xref: /linux/block/bsg.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  * bsg.c - block layer implementation of the sg v4 interface
3  *
4  * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
5  * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
6  *
7  *  This file is subject to the terms and conditions of the GNU General Public
8  *  License version 2.  See the file "COPYING" in the main directory of this
9  *  archive for more details.
10  *
11  */
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/file.h>
15 #include <linux/blkdev.h>
16 #include <linux/poll.h>
17 #include <linux/cdev.h>
18 #include <linux/jiffies.h>
19 #include <linux/percpu.h>
20 #include <linux/uio.h>
21 #include <linux/idr.h>
22 #include <linux/bsg.h>
23 #include <linux/slab.h>
24 
25 #include <scsi/scsi.h>
26 #include <scsi/scsi_ioctl.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_device.h>
29 #include <scsi/scsi_driver.h>
30 #include <scsi/sg.h>
31 
32 #define BSG_DESCRIPTION	"Block layer SCSI generic (bsg) driver"
33 #define BSG_VERSION	"0.4"
34 
35 struct bsg_device {
36 	struct request_queue *queue;
37 	spinlock_t lock;
38 	struct list_head busy_list;
39 	struct list_head done_list;
40 	struct hlist_node dev_list;
41 	atomic_t ref_count;
42 	int queued_cmds;
43 	int done_cmds;
44 	wait_queue_head_t wq_done;
45 	wait_queue_head_t wq_free;
46 	char name[20];
47 	int max_queue;
48 	unsigned long flags;
49 };
50 
51 enum {
52 	BSG_F_BLOCK		= 1,
53 };
54 
55 #define BSG_DEFAULT_CMDS	64
56 #define BSG_MAX_DEVS		32768
57 
58 #undef BSG_DEBUG
59 
60 #ifdef BSG_DEBUG
61 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __func__, ##args)
62 #else
63 #define dprintk(fmt, args...)
64 #endif
65 
66 static DEFINE_MUTEX(bsg_mutex);
67 static DEFINE_IDR(bsg_minor_idr);
68 
69 #define BSG_LIST_ARRAY_SIZE	8
70 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
71 
72 static struct class *bsg_class;
73 static int bsg_major;
74 
75 static struct kmem_cache *bsg_cmd_cachep;
76 
77 /*
78  * our internal command type
79  */
80 struct bsg_command {
81 	struct bsg_device *bd;
82 	struct list_head list;
83 	struct request *rq;
84 	struct bio *bio;
85 	struct bio *bidi_bio;
86 	int err;
87 	struct sg_io_v4 hdr;
88 };
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 blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
139 				struct sg_io_v4 *hdr, struct bsg_device *bd,
140 				fmode_t mode)
141 {
142 	struct scsi_request *req = scsi_req(rq);
143 
144 	if (hdr->request_len > BLK_MAX_CDB) {
145 		req->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
146 		if (!req->cmd)
147 			return -ENOMEM;
148 	}
149 
150 	if (copy_from_user(req->cmd, (void __user *)(unsigned long)hdr->request,
151 			   hdr->request_len))
152 		return -EFAULT;
153 
154 	if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
155 		if (blk_verify_command(req->cmd, mode))
156 			return -EPERM;
157 	} else if (!capable(CAP_SYS_RAWIO))
158 		return -EPERM;
159 
160 	/*
161 	 * fill in request structure
162 	 */
163 	req->cmd_len = hdr->request_len;
164 
165 	rq->timeout = msecs_to_jiffies(hdr->timeout);
166 	if (!rq->timeout)
167 		rq->timeout = q->sg_timeout;
168 	if (!rq->timeout)
169 		rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
170 	if (rq->timeout < BLK_MIN_SG_TIMEOUT)
171 		rq->timeout = BLK_MIN_SG_TIMEOUT;
172 
173 	return 0;
174 }
175 
176 /*
177  * Check if sg_io_v4 from user is allowed and valid
178  */
179 static int
180 bsg_validate_sgv4_hdr(struct sg_io_v4 *hdr, int *op)
181 {
182 	int ret = 0;
183 
184 	if (hdr->guard != 'Q')
185 		return -EINVAL;
186 
187 	switch (hdr->protocol) {
188 	case BSG_PROTOCOL_SCSI:
189 		switch (hdr->subprotocol) {
190 		case BSG_SUB_PROTOCOL_SCSI_CMD:
191 		case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
192 			break;
193 		default:
194 			ret = -EINVAL;
195 		}
196 		break;
197 	default:
198 		ret = -EINVAL;
199 	}
200 
201 	*op = hdr->dout_xfer_len ? REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN;
202 	return ret;
203 }
204 
205 /*
206  * map sg_io_v4 to a request.
207  */
208 static struct request *
209 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t mode)
210 {
211 	struct request_queue *q = bd->queue;
212 	struct request *rq, *next_rq = NULL;
213 	int ret;
214 	unsigned int op, dxfer_len;
215 	void __user *dxferp = NULL;
216 	struct bsg_class_device *bcd = &q->bsg_dev;
217 
218 	/* if the LLD has been removed then the bsg_unregister_queue will
219 	 * eventually be called and the class_dev was freed, so we can no
220 	 * longer use this request_queue. Return no such address.
221 	 */
222 	if (!bcd->class_dev)
223 		return ERR_PTR(-ENXIO);
224 
225 	dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
226 		hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
227 		hdr->din_xfer_len);
228 
229 	ret = bsg_validate_sgv4_hdr(hdr, &op);
230 	if (ret)
231 		return ERR_PTR(ret);
232 
233 	/*
234 	 * map scatter-gather elements separately and string them to request
235 	 */
236 	rq = blk_get_request(q, op, GFP_KERNEL);
237 	if (IS_ERR(rq))
238 		return rq;
239 
240 	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, mode);
241 	if (ret)
242 		goto out;
243 
244 	if (op == REQ_OP_SCSI_OUT && hdr->din_xfer_len) {
245 		if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
246 			ret = -EOPNOTSUPP;
247 			goto out;
248 		}
249 
250 		next_rq = blk_get_request(q, REQ_OP_SCSI_IN, GFP_KERNEL);
251 		if (IS_ERR(next_rq)) {
252 			ret = PTR_ERR(next_rq);
253 			next_rq = NULL;
254 			goto out;
255 		}
256 		rq->next_rq = next_rq;
257 
258 		dxferp = (void __user *)(unsigned long)hdr->din_xferp;
259 		ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
260 				       hdr->din_xfer_len, GFP_KERNEL);
261 		if (ret)
262 			goto out;
263 	}
264 
265 	if (hdr->dout_xfer_len) {
266 		dxfer_len = hdr->dout_xfer_len;
267 		dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
268 	} else if (hdr->din_xfer_len) {
269 		dxfer_len = hdr->din_xfer_len;
270 		dxferp = (void __user *)(unsigned long)hdr->din_xferp;
271 	} else
272 		dxfer_len = 0;
273 
274 	if (dxfer_len) {
275 		ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
276 				      GFP_KERNEL);
277 		if (ret)
278 			goto out;
279 	}
280 
281 	return rq;
282 out:
283 	scsi_req_free_cmd(scsi_req(rq));
284 	blk_put_request(rq);
285 	if (next_rq) {
286 		blk_rq_unmap_user(next_rq->bio);
287 		blk_put_request(next_rq);
288 	}
289 	return ERR_PTR(ret);
290 }
291 
292 /*
293  * async completion call-back from the block layer, when scsi/ide/whatever
294  * calls end_that_request_last() on a request
295  */
296 static void bsg_rq_end_io(struct request *rq, blk_status_t status)
297 {
298 	struct bsg_command *bc = rq->end_io_data;
299 	struct bsg_device *bd = bc->bd;
300 	unsigned long flags;
301 
302 	dprintk("%s: finished rq %p bc %p, bio %p\n",
303 		bd->name, rq, bc, bc->bio);
304 
305 	bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
306 
307 	spin_lock_irqsave(&bd->lock, flags);
308 	list_move_tail(&bc->list, &bd->done_list);
309 	bd->done_cmds++;
310 	spin_unlock_irqrestore(&bd->lock, flags);
311 
312 	wake_up(&bd->wq_done);
313 }
314 
315 /*
316  * do final setup of a 'bc' and submit the matching 'rq' to the block
317  * layer for io
318  */
319 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
320 			    struct bsg_command *bc, struct request *rq)
321 {
322 	int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
323 
324 	/*
325 	 * add bc command to busy queue and submit rq for io
326 	 */
327 	bc->rq = rq;
328 	bc->bio = rq->bio;
329 	if (rq->next_rq)
330 		bc->bidi_bio = rq->next_rq->bio;
331 	bc->hdr.duration = jiffies;
332 	spin_lock_irq(&bd->lock);
333 	list_add_tail(&bc->list, &bd->busy_list);
334 	spin_unlock_irq(&bd->lock);
335 
336 	dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
337 
338 	rq->end_io_data = bc;
339 	blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
340 }
341 
342 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
343 {
344 	struct bsg_command *bc = NULL;
345 
346 	spin_lock_irq(&bd->lock);
347 	if (bd->done_cmds) {
348 		bc = list_first_entry(&bd->done_list, struct bsg_command, list);
349 		list_del(&bc->list);
350 		bd->done_cmds--;
351 	}
352 	spin_unlock_irq(&bd->lock);
353 
354 	return bc;
355 }
356 
357 /*
358  * Get a finished command from the done list
359  */
360 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
361 {
362 	struct bsg_command *bc;
363 	int ret;
364 
365 	do {
366 		bc = bsg_next_done_cmd(bd);
367 		if (bc)
368 			break;
369 
370 		if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
371 			bc = ERR_PTR(-EAGAIN);
372 			break;
373 		}
374 
375 		ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
376 		if (ret) {
377 			bc = ERR_PTR(-ERESTARTSYS);
378 			break;
379 		}
380 	} while (1);
381 
382 	dprintk("%s: returning done %p\n", bd->name, bc);
383 
384 	return bc;
385 }
386 
387 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
388 				    struct bio *bio, struct bio *bidi_bio)
389 {
390 	struct scsi_request *req = scsi_req(rq);
391 	int ret = 0;
392 
393 	dprintk("rq %p bio %p 0x%x\n", rq, bio, req->result);
394 	/*
395 	 * fill in all the output members
396 	 */
397 	hdr->device_status = req->result & 0xff;
398 	hdr->transport_status = host_byte(req->result);
399 	hdr->driver_status = driver_byte(req->result);
400 	hdr->info = 0;
401 	if (hdr->device_status || hdr->transport_status || hdr->driver_status)
402 		hdr->info |= SG_INFO_CHECK;
403 	hdr->response_len = 0;
404 
405 	if (req->sense_len && hdr->response) {
406 		int len = min_t(unsigned int, hdr->max_response_len,
407 					req->sense_len);
408 
409 		ret = copy_to_user((void __user *)(unsigned long)hdr->response,
410 				   req->sense, len);
411 		if (!ret)
412 			hdr->response_len = len;
413 		else
414 			ret = -EFAULT;
415 	}
416 
417 	if (rq->next_rq) {
418 		hdr->dout_resid = req->resid_len;
419 		hdr->din_resid = scsi_req(rq->next_rq)->resid_len;
420 		blk_rq_unmap_user(bidi_bio);
421 		blk_put_request(rq->next_rq);
422 	} else if (rq_data_dir(rq) == READ)
423 		hdr->din_resid = req->resid_len;
424 	else
425 		hdr->dout_resid = req->resid_len;
426 
427 	/*
428 	 * If the request generated a negative error number, return it
429 	 * (providing we aren't already returning an error); if it's
430 	 * just a protocol response (i.e. non negative), that gets
431 	 * processed above.
432 	 */
433 	if (!ret && req->result < 0)
434 		ret = req->result;
435 
436 	blk_rq_unmap_user(bio);
437 	scsi_req_free_cmd(req);
438 	blk_put_request(rq);
439 
440 	return ret;
441 }
442 
443 static bool bsg_complete(struct bsg_device *bd)
444 {
445 	bool ret = false;
446 	bool spin;
447 
448 	do {
449 		spin_lock_irq(&bd->lock);
450 
451 		BUG_ON(bd->done_cmds > bd->queued_cmds);
452 
453 		/*
454 		 * All commands consumed.
455 		 */
456 		if (bd->done_cmds == bd->queued_cmds)
457 			ret = true;
458 
459 		spin = !test_bit(BSG_F_BLOCK, &bd->flags);
460 
461 		spin_unlock_irq(&bd->lock);
462 	} while (!ret && spin);
463 
464 	return ret;
465 }
466 
467 static int bsg_complete_all_commands(struct bsg_device *bd)
468 {
469 	struct bsg_command *bc;
470 	int ret, tret;
471 
472 	dprintk("%s: entered\n", bd->name);
473 
474 	/*
475 	 * wait for all commands to complete
476 	 */
477 	io_wait_event(bd->wq_done, bsg_complete(bd));
478 
479 	/*
480 	 * discard done commands
481 	 */
482 	ret = 0;
483 	do {
484 		spin_lock_irq(&bd->lock);
485 		if (!bd->queued_cmds) {
486 			spin_unlock_irq(&bd->lock);
487 			break;
488 		}
489 		spin_unlock_irq(&bd->lock);
490 
491 		bc = bsg_get_done_cmd(bd);
492 		if (IS_ERR(bc))
493 			break;
494 
495 		tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
496 						bc->bidi_bio);
497 		if (!ret)
498 			ret = tret;
499 
500 		bsg_free_command(bc);
501 	} while (1);
502 
503 	return ret;
504 }
505 
506 static int
507 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
508 	   const struct iovec *iov, ssize_t *bytes_read)
509 {
510 	struct bsg_command *bc;
511 	int nr_commands, ret;
512 
513 	if (count % sizeof(struct sg_io_v4))
514 		return -EINVAL;
515 
516 	ret = 0;
517 	nr_commands = count / sizeof(struct sg_io_v4);
518 	while (nr_commands) {
519 		bc = bsg_get_done_cmd(bd);
520 		if (IS_ERR(bc)) {
521 			ret = PTR_ERR(bc);
522 			break;
523 		}
524 
525 		/*
526 		 * this is the only case where we need to copy data back
527 		 * after completing the request. so do that here,
528 		 * bsg_complete_work() cannot do that for us
529 		 */
530 		ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
531 					       bc->bidi_bio);
532 
533 		if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
534 			ret = -EFAULT;
535 
536 		bsg_free_command(bc);
537 
538 		if (ret)
539 			break;
540 
541 		buf += sizeof(struct sg_io_v4);
542 		*bytes_read += sizeof(struct sg_io_v4);
543 		nr_commands--;
544 	}
545 
546 	return ret;
547 }
548 
549 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
550 {
551 	if (file->f_flags & O_NONBLOCK)
552 		clear_bit(BSG_F_BLOCK, &bd->flags);
553 	else
554 		set_bit(BSG_F_BLOCK, &bd->flags);
555 }
556 
557 /*
558  * Check if the error is a "real" error that we should return.
559  */
560 static inline int err_block_err(int ret)
561 {
562 	if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
563 		return 1;
564 
565 	return 0;
566 }
567 
568 static ssize_t
569 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
570 {
571 	struct bsg_device *bd = file->private_data;
572 	int ret;
573 	ssize_t bytes_read;
574 
575 	dprintk("%s: read %zd bytes\n", bd->name, count);
576 
577 	bsg_set_block(bd, file);
578 
579 	bytes_read = 0;
580 	ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
581 	*ppos = bytes_read;
582 
583 	if (!bytes_read || err_block_err(ret))
584 		bytes_read = ret;
585 
586 	return bytes_read;
587 }
588 
589 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
590 		       size_t count, ssize_t *bytes_written, fmode_t mode)
591 {
592 	struct bsg_command *bc;
593 	struct request *rq;
594 	int ret, nr_commands;
595 
596 	if (count % sizeof(struct sg_io_v4))
597 		return -EINVAL;
598 
599 	nr_commands = count / sizeof(struct sg_io_v4);
600 	rq = NULL;
601 	bc = NULL;
602 	ret = 0;
603 	while (nr_commands) {
604 		struct request_queue *q = bd->queue;
605 
606 		bc = bsg_alloc_command(bd);
607 		if (IS_ERR(bc)) {
608 			ret = PTR_ERR(bc);
609 			bc = NULL;
610 			break;
611 		}
612 
613 		if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
614 			ret = -EFAULT;
615 			break;
616 		}
617 
618 		/*
619 		 * get a request, fill in the blanks, and add to request queue
620 		 */
621 		rq = bsg_map_hdr(bd, &bc->hdr, mode);
622 		if (IS_ERR(rq)) {
623 			ret = PTR_ERR(rq);
624 			rq = NULL;
625 			break;
626 		}
627 
628 		bsg_add_command(bd, q, bc, rq);
629 		bc = NULL;
630 		rq = NULL;
631 		nr_commands--;
632 		buf += sizeof(struct sg_io_v4);
633 		*bytes_written += sizeof(struct sg_io_v4);
634 	}
635 
636 	if (bc)
637 		bsg_free_command(bc);
638 
639 	return ret;
640 }
641 
642 static ssize_t
643 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
644 {
645 	struct bsg_device *bd = file->private_data;
646 	ssize_t bytes_written;
647 	int ret;
648 
649 	dprintk("%s: write %zd bytes\n", bd->name, count);
650 
651 	if (unlikely(uaccess_kernel()))
652 		return -EINVAL;
653 
654 	bsg_set_block(bd, file);
655 
656 	bytes_written = 0;
657 	ret = __bsg_write(bd, buf, count, &bytes_written, file->f_mode);
658 
659 	*ppos = bytes_written;
660 
661 	/*
662 	 * return bytes written on non-fatal errors
663 	 */
664 	if (!bytes_written || err_block_err(ret))
665 		bytes_written = ret;
666 
667 	dprintk("%s: returning %zd\n", bd->name, bytes_written);
668 	return bytes_written;
669 }
670 
671 static struct bsg_device *bsg_alloc_device(void)
672 {
673 	struct bsg_device *bd;
674 
675 	bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
676 	if (unlikely(!bd))
677 		return NULL;
678 
679 	spin_lock_init(&bd->lock);
680 
681 	bd->max_queue = BSG_DEFAULT_CMDS;
682 
683 	INIT_LIST_HEAD(&bd->busy_list);
684 	INIT_LIST_HEAD(&bd->done_list);
685 	INIT_HLIST_NODE(&bd->dev_list);
686 
687 	init_waitqueue_head(&bd->wq_free);
688 	init_waitqueue_head(&bd->wq_done);
689 	return bd;
690 }
691 
692 static void bsg_kref_release_function(struct kref *kref)
693 {
694 	struct bsg_class_device *bcd =
695 		container_of(kref, struct bsg_class_device, ref);
696 	struct device *parent = bcd->parent;
697 
698 	if (bcd->release)
699 		bcd->release(bcd->parent);
700 
701 	put_device(parent);
702 }
703 
704 static int bsg_put_device(struct bsg_device *bd)
705 {
706 	int ret = 0, do_free;
707 	struct request_queue *q = bd->queue;
708 
709 	mutex_lock(&bsg_mutex);
710 
711 	do_free = atomic_dec_and_test(&bd->ref_count);
712 	if (!do_free) {
713 		mutex_unlock(&bsg_mutex);
714 		goto out;
715 	}
716 
717 	hlist_del(&bd->dev_list);
718 	mutex_unlock(&bsg_mutex);
719 
720 	dprintk("%s: tearing down\n", bd->name);
721 
722 	/*
723 	 * close can always block
724 	 */
725 	set_bit(BSG_F_BLOCK, &bd->flags);
726 
727 	/*
728 	 * correct error detection baddies here again. it's the responsibility
729 	 * of the app to properly reap commands before close() if it wants
730 	 * fool-proof error detection
731 	 */
732 	ret = bsg_complete_all_commands(bd);
733 
734 	kfree(bd);
735 out:
736 	kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
737 	if (do_free)
738 		blk_put_queue(q);
739 	return ret;
740 }
741 
742 static struct bsg_device *bsg_add_device(struct inode *inode,
743 					 struct request_queue *rq,
744 					 struct file *file)
745 {
746 	struct bsg_device *bd;
747 #ifdef BSG_DEBUG
748 	unsigned char buf[32];
749 #endif
750 
751 	if (!blk_queue_scsi_passthrough(rq)) {
752 		WARN_ONCE(true, "Attempt to register a non-SCSI queue\n");
753 		return ERR_PTR(-EINVAL);
754 	}
755 
756 	if (!blk_get_queue(rq))
757 		return ERR_PTR(-ENXIO);
758 
759 	bd = bsg_alloc_device();
760 	if (!bd) {
761 		blk_put_queue(rq);
762 		return ERR_PTR(-ENOMEM);
763 	}
764 
765 	bd->queue = rq;
766 
767 	bsg_set_block(bd, file);
768 
769 	atomic_set(&bd->ref_count, 1);
770 	mutex_lock(&bsg_mutex);
771 	hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
772 
773 	strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
774 	dprintk("bound to <%s>, max queue %d\n",
775 		format_dev_t(buf, inode->i_rdev), bd->max_queue);
776 
777 	mutex_unlock(&bsg_mutex);
778 	return bd;
779 }
780 
781 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
782 {
783 	struct bsg_device *bd;
784 
785 	mutex_lock(&bsg_mutex);
786 
787 	hlist_for_each_entry(bd, bsg_dev_idx_hash(minor), dev_list) {
788 		if (bd->queue == q) {
789 			atomic_inc(&bd->ref_count);
790 			goto found;
791 		}
792 	}
793 	bd = NULL;
794 found:
795 	mutex_unlock(&bsg_mutex);
796 	return bd;
797 }
798 
799 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
800 {
801 	struct bsg_device *bd;
802 	struct bsg_class_device *bcd;
803 
804 	/*
805 	 * find the class device
806 	 */
807 	mutex_lock(&bsg_mutex);
808 	bcd = idr_find(&bsg_minor_idr, iminor(inode));
809 	if (bcd)
810 		kref_get(&bcd->ref);
811 	mutex_unlock(&bsg_mutex);
812 
813 	if (!bcd)
814 		return ERR_PTR(-ENODEV);
815 
816 	bd = __bsg_get_device(iminor(inode), bcd->queue);
817 	if (bd)
818 		return bd;
819 
820 	bd = bsg_add_device(inode, bcd->queue, file);
821 	if (IS_ERR(bd))
822 		kref_put(&bcd->ref, bsg_kref_release_function);
823 
824 	return bd;
825 }
826 
827 static int bsg_open(struct inode *inode, struct file *file)
828 {
829 	struct bsg_device *bd;
830 
831 	bd = bsg_get_device(inode, file);
832 
833 	if (IS_ERR(bd))
834 		return PTR_ERR(bd);
835 
836 	file->private_data = bd;
837 	return 0;
838 }
839 
840 static int bsg_release(struct inode *inode, struct file *file)
841 {
842 	struct bsg_device *bd = file->private_data;
843 
844 	file->private_data = NULL;
845 	return bsg_put_device(bd);
846 }
847 
848 static unsigned int bsg_poll(struct file *file, poll_table *wait)
849 {
850 	struct bsg_device *bd = file->private_data;
851 	unsigned int mask = 0;
852 
853 	poll_wait(file, &bd->wq_done, wait);
854 	poll_wait(file, &bd->wq_free, wait);
855 
856 	spin_lock_irq(&bd->lock);
857 	if (!list_empty(&bd->done_list))
858 		mask |= POLLIN | POLLRDNORM;
859 	if (bd->queued_cmds < bd->max_queue)
860 		mask |= POLLOUT;
861 	spin_unlock_irq(&bd->lock);
862 
863 	return mask;
864 }
865 
866 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
867 {
868 	struct bsg_device *bd = file->private_data;
869 	int __user *uarg = (int __user *) arg;
870 	int ret;
871 
872 	switch (cmd) {
873 		/*
874 		 * our own ioctls
875 		 */
876 	case SG_GET_COMMAND_Q:
877 		return put_user(bd->max_queue, uarg);
878 	case SG_SET_COMMAND_Q: {
879 		int queue;
880 
881 		if (get_user(queue, uarg))
882 			return -EFAULT;
883 		if (queue < 1)
884 			return -EINVAL;
885 
886 		spin_lock_irq(&bd->lock);
887 		bd->max_queue = queue;
888 		spin_unlock_irq(&bd->lock);
889 		return 0;
890 	}
891 
892 	/*
893 	 * SCSI/sg ioctls
894 	 */
895 	case SG_GET_VERSION_NUM:
896 	case SCSI_IOCTL_GET_IDLUN:
897 	case SCSI_IOCTL_GET_BUS_NUMBER:
898 	case SG_SET_TIMEOUT:
899 	case SG_GET_TIMEOUT:
900 	case SG_GET_RESERVED_SIZE:
901 	case SG_SET_RESERVED_SIZE:
902 	case SG_EMULATED_HOST:
903 	case SCSI_IOCTL_SEND_COMMAND: {
904 		void __user *uarg = (void __user *) arg;
905 		return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
906 	}
907 	case SG_IO: {
908 		struct request *rq;
909 		struct bio *bio, *bidi_bio = NULL;
910 		struct sg_io_v4 hdr;
911 		int at_head;
912 
913 		if (copy_from_user(&hdr, uarg, sizeof(hdr)))
914 			return -EFAULT;
915 
916 		rq = bsg_map_hdr(bd, &hdr, file->f_mode);
917 		if (IS_ERR(rq))
918 			return PTR_ERR(rq);
919 
920 		bio = rq->bio;
921 		if (rq->next_rq)
922 			bidi_bio = rq->next_rq->bio;
923 
924 		at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
925 		blk_execute_rq(bd->queue, NULL, rq, at_head);
926 		ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
927 
928 		if (copy_to_user(uarg, &hdr, sizeof(hdr)))
929 			return -EFAULT;
930 
931 		return ret;
932 	}
933 	default:
934 		return -ENOTTY;
935 	}
936 }
937 
938 static const struct file_operations bsg_fops = {
939 	.read		=	bsg_read,
940 	.write		=	bsg_write,
941 	.poll		=	bsg_poll,
942 	.open		=	bsg_open,
943 	.release	=	bsg_release,
944 	.unlocked_ioctl	=	bsg_ioctl,
945 	.owner		=	THIS_MODULE,
946 	.llseek		=	default_llseek,
947 };
948 
949 void bsg_unregister_queue(struct request_queue *q)
950 {
951 	struct bsg_class_device *bcd = &q->bsg_dev;
952 
953 	if (!bcd->class_dev)
954 		return;
955 
956 	mutex_lock(&bsg_mutex);
957 	idr_remove(&bsg_minor_idr, bcd->minor);
958 	if (q->kobj.sd)
959 		sysfs_remove_link(&q->kobj, "bsg");
960 	device_unregister(bcd->class_dev);
961 	bcd->class_dev = NULL;
962 	kref_put(&bcd->ref, bsg_kref_release_function);
963 	mutex_unlock(&bsg_mutex);
964 }
965 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
966 
967 int bsg_register_queue(struct request_queue *q, struct device *parent,
968 		       const char *name, void (*release)(struct device *))
969 {
970 	struct bsg_class_device *bcd;
971 	dev_t dev;
972 	int ret;
973 	struct device *class_dev = NULL;
974 	const char *devname;
975 
976 	if (name)
977 		devname = name;
978 	else
979 		devname = dev_name(parent);
980 
981 	/*
982 	 * we need a proper transport to send commands, not a stacked device
983 	 */
984 	if (!queue_is_rq_based(q))
985 		return 0;
986 
987 	bcd = &q->bsg_dev;
988 	memset(bcd, 0, sizeof(*bcd));
989 
990 	mutex_lock(&bsg_mutex);
991 
992 	ret = idr_alloc(&bsg_minor_idr, bcd, 0, BSG_MAX_DEVS, GFP_KERNEL);
993 	if (ret < 0) {
994 		if (ret == -ENOSPC) {
995 			printk(KERN_ERR "bsg: too many bsg devices\n");
996 			ret = -EINVAL;
997 		}
998 		goto unlock;
999 	}
1000 
1001 	bcd->minor = ret;
1002 	bcd->queue = q;
1003 	bcd->parent = get_device(parent);
1004 	bcd->release = release;
1005 	kref_init(&bcd->ref);
1006 	dev = MKDEV(bsg_major, bcd->minor);
1007 	class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1008 	if (IS_ERR(class_dev)) {
1009 		ret = PTR_ERR(class_dev);
1010 		goto put_dev;
1011 	}
1012 	bcd->class_dev = class_dev;
1013 
1014 	if (q->kobj.sd) {
1015 		ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1016 		if (ret)
1017 			goto unregister_class_dev;
1018 	}
1019 
1020 	mutex_unlock(&bsg_mutex);
1021 	return 0;
1022 
1023 unregister_class_dev:
1024 	device_unregister(class_dev);
1025 put_dev:
1026 	put_device(parent);
1027 	idr_remove(&bsg_minor_idr, bcd->minor);
1028 unlock:
1029 	mutex_unlock(&bsg_mutex);
1030 	return ret;
1031 }
1032 EXPORT_SYMBOL_GPL(bsg_register_queue);
1033 
1034 static struct cdev bsg_cdev;
1035 
1036 static char *bsg_devnode(struct device *dev, umode_t *mode)
1037 {
1038 	return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1039 }
1040 
1041 static int __init bsg_init(void)
1042 {
1043 	int ret, i;
1044 	dev_t devid;
1045 
1046 	bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1047 				sizeof(struct bsg_command), 0, 0, NULL);
1048 	if (!bsg_cmd_cachep) {
1049 		printk(KERN_ERR "bsg: failed creating slab cache\n");
1050 		return -ENOMEM;
1051 	}
1052 
1053 	for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1054 		INIT_HLIST_HEAD(&bsg_device_list[i]);
1055 
1056 	bsg_class = class_create(THIS_MODULE, "bsg");
1057 	if (IS_ERR(bsg_class)) {
1058 		ret = PTR_ERR(bsg_class);
1059 		goto destroy_kmemcache;
1060 	}
1061 	bsg_class->devnode = bsg_devnode;
1062 
1063 	ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1064 	if (ret)
1065 		goto destroy_bsg_class;
1066 
1067 	bsg_major = MAJOR(devid);
1068 
1069 	cdev_init(&bsg_cdev, &bsg_fops);
1070 	ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1071 	if (ret)
1072 		goto unregister_chrdev;
1073 
1074 	printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1075 	       " loaded (major %d)\n", bsg_major);
1076 	return 0;
1077 unregister_chrdev:
1078 	unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1079 destroy_bsg_class:
1080 	class_destroy(bsg_class);
1081 destroy_kmemcache:
1082 	kmem_cache_destroy(bsg_cmd_cachep);
1083 	return ret;
1084 }
1085 
1086 MODULE_AUTHOR("Jens Axboe");
1087 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1088 MODULE_LICENSE("GPL");
1089 
1090 device_initcall(bsg_init);
1091