]>
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> | |
4e2872d6 FT |
32 | #include <scsi/scsi_device.h> |
33 | #include <scsi/scsi_driver.h> | |
3d6392cf JA |
34 | #include <scsi/sg.h> |
35 | ||
36 | static char bsg_version[] = "block layer sg (bsg) 0.4"; | |
37 | ||
3d6392cf | 38 | struct bsg_device { |
3d6392cf JA |
39 | 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; | |
3d6392cf JA |
48 | wait_queue_head_t wq_done; |
49 | wait_queue_head_t wq_free; | |
d351af01 | 50 | char name[BUS_ID_SIZE]; |
3d6392cf JA |
51 | int max_queue; |
52 | unsigned long flags; | |
53 | }; | |
54 | ||
55 | enum { | |
56 | BSG_F_BLOCK = 1, | |
57 | BSG_F_WRITE_PERM = 2, | |
58 | }; | |
59 | ||
5309cb38 | 60 | #define BSG_DEFAULT_CMDS 64 |
292b7f27 | 61 | #define BSG_MAX_DEVS 32768 |
3d6392cf JA |
62 | |
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 | ||
78 | static DEFINE_MUTEX(bsg_mutex); | |
292b7f27 | 79 | static int bsg_device_nr, bsg_minor_idx; |
3d6392cf JA |
80 | |
81 | #define BSG_LIST_SIZE (8) | |
82 | #define bsg_list_idx(minor) ((minor) & (BSG_LIST_SIZE - 1)) | |
83 | static struct hlist_head bsg_device_list[BSG_LIST_SIZE]; | |
84 | ||
85 | static struct class *bsg_class; | |
86 | static LIST_HEAD(bsg_class_list); | |
87 | ||
5309cb38 JA |
88 | static struct kmem_cache *bsg_cmd_cachep; |
89 | ||
3d6392cf JA |
90 | /* |
91 | * our internal command type | |
92 | */ | |
93 | struct bsg_command { | |
94 | struct bsg_device *bd; | |
95 | struct list_head list; | |
96 | struct request *rq; | |
97 | struct bio *bio; | |
98 | int err; | |
70e36ece FT |
99 | struct sg_io_v4 hdr; |
100 | struct sg_io_v4 __user *uhdr; | |
3d6392cf JA |
101 | char sense[SCSI_SENSE_BUFFERSIZE]; |
102 | }; | |
103 | ||
104 | static void bsg_free_command(struct bsg_command *bc) | |
105 | { | |
106 | struct bsg_device *bd = bc->bd; | |
3d6392cf JA |
107 | unsigned long flags; |
108 | ||
5309cb38 | 109 | kmem_cache_free(bsg_cmd_cachep, bc); |
3d6392cf JA |
110 | |
111 | spin_lock_irqsave(&bd->lock, flags); | |
112 | bd->queued_cmds--; | |
3d6392cf JA |
113 | spin_unlock_irqrestore(&bd->lock, flags); |
114 | ||
115 | wake_up(&bd->wq_free); | |
116 | } | |
117 | ||
e7d72173 | 118 | static struct bsg_command *bsg_alloc_command(struct bsg_device *bd) |
3d6392cf | 119 | { |
e7d72173 | 120 | struct bsg_command *bc = ERR_PTR(-EINVAL); |
3d6392cf JA |
121 | |
122 | spin_lock_irq(&bd->lock); | |
123 | ||
124 | if (bd->queued_cmds >= bd->max_queue) | |
125 | goto out; | |
126 | ||
3d6392cf | 127 | bd->queued_cmds++; |
3d6392cf JA |
128 | spin_unlock_irq(&bd->lock); |
129 | ||
5309cb38 JA |
130 | bc = kmem_cache_alloc(bsg_cmd_cachep, GFP_USER); |
131 | if (unlikely(!bc)) { | |
132 | spin_lock_irq(&bd->lock); | |
7e75d730 | 133 | bd->queued_cmds--; |
e7d72173 | 134 | bc = ERR_PTR(-ENOMEM); |
7e75d730 | 135 | goto out; |
5309cb38 JA |
136 | } |
137 | ||
3d6392cf JA |
138 | memset(bc, 0, sizeof(*bc)); |
139 | bc->bd = bd; | |
140 | INIT_LIST_HEAD(&bc->list); | |
5309cb38 | 141 | dprintk("%s: returning free cmd %p\n", bd->name, bc); |
3d6392cf JA |
142 | return bc; |
143 | out: | |
3d6392cf JA |
144 | spin_unlock_irq(&bd->lock); |
145 | return bc; | |
146 | } | |
147 | ||
148 | static inline void | |
149 | bsg_del_done_cmd(struct bsg_device *bd, struct bsg_command *bc) | |
150 | { | |
151 | bd->done_cmds--; | |
152 | list_del(&bc->list); | |
153 | } | |
154 | ||
155 | static inline void | |
156 | bsg_add_done_cmd(struct bsg_device *bd, struct bsg_command *bc) | |
157 | { | |
158 | bd->done_cmds++; | |
159 | list_add_tail(&bc->list, &bd->done_list); | |
160 | wake_up(&bd->wq_done); | |
161 | } | |
162 | ||
163 | static inline int bsg_io_schedule(struct bsg_device *bd, int state) | |
164 | { | |
165 | DEFINE_WAIT(wait); | |
166 | int ret = 0; | |
167 | ||
168 | spin_lock_irq(&bd->lock); | |
169 | ||
170 | BUG_ON(bd->done_cmds > bd->queued_cmds); | |
171 | ||
172 | /* | |
173 | * -ENOSPC or -ENODATA? I'm going for -ENODATA, meaning "I have no | |
174 | * work to do", even though we return -ENOSPC after this same test | |
175 | * during bsg_write() -- there, it means our buffer can't have more | |
176 | * bsg_commands added to it, thus has no space left. | |
177 | */ | |
178 | if (bd->done_cmds == bd->queued_cmds) { | |
179 | ret = -ENODATA; | |
180 | goto unlock; | |
181 | } | |
182 | ||
183 | if (!test_bit(BSG_F_BLOCK, &bd->flags)) { | |
184 | ret = -EAGAIN; | |
185 | goto unlock; | |
186 | } | |
187 | ||
188 | prepare_to_wait(&bd->wq_done, &wait, state); | |
189 | spin_unlock_irq(&bd->lock); | |
190 | io_schedule(); | |
191 | finish_wait(&bd->wq_done, &wait); | |
192 | ||
193 | if ((state == TASK_INTERRUPTIBLE) && signal_pending(current)) | |
194 | ret = -ERESTARTSYS; | |
195 | ||
196 | return ret; | |
197 | unlock: | |
198 | spin_unlock_irq(&bd->lock); | |
199 | return ret; | |
200 | } | |
201 | ||
70e36ece FT |
202 | static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq, |
203 | struct sg_io_v4 *hdr, int has_write_perm) | |
204 | { | |
205 | memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */ | |
206 | ||
207 | if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request, | |
208 | hdr->request_len)) | |
209 | return -EFAULT; | |
210 | if (blk_verify_command(rq->cmd, has_write_perm)) | |
211 | return -EPERM; | |
212 | ||
213 | /* | |
214 | * fill in request structure | |
215 | */ | |
216 | rq->cmd_len = hdr->request_len; | |
217 | rq->cmd_type = REQ_TYPE_BLOCK_PC; | |
218 | ||
219 | rq->timeout = (hdr->timeout * HZ) / 1000; | |
220 | if (!rq->timeout) | |
221 | rq->timeout = q->sg_timeout; | |
222 | if (!rq->timeout) | |
223 | rq->timeout = BLK_DEFAULT_SG_TIMEOUT; | |
224 | ||
225 | return 0; | |
226 | } | |
227 | ||
3d6392cf | 228 | /* |
70e36ece | 229 | * Check if sg_io_v4 from user is allowed and valid |
3d6392cf JA |
230 | */ |
231 | static int | |
70e36ece | 232 | bsg_validate_sgv4_hdr(request_queue_t *q, struct sg_io_v4 *hdr, int *rw) |
3d6392cf | 233 | { |
70e36ece | 234 | if (hdr->guard != 'Q') |
3d6392cf | 235 | return -EINVAL; |
70e36ece | 236 | if (hdr->request_len > BLK_MAX_CDB) |
3d6392cf | 237 | return -EINVAL; |
70e36ece FT |
238 | if (hdr->dout_xfer_len > (q->max_sectors << 9) || |
239 | hdr->din_xfer_len > (q->max_sectors << 9)) | |
3d6392cf JA |
240 | return -EIO; |
241 | ||
70e36ece FT |
242 | /* not supported currently */ |
243 | if (hdr->protocol || hdr->subprotocol) | |
244 | return -EINVAL; | |
245 | ||
3d6392cf JA |
246 | /* |
247 | * looks sane, if no data then it should be fine from our POV | |
248 | */ | |
70e36ece | 249 | if (!hdr->dout_xfer_len && !hdr->din_xfer_len) |
3d6392cf JA |
250 | return 0; |
251 | ||
70e36ece FT |
252 | /* not supported currently */ |
253 | if (hdr->dout_xfer_len && hdr->din_xfer_len) | |
254 | return -EINVAL; | |
255 | ||
256 | *rw = hdr->dout_xfer_len ? WRITE : READ; | |
3d6392cf JA |
257 | |
258 | return 0; | |
259 | } | |
260 | ||
261 | /* | |
70e36ece | 262 | * map sg_io_v4 to a request. |
3d6392cf JA |
263 | */ |
264 | static struct request * | |
70e36ece | 265 | bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr) |
3d6392cf JA |
266 | { |
267 | request_queue_t *q = bd->queue; | |
3d6392cf | 268 | struct request *rq; |
2ef7086a | 269 | int ret, rw = 0; /* shut up gcc */ |
70e36ece FT |
270 | unsigned int dxfer_len; |
271 | void *dxferp = NULL; | |
3d6392cf | 272 | |
70e36ece FT |
273 | dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp, |
274 | hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp, | |
275 | hdr->din_xfer_len); | |
3d6392cf | 276 | |
70e36ece | 277 | ret = bsg_validate_sgv4_hdr(q, hdr, &rw); |
3d6392cf JA |
278 | if (ret) |
279 | return ERR_PTR(ret); | |
280 | ||
281 | /* | |
282 | * map scatter-gather elements seperately and string them to request | |
283 | */ | |
284 | rq = blk_get_request(q, rw, GFP_KERNEL); | |
70e36ece FT |
285 | ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM, |
286 | &bd->flags)); | |
3d6392cf JA |
287 | if (ret) { |
288 | blk_put_request(rq); | |
289 | return ERR_PTR(ret); | |
290 | } | |
291 | ||
70e36ece FT |
292 | if (hdr->dout_xfer_len) { |
293 | dxfer_len = hdr->dout_xfer_len; | |
294 | dxferp = (void*)(unsigned long)hdr->dout_xferp; | |
295 | } else if (hdr->din_xfer_len) { | |
296 | dxfer_len = hdr->din_xfer_len; | |
297 | dxferp = (void*)(unsigned long)hdr->din_xferp; | |
298 | } else | |
299 | dxfer_len = 0; | |
300 | ||
301 | if (dxfer_len) { | |
302 | ret = blk_rq_map_user(q, rq, dxferp, dxfer_len); | |
303 | if (ret) { | |
304 | dprintk("failed map at %d\n", ret); | |
305 | blk_put_request(rq); | |
306 | rq = ERR_PTR(ret); | |
3d6392cf | 307 | } |
3d6392cf JA |
308 | } |
309 | ||
310 | return rq; | |
311 | } | |
312 | ||
313 | /* | |
314 | * async completion call-back from the block layer, when scsi/ide/whatever | |
315 | * calls end_that_request_last() on a request | |
316 | */ | |
317 | static void bsg_rq_end_io(struct request *rq, int uptodate) | |
318 | { | |
319 | struct bsg_command *bc = rq->end_io_data; | |
320 | struct bsg_device *bd = bc->bd; | |
321 | unsigned long flags; | |
322 | ||
5309cb38 JA |
323 | dprintk("%s: finished rq %p bc %p, bio %p stat %d\n", |
324 | bd->name, rq, bc, bc->bio, uptodate); | |
3d6392cf JA |
325 | |
326 | bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration); | |
327 | ||
328 | spin_lock_irqsave(&bd->lock, flags); | |
329 | list_del(&bc->list); | |
330 | bsg_add_done_cmd(bd, bc); | |
331 | spin_unlock_irqrestore(&bd->lock, flags); | |
332 | } | |
333 | ||
334 | /* | |
335 | * do final setup of a 'bc' and submit the matching 'rq' to the block | |
336 | * layer for io | |
337 | */ | |
338 | static void bsg_add_command(struct bsg_device *bd, request_queue_t *q, | |
339 | struct bsg_command *bc, struct request *rq) | |
340 | { | |
341 | rq->sense = bc->sense; | |
342 | rq->sense_len = 0; | |
343 | ||
344 | /* | |
345 | * add bc command to busy queue and submit rq for io | |
346 | */ | |
347 | bc->rq = rq; | |
348 | bc->bio = rq->bio; | |
349 | bc->hdr.duration = jiffies; | |
350 | spin_lock_irq(&bd->lock); | |
351 | list_add_tail(&bc->list, &bd->busy_list); | |
352 | spin_unlock_irq(&bd->lock); | |
353 | ||
354 | dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc); | |
355 | ||
356 | rq->end_io_data = bc; | |
d351af01 | 357 | blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io); |
3d6392cf JA |
358 | } |
359 | ||
360 | static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd) | |
361 | { | |
362 | struct bsg_command *bc = NULL; | |
363 | ||
364 | spin_lock_irq(&bd->lock); | |
365 | if (bd->done_cmds) { | |
366 | bc = list_entry_bc(bd->done_list.next); | |
367 | bsg_del_done_cmd(bd, bc); | |
368 | } | |
369 | spin_unlock_irq(&bd->lock); | |
370 | ||
371 | return bc; | |
372 | } | |
373 | ||
374 | /* | |
375 | * Get a finished command from the done list | |
376 | */ | |
e7d72173 | 377 | static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd) |
3d6392cf JA |
378 | { |
379 | struct bsg_command *bc; | |
380 | int ret; | |
381 | ||
382 | do { | |
383 | bc = bsg_next_done_cmd(bd); | |
384 | if (bc) | |
385 | break; | |
386 | ||
e7d72173 FT |
387 | if (!test_bit(BSG_F_BLOCK, &bd->flags)) { |
388 | bc = ERR_PTR(-EAGAIN); | |
389 | break; | |
390 | } | |
391 | ||
392 | ret = wait_event_interruptible(bd->wq_done, bd->done_cmds); | |
3d6392cf | 393 | if (ret) { |
e7d72173 | 394 | bc = ERR_PTR(-ERESTARTSYS); |
3d6392cf JA |
395 | break; |
396 | } | |
397 | } while (1); | |
398 | ||
399 | dprintk("%s: returning done %p\n", bd->name, bc); | |
400 | ||
401 | return bc; | |
402 | } | |
403 | ||
70e36ece FT |
404 | static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr, |
405 | struct bio *bio) | |
406 | { | |
407 | int ret = 0; | |
408 | ||
409 | dprintk("rq %p bio %p %u\n", rq, bio, rq->errors); | |
410 | /* | |
411 | * fill in all the output members | |
412 | */ | |
413 | hdr->device_status = status_byte(rq->errors); | |
414 | hdr->transport_status = host_byte(rq->errors); | |
415 | hdr->driver_status = driver_byte(rq->errors); | |
416 | hdr->info = 0; | |
417 | if (hdr->device_status || hdr->transport_status || hdr->driver_status) | |
418 | hdr->info |= SG_INFO_CHECK; | |
419 | hdr->din_resid = rq->data_len; | |
420 | hdr->response_len = 0; | |
421 | ||
422 | if (rq->sense_len && hdr->response) { | |
423 | int len = min((unsigned int) hdr->max_response_len, | |
424 | rq->sense_len); | |
425 | ||
426 | ret = copy_to_user((void*)(unsigned long)hdr->response, | |
427 | rq->sense, len); | |
428 | if (!ret) | |
429 | hdr->response_len = len; | |
430 | else | |
431 | ret = -EFAULT; | |
432 | } | |
433 | ||
434 | blk_rq_unmap_user(bio); | |
435 | blk_put_request(rq); | |
436 | ||
437 | return ret; | |
438 | } | |
439 | ||
3d6392cf JA |
440 | static int bsg_complete_all_commands(struct bsg_device *bd) |
441 | { | |
442 | struct bsg_command *bc; | |
443 | int ret, tret; | |
444 | ||
445 | dprintk("%s: entered\n", bd->name); | |
446 | ||
447 | set_bit(BSG_F_BLOCK, &bd->flags); | |
448 | ||
449 | /* | |
450 | * wait for all commands to complete | |
451 | */ | |
452 | ret = 0; | |
453 | do { | |
454 | ret = bsg_io_schedule(bd, TASK_UNINTERRUPTIBLE); | |
455 | /* | |
456 | * look for -ENODATA specifically -- we'll sometimes get | |
457 | * -ERESTARTSYS when we've taken a signal, but we can't | |
458 | * return until we're done freeing the queue, so ignore | |
459 | * it. The signal will get handled when we're done freeing | |
460 | * the bsg_device. | |
461 | */ | |
462 | } while (ret != -ENODATA); | |
463 | ||
464 | /* | |
465 | * discard done commands | |
466 | */ | |
467 | ret = 0; | |
468 | do { | |
e7d72173 FT |
469 | spin_lock_irq(&bd->lock); |
470 | if (!bd->queued_cmds) { | |
471 | spin_unlock_irq(&bd->lock); | |
3d6392cf JA |
472 | break; |
473 | } | |
474 | ||
e7d72173 FT |
475 | bc = bsg_get_done_cmd(bd); |
476 | if (IS_ERR(bc)) | |
477 | break; | |
478 | ||
70e36ece | 479 | tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio); |
3d6392cf JA |
480 | if (!ret) |
481 | ret = tret; | |
482 | ||
483 | bsg_free_command(bc); | |
484 | } while (1); | |
485 | ||
486 | return ret; | |
487 | } | |
488 | ||
3d6392cf | 489 | static ssize_t |
e7d72173 FT |
490 | __bsg_read(char __user *buf, size_t count, struct bsg_device *bd, |
491 | const struct iovec *iov, ssize_t *bytes_read) | |
3d6392cf JA |
492 | { |
493 | struct bsg_command *bc; | |
494 | int nr_commands, ret; | |
495 | ||
70e36ece | 496 | if (count % sizeof(struct sg_io_v4)) |
3d6392cf JA |
497 | return -EINVAL; |
498 | ||
499 | ret = 0; | |
70e36ece | 500 | nr_commands = count / sizeof(struct sg_io_v4); |
3d6392cf | 501 | while (nr_commands) { |
e7d72173 | 502 | bc = bsg_get_done_cmd(bd); |
3d6392cf JA |
503 | if (IS_ERR(bc)) { |
504 | ret = PTR_ERR(bc); | |
505 | break; | |
506 | } | |
507 | ||
508 | /* | |
509 | * this is the only case where we need to copy data back | |
510 | * after completing the request. so do that here, | |
511 | * bsg_complete_work() cannot do that for us | |
512 | */ | |
70e36ece | 513 | ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio); |
3d6392cf JA |
514 | |
515 | if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr))) | |
516 | ret = -EFAULT; | |
517 | ||
518 | bsg_free_command(bc); | |
519 | ||
520 | if (ret) | |
521 | break; | |
522 | ||
70e36ece FT |
523 | buf += sizeof(struct sg_io_v4); |
524 | *bytes_read += sizeof(struct sg_io_v4); | |
3d6392cf JA |
525 | nr_commands--; |
526 | } | |
527 | ||
528 | return ret; | |
529 | } | |
530 | ||
531 | static inline void bsg_set_block(struct bsg_device *bd, struct file *file) | |
532 | { | |
533 | if (file->f_flags & O_NONBLOCK) | |
534 | clear_bit(BSG_F_BLOCK, &bd->flags); | |
535 | else | |
536 | set_bit(BSG_F_BLOCK, &bd->flags); | |
537 | } | |
538 | ||
539 | static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file) | |
540 | { | |
541 | if (file->f_mode & FMODE_WRITE) | |
542 | set_bit(BSG_F_WRITE_PERM, &bd->flags); | |
543 | else | |
544 | clear_bit(BSG_F_WRITE_PERM, &bd->flags); | |
545 | } | |
546 | ||
547 | static inline int err_block_err(int ret) | |
548 | { | |
549 | if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN) | |
550 | return 1; | |
551 | ||
552 | return 0; | |
553 | } | |
554 | ||
555 | static ssize_t | |
556 | bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
557 | { | |
558 | struct bsg_device *bd = file->private_data; | |
559 | int ret; | |
560 | ssize_t bytes_read; | |
561 | ||
9e69fbb5 | 562 | dprintk("%s: read %Zd bytes\n", bd->name, count); |
3d6392cf JA |
563 | |
564 | bsg_set_block(bd, file); | |
565 | bytes_read = 0; | |
e7d72173 | 566 | ret = __bsg_read(buf, count, bd, NULL, &bytes_read); |
3d6392cf JA |
567 | *ppos = bytes_read; |
568 | ||
569 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
570 | bytes_read = ret; | |
571 | ||
572 | return bytes_read; | |
573 | } | |
574 | ||
575 | static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf, | |
576 | size_t count, ssize_t *bytes_read) | |
577 | { | |
578 | struct bsg_command *bc; | |
579 | struct request *rq; | |
580 | int ret, nr_commands; | |
581 | ||
70e36ece | 582 | if (count % sizeof(struct sg_io_v4)) |
3d6392cf JA |
583 | return -EINVAL; |
584 | ||
70e36ece | 585 | nr_commands = count / sizeof(struct sg_io_v4); |
3d6392cf JA |
586 | rq = NULL; |
587 | bc = NULL; | |
588 | ret = 0; | |
589 | while (nr_commands) { | |
590 | request_queue_t *q = bd->queue; | |
3d6392cf | 591 | |
e7d72173 | 592 | bc = bsg_alloc_command(bd); |
3d6392cf JA |
593 | if (IS_ERR(bc)) { |
594 | ret = PTR_ERR(bc); | |
595 | bc = NULL; | |
596 | break; | |
597 | } | |
598 | ||
70e36ece | 599 | bc->uhdr = (struct sg_io_v4 __user *) buf; |
3d6392cf JA |
600 | if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) { |
601 | ret = -EFAULT; | |
602 | break; | |
603 | } | |
604 | ||
605 | /* | |
606 | * get a request, fill in the blanks, and add to request queue | |
607 | */ | |
70e36ece | 608 | rq = bsg_map_hdr(bd, &bc->hdr); |
3d6392cf JA |
609 | if (IS_ERR(rq)) { |
610 | ret = PTR_ERR(rq); | |
611 | rq = NULL; | |
612 | break; | |
613 | } | |
614 | ||
615 | bsg_add_command(bd, q, bc, rq); | |
616 | bc = NULL; | |
617 | rq = NULL; | |
618 | nr_commands--; | |
70e36ece FT |
619 | buf += sizeof(struct sg_io_v4); |
620 | *bytes_read += sizeof(struct sg_io_v4); | |
3d6392cf JA |
621 | } |
622 | ||
3d6392cf JA |
623 | if (bc) |
624 | bsg_free_command(bc); | |
625 | ||
626 | return ret; | |
627 | } | |
628 | ||
629 | static ssize_t | |
630 | bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) | |
631 | { | |
632 | struct bsg_device *bd = file->private_data; | |
633 | ssize_t bytes_read; | |
634 | int ret; | |
635 | ||
9e69fbb5 | 636 | dprintk("%s: write %Zd bytes\n", bd->name, count); |
3d6392cf JA |
637 | |
638 | bsg_set_block(bd, file); | |
639 | bsg_set_write_perm(bd, file); | |
640 | ||
641 | bytes_read = 0; | |
642 | ret = __bsg_write(bd, buf, count, &bytes_read); | |
643 | *ppos = bytes_read; | |
644 | ||
645 | /* | |
646 | * return bytes written on non-fatal errors | |
647 | */ | |
648 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
649 | bytes_read = ret; | |
650 | ||
9e69fbb5 | 651 | dprintk("%s: returning %Zd\n", bd->name, bytes_read); |
3d6392cf JA |
652 | return bytes_read; |
653 | } | |
654 | ||
3d6392cf JA |
655 | static struct bsg_device *bsg_alloc_device(void) |
656 | { | |
3d6392cf | 657 | struct bsg_device *bd; |
3d6392cf JA |
658 | |
659 | bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL); | |
660 | if (unlikely(!bd)) | |
661 | return NULL; | |
662 | ||
663 | spin_lock_init(&bd->lock); | |
664 | ||
5309cb38 | 665 | bd->max_queue = BSG_DEFAULT_CMDS; |
3d6392cf JA |
666 | |
667 | INIT_LIST_HEAD(&bd->busy_list); | |
668 | INIT_LIST_HEAD(&bd->done_list); | |
669 | INIT_HLIST_NODE(&bd->dev_list); | |
670 | ||
671 | init_waitqueue_head(&bd->wq_free); | |
672 | init_waitqueue_head(&bd->wq_done); | |
673 | return bd; | |
3d6392cf JA |
674 | } |
675 | ||
676 | static int bsg_put_device(struct bsg_device *bd) | |
677 | { | |
678 | int ret = 0; | |
679 | ||
680 | mutex_lock(&bsg_mutex); | |
681 | ||
682 | if (!atomic_dec_and_test(&bd->ref_count)) | |
683 | goto out; | |
684 | ||
685 | dprintk("%s: tearing down\n", bd->name); | |
686 | ||
687 | /* | |
688 | * close can always block | |
689 | */ | |
690 | set_bit(BSG_F_BLOCK, &bd->flags); | |
691 | ||
692 | /* | |
693 | * correct error detection baddies here again. it's the responsibility | |
694 | * of the app to properly reap commands before close() if it wants | |
695 | * fool-proof error detection | |
696 | */ | |
697 | ret = bsg_complete_all_commands(bd); | |
698 | ||
699 | blk_put_queue(bd->queue); | |
700 | hlist_del(&bd->dev_list); | |
5309cb38 | 701 | kfree(bd); |
3d6392cf JA |
702 | out: |
703 | mutex_unlock(&bsg_mutex); | |
704 | return ret; | |
705 | } | |
706 | ||
707 | static struct bsg_device *bsg_add_device(struct inode *inode, | |
d351af01 | 708 | struct request_queue *rq, |
3d6392cf JA |
709 | struct file *file) |
710 | { | |
711 | struct bsg_device *bd = NULL; | |
712 | #ifdef BSG_DEBUG | |
713 | unsigned char buf[32]; | |
714 | #endif | |
715 | ||
716 | bd = bsg_alloc_device(); | |
717 | if (!bd) | |
718 | return ERR_PTR(-ENOMEM); | |
719 | ||
d351af01 FT |
720 | bd->queue = rq; |
721 | kobject_get(&rq->kobj); | |
3d6392cf JA |
722 | bsg_set_block(bd, file); |
723 | ||
724 | atomic_set(&bd->ref_count, 1); | |
725 | bd->minor = iminor(inode); | |
726 | mutex_lock(&bsg_mutex); | |
d351af01 | 727 | hlist_add_head(&bd->dev_list, &bsg_device_list[bsg_list_idx(bd->minor)]); |
3d6392cf | 728 | |
d351af01 | 729 | strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1); |
3d6392cf | 730 | dprintk("bound to <%s>, max queue %d\n", |
9e69fbb5 | 731 | format_dev_t(buf, inode->i_rdev), bd->max_queue); |
3d6392cf JA |
732 | |
733 | mutex_unlock(&bsg_mutex); | |
734 | return bd; | |
735 | } | |
736 | ||
737 | static struct bsg_device *__bsg_get_device(int minor) | |
738 | { | |
739 | struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)]; | |
740 | struct bsg_device *bd = NULL; | |
741 | struct hlist_node *entry; | |
742 | ||
743 | mutex_lock(&bsg_mutex); | |
744 | ||
745 | hlist_for_each(entry, list) { | |
746 | bd = hlist_entry(entry, struct bsg_device, dev_list); | |
747 | if (bd->minor == minor) { | |
748 | atomic_inc(&bd->ref_count); | |
749 | break; | |
750 | } | |
751 | ||
752 | bd = NULL; | |
753 | } | |
754 | ||
755 | mutex_unlock(&bsg_mutex); | |
756 | return bd; | |
757 | } | |
758 | ||
759 | static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file) | |
760 | { | |
761 | struct bsg_device *bd = __bsg_get_device(iminor(inode)); | |
762 | struct bsg_class_device *bcd, *__bcd; | |
763 | ||
764 | if (bd) | |
765 | return bd; | |
766 | ||
767 | /* | |
768 | * find the class device | |
769 | */ | |
770 | bcd = NULL; | |
771 | mutex_lock(&bsg_mutex); | |
772 | list_for_each_entry(__bcd, &bsg_class_list, list) { | |
773 | if (__bcd->minor == iminor(inode)) { | |
774 | bcd = __bcd; | |
775 | break; | |
776 | } | |
777 | } | |
778 | mutex_unlock(&bsg_mutex); | |
779 | ||
780 | if (!bcd) | |
781 | return ERR_PTR(-ENODEV); | |
782 | ||
d351af01 | 783 | return bsg_add_device(inode, bcd->queue, file); |
3d6392cf JA |
784 | } |
785 | ||
786 | static int bsg_open(struct inode *inode, struct file *file) | |
787 | { | |
788 | struct bsg_device *bd = bsg_get_device(inode, file); | |
789 | ||
790 | if (IS_ERR(bd)) | |
791 | return PTR_ERR(bd); | |
792 | ||
793 | file->private_data = bd; | |
794 | return 0; | |
795 | } | |
796 | ||
797 | static int bsg_release(struct inode *inode, struct file *file) | |
798 | { | |
799 | struct bsg_device *bd = file->private_data; | |
800 | ||
801 | file->private_data = NULL; | |
802 | return bsg_put_device(bd); | |
803 | } | |
804 | ||
805 | static unsigned int bsg_poll(struct file *file, poll_table *wait) | |
806 | { | |
807 | struct bsg_device *bd = file->private_data; | |
808 | unsigned int mask = 0; | |
809 | ||
810 | poll_wait(file, &bd->wq_done, wait); | |
811 | poll_wait(file, &bd->wq_free, wait); | |
812 | ||
813 | spin_lock_irq(&bd->lock); | |
814 | if (!list_empty(&bd->done_list)) | |
815 | mask |= POLLIN | POLLRDNORM; | |
816 | if (bd->queued_cmds >= bd->max_queue) | |
817 | mask |= POLLOUT; | |
818 | spin_unlock_irq(&bd->lock); | |
819 | ||
820 | return mask; | |
821 | } | |
822 | ||
823 | static int | |
824 | bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd, | |
825 | unsigned long arg) | |
826 | { | |
827 | struct bsg_device *bd = file->private_data; | |
828 | int __user *uarg = (int __user *) arg; | |
829 | ||
830 | if (!bd) | |
831 | return -ENXIO; | |
832 | ||
833 | switch (cmd) { | |
834 | /* | |
835 | * our own ioctls | |
836 | */ | |
837 | case SG_GET_COMMAND_Q: | |
838 | return put_user(bd->max_queue, uarg); | |
5309cb38 | 839 | case SG_SET_COMMAND_Q: { |
3d6392cf JA |
840 | int queue; |
841 | ||
842 | if (get_user(queue, uarg)) | |
843 | return -EFAULT; | |
5309cb38 | 844 | if (queue < 1) |
3d6392cf JA |
845 | return -EINVAL; |
846 | ||
5309cb38 | 847 | spin_lock_irq(&bd->lock); |
3d6392cf | 848 | bd->max_queue = queue; |
5309cb38 | 849 | spin_unlock_irq(&bd->lock); |
3d6392cf JA |
850 | return 0; |
851 | } | |
852 | ||
853 | /* | |
854 | * SCSI/sg ioctls | |
855 | */ | |
856 | case SG_GET_VERSION_NUM: | |
857 | case SCSI_IOCTL_GET_IDLUN: | |
858 | case SCSI_IOCTL_GET_BUS_NUMBER: | |
859 | case SG_SET_TIMEOUT: | |
860 | case SG_GET_TIMEOUT: | |
861 | case SG_GET_RESERVED_SIZE: | |
862 | case SG_SET_RESERVED_SIZE: | |
863 | case SG_EMULATED_HOST: | |
3d6392cf JA |
864 | case SCSI_IOCTL_SEND_COMMAND: { |
865 | void __user *uarg = (void __user *) arg; | |
d351af01 | 866 | return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg); |
3d6392cf | 867 | } |
10e8855b FT |
868 | case SG_IO: { |
869 | struct request *rq; | |
870 | struct bio *bio; | |
871 | struct sg_io_v4 hdr; | |
872 | ||
873 | if (copy_from_user(&hdr, uarg, sizeof(hdr))) | |
874 | return -EFAULT; | |
875 | ||
876 | rq = bsg_map_hdr(bd, &hdr); | |
877 | if (IS_ERR(rq)) | |
878 | return PTR_ERR(rq); | |
879 | ||
880 | bio = rq->bio; | |
d351af01 | 881 | blk_execute_rq(bd->queue, NULL, rq, 0); |
10e8855b FT |
882 | blk_complete_sgv4_hdr_rq(rq, &hdr, bio); |
883 | ||
884 | if (copy_to_user(uarg, &hdr, sizeof(hdr))) | |
885 | return -EFAULT; | |
b711afa6 JA |
886 | |
887 | return 0; | |
10e8855b | 888 | } |
3d6392cf JA |
889 | /* |
890 | * block device ioctls | |
891 | */ | |
892 | default: | |
893 | #if 0 | |
894 | return ioctl_by_bdev(bd->bdev, cmd, arg); | |
895 | #else | |
896 | return -ENOTTY; | |
897 | #endif | |
898 | } | |
899 | } | |
900 | ||
901 | static struct file_operations bsg_fops = { | |
902 | .read = bsg_read, | |
903 | .write = bsg_write, | |
904 | .poll = bsg_poll, | |
905 | .open = bsg_open, | |
906 | .release = bsg_release, | |
907 | .ioctl = bsg_ioctl, | |
908 | .owner = THIS_MODULE, | |
909 | }; | |
910 | ||
d351af01 | 911 | void bsg_unregister_queue(struct request_queue *q) |
3d6392cf | 912 | { |
d351af01 | 913 | struct bsg_class_device *bcd = &q->bsg_dev; |
3d6392cf JA |
914 | |
915 | if (!bcd->class_dev) | |
916 | return; | |
917 | ||
918 | mutex_lock(&bsg_mutex); | |
d351af01 | 919 | sysfs_remove_link(&q->kobj, "bsg"); |
3d6392cf JA |
920 | class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor)); |
921 | bcd->class_dev = NULL; | |
922 | list_del_init(&bcd->list); | |
292b7f27 | 923 | bsg_device_nr--; |
3d6392cf JA |
924 | mutex_unlock(&bsg_mutex); |
925 | } | |
4cf0723a | 926 | EXPORT_SYMBOL_GPL(bsg_unregister_queue); |
3d6392cf | 927 | |
4cf0723a | 928 | int bsg_register_queue(struct request_queue *q, const char *name) |
3d6392cf | 929 | { |
292b7f27 | 930 | struct bsg_class_device *bcd, *__bcd; |
3d6392cf | 931 | dev_t dev; |
292b7f27 | 932 | int ret = -EMFILE; |
4e2872d6 | 933 | struct class_device *class_dev = NULL; |
3d6392cf JA |
934 | |
935 | /* | |
936 | * we need a proper transport to send commands, not a stacked device | |
937 | */ | |
938 | if (!q->request_fn) | |
939 | return 0; | |
940 | ||
d351af01 | 941 | bcd = &q->bsg_dev; |
3d6392cf JA |
942 | memset(bcd, 0, sizeof(*bcd)); |
943 | INIT_LIST_HEAD(&bcd->list); | |
944 | ||
945 | mutex_lock(&bsg_mutex); | |
292b7f27 FT |
946 | if (bsg_device_nr == BSG_MAX_DEVS) { |
947 | printk(KERN_ERR "bsg: too many bsg devices\n"); | |
948 | goto err; | |
949 | } | |
950 | ||
951 | retry: | |
952 | list_for_each_entry(__bcd, &bsg_class_list, list) { | |
953 | if (__bcd->minor == bsg_minor_idx) { | |
954 | bsg_minor_idx++; | |
955 | if (bsg_minor_idx == BSG_MAX_DEVS) | |
956 | bsg_minor_idx = 0; | |
957 | goto retry; | |
958 | } | |
959 | } | |
960 | ||
961 | bcd->minor = bsg_minor_idx++; | |
962 | if (bsg_minor_idx == BSG_MAX_DEVS) | |
963 | bsg_minor_idx = 0; | |
964 | ||
d351af01 | 965 | bcd->queue = q; |
292b7f27 | 966 | dev = MKDEV(BSG_MAJOR, bcd->minor); |
4e2872d6 FT |
967 | class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", name); |
968 | if (IS_ERR(class_dev)) { | |
969 | ret = PTR_ERR(class_dev); | |
264a0472 | 970 | goto err; |
4e2872d6 FT |
971 | } |
972 | bcd->class_dev = class_dev; | |
973 | ||
974 | if (q->kobj.dentry) { | |
975 | ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg"); | |
976 | if (ret) | |
977 | goto err; | |
978 | } | |
979 | ||
3d6392cf | 980 | list_add_tail(&bcd->list, &bsg_class_list); |
292b7f27 | 981 | bsg_device_nr++; |
4e2872d6 | 982 | |
3d6392cf JA |
983 | mutex_unlock(&bsg_mutex); |
984 | return 0; | |
264a0472 | 985 | err: |
4e2872d6 | 986 | if (class_dev) |
264a0472 JA |
987 | class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor)); |
988 | mutex_unlock(&bsg_mutex); | |
4e2872d6 FT |
989 | return ret; |
990 | } | |
4cf0723a | 991 | EXPORT_SYMBOL_GPL(bsg_register_queue); |
4e2872d6 FT |
992 | |
993 | static int bsg_add(struct class_device *cl_dev, struct class_interface *cl_intf) | |
994 | { | |
995 | int ret; | |
996 | struct scsi_device *sdp = to_scsi_device(cl_dev->dev); | |
997 | struct request_queue *rq = sdp->request_queue; | |
998 | ||
999 | if (rq->kobj.parent) | |
1000 | ret = bsg_register_queue(rq, kobject_name(rq->kobj.parent)); | |
1001 | else | |
1002 | ret = bsg_register_queue(rq, kobject_name(&sdp->sdev_gendev.kobj)); | |
1003 | return ret; | |
3d6392cf JA |
1004 | } |
1005 | ||
4e2872d6 FT |
1006 | static void bsg_remove(struct class_device *cl_dev, struct class_interface *cl_intf) |
1007 | { | |
1008 | bsg_unregister_queue(to_scsi_device(cl_dev->dev)->request_queue); | |
1009 | } | |
1010 | ||
1011 | static struct class_interface bsg_intf = { | |
1012 | .add = bsg_add, | |
1013 | .remove = bsg_remove, | |
1014 | }; | |
1015 | ||
292b7f27 FT |
1016 | static struct cdev bsg_cdev = { |
1017 | .kobj = {.name = "bsg", }, | |
1018 | .owner = THIS_MODULE, | |
1019 | }; | |
1020 | ||
3d6392cf JA |
1021 | static int __init bsg_init(void) |
1022 | { | |
1023 | int ret, i; | |
1024 | ||
5309cb38 JA |
1025 | bsg_cmd_cachep = kmem_cache_create("bsg_cmd", |
1026 | sizeof(struct bsg_command), 0, 0, NULL, NULL); | |
1027 | if (!bsg_cmd_cachep) { | |
1028 | printk(KERN_ERR "bsg: failed creating slab cache\n"); | |
1029 | return -ENOMEM; | |
1030 | } | |
1031 | ||
3d6392cf JA |
1032 | for (i = 0; i < BSG_LIST_SIZE; i++) |
1033 | INIT_HLIST_HEAD(&bsg_device_list[i]); | |
1034 | ||
1035 | bsg_class = class_create(THIS_MODULE, "bsg"); | |
5309cb38 JA |
1036 | if (IS_ERR(bsg_class)) { |
1037 | kmem_cache_destroy(bsg_cmd_cachep); | |
3d6392cf | 1038 | return PTR_ERR(bsg_class); |
5309cb38 | 1039 | } |
3d6392cf | 1040 | |
292b7f27 FT |
1041 | ret = register_chrdev_region(MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS, "bsg"); |
1042 | if (ret) { | |
1043 | kmem_cache_destroy(bsg_cmd_cachep); | |
1044 | class_destroy(bsg_class); | |
1045 | return ret; | |
1046 | } | |
1047 | ||
1048 | cdev_init(&bsg_cdev, &bsg_fops); | |
1049 | ret = cdev_add(&bsg_cdev, MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS); | |
3d6392cf | 1050 | if (ret) { |
5309cb38 | 1051 | kmem_cache_destroy(bsg_cmd_cachep); |
3d6392cf | 1052 | class_destroy(bsg_class); |
292b7f27 | 1053 | unregister_chrdev_region(MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS); |
3d6392cf JA |
1054 | return ret; |
1055 | } | |
1056 | ||
4e2872d6 FT |
1057 | ret = scsi_register_interface(&bsg_intf); |
1058 | if (ret) { | |
1059 | printk(KERN_ERR "bsg: failed register scsi interface %d\n", ret); | |
1060 | kmem_cache_destroy(bsg_cmd_cachep); | |
1061 | class_destroy(bsg_class); | |
1062 | unregister_chrdev(BSG_MAJOR, "bsg"); | |
1063 | return ret; | |
1064 | } | |
1065 | ||
3d6392cf JA |
1066 | printk(KERN_INFO "%s loaded\n", bsg_version); |
1067 | return 0; | |
1068 | } | |
1069 | ||
1070 | MODULE_AUTHOR("Jens Axboe"); | |
1071 | MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver"); | |
1072 | MODULE_LICENSE("GPL"); | |
1073 | ||
4e2872d6 | 1074 | device_initcall(bsg_init); |