xref: /openbmc/linux/block/bsg.c (revision d7a3d85e)
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 struct bsg_device {
36 	struct request_queue *queue;
37 	spinlock_t lock;
38 	struct list_head busy_list;
39 	struct list_head done_list;
40 	struct hlist_node dev_list;
41 	atomic_t ref_count;
42 	int queued_cmds;
43 	int done_cmds;
44 	wait_queue_head_t wq_done;
45 	wait_queue_head_t wq_free;
46 	char name[20];
47 	int max_queue;
48 	unsigned long flags;
49 };
50 
51 enum {
52 	BSG_F_BLOCK		= 1,
53 };
54 
55 #define BSG_DEFAULT_CMDS	64
56 #define BSG_MAX_DEVS		32768
57 
58 #undef BSG_DEBUG
59 
60 #ifdef BSG_DEBUG
61 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __func__, ##args)
62 #else
63 #define dprintk(fmt, args...)
64 #endif
65 
66 static DEFINE_MUTEX(bsg_mutex);
67 static DEFINE_IDR(bsg_minor_idr);
68 
69 #define BSG_LIST_ARRAY_SIZE	8
70 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
71 
72 static struct class *bsg_class;
73 static int bsg_major;
74 
75 static struct kmem_cache *bsg_cmd_cachep;
76 
77 /*
78  * our internal command type
79  */
80 struct bsg_command {
81 	struct bsg_device *bd;
82 	struct list_head list;
83 	struct request *rq;
84 	struct bio *bio;
85 	struct bio *bidi_bio;
86 	int err;
87 	struct sg_io_v4 hdr;
88 	char sense[SCSI_SENSE_BUFFERSIZE];
89 };
90 
91 static void bsg_free_command(struct bsg_command *bc)
92 {
93 	struct bsg_device *bd = bc->bd;
94 	unsigned long flags;
95 
96 	kmem_cache_free(bsg_cmd_cachep, bc);
97 
98 	spin_lock_irqsave(&bd->lock, flags);
99 	bd->queued_cmds--;
100 	spin_unlock_irqrestore(&bd->lock, flags);
101 
102 	wake_up(&bd->wq_free);
103 }
104 
105 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
106 {
107 	struct bsg_command *bc = ERR_PTR(-EINVAL);
108 
109 	spin_lock_irq(&bd->lock);
110 
111 	if (bd->queued_cmds >= bd->max_queue)
112 		goto out;
113 
114 	bd->queued_cmds++;
115 	spin_unlock_irq(&bd->lock);
116 
117 	bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
118 	if (unlikely(!bc)) {
119 		spin_lock_irq(&bd->lock);
120 		bd->queued_cmds--;
121 		bc = ERR_PTR(-ENOMEM);
122 		goto out;
123 	}
124 
125 	bc->bd = bd;
126 	INIT_LIST_HEAD(&bc->list);
127 	dprintk("%s: returning free cmd %p\n", bd->name, bc);
128 	return bc;
129 out:
130 	spin_unlock_irq(&bd->lock);
131 	return bc;
132 }
133 
134 static inline struct hlist_head *bsg_dev_idx_hash(int index)
135 {
136 	return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
137 }
138 
139 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
140 				struct sg_io_v4 *hdr, struct bsg_device *bd,
141 				fmode_t has_write_perm)
142 {
143 	if (hdr->request_len > BLK_MAX_CDB) {
144 		rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
145 		if (!rq->cmd)
146 			return -ENOMEM;
147 	}
148 
149 	if (copy_from_user(rq->cmd, (void __user *)(unsigned long)hdr->request,
150 			   hdr->request_len))
151 		return -EFAULT;
152 
153 	if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
154 		if (blk_verify_command(rq->cmd, has_write_perm))
155 			return -EPERM;
156 	} else if (!capable(CAP_SYS_RAWIO))
157 		return -EPERM;
158 
159 	/*
160 	 * fill in request structure
161 	 */
162 	rq->cmd_len = hdr->request_len;
163 
164 	rq->timeout = msecs_to_jiffies(hdr->timeout);
165 	if (!rq->timeout)
166 		rq->timeout = q->sg_timeout;
167 	if (!rq->timeout)
168 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
169 	if (rq->timeout < BLK_MIN_SG_TIMEOUT)
170 		rq->timeout = BLK_MIN_SG_TIMEOUT;
171 
172 	return 0;
173 }
174 
175 /*
176  * Check if sg_io_v4 from user is allowed and valid
177  */
178 static int
179 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
180 {
181 	int ret = 0;
182 
183 	if (hdr->guard != 'Q')
184 		return -EINVAL;
185 
186 	switch (hdr->protocol) {
187 	case BSG_PROTOCOL_SCSI:
188 		switch (hdr->subprotocol) {
189 		case BSG_SUB_PROTOCOL_SCSI_CMD:
190 		case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
191 			break;
192 		default:
193 			ret = -EINVAL;
194 		}
195 		break;
196 	default:
197 		ret = -EINVAL;
198 	}
199 
200 	*rw = hdr->dout_xfer_len ? WRITE : READ;
201 	return ret;
202 }
203 
204 /*
205  * map sg_io_v4 to a request.
206  */
207 static struct request *
208 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
209 	    u8 *sense)
210 {
211 	struct request_queue *q = bd->queue;
212 	struct request *rq, *next_rq = NULL;
213 	int ret, rw;
214 	unsigned int dxfer_len;
215 	void __user *dxferp = NULL;
216 	struct bsg_class_device *bcd = &q->bsg_dev;
217 
218 	/* if the LLD has been removed then the bsg_unregister_queue will
219 	 * eventually be called and the class_dev was freed, so we can no
220 	 * longer use this request_queue. Return no such address.
221 	 */
222 	if (!bcd->class_dev)
223 		return ERR_PTR(-ENXIO);
224 
225 	dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
226 		hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
227 		hdr->din_xfer_len);
228 
229 	ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
230 	if (ret)
231 		return ERR_PTR(ret);
232 
233 	/*
234 	 * map scatter-gather elements separately and string them to request
235 	 */
236 	rq = blk_get_request(q, rw, GFP_KERNEL);
237 	if (IS_ERR(rq))
238 		return rq;
239 	blk_rq_set_block_pc(rq);
240 
241 	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
242 	if (ret)
243 		goto out;
244 
245 	if (rw == WRITE && 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, READ, GFP_KERNEL);
252 		if (IS_ERR(next_rq)) {
253 			ret = PTR_ERR(next_rq);
254 			next_rq = NULL;
255 			goto out;
256 		}
257 		rq->next_rq = next_rq;
258 		next_rq->cmd_type = rq->cmd_type;
259 
260 		dxferp = (void __user *)(unsigned long)hdr->din_xferp;
261 		ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
262 				       hdr->din_xfer_len, GFP_KERNEL);
263 		if (ret)
264 			goto out;
265 	}
266 
267 	if (hdr->dout_xfer_len) {
268 		dxfer_len = hdr->dout_xfer_len;
269 		dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
270 	} else if (hdr->din_xfer_len) {
271 		dxfer_len = hdr->din_xfer_len;
272 		dxferp = (void __user *)(unsigned long)hdr->din_xferp;
273 	} else
274 		dxfer_len = 0;
275 
276 	if (dxfer_len) {
277 		ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
278 				      GFP_KERNEL);
279 		if (ret)
280 			goto out;
281 	}
282 
283 	rq->sense = sense;
284 	rq->sense_len = 0;
285 
286 	return rq;
287 out:
288 	if (rq->cmd != rq->__cmd)
289 		kfree(rq->cmd);
290 	blk_put_request(rq);
291 	if (next_rq) {
292 		blk_rq_unmap_user(next_rq->bio);
293 		blk_put_request(next_rq);
294 	}
295 	return ERR_PTR(ret);
296 }
297 
298 /*
299  * async completion call-back from the block layer, when scsi/ide/whatever
300  * calls end_that_request_last() on a request
301  */
302 static void bsg_rq_end_io(struct request *rq, int uptodate)
303 {
304 	struct bsg_command *bc = rq->end_io_data;
305 	struct bsg_device *bd = bc->bd;
306 	unsigned long flags;
307 
308 	dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
309 		bd->name, rq, bc, bc->bio, uptodate);
310 
311 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
312 
313 	spin_lock_irqsave(&bd->lock, flags);
314 	list_move_tail(&bc->list, &bd->done_list);
315 	bd->done_cmds++;
316 	spin_unlock_irqrestore(&bd->lock, flags);
317 
318 	wake_up(&bd->wq_done);
319 }
320 
321 /*
322  * do final setup of a 'bc' and submit the matching 'rq' to the block
323  * layer for io
324  */
325 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
326 			    struct bsg_command *bc, struct request *rq)
327 {
328 	int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
329 
330 	/*
331 	 * add bc command to busy queue and submit rq for io
332 	 */
333 	bc->rq = rq;
334 	bc->bio = rq->bio;
335 	if (rq->next_rq)
336 		bc->bidi_bio = rq->next_rq->bio;
337 	bc->hdr.duration = jiffies;
338 	spin_lock_irq(&bd->lock);
339 	list_add_tail(&bc->list, &bd->busy_list);
340 	spin_unlock_irq(&bd->lock);
341 
342 	dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
343 
344 	rq->end_io_data = bc;
345 	blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
346 }
347 
348 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
349 {
350 	struct bsg_command *bc = NULL;
351 
352 	spin_lock_irq(&bd->lock);
353 	if (bd->done_cmds) {
354 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
355 		list_del(&bc->list);
356 		bd->done_cmds--;
357 	}
358 	spin_unlock_irq(&bd->lock);
359 
360 	return bc;
361 }
362 
363 /*
364  * Get a finished command from the done list
365  */
366 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
367 {
368 	struct bsg_command *bc;
369 	int ret;
370 
371 	do {
372 		bc = bsg_next_done_cmd(bd);
373 		if (bc)
374 			break;
375 
376 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
377 			bc = ERR_PTR(-EAGAIN);
378 			break;
379 		}
380 
381 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
382 		if (ret) {
383 			bc = ERR_PTR(-ERESTARTSYS);
384 			break;
385 		}
386 	} while (1);
387 
388 	dprintk("%s: returning done %p\n", bd->name, bc);
389 
390 	return bc;
391 }
392 
393 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
394 				    struct bio *bio, struct bio *bidi_bio)
395 {
396 	int ret = 0;
397 
398 	dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
399 	/*
400 	 * fill in all the output members
401 	 */
402 	hdr->device_status = rq->errors & 0xff;
403 	hdr->transport_status = host_byte(rq->errors);
404 	hdr->driver_status = driver_byte(rq->errors);
405 	hdr->info = 0;
406 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
407 		hdr->info |= SG_INFO_CHECK;
408 	hdr->response_len = 0;
409 
410 	if (rq->sense_len && hdr->response) {
411 		int len = min_t(unsigned int, hdr->max_response_len,
412 					rq->sense_len);
413 
414 		ret = copy_to_user((void __user *)(unsigned long)hdr->response,
415 				   rq->sense, len);
416 		if (!ret)
417 			hdr->response_len = len;
418 		else
419 			ret = -EFAULT;
420 	}
421 
422 	if (rq->next_rq) {
423 		hdr->dout_resid = rq->resid_len;
424 		hdr->din_resid = rq->next_rq->resid_len;
425 		blk_rq_unmap_user(bidi_bio);
426 		blk_put_request(rq->next_rq);
427 	} else if (rq_data_dir(rq) == READ)
428 		hdr->din_resid = rq->resid_len;
429 	else
430 		hdr->dout_resid = rq->resid_len;
431 
432 	/*
433 	 * If the request generated a negative error number, return it
434 	 * (providing we aren't already returning an error); if it's
435 	 * just a protocol response (i.e. non negative), that gets
436 	 * processed above.
437 	 */
438 	if (!ret && rq->errors < 0)
439 		ret = rq->errors;
440 
441 	blk_rq_unmap_user(bio);
442 	if (rq->cmd != rq->__cmd)
443 		kfree(rq->cmd);
444 	blk_put_request(rq);
445 
446 	return ret;
447 }
448 
449 static bool bsg_complete(struct bsg_device *bd)
450 {
451 	bool ret = false;
452 	bool spin;
453 
454 	do {
455 		spin_lock_irq(&bd->lock);
456 
457 		BUG_ON(bd->done_cmds > bd->queued_cmds);
458 
459 		/*
460 		 * All commands consumed.
461 		 */
462 		if (bd->done_cmds == bd->queued_cmds)
463 			ret = true;
464 
465 		spin = !test_bit(BSG_F_BLOCK, &bd->flags);
466 
467 		spin_unlock_irq(&bd->lock);
468 	} while (!ret && spin);
469 
470 	return ret;
471 }
472 
473 static int bsg_complete_all_commands(struct bsg_device *bd)
474 {
475 	struct bsg_command *bc;
476 	int ret, tret;
477 
478 	dprintk("%s: entered\n", bd->name);
479 
480 	/*
481 	 * wait for all commands to complete
482 	 */
483 	io_wait_event(bd->wq_done, bsg_complete(bd));
484 
485 	/*
486 	 * discard done commands
487 	 */
488 	ret = 0;
489 	do {
490 		spin_lock_irq(&bd->lock);
491 		if (!bd->queued_cmds) {
492 			spin_unlock_irq(&bd->lock);
493 			break;
494 		}
495 		spin_unlock_irq(&bd->lock);
496 
497 		bc = bsg_get_done_cmd(bd);
498 		if (IS_ERR(bc))
499 			break;
500 
501 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
502 						bc->bidi_bio);
503 		if (!ret)
504 			ret = tret;
505 
506 		bsg_free_command(bc);
507 	} while (1);
508 
509 	return ret;
510 }
511 
512 static int
513 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
514 	   const struct iovec *iov, ssize_t *bytes_read)
515 {
516 	struct bsg_command *bc;
517 	int nr_commands, ret;
518 
519 	if (count % sizeof(struct sg_io_v4))
520 		return -EINVAL;
521 
522 	ret = 0;
523 	nr_commands = count / sizeof(struct sg_io_v4);
524 	while (nr_commands) {
525 		bc = bsg_get_done_cmd(bd);
526 		if (IS_ERR(bc)) {
527 			ret = PTR_ERR(bc);
528 			break;
529 		}
530 
531 		/*
532 		 * this is the only case where we need to copy data back
533 		 * after completing the request. so do that here,
534 		 * bsg_complete_work() cannot do that for us
535 		 */
536 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
537 					       bc->bidi_bio);
538 
539 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
540 			ret = -EFAULT;
541 
542 		bsg_free_command(bc);
543 
544 		if (ret)
545 			break;
546 
547 		buf += sizeof(struct sg_io_v4);
548 		*bytes_read += sizeof(struct sg_io_v4);
549 		nr_commands--;
550 	}
551 
552 	return ret;
553 }
554 
555 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
556 {
557 	if (file->f_flags & O_NONBLOCK)
558 		clear_bit(BSG_F_BLOCK, &bd->flags);
559 	else
560 		set_bit(BSG_F_BLOCK, &bd->flags);
561 }
562 
563 /*
564  * Check if the error is a "real" error that we should return.
565  */
566 static inline int err_block_err(int ret)
567 {
568 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
569 		return 1;
570 
571 	return 0;
572 }
573 
574 static ssize_t
575 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
576 {
577 	struct bsg_device *bd = file->private_data;
578 	int ret;
579 	ssize_t bytes_read;
580 
581 	dprintk("%s: read %Zd bytes\n", bd->name, count);
582 
583 	bsg_set_block(bd, file);
584 
585 	bytes_read = 0;
586 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
587 	*ppos = bytes_read;
588 
589 	if (!bytes_read || err_block_err(ret))
590 		bytes_read = ret;
591 
592 	return bytes_read;
593 }
594 
595 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
596 		       size_t count, ssize_t *bytes_written,
597 		       fmode_t has_write_perm)
598 {
599 	struct bsg_command *bc;
600 	struct request *rq;
601 	int ret, nr_commands;
602 
603 	if (count % sizeof(struct sg_io_v4))
604 		return -EINVAL;
605 
606 	nr_commands = count / sizeof(struct sg_io_v4);
607 	rq = NULL;
608 	bc = NULL;
609 	ret = 0;
610 	while (nr_commands) {
611 		struct request_queue *q = bd->queue;
612 
613 		bc = bsg_alloc_command(bd);
614 		if (IS_ERR(bc)) {
615 			ret = PTR_ERR(bc);
616 			bc = NULL;
617 			break;
618 		}
619 
620 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
621 			ret = -EFAULT;
622 			break;
623 		}
624 
625 		/*
626 		 * get a request, fill in the blanks, and add to request queue
627 		 */
628 		rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
629 		if (IS_ERR(rq)) {
630 			ret = PTR_ERR(rq);
631 			rq = NULL;
632 			break;
633 		}
634 
635 		bsg_add_command(bd, q, bc, rq);
636 		bc = NULL;
637 		rq = NULL;
638 		nr_commands--;
639 		buf += sizeof(struct sg_io_v4);
640 		*bytes_written += sizeof(struct sg_io_v4);
641 	}
642 
643 	if (bc)
644 		bsg_free_command(bc);
645 
646 	return ret;
647 }
648 
649 static ssize_t
650 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
651 {
652 	struct bsg_device *bd = file->private_data;
653 	ssize_t bytes_written;
654 	int ret;
655 
656 	dprintk("%s: write %Zd bytes\n", bd->name, count);
657 
658 	bsg_set_block(bd, file);
659 
660 	bytes_written = 0;
661 	ret = __bsg_write(bd, buf, count, &bytes_written,
662 			  file->f_mode & FMODE_WRITE);
663 
664 	*ppos = bytes_written;
665 
666 	/*
667 	 * return bytes written on non-fatal errors
668 	 */
669 	if (!bytes_written || err_block_err(ret))
670 		bytes_written = ret;
671 
672 	dprintk("%s: returning %Zd\n", bd->name, bytes_written);
673 	return bytes_written;
674 }
675 
676 static struct bsg_device *bsg_alloc_device(void)
677 {
678 	struct bsg_device *bd;
679 
680 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
681 	if (unlikely(!bd))
682 		return NULL;
683 
684 	spin_lock_init(&bd->lock);
685 
686 	bd->max_queue = BSG_DEFAULT_CMDS;
687 
688 	INIT_LIST_HEAD(&bd->busy_list);
689 	INIT_LIST_HEAD(&bd->done_list);
690 	INIT_HLIST_NODE(&bd->dev_list);
691 
692 	init_waitqueue_head(&bd->wq_free);
693 	init_waitqueue_head(&bd->wq_done);
694 	return bd;
695 }
696 
697 static void bsg_kref_release_function(struct kref *kref)
698 {
699 	struct bsg_class_device *bcd =
700 		container_of(kref, struct bsg_class_device, ref);
701 	struct device *parent = bcd->parent;
702 
703 	if (bcd->release)
704 		bcd->release(bcd->parent);
705 
706 	put_device(parent);
707 }
708 
709 static int bsg_put_device(struct bsg_device *bd)
710 {
711 	int ret = 0, do_free;
712 	struct request_queue *q = bd->queue;
713 
714 	mutex_lock(&bsg_mutex);
715 
716 	do_free = atomic_dec_and_test(&bd->ref_count);
717 	if (!do_free) {
718 		mutex_unlock(&bsg_mutex);
719 		goto out;
720 	}
721 
722 	hlist_del(&bd->dev_list);
723 	mutex_unlock(&bsg_mutex);
724 
725 	dprintk("%s: tearing down\n", bd->name);
726 
727 	/*
728 	 * close can always block
729 	 */
730 	set_bit(BSG_F_BLOCK, &bd->flags);
731 
732 	/*
733 	 * correct error detection baddies here again. it's the responsibility
734 	 * of the app to properly reap commands before close() if it wants
735 	 * fool-proof error detection
736 	 */
737 	ret = bsg_complete_all_commands(bd);
738 
739 	kfree(bd);
740 out:
741 	kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
742 	if (do_free)
743 		blk_put_queue(q);
744 	return ret;
745 }
746 
747 static struct bsg_device *bsg_add_device(struct inode *inode,
748 					 struct request_queue *rq,
749 					 struct file *file)
750 {
751 	struct bsg_device *bd;
752 #ifdef BSG_DEBUG
753 	unsigned char buf[32];
754 #endif
755 	if (!blk_get_queue(rq))
756 		return ERR_PTR(-ENXIO);
757 
758 	bd = bsg_alloc_device();
759 	if (!bd) {
760 		blk_put_queue(rq);
761 		return ERR_PTR(-ENOMEM);
762 	}
763 
764 	bd->queue = rq;
765 
766 	bsg_set_block(bd, file);
767 
768 	atomic_set(&bd->ref_count, 1);
769 	mutex_lock(&bsg_mutex);
770 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
771 
772 	strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
773 	dprintk("bound to <%s>, max queue %d\n",
774 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
775 
776 	mutex_unlock(&bsg_mutex);
777 	return bd;
778 }
779 
780 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
781 {
782 	struct bsg_device *bd;
783 
784 	mutex_lock(&bsg_mutex);
785 
786 	hlist_for_each_entry(bd, bsg_dev_idx_hash(minor), dev_list) {
787 		if (bd->queue == q) {
788 			atomic_inc(&bd->ref_count);
789 			goto found;
790 		}
791 	}
792 	bd = NULL;
793 found:
794 	mutex_unlock(&bsg_mutex);
795 	return bd;
796 }
797 
798 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
799 {
800 	struct bsg_device *bd;
801 	struct bsg_class_device *bcd;
802 
803 	/*
804 	 * find the class device
805 	 */
806 	mutex_lock(&bsg_mutex);
807 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
808 	if (bcd)
809 		kref_get(&bcd->ref);
810 	mutex_unlock(&bsg_mutex);
811 
812 	if (!bcd)
813 		return ERR_PTR(-ENODEV);
814 
815 	bd = __bsg_get_device(iminor(inode), bcd->queue);
816 	if (bd)
817 		return bd;
818 
819 	bd = bsg_add_device(inode, bcd->queue, file);
820 	if (IS_ERR(bd))
821 		kref_put(&bcd->ref, bsg_kref_release_function);
822 
823 	return bd;
824 }
825 
826 static int bsg_open(struct inode *inode, struct file *file)
827 {
828 	struct bsg_device *bd;
829 
830 	bd = bsg_get_device(inode, file);
831 
832 	if (IS_ERR(bd))
833 		return PTR_ERR(bd);
834 
835 	file->private_data = bd;
836 	return 0;
837 }
838 
839 static int bsg_release(struct inode *inode, struct file *file)
840 {
841 	struct bsg_device *bd = file->private_data;
842 
843 	file->private_data = NULL;
844 	return bsg_put_device(bd);
845 }
846 
847 static unsigned int bsg_poll(struct file *file, poll_table *wait)
848 {
849 	struct bsg_device *bd = file->private_data;
850 	unsigned int mask = 0;
851 
852 	poll_wait(file, &bd->wq_done, wait);
853 	poll_wait(file, &bd->wq_free, wait);
854 
855 	spin_lock_irq(&bd->lock);
856 	if (!list_empty(&bd->done_list))
857 		mask |= POLLIN | POLLRDNORM;
858 	if (bd->queued_cmds < bd->max_queue)
859 		mask |= POLLOUT;
860 	spin_unlock_irq(&bd->lock);
861 
862 	return mask;
863 }
864 
865 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
866 {
867 	struct bsg_device *bd = file->private_data;
868 	int __user *uarg = (int __user *) arg;
869 	int ret;
870 
871 	switch (cmd) {
872 		/*
873 		 * our own ioctls
874 		 */
875 	case SG_GET_COMMAND_Q:
876 		return put_user(bd->max_queue, uarg);
877 	case SG_SET_COMMAND_Q: {
878 		int queue;
879 
880 		if (get_user(queue, uarg))
881 			return -EFAULT;
882 		if (queue < 1)
883 			return -EINVAL;
884 
885 		spin_lock_irq(&bd->lock);
886 		bd->max_queue = queue;
887 		spin_unlock_irq(&bd->lock);
888 		return 0;
889 	}
890 
891 	/*
892 	 * SCSI/sg ioctls
893 	 */
894 	case SG_GET_VERSION_NUM:
895 	case SCSI_IOCTL_GET_IDLUN:
896 	case SCSI_IOCTL_GET_BUS_NUMBER:
897 	case SG_SET_TIMEOUT:
898 	case SG_GET_TIMEOUT:
899 	case SG_GET_RESERVED_SIZE:
900 	case SG_SET_RESERVED_SIZE:
901 	case SG_EMULATED_HOST:
902 	case SCSI_IOCTL_SEND_COMMAND: {
903 		void __user *uarg = (void __user *) arg;
904 		return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
905 	}
906 	case SG_IO: {
907 		struct request *rq;
908 		struct bio *bio, *bidi_bio = NULL;
909 		struct sg_io_v4 hdr;
910 		int at_head;
911 		u8 sense[SCSI_SENSE_BUFFERSIZE];
912 
913 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
914 			return -EFAULT;
915 
916 		rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
917 		if (IS_ERR(rq))
918 			return PTR_ERR(rq);
919 
920 		bio = rq->bio;
921 		if (rq->next_rq)
922 			bidi_bio = rq->next_rq->bio;
923 
924 		at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
925 		blk_execute_rq(bd->queue, NULL, rq, at_head);
926 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
927 
928 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
929 			return -EFAULT;
930 
931 		return ret;
932 	}
933 	/*
934 	 * block device ioctls
935 	 */
936 	default:
937 #if 0
938 		return ioctl_by_bdev(bd->bdev, cmd, arg);
939 #else
940 		return -ENOTTY;
941 #endif
942 	}
943 }
944 
945 static const struct file_operations bsg_fops = {
946 	.read		=	bsg_read,
947 	.write		=	bsg_write,
948 	.poll		=	bsg_poll,
949 	.open		=	bsg_open,
950 	.release	=	bsg_release,
951 	.unlocked_ioctl	=	bsg_ioctl,
952 	.owner		=	THIS_MODULE,
953 	.llseek		=	default_llseek,
954 };
955 
956 void bsg_unregister_queue(struct request_queue *q)
957 {
958 	struct bsg_class_device *bcd = &q->bsg_dev;
959 
960 	if (!bcd->class_dev)
961 		return;
962 
963 	mutex_lock(&bsg_mutex);
964 	idr_remove(&bsg_minor_idr, bcd->minor);
965 	if (q->kobj.sd)
966 		sysfs_remove_link(&q->kobj, "bsg");
967 	device_unregister(bcd->class_dev);
968 	bcd->class_dev = NULL;
969 	kref_put(&bcd->ref, bsg_kref_release_function);
970 	mutex_unlock(&bsg_mutex);
971 }
972 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
973 
974 int bsg_register_queue(struct request_queue *q, struct device *parent,
975 		       const char *name, void (*release)(struct device *))
976 {
977 	struct bsg_class_device *bcd;
978 	dev_t dev;
979 	int ret;
980 	struct device *class_dev = NULL;
981 	const char *devname;
982 
983 	if (name)
984 		devname = name;
985 	else
986 		devname = dev_name(parent);
987 
988 	/*
989 	 * we need a proper transport to send commands, not a stacked device
990 	 */
991 	if (!queue_is_rq_based(q))
992 		return 0;
993 
994 	bcd = &q->bsg_dev;
995 	memset(bcd, 0, sizeof(*bcd));
996 
997 	mutex_lock(&bsg_mutex);
998 
999 	ret = idr_alloc(&bsg_minor_idr, bcd, 0, BSG_MAX_DEVS, GFP_KERNEL);
1000 	if (ret < 0) {
1001 		if (ret == -ENOSPC) {
1002 			printk(KERN_ERR "bsg: too many bsg devices\n");
1003 			ret = -EINVAL;
1004 		}
1005 		goto unlock;
1006 	}
1007 
1008 	bcd->minor = ret;
1009 	bcd->queue = q;
1010 	bcd->parent = get_device(parent);
1011 	bcd->release = release;
1012 	kref_init(&bcd->ref);
1013 	dev = MKDEV(bsg_major, bcd->minor);
1014 	class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1015 	if (IS_ERR(class_dev)) {
1016 		ret = PTR_ERR(class_dev);
1017 		goto put_dev;
1018 	}
1019 	bcd->class_dev = class_dev;
1020 
1021 	if (q->kobj.sd) {
1022 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1023 		if (ret)
1024 			goto unregister_class_dev;
1025 	}
1026 
1027 	mutex_unlock(&bsg_mutex);
1028 	return 0;
1029 
1030 unregister_class_dev:
1031 	device_unregister(class_dev);
1032 put_dev:
1033 	put_device(parent);
1034 	idr_remove(&bsg_minor_idr, bcd->minor);
1035 unlock:
1036 	mutex_unlock(&bsg_mutex);
1037 	return ret;
1038 }
1039 EXPORT_SYMBOL_GPL(bsg_register_queue);
1040 
1041 static struct cdev bsg_cdev;
1042 
1043 static char *bsg_devnode(struct device *dev, umode_t *mode)
1044 {
1045 	return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1046 }
1047 
1048 static int __init bsg_init(void)
1049 {
1050 	int ret, i;
1051 	dev_t devid;
1052 
1053 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1054 				sizeof(struct bsg_command), 0, 0, NULL);
1055 	if (!bsg_cmd_cachep) {
1056 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1057 		return -ENOMEM;
1058 	}
1059 
1060 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1061 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1062 
1063 	bsg_class = class_create(THIS_MODULE, "bsg");
1064 	if (IS_ERR(bsg_class)) {
1065 		ret = PTR_ERR(bsg_class);
1066 		goto destroy_kmemcache;
1067 	}
1068 	bsg_class->devnode = bsg_devnode;
1069 
1070 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1071 	if (ret)
1072 		goto destroy_bsg_class;
1073 
1074 	bsg_major = MAJOR(devid);
1075 
1076 	cdev_init(&bsg_cdev, &bsg_fops);
1077 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1078 	if (ret)
1079 		goto unregister_chrdev;
1080 
1081 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1082 	       " loaded (major %d)\n", bsg_major);
1083 	return 0;
1084 unregister_chrdev:
1085 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1086 destroy_bsg_class:
1087 	class_destroy(bsg_class);
1088 destroy_kmemcache:
1089 	kmem_cache_destroy(bsg_cmd_cachep);
1090 	return ret;
1091 }
1092 
1093 MODULE_AUTHOR("Jens Axboe");
1094 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1095 MODULE_LICENSE("GPL");
1096 
1097 device_initcall(bsg_init);
1098