Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * The Virtio 9p transport driver
4 *
5 * This is a block based transport driver based on the lguest block driver
6 * code.
7 *
8 * Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation
9 *
10 * Based on virtio console driver
11 * Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
12 */
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16#include <linux/in.h>
17#include <linux/module.h>
18#include <linux/net.h>
19#include <linux/ipv6.h>
20#include <linux/errno.h>
21#include <linux/kernel.h>
22#include <linux/un.h>
23#include <linux/uaccess.h>
24#include <linux/inet.h>
25#include <linux/idr.h>
26#include <linux/file.h>
27#include <linux/highmem.h>
28#include <linux/slab.h>
29#include <net/9p/9p.h>
30#include <linux/parser.h>
31#include <net/9p/client.h>
32#include <net/9p/transport.h>
33#include <linux/scatterlist.h>
34#include <linux/swap.h>
35#include <linux/virtio.h>
36#include <linux/virtio_9p.h>
37#include "trans_common.h"
38
39#define VIRTQUEUE_NUM 128
40
41/* a single mutex to manage channel initialization and attachment */
42static DEFINE_MUTEX(virtio_9p_lock);
43static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
44static atomic_t vp_pinned = ATOMIC_INIT(0);
45
46/**
47 * struct virtio_chan - per-instance transport information
48 * @inuse: whether the channel is in use
49 * @lock: protects multiple elements within this structure
50 * @client: client instance
51 * @vdev: virtio dev associated with this channel
52 * @vq: virtio queue associated with this channel
53 * @ring_bufs_avail: flag to indicate there is some available in the ring buf
54 * @vc_wq: wait queue for waiting for thing to be added to ring buf
55 * @p9_max_pages: maximum number of pinned pages
56 * @sg: scatter gather list which is used to pack a request (protected?)
57 * @chan_list: linked list of channels
58 *
59 * We keep all per-channel information in a structure.
60 * This structure is allocated within the devices dev->mem space.
61 * A pointer to the structure will get put in the transport private.
62 *
63 */
64
65struct virtio_chan {
66 bool inuse;
67
68 spinlock_t lock;
69
70 struct p9_client *client;
71 struct virtio_device *vdev;
72 struct virtqueue *vq;
73 int ring_bufs_avail;
74 wait_queue_head_t *vc_wq;
75 /* This is global limit. Since we don't have a global structure,
76 * will be placing it in each channel.
77 */
78 unsigned long p9_max_pages;
79 /* Scatterlist: can be too big for stack. */
80 struct scatterlist sg[VIRTQUEUE_NUM];
81 /**
82 * @tag: name to identify a mount null terminated
83 */
84 char *tag;
85
86 struct list_head chan_list;
87};
88
89static struct list_head virtio_chan_list;
90
91/* How many bytes left in this page. */
92static unsigned int rest_of_page(void *data)
93{
94 return PAGE_SIZE - offset_in_page(data);
95}
96
97/**
98 * p9_virtio_close - reclaim resources of a channel
99 * @client: client instance
100 *
101 * This reclaims a channel by freeing its resources and
102 * resetting its inuse flag.
103 *
104 */
105
106static void p9_virtio_close(struct p9_client *client)
107{
108 struct virtio_chan *chan = client->trans;
109
110 mutex_lock(&virtio_9p_lock);
111 if (chan)
112 chan->inuse = false;
113 mutex_unlock(&virtio_9p_lock);
114}
115
116/**
117 * req_done - callback which signals activity from the server
118 * @vq: virtio queue activity was received on
119 *
120 * This notifies us that the server has triggered some activity
121 * on the virtio channel - most likely a response to request we
122 * sent. Figure out which requests now have responses and wake up
123 * those threads.
124 *
125 * Bugs: could do with some additional sanity checking, but appears to work.
126 *
127 */
128
129static void req_done(struct virtqueue *vq)
130{
131 struct virtio_chan *chan = vq->vdev->priv;
132 unsigned int len;
133 struct p9_req_t *req;
134 bool need_wakeup = false;
135 unsigned long flags;
136
137 p9_debug(P9_DEBUG_TRANS, ": request done\n");
138
139 spin_lock_irqsave(&chan->lock, flags);
140 while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) {
141 if (!chan->ring_bufs_avail) {
142 chan->ring_bufs_avail = 1;
143 need_wakeup = true;
144 }
145
146 if (len) {
147 req->rc.size = len;
148 p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
149 }
150 }
151 spin_unlock_irqrestore(&chan->lock, flags);
152 /* Wakeup if anyone waiting for VirtIO ring space. */
153 if (need_wakeup)
154 wake_up(chan->vc_wq);
155}
156
157/**
158 * pack_sg_list - pack a scatter gather list from a linear buffer
159 * @sg: scatter/gather list to pack into
160 * @start: which segment of the sg_list to start at
161 * @limit: maximum segment to pack data to
162 * @data: data to pack into scatter/gather list
163 * @count: amount of data to pack into the scatter/gather list
164 *
165 * sg_lists have multiple segments of various sizes. This will pack
166 * arbitrary data into an existing scatter gather list, segmenting the
167 * data as necessary within constraints.
168 *
169 */
170
171static int pack_sg_list(struct scatterlist *sg, int start,
172 int limit, char *data, int count)
173{
174 int s;
175 int index = start;
176
177 while (count) {
178 s = rest_of_page(data);
179 if (s > count)
180 s = count;
181 BUG_ON(index >= limit);
182 /* Make sure we don't terminate early. */
183 sg_unmark_end(&sg[index]);
184 sg_set_buf(&sg[index++], data, s);
185 count -= s;
186 data += s;
187 }
188 if (index-start)
189 sg_mark_end(&sg[index - 1]);
190 return index-start;
191}
192
193/* We don't currently allow canceling of virtio requests */
194static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
195{
196 return 1;
197}
198
199/* Reply won't come, so drop req ref */
200static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req)
201{
202 p9_req_put(req);
203 return 0;
204}
205
206/**
207 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
208 * this takes a list of pages.
209 * @sg: scatter/gather list to pack into
210 * @start: which segment of the sg_list to start at
211 * @limit: maximum number of pages in sg list.
212 * @pdata: a list of pages to add into sg.
213 * @nr_pages: number of pages to pack into the scatter/gather list
214 * @offs: amount of data in the beginning of first page _not_ to pack
215 * @count: amount of data to pack into the scatter/gather list
216 */
217static int
218pack_sg_list_p(struct scatterlist *sg, int start, int limit,
219 struct page **pdata, int nr_pages, size_t offs, int count)
220{
221 int i = 0, s;
222 int data_off = offs;
223 int index = start;
224
225 BUG_ON(nr_pages > (limit - start));
226 /*
227 * if the first page doesn't start at
228 * page boundary find the offset
229 */
230 while (nr_pages) {
231 s = PAGE_SIZE - data_off;
232 if (s > count)
233 s = count;
234 BUG_ON(index >= limit);
235 /* Make sure we don't terminate early. */
236 sg_unmark_end(&sg[index]);
237 sg_set_page(&sg[index++], pdata[i++], s, data_off);
238 data_off = 0;
239 count -= s;
240 nr_pages--;
241 }
242
243 if (index-start)
244 sg_mark_end(&sg[index - 1]);
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
255static int
256p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
257{
258 int err;
259 int in, out, out_sgs, in_sgs;
260 unsigned long flags;
261 struct virtio_chan *chan = client->trans;
262 struct scatterlist *sgs[2];
263
264 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
265
266 req->status = REQ_STATUS_SENT;
267req_retry:
268 spin_lock_irqsave(&chan->lock, flags);
269
270 out_sgs = in_sgs = 0;
271 /* Handle out VirtIO ring buffers */
272 out = pack_sg_list(chan->sg, 0,
273 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
274 if (out)
275 sgs[out_sgs++] = chan->sg;
276
277 in = pack_sg_list(chan->sg, out,
278 VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity);
279 if (in)
280 sgs[out_sgs + in_sgs++] = chan->sg + out;
281
282 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
283 GFP_ATOMIC);
284 if (err < 0) {
285 if (err == -ENOSPC) {
286 chan->ring_bufs_avail = 0;
287 spin_unlock_irqrestore(&chan->lock, flags);
288 err = wait_event_killable(*chan->vc_wq,
289 chan->ring_bufs_avail);
290 if (err == -ERESTARTSYS)
291 return err;
292
293 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
294 goto req_retry;
295 } else {
296 spin_unlock_irqrestore(&chan->lock, flags);
297 p9_debug(P9_DEBUG_TRANS,
298 "virtio rpc add_sgs returned failure\n");
299 return -EIO;
300 }
301 }
302 virtqueue_kick(chan->vq);
303 spin_unlock_irqrestore(&chan->lock, flags);
304
305 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
306 return 0;
307}
308
309static int p9_get_mapped_pages(struct virtio_chan *chan,
310 struct page ***pages,
311 struct iov_iter *data,
312 int count,
313 size_t *offs,
314 int *need_drop)
315{
316 int nr_pages;
317 int err;
318
319 if (!iov_iter_count(data))
320 return 0;
321
322 if (!iov_iter_is_kvec(data)) {
323 int n;
324 /*
325 * We allow only p9_max_pages pinned. We wait for the
326 * Other zc request to finish here
327 */
328 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
329 err = wait_event_killable(vp_wq,
330 (atomic_read(&vp_pinned) < chan->p9_max_pages));
331 if (err == -ERESTARTSYS)
332 return err;
333 }
334 n = iov_iter_get_pages_alloc(data, pages, count, offs);
335 if (n < 0)
336 return n;
337 *need_drop = 1;
338 nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
339 atomic_add(nr_pages, &vp_pinned);
340 return n;
341 } else {
342 /* kernel buffer, no need to pin pages */
343 int index;
344 size_t len;
345 void *p;
346
347 /* we'd already checked that it's non-empty */
348 while (1) {
349 len = iov_iter_single_seg_count(data);
350 if (likely(len)) {
351 p = data->kvec->iov_base + data->iov_offset;
352 break;
353 }
354 iov_iter_advance(data, 0);
355 }
356 if (len > count)
357 len = count;
358
359 nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
360 (unsigned long)p / PAGE_SIZE;
361
362 *pages = kmalloc_array(nr_pages, sizeof(struct page *),
363 GFP_NOFS);
364 if (!*pages)
365 return -ENOMEM;
366
367 *need_drop = 0;
368 p -= (*offs = offset_in_page(p));
369 for (index = 0; index < nr_pages; index++) {
370 if (is_vmalloc_addr(p))
371 (*pages)[index] = vmalloc_to_page(p);
372 else
373 (*pages)[index] = kmap_to_page(p);
374 p += PAGE_SIZE;
375 }
376 return len;
377 }
378}
379
380/**
381 * p9_virtio_zc_request - issue a zero copy request
382 * @client: client instance issuing the request
383 * @req: request to be issued
384 * @uidata: user buffer that should be used for zero copy read
385 * @uodata: user buffer that should be used for zero copy write
386 * @inlen: read buffer size
387 * @outlen: write buffer size
388 * @in_hdr_len: reader header size, This is the size of response protocol data
389 *
390 */
391static int
392p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
393 struct iov_iter *uidata, struct iov_iter *uodata,
394 int inlen, int outlen, int in_hdr_len)
395{
396 int in, out, err, out_sgs, in_sgs;
397 unsigned long flags;
398 int in_nr_pages = 0, out_nr_pages = 0;
399 struct page **in_pages = NULL, **out_pages = NULL;
400 struct virtio_chan *chan = client->trans;
401 struct scatterlist *sgs[4];
402 size_t offs;
403 int need_drop = 0;
404 int kicked = 0;
405
406 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
407
408 if (uodata) {
409 __le32 sz;
410 int n = p9_get_mapped_pages(chan, &out_pages, uodata,
411 outlen, &offs, &need_drop);
412 if (n < 0) {
413 err = n;
414 goto err_out;
415 }
416 out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
417 if (n != outlen) {
418 __le32 v = cpu_to_le32(n);
419 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
420 outlen = n;
421 }
422 /* The size field of the message must include the length of the
423 * header and the length of the data. We didn't actually know
424 * the length of the data until this point so add it in now.
425 */
426 sz = cpu_to_le32(req->tc.size + outlen);
427 memcpy(&req->tc.sdata[0], &sz, sizeof(sz));
428 } else if (uidata) {
429 int n = p9_get_mapped_pages(chan, &in_pages, uidata,
430 inlen, &offs, &need_drop);
431 if (n < 0) {
432 err = n;
433 goto err_out;
434 }
435 in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
436 if (n != inlen) {
437 __le32 v = cpu_to_le32(n);
438 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
439 inlen = n;
440 }
441 }
442 req->status = REQ_STATUS_SENT;
443req_retry_pinned:
444 spin_lock_irqsave(&chan->lock, flags);
445
446 out_sgs = in_sgs = 0;
447
448 /* out data */
449 out = pack_sg_list(chan->sg, 0,
450 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
451
452 if (out)
453 sgs[out_sgs++] = chan->sg;
454
455 if (out_pages) {
456 sgs[out_sgs++] = chan->sg + out;
457 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
458 out_pages, out_nr_pages, offs, outlen);
459 }
460
461 /*
462 * Take care of in data
463 * For example TREAD have 11.
464 * 11 is the read/write header = PDU Header(7) + IO Size (4).
465 * Arrange in such a way that server places header in the
466 * allocated memory and payload onto the user buffer.
467 */
468 in = pack_sg_list(chan->sg, out,
469 VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len);
470 if (in)
471 sgs[out_sgs + in_sgs++] = chan->sg + out;
472
473 if (in_pages) {
474 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
475 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
476 in_pages, in_nr_pages, offs, inlen);
477 }
478
479 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
480 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
481 GFP_ATOMIC);
482 if (err < 0) {
483 if (err == -ENOSPC) {
484 chan->ring_bufs_avail = 0;
485 spin_unlock_irqrestore(&chan->lock, flags);
486 err = wait_event_killable(*chan->vc_wq,
487 chan->ring_bufs_avail);
488 if (err == -ERESTARTSYS)
489 goto err_out;
490
491 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
492 goto req_retry_pinned;
493 } else {
494 spin_unlock_irqrestore(&chan->lock, flags);
495 p9_debug(P9_DEBUG_TRANS,
496 "virtio rpc add_sgs returned failure\n");
497 err = -EIO;
498 goto err_out;
499 }
500 }
501 virtqueue_kick(chan->vq);
502 spin_unlock_irqrestore(&chan->lock, flags);
503 kicked = 1;
504 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
505 err = wait_event_killable(req->wq, req->status >= REQ_STATUS_RCVD);
506 /*
507 * Non kernel buffers are pinned, unpin them
508 */
509err_out:
510 if (need_drop) {
511 if (in_pages) {
512 p9_release_pages(in_pages, in_nr_pages);
513 atomic_sub(in_nr_pages, &vp_pinned);
514 }
515 if (out_pages) {
516 p9_release_pages(out_pages, out_nr_pages);
517 atomic_sub(out_nr_pages, &vp_pinned);
518 }
519 /* wakeup anybody waiting for slots to pin pages */
520 wake_up(&vp_wq);
521 }
522 kvfree(in_pages);
523 kvfree(out_pages);
524 if (!kicked) {
525 /* reply won't come */
526 p9_req_put(req);
527 }
528 return err;
529}
530
531static ssize_t p9_mount_tag_show(struct device *dev,
532 struct device_attribute *attr, char *buf)
533{
534 struct virtio_chan *chan;
535 struct virtio_device *vdev;
536 int tag_len;
537
538 vdev = dev_to_virtio(dev);
539 chan = vdev->priv;
540 tag_len = strlen(chan->tag);
541
542 memcpy(buf, chan->tag, tag_len + 1);
543
544 return tag_len + 1;
545}
546
547static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
548
549/**
550 * p9_virtio_probe - probe for existence of 9P virtio channels
551 * @vdev: virtio device to probe
552 *
553 * This probes for existing virtio channels.
554 *
555 */
556
557static int p9_virtio_probe(struct virtio_device *vdev)
558{
559 __u16 tag_len;
560 char *tag;
561 int err;
562 struct virtio_chan *chan;
563
564 if (!vdev->config->get) {
565 dev_err(&vdev->dev, "%s failure: config access disabled\n",
566 __func__);
567 return -EINVAL;
568 }
569
570 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
571 if (!chan) {
572 pr_err("Failed to allocate virtio 9P channel\n");
573 err = -ENOMEM;
574 goto fail;
575 }
576
577 chan->vdev = vdev;
578
579 /* We expect one virtqueue, for requests. */
580 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
581 if (IS_ERR(chan->vq)) {
582 err = PTR_ERR(chan->vq);
583 goto out_free_chan;
584 }
585 chan->vq->vdev->priv = chan;
586 spin_lock_init(&chan->lock);
587
588 sg_init_table(chan->sg, VIRTQUEUE_NUM);
589
590 chan->inuse = false;
591 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
592 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
593 } else {
594 err = -EINVAL;
595 goto out_free_vq;
596 }
597 tag = kzalloc(tag_len + 1, GFP_KERNEL);
598 if (!tag) {
599 err = -ENOMEM;
600 goto out_free_vq;
601 }
602
603 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
604 tag, tag_len);
605 chan->tag = tag;
606 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
607 if (err) {
608 goto out_free_tag;
609 }
610 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
611 if (!chan->vc_wq) {
612 err = -ENOMEM;
613 goto out_remove_file;
614 }
615 init_waitqueue_head(chan->vc_wq);
616 chan->ring_bufs_avail = 1;
617 /* Ceiling limit to avoid denial of service attacks */
618 chan->p9_max_pages = nr_free_buffer_pages()/4;
619
620 virtio_device_ready(vdev);
621
622 mutex_lock(&virtio_9p_lock);
623 list_add_tail(&chan->chan_list, &virtio_chan_list);
624 mutex_unlock(&virtio_9p_lock);
625
626 /* Let udev rules use the new mount_tag attribute. */
627 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
628
629 return 0;
630
631out_remove_file:
632 sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr);
633out_free_tag:
634 kfree(tag);
635out_free_vq:
636 vdev->config->del_vqs(vdev);
637out_free_chan:
638 kfree(chan);
639fail:
640 return err;
641}
642
643
644/**
645 * p9_virtio_create - allocate a new virtio channel
646 * @client: client instance invoking this transport
647 * @devname: string identifying the channel to connect to (unused)
648 * @args: args passed from sys_mount() for per-transport options (unused)
649 *
650 * This sets up a transport channel for 9p communication. Right now
651 * we only match the first available channel, but eventually we couldlook up
652 * alternate channels by matching devname versus a virtio_config entry.
653 * We use a simple reference count mechanism to ensure that only a single
654 * mount has a channel open at a time.
655 *
656 */
657
658static int
659p9_virtio_create(struct p9_client *client, const char *devname, char *args)
660{
661 struct virtio_chan *chan;
662 int ret = -ENOENT;
663 int found = 0;
664
665 if (devname == NULL)
666 return -EINVAL;
667
668 mutex_lock(&virtio_9p_lock);
669 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
670 if (!strcmp(devname, chan->tag)) {
671 if (!chan->inuse) {
672 chan->inuse = true;
673 found = 1;
674 break;
675 }
676 ret = -EBUSY;
677 }
678 }
679 mutex_unlock(&virtio_9p_lock);
680
681 if (!found) {
682 pr_err("no channels available for device %s\n", devname);
683 return ret;
684 }
685
686 client->trans = (void *)chan;
687 client->status = Connected;
688 chan->client = client;
689
690 return 0;
691}
692
693/**
694 * p9_virtio_remove - clean up resources associated with a virtio device
695 * @vdev: virtio device to remove
696 *
697 */
698
699static void p9_virtio_remove(struct virtio_device *vdev)
700{
701 struct virtio_chan *chan = vdev->priv;
702 unsigned long warning_time;
703
704 mutex_lock(&virtio_9p_lock);
705
706 /* Remove self from list so we don't get new users. */
707 list_del(&chan->chan_list);
708 warning_time = jiffies;
709
710 /* Wait for existing users to close. */
711 while (chan->inuse) {
712 mutex_unlock(&virtio_9p_lock);
713 msleep(250);
714 if (time_after(jiffies, warning_time + 10 * HZ)) {
715 dev_emerg(&vdev->dev,
716 "p9_virtio_remove: waiting for device in use.\n");
717 warning_time = jiffies;
718 }
719 mutex_lock(&virtio_9p_lock);
720 }
721
722 mutex_unlock(&virtio_9p_lock);
723
724 vdev->config->reset(vdev);
725 vdev->config->del_vqs(vdev);
726
727 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
728 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
729 kfree(chan->tag);
730 kfree(chan->vc_wq);
731 kfree(chan);
732
733}
734
735static struct virtio_device_id id_table[] = {
736 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
737 { 0 },
738};
739
740static unsigned int features[] = {
741 VIRTIO_9P_MOUNT_TAG,
742};
743
744/* The standard "struct lguest_driver": */
745static struct virtio_driver p9_virtio_drv = {
746 .feature_table = features,
747 .feature_table_size = ARRAY_SIZE(features),
748 .driver.name = KBUILD_MODNAME,
749 .driver.owner = THIS_MODULE,
750 .id_table = id_table,
751 .probe = p9_virtio_probe,
752 .remove = p9_virtio_remove,
753};
754
755static struct p9_trans_module p9_virtio_trans = {
756 .name = "virtio",
757 .create = p9_virtio_create,
758 .close = p9_virtio_close,
759 .request = p9_virtio_request,
760 .zc_request = p9_virtio_zc_request,
761 .cancel = p9_virtio_cancel,
762 .cancelled = p9_virtio_cancelled,
763 /*
764 * We leave one entry for input and one entry for response
765 * headers. We also skip one more entry to accommodate, address
766 * that are not at page boundary, that can result in an extra
767 * page in zero copy.
768 */
769 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
770 .def = 1,
771 .owner = THIS_MODULE,
772};
773
774/* The standard init function */
775static int __init p9_virtio_init(void)
776{
777 int rc;
778
779 INIT_LIST_HEAD(&virtio_chan_list);
780
781 v9fs_register_trans(&p9_virtio_trans);
782 rc = register_virtio_driver(&p9_virtio_drv);
783 if (rc)
784 v9fs_unregister_trans(&p9_virtio_trans);
785
786 return rc;
787}
788
789static void __exit p9_virtio_cleanup(void)
790{
791 unregister_virtio_driver(&p9_virtio_drv);
792 v9fs_unregister_trans(&p9_virtio_trans);
793}
794
795module_init(p9_virtio_init);
796module_exit(p9_virtio_cleanup);
797
798MODULE_DEVICE_TABLE(virtio, id_table);
799MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
800MODULE_DESCRIPTION("Virtio 9p Transport");
801MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * The Virtio 9p transport driver
4 *
5 * This is a block based transport driver based on the lguest block driver
6 * code.
7 *
8 * Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation
9 *
10 * Based on virtio console driver
11 * Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
12 */
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16#include <linux/in.h>
17#include <linux/module.h>
18#include <linux/net.h>
19#include <linux/ipv6.h>
20#include <linux/errno.h>
21#include <linux/kernel.h>
22#include <linux/un.h>
23#include <linux/uaccess.h>
24#include <linux/inet.h>
25#include <linux/file.h>
26#include <linux/highmem.h>
27#include <linux/slab.h>
28#include <net/9p/9p.h>
29#include <linux/parser.h>
30#include <net/9p/client.h>
31#include <net/9p/transport.h>
32#include <linux/scatterlist.h>
33#include <linux/swap.h>
34#include <linux/virtio.h>
35#include <linux/virtio_9p.h>
36#include "trans_common.h"
37
38#define VIRTQUEUE_NUM 128
39
40/* a single mutex to manage channel initialization and attachment */
41static DEFINE_MUTEX(virtio_9p_lock);
42static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
43static atomic_t vp_pinned = ATOMIC_INIT(0);
44
45/**
46 * struct virtio_chan - per-instance transport information
47 * @inuse: whether the channel is in use
48 * @lock: protects multiple elements within this structure
49 * @client: client instance
50 * @vdev: virtio dev associated with this channel
51 * @vq: virtio queue associated with this channel
52 * @ring_bufs_avail: flag to indicate there is some available in the ring buf
53 * @vc_wq: wait queue for waiting for thing to be added to ring buf
54 * @p9_max_pages: maximum number of pinned pages
55 * @sg: scatter gather list which is used to pack a request (protected?)
56 * @chan_list: linked list of channels
57 *
58 * We keep all per-channel information in a structure.
59 * This structure is allocated within the devices dev->mem space.
60 * A pointer to the structure will get put in the transport private.
61 *
62 */
63
64struct virtio_chan {
65 bool inuse;
66
67 spinlock_t lock;
68
69 struct p9_client *client;
70 struct virtio_device *vdev;
71 struct virtqueue *vq;
72 int ring_bufs_avail;
73 wait_queue_head_t *vc_wq;
74 /* This is global limit. Since we don't have a global structure,
75 * will be placing it in each channel.
76 */
77 unsigned long p9_max_pages;
78 /* Scatterlist: can be too big for stack. */
79 struct scatterlist sg[VIRTQUEUE_NUM];
80 /**
81 * @tag: name to identify a mount null terminated
82 */
83 char *tag;
84
85 struct list_head chan_list;
86};
87
88static struct list_head virtio_chan_list;
89
90/* How many bytes left in this page. */
91static unsigned int rest_of_page(void *data)
92{
93 return PAGE_SIZE - offset_in_page(data);
94}
95
96/**
97 * p9_virtio_close - reclaim resources of a channel
98 * @client: client instance
99 *
100 * This reclaims a channel by freeing its resources and
101 * resetting its inuse flag.
102 *
103 */
104
105static void p9_virtio_close(struct p9_client *client)
106{
107 struct virtio_chan *chan = client->trans;
108
109 mutex_lock(&virtio_9p_lock);
110 if (chan)
111 chan->inuse = false;
112 mutex_unlock(&virtio_9p_lock);
113}
114
115/**
116 * req_done - callback which signals activity from the server
117 * @vq: virtio queue activity was received on
118 *
119 * This notifies us that the server has triggered some activity
120 * on the virtio channel - most likely a response to request we
121 * sent. Figure out which requests now have responses and wake up
122 * those threads.
123 *
124 * Bugs: could do with some additional sanity checking, but appears to work.
125 *
126 */
127
128static void req_done(struct virtqueue *vq)
129{
130 struct virtio_chan *chan = vq->vdev->priv;
131 unsigned int len;
132 struct p9_req_t *req;
133 bool need_wakeup = false;
134 unsigned long flags;
135
136 p9_debug(P9_DEBUG_TRANS, ": request done\n");
137
138 spin_lock_irqsave(&chan->lock, flags);
139 while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) {
140 if (!chan->ring_bufs_avail) {
141 chan->ring_bufs_avail = 1;
142 need_wakeup = true;
143 }
144
145 if (len) {
146 req->rc.size = len;
147 p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
148 }
149 }
150 spin_unlock_irqrestore(&chan->lock, flags);
151 /* Wakeup if anyone waiting for VirtIO ring space. */
152 if (need_wakeup)
153 wake_up(chan->vc_wq);
154}
155
156/**
157 * pack_sg_list - pack a scatter gather list from a linear buffer
158 * @sg: scatter/gather list to pack into
159 * @start: which segment of the sg_list to start at
160 * @limit: maximum segment to pack data to
161 * @data: data to pack into scatter/gather list
162 * @count: amount of data to pack into the scatter/gather list
163 *
164 * sg_lists have multiple segments of various sizes. This will pack
165 * arbitrary data into an existing scatter gather list, segmenting the
166 * data as necessary within constraints.
167 *
168 */
169
170static int pack_sg_list(struct scatterlist *sg, int start,
171 int limit, char *data, int count)
172{
173 int s;
174 int index = start;
175
176 while (count) {
177 s = rest_of_page(data);
178 if (s > count)
179 s = count;
180 BUG_ON(index >= limit);
181 /* Make sure we don't terminate early. */
182 sg_unmark_end(&sg[index]);
183 sg_set_buf(&sg[index++], data, s);
184 count -= s;
185 data += s;
186 }
187 if (index-start)
188 sg_mark_end(&sg[index - 1]);
189 return index-start;
190}
191
192/* We don't currently allow canceling of virtio requests */
193static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
194{
195 return 1;
196}
197
198/* Reply won't come, so drop req ref */
199static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req)
200{
201 p9_req_put(client, req);
202 return 0;
203}
204
205/**
206 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
207 * this takes a list of pages.
208 * @sg: scatter/gather list to pack into
209 * @start: which segment of the sg_list to start at
210 * @limit: maximum number of pages in sg list.
211 * @pdata: a list of pages to add into sg.
212 * @nr_pages: number of pages to pack into the scatter/gather list
213 * @offs: amount of data in the beginning of first page _not_ to pack
214 * @count: amount of data to pack into the scatter/gather list
215 */
216static int
217pack_sg_list_p(struct scatterlist *sg, int start, int limit,
218 struct page **pdata, int nr_pages, size_t offs, int count)
219{
220 int i = 0, s;
221 int data_off = offs;
222 int index = start;
223
224 BUG_ON(nr_pages > (limit - start));
225 /*
226 * if the first page doesn't start at
227 * page boundary find the offset
228 */
229 while (nr_pages) {
230 s = PAGE_SIZE - data_off;
231 if (s > count)
232 s = count;
233 BUG_ON(index >= limit);
234 /* Make sure we don't terminate early. */
235 sg_unmark_end(&sg[index]);
236 sg_set_page(&sg[index++], pdata[i++], s, data_off);
237 data_off = 0;
238 count -= s;
239 nr_pages--;
240 }
241
242 if (index-start)
243 sg_mark_end(&sg[index - 1]);
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, out_sgs, in_sgs;
259 unsigned long flags;
260 struct virtio_chan *chan = client->trans;
261 struct scatterlist *sgs[2];
262
263 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
264
265 WRITE_ONCE(req->status, REQ_STATUS_SENT);
266req_retry:
267 spin_lock_irqsave(&chan->lock, flags);
268
269 out_sgs = in_sgs = 0;
270 /* Handle out VirtIO ring buffers */
271 out = pack_sg_list(chan->sg, 0,
272 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
273 if (out)
274 sgs[out_sgs++] = chan->sg;
275
276 in = pack_sg_list(chan->sg, out,
277 VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity);
278 if (in)
279 sgs[out_sgs + in_sgs++] = chan->sg + out;
280
281 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
282 GFP_ATOMIC);
283 if (err < 0) {
284 if (err == -ENOSPC) {
285 chan->ring_bufs_avail = 0;
286 spin_unlock_irqrestore(&chan->lock, flags);
287 err = wait_event_killable(*chan->vc_wq,
288 chan->ring_bufs_avail);
289 if (err == -ERESTARTSYS)
290 return err;
291
292 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
293 goto req_retry;
294 } else {
295 spin_unlock_irqrestore(&chan->lock, flags);
296 p9_debug(P9_DEBUG_TRANS,
297 "virtio rpc add_sgs returned failure\n");
298 return -EIO;
299 }
300 }
301 virtqueue_kick(chan->vq);
302 spin_unlock_irqrestore(&chan->lock, flags);
303
304 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
305 return 0;
306}
307
308static int p9_get_mapped_pages(struct virtio_chan *chan,
309 struct page ***pages,
310 struct iov_iter *data,
311 int count,
312 size_t *offs,
313 int *need_drop)
314{
315 int nr_pages;
316 int err;
317
318 if (!iov_iter_count(data))
319 return 0;
320
321 if (!iov_iter_is_kvec(data)) {
322 int n;
323 /*
324 * We allow only p9_max_pages pinned. We wait for the
325 * Other zc request to finish here
326 */
327 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
328 err = wait_event_killable(vp_wq,
329 (atomic_read(&vp_pinned) < chan->p9_max_pages));
330 if (err == -ERESTARTSYS)
331 return err;
332 }
333 n = iov_iter_get_pages_alloc2(data, pages, count, offs);
334 if (n < 0)
335 return n;
336 *need_drop = 1;
337 nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
338 atomic_add(nr_pages, &vp_pinned);
339 return n;
340 } else {
341 /* kernel buffer, no need to pin pages */
342 int index;
343 size_t len;
344 void *p;
345
346 /* we'd already checked that it's non-empty */
347 while (1) {
348 len = iov_iter_single_seg_count(data);
349 if (likely(len)) {
350 p = data->kvec->iov_base + data->iov_offset;
351 break;
352 }
353 iov_iter_advance(data, 0);
354 }
355 if (len > count)
356 len = count;
357
358 nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
359 (unsigned long)p / PAGE_SIZE;
360
361 *pages = kmalloc_array(nr_pages, sizeof(struct page *),
362 GFP_NOFS);
363 if (!*pages)
364 return -ENOMEM;
365
366 *need_drop = 0;
367 p -= (*offs = offset_in_page(p));
368 for (index = 0; index < nr_pages; index++) {
369 if (is_vmalloc_addr(p))
370 (*pages)[index] = vmalloc_to_page(p);
371 else
372 (*pages)[index] = kmap_to_page(p);
373 p += PAGE_SIZE;
374 }
375 iov_iter_advance(data, len);
376 return len;
377 }
378}
379
380static void handle_rerror(struct p9_req_t *req, int in_hdr_len,
381 size_t offs, struct page **pages)
382{
383 unsigned size, n;
384 void *to = req->rc.sdata + in_hdr_len;
385
386 // Fits entirely into the static data? Nothing to do.
387 if (req->rc.size < in_hdr_len)
388 return;
389
390 // Really long error message? Tough, truncate the reply. Might get
391 // rejected (we can't be arsed to adjust the size encoded in header,
392 // or string size for that matter), but it wouldn't be anything valid
393 // anyway.
394 if (unlikely(req->rc.size > P9_ZC_HDR_SZ))
395 req->rc.size = P9_ZC_HDR_SZ;
396
397 // data won't span more than two pages
398 size = req->rc.size - in_hdr_len;
399 n = PAGE_SIZE - offs;
400 if (size > n) {
401 memcpy_from_page(to, *pages++, offs, n);
402 offs = 0;
403 to += n;
404 size -= n;
405 }
406 memcpy_from_page(to, *pages, offs, size);
407}
408
409/**
410 * p9_virtio_zc_request - issue a zero copy request
411 * @client: client instance issuing the request
412 * @req: request to be issued
413 * @uidata: user buffer that should be used for zero copy read
414 * @uodata: user buffer that should be used for zero copy write
415 * @inlen: read buffer size
416 * @outlen: write buffer size
417 * @in_hdr_len: reader header size, This is the size of response protocol data
418 *
419 */
420static int
421p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
422 struct iov_iter *uidata, struct iov_iter *uodata,
423 int inlen, int outlen, int in_hdr_len)
424{
425 int in, out, err, out_sgs, in_sgs;
426 unsigned long flags;
427 int in_nr_pages = 0, out_nr_pages = 0;
428 struct page **in_pages = NULL, **out_pages = NULL;
429 struct virtio_chan *chan = client->trans;
430 struct scatterlist *sgs[4];
431 size_t offs;
432 int need_drop = 0;
433 int kicked = 0;
434
435 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
436
437 if (uodata) {
438 __le32 sz;
439 int n = p9_get_mapped_pages(chan, &out_pages, uodata,
440 outlen, &offs, &need_drop);
441 if (n < 0) {
442 err = n;
443 goto err_out;
444 }
445 out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
446 if (n != outlen) {
447 __le32 v = cpu_to_le32(n);
448 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
449 outlen = n;
450 }
451 /* The size field of the message must include the length of the
452 * header and the length of the data. We didn't actually know
453 * the length of the data until this point so add it in now.
454 */
455 sz = cpu_to_le32(req->tc.size + outlen);
456 memcpy(&req->tc.sdata[0], &sz, sizeof(sz));
457 } else if (uidata) {
458 int n = p9_get_mapped_pages(chan, &in_pages, uidata,
459 inlen, &offs, &need_drop);
460 if (n < 0) {
461 err = n;
462 goto err_out;
463 }
464 in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
465 if (n != inlen) {
466 __le32 v = cpu_to_le32(n);
467 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
468 inlen = n;
469 }
470 }
471 WRITE_ONCE(req->status, REQ_STATUS_SENT);
472req_retry_pinned:
473 spin_lock_irqsave(&chan->lock, flags);
474
475 out_sgs = in_sgs = 0;
476
477 /* out data */
478 out = pack_sg_list(chan->sg, 0,
479 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
480
481 if (out)
482 sgs[out_sgs++] = chan->sg;
483
484 if (out_pages) {
485 sgs[out_sgs++] = chan->sg + out;
486 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
487 out_pages, out_nr_pages, offs, outlen);
488 }
489
490 /*
491 * Take care of in data
492 * For example TREAD have 11.
493 * 11 is the read/write header = PDU Header(7) + IO Size (4).
494 * Arrange in such a way that server places header in the
495 * allocated memory and payload onto the user buffer.
496 */
497 in = pack_sg_list(chan->sg, out,
498 VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len);
499 if (in)
500 sgs[out_sgs + in_sgs++] = chan->sg + out;
501
502 if (in_pages) {
503 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
504 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
505 in_pages, in_nr_pages, offs, inlen);
506 }
507
508 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
509 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
510 GFP_ATOMIC);
511 if (err < 0) {
512 if (err == -ENOSPC) {
513 chan->ring_bufs_avail = 0;
514 spin_unlock_irqrestore(&chan->lock, flags);
515 err = wait_event_killable(*chan->vc_wq,
516 chan->ring_bufs_avail);
517 if (err == -ERESTARTSYS)
518 goto err_out;
519
520 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
521 goto req_retry_pinned;
522 } else {
523 spin_unlock_irqrestore(&chan->lock, flags);
524 p9_debug(P9_DEBUG_TRANS,
525 "virtio rpc add_sgs returned failure\n");
526 err = -EIO;
527 goto err_out;
528 }
529 }
530 virtqueue_kick(chan->vq);
531 spin_unlock_irqrestore(&chan->lock, flags);
532 kicked = 1;
533 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
534 err = wait_event_killable(req->wq,
535 READ_ONCE(req->status) >= REQ_STATUS_RCVD);
536 // RERROR needs reply (== error string) in static data
537 if (READ_ONCE(req->status) == REQ_STATUS_RCVD &&
538 unlikely(req->rc.sdata[4] == P9_RERROR))
539 handle_rerror(req, in_hdr_len, offs, in_pages);
540
541 /*
542 * Non kernel buffers are pinned, unpin them
543 */
544err_out:
545 if (need_drop) {
546 if (in_pages) {
547 p9_release_pages(in_pages, in_nr_pages);
548 atomic_sub(in_nr_pages, &vp_pinned);
549 }
550 if (out_pages) {
551 p9_release_pages(out_pages, out_nr_pages);
552 atomic_sub(out_nr_pages, &vp_pinned);
553 }
554 /* wakeup anybody waiting for slots to pin pages */
555 wake_up(&vp_wq);
556 }
557 kvfree(in_pages);
558 kvfree(out_pages);
559 if (!kicked) {
560 /* reply won't come */
561 p9_req_put(client, req);
562 }
563 return err;
564}
565
566static ssize_t p9_mount_tag_show(struct device *dev,
567 struct device_attribute *attr, char *buf)
568{
569 struct virtio_chan *chan;
570 struct virtio_device *vdev;
571 int tag_len;
572
573 vdev = dev_to_virtio(dev);
574 chan = vdev->priv;
575 tag_len = strlen(chan->tag);
576
577 memcpy(buf, chan->tag, tag_len + 1);
578
579 return tag_len + 1;
580}
581
582static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
583
584/**
585 * p9_virtio_probe - probe for existence of 9P virtio channels
586 * @vdev: virtio device to probe
587 *
588 * This probes for existing virtio channels.
589 *
590 */
591
592static int p9_virtio_probe(struct virtio_device *vdev)
593{
594 __u16 tag_len;
595 char *tag;
596 int err;
597 struct virtio_chan *chan;
598
599 if (!vdev->config->get) {
600 dev_err(&vdev->dev, "%s failure: config access disabled\n",
601 __func__);
602 return -EINVAL;
603 }
604
605 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
606 if (!chan) {
607 pr_err("Failed to allocate virtio 9P channel\n");
608 err = -ENOMEM;
609 goto fail;
610 }
611
612 chan->vdev = vdev;
613
614 /* We expect one virtqueue, for requests. */
615 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
616 if (IS_ERR(chan->vq)) {
617 err = PTR_ERR(chan->vq);
618 goto out_free_chan;
619 }
620 chan->vq->vdev->priv = chan;
621 spin_lock_init(&chan->lock);
622
623 sg_init_table(chan->sg, VIRTQUEUE_NUM);
624
625 chan->inuse = false;
626 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
627 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
628 } else {
629 err = -EINVAL;
630 goto out_free_vq;
631 }
632 tag = kzalloc(tag_len + 1, GFP_KERNEL);
633 if (!tag) {
634 err = -ENOMEM;
635 goto out_free_vq;
636 }
637
638 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
639 tag, tag_len);
640 chan->tag = tag;
641 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
642 if (err) {
643 goto out_free_tag;
644 }
645 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
646 if (!chan->vc_wq) {
647 err = -ENOMEM;
648 goto out_remove_file;
649 }
650 init_waitqueue_head(chan->vc_wq);
651 chan->ring_bufs_avail = 1;
652 /* Ceiling limit to avoid denial of service attacks */
653 chan->p9_max_pages = nr_free_buffer_pages()/4;
654
655 virtio_device_ready(vdev);
656
657 mutex_lock(&virtio_9p_lock);
658 list_add_tail(&chan->chan_list, &virtio_chan_list);
659 mutex_unlock(&virtio_9p_lock);
660
661 /* Let udev rules use the new mount_tag attribute. */
662 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
663
664 return 0;
665
666out_remove_file:
667 sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr);
668out_free_tag:
669 kfree(tag);
670out_free_vq:
671 vdev->config->del_vqs(vdev);
672out_free_chan:
673 kfree(chan);
674fail:
675 return err;
676}
677
678
679/**
680 * p9_virtio_create - allocate a new virtio channel
681 * @client: client instance invoking this transport
682 * @devname: string identifying the channel to connect to (unused)
683 * @args: args passed from sys_mount() for per-transport options (unused)
684 *
685 * This sets up a transport channel for 9p communication. Right now
686 * we only match the first available channel, but eventually we could look up
687 * alternate channels by matching devname versus a virtio_config entry.
688 * We use a simple reference count mechanism to ensure that only a single
689 * mount has a channel open at a time.
690 *
691 */
692
693static int
694p9_virtio_create(struct p9_client *client, const char *devname, char *args)
695{
696 struct virtio_chan *chan;
697 int ret = -ENOENT;
698 int found = 0;
699
700 if (devname == NULL)
701 return -EINVAL;
702
703 mutex_lock(&virtio_9p_lock);
704 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
705 if (!strcmp(devname, chan->tag)) {
706 if (!chan->inuse) {
707 chan->inuse = true;
708 found = 1;
709 break;
710 }
711 ret = -EBUSY;
712 }
713 }
714 mutex_unlock(&virtio_9p_lock);
715
716 if (!found) {
717 pr_err("no channels available for device %s\n", devname);
718 return ret;
719 }
720
721 client->trans = (void *)chan;
722 client->status = Connected;
723 chan->client = client;
724
725 return 0;
726}
727
728/**
729 * p9_virtio_remove - clean up resources associated with a virtio device
730 * @vdev: virtio device to remove
731 *
732 */
733
734static void p9_virtio_remove(struct virtio_device *vdev)
735{
736 struct virtio_chan *chan = vdev->priv;
737 unsigned long warning_time;
738
739 mutex_lock(&virtio_9p_lock);
740
741 /* Remove self from list so we don't get new users. */
742 list_del(&chan->chan_list);
743 warning_time = jiffies;
744
745 /* Wait for existing users to close. */
746 while (chan->inuse) {
747 mutex_unlock(&virtio_9p_lock);
748 msleep(250);
749 if (time_after(jiffies, warning_time + 10 * HZ)) {
750 dev_emerg(&vdev->dev,
751 "p9_virtio_remove: waiting for device in use.\n");
752 warning_time = jiffies;
753 }
754 mutex_lock(&virtio_9p_lock);
755 }
756
757 mutex_unlock(&virtio_9p_lock);
758
759 virtio_reset_device(vdev);
760 vdev->config->del_vqs(vdev);
761
762 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
763 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
764 kfree(chan->tag);
765 kfree(chan->vc_wq);
766 kfree(chan);
767
768}
769
770static struct virtio_device_id id_table[] = {
771 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
772 { 0 },
773};
774
775static unsigned int features[] = {
776 VIRTIO_9P_MOUNT_TAG,
777};
778
779/* The standard "struct lguest_driver": */
780static struct virtio_driver p9_virtio_drv = {
781 .feature_table = features,
782 .feature_table_size = ARRAY_SIZE(features),
783 .driver.name = KBUILD_MODNAME,
784 .driver.owner = THIS_MODULE,
785 .id_table = id_table,
786 .probe = p9_virtio_probe,
787 .remove = p9_virtio_remove,
788};
789
790static struct p9_trans_module p9_virtio_trans = {
791 .name = "virtio",
792 .create = p9_virtio_create,
793 .close = p9_virtio_close,
794 .request = p9_virtio_request,
795 .zc_request = p9_virtio_zc_request,
796 .cancel = p9_virtio_cancel,
797 .cancelled = p9_virtio_cancelled,
798 /*
799 * We leave one entry for input and one entry for response
800 * headers. We also skip one more entry to accommodate, address
801 * that are not at page boundary, that can result in an extra
802 * page in zero copy.
803 */
804 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
805 .pooled_rbuffers = false,
806 .def = 1,
807 .owner = THIS_MODULE,
808};
809
810/* The standard init function */
811static int __init p9_virtio_init(void)
812{
813 int rc;
814
815 INIT_LIST_HEAD(&virtio_chan_list);
816
817 v9fs_register_trans(&p9_virtio_trans);
818 rc = register_virtio_driver(&p9_virtio_drv);
819 if (rc)
820 v9fs_unregister_trans(&p9_virtio_trans);
821
822 return rc;
823}
824
825static void __exit p9_virtio_cleanup(void)
826{
827 unregister_virtio_driver(&p9_virtio_drv);
828 v9fs_unregister_trans(&p9_virtio_trans);
829}
830
831module_init(p9_virtio_init);
832module_exit(p9_virtio_cleanup);
833MODULE_ALIAS_9P("virtio");
834
835MODULE_DEVICE_TABLE(virtio, id_table);
836MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
837MODULE_DESCRIPTION("Virtio 9p Transport");
838MODULE_LICENSE("GPL");