev->dev, "suspend/resume unsupported\n");
 515         return 0;
 516 }
 517
 518 static void xen_9pfs_front_changed(struct xenbus_device *dev,
 519                                    enum xenbus_state backend_state)
 520 {
 521         switch (backend_state) {
 522         case XenbusStateReconfiguring:
 523         case XenbusStateReconfigured:
 524         case XenbusStateInitialising:
 525         case XenbusStateInitialised:
 526         case XenbusStateUnknown:
 527                 break;
 528
 529         case XenbusStateInitWait:
 530                 if (!xen_9pfs_front_init(dev))
 531                         xenbus_switch_state(dev, XenbusStateInitialised);
 532                 break;
 533
 534         case XenbusStateConnected:
 535                 xenbus_switch_state(dev, XenbusStateConnected);
 536                 break;
 537
 538         case XenbusStateClosed:
 539                 if (dev->state == XenbusStateClosed)
 540                         break;
 541                 /* Missed the backend's CLOSING state -- fallthrough */
 542         case XenbusStateClosing:
 543                 xenbus_frontend_closed(dev);
 544                 break;
 545         }
 546 }
 547
 548 static struct xenbus_driver xen_9pfs_front_driver = {
 549         .ids = xen_9pfs_front_ids,
 550         .probe = xen_9pfs_front_probe,
 551         .remove = xen_9pfs_front_remove,
 552         .resume = xen_9pfs_front_resume,
 553         .otherend_changed = xen_9pfs_front_changed,
 554 };
 555
 556 static int p9_trans_xen_init(void)
 557 {
 558         int rc;
 559
 560         if (!xen_domain())
 561                 return -ENODEV;
 562
 563         pr_info("Initialising Xen transport for 9pfs\n");
 564
 565         v9fs_register_trans(&p9_xen_trans);
 566         rc = xenbus_register_frontend(&xen_9pfs_front_driver);
 567         if (rc)
 568                 v9fs_unregister_trans(&p9_xen_trans);
 569
 570         return rc;
 571 }
 572 module_init(p9_trans_xen_init);
 573
 574 static void p9_trans_xen_exit(void)
 575 {
 576         v9fs_unregister_trans(&p9_xen_trans);
 577         return xenbus_unregister_driver(&xen_9pfs_front_driver);
 578 }
 579 module_exit(p9_trans_xen_exit);
 580
 581 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>");
 582 MODULE_DESCRIPTION("Xen Transport for 9P");
 583 MODULE_LICENSE("GPL");