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 * @offs: amount of data in the beginning of first page _not_ to pack
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, size_t offs, int count)
231 BUG_ON(nr_pages > (limit - start));
233 * if the first page doesn't start at
234 * page boundary find the offset
237 s = PAGE_SIZE - data_off;
240 BUG_ON(index >= limit);
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);
250 sg_mark_end(&sg[index - 1]);
251 return index - start;
255 * p9_virtio_request - issue a request
256 * @client: client instance issuing the request
257 * @req: request to be issued
262 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
265 int in, out, out_sgs, in_sgs;
267 struct virtio_chan *chan = client->trans;
268 struct scatterlist *sgs[2];
270 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
272 req->status = REQ_STATUS_SENT;
274 spin_lock_irqsave(&chan->lock, flags);
276 out_sgs = in_sgs = 0;
277 /* Handle out VirtIO ring buffers */
278 out = pack_sg_list(chan->sg, 0,
279 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
281 sgs[out_sgs++] = chan->sg;
283 in = pack_sg_list(chan->sg, out,
284 VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity);
286 sgs[out_sgs + in_sgs++] = chan->sg + out;
288 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
291 if (err == -ENOSPC) {
292 chan->ring_bufs_avail = 0;
293 spin_unlock_irqrestore(&chan->lock, flags);
294 err = wait_event_killable(*chan->vc_wq,
295 chan->ring_bufs_avail);
296 if (err == -ERESTARTSYS)
299 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
302 spin_unlock_irqrestore(&chan->lock, flags);
303 p9_debug(P9_DEBUG_TRANS,
304 "virtio rpc add_sgs returned failure\n");
308 virtqueue_kick(chan->vq);
309 spin_unlock_irqrestore(&chan->lock, flags);
311 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
315 static int p9_get_mapped_pages(struct virtio_chan *chan,
316 struct page ***pages,
317 struct iov_iter *data,
325 if (!iov_iter_count(data))
328 if (!(data->type & ITER_KVEC)) {
331 * We allow only p9_max_pages pinned. We wait for the
332 * Other zc request to finish here
334 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
335 err = wait_event_killable(vp_wq,
336 (atomic_read(&vp_pinned) < chan->p9_max_pages));
337 if (err == -ERESTARTSYS)
340 n = iov_iter_get_pages_alloc(data, pages, count, offs);
344 nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
345 atomic_add(nr_pages, &vp_pinned);
348 /* kernel buffer, no need to pin pages */
353 /* we'd already checked that it's non-empty */
355 len = iov_iter_single_seg_count(data);
357 p = data->kvec->iov_base + data->iov_offset;
360 iov_iter_advance(data, 0);
365 nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
366 (unsigned long)p / PAGE_SIZE;
368 *pages = kmalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
373 p -= (*offs = (unsigned long)p % PAGE_SIZE);
374 for (index = 0; index < nr_pages; index++) {
375 if (is_vmalloc_addr(p))
376 (*pages)[index] = vmalloc_to_page(p);
378 (*pages)[index] = kmap_to_page(p);
386 * p9_virtio_zc_request - issue a zero copy request
387 * @client: client instance issuing the request
388 * @req: request to be issued
389 * @uidata: user bffer that should be ued for zero copy read
390 * @uodata: user buffer that shoud be user for zero copy write
391 * @inlen: read buffer size
392 * @olen: write buffer size
393 * @hdrlen: reader header size, This is the size of response protocol data
397 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
398 struct iov_iter *uidata, struct iov_iter *uodata,
399 int inlen, int outlen, int in_hdr_len)
401 int in, out, err, out_sgs, in_sgs;
403 int in_nr_pages = 0, out_nr_pages = 0;
404 struct page **in_pages = NULL, **out_pages = NULL;
405 struct virtio_chan *chan = client->trans;
406 struct scatterlist *sgs[4];
410 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
414 int n = p9_get_mapped_pages(chan, &out_pages, uodata,
415 outlen, &offs, &need_drop);
418 out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
420 __le32 v = cpu_to_le32(n);
421 memcpy(&req->tc->sdata[req->tc->size - 4], &v, 4);
424 /* The size field of the message must include the length of the
425 * header and the length of the data. We didn't actually know
426 * the length of the data until this point so add it in now.
428 sz = cpu_to_le32(req->tc->size + outlen);
429 memcpy(&req->tc->sdata[0], &sz, sizeof(sz));
431 int n = p9_get_mapped_pages(chan, &in_pages, uidata,
432 inlen, &offs, &need_drop);
435 in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
437 __le32 v = cpu_to_le32(n);
438 memcpy(&req->tc->sdata[req->tc->size - 4], &v, 4);
442 req->status = REQ_STATUS_SENT;
444 spin_lock_irqsave(&chan->lock, flags);
446 out_sgs = in_sgs = 0;
449 out = pack_sg_list(chan->sg, 0,
450 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
453 sgs[out_sgs++] = chan->sg;
456 sgs[out_sgs++] = chan->sg + out;
457 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
458 out_pages, out_nr_pages, offs, outlen);
462 * Take care of in data
463 * For example TREAD have 11.
464 * 11 is the read/write header = PDU Header(7) + IO Size (4).
465 * Arrange in such a way that server places header in the
466 * alloced memory and payload onto the user buffer.
468 in = pack_sg_list(chan->sg, out,
469 VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len);
471 sgs[out_sgs + in_sgs++] = chan->sg + out;
474 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
475 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
476 in_pages, in_nr_pages, offs, inlen);
479 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
480 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
483 if (err == -ENOSPC) {
484 chan->ring_bufs_avail = 0;
485 spin_unlock_irqrestore(&chan->lock, flags);
486 err = wait_event_killable(*chan->vc_wq,
487 chan->ring_bufs_avail);
488 if (err == -ERESTARTSYS)
491 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
492 goto req_retry_pinned;
494 spin_unlock_irqrestore(&chan->lock, flags);
495 p9_debug(P9_DEBUG_TRANS,
496 "virtio rpc add_sgs returned failure\n");
501 virtqueue_kick(chan->vq);
502 spin_unlock_irqrestore(&chan->lock, flags);
503 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
504 err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD);
506 * Non kernel buffers are pinned, unpin them
511 p9_release_pages(in_pages, in_nr_pages);
512 atomic_sub(in_nr_pages, &vp_pinned);
515 p9_release_pages(out_pages, out_nr_pages);
516 atomic_sub(out_nr_pages, &vp_pinned);
518 /* wakeup anybody waiting for slots to pin pages */
526 static ssize_t p9_mount_tag_show(struct device *dev,
527 struct device_attribute *attr, char *buf)
529 struct virtio_chan *chan;
530 struct virtio_device *vdev;
532 vdev = dev_to_virtio(dev);
535 memcpy(buf, chan->tag, chan->tag_len);
536 buf[chan->tag_len] = 0;
538 return chan->tag_len + 1;
541 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
544 * p9_virtio_probe - probe for existence of 9P virtio channels
545 * @vdev: virtio device to probe
547 * This probes for existing virtio channels.
551 static int p9_virtio_probe(struct virtio_device *vdev)
556 struct virtio_chan *chan;
558 if (!vdev->config->get) {
559 dev_err(&vdev->dev, "%s failure: config access disabled\n",
564 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
566 pr_err("Failed to allocate virtio 9P channel\n");
573 /* We expect one virtqueue, for requests. */
574 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
575 if (IS_ERR(chan->vq)) {
576 err = PTR_ERR(chan->vq);
579 chan->vq->vdev->priv = chan;
580 spin_lock_init(&chan->lock);
582 sg_init_table(chan->sg, VIRTQUEUE_NUM);
585 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
586 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
591 tag = kmalloc(tag_len, GFP_KERNEL);
597 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
600 chan->tag_len = tag_len;
601 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
605 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
608 goto out_remove_file;
610 init_waitqueue_head(chan->vc_wq);
611 chan->ring_bufs_avail = 1;
612 /* Ceiling limit to avoid denial of service attacks */
613 chan->p9_max_pages = nr_free_buffer_pages()/4;
615 virtio_device_ready(vdev);
617 mutex_lock(&virtio_9p_lock);
618 list_add_tail(&chan->chan_list, &virtio_chan_list);
619 mutex_unlock(&virtio_9p_lock);
621 /* Let udev rules use the new mount_tag attribute. */
622 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
627 sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr);
631 vdev->config->del_vqs(vdev);
640 * p9_virtio_create - allocate a new virtio channel
641 * @client: client instance invoking this transport
642 * @devname: string identifying the channel to connect to (unused)
643 * @args: args passed from sys_mount() for per-transport options (unused)
645 * This sets up a transport channel for 9p communication. Right now
646 * we only match the first available channel, but eventually we couldlook up
647 * alternate channels by matching devname versus a virtio_config entry.
648 * We use a simple reference count mechanism to ensure that only a single
649 * mount has a channel open at a time.
654 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
656 struct virtio_chan *chan;
663 mutex_lock(&virtio_9p_lock);
664 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
665 if (!strncmp(devname, chan->tag, chan->tag_len) &&
666 strlen(devname) == chan->tag_len) {
675 mutex_unlock(&virtio_9p_lock);
678 pr_err("no channels available\n");
682 client->trans = (void *)chan;
683 client->status = Connected;
684 chan->client = client;
690 * p9_virtio_remove - clean up resources associated with a virtio device
691 * @vdev: virtio device to remove
695 static void p9_virtio_remove(struct virtio_device *vdev)
697 struct virtio_chan *chan = vdev->priv;
698 unsigned long warning_time;
700 mutex_lock(&virtio_9p_lock);
702 /* Remove self from list so we don't get new users. */
703 list_del(&chan->chan_list);
704 warning_time = jiffies;
706 /* Wait for existing users to close. */
707 while (chan->inuse) {
708 mutex_unlock(&virtio_9p_lock);
710 if (time_after(jiffies, warning_time + 10 * HZ)) {
711 dev_emerg(&vdev->dev,
712 "p9_virtio_remove: waiting for device in use.\n");
713 warning_time = jiffies;
715 mutex_lock(&virtio_9p_lock);
718 mutex_unlock(&virtio_9p_lock);
720 vdev->config->reset(vdev);
721 vdev->config->del_vqs(vdev);
723 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
724 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
731 static struct virtio_device_id id_table[] = {
732 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
736 static unsigned int features[] = {
740 /* The standard "struct lguest_driver": */
741 static struct virtio_driver p9_virtio_drv = {
742 .feature_table = features,
743 .feature_table_size = ARRAY_SIZE(features),
744 .driver.name = KBUILD_MODNAME,
745 .driver.owner = THIS_MODULE,
746 .id_table = id_table,
747 .probe = p9_virtio_probe,
748 .remove = p9_virtio_remove,
751 static struct p9_trans_module p9_virtio_trans = {
753 .create = p9_virtio_create,
754 .close = p9_virtio_close,
755 .request = p9_virtio_request,
756 .zc_request = p9_virtio_zc_request,
757 .cancel = p9_virtio_cancel,
759 * We leave one entry for input and one entry for response
760 * headers. We also skip one more entry to accomodate, address
761 * that are not at page boundary, that can result in an extra
764 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
766 .owner = THIS_MODULE,
769 /* The standard init function */
770 static int __init p9_virtio_init(void)
774 INIT_LIST_HEAD(&virtio_chan_list);
776 v9fs_register_trans(&p9_virtio_trans);
777 rc = register_virtio_driver(&p9_virtio_drv);
779 v9fs_unregister_trans(&p9_virtio_trans);
784 static void __exit p9_virtio_cleanup(void)
786 unregister_virtio_driver(&p9_virtio_drv);
787 v9fs_unregister_trans(&p9_virtio_trans);
790 module_init(p9_virtio_init);
791 module_exit(p9_virtio_cleanup);
793 MODULE_DEVICE_TABLE(virtio, id_table);
794 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
795 MODULE_DESCRIPTION("Virtio 9p Transport");
796 MODULE_LICENSE("GPL");