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 - ((unsigned long)data % PAGE_SIZE);
109}
110
111/**
112 * p9_virtio_close - reclaim resources of a channel
113 * @client: client instance
114 *
115 * This reclaims a channel by freeing its resources and
116 * reseting its inuse flag.
117 *
118 */
119
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	struct p9_fcall *rc;
147	unsigned int len;
148	struct p9_req_t *req;
149	unsigned long flags;
150
151	p9_debug(P9_DEBUG_TRANS, ": request done\n");
152
153	while (1) {
154		spin_lock_irqsave(&chan->lock, flags);
155		rc = virtqueue_get_buf(chan->vq, &len);
156		if (rc == NULL) {
157			spin_unlock_irqrestore(&chan->lock, flags);
158			break;
159		}
160		chan->ring_bufs_avail = 1;
161		spin_unlock_irqrestore(&chan->lock, flags);
162		/* Wakeup if anyone waiting for VirtIO ring space. */
163		wake_up(chan->vc_wq);
164		p9_debug(P9_DEBUG_TRANS, ": rc %p\n", rc);
165		p9_debug(P9_DEBUG_TRANS, ": lookup tag %d\n", rc->tag);
166		req = p9_tag_lookup(chan->client, rc->tag);
167		req->status = REQ_STATUS_RCVD;
168		p9_client_cb(chan->client, req);
169	}
170}
171
172/**
173 * pack_sg_list - pack a scatter gather list from a linear buffer
174 * @sg: scatter/gather list to pack into
175 * @start: which segment of the sg_list to start at
176 * @limit: maximum segment to pack data to
177 * @data: data to pack into scatter/gather list
178 * @count: amount of data to pack into the scatter/gather list
179 *
180 * sg_lists have multiple segments of various sizes.  This will pack
181 * arbitrary data into an existing scatter gather list, segmenting the
182 * data as necessary within constraints.
183 *
184 */
185
186static int pack_sg_list(struct scatterlist *sg, int start,
187			int limit, char *data, int count)
188{
189	int s;
190	int index = start;
191
192	while (count) {
193		s = rest_of_page(data);
194		if (s > count)
195			s = count;
196		BUG_ON(index > limit);
197		/* Make sure we don't terminate early. */
198		sg_unmark_end(&sg[index]);
199		sg_set_buf(&sg[index++], data, s);
200		count -= s;
201		data += s;
202	}
203	if (index-start)
204		sg_mark_end(&sg[index - 1]);
205	return index-start;
206}
207
208/* We don't currently allow canceling of virtio requests */
209static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
210{
211	return 1;
212}
213
214/**
215 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
216 * this takes a list of pages.
217 * @sg: scatter/gather list to pack into
218 * @start: which segment of the sg_list to start at
219 * @pdata: a list of pages to add into sg.
220 * @nr_pages: number of pages to pack into the scatter/gather list
221 * @data: data to pack into scatter/gather list
222 * @count: amount of data to pack into the scatter/gather list
223 */
224static int
225pack_sg_list_p(struct scatterlist *sg, int start, int limit,
226	       struct page **pdata, int nr_pages, char *data, int count)
227{
228	int i = 0, s;
229	int data_off;
230	int index = start;
231
232	BUG_ON(nr_pages > (limit - start));
233	/*
234	 * if the first page doesn't start at
235	 * page boundary find the offset
236	 */
237	data_off = offset_in_page(data);
238	while (nr_pages) {
239		s = rest_of_page(data);
240		if (s > count)
241			s = count;
242		/* Make sure we don't terminate early. */
243		sg_unmark_end(&sg[index]);
244		sg_set_page(&sg[index++], pdata[i++], s, data_off);
245		data_off = 0;
246		data += s;
247		count -= s;
248		nr_pages--;
249	}
250
251	if (index-start)
252		sg_mark_end(&sg[index - 1]);
253	return index - start;
254}
255
256/**
257 * p9_virtio_request - issue a request
258 * @client: client instance issuing the request
259 * @req: request to be issued
260 *
261 */
262
263static int
264p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
265{
266	int err;
267	int in, out, out_sgs, in_sgs;
268	unsigned long flags;
269	struct virtio_chan *chan = client->trans;
270	struct scatterlist *sgs[2];
271
272	p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
273
274	req->status = REQ_STATUS_SENT;
275req_retry:
276	spin_lock_irqsave(&chan->lock, flags);
277
278	out_sgs = in_sgs = 0;
279	/* Handle out VirtIO ring buffers */
280	out = pack_sg_list(chan->sg, 0,
281			   VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
282	if (out)
283		sgs[out_sgs++] = chan->sg;
284
285	in = pack_sg_list(chan->sg, out,
286			  VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity);
287	if (in)
288		sgs[out_sgs + in_sgs++] = chan->sg + out;
289
290	err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
291				GFP_ATOMIC);
292	if (err < 0) {
293		if (err == -ENOSPC) {
294			chan->ring_bufs_avail = 0;
295			spin_unlock_irqrestore(&chan->lock, flags);
296			err = wait_event_interruptible(*chan->vc_wq,
297							chan->ring_bufs_avail);
298			if (err  == -ERESTARTSYS)
299				return err;
300
301			p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
302			goto req_retry;
303		} else {
304			spin_unlock_irqrestore(&chan->lock, flags);
305			p9_debug(P9_DEBUG_TRANS,
306				 "virtio rpc add_sgs returned failure\n");
307			return -EIO;
308		}
309	}
310	virtqueue_kick(chan->vq);
311	spin_unlock_irqrestore(&chan->lock, flags);
312
313	p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
314	return 0;
315}
316
317static int p9_get_mapped_pages(struct virtio_chan *chan,
318			       struct page **pages, char *data,
319			       int nr_pages, int write, int kern_buf)
320{
321	int err;
322	if (!kern_buf) {
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_interruptible(vp_wq,
329			      (atomic_read(&vp_pinned) < chan->p9_max_pages));
330			if (err == -ERESTARTSYS)
331				return err;
332		}
333		err = p9_payload_gup(data, &nr_pages, pages, write);
334		if (err < 0)
335			return err;
336		atomic_add(nr_pages, &vp_pinned);
337	} else {
338		/* kernel buffer, no need to pin pages */
339		int s, index = 0;
340		int count = nr_pages;
341		while (nr_pages) {
342			s = rest_of_page(data);
343			pages[index++] = kmap_to_page(data);
344			data += s;
345			nr_pages--;
346		}
347		nr_pages = count;
348	}
349	return nr_pages;
350}
351
352/**
353 * p9_virtio_zc_request - issue a zero copy request
354 * @client: client instance issuing the request
355 * @req: request to be issued
356 * @uidata: user bffer that should be ued for zero copy read
357 * @uodata: user buffer that shoud be user for zero copy write
358 * @inlen: read buffer size
359 * @olen: write buffer size
360 * @hdrlen: reader header size, This is the size of response protocol data
361 *
362 */
363static int
364p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
365		     char *uidata, char *uodata, int inlen,
366		     int outlen, int in_hdr_len, int kern_buf)
367{
368	int in, out, err, out_sgs, in_sgs;
369	unsigned long flags;
370	int in_nr_pages = 0, out_nr_pages = 0;
371	struct page **in_pages = NULL, **out_pages = NULL;
372	struct virtio_chan *chan = client->trans;
373	struct scatterlist *sgs[4];
374
375	p9_debug(P9_DEBUG_TRANS, "virtio request\n");
376
377	if (uodata) {
378		out_nr_pages = p9_nr_pages(uodata, outlen);
379		out_pages = kmalloc(sizeof(struct page *) * out_nr_pages,
380				    GFP_NOFS);
381		if (!out_pages) {
382			err = -ENOMEM;
383			goto err_out;
384		}
385		out_nr_pages = p9_get_mapped_pages(chan, out_pages, uodata,
386						   out_nr_pages, 0, kern_buf);
387		if (out_nr_pages < 0) {
388			err = out_nr_pages;
389			kfree(out_pages);
390			out_pages = NULL;
391			goto err_out;
392		}
393	}
394	if (uidata) {
395		in_nr_pages = p9_nr_pages(uidata, inlen);
396		in_pages = kmalloc(sizeof(struct page *) * in_nr_pages,
397				   GFP_NOFS);
398		if (!in_pages) {
399			err = -ENOMEM;
400			goto err_out;
401		}
402		in_nr_pages = p9_get_mapped_pages(chan, in_pages, uidata,
403						  in_nr_pages, 1, kern_buf);
404		if (in_nr_pages < 0) {
405			err = in_nr_pages;
406			kfree(in_pages);
407			in_pages = NULL;
408			goto err_out;
409		}
410	}
411	req->status = REQ_STATUS_SENT;
412req_retry_pinned:
413	spin_lock_irqsave(&chan->lock, flags);
414
415	out_sgs = in_sgs = 0;
416
417	/* out data */
418	out = pack_sg_list(chan->sg, 0,
419			   VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
420
421	if (out)
422		sgs[out_sgs++] = chan->sg;
423
424	if (out_pages) {
425		sgs[out_sgs++] = chan->sg + out;
426		out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
427				      out_pages, out_nr_pages, uodata, outlen);
428	}
429
430	/*
431	 * Take care of in data
432	 * For example TREAD have 11.
433	 * 11 is the read/write header = PDU Header(7) + IO Size (4).
434	 * Arrange in such a way that server places header in the
435	 * alloced memory and payload onto the user buffer.
436	 */
437	in = pack_sg_list(chan->sg, out,
438			  VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len);
439	if (in)
440		sgs[out_sgs + in_sgs++] = chan->sg + out;
441
442	if (in_pages) {
443		sgs[out_sgs + in_sgs++] = chan->sg + out + in;
444		in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
445				     in_pages, in_nr_pages, uidata, inlen);
446	}
447
448	BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
449	err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
450				GFP_ATOMIC);
451	if (err < 0) {
452		if (err == -ENOSPC) {
453			chan->ring_bufs_avail = 0;
454			spin_unlock_irqrestore(&chan->lock, flags);
455			err = wait_event_interruptible(*chan->vc_wq,
456						       chan->ring_bufs_avail);
457			if (err  == -ERESTARTSYS)
458				goto err_out;
459
460			p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
461			goto req_retry_pinned;
462		} else {
463			spin_unlock_irqrestore(&chan->lock, flags);
464			p9_debug(P9_DEBUG_TRANS,
465				 "virtio rpc add_sgs returned failure\n");
466			err = -EIO;
467			goto err_out;
468		}
469	}
470	virtqueue_kick(chan->vq);
471	spin_unlock_irqrestore(&chan->lock, flags);
472	p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
473	err = wait_event_interruptible(*req->wq,
474				       req->status >= REQ_STATUS_RCVD);
475	/*
476	 * Non kernel buffers are pinned, unpin them
477	 */
478err_out:
479	if (!kern_buf) {
480		if (in_pages) {
481			p9_release_pages(in_pages, in_nr_pages);
482			atomic_sub(in_nr_pages, &vp_pinned);
483		}
484		if (out_pages) {
485			p9_release_pages(out_pages, out_nr_pages);
486			atomic_sub(out_nr_pages, &vp_pinned);
487		}
488		/* wakeup anybody waiting for slots to pin pages */
489		wake_up(&vp_wq);
490	}
491	kfree(in_pages);
492	kfree(out_pages);
493	return err;
494}
495
496static ssize_t p9_mount_tag_show(struct device *dev,
497				struct device_attribute *attr, char *buf)
498{
499	struct virtio_chan *chan;
500	struct virtio_device *vdev;
501
502	vdev = dev_to_virtio(dev);
503	chan = vdev->priv;
504
505	return snprintf(buf, chan->tag_len + 1, "%s", chan->tag);
506}
507
508static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
509
510/**
511 * p9_virtio_probe - probe for existence of 9P virtio channels
512 * @vdev: virtio device to probe
513 *
514 * This probes for existing virtio channels.
515 *
516 */
517
518static int p9_virtio_probe(struct virtio_device *vdev)
519{
520	__u16 tag_len;
521	char *tag;
522	int err;
523	struct virtio_chan *chan;
524
525	chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
526	if (!chan) {
527		pr_err("Failed to allocate virtio 9P channel\n");
528		err = -ENOMEM;
529		goto fail;
530	}
531
532	chan->vdev = vdev;
533
534	/* We expect one virtqueue, for requests. */
535	chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
536	if (IS_ERR(chan->vq)) {
537		err = PTR_ERR(chan->vq);
538		goto out_free_vq;
539	}
540	chan->vq->vdev->priv = chan;
541	spin_lock_init(&chan->lock);
542
543	sg_init_table(chan->sg, VIRTQUEUE_NUM);
544
545	chan->inuse = false;
546	if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
547		vdev->config->get(vdev,
548				offsetof(struct virtio_9p_config, tag_len),
549				&tag_len, sizeof(tag_len));
550	} else {
551		err = -EINVAL;
552		goto out_free_vq;
553	}
554	tag = kmalloc(tag_len, GFP_KERNEL);
555	if (!tag) {
556		err = -ENOMEM;
557		goto out_free_vq;
558	}
559	vdev->config->get(vdev, offsetof(struct virtio_9p_config, tag),
560			tag, tag_len);
561	chan->tag = tag;
562	chan->tag_len = tag_len;
563	err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
564	if (err) {
565		goto out_free_tag;
566	}
567	chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
568	if (!chan->vc_wq) {
569		err = -ENOMEM;
570		goto out_free_tag;
571	}
572	init_waitqueue_head(chan->vc_wq);
573	chan->ring_bufs_avail = 1;
574	/* Ceiling limit to avoid denial of service attacks */
575	chan->p9_max_pages = nr_free_buffer_pages()/4;
576
577	mutex_lock(&virtio_9p_lock);
578	list_add_tail(&chan->chan_list, &virtio_chan_list);
579	mutex_unlock(&virtio_9p_lock);
580	return 0;
581
582out_free_tag:
583	kfree(tag);
584out_free_vq:
585	vdev->config->del_vqs(vdev);
586	kfree(chan);
587fail:
588	return err;
589}
590
591
592/**
593 * p9_virtio_create - allocate a new virtio channel
594 * @client: client instance invoking this transport
595 * @devname: string identifying the channel to connect to (unused)
596 * @args: args passed from sys_mount() for per-transport options (unused)
597 *
598 * This sets up a transport channel for 9p communication.  Right now
599 * we only match the first available channel, but eventually we couldlook up
600 * alternate channels by matching devname versus a virtio_config entry.
601 * We use a simple reference count mechanism to ensure that only a single
602 * mount has a channel open at a time.
603 *
604 */
605
606static int
607p9_virtio_create(struct p9_client *client, const char *devname, char *args)
608{
609	struct virtio_chan *chan;
610	int ret = -ENOENT;
611	int found = 0;
612
613	mutex_lock(&virtio_9p_lock);
614	list_for_each_entry(chan, &virtio_chan_list, chan_list) {
615		if (!strncmp(devname, chan->tag, chan->tag_len) &&
616		    strlen(devname) == chan->tag_len) {
617			if (!chan->inuse) {
618				chan->inuse = true;
619				found = 1;
620				break;
621			}
622			ret = -EBUSY;
623		}
624	}
625	mutex_unlock(&virtio_9p_lock);
626
627	if (!found) {
628		pr_err("no channels available\n");
629		return ret;
630	}
631
632	client->trans = (void *)chan;
633	client->status = Connected;
634	chan->client = client;
635
636	return 0;
637}
638
639/**
640 * p9_virtio_remove - clean up resources associated with a virtio device
641 * @vdev: virtio device to remove
642 *
643 */
644
645static void p9_virtio_remove(struct virtio_device *vdev)
646{
647	struct virtio_chan *chan = vdev->priv;
648
649	if (chan->inuse)
650		p9_virtio_close(chan->client);
651	vdev->config->del_vqs(vdev);
652
653	mutex_lock(&virtio_9p_lock);
654	list_del(&chan->chan_list);
655	mutex_unlock(&virtio_9p_lock);
656	sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
657	kfree(chan->tag);
658	kfree(chan->vc_wq);
659	kfree(chan);
660
661}
662
663static struct virtio_device_id id_table[] = {
664	{ VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
665	{ 0 },
666};
667
668static unsigned int features[] = {
669	VIRTIO_9P_MOUNT_TAG,
670};
671
672/* The standard "struct lguest_driver": */
673static struct virtio_driver p9_virtio_drv = {
674	.feature_table  = features,
675	.feature_table_size = ARRAY_SIZE(features),
676	.driver.name    = KBUILD_MODNAME,
677	.driver.owner	= THIS_MODULE,
678	.id_table	= id_table,
679	.probe		= p9_virtio_probe,
680	.remove		= p9_virtio_remove,
681};
682
683static struct p9_trans_module p9_virtio_trans = {
684	.name = "virtio",
685	.create = p9_virtio_create,
686	.close = p9_virtio_close,
687	.request = p9_virtio_request,
688	.zc_request = p9_virtio_zc_request,
689	.cancel = p9_virtio_cancel,
690	/*
691	 * We leave one entry for input and one entry for response
692	 * headers. We also skip one more entry to accomodate, address
693	 * that are not at page boundary, that can result in an extra
694	 * page in zero copy.
695	 */
696	.maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
697	.def = 0,
698	.owner = THIS_MODULE,
699};
700
701/* The standard init function */
702static int __init p9_virtio_init(void)
703{
704	INIT_LIST_HEAD(&virtio_chan_list);
705
706	v9fs_register_trans(&p9_virtio_trans);
707	return register_virtio_driver(&p9_virtio_drv);
708}
709
710static void __exit p9_virtio_cleanup(void)
711{
712	unregister_virtio_driver(&p9_virtio_drv);
713	v9fs_unregister_trans(&p9_virtio_trans);
714}
715
716module_init(p9_virtio_init);
717module_exit(p9_virtio_cleanup);
718
719MODULE_DEVICE_TABLE(virtio, id_table);
720MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
721MODULE_DESCRIPTION("Virtio 9p Transport");
722MODULE_LICENSE("GPL");
723