xref: /openbmc/linux/block/bsg.c (revision a86854d0)
1 /*
2  * bsg.c - block layer implementation of the sg v4 interface
3  *
4  * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
5  * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
6  *
7  *  This file is subject to the terms and conditions of the GNU General Public
8  *  License version 2.  See the file "COPYING" in the main directory of this
9  *  archive for more details.
10  *
11  */
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/file.h>
15 #include <linux/blkdev.h>
16 #include <linux/poll.h>
17 #include <linux/cdev.h>
18 #include <linux/jiffies.h>
19 #include <linux/percpu.h>
20 #include <linux/uio.h>
21 #include <linux/idr.h>
22 #include <linux/bsg.h>
23 #include <linux/slab.h>
24 
25 #include <scsi/scsi.h>
26 #include <scsi/scsi_ioctl.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_device.h>
29 #include <scsi/scsi_driver.h>
30 #include <scsi/sg.h>
31 
32 #define BSG_DESCRIPTION	"Block layer SCSI generic (bsg) driver"
33 #define BSG_VERSION	"0.4"
34 
35 #define bsg_dbg(bd, fmt, ...) \
36 	pr_debug("%s: " fmt, (bd)->name, ##__VA_ARGS__)
37 
38 struct bsg_device {
39 	struct request_queue *queue;
40 	spinlock_t lock;
41 	struct list_head busy_list;
42 	struct list_head done_list;
43 	struct hlist_node dev_list;
44 	atomic_t ref_count;
45 	int queued_cmds;
46 	int done_cmds;
47 	wait_queue_head_t wq_done;
48 	wait_queue_head_t wq_free;
49 	char name[20];
50 	int max_queue;
51 	unsigned long flags;
52 };
53 
54 enum {
55 	BSG_F_BLOCK		= 1,
56 };
57 
58 #define BSG_DEFAULT_CMDS	64
59 #define BSG_MAX_DEVS		32768
60 
61 static DEFINE_MUTEX(bsg_mutex);
62 static DEFINE_IDR(bsg_minor_idr);
63 
64 #define BSG_LIST_ARRAY_SIZE	8
65 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
66 
67 static struct class *bsg_class;
68 static int bsg_major;
69 
70 static struct kmem_cache *bsg_cmd_cachep;
71 
72 /*
73  * our internal command type
74  */
75 struct bsg_command {
76 	struct bsg_device *bd;
77 	struct list_head list;
78 	struct request *rq;
79 	struct bio *bio;
80 	struct bio *bidi_bio;
81 	int err;
82 	struct sg_io_v4 hdr;
83 };
84 
85 static void bsg_free_command(struct bsg_command *bc)
86 {
87 	struct bsg_device *bd = bc->bd;
88 	unsigned long flags;
89 
90 	kmem_cache_free(bsg_cmd_cachep, bc);
91 
92 	spin_lock_irqsave(&bd->lock, flags);
93 	bd->queued_cmds--;
94 	spin_unlock_irqrestore(&bd->lock, flags);
95 
96 	wake_up(&bd->wq_free);
97 }
98 
99 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
100 {
101 	struct bsg_command *bc = ERR_PTR(-EINVAL);
102 
103 	spin_lock_irq(&bd->lock);
104 
105 	if (bd->queued_cmds >= bd->max_queue)
106 		goto out;
107 
108 	bd->queued_cmds++;
109 	spin_unlock_irq(&bd->lock);
110 
111 	bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
112 	if (unlikely(!bc)) {
113 		spin_lock_irq(&bd->lock);
114 		bd->queued_cmds--;
115 		bc = ERR_PTR(-ENOMEM);
116 		goto out;
117 	}
118 
119 	bc->bd = bd;
120 	INIT_LIST_HEAD(&bc->list);
121 	bsg_dbg(bd, "returning free cmd %p\n", bc);
122 	return bc;
123 out:
124 	spin_unlock_irq(&bd->lock);
125 	return bc;
126 }
127 
128 static inline struct hlist_head *bsg_dev_idx_hash(int index)
129 {
130 	return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
131 }
132 
133 #define uptr64(val) ((void __user *)(uintptr_t)(val))
134 
135 static int bsg_scsi_check_proto(struct sg_io_v4 *hdr)
136 {
137 	if (hdr->protocol != BSG_PROTOCOL_SCSI  ||
138 	    hdr->subprotocol != BSG_SUB_PROTOCOL_SCSI_CMD)
139 		return -EINVAL;
140 	return 0;
141 }
142 
143 static int bsg_scsi_fill_hdr(struct request *rq, struct sg_io_v4 *hdr,
144 		fmode_t mode)
145 {
146 	struct scsi_request *sreq = scsi_req(rq);
147 
148 	sreq->cmd_len = hdr->request_len;
149 	if (sreq->cmd_len > BLK_MAX_CDB) {
150 		sreq->cmd = kzalloc(sreq->cmd_len, GFP_KERNEL);
151 		if (!sreq->cmd)
152 			return -ENOMEM;
153 	}
154 
155 	if (copy_from_user(sreq->cmd, uptr64(hdr->request), sreq->cmd_len))
156 		return -EFAULT;
157 	if (blk_verify_command(sreq->cmd, mode))
158 		return -EPERM;
159 	return 0;
160 }
161 
162 static int bsg_scsi_complete_rq(struct request *rq, struct sg_io_v4 *hdr)
163 {
164 	struct scsi_request *sreq = scsi_req(rq);
165 	int ret = 0;
166 
167 	/*
168 	 * fill in all the output members
169 	 */
170 	hdr->device_status = sreq->result & 0xff;
171 	hdr->transport_status = host_byte(sreq->result);
172 	hdr->driver_status = driver_byte(sreq->result);
173 	hdr->info = 0;
174 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
175 		hdr->info |= SG_INFO_CHECK;
176 	hdr->response_len = 0;
177 
178 	if (sreq->sense_len && hdr->response) {
179 		int len = min_t(unsigned int, hdr->max_response_len,
180 					sreq->sense_len);
181 
182 		if (copy_to_user(uptr64(hdr->response), sreq->sense, len))
183 			ret = -EFAULT;
184 		else
185 			hdr->response_len = len;
186 	}
187 
188 	if (rq->next_rq) {
189 		hdr->dout_resid = sreq->resid_len;
190 		hdr->din_resid = scsi_req(rq->next_rq)->resid_len;
191 	} else if (rq_data_dir(rq) == READ) {
192 		hdr->din_resid = sreq->resid_len;
193 	} else {
194 		hdr->dout_resid = sreq->resid_len;
195 	}
196 
197 	return ret;
198 }
199 
200 static void bsg_scsi_free_rq(struct request *rq)
201 {
202 	scsi_req_free_cmd(scsi_req(rq));
203 }
204 
205 static const struct bsg_ops bsg_scsi_ops = {
206 	.check_proto		= bsg_scsi_check_proto,
207 	.fill_hdr		= bsg_scsi_fill_hdr,
208 	.complete_rq		= bsg_scsi_complete_rq,
209 	.free_rq		= bsg_scsi_free_rq,
210 };
211 
212 static struct request *
213 bsg_map_hdr(struct request_queue *q, struct sg_io_v4 *hdr, fmode_t mode)
214 {
215 	struct request *rq, *next_rq = NULL;
216 	int ret;
217 
218 	if (!q->bsg_dev.class_dev)
219 		return ERR_PTR(-ENXIO);
220 
221 	if (hdr->guard != 'Q')
222 		return ERR_PTR(-EINVAL);
223 
224 	ret = q->bsg_dev.ops->check_proto(hdr);
225 	if (ret)
226 		return ERR_PTR(ret);
227 
228 	rq = blk_get_request(q, hdr->dout_xfer_len ?
229 			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
230 	if (IS_ERR(rq))
231 		return rq;
232 
233 	ret = q->bsg_dev.ops->fill_hdr(rq, hdr, mode);
234 	if (ret)
235 		goto out;
236 
237 	rq->timeout = msecs_to_jiffies(hdr->timeout);
238 	if (!rq->timeout)
239 		rq->timeout = q->sg_timeout;
240 	if (!rq->timeout)
241 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
242 	if (rq->timeout < BLK_MIN_SG_TIMEOUT)
243 		rq->timeout = BLK_MIN_SG_TIMEOUT;
244 
245 	if (hdr->dout_xfer_len && hdr->din_xfer_len) {
246 		if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
247 			ret = -EOPNOTSUPP;
248 			goto out;
249 		}
250 
251 		next_rq = blk_get_request(q, REQ_OP_SCSI_IN, 0);
252 		if (IS_ERR(next_rq)) {
253 			ret = PTR_ERR(next_rq);
254 			goto out;
255 		}
256 
257 		rq->next_rq = next_rq;
258 		ret = blk_rq_map_user(q, next_rq, NULL, uptr64(hdr->din_xferp),
259 				       hdr->din_xfer_len, GFP_KERNEL);
260 		if (ret)
261 			goto out_free_nextrq;
262 	}
263 
264 	if (hdr->dout_xfer_len) {
265 		ret = blk_rq_map_user(q, rq, NULL, uptr64(hdr->dout_xferp),
266 				hdr->dout_xfer_len, GFP_KERNEL);
267 	} else if (hdr->din_xfer_len) {
268 		ret = blk_rq_map_user(q, rq, NULL, uptr64(hdr->din_xferp),
269 				hdr->din_xfer_len, GFP_KERNEL);
270 	} else {
271 		ret = blk_rq_map_user(q, rq, NULL, NULL, 0, GFP_KERNEL);
272 	}
273 
274 	if (ret)
275 		goto out_unmap_nextrq;
276 	return rq;
277 
278 out_unmap_nextrq:
279 	if (rq->next_rq)
280 		blk_rq_unmap_user(rq->next_rq->bio);
281 out_free_nextrq:
282 	if (rq->next_rq)
283 		blk_put_request(rq->next_rq);
284 out:
285 	q->bsg_dev.ops->free_rq(rq);
286 	blk_put_request(rq);
287 	return ERR_PTR(ret);
288 }
289 
290 /*
291  * async completion call-back from the block layer, when scsi/ide/whatever
292  * calls end_that_request_last() on a request
293  */
294 static void bsg_rq_end_io(struct request *rq, blk_status_t status)
295 {
296 	struct bsg_command *bc = rq->end_io_data;
297 	struct bsg_device *bd = bc->bd;
298 	unsigned long flags;
299 
300 	bsg_dbg(bd, "finished rq %p bc %p, bio %p\n",
301 		rq, bc, bc->bio);
302 
303 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
304 
305 	spin_lock_irqsave(&bd->lock, flags);
306 	list_move_tail(&bc->list, &bd->done_list);
307 	bd->done_cmds++;
308 	spin_unlock_irqrestore(&bd->lock, flags);
309 
310 	wake_up(&bd->wq_done);
311 }
312 
313 /*
314  * do final setup of a 'bc' and submit the matching 'rq' to the block
315  * layer for io
316  */
317 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
318 			    struct bsg_command *bc, struct request *rq)
319 {
320 	int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
321 
322 	/*
323 	 * add bc command to busy queue and submit rq for io
324 	 */
325 	bc->rq = rq;
326 	bc->bio = rq->bio;
327 	if (rq->next_rq)
328 		bc->bidi_bio = rq->next_rq->bio;
329 	bc->hdr.duration = jiffies;
330 	spin_lock_irq(&bd->lock);
331 	list_add_tail(&bc->list, &bd->busy_list);
332 	spin_unlock_irq(&bd->lock);
333 
334 	bsg_dbg(bd, "queueing rq %p, bc %p\n", rq, bc);
335 
336 	rq->end_io_data = bc;
337 	blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
338 }
339 
340 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
341 {
342 	struct bsg_command *bc = NULL;
343 
344 	spin_lock_irq(&bd->lock);
345 	if (bd->done_cmds) {
346 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
347 		list_del(&bc->list);
348 		bd->done_cmds--;
349 	}
350 	spin_unlock_irq(&bd->lock);
351 
352 	return bc;
353 }
354 
355 /*
356  * Get a finished command from the done list
357  */
358 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
359 {
360 	struct bsg_command *bc;
361 	int ret;
362 
363 	do {
364 		bc = bsg_next_done_cmd(bd);
365 		if (bc)
366 			break;
367 
368 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
369 			bc = ERR_PTR(-EAGAIN);
370 			break;
371 		}
372 
373 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
374 		if (ret) {
375 			bc = ERR_PTR(-ERESTARTSYS);
376 			break;
377 		}
378 	} while (1);
379 
380 	bsg_dbg(bd, "returning done %p\n", bc);
381 
382 	return bc;
383 }
384 
385 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
386 				    struct bio *bio, struct bio *bidi_bio)
387 {
388 	int ret;
389 
390 	ret = rq->q->bsg_dev.ops->complete_rq(rq, hdr);
391 
392 	if (rq->next_rq) {
393 		blk_rq_unmap_user(bidi_bio);
394 		blk_put_request(rq->next_rq);
395 	}
396 
397 	blk_rq_unmap_user(bio);
398 	rq->q->bsg_dev.ops->free_rq(rq);
399 	blk_put_request(rq);
400 	return ret;
401 }
402 
403 static bool bsg_complete(struct bsg_device *bd)
404 {
405 	bool ret = false;
406 	bool spin;
407 
408 	do {
409 		spin_lock_irq(&bd->lock);
410 
411 		BUG_ON(bd->done_cmds > bd->queued_cmds);
412 
413 		/*
414 		 * All commands consumed.
415 		 */
416 		if (bd->done_cmds == bd->queued_cmds)
417 			ret = true;
418 
419 		spin = !test_bit(BSG_F_BLOCK, &bd->flags);
420 
421 		spin_unlock_irq(&bd->lock);
422 	} while (!ret && spin);
423 
424 	return ret;
425 }
426 
427 static int bsg_complete_all_commands(struct bsg_device *bd)
428 {
429 	struct bsg_command *bc;
430 	int ret, tret;
431 
432 	bsg_dbg(bd, "entered\n");
433 
434 	/*
435 	 * wait for all commands to complete
436 	 */
437 	io_wait_event(bd->wq_done, bsg_complete(bd));
438 
439 	/*
440 	 * discard done commands
441 	 */
442 	ret = 0;
443 	do {
444 		spin_lock_irq(&bd->lock);
445 		if (!bd->queued_cmds) {
446 			spin_unlock_irq(&bd->lock);
447 			break;
448 		}
449 		spin_unlock_irq(&bd->lock);
450 
451 		bc = bsg_get_done_cmd(bd);
452 		if (IS_ERR(bc))
453 			break;
454 
455 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
456 						bc->bidi_bio);
457 		if (!ret)
458 			ret = tret;
459 
460 		bsg_free_command(bc);
461 	} while (1);
462 
463 	return ret;
464 }
465 
466 static int
467 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
468 	   const struct iovec *iov, ssize_t *bytes_read)
469 {
470 	struct bsg_command *bc;
471 	int nr_commands, ret;
472 
473 	if (count % sizeof(struct sg_io_v4))
474 		return -EINVAL;
475 
476 	ret = 0;
477 	nr_commands = count / sizeof(struct sg_io_v4);
478 	while (nr_commands) {
479 		bc = bsg_get_done_cmd(bd);
480 		if (IS_ERR(bc)) {
481 			ret = PTR_ERR(bc);
482 			break;
483 		}
484 
485 		/*
486 		 * this is the only case where we need to copy data back
487 		 * after completing the request. so do that here,
488 		 * bsg_complete_work() cannot do that for us
489 		 */
490 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
491 					       bc->bidi_bio);
492 
493 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
494 			ret = -EFAULT;
495 
496 		bsg_free_command(bc);
497 
498 		if (ret)
499 			break;
500 
501 		buf += sizeof(struct sg_io_v4);
502 		*bytes_read += sizeof(struct sg_io_v4);
503 		nr_commands--;
504 	}
505 
506 	return ret;
507 }
508 
509 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
510 {
511 	if (file->f_flags & O_NONBLOCK)
512 		clear_bit(BSG_F_BLOCK, &bd->flags);
513 	else
514 		set_bit(BSG_F_BLOCK, &bd->flags);
515 }
516 
517 /*
518  * Check if the error is a "real" error that we should return.
519  */
520 static inline int err_block_err(int ret)
521 {
522 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
523 		return 1;
524 
525 	return 0;
526 }
527 
528 static ssize_t
529 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
530 {
531 	struct bsg_device *bd = file->private_data;
532 	int ret;
533 	ssize_t bytes_read;
534 
535 	bsg_dbg(bd, "read %zd bytes\n", count);
536 
537 	bsg_set_block(bd, file);
538 
539 	bytes_read = 0;
540 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
541 	*ppos = bytes_read;
542 
543 	if (!bytes_read || err_block_err(ret))
544 		bytes_read = ret;
545 
546 	return bytes_read;
547 }
548 
549 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
550 		       size_t count, ssize_t *bytes_written, fmode_t mode)
551 {
552 	struct bsg_command *bc;
553 	struct request *rq;
554 	int ret, nr_commands;
555 
556 	if (count % sizeof(struct sg_io_v4))
557 		return -EINVAL;
558 
559 	nr_commands = count / sizeof(struct sg_io_v4);
560 	rq = NULL;
561 	bc = NULL;
562 	ret = 0;
563 	while (nr_commands) {
564 		struct request_queue *q = bd->queue;
565 
566 		bc = bsg_alloc_command(bd);
567 		if (IS_ERR(bc)) {
568 			ret = PTR_ERR(bc);
569 			bc = NULL;
570 			break;
571 		}
572 
573 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
574 			ret = -EFAULT;
575 			break;
576 		}
577 
578 		/*
579 		 * get a request, fill in the blanks, and add to request queue
580 		 */
581 		rq = bsg_map_hdr(bd->queue, &bc->hdr, mode);
582 		if (IS_ERR(rq)) {
583 			ret = PTR_ERR(rq);
584 			rq = NULL;
585 			break;
586 		}
587 
588 		bsg_add_command(bd, q, bc, rq);
589 		bc = NULL;
590 		rq = NULL;
591 		nr_commands--;
592 		buf += sizeof(struct sg_io_v4);
593 		*bytes_written += sizeof(struct sg_io_v4);
594 	}
595 
596 	if (bc)
597 		bsg_free_command(bc);
598 
599 	return ret;
600 }
601 
602 static ssize_t
603 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
604 {
605 	struct bsg_device *bd = file->private_data;
606 	ssize_t bytes_written;
607 	int ret;
608 
609 	bsg_dbg(bd, "write %zd bytes\n", count);
610 
611 	if (unlikely(uaccess_kernel()))
612 		return -EINVAL;
613 
614 	bsg_set_block(bd, file);
615 
616 	bytes_written = 0;
617 	ret = __bsg_write(bd, buf, count, &bytes_written, file->f_mode);
618 
619 	*ppos = bytes_written;
620 
621 	/*
622 	 * return bytes written on non-fatal errors
623 	 */
624 	if (!bytes_written || err_block_err(ret))
625 		bytes_written = ret;
626 
627 	bsg_dbg(bd, "returning %zd\n", bytes_written);
628 	return bytes_written;
629 }
630 
631 static struct bsg_device *bsg_alloc_device(void)
632 {
633 	struct bsg_device *bd;
634 
635 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
636 	if (unlikely(!bd))
637 		return NULL;
638 
639 	spin_lock_init(&bd->lock);
640 
641 	bd->max_queue = BSG_DEFAULT_CMDS;
642 
643 	INIT_LIST_HEAD(&bd->busy_list);
644 	INIT_LIST_HEAD(&bd->done_list);
645 	INIT_HLIST_NODE(&bd->dev_list);
646 
647 	init_waitqueue_head(&bd->wq_free);
648 	init_waitqueue_head(&bd->wq_done);
649 	return bd;
650 }
651 
652 static int bsg_put_device(struct bsg_device *bd)
653 {
654 	int ret = 0, do_free;
655 	struct request_queue *q = bd->queue;
656 
657 	mutex_lock(&bsg_mutex);
658 
659 	do_free = atomic_dec_and_test(&bd->ref_count);
660 	if (!do_free) {
661 		mutex_unlock(&bsg_mutex);
662 		goto out;
663 	}
664 
665 	hlist_del(&bd->dev_list);
666 	mutex_unlock(&bsg_mutex);
667 
668 	bsg_dbg(bd, "tearing down\n");
669 
670 	/*
671 	 * close can always block
672 	 */
673 	set_bit(BSG_F_BLOCK, &bd->flags);
674 
675 	/*
676 	 * correct error detection baddies here again. it's the responsibility
677 	 * of the app to properly reap commands before close() if it wants
678 	 * fool-proof error detection
679 	 */
680 	ret = bsg_complete_all_commands(bd);
681 
682 	kfree(bd);
683 out:
684 	if (do_free)
685 		blk_put_queue(q);
686 	return ret;
687 }
688 
689 static struct bsg_device *bsg_add_device(struct inode *inode,
690 					 struct request_queue *rq,
691 					 struct file *file)
692 {
693 	struct bsg_device *bd;
694 	unsigned char buf[32];
695 
696 	if (!blk_get_queue(rq))
697 		return ERR_PTR(-ENXIO);
698 
699 	bd = bsg_alloc_device();
700 	if (!bd) {
701 		blk_put_queue(rq);
702 		return ERR_PTR(-ENOMEM);
703 	}
704 
705 	bd->queue = rq;
706 
707 	bsg_set_block(bd, file);
708 
709 	atomic_set(&bd->ref_count, 1);
710 	mutex_lock(&bsg_mutex);
711 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
712 
713 	strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
714 	bsg_dbg(bd, "bound to <%s>, max queue %d\n",
715 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
716 
717 	mutex_unlock(&bsg_mutex);
718 	return bd;
719 }
720 
721 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
722 {
723 	struct bsg_device *bd;
724 
725 	mutex_lock(&bsg_mutex);
726 
727 	hlist_for_each_entry(bd, bsg_dev_idx_hash(minor), dev_list) {
728 		if (bd->queue == q) {
729 			atomic_inc(&bd->ref_count);
730 			goto found;
731 		}
732 	}
733 	bd = NULL;
734 found:
735 	mutex_unlock(&bsg_mutex);
736 	return bd;
737 }
738 
739 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
740 {
741 	struct bsg_device *bd;
742 	struct bsg_class_device *bcd;
743 
744 	/*
745 	 * find the class device
746 	 */
747 	mutex_lock(&bsg_mutex);
748 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
749 	mutex_unlock(&bsg_mutex);
750 
751 	if (!bcd)
752 		return ERR_PTR(-ENODEV);
753 
754 	bd = __bsg_get_device(iminor(inode), bcd->queue);
755 	if (bd)
756 		return bd;
757 
758 	bd = bsg_add_device(inode, bcd->queue, file);
759 
760 	return bd;
761 }
762 
763 static int bsg_open(struct inode *inode, struct file *file)
764 {
765 	struct bsg_device *bd;
766 
767 	bd = bsg_get_device(inode, file);
768 
769 	if (IS_ERR(bd))
770 		return PTR_ERR(bd);
771 
772 	file->private_data = bd;
773 	return 0;
774 }
775 
776 static int bsg_release(struct inode *inode, struct file *file)
777 {
778 	struct bsg_device *bd = file->private_data;
779 
780 	file->private_data = NULL;
781 	return bsg_put_device(bd);
782 }
783 
784 static __poll_t bsg_poll(struct file *file, poll_table *wait)
785 {
786 	struct bsg_device *bd = file->private_data;
787 	__poll_t mask = 0;
788 
789 	poll_wait(file, &bd->wq_done, wait);
790 	poll_wait(file, &bd->wq_free, wait);
791 
792 	spin_lock_irq(&bd->lock);
793 	if (!list_empty(&bd->done_list))
794 		mask |= EPOLLIN | EPOLLRDNORM;
795 	if (bd->queued_cmds < bd->max_queue)
796 		mask |= EPOLLOUT;
797 	spin_unlock_irq(&bd->lock);
798 
799 	return mask;
800 }
801 
802 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
803 {
804 	struct bsg_device *bd = file->private_data;
805 	int __user *uarg = (int __user *) arg;
806 	int ret;
807 
808 	switch (cmd) {
809 		/*
810 		 * our own ioctls
811 		 */
812 	case SG_GET_COMMAND_Q:
813 		return put_user(bd->max_queue, uarg);
814 	case SG_SET_COMMAND_Q: {
815 		int queue;
816 
817 		if (get_user(queue, uarg))
818 			return -EFAULT;
819 		if (queue < 1)
820 			return -EINVAL;
821 
822 		spin_lock_irq(&bd->lock);
823 		bd->max_queue = queue;
824 		spin_unlock_irq(&bd->lock);
825 		return 0;
826 	}
827 
828 	/*
829 	 * SCSI/sg ioctls
830 	 */
831 	case SG_GET_VERSION_NUM:
832 	case SCSI_IOCTL_GET_IDLUN:
833 	case SCSI_IOCTL_GET_BUS_NUMBER:
834 	case SG_SET_TIMEOUT:
835 	case SG_GET_TIMEOUT:
836 	case SG_GET_RESERVED_SIZE:
837 	case SG_SET_RESERVED_SIZE:
838 	case SG_EMULATED_HOST:
839 	case SCSI_IOCTL_SEND_COMMAND: {
840 		void __user *uarg = (void __user *) arg;
841 		return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
842 	}
843 	case SG_IO: {
844 		struct request *rq;
845 		struct bio *bio, *bidi_bio = NULL;
846 		struct sg_io_v4 hdr;
847 		int at_head;
848 
849 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
850 			return -EFAULT;
851 
852 		rq = bsg_map_hdr(bd->queue, &hdr, file->f_mode);
853 		if (IS_ERR(rq))
854 			return PTR_ERR(rq);
855 
856 		bio = rq->bio;
857 		if (rq->next_rq)
858 			bidi_bio = rq->next_rq->bio;
859 
860 		at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
861 		blk_execute_rq(bd->queue, NULL, rq, at_head);
862 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
863 
864 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
865 			return -EFAULT;
866 
867 		return ret;
868 	}
869 	default:
870 		return -ENOTTY;
871 	}
872 }
873 
874 static const struct file_operations bsg_fops = {
875 	.read		=	bsg_read,
876 	.write		=	bsg_write,
877 	.poll		=	bsg_poll,
878 	.open		=	bsg_open,
879 	.release	=	bsg_release,
880 	.unlocked_ioctl	=	bsg_ioctl,
881 	.owner		=	THIS_MODULE,
882 	.llseek		=	default_llseek,
883 };
884 
885 void bsg_unregister_queue(struct request_queue *q)
886 {
887 	struct bsg_class_device *bcd = &q->bsg_dev;
888 
889 	if (!bcd->class_dev)
890 		return;
891 
892 	mutex_lock(&bsg_mutex);
893 	idr_remove(&bsg_minor_idr, bcd->minor);
894 	if (q->kobj.sd)
895 		sysfs_remove_link(&q->kobj, "bsg");
896 	device_unregister(bcd->class_dev);
897 	bcd->class_dev = NULL;
898 	mutex_unlock(&bsg_mutex);
899 }
900 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
901 
902 int bsg_register_queue(struct request_queue *q, struct device *parent,
903 		const char *name, const struct bsg_ops *ops)
904 {
905 	struct bsg_class_device *bcd;
906 	dev_t dev;
907 	int ret;
908 	struct device *class_dev = NULL;
909 
910 	/*
911 	 * we need a proper transport to send commands, not a stacked device
912 	 */
913 	if (!queue_is_rq_based(q))
914 		return 0;
915 
916 	bcd = &q->bsg_dev;
917 	memset(bcd, 0, sizeof(*bcd));
918 
919 	mutex_lock(&bsg_mutex);
920 
921 	ret = idr_alloc(&bsg_minor_idr, bcd, 0, BSG_MAX_DEVS, GFP_KERNEL);
922 	if (ret < 0) {
923 		if (ret == -ENOSPC) {
924 			printk(KERN_ERR "bsg: too many bsg devices\n");
925 			ret = -EINVAL;
926 		}
927 		goto unlock;
928 	}
929 
930 	bcd->minor = ret;
931 	bcd->queue = q;
932 	bcd->ops = ops;
933 	dev = MKDEV(bsg_major, bcd->minor);
934 	class_dev = device_create(bsg_class, parent, dev, NULL, "%s", name);
935 	if (IS_ERR(class_dev)) {
936 		ret = PTR_ERR(class_dev);
937 		goto idr_remove;
938 	}
939 	bcd->class_dev = class_dev;
940 
941 	if (q->kobj.sd) {
942 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
943 		if (ret)
944 			goto unregister_class_dev;
945 	}
946 
947 	mutex_unlock(&bsg_mutex);
948 	return 0;
949 
950 unregister_class_dev:
951 	device_unregister(class_dev);
952 idr_remove:
953 	idr_remove(&bsg_minor_idr, bcd->minor);
954 unlock:
955 	mutex_unlock(&bsg_mutex);
956 	return ret;
957 }
958 
959 int bsg_scsi_register_queue(struct request_queue *q, struct device *parent)
960 {
961 	if (!blk_queue_scsi_passthrough(q)) {
962 		WARN_ONCE(true, "Attempt to register a non-SCSI queue\n");
963 		return -EINVAL;
964 	}
965 
966 	return bsg_register_queue(q, parent, dev_name(parent), &bsg_scsi_ops);
967 }
968 EXPORT_SYMBOL_GPL(bsg_scsi_register_queue);
969 
970 static struct cdev bsg_cdev;
971 
972 static char *bsg_devnode(struct device *dev, umode_t *mode)
973 {
974 	return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
975 }
976 
977 static int __init bsg_init(void)
978 {
979 	int ret, i;
980 	dev_t devid;
981 
982 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
983 				sizeof(struct bsg_command), 0, 0, NULL);
984 	if (!bsg_cmd_cachep) {
985 		printk(KERN_ERR "bsg: failed creating slab cache\n");
986 		return -ENOMEM;
987 	}
988 
989 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
990 		INIT_HLIST_HEAD(&bsg_device_list[i]);
991 
992 	bsg_class = class_create(THIS_MODULE, "bsg");
993 	if (IS_ERR(bsg_class)) {
994 		ret = PTR_ERR(bsg_class);
995 		goto destroy_kmemcache;
996 	}
997 	bsg_class->devnode = bsg_devnode;
998 
999 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1000 	if (ret)
1001 		goto destroy_bsg_class;
1002 
1003 	bsg_major = MAJOR(devid);
1004 
1005 	cdev_init(&bsg_cdev, &bsg_fops);
1006 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1007 	if (ret)
1008 		goto unregister_chrdev;
1009 
1010 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1011 	       " loaded (major %d)\n", bsg_major);
1012 	return 0;
1013 unregister_chrdev:
1014 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1015 destroy_bsg_class:
1016 	class_destroy(bsg_class);
1017 destroy_kmemcache:
1018 	kmem_cache_destroy(bsg_cmd_cachep);
1019 	return ret;
1020 }
1021 
1022 MODULE_AUTHOR("Jens Axboe");
1023 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1024 MODULE_LICENSE("GPL");
1025 
1026 device_initcall(bsg_init);
1027