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 49struct 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 */ 58struct 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 */ 74struct 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 84static LIST_HEAD(xen_9pfs_devs); 85static DEFINE_RWLOCK(xen_9pfs_lock); 86 87/* We don't currently allow canceling of requests */ 88static int p9_xen_cancel(struct p9_client *client, struct p9_req_t *req) 89{ 90 return 1; 91} 92 93static 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 112static 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 127static 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 139static 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 160again: 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(p9_req); 190 191 return 0; 192} 193 194static 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 251recv_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 263static 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 278static 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 289static const struct xenbus_device_id xen_9pfs_front_ids[] = { 290 { "9pfs" }, 291 { "" } 292}; 293 294static 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 struct xen_9pfs_dataring *ring = &priv->rings[i]; 304 305 cancel_work_sync(&ring->work); 306 307 if (!priv->rings[i].intf) 308 break; 309 if (priv->rings[i].irq > 0) 310 unbind_from_irqhandler(priv->rings[i].irq, priv->dev); 311 if (priv->rings[i].data.in) { 312 for (j = 0; 313 j < (1 << priv->rings[i].intf->ring_order); 314 j++) { 315 grant_ref_t ref; 316 317 ref = priv->rings[i].intf->ref[j]; 318 gnttab_end_foreign_access(ref, 0, 0); 319 } 320 free_pages_exact(priv->rings[i].data.in, 321 1UL << (priv->rings[i].intf->ring_order + 322 XEN_PAGE_SHIFT)); 323 } 324 gnttab_end_foreign_access(priv->rings[i].ref, 0, 0); 325 free_page((unsigned long)priv->rings[i].intf); 326 } 327 kfree(priv->rings); 328 kfree(priv->tag); 329 kfree(priv); 330} 331 332static int xen_9pfs_front_remove(struct xenbus_device *dev) 333{ 334 struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev); 335 336 dev_set_drvdata(&dev->dev, NULL); 337 xen_9pfs_front_free(priv); 338 return 0; 339} 340 341static int xen_9pfs_front_alloc_dataring(struct xenbus_device *dev, 342 struct xen_9pfs_dataring *ring, 343 unsigned int order) 344{ 345 int i = 0; 346 int ret = -ENOMEM; 347 void *bytes = NULL; 348 349 init_waitqueue_head(&ring->wq); 350 spin_lock_init(&ring->lock); 351 INIT_WORK(&ring->work, p9_xen_response); 352 353 ring->intf = (struct xen_9pfs_data_intf *)get_zeroed_page(GFP_KERNEL); 354 if (!ring->intf) 355 return ret; 356 ret = gnttab_grant_foreign_access(dev->otherend_id, 357 virt_to_gfn(ring->intf), 0); 358 if (ret < 0) 359 goto out; 360 ring->ref = ret; 361 bytes = alloc_pages_exact(1UL << (order + XEN_PAGE_SHIFT), 362 GFP_KERNEL | __GFP_ZERO); 363 if (!bytes) { 364 ret = -ENOMEM; 365 goto out; 366 } 367 for (; i < (1 << order); i++) { 368 ret = gnttab_grant_foreign_access( 369 dev->otherend_id, virt_to_gfn(bytes) + i, 0); 370 if (ret < 0) 371 goto out; 372 ring->intf->ref[i] = ret; 373 } 374 ring->intf->ring_order = order; 375 ring->data.in = bytes; 376 ring->data.out = bytes + XEN_FLEX_RING_SIZE(order); 377 378 ret = xenbus_alloc_evtchn(dev, &ring->evtchn); 379 if (ret) 380 goto out; 381 ring->irq = bind_evtchn_to_irqhandler(ring->evtchn, 382 xen_9pfs_front_event_handler, 383 0, "xen_9pfs-frontend", ring); 384 if (ring->irq >= 0) 385 return 0; 386 387 xenbus_free_evtchn(dev, ring->evtchn); 388 ret = ring->irq; 389out: 390 if (bytes) { 391 for (i--; i >= 0; i--) 392 gnttab_end_foreign_access(ring->intf->ref[i], 0, 0); 393 free_pages_exact(bytes, 1UL << (order + XEN_PAGE_SHIFT)); 394 } 395 gnttab_end_foreign_access(ring->ref, 0, 0); 396 free_page((unsigned long)ring->intf); 397 return ret; 398} 399 400static int xen_9pfs_front_init(struct xenbus_device *dev) 401{ 402 int ret, i; 403 struct xenbus_transaction xbt; 404 struct xen_9pfs_front_priv *priv = dev_get_drvdata(&dev->dev); 405 char *versions, *v; 406 unsigned int max_rings, max_ring_order, len = 0; 407 408 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len); 409 if (IS_ERR(versions)) 410 return PTR_ERR(versions); 411 for (v = versions; *v; v++) { 412 if (simple_strtoul(v, &v, 10) == 1) { 413 v = NULL; 414 break; 415 } 416 } 417 if (v) { 418 kfree(versions); 419 return -EINVAL; 420 } 421 kfree(versions); 422 max_rings = xenbus_read_unsigned(dev->otherend, "max-rings", 0); 423 if (max_rings < XEN_9PFS_NUM_RINGS) 424 return -EINVAL; 425 max_ring_order = xenbus_read_unsigned(dev->otherend, 426 "max-ring-page-order", 0); 427 if (max_ring_order > XEN_9PFS_RING_ORDER) 428 max_ring_order = XEN_9PFS_RING_ORDER; 429 if (p9_xen_trans.maxsize > XEN_FLEX_RING_SIZE(max_ring_order)) 430 p9_xen_trans.maxsize = XEN_FLEX_RING_SIZE(max_ring_order) / 2; 431 432 priv->num_rings = XEN_9PFS_NUM_RINGS; 433 priv->rings = kcalloc(priv->num_rings, sizeof(*priv->rings), 434 GFP_KERNEL); 435 if (!priv->rings) { 436 kfree(priv); 437 return -ENOMEM; 438 } 439 440 for (i = 0; i < priv->num_rings; i++) { 441 priv->rings[i].priv = priv; 442 ret = xen_9pfs_front_alloc_dataring(dev, &priv->rings[i], 443 max_ring_order); 444 if (ret < 0) 445 goto error; 446 } 447 448 again: 449 ret = xenbus_transaction_start(&xbt); 450 if (ret) { 451 xenbus_dev_fatal(dev, ret, "starting transaction"); 452 goto error; 453 } 454 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1); 455 if (ret) 456 goto error_xenbus; 457 ret = xenbus_printf(xbt, dev->nodename, "num-rings", "%u", 458 priv->num_rings); 459 if (ret) 460 goto error_xenbus; 461 for (i = 0; i < priv->num_rings; i++) { 462 char str[16]; 463 464 BUILD_BUG_ON(XEN_9PFS_NUM_RINGS > 9); 465 sprintf(str, "ring-ref%d", i); 466 ret = xenbus_printf(xbt, dev->nodename, str, "%d", 467 priv->rings[i].ref); 468 if (ret) 469 goto error_xenbus; 470 471 sprintf(str, "event-channel-%d", i); 472 ret = xenbus_printf(xbt, dev->nodename, str, "%u", 473 priv->rings[i].evtchn); 474 if (ret) 475 goto error_xenbus; 476 } 477 priv->tag = xenbus_read(xbt, dev->nodename, "tag", NULL); 478 if (IS_ERR(priv->tag)) { 479 ret = PTR_ERR(priv->tag); 480 goto error_xenbus; 481 } 482 ret = xenbus_transaction_end(xbt, 0); 483 if (ret) { 484 if (ret == -EAGAIN) 485 goto again; 486 xenbus_dev_fatal(dev, ret, "completing transaction"); 487 goto error; 488 } 489 490 return 0; 491 492 error_xenbus: 493 xenbus_transaction_end(xbt, 1); 494 xenbus_dev_fatal(dev, ret, "writing xenstore"); 495 error: 496 xen_9pfs_front_free(priv); 497 return ret; 498} 499 500static int xen_9pfs_front_probe(struct xenbus_device *dev, 501 const struct xenbus_device_id *id) 502{ 503 struct xen_9pfs_front_priv *priv = NULL; 504 505 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 506 if (!priv) 507 return -ENOMEM; 508 509 priv->dev = dev; 510 dev_set_drvdata(&dev->dev, priv); 511 512 write_lock(&xen_9pfs_lock); 513 list_add_tail(&priv->list, &xen_9pfs_devs); 514 write_unlock(&xen_9pfs_lock); 515 516 return 0; 517} 518 519static int xen_9pfs_front_resume(struct xenbus_device *dev) 520{ 521 dev_warn(&dev->dev, "suspend/resume unsupported\n"); 522 return 0; 523} 524 525static void xen_9pfs_front_changed(struct xenbus_device *dev, 526 enum xenbus_state backend_state) 527{ 528 switch (backend_state) { 529 case XenbusStateReconfiguring: 530 case XenbusStateReconfigured: 531 case XenbusStateInitialising: 532 case XenbusStateInitialised: 533 case XenbusStateUnknown: 534 break; 535 536 case XenbusStateInitWait: 537 if (!xen_9pfs_front_init(dev)) 538 xenbus_switch_state(dev, XenbusStateInitialised); 539 break; 540 541 case XenbusStateConnected: 542 xenbus_switch_state(dev, XenbusStateConnected); 543 break; 544 545 case XenbusStateClosed: 546 if (dev->state == XenbusStateClosed) 547 break; 548 fallthrough; /* Missed the backend's CLOSING state */ 549 case XenbusStateClosing: 550 xenbus_frontend_closed(dev); 551 break; 552 } 553} 554 555static struct xenbus_driver xen_9pfs_front_driver = { 556 .ids = xen_9pfs_front_ids, 557 .probe = xen_9pfs_front_probe, 558 .remove = xen_9pfs_front_remove, 559 .resume = xen_9pfs_front_resume, 560 .otherend_changed = xen_9pfs_front_changed, 561}; 562 563static int p9_trans_xen_init(void) 564{ 565 int rc; 566 567 if (!xen_domain()) 568 return -ENODEV; 569 570 pr_info("Initialising Xen transport for 9pfs\n"); 571 572 v9fs_register_trans(&p9_xen_trans); 573 rc = xenbus_register_frontend(&xen_9pfs_front_driver); 574 if (rc) 575 v9fs_unregister_trans(&p9_xen_trans); 576 577 return rc; 578} 579module_init(p9_trans_xen_init); 580 581static void p9_trans_xen_exit(void) 582{ 583 v9fs_unregister_trans(&p9_xen_trans); 584 return xenbus_unregister_driver(&xen_9pfs_front_driver); 585} 586module_exit(p9_trans_xen_exit); 587 588MODULE_AUTHOR("Stefano Stabellini <stefano@aporeto.com>"); 589MODULE_DESCRIPTION("Xen Transport for 9P"); 590MODULE_LICENSE("GPL"); 591