xref: /linux/block/bsg.c (revision c0c9209ddd96bc4f1d70a8b9958710671e076080)
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/percpu.h>
19 #include <linux/uio.h>
20 #include <linux/idr.h>
21 #include <linux/bsg.h>
22 #include <linux/smp_lock.h>
23 
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_ioctl.h>
26 #include <scsi/scsi_cmnd.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_driver.h>
29 #include <scsi/sg.h>
30 
31 #define BSG_DESCRIPTION	"Block layer SCSI generic (bsg) driver"
32 #define BSG_VERSION	"0.4"
33 
34 struct bsg_device {
35 	struct request_queue *queue;
36 	spinlock_t lock;
37 	struct list_head busy_list;
38 	struct list_head done_list;
39 	struct hlist_node dev_list;
40 	atomic_t ref_count;
41 	int queued_cmds;
42 	int done_cmds;
43 	wait_queue_head_t wq_done;
44 	wait_queue_head_t wq_free;
45 	char name[BUS_ID_SIZE];
46 	int max_queue;
47 	unsigned long flags;
48 };
49 
50 enum {
51 	BSG_F_BLOCK		= 1,
52 };
53 
54 #define BSG_DEFAULT_CMDS	64
55 #define BSG_MAX_DEVS		32768
56 
57 #undef BSG_DEBUG
58 
59 #ifdef BSG_DEBUG
60 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __func__, ##args)
61 #else
62 #define dprintk(fmt, args...)
63 #endif
64 
65 static DEFINE_MUTEX(bsg_mutex);
66 static DEFINE_IDR(bsg_minor_idr);
67 
68 #define BSG_LIST_ARRAY_SIZE	8
69 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
70 
71 static struct class *bsg_class;
72 static int bsg_major;
73 
74 static struct kmem_cache *bsg_cmd_cachep;
75 
76 /*
77  * our internal command type
78  */
79 struct bsg_command {
80 	struct bsg_device *bd;
81 	struct list_head list;
82 	struct request *rq;
83 	struct bio *bio;
84 	struct bio *bidi_bio;
85 	int err;
86 	struct sg_io_v4 hdr;
87 	char sense[SCSI_SENSE_BUFFERSIZE];
88 };
89 
90 static void bsg_free_command(struct bsg_command *bc)
91 {
92 	struct bsg_device *bd = bc->bd;
93 	unsigned long flags;
94 
95 	kmem_cache_free(bsg_cmd_cachep, bc);
96 
97 	spin_lock_irqsave(&bd->lock, flags);
98 	bd->queued_cmds--;
99 	spin_unlock_irqrestore(&bd->lock, flags);
100 
101 	wake_up(&bd->wq_free);
102 }
103 
104 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
105 {
106 	struct bsg_command *bc = ERR_PTR(-EINVAL);
107 
108 	spin_lock_irq(&bd->lock);
109 
110 	if (bd->queued_cmds >= bd->max_queue)
111 		goto out;
112 
113 	bd->queued_cmds++;
114 	spin_unlock_irq(&bd->lock);
115 
116 	bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
117 	if (unlikely(!bc)) {
118 		spin_lock_irq(&bd->lock);
119 		bd->queued_cmds--;
120 		bc = ERR_PTR(-ENOMEM);
121 		goto out;
122 	}
123 
124 	bc->bd = bd;
125 	INIT_LIST_HEAD(&bc->list);
126 	dprintk("%s: returning free cmd %p\n", bd->name, bc);
127 	return bc;
128 out:
129 	spin_unlock_irq(&bd->lock);
130 	return bc;
131 }
132 
133 static inline struct hlist_head *bsg_dev_idx_hash(int index)
134 {
135 	return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
136 }
137 
138 static int bsg_io_schedule(struct bsg_device *bd)
139 {
140 	DEFINE_WAIT(wait);
141 	int ret = 0;
142 
143 	spin_lock_irq(&bd->lock);
144 
145 	BUG_ON(bd->done_cmds > bd->queued_cmds);
146 
147 	/*
148 	 * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
149 	 * work to do", even though we return -ENOSPC after this same test
150 	 * during bsg_write() -- there, it means our buffer can't have more
151 	 * bsg_commands added to it, thus has no space left.
152 	 */
153 	if (bd->done_cmds == bd->queued_cmds) {
154 		ret = -ENODATA;
155 		goto unlock;
156 	}
157 
158 	if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
159 		ret = -EAGAIN;
160 		goto unlock;
161 	}
162 
163 	prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
164 	spin_unlock_irq(&bd->lock);
165 	io_schedule();
166 	finish_wait(&bd->wq_done, &wait);
167 
168 	return ret;
169 unlock:
170 	spin_unlock_irq(&bd->lock);
171 	return ret;
172 }
173 
174 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
175 				struct sg_io_v4 *hdr, struct bsg_device *bd,
176 				int has_write_perm)
177 {
178 	if (hdr->request_len > BLK_MAX_CDB) {
179 		rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
180 		if (!rq->cmd)
181 			return -ENOMEM;
182 	}
183 
184 	if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
185 			   hdr->request_len))
186 		return -EFAULT;
187 
188 	if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
189 		if (blk_verify_command(&q->cmd_filter, rq->cmd, has_write_perm))
190 			return -EPERM;
191 	} else if (!capable(CAP_SYS_RAWIO))
192 		return -EPERM;
193 
194 	/*
195 	 * fill in request structure
196 	 */
197 	rq->cmd_len = hdr->request_len;
198 	rq->cmd_type = REQ_TYPE_BLOCK_PC;
199 
200 	rq->timeout = (hdr->timeout * HZ) / 1000;
201 	if (!rq->timeout)
202 		rq->timeout = q->sg_timeout;
203 	if (!rq->timeout)
204 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
205 
206 	return 0;
207 }
208 
209 /*
210  * Check if sg_io_v4 from user is allowed and valid
211  */
212 static int
213 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
214 {
215 	int ret = 0;
216 
217 	if (hdr->guard != 'Q')
218 		return -EINVAL;
219 	if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
220 	    hdr->din_xfer_len > (q->max_sectors << 9))
221 		return -EIO;
222 
223 	switch (hdr->protocol) {
224 	case BSG_PROTOCOL_SCSI:
225 		switch (hdr->subprotocol) {
226 		case BSG_SUB_PROTOCOL_SCSI_CMD:
227 		case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
228 			break;
229 		default:
230 			ret = -EINVAL;
231 		}
232 		break;
233 	default:
234 		ret = -EINVAL;
235 	}
236 
237 	*rw = hdr->dout_xfer_len ? WRITE : READ;
238 	return ret;
239 }
240 
241 /*
242  * map sg_io_v4 to a request.
243  */
244 static struct request *
245 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, int has_write_perm)
246 {
247 	struct request_queue *q = bd->queue;
248 	struct request *rq, *next_rq = NULL;
249 	int ret, rw;
250 	unsigned int dxfer_len;
251 	void *dxferp = NULL;
252 
253 	dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
254 		hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
255 		hdr->din_xfer_len);
256 
257 	ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
258 	if (ret)
259 		return ERR_PTR(ret);
260 
261 	/*
262 	 * map scatter-gather elements seperately and string them to request
263 	 */
264 	rq = blk_get_request(q, rw, GFP_KERNEL);
265 	if (!rq)
266 		return ERR_PTR(-ENOMEM);
267 	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
268 	if (ret)
269 		goto out;
270 
271 	if (rw == WRITE && hdr->din_xfer_len) {
272 		if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
273 			ret = -EOPNOTSUPP;
274 			goto out;
275 		}
276 
277 		next_rq = blk_get_request(q, READ, GFP_KERNEL);
278 		if (!next_rq) {
279 			ret = -ENOMEM;
280 			goto out;
281 		}
282 		rq->next_rq = next_rq;
283 		next_rq->cmd_type = rq->cmd_type;
284 
285 		dxferp = (void*)(unsigned long)hdr->din_xferp;
286 		ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
287 		if (ret)
288 			goto out;
289 	}
290 
291 	if (hdr->dout_xfer_len) {
292 		dxfer_len = hdr->dout_xfer_len;
293 		dxferp = (void*)(unsigned long)hdr->dout_xferp;
294 	} else if (hdr->din_xfer_len) {
295 		dxfer_len = hdr->din_xfer_len;
296 		dxferp = (void*)(unsigned long)hdr->din_xferp;
297 	} else
298 		dxfer_len = 0;
299 
300 	if (dxfer_len) {
301 		ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
302 		if (ret)
303 			goto out;
304 	}
305 	return rq;
306 out:
307 	if (rq->cmd != rq->__cmd)
308 		kfree(rq->cmd);
309 	blk_put_request(rq);
310 	if (next_rq) {
311 		blk_rq_unmap_user(next_rq->bio);
312 		blk_put_request(next_rq);
313 	}
314 	return ERR_PTR(ret);
315 }
316 
317 /*
318  * async completion call-back from the block layer, when scsi/ide/whatever
319  * calls end_that_request_last() on a request
320  */
321 static void bsg_rq_end_io(struct request *rq, int uptodate)
322 {
323 	struct bsg_command *bc = rq->end_io_data;
324 	struct bsg_device *bd = bc->bd;
325 	unsigned long flags;
326 
327 	dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
328 		bd->name, rq, bc, bc->bio, uptodate);
329 
330 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
331 
332 	spin_lock_irqsave(&bd->lock, flags);
333 	list_move_tail(&bc->list, &bd->done_list);
334 	bd->done_cmds++;
335 	spin_unlock_irqrestore(&bd->lock, flags);
336 
337 	wake_up(&bd->wq_done);
338 }
339 
340 /*
341  * do final setup of a 'bc' and submit the matching 'rq' to the block
342  * layer for io
343  */
344 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
345 			    struct bsg_command *bc, struct request *rq)
346 {
347 	rq->sense = bc->sense;
348 	rq->sense_len = 0;
349 
350 	/*
351 	 * add bc command to busy queue and submit rq for io
352 	 */
353 	bc->rq = rq;
354 	bc->bio = rq->bio;
355 	if (rq->next_rq)
356 		bc->bidi_bio = rq->next_rq->bio;
357 	bc->hdr.duration = jiffies;
358 	spin_lock_irq(&bd->lock);
359 	list_add_tail(&bc->list, &bd->busy_list);
360 	spin_unlock_irq(&bd->lock);
361 
362 	dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
363 
364 	rq->end_io_data = bc;
365 	blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
366 }
367 
368 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
369 {
370 	struct bsg_command *bc = NULL;
371 
372 	spin_lock_irq(&bd->lock);
373 	if (bd->done_cmds) {
374 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
375 		list_del(&bc->list);
376 		bd->done_cmds--;
377 	}
378 	spin_unlock_irq(&bd->lock);
379 
380 	return bc;
381 }
382 
383 /*
384  * Get a finished command from the done list
385  */
386 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
387 {
388 	struct bsg_command *bc;
389 	int ret;
390 
391 	do {
392 		bc = bsg_next_done_cmd(bd);
393 		if (bc)
394 			break;
395 
396 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
397 			bc = ERR_PTR(-EAGAIN);
398 			break;
399 		}
400 
401 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
402 		if (ret) {
403 			bc = ERR_PTR(-ERESTARTSYS);
404 			break;
405 		}
406 	} while (1);
407 
408 	dprintk("%s: returning done %p\n", bd->name, bc);
409 
410 	return bc;
411 }
412 
413 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
414 				    struct bio *bio, struct bio *bidi_bio)
415 {
416 	int ret = 0;
417 
418 	dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
419 	/*
420 	 * fill in all the output members
421 	 */
422 	hdr->device_status = status_byte(rq->errors);
423 	hdr->transport_status = host_byte(rq->errors);
424 	hdr->driver_status = driver_byte(rq->errors);
425 	hdr->info = 0;
426 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
427 		hdr->info |= SG_INFO_CHECK;
428 	hdr->response_len = 0;
429 
430 	if (rq->sense_len && hdr->response) {
431 		int len = min_t(unsigned int, hdr->max_response_len,
432 					rq->sense_len);
433 
434 		ret = copy_to_user((void*)(unsigned long)hdr->response,
435 				   rq->sense, len);
436 		if (!ret)
437 			hdr->response_len = len;
438 		else
439 			ret = -EFAULT;
440 	}
441 
442 	if (rq->next_rq) {
443 		hdr->dout_resid = rq->data_len;
444 		hdr->din_resid = rq->next_rq->data_len;
445 		blk_rq_unmap_user(bidi_bio);
446 		blk_put_request(rq->next_rq);
447 	} else if (rq_data_dir(rq) == READ)
448 		hdr->din_resid = rq->data_len;
449 	else
450 		hdr->dout_resid = rq->data_len;
451 
452 	/*
453 	 * If the request generated a negative error number, return it
454 	 * (providing we aren't already returning an error); if it's
455 	 * just a protocol response (i.e. non negative), that gets
456 	 * processed above.
457 	 */
458 	if (!ret && rq->errors < 0)
459 		ret = rq->errors;
460 
461 	blk_rq_unmap_user(bio);
462 	if (rq->cmd != rq->__cmd)
463 		kfree(rq->cmd);
464 	blk_put_request(rq);
465 
466 	return ret;
467 }
468 
469 static int bsg_complete_all_commands(struct bsg_device *bd)
470 {
471 	struct bsg_command *bc;
472 	int ret, tret;
473 
474 	dprintk("%s: entered\n", bd->name);
475 
476 	/*
477 	 * wait for all commands to complete
478 	 */
479 	ret = 0;
480 	do {
481 		ret = bsg_io_schedule(bd);
482 		/*
483 		 * look for -ENODATA specifically -- we'll sometimes get
484 		 * -ERESTARTSYS when we've taken a signal, but we can't
485 		 * return until we're done freeing the queue, so ignore
486 		 * it.  The signal will get handled when we're done freeing
487 		 * the bsg_device.
488 		 */
489 	} while (ret != -ENODATA);
490 
491 	/*
492 	 * discard done commands
493 	 */
494 	ret = 0;
495 	do {
496 		spin_lock_irq(&bd->lock);
497 		if (!bd->queued_cmds) {
498 			spin_unlock_irq(&bd->lock);
499 			break;
500 		}
501 		spin_unlock_irq(&bd->lock);
502 
503 		bc = bsg_get_done_cmd(bd);
504 		if (IS_ERR(bc))
505 			break;
506 
507 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
508 						bc->bidi_bio);
509 		if (!ret)
510 			ret = tret;
511 
512 		bsg_free_command(bc);
513 	} while (1);
514 
515 	return ret;
516 }
517 
518 static int
519 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
520 	   const struct iovec *iov, ssize_t *bytes_read)
521 {
522 	struct bsg_command *bc;
523 	int nr_commands, ret;
524 
525 	if (count % sizeof(struct sg_io_v4))
526 		return -EINVAL;
527 
528 	ret = 0;
529 	nr_commands = count / sizeof(struct sg_io_v4);
530 	while (nr_commands) {
531 		bc = bsg_get_done_cmd(bd);
532 		if (IS_ERR(bc)) {
533 			ret = PTR_ERR(bc);
534 			break;
535 		}
536 
537 		/*
538 		 * this is the only case where we need to copy data back
539 		 * after completing the request. so do that here,
540 		 * bsg_complete_work() cannot do that for us
541 		 */
542 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
543 					       bc->bidi_bio);
544 
545 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
546 			ret = -EFAULT;
547 
548 		bsg_free_command(bc);
549 
550 		if (ret)
551 			break;
552 
553 		buf += sizeof(struct sg_io_v4);
554 		*bytes_read += sizeof(struct sg_io_v4);
555 		nr_commands--;
556 	}
557 
558 	return ret;
559 }
560 
561 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
562 {
563 	if (file->f_flags & O_NONBLOCK)
564 		clear_bit(BSG_F_BLOCK, &bd->flags);
565 	else
566 		set_bit(BSG_F_BLOCK, &bd->flags);
567 }
568 
569 /*
570  * Check if the error is a "real" error that we should return.
571  */
572 static inline int err_block_err(int ret)
573 {
574 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
575 		return 1;
576 
577 	return 0;
578 }
579 
580 static ssize_t
581 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
582 {
583 	struct bsg_device *bd = file->private_data;
584 	int ret;
585 	ssize_t bytes_read;
586 
587 	dprintk("%s: read %Zd bytes\n", bd->name, count);
588 
589 	bsg_set_block(bd, file);
590 
591 	bytes_read = 0;
592 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
593 	*ppos = bytes_read;
594 
595 	if (!bytes_read || (bytes_read && err_block_err(ret)))
596 		bytes_read = ret;
597 
598 	return bytes_read;
599 }
600 
601 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
602 		       size_t count, ssize_t *bytes_written, int has_write_perm)
603 {
604 	struct bsg_command *bc;
605 	struct request *rq;
606 	int ret, nr_commands;
607 
608 	if (count % sizeof(struct sg_io_v4))
609 		return -EINVAL;
610 
611 	nr_commands = count / sizeof(struct sg_io_v4);
612 	rq = NULL;
613 	bc = NULL;
614 	ret = 0;
615 	while (nr_commands) {
616 		struct request_queue *q = bd->queue;
617 
618 		bc = bsg_alloc_command(bd);
619 		if (IS_ERR(bc)) {
620 			ret = PTR_ERR(bc);
621 			bc = NULL;
622 			break;
623 		}
624 
625 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
626 			ret = -EFAULT;
627 			break;
628 		}
629 
630 		/*
631 		 * get a request, fill in the blanks, and add to request queue
632 		 */
633 		rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm);
634 		if (IS_ERR(rq)) {
635 			ret = PTR_ERR(rq);
636 			rq = NULL;
637 			break;
638 		}
639 
640 		bsg_add_command(bd, q, bc, rq);
641 		bc = NULL;
642 		rq = NULL;
643 		nr_commands--;
644 		buf += sizeof(struct sg_io_v4);
645 		*bytes_written += sizeof(struct sg_io_v4);
646 	}
647 
648 	if (bc)
649 		bsg_free_command(bc);
650 
651 	return ret;
652 }
653 
654 static ssize_t
655 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
656 {
657 	struct bsg_device *bd = file->private_data;
658 	ssize_t bytes_written;
659 	int ret;
660 
661 	dprintk("%s: write %Zd bytes\n", bd->name, count);
662 
663 	bsg_set_block(bd, file);
664 
665 	bytes_written = 0;
666 	ret = __bsg_write(bd, buf, count, &bytes_written,
667 			  file->f_mode & FMODE_WRITE);
668 
669 	*ppos = bytes_written;
670 
671 	/*
672 	 * return bytes written on non-fatal errors
673 	 */
674 	if (!bytes_written || (bytes_written && err_block_err(ret)))
675 		bytes_written = ret;
676 
677 	dprintk("%s: returning %Zd\n", bd->name, bytes_written);
678 	return bytes_written;
679 }
680 
681 static struct bsg_device *bsg_alloc_device(void)
682 {
683 	struct bsg_device *bd;
684 
685 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
686 	if (unlikely(!bd))
687 		return NULL;
688 
689 	spin_lock_init(&bd->lock);
690 
691 	bd->max_queue = BSG_DEFAULT_CMDS;
692 
693 	INIT_LIST_HEAD(&bd->busy_list);
694 	INIT_LIST_HEAD(&bd->done_list);
695 	INIT_HLIST_NODE(&bd->dev_list);
696 
697 	init_waitqueue_head(&bd->wq_free);
698 	init_waitqueue_head(&bd->wq_done);
699 	return bd;
700 }
701 
702 static void bsg_kref_release_function(struct kref *kref)
703 {
704 	struct bsg_class_device *bcd =
705 		container_of(kref, struct bsg_class_device, ref);
706 	struct device *parent = bcd->parent;
707 
708 	if (bcd->release)
709 		bcd->release(bcd->parent);
710 
711 	put_device(parent);
712 }
713 
714 static int bsg_put_device(struct bsg_device *bd)
715 {
716 	int ret = 0, do_free;
717 	struct request_queue *q = bd->queue;
718 
719 	mutex_lock(&bsg_mutex);
720 
721 	do_free = atomic_dec_and_test(&bd->ref_count);
722 	if (!do_free) {
723 		mutex_unlock(&bsg_mutex);
724 		goto out;
725 	}
726 
727 	hlist_del(&bd->dev_list);
728 	mutex_unlock(&bsg_mutex);
729 
730 	dprintk("%s: tearing down\n", bd->name);
731 
732 	/*
733 	 * close can always block
734 	 */
735 	set_bit(BSG_F_BLOCK, &bd->flags);
736 
737 	/*
738 	 * correct error detection baddies here again. it's the responsibility
739 	 * of the app to properly reap commands before close() if it wants
740 	 * fool-proof error detection
741 	 */
742 	ret = bsg_complete_all_commands(bd);
743 
744 	kfree(bd);
745 out:
746 	kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
747 	if (do_free)
748 		blk_put_queue(q);
749 	return ret;
750 }
751 
752 static struct bsg_device *bsg_add_device(struct inode *inode,
753 					 struct request_queue *rq,
754 					 struct file *file)
755 {
756 	struct bsg_device *bd;
757 	int ret;
758 #ifdef BSG_DEBUG
759 	unsigned char buf[32];
760 #endif
761 	ret = blk_get_queue(rq);
762 	if (ret)
763 		return ERR_PTR(-ENXIO);
764 
765 	bd = bsg_alloc_device();
766 	if (!bd) {
767 		blk_put_queue(rq);
768 		return ERR_PTR(-ENOMEM);
769 	}
770 
771 	bd->queue = rq;
772 
773 	bsg_set_block(bd, file);
774 
775 	atomic_set(&bd->ref_count, 1);
776 	mutex_lock(&bsg_mutex);
777 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
778 
779 	strncpy(bd->name, rq->bsg_dev.class_dev->bus_id, sizeof(bd->name) - 1);
780 	dprintk("bound to <%s>, max queue %d\n",
781 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
782 
783 	mutex_unlock(&bsg_mutex);
784 	return bd;
785 }
786 
787 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
788 {
789 	struct bsg_device *bd;
790 	struct hlist_node *entry;
791 
792 	mutex_lock(&bsg_mutex);
793 
794 	hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
795 		if (bd->queue == q) {
796 			atomic_inc(&bd->ref_count);
797 			goto found;
798 		}
799 	}
800 	bd = NULL;
801 found:
802 	mutex_unlock(&bsg_mutex);
803 	return bd;
804 }
805 
806 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
807 {
808 	struct bsg_device *bd;
809 	struct bsg_class_device *bcd;
810 
811 	/*
812 	 * find the class device
813 	 */
814 	mutex_lock(&bsg_mutex);
815 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
816 	if (bcd)
817 		kref_get(&bcd->ref);
818 	mutex_unlock(&bsg_mutex);
819 
820 	if (!bcd)
821 		return ERR_PTR(-ENODEV);
822 
823 	bd = __bsg_get_device(iminor(inode), bcd->queue);
824 	if (bd)
825 		return bd;
826 
827 	bd = bsg_add_device(inode, bcd->queue, file);
828 	if (IS_ERR(bd))
829 		kref_put(&bcd->ref, bsg_kref_release_function);
830 
831 	return bd;
832 }
833 
834 static int bsg_open(struct inode *inode, struct file *file)
835 {
836 	struct bsg_device *bd;
837 
838 	lock_kernel();
839 	bd = bsg_get_device(inode, file);
840 	unlock_kernel();
841 
842 	if (IS_ERR(bd))
843 		return PTR_ERR(bd);
844 
845 	file->private_data = bd;
846 	return 0;
847 }
848 
849 static int bsg_release(struct inode *inode, struct file *file)
850 {
851 	struct bsg_device *bd = file->private_data;
852 
853 	file->private_data = NULL;
854 	return bsg_put_device(bd);
855 }
856 
857 static unsigned int bsg_poll(struct file *file, poll_table *wait)
858 {
859 	struct bsg_device *bd = file->private_data;
860 	unsigned int mask = 0;
861 
862 	poll_wait(file, &bd->wq_done, wait);
863 	poll_wait(file, &bd->wq_free, wait);
864 
865 	spin_lock_irq(&bd->lock);
866 	if (!list_empty(&bd->done_list))
867 		mask |= POLLIN | POLLRDNORM;
868 	if (bd->queued_cmds >= bd->max_queue)
869 		mask |= POLLOUT;
870 	spin_unlock_irq(&bd->lock);
871 
872 	return mask;
873 }
874 
875 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
876 {
877 	struct bsg_device *bd = file->private_data;
878 	int __user *uarg = (int __user *) arg;
879 	int ret;
880 
881 	switch (cmd) {
882 		/*
883 		 * our own ioctls
884 		 */
885 	case SG_GET_COMMAND_Q:
886 		return put_user(bd->max_queue, uarg);
887 	case SG_SET_COMMAND_Q: {
888 		int queue;
889 
890 		if (get_user(queue, uarg))
891 			return -EFAULT;
892 		if (queue < 1)
893 			return -EINVAL;
894 
895 		spin_lock_irq(&bd->lock);
896 		bd->max_queue = queue;
897 		spin_unlock_irq(&bd->lock);
898 		return 0;
899 	}
900 
901 	/*
902 	 * SCSI/sg ioctls
903 	 */
904 	case SG_GET_VERSION_NUM:
905 	case SCSI_IOCTL_GET_IDLUN:
906 	case SCSI_IOCTL_GET_BUS_NUMBER:
907 	case SG_SET_TIMEOUT:
908 	case SG_GET_TIMEOUT:
909 	case SG_GET_RESERVED_SIZE:
910 	case SG_SET_RESERVED_SIZE:
911 	case SG_EMULATED_HOST:
912 	case SCSI_IOCTL_SEND_COMMAND: {
913 		void __user *uarg = (void __user *) arg;
914 		return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
915 	}
916 	case SG_IO: {
917 		struct request *rq;
918 		struct bio *bio, *bidi_bio = NULL;
919 		struct sg_io_v4 hdr;
920 
921 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
922 			return -EFAULT;
923 
924 		rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE);
925 		if (IS_ERR(rq))
926 			return PTR_ERR(rq);
927 
928 		bio = rq->bio;
929 		if (rq->next_rq)
930 			bidi_bio = rq->next_rq->bio;
931 		blk_execute_rq(bd->queue, NULL, rq, 0);
932 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
933 
934 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
935 			return -EFAULT;
936 
937 		return ret;
938 	}
939 	/*
940 	 * block device ioctls
941 	 */
942 	default:
943 #if 0
944 		return ioctl_by_bdev(bd->bdev, cmd, arg);
945 #else
946 		return -ENOTTY;
947 #endif
948 	}
949 }
950 
951 static const struct file_operations bsg_fops = {
952 	.read		=	bsg_read,
953 	.write		=	bsg_write,
954 	.poll		=	bsg_poll,
955 	.open		=	bsg_open,
956 	.release	=	bsg_release,
957 	.unlocked_ioctl	=	bsg_ioctl,
958 	.owner		=	THIS_MODULE,
959 };
960 
961 void bsg_unregister_queue(struct request_queue *q)
962 {
963 	struct bsg_class_device *bcd = &q->bsg_dev;
964 
965 	if (!bcd->class_dev)
966 		return;
967 
968 	mutex_lock(&bsg_mutex);
969 	idr_remove(&bsg_minor_idr, bcd->minor);
970 	sysfs_remove_link(&q->kobj, "bsg");
971 	device_unregister(bcd->class_dev);
972 	bcd->class_dev = NULL;
973 	kref_put(&bcd->ref, bsg_kref_release_function);
974 	mutex_unlock(&bsg_mutex);
975 }
976 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
977 
978 int bsg_register_queue(struct request_queue *q, struct device *parent,
979 		       const char *name, void (*release)(struct device *))
980 {
981 	struct bsg_class_device *bcd;
982 	dev_t dev;
983 	int ret, minor;
984 	struct device *class_dev = NULL;
985 	const char *devname;
986 
987 	if (name)
988 		devname = name;
989 	else
990 		devname = parent->bus_id;
991 
992 	/*
993 	 * we need a proper transport to send commands, not a stacked device
994 	 */
995 	if (!q->request_fn)
996 		return 0;
997 
998 	bcd = &q->bsg_dev;
999 	memset(bcd, 0, sizeof(*bcd));
1000 
1001 	mutex_lock(&bsg_mutex);
1002 
1003 	ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1004 	if (!ret) {
1005 		ret = -ENOMEM;
1006 		goto unlock;
1007 	}
1008 
1009 	ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1010 	if (ret < 0)
1011 		goto unlock;
1012 
1013 	if (minor >= BSG_MAX_DEVS) {
1014 		printk(KERN_ERR "bsg: too many bsg devices\n");
1015 		ret = -EINVAL;
1016 		goto remove_idr;
1017 	}
1018 
1019 	bcd->minor = minor;
1020 	bcd->queue = q;
1021 	bcd->parent = get_device(parent);
1022 	bcd->release = release;
1023 	kref_init(&bcd->ref);
1024 	dev = MKDEV(bsg_major, bcd->minor);
1025 	class_dev = device_create_drvdata(bsg_class, parent, dev, NULL,
1026 					  "%s", devname);
1027 	if (IS_ERR(class_dev)) {
1028 		ret = PTR_ERR(class_dev);
1029 		goto put_dev;
1030 	}
1031 	bcd->class_dev = class_dev;
1032 
1033 	if (q->kobj.sd) {
1034 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1035 		if (ret)
1036 			goto unregister_class_dev;
1037 	}
1038 
1039 	mutex_unlock(&bsg_mutex);
1040 	return 0;
1041 
1042 unregister_class_dev:
1043 	device_unregister(class_dev);
1044 put_dev:
1045 	put_device(parent);
1046 remove_idr:
1047 	idr_remove(&bsg_minor_idr, minor);
1048 unlock:
1049 	mutex_unlock(&bsg_mutex);
1050 	return ret;
1051 }
1052 EXPORT_SYMBOL_GPL(bsg_register_queue);
1053 
1054 static struct cdev bsg_cdev;
1055 
1056 static int __init bsg_init(void)
1057 {
1058 	int ret, i;
1059 	dev_t devid;
1060 
1061 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1062 				sizeof(struct bsg_command), 0, 0, NULL);
1063 	if (!bsg_cmd_cachep) {
1064 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1065 		return -ENOMEM;
1066 	}
1067 
1068 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1069 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1070 
1071 	bsg_class = class_create(THIS_MODULE, "bsg");
1072 	if (IS_ERR(bsg_class)) {
1073 		ret = PTR_ERR(bsg_class);
1074 		goto destroy_kmemcache;
1075 	}
1076 
1077 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1078 	if (ret)
1079 		goto destroy_bsg_class;
1080 
1081 	bsg_major = MAJOR(devid);
1082 
1083 	cdev_init(&bsg_cdev, &bsg_fops);
1084 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1085 	if (ret)
1086 		goto unregister_chrdev;
1087 
1088 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1089 	       " loaded (major %d)\n", bsg_major);
1090 	return 0;
1091 unregister_chrdev:
1092 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1093 destroy_bsg_class:
1094 	class_destroy(bsg_class);
1095 destroy_kmemcache:
1096 	kmem_cache_destroy(bsg_cmd_cachep);
1097 	return ret;
1098 }
1099 
1100 MODULE_AUTHOR("Jens Axboe");
1101 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1102 MODULE_LICENSE("GPL");
1103 
1104 device_initcall(bsg_init);
1105