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 * @ring_bufs_avail: flag to indicate there is some available in the ring buf
54 * @vc_wq: wait queue for waiting for thing to be added to ring buf
55 * @p9_max_pages: maximum number of pinned pages
56 * @sg: scatter gather list which is used to pack a request (protected?)
57 * @chan_list: linked list of channels
59 * We keep all per-channel information in a structure.
60 * This structure is allocated within the devices dev->mem space.
61 * A pointer to the structure will get put in the transport private.
70 struct p9_client *client;
71 struct virtio_device *vdev;
74 wait_queue_head_t *vc_wq;
75 /* This is global limit. Since we don't have a global structure,
76 * will be placing it in each channel.
78 unsigned long p9_max_pages;
79 /* Scatterlist: can be too big for stack. */
80 struct scatterlist sg[VIRTQUEUE_NUM];
82 * @tag: name to identify a mount null terminated
86 struct list_head chan_list;
89 static struct list_head virtio_chan_list;
91 /* How many bytes left in this page. */
92 static unsigned int rest_of_page(void *data)
94 return PAGE_SIZE - offset_in_page(data);
98 * p9_virtio_close - reclaim resources of a channel
99 * @client: client instance
101 * This reclaims a channel by freeing its resources and
102 * resetting its inuse flag.
106 static void p9_virtio_close(struct p9_client *client)
108 struct virtio_chan *chan = client->trans;
110 mutex_lock(&virtio_9p_lock);
113 mutex_unlock(&virtio_9p_lock);
117 * req_done - callback which signals activity from the server
118 * @vq: virtio queue activity was received on
120 * This notifies us that the server has triggered some activity
121 * on the virtio channel - most likely a response to request we
122 * sent. Figure out which requests now have responses and wake up
125 * Bugs: could do with some additional sanity checking, but appears to work.
129 static void req_done(struct virtqueue *vq)
131 struct virtio_chan *chan = vq->vdev->priv;
133 struct p9_req_t *req;
134 bool need_wakeup = false;
137 p9_debug(P9_DEBUG_TRANS, ": request done\n");
139 spin_lock_irqsave(&chan->lock, flags);
140 while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) {
141 if (!chan->ring_bufs_avail) {
142 chan->ring_bufs_avail = 1;
148 p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
151 spin_unlock_irqrestore(&chan->lock, flags);
152 /* Wakeup if anyone waiting for VirtIO ring space. */
154 wake_up(chan->vc_wq);
158 * pack_sg_list - pack a scatter gather list from a linear buffer
159 * @sg: scatter/gather list to pack into
160 * @start: which segment of the sg_list to start at
161 * @limit: maximum segment to pack data to
162 * @data: data to pack into scatter/gather list
163 * @count: amount of data to pack into the scatter/gather list
165 * sg_lists have multiple segments of various sizes. This will pack
166 * arbitrary data into an existing scatter gather list, segmenting the
167 * data as necessary within constraints.
171 static int pack_sg_list(struct scatterlist *sg, int start,
172 int limit, char *data, int count)
178 s = rest_of_page(data);
181 BUG_ON(index >= limit);
182 /* Make sure we don't terminate early. */
183 sg_unmark_end(&sg[index]);
184 sg_set_buf(&sg[index++], data, s);
189 sg_mark_end(&sg[index - 1]);
193 /* We don't currently allow canceling of virtio requests */
194 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
199 /* Reply won't come, so drop req ref */
200 static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req)
202 p9_req_put(client, req);
207 * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
208 * this takes a list of pages.
209 * @sg: scatter/gather list to pack into
210 * @start: which segment of the sg_list to start at
211 * @limit: maximum number of pages in sg list.
212 * @pdata: a list of pages to add into sg.
213 * @nr_pages: number of pages to pack into the scatter/gather list
214 * @offs: amount of data in the beginning of first page _not_ to pack
215 * @count: amount of data to pack into the scatter/gather list
218 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
219 struct page **pdata, int nr_pages, size_t offs, int count)
225 BUG_ON(nr_pages > (limit - start));
227 * if the first page doesn't start at
228 * page boundary find the offset
231 s = PAGE_SIZE - data_off;
234 BUG_ON(index >= limit);
235 /* Make sure we don't terminate early. */
236 sg_unmark_end(&sg[index]);
237 sg_set_page(&sg[index++], pdata[i++], s, data_off);
244 sg_mark_end(&sg[index - 1]);
245 return index - start;
249 * p9_virtio_request - issue a request
250 * @client: client instance issuing the request
251 * @req: request to be issued
256 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
259 int in, out, out_sgs, in_sgs;
261 struct virtio_chan *chan = client->trans;
262 struct scatterlist *sgs[2];
264 p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
266 req->status = REQ_STATUS_SENT;
268 spin_lock_irqsave(&chan->lock, flags);
270 out_sgs = in_sgs = 0;
271 /* Handle out VirtIO ring buffers */
272 out = pack_sg_list(chan->sg, 0,
273 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
275 sgs[out_sgs++] = chan->sg;
277 in = pack_sg_list(chan->sg, out,
278 VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity);
280 sgs[out_sgs + in_sgs++] = chan->sg + out;
282 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
285 if (err == -ENOSPC) {
286 chan->ring_bufs_avail = 0;
287 spin_unlock_irqrestore(&chan->lock, flags);
288 err = wait_event_killable(*chan->vc_wq,
289 chan->ring_bufs_avail);
290 if (err == -ERESTARTSYS)
293 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
296 spin_unlock_irqrestore(&chan->lock, flags);
297 p9_debug(P9_DEBUG_TRANS,
298 "virtio rpc add_sgs returned failure\n");
302 virtqueue_kick(chan->vq);
303 spin_unlock_irqrestore(&chan->lock, flags);
305 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
309 static int p9_get_mapped_pages(struct virtio_chan *chan,
310 struct page ***pages,
311 struct iov_iter *data,
319 if (!iov_iter_count(data))
322 if (!iov_iter_is_kvec(data)) {
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 n = iov_iter_get_pages_alloc(data, pages, count, offs);
338 nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
339 atomic_add(nr_pages, &vp_pinned);
342 /* kernel buffer, no need to pin pages */
347 /* we'd already checked that it's non-empty */
349 len = iov_iter_single_seg_count(data);
351 p = data->kvec->iov_base + data->iov_offset;
354 iov_iter_advance(data, 0);
359 nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
360 (unsigned long)p / PAGE_SIZE;
362 *pages = kmalloc_array(nr_pages, sizeof(struct page *),
368 p -= (*offs = offset_in_page(p));
369 for (index = 0; index < nr_pages; index++) {
370 if (is_vmalloc_addr(p))
371 (*pages)[index] = vmalloc_to_page(p);
373 (*pages)[index] = kmap_to_page(p);
380 static void handle_rerror(struct p9_req_t *req, int in_hdr_len,
381 size_t offs, struct page **pages)
384 void *to = req->rc.sdata + in_hdr_len;
386 // Fits entirely into the static data? Nothing to do.
387 if (req->rc.size < in_hdr_len)
390 // Really long error message? Tough, truncate the reply. Might get
391 // rejected (we can't be arsed to adjust the size encoded in header,
392 // or string size for that matter), but it wouldn't be anything valid
394 if (unlikely(req->rc.size > P9_ZC_HDR_SZ))
395 req->rc.size = P9_ZC_HDR_SZ;
397 // data won't span more than two pages
398 size = req->rc.size - in_hdr_len;
399 n = PAGE_SIZE - offs;
401 memcpy_from_page(to, *pages++, offs, n);
406 memcpy_from_page(to, *pages, offs, size);
410 * p9_virtio_zc_request - issue a zero copy request
411 * @client: client instance issuing the request
412 * @req: request to be issued
413 * @uidata: user buffer that should be used for zero copy read
414 * @uodata: user buffer that should be used for zero copy write
415 * @inlen: read buffer size
416 * @outlen: write buffer size
417 * @in_hdr_len: reader header size, This is the size of response protocol data
421 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
422 struct iov_iter *uidata, struct iov_iter *uodata,
423 int inlen, int outlen, int in_hdr_len)
425 int in, out, err, out_sgs, in_sgs;
427 int in_nr_pages = 0, out_nr_pages = 0;
428 struct page **in_pages = NULL, **out_pages = NULL;
429 struct virtio_chan *chan = client->trans;
430 struct scatterlist *sgs[4];
435 p9_debug(P9_DEBUG_TRANS, "virtio request\n");
439 int n = p9_get_mapped_pages(chan, &out_pages, uodata,
440 outlen, &offs, &need_drop);
445 out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
447 __le32 v = cpu_to_le32(n);
448 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
451 /* The size field of the message must include the length of the
452 * header and the length of the data. We didn't actually know
453 * the length of the data until this point so add it in now.
455 sz = cpu_to_le32(req->tc.size + outlen);
456 memcpy(&req->tc.sdata[0], &sz, sizeof(sz));
458 int n = p9_get_mapped_pages(chan, &in_pages, uidata,
459 inlen, &offs, &need_drop);
464 in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
466 __le32 v = cpu_to_le32(n);
467 memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
471 req->status = REQ_STATUS_SENT;
473 spin_lock_irqsave(&chan->lock, flags);
475 out_sgs = in_sgs = 0;
478 out = pack_sg_list(chan->sg, 0,
479 VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
482 sgs[out_sgs++] = chan->sg;
485 sgs[out_sgs++] = chan->sg + out;
486 out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
487 out_pages, out_nr_pages, offs, outlen);
491 * Take care of in data
492 * For example TREAD have 11.
493 * 11 is the read/write header = PDU Header(7) + IO Size (4).
494 * Arrange in such a way that server places header in the
495 * allocated memory and payload onto the user buffer.
497 in = pack_sg_list(chan->sg, out,
498 VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len);
500 sgs[out_sgs + in_sgs++] = chan->sg + out;
503 sgs[out_sgs + in_sgs++] = chan->sg + out + in;
504 in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
505 in_pages, in_nr_pages, offs, inlen);
508 BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
509 err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
512 if (err == -ENOSPC) {
513 chan->ring_bufs_avail = 0;
514 spin_unlock_irqrestore(&chan->lock, flags);
515 err = wait_event_killable(*chan->vc_wq,
516 chan->ring_bufs_avail);
517 if (err == -ERESTARTSYS)
520 p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
521 goto req_retry_pinned;
523 spin_unlock_irqrestore(&chan->lock, flags);
524 p9_debug(P9_DEBUG_TRANS,
525 "virtio rpc add_sgs returned failure\n");
530 virtqueue_kick(chan->vq);
531 spin_unlock_irqrestore(&chan->lock, flags);
533 p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
534 err = wait_event_killable(req->wq, req->status >= REQ_STATUS_RCVD);
535 // RERROR needs reply (== error string) in static data
536 if (req->status == REQ_STATUS_RCVD &&
537 unlikely(req->rc.sdata[4] == P9_RERROR))
538 handle_rerror(req, in_hdr_len, offs, in_pages);
541 * Non kernel buffers are pinned, unpin them
546 p9_release_pages(in_pages, in_nr_pages);
547 atomic_sub(in_nr_pages, &vp_pinned);
550 p9_release_pages(out_pages, out_nr_pages);
551 atomic_sub(out_nr_pages, &vp_pinned);
553 /* wakeup anybody waiting for slots to pin pages */
559 /* reply won't come */
560 p9_req_put(client, req);
565 static ssize_t p9_mount_tag_show(struct device *dev,
566 struct device_attribute *attr, char *buf)
568 struct virtio_chan *chan;
569 struct virtio_device *vdev;
572 vdev = dev_to_virtio(dev);
574 tag_len = strlen(chan->tag);
576 memcpy(buf, chan->tag, tag_len + 1);
581 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
584 * p9_virtio_probe - probe for existence of 9P virtio channels
585 * @vdev: virtio device to probe
587 * This probes for existing virtio channels.
591 static int p9_virtio_probe(struct virtio_device *vdev)
596 struct virtio_chan *chan;
598 if (!vdev->config->get) {
599 dev_err(&vdev->dev, "%s failure: config access disabled\n",
604 chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
606 pr_err("Failed to allocate virtio 9P channel\n");
613 /* We expect one virtqueue, for requests. */
614 chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
615 if (IS_ERR(chan->vq)) {
616 err = PTR_ERR(chan->vq);
619 chan->vq->vdev->priv = chan;
620 spin_lock_init(&chan->lock);
622 sg_init_table(chan->sg, VIRTQUEUE_NUM);
625 if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
626 virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
631 tag = kzalloc(tag_len + 1, GFP_KERNEL);
637 virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
640 err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
644 chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
647 goto out_remove_file;
649 init_waitqueue_head(chan->vc_wq);
650 chan->ring_bufs_avail = 1;
651 /* Ceiling limit to avoid denial of service attacks */
652 chan->p9_max_pages = nr_free_buffer_pages()/4;
654 virtio_device_ready(vdev);
656 mutex_lock(&virtio_9p_lock);
657 list_add_tail(&chan->chan_list, &virtio_chan_list);
658 mutex_unlock(&virtio_9p_lock);
660 /* Let udev rules use the new mount_tag attribute. */
661 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
666 sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr);
670 vdev->config->del_vqs(vdev);
679 * p9_virtio_create - allocate a new virtio channel
680 * @client: client instance invoking this transport
681 * @devname: string identifying the channel to connect to (unused)
682 * @args: args passed from sys_mount() for per-transport options (unused)
684 * This sets up a transport channel for 9p communication. Right now
685 * we only match the first available channel, but eventually we could look up
686 * alternate channels by matching devname versus a virtio_config entry.
687 * We use a simple reference count mechanism to ensure that only a single
688 * mount has a channel open at a time.
693 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
695 struct virtio_chan *chan;
702 mutex_lock(&virtio_9p_lock);
703 list_for_each_entry(chan, &virtio_chan_list, chan_list) {
704 if (!strcmp(devname, chan->tag)) {
713 mutex_unlock(&virtio_9p_lock);
716 pr_err("no channels available for device %s\n", devname);
720 client->trans = (void *)chan;
721 client->status = Connected;
722 chan->client = client;
728 * p9_virtio_remove - clean up resources associated with a virtio device
729 * @vdev: virtio device to remove
733 static void p9_virtio_remove(struct virtio_device *vdev)
735 struct virtio_chan *chan = vdev->priv;
736 unsigned long warning_time;
738 mutex_lock(&virtio_9p_lock);
740 /* Remove self from list so we don't get new users. */
741 list_del(&chan->chan_list);
742 warning_time = jiffies;
744 /* Wait for existing users to close. */
745 while (chan->inuse) {
746 mutex_unlock(&virtio_9p_lock);
748 if (time_after(jiffies, warning_time + 10 * HZ)) {
749 dev_emerg(&vdev->dev,
750 "p9_virtio_remove: waiting for device in use.\n");
751 warning_time = jiffies;
753 mutex_lock(&virtio_9p_lock);
756 mutex_unlock(&virtio_9p_lock);
758 virtio_reset_device(vdev);
759 vdev->config->del_vqs(vdev);
761 sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
762 kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
769 static struct virtio_device_id id_table[] = {
770 { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
774 static unsigned int features[] = {
778 /* The standard "struct lguest_driver": */
779 static struct virtio_driver p9_virtio_drv = {
780 .feature_table = features,
781 .feature_table_size = ARRAY_SIZE(features),
782 .driver.name = KBUILD_MODNAME,
783 .driver.owner = THIS_MODULE,
784 .id_table = id_table,
785 .probe = p9_virtio_probe,
786 .remove = p9_virtio_remove,
789 static struct p9_trans_module p9_virtio_trans = {
791 .create = p9_virtio_create,
792 .close = p9_virtio_close,
793 .request = p9_virtio_request,
794 .zc_request = p9_virtio_zc_request,
795 .cancel = p9_virtio_cancel,
796 .cancelled = p9_virtio_cancelled,
798 * We leave one entry for input and one entry for response
799 * headers. We also skip one more entry to accommodate, address
800 * that are not at page boundary, that can result in an extra
803 .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
805 .owner = THIS_MODULE,
808 /* The standard init function */
809 static int __init p9_virtio_init(void)
813 INIT_LIST_HEAD(&virtio_chan_list);
815 v9fs_register_trans(&p9_virtio_trans);
816 rc = register_virtio_driver(&p9_virtio_drv);
818 v9fs_unregister_trans(&p9_virtio_trans);
823 static void __exit p9_virtio_cleanup(void)
825 unregister_virtio_driver(&p9_virtio_drv);
826 v9fs_unregister_trans(&p9_virtio_trans);
829 module_init(p9_virtio_init);
830 module_exit(p9_virtio_cleanup);
831 MODULE_ALIAS_9P("virtio");
833 MODULE_DEVICE_TABLE(virtio, id_table);
834 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
835 MODULE_DESCRIPTION("Virtio 9p Transport");
836 MODULE_LICENSE("GPL");