xref: /openbmc/linux/block/bsg.c (revision c21b37f6)
1 /*
2  * bsg.c - block layer implementation of the sg v3 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 
23 #include <scsi/scsi.h>
24 #include <scsi/scsi_ioctl.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_device.h>
27 #include <scsi/scsi_driver.h>
28 #include <scsi/sg.h>
29 
30 #define BSG_DESCRIPTION	"Block layer SCSI generic (bsg) driver"
31 #define BSG_VERSION	"0.4"
32 
33 struct bsg_device {
34 	struct request_queue *queue;
35 	spinlock_t lock;
36 	struct list_head busy_list;
37 	struct list_head done_list;
38 	struct hlist_node dev_list;
39 	atomic_t ref_count;
40 	int minor;
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 	BSG_F_WRITE_PERM	= 2,
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, __FUNCTION__, ##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 bsg_io_schedule(struct bsg_device *bd)
140 {
141 	DEFINE_WAIT(wait);
142 	int ret = 0;
143 
144 	spin_lock_irq(&bd->lock);
145 
146 	BUG_ON(bd->done_cmds > bd->queued_cmds);
147 
148 	/*
149 	 * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
150 	 * work to do", even though we return -ENOSPC after this same test
151 	 * during bsg_write() -- there, it means our buffer can't have more
152 	 * bsg_commands added to it, thus has no space left.
153 	 */
154 	if (bd->done_cmds == bd->queued_cmds) {
155 		ret = -ENODATA;
156 		goto unlock;
157 	}
158 
159 	if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
160 		ret = -EAGAIN;
161 		goto unlock;
162 	}
163 
164 	prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
165 	spin_unlock_irq(&bd->lock);
166 	io_schedule();
167 	finish_wait(&bd->wq_done, &wait);
168 
169 	return ret;
170 unlock:
171 	spin_unlock_irq(&bd->lock);
172 	return ret;
173 }
174 
175 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
176 				struct sg_io_v4 *hdr, int has_write_perm)
177 {
178 	memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
179 
180 	if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
181 			   hdr->request_len))
182 		return -EFAULT;
183 
184 	if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
185 		if (blk_verify_command(rq->cmd, has_write_perm))
186 			return -EPERM;
187 	} else if (!capable(CAP_SYS_RAWIO))
188 		return -EPERM;
189 
190 	/*
191 	 * fill in request structure
192 	 */
193 	rq->cmd_len = hdr->request_len;
194 	rq->cmd_type = REQ_TYPE_BLOCK_PC;
195 
196 	rq->timeout = (hdr->timeout * HZ) / 1000;
197 	if (!rq->timeout)
198 		rq->timeout = q->sg_timeout;
199 	if (!rq->timeout)
200 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
201 
202 	return 0;
203 }
204 
205 /*
206  * Check if sg_io_v4 from user is allowed and valid
207  */
208 static int
209 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
210 {
211 	int ret = 0;
212 
213 	if (hdr->guard != 'Q')
214 		return -EINVAL;
215 	if (hdr->request_len > BLK_MAX_CDB)
216 		return -EINVAL;
217 	if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
218 	    hdr->din_xfer_len > (q->max_sectors << 9))
219 		return -EIO;
220 
221 	switch (hdr->protocol) {
222 	case BSG_PROTOCOL_SCSI:
223 		switch (hdr->subprotocol) {
224 		case BSG_SUB_PROTOCOL_SCSI_CMD:
225 		case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
226 			break;
227 		default:
228 			ret = -EINVAL;
229 		}
230 		break;
231 	default:
232 		ret = -EINVAL;
233 	}
234 
235 	*rw = hdr->dout_xfer_len ? WRITE : READ;
236 	return ret;
237 }
238 
239 /*
240  * map sg_io_v4 to a request.
241  */
242 static struct request *
243 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
244 {
245 	struct request_queue *q = bd->queue;
246 	struct request *rq, *next_rq = NULL;
247 	int ret, rw;
248 	unsigned int dxfer_len;
249 	void *dxferp = NULL;
250 
251 	dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
252 		hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
253 		hdr->din_xfer_len);
254 
255 	ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
256 	if (ret)
257 		return ERR_PTR(ret);
258 
259 	/*
260 	 * map scatter-gather elements seperately and string them to request
261 	 */
262 	rq = blk_get_request(q, rw, GFP_KERNEL);
263 	if (!rq)
264 		return ERR_PTR(-ENOMEM);
265 	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
266 						       &bd->flags));
267 	if (ret)
268 		goto out;
269 
270 	if (rw == WRITE && hdr->din_xfer_len) {
271 		if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
272 			ret = -EOPNOTSUPP;
273 			goto out;
274 		}
275 
276 		next_rq = blk_get_request(q, READ, GFP_KERNEL);
277 		if (!next_rq) {
278 			ret = -ENOMEM;
279 			goto out;
280 		}
281 		rq->next_rq = next_rq;
282 
283 		dxferp = (void*)(unsigned long)hdr->din_xferp;
284 		ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
285 		if (ret)
286 			goto out;
287 	}
288 
289 	if (hdr->dout_xfer_len) {
290 		dxfer_len = hdr->dout_xfer_len;
291 		dxferp = (void*)(unsigned long)hdr->dout_xferp;
292 	} else if (hdr->din_xfer_len) {
293 		dxfer_len = hdr->din_xfer_len;
294 		dxferp = (void*)(unsigned long)hdr->din_xferp;
295 	} else
296 		dxfer_len = 0;
297 
298 	if (dxfer_len) {
299 		ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
300 		if (ret)
301 			goto out;
302 	}
303 	return rq;
304 out:
305 	blk_put_request(rq);
306 	if (next_rq) {
307 		blk_rq_unmap_user(next_rq->bio);
308 		blk_put_request(next_rq);
309 	}
310 	return ERR_PTR(ret);
311 }
312 
313 /*
314  * async completion call-back from the block layer, when scsi/ide/whatever
315  * calls end_that_request_last() on a request
316  */
317 static void bsg_rq_end_io(struct request *rq, int uptodate)
318 {
319 	struct bsg_command *bc = rq->end_io_data;
320 	struct bsg_device *bd = bc->bd;
321 	unsigned long flags;
322 
323 	dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
324 		bd->name, rq, bc, bc->bio, uptodate);
325 
326 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
327 
328 	spin_lock_irqsave(&bd->lock, flags);
329 	list_move_tail(&bc->list, &bd->done_list);
330 	bd->done_cmds++;
331 	spin_unlock_irqrestore(&bd->lock, flags);
332 
333 	wake_up(&bd->wq_done);
334 }
335 
336 /*
337  * do final setup of a 'bc' and submit the matching 'rq' to the block
338  * layer for io
339  */
340 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
341 			    struct bsg_command *bc, struct request *rq)
342 {
343 	rq->sense = bc->sense;
344 	rq->sense_len = 0;
345 
346 	/*
347 	 * add bc command to busy queue and submit rq for io
348 	 */
349 	bc->rq = rq;
350 	bc->bio = rq->bio;
351 	if (rq->next_rq)
352 		bc->bidi_bio = rq->next_rq->bio;
353 	bc->hdr.duration = jiffies;
354 	spin_lock_irq(&bd->lock);
355 	list_add_tail(&bc->list, &bd->busy_list);
356 	spin_unlock_irq(&bd->lock);
357 
358 	dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
359 
360 	rq->end_io_data = bc;
361 	blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
362 }
363 
364 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
365 {
366 	struct bsg_command *bc = NULL;
367 
368 	spin_lock_irq(&bd->lock);
369 	if (bd->done_cmds) {
370 		bc = list_entry(bd->done_list.next, struct bsg_command, list);
371 		list_del(&bc->list);
372 		bd->done_cmds--;
373 	}
374 	spin_unlock_irq(&bd->lock);
375 
376 	return bc;
377 }
378 
379 /*
380  * Get a finished command from the done list
381  */
382 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
383 {
384 	struct bsg_command *bc;
385 	int ret;
386 
387 	do {
388 		bc = bsg_next_done_cmd(bd);
389 		if (bc)
390 			break;
391 
392 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
393 			bc = ERR_PTR(-EAGAIN);
394 			break;
395 		}
396 
397 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
398 		if (ret) {
399 			bc = ERR_PTR(-ERESTARTSYS);
400 			break;
401 		}
402 	} while (1);
403 
404 	dprintk("%s: returning done %p\n", bd->name, bc);
405 
406 	return bc;
407 }
408 
409 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
410 				    struct bio *bio, struct bio *bidi_bio)
411 {
412 	int ret = 0;
413 
414 	dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
415 	/*
416 	 * fill in all the output members
417 	 */
418 	hdr->device_status = status_byte(rq->errors);
419 	hdr->transport_status = host_byte(rq->errors);
420 	hdr->driver_status = driver_byte(rq->errors);
421 	hdr->info = 0;
422 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
423 		hdr->info |= SG_INFO_CHECK;
424 	hdr->din_resid = rq->data_len;
425 	hdr->response_len = 0;
426 
427 	if (rq->sense_len && hdr->response) {
428 		int len = min_t(unsigned int, hdr->max_response_len,
429 					rq->sense_len);
430 
431 		ret = copy_to_user((void*)(unsigned long)hdr->response,
432 				   rq->sense, len);
433 		if (!ret)
434 			hdr->response_len = len;
435 		else
436 			ret = -EFAULT;
437 	}
438 
439 	if (rq->next_rq) {
440 		blk_rq_unmap_user(bidi_bio);
441 		blk_put_request(rq->next_rq);
442 	}
443 
444 	blk_rq_unmap_user(bio);
445 	blk_put_request(rq);
446 
447 	return ret;
448 }
449 
450 static int bsg_complete_all_commands(struct bsg_device *bd)
451 {
452 	struct bsg_command *bc;
453 	int ret, tret;
454 
455 	dprintk("%s: entered\n", bd->name);
456 
457 	set_bit(BSG_F_BLOCK, &bd->flags);
458 
459 	/*
460 	 * wait for all commands to complete
461 	 */
462 	ret = 0;
463 	do {
464 		ret = bsg_io_schedule(bd);
465 		/*
466 		 * look for -ENODATA specifically -- we'll sometimes get
467 		 * -ERESTARTSYS when we've taken a signal, but we can't
468 		 * return until we're done freeing the queue, so ignore
469 		 * it.  The signal will get handled when we're done freeing
470 		 * the bsg_device.
471 		 */
472 	} while (ret != -ENODATA);
473 
474 	/*
475 	 * discard done commands
476 	 */
477 	ret = 0;
478 	do {
479 		spin_lock_irq(&bd->lock);
480 		if (!bd->queued_cmds) {
481 			spin_unlock_irq(&bd->lock);
482 			break;
483 		}
484 		spin_unlock_irq(&bd->lock);
485 
486 		bc = bsg_get_done_cmd(bd);
487 		if (IS_ERR(bc))
488 			break;
489 
490 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
491 						bc->bidi_bio);
492 		if (!ret)
493 			ret = tret;
494 
495 		bsg_free_command(bc);
496 	} while (1);
497 
498 	return ret;
499 }
500 
501 static int
502 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
503 	   const struct iovec *iov, ssize_t *bytes_read)
504 {
505 	struct bsg_command *bc;
506 	int nr_commands, ret;
507 
508 	if (count % sizeof(struct sg_io_v4))
509 		return -EINVAL;
510 
511 	ret = 0;
512 	nr_commands = count / sizeof(struct sg_io_v4);
513 	while (nr_commands) {
514 		bc = bsg_get_done_cmd(bd);
515 		if (IS_ERR(bc)) {
516 			ret = PTR_ERR(bc);
517 			break;
518 		}
519 
520 		/*
521 		 * this is the only case where we need to copy data back
522 		 * after completing the request. so do that here,
523 		 * bsg_complete_work() cannot do that for us
524 		 */
525 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
526 					       bc->bidi_bio);
527 
528 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
529 			ret = -EFAULT;
530 
531 		bsg_free_command(bc);
532 
533 		if (ret)
534 			break;
535 
536 		buf += sizeof(struct sg_io_v4);
537 		*bytes_read += sizeof(struct sg_io_v4);
538 		nr_commands--;
539 	}
540 
541 	return ret;
542 }
543 
544 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
545 {
546 	if (file->f_flags & O_NONBLOCK)
547 		clear_bit(BSG_F_BLOCK, &bd->flags);
548 	else
549 		set_bit(BSG_F_BLOCK, &bd->flags);
550 }
551 
552 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
553 {
554 	if (file->f_mode & FMODE_WRITE)
555 		set_bit(BSG_F_WRITE_PERM, &bd->flags);
556 	else
557 		clear_bit(BSG_F_WRITE_PERM, &bd->flags);
558 }
559 
560 /*
561  * Check if the error is a "real" error that we should return.
562  */
563 static inline int err_block_err(int ret)
564 {
565 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
566 		return 1;
567 
568 	return 0;
569 }
570 
571 static ssize_t
572 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
573 {
574 	struct bsg_device *bd = file->private_data;
575 	int ret;
576 	ssize_t bytes_read;
577 
578 	dprintk("%s: read %Zd bytes\n", bd->name, count);
579 
580 	bsg_set_block(bd, file);
581 	bytes_read = 0;
582 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
583 	*ppos = bytes_read;
584 
585 	if (!bytes_read || (bytes_read && err_block_err(ret)))
586 		bytes_read = ret;
587 
588 	return bytes_read;
589 }
590 
591 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
592 		       size_t count, ssize_t *bytes_written)
593 {
594 	struct bsg_command *bc;
595 	struct request *rq;
596 	int ret, nr_commands;
597 
598 	if (count % sizeof(struct sg_io_v4))
599 		return -EINVAL;
600 
601 	nr_commands = count / sizeof(struct sg_io_v4);
602 	rq = NULL;
603 	bc = NULL;
604 	ret = 0;
605 	while (nr_commands) {
606 		struct request_queue *q = bd->queue;
607 
608 		bc = bsg_alloc_command(bd);
609 		if (IS_ERR(bc)) {
610 			ret = PTR_ERR(bc);
611 			bc = NULL;
612 			break;
613 		}
614 
615 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
616 			ret = -EFAULT;
617 			break;
618 		}
619 
620 		/*
621 		 * get a request, fill in the blanks, and add to request queue
622 		 */
623 		rq = bsg_map_hdr(bd, &bc->hdr);
624 		if (IS_ERR(rq)) {
625 			ret = PTR_ERR(rq);
626 			rq = NULL;
627 			break;
628 		}
629 
630 		bsg_add_command(bd, q, bc, rq);
631 		bc = NULL;
632 		rq = NULL;
633 		nr_commands--;
634 		buf += sizeof(struct sg_io_v4);
635 		*bytes_written += sizeof(struct sg_io_v4);
636 	}
637 
638 	if (bc)
639 		bsg_free_command(bc);
640 
641 	return ret;
642 }
643 
644 static ssize_t
645 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
646 {
647 	struct bsg_device *bd = file->private_data;
648 	ssize_t bytes_written;
649 	int ret;
650 
651 	dprintk("%s: write %Zd bytes\n", bd->name, count);
652 
653 	bsg_set_block(bd, file);
654 	bsg_set_write_perm(bd, file);
655 
656 	bytes_written = 0;
657 	ret = __bsg_write(bd, buf, count, &bytes_written);
658 	*ppos = bytes_written;
659 
660 	/*
661 	 * return bytes written on non-fatal errors
662 	 */
663 	if (!bytes_written || (bytes_written && err_block_err(ret)))
664 		bytes_written = ret;
665 
666 	dprintk("%s: returning %Zd\n", bd->name, bytes_written);
667 	return bytes_written;
668 }
669 
670 static struct bsg_device *bsg_alloc_device(void)
671 {
672 	struct bsg_device *bd;
673 
674 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
675 	if (unlikely(!bd))
676 		return NULL;
677 
678 	spin_lock_init(&bd->lock);
679 
680 	bd->max_queue = BSG_DEFAULT_CMDS;
681 
682 	INIT_LIST_HEAD(&bd->busy_list);
683 	INIT_LIST_HEAD(&bd->done_list);
684 	INIT_HLIST_NODE(&bd->dev_list);
685 
686 	init_waitqueue_head(&bd->wq_free);
687 	init_waitqueue_head(&bd->wq_done);
688 	return bd;
689 }
690 
691 static int bsg_put_device(struct bsg_device *bd)
692 {
693 	int ret = 0;
694 
695 	mutex_lock(&bsg_mutex);
696 
697 	if (!atomic_dec_and_test(&bd->ref_count))
698 		goto out;
699 
700 	dprintk("%s: tearing down\n", bd->name);
701 
702 	/*
703 	 * close can always block
704 	 */
705 	set_bit(BSG_F_BLOCK, &bd->flags);
706 
707 	/*
708 	 * correct error detection baddies here again. it's the responsibility
709 	 * of the app to properly reap commands before close() if it wants
710 	 * fool-proof error detection
711 	 */
712 	ret = bsg_complete_all_commands(bd);
713 
714 	blk_put_queue(bd->queue);
715 	hlist_del(&bd->dev_list);
716 	kfree(bd);
717 out:
718 	mutex_unlock(&bsg_mutex);
719 	return ret;
720 }
721 
722 static struct bsg_device *bsg_add_device(struct inode *inode,
723 					 struct request_queue *rq,
724 					 struct file *file)
725 {
726 	struct bsg_device *bd;
727 #ifdef BSG_DEBUG
728 	unsigned char buf[32];
729 #endif
730 
731 	bd = bsg_alloc_device();
732 	if (!bd)
733 		return ERR_PTR(-ENOMEM);
734 
735 	bd->queue = rq;
736 	kobject_get(&rq->kobj);
737 	bsg_set_block(bd, file);
738 
739 	atomic_set(&bd->ref_count, 1);
740 	bd->minor = iminor(inode);
741 	mutex_lock(&bsg_mutex);
742 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(bd->minor));
743 
744 	strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
745 	dprintk("bound to <%s>, max queue %d\n",
746 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
747 
748 	mutex_unlock(&bsg_mutex);
749 	return bd;
750 }
751 
752 static struct bsg_device *__bsg_get_device(int minor)
753 {
754 	struct bsg_device *bd = NULL;
755 	struct hlist_node *entry;
756 
757 	mutex_lock(&bsg_mutex);
758 
759 	hlist_for_each(entry, bsg_dev_idx_hash(minor)) {
760 		bd = hlist_entry(entry, struct bsg_device, dev_list);
761 		if (bd->minor == minor) {
762 			atomic_inc(&bd->ref_count);
763 			break;
764 		}
765 
766 		bd = NULL;
767 	}
768 
769 	mutex_unlock(&bsg_mutex);
770 	return bd;
771 }
772 
773 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
774 {
775 	struct bsg_device *bd;
776 	struct bsg_class_device *bcd;
777 
778 	bd = __bsg_get_device(iminor(inode));
779 	if (bd)
780 		return bd;
781 
782 	/*
783 	 * find the class device
784 	 */
785 	mutex_lock(&bsg_mutex);
786 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
787 	mutex_unlock(&bsg_mutex);
788 
789 	if (!bcd)
790 		return ERR_PTR(-ENODEV);
791 
792 	return bsg_add_device(inode, bcd->queue, file);
793 }
794 
795 static int bsg_open(struct inode *inode, struct file *file)
796 {
797 	struct bsg_device *bd = bsg_get_device(inode, file);
798 
799 	if (IS_ERR(bd))
800 		return PTR_ERR(bd);
801 
802 	file->private_data = bd;
803 	return 0;
804 }
805 
806 static int bsg_release(struct inode *inode, struct file *file)
807 {
808 	struct bsg_device *bd = file->private_data;
809 
810 	file->private_data = NULL;
811 	return bsg_put_device(bd);
812 }
813 
814 static unsigned int bsg_poll(struct file *file, poll_table *wait)
815 {
816 	struct bsg_device *bd = file->private_data;
817 	unsigned int mask = 0;
818 
819 	poll_wait(file, &bd->wq_done, wait);
820 	poll_wait(file, &bd->wq_free, wait);
821 
822 	spin_lock_irq(&bd->lock);
823 	if (!list_empty(&bd->done_list))
824 		mask |= POLLIN | POLLRDNORM;
825 	if (bd->queued_cmds >= bd->max_queue)
826 		mask |= POLLOUT;
827 	spin_unlock_irq(&bd->lock);
828 
829 	return mask;
830 }
831 
832 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
833 {
834 	struct bsg_device *bd = file->private_data;
835 	int __user *uarg = (int __user *) arg;
836 
837 	switch (cmd) {
838 		/*
839 		 * our own ioctls
840 		 */
841 	case SG_GET_COMMAND_Q:
842 		return put_user(bd->max_queue, uarg);
843 	case SG_SET_COMMAND_Q: {
844 		int queue;
845 
846 		if (get_user(queue, uarg))
847 			return -EFAULT;
848 		if (queue < 1)
849 			return -EINVAL;
850 
851 		spin_lock_irq(&bd->lock);
852 		bd->max_queue = queue;
853 		spin_unlock_irq(&bd->lock);
854 		return 0;
855 	}
856 
857 	/*
858 	 * SCSI/sg ioctls
859 	 */
860 	case SG_GET_VERSION_NUM:
861 	case SCSI_IOCTL_GET_IDLUN:
862 	case SCSI_IOCTL_GET_BUS_NUMBER:
863 	case SG_SET_TIMEOUT:
864 	case SG_GET_TIMEOUT:
865 	case SG_GET_RESERVED_SIZE:
866 	case SG_SET_RESERVED_SIZE:
867 	case SG_EMULATED_HOST:
868 	case SCSI_IOCTL_SEND_COMMAND: {
869 		void __user *uarg = (void __user *) arg;
870 		return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
871 	}
872 	case SG_IO: {
873 		struct request *rq;
874 		struct bio *bio, *bidi_bio = NULL;
875 		struct sg_io_v4 hdr;
876 
877 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
878 			return -EFAULT;
879 
880 		rq = bsg_map_hdr(bd, &hdr);
881 		if (IS_ERR(rq))
882 			return PTR_ERR(rq);
883 
884 		bio = rq->bio;
885 		if (rq->next_rq)
886 			bidi_bio = rq->next_rq->bio;
887 		blk_execute_rq(bd->queue, NULL, rq, 0);
888 		blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
889 
890 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
891 			return -EFAULT;
892 
893 		return 0;
894 	}
895 	/*
896 	 * block device ioctls
897 	 */
898 	default:
899 #if 0
900 		return ioctl_by_bdev(bd->bdev, cmd, arg);
901 #else
902 		return -ENOTTY;
903 #endif
904 	}
905 }
906 
907 static struct file_operations bsg_fops = {
908 	.read		=	bsg_read,
909 	.write		=	bsg_write,
910 	.poll		=	bsg_poll,
911 	.open		=	bsg_open,
912 	.release	=	bsg_release,
913 	.unlocked_ioctl	=	bsg_ioctl,
914 	.owner		=	THIS_MODULE,
915 };
916 
917 void bsg_unregister_queue(struct request_queue *q)
918 {
919 	struct bsg_class_device *bcd = &q->bsg_dev;
920 
921 	if (!bcd->class_dev)
922 		return;
923 
924 	mutex_lock(&bsg_mutex);
925 	idr_remove(&bsg_minor_idr, bcd->minor);
926 	sysfs_remove_link(&q->kobj, "bsg");
927 	class_device_unregister(bcd->class_dev);
928 	put_device(bcd->dev);
929 	bcd->class_dev = NULL;
930 	bcd->dev = NULL;
931 	mutex_unlock(&bsg_mutex);
932 }
933 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
934 
935 int bsg_register_queue(struct request_queue *q, struct device *gdev,
936 		       const char *name)
937 {
938 	struct bsg_class_device *bcd;
939 	dev_t dev;
940 	int ret, minor;
941 	struct class_device *class_dev = NULL;
942 	const char *devname;
943 
944 	if (name)
945 		devname = name;
946 	else
947 		devname = gdev->bus_id;
948 
949 	/*
950 	 * we need a proper transport to send commands, not a stacked device
951 	 */
952 	if (!q->request_fn)
953 		return 0;
954 
955 	bcd = &q->bsg_dev;
956 	memset(bcd, 0, sizeof(*bcd));
957 
958 	mutex_lock(&bsg_mutex);
959 
960 	ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
961 	if (!ret) {
962 		ret = -ENOMEM;
963 		goto unlock;
964 	}
965 
966 	ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
967 	if (ret < 0)
968 		goto unlock;
969 
970 	if (minor >= BSG_MAX_DEVS) {
971 		printk(KERN_ERR "bsg: too many bsg devices\n");
972 		ret = -EINVAL;
973 		goto remove_idr;
974 	}
975 
976 	bcd->minor = minor;
977 	bcd->queue = q;
978 	bcd->dev = get_device(gdev);
979 	dev = MKDEV(bsg_major, bcd->minor);
980 	class_dev = class_device_create(bsg_class, NULL, dev, gdev, "%s",
981 					devname);
982 	if (IS_ERR(class_dev)) {
983 		ret = PTR_ERR(class_dev);
984 		goto put_dev;
985 	}
986 	bcd->class_dev = class_dev;
987 
988 	if (q->kobj.sd) {
989 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
990 		if (ret)
991 			goto unregister_class_dev;
992 	}
993 
994 	mutex_unlock(&bsg_mutex);
995 	return 0;
996 
997 unregister_class_dev:
998 	class_device_unregister(class_dev);
999 put_dev:
1000 	put_device(gdev);
1001 remove_idr:
1002 	idr_remove(&bsg_minor_idr, minor);
1003 unlock:
1004 	mutex_unlock(&bsg_mutex);
1005 	return ret;
1006 }
1007 EXPORT_SYMBOL_GPL(bsg_register_queue);
1008 
1009 static struct cdev bsg_cdev = {
1010 	.kobj   = {.name = "bsg", },
1011 	.owner  = THIS_MODULE,
1012 };
1013 
1014 static int __init bsg_init(void)
1015 {
1016 	int ret, i;
1017 	dev_t devid;
1018 
1019 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1020 				sizeof(struct bsg_command), 0, 0, NULL);
1021 	if (!bsg_cmd_cachep) {
1022 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1023 		return -ENOMEM;
1024 	}
1025 
1026 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1027 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1028 
1029 	bsg_class = class_create(THIS_MODULE, "bsg");
1030 	if (IS_ERR(bsg_class)) {
1031 		ret = PTR_ERR(bsg_class);
1032 		goto destroy_kmemcache;
1033 	}
1034 
1035 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1036 	if (ret)
1037 		goto destroy_bsg_class;
1038 
1039 	bsg_major = MAJOR(devid);
1040 
1041 	cdev_init(&bsg_cdev, &bsg_fops);
1042 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1043 	if (ret)
1044 		goto unregister_chrdev;
1045 
1046 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1047 	       " loaded (major %d)\n", bsg_major);
1048 	return 0;
1049 unregister_chrdev:
1050 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1051 destroy_bsg_class:
1052 	class_destroy(bsg_class);
1053 destroy_kmemcache:
1054 	kmem_cache_destroy(bsg_cmd_cachep);
1055 	return ret;
1056 }
1057 
1058 MODULE_AUTHOR("Jens Axboe");
1059 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1060 MODULE_LICENSE("GPL");
1061 
1062 device_initcall(bsg_init);
1063