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
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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"
54 #define VIRTQUEUE_NUM 128
56 /* a single mutex to manage channel initialization and attachment */
57 static DEFINE_MUTEX(virtio_9p_lock
);
58 static DECLARE_WAIT_QUEUE_HEAD(vp_wq
);
59 static atomic_t vp_pinned
= ATOMIC_INIT(0);
62 * struct virtio_chan - per-instance transport information
63 * @inuse: whether the channel is in use
64 * @lock: protects multiple elements within this structure
65 * @client: client instance
66 * @vdev: virtio dev associated with this channel
67 * @vq: virtio queue associated with this channel
68 * @sg: scatter gather list which is used to pack a request (protected?)
70 * We keep all per-channel information in a structure.
71 * This structure is allocated within the devices dev->mem space.
72 * A pointer to the structure will get put in the transport private.
81 struct p9_client
*client
;
82 struct virtio_device
*vdev
;
85 wait_queue_head_t
*vc_wq
;
86 /* This is global limit. Since we don't have a global structure,
87 * will be placing it in each channel.
89 unsigned long p9_max_pages
;
90 /* Scatterlist: can be too big for stack. */
91 struct scatterlist sg
[VIRTQUEUE_NUM
];
93 * tag name to identify a mount null terminated
97 struct list_head chan_list
;
100 static struct list_head virtio_chan_list
;
102 /* How many bytes left in this page. */
103 static unsigned int rest_of_page(void *data
)
105 return PAGE_SIZE
- offset_in_page(data
);
109 * p9_virtio_close - reclaim resources of a channel
110 * @client: client instance
112 * This reclaims a channel by freeing its resources and
113 * reseting its inuse flag.
117 static void p9_virtio_close(struct p9_client
*client
)
119 struct virtio_chan
*chan
= client
->trans
;
121 mutex_lock(&virtio_9p_lock
);
124 mutex_unlock(&virtio_9p_lock
);
128 * req_done - callback which signals activity from the server
129 * @vq: virtio queue activity was received on
131 * This notifies us that the server has triggered some activity
132 * on the virtio channel - most likely a response to request we
133 * sent. Figure out which requests now have responses and wake up
136 * Bugs: could do with some additional sanity checking, but appears to work.
140 static void req_done(struct virtqueue
*vq
)
142 struct virtio_chan
*chan
= vq
->vdev
->priv
;
144 struct p9_req_t
*req
;
145 bool need_wakeup
= false;
148 p9_debug(P9_DEBUG_TRANS
, ": request done\n");
150 spin_lock_irqsave(&chan
->lock
, flags
);
151 while ((req
= virtqueue_get_buf(chan
->vq
, &len
)) != NULL
) {
152 if (!chan
->ring_bufs_avail
) {
153 chan
->ring_bufs_avail
= 1;
159 p9_client_cb(chan
->client
, req
, REQ_STATUS_RCVD
);
162 spin_unlock_irqrestore(&chan
->lock
, flags
);
163 /* Wakeup if anyone waiting for VirtIO ring space. */
165 wake_up(chan
->vc_wq
);
169 * pack_sg_list - pack a scatter gather list from a linear buffer
170 * @sg: scatter/gather list to pack into
171 * @start: which segment of the sg_list to start at
172 * @limit: maximum segment to pack data to
173 * @data: data to pack into scatter/gather list
174 * @count: amount of data to pack into the scatter/gather list
176 * sg_lists have multiple segments of various sizes. This will pack
177 * arbitrary data into an existing scatter gather list, segmenting the
178 * data as necessary within constraints.
182 static int pack_sg_list(struct scatterlist
*sg
, int start
,
183 int limit
, char *data
, int count
)
189 s
= rest_of_page(data
);
192 BUG_ON(index
>= limit
);
193 /* Make sure we don't terminate early. */
194 sg_unmark_end(&sg
[index
]);
195 sg_set_buf(&sg
[index
++], data
, s
);
200 sg_mark_end(&sg
[index
- 1]);
204 /* We don't currently allow canceling of virtio requests */
205 static int p9_virtio_cancel(struct p9_client
*client
, struct p9_req_t
*req
)
210 /* Reply won't come, so drop req ref */
211 static int p9_virtio_cancelled(struct p9_client
*client
, struct p9_req_t
*req
)
218 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
219 * this takes a list of pages.
220 * @sg: scatter/gather list to pack into
221 * @start: which segment of the sg_list to start at
222 * @pdata: a list of pages to add into sg.
223 * @nr_pages: number of pages to pack into the scatter/gather list
224 * @offs: amount of data in the beginning of first page _not_ to pack
225 * @count: amount of data to pack into the scatter/gather list
228 pack_sg_list_p(struct scatterlist
*sg
, int start
, int limit
,
229 struct page
**pdata
, int nr_pages
, size_t offs
, int count
)
235 BUG_ON(nr_pages
> (limit
- start
));
237 * if the first page doesn't start at
238 * page boundary find the offset
241 s
= PAGE_SIZE
- data_off
;
244 BUG_ON(index
>= limit
);
245 /* Make sure we don't terminate early. */
246 sg_unmark_end(&sg
[index
]);
247 sg_set_page(&sg
[index
++], pdata
[i
++], s
, data_off
);
254 sg_mark_end(&sg
[index
- 1]);
255 return index
- start
;
259 * p9_virtio_request - issue a request
260 * @client: client instance issuing the request
261 * @req: request to be issued
266 p9_virtio_request(struct p9_client
*client
, struct p9_req_t
*req
)
269 int in
, out
, out_sgs
, in_sgs
;
271 struct virtio_chan
*chan
= client
->trans
;
272 struct scatterlist
*sgs
[2];
274 p9_debug(P9_DEBUG_TRANS
, "9p debug: virtio request\n");
276 req
->status
= REQ_STATUS_SENT
;
278 spin_lock_irqsave(&chan
->lock
, flags
);
280 out_sgs
= in_sgs
= 0;
281 /* Handle out VirtIO ring buffers */
282 out
= pack_sg_list(chan
->sg
, 0,
283 VIRTQUEUE_NUM
, req
->tc
.sdata
, req
->tc
.size
);
285 sgs
[out_sgs
++] = chan
->sg
;
287 in
= pack_sg_list(chan
->sg
, out
,
288 VIRTQUEUE_NUM
, req
->rc
.sdata
, req
->rc
.capacity
);
290 sgs
[out_sgs
+ in_sgs
++] = chan
->sg
+ out
;
292 err
= virtqueue_add_sgs(chan
->vq
, sgs
, out_sgs
, in_sgs
, req
,
295 if (err
== -ENOSPC
) {
296 chan
->ring_bufs_avail
= 0;
297 spin_unlock_irqrestore(&chan
->lock
, flags
);
298 err
= wait_event_killable(*chan
->vc_wq
,
299 chan
->ring_bufs_avail
);
300 if (err
== -ERESTARTSYS
)
303 p9_debug(P9_DEBUG_TRANS
, "Retry virtio request\n");
306 spin_unlock_irqrestore(&chan
->lock
, flags
);
307 p9_debug(P9_DEBUG_TRANS
,
308 "virtio rpc add_sgs returned failure\n");
312 virtqueue_kick(chan
->vq
);
313 spin_unlock_irqrestore(&chan
->lock
, flags
);
315 p9_debug(P9_DEBUG_TRANS
, "virtio request kicked\n");
319 static int p9_get_mapped_pages(struct virtio_chan
*chan
,
320 struct page
***pages
,
321 struct iov_iter
*data
,
329 if (!iov_iter_count(data
))
332 if (!(data
->type
& ITER_KVEC
)) {
335 * We allow only p9_max_pages pinned. We wait for the
336 * Other zc request to finish here
338 if (atomic_read(&vp_pinned
) >= chan
->p9_max_pages
) {
339 err
= wait_event_killable(vp_wq
,
340 (atomic_read(&vp_pinned
) < chan
->p9_max_pages
));
341 if (err
== -ERESTARTSYS
)
344 n
= iov_iter_get_pages_alloc(data
, pages
, count
, offs
);
348 nr_pages
= DIV_ROUND_UP(n
+ *offs
, PAGE_SIZE
);
349 atomic_add(nr_pages
, &vp_pinned
);
352 /* kernel buffer, no need to pin pages */
357 /* we'd already checked that it's non-empty */
359 len
= iov_iter_single_seg_count(data
);
361 p
= data
->kvec
->iov_base
+ data
->iov_offset
;
364 iov_iter_advance(data
, 0);
369 nr_pages
= DIV_ROUND_UP((unsigned long)p
+ len
, PAGE_SIZE
) -
370 (unsigned long)p
/ PAGE_SIZE
;
372 *pages
= kmalloc_array(nr_pages
, sizeof(struct page
*),
378 p
-= (*offs
= offset_in_page(p
));
379 for (index
= 0; index
< nr_pages
; index
++) {
380 if (is_vmalloc_addr(p
))
381 (*pages
)[index
] = vmalloc_to_page(p
);
383 (*pages
)[index
] = kmap_to_page(p
);
391 * p9_virtio_zc_request - issue a zero copy request
392 * @client: client instance issuing the request
393 * @req: request to be issued
394 * @uidata: user buffer that should be used for zero copy read
395 * @uodata: user buffer that should be used for zero copy write
396 * @inlen: read buffer size
397 * @outlen: write buffer size
398 * @in_hdr_len: reader header size, This is the size of response protocol data
402 p9_virtio_zc_request(struct p9_client
*client
, struct p9_req_t
*req
,
403 struct iov_iter
*uidata
, struct iov_iter
*uodata
,
404 int inlen
, int outlen
, int in_hdr_len
)
406 int in
, out
, err
, out_sgs
, in_sgs
;
408 int in_nr_pages
= 0, out_nr_pages
= 0;
409 struct page
**in_pages
= NULL
, **out_pages
= NULL
;
410 struct virtio_chan
*chan
= client
->trans
;
411 struct scatterlist
*sgs
[4];
416 p9_debug(P9_DEBUG_TRANS
, "virtio request\n");
420 int n
= p9_get_mapped_pages(chan
, &out_pages
, uodata
,
421 outlen
, &offs
, &need_drop
);
426 out_nr_pages
= DIV_ROUND_UP(n
+ offs
, PAGE_SIZE
);
428 __le32 v
= cpu_to_le32(n
);
429 memcpy(&req
->tc
.sdata
[req
->tc
.size
- 4], &v
, 4);
432 /* The size field of the message must include the length of the
433 * header and the length of the data. We didn't actually know
434 * the length of the data until this point so add it in now.
436 sz
= cpu_to_le32(req
->tc
.size
+ outlen
);
437 memcpy(&req
->tc
.sdata
[0], &sz
, sizeof(sz
));
439 int n
= p9_get_mapped_pages(chan
, &in_pages
, uidata
,
440 inlen
, &offs
, &need_drop
);
445 in_nr_pages
= DIV_ROUND_UP(n
+ offs
, PAGE_SIZE
);
447 __le32 v
= cpu_to_le32(n
);
448 memcpy(&req
->tc
.sdata
[req
->tc
.size
- 4], &v
, 4);
452 req
->status
= REQ_STATUS_SENT
;
454 spin_lock_irqsave(&chan
->lock
, flags
);
456 out_sgs
= in_sgs
= 0;
459 out
= pack_sg_list(chan
->sg
, 0,
460 VIRTQUEUE_NUM
, req
->tc
.sdata
, req
->tc
.size
);
463 sgs
[out_sgs
++] = chan
->sg
;
466 sgs
[out_sgs
++] = chan
->sg
+ out
;
467 out
+= pack_sg_list_p(chan
->sg
, out
, VIRTQUEUE_NUM
,
468 out_pages
, out_nr_pages
, offs
, outlen
);
472 * Take care of in data
473 * For example TREAD have 11.
474 * 11 is the read/write header = PDU Header(7) + IO Size (4).
475 * Arrange in such a way that server places header in the
476 * alloced memory and payload onto the user buffer.
478 in
= pack_sg_list(chan
->sg
, out
,
479 VIRTQUEUE_NUM
, req
->rc
.sdata
, in_hdr_len
);
481 sgs
[out_sgs
+ in_sgs
++] = chan
->sg
+ out
;
484 sgs
[out_sgs
+ in_sgs
++] = chan
->sg
+ out
+ in
;
485 in
+= pack_sg_list_p(chan
->sg
, out
+ in
, VIRTQUEUE_NUM
,
486 in_pages
, in_nr_pages
, offs
, inlen
);
489 BUG_ON(out_sgs
+ in_sgs
> ARRAY_SIZE(sgs
));
490 err
= virtqueue_add_sgs(chan
->vq
, sgs
, out_sgs
, in_sgs
, req
,
493 if (err
== -ENOSPC
) {
494 chan
->ring_bufs_avail
= 0;
495 spin_unlock_irqrestore(&chan
->lock
, flags
);
496 err
= wait_event_killable(*chan
->vc_wq
,
497 chan
->ring_bufs_avail
);
498 if (err
== -ERESTARTSYS
)
501 p9_debug(P9_DEBUG_TRANS
, "Retry virtio request\n");
502 goto req_retry_pinned
;
504 spin_unlock_irqrestore(&chan
->lock
, flags
);
505 p9_debug(P9_DEBUG_TRANS
,
506 "virtio rpc add_sgs returned failure\n");
511 virtqueue_kick(chan
->vq
);
512 spin_unlock_irqrestore(&chan
->lock
, flags
);
514 p9_debug(P9_DEBUG_TRANS
, "virtio request kicked\n");
515 err
= wait_event_killable(req
->wq
, req
->status
>= REQ_STATUS_RCVD
);
517 * Non kernel buffers are pinned, unpin them
522 p9_release_pages(in_pages
, in_nr_pages
);
523 atomic_sub(in_nr_pages
, &vp_pinned
);
526 p9_release_pages(out_pages
, out_nr_pages
);
527 atomic_sub(out_nr_pages
, &vp_pinned
);
529 /* wakeup anybody waiting for slots to pin pages */
535 /* reply won't come */
541 static ssize_t
p9_mount_tag_show(struct device
*dev
,
542 struct device_attribute
*attr
, char *buf
)
544 struct virtio_chan
*chan
;
545 struct virtio_device
*vdev
;
548 vdev
= dev_to_virtio(dev
);
550 tag_len
= strlen(chan
->tag
);
552 memcpy(buf
, chan
->tag
, tag_len
+ 1);
557 static DEVICE_ATTR(mount_tag
, 0444, p9_mount_tag_show
, NULL
);
560 * p9_virtio_probe - probe for existence of 9P virtio channels
561 * @vdev: virtio device to probe
563 * This probes for existing virtio channels.
567 static int p9_virtio_probe(struct virtio_device
*vdev
)
572 struct virtio_chan
*chan
;
574 if (!vdev
->config
->get
) {
575 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
580 chan
= kmalloc(sizeof(struct virtio_chan
), GFP_KERNEL
);
582 pr_err("Failed to allocate virtio 9P channel\n");
589 /* We expect one virtqueue, for requests. */
590 chan
->vq
= virtio_find_single_vq(vdev
, req_done
, "requests");
591 if (IS_ERR(chan
->vq
)) {
592 err
= PTR_ERR(chan
->vq
);
595 chan
->vq
->vdev
->priv
= chan
;
596 spin_lock_init(&chan
->lock
);
598 sg_init_table(chan
->sg
, VIRTQUEUE_NUM
);
601 if (virtio_has_feature(vdev
, VIRTIO_9P_MOUNT_TAG
)) {
602 virtio_cread(vdev
, struct virtio_9p_config
, tag_len
, &tag_len
);
607 tag
= kzalloc(tag_len
+ 1, GFP_KERNEL
);
613 virtio_cread_bytes(vdev
, offsetof(struct virtio_9p_config
, tag
),
616 err
= sysfs_create_file(&(vdev
->dev
.kobj
), &dev_attr_mount_tag
.attr
);
620 chan
->vc_wq
= kmalloc(sizeof(wait_queue_head_t
), GFP_KERNEL
);
625 init_waitqueue_head(chan
->vc_wq
);
626 chan
->ring_bufs_avail
= 1;
627 /* Ceiling limit to avoid denial of service attacks */
628 chan
->p9_max_pages
= nr_free_buffer_pages()/4;
630 virtio_device_ready(vdev
);
632 mutex_lock(&virtio_9p_lock
);
633 list_add_tail(&chan
->chan_list
, &virtio_chan_list
);
634 mutex_unlock(&virtio_9p_lock
);
636 /* Let udev rules use the new mount_tag attribute. */
637 kobject_uevent(&(vdev
->dev
.kobj
), KOBJ_CHANGE
);
644 vdev
->config
->del_vqs(vdev
);
653 * p9_virtio_create - allocate a new virtio channel
654 * @client: client instance invoking this transport
655 * @devname: string identifying the channel to connect to (unused)
656 * @args: args passed from sys_mount() for per-transport options (unused)
658 * This sets up a transport channel for 9p communication. Right now
659 * we only match the first available channel, but eventually we couldlook up
660 * alternate channels by matching devname versus a virtio_config entry.
661 * We use a simple reference count mechanism to ensure that only a single
662 * mount has a channel open at a time.
667 p9_virtio_create(struct p9_client
*client
, const char *devname
, char *args
)
669 struct virtio_chan
*chan
;
676 mutex_lock(&virtio_9p_lock
);
677 list_for_each_entry(chan
, &virtio_chan_list
, chan_list
) {
678 if (!strcmp(devname
, chan
->tag
)) {
687 mutex_unlock(&virtio_9p_lock
);
690 pr_err("no channels available for device %s\n", devname
);
694 client
->trans
= (void *)chan
;
695 client
->status
= Connected
;
696 chan
->client
= client
;
702 * p9_virtio_remove - clean up resources associated with a virtio device
703 * @vdev: virtio device to remove
707 static void p9_virtio_remove(struct virtio_device
*vdev
)
709 struct virtio_chan
*chan
= vdev
->priv
;
710 unsigned long warning_time
;
712 mutex_lock(&virtio_9p_lock
);
714 /* Remove self from list so we don't get new users. */
715 list_del(&chan
->chan_list
);
716 warning_time
= jiffies
;
718 /* Wait for existing users to close. */
719 while (chan
->inuse
) {
720 mutex_unlock(&virtio_9p_lock
);
722 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
723 dev_emerg(&vdev
->dev
,
724 "p9_virtio_remove: waiting for device in use.\n");
725 warning_time
= jiffies
;
727 mutex_lock(&virtio_9p_lock
);
730 mutex_unlock(&virtio_9p_lock
);
732 vdev
->config
->reset(vdev
);
733 vdev
->config
->del_vqs(vdev
);
735 sysfs_remove_file(&(vdev
->dev
.kobj
), &dev_attr_mount_tag
.attr
);
736 kobject_uevent(&(vdev
->dev
.kobj
), KOBJ_CHANGE
);
743 static struct virtio_device_id id_table
[] = {
744 { VIRTIO_ID_9P
, VIRTIO_DEV_ANY_ID
},
748 static unsigned int features
[] = {
752 /* The standard "struct lguest_driver": */
753 static struct virtio_driver p9_virtio_drv
= {
754 .feature_table
= features
,
755 .feature_table_size
= ARRAY_SIZE(features
),
756 .driver
.name
= KBUILD_MODNAME
,
757 .driver
.owner
= THIS_MODULE
,
758 .id_table
= id_table
,
759 .probe
= p9_virtio_probe
,
760 .remove
= p9_virtio_remove
,
763 static struct p9_trans_module p9_virtio_trans
= {
765 .create
= p9_virtio_create
,
766 .close
= p9_virtio_close
,
767 .request
= p9_virtio_request
,
768 .zc_request
= p9_virtio_zc_request
,
769 .cancel
= p9_virtio_cancel
,
770 .cancelled
= p9_virtio_cancelled
,
772 * We leave one entry for input and one entry for response
773 * headers. We also skip one more entry to accomodate, address
774 * that are not at page boundary, that can result in an extra
777 .maxsize
= PAGE_SIZE
* (VIRTQUEUE_NUM
- 3),
779 .owner
= THIS_MODULE
,
782 /* The standard init function */
783 static int __init
p9_virtio_init(void)
787 INIT_LIST_HEAD(&virtio_chan_list
);
789 v9fs_register_trans(&p9_virtio_trans
);
790 rc
= register_virtio_driver(&p9_virtio_drv
);
792 v9fs_unregister_trans(&p9_virtio_trans
);
797 static void __exit
p9_virtio_cleanup(void)
799 unregister_virtio_driver(&p9_virtio_drv
);
800 v9fs_unregister_trans(&p9_virtio_trans
);
803 module_init(p9_virtio_init
);
804 module_exit(p9_virtio_cleanup
);
806 MODULE_DEVICE_TABLE(virtio
, id_table
);
807 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
808 MODULE_DESCRIPTION("Virtio 9p Transport");
809 MODULE_LICENSE("GPL");