xref: /openbmc/linux/block/bsg.c (revision 133f9794)
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 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
134 				struct sg_io_v4 *hdr, struct bsg_device *bd,
135 				fmode_t mode)
136 {
137 	struct scsi_request *req = scsi_req(rq);
138 
139 	if (hdr->request_len > BLK_MAX_CDB) {
140 		req->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
141 		if (!req->cmd)
142 			return -ENOMEM;
143 	}
144 
145 	if (copy_from_user(req->cmd, (void __user *)(unsigned long)hdr->request,
146 			   hdr->request_len))
147 		return -EFAULT;
148 
149 	if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
150 		if (blk_verify_command(req->cmd, mode))
151 			return -EPERM;
152 	} else if (!capable(CAP_SYS_RAWIO))
153 		return -EPERM;
154 
155 	/*
156 	 * fill in request structure
157 	 */
158 	req->cmd_len = hdr->request_len;
159 
160 	rq->timeout = msecs_to_jiffies(hdr->timeout);
161 	if (!rq->timeout)
162 		rq->timeout = q->sg_timeout;
163 	if (!rq->timeout)
164 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
165 	if (rq->timeout < BLK_MIN_SG_TIMEOUT)
166 		rq->timeout = BLK_MIN_SG_TIMEOUT;
167 
168 	return 0;
169 }
170 
171 /*
172  * Check if sg_io_v4 from user is allowed and valid
173  */
174 static int
175 bsg_validate_sgv4_hdr(struct sg_io_v4 *hdr, int *op)
176 {
177 	int ret = 0;
178 
179 	if (hdr->guard != 'Q')
180 		return -EINVAL;
181 
182 	switch (hdr->protocol) {
183 	case BSG_PROTOCOL_SCSI:
184 		switch (hdr->subprotocol) {
185 		case BSG_SUB_PROTOCOL_SCSI_CMD:
186 		case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
187 			break;
188 		default:
189 			ret = -EINVAL;
190 		}
191 		break;
192 	default:
193 		ret = -EINVAL;
194 	}
195 
196 	*op = hdr->dout_xfer_len ? REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN;
197 	return ret;
198 }
199 
200 /*
201  * map sg_io_v4 to a request.
202  */
203 static struct request *
204 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t mode)
205 {
206 	struct request_queue *q = bd->queue;
207 	struct request *rq, *next_rq = NULL;
208 	int ret;
209 	unsigned int op, dxfer_len;
210 	void __user *dxferp = NULL;
211 	struct bsg_class_device *bcd = &q->bsg_dev;
212 
213 	/* if the LLD has been removed then the bsg_unregister_queue will
214 	 * eventually be called and the class_dev was freed, so we can no
215 	 * longer use this request_queue. Return no such address.
216 	 */
217 	if (!bcd->class_dev)
218 		return ERR_PTR(-ENXIO);
219 
220 	bsg_dbg(bd, "map hdr %llx/%u %llx/%u\n",
221 		(unsigned long long) hdr->dout_xferp,
222 		hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
223 		hdr->din_xfer_len);
224 
225 	ret = bsg_validate_sgv4_hdr(hdr, &op);
226 	if (ret)
227 		return ERR_PTR(ret);
228 
229 	/*
230 	 * map scatter-gather elements separately and string them to request
231 	 */
232 	rq = blk_get_request(q, op, GFP_KERNEL);
233 	if (IS_ERR(rq))
234 		return rq;
235 
236 	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, mode);
237 	if (ret)
238 		goto out;
239 
240 	if (op == REQ_OP_SCSI_OUT && hdr->din_xfer_len) {
241 		if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
242 			ret = -EOPNOTSUPP;
243 			goto out;
244 		}
245 
246 		next_rq = blk_get_request(q, REQ_OP_SCSI_IN, GFP_KERNEL);
247 		if (IS_ERR(next_rq)) {
248 			ret = PTR_ERR(next_rq);
249 			next_rq = NULL;
250 			goto out;
251 		}
252 		rq->next_rq = next_rq;
253 
254 		dxferp = (void __user *)(unsigned long)hdr->din_xferp;
255 		ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
256 				       hdr->din_xfer_len, GFP_KERNEL);
257 		if (ret)
258 			goto out;
259 	}
260 
261 	if (hdr->dout_xfer_len) {
262 		dxfer_len = hdr->dout_xfer_len;
263 		dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
264 	} else if (hdr->din_xfer_len) {
265 		dxfer_len = hdr->din_xfer_len;
266 		dxferp = (void __user *)(unsigned long)hdr->din_xferp;
267 	} else
268 		dxfer_len = 0;
269 
270 	if (dxfer_len) {
271 		ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
272 				      GFP_KERNEL);
273 		if (ret)
274 			goto out;
275 	}
276 
277 	return rq;
278 out:
279 	scsi_req_free_cmd(scsi_req(rq));
280 	blk_put_request(rq);
281 	if (next_rq) {
282 		blk_rq_unmap_user(next_rq->bio);
283 		blk_put_request(next_rq);
284 	}
285 	return ERR_PTR(ret);
286 }
287 
288 /*
289  * async completion call-back from the block layer, when scsi/ide/whatever
290  * calls end_that_request_last() on a request
291  */
292 static void bsg_rq_end_io(struct request *rq, blk_status_t status)
293 {
294 	struct bsg_command *bc = rq->end_io_data;
295 	struct bsg_device *bd = bc->bd;
296 	unsigned long flags;
297 
298 	bsg_dbg(bd, "finished rq %p bc %p, bio %p\n",
299 		rq, bc, bc->bio);
300 
301 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
302 
303 	spin_lock_irqsave(&bd->lock, flags);
304 	list_move_tail(&bc->list, &bd->done_list);
305 	bd->done_cmds++;
306 	spin_unlock_irqrestore(&bd->lock, flags);
307 
308 	wake_up(&bd->wq_done);
309 }
310 
311 /*
312  * do final setup of a 'bc' and submit the matching 'rq' to the block
313  * layer for io
314  */
315 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
316 			    struct bsg_command *bc, struct request *rq)
317 {
318 	int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
319 
320 	/*
321 	 * add bc command to busy queue and submit rq for io
322 	 */
323 	bc->rq = rq;
324 	bc->bio = rq->bio;
325 	if (rq->next_rq)
326 		bc->bidi_bio = rq->next_rq->bio;
327 	bc->hdr.duration = jiffies;
328 	spin_lock_irq(&bd->lock);
329 	list_add_tail(&bc->list, &bd->busy_list);
330 	spin_unlock_irq(&bd->lock);
331 
332 	bsg_dbg(bd, "queueing rq %p, bc %p\n", rq, bc);
333 
334 	rq->end_io_data = bc;
335 	blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
336 }
337 
338 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
339 {
340 	struct bsg_command *bc = NULL;
341 
342 	spin_lock_irq(&bd->lock);
343 	if (bd->done_cmds) {
344 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
345 		list_del(&bc->list);
346 		bd->done_cmds--;
347 	}
348 	spin_unlock_irq(&bd->lock);
349 
350 	return bc;
351 }
352 
353 /*
354  * Get a finished command from the done list
355  */
356 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
357 {
358 	struct bsg_command *bc;
359 	int ret;
360 
361 	do {
362 		bc = bsg_next_done_cmd(bd);
363 		if (bc)
364 			break;
365 
366 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
367 			bc = ERR_PTR(-EAGAIN);
368 			break;
369 		}
370 
371 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
372 		if (ret) {
373 			bc = ERR_PTR(-ERESTARTSYS);
374 			break;
375 		}
376 	} while (1);
377 
378 	bsg_dbg(bd, "returning done %p\n", bc);
379 
380 	return bc;
381 }
382 
383 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
384 				    struct bio *bio, struct bio *bidi_bio)
385 {
386 	struct scsi_request *req = scsi_req(rq);
387 	int ret = 0;
388 
389 	pr_debug("rq %p bio %p 0x%x\n", rq, bio, req->result);
390 	/*
391 	 * fill in all the output members
392 	 */
393 	hdr->device_status = req->result & 0xff;
394 	hdr->transport_status = host_byte(req->result);
395 	hdr->driver_status = driver_byte(req->result);
396 	hdr->info = 0;
397 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
398 		hdr->info |= SG_INFO_CHECK;
399 	hdr->response_len = 0;
400 
401 	if (req->sense_len && hdr->response) {
402 		int len = min_t(unsigned int, hdr->max_response_len,
403 					req->sense_len);
404 
405 		ret = copy_to_user((void __user *)(unsigned long)hdr->response,
406 				   req->sense, len);
407 		if (!ret)
408 			hdr->response_len = len;
409 		else
410 			ret = -EFAULT;
411 	}
412 
413 	if (rq->next_rq) {
414 		hdr->dout_resid = req->resid_len;
415 		hdr->din_resid = scsi_req(rq->next_rq)->resid_len;
416 		blk_rq_unmap_user(bidi_bio);
417 		blk_put_request(rq->next_rq);
418 	} else if (rq_data_dir(rq) == READ)
419 		hdr->din_resid = req->resid_len;
420 	else
421 		hdr->dout_resid = req->resid_len;
422 
423 	/*
424 	 * If the request generated a negative error number, return it
425 	 * (providing we aren't already returning an error); if it's
426 	 * just a protocol response (i.e. non negative), that gets
427 	 * processed above.
428 	 */
429 	if (!ret && req->result < 0)
430 		ret = req->result;
431 
432 	blk_rq_unmap_user(bio);
433 	scsi_req_free_cmd(req);
434 	blk_put_request(rq);
435 
436 	return ret;
437 }
438 
439 static bool bsg_complete(struct bsg_device *bd)
440 {
441 	bool ret = false;
442 	bool spin;
443 
444 	do {
445 		spin_lock_irq(&bd->lock);
446 
447 		BUG_ON(bd->done_cmds > bd->queued_cmds);
448 
449 		/*
450 		 * All commands consumed.
451 		 */
452 		if (bd->done_cmds == bd->queued_cmds)
453 			ret = true;
454 
455 		spin = !test_bit(BSG_F_BLOCK, &bd->flags);
456 
457 		spin_unlock_irq(&bd->lock);
458 	} while (!ret && spin);
459 
460 	return ret;
461 }
462 
463 static int bsg_complete_all_commands(struct bsg_device *bd)
464 {
465 	struct bsg_command *bc;
466 	int ret, tret;
467 
468 	bsg_dbg(bd, "entered\n");
469 
470 	/*
471 	 * wait for all commands to complete
472 	 */
473 	io_wait_event(bd->wq_done, bsg_complete(bd));
474 
475 	/*
476 	 * discard done commands
477 	 */
478 	ret = 0;
479 	do {
480 		spin_lock_irq(&bd->lock);
481 		if (!bd->queued_cmds) {
482 			spin_unlock_irq(&bd->lock);
483 			break;
484 		}
485 		spin_unlock_irq(&bd->lock);
486 
487 		bc = bsg_get_done_cmd(bd);
488 		if (IS_ERR(bc))
489 			break;
490 
491 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
492 						bc->bidi_bio);
493 		if (!ret)
494 			ret = tret;
495 
496 		bsg_free_command(bc);
497 	} while (1);
498 
499 	return ret;
500 }
501 
502 static int
503 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
504 	   const struct iovec *iov, ssize_t *bytes_read)
505 {
506 	struct bsg_command *bc;
507 	int nr_commands, ret;
508 
509 	if (count % sizeof(struct sg_io_v4))
510 		return -EINVAL;
511 
512 	ret = 0;
513 	nr_commands = count / sizeof(struct sg_io_v4);
514 	while (nr_commands) {
515 		bc = bsg_get_done_cmd(bd);
516 		if (IS_ERR(bc)) {
517 			ret = PTR_ERR(bc);
518 			break;
519 		}
520 
521 		/*
522 		 * this is the only case where we need to copy data back
523 		 * after completing the request. so do that here,
524 		 * bsg_complete_work() cannot do that for us
525 		 */
526 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
527 					       bc->bidi_bio);
528 
529 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
530 			ret = -EFAULT;
531 
532 		bsg_free_command(bc);
533 
534 		if (ret)
535 			break;
536 
537 		buf += sizeof(struct sg_io_v4);
538 		*bytes_read += sizeof(struct sg_io_v4);
539 		nr_commands--;
540 	}
541 
542 	return ret;
543 }
544 
545 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
546 {
547 	if (file->f_flags & O_NONBLOCK)
548 		clear_bit(BSG_F_BLOCK, &bd->flags);
549 	else
550 		set_bit(BSG_F_BLOCK, &bd->flags);
551 }
552 
553 /*
554  * Check if the error is a "real" error that we should return.
555  */
556 static inline int err_block_err(int ret)
557 {
558 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
559 		return 1;
560 
561 	return 0;
562 }
563 
564 static ssize_t
565 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
566 {
567 	struct bsg_device *bd = file->private_data;
568 	int ret;
569 	ssize_t bytes_read;
570 
571 	bsg_dbg(bd, "read %zd bytes\n", count);
572 
573 	bsg_set_block(bd, file);
574 
575 	bytes_read = 0;
576 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
577 	*ppos = bytes_read;
578 
579 	if (!bytes_read || err_block_err(ret))
580 		bytes_read = ret;
581 
582 	return bytes_read;
583 }
584 
585 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
586 		       size_t count, ssize_t *bytes_written, fmode_t mode)
587 {
588 	struct bsg_command *bc;
589 	struct request *rq;
590 	int ret, nr_commands;
591 
592 	if (count % sizeof(struct sg_io_v4))
593 		return -EINVAL;
594 
595 	nr_commands = count / sizeof(struct sg_io_v4);
596 	rq = NULL;
597 	bc = NULL;
598 	ret = 0;
599 	while (nr_commands) {
600 		struct request_queue *q = bd->queue;
601 
602 		bc = bsg_alloc_command(bd);
603 		if (IS_ERR(bc)) {
604 			ret = PTR_ERR(bc);
605 			bc = NULL;
606 			break;
607 		}
608 
609 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
610 			ret = -EFAULT;
611 			break;
612 		}
613 
614 		/*
615 		 * get a request, fill in the blanks, and add to request queue
616 		 */
617 		rq = bsg_map_hdr(bd, &bc->hdr, mode);
618 		if (IS_ERR(rq)) {
619 			ret = PTR_ERR(rq);
620 			rq = NULL;
621 			break;
622 		}
623 
624 		bsg_add_command(bd, q, bc, rq);
625 		bc = NULL;
626 		rq = NULL;
627 		nr_commands--;
628 		buf += sizeof(struct sg_io_v4);
629 		*bytes_written += sizeof(struct sg_io_v4);
630 	}
631 
632 	if (bc)
633 		bsg_free_command(bc);
634 
635 	return ret;
636 }
637 
638 static ssize_t
639 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
640 {
641 	struct bsg_device *bd = file->private_data;
642 	ssize_t bytes_written;
643 	int ret;
644 
645 	bsg_dbg(bd, "write %zd bytes\n", count);
646 
647 	if (unlikely(uaccess_kernel()))
648 		return -EINVAL;
649 
650 	bsg_set_block(bd, file);
651 
652 	bytes_written = 0;
653 	ret = __bsg_write(bd, buf, count, &bytes_written, file->f_mode);
654 
655 	*ppos = bytes_written;
656 
657 	/*
658 	 * return bytes written on non-fatal errors
659 	 */
660 	if (!bytes_written || err_block_err(ret))
661 		bytes_written = ret;
662 
663 	bsg_dbg(bd, "returning %zd\n", bytes_written);
664 	return bytes_written;
665 }
666 
667 static struct bsg_device *bsg_alloc_device(void)
668 {
669 	struct bsg_device *bd;
670 
671 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
672 	if (unlikely(!bd))
673 		return NULL;
674 
675 	spin_lock_init(&bd->lock);
676 
677 	bd->max_queue = BSG_DEFAULT_CMDS;
678 
679 	INIT_LIST_HEAD(&bd->busy_list);
680 	INIT_LIST_HEAD(&bd->done_list);
681 	INIT_HLIST_NODE(&bd->dev_list);
682 
683 	init_waitqueue_head(&bd->wq_free);
684 	init_waitqueue_head(&bd->wq_done);
685 	return bd;
686 }
687 
688 static void bsg_kref_release_function(struct kref *kref)
689 {
690 	struct bsg_class_device *bcd =
691 		container_of(kref, struct bsg_class_device, ref);
692 	struct device *parent = bcd->parent;
693 
694 	if (bcd->release)
695 		bcd->release(bcd->parent);
696 
697 	put_device(parent);
698 }
699 
700 static int bsg_put_device(struct bsg_device *bd)
701 {
702 	int ret = 0, do_free;
703 	struct request_queue *q = bd->queue;
704 
705 	mutex_lock(&bsg_mutex);
706 
707 	do_free = atomic_dec_and_test(&bd->ref_count);
708 	if (!do_free) {
709 		mutex_unlock(&bsg_mutex);
710 		goto out;
711 	}
712 
713 	hlist_del(&bd->dev_list);
714 	mutex_unlock(&bsg_mutex);
715 
716 	bsg_dbg(bd, "tearing down\n");
717 
718 	/*
719 	 * close can always block
720 	 */
721 	set_bit(BSG_F_BLOCK, &bd->flags);
722 
723 	/*
724 	 * correct error detection baddies here again. it's the responsibility
725 	 * of the app to properly reap commands before close() if it wants
726 	 * fool-proof error detection
727 	 */
728 	ret = bsg_complete_all_commands(bd);
729 
730 	kfree(bd);
731 out:
732 	kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
733 	if (do_free)
734 		blk_put_queue(q);
735 	return ret;
736 }
737 
738 static struct bsg_device *bsg_add_device(struct inode *inode,
739 					 struct request_queue *rq,
740 					 struct file *file)
741 {
742 	struct bsg_device *bd;
743 	unsigned char buf[32];
744 
745 	if (!blk_queue_scsi_passthrough(rq)) {
746 		WARN_ONCE(true, "Attempt to register a non-SCSI queue\n");
747 		return ERR_PTR(-EINVAL);
748 	}
749 
750 	if (!blk_get_queue(rq))
751 		return ERR_PTR(-ENXIO);
752 
753 	bd = bsg_alloc_device();
754 	if (!bd) {
755 		blk_put_queue(rq);
756 		return ERR_PTR(-ENOMEM);
757 	}
758 
759 	bd->queue = rq;
760 
761 	bsg_set_block(bd, file);
762 
763 	atomic_set(&bd->ref_count, 1);
764 	mutex_lock(&bsg_mutex);
765 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
766 
767 	strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
768 	bsg_dbg(bd, "bound to <%s>, max queue %d\n",
769 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
770 
771 	mutex_unlock(&bsg_mutex);
772 	return bd;
773 }
774 
775 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
776 {
777 	struct bsg_device *bd;
778 
779 	mutex_lock(&bsg_mutex);
780 
781 	hlist_for_each_entry(bd, bsg_dev_idx_hash(minor), dev_list) {
782 		if (bd->queue == q) {
783 			atomic_inc(&bd->ref_count);
784 			goto found;
785 		}
786 	}
787 	bd = NULL;
788 found:
789 	mutex_unlock(&bsg_mutex);
790 	return bd;
791 }
792 
793 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
794 {
795 	struct bsg_device *bd;
796 	struct bsg_class_device *bcd;
797 
798 	/*
799 	 * find the class device
800 	 */
801 	mutex_lock(&bsg_mutex);
802 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
803 	if (bcd)
804 		kref_get(&bcd->ref);
805 	mutex_unlock(&bsg_mutex);
806 
807 	if (!bcd)
808 		return ERR_PTR(-ENODEV);
809 
810 	bd = __bsg_get_device(iminor(inode), bcd->queue);
811 	if (bd)
812 		return bd;
813 
814 	bd = bsg_add_device(inode, bcd->queue, file);
815 	if (IS_ERR(bd))
816 		kref_put(&bcd->ref, bsg_kref_release_function);
817 
818 	return bd;
819 }
820 
821 static int bsg_open(struct inode *inode, struct file *file)
822 {
823 	struct bsg_device *bd;
824 
825 	bd = bsg_get_device(inode, file);
826 
827 	if (IS_ERR(bd))
828 		return PTR_ERR(bd);
829 
830 	file->private_data = bd;
831 	return 0;
832 }
833 
834 static int bsg_release(struct inode *inode, struct file *file)
835 {
836 	struct bsg_device *bd = file->private_data;
837 
838 	file->private_data = NULL;
839 	return bsg_put_device(bd);
840 }
841 
842 static __poll_t bsg_poll(struct file *file, poll_table *wait)
843 {
844 	struct bsg_device *bd = file->private_data;
845 	__poll_t mask = 0;
846 
847 	poll_wait(file, &bd->wq_done, wait);
848 	poll_wait(file, &bd->wq_free, wait);
849 
850 	spin_lock_irq(&bd->lock);
851 	if (!list_empty(&bd->done_list))
852 		mask |= EPOLLIN | EPOLLRDNORM;
853 	if (bd->queued_cmds < bd->max_queue)
854 		mask |= EPOLLOUT;
855 	spin_unlock_irq(&bd->lock);
856 
857 	return mask;
858 }
859 
860 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
861 {
862 	struct bsg_device *bd = file->private_data;
863 	int __user *uarg = (int __user *) arg;
864 	int ret;
865 
866 	switch (cmd) {
867 		/*
868 		 * our own ioctls
869 		 */
870 	case SG_GET_COMMAND_Q:
871 		return put_user(bd->max_queue, uarg);
872 	case SG_SET_COMMAND_Q: {
873 		int queue;
874 
875 		if (get_user(queue, uarg))
876 			return -EFAULT;
877 		if (queue < 1)
878 			return -EINVAL;
879 
880 		spin_lock_irq(&bd->lock);
881 		bd->max_queue = queue;
882 		spin_unlock_irq(&bd->lock);
883 		return 0;
884 	}
885 
886 	/*
887 	 * SCSI/sg ioctls
888 	 */
889 	case SG_GET_VERSION_NUM:
890 	case SCSI_IOCTL_GET_IDLUN:
891 	case SCSI_IOCTL_GET_BUS_NUMBER:
892 	case SG_SET_TIMEOUT:
893 	case SG_GET_TIMEOUT:
894 	case SG_GET_RESERVED_SIZE:
895 	case SG_SET_RESERVED_SIZE:
896 	case SG_EMULATED_HOST:
897 	case SCSI_IOCTL_SEND_COMMAND: {
898 		void __user *uarg = (void __user *) arg;
899 		return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
900 	}
901 	case SG_IO: {
902 		struct request *rq;
903 		struct bio *bio, *bidi_bio = NULL;
904 		struct sg_io_v4 hdr;
905 		int at_head;
906 
907 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
908 			return -EFAULT;
909 
910 		rq = bsg_map_hdr(bd, &hdr, file->f_mode);
911 		if (IS_ERR(rq))
912 			return PTR_ERR(rq);
913 
914 		bio = rq->bio;
915 		if (rq->next_rq)
916 			bidi_bio = rq->next_rq->bio;
917 
918 		at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
919 		blk_execute_rq(bd->queue, NULL, rq, at_head);
920 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
921 
922 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
923 			return -EFAULT;
924 
925 		return ret;
926 	}
927 	default:
928 		return -ENOTTY;
929 	}
930 }
931 
932 static const struct file_operations bsg_fops = {
933 	.read		=	bsg_read,
934 	.write		=	bsg_write,
935 	.poll		=	bsg_poll,
936 	.open		=	bsg_open,
937 	.release	=	bsg_release,
938 	.unlocked_ioctl	=	bsg_ioctl,
939 	.owner		=	THIS_MODULE,
940 	.llseek		=	default_llseek,
941 };
942 
943 void bsg_unregister_queue(struct request_queue *q)
944 {
945 	struct bsg_class_device *bcd = &q->bsg_dev;
946 
947 	if (!bcd->class_dev)
948 		return;
949 
950 	mutex_lock(&bsg_mutex);
951 	idr_remove(&bsg_minor_idr, bcd->minor);
952 	if (q->kobj.sd)
953 		sysfs_remove_link(&q->kobj, "bsg");
954 	device_unregister(bcd->class_dev);
955 	bcd->class_dev = NULL;
956 	kref_put(&bcd->ref, bsg_kref_release_function);
957 	mutex_unlock(&bsg_mutex);
958 }
959 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
960 
961 int bsg_register_queue(struct request_queue *q, struct device *parent,
962 		       const char *name, void (*release)(struct device *))
963 {
964 	struct bsg_class_device *bcd;
965 	dev_t dev;
966 	int ret;
967 	struct device *class_dev = NULL;
968 	const char *devname;
969 
970 	if (name)
971 		devname = name;
972 	else
973 		devname = dev_name(parent);
974 
975 	/*
976 	 * we need a proper transport to send commands, not a stacked device
977 	 */
978 	if (!queue_is_rq_based(q))
979 		return 0;
980 
981 	bcd = &q->bsg_dev;
982 	memset(bcd, 0, sizeof(*bcd));
983 
984 	mutex_lock(&bsg_mutex);
985 
986 	ret = idr_alloc(&bsg_minor_idr, bcd, 0, BSG_MAX_DEVS, GFP_KERNEL);
987 	if (ret < 0) {
988 		if (ret == -ENOSPC) {
989 			printk(KERN_ERR "bsg: too many bsg devices\n");
990 			ret = -EINVAL;
991 		}
992 		goto unlock;
993 	}
994 
995 	bcd->minor = ret;
996 	bcd->queue = q;
997 	bcd->parent = get_device(parent);
998 	bcd->release = release;
999 	kref_init(&bcd->ref);
1000 	dev = MKDEV(bsg_major, bcd->minor);
1001 	class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1002 	if (IS_ERR(class_dev)) {
1003 		ret = PTR_ERR(class_dev);
1004 		goto put_dev;
1005 	}
1006 	bcd->class_dev = class_dev;
1007 
1008 	if (q->kobj.sd) {
1009 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1010 		if (ret)
1011 			goto unregister_class_dev;
1012 	}
1013 
1014 	mutex_unlock(&bsg_mutex);
1015 	return 0;
1016 
1017 unregister_class_dev:
1018 	device_unregister(class_dev);
1019 put_dev:
1020 	put_device(parent);
1021 	idr_remove(&bsg_minor_idr, bcd->minor);
1022 unlock:
1023 	mutex_unlock(&bsg_mutex);
1024 	return ret;
1025 }
1026 EXPORT_SYMBOL_GPL(bsg_register_queue);
1027 
1028 static struct cdev bsg_cdev;
1029 
1030 static char *bsg_devnode(struct device *dev, umode_t *mode)
1031 {
1032 	return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1033 }
1034 
1035 static int __init bsg_init(void)
1036 {
1037 	int ret, i;
1038 	dev_t devid;
1039 
1040 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1041 				sizeof(struct bsg_command), 0, 0, NULL);
1042 	if (!bsg_cmd_cachep) {
1043 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1044 		return -ENOMEM;
1045 	}
1046 
1047 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1048 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1049 
1050 	bsg_class = class_create(THIS_MODULE, "bsg");
1051 	if (IS_ERR(bsg_class)) {
1052 		ret = PTR_ERR(bsg_class);
1053 		goto destroy_kmemcache;
1054 	}
1055 	bsg_class->devnode = bsg_devnode;
1056 
1057 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1058 	if (ret)
1059 		goto destroy_bsg_class;
1060 
1061 	bsg_major = MAJOR(devid);
1062 
1063 	cdev_init(&bsg_cdev, &bsg_fops);
1064 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1065 	if (ret)
1066 		goto unregister_chrdev;
1067 
1068 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1069 	       " loaded (major %d)\n", bsg_major);
1070 	return 0;
1071 unregister_chrdev:
1072 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1073 destroy_bsg_class:
1074 	class_destroy(bsg_class);
1075 destroy_kmemcache:
1076 	kmem_cache_destroy(bsg_cmd_cachep);
1077 	return ret;
1078 }
1079 
1080 MODULE_AUTHOR("Jens Axboe");
1081 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1082 MODULE_LICENSE("GPL");
1083 
1084 device_initcall(bsg_init);
1085