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/highmem.h>
41 #include <linux/slab.h>
42 #include <net/9p/9p.h>
43 #include <linux/parser.h>
44 #include <net/9p/client.h>
45 #include <net/9p/transport.h>
46 #include <linux/scatterlist.h>
47 #include <linux/swap.h>
48 #include <linux/virtio.h>
49 #include <linux/virtio_9p.h>
50 #include "trans_common.h"
52 #define VIRTQUEUE_NUM 128
54 /* a single mutex to manage channel initialization and attachment */
55 static DEFINE_MUTEX(virtio_9p_lock);
56 static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
57 static atomic_t vp_pinned = ATOMIC_INIT(0);
60 * struct virtio_chan - per-instance transport information
61 * @initialized: whether the channel is initialized
62 * @inuse: whether the channel is in use
63 * @lock: protects multiple elements within this structure
64 * @client: client instance
65 * @vdev: virtio dev associated with this channel
66 * @vq: virtio queue associated with this channel
67 * @sg: scatter gather list which is used to pack a request (protected?)
69 * We keep all per-channel information in a structure.
70 * This structure is allocated within the devices dev->mem space.
71 * A pointer to the structure will get put in the transport private.
80 struct p9_client *client;
81 struct virtio_device *vdev;
84 wait_queue_head_t *vc_wq;
85 /* This is global limit. Since we don't have a global structure,
86 * will be placing it in each channel.
89 /* Scatterlist: can be too big for stack. */
90 struct scatterlist sg[VIRTQUEUE_NUM];
94 * tag name to identify a mount Non-null terminated
98 struct list_head chan_list;
101 static struct list_head virtio_chan_list;
103 /* How many bytes left in this page. */
104 static unsigned int rest_of_page(void *data)
106 return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE);
110 * p9_virtio_close - reclaim resources of a channel
111 * @client: client instance
113 * This reclaims a channel by freeing its resources and
114 * reseting its inuse flag.
118 static void p9_virtio_close(struct p9_client *client)
120 struct virtio_chan *chan = client->trans;
122 mutex_lock(&virtio_9p_lock);
125 mutex_unlock(&virtio_9p_lock);
129 * req_done - callback which signals activity from the server
130 * @vq: virtio queue activity was received on
132 * This notifies us that the server has triggered some activity
133 * on the virtio channel - most likely a response to request we
134 * sent. Figure out which requests now have responses and wake up
137 * Bugs: could do with some additional sanity checking, but appears to work.
141 static void req_done(struct virtqueue *vq)
143 struct virtio_chan *chan = vq->vdev->priv;
146 struct p9_req_t *req;
149 P9_DPRINTK(P9_DEBUG_TRANS, ": request done\n");
152 spin_lock_irqsave(&chan->lock, flags);
153 rc = virtqueue_get_buf(chan->vq, &len);
155 spin_unlock_irqrestore(&chan->lock, flags);
158 chan->ring_bufs_avail = 1;
159 spin_unlock_irqrestore(&chan->lock, flags);
160 /* Wakeup if anyone waiting for VirtIO ring space. */
161 wake_up(chan->vc_wq);
162 P9_DPRINTK(P9_DEBUG_TRANS, ": rc %p\n", rc);
163 P9_DPRINTK(P9_DEBUG_TRANS, ": lookup tag %d\n", rc->tag);
164 req = p9_tag_lookup(chan->client, rc->tag);
165 req->status = REQ_STATUS_RCVD;
166 p9_client_cb(chan->client, req);
171 * pack_sg_list - pack a scatter gather list from a linear buffer
172 * @sg: scatter/gather list to pack into
173 * @start: which segment of the sg_list to start at
174 * @limit: maximum segment to pack data to
175 * @data: data to pack into scatter/gather list
176 * @count: amount of data to pack into the scatter/gather list
178 * sg_lists have multiple segments of various sizes. This will pack
179 * arbitrary data into an existing scatter gather list, segmenting the
180 * data as necessary within constraints.
184 static int pack_sg_list(struct scatterlist *sg, int start,
185 int limit, char *data, int count)
191 s = rest_of_page(data);
194 sg_set_buf(&sg[index++], data, s);
197 BUG_ON(index > limit);
203 /* We don't currently allow canceling of virtio requests */
204 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
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 * @data: data to pack into scatter/gather list
217 * @count: amount of data to pack into the scatter/gather list
220 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
221 struct page **pdata, int nr_pages, char *data, int count)
227 BUG_ON(nr_pages > (limit - start));
229 * if the first page doesn't start at
230 * page boundary find the offset
232 data_off = offset_in_page(data);
234 s = rest_of_page(data);
237 sg_set_page(&sg[index++], pdata[i++], s, data_off);
243 return index - start;
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)
259 struct virtio_chan *chan = client->trans;
261 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request\n");
263 req->status = REQ_STATUS_SENT;
265 spin_lock_irqsave(&chan->lock, flags);
267 /* Handle out VirtIO ring buffers */
268 out = pack_sg_list(chan->sg, 0,
269 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
271 in = pack_sg_list(chan->sg, out,
272 VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity);
274 err = virtqueue_add_buf(chan->vq, chan->sg, out, in, req->tc);
276 if (err == -ENOSPC) {
277 chan->ring_bufs_avail = 0;
278 spin_unlock_irqrestore(&chan->lock, flags);
279 err = wait_event_interruptible(*chan->vc_wq,
280 chan->ring_bufs_avail);
281 if (err == -ERESTARTSYS)
284 P9_DPRINTK(P9_DEBUG_TRANS, "9p:Retry virtio request\n");
287 spin_unlock_irqrestore(&chan->lock, flags);
288 P9_DPRINTK(P9_DEBUG_TRANS,
290 "virtio rpc add_buf returned failure");
294 virtqueue_kick(chan->vq);
295 spin_unlock_irqrestore(&chan->lock, flags);
297 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request kicked\n");
301 static int p9_get_mapped_pages(struct virtio_chan *chan,
302 struct page **pages, char *data,
303 int nr_pages, int write, int kern_buf)
308 * We allow only p9_max_pages pinned. We wait for the
309 * Other zc request to finish here
311 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
312 err = wait_event_interruptible(vp_wq,
313 (atomic_read(&vp_pinned) < chan->p9_max_pages));
314 if (err == -ERESTARTSYS)
317 err = p9_payload_gup(data, &nr_pages, pages, write);
320 atomic_add(nr_pages, &vp_pinned);
322 /* kernel buffer, no need to pin pages */
324 int count = nr_pages;
326 s = rest_of_page(data);
327 pages[index++] = kmap_to_page(data);
337 * p9_virtio_zc_request - issue a zero copy request
338 * @client: client instance issuing the request
339 * @req: request to be issued
340 * @uidata: user bffer that should be ued for zero copy read
341 * @uodata: user buffer that shoud be user for zero copy write
342 * @inlen: read buffer size
343 * @olen: write buffer size
344 * @hdrlen: reader header size, This is the size of response protocol data
348 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
349 char *uidata, char *uodata, int inlen,
350 int outlen, int in_hdr_len, int kern_buf)
354 int in_nr_pages = 0, out_nr_pages = 0;
355 struct page **in_pages = NULL, **out_pages = NULL;
356 struct virtio_chan *chan = client->trans;
358 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request\n");
361 out_nr_pages = p9_nr_pages(uodata, outlen);
362 out_pages = kmalloc(sizeof(struct page *) * out_nr_pages,
368 out_nr_pages = p9_get_mapped_pages(chan, out_pages, uodata,
369 out_nr_pages, 0, kern_buf);
370 if (out_nr_pages < 0) {
378 in_nr_pages = p9_nr_pages(uidata, inlen);
379 in_pages = kmalloc(sizeof(struct page *) * in_nr_pages,
385 in_nr_pages = p9_get_mapped_pages(chan, in_pages, uidata,
386 in_nr_pages, 1, kern_buf);
387 if (in_nr_pages < 0) {
394 req->status = REQ_STATUS_SENT;
396 spin_lock_irqsave(&chan->lock, flags);
398 out = pack_sg_list(chan->sg, 0,
399 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
402 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
403 out_pages, out_nr_pages, uodata, outlen);
405 * Take care of in data
406 * For example TREAD have 11.
407 * 11 is the read/write header = PDU Header(7) + IO Size (4).
408 * Arrange in such a way that server places header in the
409 * alloced memory and payload onto the user buffer.
411 in = pack_sg_list(chan->sg, out,
412 VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len);
414 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
415 in_pages, in_nr_pages, uidata, inlen);
417 err = virtqueue_add_buf(chan->vq, chan->sg, out, in, req->tc);
419 if (err == -ENOSPC) {
420 chan->ring_bufs_avail = 0;
421 spin_unlock_irqrestore(&chan->lock, flags);
422 err = wait_event_interruptible(*chan->vc_wq,
423 chan->ring_bufs_avail);
424 if (err == -ERESTARTSYS)
427 P9_DPRINTK(P9_DEBUG_TRANS, "9p:Retry virtio request\n");
428 goto req_retry_pinned;
430 spin_unlock_irqrestore(&chan->lock, flags);
431 P9_DPRINTK(P9_DEBUG_TRANS,
433 "virtio rpc add_buf returned failure");
438 virtqueue_kick(chan->vq);
439 spin_unlock_irqrestore(&chan->lock, flags);
440 P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio request kicked\n");
441 err = wait_event_interruptible(*req->wq,
442 req->status >= REQ_STATUS_RCVD);
444 * Non kernel buffers are pinned, unpin them
449 p9_release_pages(in_pages, in_nr_pages);
450 atomic_sub(in_nr_pages, &vp_pinned);
453 p9_release_pages(out_pages, out_nr_pages);
454 atomic_sub(out_nr_pages, &vp_pinned);
456 /* wakeup anybody waiting for slots to pin pages */
464 static ssize_t p9_mount_tag_show(struct device *dev,
465 struct device_attribute *attr, char *buf)
467 struct virtio_chan *chan;
468 struct virtio_device *vdev;
470 vdev = dev_to_virtio(dev);
473 return snprintf(buf, chan->tag_len + 1, "%s", chan->tag);
476 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
479 * p9_virtio_probe - probe for existence of 9P virtio channels
480 * @vdev: virtio device to probe
482 * This probes for existing virtio channels.
486 static int p9_virtio_probe(struct virtio_device *vdev)
491 struct virtio_chan *chan;
493 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
495 printk(KERN_ERR "9p: Failed to allocate virtio 9P channel\n");
502 /* We expect one virtqueue, for requests. */
503 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
504 if (IS_ERR(chan->vq)) {
505 err = PTR_ERR(chan->vq);
508 chan->vq->vdev->priv = chan;
509 spin_lock_init(&chan->lock);
511 sg_init_table(chan->sg, VIRTQUEUE_NUM);
514 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
515 vdev->config->get(vdev,
516 offsetof(struct virtio_9p_config, tag_len),
517 &tag_len, sizeof(tag_len));
522 tag = kmalloc(tag_len, GFP_KERNEL);
527 vdev->config->get(vdev, offsetof(struct virtio_9p_config, tag),
530 chan->tag_len = tag_len;
531 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
535 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
540 init_waitqueue_head(chan->vc_wq);
541 chan->ring_bufs_avail = 1;
542 /* Ceiling limit to avoid denial of service attacks */
543 chan->p9_max_pages = nr_free_buffer_pages()/4;
545 mutex_lock(&virtio_9p_lock);
546 list_add_tail(&chan->chan_list, &virtio_chan_list);
547 mutex_unlock(&virtio_9p_lock);
553 vdev->config->del_vqs(vdev);
561 * p9_virtio_create - allocate a new virtio channel
562 * @client: client instance invoking this transport
563 * @devname: string identifying the channel to connect to (unused)
564 * @args: args passed from sys_mount() for per-transport options (unused)
566 * This sets up a transport channel for 9p communication. Right now
567 * we only match the first available channel, but eventually we couldlook up
568 * alternate channels by matching devname versus a virtio_config entry.
569 * We use a simple reference count mechanism to ensure that only a single
570 * mount has a channel open at a time.
575 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
577 struct virtio_chan *chan;
581 mutex_lock(&virtio_9p_lock);
582 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
583 if (!strncmp(devname, chan->tag, chan->tag_len) &&
584 strlen(devname) == chan->tag_len) {
593 mutex_unlock(&virtio_9p_lock);
596 printk(KERN_ERR "9p: no channels available\n");
600 client->trans = (void *)chan;
601 client->status = Connected;
602 chan->client = client;
608 * p9_virtio_remove - clean up resources associated with a virtio device
609 * @vdev: virtio device to remove
613 static void p9_virtio_remove(struct virtio_device *vdev)
615 struct virtio_chan *chan = vdev->priv;
618 vdev->config->del_vqs(vdev);
620 mutex_lock(&virtio_9p_lock);
621 list_del(&chan->chan_list);
622 mutex_unlock(&virtio_9p_lock);
623 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
630 static struct virtio_device_id id_table[] = {
631 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
635 static unsigned int features[] = {
639 /* The standard "struct lguest_driver": */
640 static struct virtio_driver p9_virtio_drv = {
641 .feature_table = features,
642 .feature_table_size = ARRAY_SIZE(features),
643 .driver.name = KBUILD_MODNAME,
644 .driver.owner = THIS_MODULE,
645 .id_table = id_table,
646 .probe = p9_virtio_probe,
647 .remove = p9_virtio_remove,
650 static struct p9_trans_module p9_virtio_trans = {
652 .create = p9_virtio_create,
653 .close = p9_virtio_close,
654 .request = p9_virtio_request,
655 .zc_request = p9_virtio_zc_request,
656 .cancel = p9_virtio_cancel,
658 * We leave one entry for input and one entry for response
659 * headers. We also skip one more entry to accomodate, address
660 * that are not at page boundary, that can result in an extra
663 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
665 .owner = THIS_MODULE,
668 /* The standard init function */
669 static int __init p9_virtio_init(void)
671 INIT_LIST_HEAD(&virtio_chan_list);
673 v9fs_register_trans(&p9_virtio_trans);
674 return register_virtio_driver(&p9_virtio_drv);
677 static void __exit p9_virtio_cleanup(void)
679 unregister_virtio_driver(&p9_virtio_drv);
680 v9fs_unregister_trans(&p9_virtio_trans);
683 module_init(p9_virtio_init);
684 module_exit(p9_virtio_cleanup);
686 MODULE_DEVICE_TABLE(virtio, id_table);
687 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
688 MODULE_DESCRIPTION("Virtio 9p Transport");
689 MODULE_LICENSE("GPL");