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);
168 p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
173 * pack_sg_list - pack a scatter gather list from a linear buffer
174 * @sg: scatter/gather list to pack into
175 * @start: which segment of the sg_list to start at
176 * @limit: maximum segment to pack data to
177 * @data: data to pack into scatter/gather list
178 * @count: amount of data to pack into the scatter/gather list
180 * sg_lists have multiple segments of various sizes. This will pack
181 * arbitrary data into an existing scatter gather list, segmenting the
182 * data as necessary within constraints.
186 static int pack_sg_list(struct scatterlist *sg, int start,
187 int limit, char *data, int count)
193 s = rest_of_page(data);
196 BUG_ON(index >= limit);
197 /* Make sure we don't terminate early. */
198 sg_unmark_end(&sg[index]);
199 sg_set_buf(&sg[index++], data, s);
204 sg_mark_end(&sg[index - 1]);
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: a list of pages to add into sg.
220 * @nr_pages: number of pages to pack into the scatter/gather list
221 * @data: data to pack into scatter/gather list
222 * @count: amount of data to pack into the scatter/gather list
225 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
226 struct page **pdata, int nr_pages, char *data, int count)
232 BUG_ON(nr_pages > (limit - start));
234 * if the first page doesn't start at
235 * page boundary find the offset
237 data_off = offset_in_page(data);
239 s = rest_of_page(data);
242 BUG_ON(index >= limit);
243 /* Make sure we don't terminate early. */
244 sg_unmark_end(&sg[index]);
245 sg_set_page(&sg[index++], pdata[i++], s, data_off);
253 sg_mark_end(&sg[index - 1]);
254 return index - start;
258 * p9_virtio_request - issue a request
259 * @client: client instance issuing the request
260 * @req: request to be issued
265 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
268 int in, out, out_sgs, in_sgs;
270 struct virtio_chan *chan = client->trans;
271 struct scatterlist *sgs[2];
273 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
275 req->status = REQ_STATUS_SENT;
277 spin_lock_irqsave(&chan->lock, flags);
279 out_sgs = in_sgs = 0;
280 /* Handle out VirtIO ring buffers */
281 out = pack_sg_list(chan->sg, 0,
282 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
284 sgs[out_sgs++] = chan->sg;
286 in = pack_sg_list(chan->sg, out,
287 VIRTQUEUE_NUM, req->rc->sdata, req->rc->capacity);
289 sgs[out_sgs + in_sgs++] = chan->sg + out;
291 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
294 if (err == -ENOSPC) {
295 chan->ring_bufs_avail = 0;
296 spin_unlock_irqrestore(&chan->lock, flags);
297 err = wait_event_killable(*chan->vc_wq,
298 chan->ring_bufs_avail);
299 if (err == -ERESTARTSYS)
302 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
305 spin_unlock_irqrestore(&chan->lock, flags);
306 p9_debug(P9_DEBUG_TRANS,
307 "virtio rpc add_sgs returned failure\n");
311 virtqueue_kick(chan->vq);
312 spin_unlock_irqrestore(&chan->lock, flags);
314 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
318 static int p9_get_mapped_pages(struct virtio_chan *chan,
319 struct page **pages, char *data,
320 int nr_pages, int write, int kern_buf)
325 * We allow only p9_max_pages pinned. We wait for the
326 * Other zc request to finish here
328 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
329 err = wait_event_killable(vp_wq,
330 (atomic_read(&vp_pinned) < chan->p9_max_pages));
331 if (err == -ERESTARTSYS)
334 err = p9_payload_gup(data, &nr_pages, pages, write);
337 atomic_add(nr_pages, &vp_pinned);
339 /* kernel buffer, no need to pin pages */
341 int count = nr_pages;
343 s = rest_of_page(data);
344 if (is_vmalloc_addr(data))
345 pages[index++] = vmalloc_to_page(data);
347 pages[index++] = kmap_to_page(data);
357 * p9_virtio_zc_request - issue a zero copy request
358 * @client: client instance issuing the request
359 * @req: request to be issued
360 * @uidata: user bffer that should be ued for zero copy read
361 * @uodata: user buffer that shoud be user for zero copy write
362 * @inlen: read buffer size
363 * @olen: write buffer size
364 * @hdrlen: reader header size, This is the size of response protocol data
368 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
369 char *uidata, char *uodata, int inlen,
370 int outlen, int in_hdr_len, int kern_buf)
372 int in, out, err, out_sgs, in_sgs;
374 int in_nr_pages = 0, out_nr_pages = 0;
375 struct page **in_pages = NULL, **out_pages = NULL;
376 struct virtio_chan *chan = client->trans;
377 struct scatterlist *sgs[4];
379 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
383 out_nr_pages = p9_nr_pages(uodata, outlen);
384 out_pages = kmalloc(sizeof(struct page *) * out_nr_pages,
390 out_nr_pages = p9_get_mapped_pages(chan, out_pages, uodata,
391 out_nr_pages, 0, kern_buf);
392 if (out_nr_pages < 0) {
398 /* The size field of the message must include the length of the
399 * header and the length of the data. We didn't actually know
400 * the length of the data until this point so add it in now.
402 sz = cpu_to_le32(req->tc->size + outlen);
403 memcpy(&req->tc->sdata[0], &sz, sizeof(sz));
406 in_nr_pages = p9_nr_pages(uidata, inlen);
407 in_pages = kmalloc(sizeof(struct page *) * in_nr_pages,
413 in_nr_pages = p9_get_mapped_pages(chan, in_pages, uidata,
414 in_nr_pages, 1, kern_buf);
415 if (in_nr_pages < 0) {
422 req->status = REQ_STATUS_SENT;
424 spin_lock_irqsave(&chan->lock, flags);
426 out_sgs = in_sgs = 0;
429 out = pack_sg_list(chan->sg, 0,
430 VIRTQUEUE_NUM, req->tc->sdata, req->tc->size);
433 sgs[out_sgs++] = chan->sg;
436 sgs[out_sgs++] = chan->sg + out;
437 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
438 out_pages, out_nr_pages, uodata, outlen);
442 * Take care of in data
443 * For example TREAD have 11.
444 * 11 is the read/write header = PDU Header(7) + IO Size (4).
445 * Arrange in such a way that server places header in the
446 * alloced memory and payload onto the user buffer.
448 in = pack_sg_list(chan->sg, out,
449 VIRTQUEUE_NUM, req->rc->sdata, in_hdr_len);
451 sgs[out_sgs + in_sgs++] = chan->sg + out;
454 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
455 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
456 in_pages, in_nr_pages, uidata, inlen);
459 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
460 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req->tc,
463 if (err == -ENOSPC) {
464 chan->ring_bufs_avail = 0;
465 spin_unlock_irqrestore(&chan->lock, flags);
466 err = wait_event_killable(*chan->vc_wq,
467 chan->ring_bufs_avail);
468 if (err == -ERESTARTSYS)
471 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
472 goto req_retry_pinned;
474 spin_unlock_irqrestore(&chan->lock, flags);
475 p9_debug(P9_DEBUG_TRANS,
476 "virtio rpc add_sgs returned failure\n");
481 virtqueue_kick(chan->vq);
482 spin_unlock_irqrestore(&chan->lock, flags);
483 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
484 err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD);
486 * Non kernel buffers are pinned, unpin them
491 p9_release_pages(in_pages, in_nr_pages);
492 atomic_sub(in_nr_pages, &vp_pinned);
495 p9_release_pages(out_pages, out_nr_pages);
496 atomic_sub(out_nr_pages, &vp_pinned);
498 /* wakeup anybody waiting for slots to pin pages */
506 static ssize_t p9_mount_tag_show(struct device *dev,
507 struct device_attribute *attr, char *buf)
509 struct virtio_chan *chan;
510 struct virtio_device *vdev;
512 vdev = dev_to_virtio(dev);
515 return snprintf(buf, chan->tag_len + 1, "%s", chan->tag);
518 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
521 * p9_virtio_probe - probe for existence of 9P virtio channels
522 * @vdev: virtio device to probe
524 * This probes for existing virtio channels.
528 static int p9_virtio_probe(struct virtio_device *vdev)
533 struct virtio_chan *chan;
535 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
537 pr_err("Failed to allocate virtio 9P channel\n");
544 /* We expect one virtqueue, for requests. */
545 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
546 if (IS_ERR(chan->vq)) {
547 err = PTR_ERR(chan->vq);
550 chan->vq->vdev->priv = chan;
551 spin_lock_init(&chan->lock);
553 sg_init_table(chan->sg, VIRTQUEUE_NUM);
556 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
557 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
562 tag = kmalloc(tag_len, GFP_KERNEL);
568 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
571 chan->tag_len = tag_len;
572 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
576 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
581 init_waitqueue_head(chan->vc_wq);
582 chan->ring_bufs_avail = 1;
583 /* Ceiling limit to avoid denial of service attacks */
584 chan->p9_max_pages = nr_free_buffer_pages()/4;
586 mutex_lock(&virtio_9p_lock);
587 list_add_tail(&chan->chan_list, &virtio_chan_list);
588 mutex_unlock(&virtio_9p_lock);
590 /* Let udev rules use the new mount_tag attribute. */
591 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
598 vdev->config->del_vqs(vdev);
606 * p9_virtio_create - allocate a new virtio channel
607 * @client: client instance invoking this transport
608 * @devname: string identifying the channel to connect to (unused)
609 * @args: args passed from sys_mount() for per-transport options (unused)
611 * This sets up a transport channel for 9p communication. Right now
612 * we only match the first available channel, but eventually we couldlook up
613 * alternate channels by matching devname versus a virtio_config entry.
614 * We use a simple reference count mechanism to ensure that only a single
615 * mount has a channel open at a time.
620 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
622 struct virtio_chan *chan;
629 mutex_lock(&virtio_9p_lock);
630 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
631 if (!strncmp(devname, chan->tag, chan->tag_len) &&
632 strlen(devname) == chan->tag_len) {
641 mutex_unlock(&virtio_9p_lock);
644 pr_err("no channels available\n");
648 client->trans = (void *)chan;
649 client->status = Connected;
650 chan->client = client;
656 * p9_virtio_remove - clean up resources associated with a virtio device
657 * @vdev: virtio device to remove
661 static void p9_virtio_remove(struct virtio_device *vdev)
663 struct virtio_chan *chan = vdev->priv;
666 p9_virtio_close(chan->client);
667 vdev->config->del_vqs(vdev);
669 mutex_lock(&virtio_9p_lock);
670 list_del(&chan->chan_list);
671 mutex_unlock(&virtio_9p_lock);
672 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
673 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
680 static struct virtio_device_id id_table[] = {
681 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
685 static unsigned int features[] = {
689 /* The standard "struct lguest_driver": */
690 static struct virtio_driver p9_virtio_drv = {
691 .feature_table = features,
692 .feature_table_size = ARRAY_SIZE(features),
693 .driver.name = KBUILD_MODNAME,
694 .driver.owner = THIS_MODULE,
695 .id_table = id_table,
696 .probe = p9_virtio_probe,
697 .remove = p9_virtio_remove,
700 static struct p9_trans_module p9_virtio_trans = {
702 .create = p9_virtio_create,
703 .close = p9_virtio_close,
704 .request = p9_virtio_request,
705 .zc_request = p9_virtio_zc_request,
706 .cancel = p9_virtio_cancel,
708 * We leave one entry for input and one entry for response
709 * headers. We also skip one more entry to accomodate, address
710 * that are not at page boundary, that can result in an extra
713 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
715 .owner = THIS_MODULE,
718 /* The standard init function */
719 static int __init p9_virtio_init(void)
723 INIT_LIST_HEAD(&virtio_chan_list);
725 v9fs_register_trans(&p9_virtio_trans);
726 rc = register_virtio_driver(&p9_virtio_drv);
728 v9fs_unregister_trans(&p9_virtio_trans);
733 static void __exit p9_virtio_cleanup(void)
735 unregister_virtio_driver(&p9_virtio_drv);
736 v9fs_unregister_trans(&p9_virtio_trans);
739 module_init(p9_virtio_init);
740 module_exit(p9_virtio_cleanup);
742 MODULE_DEVICE_TABLE(virtio, id_table);
743 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
744 MODULE_DESCRIPTION("Virtio 9p Transport");
745 MODULE_LICENSE("GPL");