xref: /openbmc/linux/block/bsg.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
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/smp_lock.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 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, struct bsg_device *bd,
177 				fmode_t has_write_perm)
178 {
179 	if (hdr->request_len > BLK_MAX_CDB) {
180 		rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
181 		if (!rq->cmd)
182 			return -ENOMEM;
183 	}
184 
185 	if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
186 			   hdr->request_len))
187 		return -EFAULT;
188 
189 	if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
190 		if (blk_verify_command(rq->cmd, has_write_perm))
191 			return -EPERM;
192 	} else if (!capable(CAP_SYS_RAWIO))
193 		return -EPERM;
194 
195 	/*
196 	 * fill in request structure
197 	 */
198 	rq->cmd_len = hdr->request_len;
199 	rq->cmd_type = REQ_TYPE_BLOCK_PC;
200 
201 	rq->timeout = msecs_to_jiffies(hdr->timeout);
202 	if (!rq->timeout)
203 		rq->timeout = q->sg_timeout;
204 	if (!rq->timeout)
205 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
206 	if (rq->timeout < BLK_MIN_SG_TIMEOUT)
207 		rq->timeout = BLK_MIN_SG_TIMEOUT;
208 
209 	return 0;
210 }
211 
212 /*
213  * Check if sg_io_v4 from user is allowed and valid
214  */
215 static int
216 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
217 {
218 	int ret = 0;
219 
220 	if (hdr->guard != 'Q')
221 		return -EINVAL;
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 	    u8 *sense)
247 {
248 	struct request_queue *q = bd->queue;
249 	struct request *rq, *next_rq = NULL;
250 	int ret, rw;
251 	unsigned int dxfer_len;
252 	void *dxferp = NULL;
253 
254 	dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
255 		hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
256 		hdr->din_xfer_len);
257 
258 	ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
259 	if (ret)
260 		return ERR_PTR(ret);
261 
262 	/*
263 	 * map scatter-gather elements seperately and string them to request
264 	 */
265 	rq = blk_get_request(q, rw, GFP_KERNEL);
266 	if (!rq)
267 		return ERR_PTR(-ENOMEM);
268 	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
269 	if (ret)
270 		goto out;
271 
272 	if (rw == WRITE && hdr->din_xfer_len) {
273 		if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
274 			ret = -EOPNOTSUPP;
275 			goto out;
276 		}
277 
278 		next_rq = blk_get_request(q, READ, GFP_KERNEL);
279 		if (!next_rq) {
280 			ret = -ENOMEM;
281 			goto out;
282 		}
283 		rq->next_rq = next_rq;
284 		next_rq->cmd_type = rq->cmd_type;
285 
286 		dxferp = (void*)(unsigned long)hdr->din_xferp;
287 		ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
288 				       hdr->din_xfer_len, GFP_KERNEL);
289 		if (ret)
290 			goto out;
291 	}
292 
293 	if (hdr->dout_xfer_len) {
294 		dxfer_len = hdr->dout_xfer_len;
295 		dxferp = (void*)(unsigned long)hdr->dout_xferp;
296 	} else if (hdr->din_xfer_len) {
297 		dxfer_len = hdr->din_xfer_len;
298 		dxferp = (void*)(unsigned long)hdr->din_xferp;
299 	} else
300 		dxfer_len = 0;
301 
302 	if (dxfer_len) {
303 		ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
304 				      GFP_KERNEL);
305 		if (ret)
306 			goto out;
307 	}
308 
309 	rq->sense = sense;
310 	rq->sense_len = 0;
311 
312 	return rq;
313 out:
314 	if (rq->cmd != rq->__cmd)
315 		kfree(rq->cmd);
316 	blk_put_request(rq);
317 	if (next_rq) {
318 		blk_rq_unmap_user(next_rq->bio);
319 		blk_put_request(next_rq);
320 	}
321 	return ERR_PTR(ret);
322 }
323 
324 /*
325  * async completion call-back from the block layer, when scsi/ide/whatever
326  * calls end_that_request_last() on a request
327  */
328 static void bsg_rq_end_io(struct request *rq, int uptodate)
329 {
330 	struct bsg_command *bc = rq->end_io_data;
331 	struct bsg_device *bd = bc->bd;
332 	unsigned long flags;
333 
334 	dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
335 		bd->name, rq, bc, bc->bio, uptodate);
336 
337 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
338 
339 	spin_lock_irqsave(&bd->lock, flags);
340 	list_move_tail(&bc->list, &bd->done_list);
341 	bd->done_cmds++;
342 	spin_unlock_irqrestore(&bd->lock, flags);
343 
344 	wake_up(&bd->wq_done);
345 }
346 
347 /*
348  * do final setup of a 'bc' and submit the matching 'rq' to the block
349  * layer for io
350  */
351 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
352 			    struct bsg_command *bc, struct request *rq)
353 {
354 	int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
355 
356 	/*
357 	 * add bc command to busy queue and submit rq for io
358 	 */
359 	bc->rq = rq;
360 	bc->bio = rq->bio;
361 	if (rq->next_rq)
362 		bc->bidi_bio = rq->next_rq->bio;
363 	bc->hdr.duration = jiffies;
364 	spin_lock_irq(&bd->lock);
365 	list_add_tail(&bc->list, &bd->busy_list);
366 	spin_unlock_irq(&bd->lock);
367 
368 	dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
369 
370 	rq->end_io_data = bc;
371 	blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
372 }
373 
374 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
375 {
376 	struct bsg_command *bc = NULL;
377 
378 	spin_lock_irq(&bd->lock);
379 	if (bd->done_cmds) {
380 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
381 		list_del(&bc->list);
382 		bd->done_cmds--;
383 	}
384 	spin_unlock_irq(&bd->lock);
385 
386 	return bc;
387 }
388 
389 /*
390  * Get a finished command from the done list
391  */
392 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
393 {
394 	struct bsg_command *bc;
395 	int ret;
396 
397 	do {
398 		bc = bsg_next_done_cmd(bd);
399 		if (bc)
400 			break;
401 
402 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
403 			bc = ERR_PTR(-EAGAIN);
404 			break;
405 		}
406 
407 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
408 		if (ret) {
409 			bc = ERR_PTR(-ERESTARTSYS);
410 			break;
411 		}
412 	} while (1);
413 
414 	dprintk("%s: returning done %p\n", bd->name, bc);
415 
416 	return bc;
417 }
418 
419 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
420 				    struct bio *bio, struct bio *bidi_bio)
421 {
422 	int ret = 0;
423 
424 	dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
425 	/*
426 	 * fill in all the output members
427 	 */
428 	hdr->device_status = status_byte(rq->errors);
429 	hdr->transport_status = host_byte(rq->errors);
430 	hdr->driver_status = driver_byte(rq->errors);
431 	hdr->info = 0;
432 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
433 		hdr->info |= SG_INFO_CHECK;
434 	hdr->response_len = 0;
435 
436 	if (rq->sense_len && hdr->response) {
437 		int len = min_t(unsigned int, hdr->max_response_len,
438 					rq->sense_len);
439 
440 		ret = copy_to_user((void*)(unsigned long)hdr->response,
441 				   rq->sense, len);
442 		if (!ret)
443 			hdr->response_len = len;
444 		else
445 			ret = -EFAULT;
446 	}
447 
448 	if (rq->next_rq) {
449 		hdr->dout_resid = rq->resid_len;
450 		hdr->din_resid = rq->next_rq->resid_len;
451 		blk_rq_unmap_user(bidi_bio);
452 		blk_put_request(rq->next_rq);
453 	} else if (rq_data_dir(rq) == READ)
454 		hdr->din_resid = rq->resid_len;
455 	else
456 		hdr->dout_resid = rq->resid_len;
457 
458 	/*
459 	 * If the request generated a negative error number, return it
460 	 * (providing we aren't already returning an error); if it's
461 	 * just a protocol response (i.e. non negative), that gets
462 	 * processed above.
463 	 */
464 	if (!ret && rq->errors < 0)
465 		ret = rq->errors;
466 
467 	blk_rq_unmap_user(bio);
468 	if (rq->cmd != rq->__cmd)
469 		kfree(rq->cmd);
470 	blk_put_request(rq);
471 
472 	return ret;
473 }
474 
475 static int bsg_complete_all_commands(struct bsg_device *bd)
476 {
477 	struct bsg_command *bc;
478 	int ret, tret;
479 
480 	dprintk("%s: entered\n", bd->name);
481 
482 	/*
483 	 * wait for all commands to complete
484 	 */
485 	ret = 0;
486 	do {
487 		ret = bsg_io_schedule(bd);
488 		/*
489 		 * look for -ENODATA specifically -- we'll sometimes get
490 		 * -ERESTARTSYS when we've taken a signal, but we can't
491 		 * return until we're done freeing the queue, so ignore
492 		 * it.  The signal will get handled when we're done freeing
493 		 * the bsg_device.
494 		 */
495 	} while (ret != -ENODATA);
496 
497 	/*
498 	 * discard done commands
499 	 */
500 	ret = 0;
501 	do {
502 		spin_lock_irq(&bd->lock);
503 		if (!bd->queued_cmds) {
504 			spin_unlock_irq(&bd->lock);
505 			break;
506 		}
507 		spin_unlock_irq(&bd->lock);
508 
509 		bc = bsg_get_done_cmd(bd);
510 		if (IS_ERR(bc))
511 			break;
512 
513 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
514 						bc->bidi_bio);
515 		if (!ret)
516 			ret = tret;
517 
518 		bsg_free_command(bc);
519 	} while (1);
520 
521 	return ret;
522 }
523 
524 static int
525 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
526 	   const struct iovec *iov, ssize_t *bytes_read)
527 {
528 	struct bsg_command *bc;
529 	int nr_commands, ret;
530 
531 	if (count % sizeof(struct sg_io_v4))
532 		return -EINVAL;
533 
534 	ret = 0;
535 	nr_commands = count / sizeof(struct sg_io_v4);
536 	while (nr_commands) {
537 		bc = bsg_get_done_cmd(bd);
538 		if (IS_ERR(bc)) {
539 			ret = PTR_ERR(bc);
540 			break;
541 		}
542 
543 		/*
544 		 * this is the only case where we need to copy data back
545 		 * after completing the request. so do that here,
546 		 * bsg_complete_work() cannot do that for us
547 		 */
548 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
549 					       bc->bidi_bio);
550 
551 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
552 			ret = -EFAULT;
553 
554 		bsg_free_command(bc);
555 
556 		if (ret)
557 			break;
558 
559 		buf += sizeof(struct sg_io_v4);
560 		*bytes_read += sizeof(struct sg_io_v4);
561 		nr_commands--;
562 	}
563 
564 	return ret;
565 }
566 
567 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
568 {
569 	if (file->f_flags & O_NONBLOCK)
570 		clear_bit(BSG_F_BLOCK, &bd->flags);
571 	else
572 		set_bit(BSG_F_BLOCK, &bd->flags);
573 }
574 
575 /*
576  * Check if the error is a "real" error that we should return.
577  */
578 static inline int err_block_err(int ret)
579 {
580 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
581 		return 1;
582 
583 	return 0;
584 }
585 
586 static ssize_t
587 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
588 {
589 	struct bsg_device *bd = file->private_data;
590 	int ret;
591 	ssize_t bytes_read;
592 
593 	dprintk("%s: read %Zd bytes\n", bd->name, count);
594 
595 	bsg_set_block(bd, file);
596 
597 	bytes_read = 0;
598 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
599 	*ppos = bytes_read;
600 
601 	if (!bytes_read || (bytes_read && err_block_err(ret)))
602 		bytes_read = ret;
603 
604 	return bytes_read;
605 }
606 
607 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
608 		       size_t count, ssize_t *bytes_written,
609 		       fmode_t has_write_perm)
610 {
611 	struct bsg_command *bc;
612 	struct request *rq;
613 	int ret, nr_commands;
614 
615 	if (count % sizeof(struct sg_io_v4))
616 		return -EINVAL;
617 
618 	nr_commands = count / sizeof(struct sg_io_v4);
619 	rq = NULL;
620 	bc = NULL;
621 	ret = 0;
622 	while (nr_commands) {
623 		struct request_queue *q = bd->queue;
624 
625 		bc = bsg_alloc_command(bd);
626 		if (IS_ERR(bc)) {
627 			ret = PTR_ERR(bc);
628 			bc = NULL;
629 			break;
630 		}
631 
632 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
633 			ret = -EFAULT;
634 			break;
635 		}
636 
637 		/*
638 		 * get a request, fill in the blanks, and add to request queue
639 		 */
640 		rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
641 		if (IS_ERR(rq)) {
642 			ret = PTR_ERR(rq);
643 			rq = NULL;
644 			break;
645 		}
646 
647 		bsg_add_command(bd, q, bc, rq);
648 		bc = NULL;
649 		rq = NULL;
650 		nr_commands--;
651 		buf += sizeof(struct sg_io_v4);
652 		*bytes_written += sizeof(struct sg_io_v4);
653 	}
654 
655 	if (bc)
656 		bsg_free_command(bc);
657 
658 	return ret;
659 }
660 
661 static ssize_t
662 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
663 {
664 	struct bsg_device *bd = file->private_data;
665 	ssize_t bytes_written;
666 	int ret;
667 
668 	dprintk("%s: write %Zd bytes\n", bd->name, count);
669 
670 	bsg_set_block(bd, file);
671 
672 	bytes_written = 0;
673 	ret = __bsg_write(bd, buf, count, &bytes_written,
674 			  file->f_mode & FMODE_WRITE);
675 
676 	*ppos = bytes_written;
677 
678 	/*
679 	 * return bytes written on non-fatal errors
680 	 */
681 	if (!bytes_written || (bytes_written && err_block_err(ret)))
682 		bytes_written = ret;
683 
684 	dprintk("%s: returning %Zd\n", bd->name, bytes_written);
685 	return bytes_written;
686 }
687 
688 static struct bsg_device *bsg_alloc_device(void)
689 {
690 	struct bsg_device *bd;
691 
692 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
693 	if (unlikely(!bd))
694 		return NULL;
695 
696 	spin_lock_init(&bd->lock);
697 
698 	bd->max_queue = BSG_DEFAULT_CMDS;
699 
700 	INIT_LIST_HEAD(&bd->busy_list);
701 	INIT_LIST_HEAD(&bd->done_list);
702 	INIT_HLIST_NODE(&bd->dev_list);
703 
704 	init_waitqueue_head(&bd->wq_free);
705 	init_waitqueue_head(&bd->wq_done);
706 	return bd;
707 }
708 
709 static void bsg_kref_release_function(struct kref *kref)
710 {
711 	struct bsg_class_device *bcd =
712 		container_of(kref, struct bsg_class_device, ref);
713 	struct device *parent = bcd->parent;
714 
715 	if (bcd->release)
716 		bcd->release(bcd->parent);
717 
718 	put_device(parent);
719 }
720 
721 static int bsg_put_device(struct bsg_device *bd)
722 {
723 	int ret = 0, do_free;
724 	struct request_queue *q = bd->queue;
725 
726 	mutex_lock(&bsg_mutex);
727 
728 	do_free = atomic_dec_and_test(&bd->ref_count);
729 	if (!do_free) {
730 		mutex_unlock(&bsg_mutex);
731 		goto out;
732 	}
733 
734 	hlist_del(&bd->dev_list);
735 	mutex_unlock(&bsg_mutex);
736 
737 	dprintk("%s: tearing down\n", bd->name);
738 
739 	/*
740 	 * close can always block
741 	 */
742 	set_bit(BSG_F_BLOCK, &bd->flags);
743 
744 	/*
745 	 * correct error detection baddies here again. it's the responsibility
746 	 * of the app to properly reap commands before close() if it wants
747 	 * fool-proof error detection
748 	 */
749 	ret = bsg_complete_all_commands(bd);
750 
751 	kfree(bd);
752 out:
753 	kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
754 	if (do_free)
755 		blk_put_queue(q);
756 	return ret;
757 }
758 
759 static struct bsg_device *bsg_add_device(struct inode *inode,
760 					 struct request_queue *rq,
761 					 struct file *file)
762 {
763 	struct bsg_device *bd;
764 	int ret;
765 #ifdef BSG_DEBUG
766 	unsigned char buf[32];
767 #endif
768 	ret = blk_get_queue(rq);
769 	if (ret)
770 		return ERR_PTR(-ENXIO);
771 
772 	bd = bsg_alloc_device();
773 	if (!bd) {
774 		blk_put_queue(rq);
775 		return ERR_PTR(-ENOMEM);
776 	}
777 
778 	bd->queue = rq;
779 
780 	bsg_set_block(bd, file);
781 
782 	atomic_set(&bd->ref_count, 1);
783 	mutex_lock(&bsg_mutex);
784 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
785 
786 	strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
787 	dprintk("bound to <%s>, max queue %d\n",
788 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
789 
790 	mutex_unlock(&bsg_mutex);
791 	return bd;
792 }
793 
794 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
795 {
796 	struct bsg_device *bd;
797 	struct hlist_node *entry;
798 
799 	mutex_lock(&bsg_mutex);
800 
801 	hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
802 		if (bd->queue == q) {
803 			atomic_inc(&bd->ref_count);
804 			goto found;
805 		}
806 	}
807 	bd = NULL;
808 found:
809 	mutex_unlock(&bsg_mutex);
810 	return bd;
811 }
812 
813 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
814 {
815 	struct bsg_device *bd;
816 	struct bsg_class_device *bcd;
817 
818 	/*
819 	 * find the class device
820 	 */
821 	mutex_lock(&bsg_mutex);
822 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
823 	if (bcd)
824 		kref_get(&bcd->ref);
825 	mutex_unlock(&bsg_mutex);
826 
827 	if (!bcd)
828 		return ERR_PTR(-ENODEV);
829 
830 	bd = __bsg_get_device(iminor(inode), bcd->queue);
831 	if (bd)
832 		return bd;
833 
834 	bd = bsg_add_device(inode, bcd->queue, file);
835 	if (IS_ERR(bd))
836 		kref_put(&bcd->ref, bsg_kref_release_function);
837 
838 	return bd;
839 }
840 
841 static int bsg_open(struct inode *inode, struct file *file)
842 {
843 	struct bsg_device *bd;
844 
845 	lock_kernel();
846 	bd = bsg_get_device(inode, file);
847 	unlock_kernel();
848 
849 	if (IS_ERR(bd))
850 		return PTR_ERR(bd);
851 
852 	file->private_data = bd;
853 	return 0;
854 }
855 
856 static int bsg_release(struct inode *inode, struct file *file)
857 {
858 	struct bsg_device *bd = file->private_data;
859 
860 	file->private_data = NULL;
861 	return bsg_put_device(bd);
862 }
863 
864 static unsigned int bsg_poll(struct file *file, poll_table *wait)
865 {
866 	struct bsg_device *bd = file->private_data;
867 	unsigned int mask = 0;
868 
869 	poll_wait(file, &bd->wq_done, wait);
870 	poll_wait(file, &bd->wq_free, wait);
871 
872 	spin_lock_irq(&bd->lock);
873 	if (!list_empty(&bd->done_list))
874 		mask |= POLLIN | POLLRDNORM;
875 	if (bd->queued_cmds >= bd->max_queue)
876 		mask |= POLLOUT;
877 	spin_unlock_irq(&bd->lock);
878 
879 	return mask;
880 }
881 
882 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
883 {
884 	struct bsg_device *bd = file->private_data;
885 	int __user *uarg = (int __user *) arg;
886 	int ret;
887 
888 	switch (cmd) {
889 		/*
890 		 * our own ioctls
891 		 */
892 	case SG_GET_COMMAND_Q:
893 		return put_user(bd->max_queue, uarg);
894 	case SG_SET_COMMAND_Q: {
895 		int queue;
896 
897 		if (get_user(queue, uarg))
898 			return -EFAULT;
899 		if (queue < 1)
900 			return -EINVAL;
901 
902 		spin_lock_irq(&bd->lock);
903 		bd->max_queue = queue;
904 		spin_unlock_irq(&bd->lock);
905 		return 0;
906 	}
907 
908 	/*
909 	 * SCSI/sg ioctls
910 	 */
911 	case SG_GET_VERSION_NUM:
912 	case SCSI_IOCTL_GET_IDLUN:
913 	case SCSI_IOCTL_GET_BUS_NUMBER:
914 	case SG_SET_TIMEOUT:
915 	case SG_GET_TIMEOUT:
916 	case SG_GET_RESERVED_SIZE:
917 	case SG_SET_RESERVED_SIZE:
918 	case SG_EMULATED_HOST:
919 	case SCSI_IOCTL_SEND_COMMAND: {
920 		void __user *uarg = (void __user *) arg;
921 		return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
922 	}
923 	case SG_IO: {
924 		struct request *rq;
925 		struct bio *bio, *bidi_bio = NULL;
926 		struct sg_io_v4 hdr;
927 		int at_head;
928 		u8 sense[SCSI_SENSE_BUFFERSIZE];
929 
930 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
931 			return -EFAULT;
932 
933 		rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
934 		if (IS_ERR(rq))
935 			return PTR_ERR(rq);
936 
937 		bio = rq->bio;
938 		if (rq->next_rq)
939 			bidi_bio = rq->next_rq->bio;
940 
941 		at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
942 		blk_execute_rq(bd->queue, NULL, rq, at_head);
943 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
944 
945 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
946 			return -EFAULT;
947 
948 		return ret;
949 	}
950 	/*
951 	 * block device ioctls
952 	 */
953 	default:
954 #if 0
955 		return ioctl_by_bdev(bd->bdev, cmd, arg);
956 #else
957 		return -ENOTTY;
958 #endif
959 	}
960 }
961 
962 static const struct file_operations bsg_fops = {
963 	.read		=	bsg_read,
964 	.write		=	bsg_write,
965 	.poll		=	bsg_poll,
966 	.open		=	bsg_open,
967 	.release	=	bsg_release,
968 	.unlocked_ioctl	=	bsg_ioctl,
969 	.owner		=	THIS_MODULE,
970 };
971 
972 void bsg_unregister_queue(struct request_queue *q)
973 {
974 	struct bsg_class_device *bcd = &q->bsg_dev;
975 
976 	if (!bcd->class_dev)
977 		return;
978 
979 	mutex_lock(&bsg_mutex);
980 	idr_remove(&bsg_minor_idr, bcd->minor);
981 	sysfs_remove_link(&q->kobj, "bsg");
982 	device_unregister(bcd->class_dev);
983 	bcd->class_dev = NULL;
984 	kref_put(&bcd->ref, bsg_kref_release_function);
985 	mutex_unlock(&bsg_mutex);
986 }
987 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
988 
989 int bsg_register_queue(struct request_queue *q, struct device *parent,
990 		       const char *name, void (*release)(struct device *))
991 {
992 	struct bsg_class_device *bcd;
993 	dev_t dev;
994 	int ret, minor;
995 	struct device *class_dev = NULL;
996 	const char *devname;
997 
998 	if (name)
999 		devname = name;
1000 	else
1001 		devname = dev_name(parent);
1002 
1003 	/*
1004 	 * we need a proper transport to send commands, not a stacked device
1005 	 */
1006 	if (!q->request_fn)
1007 		return 0;
1008 
1009 	bcd = &q->bsg_dev;
1010 	memset(bcd, 0, sizeof(*bcd));
1011 
1012 	mutex_lock(&bsg_mutex);
1013 
1014 	ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1015 	if (!ret) {
1016 		ret = -ENOMEM;
1017 		goto unlock;
1018 	}
1019 
1020 	ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1021 	if (ret < 0)
1022 		goto unlock;
1023 
1024 	if (minor >= BSG_MAX_DEVS) {
1025 		printk(KERN_ERR "bsg: too many bsg devices\n");
1026 		ret = -EINVAL;
1027 		goto remove_idr;
1028 	}
1029 
1030 	bcd->minor = minor;
1031 	bcd->queue = q;
1032 	bcd->parent = get_device(parent);
1033 	bcd->release = release;
1034 	kref_init(&bcd->ref);
1035 	dev = MKDEV(bsg_major, bcd->minor);
1036 	class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1037 	if (IS_ERR(class_dev)) {
1038 		ret = PTR_ERR(class_dev);
1039 		goto put_dev;
1040 	}
1041 	bcd->class_dev = class_dev;
1042 
1043 	if (q->kobj.sd) {
1044 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1045 		if (ret)
1046 			goto unregister_class_dev;
1047 	}
1048 
1049 	mutex_unlock(&bsg_mutex);
1050 	return 0;
1051 
1052 unregister_class_dev:
1053 	device_unregister(class_dev);
1054 put_dev:
1055 	put_device(parent);
1056 remove_idr:
1057 	idr_remove(&bsg_minor_idr, minor);
1058 unlock:
1059 	mutex_unlock(&bsg_mutex);
1060 	return ret;
1061 }
1062 EXPORT_SYMBOL_GPL(bsg_register_queue);
1063 
1064 static struct cdev bsg_cdev;
1065 
1066 static char *bsg_devnode(struct device *dev, mode_t *mode)
1067 {
1068 	return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1069 }
1070 
1071 static int __init bsg_init(void)
1072 {
1073 	int ret, i;
1074 	dev_t devid;
1075 
1076 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1077 				sizeof(struct bsg_command), 0, 0, NULL);
1078 	if (!bsg_cmd_cachep) {
1079 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1080 		return -ENOMEM;
1081 	}
1082 
1083 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1084 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1085 
1086 	bsg_class = class_create(THIS_MODULE, "bsg");
1087 	if (IS_ERR(bsg_class)) {
1088 		ret = PTR_ERR(bsg_class);
1089 		goto destroy_kmemcache;
1090 	}
1091 	bsg_class->devnode = bsg_devnode;
1092 
1093 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1094 	if (ret)
1095 		goto destroy_bsg_class;
1096 
1097 	bsg_major = MAJOR(devid);
1098 
1099 	cdev_init(&bsg_cdev, &bsg_fops);
1100 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1101 	if (ret)
1102 		goto unregister_chrdev;
1103 
1104 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1105 	       " loaded (major %d)\n", bsg_major);
1106 	return 0;
1107 unregister_chrdev:
1108 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1109 destroy_bsg_class:
1110 	class_destroy(bsg_class);
1111 destroy_kmemcache:
1112 	kmem_cache_destroy(bsg_cmd_cachep);
1113 	return ret;
1114 }
1115 
1116 MODULE_AUTHOR("Jens Axboe");
1117 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1118 MODULE_LICENSE("GPL");
1119 
1120 device_initcall(bsg_init);
1121