]>
Commit | Line | Data |
---|---|---|
3d6392cf JA |
1 | /* |
2 | * bsg.c - block layer implementation of the sg v3 interface | |
3 | * | |
4 | * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs | |
5 | * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com> | |
6 | * | |
7 | * This file is subject to the terms and conditions of the GNU General Public | |
8 | * License version 2. See the file "COPYING" in the main directory of this | |
9 | * archive for more details. | |
10 | * | |
11 | */ | |
12 | /* | |
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 | */ | |
3d6392cf JA |
19 | #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> | |
32 | #include <scsi/sg.h> | |
33 | ||
34 | static char bsg_version[] = "block layer sg (bsg) 0.4"; | |
35 | ||
3d6392cf JA |
36 | struct bsg_device { |
37 | struct gendisk *disk; | |
38 | request_queue_t *queue; | |
39 | spinlock_t lock; | |
40 | struct list_head busy_list; | |
41 | struct list_head done_list; | |
42 | struct hlist_node dev_list; | |
43 | atomic_t ref_count; | |
44 | int minor; | |
45 | int queued_cmds; | |
46 | int done_cmds; | |
3d6392cf JA |
47 | wait_queue_head_t wq_done; |
48 | wait_queue_head_t wq_free; | |
49 | char name[BDEVNAME_SIZE]; | |
50 | int max_queue; | |
51 | unsigned long flags; | |
52 | }; | |
53 | ||
54 | enum { | |
55 | BSG_F_BLOCK = 1, | |
56 | BSG_F_WRITE_PERM = 2, | |
57 | }; | |
58 | ||
5309cb38 | 59 | #define BSG_DEFAULT_CMDS 64 |
3d6392cf JA |
60 | |
61 | #undef BSG_DEBUG | |
62 | ||
63 | #ifdef BSG_DEBUG | |
64 | #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args) | |
65 | #else | |
66 | #define dprintk(fmt, args...) | |
67 | #endif | |
68 | ||
69 | #define list_entry_bc(entry) list_entry((entry), struct bsg_command, list) | |
70 | ||
71 | /* | |
72 | * just for testing | |
73 | */ | |
74 | #define BSG_MAJOR (240) | |
75 | ||
76 | static DEFINE_MUTEX(bsg_mutex); | |
77 | static int bsg_device_nr; | |
78 | ||
79 | #define BSG_LIST_SIZE (8) | |
80 | #define bsg_list_idx(minor) ((minor) & (BSG_LIST_SIZE - 1)) | |
81 | static struct hlist_head bsg_device_list[BSG_LIST_SIZE]; | |
82 | ||
83 | static struct class *bsg_class; | |
84 | static LIST_HEAD(bsg_class_list); | |
85 | ||
5309cb38 JA |
86 | static struct kmem_cache *bsg_cmd_cachep; |
87 | ||
3d6392cf JA |
88 | /* |
89 | * our internal command type | |
90 | */ | |
91 | struct bsg_command { | |
92 | struct bsg_device *bd; | |
93 | struct list_head list; | |
94 | struct request *rq; | |
95 | struct bio *bio; | |
96 | int err; | |
70e36ece FT |
97 | struct sg_io_v4 hdr; |
98 | struct sg_io_v4 __user *uhdr; | |
3d6392cf JA |
99 | char sense[SCSI_SENSE_BUFFERSIZE]; |
100 | }; | |
101 | ||
102 | static void bsg_free_command(struct bsg_command *bc) | |
103 | { | |
104 | struct bsg_device *bd = bc->bd; | |
3d6392cf JA |
105 | unsigned long flags; |
106 | ||
5309cb38 | 107 | kmem_cache_free(bsg_cmd_cachep, bc); |
3d6392cf JA |
108 | |
109 | spin_lock_irqsave(&bd->lock, flags); | |
110 | bd->queued_cmds--; | |
3d6392cf JA |
111 | spin_unlock_irqrestore(&bd->lock, flags); |
112 | ||
113 | wake_up(&bd->wq_free); | |
114 | } | |
115 | ||
116 | static struct bsg_command *__bsg_alloc_command(struct bsg_device *bd) | |
117 | { | |
118 | struct bsg_command *bc = NULL; | |
3d6392cf JA |
119 | |
120 | spin_lock_irq(&bd->lock); | |
121 | ||
122 | if (bd->queued_cmds >= bd->max_queue) | |
123 | goto out; | |
124 | ||
3d6392cf | 125 | bd->queued_cmds++; |
3d6392cf JA |
126 | spin_unlock_irq(&bd->lock); |
127 | ||
5309cb38 JA |
128 | bc = kmem_cache_alloc(bsg_cmd_cachep, GFP_USER); |
129 | if (unlikely(!bc)) { | |
130 | spin_lock_irq(&bd->lock); | |
7e75d730 FT |
131 | bd->queued_cmds--; |
132 | goto out; | |
5309cb38 JA |
133 | } |
134 | ||
3d6392cf JA |
135 | memset(bc, 0, sizeof(*bc)); |
136 | bc->bd = bd; | |
137 | INIT_LIST_HEAD(&bc->list); | |
5309cb38 | 138 | dprintk("%s: returning free cmd %p\n", bd->name, bc); |
3d6392cf JA |
139 | return bc; |
140 | out: | |
3d6392cf JA |
141 | spin_unlock_irq(&bd->lock); |
142 | return bc; | |
143 | } | |
144 | ||
145 | static inline void | |
146 | bsg_del_done_cmd(struct bsg_device *bd, struct bsg_command *bc) | |
147 | { | |
148 | bd->done_cmds--; | |
149 | list_del(&bc->list); | |
150 | } | |
151 | ||
152 | static inline void | |
153 | bsg_add_done_cmd(struct bsg_device *bd, struct bsg_command *bc) | |
154 | { | |
155 | bd->done_cmds++; | |
156 | list_add_tail(&bc->list, &bd->done_list); | |
157 | wake_up(&bd->wq_done); | |
158 | } | |
159 | ||
160 | static inline int bsg_io_schedule(struct bsg_device *bd, int state) | |
161 | { | |
162 | DEFINE_WAIT(wait); | |
163 | int ret = 0; | |
164 | ||
165 | spin_lock_irq(&bd->lock); | |
166 | ||
167 | BUG_ON(bd->done_cmds > bd->queued_cmds); | |
168 | ||
169 | /* | |
170 | * -ENOSPC or -ENODATA? I'm going for -ENODATA, meaning "I have no | |
171 | * work to do", even though we return -ENOSPC after this same test | |
172 | * during bsg_write() -- there, it means our buffer can't have more | |
173 | * bsg_commands added to it, thus has no space left. | |
174 | */ | |
175 | if (bd->done_cmds == bd->queued_cmds) { | |
176 | ret = -ENODATA; | |
177 | goto unlock; | |
178 | } | |
179 | ||
180 | if (!test_bit(BSG_F_BLOCK, &bd->flags)) { | |
181 | ret = -EAGAIN; | |
182 | goto unlock; | |
183 | } | |
184 | ||
185 | prepare_to_wait(&bd->wq_done, &wait, state); | |
186 | spin_unlock_irq(&bd->lock); | |
187 | io_schedule(); | |
188 | finish_wait(&bd->wq_done, &wait); | |
189 | ||
190 | if ((state == TASK_INTERRUPTIBLE) && signal_pending(current)) | |
191 | ret = -ERESTARTSYS; | |
192 | ||
193 | return ret; | |
194 | unlock: | |
195 | spin_unlock_irq(&bd->lock); | |
196 | return ret; | |
197 | } | |
198 | ||
199 | /* | |
200 | * get a new free command, blocking if needed and specified | |
201 | */ | |
202 | static struct bsg_command *bsg_get_command(struct bsg_device *bd) | |
203 | { | |
204 | struct bsg_command *bc; | |
205 | int ret; | |
206 | ||
207 | do { | |
208 | bc = __bsg_alloc_command(bd); | |
209 | if (bc) | |
210 | break; | |
211 | ||
212 | ret = bsg_io_schedule(bd, TASK_INTERRUPTIBLE); | |
213 | if (ret) { | |
214 | bc = ERR_PTR(ret); | |
215 | break; | |
216 | } | |
217 | ||
218 | } while (1); | |
219 | ||
220 | return bc; | |
221 | } | |
222 | ||
70e36ece FT |
223 | static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq, |
224 | struct sg_io_v4 *hdr, int has_write_perm) | |
225 | { | |
226 | memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */ | |
227 | ||
228 | if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request, | |
229 | hdr->request_len)) | |
230 | return -EFAULT; | |
231 | if (blk_verify_command(rq->cmd, has_write_perm)) | |
232 | return -EPERM; | |
233 | ||
234 | /* | |
235 | * fill in request structure | |
236 | */ | |
237 | rq->cmd_len = hdr->request_len; | |
238 | rq->cmd_type = REQ_TYPE_BLOCK_PC; | |
239 | ||
240 | rq->timeout = (hdr->timeout * HZ) / 1000; | |
241 | if (!rq->timeout) | |
242 | rq->timeout = q->sg_timeout; | |
243 | if (!rq->timeout) | |
244 | rq->timeout = BLK_DEFAULT_SG_TIMEOUT; | |
245 | ||
246 | return 0; | |
247 | } | |
248 | ||
3d6392cf | 249 | /* |
70e36ece | 250 | * Check if sg_io_v4 from user is allowed and valid |
3d6392cf JA |
251 | */ |
252 | static int | |
70e36ece | 253 | bsg_validate_sgv4_hdr(request_queue_t *q, struct sg_io_v4 *hdr, int *rw) |
3d6392cf | 254 | { |
70e36ece | 255 | if (hdr->guard != 'Q') |
3d6392cf | 256 | return -EINVAL; |
70e36ece | 257 | if (hdr->request_len > BLK_MAX_CDB) |
3d6392cf | 258 | return -EINVAL; |
70e36ece FT |
259 | if (hdr->dout_xfer_len > (q->max_sectors << 9) || |
260 | hdr->din_xfer_len > (q->max_sectors << 9)) | |
3d6392cf JA |
261 | return -EIO; |
262 | ||
70e36ece FT |
263 | /* not supported currently */ |
264 | if (hdr->protocol || hdr->subprotocol) | |
265 | return -EINVAL; | |
266 | ||
3d6392cf JA |
267 | /* |
268 | * looks sane, if no data then it should be fine from our POV | |
269 | */ | |
70e36ece | 270 | if (!hdr->dout_xfer_len && !hdr->din_xfer_len) |
3d6392cf JA |
271 | return 0; |
272 | ||
70e36ece FT |
273 | /* not supported currently */ |
274 | if (hdr->dout_xfer_len && hdr->din_xfer_len) | |
275 | return -EINVAL; | |
276 | ||
277 | *rw = hdr->dout_xfer_len ? WRITE : READ; | |
3d6392cf JA |
278 | |
279 | return 0; | |
280 | } | |
281 | ||
282 | /* | |
70e36ece | 283 | * map sg_io_v4 to a request. |
3d6392cf JA |
284 | */ |
285 | static struct request * | |
70e36ece | 286 | bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr) |
3d6392cf JA |
287 | { |
288 | request_queue_t *q = bd->queue; | |
3d6392cf | 289 | struct request *rq; |
2ef7086a | 290 | int ret, rw = 0; /* shut up gcc */ |
70e36ece FT |
291 | unsigned int dxfer_len; |
292 | void *dxferp = NULL; | |
3d6392cf | 293 | |
70e36ece FT |
294 | dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp, |
295 | hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp, | |
296 | hdr->din_xfer_len); | |
3d6392cf | 297 | |
70e36ece | 298 | ret = bsg_validate_sgv4_hdr(q, hdr, &rw); |
3d6392cf JA |
299 | if (ret) |
300 | return ERR_PTR(ret); | |
301 | ||
302 | /* | |
303 | * map scatter-gather elements seperately and string them to request | |
304 | */ | |
305 | rq = blk_get_request(q, rw, GFP_KERNEL); | |
70e36ece FT |
306 | ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM, |
307 | &bd->flags)); | |
3d6392cf JA |
308 | if (ret) { |
309 | blk_put_request(rq); | |
310 | return ERR_PTR(ret); | |
311 | } | |
312 | ||
70e36ece FT |
313 | if (hdr->dout_xfer_len) { |
314 | dxfer_len = hdr->dout_xfer_len; | |
315 | dxferp = (void*)(unsigned long)hdr->dout_xferp; | |
316 | } else if (hdr->din_xfer_len) { | |
317 | dxfer_len = hdr->din_xfer_len; | |
318 | dxferp = (void*)(unsigned long)hdr->din_xferp; | |
319 | } else | |
320 | dxfer_len = 0; | |
321 | ||
322 | if (dxfer_len) { | |
323 | ret = blk_rq_map_user(q, rq, dxferp, dxfer_len); | |
324 | if (ret) { | |
325 | dprintk("failed map at %d\n", ret); | |
326 | blk_put_request(rq); | |
327 | rq = ERR_PTR(ret); | |
3d6392cf | 328 | } |
3d6392cf JA |
329 | } |
330 | ||
331 | return rq; | |
332 | } | |
333 | ||
334 | /* | |
335 | * async completion call-back from the block layer, when scsi/ide/whatever | |
336 | * calls end_that_request_last() on a request | |
337 | */ | |
338 | static void bsg_rq_end_io(struct request *rq, int uptodate) | |
339 | { | |
340 | struct bsg_command *bc = rq->end_io_data; | |
341 | struct bsg_device *bd = bc->bd; | |
342 | unsigned long flags; | |
343 | ||
5309cb38 JA |
344 | dprintk("%s: finished rq %p bc %p, bio %p stat %d\n", |
345 | bd->name, rq, bc, bc->bio, uptodate); | |
3d6392cf JA |
346 | |
347 | bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration); | |
348 | ||
349 | spin_lock_irqsave(&bd->lock, flags); | |
350 | list_del(&bc->list); | |
351 | bsg_add_done_cmd(bd, bc); | |
352 | spin_unlock_irqrestore(&bd->lock, flags); | |
353 | } | |
354 | ||
355 | /* | |
356 | * do final setup of a 'bc' and submit the matching 'rq' to the block | |
357 | * layer for io | |
358 | */ | |
359 | static void bsg_add_command(struct bsg_device *bd, request_queue_t *q, | |
360 | struct bsg_command *bc, struct request *rq) | |
361 | { | |
362 | rq->sense = bc->sense; | |
363 | rq->sense_len = 0; | |
364 | ||
365 | /* | |
366 | * add bc command to busy queue and submit rq for io | |
367 | */ | |
368 | bc->rq = rq; | |
369 | bc->bio = rq->bio; | |
370 | bc->hdr.duration = jiffies; | |
371 | spin_lock_irq(&bd->lock); | |
372 | list_add_tail(&bc->list, &bd->busy_list); | |
373 | spin_unlock_irq(&bd->lock); | |
374 | ||
375 | dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc); | |
376 | ||
377 | rq->end_io_data = bc; | |
378 | blk_execute_rq_nowait(q, bd->disk, rq, 1, bsg_rq_end_io); | |
379 | } | |
380 | ||
381 | static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd) | |
382 | { | |
383 | struct bsg_command *bc = NULL; | |
384 | ||
385 | spin_lock_irq(&bd->lock); | |
386 | if (bd->done_cmds) { | |
387 | bc = list_entry_bc(bd->done_list.next); | |
388 | bsg_del_done_cmd(bd, bc); | |
389 | } | |
390 | spin_unlock_irq(&bd->lock); | |
391 | ||
392 | return bc; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Get a finished command from the done list | |
397 | */ | |
398 | static struct bsg_command *__bsg_get_done_cmd(struct bsg_device *bd, int state) | |
399 | { | |
400 | struct bsg_command *bc; | |
401 | int ret; | |
402 | ||
403 | do { | |
404 | bc = bsg_next_done_cmd(bd); | |
405 | if (bc) | |
406 | break; | |
407 | ||
408 | ret = bsg_io_schedule(bd, state); | |
409 | if (ret) { | |
410 | bc = ERR_PTR(ret); | |
411 | break; | |
412 | } | |
413 | } while (1); | |
414 | ||
415 | dprintk("%s: returning done %p\n", bd->name, bc); | |
416 | ||
417 | return bc; | |
418 | } | |
419 | ||
420 | static struct bsg_command * | |
421 | bsg_get_done_cmd(struct bsg_device *bd, const struct iovec *iov) | |
422 | { | |
423 | return __bsg_get_done_cmd(bd, TASK_INTERRUPTIBLE); | |
424 | } | |
425 | ||
426 | static struct bsg_command * | |
427 | bsg_get_done_cmd_nosignals(struct bsg_device *bd) | |
428 | { | |
429 | return __bsg_get_done_cmd(bd, TASK_UNINTERRUPTIBLE); | |
430 | } | |
431 | ||
70e36ece FT |
432 | static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr, |
433 | struct bio *bio) | |
434 | { | |
435 | int ret = 0; | |
436 | ||
437 | dprintk("rq %p bio %p %u\n", rq, bio, rq->errors); | |
438 | /* | |
439 | * fill in all the output members | |
440 | */ | |
441 | hdr->device_status = status_byte(rq->errors); | |
442 | hdr->transport_status = host_byte(rq->errors); | |
443 | hdr->driver_status = driver_byte(rq->errors); | |
444 | hdr->info = 0; | |
445 | if (hdr->device_status || hdr->transport_status || hdr->driver_status) | |
446 | hdr->info |= SG_INFO_CHECK; | |
447 | hdr->din_resid = rq->data_len; | |
448 | hdr->response_len = 0; | |
449 | ||
450 | if (rq->sense_len && hdr->response) { | |
451 | int len = min((unsigned int) hdr->max_response_len, | |
452 | rq->sense_len); | |
453 | ||
454 | ret = copy_to_user((void*)(unsigned long)hdr->response, | |
455 | rq->sense, len); | |
456 | if (!ret) | |
457 | hdr->response_len = len; | |
458 | else | |
459 | ret = -EFAULT; | |
460 | } | |
461 | ||
462 | blk_rq_unmap_user(bio); | |
463 | blk_put_request(rq); | |
464 | ||
465 | return ret; | |
466 | } | |
467 | ||
3d6392cf JA |
468 | static int bsg_complete_all_commands(struct bsg_device *bd) |
469 | { | |
470 | struct bsg_command *bc; | |
471 | int ret, tret; | |
472 | ||
473 | dprintk("%s: entered\n", bd->name); | |
474 | ||
475 | set_bit(BSG_F_BLOCK, &bd->flags); | |
476 | ||
477 | /* | |
478 | * wait for all commands to complete | |
479 | */ | |
480 | ret = 0; | |
481 | do { | |
482 | ret = bsg_io_schedule(bd, TASK_UNINTERRUPTIBLE); | |
483 | /* | |
484 | * look for -ENODATA specifically -- we'll sometimes get | |
485 | * -ERESTARTSYS when we've taken a signal, but we can't | |
486 | * return until we're done freeing the queue, so ignore | |
487 | * it. The signal will get handled when we're done freeing | |
488 | * the bsg_device. | |
489 | */ | |
490 | } while (ret != -ENODATA); | |
491 | ||
492 | /* | |
493 | * discard done commands | |
494 | */ | |
495 | ret = 0; | |
496 | do { | |
497 | bc = bsg_get_done_cmd_nosignals(bd); | |
498 | ||
499 | /* | |
500 | * we _must_ complete before restarting, because | |
501 | * bsg_release can't handle this failing. | |
502 | */ | |
503 | if (PTR_ERR(bc) == -ERESTARTSYS) | |
504 | continue; | |
505 | if (IS_ERR(bc)) { | |
506 | ret = PTR_ERR(bc); | |
507 | break; | |
508 | } | |
509 | ||
70e36ece | 510 | tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio); |
3d6392cf JA |
511 | if (!ret) |
512 | ret = tret; | |
513 | ||
514 | bsg_free_command(bc); | |
515 | } while (1); | |
516 | ||
517 | return ret; | |
518 | } | |
519 | ||
520 | typedef struct bsg_command *(*bsg_command_callback)(struct bsg_device *bd, const struct iovec *iov); | |
521 | ||
522 | static ssize_t | |
523 | __bsg_read(char __user *buf, size_t count, bsg_command_callback get_bc, | |
524 | struct bsg_device *bd, const struct iovec *iov, ssize_t *bytes_read) | |
525 | { | |
526 | struct bsg_command *bc; | |
527 | int nr_commands, ret; | |
528 | ||
70e36ece | 529 | if (count % sizeof(struct sg_io_v4)) |
3d6392cf JA |
530 | return -EINVAL; |
531 | ||
532 | ret = 0; | |
70e36ece | 533 | nr_commands = count / sizeof(struct sg_io_v4); |
3d6392cf JA |
534 | while (nr_commands) { |
535 | bc = get_bc(bd, iov); | |
536 | if (IS_ERR(bc)) { | |
537 | ret = PTR_ERR(bc); | |
538 | break; | |
539 | } | |
540 | ||
541 | /* | |
542 | * this is the only case where we need to copy data back | |
543 | * after completing the request. so do that here, | |
544 | * bsg_complete_work() cannot do that for us | |
545 | */ | |
70e36ece | 546 | ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio); |
3d6392cf JA |
547 | |
548 | if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr))) | |
549 | ret = -EFAULT; | |
550 | ||
551 | bsg_free_command(bc); | |
552 | ||
553 | if (ret) | |
554 | break; | |
555 | ||
70e36ece FT |
556 | buf += sizeof(struct sg_io_v4); |
557 | *bytes_read += sizeof(struct sg_io_v4); | |
3d6392cf JA |
558 | nr_commands--; |
559 | } | |
560 | ||
561 | return ret; | |
562 | } | |
563 | ||
564 | static inline void bsg_set_block(struct bsg_device *bd, struct file *file) | |
565 | { | |
566 | if (file->f_flags & O_NONBLOCK) | |
567 | clear_bit(BSG_F_BLOCK, &bd->flags); | |
568 | else | |
569 | set_bit(BSG_F_BLOCK, &bd->flags); | |
570 | } | |
571 | ||
572 | static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file) | |
573 | { | |
574 | if (file->f_mode & FMODE_WRITE) | |
575 | set_bit(BSG_F_WRITE_PERM, &bd->flags); | |
576 | else | |
577 | clear_bit(BSG_F_WRITE_PERM, &bd->flags); | |
578 | } | |
579 | ||
580 | static inline int err_block_err(int ret) | |
581 | { | |
582 | if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN) | |
583 | return 1; | |
584 | ||
585 | return 0; | |
586 | } | |
587 | ||
588 | static ssize_t | |
589 | bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
590 | { | |
591 | struct bsg_device *bd = file->private_data; | |
592 | int ret; | |
593 | ssize_t bytes_read; | |
594 | ||
9e69fbb5 | 595 | dprintk("%s: read %Zd bytes\n", bd->name, count); |
3d6392cf JA |
596 | |
597 | bsg_set_block(bd, file); | |
598 | bytes_read = 0; | |
599 | ret = __bsg_read(buf, count, bsg_get_done_cmd, | |
600 | bd, NULL, &bytes_read); | |
601 | *ppos = bytes_read; | |
602 | ||
603 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
604 | bytes_read = ret; | |
605 | ||
606 | return bytes_read; | |
607 | } | |
608 | ||
609 | static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf, | |
610 | size_t count, ssize_t *bytes_read) | |
611 | { | |
612 | struct bsg_command *bc; | |
613 | struct request *rq; | |
614 | int ret, nr_commands; | |
615 | ||
70e36ece | 616 | if (count % sizeof(struct sg_io_v4)) |
3d6392cf JA |
617 | return -EINVAL; |
618 | ||
70e36ece | 619 | nr_commands = count / sizeof(struct sg_io_v4); |
3d6392cf JA |
620 | rq = NULL; |
621 | bc = NULL; | |
622 | ret = 0; | |
623 | while (nr_commands) { | |
624 | request_queue_t *q = bd->queue; | |
3d6392cf JA |
625 | |
626 | bc = bsg_get_command(bd); | |
627 | if (!bc) | |
628 | break; | |
629 | if (IS_ERR(bc)) { | |
630 | ret = PTR_ERR(bc); | |
631 | bc = NULL; | |
632 | break; | |
633 | } | |
634 | ||
70e36ece | 635 | bc->uhdr = (struct sg_io_v4 __user *) buf; |
3d6392cf JA |
636 | if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) { |
637 | ret = -EFAULT; | |
638 | break; | |
639 | } | |
640 | ||
641 | /* | |
642 | * get a request, fill in the blanks, and add to request queue | |
643 | */ | |
70e36ece | 644 | rq = bsg_map_hdr(bd, &bc->hdr); |
3d6392cf JA |
645 | if (IS_ERR(rq)) { |
646 | ret = PTR_ERR(rq); | |
647 | rq = NULL; | |
648 | break; | |
649 | } | |
650 | ||
651 | bsg_add_command(bd, q, bc, rq); | |
652 | bc = NULL; | |
653 | rq = NULL; | |
654 | nr_commands--; | |
70e36ece FT |
655 | buf += sizeof(struct sg_io_v4); |
656 | *bytes_read += sizeof(struct sg_io_v4); | |
3d6392cf JA |
657 | } |
658 | ||
3d6392cf JA |
659 | if (bc) |
660 | bsg_free_command(bc); | |
661 | ||
662 | return ret; | |
663 | } | |
664 | ||
665 | static ssize_t | |
666 | bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) | |
667 | { | |
668 | struct bsg_device *bd = file->private_data; | |
669 | ssize_t bytes_read; | |
670 | int ret; | |
671 | ||
9e69fbb5 | 672 | dprintk("%s: write %Zd bytes\n", bd->name, count); |
3d6392cf JA |
673 | |
674 | bsg_set_block(bd, file); | |
675 | bsg_set_write_perm(bd, file); | |
676 | ||
677 | bytes_read = 0; | |
678 | ret = __bsg_write(bd, buf, count, &bytes_read); | |
679 | *ppos = bytes_read; | |
680 | ||
681 | /* | |
682 | * return bytes written on non-fatal errors | |
683 | */ | |
684 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
685 | bytes_read = ret; | |
686 | ||
9e69fbb5 | 687 | dprintk("%s: returning %Zd\n", bd->name, bytes_read); |
3d6392cf JA |
688 | return bytes_read; |
689 | } | |
690 | ||
3d6392cf JA |
691 | static struct bsg_device *bsg_alloc_device(void) |
692 | { | |
3d6392cf | 693 | struct bsg_device *bd; |
3d6392cf JA |
694 | |
695 | bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL); | |
696 | if (unlikely(!bd)) | |
697 | return NULL; | |
698 | ||
699 | spin_lock_init(&bd->lock); | |
700 | ||
5309cb38 | 701 | bd->max_queue = BSG_DEFAULT_CMDS; |
3d6392cf JA |
702 | |
703 | INIT_LIST_HEAD(&bd->busy_list); | |
704 | INIT_LIST_HEAD(&bd->done_list); | |
705 | INIT_HLIST_NODE(&bd->dev_list); | |
706 | ||
707 | init_waitqueue_head(&bd->wq_free); | |
708 | init_waitqueue_head(&bd->wq_done); | |
709 | return bd; | |
3d6392cf JA |
710 | } |
711 | ||
712 | static int bsg_put_device(struct bsg_device *bd) | |
713 | { | |
714 | int ret = 0; | |
715 | ||
716 | mutex_lock(&bsg_mutex); | |
717 | ||
718 | if (!atomic_dec_and_test(&bd->ref_count)) | |
719 | goto out; | |
720 | ||
721 | dprintk("%s: tearing down\n", bd->name); | |
722 | ||
723 | /* | |
724 | * close can always block | |
725 | */ | |
726 | set_bit(BSG_F_BLOCK, &bd->flags); | |
727 | ||
728 | /* | |
729 | * correct error detection baddies here again. it's the responsibility | |
730 | * of the app to properly reap commands before close() if it wants | |
731 | * fool-proof error detection | |
732 | */ | |
733 | ret = bsg_complete_all_commands(bd); | |
734 | ||
735 | blk_put_queue(bd->queue); | |
736 | hlist_del(&bd->dev_list); | |
5309cb38 | 737 | kfree(bd); |
3d6392cf JA |
738 | out: |
739 | mutex_unlock(&bsg_mutex); | |
740 | return ret; | |
741 | } | |
742 | ||
743 | static struct bsg_device *bsg_add_device(struct inode *inode, | |
744 | struct gendisk *disk, | |
745 | struct file *file) | |
746 | { | |
747 | struct bsg_device *bd = NULL; | |
748 | #ifdef BSG_DEBUG | |
749 | unsigned char buf[32]; | |
750 | #endif | |
751 | ||
752 | bd = bsg_alloc_device(); | |
753 | if (!bd) | |
754 | return ERR_PTR(-ENOMEM); | |
755 | ||
756 | bd->disk = disk; | |
757 | bd->queue = disk->queue; | |
758 | kobject_get(&disk->queue->kobj); | |
759 | bsg_set_block(bd, file); | |
760 | ||
761 | atomic_set(&bd->ref_count, 1); | |
762 | bd->minor = iminor(inode); | |
763 | mutex_lock(&bsg_mutex); | |
764 | hlist_add_head(&bd->dev_list,&bsg_device_list[bsg_list_idx(bd->minor)]); | |
765 | ||
766 | strncpy(bd->name, disk->disk_name, sizeof(bd->name) - 1); | |
767 | dprintk("bound to <%s>, max queue %d\n", | |
9e69fbb5 | 768 | format_dev_t(buf, inode->i_rdev), bd->max_queue); |
3d6392cf JA |
769 | |
770 | mutex_unlock(&bsg_mutex); | |
771 | return bd; | |
772 | } | |
773 | ||
774 | static struct bsg_device *__bsg_get_device(int minor) | |
775 | { | |
776 | struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)]; | |
777 | struct bsg_device *bd = NULL; | |
778 | struct hlist_node *entry; | |
779 | ||
780 | mutex_lock(&bsg_mutex); | |
781 | ||
782 | hlist_for_each(entry, list) { | |
783 | bd = hlist_entry(entry, struct bsg_device, dev_list); | |
784 | if (bd->minor == minor) { | |
785 | atomic_inc(&bd->ref_count); | |
786 | break; | |
787 | } | |
788 | ||
789 | bd = NULL; | |
790 | } | |
791 | ||
792 | mutex_unlock(&bsg_mutex); | |
793 | return bd; | |
794 | } | |
795 | ||
796 | static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file) | |
797 | { | |
798 | struct bsg_device *bd = __bsg_get_device(iminor(inode)); | |
799 | struct bsg_class_device *bcd, *__bcd; | |
800 | ||
801 | if (bd) | |
802 | return bd; | |
803 | ||
804 | /* | |
805 | * find the class device | |
806 | */ | |
807 | bcd = NULL; | |
808 | mutex_lock(&bsg_mutex); | |
809 | list_for_each_entry(__bcd, &bsg_class_list, list) { | |
810 | if (__bcd->minor == iminor(inode)) { | |
811 | bcd = __bcd; | |
812 | break; | |
813 | } | |
814 | } | |
815 | mutex_unlock(&bsg_mutex); | |
816 | ||
817 | if (!bcd) | |
818 | return ERR_PTR(-ENODEV); | |
819 | ||
820 | return bsg_add_device(inode, bcd->disk, file); | |
821 | } | |
822 | ||
823 | static int bsg_open(struct inode *inode, struct file *file) | |
824 | { | |
825 | struct bsg_device *bd = bsg_get_device(inode, file); | |
826 | ||
827 | if (IS_ERR(bd)) | |
828 | return PTR_ERR(bd); | |
829 | ||
830 | file->private_data = bd; | |
831 | return 0; | |
832 | } | |
833 | ||
834 | static int bsg_release(struct inode *inode, struct file *file) | |
835 | { | |
836 | struct bsg_device *bd = file->private_data; | |
837 | ||
838 | file->private_data = NULL; | |
839 | return bsg_put_device(bd); | |
840 | } | |
841 | ||
842 | static unsigned int bsg_poll(struct file *file, poll_table *wait) | |
843 | { | |
844 | struct bsg_device *bd = file->private_data; | |
845 | unsigned int mask = 0; | |
846 | ||
847 | poll_wait(file, &bd->wq_done, wait); | |
848 | poll_wait(file, &bd->wq_free, wait); | |
849 | ||
850 | spin_lock_irq(&bd->lock); | |
851 | if (!list_empty(&bd->done_list)) | |
852 | mask |= POLLIN | POLLRDNORM; | |
853 | if (bd->queued_cmds >= bd->max_queue) | |
854 | mask |= POLLOUT; | |
855 | spin_unlock_irq(&bd->lock); | |
856 | ||
857 | return mask; | |
858 | } | |
859 | ||
860 | static int | |
861 | bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd, | |
862 | unsigned long arg) | |
863 | { | |
864 | struct bsg_device *bd = file->private_data; | |
865 | int __user *uarg = (int __user *) arg; | |
866 | ||
867 | if (!bd) | |
868 | return -ENXIO; | |
869 | ||
870 | switch (cmd) { | |
871 | /* | |
872 | * our own ioctls | |
873 | */ | |
874 | case SG_GET_COMMAND_Q: | |
875 | return put_user(bd->max_queue, uarg); | |
5309cb38 | 876 | case SG_SET_COMMAND_Q: { |
3d6392cf JA |
877 | int queue; |
878 | ||
879 | if (get_user(queue, uarg)) | |
880 | return -EFAULT; | |
5309cb38 | 881 | if (queue < 1) |
3d6392cf JA |
882 | return -EINVAL; |
883 | ||
5309cb38 | 884 | spin_lock_irq(&bd->lock); |
3d6392cf | 885 | bd->max_queue = queue; |
5309cb38 | 886 | spin_unlock_irq(&bd->lock); |
3d6392cf JA |
887 | return 0; |
888 | } | |
889 | ||
890 | /* | |
891 | * SCSI/sg ioctls | |
892 | */ | |
893 | case SG_GET_VERSION_NUM: | |
894 | case SCSI_IOCTL_GET_IDLUN: | |
895 | case SCSI_IOCTL_GET_BUS_NUMBER: | |
896 | case SG_SET_TIMEOUT: | |
897 | case SG_GET_TIMEOUT: | |
898 | case SG_GET_RESERVED_SIZE: | |
899 | case SG_SET_RESERVED_SIZE: | |
900 | case SG_EMULATED_HOST: | |
3d6392cf JA |
901 | case SCSI_IOCTL_SEND_COMMAND: { |
902 | void __user *uarg = (void __user *) arg; | |
45e79a3a | 903 | return scsi_cmd_ioctl(file, bd->queue, bd->disk, cmd, uarg); |
3d6392cf | 904 | } |
10e8855b FT |
905 | case SG_IO: { |
906 | struct request *rq; | |
907 | struct bio *bio; | |
908 | struct sg_io_v4 hdr; | |
909 | ||
910 | if (copy_from_user(&hdr, uarg, sizeof(hdr))) | |
911 | return -EFAULT; | |
912 | ||
913 | rq = bsg_map_hdr(bd, &hdr); | |
914 | if (IS_ERR(rq)) | |
915 | return PTR_ERR(rq); | |
916 | ||
917 | bio = rq->bio; | |
918 | blk_execute_rq(bd->queue, bd->disk, rq, 0); | |
919 | blk_complete_sgv4_hdr_rq(rq, &hdr, bio); | |
920 | ||
921 | if (copy_to_user(uarg, &hdr, sizeof(hdr))) | |
922 | return -EFAULT; | |
b711afa6 JA |
923 | |
924 | return 0; | |
10e8855b | 925 | } |
3d6392cf JA |
926 | /* |
927 | * block device ioctls | |
928 | */ | |
929 | default: | |
930 | #if 0 | |
931 | return ioctl_by_bdev(bd->bdev, cmd, arg); | |
932 | #else | |
933 | return -ENOTTY; | |
934 | #endif | |
935 | } | |
936 | } | |
937 | ||
938 | static 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 | .ioctl = bsg_ioctl, | |
945 | .owner = THIS_MODULE, | |
946 | }; | |
947 | ||
948 | void bsg_unregister_disk(struct gendisk *disk) | |
949 | { | |
950 | struct bsg_class_device *bcd = &disk->bsg_dev; | |
951 | ||
952 | if (!bcd->class_dev) | |
953 | return; | |
954 | ||
955 | mutex_lock(&bsg_mutex); | |
956 | sysfs_remove_link(&bcd->disk->queue->kobj, "bsg"); | |
957 | class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor)); | |
958 | bcd->class_dev = NULL; | |
959 | list_del_init(&bcd->list); | |
960 | mutex_unlock(&bsg_mutex); | |
961 | } | |
962 | ||
963 | int bsg_register_disk(struct gendisk *disk) | |
964 | { | |
965 | request_queue_t *q = disk->queue; | |
966 | struct bsg_class_device *bcd; | |
967 | dev_t dev; | |
968 | ||
969 | /* | |
970 | * we need a proper transport to send commands, not a stacked device | |
971 | */ | |
972 | if (!q->request_fn) | |
973 | return 0; | |
974 | ||
975 | bcd = &disk->bsg_dev; | |
976 | memset(bcd, 0, sizeof(*bcd)); | |
977 | INIT_LIST_HEAD(&bcd->list); | |
978 | ||
979 | mutex_lock(&bsg_mutex); | |
980 | dev = MKDEV(BSG_MAJOR, bsg_device_nr); | |
981 | bcd->minor = bsg_device_nr; | |
982 | bsg_device_nr++; | |
983 | bcd->disk = disk; | |
984 | bcd->class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", disk->disk_name); | |
264a0472 JA |
985 | if (!bcd->class_dev) |
986 | goto err; | |
3d6392cf | 987 | list_add_tail(&bcd->list, &bsg_class_list); |
264a0472 JA |
988 | if (sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg")) |
989 | goto err; | |
3d6392cf JA |
990 | mutex_unlock(&bsg_mutex); |
991 | return 0; | |
264a0472 JA |
992 | err: |
993 | bsg_device_nr--; | |
994 | if (bcd->class_dev) | |
995 | class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor)); | |
996 | mutex_unlock(&bsg_mutex); | |
997 | return -ENOMEM; | |
3d6392cf JA |
998 | } |
999 | ||
1000 | static int __init bsg_init(void) | |
1001 | { | |
1002 | int ret, i; | |
1003 | ||
5309cb38 JA |
1004 | bsg_cmd_cachep = kmem_cache_create("bsg_cmd", |
1005 | sizeof(struct bsg_command), 0, 0, NULL, NULL); | |
1006 | if (!bsg_cmd_cachep) { | |
1007 | printk(KERN_ERR "bsg: failed creating slab cache\n"); | |
1008 | return -ENOMEM; | |
1009 | } | |
1010 | ||
3d6392cf JA |
1011 | for (i = 0; i < BSG_LIST_SIZE; i++) |
1012 | INIT_HLIST_HEAD(&bsg_device_list[i]); | |
1013 | ||
1014 | bsg_class = class_create(THIS_MODULE, "bsg"); | |
5309cb38 JA |
1015 | if (IS_ERR(bsg_class)) { |
1016 | kmem_cache_destroy(bsg_cmd_cachep); | |
3d6392cf | 1017 | return PTR_ERR(bsg_class); |
5309cb38 | 1018 | } |
3d6392cf JA |
1019 | |
1020 | ret = register_chrdev(BSG_MAJOR, "bsg", &bsg_fops); | |
1021 | if (ret) { | |
5309cb38 | 1022 | kmem_cache_destroy(bsg_cmd_cachep); |
3d6392cf JA |
1023 | class_destroy(bsg_class); |
1024 | return ret; | |
1025 | } | |
1026 | ||
1027 | printk(KERN_INFO "%s loaded\n", bsg_version); | |
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | MODULE_AUTHOR("Jens Axboe"); | |
1032 | MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver"); | |
1033 | MODULE_LICENSE("GPL"); | |
1034 | ||
1035 | subsys_initcall(bsg_init); |