]>
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 | ||
36 | struct bsg_command; | |
37 | ||
38 | struct bsg_device { | |
39 | struct gendisk *disk; | |
40 | request_queue_t *queue; | |
41 | spinlock_t lock; | |
42 | struct list_head busy_list; | |
43 | struct list_head done_list; | |
44 | struct hlist_node dev_list; | |
45 | atomic_t ref_count; | |
46 | int minor; | |
47 | int queued_cmds; | |
48 | int done_cmds; | |
49 | unsigned long *cmd_bitmap; | |
50 | struct bsg_command *cmd_map; | |
51 | wait_queue_head_t wq_done; | |
52 | wait_queue_head_t wq_free; | |
53 | char name[BDEVNAME_SIZE]; | |
54 | int max_queue; | |
55 | unsigned long flags; | |
56 | }; | |
57 | ||
58 | enum { | |
59 | BSG_F_BLOCK = 1, | |
60 | BSG_F_WRITE_PERM = 2, | |
61 | }; | |
62 | ||
63 | /* | |
64 | * command allocation bitmap defines | |
65 | */ | |
66 | #define BSG_CMDS_PAGE_ORDER (1) | |
67 | #define BSG_CMDS_PER_LONG (sizeof(unsigned long) * 8) | |
68 | #define BSG_CMDS_MASK (BSG_CMDS_PER_LONG - 1) | |
69 | #define BSG_CMDS_BYTES (PAGE_SIZE * (1 << BSG_CMDS_PAGE_ORDER)) | |
70 | #define BSG_CMDS (BSG_CMDS_BYTES / sizeof(struct bsg_command)) | |
71 | ||
72 | #undef BSG_DEBUG | |
73 | ||
74 | #ifdef BSG_DEBUG | |
75 | #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args) | |
76 | #else | |
77 | #define dprintk(fmt, args...) | |
78 | #endif | |
79 | ||
80 | #define list_entry_bc(entry) list_entry((entry), struct bsg_command, list) | |
81 | ||
82 | /* | |
83 | * just for testing | |
84 | */ | |
85 | #define BSG_MAJOR (240) | |
86 | ||
87 | static DEFINE_MUTEX(bsg_mutex); | |
88 | static int bsg_device_nr; | |
89 | ||
90 | #define BSG_LIST_SIZE (8) | |
91 | #define bsg_list_idx(minor) ((minor) & (BSG_LIST_SIZE - 1)) | |
92 | static struct hlist_head bsg_device_list[BSG_LIST_SIZE]; | |
93 | ||
94 | static struct class *bsg_class; | |
95 | static LIST_HEAD(bsg_class_list); | |
96 | ||
97 | /* | |
98 | * our internal command type | |
99 | */ | |
100 | struct bsg_command { | |
101 | struct bsg_device *bd; | |
102 | struct list_head list; | |
103 | struct request *rq; | |
104 | struct bio *bio; | |
105 | int err; | |
106 | struct sg_io_hdr hdr; | |
107 | struct sg_io_hdr __user *uhdr; | |
108 | char sense[SCSI_SENSE_BUFFERSIZE]; | |
109 | }; | |
110 | ||
111 | static void bsg_free_command(struct bsg_command *bc) | |
112 | { | |
113 | struct bsg_device *bd = bc->bd; | |
114 | unsigned long bitnr = bc - bd->cmd_map; | |
115 | unsigned long flags; | |
116 | ||
117 | dprintk("%s: command bit offset %lu\n", bd->name, bitnr); | |
118 | ||
119 | spin_lock_irqsave(&bd->lock, flags); | |
120 | bd->queued_cmds--; | |
121 | __clear_bit(bitnr, bd->cmd_bitmap); | |
122 | spin_unlock_irqrestore(&bd->lock, flags); | |
123 | ||
124 | wake_up(&bd->wq_free); | |
125 | } | |
126 | ||
127 | static struct bsg_command *__bsg_alloc_command(struct bsg_device *bd) | |
128 | { | |
129 | struct bsg_command *bc = NULL; | |
130 | unsigned long *map; | |
131 | int free_nr; | |
132 | ||
133 | spin_lock_irq(&bd->lock); | |
134 | ||
135 | if (bd->queued_cmds >= bd->max_queue) | |
136 | goto out; | |
137 | ||
138 | for (free_nr = 0, map = bd->cmd_bitmap; *map == ~0UL; map++) | |
139 | free_nr += BSG_CMDS_PER_LONG; | |
140 | ||
141 | BUG_ON(*map == ~0UL); | |
142 | ||
143 | bd->queued_cmds++; | |
144 | free_nr += ffz(*map); | |
145 | __set_bit(free_nr, bd->cmd_bitmap); | |
146 | spin_unlock_irq(&bd->lock); | |
147 | ||
148 | bc = bd->cmd_map + free_nr; | |
149 | memset(bc, 0, sizeof(*bc)); | |
150 | bc->bd = bd; | |
151 | INIT_LIST_HEAD(&bc->list); | |
152 | dprintk("%s: returning free cmd %p (bit %d)\n", bd->name, bc, free_nr); | |
153 | return bc; | |
154 | out: | |
155 | dprintk("%s: failed (depth %d)\n", bd->name, bd->queued_cmds); | |
156 | spin_unlock_irq(&bd->lock); | |
157 | return bc; | |
158 | } | |
159 | ||
160 | static inline void | |
161 | bsg_del_done_cmd(struct bsg_device *bd, struct bsg_command *bc) | |
162 | { | |
163 | bd->done_cmds--; | |
164 | list_del(&bc->list); | |
165 | } | |
166 | ||
167 | static inline void | |
168 | bsg_add_done_cmd(struct bsg_device *bd, struct bsg_command *bc) | |
169 | { | |
170 | bd->done_cmds++; | |
171 | list_add_tail(&bc->list, &bd->done_list); | |
172 | wake_up(&bd->wq_done); | |
173 | } | |
174 | ||
175 | static inline int bsg_io_schedule(struct bsg_device *bd, int state) | |
176 | { | |
177 | DEFINE_WAIT(wait); | |
178 | int ret = 0; | |
179 | ||
180 | spin_lock_irq(&bd->lock); | |
181 | ||
182 | BUG_ON(bd->done_cmds > bd->queued_cmds); | |
183 | ||
184 | /* | |
185 | * -ENOSPC or -ENODATA? I'm going for -ENODATA, meaning "I have no | |
186 | * work to do", even though we return -ENOSPC after this same test | |
187 | * during bsg_write() -- there, it means our buffer can't have more | |
188 | * bsg_commands added to it, thus has no space left. | |
189 | */ | |
190 | if (bd->done_cmds == bd->queued_cmds) { | |
191 | ret = -ENODATA; | |
192 | goto unlock; | |
193 | } | |
194 | ||
195 | if (!test_bit(BSG_F_BLOCK, &bd->flags)) { | |
196 | ret = -EAGAIN; | |
197 | goto unlock; | |
198 | } | |
199 | ||
200 | prepare_to_wait(&bd->wq_done, &wait, state); | |
201 | spin_unlock_irq(&bd->lock); | |
202 | io_schedule(); | |
203 | finish_wait(&bd->wq_done, &wait); | |
204 | ||
205 | if ((state == TASK_INTERRUPTIBLE) && signal_pending(current)) | |
206 | ret = -ERESTARTSYS; | |
207 | ||
208 | return ret; | |
209 | unlock: | |
210 | spin_unlock_irq(&bd->lock); | |
211 | return ret; | |
212 | } | |
213 | ||
214 | /* | |
215 | * get a new free command, blocking if needed and specified | |
216 | */ | |
217 | static struct bsg_command *bsg_get_command(struct bsg_device *bd) | |
218 | { | |
219 | struct bsg_command *bc; | |
220 | int ret; | |
221 | ||
222 | do { | |
223 | bc = __bsg_alloc_command(bd); | |
224 | if (bc) | |
225 | break; | |
226 | ||
227 | ret = bsg_io_schedule(bd, TASK_INTERRUPTIBLE); | |
228 | if (ret) { | |
229 | bc = ERR_PTR(ret); | |
230 | break; | |
231 | } | |
232 | ||
233 | } while (1); | |
234 | ||
235 | return bc; | |
236 | } | |
237 | ||
238 | /* | |
239 | * Check if sg_io_hdr from user is allowed and valid | |
240 | */ | |
241 | static int | |
242 | bsg_validate_sghdr(request_queue_t *q, struct sg_io_hdr *hdr, int *rw) | |
243 | { | |
244 | if (hdr->interface_id != 'S') | |
245 | return -EINVAL; | |
246 | if (hdr->cmd_len > BLK_MAX_CDB) | |
247 | return -EINVAL; | |
248 | if (hdr->dxfer_len > (q->max_sectors << 9)) | |
249 | return -EIO; | |
250 | ||
251 | /* | |
252 | * looks sane, if no data then it should be fine from our POV | |
253 | */ | |
254 | if (!hdr->dxfer_len) | |
255 | return 0; | |
256 | ||
257 | switch (hdr->dxfer_direction) { | |
258 | case SG_DXFER_TO_FROM_DEV: | |
259 | case SG_DXFER_FROM_DEV: | |
260 | *rw = READ; | |
261 | break; | |
262 | case SG_DXFER_TO_DEV: | |
263 | *rw = WRITE; | |
264 | break; | |
265 | default: | |
266 | return -EINVAL; | |
267 | } | |
268 | ||
269 | return 0; | |
270 | } | |
271 | ||
272 | /* | |
273 | * map sg_io_hdr to a request. for scatter-gather sg_io_hdr, we map | |
274 | * each segment to a bio and string multiple bio's to the request | |
275 | */ | |
276 | static struct request * | |
277 | bsg_map_hdr(struct bsg_device *bd, int rw, struct sg_io_hdr *hdr) | |
278 | { | |
279 | request_queue_t *q = bd->queue; | |
280 | struct sg_iovec iov; | |
281 | struct sg_iovec __user *u_iov; | |
282 | struct request *rq; | |
283 | int ret, i = 0; | |
284 | ||
285 | dprintk("map hdr %p/%d/%d\n", hdr->dxferp, hdr->dxfer_len, | |
286 | hdr->iovec_count); | |
287 | ||
288 | ret = bsg_validate_sghdr(q, hdr, &rw); | |
289 | if (ret) | |
290 | return ERR_PTR(ret); | |
291 | ||
292 | /* | |
293 | * map scatter-gather elements seperately and string them to request | |
294 | */ | |
295 | rq = blk_get_request(q, rw, GFP_KERNEL); | |
296 | ret = blk_fill_sghdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM, | |
297 | &bd->flags)); | |
298 | if (ret) { | |
299 | blk_put_request(rq); | |
300 | return ERR_PTR(ret); | |
301 | } | |
302 | ||
303 | if (!hdr->iovec_count) { | |
304 | ret = blk_rq_map_user(q, rq, hdr->dxferp, hdr->dxfer_len); | |
305 | if (ret) | |
306 | goto out; | |
307 | } | |
308 | ||
309 | u_iov = hdr->dxferp; | |
310 | for (ret = 0, i = 0; i < hdr->iovec_count; i++, u_iov++) { | |
311 | if (copy_from_user(&iov, u_iov, sizeof(iov))) { | |
312 | ret = -EFAULT; | |
313 | break; | |
314 | } | |
315 | ||
316 | if (!iov.iov_len || !iov.iov_base) { | |
317 | ret = -EINVAL; | |
318 | break; | |
319 | } | |
320 | ||
321 | ret = blk_rq_map_user(q, rq, iov.iov_base, iov.iov_len); | |
322 | if (ret) | |
323 | break; | |
324 | } | |
325 | ||
326 | /* | |
327 | * bugger, cleanup | |
328 | */ | |
329 | if (ret) { | |
330 | out: | |
331 | dprintk("failed map at %d: %d\n", i, ret); | |
332 | blk_unmap_sghdr_rq(rq, hdr); | |
333 | rq = ERR_PTR(ret); | |
334 | } | |
335 | ||
336 | return rq; | |
337 | } | |
338 | ||
339 | /* | |
340 | * async completion call-back from the block layer, when scsi/ide/whatever | |
341 | * calls end_that_request_last() on a request | |
342 | */ | |
343 | static void bsg_rq_end_io(struct request *rq, int uptodate) | |
344 | { | |
345 | struct bsg_command *bc = rq->end_io_data; | |
346 | struct bsg_device *bd = bc->bd; | |
347 | unsigned long flags; | |
348 | ||
9e69fbb5 FT |
349 | dprintk("%s: finished rq %p bc %p, bio %p offset %d stat %d\n", |
350 | bd->name, rq, bc, bc->bio, bc - bd->cmd_map, uptodate); | |
3d6392cf JA |
351 | |
352 | bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration); | |
353 | ||
354 | spin_lock_irqsave(&bd->lock, flags); | |
355 | list_del(&bc->list); | |
356 | bsg_add_done_cmd(bd, bc); | |
357 | spin_unlock_irqrestore(&bd->lock, flags); | |
358 | } | |
359 | ||
360 | /* | |
361 | * do final setup of a 'bc' and submit the matching 'rq' to the block | |
362 | * layer for io | |
363 | */ | |
364 | static void bsg_add_command(struct bsg_device *bd, request_queue_t *q, | |
365 | struct bsg_command *bc, struct request *rq) | |
366 | { | |
367 | rq->sense = bc->sense; | |
368 | rq->sense_len = 0; | |
369 | ||
370 | /* | |
371 | * add bc command to busy queue and submit rq for io | |
372 | */ | |
373 | bc->rq = rq; | |
374 | bc->bio = rq->bio; | |
375 | bc->hdr.duration = jiffies; | |
376 | spin_lock_irq(&bd->lock); | |
377 | list_add_tail(&bc->list, &bd->busy_list); | |
378 | spin_unlock_irq(&bd->lock); | |
379 | ||
380 | dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc); | |
381 | ||
382 | rq->end_io_data = bc; | |
383 | blk_execute_rq_nowait(q, bd->disk, rq, 1, bsg_rq_end_io); | |
384 | } | |
385 | ||
386 | static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd) | |
387 | { | |
388 | struct bsg_command *bc = NULL; | |
389 | ||
390 | spin_lock_irq(&bd->lock); | |
391 | if (bd->done_cmds) { | |
392 | bc = list_entry_bc(bd->done_list.next); | |
393 | bsg_del_done_cmd(bd, bc); | |
394 | } | |
395 | spin_unlock_irq(&bd->lock); | |
396 | ||
397 | return bc; | |
398 | } | |
399 | ||
400 | /* | |
401 | * Get a finished command from the done list | |
402 | */ | |
403 | static struct bsg_command *__bsg_get_done_cmd(struct bsg_device *bd, int state) | |
404 | { | |
405 | struct bsg_command *bc; | |
406 | int ret; | |
407 | ||
408 | do { | |
409 | bc = bsg_next_done_cmd(bd); | |
410 | if (bc) | |
411 | break; | |
412 | ||
413 | ret = bsg_io_schedule(bd, state); | |
414 | if (ret) { | |
415 | bc = ERR_PTR(ret); | |
416 | break; | |
417 | } | |
418 | } while (1); | |
419 | ||
420 | dprintk("%s: returning done %p\n", bd->name, bc); | |
421 | ||
422 | return bc; | |
423 | } | |
424 | ||
425 | static struct bsg_command * | |
426 | bsg_get_done_cmd(struct bsg_device *bd, const struct iovec *iov) | |
427 | { | |
428 | return __bsg_get_done_cmd(bd, TASK_INTERRUPTIBLE); | |
429 | } | |
430 | ||
431 | static struct bsg_command * | |
432 | bsg_get_done_cmd_nosignals(struct bsg_device *bd) | |
433 | { | |
434 | return __bsg_get_done_cmd(bd, TASK_UNINTERRUPTIBLE); | |
435 | } | |
436 | ||
437 | static int bsg_complete_all_commands(struct bsg_device *bd) | |
438 | { | |
439 | struct bsg_command *bc; | |
440 | int ret, tret; | |
441 | ||
442 | dprintk("%s: entered\n", bd->name); | |
443 | ||
444 | set_bit(BSG_F_BLOCK, &bd->flags); | |
445 | ||
446 | /* | |
447 | * wait for all commands to complete | |
448 | */ | |
449 | ret = 0; | |
450 | do { | |
451 | ret = bsg_io_schedule(bd, TASK_UNINTERRUPTIBLE); | |
452 | /* | |
453 | * look for -ENODATA specifically -- we'll sometimes get | |
454 | * -ERESTARTSYS when we've taken a signal, but we can't | |
455 | * return until we're done freeing the queue, so ignore | |
456 | * it. The signal will get handled when we're done freeing | |
457 | * the bsg_device. | |
458 | */ | |
459 | } while (ret != -ENODATA); | |
460 | ||
461 | /* | |
462 | * discard done commands | |
463 | */ | |
464 | ret = 0; | |
465 | do { | |
466 | bc = bsg_get_done_cmd_nosignals(bd); | |
467 | ||
468 | /* | |
469 | * we _must_ complete before restarting, because | |
470 | * bsg_release can't handle this failing. | |
471 | */ | |
472 | if (PTR_ERR(bc) == -ERESTARTSYS) | |
473 | continue; | |
474 | if (IS_ERR(bc)) { | |
475 | ret = PTR_ERR(bc); | |
476 | break; | |
477 | } | |
478 | ||
479 | tret = blk_complete_sghdr_rq(bc->rq, &bc->hdr, bc->bio); | |
480 | if (!ret) | |
481 | ret = tret; | |
482 | ||
483 | bsg_free_command(bc); | |
484 | } while (1); | |
485 | ||
486 | return ret; | |
487 | } | |
488 | ||
489 | typedef struct bsg_command *(*bsg_command_callback)(struct bsg_device *bd, const struct iovec *iov); | |
490 | ||
491 | static ssize_t | |
492 | __bsg_read(char __user *buf, size_t count, bsg_command_callback get_bc, | |
493 | struct bsg_device *bd, const struct iovec *iov, ssize_t *bytes_read) | |
494 | { | |
495 | struct bsg_command *bc; | |
496 | int nr_commands, ret; | |
497 | ||
498 | if (count % sizeof(struct sg_io_hdr)) | |
499 | return -EINVAL; | |
500 | ||
501 | ret = 0; | |
502 | nr_commands = count / sizeof(struct sg_io_hdr); | |
503 | while (nr_commands) { | |
504 | bc = get_bc(bd, iov); | |
505 | if (IS_ERR(bc)) { | |
506 | ret = PTR_ERR(bc); | |
507 | break; | |
508 | } | |
509 | ||
510 | /* | |
511 | * this is the only case where we need to copy data back | |
512 | * after completing the request. so do that here, | |
513 | * bsg_complete_work() cannot do that for us | |
514 | */ | |
515 | ret = blk_complete_sghdr_rq(bc->rq, &bc->hdr, bc->bio); | |
516 | ||
517 | if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr))) | |
518 | ret = -EFAULT; | |
519 | ||
520 | bsg_free_command(bc); | |
521 | ||
522 | if (ret) | |
523 | break; | |
524 | ||
525 | buf += sizeof(struct sg_io_hdr); | |
526 | *bytes_read += sizeof(struct sg_io_hdr); | |
527 | nr_commands--; | |
528 | } | |
529 | ||
530 | return ret; | |
531 | } | |
532 | ||
533 | static inline void bsg_set_block(struct bsg_device *bd, struct file *file) | |
534 | { | |
535 | if (file->f_flags & O_NONBLOCK) | |
536 | clear_bit(BSG_F_BLOCK, &bd->flags); | |
537 | else | |
538 | set_bit(BSG_F_BLOCK, &bd->flags); | |
539 | } | |
540 | ||
541 | static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file) | |
542 | { | |
543 | if (file->f_mode & FMODE_WRITE) | |
544 | set_bit(BSG_F_WRITE_PERM, &bd->flags); | |
545 | else | |
546 | clear_bit(BSG_F_WRITE_PERM, &bd->flags); | |
547 | } | |
548 | ||
549 | static inline int err_block_err(int ret) | |
550 | { | |
551 | if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN) | |
552 | return 1; | |
553 | ||
554 | return 0; | |
555 | } | |
556 | ||
557 | static ssize_t | |
558 | bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) | |
559 | { | |
560 | struct bsg_device *bd = file->private_data; | |
561 | int ret; | |
562 | ssize_t bytes_read; | |
563 | ||
9e69fbb5 | 564 | dprintk("%s: read %Zd bytes\n", bd->name, count); |
3d6392cf JA |
565 | |
566 | bsg_set_block(bd, file); | |
567 | bytes_read = 0; | |
568 | ret = __bsg_read(buf, count, bsg_get_done_cmd, | |
569 | bd, NULL, &bytes_read); | |
570 | *ppos = bytes_read; | |
571 | ||
572 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
573 | bytes_read = ret; | |
574 | ||
575 | return bytes_read; | |
576 | } | |
577 | ||
578 | static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf, | |
579 | size_t count, ssize_t *bytes_read) | |
580 | { | |
581 | struct bsg_command *bc; | |
582 | struct request *rq; | |
583 | int ret, nr_commands; | |
584 | ||
585 | if (count % sizeof(struct sg_io_hdr)) | |
586 | return -EINVAL; | |
587 | ||
588 | nr_commands = count / sizeof(struct sg_io_hdr); | |
589 | rq = NULL; | |
590 | bc = NULL; | |
591 | ret = 0; | |
592 | while (nr_commands) { | |
593 | request_queue_t *q = bd->queue; | |
594 | int rw = READ; | |
595 | ||
596 | bc = bsg_get_command(bd); | |
597 | if (!bc) | |
598 | break; | |
599 | if (IS_ERR(bc)) { | |
600 | ret = PTR_ERR(bc); | |
601 | bc = NULL; | |
602 | break; | |
603 | } | |
604 | ||
605 | bc->uhdr = (struct sg_io_hdr __user *) buf; | |
606 | if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) { | |
607 | ret = -EFAULT; | |
608 | break; | |
609 | } | |
610 | ||
611 | /* | |
612 | * get a request, fill in the blanks, and add to request queue | |
613 | */ | |
614 | rq = bsg_map_hdr(bd, rw, &bc->hdr); | |
615 | if (IS_ERR(rq)) { | |
616 | ret = PTR_ERR(rq); | |
617 | rq = NULL; | |
618 | break; | |
619 | } | |
620 | ||
621 | bsg_add_command(bd, q, bc, rq); | |
622 | bc = NULL; | |
623 | rq = NULL; | |
624 | nr_commands--; | |
625 | buf += sizeof(struct sg_io_hdr); | |
626 | *bytes_read += sizeof(struct sg_io_hdr); | |
627 | } | |
628 | ||
629 | if (rq) | |
630 | blk_unmap_sghdr_rq(rq, &bc->hdr); | |
631 | if (bc) | |
632 | bsg_free_command(bc); | |
633 | ||
634 | return ret; | |
635 | } | |
636 | ||
637 | static ssize_t | |
638 | bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) | |
639 | { | |
640 | struct bsg_device *bd = file->private_data; | |
641 | ssize_t bytes_read; | |
642 | int ret; | |
643 | ||
9e69fbb5 | 644 | dprintk("%s: write %Zd bytes\n", bd->name, count); |
3d6392cf JA |
645 | |
646 | bsg_set_block(bd, file); | |
647 | bsg_set_write_perm(bd, file); | |
648 | ||
649 | bytes_read = 0; | |
650 | ret = __bsg_write(bd, buf, count, &bytes_read); | |
651 | *ppos = bytes_read; | |
652 | ||
653 | /* | |
654 | * return bytes written on non-fatal errors | |
655 | */ | |
656 | if (!bytes_read || (bytes_read && err_block_err(ret))) | |
657 | bytes_read = ret; | |
658 | ||
9e69fbb5 | 659 | dprintk("%s: returning %Zd\n", bd->name, bytes_read); |
3d6392cf JA |
660 | return bytes_read; |
661 | } | |
662 | ||
663 | static void bsg_free_device(struct bsg_device *bd) | |
664 | { | |
665 | if (bd->cmd_map) | |
666 | free_pages((unsigned long) bd->cmd_map, BSG_CMDS_PAGE_ORDER); | |
667 | ||
668 | kfree(bd->cmd_bitmap); | |
669 | kfree(bd); | |
670 | } | |
671 | ||
672 | static struct bsg_device *bsg_alloc_device(void) | |
673 | { | |
674 | struct bsg_command *cmd_map; | |
675 | unsigned long *cmd_bitmap; | |
676 | struct bsg_device *bd; | |
677 | int bits; | |
678 | ||
679 | bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL); | |
680 | if (unlikely(!bd)) | |
681 | return NULL; | |
682 | ||
683 | spin_lock_init(&bd->lock); | |
684 | ||
685 | bd->max_queue = BSG_CMDS; | |
686 | ||
687 | bits = (BSG_CMDS / BSG_CMDS_PER_LONG) + 1; | |
688 | cmd_bitmap = kzalloc(bits * sizeof(unsigned long), GFP_KERNEL); | |
689 | if (!cmd_bitmap) | |
690 | goto out_free_bd; | |
691 | bd->cmd_bitmap = cmd_bitmap; | |
692 | ||
693 | cmd_map = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO, | |
694 | BSG_CMDS_PAGE_ORDER); | |
695 | if (!cmd_map) | |
696 | goto out_free_bitmap; | |
697 | bd->cmd_map = cmd_map; | |
698 | ||
699 | INIT_LIST_HEAD(&bd->busy_list); | |
700 | INIT_LIST_HEAD(&bd->done_list); | |
701 | INIT_HLIST_NODE(&bd->dev_list); | |
702 | ||
703 | init_waitqueue_head(&bd->wq_free); | |
704 | init_waitqueue_head(&bd->wq_done); | |
705 | return bd; | |
706 | ||
707 | out_free_bitmap: | |
708 | kfree(cmd_bitmap); | |
709 | out_free_bd: | |
710 | kfree(bd); | |
711 | return NULL; | |
712 | } | |
713 | ||
714 | static int bsg_put_device(struct bsg_device *bd) | |
715 | { | |
716 | int ret = 0; | |
717 | ||
718 | mutex_lock(&bsg_mutex); | |
719 | ||
720 | if (!atomic_dec_and_test(&bd->ref_count)) | |
721 | goto out; | |
722 | ||
723 | dprintk("%s: tearing down\n", bd->name); | |
724 | ||
725 | /* | |
726 | * close can always block | |
727 | */ | |
728 | set_bit(BSG_F_BLOCK, &bd->flags); | |
729 | ||
730 | /* | |
731 | * correct error detection baddies here again. it's the responsibility | |
732 | * of the app to properly reap commands before close() if it wants | |
733 | * fool-proof error detection | |
734 | */ | |
735 | ret = bsg_complete_all_commands(bd); | |
736 | ||
737 | blk_put_queue(bd->queue); | |
738 | hlist_del(&bd->dev_list); | |
739 | bsg_free_device(bd); | |
740 | out: | |
741 | mutex_unlock(&bsg_mutex); | |
742 | return ret; | |
743 | } | |
744 | ||
745 | static struct bsg_device *bsg_add_device(struct inode *inode, | |
746 | struct gendisk *disk, | |
747 | struct file *file) | |
748 | { | |
749 | struct bsg_device *bd = NULL; | |
750 | #ifdef BSG_DEBUG | |
751 | unsigned char buf[32]; | |
752 | #endif | |
753 | ||
754 | bd = bsg_alloc_device(); | |
755 | if (!bd) | |
756 | return ERR_PTR(-ENOMEM); | |
757 | ||
758 | bd->disk = disk; | |
759 | bd->queue = disk->queue; | |
760 | kobject_get(&disk->queue->kobj); | |
761 | bsg_set_block(bd, file); | |
762 | ||
763 | atomic_set(&bd->ref_count, 1); | |
764 | bd->minor = iminor(inode); | |
765 | mutex_lock(&bsg_mutex); | |
766 | hlist_add_head(&bd->dev_list,&bsg_device_list[bsg_list_idx(bd->minor)]); | |
767 | ||
768 | strncpy(bd->name, disk->disk_name, sizeof(bd->name) - 1); | |
769 | dprintk("bound to <%s>, max queue %d\n", | |
9e69fbb5 | 770 | format_dev_t(buf, inode->i_rdev), bd->max_queue); |
3d6392cf JA |
771 | |
772 | mutex_unlock(&bsg_mutex); | |
773 | return bd; | |
774 | } | |
775 | ||
776 | static struct bsg_device *__bsg_get_device(int minor) | |
777 | { | |
778 | struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)]; | |
779 | struct bsg_device *bd = NULL; | |
780 | struct hlist_node *entry; | |
781 | ||
782 | mutex_lock(&bsg_mutex); | |
783 | ||
784 | hlist_for_each(entry, list) { | |
785 | bd = hlist_entry(entry, struct bsg_device, dev_list); | |
786 | if (bd->minor == minor) { | |
787 | atomic_inc(&bd->ref_count); | |
788 | break; | |
789 | } | |
790 | ||
791 | bd = NULL; | |
792 | } | |
793 | ||
794 | mutex_unlock(&bsg_mutex); | |
795 | return bd; | |
796 | } | |
797 | ||
798 | static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file) | |
799 | { | |
800 | struct bsg_device *bd = __bsg_get_device(iminor(inode)); | |
801 | struct bsg_class_device *bcd, *__bcd; | |
802 | ||
803 | if (bd) | |
804 | return bd; | |
805 | ||
806 | /* | |
807 | * find the class device | |
808 | */ | |
809 | bcd = NULL; | |
810 | mutex_lock(&bsg_mutex); | |
811 | list_for_each_entry(__bcd, &bsg_class_list, list) { | |
812 | if (__bcd->minor == iminor(inode)) { | |
813 | bcd = __bcd; | |
814 | break; | |
815 | } | |
816 | } | |
817 | mutex_unlock(&bsg_mutex); | |
818 | ||
819 | if (!bcd) | |
820 | return ERR_PTR(-ENODEV); | |
821 | ||
822 | return bsg_add_device(inode, bcd->disk, file); | |
823 | } | |
824 | ||
825 | static int bsg_open(struct inode *inode, struct file *file) | |
826 | { | |
827 | struct bsg_device *bd = bsg_get_device(inode, file); | |
828 | ||
829 | if (IS_ERR(bd)) | |
830 | return PTR_ERR(bd); | |
831 | ||
832 | file->private_data = bd; | |
833 | return 0; | |
834 | } | |
835 | ||
836 | static int bsg_release(struct inode *inode, struct file *file) | |
837 | { | |
838 | struct bsg_device *bd = file->private_data; | |
839 | ||
840 | file->private_data = NULL; | |
841 | return bsg_put_device(bd); | |
842 | } | |
843 | ||
844 | static unsigned int bsg_poll(struct file *file, poll_table *wait) | |
845 | { | |
846 | struct bsg_device *bd = file->private_data; | |
847 | unsigned int mask = 0; | |
848 | ||
849 | poll_wait(file, &bd->wq_done, wait); | |
850 | poll_wait(file, &bd->wq_free, wait); | |
851 | ||
852 | spin_lock_irq(&bd->lock); | |
853 | if (!list_empty(&bd->done_list)) | |
854 | mask |= POLLIN | POLLRDNORM; | |
855 | if (bd->queued_cmds >= bd->max_queue) | |
856 | mask |= POLLOUT; | |
857 | spin_unlock_irq(&bd->lock); | |
858 | ||
859 | return mask; | |
860 | } | |
861 | ||
862 | static int | |
863 | bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd, | |
864 | unsigned long arg) | |
865 | { | |
866 | struct bsg_device *bd = file->private_data; | |
867 | int __user *uarg = (int __user *) arg; | |
868 | ||
869 | if (!bd) | |
870 | return -ENXIO; | |
871 | ||
872 | switch (cmd) { | |
873 | /* | |
874 | * our own ioctls | |
875 | */ | |
876 | case SG_GET_COMMAND_Q: | |
877 | return put_user(bd->max_queue, uarg); | |
878 | case SG_SET_COMMAND_Q: { | |
879 | int queue; | |
880 | ||
881 | if (get_user(queue, uarg)) | |
882 | return -EFAULT; | |
883 | if (queue > BSG_CMDS || queue < 1) | |
884 | return -EINVAL; | |
885 | ||
886 | bd->max_queue = queue; | |
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: | |
901 | case SG_IO: | |
902 | case SCSI_IOCTL_SEND_COMMAND: { | |
903 | void __user *uarg = (void __user *) arg; | |
904 | return scsi_cmd_ioctl(file, bd->disk, cmd, uarg); | |
905 | } | |
906 | /* | |
907 | * block device ioctls | |
908 | */ | |
909 | default: | |
910 | #if 0 | |
911 | return ioctl_by_bdev(bd->bdev, cmd, arg); | |
912 | #else | |
913 | return -ENOTTY; | |
914 | #endif | |
915 | } | |
916 | } | |
917 | ||
918 | static struct file_operations bsg_fops = { | |
919 | .read = bsg_read, | |
920 | .write = bsg_write, | |
921 | .poll = bsg_poll, | |
922 | .open = bsg_open, | |
923 | .release = bsg_release, | |
924 | .ioctl = bsg_ioctl, | |
925 | .owner = THIS_MODULE, | |
926 | }; | |
927 | ||
928 | void bsg_unregister_disk(struct gendisk *disk) | |
929 | { | |
930 | struct bsg_class_device *bcd = &disk->bsg_dev; | |
931 | ||
932 | if (!bcd->class_dev) | |
933 | return; | |
934 | ||
935 | mutex_lock(&bsg_mutex); | |
936 | sysfs_remove_link(&bcd->disk->queue->kobj, "bsg"); | |
937 | class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor)); | |
938 | bcd->class_dev = NULL; | |
939 | list_del_init(&bcd->list); | |
940 | mutex_unlock(&bsg_mutex); | |
941 | } | |
942 | ||
943 | int bsg_register_disk(struct gendisk *disk) | |
944 | { | |
945 | request_queue_t *q = disk->queue; | |
946 | struct bsg_class_device *bcd; | |
947 | dev_t dev; | |
948 | ||
949 | /* | |
950 | * we need a proper transport to send commands, not a stacked device | |
951 | */ | |
952 | if (!q->request_fn) | |
953 | return 0; | |
954 | ||
955 | bcd = &disk->bsg_dev; | |
956 | memset(bcd, 0, sizeof(*bcd)); | |
957 | INIT_LIST_HEAD(&bcd->list); | |
958 | ||
959 | mutex_lock(&bsg_mutex); | |
960 | dev = MKDEV(BSG_MAJOR, bsg_device_nr); | |
961 | bcd->minor = bsg_device_nr; | |
962 | bsg_device_nr++; | |
963 | bcd->disk = disk; | |
964 | bcd->class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", disk->disk_name); | |
965 | list_add_tail(&bcd->list, &bsg_class_list); | |
966 | sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg"); | |
967 | mutex_unlock(&bsg_mutex); | |
968 | return 0; | |
969 | } | |
970 | ||
971 | static int __init bsg_init(void) | |
972 | { | |
973 | int ret, i; | |
974 | ||
975 | for (i = 0; i < BSG_LIST_SIZE; i++) | |
976 | INIT_HLIST_HEAD(&bsg_device_list[i]); | |
977 | ||
978 | bsg_class = class_create(THIS_MODULE, "bsg"); | |
979 | if (IS_ERR(bsg_class)) | |
980 | return PTR_ERR(bsg_class); | |
981 | ||
982 | ret = register_chrdev(BSG_MAJOR, "bsg", &bsg_fops); | |
983 | if (ret) { | |
984 | class_destroy(bsg_class); | |
985 | return ret; | |
986 | } | |
987 | ||
988 | printk(KERN_INFO "%s loaded\n", bsg_version); | |
989 | return 0; | |
990 | } | |
991 | ||
992 | MODULE_AUTHOR("Jens Axboe"); | |
993 | MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver"); | |
994 | MODULE_LICENSE("GPL"); | |
995 | ||
996 | subsys_initcall(bsg_init); |