2 * The Virtio 9p transport driver
4 * This is a block based transport driver based on the lguest block driver
7 * Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation
9 * Based on virtio console driver
10 * Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
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.
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.
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
30 #include <linux/module.h>
31 #include <linux/net.h>
32 #include <linux/ipv6.h>
33 #include <linux/errno.h>
34 #include <linux/kernel.h>
36 #include <linux/uaccess.h>
37 #include <linux/inet.h>
38 #include <linux/idr.h>
39 #include <linux/file.h>
40 #include <linux/slab.h>
41 #include <net/9p/9p.h>
42 #include <linux/parser.h>
43 #include <net/9p/client.h>
44 #include <net/9p/transport.h>
45 #include <linux/scatterlist.h>
46 #include <linux/virtio.h>
47 #include <linux/virtio_9p.h>
48 #include "trans_common.h"
50 #define VIRTQUEUE_NUM 128
52 /* a single mutex to manage channel initialization and attachment */
53 static DEFINE_MUTEX(virtio_9p_lock
);
56 * struct virtio_chan - per-instance transport information
57 * @initialized: whether the channel is initialized
58 * @inuse: whether the channel is in use
59 * @lock: protects multiple elements within this structure
60 * @client: client instance
61 * @vdev: virtio dev associated with this channel
62 * @vq: virtio queue associated with this channel
63 * @sg: scatter gather list which is used to pack a request (protected?)
65 * We keep all per-channel information in a structure.
66 * This structure is allocated within the devices dev->mem space.
67 * A pointer to the structure will get put in the transport private.
76 struct p9_client
*client
;
77 struct virtio_device
*vdev
;
80 wait_queue_head_t
*vc_wq
;
82 /* Scatterlist: can be too big for stack. */
83 struct scatterlist sg
[VIRTQUEUE_NUM
];
87 * tag name to identify a mount Non-null terminated
91 struct list_head chan_list
;
94 static struct list_head virtio_chan_list
;
96 /* How many bytes left in this page. */
97 static unsigned int rest_of_page(void *data
)
99 return PAGE_SIZE
- ((unsigned long)data
% PAGE_SIZE
);
103 * p9_virtio_close - reclaim resources of a channel
104 * @client: client instance
106 * This reclaims a channel by freeing its resources and
107 * reseting its inuse flag.
111 static void p9_virtio_close(struct p9_client
*client
)
113 struct virtio_chan
*chan
= client
->trans
;
115 mutex_lock(&virtio_9p_lock
);
118 mutex_unlock(&virtio_9p_lock
);
122 * req_done - callback which signals activity from the server
123 * @vq: virtio queue activity was received on
125 * This notifies us that the server has triggered some activity
126 * on the virtio channel - most likely a response to request we
127 * sent. Figure out which requests now have responses and wake up
130 * Bugs: could do with some additional sanity checking, but appears to work.
134 static void req_done(struct virtqueue
*vq
)
136 struct virtio_chan
*chan
= vq
->vdev
->priv
;
139 struct p9_req_t
*req
;
142 P9_DPRINTK(P9_DEBUG_TRANS
, ": request done\n");
145 spin_lock_irqsave(&chan
->lock
, flags
);
146 rc
= virtqueue_get_buf(chan
->vq
, &len
);
149 if (!chan
->ring_bufs_avail
) {
150 chan
->ring_bufs_avail
= 1;
151 wake_up(chan
->vc_wq
);
153 spin_unlock_irqrestore(&chan
->lock
, flags
);
154 P9_DPRINTK(P9_DEBUG_TRANS
, ": rc %p\n", rc
);
155 P9_DPRINTK(P9_DEBUG_TRANS
, ": lookup tag %d\n",
157 req
= p9_tag_lookup(chan
->client
, rc
->tag
);
158 req
->status
= REQ_STATUS_RCVD
;
159 if (req
->tc
->private) {
160 struct trans_rpage_info
*rp
= req
->tc
->private;
162 p9_release_req_pages(rp
);
165 req
->tc
->private = NULL
;
167 p9_client_cb(chan
->client
, req
);
169 spin_unlock_irqrestore(&chan
->lock
, flags
);
171 } while (rc
!= NULL
);
175 * pack_sg_list - pack a scatter gather list from a linear buffer
176 * @sg: scatter/gather list to pack into
177 * @start: which segment of the sg_list to start at
178 * @limit: maximum segment to pack data to
179 * @data: data to pack into scatter/gather list
180 * @count: amount of data to pack into the scatter/gather list
182 * sg_lists have multiple segments of various sizes. This will pack
183 * arbitrary data into an existing scatter gather list, segmenting the
184 * data as necessary within constraints.
189 pack_sg_list(struct scatterlist
*sg
, int start
, int limit
, char *data
,
196 s
= rest_of_page(data
);
199 sg_set_buf(&sg
[index
++], data
, s
);
202 BUG_ON(index
> limit
);
208 /* We don't currently allow canceling of virtio requests */
209 static int p9_virtio_cancel(struct p9_client
*client
, struct p9_req_t
*req
)
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_off: Offset into the first page
220 * @**pdata: a list of pages to add into sg.
221 * @count: amount of data to pack into the scatter/gather list
224 pack_sg_list_p(struct scatterlist
*sg
, int start
, int limit
, size_t pdata_off
,
225 struct page
**pdata
, int count
)
232 s
= min((int)(PAGE_SIZE
- pdata_off
), count
);
233 sg_set_page(&sg
[index
++], pdata
[i
++], s
, pdata_off
);
238 BUG_ON(index
> limit
);
239 s
= min((int)PAGE_SIZE
, count
);
240 sg_set_page(&sg
[index
++], pdata
[i
++], s
, 0);
247 * p9_virtio_request - issue a request
248 * @client: client instance issuing the request
249 * @req: request to be issued
254 p9_virtio_request(struct p9_client
*client
, struct p9_req_t
*req
)
256 int in
, out
, inp
, outp
;
257 struct virtio_chan
*chan
= client
->trans
;
258 char *rdata
= (char *)req
->rc
+sizeof(struct p9_fcall
);
260 size_t pdata_off
= 0;
261 struct trans_rpage_info
*rpinfo
= NULL
;
262 int err
, pdata_len
= 0;
264 P9_DPRINTK(P9_DEBUG_TRANS
, "9p debug: virtio request\n");
267 req
->status
= REQ_STATUS_SENT
;
269 if (req
->tc
->pbuf_size
&& (req
->tc
->pubuf
&& P9_IS_USER_CONTEXT
)) {
270 int nr_pages
= p9_nr_pages(req
);
271 int rpinfo_size
= sizeof(struct trans_rpage_info
) +
272 sizeof(struct page
*) * nr_pages
;
274 if (rpinfo_size
<= (req
->tc
->capacity
- req
->tc
->size
)) {
275 /* We can use sdata */
276 req
->tc
->private = req
->tc
->sdata
+ req
->tc
->size
;
277 rpinfo
= (struct trans_rpage_info
*)req
->tc
->private;
278 rpinfo
->rp_alloc
= 0;
280 req
->tc
->private = kmalloc(rpinfo_size
, GFP_NOFS
);
281 if (!req
->tc
->private) {
282 P9_DPRINTK(P9_DEBUG_TRANS
, "9p debug: "
283 "private kmalloc returned NULL");
286 rpinfo
= (struct trans_rpage_info
*)req
->tc
->private;
287 rpinfo
->rp_alloc
= 1;
290 err
= p9_payload_gup(req
, &pdata_off
, &pdata_len
, nr_pages
,
291 req
->tc
->id
== P9_TREAD
? 1 : 0);
293 if (rpinfo
->rp_alloc
)
299 spin_lock_irqsave(&chan
->lock
, flags
);
301 /* Handle out VirtIO ring buffers */
302 out
= pack_sg_list(chan
->sg
, 0, VIRTQUEUE_NUM
, req
->tc
->sdata
,
305 if (req
->tc
->pbuf_size
&& (req
->tc
->id
== P9_TWRITE
)) {
306 /* We have additional write payload buffer to take care */
307 if (req
->tc
->pubuf
&& P9_IS_USER_CONTEXT
) {
308 outp
= pack_sg_list_p(chan
->sg
, out
, VIRTQUEUE_NUM
,
309 pdata_off
, rpinfo
->rp_data
, pdata_len
);
311 char *pbuf
= req
->tc
->pubuf
? req
->tc
->pubuf
:
313 outp
= pack_sg_list(chan
->sg
, out
, VIRTQUEUE_NUM
, pbuf
,
319 /* Handle in VirtIO ring buffers */
320 if (req
->tc
->pbuf_size
&&
321 ((req
->tc
->id
== P9_TREAD
) || (req
->tc
->id
== P9_TREADDIR
))) {
323 * Take care of additional Read payload.
324 * 11 is the read/write header = PDU Header(7) + IO Size (4).
325 * Arrange in such a way that server places header in the
326 * alloced memory and payload onto the user buffer.
328 inp
= pack_sg_list(chan
->sg
, out
, VIRTQUEUE_NUM
, rdata
, 11);
330 * Running executables in the filesystem may result in
331 * a read request with kernel buffer as opposed to user buffer.
333 if (req
->tc
->pubuf
&& P9_IS_USER_CONTEXT
) {
334 in
= pack_sg_list_p(chan
->sg
, out
+inp
, VIRTQUEUE_NUM
,
335 pdata_off
, rpinfo
->rp_data
, pdata_len
);
337 char *pbuf
= req
->tc
->pubuf
? req
->tc
->pubuf
:
339 in
= pack_sg_list(chan
->sg
, out
+inp
, VIRTQUEUE_NUM
,
340 pbuf
, req
->tc
->pbuf_size
);
344 in
= pack_sg_list(chan
->sg
, out
, VIRTQUEUE_NUM
, rdata
,
348 err
= virtqueue_add_buf(chan
->vq
, chan
->sg
, out
, in
, req
->tc
);
350 if (err
== -ENOSPC
) {
351 chan
->ring_bufs_avail
= 0;
352 spin_unlock_irqrestore(&chan
->lock
, flags
);
353 err
= wait_event_interruptible(*chan
->vc_wq
,
354 chan
->ring_bufs_avail
);
355 if (err
== -ERESTARTSYS
)
358 P9_DPRINTK(P9_DEBUG_TRANS
, "9p:Retry virtio request\n");
361 spin_unlock_irqrestore(&chan
->lock
, flags
);
362 P9_DPRINTK(P9_DEBUG_TRANS
,
364 "virtio rpc add_buf returned failure");
365 if (rpinfo
&& rpinfo
->rp_alloc
)
371 virtqueue_kick(chan
->vq
);
372 spin_unlock_irqrestore(&chan
->lock
, flags
);
374 P9_DPRINTK(P9_DEBUG_TRANS
, "9p debug: virtio request kicked\n");
378 static ssize_t
p9_mount_tag_show(struct device
*dev
,
379 struct device_attribute
*attr
, char *buf
)
381 struct virtio_chan
*chan
;
382 struct virtio_device
*vdev
;
384 vdev
= dev_to_virtio(dev
);
387 return snprintf(buf
, chan
->tag_len
+ 1, "%s", chan
->tag
);
390 static DEVICE_ATTR(mount_tag
, 0444, p9_mount_tag_show
, NULL
);
393 * p9_virtio_probe - probe for existence of 9P virtio channels
394 * @vdev: virtio device to probe
396 * This probes for existing virtio channels.
400 static int p9_virtio_probe(struct virtio_device
*vdev
)
405 struct virtio_chan
*chan
;
407 chan
= kmalloc(sizeof(struct virtio_chan
), GFP_KERNEL
);
409 printk(KERN_ERR
"9p: Failed to allocate virtio 9P channel\n");
416 /* We expect one virtqueue, for requests. */
417 chan
->vq
= virtio_find_single_vq(vdev
, req_done
, "requests");
418 if (IS_ERR(chan
->vq
)) {
419 err
= PTR_ERR(chan
->vq
);
422 chan
->vq
->vdev
->priv
= chan
;
423 spin_lock_init(&chan
->lock
);
425 sg_init_table(chan
->sg
, VIRTQUEUE_NUM
);
428 if (virtio_has_feature(vdev
, VIRTIO_9P_MOUNT_TAG
)) {
429 vdev
->config
->get(vdev
,
430 offsetof(struct virtio_9p_config
, tag_len
),
431 &tag_len
, sizeof(tag_len
));
436 tag
= kmalloc(tag_len
, GFP_KERNEL
);
441 vdev
->config
->get(vdev
, offsetof(struct virtio_9p_config
, tag
),
444 chan
->tag_len
= tag_len
;
445 err
= sysfs_create_file(&(vdev
->dev
.kobj
), &dev_attr_mount_tag
.attr
);
449 chan
->vc_wq
= kmalloc(sizeof(wait_queue_head_t
), GFP_KERNEL
);
454 init_waitqueue_head(chan
->vc_wq
);
455 chan
->ring_bufs_avail
= 1;
457 mutex_lock(&virtio_9p_lock
);
458 list_add_tail(&chan
->chan_list
, &virtio_chan_list
);
459 mutex_unlock(&virtio_9p_lock
);
465 vdev
->config
->del_vqs(vdev
);
473 * p9_virtio_create - allocate a new virtio channel
474 * @client: client instance invoking this transport
475 * @devname: string identifying the channel to connect to (unused)
476 * @args: args passed from sys_mount() for per-transport options (unused)
478 * This sets up a transport channel for 9p communication. Right now
479 * we only match the first available channel, but eventually we couldlook up
480 * alternate channels by matching devname versus a virtio_config entry.
481 * We use a simple reference count mechanism to ensure that only a single
482 * mount has a channel open at a time.
487 p9_virtio_create(struct p9_client
*client
, const char *devname
, char *args
)
489 struct virtio_chan
*chan
;
493 mutex_lock(&virtio_9p_lock
);
494 list_for_each_entry(chan
, &virtio_chan_list
, chan_list
) {
495 if (!strncmp(devname
, chan
->tag
, chan
->tag_len
) &&
496 strlen(devname
) == chan
->tag_len
) {
505 mutex_unlock(&virtio_9p_lock
);
508 printk(KERN_ERR
"9p: no channels available\n");
512 client
->trans
= (void *)chan
;
513 client
->status
= Connected
;
514 chan
->client
= client
;
520 * p9_virtio_remove - clean up resources associated with a virtio device
521 * @vdev: virtio device to remove
525 static void p9_virtio_remove(struct virtio_device
*vdev
)
527 struct virtio_chan
*chan
= vdev
->priv
;
530 vdev
->config
->del_vqs(vdev
);
532 mutex_lock(&virtio_9p_lock
);
533 list_del(&chan
->chan_list
);
534 mutex_unlock(&virtio_9p_lock
);
535 sysfs_remove_file(&(vdev
->dev
.kobj
), &dev_attr_mount_tag
.attr
);
542 static struct virtio_device_id id_table
[] = {
543 { VIRTIO_ID_9P
, VIRTIO_DEV_ANY_ID
},
547 static unsigned int features
[] = {
551 /* The standard "struct lguest_driver": */
552 static struct virtio_driver p9_virtio_drv
= {
553 .feature_table
= features
,
554 .feature_table_size
= ARRAY_SIZE(features
),
555 .driver
.name
= KBUILD_MODNAME
,
556 .driver
.owner
= THIS_MODULE
,
557 .id_table
= id_table
,
558 .probe
= p9_virtio_probe
,
559 .remove
= p9_virtio_remove
,
562 static struct p9_trans_module p9_virtio_trans
= {
564 .create
= p9_virtio_create
,
565 .close
= p9_virtio_close
,
566 .request
= p9_virtio_request
,
567 .cancel
= p9_virtio_cancel
,
568 .maxsize
= PAGE_SIZE
*16,
569 .pref
= P9_TRANS_PREF_PAYLOAD_SEP
,
571 .owner
= THIS_MODULE
,
574 /* The standard init function */
575 static int __init
p9_virtio_init(void)
577 INIT_LIST_HEAD(&virtio_chan_list
);
579 v9fs_register_trans(&p9_virtio_trans
);
580 return register_virtio_driver(&p9_virtio_drv
);
583 static void __exit
p9_virtio_cleanup(void)
585 unregister_virtio_driver(&p9_virtio_drv
);
586 v9fs_unregister_trans(&p9_virtio_trans
);
589 module_init(p9_virtio_init
);
590 module_exit(p9_virtio_cleanup
);
592 MODULE_DEVICE_TABLE(virtio
, id_table
);
593 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
594 MODULE_DESCRIPTION("Virtio 9p Transport");
595 MODULE_LICENSE("GPL");