xref: /linux/net/9p/trans_virtio.c (revision bf4afc53b77aeaa48b5409da5c8da6bb4eff7f43)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * The Virtio 9p transport driver
4  *
5  * This is a block based transport driver based on the lguest block driver
6  * code.
7  *
8  *  Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation
9  *
10  *  Based on virtio console driver
11  *  Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
12  */
13 
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 
16 #include <linux/in.h>
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>
22 #include <linux/un.h>
23 #include <linux/uaccess.h>
24 #include <linux/inet.h>
25 #include <linux/file.h>
26 #include <linux/highmem.h>
27 #include <linux/slab.h>
28 #include <net/9p/9p.h>
29 #include <linux/fs_context.h>
30 #include <net/9p/client.h>
31 #include <net/9p/transport.h>
32 #include <linux/scatterlist.h>
33 #include <linux/swap.h>
34 #include <linux/virtio.h>
35 #include <linux/virtio_9p.h>
36 #include "trans_common.h"
37 
38 #define VIRTQUEUE_NUM	128
39 
40 /* a single mutex to manage channel initialization and attachment */
41 static DEFINE_MUTEX(virtio_9p_lock);
42 static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
43 static atomic_t vp_pinned = ATOMIC_INIT(0);
44 
45 /**
46  * struct virtio_chan - per-instance transport information
47  * @inuse: whether the channel is in use
48  * @lock: protects multiple elements within this structure
49  * @client: client instance
50  * @vdev: virtio dev associated with this channel
51  * @vq: virtio queue associated with this channel
52  * @ring_bufs_avail: flag to indicate there is some available in the ring buf
53  * @vc_wq: wait queue for waiting for thing to be added to ring buf
54  * @p9_max_pages: maximum number of pinned pages
55  * @sg: scatter gather list which is used to pack a request (protected?)
56  * @chan_list: linked list of channels
57  *
58  * We keep all per-channel information in a structure.
59  * This structure is allocated within the devices dev->mem space.
60  * A pointer to the structure will get put in the transport private.
61  *
62  */
63 
64 struct virtio_chan {
65 	bool inuse;
66 
67 	spinlock_t lock;
68 
69 	struct p9_client *client;
70 	struct virtio_device *vdev;
71 	struct virtqueue *vq;
72 	int ring_bufs_avail;
73 	wait_queue_head_t *vc_wq;
74 	/* This is global limit. Since we don't have a global structure,
75 	 * will be placing it in each channel.
76 	 */
77 	unsigned long p9_max_pages;
78 	/* Scatterlist: can be too big for stack. */
79 	struct scatterlist sg[VIRTQUEUE_NUM];
80 	/**
81 	 * @tag: name to identify a mount null terminated
82 	 */
83 	char *tag;
84 
85 	struct list_head chan_list;
86 };
87 
88 static struct list_head virtio_chan_list;
89 
90 /* How many bytes left in this page. */
rest_of_page(void * data)91 static unsigned int rest_of_page(void *data)
92 {
93 	return PAGE_SIZE - offset_in_page(data);
94 }
95 
96 /**
97  * p9_virtio_close - reclaim resources of a channel
98  * @client: client instance
99  *
100  * This reclaims a channel by freeing its resources and
101  * resetting its inuse flag.
102  *
103  */
104 
p9_virtio_close(struct p9_client * client)105 static void p9_virtio_close(struct p9_client *client)
106 {
107 	struct virtio_chan *chan = client->trans;
108 
109 	mutex_lock(&virtio_9p_lock);
110 	if (chan)
111 		chan->inuse = false;
112 	mutex_unlock(&virtio_9p_lock);
113 }
114 
115 /**
116  * req_done - callback which signals activity from the server
117  * @vq: virtio queue activity was received on
118  *
119  * This notifies us that the server has triggered some activity
120  * on the virtio channel - most likely a response to request we
121  * sent.  Figure out which requests now have responses and wake up
122  * those threads.
123  *
124  * Bugs: could do with some additional sanity checking, but appears to work.
125  *
126  */
127 
req_done(struct virtqueue * vq)128 static void req_done(struct virtqueue *vq)
129 {
130 	struct virtio_chan *chan = vq->vdev->priv;
131 	unsigned int len;
132 	struct p9_req_t *req;
133 	bool need_wakeup = false;
134 	unsigned long flags;
135 
136 	p9_debug(P9_DEBUG_TRANS, ": request done\n");
137 
138 	spin_lock_irqsave(&chan->lock, flags);
139 	while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) {
140 		if (!chan->ring_bufs_avail) {
141 			chan->ring_bufs_avail = 1;
142 			need_wakeup = true;
143 		}
144 
145 		if (len) {
146 			req->rc.size = len;
147 			p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
148 		}
149 	}
150 	spin_unlock_irqrestore(&chan->lock, flags);
151 	/* Wakeup if anyone waiting for VirtIO ring space. */
152 	if (need_wakeup)
153 		wake_up(chan->vc_wq);
154 }
155 
156 /**
157  * pack_sg_list - pack a scatter gather list from a linear buffer
158  * @sg: scatter/gather list to pack into
159  * @start: which segment of the sg_list to start at
160  * @limit: maximum segment to pack data to
161  * @data: data to pack into scatter/gather list
162  * @count: amount of data to pack into the scatter/gather list
163  *
164  * sg_lists have multiple segments of various sizes.  This will pack
165  * arbitrary data into an existing scatter gather list, segmenting the
166  * data as necessary within constraints.
167  *
168  */
169 
pack_sg_list(struct scatterlist * sg,int start,int limit,char * data,int count)170 static int pack_sg_list(struct scatterlist *sg, int start,
171 			int limit, char *data, int count)
172 {
173 	int s;
174 	int index = start;
175 
176 	while (count) {
177 		s = rest_of_page(data);
178 		if (s > count)
179 			s = count;
180 		BUG_ON(index >= limit);
181 		/* Make sure we don't terminate early. */
182 		sg_unmark_end(&sg[index]);
183 		sg_set_buf(&sg[index++], data, s);
184 		count -= s;
185 		data += s;
186 	}
187 	if (index-start)
188 		sg_mark_end(&sg[index - 1]);
189 	return index-start;
190 }
191 
192 /* We don't currently allow canceling of virtio requests */
p9_virtio_cancel(struct p9_client * client,struct p9_req_t * req)193 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
194 {
195 	return 1;
196 }
197 
198 /* Reply won't come, so drop req ref */
p9_virtio_cancelled(struct p9_client * client,struct p9_req_t * req)199 static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req)
200 {
201 	p9_req_put(client, req);
202 	return 0;
203 }
204 
205 /**
206  * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
207  * this takes a list of pages.
208  * @sg: scatter/gather list to pack into
209  * @start: which segment of the sg_list to start at
210  * @limit: maximum number of pages in sg list.
211  * @pdata: a list of pages to add into sg.
212  * @nr_pages: number of pages to pack into the scatter/gather list
213  * @offs: amount of data in the beginning of first page _not_ to pack
214  * @count: amount of data to pack into the scatter/gather list
215  */
216 static int
pack_sg_list_p(struct scatterlist * sg,int start,int limit,struct page ** pdata,int nr_pages,size_t offs,int count)217 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
218 	       struct page **pdata, int nr_pages, size_t offs, int count)
219 {
220 	int i = 0, s;
221 	int data_off = offs;
222 	int index = start;
223 
224 	BUG_ON(nr_pages > (limit - start));
225 	/*
226 	 * if the first page doesn't start at
227 	 * page boundary find the offset
228 	 */
229 	while (nr_pages) {
230 		s = PAGE_SIZE - data_off;
231 		if (s > count)
232 			s = count;
233 		BUG_ON(index >= limit);
234 		/* Make sure we don't terminate early. */
235 		sg_unmark_end(&sg[index]);
236 		sg_set_page(&sg[index++], pdata[i++], s, data_off);
237 		data_off = 0;
238 		count -= s;
239 		nr_pages--;
240 	}
241 
242 	if (index-start)
243 		sg_mark_end(&sg[index - 1]);
244 	return index - start;
245 }
246 
247 /**
248  * p9_virtio_request - issue a request
249  * @client: client instance issuing the request
250  * @req: request to be issued
251  *
252  */
253 
254 static int
p9_virtio_request(struct p9_client * client,struct p9_req_t * req)255 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
256 {
257 	int err;
258 	int in, out, out_sgs, in_sgs;
259 	unsigned long flags;
260 	struct virtio_chan *chan = client->trans;
261 	struct scatterlist *sgs[2];
262 
263 	p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
264 
265 	WRITE_ONCE(req->status, REQ_STATUS_SENT);
266 req_retry:
267 	spin_lock_irqsave(&chan->lock, flags);
268 
269 	out_sgs = in_sgs = 0;
270 	/* Handle out VirtIO ring buffers */
271 	out = pack_sg_list(chan->sg, 0,
272 			   VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
273 	if (out)
274 		sgs[out_sgs++] = chan->sg;
275 
276 	in = pack_sg_list(chan->sg, out,
277 			  VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity);
278 	if (in)
279 		sgs[out_sgs + in_sgs++] = chan->sg + out;
280 
281 	err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
282 				GFP_ATOMIC);
283 	if (err < 0) {
284 		if (err == -ENOSPC) {
285 			chan->ring_bufs_avail = 0;
286 			spin_unlock_irqrestore(&chan->lock, flags);
287 			err = io_wait_event_killable(*chan->vc_wq,
288 						     chan->ring_bufs_avail);
289 			if (err  == -ERESTARTSYS)
290 				return err;
291 
292 			p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
293 			goto req_retry;
294 		} else {
295 			spin_unlock_irqrestore(&chan->lock, flags);
296 			p9_debug(P9_DEBUG_TRANS,
297 				 "virtio rpc add_sgs returned failure\n");
298 			return -EIO;
299 		}
300 	}
301 	virtqueue_kick(chan->vq);
302 	spin_unlock_irqrestore(&chan->lock, flags);
303 
304 	p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
305 	return 0;
306 }
307 
p9_get_mapped_pages(struct virtio_chan * chan,struct page *** pages,struct iov_iter * data,int count,size_t * offs,int * need_drop)308 static int p9_get_mapped_pages(struct virtio_chan *chan,
309 			       struct page ***pages,
310 			       struct iov_iter *data,
311 			       int count,
312 			       size_t *offs,
313 			       int *need_drop)
314 {
315 	int nr_pages;
316 	int err;
317 
318 	if (!iov_iter_count(data))
319 		return 0;
320 
321 	if (!iov_iter_is_kvec(data)) {
322 		int n;
323 		/*
324 		 * We allow only p9_max_pages pinned. We wait for the
325 		 * Other zc request to finish here
326 		 */
327 		if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
328 			err = io_wait_event_killable(vp_wq,
329 			      (atomic_read(&vp_pinned) < chan->p9_max_pages));
330 			if (err == -ERESTARTSYS)
331 				return err;
332 		}
333 		n = iov_iter_get_pages_alloc2(data, pages, count, offs);
334 		if (n < 0)
335 			return n;
336 		*need_drop = 1;
337 		nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
338 		atomic_add(nr_pages, &vp_pinned);
339 		return n;
340 	} else {
341 		/* kernel buffer, no need to pin pages */
342 		int index;
343 		size_t len;
344 		void *p;
345 
346 		/* we'd already checked that it's non-empty */
347 		while (1) {
348 			len = iov_iter_single_seg_count(data);
349 			if (likely(len)) {
350 				p = data->kvec->iov_base + data->iov_offset;
351 				break;
352 			}
353 			iov_iter_advance(data, 0);
354 		}
355 		if (len > count)
356 			len = count;
357 
358 		nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
359 			   (unsigned long)p / PAGE_SIZE;
360 
361 		*pages = kmalloc_objs(struct page *, nr_pages, GFP_NOFS);
362 		if (!*pages)
363 			return -ENOMEM;
364 
365 		*need_drop = 0;
366 		p -= (*offs = offset_in_page(p));
367 		for (index = 0; index < nr_pages; index++) {
368 			if (is_vmalloc_addr(p))
369 				(*pages)[index] = vmalloc_to_page(p);
370 			else
371 				(*pages)[index] = kmap_to_page(p);
372 			p += PAGE_SIZE;
373 		}
374 		iov_iter_advance(data, len);
375 		return len;
376 	}
377 }
378 
handle_rerror(struct p9_req_t * req,int in_hdr_len,size_t offs,struct page ** pages)379 static void handle_rerror(struct p9_req_t *req, int in_hdr_len,
380 			  size_t offs, struct page **pages)
381 {
382 	unsigned size, n;
383 	void *to = req->rc.sdata + in_hdr_len;
384 
385 	// Fits entirely into the static data?  Nothing to do.
386 	if (req->rc.size < in_hdr_len || !pages)
387 		return;
388 
389 	// Really long error message?  Tough, truncate the reply.  Might get
390 	// rejected (we can't be arsed to adjust the size encoded in header,
391 	// or string size for that matter), but it wouldn't be anything valid
392 	// anyway.
393 	if (unlikely(req->rc.size > P9_ZC_HDR_SZ))
394 		req->rc.size = P9_ZC_HDR_SZ;
395 
396 	// data won't span more than two pages
397 	size = req->rc.size - in_hdr_len;
398 	n = PAGE_SIZE - offs;
399 	if (size > n) {
400 		memcpy_from_page(to, *pages++, offs, n);
401 		offs = 0;
402 		to += n;
403 		size -= n;
404 	}
405 	memcpy_from_page(to, *pages, offs, size);
406 }
407 
408 /**
409  * p9_virtio_zc_request - issue a zero copy request
410  * @client: client instance issuing the request
411  * @req: request to be issued
412  * @uidata: user buffer that should be used for zero copy read
413  * @uodata: user buffer that should be used for zero copy write
414  * @inlen: read buffer size
415  * @outlen: write buffer size
416  * @in_hdr_len: reader header size, This is the size of response protocol data
417  *
418  */
419 static int
p9_virtio_zc_request(struct p9_client * client,struct p9_req_t * req,struct iov_iter * uidata,struct iov_iter * uodata,int inlen,int outlen,int in_hdr_len)420 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
421 		     struct iov_iter *uidata, struct iov_iter *uodata,
422 		     int inlen, int outlen, int in_hdr_len)
423 {
424 	int in, out, err, out_sgs, in_sgs;
425 	unsigned long flags;
426 	int in_nr_pages = 0, out_nr_pages = 0;
427 	struct page **in_pages = NULL, **out_pages = NULL;
428 	struct virtio_chan *chan = client->trans;
429 	struct scatterlist *sgs[4];
430 	size_t offs = 0;
431 	int need_drop = 0;
432 	int kicked = 0;
433 
434 	p9_debug(P9_DEBUG_TRANS, "virtio request\n");
435 
436 	if (uodata) {
437 		__le32 sz;
438 		int n = p9_get_mapped_pages(chan, &out_pages, uodata,
439 					    outlen, &offs, &need_drop);
440 		if (n < 0) {
441 			err = n;
442 			goto err_out;
443 		}
444 		out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
445 		if (n != outlen) {
446 			__le32 v = cpu_to_le32(n);
447 			memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
448 			outlen = n;
449 		}
450 		/* The size field of the message must include the length of the
451 		 * header and the length of the data.  We didn't actually know
452 		 * the length of the data until this point so add it in now.
453 		 */
454 		sz = cpu_to_le32(req->tc.size + outlen);
455 		memcpy(&req->tc.sdata[0], &sz, sizeof(sz));
456 	} else if (uidata) {
457 		int n = p9_get_mapped_pages(chan, &in_pages, uidata,
458 					    inlen, &offs, &need_drop);
459 		if (n < 0) {
460 			err = n;
461 			goto err_out;
462 		}
463 		in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
464 		if (n != inlen) {
465 			__le32 v = cpu_to_le32(n);
466 			memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
467 			inlen = n;
468 		}
469 	}
470 	WRITE_ONCE(req->status, REQ_STATUS_SENT);
471 req_retry_pinned:
472 	spin_lock_irqsave(&chan->lock, flags);
473 
474 	out_sgs = in_sgs = 0;
475 
476 	/* out data */
477 	out = pack_sg_list(chan->sg, 0,
478 			   VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
479 
480 	if (out)
481 		sgs[out_sgs++] = chan->sg;
482 
483 	if (out_pages) {
484 		sgs[out_sgs++] = chan->sg + out;
485 		out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
486 				      out_pages, out_nr_pages, offs, outlen);
487 	}
488 
489 	/*
490 	 * Take care of in data
491 	 * For example TREAD have 11.
492 	 * 11 is the read/write header = PDU Header(7) + IO Size (4).
493 	 * Arrange in such a way that server places header in the
494 	 * allocated memory and payload onto the user buffer.
495 	 */
496 	in = pack_sg_list(chan->sg, out,
497 			  VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len);
498 	if (in)
499 		sgs[out_sgs + in_sgs++] = chan->sg + out;
500 
501 	if (in_pages) {
502 		sgs[out_sgs + in_sgs++] = chan->sg + out + in;
503 		pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
504 			       in_pages, in_nr_pages, offs, inlen);
505 	}
506 
507 	BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
508 	err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
509 				GFP_ATOMIC);
510 	if (err < 0) {
511 		if (err == -ENOSPC) {
512 			chan->ring_bufs_avail = 0;
513 			spin_unlock_irqrestore(&chan->lock, flags);
514 			err = io_wait_event_killable(*chan->vc_wq,
515 						     chan->ring_bufs_avail);
516 			if (err  == -ERESTARTSYS)
517 				goto err_out;
518 
519 			p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
520 			goto req_retry_pinned;
521 		} else {
522 			spin_unlock_irqrestore(&chan->lock, flags);
523 			p9_debug(P9_DEBUG_TRANS,
524 				 "virtio rpc add_sgs returned failure\n");
525 			err = -EIO;
526 			goto err_out;
527 		}
528 	}
529 	virtqueue_kick(chan->vq);
530 	spin_unlock_irqrestore(&chan->lock, flags);
531 	kicked = 1;
532 	p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
533 	err = io_wait_event_killable(req->wq,
534 				     READ_ONCE(req->status) >= REQ_STATUS_RCVD);
535 	// RERROR needs reply (== error string) in static data
536 	if (READ_ONCE(req->status) == REQ_STATUS_RCVD &&
537 	    unlikely(req->rc.sdata[4] == P9_RERROR))
538 		handle_rerror(req, in_hdr_len, offs, in_pages);
539 
540 	/*
541 	 * Non kernel buffers are pinned, unpin them
542 	 */
543 err_out:
544 	if (need_drop) {
545 		if (in_pages) {
546 			p9_release_pages(in_pages, in_nr_pages);
547 			atomic_sub(in_nr_pages, &vp_pinned);
548 		}
549 		if (out_pages) {
550 			p9_release_pages(out_pages, out_nr_pages);
551 			atomic_sub(out_nr_pages, &vp_pinned);
552 		}
553 		/* wakeup anybody waiting for slots to pin pages */
554 		wake_up(&vp_wq);
555 	}
556 	kvfree(in_pages);
557 	kvfree(out_pages);
558 	if (!kicked) {
559 		/* reply won't come */
560 		p9_req_put(client, req);
561 	}
562 	return err;
563 }
564 
p9_mount_tag_show(struct device * dev,struct device_attribute * attr,char * buf)565 static ssize_t p9_mount_tag_show(struct device *dev,
566 				struct device_attribute *attr, char *buf)
567 {
568 	struct virtio_chan *chan;
569 	struct virtio_device *vdev;
570 	int tag_len;
571 
572 	vdev = dev_to_virtio(dev);
573 	chan = vdev->priv;
574 	tag_len = strlen(chan->tag);
575 
576 	memcpy(buf, chan->tag, tag_len + 1);
577 
578 	return tag_len + 1;
579 }
580 
581 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
582 
583 /**
584  * p9_virtio_probe - probe for existence of 9P virtio channels
585  * @vdev: virtio device to probe
586  *
587  * This probes for existing virtio channels.
588  *
589  */
590 
p9_virtio_probe(struct virtio_device * vdev)591 static int p9_virtio_probe(struct virtio_device *vdev)
592 {
593 	__u16 tag_len;
594 	char *tag;
595 	int err;
596 	struct virtio_chan *chan;
597 
598 	if (!vdev->config->get) {
599 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
600 			__func__);
601 		return -EINVAL;
602 	}
603 
604 	chan = kmalloc_obj(struct virtio_chan);
605 	if (!chan) {
606 		pr_err("Failed to allocate virtio 9P channel\n");
607 		err = -ENOMEM;
608 		goto fail;
609 	}
610 
611 	chan->vdev = vdev;
612 
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);
617 		goto out_free_chan;
618 	}
619 	chan->vq->vdev->priv = chan;
620 	spin_lock_init(&chan->lock);
621 
622 	sg_init_table(chan->sg, VIRTQUEUE_NUM);
623 
624 	chan->inuse = false;
625 	if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
626 		virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
627 	} else {
628 		err = -EINVAL;
629 		goto out_free_vq;
630 	}
631 	tag = kzalloc(tag_len + 1, GFP_KERNEL);
632 	if (!tag) {
633 		err = -ENOMEM;
634 		goto out_free_vq;
635 	}
636 
637 	virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
638 			   tag, tag_len);
639 	chan->tag = tag;
640 	err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
641 	if (err) {
642 		goto out_free_tag;
643 	}
644 	chan->vc_wq = kmalloc_obj(wait_queue_head_t);
645 	if (!chan->vc_wq) {
646 		err = -ENOMEM;
647 		goto out_remove_file;
648 	}
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;
653 
654 	virtio_device_ready(vdev);
655 
656 	mutex_lock(&virtio_9p_lock);
657 	list_add_tail(&chan->chan_list, &virtio_chan_list);
658 	mutex_unlock(&virtio_9p_lock);
659 
660 	/* Let udev rules use the new mount_tag attribute. */
661 	kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
662 
663 	return 0;
664 
665 out_remove_file:
666 	sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr);
667 out_free_tag:
668 	kfree(tag);
669 out_free_vq:
670 	vdev->config->del_vqs(vdev);
671 out_free_chan:
672 	kfree(chan);
673 fail:
674 	return err;
675 }
676 
677 
678 /**
679  * p9_virtio_create - allocate a new virtio channel
680  * @client: client instance invoking this transport
681  * @fc: the filesystem context
682  *
683  * This sets up a transport channel for 9p communication.  Right now
684  * we only match the first available channel, but eventually we could look up
685  * alternate channels by matching devname versus a virtio_config entry.
686  * We use a simple reference count mechanism to ensure that only a single
687  * mount has a channel open at a time.
688  *
689  */
690 
691 static int
p9_virtio_create(struct p9_client * client,struct fs_context * fc)692 p9_virtio_create(struct p9_client *client, struct fs_context *fc)
693 {
694 	const char *devname = fc->source;
695 	struct virtio_chan *chan;
696 	int ret = -ENOENT;
697 	int found = 0;
698 
699 	if (devname == NULL)
700 		return -EINVAL;
701 
702 	mutex_lock(&virtio_9p_lock);
703 	list_for_each_entry(chan, &virtio_chan_list, chan_list) {
704 		if (!strcmp(devname, chan->tag)) {
705 			if (!chan->inuse) {
706 				chan->inuse = true;
707 				found = 1;
708 				break;
709 			}
710 			ret = -EBUSY;
711 		}
712 	}
713 	mutex_unlock(&virtio_9p_lock);
714 
715 	if (!found) {
716 		pr_err("no channels available for device %s\n", devname);
717 		return ret;
718 	}
719 
720 	client->trans = (void *)chan;
721 	client->status = Connected;
722 	chan->client = client;
723 
724 	return 0;
725 }
726 
727 /**
728  * p9_virtio_remove - clean up resources associated with a virtio device
729  * @vdev: virtio device to remove
730  *
731  */
732 
p9_virtio_remove(struct virtio_device * vdev)733 static void p9_virtio_remove(struct virtio_device *vdev)
734 {
735 	struct virtio_chan *chan = vdev->priv;
736 	unsigned long warning_time;
737 
738 	mutex_lock(&virtio_9p_lock);
739 
740 	/* Remove self from list so we don't get new users. */
741 	list_del(&chan->chan_list);
742 	warning_time = jiffies;
743 
744 	/* Wait for existing users to close. */
745 	while (chan->inuse) {
746 		mutex_unlock(&virtio_9p_lock);
747 		msleep(250);
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;
752 		}
753 		mutex_lock(&virtio_9p_lock);
754 	}
755 
756 	mutex_unlock(&virtio_9p_lock);
757 
758 	virtio_reset_device(vdev);
759 	vdev->config->del_vqs(vdev);
760 
761 	sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
762 	kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
763 	kfree(chan->tag);
764 	kfree(chan->vc_wq);
765 	kfree(chan);
766 
767 }
768 
769 static struct virtio_device_id id_table[] = {
770 	{ VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
771 	{ 0 },
772 };
773 
774 static unsigned int features[] = {
775 	VIRTIO_9P_MOUNT_TAG,
776 };
777 
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 	.id_table	= id_table,
784 	.probe		= p9_virtio_probe,
785 	.remove		= p9_virtio_remove,
786 };
787 
788 static struct p9_trans_module p9_virtio_trans = {
789 	.name = "virtio",
790 	.create = p9_virtio_create,
791 	.close = p9_virtio_close,
792 	.request = p9_virtio_request,
793 	.zc_request = p9_virtio_zc_request,
794 	.cancel = p9_virtio_cancel,
795 	.cancelled = p9_virtio_cancelled,
796 	/*
797 	 * We leave one entry for input and one entry for response
798 	 * headers. We also skip one more entry to accommodate, address
799 	 * that are not at page boundary, that can result in an extra
800 	 * page in zero copy.
801 	 */
802 	.maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
803 	.pooled_rbuffers = false,
804 	.def = true,
805 	.supports_vmalloc = false,
806 	.owner = THIS_MODULE,
807 };
808 
809 /* The standard init function */
p9_virtio_init(void)810 static int __init p9_virtio_init(void)
811 {
812 	int rc;
813 
814 	INIT_LIST_HEAD(&virtio_chan_list);
815 
816 	v9fs_register_trans(&p9_virtio_trans);
817 	rc = register_virtio_driver(&p9_virtio_drv);
818 	if (rc)
819 		v9fs_unregister_trans(&p9_virtio_trans);
820 
821 	return rc;
822 }
823 
p9_virtio_cleanup(void)824 static void __exit p9_virtio_cleanup(void)
825 {
826 	unregister_virtio_driver(&p9_virtio_drv);
827 	v9fs_unregister_trans(&p9_virtio_trans);
828 }
829 
830 module_init(p9_virtio_init);
831 module_exit(p9_virtio_cleanup);
832 MODULE_ALIAS_9P("virtio");
833 
834 MODULE_DEVICE_TABLE(virtio, id_table);
835 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
836 MODULE_DESCRIPTION("Virtio 9p Transport");
837 MODULE_LICENSE("GPL");
838