xref: /openbmc/linux/block/bsg.c (revision 367b8112)
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 				fmode_t 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, fmode_t 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, NULL, dxferp,
287 				       hdr->din_xfer_len, GFP_KERNEL);
288 		if (ret)
289 			goto out;
290 	}
291 
292 	if (hdr->dout_xfer_len) {
293 		dxfer_len = hdr->dout_xfer_len;
294 		dxferp = (void*)(unsigned long)hdr->dout_xferp;
295 	} else if (hdr->din_xfer_len) {
296 		dxfer_len = hdr->din_xfer_len;
297 		dxferp = (void*)(unsigned long)hdr->din_xferp;
298 	} else
299 		dxfer_len = 0;
300 
301 	if (dxfer_len) {
302 		ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
303 				      GFP_KERNEL);
304 		if (ret)
305 			goto out;
306 	}
307 	return rq;
308 out:
309 	if (rq->cmd != rq->__cmd)
310 		kfree(rq->cmd);
311 	blk_put_request(rq);
312 	if (next_rq) {
313 		blk_rq_unmap_user(next_rq->bio);
314 		blk_put_request(next_rq);
315 	}
316 	return ERR_PTR(ret);
317 }
318 
319 /*
320  * async completion call-back from the block layer, when scsi/ide/whatever
321  * calls end_that_request_last() on a request
322  */
323 static void bsg_rq_end_io(struct request *rq, int uptodate)
324 {
325 	struct bsg_command *bc = rq->end_io_data;
326 	struct bsg_device *bd = bc->bd;
327 	unsigned long flags;
328 
329 	dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
330 		bd->name, rq, bc, bc->bio, uptodate);
331 
332 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
333 
334 	spin_lock_irqsave(&bd->lock, flags);
335 	list_move_tail(&bc->list, &bd->done_list);
336 	bd->done_cmds++;
337 	spin_unlock_irqrestore(&bd->lock, flags);
338 
339 	wake_up(&bd->wq_done);
340 }
341 
342 /*
343  * do final setup of a 'bc' and submit the matching 'rq' to the block
344  * layer for io
345  */
346 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
347 			    struct bsg_command *bc, struct request *rq)
348 {
349 	rq->sense = bc->sense;
350 	rq->sense_len = 0;
351 
352 	/*
353 	 * add bc command to busy queue and submit rq for io
354 	 */
355 	bc->rq = rq;
356 	bc->bio = rq->bio;
357 	if (rq->next_rq)
358 		bc->bidi_bio = rq->next_rq->bio;
359 	bc->hdr.duration = jiffies;
360 	spin_lock_irq(&bd->lock);
361 	list_add_tail(&bc->list, &bd->busy_list);
362 	spin_unlock_irq(&bd->lock);
363 
364 	dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
365 
366 	rq->end_io_data = bc;
367 	blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
368 }
369 
370 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
371 {
372 	struct bsg_command *bc = NULL;
373 
374 	spin_lock_irq(&bd->lock);
375 	if (bd->done_cmds) {
376 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
377 		list_del(&bc->list);
378 		bd->done_cmds--;
379 	}
380 	spin_unlock_irq(&bd->lock);
381 
382 	return bc;
383 }
384 
385 /*
386  * Get a finished command from the done list
387  */
388 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
389 {
390 	struct bsg_command *bc;
391 	int ret;
392 
393 	do {
394 		bc = bsg_next_done_cmd(bd);
395 		if (bc)
396 			break;
397 
398 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
399 			bc = ERR_PTR(-EAGAIN);
400 			break;
401 		}
402 
403 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
404 		if (ret) {
405 			bc = ERR_PTR(-ERESTARTSYS);
406 			break;
407 		}
408 	} while (1);
409 
410 	dprintk("%s: returning done %p\n", bd->name, bc);
411 
412 	return bc;
413 }
414 
415 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
416 				    struct bio *bio, struct bio *bidi_bio)
417 {
418 	int ret = 0;
419 
420 	dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
421 	/*
422 	 * fill in all the output members
423 	 */
424 	hdr->device_status = status_byte(rq->errors);
425 	hdr->transport_status = host_byte(rq->errors);
426 	hdr->driver_status = driver_byte(rq->errors);
427 	hdr->info = 0;
428 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
429 		hdr->info |= SG_INFO_CHECK;
430 	hdr->response_len = 0;
431 
432 	if (rq->sense_len && hdr->response) {
433 		int len = min_t(unsigned int, hdr->max_response_len,
434 					rq->sense_len);
435 
436 		ret = copy_to_user((void*)(unsigned long)hdr->response,
437 				   rq->sense, len);
438 		if (!ret)
439 			hdr->response_len = len;
440 		else
441 			ret = -EFAULT;
442 	}
443 
444 	if (rq->next_rq) {
445 		hdr->dout_resid = rq->data_len;
446 		hdr->din_resid = rq->next_rq->data_len;
447 		blk_rq_unmap_user(bidi_bio);
448 		blk_put_request(rq->next_rq);
449 	} else if (rq_data_dir(rq) == READ)
450 		hdr->din_resid = rq->data_len;
451 	else
452 		hdr->dout_resid = rq->data_len;
453 
454 	/*
455 	 * If the request generated a negative error number, return it
456 	 * (providing we aren't already returning an error); if it's
457 	 * just a protocol response (i.e. non negative), that gets
458 	 * processed above.
459 	 */
460 	if (!ret && rq->errors < 0)
461 		ret = rq->errors;
462 
463 	blk_rq_unmap_user(bio);
464 	if (rq->cmd != rq->__cmd)
465 		kfree(rq->cmd);
466 	blk_put_request(rq);
467 
468 	return ret;
469 }
470 
471 static int bsg_complete_all_commands(struct bsg_device *bd)
472 {
473 	struct bsg_command *bc;
474 	int ret, tret;
475 
476 	dprintk("%s: entered\n", bd->name);
477 
478 	/*
479 	 * wait for all commands to complete
480 	 */
481 	ret = 0;
482 	do {
483 		ret = bsg_io_schedule(bd);
484 		/*
485 		 * look for -ENODATA specifically -- we'll sometimes get
486 		 * -ERESTARTSYS when we've taken a signal, but we can't
487 		 * return until we're done freeing the queue, so ignore
488 		 * it.  The signal will get handled when we're done freeing
489 		 * the bsg_device.
490 		 */
491 	} while (ret != -ENODATA);
492 
493 	/*
494 	 * discard done commands
495 	 */
496 	ret = 0;
497 	do {
498 		spin_lock_irq(&bd->lock);
499 		if (!bd->queued_cmds) {
500 			spin_unlock_irq(&bd->lock);
501 			break;
502 		}
503 		spin_unlock_irq(&bd->lock);
504 
505 		bc = bsg_get_done_cmd(bd);
506 		if (IS_ERR(bc))
507 			break;
508 
509 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
510 						bc->bidi_bio);
511 		if (!ret)
512 			ret = tret;
513 
514 		bsg_free_command(bc);
515 	} while (1);
516 
517 	return ret;
518 }
519 
520 static int
521 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
522 	   const struct iovec *iov, ssize_t *bytes_read)
523 {
524 	struct bsg_command *bc;
525 	int nr_commands, ret;
526 
527 	if (count % sizeof(struct sg_io_v4))
528 		return -EINVAL;
529 
530 	ret = 0;
531 	nr_commands = count / sizeof(struct sg_io_v4);
532 	while (nr_commands) {
533 		bc = bsg_get_done_cmd(bd);
534 		if (IS_ERR(bc)) {
535 			ret = PTR_ERR(bc);
536 			break;
537 		}
538 
539 		/*
540 		 * this is the only case where we need to copy data back
541 		 * after completing the request. so do that here,
542 		 * bsg_complete_work() cannot do that for us
543 		 */
544 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
545 					       bc->bidi_bio);
546 
547 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
548 			ret = -EFAULT;
549 
550 		bsg_free_command(bc);
551 
552 		if (ret)
553 			break;
554 
555 		buf += sizeof(struct sg_io_v4);
556 		*bytes_read += sizeof(struct sg_io_v4);
557 		nr_commands--;
558 	}
559 
560 	return ret;
561 }
562 
563 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
564 {
565 	if (file->f_flags & O_NONBLOCK)
566 		clear_bit(BSG_F_BLOCK, &bd->flags);
567 	else
568 		set_bit(BSG_F_BLOCK, &bd->flags);
569 }
570 
571 /*
572  * Check if the error is a "real" error that we should return.
573  */
574 static inline int err_block_err(int ret)
575 {
576 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
577 		return 1;
578 
579 	return 0;
580 }
581 
582 static ssize_t
583 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
584 {
585 	struct bsg_device *bd = file->private_data;
586 	int ret;
587 	ssize_t bytes_read;
588 
589 	dprintk("%s: read %Zd bytes\n", bd->name, count);
590 
591 	bsg_set_block(bd, file);
592 
593 	bytes_read = 0;
594 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
595 	*ppos = bytes_read;
596 
597 	if (!bytes_read || (bytes_read && err_block_err(ret)))
598 		bytes_read = ret;
599 
600 	return bytes_read;
601 }
602 
603 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
604 		       size_t count, ssize_t *bytes_written,
605 		       fmode_t has_write_perm)
606 {
607 	struct bsg_command *bc;
608 	struct request *rq;
609 	int ret, nr_commands;
610 
611 	if (count % sizeof(struct sg_io_v4))
612 		return -EINVAL;
613 
614 	nr_commands = count / sizeof(struct sg_io_v4);
615 	rq = NULL;
616 	bc = NULL;
617 	ret = 0;
618 	while (nr_commands) {
619 		struct request_queue *q = bd->queue;
620 
621 		bc = bsg_alloc_command(bd);
622 		if (IS_ERR(bc)) {
623 			ret = PTR_ERR(bc);
624 			bc = NULL;
625 			break;
626 		}
627 
628 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
629 			ret = -EFAULT;
630 			break;
631 		}
632 
633 		/*
634 		 * get a request, fill in the blanks, and add to request queue
635 		 */
636 		rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm);
637 		if (IS_ERR(rq)) {
638 			ret = PTR_ERR(rq);
639 			rq = NULL;
640 			break;
641 		}
642 
643 		bsg_add_command(bd, q, bc, rq);
644 		bc = NULL;
645 		rq = NULL;
646 		nr_commands--;
647 		buf += sizeof(struct sg_io_v4);
648 		*bytes_written += sizeof(struct sg_io_v4);
649 	}
650 
651 	if (bc)
652 		bsg_free_command(bc);
653 
654 	return ret;
655 }
656 
657 static ssize_t
658 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
659 {
660 	struct bsg_device *bd = file->private_data;
661 	ssize_t bytes_written;
662 	int ret;
663 
664 	dprintk("%s: write %Zd bytes\n", bd->name, count);
665 
666 	bsg_set_block(bd, file);
667 
668 	bytes_written = 0;
669 	ret = __bsg_write(bd, buf, count, &bytes_written,
670 			  file->f_mode & FMODE_WRITE);
671 
672 	*ppos = bytes_written;
673 
674 	/*
675 	 * return bytes written on non-fatal errors
676 	 */
677 	if (!bytes_written || (bytes_written && err_block_err(ret)))
678 		bytes_written = ret;
679 
680 	dprintk("%s: returning %Zd\n", bd->name, bytes_written);
681 	return bytes_written;
682 }
683 
684 static struct bsg_device *bsg_alloc_device(void)
685 {
686 	struct bsg_device *bd;
687 
688 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
689 	if (unlikely(!bd))
690 		return NULL;
691 
692 	spin_lock_init(&bd->lock);
693 
694 	bd->max_queue = BSG_DEFAULT_CMDS;
695 
696 	INIT_LIST_HEAD(&bd->busy_list);
697 	INIT_LIST_HEAD(&bd->done_list);
698 	INIT_HLIST_NODE(&bd->dev_list);
699 
700 	init_waitqueue_head(&bd->wq_free);
701 	init_waitqueue_head(&bd->wq_done);
702 	return bd;
703 }
704 
705 static void bsg_kref_release_function(struct kref *kref)
706 {
707 	struct bsg_class_device *bcd =
708 		container_of(kref, struct bsg_class_device, ref);
709 	struct device *parent = bcd->parent;
710 
711 	if (bcd->release)
712 		bcd->release(bcd->parent);
713 
714 	put_device(parent);
715 }
716 
717 static int bsg_put_device(struct bsg_device *bd)
718 {
719 	int ret = 0, do_free;
720 	struct request_queue *q = bd->queue;
721 
722 	mutex_lock(&bsg_mutex);
723 
724 	do_free = atomic_dec_and_test(&bd->ref_count);
725 	if (!do_free) {
726 		mutex_unlock(&bsg_mutex);
727 		goto out;
728 	}
729 
730 	hlist_del(&bd->dev_list);
731 	mutex_unlock(&bsg_mutex);
732 
733 	dprintk("%s: tearing down\n", bd->name);
734 
735 	/*
736 	 * close can always block
737 	 */
738 	set_bit(BSG_F_BLOCK, &bd->flags);
739 
740 	/*
741 	 * correct error detection baddies here again. it's the responsibility
742 	 * of the app to properly reap commands before close() if it wants
743 	 * fool-proof error detection
744 	 */
745 	ret = bsg_complete_all_commands(bd);
746 
747 	kfree(bd);
748 out:
749 	kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
750 	if (do_free)
751 		blk_put_queue(q);
752 	return ret;
753 }
754 
755 static struct bsg_device *bsg_add_device(struct inode *inode,
756 					 struct request_queue *rq,
757 					 struct file *file)
758 {
759 	struct bsg_device *bd;
760 	int ret;
761 #ifdef BSG_DEBUG
762 	unsigned char buf[32];
763 #endif
764 	ret = blk_get_queue(rq);
765 	if (ret)
766 		return ERR_PTR(-ENXIO);
767 
768 	bd = bsg_alloc_device();
769 	if (!bd) {
770 		blk_put_queue(rq);
771 		return ERR_PTR(-ENOMEM);
772 	}
773 
774 	bd->queue = rq;
775 
776 	bsg_set_block(bd, file);
777 
778 	atomic_set(&bd->ref_count, 1);
779 	mutex_lock(&bsg_mutex);
780 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
781 
782 	strncpy(bd->name, rq->bsg_dev.class_dev->bus_id, sizeof(bd->name) - 1);
783 	dprintk("bound to <%s>, max queue %d\n",
784 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
785 
786 	mutex_unlock(&bsg_mutex);
787 	return bd;
788 }
789 
790 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
791 {
792 	struct bsg_device *bd;
793 	struct hlist_node *entry;
794 
795 	mutex_lock(&bsg_mutex);
796 
797 	hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
798 		if (bd->queue == q) {
799 			atomic_inc(&bd->ref_count);
800 			goto found;
801 		}
802 	}
803 	bd = NULL;
804 found:
805 	mutex_unlock(&bsg_mutex);
806 	return bd;
807 }
808 
809 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
810 {
811 	struct bsg_device *bd;
812 	struct bsg_class_device *bcd;
813 
814 	/*
815 	 * find the class device
816 	 */
817 	mutex_lock(&bsg_mutex);
818 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
819 	if (bcd)
820 		kref_get(&bcd->ref);
821 	mutex_unlock(&bsg_mutex);
822 
823 	if (!bcd)
824 		return ERR_PTR(-ENODEV);
825 
826 	bd = __bsg_get_device(iminor(inode), bcd->queue);
827 	if (bd)
828 		return bd;
829 
830 	bd = bsg_add_device(inode, bcd->queue, file);
831 	if (IS_ERR(bd))
832 		kref_put(&bcd->ref, bsg_kref_release_function);
833 
834 	return bd;
835 }
836 
837 static int bsg_open(struct inode *inode, struct file *file)
838 {
839 	struct bsg_device *bd;
840 
841 	lock_kernel();
842 	bd = bsg_get_device(inode, file);
843 	unlock_kernel();
844 
845 	if (IS_ERR(bd))
846 		return PTR_ERR(bd);
847 
848 	file->private_data = bd;
849 	return 0;
850 }
851 
852 static int bsg_release(struct inode *inode, struct file *file)
853 {
854 	struct bsg_device *bd = file->private_data;
855 
856 	file->private_data = NULL;
857 	return bsg_put_device(bd);
858 }
859 
860 static unsigned int bsg_poll(struct file *file, poll_table *wait)
861 {
862 	struct bsg_device *bd = file->private_data;
863 	unsigned int mask = 0;
864 
865 	poll_wait(file, &bd->wq_done, wait);
866 	poll_wait(file, &bd->wq_free, wait);
867 
868 	spin_lock_irq(&bd->lock);
869 	if (!list_empty(&bd->done_list))
870 		mask |= POLLIN | POLLRDNORM;
871 	if (bd->queued_cmds >= bd->max_queue)
872 		mask |= POLLOUT;
873 	spin_unlock_irq(&bd->lock);
874 
875 	return mask;
876 }
877 
878 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
879 {
880 	struct bsg_device *bd = file->private_data;
881 	int __user *uarg = (int __user *) arg;
882 	int ret;
883 
884 	switch (cmd) {
885 		/*
886 		 * our own ioctls
887 		 */
888 	case SG_GET_COMMAND_Q:
889 		return put_user(bd->max_queue, uarg);
890 	case SG_SET_COMMAND_Q: {
891 		int queue;
892 
893 		if (get_user(queue, uarg))
894 			return -EFAULT;
895 		if (queue < 1)
896 			return -EINVAL;
897 
898 		spin_lock_irq(&bd->lock);
899 		bd->max_queue = queue;
900 		spin_unlock_irq(&bd->lock);
901 		return 0;
902 	}
903 
904 	/*
905 	 * SCSI/sg ioctls
906 	 */
907 	case SG_GET_VERSION_NUM:
908 	case SCSI_IOCTL_GET_IDLUN:
909 	case SCSI_IOCTL_GET_BUS_NUMBER:
910 	case SG_SET_TIMEOUT:
911 	case SG_GET_TIMEOUT:
912 	case SG_GET_RESERVED_SIZE:
913 	case SG_SET_RESERVED_SIZE:
914 	case SG_EMULATED_HOST:
915 	case SCSI_IOCTL_SEND_COMMAND: {
916 		void __user *uarg = (void __user *) arg;
917 		return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
918 	}
919 	case SG_IO: {
920 		struct request *rq;
921 		struct bio *bio, *bidi_bio = NULL;
922 		struct sg_io_v4 hdr;
923 
924 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
925 			return -EFAULT;
926 
927 		rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE);
928 		if (IS_ERR(rq))
929 			return PTR_ERR(rq);
930 
931 		bio = rq->bio;
932 		if (rq->next_rq)
933 			bidi_bio = rq->next_rq->bio;
934 		blk_execute_rq(bd->queue, NULL, rq, 0);
935 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
936 
937 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
938 			return -EFAULT;
939 
940 		return ret;
941 	}
942 	/*
943 	 * block device ioctls
944 	 */
945 	default:
946 #if 0
947 		return ioctl_by_bdev(bd->bdev, cmd, arg);
948 #else
949 		return -ENOTTY;
950 #endif
951 	}
952 }
953 
954 static const struct file_operations bsg_fops = {
955 	.read		=	bsg_read,
956 	.write		=	bsg_write,
957 	.poll		=	bsg_poll,
958 	.open		=	bsg_open,
959 	.release	=	bsg_release,
960 	.unlocked_ioctl	=	bsg_ioctl,
961 	.owner		=	THIS_MODULE,
962 };
963 
964 void bsg_unregister_queue(struct request_queue *q)
965 {
966 	struct bsg_class_device *bcd = &q->bsg_dev;
967 
968 	if (!bcd->class_dev)
969 		return;
970 
971 	mutex_lock(&bsg_mutex);
972 	idr_remove(&bsg_minor_idr, bcd->minor);
973 	sysfs_remove_link(&q->kobj, "bsg");
974 	device_unregister(bcd->class_dev);
975 	bcd->class_dev = NULL;
976 	kref_put(&bcd->ref, bsg_kref_release_function);
977 	mutex_unlock(&bsg_mutex);
978 }
979 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
980 
981 int bsg_register_queue(struct request_queue *q, struct device *parent,
982 		       const char *name, void (*release)(struct device *))
983 {
984 	struct bsg_class_device *bcd;
985 	dev_t dev;
986 	int ret, minor;
987 	struct device *class_dev = NULL;
988 	const char *devname;
989 
990 	if (name)
991 		devname = name;
992 	else
993 		devname = parent->bus_id;
994 
995 	/*
996 	 * we need a proper transport to send commands, not a stacked device
997 	 */
998 	if (!q->request_fn)
999 		return 0;
1000 
1001 	bcd = &q->bsg_dev;
1002 	memset(bcd, 0, sizeof(*bcd));
1003 
1004 	mutex_lock(&bsg_mutex);
1005 
1006 	ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1007 	if (!ret) {
1008 		ret = -ENOMEM;
1009 		goto unlock;
1010 	}
1011 
1012 	ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1013 	if (ret < 0)
1014 		goto unlock;
1015 
1016 	if (minor >= BSG_MAX_DEVS) {
1017 		printk(KERN_ERR "bsg: too many bsg devices\n");
1018 		ret = -EINVAL;
1019 		goto remove_idr;
1020 	}
1021 
1022 	bcd->minor = minor;
1023 	bcd->queue = q;
1024 	bcd->parent = get_device(parent);
1025 	bcd->release = release;
1026 	kref_init(&bcd->ref);
1027 	dev = MKDEV(bsg_major, bcd->minor);
1028 	class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1029 	if (IS_ERR(class_dev)) {
1030 		ret = PTR_ERR(class_dev);
1031 		goto put_dev;
1032 	}
1033 	bcd->class_dev = class_dev;
1034 
1035 	if (q->kobj.sd) {
1036 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1037 		if (ret)
1038 			goto unregister_class_dev;
1039 	}
1040 
1041 	mutex_unlock(&bsg_mutex);
1042 	return 0;
1043 
1044 unregister_class_dev:
1045 	device_unregister(class_dev);
1046 put_dev:
1047 	put_device(parent);
1048 remove_idr:
1049 	idr_remove(&bsg_minor_idr, minor);
1050 unlock:
1051 	mutex_unlock(&bsg_mutex);
1052 	return ret;
1053 }
1054 EXPORT_SYMBOL_GPL(bsg_register_queue);
1055 
1056 static struct cdev bsg_cdev;
1057 
1058 static int __init bsg_init(void)
1059 {
1060 	int ret, i;
1061 	dev_t devid;
1062 
1063 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1064 				sizeof(struct bsg_command), 0, 0, NULL);
1065 	if (!bsg_cmd_cachep) {
1066 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1067 		return -ENOMEM;
1068 	}
1069 
1070 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1071 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1072 
1073 	bsg_class = class_create(THIS_MODULE, "bsg");
1074 	if (IS_ERR(bsg_class)) {
1075 		ret = PTR_ERR(bsg_class);
1076 		goto destroy_kmemcache;
1077 	}
1078 
1079 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1080 	if (ret)
1081 		goto destroy_bsg_class;
1082 
1083 	bsg_major = MAJOR(devid);
1084 
1085 	cdev_init(&bsg_cdev, &bsg_fops);
1086 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1087 	if (ret)
1088 		goto unregister_chrdev;
1089 
1090 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1091 	       " loaded (major %d)\n", bsg_major);
1092 	return 0;
1093 unregister_chrdev:
1094 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1095 destroy_bsg_class:
1096 	class_destroy(bsg_class);
1097 destroy_kmemcache:
1098 	kmem_cache_destroy(bsg_cmd_cachep);
1099 	return ret;
1100 }
1101 
1102 MODULE_AUTHOR("Jens Axboe");
1103 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1104 MODULE_LICENSE("GPL");
1105 
1106 device_initcall(bsg_init);
1107