Commit | Line | Data |
---|---|---|
3d6392cf | 1 | /* |
0c6a89ba | 2 | * bsg.c - block layer implementation of the sg v4 interface |
3d6392cf JA |
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 | */ | |
3d6392cf JA |
12 | #include <linux/module.h> |
13 | #include <linux/init.h> | |
14 | #include <linux/file.h> | |
15 | #include <linux/blkdev.h> | |
16 | #include <linux/poll.h> | |
17 | #include <linux/cdev.h> | |
18 | #include <linux/percpu.h> | |
19 | #include <linux/uio.h> | |
598443a2 | 20 | #include <linux/idr.h> |
3d6392cf JA |
21 | #include <linux/bsg.h> |
22 | ||
23 | #include <scsi/scsi.h> | |
24 | #include <scsi/scsi_ioctl.h> | |
25 | #include <scsi/scsi_cmnd.h> | |
4e2872d6 FT |
26 | #include <scsi/scsi_device.h> |
27 | #include <scsi/scsi_driver.h> | |
3d6392cf JA |
28 | #include <scsi/sg.h> |
29 | ||
0ed081ce FT |
30 | #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver" |
31 | #define BSG_VERSION "0.4" | |
3d6392cf | 32 | |
3d6392cf | 33 | struct bsg_device { |
165125e1 | 34 | struct request_queue *queue; |
3d6392cf JA |
35 | spinlock_t lock; |
36 | struct list_head busy_list; | |
37 | struct list_head done_list; | |
38 | struct hlist_node dev_list; | |
39 | atomic_t ref_count; | |
40 | int minor; | |
41 | int queued_cmds; | |
42 | int done_cmds; | |
3d6392cf JA |
43 | wait_queue_head_t wq_done; |
44 | wait_queue_head_t wq_free; | |
d351af01 | 45 | char name[BUS_ID_SIZE]; |
3d6392cf JA |
46 | int max_queue; |
47 | unsigned long flags; | |
48 | }; | |
49 | ||
50 | enum { | |
51 | BSG_F_BLOCK = 1, | |
52 | BSG_F_WRITE_PERM = 2, | |
53 | }; | |
54 | ||
5309cb38 | 55 | #define BSG_DEFAULT_CMDS 64 |
292b7f27 | 56 | #define BSG_MAX_DEVS 32768 |
3d6392cf JA |
57 | |
58 | #undef BSG_DEBUG | |
59 | ||
60 | #ifdef BSG_DEBUG | |
61 | #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args) | |
62 | #else | |
63 | #define dprintk(fmt, args...) | |
64 | #endif | |
65 | ||
3d6392cf | 66 | static DEFINE_MUTEX(bsg_mutex); |
598443a2 | 67 | static DEFINE_IDR(bsg_minor_idr); |
3d6392cf | 68 | |
25fd1643 | 69 | #define BSG_LIST_ARRAY_SIZE 8 |
25fd1643 | 70 | static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE]; |
3d6392cf JA |
71 | |
72 | static struct class *bsg_class; | |
46f6ef4a | 73 | static int bsg_major; |
3d6392cf | 74 | |
5309cb38 JA |
75 | static struct kmem_cache *bsg_cmd_cachep; |
76 | ||
3d6392cf JA |
77 | /* |
78 | * our internal command type | |
79 | */ | |
80 | struct bsg_command { | |
81 | struct bsg_device *bd; | |
82 | struct list_head list; | |
83 | struct request *rq; | |
84 | struct bio *bio; | |
2c9ecdf4 | 85 | struct bio *bidi_bio; |
3d6392cf | 86 | int err; |
70e36ece | 87 | struct sg_io_v4 hdr; |
3d6392cf JA |
88 | char sense[SCSI_SENSE_BUFFERSIZE]; |
89 | }; | |
90 | ||
91 | static void bsg_free_command(struct bsg_command *bc) | |
92 | { | |
93 | struct bsg_device *bd = bc->bd; | |
3d6392cf JA |
94 | unsigned long flags; |
95 | ||
5309cb38 | 96 | kmem_cache_free(bsg_cmd_cachep, bc); |
3d6392cf JA |
97 | |
98 | spin_lock_irqsave(&bd->lock, flags); | |
99 | bd->queued_cmds--; | |
3d6392cf JA |
100 | spin_unlock_irqrestore(&bd->lock, flags); |
101 | ||
102 | wake_up(&bd->wq_free); | |
103 | } | |
104 | ||
e7d72173 | 105 | static struct bsg_command *bsg_alloc_command(struct bsg_device *bd) |
3d6392cf | 106 | { |
e7d72173 | 107 | struct bsg_command *bc = ERR_PTR(-EINVAL); |
3d6392cf JA |
108 | |
109 | spin_lock_irq(&bd->lock); | |
110 | ||
111 | if (bd->queued_cmds >= bd->max_queue) | |
112 | goto out; | |
113 | ||
3d6392cf | 114 | bd->queued_cmds++; |
3d6392cf JA |
115 | spin_unlock_irq(&bd->lock); |
116 | ||
25fd1643 | 117 | bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL); |
5309cb38 JA |
118 | if (unlikely(!bc)) { |
119 | spin_lock_irq(&bd->lock); | |
7e75d730 | 120 | bd->queued_cmds--; |
e7d72173 | 121 | bc = ERR_PTR(-ENOMEM); |
7e75d730 | 122 | goto out; |
5309cb38 JA |
123 | } |
124 | ||
3d6392cf JA |
125 | bc->bd = bd; |
126 | INIT_LIST_HEAD(&bc->list); | |
5309cb38 | 127 | dprintk("%s: returning free cmd %p\n", bd->name, bc); |
3d6392cf JA |
128 | return bc; |
129 | out: | |
3d6392cf JA |
130 | spin_unlock_irq(&bd->lock); |
131 | return bc; | |
132 | } | |
133 | ||
1c1133e1 | 134 | static inline struct hlist_head *bsg_dev_idx_hash(int index) |
3d6392cf | 135 | { |
1c1133e1 | 136 | return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)]; |
3d6392cf JA |
137 | } |
138 | ||
25fd1643 | 139 | static int bsg_io_schedule(struct bsg_device *bd) |
3d6392cf JA |
140 | { |
141 | DEFINE_WAIT(wait); | |
142 | int ret = 0; | |
143 | ||
144 | spin_lock_irq(&bd->lock); | |
145 | ||
146 | BUG_ON(bd->done_cmds > bd->queued_cmds); | |
147 | ||
148 | /* | |
149 | * -ENOSPC or -ENODATA? I'm going for -ENODATA, meaning "I have no | |
150 | * work to do", even though we return -ENOSPC after this same test | |
151 | * during bsg_write() -- there, it means our buffer can't have more | |
152 | * bsg_commands added to it, thus has no space left. | |
153 | */ | |
154 | if (bd->done_cmds == bd->queued_cmds) { | |
155 | ret = -ENODATA; | |
156 | goto unlock; | |
157 | } | |
158 | ||
159 | if (!test_bit(BSG_F_BLOCK, &bd->flags)) { | |
160 | ret = -EAGAIN; | |
161 | goto unlock; | |
162 | } | |
163 | ||
25fd1643 | 164 | prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE); |
3d6392cf JA |
165 | spin_unlock_irq(&bd->lock); |
166 | io_schedule(); | |
167 | finish_wait(&bd->wq_done, &wait); | |
168 | ||
3d6392cf JA |
169 | return ret; |
170 | unlock: | |
171 | spin_unlock_irq(&bd->lock); | |
172 | return ret; | |
173 | } | |
174 | ||
165125e1 | 175 | static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq, |
70e36ece FT |
176 | struct sg_io_v4 *hdr, int has_write_perm) |
177 | { | |
178 | memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */ | |
179 | ||
180 | if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request, | |
181 | hdr->request_len)) | |
182 | return -EFAULT; | |
15d10b61 FT |
183 | |
184 | if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) { | |
185 | if (blk_verify_command(rq->cmd, has_write_perm)) | |
186 | return -EPERM; | |
187 | } else if (!capable(CAP_SYS_RAWIO)) | |
70e36ece FT |
188 | return -EPERM; |
189 | ||
190 | /* | |
191 | * fill in request structure | |
192 | */ | |
193 | rq->cmd_len = hdr->request_len; | |
194 | rq->cmd_type = REQ_TYPE_BLOCK_PC; | |
195 | ||
196 | rq->timeout = (hdr->timeout * HZ) / 1000; | |
197 | if (!rq->timeout) | |
198 | rq->timeout = q->sg_timeout; | |
199 | if (!rq->timeout) | |
200 | rq->timeout = BLK_DEFAULT_SG_TIMEOUT; | |
201 | ||
202 | return 0; | |
203 | } | |
204 | ||
3d6392cf | 205 | /* |
70e36ece | 206 | * Check if sg_io_v4 from user is allowed and valid |
3d6392cf JA |
207 | */ |
208 | static int | |
165125e1 | 209 | bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw) |
3d6392cf | 210 | { |
15d10b61 FT |
211 | int ret = 0; |
212 | ||
70e36ece | 213 | if (hdr->guard != 'Q') |
3d6392cf | 214 | return -EINVAL; |
70e36ece | 215 | if (hdr->request_len > BLK_MAX_CDB) |
3d6392cf | 216 | return -EINVAL; |
70e36ece FT |
217 | if (hdr->dout_xfer_len > (q->max_sectors << 9) || |
218 | hdr->din_xfer_len > (q->max_sectors << 9)) | |
3d6392cf JA |
219 | return -EIO; |
220 | ||
15d10b61 FT |
221 | switch (hdr->protocol) { |
222 | case BSG_PROTOCOL_SCSI: | |
223 | switch (hdr->subprotocol) { | |
224 | case BSG_SUB_PROTOCOL_SCSI_CMD: | |
225 | case BSG_SUB_PROTOCOL_SCSI_TRANSPORT: | |
226 | break; | |
227 | default: | |
228 | ret = -EINVAL; | |
229 | } | |
230 | break; | |
231 | default: | |
232 | ret = -EINVAL; | |
233 | } | |
70e36ece | 234 | |
70e36ece | 235 | *rw = hdr->dout_xfer_len ? WRITE : READ; |
15d10b61 | 236 | return ret; |
3d6392cf JA |
237 | } |
238 | ||
239 | /* | |
70e36ece | 240 | * map sg_io_v4 to a request. |
3d6392cf JA |
241 | */ |
242 | static struct request * | |
70e36ece | 243 | bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr) |
3d6392cf | 244 | { |
165125e1 | 245 | struct request_queue *q = bd->queue; |
2c9ecdf4 | 246 | struct request *rq, *next_rq = NULL; |
25fd1643 | 247 | int ret, rw; |
70e36ece FT |
248 | unsigned int dxfer_len; |
249 | void *dxferp = NULL; | |
3d6392cf | 250 | |
70e36ece FT |
251 | dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp, |
252 | hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp, | |
253 | hdr->din_xfer_len); | |
3d6392cf | 254 | |
70e36ece | 255 | ret = bsg_validate_sgv4_hdr(q, hdr, &rw); |
3d6392cf JA |
256 | if (ret) |
257 | return ERR_PTR(ret); | |
258 | ||
259 | /* | |
260 | * map scatter-gather elements seperately and string them to request | |
261 | */ | |
262 | rq = blk_get_request(q, rw, GFP_KERNEL); | |
2c9ecdf4 FT |
263 | if (!rq) |
264 | return ERR_PTR(-ENOMEM); | |
70e36ece FT |
265 | ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM, |
266 | &bd->flags)); | |
2c9ecdf4 FT |
267 | if (ret) |
268 | goto out; | |
269 | ||
270 | if (rw == WRITE && hdr->din_xfer_len) { | |
271 | if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) { | |
272 | ret = -EOPNOTSUPP; | |
273 | goto out; | |
274 | } | |
275 | ||
276 | next_rq = blk_get_request(q, READ, GFP_KERNEL); | |
277 | if (!next_rq) { | |
278 | ret = -ENOMEM; | |
279 | goto out; | |
280 | } | |
281 | rq->next_rq = next_rq; | |
282 | ||
283 | dxferp = (void*)(unsigned long)hdr->din_xferp; | |
284 | ret = blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len); | |
285 | if (ret) | |
286 | goto out; | |
3d6392cf JA |
287 | } |
288 | ||
70e36ece FT |
289 | if (hdr->dout_xfer_len) { |
290 | dxfer_len = hdr->dout_xfer_len; | |
291 | dxferp = (void*)(unsigned long)hdr->dout_xferp; | |
292 | } else if (hdr->din_xfer_len) { | |
293 | dxfer_len = hdr->din_xfer_len; | |
294 | dxferp = (void*)(unsigned long)hdr->din_xferp; | |
295 | } else | |
296 | dxfer_len = 0; | |
297 | ||
298 | if (dxfer_len) { | |
299 | ret = blk_rq_map_user(q, rq, dxferp, dxfer_len); | |
2c9ecdf4 FT |
300 | if (ret) |
301 | goto out; | |
3d6392cf | 302 | } |
3d6392cf | 303 | return rq; |
2c9ecdf4 FT |
304 | out: |
305 | blk_put_request(rq); | |
306 | if (next_rq) { | |
307 | blk_rq_unmap_user(next_rq->bio); | |
308 | blk_put_request(next_rq); | |
309 | } | |
310 | return ERR_PTR(ret); | |
3d6392cf JA |
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); | |
25fd1643 JA |
329 | list_move_tail(&bc->list, &bd->done_list); |
330 | bd->done_cmds++; | |
3d6392cf | 331 | spin_unlock_irqrestore(&bd->lock, flags); |
25fd1643 JA |
332 | |
333 | wake_up(&bd->wq_done); | |
3d6392cf JA |
334 | } |
335 | ||
336 | /* | |
337 | * do final setup of a 'bc' and submit the matching 'rq' to the block | |
338 | * layer for io | |
339 | */ | |
165125e1 | 340 | static void bsg_add_command(struct bsg_device *bd, struct request_queue *q, |
3d6392cf JA |
341 | struct bsg_command *bc, struct request *rq) |
342 | { | |
343 | rq->sense = bc->sense; | |
344 | rq->sense_len = 0; | |
345 | ||
346 | /* | |
347 | * add bc command to busy queue and submit rq for io | |
348 | */ | |
349 | bc->rq = rq; | |
350 | bc->bio = rq->bio; | |
2c9ecdf4 FT |
351 | if (rq->next_rq) |
352 | bc->bidi_bio = rq->next_rq->bio; | |
3d6392cf JA |
353 | bc->hdr.duration = jiffies; |
354 | spin_lock_irq(&bd->lock); | |
355 | list_add_tail(&bc->list, &bd->busy_list); | |
356 | spin_unlock_irq(&bd->lock); | |
357 | ||
358 | dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc); | |
359 | ||
360 | rq->end_io_data = bc; | |
d351af01 | 361 | blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io); |
3d6392cf JA |
362 | } |
363 | ||
25fd1643 | 364 | static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd) |
3d6392cf JA |
365 | { |
366 | struct bsg_command *bc = NULL; | |
367 | ||
368 | spin_lock_irq(&bd->lock); | |
369 | if (bd->done_cmds) { | |
25fd1643 JA |
370 | bc = list_entry(bd->done_list.next, struct bsg_command, list); |
371 | list_del(&bc->list); | |
372 | bd->done_cmds--; | |
3d6392cf JA |
373 | } |
374 | spin_unlock_irq(&bd->lock); | |
375 | ||
376 | return bc; | |
377 | } | |
378 | ||
379 | /* | |
380 | * Get a finished command from the done list | |
381 | */ | |
e7d72173 | 382 | static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd) |
3d6392cf JA |
383 | { |
384 | struct bsg_command *bc; | |
385 | int ret; | |
386 | ||
387 | do { | |
388 | bc = bsg_next_done_cmd(bd); | |
389 | if (bc) | |
390 | break; | |
391 | ||
e7d72173 FT |
392 | if (!test_bit(BSG_F_BLOCK, &bd->flags)) { |
393 | bc = ERR_PTR(-EAGAIN); | |
394 | break; | |
395 | } | |
396 | ||
397 | ret = wait_event_interruptible(bd->wq_done, bd->done_cmds); | |
3d6392cf | 398 | if (ret) { |
e7d72173 | 399 | bc = ERR_PTR(-ERESTARTSYS); |
3d6392cf JA |
400 | break; |
401 | } | |
402 | } while (1); | |
403 | ||
404 | dprintk("%s: returning done %p\n", bd->name, bc); | |
405 | ||
406 | return bc; | |
407 | } | |
408 | ||
70e36ece | 409 | static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr, |
2c9ecdf4 | 410 | struct bio *bio, struct bio *bidi_bio) |
70e36ece FT |
411 | { |
412 | int ret = 0; | |
413 | ||
414 | dprintk("rq %p bio %p %u\n", rq, bio, rq->errors); | |
415 | /* | |
416 | * fill in all the output members | |
417 | */ | |
418 | hdr->device_status = status_byte(rq->errors); | |
419 | hdr->transport_status = host_byte(rq->errors); | |
420 | hdr->driver_status = driver_byte(rq->errors); | |
421 | hdr->info = 0; | |
422 | if (hdr->device_status || hdr->transport_status || hdr->driver_status) | |
423 | hdr->info |= SG_INFO_CHECK; | |
70e36ece FT |
424 | hdr->response_len = 0; |
425 | ||
426 | if (rq->sense_len && hdr->response) { | |
25fd1643 JA |
427 | int len = min_t(unsigned int, hdr->max_response_len, |
428 | rq->sense_len); | |
70e36ece FT |
429 | |
430 | ret = copy_to_user((void*)(unsigned long)hdr->response, | |
431 | rq->sense, len); | |
432 | if (!ret) | |
433 | hdr->response_len = len; | |
434 | else | |
435 | ret = -EFAULT; | |
436 | } | |
437 | ||
2c9ecdf4 | 438 | if (rq->next_rq) { |
0c6a89ba FT |
439 | hdr->dout_resid = rq->data_len; |
440 | hdr->din_resid = rq->next_rq->data_len; | |
2c9ecdf4 FT |
441 | blk_rq_unmap_user(bidi_bio); |
442 | blk_put_request(rq->next_rq); | |
0c6a89ba FT |
443 | } else if (rq_data_dir(rq) == READ) |
444 | hdr->din_resid = rq->data_len; | |
445 | else | |
446 | hdr->dout_resid = rq->data_len; | |
2c9ecdf4 | 447 | |
2d507a01 JB |
448 | /* |
449 | * If the request generated a negative error number, return it | |
450 | * (providing we aren't already returning an error); if it's | |
451 | * just a protocol response (i.e. non negative), that gets | |
452 | * processed above. | |
453 | */ | |
454 | if (!ret && rq->errors < 0) | |
455 | ret = rq->errors; | |
456 | ||
70e36ece FT |
457 | blk_rq_unmap_user(bio); |
458 | blk_put_request(rq); | |
459 | ||
460 | return ret; | |
461 | } | |
462 | ||
3d6392cf JA |
463 | static int bsg_complete_all_commands(struct bsg_device *bd) |
464 | { | |
465 | struct bsg_command *bc; | |
466 | int ret, tret; | |
467 | ||
468 | dprintk("%s: entered\n", bd->name); | |
469 | ||
470 | set_bit(BSG_F_BLOCK, &bd->flags); | |
471 | ||
472 | /* | |
473 | * wait for all commands to complete | |
474 | */ | |
475 | ret = 0; | |
476 | do { | |
25fd1643 | 477 | ret = bsg_io_schedule(bd); |
3d6392cf JA |
478 | /* |
479 | * look for -ENODATA specifically -- we'll sometimes get | |
480 | * -ERESTARTSYS when we've taken a signal, but we can't | |
481 | * return until we're done freeing the queue, so ignore | |
482 | * it. The signal will get handled when we're done freeing | |
483 | * the bsg_device. | |
484 | */ | |
485 | } while (ret != -ENODATA); | |
486 | ||
487 | /* | |
488 | * discard done commands | |
489 | */ | |
490 | ret = 0; | |
491 | do { | |
e7d72173 FT |
492 | spin_lock_irq(&bd->lock); |
493 | if (!bd->queued_cmds) { | |
494 | spin_unlock_irq(&bd->lock); | |
3d6392cf JA |
495 | break; |
496 | } | |
efba1a31 | 497 | spin_unlock_irq(&bd->lock); |
3d6392cf | 498 | |
e7d72173 FT |
499 | bc = bsg_get_done_cmd(bd); |
500 | if (IS_ERR(bc)) | |
501 | break; | |
502 | ||
2c9ecdf4 FT |
503 | tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio, |
504 | bc->bidi_bio); | |
3d6392cf JA |
505 | if (!ret) |
506 | ret = tret; | |
507 | ||
508 | bsg_free_command(bc); | |
509 | } while (1); | |
510 | ||
511 | return ret; | |
512 | } | |
513 | ||
25fd1643 | 514 | static int |
e7d72173 FT |
515 | __bsg_read(char __user *buf, size_t count, struct bsg_device *bd, |
516 | const struct iovec *iov, ssize_t *bytes_read) | |
3d6392cf JA |
517 | { |
518 | struct bsg_command *bc; | |
519 | int nr_commands, ret; | |
520 | ||
70e36ece | 521 | if (count % sizeof(struct sg_io_v4)) |
3d6392cf JA |
522 | return -EINVAL; |
523 | ||
524 | ret = 0; | |
70e36ece | 525 | nr_commands = count / sizeof(struct sg_io_v4); |
3d6392cf | 526 | while (nr_commands) { |
e7d72173 | 527 | bc = bsg_get_done_cmd(bd); |
3d6392cf JA |
528 | if (IS_ERR(bc)) { |
529 | ret = PTR_ERR(bc); | |
530 | break; | |
531 | } | |
532 | ||
533 | /* | |
534 | * this is the only case where we need to copy data back | |
535 | * after completing the request. so do that here, | |
536 | * bsg_complete_work() cannot do that for us | |
537 | */ | |
2c9ecdf4 FT |
538 | ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio, |
539 | bc->bidi_bio); | |
3d6392cf | 540 | |
25fd1643 | 541 | if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr))) |
3d6392cf JA |
542 | ret = -EFAULT; |
543 | ||
544 | bsg_free_command(bc); | |
545 | ||
546 | if (ret) | |
547 | break; | |
548 | ||
70e36ece FT |
549 | buf += sizeof(struct sg_io_v4); |
550 | *bytes_read += sizeof(struct sg_io_v4); | |
3d6392cf JA |
551 | nr_commands--; |
552 | } | |
553 | ||
554 | return ret; | |
555 | } | |
556 | ||
557 | static inline void bsg_set_block(struct bsg_device *bd, struct file *file) | |
558 | { | |
559 | if (file->f_flags & O_NONBLOCK) | |
560 | clear_bit(BSG_F_BLOCK, &bd->flags); | |
561 | else | |
562 | set_bit(BSG_F_BLOCK, &bd->flags); | |
563 | } | |
564 | ||
565 | static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file) | |
566 | { | |
567 | if (file->f_mode & FMODE_WRITE) | |
568 | set_bit(BSG_F_WRITE_PERM, &bd->flags); | |
569 | else | |
570 | clear_bit(BSG_F_WRITE_PERM, &bd->flags); | |
571 | } | |
572 | ||
25fd1643 JA |
573 | /* |
574 | * Check if the error is a "real" error that we should return. | |
575 | */ | |
3d6392cf JA |
576 | static inline int err_block_err(int ret) |
577 | { | |
578 | if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN) | |
579 | return 1; | |
580 | ||
581 | return 0; | |
582 | } | |
583 | ||
584 | static ssize_t | |
585 | bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
586 | { | |
587 | struct bsg_device *bd = file->private_data; | |
588 | int ret; | |
589 | ssize_t bytes_read; | |
590 | ||
9e69fbb5 | 591 | dprintk("%s: read %Zd bytes\n", bd->name, count); |
3d6392cf JA |
592 | |
593 | bsg_set_block(bd, file); | |
594 | bytes_read = 0; | |
e7d72173 | 595 | ret = __bsg_read(buf, count, bd, NULL, &bytes_read); |
3d6392cf JA |
596 | *ppos = bytes_read; |
597 | ||
598 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
599 | bytes_read = ret; | |
600 | ||
601 | return bytes_read; | |
602 | } | |
603 | ||
25fd1643 JA |
604 | static int __bsg_write(struct bsg_device *bd, const char __user *buf, |
605 | size_t count, ssize_t *bytes_written) | |
3d6392cf JA |
606 | { |
607 | struct bsg_command *bc; | |
608 | struct request *rq; | |
609 | int ret, nr_commands; | |
610 | ||
70e36ece | 611 | if (count % sizeof(struct sg_io_v4)) |
3d6392cf JA |
612 | return -EINVAL; |
613 | ||
70e36ece | 614 | nr_commands = count / sizeof(struct sg_io_v4); |
3d6392cf JA |
615 | rq = NULL; |
616 | bc = NULL; | |
617 | ret = 0; | |
618 | while (nr_commands) { | |
165125e1 | 619 | struct request_queue *q = bd->queue; |
3d6392cf | 620 | |
e7d72173 | 621 | bc = bsg_alloc_command(bd); |
3d6392cf JA |
622 | if (IS_ERR(bc)) { |
623 | ret = PTR_ERR(bc); | |
624 | bc = NULL; | |
625 | break; | |
626 | } | |
627 | ||
3d6392cf JA |
628 | if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) { |
629 | ret = -EFAULT; | |
630 | break; | |
631 | } | |
632 | ||
633 | /* | |
634 | * get a request, fill in the blanks, and add to request queue | |
635 | */ | |
70e36ece | 636 | rq = bsg_map_hdr(bd, &bc->hdr); |
3d6392cf JA |
637 | if (IS_ERR(rq)) { |
638 | ret = PTR_ERR(rq); | |
639 | rq = NULL; | |
640 | break; | |
641 | } | |
642 | ||
643 | bsg_add_command(bd, q, bc, rq); | |
644 | bc = NULL; | |
645 | rq = NULL; | |
646 | nr_commands--; | |
70e36ece | 647 | buf += sizeof(struct sg_io_v4); |
25fd1643 | 648 | *bytes_written += sizeof(struct sg_io_v4); |
3d6392cf JA |
649 | } |
650 | ||
3d6392cf JA |
651 | if (bc) |
652 | bsg_free_command(bc); | |
653 | ||
654 | return ret; | |
655 | } | |
656 | ||
657 | static ssize_t | |
658 | bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) | |
659 | { | |
660 | struct bsg_device *bd = file->private_data; | |
25fd1643 | 661 | ssize_t bytes_written; |
3d6392cf JA |
662 | int ret; |
663 | ||
9e69fbb5 | 664 | dprintk("%s: write %Zd bytes\n", bd->name, count); |
3d6392cf JA |
665 | |
666 | bsg_set_block(bd, file); | |
667 | bsg_set_write_perm(bd, file); | |
668 | ||
25fd1643 JA |
669 | bytes_written = 0; |
670 | ret = __bsg_write(bd, buf, count, &bytes_written); | |
671 | *ppos = bytes_written; | |
3d6392cf JA |
672 | |
673 | /* | |
674 | * return bytes written on non-fatal errors | |
675 | */ | |
25fd1643 JA |
676 | if (!bytes_written || (bytes_written && err_block_err(ret))) |
677 | bytes_written = ret; | |
3d6392cf | 678 | |
25fd1643 JA |
679 | dprintk("%s: returning %Zd\n", bd->name, bytes_written); |
680 | return bytes_written; | |
3d6392cf JA |
681 | } |
682 | ||
3d6392cf JA |
683 | static struct bsg_device *bsg_alloc_device(void) |
684 | { | |
3d6392cf | 685 | struct bsg_device *bd; |
3d6392cf JA |
686 | |
687 | bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL); | |
688 | if (unlikely(!bd)) | |
689 | return NULL; | |
690 | ||
691 | spin_lock_init(&bd->lock); | |
692 | ||
5309cb38 | 693 | bd->max_queue = BSG_DEFAULT_CMDS; |
3d6392cf JA |
694 | |
695 | INIT_LIST_HEAD(&bd->busy_list); | |
696 | INIT_LIST_HEAD(&bd->done_list); | |
697 | INIT_HLIST_NODE(&bd->dev_list); | |
698 | ||
699 | init_waitqueue_head(&bd->wq_free); | |
700 | init_waitqueue_head(&bd->wq_done); | |
701 | return bd; | |
3d6392cf JA |
702 | } |
703 | ||
704 | static int bsg_put_device(struct bsg_device *bd) | |
705 | { | |
706 | int ret = 0; | |
707 | ||
708 | mutex_lock(&bsg_mutex); | |
709 | ||
710 | if (!atomic_dec_and_test(&bd->ref_count)) | |
711 | goto out; | |
712 | ||
713 | dprintk("%s: tearing down\n", bd->name); | |
714 | ||
715 | /* | |
716 | * close can always block | |
717 | */ | |
718 | set_bit(BSG_F_BLOCK, &bd->flags); | |
719 | ||
720 | /* | |
721 | * correct error detection baddies here again. it's the responsibility | |
722 | * of the app to properly reap commands before close() if it wants | |
723 | * fool-proof error detection | |
724 | */ | |
725 | ret = bsg_complete_all_commands(bd); | |
726 | ||
727 | blk_put_queue(bd->queue); | |
728 | hlist_del(&bd->dev_list); | |
5309cb38 | 729 | kfree(bd); |
3d6392cf JA |
730 | out: |
731 | mutex_unlock(&bsg_mutex); | |
732 | return ret; | |
733 | } | |
734 | ||
735 | static struct bsg_device *bsg_add_device(struct inode *inode, | |
d351af01 | 736 | struct request_queue *rq, |
3d6392cf JA |
737 | struct file *file) |
738 | { | |
25fd1643 | 739 | struct bsg_device *bd; |
3d6392cf JA |
740 | #ifdef BSG_DEBUG |
741 | unsigned char buf[32]; | |
742 | #endif | |
743 | ||
744 | bd = bsg_alloc_device(); | |
745 | if (!bd) | |
746 | return ERR_PTR(-ENOMEM); | |
747 | ||
d351af01 FT |
748 | bd->queue = rq; |
749 | kobject_get(&rq->kobj); | |
3d6392cf JA |
750 | bsg_set_block(bd, file); |
751 | ||
752 | atomic_set(&bd->ref_count, 1); | |
753 | bd->minor = iminor(inode); | |
754 | mutex_lock(&bsg_mutex); | |
1c1133e1 | 755 | hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(bd->minor)); |
3d6392cf | 756 | |
d351af01 | 757 | strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1); |
3d6392cf | 758 | dprintk("bound to <%s>, max queue %d\n", |
9e69fbb5 | 759 | format_dev_t(buf, inode->i_rdev), bd->max_queue); |
3d6392cf JA |
760 | |
761 | mutex_unlock(&bsg_mutex); | |
762 | return bd; | |
763 | } | |
764 | ||
765 | static struct bsg_device *__bsg_get_device(int minor) | |
766 | { | |
3d6392cf JA |
767 | struct bsg_device *bd = NULL; |
768 | struct hlist_node *entry; | |
769 | ||
770 | mutex_lock(&bsg_mutex); | |
771 | ||
1c1133e1 | 772 | hlist_for_each(entry, bsg_dev_idx_hash(minor)) { |
3d6392cf JA |
773 | bd = hlist_entry(entry, struct bsg_device, dev_list); |
774 | if (bd->minor == minor) { | |
775 | atomic_inc(&bd->ref_count); | |
776 | break; | |
777 | } | |
778 | ||
779 | bd = NULL; | |
780 | } | |
781 | ||
782 | mutex_unlock(&bsg_mutex); | |
783 | return bd; | |
784 | } | |
785 | ||
786 | static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file) | |
787 | { | |
598443a2 FT |
788 | struct bsg_device *bd; |
789 | struct bsg_class_device *bcd; | |
3d6392cf | 790 | |
598443a2 | 791 | bd = __bsg_get_device(iminor(inode)); |
3d6392cf JA |
792 | if (bd) |
793 | return bd; | |
794 | ||
795 | /* | |
796 | * find the class device | |
797 | */ | |
3d6392cf | 798 | mutex_lock(&bsg_mutex); |
598443a2 | 799 | bcd = idr_find(&bsg_minor_idr, iminor(inode)); |
3d6392cf JA |
800 | mutex_unlock(&bsg_mutex); |
801 | ||
802 | if (!bcd) | |
803 | return ERR_PTR(-ENODEV); | |
804 | ||
d351af01 | 805 | return bsg_add_device(inode, bcd->queue, file); |
3d6392cf JA |
806 | } |
807 | ||
808 | static int bsg_open(struct inode *inode, struct file *file) | |
809 | { | |
810 | struct bsg_device *bd = bsg_get_device(inode, file); | |
811 | ||
812 | if (IS_ERR(bd)) | |
813 | return PTR_ERR(bd); | |
814 | ||
815 | file->private_data = bd; | |
816 | return 0; | |
817 | } | |
818 | ||
819 | static int bsg_release(struct inode *inode, struct file *file) | |
820 | { | |
821 | struct bsg_device *bd = file->private_data; | |
822 | ||
823 | file->private_data = NULL; | |
824 | return bsg_put_device(bd); | |
825 | } | |
826 | ||
827 | static unsigned int bsg_poll(struct file *file, poll_table *wait) | |
828 | { | |
829 | struct bsg_device *bd = file->private_data; | |
830 | unsigned int mask = 0; | |
831 | ||
832 | poll_wait(file, &bd->wq_done, wait); | |
833 | poll_wait(file, &bd->wq_free, wait); | |
834 | ||
835 | spin_lock_irq(&bd->lock); | |
836 | if (!list_empty(&bd->done_list)) | |
837 | mask |= POLLIN | POLLRDNORM; | |
838 | if (bd->queued_cmds >= bd->max_queue) | |
839 | mask |= POLLOUT; | |
840 | spin_unlock_irq(&bd->lock); | |
841 | ||
842 | return mask; | |
843 | } | |
844 | ||
25fd1643 | 845 | static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg) |
3d6392cf JA |
846 | { |
847 | struct bsg_device *bd = file->private_data; | |
848 | int __user *uarg = (int __user *) arg; | |
2d507a01 | 849 | int ret; |
3d6392cf | 850 | |
3d6392cf JA |
851 | switch (cmd) { |
852 | /* | |
853 | * our own ioctls | |
854 | */ | |
855 | case SG_GET_COMMAND_Q: | |
856 | return put_user(bd->max_queue, uarg); | |
5309cb38 | 857 | case SG_SET_COMMAND_Q: { |
3d6392cf JA |
858 | int queue; |
859 | ||
860 | if (get_user(queue, uarg)) | |
861 | return -EFAULT; | |
5309cb38 | 862 | if (queue < 1) |
3d6392cf JA |
863 | return -EINVAL; |
864 | ||
5309cb38 | 865 | spin_lock_irq(&bd->lock); |
3d6392cf | 866 | bd->max_queue = queue; |
5309cb38 | 867 | spin_unlock_irq(&bd->lock); |
3d6392cf JA |
868 | return 0; |
869 | } | |
870 | ||
871 | /* | |
872 | * SCSI/sg ioctls | |
873 | */ | |
874 | case SG_GET_VERSION_NUM: | |
875 | case SCSI_IOCTL_GET_IDLUN: | |
876 | case SCSI_IOCTL_GET_BUS_NUMBER: | |
877 | case SG_SET_TIMEOUT: | |
878 | case SG_GET_TIMEOUT: | |
879 | case SG_GET_RESERVED_SIZE: | |
880 | case SG_SET_RESERVED_SIZE: | |
881 | case SG_EMULATED_HOST: | |
3d6392cf JA |
882 | case SCSI_IOCTL_SEND_COMMAND: { |
883 | void __user *uarg = (void __user *) arg; | |
d351af01 | 884 | return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg); |
3d6392cf | 885 | } |
10e8855b FT |
886 | case SG_IO: { |
887 | struct request *rq; | |
2c9ecdf4 | 888 | struct bio *bio, *bidi_bio = NULL; |
10e8855b FT |
889 | struct sg_io_v4 hdr; |
890 | ||
891 | if (copy_from_user(&hdr, uarg, sizeof(hdr))) | |
892 | return -EFAULT; | |
893 | ||
894 | rq = bsg_map_hdr(bd, &hdr); | |
895 | if (IS_ERR(rq)) | |
896 | return PTR_ERR(rq); | |
897 | ||
898 | bio = rq->bio; | |
2c9ecdf4 FT |
899 | if (rq->next_rq) |
900 | bidi_bio = rq->next_rq->bio; | |
d351af01 | 901 | blk_execute_rq(bd->queue, NULL, rq, 0); |
2d507a01 | 902 | ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio); |
10e8855b FT |
903 | |
904 | if (copy_to_user(uarg, &hdr, sizeof(hdr))) | |
905 | return -EFAULT; | |
b711afa6 | 906 | |
2d507a01 | 907 | return ret; |
10e8855b | 908 | } |
3d6392cf JA |
909 | /* |
910 | * block device ioctls | |
911 | */ | |
912 | default: | |
913 | #if 0 | |
914 | return ioctl_by_bdev(bd->bdev, cmd, arg); | |
915 | #else | |
916 | return -ENOTTY; | |
917 | #endif | |
918 | } | |
919 | } | |
920 | ||
7344be05 | 921 | static const struct file_operations bsg_fops = { |
3d6392cf JA |
922 | .read = bsg_read, |
923 | .write = bsg_write, | |
924 | .poll = bsg_poll, | |
925 | .open = bsg_open, | |
926 | .release = bsg_release, | |
25fd1643 | 927 | .unlocked_ioctl = bsg_ioctl, |
3d6392cf JA |
928 | .owner = THIS_MODULE, |
929 | }; | |
930 | ||
d351af01 | 931 | void bsg_unregister_queue(struct request_queue *q) |
3d6392cf | 932 | { |
d351af01 | 933 | struct bsg_class_device *bcd = &q->bsg_dev; |
3d6392cf | 934 | |
df468820 FT |
935 | if (!bcd->class_dev) |
936 | return; | |
3d6392cf JA |
937 | |
938 | mutex_lock(&bsg_mutex); | |
598443a2 | 939 | idr_remove(&bsg_minor_idr, bcd->minor); |
d351af01 | 940 | sysfs_remove_link(&q->kobj, "bsg"); |
39dca558 JB |
941 | class_device_unregister(bcd->class_dev); |
942 | put_device(bcd->dev); | |
3d6392cf | 943 | bcd->class_dev = NULL; |
39dca558 | 944 | bcd->dev = NULL; |
3d6392cf JA |
945 | mutex_unlock(&bsg_mutex); |
946 | } | |
4cf0723a | 947 | EXPORT_SYMBOL_GPL(bsg_unregister_queue); |
3d6392cf | 948 | |
39dca558 JB |
949 | int bsg_register_queue(struct request_queue *q, struct device *gdev, |
950 | const char *name) | |
3d6392cf | 951 | { |
598443a2 | 952 | struct bsg_class_device *bcd; |
3d6392cf | 953 | dev_t dev; |
598443a2 | 954 | int ret, minor; |
4e2872d6 | 955 | struct class_device *class_dev = NULL; |
39dca558 JB |
956 | const char *devname; |
957 | ||
958 | if (name) | |
959 | devname = name; | |
960 | else | |
961 | devname = gdev->bus_id; | |
3d6392cf JA |
962 | |
963 | /* | |
964 | * we need a proper transport to send commands, not a stacked device | |
965 | */ | |
966 | if (!q->request_fn) | |
967 | return 0; | |
968 | ||
d351af01 | 969 | bcd = &q->bsg_dev; |
3d6392cf | 970 | memset(bcd, 0, sizeof(*bcd)); |
3d6392cf JA |
971 | |
972 | mutex_lock(&bsg_mutex); | |
292b7f27 | 973 | |
598443a2 FT |
974 | ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL); |
975 | if (!ret) { | |
976 | ret = -ENOMEM; | |
977 | goto unlock; | |
292b7f27 FT |
978 | } |
979 | ||
598443a2 FT |
980 | ret = idr_get_new(&bsg_minor_idr, bcd, &minor); |
981 | if (ret < 0) | |
982 | goto unlock; | |
292b7f27 | 983 | |
598443a2 FT |
984 | if (minor >= BSG_MAX_DEVS) { |
985 | printk(KERN_ERR "bsg: too many bsg devices\n"); | |
986 | ret = -EINVAL; | |
987 | goto remove_idr; | |
988 | } | |
989 | ||
990 | bcd->minor = minor; | |
d351af01 | 991 | bcd->queue = q; |
39dca558 | 992 | bcd->dev = get_device(gdev); |
46f6ef4a | 993 | dev = MKDEV(bsg_major, bcd->minor); |
39dca558 JB |
994 | class_dev = class_device_create(bsg_class, NULL, dev, gdev, "%s", |
995 | devname); | |
4e2872d6 FT |
996 | if (IS_ERR(class_dev)) { |
997 | ret = PTR_ERR(class_dev); | |
598443a2 | 998 | goto put_dev; |
4e2872d6 FT |
999 | } |
1000 | bcd->class_dev = class_dev; | |
1001 | ||
abce891a | 1002 | if (q->kobj.sd) { |
4e2872d6 FT |
1003 | ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg"); |
1004 | if (ret) | |
598443a2 | 1005 | goto unregister_class_dev; |
4e2872d6 FT |
1006 | } |
1007 | ||
3d6392cf JA |
1008 | mutex_unlock(&bsg_mutex); |
1009 | return 0; | |
6826ee4f | 1010 | |
598443a2 | 1011 | unregister_class_dev: |
6826ee4f | 1012 | class_device_unregister(class_dev); |
598443a2 | 1013 | put_dev: |
39dca558 | 1014 | put_device(gdev); |
598443a2 FT |
1015 | remove_idr: |
1016 | idr_remove(&bsg_minor_idr, minor); | |
1017 | unlock: | |
264a0472 | 1018 | mutex_unlock(&bsg_mutex); |
4e2872d6 FT |
1019 | return ret; |
1020 | } | |
4cf0723a | 1021 | EXPORT_SYMBOL_GPL(bsg_register_queue); |
4e2872d6 | 1022 | |
7e7654a9 | 1023 | static struct cdev bsg_cdev; |
292b7f27 | 1024 | |
3d6392cf JA |
1025 | static int __init bsg_init(void) |
1026 | { | |
1027 | int ret, i; | |
46f6ef4a | 1028 | dev_t devid; |
3d6392cf | 1029 | |
5309cb38 | 1030 | bsg_cmd_cachep = kmem_cache_create("bsg_cmd", |
20c2df83 | 1031 | sizeof(struct bsg_command), 0, 0, NULL); |
5309cb38 JA |
1032 | if (!bsg_cmd_cachep) { |
1033 | printk(KERN_ERR "bsg: failed creating slab cache\n"); | |
1034 | return -ENOMEM; | |
1035 | } | |
1036 | ||
25fd1643 | 1037 | for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++) |
3d6392cf JA |
1038 | INIT_HLIST_HEAD(&bsg_device_list[i]); |
1039 | ||
1040 | bsg_class = class_create(THIS_MODULE, "bsg"); | |
5309cb38 | 1041 | if (IS_ERR(bsg_class)) { |
9b9f770c FT |
1042 | ret = PTR_ERR(bsg_class); |
1043 | goto destroy_kmemcache; | |
5309cb38 | 1044 | } |
3d6392cf | 1045 | |
46f6ef4a | 1046 | ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg"); |
9b9f770c FT |
1047 | if (ret) |
1048 | goto destroy_bsg_class; | |
292b7f27 | 1049 | |
46f6ef4a JA |
1050 | bsg_major = MAJOR(devid); |
1051 | ||
292b7f27 | 1052 | cdev_init(&bsg_cdev, &bsg_fops); |
46f6ef4a | 1053 | ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS); |
9b9f770c FT |
1054 | if (ret) |
1055 | goto unregister_chrdev; | |
3d6392cf | 1056 | |
5d3a8cd3 | 1057 | printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION |
0ed081ce | 1058 | " loaded (major %d)\n", bsg_major); |
3d6392cf | 1059 | return 0; |
9b9f770c FT |
1060 | unregister_chrdev: |
1061 | unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS); | |
1062 | destroy_bsg_class: | |
1063 | class_destroy(bsg_class); | |
1064 | destroy_kmemcache: | |
1065 | kmem_cache_destroy(bsg_cmd_cachep); | |
1066 | return ret; | |
3d6392cf JA |
1067 | } |
1068 | ||
1069 | MODULE_AUTHOR("Jens Axboe"); | |
0ed081ce | 1070 | MODULE_DESCRIPTION(BSG_DESCRIPTION); |
3d6392cf JA |
1071 | MODULE_LICENSE("GPL"); |
1072 | ||
4e2872d6 | 1073 | device_initcall(bsg_init); |