1 // SPDX-License-Identifier: GPL-2.0-only
3 * The Virtio 9p transport driver
5 * This is a block based transport driver based on the lguest block driver
8 * Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation
10 * Based on virtio console driver
11 * Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/module.h>
18 #include <linux/net.h>
19 #include <linux/ipv6.h>
20 #include <linux/errno.h>
21 #include <linux/kernel.h>
23 #include <linux/uaccess.h>
24 #include <linux/inet.h>
25 #include <linux/idr.h>
26 #include <linux/file.h>
27 #include <linux/highmem.h>
28 #include <linux/slab.h>
29 #include <net/9p/9p.h>
30 #include <linux/parser.h>
31 #include <net/9p/client.h>
32 #include <net/9p/transport.h>
33 #include <linux/scatterlist.h>
34 #include <linux/swap.h>
35 #include <linux/virtio.h>
36 #include <linux/virtio_9p.h>
37 #include "trans_common.h"
39 #define VIRTQUEUE_NUM 128
41 /* a single mutex to manage channel initialization and attachment */
42 static DEFINE_MUTEX(virtio_9p_lock);
43 static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
44 static atomic_t vp_pinned = ATOMIC_INIT(0);
47 * struct virtio_chan - per-instance transport information
48 * @inuse: whether the channel is in use
49 * @lock: protects multiple elements within this structure
50 * @client: client instance
51 * @vdev: virtio dev associated with this channel
52 * @vq: virtio queue associated with this channel
53 * @sg: scatter gather list which is used to pack a request (protected?)
55 * We keep all per-channel information in a structure.
56 * This structure is allocated within the devices dev->mem space.
57 * A pointer to the structure will get put in the transport private.
66 struct p9_client *client;
67 struct virtio_device *vdev;
70 wait_queue_head_t *vc_wq;
71 /* This is global limit. Since we don't have a global structure,
72 * will be placing it in each channel.
74 unsigned long p9_max_pages;
75 /* Scatterlist: can be too big for stack. */
76 struct scatterlist sg[VIRTQUEUE_NUM];
78 * tag name to identify a mount null terminated
82 struct list_head chan_list;
85 static struct list_head virtio_chan_list;
87 /* How many bytes left in this page. */
88 static unsigned int rest_of_page(void *data)
90 return PAGE_SIZE - offset_in_page(data);
94 * p9_virtio_close - reclaim resources of a channel
95 * @client: client instance
97 * This reclaims a channel by freeing its resources and
98 * reseting its inuse flag.
102 static void p9_virtio_close(struct p9_client *client)
104 struct virtio_chan *chan = client->trans;
106 mutex_lock(&virtio_9p_lock);
109 mutex_unlock(&virtio_9p_lock);
113 * req_done - callback which signals activity from the server
114 * @vq: virtio queue activity was received on
116 * This notifies us that the server has triggered some activity
117 * on the virtio channel - most likely a response to request we
118 * sent. Figure out which requests now have responses and wake up
121 * Bugs: could do with some additional sanity checking, but appears to work.
125 static void req_done(struct virtqueue *vq)
127 struct virtio_chan *chan = vq->vdev->priv;
129 struct p9_req_t *req;
130 bool need_wakeup = false;
133 p9_debug(P9_DEBUG_TRANS, ": request done\n");
135 spin_lock_irqsave(&chan->lock, flags);
136 while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) {
137 if (!chan->ring_bufs_avail) {
138 chan->ring_bufs_avail = 1;
144 p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
147 spin_unlock_irqrestore(&chan->lock, flags);
148 /* Wakeup if anyone waiting for VirtIO ring space. */
150 wake_up(chan->vc_wq);
154 * pack_sg_list - pack a scatter gather list from a linear buffer
155 * @sg: scatter/gather list to pack into
156 * @start: which segment of the sg_list to start at
157 * @limit: maximum segment to pack data to
158 * @data: data to pack into scatter/gather list
159 * @count: amount of data to pack into the scatter/gather list
161 * sg_lists have multiple segments of various sizes. This will pack
162 * arbitrary data into an existing scatter gather list, segmenting the
163 * data as necessary within constraints.
167 static int pack_sg_list(struct scatterlist *sg, int start,
168 int limit, char *data, int count)
174 s = rest_of_page(data);
177 BUG_ON(index >= limit);
178 /* Make sure we don't terminate early. */
179 sg_unmark_end(&sg[index]);
180 sg_set_buf(&sg[index++], data, s);
185 sg_mark_end(&sg[index - 1]);
189 /* We don't currently allow canceling of virtio requests */
190 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
195 /* Reply won't come, so drop req ref */
196 static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req)
203 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
204 * this takes a list of pages.
205 * @sg: scatter/gather list to pack into
206 * @start: which segment of the sg_list to start at
207 * @pdata: a list of pages to add into sg.
208 * @nr_pages: number of pages to pack into the scatter/gather list
209 * @offs: amount of data in the beginning of first page _not_ to pack
210 * @count: amount of data to pack into the scatter/gather list
213 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
214 struct page **pdata, int nr_pages, size_t offs, int count)
220 BUG_ON(nr_pages > (limit - start));
222 * if the first page doesn't start at
223 * page boundary find the offset
226 s = PAGE_SIZE - data_off;
229 BUG_ON(index >= limit);
230 /* Make sure we don't terminate early. */
231 sg_unmark_end(&sg[index]);
232 sg_set_page(&sg[index++], pdata[i++], s, data_off);
239 sg_mark_end(&sg[index - 1]);
240 return index - start;
244 * p9_virtio_request - issue a request
245 * @client: client instance issuing the request
246 * @req: request to be issued
251 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
254 int in, out, out_sgs, in_sgs;
256 struct virtio_chan *chan = client->trans;
257 struct scatterlist *sgs[2];
259 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
261 req->status = REQ_STATUS_SENT;
263 spin_lock_irqsave(&chan->lock, flags);
265 out_sgs = in_sgs = 0;
266 /* Handle out VirtIO ring buffers */
267 out = pack_sg_list(chan->sg, 0,
268 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
270 sgs[out_sgs++] = chan->sg;
272 in = pack_sg_list(chan->sg, out,
273 VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity);
275 sgs[out_sgs + in_sgs++] = chan->sg + out;
277 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
280 if (err == -ENOSPC) {
281 chan->ring_bufs_avail = 0;
282 spin_unlock_irqrestore(&chan->lock, flags);
283 err = wait_event_killable(*chan->vc_wq,
284 chan->ring_bufs_avail);
285 if (err == -ERESTARTSYS)
288 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
291 spin_unlock_irqrestore(&chan->lock, flags);
292 p9_debug(P9_DEBUG_TRANS,
293 "virtio rpc add_sgs returned failure\n");
297 virtqueue_kick(chan->vq);
298 spin_unlock_irqrestore(&chan->lock, flags);
300 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
304 static int p9_get_mapped_pages(struct virtio_chan *chan,
305 struct page ***pages,
306 struct iov_iter *data,
314 if (!iov_iter_count(data))
317 if (!iov_iter_is_kvec(data)) {
320 * We allow only p9_max_pages pinned. We wait for the
321 * Other zc request to finish here
323 if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
324 err = wait_event_killable(vp_wq,
325 (atomic_read(&vp_pinned) < chan->p9_max_pages));
326 if (err == -ERESTARTSYS)
329 n = iov_iter_get_pages_alloc(data, pages, count, offs);
333 nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
334 atomic_add(nr_pages, &vp_pinned);
337 /* kernel buffer, no need to pin pages */
342 /* we'd already checked that it's non-empty */
344 len = iov_iter_single_seg_count(data);
346 p = data->kvec->iov_base + data->iov_offset;
349 iov_iter_advance(data, 0);
354 nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
355 (unsigned long)p / PAGE_SIZE;
357 *pages = kmalloc_array(nr_pages, sizeof(struct page *),
363 p -= (*offs = offset_in_page(p));
364 for (index = 0; index < nr_pages; index++) {
365 if (is_vmalloc_addr(p))
366 (*pages)[index] = vmalloc_to_page(p);
368 (*pages)[index] = kmap_to_page(p);
376 * p9_virtio_zc_request - issue a zero copy request
377 * @client: client instance issuing the request
378 * @req: request to be issued
379 * @uidata: user buffer that should be used for zero copy read
380 * @uodata: user buffer that should be used for zero copy write
381 * @inlen: read buffer size
382 * @outlen: write buffer size
383 * @in_hdr_len: reader header size, This is the size of response protocol data
387 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
388 struct iov_iter *uidata, struct iov_iter *uodata,
389 int inlen, int outlen, int in_hdr_len)
391 int in, out, err, out_sgs, in_sgs;
393 int in_nr_pages = 0, out_nr_pages = 0;
394 struct page **in_pages = NULL, **out_pages = NULL;
395 struct virtio_chan *chan = client->trans;
396 struct scatterlist *sgs[4];
401 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
405 int n = p9_get_mapped_pages(chan, &out_pages, uodata,
406 outlen, &offs, &need_drop);
411 out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
413 __le32 v = cpu_to_le32(n);
414 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
417 /* The size field of the message must include the length of the
418 * header and the length of the data. We didn't actually know
419 * the length of the data until this point so add it in now.
421 sz = cpu_to_le32(req->tc.size + outlen);
422 memcpy(&req->tc.sdata[0], &sz, sizeof(sz));
424 int n = p9_get_mapped_pages(chan, &in_pages, uidata,
425 inlen, &offs, &need_drop);
430 in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
432 __le32 v = cpu_to_le32(n);
433 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
437 req->status = REQ_STATUS_SENT;
439 spin_lock_irqsave(&chan->lock, flags);
441 out_sgs = in_sgs = 0;
444 out = pack_sg_list(chan->sg, 0,
445 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
448 sgs[out_sgs++] = chan->sg;
451 sgs[out_sgs++] = chan->sg + out;
452 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
453 out_pages, out_nr_pages, offs, outlen);
457 * Take care of in data
458 * For example TREAD have 11.
459 * 11 is the read/write header = PDU Header(7) + IO Size (4).
460 * Arrange in such a way that server places header in the
461 * alloced memory and payload onto the user buffer.
463 in = pack_sg_list(chan->sg, out,
464 VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len);
466 sgs[out_sgs + in_sgs++] = chan->sg + out;
469 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
470 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
471 in_pages, in_nr_pages, offs, inlen);
474 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
475 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
478 if (err == -ENOSPC) {
479 chan->ring_bufs_avail = 0;
480 spin_unlock_irqrestore(&chan->lock, flags);
481 err = wait_event_killable(*chan->vc_wq,
482 chan->ring_bufs_avail);
483 if (err == -ERESTARTSYS)
486 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
487 goto req_retry_pinned;
489 spin_unlock_irqrestore(&chan->lock, flags);
490 p9_debug(P9_DEBUG_TRANS,
491 "virtio rpc add_sgs returned failure\n");
496 virtqueue_kick(chan->vq);
497 spin_unlock_irqrestore(&chan->lock, flags);
499 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
500 err = wait_event_killable(req->wq, req->status >= REQ_STATUS_RCVD);
502 * Non kernel buffers are pinned, unpin them
507 p9_release_pages(in_pages, in_nr_pages);
508 atomic_sub(in_nr_pages, &vp_pinned);
511 p9_release_pages(out_pages, out_nr_pages);
512 atomic_sub(out_nr_pages, &vp_pinned);
514 /* wakeup anybody waiting for slots to pin pages */
520 /* reply won't come */
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;
533 vdev = dev_to_virtio(dev);
535 tag_len = strlen(chan->tag);
537 memcpy(buf, chan->tag, tag_len + 1);
542 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
545 * p9_virtio_probe - probe for existence of 9P virtio channels
546 * @vdev: virtio device to probe
548 * This probes for existing virtio channels.
552 static int p9_virtio_probe(struct virtio_device *vdev)
557 struct virtio_chan *chan;
559 if (!vdev->config->get) {
560 dev_err(&vdev->dev, "%s failure: config access disabled\n",
565 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
567 pr_err("Failed to allocate virtio 9P channel\n");
574 /* We expect one virtqueue, for requests. */
575 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
576 if (IS_ERR(chan->vq)) {
577 err = PTR_ERR(chan->vq);
580 chan->vq->vdev->priv = chan;
581 spin_lock_init(&chan->lock);
583 sg_init_table(chan->sg, VIRTQUEUE_NUM);
586 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
587 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
592 tag = kzalloc(tag_len + 1, GFP_KERNEL);
598 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
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);
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);
629 vdev->config->del_vqs(vdev);
638 * p9_virtio_create - allocate a new virtio channel
639 * @client: client instance invoking this transport
640 * @devname: string identifying the channel to connect to (unused)
641 * @args: args passed from sys_mount() for per-transport options (unused)
643 * This sets up a transport channel for 9p communication. Right now
644 * we only match the first available channel, but eventually we couldlook up
645 * alternate channels by matching devname versus a virtio_config entry.
646 * We use a simple reference count mechanism to ensure that only a single
647 * mount has a channel open at a time.
652 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
654 struct virtio_chan *chan;
661 mutex_lock(&virtio_9p_lock);
662 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
663 if (!strcmp(devname, chan->tag)) {
672 mutex_unlock(&virtio_9p_lock);
675 pr_err("no channels available for device %s\n", devname);
679 client->trans = (void *)chan;
680 client->status = Connected;
681 chan->client = client;
687 * p9_virtio_remove - clean up resources associated with a virtio device
688 * @vdev: virtio device to remove
692 static void p9_virtio_remove(struct virtio_device *vdev)
694 struct virtio_chan *chan = vdev->priv;
695 unsigned long warning_time;
697 mutex_lock(&virtio_9p_lock);
699 /* Remove self from list so we don't get new users. */
700 list_del(&chan->chan_list);
701 warning_time = jiffies;
703 /* Wait for existing users to close. */
704 while (chan->inuse) {
705 mutex_unlock(&virtio_9p_lock);
707 if (time_after(jiffies, warning_time + 10 * HZ)) {
708 dev_emerg(&vdev->dev,
709 "p9_virtio_remove: waiting for device in use.\n");
710 warning_time = jiffies;
712 mutex_lock(&virtio_9p_lock);
715 mutex_unlock(&virtio_9p_lock);
717 vdev->config->reset(vdev);
718 vdev->config->del_vqs(vdev);
720 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
721 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
728 static struct virtio_device_id id_table[] = {
729 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
733 static unsigned int features[] = {
737 /* The standard "struct lguest_driver": */
738 static struct virtio_driver p9_virtio_drv = {
739 .feature_table = features,
740 .feature_table_size = ARRAY_SIZE(features),
741 .driver.name = KBUILD_MODNAME,
742 .driver.owner = THIS_MODULE,
743 .id_table = id_table,
744 .probe = p9_virtio_probe,
745 .remove = p9_virtio_remove,
748 static struct p9_trans_module p9_virtio_trans = {
750 .create = p9_virtio_create,
751 .close = p9_virtio_close,
752 .request = p9_virtio_request,
753 .zc_request = p9_virtio_zc_request,
754 .cancel = p9_virtio_cancel,
755 .cancelled = p9_virtio_cancelled,
757 * We leave one entry for input and one entry for response
758 * headers. We also skip one more entry to accomodate, address
759 * that are not at page boundary, that can result in an extra
762 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
764 .owner = THIS_MODULE,
767 /* The standard init function */
768 static int __init p9_virtio_init(void)
770 INIT_LIST_HEAD(&virtio_chan_list);
772 v9fs_register_trans(&p9_virtio_trans);
773 return register_virtio_driver(&p9_virtio_drv);
776 static void __exit p9_virtio_cleanup(void)
778 unregister_virtio_driver(&p9_virtio_drv);
779 v9fs_unregister_trans(&p9_virtio_trans);
782 module_init(p9_virtio_init);
783 module_exit(p9_virtio_cleanup);
785 MODULE_DEVICE_TABLE(virtio, id_table);
786 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
787 MODULE_DESCRIPTION("Virtio 9p Transport");
788 MODULE_LICENSE("GPL");