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 <linux/fs_context.h> 19 #include <net/9p/9p.h> 20 #include <net/9p/client.h> 21 #include <net/9p/transport.h> 22 23 #define XEN_9PFS_NUM_RINGS 2 24 #define XEN_9PFS_RING_ORDER 9 25 #define XEN_9PFS_RING_SIZE(ring) XEN_FLEX_RING_SIZE(ring->intf->ring_order) 26 27 struct xen_9pfs_header { 28 uint32_t size; 29 uint8_t id; 30 uint16_t tag; 31 32 /* uint8_t sdata[]; */ 33 } __attribute__((packed)); 34 35 /* One per ring, more than one per 9pfs share */ 36 struct xen_9pfs_dataring { 37 struct xen_9pfs_front_priv *priv; 38 39 struct xen_9pfs_data_intf *intf; 40 grant_ref_t ref; 41 int evtchn; 42 int irq; 43 /* protect a ring from concurrent accesses */ 44 spinlock_t lock; 45 46 struct xen_9pfs_data data; 47 wait_queue_head_t wq; 48 struct work_struct work; 49 }; 50 51 /* One per 9pfs share */ 52 struct xen_9pfs_front_priv { 53 struct list_head list; 54 struct xenbus_device *dev; 55 char *tag; 56 struct p9_client *client; 57 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, struct fs_context *fc) 71 { 72 const char *addr = fc->source; 73 struct xen_9pfs_front_priv *priv; 74 75 if (addr == NULL) 76 return -EINVAL; 77 78 read_lock(&xen_9pfs_lock); 79 list_for_each_entry(priv, &xen_9pfs_devs, list) { 80 if (!strcmp(priv->tag, addr)) { 81 priv->client = client; 82 read_unlock(&xen_9pfs_lock); 83 return 0; 84 } 85 } 86 read_unlock(&xen_9pfs_lock); 87 return -EINVAL; 88 } 89 90 static void p9_xen_close(struct p9_client *client) 91 { 92 struct xen_9pfs_front_priv *priv; 93 94 read_lock(&xen_9pfs_lock); 95 list_for_each_entry(priv, &xen_9pfs_devs, list) { 96 if (priv->client == client) { 97 priv->client = NULL; 98 read_unlock(&xen_9pfs_lock); 99 return; 100 } 101 } 102 read_unlock(&xen_9pfs_lock); 103 } 104 105 static bool p9_xen_write_todo(struct xen_9pfs_dataring *ring, RING_IDX size) 106 { 107 RING_IDX cons, prod; 108 109 cons = ring->intf->out_cons; 110 prod = ring->intf->out_prod; 111 virt_mb(); 112 113 return XEN_9PFS_RING_SIZE(ring) - 114 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) >= size; 115 } 116 117 static int p9_xen_request(struct p9_client *client, struct p9_req_t *p9_req) 118 { 119 struct xen_9pfs_front_priv *priv; 120 RING_IDX cons, prod, masked_cons, masked_prod; 121 unsigned long flags; 122 u32 size = p9_req->tc.size; 123 struct xen_9pfs_dataring *ring; 124 int num; 125 126 read_lock(&xen_9pfs_lock); 127 list_for_each_entry(priv, &xen_9pfs_devs, list) { 128 if (priv->client == client) 129 break; 130 } 131 read_unlock(&xen_9pfs_lock); 132 if (list_entry_is_head(priv, &xen_9pfs_devs, list)) 133 return -EINVAL; 134 135 num = p9_req->tc.tag % XEN_9PFS_NUM_RINGS; 136 ring = &priv->rings[num]; 137 138 again: 139 while (io_wait_event_killable(ring->wq, 140 p9_xen_write_todo(ring, size)) != 0) 141 ; 142 143 spin_lock_irqsave(&ring->lock, flags); 144 cons = ring->intf->out_cons; 145 prod = ring->intf->out_prod; 146 virt_mb(); 147 148 if (XEN_9PFS_RING_SIZE(ring) - 149 xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) < size) { 150 spin_unlock_irqrestore(&ring->lock, flags); 151 goto again; 152 } 153 154 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring)); 155 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring)); 156 157 xen_9pfs_write_packet(ring->data.out, p9_req->tc.sdata, size, 158 &masked_prod, masked_cons, 159 XEN_9PFS_RING_SIZE(ring)); 160 161 WRITE_ONCE(p9_req->status, REQ_STATUS_SENT); 162 virt_wmb(); /* write ring before updating pointer */ 163 prod += size; 164 ring->intf->out_prod = prod; 165 spin_unlock_irqrestore(&ring->lock, flags); 166 notify_remote_via_irq(ring->irq); 167 p9_req_put(client, p9_req); 168 169 return 0; 170 } 171 172 static void p9_xen_response(struct work_struct *work) 173 { 174 struct xen_9pfs_front_priv *priv; 175 struct xen_9pfs_dataring *ring; 176 RING_IDX cons, prod, masked_cons, masked_prod; 177 struct xen_9pfs_header h; 178 struct p9_req_t *req; 179 int status; 180 181 ring = container_of(work, struct xen_9pfs_dataring, work); 182 priv = ring->priv; 183 184 while (1) { 185 cons = ring->intf->in_cons; 186 prod = ring->intf->in_prod; 187 virt_rmb(); 188 189 if (xen_9pfs_queued(prod, cons, XEN_9PFS_RING_SIZE(ring)) < 190 sizeof(h)) { 191 notify_remote_via_irq(ring->irq); 192 return; 193 } 194 195 masked_prod = xen_9pfs_mask(prod, XEN_9PFS_RING_SIZE(ring)); 196 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring)); 197 198 /* First, read just the header */ 199 xen_9pfs_read_packet(&h, ring->data.in, sizeof(h), 200 masked_prod, &masked_cons, 201 XEN_9PFS_RING_SIZE(ring)); 202 203 req = p9_tag_lookup(priv->client, h.tag); 204 if (!req || req->status != REQ_STATUS_SENT) { 205 dev_warn(&priv->dev->dev, "Wrong req tag=%x\n", h.tag); 206 cons += h.size; 207 virt_mb(); 208 ring->intf->in_cons = cons; 209 continue; 210 } 211 212 if (h.size > req->rc.capacity) { 213 dev_warn(&priv->dev->dev, 214 "requested packet size too big: %d for tag %d with capacity %zd\n", 215 h.size, h.tag, req->rc.capacity); 216 WRITE_ONCE(req->status, REQ_STATUS_ERROR); 217 goto recv_error; 218 } 219 220 req->rc.size = h.size; 221 req->rc.id = h.id; 222 req->rc.tag = h.tag; 223 req->rc.offset = 0; 224 225 masked_cons = xen_9pfs_mask(cons, XEN_9PFS_RING_SIZE(ring)); 226 /* Then, read the whole packet (including the header) */ 227 xen_9pfs_read_packet(req->rc.sdata, ring->data.in, h.size, 228 masked_prod, &masked_cons, 229 XEN_9PFS_RING_SIZE(ring)); 230 231 recv_error: 232 virt_mb(); 233 cons += h.size; 234 ring->intf->in_cons = cons; 235 236 status = (req->status != REQ_STATUS_ERROR) ? 237 REQ_STATUS_RCVD : REQ_STATUS_ERROR; 238 239 p9_client_cb(priv->client, req, status); 240 } 241 } 242 243 static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r) 244 { 245 struct xen_9pfs_dataring *ring = r; 246 247 if (!ring || !ring->priv->client) { 248 /* ignore spurious interrupt */ 249 return IRQ_HANDLED; 250 } 251 252 wake_up_interruptible(&ring->wq); 253 schedule_work(&ring->work); 254 255 return IRQ_HANDLED; 256 } 257 258 static struct p9_trans_module p9_xen_trans = { 259 .name = "xen", 260 .maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2), 261 .pooled_rbuffers = false, 262 .def = true, 263 .supports_vmalloc = false, 264 .create = p9_xen_create, 265 .close = p9_xen_close, 266 .request = p9_xen_request, 267 .cancel = p9_xen_cancel, 268 .owner = THIS_MODULE, 269 }; 270 271 static const struct xenbus_device_id xen_9pfs_front_ids[] = { 272 { "9pfs" }, 273 { "" } 274 }; 275 276 static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv) 277 { 278 int i, j; 279 280 if (priv->rings) { 281 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) { 282 struct xen_9pfs_dataring *ring = &priv->rings[i]; 283 284 cancel_work_sync(&ring->work); 285 286 if (!priv->rings[i].intf) 287 break; 288 if (priv->rings[i].irq > 0) 289 unbind_from_irqhandler(priv->rings[i].irq, ring); 290 if (priv->rings[i].data.in) { 291 for (j = 0; 292 j < (1 << priv->rings[i].intf->ring_order); 293 j++) { 294 grant_ref_t ref; 295 296 ref = priv->rings[i].intf->ref[j]; 297 gnttab_end_foreign_access(ref, NULL); 298 } 299 free_pages_exact(priv->rings[i].data.in, 300 1UL << (priv->rings[i].intf->ring_order + 301 XEN_PAGE_SHIFT)); 302 } 303 gnttab_end_foreign_access(priv->rings[i].ref, NULL); 304 free_page((unsigned long)priv->rings[i].intf); 305 } 306 kfree(priv->rings); 307 } 308 kfree(priv->tag); 309 kfree(priv); 310 } 311 312 static void xen_9pfs_front_remove(struct xenbus_device *dev) 313 { 314 struct xen_9pfs_front_priv *priv; 315 316 write_lock(&xen_9pfs_lock); 317 priv = dev_get_drvdata(&dev->dev); 318 if (priv == NULL) { 319 write_unlock(&xen_9pfs_lock); 320 return; 321 } 322 dev_set_drvdata(&dev->dev, NULL); 323 list_del(&priv->list); 324 write_unlock(&xen_9pfs_lock); 325 326 xen_9pfs_front_free(priv); 327 } 328 329 static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev, 330 struct xen_9pfs_dataring *ring, 331 unsigned int order) 332 { 333 int i = 0; 334 int ret = -ENOMEM; 335 void *bytes = NULL; 336 337 init_waitqueue_head(&ring->wq); 338 spin_lock_init(&ring->lock); 339 INIT_WORK(&ring->work, p9_xen_response); 340 341 ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL); 342 if (!ring->intf) 343 return ret; 344 ret = gnttab_grant_foreign_access(dev->otherend_id, 345 virt_to_gfn(ring->intf), 0); 346 if (ret < 0) 347 goto out; 348 ring->ref = ret; 349 bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT), 350 GFP_KERNEL | __GFP_ZERO); 351 if (!bytes) { 352 ret = -ENOMEM; 353 goto out; 354 } 355 for (; i < (1 << order); i++) { 356 ret = gnttab_grant_foreign_access( 357 dev->otherend_id, virt_to_gfn(bytes) + i, 0); 358 if (ret < 0) 359 goto out; 360 ring->intf->ref[i] = ret; 361 } 362 ring->intf->ring_order = order; 363 ring->data.in = bytes; 364 ring->data.out = bytes + XEN_FLEX_RING_SIZE(order); 365 366 ret = xenbus_alloc_evtchn(dev, &ring->evtchn); 367 if (ret) 368 goto out; 369 ring->irq = bind_evtchn_to_irqhandler(ring->evtchn, 370 xen_9pfs_front_event_handler, 371 0, "xen_9pfs-frontend", ring); 372 if (ring->irq >= 0) 373 return 0; 374 375 xenbus_free_evtchn(dev, ring->evtchn); 376 ret = ring->irq; 377 out: 378 if (bytes) { 379 for (i--; i >= 0; i--) 380 gnttab_end_foreign_access(ring->intf->ref[i], NULL); 381 free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT)); 382 } 383 gnttab_end_foreign_access(ring->ref, NULL); 384 free_page((unsigned long)ring->intf); 385 return ret; 386 } 387 388 static int xen_9pfs_front_init(struct xenbus_device *dev) 389 { 390 int ret, i; 391 struct xenbus_transaction xbt; 392 struct xen_9pfs_front_priv *priv; 393 char *versions, *v; 394 unsigned int max_rings, max_ring_order, len = 0; 395 396 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len); 397 if (IS_ERR(versions)) 398 return PTR_ERR(versions); 399 for (v = versions; *v; v++) { 400 if (simple_strtoul(v, &v, 10) == 1) { 401 v = NULL; 402 break; 403 } 404 } 405 if (v) { 406 kfree(versions); 407 return -EINVAL; 408 } 409 kfree(versions); 410 max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0); 411 if (max_rings < XEN_9PFS_NUM_RINGS) 412 return -EINVAL; 413 max_ring_order = xenbus_read_unsigned(dev->otherend, 414 "max-ring-page-order", 0); 415 if (max_ring_order > XEN_9PFS_RING_ORDER) 416 max_ring_order = XEN_9PFS_RING_ORDER; 417 if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order)) 418 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2; 419 420 priv = kzalloc_obj(*priv); 421 if (!priv) 422 return -ENOMEM; 423 priv->dev = dev; 424 priv->rings = kzalloc_objs(*priv->rings, XEN_9PFS_NUM_RINGS); 425 if (!priv->rings) { 426 kfree(priv); 427 return -ENOMEM; 428 } 429 430 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) { 431 priv->rings[i].priv = priv; 432 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i], 433 max_ring_order); 434 if (ret < 0) 435 goto error; 436 } 437 438 again: 439 ret = xenbus_transaction_start(&xbt); 440 if (ret) { 441 xenbus_dev_fatal(dev, ret, "starting transaction"); 442 goto error; 443 } 444 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1); 445 if (ret) 446 goto error_xenbus; 447 ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u", 448 XEN_9PFS_NUM_RINGS); 449 if (ret) 450 goto error_xenbus; 451 452 for (i = 0; i < XEN_9PFS_NUM_RINGS; i++) { 453 char str[16]; 454 455 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9); 456 sprintf(str, "ring-ref%d", i); 457 ret = xenbus_printf(xbt, dev->nodename, str, "%d", 458 priv->rings[i].ref); 459 if (ret) 460 goto error_xenbus; 461 462 sprintf(str, "event-channel-%d", i); 463 ret = xenbus_printf(xbt, dev->nodename, str, "%u", 464 priv->rings[i].evtchn); 465 if (ret) 466 goto error_xenbus; 467 } 468 priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL); 469 if (IS_ERR(priv->tag)) { 470 ret = PTR_ERR(priv->tag); 471 goto error_xenbus; 472 } 473 ret = xenbus_transaction_end(xbt, 0); 474 if (ret) { 475 if (ret == -EAGAIN) 476 goto again; 477 xenbus_dev_fatal(dev, ret, "completing transaction"); 478 goto error; 479 } 480 481 write_lock(&xen_9pfs_lock); 482 dev_set_drvdata(&dev->dev, priv); 483 list_add_tail(&priv->list, &xen_9pfs_devs); 484 write_unlock(&xen_9pfs_lock); 485 486 xenbus_switch_state(dev, XenbusStateInitialised); 487 return 0; 488 489 error_xenbus: 490 xenbus_transaction_end(xbt, 1); 491 xenbus_dev_fatal(dev, ret, "writing xenstore"); 492 error: 493 xen_9pfs_front_free(priv); 494 return ret; 495 } 496 497 static int xen_9pfs_front_probe(struct xenbus_device *dev, 498 const struct xenbus_device_id *id) 499 { 500 return 0; 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 if (dev->state != XenbusStateInitialising) 522 break; 523 524 xen_9pfs_front_init(dev); 525 break; 526 527 case XenbusStateConnected: 528 xenbus_switch_state(dev, XenbusStateConnected); 529 break; 530 531 case XenbusStateClosed: 532 if (dev->state == XenbusStateClosed) 533 break; 534 fallthrough; /* Missed the backend's CLOSING state */ 535 case XenbusStateClosing: 536 xenbus_frontend_closed(dev); 537 break; 538 } 539 } 540 541 static struct xenbus_driver xen_9pfs_front_driver = { 542 .ids = xen_9pfs_front_ids, 543 .probe = xen_9pfs_front_probe, 544 .remove = xen_9pfs_front_remove, 545 .resume = xen_9pfs_front_resume, 546 .otherend_changed = xen_9pfs_front_changed, 547 }; 548 549 static int __init p9_trans_xen_init(void) 550 { 551 int rc; 552 553 if (!xen_domain()) 554 return -ENODEV; 555 556 pr_info("Initialising Xen transport for 9pfs\n"); 557 558 v9fs_register_trans(&p9_xen_trans); 559 rc = xenbus_register_frontend(&xen_9pfs_front_driver); 560 if (rc) 561 v9fs_unregister_trans(&p9_xen_trans); 562 563 return rc; 564 } 565 module_init(p9_trans_xen_init); 566 MODULE_ALIAS_9P("xen"); 567 568 static void __exit p9_trans_xen_exit(void) 569 { 570 v9fs_unregister_trans(&p9_xen_trans); 571 return xenbus_unregister_driver(&xen_9pfs_front_driver); 572 } 573 module_exit(p9_trans_xen_exit); 574 575 MODULE_ALIAS("xen:9pfs"); 576 MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>"); 577 MODULE_DESCRIPTION("Xen Transport for 9P"); 578 MODULE_LICENSE("GPL"); 579