Loading...
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
35struct 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
51enum {
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
66static DEFINE_MUTEX(bsg_mutex);
67static DEFINE_IDR(bsg_minor_idr);
68
69#define BSG_LIST_ARRAY_SIZE 8
70static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
71
72static struct class *bsg_class;
73static int bsg_major;
74
75static struct kmem_cache *bsg_cmd_cachep;
76
77/*
78 * our internal command type
79 */
80struct 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
91static 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
105static 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;
129out:
130 spin_unlock_irq(&bd->lock);
131 return bc;
132}
133
134static inline struct hlist_head *bsg_dev_idx_hash(int index)
135{
136 return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
137}
138
139static 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;
170unlock:
171 spin_unlock_irq(&bd->lock);
172 return ret;
173}
174
175static 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 __user *)(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 */
215static int
216bsg_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 */
244static struct request *
245bsg_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 __user *dxferp = NULL;
253 struct bsg_class_device *bcd = &q->bsg_dev;
254
255 /* if the LLD has been removed then the bsg_unregister_queue will
256 * eventually be called and the class_dev was freed, so we can no
257 * longer use this request_queue. Return no such address.
258 */
259 if (!bcd->class_dev)
260 return ERR_PTR(-ENXIO);
261
262 dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
263 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
264 hdr->din_xfer_len);
265
266 ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
267 if (ret)
268 return ERR_PTR(ret);
269
270 /*
271 * map scatter-gather elements separately and string them to request
272 */
273 rq = blk_get_request(q, rw, GFP_KERNEL);
274 if (!rq)
275 return ERR_PTR(-ENOMEM);
276 ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
277 if (ret)
278 goto out;
279
280 if (rw == WRITE && hdr->din_xfer_len) {
281 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
282 ret = -EOPNOTSUPP;
283 goto out;
284 }
285
286 next_rq = blk_get_request(q, READ, GFP_KERNEL);
287 if (!next_rq) {
288 ret = -ENOMEM;
289 goto out;
290 }
291 rq->next_rq = next_rq;
292 next_rq->cmd_type = rq->cmd_type;
293
294 dxferp = (void __user *)(unsigned long)hdr->din_xferp;
295 ret = blk_rq_map_user(q, next_rq, NULL, dxferp,
296 hdr->din_xfer_len, GFP_KERNEL);
297 if (ret)
298 goto out;
299 }
300
301 if (hdr->dout_xfer_len) {
302 dxfer_len = hdr->dout_xfer_len;
303 dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
304 } else if (hdr->din_xfer_len) {
305 dxfer_len = hdr->din_xfer_len;
306 dxferp = (void __user *)(unsigned long)hdr->din_xferp;
307 } else
308 dxfer_len = 0;
309
310 if (dxfer_len) {
311 ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
312 GFP_KERNEL);
313 if (ret)
314 goto out;
315 }
316
317 rq->sense = sense;
318 rq->sense_len = 0;
319
320 return rq;
321out:
322 if (rq->cmd != rq->__cmd)
323 kfree(rq->cmd);
324 blk_put_request(rq);
325 if (next_rq) {
326 blk_rq_unmap_user(next_rq->bio);
327 blk_put_request(next_rq);
328 }
329 return ERR_PTR(ret);
330}
331
332/*
333 * async completion call-back from the block layer, when scsi/ide/whatever
334 * calls end_that_request_last() on a request
335 */
336static void bsg_rq_end_io(struct request *rq, int uptodate)
337{
338 struct bsg_command *bc = rq->end_io_data;
339 struct bsg_device *bd = bc->bd;
340 unsigned long flags;
341
342 dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
343 bd->name, rq, bc, bc->bio, uptodate);
344
345 bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
346
347 spin_lock_irqsave(&bd->lock, flags);
348 list_move_tail(&bc->list, &bd->done_list);
349 bd->done_cmds++;
350 spin_unlock_irqrestore(&bd->lock, flags);
351
352 wake_up(&bd->wq_done);
353}
354
355/*
356 * do final setup of a 'bc' and submit the matching 'rq' to the block
357 * layer for io
358 */
359static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
360 struct bsg_command *bc, struct request *rq)
361{
362 int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
363
364 /*
365 * add bc command to busy queue and submit rq for io
366 */
367 bc->rq = rq;
368 bc->bio = rq->bio;
369 if (rq->next_rq)
370 bc->bidi_bio = rq->next_rq->bio;
371 bc->hdr.duration = jiffies;
372 spin_lock_irq(&bd->lock);
373 list_add_tail(&bc->list, &bd->busy_list);
374 spin_unlock_irq(&bd->lock);
375
376 dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
377
378 rq->end_io_data = bc;
379 blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
380}
381
382static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
383{
384 struct bsg_command *bc = NULL;
385
386 spin_lock_irq(&bd->lock);
387 if (bd->done_cmds) {
388 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
389 list_del(&bc->list);
390 bd->done_cmds--;
391 }
392 spin_unlock_irq(&bd->lock);
393
394 return bc;
395}
396
397/*
398 * Get a finished command from the done list
399 */
400static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
401{
402 struct bsg_command *bc;
403 int ret;
404
405 do {
406 bc = bsg_next_done_cmd(bd);
407 if (bc)
408 break;
409
410 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
411 bc = ERR_PTR(-EAGAIN);
412 break;
413 }
414
415 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
416 if (ret) {
417 bc = ERR_PTR(-ERESTARTSYS);
418 break;
419 }
420 } while (1);
421
422 dprintk("%s: returning done %p\n", bd->name, bc);
423
424 return bc;
425}
426
427static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
428 struct bio *bio, struct bio *bidi_bio)
429{
430 int ret = 0;
431
432 dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
433 /*
434 * fill in all the output members
435 */
436 hdr->device_status = rq->errors & 0xff;
437 hdr->transport_status = host_byte(rq->errors);
438 hdr->driver_status = driver_byte(rq->errors);
439 hdr->info = 0;
440 if (hdr->device_status || hdr->transport_status || hdr->driver_status)
441 hdr->info |= SG_INFO_CHECK;
442 hdr->response_len = 0;
443
444 if (rq->sense_len && hdr->response) {
445 int len = min_t(unsigned int, hdr->max_response_len,
446 rq->sense_len);
447
448 ret = copy_to_user((void __user *)(unsigned long)hdr->response,
449 rq->sense, len);
450 if (!ret)
451 hdr->response_len = len;
452 else
453 ret = -EFAULT;
454 }
455
456 if (rq->next_rq) {
457 hdr->dout_resid = rq->resid_len;
458 hdr->din_resid = rq->next_rq->resid_len;
459 blk_rq_unmap_user(bidi_bio);
460 blk_put_request(rq->next_rq);
461 } else if (rq_data_dir(rq) == READ)
462 hdr->din_resid = rq->resid_len;
463 else
464 hdr->dout_resid = rq->resid_len;
465
466 /*
467 * If the request generated a negative error number, return it
468 * (providing we aren't already returning an error); if it's
469 * just a protocol response (i.e. non negative), that gets
470 * processed above.
471 */
472 if (!ret && rq->errors < 0)
473 ret = rq->errors;
474
475 blk_rq_unmap_user(bio);
476 if (rq->cmd != rq->__cmd)
477 kfree(rq->cmd);
478 blk_put_request(rq);
479
480 return ret;
481}
482
483static int bsg_complete_all_commands(struct bsg_device *bd)
484{
485 struct bsg_command *bc;
486 int ret, tret;
487
488 dprintk("%s: entered\n", bd->name);
489
490 /*
491 * wait for all commands to complete
492 */
493 ret = 0;
494 do {
495 ret = bsg_io_schedule(bd);
496 /*
497 * look for -ENODATA specifically -- we'll sometimes get
498 * -ERESTARTSYS when we've taken a signal, but we can't
499 * return until we're done freeing the queue, so ignore
500 * it. The signal will get handled when we're done freeing
501 * the bsg_device.
502 */
503 } while (ret != -ENODATA);
504
505 /*
506 * discard done commands
507 */
508 ret = 0;
509 do {
510 spin_lock_irq(&bd->lock);
511 if (!bd->queued_cmds) {
512 spin_unlock_irq(&bd->lock);
513 break;
514 }
515 spin_unlock_irq(&bd->lock);
516
517 bc = bsg_get_done_cmd(bd);
518 if (IS_ERR(bc))
519 break;
520
521 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
522 bc->bidi_bio);
523 if (!ret)
524 ret = tret;
525
526 bsg_free_command(bc);
527 } while (1);
528
529 return ret;
530}
531
532static int
533__bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
534 const struct iovec *iov, ssize_t *bytes_read)
535{
536 struct bsg_command *bc;
537 int nr_commands, ret;
538
539 if (count % sizeof(struct sg_io_v4))
540 return -EINVAL;
541
542 ret = 0;
543 nr_commands = count / sizeof(struct sg_io_v4);
544 while (nr_commands) {
545 bc = bsg_get_done_cmd(bd);
546 if (IS_ERR(bc)) {
547 ret = PTR_ERR(bc);
548 break;
549 }
550
551 /*
552 * this is the only case where we need to copy data back
553 * after completing the request. so do that here,
554 * bsg_complete_work() cannot do that for us
555 */
556 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
557 bc->bidi_bio);
558
559 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
560 ret = -EFAULT;
561
562 bsg_free_command(bc);
563
564 if (ret)
565 break;
566
567 buf += sizeof(struct sg_io_v4);
568 *bytes_read += sizeof(struct sg_io_v4);
569 nr_commands--;
570 }
571
572 return ret;
573}
574
575static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
576{
577 if (file->f_flags & O_NONBLOCK)
578 clear_bit(BSG_F_BLOCK, &bd->flags);
579 else
580 set_bit(BSG_F_BLOCK, &bd->flags);
581}
582
583/*
584 * Check if the error is a "real" error that we should return.
585 */
586static inline int err_block_err(int ret)
587{
588 if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
589 return 1;
590
591 return 0;
592}
593
594static ssize_t
595bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
596{
597 struct bsg_device *bd = file->private_data;
598 int ret;
599 ssize_t bytes_read;
600
601 dprintk("%s: read %Zd bytes\n", bd->name, count);
602
603 bsg_set_block(bd, file);
604
605 bytes_read = 0;
606 ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
607 *ppos = bytes_read;
608
609 if (!bytes_read || err_block_err(ret))
610 bytes_read = ret;
611
612 return bytes_read;
613}
614
615static int __bsg_write(struct bsg_device *bd, const char __user *buf,
616 size_t count, ssize_t *bytes_written,
617 fmode_t has_write_perm)
618{
619 struct bsg_command *bc;
620 struct request *rq;
621 int ret, nr_commands;
622
623 if (count % sizeof(struct sg_io_v4))
624 return -EINVAL;
625
626 nr_commands = count / sizeof(struct sg_io_v4);
627 rq = NULL;
628 bc = NULL;
629 ret = 0;
630 while (nr_commands) {
631 struct request_queue *q = bd->queue;
632
633 bc = bsg_alloc_command(bd);
634 if (IS_ERR(bc)) {
635 ret = PTR_ERR(bc);
636 bc = NULL;
637 break;
638 }
639
640 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
641 ret = -EFAULT;
642 break;
643 }
644
645 /*
646 * get a request, fill in the blanks, and add to request queue
647 */
648 rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
649 if (IS_ERR(rq)) {
650 ret = PTR_ERR(rq);
651 rq = NULL;
652 break;
653 }
654
655 bsg_add_command(bd, q, bc, rq);
656 bc = NULL;
657 rq = NULL;
658 nr_commands--;
659 buf += sizeof(struct sg_io_v4);
660 *bytes_written += sizeof(struct sg_io_v4);
661 }
662
663 if (bc)
664 bsg_free_command(bc);
665
666 return ret;
667}
668
669static ssize_t
670bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
671{
672 struct bsg_device *bd = file->private_data;
673 ssize_t bytes_written;
674 int ret;
675
676 dprintk("%s: write %Zd bytes\n", bd->name, count);
677
678 bsg_set_block(bd, file);
679
680 bytes_written = 0;
681 ret = __bsg_write(bd, buf, count, &bytes_written,
682 file->f_mode & FMODE_WRITE);
683
684 *ppos = bytes_written;
685
686 /*
687 * return bytes written on non-fatal errors
688 */
689 if (!bytes_written || err_block_err(ret))
690 bytes_written = ret;
691
692 dprintk("%s: returning %Zd\n", bd->name, bytes_written);
693 return bytes_written;
694}
695
696static struct bsg_device *bsg_alloc_device(void)
697{
698 struct bsg_device *bd;
699
700 bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
701 if (unlikely(!bd))
702 return NULL;
703
704 spin_lock_init(&bd->lock);
705
706 bd->max_queue = BSG_DEFAULT_CMDS;
707
708 INIT_LIST_HEAD(&bd->busy_list);
709 INIT_LIST_HEAD(&bd->done_list);
710 INIT_HLIST_NODE(&bd->dev_list);
711
712 init_waitqueue_head(&bd->wq_free);
713 init_waitqueue_head(&bd->wq_done);
714 return bd;
715}
716
717static void bsg_kref_release_function(struct kref *kref)
718{
719 struct bsg_class_device *bcd =
720 container_of(kref, struct bsg_class_device, ref);
721 struct device *parent = bcd->parent;
722
723 if (bcd->release)
724 bcd->release(bcd->parent);
725
726 put_device(parent);
727}
728
729static int bsg_put_device(struct bsg_device *bd)
730{
731 int ret = 0, do_free;
732 struct request_queue *q = bd->queue;
733
734 mutex_lock(&bsg_mutex);
735
736 do_free = atomic_dec_and_test(&bd->ref_count);
737 if (!do_free) {
738 mutex_unlock(&bsg_mutex);
739 goto out;
740 }
741
742 hlist_del(&bd->dev_list);
743 mutex_unlock(&bsg_mutex);
744
745 dprintk("%s: tearing down\n", bd->name);
746
747 /*
748 * close can always block
749 */
750 set_bit(BSG_F_BLOCK, &bd->flags);
751
752 /*
753 * correct error detection baddies here again. it's the responsibility
754 * of the app to properly reap commands before close() if it wants
755 * fool-proof error detection
756 */
757 ret = bsg_complete_all_commands(bd);
758
759 kfree(bd);
760out:
761 kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
762 if (do_free)
763 blk_put_queue(q);
764 return ret;
765}
766
767static struct bsg_device *bsg_add_device(struct inode *inode,
768 struct request_queue *rq,
769 struct file *file)
770{
771 struct bsg_device *bd;
772 int ret;
773#ifdef BSG_DEBUG
774 unsigned char buf[32];
775#endif
776 ret = blk_get_queue(rq);
777 if (ret)
778 return ERR_PTR(-ENXIO);
779
780 bd = bsg_alloc_device();
781 if (!bd) {
782 blk_put_queue(rq);
783 return ERR_PTR(-ENOMEM);
784 }
785
786 bd->queue = rq;
787
788 bsg_set_block(bd, file);
789
790 atomic_set(&bd->ref_count, 1);
791 mutex_lock(&bsg_mutex);
792 hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
793
794 strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
795 dprintk("bound to <%s>, max queue %d\n",
796 format_dev_t(buf, inode->i_rdev), bd->max_queue);
797
798 mutex_unlock(&bsg_mutex);
799 return bd;
800}
801
802static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
803{
804 struct bsg_device *bd;
805 struct hlist_node *entry;
806
807 mutex_lock(&bsg_mutex);
808
809 hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
810 if (bd->queue == q) {
811 atomic_inc(&bd->ref_count);
812 goto found;
813 }
814 }
815 bd = NULL;
816found:
817 mutex_unlock(&bsg_mutex);
818 return bd;
819}
820
821static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
822{
823 struct bsg_device *bd;
824 struct bsg_class_device *bcd;
825
826 /*
827 * find the class device
828 */
829 mutex_lock(&bsg_mutex);
830 bcd = idr_find(&bsg_minor_idr, iminor(inode));
831 if (bcd)
832 kref_get(&bcd->ref);
833 mutex_unlock(&bsg_mutex);
834
835 if (!bcd)
836 return ERR_PTR(-ENODEV);
837
838 bd = __bsg_get_device(iminor(inode), bcd->queue);
839 if (bd)
840 return bd;
841
842 bd = bsg_add_device(inode, bcd->queue, file);
843 if (IS_ERR(bd))
844 kref_put(&bcd->ref, bsg_kref_release_function);
845
846 return bd;
847}
848
849static int bsg_open(struct inode *inode, struct file *file)
850{
851 struct bsg_device *bd;
852
853 bd = bsg_get_device(inode, file);
854
855 if (IS_ERR(bd))
856 return PTR_ERR(bd);
857
858 file->private_data = bd;
859 return 0;
860}
861
862static int bsg_release(struct inode *inode, struct file *file)
863{
864 struct bsg_device *bd = file->private_data;
865
866 file->private_data = NULL;
867 return bsg_put_device(bd);
868}
869
870static unsigned int bsg_poll(struct file *file, poll_table *wait)
871{
872 struct bsg_device *bd = file->private_data;
873 unsigned int mask = 0;
874
875 poll_wait(file, &bd->wq_done, wait);
876 poll_wait(file, &bd->wq_free, wait);
877
878 spin_lock_irq(&bd->lock);
879 if (!list_empty(&bd->done_list))
880 mask |= POLLIN | POLLRDNORM;
881 if (bd->queued_cmds < bd->max_queue)
882 mask |= POLLOUT;
883 spin_unlock_irq(&bd->lock);
884
885 return mask;
886}
887
888static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
889{
890 struct bsg_device *bd = file->private_data;
891 int __user *uarg = (int __user *) arg;
892 int ret;
893
894 switch (cmd) {
895 /*
896 * our own ioctls
897 */
898 case SG_GET_COMMAND_Q:
899 return put_user(bd->max_queue, uarg);
900 case SG_SET_COMMAND_Q: {
901 int queue;
902
903 if (get_user(queue, uarg))
904 return -EFAULT;
905 if (queue < 1)
906 return -EINVAL;
907
908 spin_lock_irq(&bd->lock);
909 bd->max_queue = queue;
910 spin_unlock_irq(&bd->lock);
911 return 0;
912 }
913
914 /*
915 * SCSI/sg ioctls
916 */
917 case SG_GET_VERSION_NUM:
918 case SCSI_IOCTL_GET_IDLUN:
919 case SCSI_IOCTL_GET_BUS_NUMBER:
920 case SG_SET_TIMEOUT:
921 case SG_GET_TIMEOUT:
922 case SG_GET_RESERVED_SIZE:
923 case SG_SET_RESERVED_SIZE:
924 case SG_EMULATED_HOST:
925 case SCSI_IOCTL_SEND_COMMAND: {
926 void __user *uarg = (void __user *) arg;
927 return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
928 }
929 case SG_IO: {
930 struct request *rq;
931 struct bio *bio, *bidi_bio = NULL;
932 struct sg_io_v4 hdr;
933 int at_head;
934 u8 sense[SCSI_SENSE_BUFFERSIZE];
935
936 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
937 return -EFAULT;
938
939 rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
940 if (IS_ERR(rq))
941 return PTR_ERR(rq);
942
943 bio = rq->bio;
944 if (rq->next_rq)
945 bidi_bio = rq->next_rq->bio;
946
947 at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
948 blk_execute_rq(bd->queue, NULL, rq, at_head);
949 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
950
951 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
952 return -EFAULT;
953
954 return ret;
955 }
956 /*
957 * block device ioctls
958 */
959 default:
960#if 0
961 return ioctl_by_bdev(bd->bdev, cmd, arg);
962#else
963 return -ENOTTY;
964#endif
965 }
966}
967
968static const struct file_operations bsg_fops = {
969 .read = bsg_read,
970 .write = bsg_write,
971 .poll = bsg_poll,
972 .open = bsg_open,
973 .release = bsg_release,
974 .unlocked_ioctl = bsg_ioctl,
975 .owner = THIS_MODULE,
976 .llseek = default_llseek,
977};
978
979void bsg_unregister_queue(struct request_queue *q)
980{
981 struct bsg_class_device *bcd = &q->bsg_dev;
982
983 if (!bcd->class_dev)
984 return;
985
986 mutex_lock(&bsg_mutex);
987 idr_remove(&bsg_minor_idr, bcd->minor);
988 sysfs_remove_link(&q->kobj, "bsg");
989 device_unregister(bcd->class_dev);
990 bcd->class_dev = NULL;
991 kref_put(&bcd->ref, bsg_kref_release_function);
992 mutex_unlock(&bsg_mutex);
993}
994EXPORT_SYMBOL_GPL(bsg_unregister_queue);
995
996int bsg_register_queue(struct request_queue *q, struct device *parent,
997 const char *name, void (*release)(struct device *))
998{
999 struct bsg_class_device *bcd;
1000 dev_t dev;
1001 int ret, minor;
1002 struct device *class_dev = NULL;
1003 const char *devname;
1004
1005 if (name)
1006 devname = name;
1007 else
1008 devname = dev_name(parent);
1009
1010 /*
1011 * we need a proper transport to send commands, not a stacked device
1012 */
1013 if (!q->request_fn)
1014 return 0;
1015
1016 bcd = &q->bsg_dev;
1017 memset(bcd, 0, sizeof(*bcd));
1018
1019 mutex_lock(&bsg_mutex);
1020
1021 ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1022 if (!ret) {
1023 ret = -ENOMEM;
1024 goto unlock;
1025 }
1026
1027 ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1028 if (ret < 0)
1029 goto unlock;
1030
1031 if (minor >= BSG_MAX_DEVS) {
1032 printk(KERN_ERR "bsg: too many bsg devices\n");
1033 ret = -EINVAL;
1034 goto remove_idr;
1035 }
1036
1037 bcd->minor = minor;
1038 bcd->queue = q;
1039 bcd->parent = get_device(parent);
1040 bcd->release = release;
1041 kref_init(&bcd->ref);
1042 dev = MKDEV(bsg_major, bcd->minor);
1043 class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1044 if (IS_ERR(class_dev)) {
1045 ret = PTR_ERR(class_dev);
1046 goto put_dev;
1047 }
1048 bcd->class_dev = class_dev;
1049
1050 if (q->kobj.sd) {
1051 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1052 if (ret)
1053 goto unregister_class_dev;
1054 }
1055
1056 mutex_unlock(&bsg_mutex);
1057 return 0;
1058
1059unregister_class_dev:
1060 device_unregister(class_dev);
1061put_dev:
1062 put_device(parent);
1063remove_idr:
1064 idr_remove(&bsg_minor_idr, minor);
1065unlock:
1066 mutex_unlock(&bsg_mutex);
1067 return ret;
1068}
1069EXPORT_SYMBOL_GPL(bsg_register_queue);
1070
1071static struct cdev bsg_cdev;
1072
1073static char *bsg_devnode(struct device *dev, mode_t *mode)
1074{
1075 return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1076}
1077
1078static int __init bsg_init(void)
1079{
1080 int ret, i;
1081 dev_t devid;
1082
1083 bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1084 sizeof(struct bsg_command), 0, 0, NULL);
1085 if (!bsg_cmd_cachep) {
1086 printk(KERN_ERR "bsg: failed creating slab cache\n");
1087 return -ENOMEM;
1088 }
1089
1090 for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1091 INIT_HLIST_HEAD(&bsg_device_list[i]);
1092
1093 bsg_class = class_create(THIS_MODULE, "bsg");
1094 if (IS_ERR(bsg_class)) {
1095 ret = PTR_ERR(bsg_class);
1096 goto destroy_kmemcache;
1097 }
1098 bsg_class->devnode = bsg_devnode;
1099
1100 ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1101 if (ret)
1102 goto destroy_bsg_class;
1103
1104 bsg_major = MAJOR(devid);
1105
1106 cdev_init(&bsg_cdev, &bsg_fops);
1107 ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1108 if (ret)
1109 goto unregister_chrdev;
1110
1111 printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1112 " loaded (major %d)\n", bsg_major);
1113 return 0;
1114unregister_chrdev:
1115 unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1116destroy_bsg_class:
1117 class_destroy(bsg_class);
1118destroy_kmemcache:
1119 kmem_cache_destroy(bsg_cmd_cachep);
1120 return ret;
1121}
1122
1123MODULE_AUTHOR("Jens Axboe");
1124MODULE_DESCRIPTION(BSG_DESCRIPTION);
1125MODULE_LICENSE("GPL");
1126
1127device_initcall(bsg_init);
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
35struct 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
51enum {
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
66static DEFINE_MUTEX(bsg_mutex);
67static DEFINE_IDR(bsg_minor_idr);
68
69#define BSG_LIST_ARRAY_SIZE 8
70static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
71
72static struct class *bsg_class;
73static int bsg_major;
74
75static struct kmem_cache *bsg_cmd_cachep;
76
77/*
78 * our internal command type
79 */
80struct 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
91static 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
105static 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;
129out:
130 spin_unlock_irq(&bd->lock);
131 return bc;
132}
133
134static inline struct hlist_head *bsg_dev_idx_hash(int index)
135{
136 return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
137}
138
139static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
140 struct sg_io_v4 *hdr, struct bsg_device *bd,
141 fmode_t has_write_perm)
142{
143 if (hdr->request_len > BLK_MAX_CDB) {
144 rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
145 if (!rq->cmd)
146 return -ENOMEM;
147 }
148
149 if (copy_from_user(rq->cmd, (void __user *)(unsigned long)hdr->request,
150 hdr->request_len))
151 return -EFAULT;
152
153 if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
154 if (blk_verify_command(rq->cmd, has_write_perm))
155 return -EPERM;
156 } else if (!capable(CAP_SYS_RAWIO))
157 return -EPERM;
158
159 /*
160 * fill in request structure
161 */
162 rq->cmd_len = hdr->request_len;
163
164 rq->timeout = msecs_to_jiffies(hdr->timeout);
165 if (!rq->timeout)
166 rq->timeout = q->sg_timeout;
167 if (!rq->timeout)
168 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
169 if (rq->timeout < BLK_MIN_SG_TIMEOUT)
170 rq->timeout = BLK_MIN_SG_TIMEOUT;
171
172 return 0;
173}
174
175/*
176 * Check if sg_io_v4 from user is allowed and valid
177 */
178static int
179bsg_validate_sgv4_hdr(struct sg_io_v4 *hdr, int *rw)
180{
181 int ret = 0;
182
183 if (hdr->guard != 'Q')
184 return -EINVAL;
185
186 switch (hdr->protocol) {
187 case BSG_PROTOCOL_SCSI:
188 switch (hdr->subprotocol) {
189 case BSG_SUB_PROTOCOL_SCSI_CMD:
190 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
191 break;
192 default:
193 ret = -EINVAL;
194 }
195 break;
196 default:
197 ret = -EINVAL;
198 }
199
200 *rw = hdr->dout_xfer_len ? WRITE : READ;
201 return ret;
202}
203
204/*
205 * map sg_io_v4 to a request.
206 */
207static struct request *
208bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
209 u8 *sense)
210{
211 struct request_queue *q = bd->queue;
212 struct request *rq, *next_rq = NULL;
213 int ret, rw;
214 unsigned int 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, &rw);
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, rw, GFP_KERNEL);
237 if (IS_ERR(rq))
238 return rq;
239 blk_rq_set_block_pc(rq);
240
241 ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
242 if (ret)
243 goto out;
244
245 if (rw == WRITE && hdr->din_xfer_len) {
246 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
247 ret = -EOPNOTSUPP;
248 goto out;
249 }
250
251 next_rq = blk_get_request(q, READ, GFP_KERNEL);
252 if (IS_ERR(next_rq)) {
253 ret = PTR_ERR(next_rq);
254 next_rq = NULL;
255 goto out;
256 }
257 rq->next_rq = next_rq;
258 next_rq->cmd_type = rq->cmd_type;
259
260 dxferp = (void __user *)(unsigned long)hdr->din_xferp;
261 ret = blk_rq_map_user(q, next_rq, NULL, dxferp,
262 hdr->din_xfer_len, GFP_KERNEL);
263 if (ret)
264 goto out;
265 }
266
267 if (hdr->dout_xfer_len) {
268 dxfer_len = hdr->dout_xfer_len;
269 dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
270 } else if (hdr->din_xfer_len) {
271 dxfer_len = hdr->din_xfer_len;
272 dxferp = (void __user *)(unsigned long)hdr->din_xferp;
273 } else
274 dxfer_len = 0;
275
276 if (dxfer_len) {
277 ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
278 GFP_KERNEL);
279 if (ret)
280 goto out;
281 }
282
283 rq->sense = sense;
284 rq->sense_len = 0;
285
286 return rq;
287out:
288 if (rq->cmd != rq->__cmd)
289 kfree(rq->cmd);
290 blk_put_request(rq);
291 if (next_rq) {
292 blk_rq_unmap_user(next_rq->bio);
293 blk_put_request(next_rq);
294 }
295 return ERR_PTR(ret);
296}
297
298/*
299 * async completion call-back from the block layer, when scsi/ide/whatever
300 * calls end_that_request_last() on a request
301 */
302static void bsg_rq_end_io(struct request *rq, int uptodate)
303{
304 struct bsg_command *bc = rq->end_io_data;
305 struct bsg_device *bd = bc->bd;
306 unsigned long flags;
307
308 dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
309 bd->name, rq, bc, bc->bio, uptodate);
310
311 bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
312
313 spin_lock_irqsave(&bd->lock, flags);
314 list_move_tail(&bc->list, &bd->done_list);
315 bd->done_cmds++;
316 spin_unlock_irqrestore(&bd->lock, flags);
317
318 wake_up(&bd->wq_done);
319}
320
321/*
322 * do final setup of a 'bc' and submit the matching 'rq' to the block
323 * layer for io
324 */
325static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
326 struct bsg_command *bc, struct request *rq)
327{
328 int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
329
330 /*
331 * add bc command to busy queue and submit rq for io
332 */
333 bc->rq = rq;
334 bc->bio = rq->bio;
335 if (rq->next_rq)
336 bc->bidi_bio = rq->next_rq->bio;
337 bc->hdr.duration = jiffies;
338 spin_lock_irq(&bd->lock);
339 list_add_tail(&bc->list, &bd->busy_list);
340 spin_unlock_irq(&bd->lock);
341
342 dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
343
344 rq->end_io_data = bc;
345 blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
346}
347
348static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
349{
350 struct bsg_command *bc = NULL;
351
352 spin_lock_irq(&bd->lock);
353 if (bd->done_cmds) {
354 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
355 list_del(&bc->list);
356 bd->done_cmds--;
357 }
358 spin_unlock_irq(&bd->lock);
359
360 return bc;
361}
362
363/*
364 * Get a finished command from the done list
365 */
366static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
367{
368 struct bsg_command *bc;
369 int ret;
370
371 do {
372 bc = bsg_next_done_cmd(bd);
373 if (bc)
374 break;
375
376 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
377 bc = ERR_PTR(-EAGAIN);
378 break;
379 }
380
381 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
382 if (ret) {
383 bc = ERR_PTR(-ERESTARTSYS);
384 break;
385 }
386 } while (1);
387
388 dprintk("%s: returning done %p\n", bd->name, bc);
389
390 return bc;
391}
392
393static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
394 struct bio *bio, struct bio *bidi_bio)
395{
396 int ret = 0;
397
398 dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
399 /*
400 * fill in all the output members
401 */
402 hdr->device_status = rq->errors & 0xff;
403 hdr->transport_status = host_byte(rq->errors);
404 hdr->driver_status = driver_byte(rq->errors);
405 hdr->info = 0;
406 if (hdr->device_status || hdr->transport_status || hdr->driver_status)
407 hdr->info |= SG_INFO_CHECK;
408 hdr->response_len = 0;
409
410 if (rq->sense_len && hdr->response) {
411 int len = min_t(unsigned int, hdr->max_response_len,
412 rq->sense_len);
413
414 ret = copy_to_user((void __user *)(unsigned long)hdr->response,
415 rq->sense, len);
416 if (!ret)
417 hdr->response_len = len;
418 else
419 ret = -EFAULT;
420 }
421
422 if (rq->next_rq) {
423 hdr->dout_resid = rq->resid_len;
424 hdr->din_resid = rq->next_rq->resid_len;
425 blk_rq_unmap_user(bidi_bio);
426 blk_put_request(rq->next_rq);
427 } else if (rq_data_dir(rq) == READ)
428 hdr->din_resid = rq->resid_len;
429 else
430 hdr->dout_resid = rq->resid_len;
431
432 /*
433 * If the request generated a negative error number, return it
434 * (providing we aren't already returning an error); if it's
435 * just a protocol response (i.e. non negative), that gets
436 * processed above.
437 */
438 if (!ret && rq->errors < 0)
439 ret = rq->errors;
440
441 blk_rq_unmap_user(bio);
442 if (rq->cmd != rq->__cmd)
443 kfree(rq->cmd);
444 blk_put_request(rq);
445
446 return ret;
447}
448
449static bool bsg_complete(struct bsg_device *bd)
450{
451 bool ret = false;
452 bool spin;
453
454 do {
455 spin_lock_irq(&bd->lock);
456
457 BUG_ON(bd->done_cmds > bd->queued_cmds);
458
459 /*
460 * All commands consumed.
461 */
462 if (bd->done_cmds == bd->queued_cmds)
463 ret = true;
464
465 spin = !test_bit(BSG_F_BLOCK, &bd->flags);
466
467 spin_unlock_irq(&bd->lock);
468 } while (!ret && spin);
469
470 return ret;
471}
472
473static int bsg_complete_all_commands(struct bsg_device *bd)
474{
475 struct bsg_command *bc;
476 int ret, tret;
477
478 dprintk("%s: entered\n", bd->name);
479
480 /*
481 * wait for all commands to complete
482 */
483 io_wait_event(bd->wq_done, bsg_complete(bd));
484
485 /*
486 * discard done commands
487 */
488 ret = 0;
489 do {
490 spin_lock_irq(&bd->lock);
491 if (!bd->queued_cmds) {
492 spin_unlock_irq(&bd->lock);
493 break;
494 }
495 spin_unlock_irq(&bd->lock);
496
497 bc = bsg_get_done_cmd(bd);
498 if (IS_ERR(bc))
499 break;
500
501 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
502 bc->bidi_bio);
503 if (!ret)
504 ret = tret;
505
506 bsg_free_command(bc);
507 } while (1);
508
509 return ret;
510}
511
512static int
513__bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
514 const struct iovec *iov, ssize_t *bytes_read)
515{
516 struct bsg_command *bc;
517 int nr_commands, ret;
518
519 if (count % sizeof(struct sg_io_v4))
520 return -EINVAL;
521
522 ret = 0;
523 nr_commands = count / sizeof(struct sg_io_v4);
524 while (nr_commands) {
525 bc = bsg_get_done_cmd(bd);
526 if (IS_ERR(bc)) {
527 ret = PTR_ERR(bc);
528 break;
529 }
530
531 /*
532 * this is the only case where we need to copy data back
533 * after completing the request. so do that here,
534 * bsg_complete_work() cannot do that for us
535 */
536 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
537 bc->bidi_bio);
538
539 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
540 ret = -EFAULT;
541
542 bsg_free_command(bc);
543
544 if (ret)
545 break;
546
547 buf += sizeof(struct sg_io_v4);
548 *bytes_read += sizeof(struct sg_io_v4);
549 nr_commands--;
550 }
551
552 return ret;
553}
554
555static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
556{
557 if (file->f_flags & O_NONBLOCK)
558 clear_bit(BSG_F_BLOCK, &bd->flags);
559 else
560 set_bit(BSG_F_BLOCK, &bd->flags);
561}
562
563/*
564 * Check if the error is a "real" error that we should return.
565 */
566static inline int err_block_err(int ret)
567{
568 if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
569 return 1;
570
571 return 0;
572}
573
574static ssize_t
575bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
576{
577 struct bsg_device *bd = file->private_data;
578 int ret;
579 ssize_t bytes_read;
580
581 dprintk("%s: read %Zd bytes\n", bd->name, count);
582
583 bsg_set_block(bd, file);
584
585 bytes_read = 0;
586 ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
587 *ppos = bytes_read;
588
589 if (!bytes_read || err_block_err(ret))
590 bytes_read = ret;
591
592 return bytes_read;
593}
594
595static int __bsg_write(struct bsg_device *bd, const char __user *buf,
596 size_t count, ssize_t *bytes_written,
597 fmode_t has_write_perm)
598{
599 struct bsg_command *bc;
600 struct request *rq;
601 int ret, nr_commands;
602
603 if (count % sizeof(struct sg_io_v4))
604 return -EINVAL;
605
606 nr_commands = count / sizeof(struct sg_io_v4);
607 rq = NULL;
608 bc = NULL;
609 ret = 0;
610 while (nr_commands) {
611 struct request_queue *q = bd->queue;
612
613 bc = bsg_alloc_command(bd);
614 if (IS_ERR(bc)) {
615 ret = PTR_ERR(bc);
616 bc = NULL;
617 break;
618 }
619
620 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
621 ret = -EFAULT;
622 break;
623 }
624
625 /*
626 * get a request, fill in the blanks, and add to request queue
627 */
628 rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
629 if (IS_ERR(rq)) {
630 ret = PTR_ERR(rq);
631 rq = NULL;
632 break;
633 }
634
635 bsg_add_command(bd, q, bc, rq);
636 bc = NULL;
637 rq = NULL;
638 nr_commands--;
639 buf += sizeof(struct sg_io_v4);
640 *bytes_written += sizeof(struct sg_io_v4);
641 }
642
643 if (bc)
644 bsg_free_command(bc);
645
646 return ret;
647}
648
649static ssize_t
650bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
651{
652 struct bsg_device *bd = file->private_data;
653 ssize_t bytes_written;
654 int ret;
655
656 dprintk("%s: write %Zd bytes\n", bd->name, count);
657
658 if (unlikely(segment_eq(get_fs(), KERNEL_DS)))
659 return -EINVAL;
660
661 bsg_set_block(bd, file);
662
663 bytes_written = 0;
664 ret = __bsg_write(bd, buf, count, &bytes_written,
665 file->f_mode & FMODE_WRITE);
666
667 *ppos = bytes_written;
668
669 /*
670 * return bytes written on non-fatal errors
671 */
672 if (!bytes_written || err_block_err(ret))
673 bytes_written = ret;
674
675 dprintk("%s: returning %Zd\n", bd->name, bytes_written);
676 return bytes_written;
677}
678
679static struct bsg_device *bsg_alloc_device(void)
680{
681 struct bsg_device *bd;
682
683 bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
684 if (unlikely(!bd))
685 return NULL;
686
687 spin_lock_init(&bd->lock);
688
689 bd->max_queue = BSG_DEFAULT_CMDS;
690
691 INIT_LIST_HEAD(&bd->busy_list);
692 INIT_LIST_HEAD(&bd->done_list);
693 INIT_HLIST_NODE(&bd->dev_list);
694
695 init_waitqueue_head(&bd->wq_free);
696 init_waitqueue_head(&bd->wq_done);
697 return bd;
698}
699
700static void bsg_kref_release_function(struct kref *kref)
701{
702 struct bsg_class_device *bcd =
703 container_of(kref, struct bsg_class_device, ref);
704 struct device *parent = bcd->parent;
705
706 if (bcd->release)
707 bcd->release(bcd->parent);
708
709 put_device(parent);
710}
711
712static int bsg_put_device(struct bsg_device *bd)
713{
714 int ret = 0, do_free;
715 struct request_queue *q = bd->queue;
716
717 mutex_lock(&bsg_mutex);
718
719 do_free = atomic_dec_and_test(&bd->ref_count);
720 if (!do_free) {
721 mutex_unlock(&bsg_mutex);
722 goto out;
723 }
724
725 hlist_del(&bd->dev_list);
726 mutex_unlock(&bsg_mutex);
727
728 dprintk("%s: tearing down\n", bd->name);
729
730 /*
731 * close can always block
732 */
733 set_bit(BSG_F_BLOCK, &bd->flags);
734
735 /*
736 * correct error detection baddies here again. it's the responsibility
737 * of the app to properly reap commands before close() if it wants
738 * fool-proof error detection
739 */
740 ret = bsg_complete_all_commands(bd);
741
742 kfree(bd);
743out:
744 kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
745 if (do_free)
746 blk_put_queue(q);
747 return ret;
748}
749
750static struct bsg_device *bsg_add_device(struct inode *inode,
751 struct request_queue *rq,
752 struct file *file)
753{
754 struct bsg_device *bd;
755#ifdef BSG_DEBUG
756 unsigned char buf[32];
757#endif
758 if (!blk_get_queue(rq))
759 return ERR_PTR(-ENXIO);
760
761 bd = bsg_alloc_device();
762 if (!bd) {
763 blk_put_queue(rq);
764 return ERR_PTR(-ENOMEM);
765 }
766
767 bd->queue = rq;
768
769 bsg_set_block(bd, file);
770
771 atomic_set(&bd->ref_count, 1);
772 mutex_lock(&bsg_mutex);
773 hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
774
775 strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
776 dprintk("bound to <%s>, max queue %d\n",
777 format_dev_t(buf, inode->i_rdev), bd->max_queue);
778
779 mutex_unlock(&bsg_mutex);
780 return bd;
781}
782
783static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
784{
785 struct bsg_device *bd;
786
787 mutex_lock(&bsg_mutex);
788
789 hlist_for_each_entry(bd, bsg_dev_idx_hash(minor), dev_list) {
790 if (bd->queue == q) {
791 atomic_inc(&bd->ref_count);
792 goto found;
793 }
794 }
795 bd = NULL;
796found:
797 mutex_unlock(&bsg_mutex);
798 return bd;
799}
800
801static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
802{
803 struct bsg_device *bd;
804 struct bsg_class_device *bcd;
805
806 /*
807 * find the class device
808 */
809 mutex_lock(&bsg_mutex);
810 bcd = idr_find(&bsg_minor_idr, iminor(inode));
811 if (bcd)
812 kref_get(&bcd->ref);
813 mutex_unlock(&bsg_mutex);
814
815 if (!bcd)
816 return ERR_PTR(-ENODEV);
817
818 bd = __bsg_get_device(iminor(inode), bcd->queue);
819 if (bd)
820 return bd;
821
822 bd = bsg_add_device(inode, bcd->queue, file);
823 if (IS_ERR(bd))
824 kref_put(&bcd->ref, bsg_kref_release_function);
825
826 return bd;
827}
828
829static int bsg_open(struct inode *inode, struct file *file)
830{
831 struct bsg_device *bd;
832
833 bd = bsg_get_device(inode, file);
834
835 if (IS_ERR(bd))
836 return PTR_ERR(bd);
837
838 file->private_data = bd;
839 return 0;
840}
841
842static int bsg_release(struct inode *inode, struct file *file)
843{
844 struct bsg_device *bd = file->private_data;
845
846 file->private_data = NULL;
847 return bsg_put_device(bd);
848}
849
850static unsigned int bsg_poll(struct file *file, poll_table *wait)
851{
852 struct bsg_device *bd = file->private_data;
853 unsigned int mask = 0;
854
855 poll_wait(file, &bd->wq_done, wait);
856 poll_wait(file, &bd->wq_free, wait);
857
858 spin_lock_irq(&bd->lock);
859 if (!list_empty(&bd->done_list))
860 mask |= POLLIN | POLLRDNORM;
861 if (bd->queued_cmds < bd->max_queue)
862 mask |= POLLOUT;
863 spin_unlock_irq(&bd->lock);
864
865 return mask;
866}
867
868static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
869{
870 struct bsg_device *bd = file->private_data;
871 int __user *uarg = (int __user *) arg;
872 int ret;
873
874 switch (cmd) {
875 /*
876 * our own ioctls
877 */
878 case SG_GET_COMMAND_Q:
879 return put_user(bd->max_queue, uarg);
880 case SG_SET_COMMAND_Q: {
881 int queue;
882
883 if (get_user(queue, uarg))
884 return -EFAULT;
885 if (queue < 1)
886 return -EINVAL;
887
888 spin_lock_irq(&bd->lock);
889 bd->max_queue = queue;
890 spin_unlock_irq(&bd->lock);
891 return 0;
892 }
893
894 /*
895 * SCSI/sg ioctls
896 */
897 case SG_GET_VERSION_NUM:
898 case SCSI_IOCTL_GET_IDLUN:
899 case SCSI_IOCTL_GET_BUS_NUMBER:
900 case SG_SET_TIMEOUT:
901 case SG_GET_TIMEOUT:
902 case SG_GET_RESERVED_SIZE:
903 case SG_SET_RESERVED_SIZE:
904 case SG_EMULATED_HOST:
905 case SCSI_IOCTL_SEND_COMMAND: {
906 void __user *uarg = (void __user *) arg;
907 return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
908 }
909 case SG_IO: {
910 struct request *rq;
911 struct bio *bio, *bidi_bio = NULL;
912 struct sg_io_v4 hdr;
913 int at_head;
914 u8 sense[SCSI_SENSE_BUFFERSIZE];
915
916 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
917 return -EFAULT;
918
919 rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
920 if (IS_ERR(rq))
921 return PTR_ERR(rq);
922
923 bio = rq->bio;
924 if (rq->next_rq)
925 bidi_bio = rq->next_rq->bio;
926
927 at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
928 blk_execute_rq(bd->queue, NULL, rq, at_head);
929 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
930
931 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
932 return -EFAULT;
933
934 return ret;
935 }
936 /*
937 * block device ioctls
938 */
939 default:
940#if 0
941 return ioctl_by_bdev(bd->bdev, cmd, arg);
942#else
943 return -ENOTTY;
944#endif
945 }
946}
947
948static const struct file_operations bsg_fops = {
949 .read = bsg_read,
950 .write = bsg_write,
951 .poll = bsg_poll,
952 .open = bsg_open,
953 .release = bsg_release,
954 .unlocked_ioctl = bsg_ioctl,
955 .owner = THIS_MODULE,
956 .llseek = default_llseek,
957};
958
959void bsg_unregister_queue(struct request_queue *q)
960{
961 struct bsg_class_device *bcd = &q->bsg_dev;
962
963 if (!bcd->class_dev)
964 return;
965
966 mutex_lock(&bsg_mutex);
967 idr_remove(&bsg_minor_idr, bcd->minor);
968 if (q->kobj.sd)
969 sysfs_remove_link(&q->kobj, "bsg");
970 device_unregister(bcd->class_dev);
971 bcd->class_dev = NULL;
972 kref_put(&bcd->ref, bsg_kref_release_function);
973 mutex_unlock(&bsg_mutex);
974}
975EXPORT_SYMBOL_GPL(bsg_unregister_queue);
976
977int bsg_register_queue(struct request_queue *q, struct device *parent,
978 const char *name, void (*release)(struct device *))
979{
980 struct bsg_class_device *bcd;
981 dev_t dev;
982 int ret;
983 struct device *class_dev = NULL;
984 const char *devname;
985
986 if (name)
987 devname = name;
988 else
989 devname = dev_name(parent);
990
991 /*
992 * we need a proper transport to send commands, not a stacked device
993 */
994 if (!queue_is_rq_based(q))
995 return 0;
996
997 bcd = &q->bsg_dev;
998 memset(bcd, 0, sizeof(*bcd));
999
1000 mutex_lock(&bsg_mutex);
1001
1002 ret = idr_alloc(&bsg_minor_idr, bcd, 0, BSG_MAX_DEVS, GFP_KERNEL);
1003 if (ret < 0) {
1004 if (ret == -ENOSPC) {
1005 printk(KERN_ERR "bsg: too many bsg devices\n");
1006 ret = -EINVAL;
1007 }
1008 goto unlock;
1009 }
1010
1011 bcd->minor = ret;
1012 bcd->queue = q;
1013 bcd->parent = get_device(parent);
1014 bcd->release = release;
1015 kref_init(&bcd->ref);
1016 dev = MKDEV(bsg_major, bcd->minor);
1017 class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1018 if (IS_ERR(class_dev)) {
1019 ret = PTR_ERR(class_dev);
1020 goto put_dev;
1021 }
1022 bcd->class_dev = class_dev;
1023
1024 if (q->kobj.sd) {
1025 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1026 if (ret)
1027 goto unregister_class_dev;
1028 }
1029
1030 mutex_unlock(&bsg_mutex);
1031 return 0;
1032
1033unregister_class_dev:
1034 device_unregister(class_dev);
1035put_dev:
1036 put_device(parent);
1037 idr_remove(&bsg_minor_idr, bcd->minor);
1038unlock:
1039 mutex_unlock(&bsg_mutex);
1040 return ret;
1041}
1042EXPORT_SYMBOL_GPL(bsg_register_queue);
1043
1044static struct cdev bsg_cdev;
1045
1046static char *bsg_devnode(struct device *dev, umode_t *mode)
1047{
1048 return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1049}
1050
1051static int __init bsg_init(void)
1052{
1053 int ret, i;
1054 dev_t devid;
1055
1056 bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1057 sizeof(struct bsg_command), 0, 0, NULL);
1058 if (!bsg_cmd_cachep) {
1059 printk(KERN_ERR "bsg: failed creating slab cache\n");
1060 return -ENOMEM;
1061 }
1062
1063 for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1064 INIT_HLIST_HEAD(&bsg_device_list[i]);
1065
1066 bsg_class = class_create(THIS_MODULE, "bsg");
1067 if (IS_ERR(bsg_class)) {
1068 ret = PTR_ERR(bsg_class);
1069 goto destroy_kmemcache;
1070 }
1071 bsg_class->devnode = bsg_devnode;
1072
1073 ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1074 if (ret)
1075 goto destroy_bsg_class;
1076
1077 bsg_major = MAJOR(devid);
1078
1079 cdev_init(&bsg_cdev, &bsg_fops);
1080 ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1081 if (ret)
1082 goto unregister_chrdev;
1083
1084 printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1085 " loaded (major %d)\n", bsg_major);
1086 return 0;
1087unregister_chrdev:
1088 unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1089destroy_bsg_class:
1090 class_destroy(bsg_class);
1091destroy_kmemcache:
1092 kmem_cache_destroy(bsg_cmd_cachep);
1093 return ret;
1094}
1095
1096MODULE_AUTHOR("Jens Axboe");
1097MODULE_DESCRIPTION(BSG_DESCRIPTION);
1098MODULE_LICENSE("GPL");
1099
1100device_initcall(bsg_init);