xref: /openbmc/linux/fs/fuse/virtio_fs.c (revision b52455a7)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * virtio-fs: Virtio Filesystem
4  * Copyright (C) 2018 Red Hat, Inc.
5  */
6 
7 #include <linux/fs.h>
8 #include <linux/dax.h>
9 #include <linux/pci.h>
10 #include <linux/pfn_t.h>
11 #include <linux/memremap.h>
12 #include <linux/module.h>
13 #include <linux/virtio.h>
14 #include <linux/virtio_fs.h>
15 #include <linux/delay.h>
16 #include <linux/fs_context.h>
17 #include <linux/fs_parser.h>
18 #include <linux/highmem.h>
19 #include <linux/uio.h>
20 #include "fuse_i.h"
21 
22 /* Used to help calculate the FUSE connection's max_pages limit for a request's
23  * size. Parts of the struct fuse_req are sliced into scattergather lists in
24  * addition to the pages used, so this can help account for that overhead.
25  */
26 #define FUSE_HEADER_OVERHEAD    4
27 
28 /* List of virtio-fs device instances and a lock for the list. Also provides
29  * mutual exclusion in device removal and mounting path
30  */
31 static DEFINE_MUTEX(virtio_fs_mutex);
32 static LIST_HEAD(virtio_fs_instances);
33 
34 enum {
35 	VQ_HIPRIO,
36 	VQ_REQUEST
37 };
38 
39 #define VQ_NAME_LEN	24
40 
41 /* Per-virtqueue state */
42 struct virtio_fs_vq {
43 	spinlock_t lock;
44 	struct virtqueue *vq;     /* protected by ->lock */
45 	struct work_struct done_work;
46 	struct list_head queued_reqs;
47 	struct list_head end_reqs;	/* End these requests */
48 	struct delayed_work dispatch_work;
49 	struct fuse_dev *fud;
50 	bool connected;
51 	long in_flight;
52 	struct completion in_flight_zero; /* No inflight requests */
53 	char name[VQ_NAME_LEN];
54 } ____cacheline_aligned_in_smp;
55 
56 /* A virtio-fs device instance */
57 struct virtio_fs {
58 	struct kref refcount;
59 	struct list_head list;    /* on virtio_fs_instances */
60 	char *tag;
61 	struct virtio_fs_vq *vqs;
62 	unsigned int nvqs;               /* number of virtqueues */
63 	unsigned int num_request_queues; /* number of request queues */
64 	struct dax_device *dax_dev;
65 
66 	/* DAX memory window where file contents are mapped */
67 	void *window_kaddr;
68 	phys_addr_t window_phys_addr;
69 	size_t window_len;
70 };
71 
72 struct virtio_fs_forget_req {
73 	struct fuse_in_header ih;
74 	struct fuse_forget_in arg;
75 };
76 
77 struct virtio_fs_forget {
78 	/* This request can be temporarily queued on virt queue */
79 	struct list_head list;
80 	struct virtio_fs_forget_req req;
81 };
82 
83 struct virtio_fs_req_work {
84 	struct fuse_req *req;
85 	struct virtio_fs_vq *fsvq;
86 	struct work_struct done_work;
87 };
88 
89 static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq,
90 				 struct fuse_req *req, bool in_flight);
91 
92 static const struct constant_table dax_param_enums[] = {
93 	{"always",	FUSE_DAX_ALWAYS },
94 	{"never",	FUSE_DAX_NEVER },
95 	{"inode",	FUSE_DAX_INODE_USER },
96 	{}
97 };
98 
99 enum {
100 	OPT_DAX,
101 	OPT_DAX_ENUM,
102 };
103 
104 static const struct fs_parameter_spec virtio_fs_parameters[] = {
105 	fsparam_flag("dax", OPT_DAX),
106 	fsparam_enum("dax", OPT_DAX_ENUM, dax_param_enums),
107 	{}
108 };
109 
110 static int virtio_fs_parse_param(struct fs_context *fsc,
111 				 struct fs_parameter *param)
112 {
113 	struct fs_parse_result result;
114 	struct fuse_fs_context *ctx = fsc->fs_private;
115 	int opt;
116 
117 	opt = fs_parse(fsc, virtio_fs_parameters, param, &result);
118 	if (opt < 0)
119 		return opt;
120 
121 	switch (opt) {
122 	case OPT_DAX:
123 		ctx->dax_mode = FUSE_DAX_ALWAYS;
124 		break;
125 	case OPT_DAX_ENUM:
126 		ctx->dax_mode = result.uint_32;
127 		break;
128 	default:
129 		return -EINVAL;
130 	}
131 
132 	return 0;
133 }
134 
135 static void virtio_fs_free_fsc(struct fs_context *fsc)
136 {
137 	struct fuse_fs_context *ctx = fsc->fs_private;
138 
139 	kfree(ctx);
140 }
141 
142 static inline struct virtio_fs_vq *vq_to_fsvq(struct virtqueue *vq)
143 {
144 	struct virtio_fs *fs = vq->vdev->priv;
145 
146 	return &fs->vqs[vq->index];
147 }
148 
149 /* Should be called with fsvq->lock held. */
150 static inline void inc_in_flight_req(struct virtio_fs_vq *fsvq)
151 {
152 	fsvq->in_flight++;
153 }
154 
155 /* Should be called with fsvq->lock held. */
156 static inline void dec_in_flight_req(struct virtio_fs_vq *fsvq)
157 {
158 	WARN_ON(fsvq->in_flight <= 0);
159 	fsvq->in_flight--;
160 	if (!fsvq->in_flight)
161 		complete(&fsvq->in_flight_zero);
162 }
163 
164 static void release_virtio_fs_obj(struct kref *ref)
165 {
166 	struct virtio_fs *vfs = container_of(ref, struct virtio_fs, refcount);
167 
168 	kfree(vfs->vqs);
169 	kfree(vfs);
170 }
171 
172 /* Make sure virtiofs_mutex is held */
173 static void virtio_fs_put(struct virtio_fs *fs)
174 {
175 	kref_put(&fs->refcount, release_virtio_fs_obj);
176 }
177 
178 static void virtio_fs_fiq_release(struct fuse_iqueue *fiq)
179 {
180 	struct virtio_fs *vfs = fiq->priv;
181 
182 	mutex_lock(&virtio_fs_mutex);
183 	virtio_fs_put(vfs);
184 	mutex_unlock(&virtio_fs_mutex);
185 }
186 
187 static void virtio_fs_drain_queue(struct virtio_fs_vq *fsvq)
188 {
189 	WARN_ON(fsvq->in_flight < 0);
190 
191 	/* Wait for in flight requests to finish.*/
192 	spin_lock(&fsvq->lock);
193 	if (fsvq->in_flight) {
194 		/* We are holding virtio_fs_mutex. There should not be any
195 		 * waiters waiting for completion.
196 		 */
197 		reinit_completion(&fsvq->in_flight_zero);
198 		spin_unlock(&fsvq->lock);
199 		wait_for_completion(&fsvq->in_flight_zero);
200 	} else {
201 		spin_unlock(&fsvq->lock);
202 	}
203 
204 	flush_work(&fsvq->done_work);
205 	flush_delayed_work(&fsvq->dispatch_work);
206 }
207 
208 static void virtio_fs_drain_all_queues_locked(struct virtio_fs *fs)
209 {
210 	struct virtio_fs_vq *fsvq;
211 	int i;
212 
213 	for (i = 0; i < fs->nvqs; i++) {
214 		fsvq = &fs->vqs[i];
215 		virtio_fs_drain_queue(fsvq);
216 	}
217 }
218 
219 static void virtio_fs_drain_all_queues(struct virtio_fs *fs)
220 {
221 	/* Provides mutual exclusion between ->remove and ->kill_sb
222 	 * paths. We don't want both of these draining queue at the
223 	 * same time. Current completion logic reinits completion
224 	 * and that means there should not be any other thread
225 	 * doing reinit or waiting for completion already.
226 	 */
227 	mutex_lock(&virtio_fs_mutex);
228 	virtio_fs_drain_all_queues_locked(fs);
229 	mutex_unlock(&virtio_fs_mutex);
230 }
231 
232 static void virtio_fs_start_all_queues(struct virtio_fs *fs)
233 {
234 	struct virtio_fs_vq *fsvq;
235 	int i;
236 
237 	for (i = 0; i < fs->nvqs; i++) {
238 		fsvq = &fs->vqs[i];
239 		spin_lock(&fsvq->lock);
240 		fsvq->connected = true;
241 		spin_unlock(&fsvq->lock);
242 	}
243 }
244 
245 /* Add a new instance to the list or return -EEXIST if tag name exists*/
246 static int virtio_fs_add_instance(struct virtio_fs *fs)
247 {
248 	struct virtio_fs *fs2;
249 	bool duplicate = false;
250 
251 	mutex_lock(&virtio_fs_mutex);
252 
253 	list_for_each_entry(fs2, &virtio_fs_instances, list) {
254 		if (strcmp(fs->tag, fs2->tag) == 0)
255 			duplicate = true;
256 	}
257 
258 	if (!duplicate)
259 		list_add_tail(&fs->list, &virtio_fs_instances);
260 
261 	mutex_unlock(&virtio_fs_mutex);
262 
263 	if (duplicate)
264 		return -EEXIST;
265 	return 0;
266 }
267 
268 /* Return the virtio_fs with a given tag, or NULL */
269 static struct virtio_fs *virtio_fs_find_instance(const char *tag)
270 {
271 	struct virtio_fs *fs;
272 
273 	mutex_lock(&virtio_fs_mutex);
274 
275 	list_for_each_entry(fs, &virtio_fs_instances, list) {
276 		if (strcmp(fs->tag, tag) == 0) {
277 			kref_get(&fs->refcount);
278 			goto found;
279 		}
280 	}
281 
282 	fs = NULL; /* not found */
283 
284 found:
285 	mutex_unlock(&virtio_fs_mutex);
286 
287 	return fs;
288 }
289 
290 static void virtio_fs_free_devs(struct virtio_fs *fs)
291 {
292 	unsigned int i;
293 
294 	for (i = 0; i < fs->nvqs; i++) {
295 		struct virtio_fs_vq *fsvq = &fs->vqs[i];
296 
297 		if (!fsvq->fud)
298 			continue;
299 
300 		fuse_dev_free(fsvq->fud);
301 		fsvq->fud = NULL;
302 	}
303 }
304 
305 /* Read filesystem name from virtio config into fs->tag (must kfree()). */
306 static int virtio_fs_read_tag(struct virtio_device *vdev, struct virtio_fs *fs)
307 {
308 	char tag_buf[sizeof_field(struct virtio_fs_config, tag)];
309 	char *end;
310 	size_t len;
311 
312 	virtio_cread_bytes(vdev, offsetof(struct virtio_fs_config, tag),
313 			   &tag_buf, sizeof(tag_buf));
314 	end = memchr(tag_buf, '\0', sizeof(tag_buf));
315 	if (end == tag_buf)
316 		return -EINVAL; /* empty tag */
317 	if (!end)
318 		end = &tag_buf[sizeof(tag_buf)];
319 
320 	len = end - tag_buf;
321 	fs->tag = devm_kmalloc(&vdev->dev, len + 1, GFP_KERNEL);
322 	if (!fs->tag)
323 		return -ENOMEM;
324 	memcpy(fs->tag, tag_buf, len);
325 	fs->tag[len] = '\0';
326 	return 0;
327 }
328 
329 /* Work function for hiprio completion */
330 static void virtio_fs_hiprio_done_work(struct work_struct *work)
331 {
332 	struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq,
333 						 done_work);
334 	struct virtqueue *vq = fsvq->vq;
335 
336 	/* Free completed FUSE_FORGET requests */
337 	spin_lock(&fsvq->lock);
338 	do {
339 		unsigned int len;
340 		void *req;
341 
342 		virtqueue_disable_cb(vq);
343 
344 		while ((req = virtqueue_get_buf(vq, &len)) != NULL) {
345 			kfree(req);
346 			dec_in_flight_req(fsvq);
347 		}
348 	} while (!virtqueue_enable_cb(vq) && likely(!virtqueue_is_broken(vq)));
349 	spin_unlock(&fsvq->lock);
350 }
351 
352 static void virtio_fs_request_dispatch_work(struct work_struct *work)
353 {
354 	struct fuse_req *req;
355 	struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq,
356 						 dispatch_work.work);
357 	int ret;
358 
359 	pr_debug("virtio-fs: worker %s called.\n", __func__);
360 	while (1) {
361 		spin_lock(&fsvq->lock);
362 		req = list_first_entry_or_null(&fsvq->end_reqs, struct fuse_req,
363 					       list);
364 		if (!req) {
365 			spin_unlock(&fsvq->lock);
366 			break;
367 		}
368 
369 		list_del_init(&req->list);
370 		spin_unlock(&fsvq->lock);
371 		fuse_request_end(req);
372 	}
373 
374 	/* Dispatch pending requests */
375 	while (1) {
376 		spin_lock(&fsvq->lock);
377 		req = list_first_entry_or_null(&fsvq->queued_reqs,
378 					       struct fuse_req, list);
379 		if (!req) {
380 			spin_unlock(&fsvq->lock);
381 			return;
382 		}
383 		list_del_init(&req->list);
384 		spin_unlock(&fsvq->lock);
385 
386 		ret = virtio_fs_enqueue_req(fsvq, req, true);
387 		if (ret < 0) {
388 			if (ret == -ENOMEM || ret == -ENOSPC) {
389 				spin_lock(&fsvq->lock);
390 				list_add_tail(&req->list, &fsvq->queued_reqs);
391 				schedule_delayed_work(&fsvq->dispatch_work,
392 						      msecs_to_jiffies(1));
393 				spin_unlock(&fsvq->lock);
394 				return;
395 			}
396 			req->out.h.error = ret;
397 			spin_lock(&fsvq->lock);
398 			dec_in_flight_req(fsvq);
399 			spin_unlock(&fsvq->lock);
400 			pr_err("virtio-fs: virtio_fs_enqueue_req() failed %d\n",
401 			       ret);
402 			fuse_request_end(req);
403 		}
404 	}
405 }
406 
407 /*
408  * Returns 1 if queue is full and sender should wait a bit before sending
409  * next request, 0 otherwise.
410  */
411 static int send_forget_request(struct virtio_fs_vq *fsvq,
412 			       struct virtio_fs_forget *forget,
413 			       bool in_flight)
414 {
415 	struct scatterlist sg;
416 	struct virtqueue *vq;
417 	int ret = 0;
418 	bool notify;
419 	struct virtio_fs_forget_req *req = &forget->req;
420 
421 	spin_lock(&fsvq->lock);
422 	if (!fsvq->connected) {
423 		if (in_flight)
424 			dec_in_flight_req(fsvq);
425 		kfree(forget);
426 		goto out;
427 	}
428 
429 	sg_init_one(&sg, req, sizeof(*req));
430 	vq = fsvq->vq;
431 	dev_dbg(&vq->vdev->dev, "%s\n", __func__);
432 
433 	ret = virtqueue_add_outbuf(vq, &sg, 1, forget, GFP_ATOMIC);
434 	if (ret < 0) {
435 		if (ret == -ENOMEM || ret == -ENOSPC) {
436 			pr_debug("virtio-fs: Could not queue FORGET: err=%d. Will try later\n",
437 				 ret);
438 			list_add_tail(&forget->list, &fsvq->queued_reqs);
439 			schedule_delayed_work(&fsvq->dispatch_work,
440 					      msecs_to_jiffies(1));
441 			if (!in_flight)
442 				inc_in_flight_req(fsvq);
443 			/* Queue is full */
444 			ret = 1;
445 		} else {
446 			pr_debug("virtio-fs: Could not queue FORGET: err=%d. Dropping it.\n",
447 				 ret);
448 			kfree(forget);
449 			if (in_flight)
450 				dec_in_flight_req(fsvq);
451 		}
452 		goto out;
453 	}
454 
455 	if (!in_flight)
456 		inc_in_flight_req(fsvq);
457 	notify = virtqueue_kick_prepare(vq);
458 	spin_unlock(&fsvq->lock);
459 
460 	if (notify)
461 		virtqueue_notify(vq);
462 	return ret;
463 out:
464 	spin_unlock(&fsvq->lock);
465 	return ret;
466 }
467 
468 static void virtio_fs_hiprio_dispatch_work(struct work_struct *work)
469 {
470 	struct virtio_fs_forget *forget;
471 	struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq,
472 						 dispatch_work.work);
473 	pr_debug("virtio-fs: worker %s called.\n", __func__);
474 	while (1) {
475 		spin_lock(&fsvq->lock);
476 		forget = list_first_entry_or_null(&fsvq->queued_reqs,
477 					struct virtio_fs_forget, list);
478 		if (!forget) {
479 			spin_unlock(&fsvq->lock);
480 			return;
481 		}
482 
483 		list_del(&forget->list);
484 		spin_unlock(&fsvq->lock);
485 		if (send_forget_request(fsvq, forget, true))
486 			return;
487 	}
488 }
489 
490 /* Allocate and copy args into req->argbuf */
491 static int copy_args_to_argbuf(struct fuse_req *req)
492 {
493 	struct fuse_args *args = req->args;
494 	unsigned int offset = 0;
495 	unsigned int num_in;
496 	unsigned int num_out;
497 	unsigned int len;
498 	unsigned int i;
499 
500 	num_in = args->in_numargs - args->in_pages;
501 	num_out = args->out_numargs - args->out_pages;
502 	len = fuse_len_args(num_in, (struct fuse_arg *) args->in_args) +
503 	      fuse_len_args(num_out, args->out_args);
504 
505 	req->argbuf = kmalloc(len, GFP_ATOMIC);
506 	if (!req->argbuf)
507 		return -ENOMEM;
508 
509 	for (i = 0; i < num_in; i++) {
510 		memcpy(req->argbuf + offset,
511 		       args->in_args[i].value,
512 		       args->in_args[i].size);
513 		offset += args->in_args[i].size;
514 	}
515 
516 	return 0;
517 }
518 
519 /* Copy args out of and free req->argbuf */
520 static void copy_args_from_argbuf(struct fuse_args *args, struct fuse_req *req)
521 {
522 	unsigned int remaining;
523 	unsigned int offset;
524 	unsigned int num_in;
525 	unsigned int num_out;
526 	unsigned int i;
527 
528 	remaining = req->out.h.len - sizeof(req->out.h);
529 	num_in = args->in_numargs - args->in_pages;
530 	num_out = args->out_numargs - args->out_pages;
531 	offset = fuse_len_args(num_in, (struct fuse_arg *)args->in_args);
532 
533 	for (i = 0; i < num_out; i++) {
534 		unsigned int argsize = args->out_args[i].size;
535 
536 		if (args->out_argvar &&
537 		    i == args->out_numargs - 1 &&
538 		    argsize > remaining) {
539 			argsize = remaining;
540 		}
541 
542 		memcpy(args->out_args[i].value, req->argbuf + offset, argsize);
543 		offset += argsize;
544 
545 		if (i != args->out_numargs - 1)
546 			remaining -= argsize;
547 	}
548 
549 	/* Store the actual size of the variable-length arg */
550 	if (args->out_argvar)
551 		args->out_args[args->out_numargs - 1].size = remaining;
552 
553 	kfree(req->argbuf);
554 	req->argbuf = NULL;
555 }
556 
557 /* Work function for request completion */
558 static void virtio_fs_request_complete(struct fuse_req *req,
559 				       struct virtio_fs_vq *fsvq)
560 {
561 	struct fuse_pqueue *fpq = &fsvq->fud->pq;
562 	struct fuse_args *args;
563 	struct fuse_args_pages *ap;
564 	unsigned int len, i, thislen;
565 	struct page *page;
566 
567 	/*
568 	 * TODO verify that server properly follows FUSE protocol
569 	 * (oh.uniq, oh.len)
570 	 */
571 	args = req->args;
572 	copy_args_from_argbuf(args, req);
573 
574 	if (args->out_pages && args->page_zeroing) {
575 		len = args->out_args[args->out_numargs - 1].size;
576 		ap = container_of(args, typeof(*ap), args);
577 		for (i = 0; i < ap->num_pages; i++) {
578 			thislen = ap->descs[i].length;
579 			if (len < thislen) {
580 				WARN_ON(ap->descs[i].offset);
581 				page = ap->pages[i];
582 				zero_user_segment(page, len, thislen);
583 				len = 0;
584 			} else {
585 				len -= thislen;
586 			}
587 		}
588 	}
589 
590 	spin_lock(&fpq->lock);
591 	clear_bit(FR_SENT, &req->flags);
592 	spin_unlock(&fpq->lock);
593 
594 	fuse_request_end(req);
595 	spin_lock(&fsvq->lock);
596 	dec_in_flight_req(fsvq);
597 	spin_unlock(&fsvq->lock);
598 }
599 
600 static void virtio_fs_complete_req_work(struct work_struct *work)
601 {
602 	struct virtio_fs_req_work *w =
603 		container_of(work, typeof(*w), done_work);
604 
605 	virtio_fs_request_complete(w->req, w->fsvq);
606 	kfree(w);
607 }
608 
609 static void virtio_fs_requests_done_work(struct work_struct *work)
610 {
611 	struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq,
612 						 done_work);
613 	struct fuse_pqueue *fpq = &fsvq->fud->pq;
614 	struct virtqueue *vq = fsvq->vq;
615 	struct fuse_req *req;
616 	struct fuse_req *next;
617 	unsigned int len;
618 	LIST_HEAD(reqs);
619 
620 	/* Collect completed requests off the virtqueue */
621 	spin_lock(&fsvq->lock);
622 	do {
623 		virtqueue_disable_cb(vq);
624 
625 		while ((req = virtqueue_get_buf(vq, &len)) != NULL) {
626 			spin_lock(&fpq->lock);
627 			list_move_tail(&req->list, &reqs);
628 			spin_unlock(&fpq->lock);
629 		}
630 	} while (!virtqueue_enable_cb(vq) && likely(!virtqueue_is_broken(vq)));
631 	spin_unlock(&fsvq->lock);
632 
633 	/* End requests */
634 	list_for_each_entry_safe(req, next, &reqs, list) {
635 		list_del_init(&req->list);
636 
637 		/* blocking async request completes in a worker context */
638 		if (req->args->may_block) {
639 			struct virtio_fs_req_work *w;
640 
641 			w = kzalloc(sizeof(*w), GFP_NOFS | __GFP_NOFAIL);
642 			INIT_WORK(&w->done_work, virtio_fs_complete_req_work);
643 			w->fsvq = fsvq;
644 			w->req = req;
645 			schedule_work(&w->done_work);
646 		} else {
647 			virtio_fs_request_complete(req, fsvq);
648 		}
649 	}
650 }
651 
652 /* Virtqueue interrupt handler */
653 static void virtio_fs_vq_done(struct virtqueue *vq)
654 {
655 	struct virtio_fs_vq *fsvq = vq_to_fsvq(vq);
656 
657 	dev_dbg(&vq->vdev->dev, "%s %s\n", __func__, fsvq->name);
658 
659 	schedule_work(&fsvq->done_work);
660 }
661 
662 static void virtio_fs_init_vq(struct virtio_fs_vq *fsvq, char *name,
663 			      int vq_type)
664 {
665 	strscpy(fsvq->name, name, VQ_NAME_LEN);
666 	spin_lock_init(&fsvq->lock);
667 	INIT_LIST_HEAD(&fsvq->queued_reqs);
668 	INIT_LIST_HEAD(&fsvq->end_reqs);
669 	init_completion(&fsvq->in_flight_zero);
670 
671 	if (vq_type == VQ_REQUEST) {
672 		INIT_WORK(&fsvq->done_work, virtio_fs_requests_done_work);
673 		INIT_DELAYED_WORK(&fsvq->dispatch_work,
674 				  virtio_fs_request_dispatch_work);
675 	} else {
676 		INIT_WORK(&fsvq->done_work, virtio_fs_hiprio_done_work);
677 		INIT_DELAYED_WORK(&fsvq->dispatch_work,
678 				  virtio_fs_hiprio_dispatch_work);
679 	}
680 }
681 
682 /* Initialize virtqueues */
683 static int virtio_fs_setup_vqs(struct virtio_device *vdev,
684 			       struct virtio_fs *fs)
685 {
686 	struct virtqueue **vqs;
687 	vq_callback_t **callbacks;
688 	const char **names;
689 	unsigned int i;
690 	int ret = 0;
691 
692 	virtio_cread_le(vdev, struct virtio_fs_config, num_request_queues,
693 			&fs->num_request_queues);
694 	if (fs->num_request_queues == 0)
695 		return -EINVAL;
696 
697 	fs->nvqs = VQ_REQUEST + fs->num_request_queues;
698 	fs->vqs = kcalloc(fs->nvqs, sizeof(fs->vqs[VQ_HIPRIO]), GFP_KERNEL);
699 	if (!fs->vqs)
700 		return -ENOMEM;
701 
702 	vqs = kmalloc_array(fs->nvqs, sizeof(vqs[VQ_HIPRIO]), GFP_KERNEL);
703 	callbacks = kmalloc_array(fs->nvqs, sizeof(callbacks[VQ_HIPRIO]),
704 					GFP_KERNEL);
705 	names = kmalloc_array(fs->nvqs, sizeof(names[VQ_HIPRIO]), GFP_KERNEL);
706 	if (!vqs || !callbacks || !names) {
707 		ret = -ENOMEM;
708 		goto out;
709 	}
710 
711 	/* Initialize the hiprio/forget request virtqueue */
712 	callbacks[VQ_HIPRIO] = virtio_fs_vq_done;
713 	virtio_fs_init_vq(&fs->vqs[VQ_HIPRIO], "hiprio", VQ_HIPRIO);
714 	names[VQ_HIPRIO] = fs->vqs[VQ_HIPRIO].name;
715 
716 	/* Initialize the requests virtqueues */
717 	for (i = VQ_REQUEST; i < fs->nvqs; i++) {
718 		char vq_name[VQ_NAME_LEN];
719 
720 		snprintf(vq_name, VQ_NAME_LEN, "requests.%u", i - VQ_REQUEST);
721 		virtio_fs_init_vq(&fs->vqs[i], vq_name, VQ_REQUEST);
722 		callbacks[i] = virtio_fs_vq_done;
723 		names[i] = fs->vqs[i].name;
724 	}
725 
726 	ret = virtio_find_vqs(vdev, fs->nvqs, vqs, callbacks, names, NULL);
727 	if (ret < 0)
728 		goto out;
729 
730 	for (i = 0; i < fs->nvqs; i++)
731 		fs->vqs[i].vq = vqs[i];
732 
733 	virtio_fs_start_all_queues(fs);
734 out:
735 	kfree(names);
736 	kfree(callbacks);
737 	kfree(vqs);
738 	if (ret)
739 		kfree(fs->vqs);
740 	return ret;
741 }
742 
743 /* Free virtqueues (device must already be reset) */
744 static void virtio_fs_cleanup_vqs(struct virtio_device *vdev,
745 				  struct virtio_fs *fs)
746 {
747 	vdev->config->del_vqs(vdev);
748 }
749 
750 /* Map a window offset to a page frame number.  The window offset will have
751  * been produced by .iomap_begin(), which maps a file offset to a window
752  * offset.
753  */
754 static long virtio_fs_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
755 				    long nr_pages, void **kaddr, pfn_t *pfn)
756 {
757 	struct virtio_fs *fs = dax_get_private(dax_dev);
758 	phys_addr_t offset = PFN_PHYS(pgoff);
759 	size_t max_nr_pages = fs->window_len/PAGE_SIZE - pgoff;
760 
761 	if (kaddr)
762 		*kaddr = fs->window_kaddr + offset;
763 	if (pfn)
764 		*pfn = phys_to_pfn_t(fs->window_phys_addr + offset,
765 					PFN_DEV | PFN_MAP);
766 	return nr_pages > max_nr_pages ? max_nr_pages : nr_pages;
767 }
768 
769 static int virtio_fs_zero_page_range(struct dax_device *dax_dev,
770 				     pgoff_t pgoff, size_t nr_pages)
771 {
772 	long rc;
773 	void *kaddr;
774 
775 	rc = dax_direct_access(dax_dev, pgoff, nr_pages, &kaddr, NULL);
776 	if (rc < 0)
777 		return rc;
778 	memset(kaddr, 0, nr_pages << PAGE_SHIFT);
779 	dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT);
780 	return 0;
781 }
782 
783 static const struct dax_operations virtio_fs_dax_ops = {
784 	.direct_access = virtio_fs_direct_access,
785 	.zero_page_range = virtio_fs_zero_page_range,
786 };
787 
788 static void virtio_fs_cleanup_dax(void *data)
789 {
790 	struct dax_device *dax_dev = data;
791 
792 	kill_dax(dax_dev);
793 	put_dax(dax_dev);
794 }
795 
796 static int virtio_fs_setup_dax(struct virtio_device *vdev, struct virtio_fs *fs)
797 {
798 	struct virtio_shm_region cache_reg;
799 	struct dev_pagemap *pgmap;
800 	bool have_cache;
801 
802 	if (!IS_ENABLED(CONFIG_FUSE_DAX))
803 		return 0;
804 
805 	/* Get cache region */
806 	have_cache = virtio_get_shm_region(vdev, &cache_reg,
807 					   (u8)VIRTIO_FS_SHMCAP_ID_CACHE);
808 	if (!have_cache) {
809 		dev_notice(&vdev->dev, "%s: No cache capability\n", __func__);
810 		return 0;
811 	}
812 
813 	if (!devm_request_mem_region(&vdev->dev, cache_reg.addr, cache_reg.len,
814 				     dev_name(&vdev->dev))) {
815 		dev_warn(&vdev->dev, "could not reserve region addr=0x%llx len=0x%llx\n",
816 			 cache_reg.addr, cache_reg.len);
817 		return -EBUSY;
818 	}
819 
820 	dev_notice(&vdev->dev, "Cache len: 0x%llx @ 0x%llx\n", cache_reg.len,
821 		   cache_reg.addr);
822 
823 	pgmap = devm_kzalloc(&vdev->dev, sizeof(*pgmap), GFP_KERNEL);
824 	if (!pgmap)
825 		return -ENOMEM;
826 
827 	pgmap->type = MEMORY_DEVICE_FS_DAX;
828 
829 	/* Ideally we would directly use the PCI BAR resource but
830 	 * devm_memremap_pages() wants its own copy in pgmap.  So
831 	 * initialize a struct resource from scratch (only the start
832 	 * and end fields will be used).
833 	 */
834 	pgmap->range = (struct range) {
835 		.start = (phys_addr_t) cache_reg.addr,
836 		.end = (phys_addr_t) cache_reg.addr + cache_reg.len - 1,
837 	};
838 	pgmap->nr_range = 1;
839 
840 	fs->window_kaddr = devm_memremap_pages(&vdev->dev, pgmap);
841 	if (IS_ERR(fs->window_kaddr))
842 		return PTR_ERR(fs->window_kaddr);
843 
844 	fs->window_phys_addr = (phys_addr_t) cache_reg.addr;
845 	fs->window_len = (phys_addr_t) cache_reg.len;
846 
847 	dev_dbg(&vdev->dev, "%s: window kaddr 0x%px phys_addr 0x%llx len 0x%llx\n",
848 		__func__, fs->window_kaddr, cache_reg.addr, cache_reg.len);
849 
850 	fs->dax_dev = alloc_dax(fs, &virtio_fs_dax_ops);
851 	if (IS_ERR(fs->dax_dev))
852 		return PTR_ERR(fs->dax_dev);
853 
854 	return devm_add_action_or_reset(&vdev->dev, virtio_fs_cleanup_dax,
855 					fs->dax_dev);
856 }
857 
858 static int virtio_fs_probe(struct virtio_device *vdev)
859 {
860 	struct virtio_fs *fs;
861 	int ret;
862 
863 	fs = kzalloc(sizeof(*fs), GFP_KERNEL);
864 	if (!fs)
865 		return -ENOMEM;
866 	kref_init(&fs->refcount);
867 	vdev->priv = fs;
868 
869 	ret = virtio_fs_read_tag(vdev, fs);
870 	if (ret < 0)
871 		goto out;
872 
873 	ret = virtio_fs_setup_vqs(vdev, fs);
874 	if (ret < 0)
875 		goto out;
876 
877 	/* TODO vq affinity */
878 
879 	ret = virtio_fs_setup_dax(vdev, fs);
880 	if (ret < 0)
881 		goto out_vqs;
882 
883 	/* Bring the device online in case the filesystem is mounted and
884 	 * requests need to be sent before we return.
885 	 */
886 	virtio_device_ready(vdev);
887 
888 	ret = virtio_fs_add_instance(fs);
889 	if (ret < 0)
890 		goto out_vqs;
891 
892 	return 0;
893 
894 out_vqs:
895 	virtio_reset_device(vdev);
896 	virtio_fs_cleanup_vqs(vdev, fs);
897 	kfree(fs->vqs);
898 
899 out:
900 	vdev->priv = NULL;
901 	kfree(fs);
902 	return ret;
903 }
904 
905 static void virtio_fs_stop_all_queues(struct virtio_fs *fs)
906 {
907 	struct virtio_fs_vq *fsvq;
908 	int i;
909 
910 	for (i = 0; i < fs->nvqs; i++) {
911 		fsvq = &fs->vqs[i];
912 		spin_lock(&fsvq->lock);
913 		fsvq->connected = false;
914 		spin_unlock(&fsvq->lock);
915 	}
916 }
917 
918 static void virtio_fs_remove(struct virtio_device *vdev)
919 {
920 	struct virtio_fs *fs = vdev->priv;
921 
922 	mutex_lock(&virtio_fs_mutex);
923 	/* This device is going away. No one should get new reference */
924 	list_del_init(&fs->list);
925 	virtio_fs_stop_all_queues(fs);
926 	virtio_fs_drain_all_queues_locked(fs);
927 	virtio_reset_device(vdev);
928 	virtio_fs_cleanup_vqs(vdev, fs);
929 
930 	vdev->priv = NULL;
931 	/* Put device reference on virtio_fs object */
932 	virtio_fs_put(fs);
933 	mutex_unlock(&virtio_fs_mutex);
934 }
935 
936 #ifdef CONFIG_PM_SLEEP
937 static int virtio_fs_freeze(struct virtio_device *vdev)
938 {
939 	/* TODO need to save state here */
940 	pr_warn("virtio-fs: suspend/resume not yet supported\n");
941 	return -EOPNOTSUPP;
942 }
943 
944 static int virtio_fs_restore(struct virtio_device *vdev)
945 {
946 	 /* TODO need to restore state here */
947 	return 0;
948 }
949 #endif /* CONFIG_PM_SLEEP */
950 
951 static const struct virtio_device_id id_table[] = {
952 	{ VIRTIO_ID_FS, VIRTIO_DEV_ANY_ID },
953 	{},
954 };
955 
956 static const unsigned int feature_table[] = {};
957 
958 static struct virtio_driver virtio_fs_driver = {
959 	.driver.name		= KBUILD_MODNAME,
960 	.driver.owner		= THIS_MODULE,
961 	.id_table		= id_table,
962 	.feature_table		= feature_table,
963 	.feature_table_size	= ARRAY_SIZE(feature_table),
964 	.probe			= virtio_fs_probe,
965 	.remove			= virtio_fs_remove,
966 #ifdef CONFIG_PM_SLEEP
967 	.freeze			= virtio_fs_freeze,
968 	.restore		= virtio_fs_restore,
969 #endif
970 };
971 
972 static void virtio_fs_wake_forget_and_unlock(struct fuse_iqueue *fiq)
973 __releases(fiq->lock)
974 {
975 	struct fuse_forget_link *link;
976 	struct virtio_fs_forget *forget;
977 	struct virtio_fs_forget_req *req;
978 	struct virtio_fs *fs;
979 	struct virtio_fs_vq *fsvq;
980 	u64 unique;
981 
982 	link = fuse_dequeue_forget(fiq, 1, NULL);
983 	unique = fuse_get_unique(fiq);
984 
985 	fs = fiq->priv;
986 	fsvq = &fs->vqs[VQ_HIPRIO];
987 	spin_unlock(&fiq->lock);
988 
989 	/* Allocate a buffer for the request */
990 	forget = kmalloc(sizeof(*forget), GFP_NOFS | __GFP_NOFAIL);
991 	req = &forget->req;
992 
993 	req->ih = (struct fuse_in_header){
994 		.opcode = FUSE_FORGET,
995 		.nodeid = link->forget_one.nodeid,
996 		.unique = unique,
997 		.len = sizeof(*req),
998 	};
999 	req->arg = (struct fuse_forget_in){
1000 		.nlookup = link->forget_one.nlookup,
1001 	};
1002 
1003 	send_forget_request(fsvq, forget, false);
1004 	kfree(link);
1005 }
1006 
1007 static void virtio_fs_wake_interrupt_and_unlock(struct fuse_iqueue *fiq)
1008 __releases(fiq->lock)
1009 {
1010 	/*
1011 	 * TODO interrupts.
1012 	 *
1013 	 * Normal fs operations on a local filesystems aren't interruptible.
1014 	 * Exceptions are blocking lock operations; for example fcntl(F_SETLKW)
1015 	 * with shared lock between host and guest.
1016 	 */
1017 	spin_unlock(&fiq->lock);
1018 }
1019 
1020 /* Count number of scatter-gather elements required */
1021 static unsigned int sg_count_fuse_pages(struct fuse_page_desc *page_descs,
1022 				       unsigned int num_pages,
1023 				       unsigned int total_len)
1024 {
1025 	unsigned int i;
1026 	unsigned int this_len;
1027 
1028 	for (i = 0; i < num_pages && total_len; i++) {
1029 		this_len =  min(page_descs[i].length, total_len);
1030 		total_len -= this_len;
1031 	}
1032 
1033 	return i;
1034 }
1035 
1036 /* Return the number of scatter-gather list elements required */
1037 static unsigned int sg_count_fuse_req(struct fuse_req *req)
1038 {
1039 	struct fuse_args *args = req->args;
1040 	struct fuse_args_pages *ap = container_of(args, typeof(*ap), args);
1041 	unsigned int size, total_sgs = 1 /* fuse_in_header */;
1042 
1043 	if (args->in_numargs - args->in_pages)
1044 		total_sgs += 1;
1045 
1046 	if (args->in_pages) {
1047 		size = args->in_args[args->in_numargs - 1].size;
1048 		total_sgs += sg_count_fuse_pages(ap->descs, ap->num_pages,
1049 						 size);
1050 	}
1051 
1052 	if (!test_bit(FR_ISREPLY, &req->flags))
1053 		return total_sgs;
1054 
1055 	total_sgs += 1 /* fuse_out_header */;
1056 
1057 	if (args->out_numargs - args->out_pages)
1058 		total_sgs += 1;
1059 
1060 	if (args->out_pages) {
1061 		size = args->out_args[args->out_numargs - 1].size;
1062 		total_sgs += sg_count_fuse_pages(ap->descs, ap->num_pages,
1063 						 size);
1064 	}
1065 
1066 	return total_sgs;
1067 }
1068 
1069 /* Add pages to scatter-gather list and return number of elements used */
1070 static unsigned int sg_init_fuse_pages(struct scatterlist *sg,
1071 				       struct page **pages,
1072 				       struct fuse_page_desc *page_descs,
1073 				       unsigned int num_pages,
1074 				       unsigned int total_len)
1075 {
1076 	unsigned int i;
1077 	unsigned int this_len;
1078 
1079 	for (i = 0; i < num_pages && total_len; i++) {
1080 		sg_init_table(&sg[i], 1);
1081 		this_len =  min(page_descs[i].length, total_len);
1082 		sg_set_page(&sg[i], pages[i], this_len, page_descs[i].offset);
1083 		total_len -= this_len;
1084 	}
1085 
1086 	return i;
1087 }
1088 
1089 /* Add args to scatter-gather list and return number of elements used */
1090 static unsigned int sg_init_fuse_args(struct scatterlist *sg,
1091 				      struct fuse_req *req,
1092 				      struct fuse_arg *args,
1093 				      unsigned int numargs,
1094 				      bool argpages,
1095 				      void *argbuf,
1096 				      unsigned int *len_used)
1097 {
1098 	struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args);
1099 	unsigned int total_sgs = 0;
1100 	unsigned int len;
1101 
1102 	len = fuse_len_args(numargs - argpages, args);
1103 	if (len)
1104 		sg_init_one(&sg[total_sgs++], argbuf, len);
1105 
1106 	if (argpages)
1107 		total_sgs += sg_init_fuse_pages(&sg[total_sgs],
1108 						ap->pages, ap->descs,
1109 						ap->num_pages,
1110 						args[numargs - 1].size);
1111 
1112 	if (len_used)
1113 		*len_used = len;
1114 
1115 	return total_sgs;
1116 }
1117 
1118 /* Add a request to a virtqueue and kick the device */
1119 static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq,
1120 				 struct fuse_req *req, bool in_flight)
1121 {
1122 	/* requests need at least 4 elements */
1123 	struct scatterlist *stack_sgs[6];
1124 	struct scatterlist stack_sg[ARRAY_SIZE(stack_sgs)];
1125 	struct scatterlist **sgs = stack_sgs;
1126 	struct scatterlist *sg = stack_sg;
1127 	struct virtqueue *vq;
1128 	struct fuse_args *args = req->args;
1129 	unsigned int argbuf_used = 0;
1130 	unsigned int out_sgs = 0;
1131 	unsigned int in_sgs = 0;
1132 	unsigned int total_sgs;
1133 	unsigned int i;
1134 	int ret;
1135 	bool notify;
1136 	struct fuse_pqueue *fpq;
1137 
1138 	/* Does the sglist fit on the stack? */
1139 	total_sgs = sg_count_fuse_req(req);
1140 	if (total_sgs > ARRAY_SIZE(stack_sgs)) {
1141 		sgs = kmalloc_array(total_sgs, sizeof(sgs[0]), GFP_ATOMIC);
1142 		sg = kmalloc_array(total_sgs, sizeof(sg[0]), GFP_ATOMIC);
1143 		if (!sgs || !sg) {
1144 			ret = -ENOMEM;
1145 			goto out;
1146 		}
1147 	}
1148 
1149 	/* Use a bounce buffer since stack args cannot be mapped */
1150 	ret = copy_args_to_argbuf(req);
1151 	if (ret < 0)
1152 		goto out;
1153 
1154 	/* Request elements */
1155 	sg_init_one(&sg[out_sgs++], &req->in.h, sizeof(req->in.h));
1156 	out_sgs += sg_init_fuse_args(&sg[out_sgs], req,
1157 				     (struct fuse_arg *)args->in_args,
1158 				     args->in_numargs, args->in_pages,
1159 				     req->argbuf, &argbuf_used);
1160 
1161 	/* Reply elements */
1162 	if (test_bit(FR_ISREPLY, &req->flags)) {
1163 		sg_init_one(&sg[out_sgs + in_sgs++],
1164 			    &req->out.h, sizeof(req->out.h));
1165 		in_sgs += sg_init_fuse_args(&sg[out_sgs + in_sgs], req,
1166 					    args->out_args, args->out_numargs,
1167 					    args->out_pages,
1168 					    req->argbuf + argbuf_used, NULL);
1169 	}
1170 
1171 	WARN_ON(out_sgs + in_sgs != total_sgs);
1172 
1173 	for (i = 0; i < total_sgs; i++)
1174 		sgs[i] = &sg[i];
1175 
1176 	spin_lock(&fsvq->lock);
1177 
1178 	if (!fsvq->connected) {
1179 		spin_unlock(&fsvq->lock);
1180 		ret = -ENOTCONN;
1181 		goto out;
1182 	}
1183 
1184 	vq = fsvq->vq;
1185 	ret = virtqueue_add_sgs(vq, sgs, out_sgs, in_sgs, req, GFP_ATOMIC);
1186 	if (ret < 0) {
1187 		spin_unlock(&fsvq->lock);
1188 		goto out;
1189 	}
1190 
1191 	/* Request successfully sent. */
1192 	fpq = &fsvq->fud->pq;
1193 	spin_lock(&fpq->lock);
1194 	list_add_tail(&req->list, fpq->processing);
1195 	spin_unlock(&fpq->lock);
1196 	set_bit(FR_SENT, &req->flags);
1197 	/* matches barrier in request_wait_answer() */
1198 	smp_mb__after_atomic();
1199 
1200 	if (!in_flight)
1201 		inc_in_flight_req(fsvq);
1202 	notify = virtqueue_kick_prepare(vq);
1203 
1204 	spin_unlock(&fsvq->lock);
1205 
1206 	if (notify)
1207 		virtqueue_notify(vq);
1208 
1209 out:
1210 	if (ret < 0 && req->argbuf) {
1211 		kfree(req->argbuf);
1212 		req->argbuf = NULL;
1213 	}
1214 	if (sgs != stack_sgs) {
1215 		kfree(sgs);
1216 		kfree(sg);
1217 	}
1218 
1219 	return ret;
1220 }
1221 
1222 static void virtio_fs_wake_pending_and_unlock(struct fuse_iqueue *fiq)
1223 __releases(fiq->lock)
1224 {
1225 	unsigned int queue_id = VQ_REQUEST; /* TODO multiqueue */
1226 	struct virtio_fs *fs;
1227 	struct fuse_req *req;
1228 	struct virtio_fs_vq *fsvq;
1229 	int ret;
1230 
1231 	WARN_ON(list_empty(&fiq->pending));
1232 	req = list_last_entry(&fiq->pending, struct fuse_req, list);
1233 	clear_bit(FR_PENDING, &req->flags);
1234 	list_del_init(&req->list);
1235 	WARN_ON(!list_empty(&fiq->pending));
1236 	spin_unlock(&fiq->lock);
1237 
1238 	fs = fiq->priv;
1239 
1240 	pr_debug("%s: opcode %u unique %#llx nodeid %#llx in.len %u out.len %u\n",
1241 		  __func__, req->in.h.opcode, req->in.h.unique,
1242 		 req->in.h.nodeid, req->in.h.len,
1243 		 fuse_len_args(req->args->out_numargs, req->args->out_args));
1244 
1245 	fsvq = &fs->vqs[queue_id];
1246 	ret = virtio_fs_enqueue_req(fsvq, req, false);
1247 	if (ret < 0) {
1248 		if (ret == -ENOMEM || ret == -ENOSPC) {
1249 			/*
1250 			 * Virtqueue full. Retry submission from worker
1251 			 * context as we might be holding fc->bg_lock.
1252 			 */
1253 			spin_lock(&fsvq->lock);
1254 			list_add_tail(&req->list, &fsvq->queued_reqs);
1255 			inc_in_flight_req(fsvq);
1256 			schedule_delayed_work(&fsvq->dispatch_work,
1257 						msecs_to_jiffies(1));
1258 			spin_unlock(&fsvq->lock);
1259 			return;
1260 		}
1261 		req->out.h.error = ret;
1262 		pr_err("virtio-fs: virtio_fs_enqueue_req() failed %d\n", ret);
1263 
1264 		/* Can't end request in submission context. Use a worker */
1265 		spin_lock(&fsvq->lock);
1266 		list_add_tail(&req->list, &fsvq->end_reqs);
1267 		schedule_delayed_work(&fsvq->dispatch_work, 0);
1268 		spin_unlock(&fsvq->lock);
1269 		return;
1270 	}
1271 }
1272 
1273 static const struct fuse_iqueue_ops virtio_fs_fiq_ops = {
1274 	.wake_forget_and_unlock		= virtio_fs_wake_forget_and_unlock,
1275 	.wake_interrupt_and_unlock	= virtio_fs_wake_interrupt_and_unlock,
1276 	.wake_pending_and_unlock	= virtio_fs_wake_pending_and_unlock,
1277 	.release			= virtio_fs_fiq_release,
1278 };
1279 
1280 static inline void virtio_fs_ctx_set_defaults(struct fuse_fs_context *ctx)
1281 {
1282 	ctx->rootmode = S_IFDIR;
1283 	ctx->default_permissions = 1;
1284 	ctx->allow_other = 1;
1285 	ctx->max_read = UINT_MAX;
1286 	ctx->blksize = 512;
1287 	ctx->destroy = true;
1288 	ctx->no_control = true;
1289 	ctx->no_force_umount = true;
1290 }
1291 
1292 static int virtio_fs_fill_super(struct super_block *sb, struct fs_context *fsc)
1293 {
1294 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1295 	struct fuse_conn *fc = fm->fc;
1296 	struct virtio_fs *fs = fc->iq.priv;
1297 	struct fuse_fs_context *ctx = fsc->fs_private;
1298 	unsigned int i;
1299 	int err;
1300 
1301 	virtio_fs_ctx_set_defaults(ctx);
1302 	mutex_lock(&virtio_fs_mutex);
1303 
1304 	/* After holding mutex, make sure virtiofs device is still there.
1305 	 * Though we are holding a reference to it, drive ->remove might
1306 	 * still have cleaned up virtual queues. In that case bail out.
1307 	 */
1308 	err = -EINVAL;
1309 	if (list_empty(&fs->list)) {
1310 		pr_info("virtio-fs: tag <%s> not found\n", fs->tag);
1311 		goto err;
1312 	}
1313 
1314 	err = -ENOMEM;
1315 	/* Allocate fuse_dev for hiprio and notification queues */
1316 	for (i = 0; i < fs->nvqs; i++) {
1317 		struct virtio_fs_vq *fsvq = &fs->vqs[i];
1318 
1319 		fsvq->fud = fuse_dev_alloc();
1320 		if (!fsvq->fud)
1321 			goto err_free_fuse_devs;
1322 	}
1323 
1324 	/* virtiofs allocates and installs its own fuse devices */
1325 	ctx->fudptr = NULL;
1326 	if (ctx->dax_mode != FUSE_DAX_NEVER) {
1327 		if (ctx->dax_mode == FUSE_DAX_ALWAYS && !fs->dax_dev) {
1328 			err = -EINVAL;
1329 			pr_err("virtio-fs: dax can't be enabled as filesystem"
1330 			       " device does not support it.\n");
1331 			goto err_free_fuse_devs;
1332 		}
1333 		ctx->dax_dev = fs->dax_dev;
1334 	}
1335 	err = fuse_fill_super_common(sb, ctx);
1336 	if (err < 0)
1337 		goto err_free_fuse_devs;
1338 
1339 	for (i = 0; i < fs->nvqs; i++) {
1340 		struct virtio_fs_vq *fsvq = &fs->vqs[i];
1341 
1342 		fuse_dev_install(fsvq->fud, fc);
1343 	}
1344 
1345 	/* Previous unmount will stop all queues. Start these again */
1346 	virtio_fs_start_all_queues(fs);
1347 	fuse_send_init(fm);
1348 	mutex_unlock(&virtio_fs_mutex);
1349 	return 0;
1350 
1351 err_free_fuse_devs:
1352 	virtio_fs_free_devs(fs);
1353 err:
1354 	mutex_unlock(&virtio_fs_mutex);
1355 	return err;
1356 }
1357 
1358 static void virtio_fs_conn_destroy(struct fuse_mount *fm)
1359 {
1360 	struct fuse_conn *fc = fm->fc;
1361 	struct virtio_fs *vfs = fc->iq.priv;
1362 	struct virtio_fs_vq *fsvq = &vfs->vqs[VQ_HIPRIO];
1363 
1364 	/* Stop dax worker. Soon evict_inodes() will be called which
1365 	 * will free all memory ranges belonging to all inodes.
1366 	 */
1367 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1368 		fuse_dax_cancel_work(fc);
1369 
1370 	/* Stop forget queue. Soon destroy will be sent */
1371 	spin_lock(&fsvq->lock);
1372 	fsvq->connected = false;
1373 	spin_unlock(&fsvq->lock);
1374 	virtio_fs_drain_all_queues(vfs);
1375 
1376 	fuse_conn_destroy(fm);
1377 
1378 	/* fuse_conn_destroy() must have sent destroy. Stop all queues
1379 	 * and drain one more time and free fuse devices. Freeing fuse
1380 	 * devices will drop their reference on fuse_conn and that in
1381 	 * turn will drop its reference on virtio_fs object.
1382 	 */
1383 	virtio_fs_stop_all_queues(vfs);
1384 	virtio_fs_drain_all_queues(vfs);
1385 	virtio_fs_free_devs(vfs);
1386 }
1387 
1388 static void virtio_kill_sb(struct super_block *sb)
1389 {
1390 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1391 	bool last;
1392 
1393 	/* If mount failed, we can still be called without any fc */
1394 	if (sb->s_root) {
1395 		last = fuse_mount_remove(fm);
1396 		if (last)
1397 			virtio_fs_conn_destroy(fm);
1398 	}
1399 	kill_anon_super(sb);
1400 	fuse_mount_destroy(fm);
1401 }
1402 
1403 static int virtio_fs_test_super(struct super_block *sb,
1404 				struct fs_context *fsc)
1405 {
1406 	struct fuse_mount *fsc_fm = fsc->s_fs_info;
1407 	struct fuse_mount *sb_fm = get_fuse_mount_super(sb);
1408 
1409 	return fsc_fm->fc->iq.priv == sb_fm->fc->iq.priv;
1410 }
1411 
1412 static int virtio_fs_get_tree(struct fs_context *fsc)
1413 {
1414 	struct virtio_fs *fs;
1415 	struct super_block *sb;
1416 	struct fuse_conn *fc = NULL;
1417 	struct fuse_mount *fm;
1418 	unsigned int virtqueue_size;
1419 	int err = -EIO;
1420 
1421 	/* This gets a reference on virtio_fs object. This ptr gets installed
1422 	 * in fc->iq->priv. Once fuse_conn is going away, it calls ->put()
1423 	 * to drop the reference to this object.
1424 	 */
1425 	fs = virtio_fs_find_instance(fsc->source);
1426 	if (!fs) {
1427 		pr_info("virtio-fs: tag <%s> not found\n", fsc->source);
1428 		return -EINVAL;
1429 	}
1430 
1431 	virtqueue_size = virtqueue_get_vring_size(fs->vqs[VQ_REQUEST].vq);
1432 	if (WARN_ON(virtqueue_size <= FUSE_HEADER_OVERHEAD))
1433 		goto out_err;
1434 
1435 	err = -ENOMEM;
1436 	fc = kzalloc(sizeof(struct fuse_conn), GFP_KERNEL);
1437 	if (!fc)
1438 		goto out_err;
1439 
1440 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1441 	if (!fm)
1442 		goto out_err;
1443 
1444 	fuse_conn_init(fc, fm, fsc->user_ns, &virtio_fs_fiq_ops, fs);
1445 	fc->release = fuse_free_conn;
1446 	fc->delete_stale = true;
1447 	fc->auto_submounts = true;
1448 	fc->sync_fs = true;
1449 
1450 	/* Tell FUSE to split requests that exceed the virtqueue's size */
1451 	fc->max_pages_limit = min_t(unsigned int, fc->max_pages_limit,
1452 				    virtqueue_size - FUSE_HEADER_OVERHEAD);
1453 
1454 	fsc->s_fs_info = fm;
1455 	sb = sget_fc(fsc, virtio_fs_test_super, set_anon_super_fc);
1456 	if (fsc->s_fs_info)
1457 		fuse_mount_destroy(fm);
1458 	if (IS_ERR(sb))
1459 		return PTR_ERR(sb);
1460 
1461 	if (!sb->s_root) {
1462 		err = virtio_fs_fill_super(sb, fsc);
1463 		if (err) {
1464 			deactivate_locked_super(sb);
1465 			return err;
1466 		}
1467 
1468 		sb->s_flags |= SB_ACTIVE;
1469 	}
1470 
1471 	WARN_ON(fsc->root);
1472 	fsc->root = dget(sb->s_root);
1473 	return 0;
1474 
1475 out_err:
1476 	kfree(fc);
1477 	mutex_lock(&virtio_fs_mutex);
1478 	virtio_fs_put(fs);
1479 	mutex_unlock(&virtio_fs_mutex);
1480 	return err;
1481 }
1482 
1483 static const struct fs_context_operations virtio_fs_context_ops = {
1484 	.free		= virtio_fs_free_fsc,
1485 	.parse_param	= virtio_fs_parse_param,
1486 	.get_tree	= virtio_fs_get_tree,
1487 };
1488 
1489 static int virtio_fs_init_fs_context(struct fs_context *fsc)
1490 {
1491 	struct fuse_fs_context *ctx;
1492 
1493 	if (fsc->purpose == FS_CONTEXT_FOR_SUBMOUNT)
1494 		return fuse_init_fs_context_submount(fsc);
1495 
1496 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1497 	if (!ctx)
1498 		return -ENOMEM;
1499 	fsc->fs_private = ctx;
1500 	fsc->ops = &virtio_fs_context_ops;
1501 	return 0;
1502 }
1503 
1504 static struct file_system_type virtio_fs_type = {
1505 	.owner		= THIS_MODULE,
1506 	.name		= "virtiofs",
1507 	.init_fs_context = virtio_fs_init_fs_context,
1508 	.kill_sb	= virtio_kill_sb,
1509 };
1510 
1511 static int __init virtio_fs_init(void)
1512 {
1513 	int ret;
1514 
1515 	ret = register_virtio_driver(&virtio_fs_driver);
1516 	if (ret < 0)
1517 		return ret;
1518 
1519 	ret = register_filesystem(&virtio_fs_type);
1520 	if (ret < 0) {
1521 		unregister_virtio_driver(&virtio_fs_driver);
1522 		return ret;
1523 	}
1524 
1525 	return 0;
1526 }
1527 module_init(virtio_fs_init);
1528 
1529 static void __exit virtio_fs_exit(void)
1530 {
1531 	unregister_filesystem(&virtio_fs_type);
1532 	unregister_virtio_driver(&virtio_fs_driver);
1533 }
1534 module_exit(virtio_fs_exit);
1535 
1536 MODULE_AUTHOR("Stefan Hajnoczi <stefanha@redhat.com>");
1537 MODULE_DESCRIPTION("Virtio Filesystem");
1538 MODULE_LICENSE("GPL");
1539 MODULE_ALIAS_FS(KBUILD_MODNAME);
1540 MODULE_DEVICE_TABLE(virtio, id_table);
1541