1 /* 2 * The Virtio 9p transport driver 3 * 4 * This is a block based transport driver based on the lguest block driver 5 * code. 6 * 7 * Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation 8 * 9 * Based on virtio console driver 10 * Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation 11 * 12 * This program is free software; you can redistribute it and/or modify 13 * it under the terms of the GNU General Public License version 2 14 * as published by the Free Software Foundation. 15 * 16 * This program is distributed in the hope that it will be useful, 17 * but WITHOUT ANY WARRANTY; without even the implied warranty of 18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 19 * GNU General Public License for more details. 20 * 21 * You should have received a copy of the GNU General Public License 22 * along with this program; if not, write to: 23 * Free Software Foundation 24 * 51 Franklin Street, Fifth Floor 25 * Boston, MA 02111-1301 USA 26 * 27 */ 28 29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 30 31 #include <linux/in.h> 32 #include <linux/module.h> 33 #include <linux/net.h> 34 #include <linux/ipv6.h> 35 #include <linux/errno.h> 36 #include <linux/kernel.h> 37 #include <linux/un.h> 38 #include <linux/uaccess.h> 39 #include <linux/inet.h> 40 #include <linux/idr.h> 41 #include <linux/file.h> 42 #include <linux/highmem.h> 43 #include <linux/slab.h> 44 #include <net/9p/9p.h> 45 #include <linux/parser.h> 46 #include <net/9p/client.h> 47 #include <net/9p/transport.h> 48 #include <linux/scatterlist.h> 49 #include <linux/swap.h> 50 #include <linux/virtio.h> 51 #include <linux/virtio_9p.h> 52 #include "trans_common.h" 53 54 #define VIRTQUEUE_NUM 128 55 56 /* a single mutex to manage channel initialization and attachment */ 57 static DEFINE_MUTEX(virtio_9p_lock); 58 static DECLARE_WAIT_QUEUE_HEAD(vp_wq); 59 static atomic_t vp_pinned = ATOMIC_INIT(0); 60 61 /** 62 * struct virtio_chan - per-instance transport information 63 * @initialized: whether the channel is initialized 64 * @inuse: whether the channel is in use 65 * @lock: protects multiple elements within this structure 66 * @client: client instance 67 * @vdev: virtio dev associated with this channel 68 * @vq: virtio queue associated with this channel 69 * @sg: scatter gather list which is used to pack a request (protected?) 70 * 71 * We keep all per-channel information in a structure. 72 * This structure is allocated within the devices dev->mem space. 73 * A pointer to the structure will get put in the transport private. 74 * 75 */ 76 77 struct virtio_chan { 78 bool inuse; 79 80 spinlock_t lock; 81 82 struct p9_client *client; 83 struct virtio_device *vdev; 84 struct virtqueue *vq; 85 int ring_bufs_avail; 86 wait_queue_head_t *vc_wq; 87 /* This is global limit. Since we don't have a global structure, 88 * will be placing it in each channel. 89 */ 90 int p9_max_pages; 91 /* Scatterlist: can be too big for stack. */ 92 struct scatterlist sg[VIRTQUEUE_NUM]; 93 94 int tag_len; 95 /* 96 * tag name to identify a mount Non-null terminated 97 */ 98 char *tag; 99 100 struct list_head chan_list; 101 }; 102 103 static struct list_head virtio_chan_list; 104 105 /* How many bytes left in this page. */ 106 static unsigned int rest_of_page(void *data) 107 { 108 return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE); 109 } 110 111 /** 112 * p9_virtio_close - reclaim resources of a channel 113 * @client: client instance 114 * 115 * This reclaims a channel by freeing its resources and 116 * reseting its inuse flag. 117 * 118 */ 119 120 static void p9_virtio_close(struct p9_client *client) 121 { 122 struct virtio_chan *chan = client->trans; 123 124 mutex_lock(&virtio_9p_lock); 125 if (chan) 126 chan->inuse = false; 127 mutex_unlock(&virtio_9p_lock); 128 } 129 130 /** 131 * req_done - callback which signals activity from the server 132 * @vq: virtio queue activity was received on 133 * 134 * This notifies us that the server has triggered some activity 135 * on the virtio channel - most likely a response to request we 136 * sent. Figure out which requests now have responses and wake up 137 * those threads. 138 * 139 * Bugs: could do with some additional sanity checking, but appears to work. 140 * 141 */ 142 143 static void req_done(struct virtqueue *vq) 144 { 145 struct virtio_chan *chan = vq->vdev->priv; 146 struct p9_fcall *rc; 147 unsigned int len; 148 struct p9_req_t *req; 149 unsigned long flags; 150 151 p9_debug(P9_DEBUG_TRANS, ": request done\n"); 152 153 while (1) { 154 spin_lock_irqsave(&chan->lock, flags); 155 rc = virtqueue_get_buf(chan->vq, &len); 156 if (rc == NULL) { 157 spin_unlock_irqrestore(&chan->lock, flags); 158 break; 159 } 160 chan->ring_bufs_avail = 1; 161 spin_unlock_irqrestore(&chan->lock, flags); 162 /* Wakeup if anyone waiting for VirtIO ring space. */ 163 wake_up(chan->vc_wq); 164 p9_debug(P9_DEBUG_TRANS, ": rc %p\n", rc); 165 p9_debug(P9_DEBUG_TRANS, ": lookup tag %d\n", rc->tag); 166 req = p9_tag_lookup(chan->client, rc->tag); 167 req->status = REQ_STATUS_RCVD; 168 p9_client_cb(chan->client, req); 169 } 170 } 171 172 /** 173 * pack_sg_list - pack a scatter gather list from a linear buffer 174 * @sg: scatter/gather list to pack into 175 * @start: which segment of the sg_list to start at 176 * @limit: maximum segment to pack data to 177 * @data: data to pack into scatter/gather list 178 * @count: amount of data to pack into the scatter/gather list 179 * 180 * sg_lists have multiple segments of various sizes. This will pack 181 * arbitrary data into an existing scatter gather list, segmenting the 182 * data as necessary within constraints. 183 * 184 */ 185 186 static int pack_sg_list(struct scatterlist *sg, int start, 187 int limit, char *data, int count) 188 { 189 int s; 190 int index = start; 191 192 while (count) { 193 s = rest_of_page(data); 194 if (s > count) 195 s = count; 196 BUG_ON(index > limit); 197 sg_set_buf(&sg[index++], data, s); 198 count -= s; 199 data += s; 200 } 201 202 return index-start; 203 } 204 205 /* We don't currently allow canceling of virtio requests */ 206 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req) 207 { 208 return 1; 209 } 210 211 /** 212 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer, 213 * this takes a list of pages. 214 * @sg: scatter/gather list to pack into 215 * @start: which segment of the sg_list to start at 216 * @pdata: a list of pages to add into sg. 217 * @nr_pages: number of pages to pack into the scatter/gather list 218 * @data: data to pack into scatter/gather list 219 * @count: amount of data to pack into the scatter/gather list 220 */ 221 static int 222 pack_sg_list_p(struct scatterlist *sg, int start, int limit, 223 struct page **pdata, int nr_pages, char *data, int count) 224 { 225 int i = 0, s; 226 int data_off; 227 int index = start; 228 229 BUG_ON(nr_pages > (limit - start)); 230 /* 231 * if the first page doesn't start at 232 * page boundary find the offset 233 */ 234 data_off = offset_in_page(data); 235 while (nr_pages) { 236 s = rest_of_page(data); 237 if (s > count) 238 s = count; 239 sg_set_page(&sg[index++], pdata[i++], s, data_off); 240 data_off = 0; 241 data += s; 242 count -= s; 243 nr_pages--; 244 } 245 return index - start; 246 } 247 248 /** 249 * p9_virtio_request - issue a request 250 * @client: client instance issuing the request 251 * @req: request to be issued 252 * 253 */ 254 255 static int 256 p9_virtio_request(struct p9_client *client, struct p9_req_t *req) 257 { 258 int err; 259 int in, out; 260 unsigned long flags; 261 struct virtio_chan *chan = client->trans; 262 263 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n"); 264 265 req->status = REQ_STATUS_SENT; 266 req_retry: 267 spin_lock_irqsave(&chan->lock, flags); 268 269 /* Handle out VirtIO ring buffers */ 270 out = pack_sg_list(chan->sg, 0, 271 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size); 272 273 in = pack_sg_list(chan->sg, out, 274 VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity); 275 276 err = virtqueue_add_buf(chan->vq, chan->sg, out, in, req->tc, 277 GFP_ATOMIC); 278 if (err < 0) { 279 if (err == -ENOSPC) { 280 chan->ring_bufs_avail = 0; 281 spin_unlock_irqrestore(&chan->lock, flags); 282 err = wait_event_interruptible(*chan->vc_wq, 283 chan->ring_bufs_avail); 284 if (err == -ERESTARTSYS) 285 return err; 286 287 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n"); 288 goto req_retry; 289 } else { 290 spin_unlock_irqrestore(&chan->lock, flags); 291 p9_debug(P9_DEBUG_TRANS, 292 "virtio rpc add_buf returned failure\n"); 293 return -EIO; 294 } 295 } 296 virtqueue_kick(chan->vq); 297 spin_unlock_irqrestore(&chan->lock, flags); 298 299 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); 300 return 0; 301 } 302 303 static int p9_get_mapped_pages(struct virtio_chan *chan, 304 struct page **pages, char *data, 305 int nr_pages, int write, int kern_buf) 306 { 307 int err; 308 if (!kern_buf) { 309 /* 310 * We allow only p9_max_pages pinned. We wait for the 311 * Other zc request to finish here 312 */ 313 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { 314 err = wait_event_interruptible(vp_wq, 315 (atomic_read(&vp_pinned) < chan->p9_max_pages)); 316 if (err == -ERESTARTSYS) 317 return err; 318 } 319 err = p9_payload_gup(data, &nr_pages, pages, write); 320 if (err < 0) 321 return err; 322 atomic_add(nr_pages, &vp_pinned); 323 } else { 324 /* kernel buffer, no need to pin pages */ 325 int s, index = 0; 326 int count = nr_pages; 327 while (nr_pages) { 328 s = rest_of_page(data); 329 pages[index++] = kmap_to_page(data); 330 data += s; 331 nr_pages--; 332 } 333 nr_pages = count; 334 } 335 return nr_pages; 336 } 337 338 /** 339 * p9_virtio_zc_request - issue a zero copy request 340 * @client: client instance issuing the request 341 * @req: request to be issued 342 * @uidata: user bffer that should be ued for zero copy read 343 * @uodata: user buffer that shoud be user for zero copy write 344 * @inlen: read buffer size 345 * @olen: write buffer size 346 * @hdrlen: reader header size, This is the size of response protocol data 347 * 348 */ 349 static int 350 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req, 351 char *uidata, char *uodata, int inlen, 352 int outlen, int in_hdr_len, int kern_buf) 353 { 354 int in, out, err; 355 unsigned long flags; 356 int in_nr_pages = 0, out_nr_pages = 0; 357 struct page **in_pages = NULL, **out_pages = NULL; 358 struct virtio_chan *chan = client->trans; 359 360 p9_debug(P9_DEBUG_TRANS, "virtio request\n"); 361 362 if (uodata) { 363 out_nr_pages = p9_nr_pages(uodata, outlen); 364 out_pages = kmalloc(sizeof(struct page *) * out_nr_pages, 365 GFP_NOFS); 366 if (!out_pages) { 367 err = -ENOMEM; 368 goto err_out; 369 } 370 out_nr_pages = p9_get_mapped_pages(chan, out_pages, uodata, 371 out_nr_pages, 0, kern_buf); 372 if (out_nr_pages < 0) { 373 err = out_nr_pages; 374 kfree(out_pages); 375 out_pages = NULL; 376 goto err_out; 377 } 378 } 379 if (uidata) { 380 in_nr_pages = p9_nr_pages(uidata, inlen); 381 in_pages = kmalloc(sizeof(struct page *) * in_nr_pages, 382 GFP_NOFS); 383 if (!in_pages) { 384 err = -ENOMEM; 385 goto err_out; 386 } 387 in_nr_pages = p9_get_mapped_pages(chan, in_pages, uidata, 388 in_nr_pages, 1, kern_buf); 389 if (in_nr_pages < 0) { 390 err = in_nr_pages; 391 kfree(in_pages); 392 in_pages = NULL; 393 goto err_out; 394 } 395 } 396 req->status = REQ_STATUS_SENT; 397 req_retry_pinned: 398 spin_lock_irqsave(&chan->lock, flags); 399 /* out data */ 400 out = pack_sg_list(chan->sg, 0, 401 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size); 402 403 if (out_pages) 404 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM, 405 out_pages, out_nr_pages, uodata, outlen); 406 /* 407 * Take care of in data 408 * For example TREAD have 11. 409 * 11 is the read/write header = PDU Header(7) + IO Size (4). 410 * Arrange in such a way that server places header in the 411 * alloced memory and payload onto the user buffer. 412 */ 413 in = pack_sg_list(chan->sg, out, 414 VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len); 415 if (in_pages) 416 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM, 417 in_pages, in_nr_pages, uidata, inlen); 418 419 err = virtqueue_add_buf(chan->vq, chan->sg, out, in, req->tc, 420 GFP_ATOMIC); 421 if (err < 0) { 422 if (err == -ENOSPC) { 423 chan->ring_bufs_avail = 0; 424 spin_unlock_irqrestore(&chan->lock, flags); 425 err = wait_event_interruptible(*chan->vc_wq, 426 chan->ring_bufs_avail); 427 if (err == -ERESTARTSYS) 428 goto err_out; 429 430 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n"); 431 goto req_retry_pinned; 432 } else { 433 spin_unlock_irqrestore(&chan->lock, flags); 434 p9_debug(P9_DEBUG_TRANS, 435 "virtio rpc add_buf returned failure\n"); 436 err = -EIO; 437 goto err_out; 438 } 439 } 440 virtqueue_kick(chan->vq); 441 spin_unlock_irqrestore(&chan->lock, flags); 442 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); 443 err = wait_event_interruptible(*req->wq, 444 req->status >= REQ_STATUS_RCVD); 445 /* 446 * Non kernel buffers are pinned, unpin them 447 */ 448 err_out: 449 if (!kern_buf) { 450 if (in_pages) { 451 p9_release_pages(in_pages, in_nr_pages); 452 atomic_sub(in_nr_pages, &vp_pinned); 453 } 454 if (out_pages) { 455 p9_release_pages(out_pages, out_nr_pages); 456 atomic_sub(out_nr_pages, &vp_pinned); 457 } 458 /* wakeup anybody waiting for slots to pin pages */ 459 wake_up(&vp_wq); 460 } 461 kfree(in_pages); 462 kfree(out_pages); 463 return err; 464 } 465 466 static ssize_t p9_mount_tag_show(struct device *dev, 467 struct device_attribute *attr, char *buf) 468 { 469 struct virtio_chan *chan; 470 struct virtio_device *vdev; 471 472 vdev = dev_to_virtio(dev); 473 chan = vdev->priv; 474 475 return snprintf(buf, chan->tag_len + 1, "%s", chan->tag); 476 } 477 478 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL); 479 480 /** 481 * p9_virtio_probe - probe for existence of 9P virtio channels 482 * @vdev: virtio device to probe 483 * 484 * This probes for existing virtio channels. 485 * 486 */ 487 488 static int p9_virtio_probe(struct virtio_device *vdev) 489 { 490 __u16 tag_len; 491 char *tag; 492 int err; 493 struct virtio_chan *chan; 494 495 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL); 496 if (!chan) { 497 pr_err("Failed to allocate virtio 9P channel\n"); 498 err = -ENOMEM; 499 goto fail; 500 } 501 502 chan->vdev = vdev; 503 504 /* We expect one virtqueue, for requests. */ 505 chan->vq = virtio_find_single_vq(vdev, req_done, "requests"); 506 if (IS_ERR(chan->vq)) { 507 err = PTR_ERR(chan->vq); 508 goto out_free_vq; 509 } 510 chan->vq->vdev->priv = chan; 511 spin_lock_init(&chan->lock); 512 513 sg_init_table(chan->sg, VIRTQUEUE_NUM); 514 515 chan->inuse = false; 516 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) { 517 vdev->config->get(vdev, 518 offsetof(struct virtio_9p_config, tag_len), 519 &tag_len, sizeof(tag_len)); 520 } else { 521 err = -EINVAL; 522 goto out_free_vq; 523 } 524 tag = kmalloc(tag_len, GFP_KERNEL); 525 if (!tag) { 526 err = -ENOMEM; 527 goto out_free_vq; 528 } 529 vdev->config->get(vdev, offsetof(struct virtio_9p_config, tag), 530 tag, tag_len); 531 chan->tag = tag; 532 chan->tag_len = tag_len; 533 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr); 534 if (err) { 535 goto out_free_tag; 536 } 537 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL); 538 if (!chan->vc_wq) { 539 err = -ENOMEM; 540 goto out_free_tag; 541 } 542 init_waitqueue_head(chan->vc_wq); 543 chan->ring_bufs_avail = 1; 544 /* Ceiling limit to avoid denial of service attacks */ 545 chan->p9_max_pages = nr_free_buffer_pages()/4; 546 547 mutex_lock(&virtio_9p_lock); 548 list_add_tail(&chan->chan_list, &virtio_chan_list); 549 mutex_unlock(&virtio_9p_lock); 550 return 0; 551 552 out_free_tag: 553 kfree(tag); 554 out_free_vq: 555 vdev->config->del_vqs(vdev); 556 kfree(chan); 557 fail: 558 return err; 559 } 560 561 562 /** 563 * p9_virtio_create - allocate a new virtio channel 564 * @client: client instance invoking this transport 565 * @devname: string identifying the channel to connect to (unused) 566 * @args: args passed from sys_mount() for per-transport options (unused) 567 * 568 * This sets up a transport channel for 9p communication. Right now 569 * we only match the first available channel, but eventually we couldlook up 570 * alternate channels by matching devname versus a virtio_config entry. 571 * We use a simple reference count mechanism to ensure that only a single 572 * mount has a channel open at a time. 573 * 574 */ 575 576 static int 577 p9_virtio_create(struct p9_client *client, const char *devname, char *args) 578 { 579 struct virtio_chan *chan; 580 int ret = -ENOENT; 581 int found = 0; 582 583 mutex_lock(&virtio_9p_lock); 584 list_for_each_entry(chan, &virtio_chan_list, chan_list) { 585 if (!strncmp(devname, chan->tag, chan->tag_len) && 586 strlen(devname) == chan->tag_len) { 587 if (!chan->inuse) { 588 chan->inuse = true; 589 found = 1; 590 break; 591 } 592 ret = -EBUSY; 593 } 594 } 595 mutex_unlock(&virtio_9p_lock); 596 597 if (!found) { 598 pr_err("no channels available\n"); 599 return ret; 600 } 601 602 client->trans = (void *)chan; 603 client->status = Connected; 604 chan->client = client; 605 606 return 0; 607 } 608 609 /** 610 * p9_virtio_remove - clean up resources associated with a virtio device 611 * @vdev: virtio device to remove 612 * 613 */ 614 615 static void p9_virtio_remove(struct virtio_device *vdev) 616 { 617 struct virtio_chan *chan = vdev->priv; 618 619 if (chan->inuse) 620 p9_virtio_close(chan->client); 621 vdev->config->del_vqs(vdev); 622 623 mutex_lock(&virtio_9p_lock); 624 list_del(&chan->chan_list); 625 mutex_unlock(&virtio_9p_lock); 626 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr); 627 kfree(chan->tag); 628 kfree(chan->vc_wq); 629 kfree(chan); 630 631 } 632 633 static struct virtio_device_id id_table[] = { 634 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID }, 635 { 0 }, 636 }; 637 638 static unsigned int features[] = { 639 VIRTIO_9P_MOUNT_TAG, 640 }; 641 642 /* The standard "struct lguest_driver": */ 643 static struct virtio_driver p9_virtio_drv = { 644 .feature_table = features, 645 .feature_table_size = ARRAY_SIZE(features), 646 .driver.name = KBUILD_MODNAME, 647 .driver.owner = THIS_MODULE, 648 .id_table = id_table, 649 .probe = p9_virtio_probe, 650 .remove = p9_virtio_remove, 651 }; 652 653 static struct p9_trans_module p9_virtio_trans = { 654 .name = "virtio", 655 .create = p9_virtio_create, 656 .close = p9_virtio_close, 657 .request = p9_virtio_request, 658 .zc_request = p9_virtio_zc_request, 659 .cancel = p9_virtio_cancel, 660 /* 661 * We leave one entry for input and one entry for response 662 * headers. We also skip one more entry to accomodate, address 663 * that are not at page boundary, that can result in an extra 664 * page in zero copy. 665 */ 666 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3), 667 .def = 0, 668 .owner = THIS_MODULE, 669 }; 670 671 /* The standard init function */ 672 static int __init p9_virtio_init(void) 673 { 674 INIT_LIST_HEAD(&virtio_chan_list); 675 676 v9fs_register_trans(&p9_virtio_trans); 677 return register_virtio_driver(&p9_virtio_drv); 678 } 679 680 static void __exit p9_virtio_cleanup(void) 681 { 682 unregister_virtio_driver(&p9_virtio_drv); 683 v9fs_unregister_trans(&p9_virtio_trans); 684 } 685 686 module_init(p9_virtio_init); 687 module_exit(p9_virtio_cleanup); 688 689 MODULE_DEVICE_TABLE(virtio, id_table); 690 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>"); 691 MODULE_DESCRIPTION("Virtio 9p Transport"); 692 MODULE_LICENSE("GPL"); 693