xref: /openbmc/linux/net/9p/trans_virtio.c (revision 93dc544c)
1 /*
2  * The Guest 9p transport driver
3  *
4  * This is a block based transport driver based on the lguest block driver
5  * code.
6  *
7  */
8 /*
9  *  Copyright (C) 2007 Eric Van Hensbergen, IBM Corporation
10  *
11  *  Based on virtio console driver
12  *  Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
13  *
14  *  This program is free software; you can redistribute it and/or modify
15  *  it under the terms of the GNU General Public License version 2
16  *  as published by the Free Software Foundation.
17  *
18  *  This program is distributed in the hope that it will be useful,
19  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
20  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21  *  GNU General Public License for more details.
22  *
23  *  You should have received a copy of the GNU General Public License
24  *  along with this program; if not, write to:
25  *  Free Software Foundation
26  *  51 Franklin Street, Fifth Floor
27  *  Boston, MA  02111-1301  USA
28  *
29  */
30 
31 #include <linux/in.h>
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>
37 #include <linux/un.h>
38 #include <linux/uaccess.h>
39 #include <linux/inet.h>
40 #include <linux/idr.h>
41 #include <linux/file.h>
42 #include <net/9p/9p.h>
43 #include <linux/parser.h>
44 #include <net/9p/transport.h>
45 #include <linux/scatterlist.h>
46 #include <linux/virtio.h>
47 #include <linux/virtio_9p.h>
48 
49 #define VIRTQUEUE_NUM	128
50 
51 /* a single mutex to manage channel initialization and attachment */
52 static DEFINE_MUTEX(virtio_9p_lock);
53 /* global which tracks highest initialized channel */
54 static int chan_index;
55 
56 #define P9_INIT_MAXTAG	16
57 
58 
59 /**
60  * enum p9_req_status_t - virtio request status
61  * @REQ_STATUS_IDLE: request slot unused
62  * @REQ_STATUS_SENT: request sent to server
63  * @REQ_STATUS_RCVD: response received from server
64  * @REQ_STATUS_FLSH: request has been flushed
65  *
66  * The @REQ_STATUS_IDLE state is used to mark a request slot as unused
67  * but use is actually tracked by the idpool structure which handles tag
68  * id allocation.
69  *
70  */
71 
72 enum p9_req_status_t {
73 	REQ_STATUS_IDLE,
74 	REQ_STATUS_SENT,
75 	REQ_STATUS_RCVD,
76 	REQ_STATUS_FLSH,
77 };
78 
79 /**
80  * struct p9_req_t - virtio request slots
81  * @status: status of this request slot
82  * @wq: wait_queue for the client to block on for this request
83  *
84  * The virtio transport uses an array to track outstanding requests
85  * instead of a list.  While this may incurr overhead during initial
86  * allocation or expansion, it makes request lookup much easier as the
87  * tag id is a index into an array.  (We use tag+1 so that we can accomodate
88  * the -1 tag for the T_VERSION request).
89  * This also has the nice effect of only having to allocate wait_queues
90  * once, instead of constantly allocating and freeing them.  Its possible
91  * other resources could benefit from this scheme as well.
92  *
93  */
94 
95 struct p9_req_t {
96 	int status;
97 	wait_queue_head_t *wq;
98 };
99 
100 /**
101  * struct virtio_chan - per-instance transport information
102  * @initialized: whether the channel is initialized
103  * @inuse: whether the channel is in use
104  * @lock: protects multiple elements within this structure
105  * @vdev: virtio dev associated with this channel
106  * @vq: virtio queue associated with this channel
107  * @tagpool: accounting for tag ids (and request slots)
108  * @reqs: array of request slots
109  * @max_tag: current number of request_slots allocated
110  * @sg: scatter gather list which is used to pack a request (protected?)
111  *
112  * We keep all per-channel information in a structure.
113  * This structure is allocated within the devices dev->mem space.
114  * A pointer to the structure will get put in the transport private.
115  *
116  */
117 
118 static struct virtio_chan {
119 	bool initialized;
120 	bool inuse;
121 
122 	spinlock_t lock;
123 
124 	struct virtio_device *vdev;
125 	struct virtqueue *vq;
126 
127 	struct p9_idpool *tagpool;
128 	struct p9_req_t *reqs;
129 	int max_tag;
130 
131 	/* Scatterlist: can be too big for stack. */
132 	struct scatterlist sg[VIRTQUEUE_NUM];
133 } channels[MAX_9P_CHAN];
134 
135 /**
136  * p9_lookup_tag - Lookup requests by tag
137  * @c: virtio channel to lookup tag within
138  * @tag: numeric id for transaction
139  *
140  * this is a simple array lookup, but will grow the
141  * request_slots as necessary to accomodate transaction
142  * ids which did not previously have a slot.
143  *
144  * Bugs: there is currently no upper limit on request slots set
145  * here, but that should be constrained by the id accounting.
146  */
147 
148 static struct p9_req_t *p9_lookup_tag(struct virtio_chan *c, u16 tag)
149 {
150 	/* This looks up the original request by tag so we know which
151 	 * buffer to read the data into */
152 	tag++;
153 
154 	while (tag >= c->max_tag) {
155 		int old_max = c->max_tag;
156 		int count;
157 
158 		if (c->max_tag)
159 			c->max_tag *= 2;
160 		else
161 			c->max_tag = P9_INIT_MAXTAG;
162 
163 		c->reqs = krealloc(c->reqs, sizeof(struct p9_req_t)*c->max_tag,
164 								GFP_ATOMIC);
165 		if (!c->reqs) {
166 			printk(KERN_ERR "Couldn't grow tag array\n");
167 			BUG();
168 		}
169 		for (count = old_max; count < c->max_tag; count++) {
170 			c->reqs[count].status = REQ_STATUS_IDLE;
171 			c->reqs[count].wq = kmalloc(sizeof(wait_queue_head_t),
172 								GFP_ATOMIC);
173 			if (!c->reqs[count].wq) {
174 				printk(KERN_ERR "Couldn't grow tag array\n");
175 				BUG();
176 			}
177 			init_waitqueue_head(c->reqs[count].wq);
178 		}
179 	}
180 
181 	return &c->reqs[tag];
182 }
183 
184 
185 /* How many bytes left in this page. */
186 static unsigned int rest_of_page(void *data)
187 {
188 	return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE);
189 }
190 
191 /**
192  * p9_virtio_close - reclaim resources of a channel
193  * @trans: transport state
194  *
195  * This reclaims a channel by freeing its resources and
196  * reseting its inuse flag.
197  *
198  */
199 
200 static void p9_virtio_close(struct p9_trans *trans)
201 {
202 	struct virtio_chan *chan = trans->priv;
203 	int count;
204 	unsigned long flags;
205 
206 	spin_lock_irqsave(&chan->lock, flags);
207 	p9_idpool_destroy(chan->tagpool);
208 	for (count = 0; count < chan->max_tag; count++)
209 		kfree(chan->reqs[count].wq);
210 	kfree(chan->reqs);
211 	chan->max_tag = 0;
212 	spin_unlock_irqrestore(&chan->lock, flags);
213 
214 	mutex_lock(&virtio_9p_lock);
215 	chan->inuse = false;
216 	mutex_unlock(&virtio_9p_lock);
217 
218 	kfree(trans);
219 }
220 
221 /**
222  * req_done - callback which signals activity from the server
223  * @vq: virtio queue activity was received on
224  *
225  * This notifies us that the server has triggered some activity
226  * on the virtio channel - most likely a response to request we
227  * sent.  Figure out which requests now have responses and wake up
228  * those threads.
229  *
230  * Bugs: could do with some additional sanity checking, but appears to work.
231  *
232  */
233 
234 static void req_done(struct virtqueue *vq)
235 {
236 	struct virtio_chan *chan = vq->vdev->priv;
237 	struct p9_fcall *rc;
238 	unsigned int len;
239 	unsigned long flags;
240 	struct p9_req_t *req;
241 
242 	spin_lock_irqsave(&chan->lock, flags);
243 	while ((rc = chan->vq->vq_ops->get_buf(chan->vq, &len)) != NULL) {
244 		req = p9_lookup_tag(chan, rc->tag);
245 		req->status = REQ_STATUS_RCVD;
246 		wake_up(req->wq);
247 	}
248 	/* In case queue is stopped waiting for more buffers. */
249 	spin_unlock_irqrestore(&chan->lock, flags);
250 }
251 
252 /**
253  * pack_sg_list - pack a scatter gather list from a linear buffer
254  * @sg: scatter/gather list to pack into
255  * @start: which segment of the sg_list to start at
256  * @limit: maximum segment to pack data to
257  * @data: data to pack into scatter/gather list
258  * @count: amount of data to pack into the scatter/gather list
259  *
260  * sg_lists have multiple segments of various sizes.  This will pack
261  * arbitrary data into an existing scatter gather list, segmenting the
262  * data as necessary within constraints.
263  *
264  */
265 
266 static int
267 pack_sg_list(struct scatterlist *sg, int start, int limit, char *data,
268 								int count)
269 {
270 	int s;
271 	int index = start;
272 
273 	while (count) {
274 		s = rest_of_page(data);
275 		if (s > count)
276 			s = count;
277 		sg_set_buf(&sg[index++], data, s);
278 		count -= s;
279 		data += s;
280 		BUG_ON(index > limit);
281 	}
282 
283 	return index-start;
284 }
285 
286 /**
287  * p9_virtio_rpc - issue a request and wait for a response
288  * @t: transport state
289  * @tc: &p9_fcall request to transmit
290  * @rc: &p9_fcall to put reponse into
291  *
292  */
293 
294 static int
295 p9_virtio_rpc(struct p9_trans *t, struct p9_fcall *tc, struct p9_fcall **rc)
296 {
297 	int in, out;
298 	int n, err, size;
299 	struct virtio_chan *chan = t->priv;
300 	char *rdata;
301 	struct p9_req_t *req;
302 	unsigned long flags;
303 
304 	if (*rc == NULL) {
305 		*rc = kmalloc(sizeof(struct p9_fcall) + t->msize, GFP_KERNEL);
306 		if (!*rc)
307 			return -ENOMEM;
308 	}
309 
310 	rdata = (char *)*rc+sizeof(struct p9_fcall);
311 
312 	n = P9_NOTAG;
313 	if (tc->id != P9_TVERSION) {
314 		n = p9_idpool_get(chan->tagpool);
315 		if (n < 0)
316 			return -ENOMEM;
317 	}
318 
319 	spin_lock_irqsave(&chan->lock, flags);
320 	req = p9_lookup_tag(chan, n);
321 	spin_unlock_irqrestore(&chan->lock, flags);
322 
323 	p9_set_tag(tc, n);
324 
325 	P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio rpc tag %d\n", n);
326 
327 	out = pack_sg_list(chan->sg, 0, VIRTQUEUE_NUM, tc->sdata, tc->size);
328 	in = pack_sg_list(chan->sg, out, VIRTQUEUE_NUM-out, rdata, t->msize);
329 
330 	req->status = REQ_STATUS_SENT;
331 
332 	if (chan->vq->vq_ops->add_buf(chan->vq, chan->sg, out, in, tc)) {
333 		P9_DPRINTK(P9_DEBUG_TRANS,
334 			"9p debug: virtio rpc add_buf returned failure");
335 		return -EIO;
336 	}
337 
338 	chan->vq->vq_ops->kick(chan->vq);
339 
340 	wait_event(*req->wq, req->status == REQ_STATUS_RCVD);
341 
342 	size = le32_to_cpu(*(__le32 *) rdata);
343 
344 	err = p9_deserialize_fcall(rdata, size, *rc, t->extended);
345 	if (err < 0) {
346 		P9_DPRINTK(P9_DEBUG_TRANS,
347 			"9p debug: virtio rpc deserialize returned %d\n", err);
348 		return err;
349 	}
350 
351 #ifdef CONFIG_NET_9P_DEBUG
352 	if ((p9_debug_level&P9_DEBUG_FCALL) == P9_DEBUG_FCALL) {
353 		char buf[150];
354 
355 		p9_printfcall(buf, sizeof(buf), *rc, t->extended);
356 		printk(KERN_NOTICE ">>> %p %s\n", t, buf);
357 	}
358 #endif
359 
360 	if (n != P9_NOTAG && p9_idpool_check(n, chan->tagpool))
361 		p9_idpool_put(n, chan->tagpool);
362 
363 	req->status = REQ_STATUS_IDLE;
364 
365 	return 0;
366 }
367 
368 /**
369  * p9_virtio_probe - probe for existence of 9P virtio channels
370  * @vdev: virtio device to probe
371  *
372  * This probes for existing virtio channels.  At present only
373  * a single channel is in use, so in the future more work may need
374  * to be done here.
375  *
376  */
377 
378 static int p9_virtio_probe(struct virtio_device *vdev)
379 {
380 	int err;
381 	struct virtio_chan *chan;
382 	int index;
383 
384 	mutex_lock(&virtio_9p_lock);
385 	index = chan_index++;
386 	chan = &channels[index];
387 	mutex_unlock(&virtio_9p_lock);
388 
389 	if (chan_index > MAX_9P_CHAN) {
390 		printk(KERN_ERR "9p: virtio: Maximum channels exceeded\n");
391 		BUG();
392 		err = -ENOMEM;
393 		goto fail;
394 	}
395 
396 	chan->vdev = vdev;
397 
398 	/* We expect one virtqueue, for requests. */
399 	chan->vq = vdev->config->find_vq(vdev, 0, req_done);
400 	if (IS_ERR(chan->vq)) {
401 		err = PTR_ERR(chan->vq);
402 		goto out_free_vq;
403 	}
404 	chan->vq->vdev->priv = chan;
405 	spin_lock_init(&chan->lock);
406 
407 	sg_init_table(chan->sg, VIRTQUEUE_NUM);
408 
409 	chan->inuse = false;
410 	chan->initialized = true;
411 	return 0;
412 
413 out_free_vq:
414 	vdev->config->del_vq(chan->vq);
415 fail:
416 	mutex_lock(&virtio_9p_lock);
417 	chan_index--;
418 	mutex_unlock(&virtio_9p_lock);
419 	return err;
420 }
421 
422 
423 /**
424  * p9_virtio_create - allocate a new virtio channel
425  * @devname: string identifying the channel to connect to (unused)
426  * @args: args passed from sys_mount() for per-transport options (unused)
427  * @msize: requested maximum packet size
428  * @extended: 9p2000.u enabled flag
429  *
430  * This sets up a transport channel for 9p communication.  Right now
431  * we only match the first available channel, but eventually we couldlook up
432  * alternate channels by matching devname versus a virtio_config entry.
433  * We use a simple reference count mechanism to ensure that only a single
434  * mount has a channel open at a time.
435  *
436  * Bugs: doesn't allow identification of a specific channel
437  * to allocate, channels are allocated sequentially. This was
438  * a pragmatic decision to get things rolling, but ideally some
439  * way of identifying the channel to attach to would be nice
440  * if we are going to support multiple channels.
441  *
442  */
443 
444 static struct p9_trans *
445 p9_virtio_create(const char *devname, char *args, int msize,
446 							unsigned char extended)
447 {
448 	struct p9_trans *trans;
449 	struct virtio_chan *chan = channels;
450 	int index = 0;
451 
452 	mutex_lock(&virtio_9p_lock);
453 	while (index < MAX_9P_CHAN) {
454 		if (chan->initialized && !chan->inuse) {
455 			chan->inuse = true;
456 			break;
457 		} else {
458 			index++;
459 			chan = &channels[index];
460 		}
461 	}
462 	mutex_unlock(&virtio_9p_lock);
463 
464 	if (index >= MAX_9P_CHAN) {
465 		printk(KERN_ERR "9p: no channels available\n");
466 		return ERR_PTR(-ENODEV);
467 	}
468 
469 	chan->tagpool = p9_idpool_create();
470 	if (IS_ERR(chan->tagpool)) {
471 		printk(KERN_ERR "9p: couldn't allocate tagpool\n");
472 		return ERR_PTR(-ENOMEM);
473 	}
474 	p9_idpool_get(chan->tagpool); /* reserve tag 0 */
475 	chan->max_tag = 0;
476 	chan->reqs = NULL;
477 
478 	trans = kmalloc(sizeof(struct p9_trans), GFP_KERNEL);
479 	if (!trans) {
480 		printk(KERN_ERR "9p: couldn't allocate transport\n");
481 		return ERR_PTR(-ENOMEM);
482 	}
483 	trans->extended = extended;
484 	trans->msize = msize;
485 	trans->close = p9_virtio_close;
486 	trans->rpc = p9_virtio_rpc;
487 	trans->priv = chan;
488 
489 	return trans;
490 }
491 
492 /**
493  * p9_virtio_remove - clean up resources associated with a virtio device
494  * @vdev: virtio device to remove
495  *
496  */
497 
498 static void p9_virtio_remove(struct virtio_device *vdev)
499 {
500 	struct virtio_chan *chan = vdev->priv;
501 
502 	BUG_ON(chan->inuse);
503 
504 	if (chan->initialized) {
505 		vdev->config->del_vq(chan->vq);
506 		chan->initialized = false;
507 	}
508 }
509 
510 #define VIRTIO_ID_9P 9
511 
512 static struct virtio_device_id id_table[] = {
513 	{ VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
514 	{ 0 },
515 };
516 
517 /* The standard "struct lguest_driver": */
518 static struct virtio_driver p9_virtio_drv = {
519 	.driver.name = 	KBUILD_MODNAME,
520 	.driver.owner = THIS_MODULE,
521 	.id_table =	id_table,
522 	.probe = 	p9_virtio_probe,
523 	.remove =	p9_virtio_remove,
524 };
525 
526 static struct p9_trans_module p9_virtio_trans = {
527 	.name = "virtio",
528 	.create = p9_virtio_create,
529 	.maxsize = PAGE_SIZE*16,
530 	.def = 0,
531 };
532 
533 /* The standard init function */
534 static int __init p9_virtio_init(void)
535 {
536 	int count;
537 
538 	for (count = 0; count < MAX_9P_CHAN; count++)
539 		channels[count].initialized = false;
540 
541 	v9fs_register_trans(&p9_virtio_trans);
542 	return register_virtio_driver(&p9_virtio_drv);
543 }
544 
545 static void __exit p9_virtio_cleanup(void)
546 {
547 	unregister_virtio_driver(&p9_virtio_drv);
548 }
549 
550 module_init(p9_virtio_init);
551 module_exit(p9_virtio_cleanup);
552 
553 MODULE_DEVICE_TABLE(virtio, id_table);
554 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
555 MODULE_DESCRIPTION("Virtio 9p Transport");
556 MODULE_LICENSE("GPL");
557