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 #include <linux/in.h> 30 #include <linux/module.h> 31 #include <linux/net.h> 32 #include <linux/ipv6.h> 33 #include <linux/errno.h> 34 #include <linux/kernel.h> 35 #include <linux/un.h> 36 #include <linux/uaccess.h> 37 #include <linux/inet.h> 38 #include <linux/idr.h> 39 #include <linux/file.h> 40 #include <linux/slab.h> 41 #include <net/9p/9p.h> 42 #include <linux/parser.h> 43 #include <net/9p/client.h> 44 #include <net/9p/transport.h> 45 #include <linux/scatterlist.h> 46 #include <linux/swap.h> 47 #include <linux/virtio.h> 48 #include <linux/virtio_9p.h> 49 #include "trans_common.h" 50 51 #define VIRTQUEUE_NUM 128 52 53 /* a single mutex to manage channel initialization and attachment */ 54 static DEFINE_MUTEX(virtio_9p_lock); 55 static DECLARE_WAIT_QUEUE_HEAD(vp_wq); 56 static atomic_t vp_pinned = ATOMIC_INIT(0); 57 58 /** 59 * struct virtio_chan - per-instance transport information 60 * @initialized: whether the channel is initialized 61 * @inuse: whether the channel is in use 62 * @lock: protects multiple elements within this structure 63 * @client: client instance 64 * @vdev: virtio dev associated with this channel 65 * @vq: virtio queue associated with this channel 66 * @sg: scatter gather list which is used to pack a request (protected?) 67 * 68 * We keep all per-channel information in a structure. 69 * This structure is allocated within the devices dev->mem space. 70 * A pointer to the structure will get put in the transport private. 71 * 72 */ 73 74 struct virtio_chan { 75 bool inuse; 76 77 spinlock_t lock; 78 79 struct p9_client *client; 80 struct virtio_device *vdev; 81 struct virtqueue *vq; 82 int ring_bufs_avail; 83 wait_queue_head_t *vc_wq; 84 /* This is global limit. Since we don't have a global structure, 85 * will be placing it in each channel. 86 */ 87 int p9_max_pages; 88 /* Scatterlist: can be too big for stack. */ 89 struct scatterlist sg[VIRTQUEUE_NUM]; 90 91 int tag_len; 92 /* 93 * tag name to identify a mount Non-null terminated 94 */ 95 char *tag; 96 97 struct list_head chan_list; 98 }; 99 100 static struct list_head virtio_chan_list; 101 102 /* How many bytes left in this page. */ 103 static unsigned int rest_of_page(void *data) 104 { 105 return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE); 106 } 107 108 /** 109 * p9_virtio_close - reclaim resources of a channel 110 * @client: client instance 111 * 112 * This reclaims a channel by freeing its resources and 113 * reseting its inuse flag. 114 * 115 */ 116 117 static void p9_virtio_close(struct p9_client *client) 118 { 119 struct virtio_chan *chan = client->trans; 120 121 mutex_lock(&virtio_9p_lock); 122 if (chan) 123 chan->inuse = false; 124 mutex_unlock(&virtio_9p_lock); 125 } 126 127 /** 128 * req_done - callback which signals activity from the server 129 * @vq: virtio queue activity was received on 130 * 131 * This notifies us that the server has triggered some activity 132 * on the virtio channel - most likely a response to request we 133 * sent. Figure out which requests now have responses and wake up 134 * those threads. 135 * 136 * Bugs: could do with some additional sanity checking, but appears to work. 137 * 138 */ 139 140 static void req_done(struct virtqueue *vq) 141 { 142 struct virtio_chan *chan = vq->vdev->priv; 143 struct p9_fcall *rc; 144 unsigned int len; 145 struct p9_req_t *req; 146 unsigned long flags; 147 148 P9_DPRINTK(P9_DEBUG_TRANS, ": request done\n"); 149 150 while (1) { 151 spin_lock_irqsave(&chan->lock, flags); 152 rc = virtqueue_get_buf(chan->vq, &len); 153 if (rc == NULL) { 154 spin_unlock_irqrestore(&chan->lock, flags); 155 break; 156 } 157 chan->ring_bufs_avail = 1; 158 spin_unlock_irqrestore(&chan->lock, flags); 159 /* Wakeup if anyone waiting for VirtIO ring space. */ 160 wake_up(chan->vc_wq); 161 P9_DPRINTK(P9_DEBUG_TRANS, ": rc %p\n", rc); 162 P9_DPRINTK(P9_DEBUG_TRANS, ": lookup tag %d\n", rc->tag); 163 req = p9_tag_lookup(chan->client, rc->tag); 164 req->status = REQ_STATUS_RCVD; 165 p9_client_cb(chan->client, req); 166 } 167 } 168 169 /** 170 * pack_sg_list - pack a scatter gather list from a linear buffer 171 * @sg: scatter/gather list to pack into 172 * @start: which segment of the sg_list to start at 173 * @limit: maximum segment to pack data to 174 * @data: data to pack into scatter/gather list 175 * @count: amount of data to pack into the scatter/gather list 176 * 177 * sg_lists have multiple segments of various sizes. This will pack 178 * arbitrary data into an existing scatter gather list, segmenting the 179 * data as necessary within constraints. 180 * 181 */ 182 183 static int pack_sg_list(struct scatterlist *sg, int start, 184 int limit, char *data, int count) 185 { 186 int s; 187 int index = start; 188 189 while (count) { 190 s = rest_of_page(data); 191 if (s > count) 192 s = count; 193 sg_set_buf(&sg[index++], data, s); 194 count -= s; 195 data += s; 196 BUG_ON(index > limit); 197 } 198 199 return index-start; 200 } 201 202 /* We don't currently allow canceling of virtio requests */ 203 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req) 204 { 205 return 1; 206 } 207 208 /** 209 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer, 210 * this takes a list of pages. 211 * @sg: scatter/gather list to pack into 212 * @start: which segment of the sg_list to start at 213 * @**pdata: a list of pages to add into sg. 214 * @nr_pages: number of pages to pack into the scatter/gather list 215 * @data: data to pack into scatter/gather list 216 * @count: amount of data to pack into the scatter/gather list 217 */ 218 static int 219 pack_sg_list_p(struct scatterlist *sg, int start, int limit, 220 struct page **pdata, int nr_pages, char *data, int count) 221 { 222 int i = 0, s; 223 int data_off; 224 int index = start; 225 226 BUG_ON(nr_pages > (limit - start)); 227 /* 228 * if the first page doesn't start at 229 * page boundary find the offset 230 */ 231 data_off = offset_in_page(data); 232 while (nr_pages) { 233 s = rest_of_page(data); 234 if (s > count) 235 s = count; 236 sg_set_page(&sg[index++], pdata[i++], s, data_off); 237 data_off = 0; 238 data += s; 239 count -= s; 240 nr_pages--; 241 } 242 return index - start; 243 } 244 245 /** 246 * p9_virtio_request - issue a request 247 * @client: client instance issuing the request 248 * @req: request to be issued 249 * 250 */ 251 252 static int 253 p9_virtio_request(struct p9_client *client, struct p9_req_t *req) 254 { 255 int err; 256 int in, out; 257 unsigned long flags; 258 struct virtio_chan *chan = client->trans; 259 260 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request\n"); 261 262 req->status = REQ_STATUS_SENT; 263 req_retry: 264 spin_lock_irqsave(&chan->lock, flags); 265 266 /* Handle out VirtIO ring buffers */ 267 out = pack_sg_list(chan->sg, 0, 268 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size); 269 270 in = pack_sg_list(chan->sg, out, 271 VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity); 272 273 err = virtqueue_add_buf(chan->vq, chan->sg, out, in, req->tc); 274 if (err < 0) { 275 if (err == -ENOSPC) { 276 chan->ring_bufs_avail = 0; 277 spin_unlock_irqrestore(&chan->lock, flags); 278 err = wait_event_interruptible(*chan->vc_wq, 279 chan->ring_bufs_avail); 280 if (err == -ERESTARTSYS) 281 return err; 282 283 P9_DPRINTK(P9_DEBUG_TRANS, "9p:Retry virtio request\n"); 284 goto req_retry; 285 } else { 286 spin_unlock_irqrestore(&chan->lock, flags); 287 P9_DPRINTK(P9_DEBUG_TRANS, 288 "9p debug: " 289 "virtio rpc add_buf returned failure"); 290 return -EIO; 291 } 292 } 293 virtqueue_kick(chan->vq); 294 spin_unlock_irqrestore(&chan->lock, flags); 295 296 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request kicked\n"); 297 return 0; 298 } 299 300 static int p9_get_mapped_pages(struct virtio_chan *chan, 301 struct page **pages, char *data, 302 int nr_pages, int write, int kern_buf) 303 { 304 int err; 305 if (!kern_buf) { 306 /* 307 * We allow only p9_max_pages pinned. We wait for the 308 * Other zc request to finish here 309 */ 310 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { 311 err = wait_event_interruptible(vp_wq, 312 (atomic_read(&vp_pinned) < chan->p9_max_pages)); 313 if (err == -ERESTARTSYS) 314 return err; 315 } 316 err = p9_payload_gup(data, &nr_pages, pages, write); 317 if (err < 0) 318 return err; 319 atomic_add(nr_pages, &vp_pinned); 320 } else { 321 /* kernel buffer, no need to pin pages */ 322 int s, index = 0; 323 int count = nr_pages; 324 while (nr_pages) { 325 s = rest_of_page(data); 326 pages[index++] = virt_to_page(data); 327 data += s; 328 nr_pages--; 329 } 330 nr_pages = count; 331 } 332 return nr_pages; 333 } 334 335 /** 336 * p9_virtio_zc_request - issue a zero copy request 337 * @client: client instance issuing the request 338 * @req: request to be issued 339 * @uidata: user bffer that should be ued for zero copy read 340 * @uodata: user buffer that shoud be user for zero copy write 341 * @inlen: read buffer size 342 * @olen: write buffer size 343 * @hdrlen: reader header size, This is the size of response protocol data 344 * 345 */ 346 static int 347 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req, 348 char *uidata, char *uodata, int inlen, 349 int outlen, int in_hdr_len, int kern_buf) 350 { 351 int in, out, err; 352 unsigned long flags; 353 int in_nr_pages = 0, out_nr_pages = 0; 354 struct page **in_pages = NULL, **out_pages = NULL; 355 struct virtio_chan *chan = client->trans; 356 357 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request\n"); 358 359 if (uodata) { 360 out_nr_pages = p9_nr_pages(uodata, outlen); 361 out_pages = kmalloc(sizeof(struct page *) * out_nr_pages, 362 GFP_NOFS); 363 if (!out_pages) { 364 err = -ENOMEM; 365 goto err_out; 366 } 367 out_nr_pages = p9_get_mapped_pages(chan, out_pages, uodata, 368 out_nr_pages, 0, kern_buf); 369 if (out_nr_pages < 0) { 370 err = out_nr_pages; 371 kfree(out_pages); 372 out_pages = NULL; 373 goto err_out; 374 } 375 } 376 if (uidata) { 377 in_nr_pages = p9_nr_pages(uidata, inlen); 378 in_pages = kmalloc(sizeof(struct page *) * in_nr_pages, 379 GFP_NOFS); 380 if (!in_pages) { 381 err = -ENOMEM; 382 goto err_out; 383 } 384 in_nr_pages = p9_get_mapped_pages(chan, in_pages, uidata, 385 in_nr_pages, 1, kern_buf); 386 if (in_nr_pages < 0) { 387 err = in_nr_pages; 388 kfree(in_pages); 389 in_pages = NULL; 390 goto err_out; 391 } 392 } 393 req->status = REQ_STATUS_SENT; 394 req_retry_pinned: 395 spin_lock_irqsave(&chan->lock, flags); 396 /* out data */ 397 out = pack_sg_list(chan->sg, 0, 398 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size); 399 400 if (out_pages) 401 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM, 402 out_pages, out_nr_pages, uodata, outlen); 403 /* 404 * Take care of in data 405 * For example TREAD have 11. 406 * 11 is the read/write header = PDU Header(7) + IO Size (4). 407 * Arrange in such a way that server places header in the 408 * alloced memory and payload onto the user buffer. 409 */ 410 in = pack_sg_list(chan->sg, out, 411 VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len); 412 if (in_pages) 413 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM, 414 in_pages, in_nr_pages, uidata, inlen); 415 416 err = virtqueue_add_buf(chan->vq, chan->sg, out, in, req->tc); 417 if (err < 0) { 418 if (err == -ENOSPC) { 419 chan->ring_bufs_avail = 0; 420 spin_unlock_irqrestore(&chan->lock, flags); 421 err = wait_event_interruptible(*chan->vc_wq, 422 chan->ring_bufs_avail); 423 if (err == -ERESTARTSYS) 424 goto err_out; 425 426 P9_DPRINTK(P9_DEBUG_TRANS, "9p:Retry virtio request\n"); 427 goto req_retry_pinned; 428 } else { 429 spin_unlock_irqrestore(&chan->lock, flags); 430 P9_DPRINTK(P9_DEBUG_TRANS, 431 "9p debug: " 432 "virtio rpc add_buf returned failure"); 433 err = -EIO; 434 goto err_out; 435 } 436 } 437 virtqueue_kick(chan->vq); 438 spin_unlock_irqrestore(&chan->lock, flags); 439 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request kicked\n"); 440 err = wait_event_interruptible(*req->wq, 441 req->status >= REQ_STATUS_RCVD); 442 /* 443 * Non kernel buffers are pinned, unpin them 444 */ 445 err_out: 446 if (!kern_buf) { 447 if (in_pages) { 448 p9_release_pages(in_pages, in_nr_pages); 449 atomic_sub(in_nr_pages, &vp_pinned); 450 } 451 if (out_pages) { 452 p9_release_pages(out_pages, out_nr_pages); 453 atomic_sub(out_nr_pages, &vp_pinned); 454 } 455 /* wakeup anybody waiting for slots to pin pages */ 456 wake_up(&vp_wq); 457 } 458 kfree(in_pages); 459 kfree(out_pages); 460 return err; 461 } 462 463 static ssize_t p9_mount_tag_show(struct device *dev, 464 struct device_attribute *attr, char *buf) 465 { 466 struct virtio_chan *chan; 467 struct virtio_device *vdev; 468 469 vdev = dev_to_virtio(dev); 470 chan = vdev->priv; 471 472 return snprintf(buf, chan->tag_len + 1, "%s", chan->tag); 473 } 474 475 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL); 476 477 /** 478 * p9_virtio_probe - probe for existence of 9P virtio channels 479 * @vdev: virtio device to probe 480 * 481 * This probes for existing virtio channels. 482 * 483 */ 484 485 static int p9_virtio_probe(struct virtio_device *vdev) 486 { 487 __u16 tag_len; 488 char *tag; 489 int err; 490 struct virtio_chan *chan; 491 492 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL); 493 if (!chan) { 494 printk(KERN_ERR "9p: Failed to allocate virtio 9P channel\n"); 495 err = -ENOMEM; 496 goto fail; 497 } 498 499 chan->vdev = vdev; 500 501 /* We expect one virtqueue, for requests. */ 502 chan->vq = virtio_find_single_vq(vdev, req_done, "requests"); 503 if (IS_ERR(chan->vq)) { 504 err = PTR_ERR(chan->vq); 505 goto out_free_vq; 506 } 507 chan->vq->vdev->priv = chan; 508 spin_lock_init(&chan->lock); 509 510 sg_init_table(chan->sg, VIRTQUEUE_NUM); 511 512 chan->inuse = false; 513 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) { 514 vdev->config->get(vdev, 515 offsetof(struct virtio_9p_config, tag_len), 516 &tag_len, sizeof(tag_len)); 517 } else { 518 err = -EINVAL; 519 goto out_free_vq; 520 } 521 tag = kmalloc(tag_len, GFP_KERNEL); 522 if (!tag) { 523 err = -ENOMEM; 524 goto out_free_vq; 525 } 526 vdev->config->get(vdev, offsetof(struct virtio_9p_config, tag), 527 tag, tag_len); 528 chan->tag = tag; 529 chan->tag_len = tag_len; 530 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr); 531 if (err) { 532 goto out_free_tag; 533 } 534 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL); 535 if (!chan->vc_wq) { 536 err = -ENOMEM; 537 goto out_free_tag; 538 } 539 init_waitqueue_head(chan->vc_wq); 540 chan->ring_bufs_avail = 1; 541 /* Ceiling limit to avoid denial of service attacks */ 542 chan->p9_max_pages = nr_free_buffer_pages()/4; 543 544 mutex_lock(&virtio_9p_lock); 545 list_add_tail(&chan->chan_list, &virtio_chan_list); 546 mutex_unlock(&virtio_9p_lock); 547 return 0; 548 549 out_free_tag: 550 kfree(tag); 551 out_free_vq: 552 vdev->config->del_vqs(vdev); 553 kfree(chan); 554 fail: 555 return err; 556 } 557 558 559 /** 560 * p9_virtio_create - allocate a new virtio channel 561 * @client: client instance invoking this transport 562 * @devname: string identifying the channel to connect to (unused) 563 * @args: args passed from sys_mount() for per-transport options (unused) 564 * 565 * This sets up a transport channel for 9p communication. Right now 566 * we only match the first available channel, but eventually we couldlook up 567 * alternate channels by matching devname versus a virtio_config entry. 568 * We use a simple reference count mechanism to ensure that only a single 569 * mount has a channel open at a time. 570 * 571 */ 572 573 static int 574 p9_virtio_create(struct p9_client *client, const char *devname, char *args) 575 { 576 struct virtio_chan *chan; 577 int ret = -ENOENT; 578 int found = 0; 579 580 mutex_lock(&virtio_9p_lock); 581 list_for_each_entry(chan, &virtio_chan_list, chan_list) { 582 if (!strncmp(devname, chan->tag, chan->tag_len) && 583 strlen(devname) == chan->tag_len) { 584 if (!chan->inuse) { 585 chan->inuse = true; 586 found = 1; 587 break; 588 } 589 ret = -EBUSY; 590 } 591 } 592 mutex_unlock(&virtio_9p_lock); 593 594 if (!found) { 595 printk(KERN_ERR "9p: no channels available\n"); 596 return ret; 597 } 598 599 client->trans = (void *)chan; 600 client->status = Connected; 601 chan->client = client; 602 603 return 0; 604 } 605 606 /** 607 * p9_virtio_remove - clean up resources associated with a virtio device 608 * @vdev: virtio device to remove 609 * 610 */ 611 612 static void p9_virtio_remove(struct virtio_device *vdev) 613 { 614 struct virtio_chan *chan = vdev->priv; 615 616 BUG_ON(chan->inuse); 617 vdev->config->del_vqs(vdev); 618 619 mutex_lock(&virtio_9p_lock); 620 list_del(&chan->chan_list); 621 mutex_unlock(&virtio_9p_lock); 622 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr); 623 kfree(chan->tag); 624 kfree(chan->vc_wq); 625 kfree(chan); 626 627 } 628 629 static struct virtio_device_id id_table[] = { 630 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID }, 631 { 0 }, 632 }; 633 634 static unsigned int features[] = { 635 VIRTIO_9P_MOUNT_TAG, 636 }; 637 638 /* The standard "struct lguest_driver": */ 639 static struct virtio_driver p9_virtio_drv = { 640 .feature_table = features, 641 .feature_table_size = ARRAY_SIZE(features), 642 .driver.name = KBUILD_MODNAME, 643 .driver.owner = THIS_MODULE, 644 .id_table = id_table, 645 .probe = p9_virtio_probe, 646 .remove = p9_virtio_remove, 647 }; 648 649 static struct p9_trans_module p9_virtio_trans = { 650 .name = "virtio", 651 .create = p9_virtio_create, 652 .close = p9_virtio_close, 653 .request = p9_virtio_request, 654 .zc_request = p9_virtio_zc_request, 655 .cancel = p9_virtio_cancel, 656 /* 657 * We leave one entry for input and one entry for response 658 * headers. We also skip one more entry to accomodate, address 659 * that are not at page boundary, that can result in an extra 660 * page in zero copy. 661 */ 662 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3), 663 .def = 0, 664 .owner = THIS_MODULE, 665 }; 666 667 /* The standard init function */ 668 static int __init p9_virtio_init(void) 669 { 670 INIT_LIST_HEAD(&virtio_chan_list); 671 672 v9fs_register_trans(&p9_virtio_trans); 673 return register_virtio_driver(&p9_virtio_drv); 674 } 675 676 static void __exit p9_virtio_cleanup(void) 677 { 678 unregister_virtio_driver(&p9_virtio_drv); 679 v9fs_unregister_trans(&p9_virtio_trans); 680 } 681 682 module_init(p9_virtio_init); 683 module_exit(p9_virtio_cleanup); 684 685 MODULE_DEVICE_TABLE(virtio, id_table); 686 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>"); 687 MODULE_DESCRIPTION("Virtio 9p Transport"); 688 MODULE_LICENSE("GPL"); 689