blob: a333c9337093f8c580105c45e89ab49690f7d242 [file] [log] [blame]
Jens Axboe3d6392c2007-07-09 12:38:05 +02001/*
2 * bsg.c - block layer implementation of the sg v3 interface
3 *
4 * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
5 * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
6 *
7 * This file is subject to the terms and conditions of the GNU General Public
8 * License version 2. See the file "COPYING" in the main directory of this
9 * archive for more details.
10 *
11 */
12/*
13 * TODO
14 * - Should this get merged, block/scsi_ioctl.c will be migrated into
15 * this file. To keep maintenance down, it's easier to have them
16 * seperated right now.
17 *
18 */
Jens Axboe3d6392c2007-07-09 12:38:05 +020019#include <linux/module.h>
20#include <linux/init.h>
21#include <linux/file.h>
22#include <linux/blkdev.h>
23#include <linux/poll.h>
24#include <linux/cdev.h>
25#include <linux/percpu.h>
26#include <linux/uio.h>
27#include <linux/bsg.h>
28
29#include <scsi/scsi.h>
30#include <scsi/scsi_ioctl.h>
31#include <scsi/scsi_cmnd.h>
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +020032#include <scsi/scsi_device.h>
33#include <scsi/scsi_driver.h>
Jens Axboe3d6392c2007-07-09 12:38:05 +020034#include <scsi/sg.h>
35
36static char bsg_version[] = "block layer sg (bsg) 0.4";
37
Jens Axboe3d6392c2007-07-09 12:38:05 +020038struct bsg_device {
Jens Axboe3d6392c2007-07-09 12:38:05 +020039 request_queue_t *queue;
40 spinlock_t lock;
41 struct list_head busy_list;
42 struct list_head done_list;
43 struct hlist_node dev_list;
44 atomic_t ref_count;
45 int minor;
46 int queued_cmds;
47 int done_cmds;
Jens Axboe3d6392c2007-07-09 12:38:05 +020048 wait_queue_head_t wq_done;
49 wait_queue_head_t wq_free;
FUJITA Tomonorid351af02007-07-09 12:40:35 +020050 char name[BUS_ID_SIZE];
Jens Axboe3d6392c2007-07-09 12:38:05 +020051 int max_queue;
52 unsigned long flags;
53};
54
55enum {
56 BSG_F_BLOCK = 1,
57 BSG_F_WRITE_PERM = 2,
58};
59
Jens Axboe5309cb32007-01-23 16:24:41 +010060#define BSG_DEFAULT_CMDS 64
FUJITA Tomonori292b7f22007-03-28 13:29:58 +020061#define BSG_MAX_DEVS 32768
Jens Axboe3d6392c2007-07-09 12:38:05 +020062
63#undef BSG_DEBUG
64
65#ifdef BSG_DEBUG
66#define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
67#else
68#define dprintk(fmt, args...)
69#endif
70
71#define list_entry_bc(entry) list_entry((entry), struct bsg_command, list)
72
73/*
74 * just for testing
75 */
76#define BSG_MAJOR (240)
77
78static DEFINE_MUTEX(bsg_mutex);
FUJITA Tomonori292b7f22007-03-28 13:29:58 +020079static int bsg_device_nr, bsg_minor_idx;
Jens Axboe3d6392c2007-07-09 12:38:05 +020080
81#define BSG_LIST_SIZE (8)
82#define bsg_list_idx(minor) ((minor) & (BSG_LIST_SIZE - 1))
83static struct hlist_head bsg_device_list[BSG_LIST_SIZE];
84
85static struct class *bsg_class;
86static LIST_HEAD(bsg_class_list);
87
Jens Axboe5309cb32007-01-23 16:24:41 +010088static struct kmem_cache *bsg_cmd_cachep;
89
Jens Axboe3d6392c2007-07-09 12:38:05 +020090/*
91 * our internal command type
92 */
93struct bsg_command {
94 struct bsg_device *bd;
95 struct list_head list;
96 struct request *rq;
97 struct bio *bio;
98 int err;
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +010099 struct sg_io_v4 hdr;
100 struct sg_io_v4 __user *uhdr;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200101 char sense[SCSI_SENSE_BUFFERSIZE];
102};
103
104static void bsg_free_command(struct bsg_command *bc)
105{
106 struct bsg_device *bd = bc->bd;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200107 unsigned long flags;
108
Jens Axboe5309cb32007-01-23 16:24:41 +0100109 kmem_cache_free(bsg_cmd_cachep, bc);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200110
111 spin_lock_irqsave(&bd->lock, flags);
112 bd->queued_cmds--;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200113 spin_unlock_irqrestore(&bd->lock, flags);
114
115 wake_up(&bd->wq_free);
116}
117
118static struct bsg_command *__bsg_alloc_command(struct bsg_device *bd)
119{
120 struct bsg_command *bc = NULL;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200121
122 spin_lock_irq(&bd->lock);
123
124 if (bd->queued_cmds >= bd->max_queue)
125 goto out;
126
Jens Axboe3d6392c2007-07-09 12:38:05 +0200127 bd->queued_cmds++;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200128 spin_unlock_irq(&bd->lock);
129
Jens Axboe5309cb32007-01-23 16:24:41 +0100130 bc = kmem_cache_alloc(bsg_cmd_cachep, GFP_USER);
131 if (unlikely(!bc)) {
132 spin_lock_irq(&bd->lock);
FUJITA Tomonori7e75d732007-01-24 09:05:54 +0100133 bd->queued_cmds--;
134 goto out;
Jens Axboe5309cb32007-01-23 16:24:41 +0100135 }
136
Jens Axboe3d6392c2007-07-09 12:38:05 +0200137 memset(bc, 0, sizeof(*bc));
138 bc->bd = bd;
139 INIT_LIST_HEAD(&bc->list);
Jens Axboe5309cb32007-01-23 16:24:41 +0100140 dprintk("%s: returning free cmd %p\n", bd->name, bc);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200141 return bc;
142out:
Jens Axboe3d6392c2007-07-09 12:38:05 +0200143 spin_unlock_irq(&bd->lock);
144 return bc;
145}
146
147static inline void
148bsg_del_done_cmd(struct bsg_device *bd, struct bsg_command *bc)
149{
150 bd->done_cmds--;
151 list_del(&bc->list);
152}
153
154static inline void
155bsg_add_done_cmd(struct bsg_device *bd, struct bsg_command *bc)
156{
157 bd->done_cmds++;
158 list_add_tail(&bc->list, &bd->done_list);
159 wake_up(&bd->wq_done);
160}
161
162static inline int bsg_io_schedule(struct bsg_device *bd, int state)
163{
164 DEFINE_WAIT(wait);
165 int ret = 0;
166
167 spin_lock_irq(&bd->lock);
168
169 BUG_ON(bd->done_cmds > bd->queued_cmds);
170
171 /*
172 * -ENOSPC or -ENODATA? I'm going for -ENODATA, meaning "I have no
173 * work to do", even though we return -ENOSPC after this same test
174 * during bsg_write() -- there, it means our buffer can't have more
175 * bsg_commands added to it, thus has no space left.
176 */
177 if (bd->done_cmds == bd->queued_cmds) {
178 ret = -ENODATA;
179 goto unlock;
180 }
181
182 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
183 ret = -EAGAIN;
184 goto unlock;
185 }
186
187 prepare_to_wait(&bd->wq_done, &wait, state);
188 spin_unlock_irq(&bd->lock);
189 io_schedule();
190 finish_wait(&bd->wq_done, &wait);
191
192 if ((state == TASK_INTERRUPTIBLE) && signal_pending(current))
193 ret = -ERESTARTSYS;
194
195 return ret;
196unlock:
197 spin_unlock_irq(&bd->lock);
198 return ret;
199}
200
201/*
202 * get a new free command, blocking if needed and specified
203 */
204static struct bsg_command *bsg_get_command(struct bsg_device *bd)
205{
206 struct bsg_command *bc;
207 int ret;
208
209 do {
210 bc = __bsg_alloc_command(bd);
211 if (bc)
212 break;
213
214 ret = bsg_io_schedule(bd, TASK_INTERRUPTIBLE);
215 if (ret) {
216 bc = ERR_PTR(ret);
217 break;
218 }
219
220 } while (1);
221
222 return bc;
223}
224
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100225static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq,
226 struct sg_io_v4 *hdr, int has_write_perm)
Jens Axboe3d6392c2007-07-09 12:38:05 +0200227{
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100228 memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
229
230 if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
231 hdr->request_len))
232 return -EFAULT;
233 if (blk_verify_command(rq->cmd, has_write_perm))
234 return -EPERM;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200235
236 /*
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100237 * fill in request structure
Jens Axboe3d6392c2007-07-09 12:38:05 +0200238 */
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100239 rq->cmd_len = hdr->request_len;
240 rq->cmd_type = REQ_TYPE_BLOCK_PC;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200241
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100242 rq->timeout = (hdr->timeout * HZ) / 1000;
243 if (!rq->timeout)
244 rq->timeout = q->sg_timeout;
245 if (!rq->timeout)
246 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200247
248 return 0;
249}
250
251/*
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100252 * Check if sg_io_v4 from user is allowed and valid
253 */
254static int
255bsg_validate_sgv4_hdr(request_queue_t *q, struct sg_io_v4 *hdr, int *rw)
256{
257 if (hdr->guard != 'Q')
258 return -EINVAL;
259 if (hdr->request_len > BLK_MAX_CDB)
260 return -EINVAL;
261 if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
262 hdr->din_xfer_len > (q->max_sectors << 9))
263 return -EIO;
264
265 /* not supported currently */
266 if (hdr->protocol || hdr->subprotocol)
267 return -EINVAL;
268
269 /*
270 * looks sane, if no data then it should be fine from our POV
271 */
272 if (!hdr->dout_xfer_len && !hdr->din_xfer_len)
273 return 0;
274
275 /* not supported currently */
276 if (hdr->dout_xfer_len && hdr->din_xfer_len)
277 return -EINVAL;
278
279 *rw = hdr->dout_xfer_len ? WRITE : READ;
280
281 return 0;
282}
283
284/*
285 * map sg_io_v4 to a request.
Jens Axboe3d6392c2007-07-09 12:38:05 +0200286 */
287static struct request *
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100288bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
Jens Axboe3d6392c2007-07-09 12:38:05 +0200289{
290 request_queue_t *q = bd->queue;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200291 struct request *rq;
Jens Axboe2ef70862006-12-20 11:26:11 +0100292 int ret, rw = 0; /* shut up gcc */
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100293 unsigned int dxfer_len;
294 void *dxferp = NULL;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200295
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100296 dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
297 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
298 hdr->din_xfer_len);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200299
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100300 ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200301 if (ret)
302 return ERR_PTR(ret);
303
304 /*
305 * map scatter-gather elements seperately and string them to request
306 */
307 rq = blk_get_request(q, rw, GFP_KERNEL);
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100308 ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
309 &bd->flags));
Jens Axboe3d6392c2007-07-09 12:38:05 +0200310 if (ret) {
311 blk_put_request(rq);
312 return ERR_PTR(ret);
313 }
314
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100315 if (hdr->dout_xfer_len) {
316 dxfer_len = hdr->dout_xfer_len;
317 dxferp = (void*)(unsigned long)hdr->dout_xferp;
318 } else if (hdr->din_xfer_len) {
319 dxfer_len = hdr->din_xfer_len;
320 dxferp = (void*)(unsigned long)hdr->din_xferp;
321 } else
322 dxfer_len = 0;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200323
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100324 if (dxfer_len) {
325 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
326 if (ret) {
327 dprintk("failed map at %d\n", ret);
328 blk_put_request(rq);
329 rq = ERR_PTR(ret);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200330 }
Jens Axboe3d6392c2007-07-09 12:38:05 +0200331 }
332
333 return rq;
334}
335
336/*
337 * async completion call-back from the block layer, when scsi/ide/whatever
338 * calls end_that_request_last() on a request
339 */
340static void bsg_rq_end_io(struct request *rq, int uptodate)
341{
342 struct bsg_command *bc = rq->end_io_data;
343 struct bsg_device *bd = bc->bd;
344 unsigned long flags;
345
Jens Axboe5309cb32007-01-23 16:24:41 +0100346 dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
347 bd->name, rq, bc, bc->bio, uptodate);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200348
349 bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
350
351 spin_lock_irqsave(&bd->lock, flags);
352 list_del(&bc->list);
353 bsg_add_done_cmd(bd, bc);
354 spin_unlock_irqrestore(&bd->lock, flags);
355}
356
357/*
358 * do final setup of a 'bc' and submit the matching 'rq' to the block
359 * layer for io
360 */
361static void bsg_add_command(struct bsg_device *bd, request_queue_t *q,
362 struct bsg_command *bc, struct request *rq)
363{
364 rq->sense = bc->sense;
365 rq->sense_len = 0;
366
367 /*
368 * add bc command to busy queue and submit rq for io
369 */
370 bc->rq = rq;
371 bc->bio = rq->bio;
372 bc->hdr.duration = jiffies;
373 spin_lock_irq(&bd->lock);
374 list_add_tail(&bc->list, &bd->busy_list);
375 spin_unlock_irq(&bd->lock);
376
377 dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
378
379 rq->end_io_data = bc;
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200380 blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200381}
382
383static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
384{
385 struct bsg_command *bc = NULL;
386
387 spin_lock_irq(&bd->lock);
388 if (bd->done_cmds) {
389 bc = list_entry_bc(bd->done_list.next);
390 bsg_del_done_cmd(bd, bc);
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, int state)
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 ret = bsg_io_schedule(bd, state);
411 if (ret) {
412 bc = ERR_PTR(ret);
413 break;
414 }
415 } while (1);
416
417 dprintk("%s: returning done %p\n", bd->name, bc);
418
419 return bc;
420}
421
422static struct bsg_command *
423bsg_get_done_cmd(struct bsg_device *bd, const struct iovec *iov)
424{
425 return __bsg_get_done_cmd(bd, TASK_INTERRUPTIBLE);
426}
427
428static struct bsg_command *
429bsg_get_done_cmd_nosignals(struct bsg_device *bd)
430{
431 return __bsg_get_done_cmd(bd, TASK_UNINTERRUPTIBLE);
432}
433
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100434static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
435 struct bio *bio)
436{
437 int ret = 0;
438
439 dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
440 /*
441 * fill in all the output members
442 */
443 hdr->device_status = status_byte(rq->errors);
444 hdr->transport_status = host_byte(rq->errors);
445 hdr->driver_status = driver_byte(rq->errors);
446 hdr->info = 0;
447 if (hdr->device_status || hdr->transport_status || hdr->driver_status)
448 hdr->info |= SG_INFO_CHECK;
449 hdr->din_resid = rq->data_len;
450 hdr->response_len = 0;
451
452 if (rq->sense_len && hdr->response) {
453 int len = min((unsigned int) hdr->max_response_len,
454 rq->sense_len);
455
456 ret = copy_to_user((void*)(unsigned long)hdr->response,
457 rq->sense, len);
458 if (!ret)
459 hdr->response_len = len;
460 else
461 ret = -EFAULT;
462 }
463
464 blk_rq_unmap_user(bio);
465 blk_put_request(rq);
466
467 return ret;
468}
469
Jens Axboe3d6392c2007-07-09 12:38:05 +0200470static int bsg_complete_all_commands(struct bsg_device *bd)
471{
472 struct bsg_command *bc;
473 int ret, tret;
474
475 dprintk("%s: entered\n", bd->name);
476
477 set_bit(BSG_F_BLOCK, &bd->flags);
478
479 /*
480 * wait for all commands to complete
481 */
482 ret = 0;
483 do {
484 ret = bsg_io_schedule(bd, TASK_UNINTERRUPTIBLE);
485 /*
486 * look for -ENODATA specifically -- we'll sometimes get
487 * -ERESTARTSYS when we've taken a signal, but we can't
488 * return until we're done freeing the queue, so ignore
489 * it. The signal will get handled when we're done freeing
490 * the bsg_device.
491 */
492 } while (ret != -ENODATA);
493
494 /*
495 * discard done commands
496 */
497 ret = 0;
498 do {
499 bc = bsg_get_done_cmd_nosignals(bd);
500
501 /*
502 * we _must_ complete before restarting, because
503 * bsg_release can't handle this failing.
504 */
505 if (PTR_ERR(bc) == -ERESTARTSYS)
506 continue;
507 if (IS_ERR(bc)) {
508 ret = PTR_ERR(bc);
509 break;
510 }
511
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100512 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200513 if (!ret)
514 ret = tret;
515
516 bsg_free_command(bc);
517 } while (1);
518
519 return ret;
520}
521
522typedef struct bsg_command *(*bsg_command_callback)(struct bsg_device *bd, const struct iovec *iov);
523
524static ssize_t
525__bsg_read(char __user *buf, size_t count, bsg_command_callback get_bc,
526 struct bsg_device *bd, const struct iovec *iov, ssize_t *bytes_read)
527{
528 struct bsg_command *bc;
529 int nr_commands, ret;
530
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100531 if (count % sizeof(struct sg_io_v4))
Jens Axboe3d6392c2007-07-09 12:38:05 +0200532 return -EINVAL;
533
534 ret = 0;
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100535 nr_commands = count / sizeof(struct sg_io_v4);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200536 while (nr_commands) {
537 bc = get_bc(bd, iov);
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 */
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100548 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200549
550 if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr)))
551 ret = -EFAULT;
552
553 bsg_free_command(bc);
554
555 if (ret)
556 break;
557
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100558 buf += sizeof(struct sg_io_v4);
559 *bytes_read += sizeof(struct sg_io_v4);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200560 nr_commands--;
561 }
562
563 return ret;
564}
565
566static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
567{
568 if (file->f_flags & O_NONBLOCK)
569 clear_bit(BSG_F_BLOCK, &bd->flags);
570 else
571 set_bit(BSG_F_BLOCK, &bd->flags);
572}
573
574static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
575{
576 if (file->f_mode & FMODE_WRITE)
577 set_bit(BSG_F_WRITE_PERM, &bd->flags);
578 else
579 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
580}
581
582static inline int err_block_err(int ret)
583{
584 if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
585 return 1;
586
587 return 0;
588}
589
590static ssize_t
591bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
592{
593 struct bsg_device *bd = file->private_data;
594 int ret;
595 ssize_t bytes_read;
596
FUJITA Tomonori9e69fbb2006-12-20 11:18:22 +0100597 dprintk("%s: read %Zd bytes\n", bd->name, count);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200598
599 bsg_set_block(bd, file);
600 bytes_read = 0;
601 ret = __bsg_read(buf, count, bsg_get_done_cmd,
602 bd, NULL, &bytes_read);
603 *ppos = bytes_read;
604
605 if (!bytes_read || (bytes_read && err_block_err(ret)))
606 bytes_read = ret;
607
608 return bytes_read;
609}
610
611static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf,
612 size_t count, ssize_t *bytes_read)
613{
614 struct bsg_command *bc;
615 struct request *rq;
616 int ret, nr_commands;
617
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100618 if (count % sizeof(struct sg_io_v4))
Jens Axboe3d6392c2007-07-09 12:38:05 +0200619 return -EINVAL;
620
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100621 nr_commands = count / sizeof(struct sg_io_v4);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200622 rq = NULL;
623 bc = NULL;
624 ret = 0;
625 while (nr_commands) {
626 request_queue_t *q = bd->queue;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200627
628 bc = bsg_get_command(bd);
629 if (!bc)
630 break;
631 if (IS_ERR(bc)) {
632 ret = PTR_ERR(bc);
633 bc = NULL;
634 break;
635 }
636
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100637 bc->uhdr = (struct sg_io_v4 __user *) buf;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200638 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
639 ret = -EFAULT;
640 break;
641 }
642
643 /*
644 * get a request, fill in the blanks, and add to request queue
645 */
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100646 rq = bsg_map_hdr(bd, &bc->hdr);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200647 if (IS_ERR(rq)) {
648 ret = PTR_ERR(rq);
649 rq = NULL;
650 break;
651 }
652
653 bsg_add_command(bd, q, bc, rq);
654 bc = NULL;
655 rq = NULL;
656 nr_commands--;
FUJITA Tomonori70e36ec2006-12-20 11:20:15 +0100657 buf += sizeof(struct sg_io_v4);
658 *bytes_read += sizeof(struct sg_io_v4);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200659 }
660
Jens Axboe3d6392c2007-07-09 12:38:05 +0200661 if (bc)
662 bsg_free_command(bc);
663
664 return ret;
665}
666
667static ssize_t
668bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
669{
670 struct bsg_device *bd = file->private_data;
671 ssize_t bytes_read;
672 int ret;
673
FUJITA Tomonori9e69fbb2006-12-20 11:18:22 +0100674 dprintk("%s: write %Zd bytes\n", bd->name, count);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200675
676 bsg_set_block(bd, file);
677 bsg_set_write_perm(bd, file);
678
679 bytes_read = 0;
680 ret = __bsg_write(bd, buf, count, &bytes_read);
681 *ppos = bytes_read;
682
683 /*
684 * return bytes written on non-fatal errors
685 */
686 if (!bytes_read || (bytes_read && err_block_err(ret)))
687 bytes_read = ret;
688
FUJITA Tomonori9e69fbb2006-12-20 11:18:22 +0100689 dprintk("%s: returning %Zd\n", bd->name, bytes_read);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200690 return bytes_read;
691}
692
Jens Axboe3d6392c2007-07-09 12:38:05 +0200693static struct bsg_device *bsg_alloc_device(void)
694{
Jens Axboe3d6392c2007-07-09 12:38:05 +0200695 struct bsg_device *bd;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200696
697 bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
698 if (unlikely(!bd))
699 return NULL;
700
701 spin_lock_init(&bd->lock);
702
Jens Axboe5309cb32007-01-23 16:24:41 +0100703 bd->max_queue = BSG_DEFAULT_CMDS;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200704
705 INIT_LIST_HEAD(&bd->busy_list);
706 INIT_LIST_HEAD(&bd->done_list);
707 INIT_HLIST_NODE(&bd->dev_list);
708
709 init_waitqueue_head(&bd->wq_free);
710 init_waitqueue_head(&bd->wq_done);
711 return bd;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200712}
713
714static int bsg_put_device(struct bsg_device *bd)
715{
716 int ret = 0;
717
718 mutex_lock(&bsg_mutex);
719
720 if (!atomic_dec_and_test(&bd->ref_count))
721 goto out;
722
723 dprintk("%s: tearing down\n", bd->name);
724
725 /*
726 * close can always block
727 */
728 set_bit(BSG_F_BLOCK, &bd->flags);
729
730 /*
731 * correct error detection baddies here again. it's the responsibility
732 * of the app to properly reap commands before close() if it wants
733 * fool-proof error detection
734 */
735 ret = bsg_complete_all_commands(bd);
736
737 blk_put_queue(bd->queue);
738 hlist_del(&bd->dev_list);
Jens Axboe5309cb32007-01-23 16:24:41 +0100739 kfree(bd);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200740out:
741 mutex_unlock(&bsg_mutex);
742 return ret;
743}
744
745static struct bsg_device *bsg_add_device(struct inode *inode,
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200746 struct request_queue *rq,
Jens Axboe3d6392c2007-07-09 12:38:05 +0200747 struct file *file)
748{
749 struct bsg_device *bd = NULL;
750#ifdef BSG_DEBUG
751 unsigned char buf[32];
752#endif
753
754 bd = bsg_alloc_device();
755 if (!bd)
756 return ERR_PTR(-ENOMEM);
757
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200758 bd->queue = rq;
759 kobject_get(&rq->kobj);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200760 bsg_set_block(bd, file);
761
762 atomic_set(&bd->ref_count, 1);
763 bd->minor = iminor(inode);
764 mutex_lock(&bsg_mutex);
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200765 hlist_add_head(&bd->dev_list, &bsg_device_list[bsg_list_idx(bd->minor)]);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200766
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200767 strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200768 dprintk("bound to <%s>, max queue %d\n",
FUJITA Tomonori9e69fbb2006-12-20 11:18:22 +0100769 format_dev_t(buf, inode->i_rdev), bd->max_queue);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200770
771 mutex_unlock(&bsg_mutex);
772 return bd;
773}
774
775static struct bsg_device *__bsg_get_device(int minor)
776{
777 struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)];
778 struct bsg_device *bd = NULL;
779 struct hlist_node *entry;
780
781 mutex_lock(&bsg_mutex);
782
783 hlist_for_each(entry, list) {
784 bd = hlist_entry(entry, struct bsg_device, dev_list);
785 if (bd->minor == minor) {
786 atomic_inc(&bd->ref_count);
787 break;
788 }
789
790 bd = NULL;
791 }
792
793 mutex_unlock(&bsg_mutex);
794 return bd;
795}
796
797static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
798{
799 struct bsg_device *bd = __bsg_get_device(iminor(inode));
800 struct bsg_class_device *bcd, *__bcd;
801
802 if (bd)
803 return bd;
804
805 /*
806 * find the class device
807 */
808 bcd = NULL;
809 mutex_lock(&bsg_mutex);
810 list_for_each_entry(__bcd, &bsg_class_list, list) {
811 if (__bcd->minor == iminor(inode)) {
812 bcd = __bcd;
813 break;
814 }
815 }
816 mutex_unlock(&bsg_mutex);
817
818 if (!bcd)
819 return ERR_PTR(-ENODEV);
820
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200821 return bsg_add_device(inode, bcd->queue, file);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200822}
823
824static int bsg_open(struct inode *inode, struct file *file)
825{
826 struct bsg_device *bd = bsg_get_device(inode, file);
827
828 if (IS_ERR(bd))
829 return PTR_ERR(bd);
830
831 file->private_data = bd;
832 return 0;
833}
834
835static int bsg_release(struct inode *inode, struct file *file)
836{
837 struct bsg_device *bd = file->private_data;
838
839 file->private_data = NULL;
840 return bsg_put_device(bd);
841}
842
843static unsigned int bsg_poll(struct file *file, poll_table *wait)
844{
845 struct bsg_device *bd = file->private_data;
846 unsigned int mask = 0;
847
848 poll_wait(file, &bd->wq_done, wait);
849 poll_wait(file, &bd->wq_free, wait);
850
851 spin_lock_irq(&bd->lock);
852 if (!list_empty(&bd->done_list))
853 mask |= POLLIN | POLLRDNORM;
854 if (bd->queued_cmds >= bd->max_queue)
855 mask |= POLLOUT;
856 spin_unlock_irq(&bd->lock);
857
858 return mask;
859}
860
861static int
862bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
863 unsigned long arg)
864{
865 struct bsg_device *bd = file->private_data;
866 int __user *uarg = (int __user *) arg;
867
868 if (!bd)
869 return -ENXIO;
870
871 switch (cmd) {
872 /*
873 * our own ioctls
874 */
875 case SG_GET_COMMAND_Q:
876 return put_user(bd->max_queue, uarg);
Jens Axboe5309cb32007-01-23 16:24:41 +0100877 case SG_SET_COMMAND_Q: {
Jens Axboe3d6392c2007-07-09 12:38:05 +0200878 int queue;
879
880 if (get_user(queue, uarg))
881 return -EFAULT;
Jens Axboe5309cb32007-01-23 16:24:41 +0100882 if (queue < 1)
Jens Axboe3d6392c2007-07-09 12:38:05 +0200883 return -EINVAL;
884
Jens Axboe5309cb32007-01-23 16:24:41 +0100885 spin_lock_irq(&bd->lock);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200886 bd->max_queue = queue;
Jens Axboe5309cb32007-01-23 16:24:41 +0100887 spin_unlock_irq(&bd->lock);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200888 return 0;
889 }
890
891 /*
892 * SCSI/sg ioctls
893 */
894 case SG_GET_VERSION_NUM:
895 case SCSI_IOCTL_GET_IDLUN:
896 case SCSI_IOCTL_GET_BUS_NUMBER:
897 case SG_SET_TIMEOUT:
898 case SG_GET_TIMEOUT:
899 case SG_GET_RESERVED_SIZE:
900 case SG_SET_RESERVED_SIZE:
901 case SG_EMULATED_HOST:
Jens Axboe3d6392c2007-07-09 12:38:05 +0200902 case SCSI_IOCTL_SEND_COMMAND: {
903 void __user *uarg = (void __user *) arg;
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200904 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200905 }
FUJITA Tomonori10e88552006-12-20 11:20:57 +0100906 case SG_IO: {
907 struct request *rq;
908 struct bio *bio;
909 struct sg_io_v4 hdr;
910
911 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
912 return -EFAULT;
913
914 rq = bsg_map_hdr(bd, &hdr);
915 if (IS_ERR(rq))
916 return PTR_ERR(rq);
917
918 bio = rq->bio;
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200919 blk_execute_rq(bd->queue, NULL, rq, 0);
FUJITA Tomonori10e88552006-12-20 11:20:57 +0100920 blk_complete_sgv4_hdr_rq(rq, &hdr, bio);
921
922 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
923 return -EFAULT;
Jens Axboeb711afa2006-12-20 11:25:23 +0100924
925 return 0;
FUJITA Tomonori10e88552006-12-20 11:20:57 +0100926 }
Jens Axboe3d6392c2007-07-09 12:38:05 +0200927 /*
928 * block device ioctls
929 */
930 default:
931#if 0
932 return ioctl_by_bdev(bd->bdev, cmd, arg);
933#else
934 return -ENOTTY;
935#endif
936 }
937}
938
939static struct file_operations bsg_fops = {
940 .read = bsg_read,
941 .write = bsg_write,
942 .poll = bsg_poll,
943 .open = bsg_open,
944 .release = bsg_release,
945 .ioctl = bsg_ioctl,
946 .owner = THIS_MODULE,
947};
948
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200949void bsg_unregister_queue(struct request_queue *q)
Jens Axboe3d6392c2007-07-09 12:38:05 +0200950{
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200951 struct bsg_class_device *bcd = &q->bsg_dev;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200952
953 if (!bcd->class_dev)
954 return;
955
956 mutex_lock(&bsg_mutex);
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200957 sysfs_remove_link(&q->kobj, "bsg");
Jens Axboe3d6392c2007-07-09 12:38:05 +0200958 class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor));
959 bcd->class_dev = NULL;
960 list_del_init(&bcd->list);
FUJITA Tomonori292b7f22007-03-28 13:29:58 +0200961 bsg_device_nr--;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200962 mutex_unlock(&bsg_mutex);
963}
FUJITA Tomonori4cf07232007-03-30 11:19:39 +0200964EXPORT_SYMBOL_GPL(bsg_unregister_queue);
Jens Axboe3d6392c2007-07-09 12:38:05 +0200965
FUJITA Tomonori4cf07232007-03-30 11:19:39 +0200966int bsg_register_queue(struct request_queue *q, const char *name)
Jens Axboe3d6392c2007-07-09 12:38:05 +0200967{
FUJITA Tomonori292b7f22007-03-28 13:29:58 +0200968 struct bsg_class_device *bcd, *__bcd;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200969 dev_t dev;
FUJITA Tomonori292b7f22007-03-28 13:29:58 +0200970 int ret = -EMFILE;
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +0200971 struct class_device *class_dev = NULL;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200972
973 /*
974 * we need a proper transport to send commands, not a stacked device
975 */
976 if (!q->request_fn)
977 return 0;
978
FUJITA Tomonorid351af02007-07-09 12:40:35 +0200979 bcd = &q->bsg_dev;
Jens Axboe3d6392c2007-07-09 12:38:05 +0200980 memset(bcd, 0, sizeof(*bcd));
981 INIT_LIST_HEAD(&bcd->list);
982
983 mutex_lock(&bsg_mutex);
FUJITA Tomonori292b7f22007-03-28 13:29:58 +0200984 if (bsg_device_nr == BSG_MAX_DEVS) {
985 printk(KERN_ERR "bsg: too many bsg devices\n");
986 goto err;
987 }
988
989retry:
990 list_for_each_entry(__bcd, &bsg_class_list, list) {
991 if (__bcd->minor == bsg_minor_idx) {
992 bsg_minor_idx++;
993 if (bsg_minor_idx == BSG_MAX_DEVS)
994 bsg_minor_idx = 0;
995 goto retry;
996 }
997 }
998
999 bcd->minor = bsg_minor_idx++;
1000 if (bsg_minor_idx == BSG_MAX_DEVS)
1001 bsg_minor_idx = 0;
1002
FUJITA Tomonorid351af02007-07-09 12:40:35 +02001003 bcd->queue = q;
FUJITA Tomonori292b7f22007-03-28 13:29:58 +02001004 dev = MKDEV(BSG_MAJOR, bcd->minor);
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001005 class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", name);
1006 if (IS_ERR(class_dev)) {
1007 ret = PTR_ERR(class_dev);
Jens Axboe264a0472007-01-23 16:30:17 +01001008 goto err;
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001009 }
1010 bcd->class_dev = class_dev;
1011
1012 if (q->kobj.dentry) {
1013 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1014 if (ret)
1015 goto err;
1016 }
1017
Jens Axboe3d6392c2007-07-09 12:38:05 +02001018 list_add_tail(&bcd->list, &bsg_class_list);
FUJITA Tomonori292b7f22007-03-28 13:29:58 +02001019 bsg_device_nr++;
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001020
Jens Axboe3d6392c2007-07-09 12:38:05 +02001021 mutex_unlock(&bsg_mutex);
1022 return 0;
Jens Axboe264a0472007-01-23 16:30:17 +01001023err:
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001024 if (class_dev)
Jens Axboe264a0472007-01-23 16:30:17 +01001025 class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor));
1026 mutex_unlock(&bsg_mutex);
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001027 return ret;
Jens Axboe3d6392c2007-07-09 12:38:05 +02001028}
FUJITA Tomonori4cf07232007-03-30 11:19:39 +02001029EXPORT_SYMBOL_GPL(bsg_register_queue);
Jens Axboe3d6392c2007-07-09 12:38:05 +02001030
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001031static int bsg_add(struct class_device *cl_dev, struct class_interface *cl_intf)
1032{
1033 int ret;
1034 struct scsi_device *sdp = to_scsi_device(cl_dev->dev);
1035 struct request_queue *rq = sdp->request_queue;
1036
1037 if (rq->kobj.parent)
1038 ret = bsg_register_queue(rq, kobject_name(rq->kobj.parent));
1039 else
1040 ret = bsg_register_queue(rq, kobject_name(&sdp->sdev_gendev.kobj));
1041 return ret;
1042}
1043
1044static void bsg_remove(struct class_device *cl_dev, struct class_interface *cl_intf)
1045{
1046 bsg_unregister_queue(to_scsi_device(cl_dev->dev)->request_queue);
1047}
1048
1049static struct class_interface bsg_intf = {
1050 .add = bsg_add,
1051 .remove = bsg_remove,
1052};
1053
FUJITA Tomonori292b7f22007-03-28 13:29:58 +02001054static struct cdev bsg_cdev = {
1055 .kobj = {.name = "bsg", },
1056 .owner = THIS_MODULE,
1057};
1058
Jens Axboe3d6392c2007-07-09 12:38:05 +02001059static int __init bsg_init(void)
1060{
1061 int ret, i;
1062
Jens Axboe5309cb32007-01-23 16:24:41 +01001063 bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1064 sizeof(struct bsg_command), 0, 0, NULL, NULL);
1065 if (!bsg_cmd_cachep) {
1066 printk(KERN_ERR "bsg: failed creating slab cache\n");
1067 return -ENOMEM;
1068 }
1069
Jens Axboe3d6392c2007-07-09 12:38:05 +02001070 for (i = 0; i < BSG_LIST_SIZE; i++)
1071 INIT_HLIST_HEAD(&bsg_device_list[i]);
1072
1073 bsg_class = class_create(THIS_MODULE, "bsg");
Jens Axboe5309cb32007-01-23 16:24:41 +01001074 if (IS_ERR(bsg_class)) {
1075 kmem_cache_destroy(bsg_cmd_cachep);
Jens Axboe3d6392c2007-07-09 12:38:05 +02001076 return PTR_ERR(bsg_class);
Jens Axboe5309cb32007-01-23 16:24:41 +01001077 }
Jens Axboe3d6392c2007-07-09 12:38:05 +02001078
FUJITA Tomonori292b7f22007-03-28 13:29:58 +02001079 ret = register_chrdev_region(MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS, "bsg");
Jens Axboe3d6392c2007-07-09 12:38:05 +02001080 if (ret) {
Jens Axboe5309cb32007-01-23 16:24:41 +01001081 kmem_cache_destroy(bsg_cmd_cachep);
Jens Axboe3d6392c2007-07-09 12:38:05 +02001082 class_destroy(bsg_class);
1083 return ret;
1084 }
1085
FUJITA Tomonori292b7f22007-03-28 13:29:58 +02001086 cdev_init(&bsg_cdev, &bsg_fops);
1087 ret = cdev_add(&bsg_cdev, MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS);
1088 if (ret) {
1089 kmem_cache_destroy(bsg_cmd_cachep);
1090 class_destroy(bsg_class);
1091 unregister_chrdev_region(MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS);
1092 return ret;
1093 }
1094
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001095 ret = scsi_register_interface(&bsg_intf);
1096 if (ret) {
1097 printk(KERN_ERR "bsg: failed register scsi interface %d\n", ret);
1098 kmem_cache_destroy(bsg_cmd_cachep);
1099 class_destroy(bsg_class);
1100 unregister_chrdev(BSG_MAJOR, "bsg");
1101 return ret;
1102 }
1103
Jens Axboe3d6392c2007-07-09 12:38:05 +02001104 printk(KERN_INFO "%s loaded\n", bsg_version);
1105 return 0;
1106}
1107
1108MODULE_AUTHOR("Jens Axboe");
1109MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver");
1110MODULE_LICENSE("GPL");
1111
FUJITA Tomonori4e2872d2007-03-28 13:29:24 +02001112device_initcall(bsg_init);