Linux Audio

Check our new training course

Loading...
v4.10.11
 
  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 */
 57static DEFINE_MUTEX(virtio_9p_lock);
 58static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
 59static 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
 77struct 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	unsigned long 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
103static struct list_head virtio_chan_list;
104
105/* How many bytes left in this page. */
106static unsigned int rest_of_page(void *data)
107{
108	return PAGE_SIZE - offset_in_page(data);
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
120static 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
143static void req_done(struct virtqueue *vq)
144{
145	struct virtio_chan *chan = vq->vdev->priv;
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		req = virtqueue_get_buf(chan->vq, &len);
155		if (req == 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_client_cb(chan->client, req, REQ_STATUS_RCVD);
164	}
 
 
 
 
165}
166
167/**
168 * pack_sg_list - pack a scatter gather list from a linear buffer
169 * @sg: scatter/gather list to pack into
170 * @start: which segment of the sg_list to start at
171 * @limit: maximum segment to pack data to
172 * @data: data to pack into scatter/gather list
173 * @count: amount of data to pack into the scatter/gather list
174 *
175 * sg_lists have multiple segments of various sizes.  This will pack
176 * arbitrary data into an existing scatter gather list, segmenting the
177 * data as necessary within constraints.
178 *
179 */
180
181static int pack_sg_list(struct scatterlist *sg, int start,
182			int limit, char *data, int count)
183{
184	int s;
185	int index = start;
186
187	while (count) {
188		s = rest_of_page(data);
189		if (s > count)
190			s = count;
191		BUG_ON(index > limit);
192		/* Make sure we don't terminate early. */
193		sg_unmark_end(&sg[index]);
194		sg_set_buf(&sg[index++], data, s);
195		count -= s;
196		data += s;
197	}
198	if (index-start)
199		sg_mark_end(&sg[index - 1]);
200	return index-start;
201}
202
203/* We don't currently allow canceling of virtio requests */
204static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
205{
206	return 1;
207}
208
 
 
 
 
 
 
 
209/**
210 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
211 * this takes a list of pages.
212 * @sg: scatter/gather list to pack into
213 * @start: which segment of the sg_list to start at
 
214 * @pdata: a list of pages to add into sg.
215 * @nr_pages: number of pages to pack into the scatter/gather list
216 * @offs: amount of data in the beginning of first page _not_ to pack
217 * @count: amount of data to pack into the scatter/gather list
218 */
219static int
220pack_sg_list_p(struct scatterlist *sg, int start, int limit,
221	       struct page **pdata, int nr_pages, size_t offs, int count)
222{
223	int i = 0, s;
224	int data_off = offs;
225	int index = start;
226
227	BUG_ON(nr_pages > (limit - start));
228	/*
229	 * if the first page doesn't start at
230	 * page boundary find the offset
231	 */
232	while (nr_pages) {
233		s = PAGE_SIZE - data_off;
234		if (s > count)
235			s = count;
 
236		/* Make sure we don't terminate early. */
237		sg_unmark_end(&sg[index]);
238		sg_set_page(&sg[index++], pdata[i++], s, data_off);
239		data_off = 0;
240		count -= s;
241		nr_pages--;
242	}
243
244	if (index-start)
245		sg_mark_end(&sg[index - 1]);
246	return index - start;
247}
248
249/**
250 * p9_virtio_request - issue a request
251 * @client: client instance issuing the request
252 * @req: request to be issued
253 *
254 */
255
256static int
257p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
258{
259	int err;
260	int in, out, out_sgs, in_sgs;
261	unsigned long flags;
262	struct virtio_chan *chan = client->trans;
263	struct scatterlist *sgs[2];
264
265	p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
266
267	req->status = REQ_STATUS_SENT;
268req_retry:
269	spin_lock_irqsave(&chan->lock, flags);
270
271	out_sgs = in_sgs = 0;
272	/* Handle out VirtIO ring buffers */
273	out = pack_sg_list(chan->sg, 0,
274			   VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
275	if (out)
276		sgs[out_sgs++] = chan->sg;
277
278	in = pack_sg_list(chan->sg, out,
279			  VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity);
280	if (in)
281		sgs[out_sgs + in_sgs++] = chan->sg + out;
282
283	err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
284				GFP_ATOMIC);
285	if (err < 0) {
286		if (err == -ENOSPC) {
287			chan->ring_bufs_avail = 0;
288			spin_unlock_irqrestore(&chan->lock, flags);
289			err = wait_event_interruptible(*chan->vc_wq,
290							chan->ring_bufs_avail);
291			if (err  == -ERESTARTSYS)
292				return err;
293
294			p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
295			goto req_retry;
296		} else {
297			spin_unlock_irqrestore(&chan->lock, flags);
298			p9_debug(P9_DEBUG_TRANS,
299				 "virtio rpc add_sgs returned failure\n");
300			return -EIO;
301		}
302	}
303	virtqueue_kick(chan->vq);
304	spin_unlock_irqrestore(&chan->lock, flags);
305
306	p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
307	return 0;
308}
309
310static int p9_get_mapped_pages(struct virtio_chan *chan,
311			       struct page ***pages,
312			       struct iov_iter *data,
313			       int count,
314			       size_t *offs,
315			       int *need_drop)
316{
317	int nr_pages;
318	int err;
319
320	if (!iov_iter_count(data))
321		return 0;
322
323	if (!(data->type & ITER_KVEC)) {
324		int n;
325		/*
326		 * We allow only p9_max_pages pinned. We wait for the
327		 * Other zc request to finish here
328		 */
329		if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
330			err = wait_event_interruptible(vp_wq,
331			      (atomic_read(&vp_pinned) < chan->p9_max_pages));
332			if (err == -ERESTARTSYS)
333				return err;
334		}
335		n = iov_iter_get_pages_alloc(data, pages, count, offs);
336		if (n < 0)
337			return n;
338		*need_drop = 1;
339		nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
340		atomic_add(nr_pages, &vp_pinned);
341		return n;
342	} else {
343		/* kernel buffer, no need to pin pages */
344		int index;
345		size_t len;
346		void *p;
347
348		/* we'd already checked that it's non-empty */
349		while (1) {
350			len = iov_iter_single_seg_count(data);
351			if (likely(len)) {
352				p = data->kvec->iov_base + data->iov_offset;
353				break;
354			}
355			iov_iter_advance(data, 0);
356		}
357		if (len > count)
358			len = count;
359
360		nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
361			   (unsigned long)p / PAGE_SIZE;
362
363		*pages = kmalloc(sizeof(struct page *) * nr_pages, 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 bffer that should be ued for zero copy read
385 * @uodata: user buffer that shoud be user for zero copy write
386 * @inlen: read buffer size
387 * @olen: write buffer size
388 * @hdrlen: 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
405	p9_debug(P9_DEBUG_TRANS, "virtio request\n");
406
407	if (uodata) {
 
408		int n = p9_get_mapped_pages(chan, &out_pages, uodata,
409					    outlen, &offs, &need_drop);
410		if (n < 0)
411			return n;
 
 
412		out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
413		if (n != outlen) {
414			__le32 v = cpu_to_le32(n);
415			memcpy(&req->tc->sdata[req->tc->size - 4], &v, 4);
416			outlen = n;
417		}
 
 
 
 
 
 
418	} else if (uidata) {
419		int n = p9_get_mapped_pages(chan, &in_pages, uidata,
420					    inlen, &offs, &need_drop);
421		if (n < 0)
422			return n;
 
 
423		in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
424		if (n != inlen) {
425			__le32 v = cpu_to_le32(n);
426			memcpy(&req->tc->sdata[req->tc->size - 4], &v, 4);
427			inlen = n;
428		}
429	}
430	req->status = REQ_STATUS_SENT;
431req_retry_pinned:
432	spin_lock_irqsave(&chan->lock, flags);
433
434	out_sgs = in_sgs = 0;
435
436	/* out data */
437	out = pack_sg_list(chan->sg, 0,
438			   VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
439
440	if (out)
441		sgs[out_sgs++] = chan->sg;
442
443	if (out_pages) {
444		sgs[out_sgs++] = chan->sg + out;
445		out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
446				      out_pages, out_nr_pages, offs, outlen);
447	}
448		
449	/*
450	 * Take care of in data
451	 * For example TREAD have 11.
452	 * 11 is the read/write header = PDU Header(7) + IO Size (4).
453	 * Arrange in such a way that server places header in the
454	 * alloced memory and payload onto the user buffer.
455	 */
456	in = pack_sg_list(chan->sg, out,
457			  VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len);
458	if (in)
459		sgs[out_sgs + in_sgs++] = chan->sg + out;
460
461	if (in_pages) {
462		sgs[out_sgs + in_sgs++] = chan->sg + out + in;
463		in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
464				     in_pages, in_nr_pages, offs, inlen);
465	}
466
467	BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
468	err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
469				GFP_ATOMIC);
470	if (err < 0) {
471		if (err == -ENOSPC) {
472			chan->ring_bufs_avail = 0;
473			spin_unlock_irqrestore(&chan->lock, flags);
474			err = wait_event_interruptible(*chan->vc_wq,
475						       chan->ring_bufs_avail);
476			if (err  == -ERESTARTSYS)
477				goto err_out;
478
479			p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
480			goto req_retry_pinned;
481		} else {
482			spin_unlock_irqrestore(&chan->lock, flags);
483			p9_debug(P9_DEBUG_TRANS,
484				 "virtio rpc add_sgs returned failure\n");
485			err = -EIO;
486			goto err_out;
487		}
488	}
489	virtqueue_kick(chan->vq);
490	spin_unlock_irqrestore(&chan->lock, flags);
 
491	p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
492	err = wait_event_interruptible(*req->wq,
493				       req->status >= REQ_STATUS_RCVD);
 
 
 
 
 
494	/*
495	 * Non kernel buffers are pinned, unpin them
496	 */
497err_out:
498	if (need_drop) {
499		if (in_pages) {
500			p9_release_pages(in_pages, in_nr_pages);
501			atomic_sub(in_nr_pages, &vp_pinned);
502		}
503		if (out_pages) {
504			p9_release_pages(out_pages, out_nr_pages);
505			atomic_sub(out_nr_pages, &vp_pinned);
506		}
507		/* wakeup anybody waiting for slots to pin pages */
508		wake_up(&vp_wq);
509	}
510	kvfree(in_pages);
511	kvfree(out_pages);
 
 
 
 
512	return err;
513}
514
515static ssize_t p9_mount_tag_show(struct device *dev,
516				struct device_attribute *attr, char *buf)
517{
518	struct virtio_chan *chan;
519	struct virtio_device *vdev;
 
520
521	vdev = dev_to_virtio(dev);
522	chan = vdev->priv;
 
523
524	memcpy(buf, chan->tag, chan->tag_len);
525	buf[chan->tag_len] = 0;
526
527	return chan->tag_len + 1;
528}
529
530static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
531
532/**
533 * p9_virtio_probe - probe for existence of 9P virtio channels
534 * @vdev: virtio device to probe
535 *
536 * This probes for existing virtio channels.
537 *
538 */
539
540static int p9_virtio_probe(struct virtio_device *vdev)
541{
542	__u16 tag_len;
543	char *tag;
544	int err;
545	struct virtio_chan *chan;
546
547	if (!vdev->config->get) {
548		dev_err(&vdev->dev, "%s failure: config access disabled\n",
549			__func__);
550		return -EINVAL;
551	}
552
553	chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
554	if (!chan) {
555		pr_err("Failed to allocate virtio 9P channel\n");
556		err = -ENOMEM;
557		goto fail;
558	}
559
560	chan->vdev = vdev;
561
562	/* We expect one virtqueue, for requests. */
563	chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
564	if (IS_ERR(chan->vq)) {
565		err = PTR_ERR(chan->vq);
566		goto out_free_vq;
567	}
568	chan->vq->vdev->priv = chan;
569	spin_lock_init(&chan->lock);
570
571	sg_init_table(chan->sg, VIRTQUEUE_NUM);
572
573	chan->inuse = false;
574	if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
575		virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
576	} else {
577		err = -EINVAL;
578		goto out_free_vq;
579	}
580	tag = kmalloc(tag_len, GFP_KERNEL);
581	if (!tag) {
582		err = -ENOMEM;
583		goto out_free_vq;
584	}
585
586	virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
587			   tag, tag_len);
588	chan->tag = tag;
589	chan->tag_len = tag_len;
590	err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
591	if (err) {
592		goto out_free_tag;
593	}
594	chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
595	if (!chan->vc_wq) {
596		err = -ENOMEM;
597		goto out_free_tag;
598	}
599	init_waitqueue_head(chan->vc_wq);
600	chan->ring_bufs_avail = 1;
601	/* Ceiling limit to avoid denial of service attacks */
602	chan->p9_max_pages = nr_free_buffer_pages()/4;
603
604	virtio_device_ready(vdev);
605
606	mutex_lock(&virtio_9p_lock);
607	list_add_tail(&chan->chan_list, &virtio_chan_list);
608	mutex_unlock(&virtio_9p_lock);
609
610	/* Let udev rules use the new mount_tag attribute. */
611	kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
612
613	return 0;
614
 
 
615out_free_tag:
616	kfree(tag);
617out_free_vq:
618	vdev->config->del_vqs(vdev);
 
619	kfree(chan);
620fail:
621	return err;
622}
623
624
625/**
626 * p9_virtio_create - allocate a new virtio channel
627 * @client: client instance invoking this transport
628 * @devname: string identifying the channel to connect to (unused)
629 * @args: args passed from sys_mount() for per-transport options (unused)
630 *
631 * This sets up a transport channel for 9p communication.  Right now
632 * we only match the first available channel, but eventually we couldlook up
633 * alternate channels by matching devname versus a virtio_config entry.
634 * We use a simple reference count mechanism to ensure that only a single
635 * mount has a channel open at a time.
636 *
637 */
638
639static int
640p9_virtio_create(struct p9_client *client, const char *devname, char *args)
641{
642	struct virtio_chan *chan;
643	int ret = -ENOENT;
644	int found = 0;
645
 
 
 
646	mutex_lock(&virtio_9p_lock);
647	list_for_each_entry(chan, &virtio_chan_list, chan_list) {
648		if (!strncmp(devname, chan->tag, chan->tag_len) &&
649		    strlen(devname) == chan->tag_len) {
650			if (!chan->inuse) {
651				chan->inuse = true;
652				found = 1;
653				break;
654			}
655			ret = -EBUSY;
656		}
657	}
658	mutex_unlock(&virtio_9p_lock);
659
660	if (!found) {
661		pr_err("no channels available for device %s\n", devname);
662		return ret;
663	}
664
665	client->trans = (void *)chan;
666	client->status = Connected;
667	chan->client = client;
668
669	return 0;
670}
671
672/**
673 * p9_virtio_remove - clean up resources associated with a virtio device
674 * @vdev: virtio device to remove
675 *
676 */
677
678static void p9_virtio_remove(struct virtio_device *vdev)
679{
680	struct virtio_chan *chan = vdev->priv;
681	unsigned long warning_time;
682
683	mutex_lock(&virtio_9p_lock);
684
685	/* Remove self from list so we don't get new users. */
686	list_del(&chan->chan_list);
687	warning_time = jiffies;
688
689	/* Wait for existing users to close. */
690	while (chan->inuse) {
691		mutex_unlock(&virtio_9p_lock);
692		msleep(250);
693		if (time_after(jiffies, warning_time + 10 * HZ)) {
694			dev_emerg(&vdev->dev,
695				  "p9_virtio_remove: waiting for device in use.\n");
696			warning_time = jiffies;
697		}
698		mutex_lock(&virtio_9p_lock);
699	}
700
701	mutex_unlock(&virtio_9p_lock);
702
703	vdev->config->reset(vdev);
704	vdev->config->del_vqs(vdev);
705
706	sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
707	kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
708	kfree(chan->tag);
709	kfree(chan->vc_wq);
710	kfree(chan);
711
712}
713
714static struct virtio_device_id id_table[] = {
715	{ VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
716	{ 0 },
717};
718
719static unsigned int features[] = {
720	VIRTIO_9P_MOUNT_TAG,
721};
722
723/* The standard "struct lguest_driver": */
724static struct virtio_driver p9_virtio_drv = {
725	.feature_table  = features,
726	.feature_table_size = ARRAY_SIZE(features),
727	.driver.name    = KBUILD_MODNAME,
728	.driver.owner	= THIS_MODULE,
729	.id_table	= id_table,
730	.probe		= p9_virtio_probe,
731	.remove		= p9_virtio_remove,
732};
733
734static struct p9_trans_module p9_virtio_trans = {
735	.name = "virtio",
736	.create = p9_virtio_create,
737	.close = p9_virtio_close,
738	.request = p9_virtio_request,
739	.zc_request = p9_virtio_zc_request,
740	.cancel = p9_virtio_cancel,
 
741	/*
742	 * We leave one entry for input and one entry for response
743	 * headers. We also skip one more entry to accomodate, address
744	 * that are not at page boundary, that can result in an extra
745	 * page in zero copy.
746	 */
747	.maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
 
748	.def = 1,
749	.owner = THIS_MODULE,
750};
751
752/* The standard init function */
753static int __init p9_virtio_init(void)
754{
 
 
755	INIT_LIST_HEAD(&virtio_chan_list);
756
757	v9fs_register_trans(&p9_virtio_trans);
758	return register_virtio_driver(&p9_virtio_drv);
 
 
 
 
759}
760
761static void __exit p9_virtio_cleanup(void)
762{
763	unregister_virtio_driver(&p9_virtio_drv);
764	v9fs_unregister_trans(&p9_virtio_trans);
765}
766
767module_init(p9_virtio_init);
768module_exit(p9_virtio_cleanup);
 
769
770MODULE_DEVICE_TABLE(virtio, id_table);
771MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
772MODULE_DESCRIPTION("Virtio 9p Transport");
773MODULE_LICENSE("GPL");
v6.8
  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 || !pages)
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 = 0;
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		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");