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