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 * @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?)
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.
82 struct p9_client *client;
83 struct virtio_device *vdev;
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.
90 unsigned long p9_max_pages;
91 /* Scatterlist: can be too big for stack. */
92 struct scatterlist sg[VIRTQUEUE_NUM];
96 * tag name to identify a mount Non-null terminated
100 struct list_head chan_list;
103 static struct list_head virtio_chan_list;
105 /* How many bytes left in this page. */
106 static unsigned int rest_of_page(void *data)
108 return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE);
112 * p9_virtio_close - reclaim resources of a channel
113 * @client: client instance
115 * This reclaims a channel by freeing its resources and
116 * reseting its inuse flag.
120 static void p9_virtio_close(struct p9_client *client)
122 struct virtio_chan *chan = client->trans;
124 mutex_lock(&virtio_9p_lock);
127 mutex_unlock(&virtio_9p_lock);
131 * req_done - callback which signals activity from the server
132 * @vq: virtio queue activity was received on
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
139 * Bugs: could do with some additional sanity checking, but appears to work.
143 static void req_done(struct virtqueue *vq)
145 struct virtio_chan *chan = vq->vdev->priv;
148 struct p9_req_t *req;
151 p9_debug(P9_DEBUG_TRANS, ": request done\n");
154 spin_lock_irqsave(&chan->lock, flags);
155 rc = virtqueue_get_buf(chan->vq, &len);
157 spin_unlock_irqrestore(&chan->lock, flags);
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 p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
172 * pack_sg_list - pack a scatter gather list from a linear buffer
173 * @sg: scatter/gather list to pack into
174 * @start: which segment of the sg_list to start at
175 * @limit: maximum segment to pack data to
176 * @data: data to pack into scatter/gather list
177 * @count: amount of data to pack into the scatter/gather list
179 * sg_lists have multiple segments of various sizes. This will pack
180 * arbitrary data into an existing scatter gather list, segmenting the
181 * data as necessary within constraints.
185 static int pack_sg_list(struct scatterlist *sg, int start,
186 int limit, char *data, int count)
192 s = rest_of_page(data);
195 BUG_ON(index > limit);
196 /* Make sure we don't terminate early. */
197 sg_unmark_end(&sg[index]);
198 sg_set_buf(&sg[index++], data, s);
203 sg_mark_end(&sg[index - 1]);
207 /* We don't currently allow canceling of virtio requests */
208 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
214 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
215 * this takes a list of pages.
216 * @sg: scatter/gather list to pack into
217 * @start: which segment of the sg_list to start at
218 * @pdata: a list of pages to add into sg.
219 * @nr_pages: number of pages to pack into the scatter/gather list
220 * @data: data to pack into scatter/gather list
221 * @count: amount of data to pack into the scatter/gather list
224 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
225 struct page **pdata, int nr_pages, char *data, int count)
231 BUG_ON(nr_pages > (limit - start));
233 * if the first page doesn't start at
234 * page boundary find the offset
236 data_off = offset_in_page(data);
238 s = rest_of_page(data);
241 /* Make sure we don't terminate early. */
242 sg_unmark_end(&sg[index]);
243 sg_set_page(&sg[index++], pdata[i++], s, data_off);
251 sg_mark_end(&sg[index - 1]);
252 return index - start;
256 * p9_virtio_request - issue a request
257 * @client: client instance issuing the request
258 * @req: request to be issued
263 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
266 int in, out, out_sgs, in_sgs;
268 struct virtio_chan *chan = client->trans;
269 struct scatterlist *sgs[2];
271 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
273 req->status = REQ_STATUS_SENT;
275 spin_lock_irqsave(&chan->lock, flags);
277 out_sgs = in_sgs = 0;
278 /* Handle out VirtIO ring buffers */
279 out = pack_sg_list(chan->sg, 0,
280 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
282 sgs[out_sgs++] = chan->sg;
284 in = pack_sg_list(chan->sg, out,
285 VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity);
287 sgs[out_sgs + in_sgs++] = chan->sg + out;
289 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
292 if (err == -ENOSPC) {
293 chan->ring_bufs_avail = 0;
294 spin_unlock_irqrestore(&chan->lock, flags);
295 err = wait_event_interruptible(*chan->vc_wq,
296 chan->ring_bufs_avail);
297 if (err == -ERESTARTSYS)
300 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
303 spin_unlock_irqrestore(&chan->lock, flags);
304 p9_debug(P9_DEBUG_TRANS,
305 "virtio rpc add_sgs returned failure\n");
309 virtqueue_kick(chan->vq);
310 spin_unlock_irqrestore(&chan->lock, flags);
312 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
316 static int p9_get_mapped_pages(struct virtio_chan *chan,
317 struct page **pages, char *data,
318 int nr_pages, int write, int kern_buf)
323 * We allow only p9_max_pages pinned. We wait for the
324 * Other zc request to finish here
326 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
327 err = wait_event_interruptible(vp_wq,
328 (atomic_read(&vp_pinned) < chan->p9_max_pages));
329 if (err == -ERESTARTSYS)
332 err = p9_payload_gup(data, &nr_pages, pages, write);
335 atomic_add(nr_pages, &vp_pinned);
337 /* kernel buffer, no need to pin pages */
339 int count = nr_pages;
341 s = rest_of_page(data);
342 if (is_vmalloc_addr(data))
343 pages[index++] = vmalloc_to_page(data);
345 pages[index++] = kmap_to_page(data);
355 * p9_virtio_zc_request - issue a zero copy request
356 * @client: client instance issuing the request
357 * @req: request to be issued
358 * @uidata: user bffer that should be ued for zero copy read
359 * @uodata: user buffer that shoud be user for zero copy write
360 * @inlen: read buffer size
361 * @olen: write buffer size
362 * @hdrlen: reader header size, This is the size of response protocol data
366 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
367 char *uidata, char *uodata, int inlen,
368 int outlen, int in_hdr_len, int kern_buf)
370 int in, out, err, out_sgs, in_sgs;
372 int in_nr_pages = 0, out_nr_pages = 0;
373 struct page **in_pages = NULL, **out_pages = NULL;
374 struct virtio_chan *chan = client->trans;
375 struct scatterlist *sgs[4];
377 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
380 out_nr_pages = p9_nr_pages(uodata, outlen);
381 out_pages = kmalloc(sizeof(struct page *) * out_nr_pages,
387 out_nr_pages = p9_get_mapped_pages(chan, out_pages, uodata,
388 out_nr_pages, 0, kern_buf);
389 if (out_nr_pages < 0) {
397 in_nr_pages = p9_nr_pages(uidata, inlen);
398 in_pages = kmalloc(sizeof(struct page *) * in_nr_pages,
404 in_nr_pages = p9_get_mapped_pages(chan, in_pages, uidata,
405 in_nr_pages, 1, kern_buf);
406 if (in_nr_pages < 0) {
413 req->status = REQ_STATUS_SENT;
415 spin_lock_irqsave(&chan->lock, flags);
417 out_sgs = in_sgs = 0;
420 out = pack_sg_list(chan->sg, 0,
421 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
424 sgs[out_sgs++] = chan->sg;
427 sgs[out_sgs++] = chan->sg + out;
428 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
429 out_pages, out_nr_pages, uodata, outlen);
433 * Take care of in data
434 * For example TREAD have 11.
435 * 11 is the read/write header = PDU Header(7) + IO Size (4).
436 * Arrange in such a way that server places header in the
437 * alloced memory and payload onto the user buffer.
439 in = pack_sg_list(chan->sg, out,
440 VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len);
442 sgs[out_sgs + in_sgs++] = chan->sg + out;
445 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
446 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
447 in_pages, in_nr_pages, uidata, inlen);
450 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
451 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
454 if (err == -ENOSPC) {
455 chan->ring_bufs_avail = 0;
456 spin_unlock_irqrestore(&chan->lock, flags);
457 err = wait_event_interruptible(*chan->vc_wq,
458 chan->ring_bufs_avail);
459 if (err == -ERESTARTSYS)
462 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
463 goto req_retry_pinned;
465 spin_unlock_irqrestore(&chan->lock, flags);
466 p9_debug(P9_DEBUG_TRANS,
467 "virtio rpc add_sgs returned failure\n");
472 virtqueue_kick(chan->vq);
473 spin_unlock_irqrestore(&chan->lock, flags);
474 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
475 err = wait_event_interruptible(*req->wq,
476 req->status >= REQ_STATUS_RCVD);
478 * Non kernel buffers are pinned, unpin them
483 p9_release_pages(in_pages, in_nr_pages);
484 atomic_sub(in_nr_pages, &vp_pinned);
487 p9_release_pages(out_pages, out_nr_pages);
488 atomic_sub(out_nr_pages, &vp_pinned);
490 /* wakeup anybody waiting for slots to pin pages */
498 static ssize_t p9_mount_tag_show(struct device *dev,
499 struct device_attribute *attr, char *buf)
501 struct virtio_chan *chan;
502 struct virtio_device *vdev;
504 vdev = dev_to_virtio(dev);
507 return snprintf(buf, chan->tag_len + 1, "%s", chan->tag);
510 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
513 * p9_virtio_probe - probe for existence of 9P virtio channels
514 * @vdev: virtio device to probe
516 * This probes for existing virtio channels.
520 static int p9_virtio_probe(struct virtio_device *vdev)
525 struct virtio_chan *chan;
527 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
529 pr_err("Failed to allocate virtio 9P channel\n");
536 /* We expect one virtqueue, for requests. */
537 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
538 if (IS_ERR(chan->vq)) {
539 err = PTR_ERR(chan->vq);
542 chan->vq->vdev->priv = chan;
543 spin_lock_init(&chan->lock);
545 sg_init_table(chan->sg, VIRTQUEUE_NUM);
548 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
549 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
554 tag = kmalloc(tag_len, GFP_KERNEL);
560 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
563 chan->tag_len = tag_len;
564 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
568 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
573 init_waitqueue_head(chan->vc_wq);
574 chan->ring_bufs_avail = 1;
575 /* Ceiling limit to avoid denial of service attacks */
576 chan->p9_max_pages = nr_free_buffer_pages()/4;
578 mutex_lock(&virtio_9p_lock);
579 list_add_tail(&chan->chan_list, &virtio_chan_list);
580 mutex_unlock(&virtio_9p_lock);
582 /* Let udev rules use the new mount_tag attribute. */
583 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
590 vdev->config->del_vqs(vdev);
598 * p9_virtio_create - allocate a new virtio channel
599 * @client: client instance invoking this transport
600 * @devname: string identifying the channel to connect to (unused)
601 * @args: args passed from sys_mount() for per-transport options (unused)
603 * This sets up a transport channel for 9p communication. Right now
604 * we only match the first available channel, but eventually we couldlook up
605 * alternate channels by matching devname versus a virtio_config entry.
606 * We use a simple reference count mechanism to ensure that only a single
607 * mount has a channel open at a time.
612 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
614 struct virtio_chan *chan;
618 mutex_lock(&virtio_9p_lock);
619 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
620 if (!strncmp(devname, chan->tag, chan->tag_len) &&
621 strlen(devname) == chan->tag_len) {
630 mutex_unlock(&virtio_9p_lock);
633 pr_err("no channels available\n");
637 client->trans = (void *)chan;
638 client->status = Connected;
639 chan->client = client;
645 * p9_virtio_remove - clean up resources associated with a virtio device
646 * @vdev: virtio device to remove
650 static void p9_virtio_remove(struct virtio_device *vdev)
652 struct virtio_chan *chan = vdev->priv;
655 p9_virtio_close(chan->client);
656 vdev->config->del_vqs(vdev);
658 mutex_lock(&virtio_9p_lock);
659 list_del(&chan->chan_list);
660 mutex_unlock(&virtio_9p_lock);
661 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
662 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
669 static struct virtio_device_id id_table[] = {
670 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
674 static unsigned int features[] = {
678 /* The standard "struct lguest_driver": */
679 static struct virtio_driver p9_virtio_drv = {
680 .feature_table = features,
681 .feature_table_size = ARRAY_SIZE(features),
682 .driver.name = KBUILD_MODNAME,
683 .driver.owner = THIS_MODULE,
684 .id_table = id_table,
685 .probe = p9_virtio_probe,
686 .remove = p9_virtio_remove,
689 static struct p9_trans_module p9_virtio_trans = {
691 .create = p9_virtio_create,
692 .close = p9_virtio_close,
693 .request = p9_virtio_request,
694 .zc_request = p9_virtio_zc_request,
695 .cancel = p9_virtio_cancel,
697 * We leave one entry for input and one entry for response
698 * headers. We also skip one more entry to accomodate, address
699 * that are not at page boundary, that can result in an extra
702 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
704 .owner = THIS_MODULE,
707 /* The standard init function */
708 static int __init p9_virtio_init(void)
710 INIT_LIST_HEAD(&virtio_chan_list);
712 v9fs_register_trans(&p9_virtio_trans);
713 return register_virtio_driver(&p9_virtio_drv);
716 static void __exit p9_virtio_cleanup(void)
718 unregister_virtio_driver(&p9_virtio_drv);
719 v9fs_unregister_trans(&p9_virtio_trans);
722 module_init(p9_virtio_init);
723 module_exit(p9_virtio_cleanup);
725 MODULE_DEVICE_TABLE(virtio, id_table);
726 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
727 MODULE_DESCRIPTION("Virtio 9p Transport");
728 MODULE_LICENSE("GPL");