xref: /openbmc/linux/drivers/scsi/sd.c (revision 63dc02bd)
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *	Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17  *	   sd_init and cleanups.
18  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *	   not being read in sd_open. Fix problem where removable media
20  *	   could be ejected after sd_open.
21  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24  *	   Support 32k/1M disks.
25  *
26  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *	Note: when the logging level is set by the user, it must be greater
32  *	than the level indicated above to trigger output.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <asm/uaccess.h>
55 #include <asm/unaligned.h>
56 
57 #include <scsi/scsi.h>
58 #include <scsi/scsi_cmnd.h>
59 #include <scsi/scsi_dbg.h>
60 #include <scsi/scsi_device.h>
61 #include <scsi/scsi_driver.h>
62 #include <scsi/scsi_eh.h>
63 #include <scsi/scsi_host.h>
64 #include <scsi/scsi_ioctl.h>
65 #include <scsi/scsicam.h>
66 
67 #include "sd.h"
68 #include "scsi_logging.h"
69 
70 MODULE_AUTHOR("Eric Youngdale");
71 MODULE_DESCRIPTION("SCSI disk (sd) driver");
72 MODULE_LICENSE("GPL");
73 
74 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
90 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
93 
94 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
95 #define SD_MINORS	16
96 #else
97 #define SD_MINORS	0
98 #endif
99 
100 static void sd_config_discard(struct scsi_disk *, unsigned int);
101 static int  sd_revalidate_disk(struct gendisk *);
102 static void sd_unlock_native_capacity(struct gendisk *disk);
103 static int  sd_probe(struct device *);
104 static int  sd_remove(struct device *);
105 static void sd_shutdown(struct device *);
106 static int sd_suspend(struct device *, pm_message_t state);
107 static int sd_resume(struct device *);
108 static void sd_rescan(struct device *);
109 static int sd_done(struct scsi_cmnd *);
110 static int sd_eh_action(struct scsi_cmnd *, unsigned char *, int, int);
111 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
112 static void scsi_disk_release(struct device *cdev);
113 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
114 static void sd_print_result(struct scsi_disk *, int);
115 
116 static DEFINE_SPINLOCK(sd_index_lock);
117 static DEFINE_IDA(sd_index_ida);
118 
119 /* This semaphore is used to mediate the 0->1 reference get in the
120  * face of object destruction (i.e. we can't allow a get on an
121  * object after last put) */
122 static DEFINE_MUTEX(sd_ref_mutex);
123 
124 static struct kmem_cache *sd_cdb_cache;
125 static mempool_t *sd_cdb_pool;
126 
127 static const char *sd_cache_types[] = {
128 	"write through", "none", "write back",
129 	"write back, no read (daft)"
130 };
131 
132 static ssize_t
133 sd_store_cache_type(struct device *dev, struct device_attribute *attr,
134 		    const char *buf, size_t count)
135 {
136 	int i, ct = -1, rcd, wce, sp;
137 	struct scsi_disk *sdkp = to_scsi_disk(dev);
138 	struct scsi_device *sdp = sdkp->device;
139 	char buffer[64];
140 	char *buffer_data;
141 	struct scsi_mode_data data;
142 	struct scsi_sense_hdr sshdr;
143 	int len;
144 
145 	if (sdp->type != TYPE_DISK)
146 		/* no cache control on RBC devices; theoretically they
147 		 * can do it, but there's probably so many exceptions
148 		 * it's not worth the risk */
149 		return -EINVAL;
150 
151 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
152 		len = strlen(sd_cache_types[i]);
153 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
154 		    buf[len] == '\n') {
155 			ct = i;
156 			break;
157 		}
158 	}
159 	if (ct < 0)
160 		return -EINVAL;
161 	rcd = ct & 0x01 ? 1 : 0;
162 	wce = ct & 0x02 ? 1 : 0;
163 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
164 			    SD_MAX_RETRIES, &data, NULL))
165 		return -EINVAL;
166 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
167 		  data.block_descriptor_length);
168 	buffer_data = buffer + data.header_length +
169 		data.block_descriptor_length;
170 	buffer_data[2] &= ~0x05;
171 	buffer_data[2] |= wce << 2 | rcd;
172 	sp = buffer_data[0] & 0x80 ? 1 : 0;
173 
174 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
175 			     SD_MAX_RETRIES, &data, &sshdr)) {
176 		if (scsi_sense_valid(&sshdr))
177 			sd_print_sense_hdr(sdkp, &sshdr);
178 		return -EINVAL;
179 	}
180 	revalidate_disk(sdkp->disk);
181 	return count;
182 }
183 
184 static ssize_t
185 sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
186 			   const char *buf, size_t count)
187 {
188 	struct scsi_disk *sdkp = to_scsi_disk(dev);
189 	struct scsi_device *sdp = sdkp->device;
190 
191 	if (!capable(CAP_SYS_ADMIN))
192 		return -EACCES;
193 
194 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
195 
196 	return count;
197 }
198 
199 static ssize_t
200 sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
201 		       const char *buf, size_t count)
202 {
203 	struct scsi_disk *sdkp = to_scsi_disk(dev);
204 	struct scsi_device *sdp = sdkp->device;
205 
206 	if (!capable(CAP_SYS_ADMIN))
207 		return -EACCES;
208 
209 	if (sdp->type != TYPE_DISK)
210 		return -EINVAL;
211 
212 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
213 
214 	return count;
215 }
216 
217 static ssize_t
218 sd_show_cache_type(struct device *dev, struct device_attribute *attr,
219 		   char *buf)
220 {
221 	struct scsi_disk *sdkp = to_scsi_disk(dev);
222 	int ct = sdkp->RCD + 2*sdkp->WCE;
223 
224 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
225 }
226 
227 static ssize_t
228 sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
229 {
230 	struct scsi_disk *sdkp = to_scsi_disk(dev);
231 
232 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
233 }
234 
235 static ssize_t
236 sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
237 			  char *buf)
238 {
239 	struct scsi_disk *sdkp = to_scsi_disk(dev);
240 	struct scsi_device *sdp = sdkp->device;
241 
242 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
243 }
244 
245 static ssize_t
246 sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
247 		      char *buf)
248 {
249 	struct scsi_disk *sdkp = to_scsi_disk(dev);
250 
251 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
252 }
253 
254 static ssize_t
255 sd_show_protection_type(struct device *dev, struct device_attribute *attr,
256 			char *buf)
257 {
258 	struct scsi_disk *sdkp = to_scsi_disk(dev);
259 
260 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
261 }
262 
263 static ssize_t
264 sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
265 			char *buf)
266 {
267 	struct scsi_disk *sdkp = to_scsi_disk(dev);
268 	struct scsi_device *sdp = sdkp->device;
269 	unsigned int dif, dix;
270 
271 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
272 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
273 
274 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
275 		dif = 0;
276 		dix = 1;
277 	}
278 
279 	if (!dif && !dix)
280 		return snprintf(buf, 20, "none\n");
281 
282 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
283 }
284 
285 static ssize_t
286 sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
287 		    char *buf)
288 {
289 	struct scsi_disk *sdkp = to_scsi_disk(dev);
290 
291 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
292 }
293 
294 static ssize_t
295 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
296 			  char *buf)
297 {
298 	struct scsi_disk *sdkp = to_scsi_disk(dev);
299 
300 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
301 }
302 
303 static const char *lbp_mode[] = {
304 	[SD_LBP_FULL]		= "full",
305 	[SD_LBP_UNMAP]		= "unmap",
306 	[SD_LBP_WS16]		= "writesame_16",
307 	[SD_LBP_WS10]		= "writesame_10",
308 	[SD_LBP_ZERO]		= "writesame_zero",
309 	[SD_LBP_DISABLE]	= "disabled",
310 };
311 
312 static ssize_t
313 sd_show_provisioning_mode(struct device *dev, struct device_attribute *attr,
314 			  char *buf)
315 {
316 	struct scsi_disk *sdkp = to_scsi_disk(dev);
317 
318 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
319 }
320 
321 static ssize_t
322 sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
323 			   const char *buf, size_t count)
324 {
325 	struct scsi_disk *sdkp = to_scsi_disk(dev);
326 	struct scsi_device *sdp = sdkp->device;
327 
328 	if (!capable(CAP_SYS_ADMIN))
329 		return -EACCES;
330 
331 	if (sdp->type != TYPE_DISK)
332 		return -EINVAL;
333 
334 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
335 		sd_config_discard(sdkp, SD_LBP_UNMAP);
336 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
337 		sd_config_discard(sdkp, SD_LBP_WS16);
338 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
339 		sd_config_discard(sdkp, SD_LBP_WS10);
340 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
341 		sd_config_discard(sdkp, SD_LBP_ZERO);
342 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
343 		sd_config_discard(sdkp, SD_LBP_DISABLE);
344 	else
345 		return -EINVAL;
346 
347 	return count;
348 }
349 
350 static ssize_t
351 sd_show_max_medium_access_timeouts(struct device *dev,
352 				   struct device_attribute *attr, char *buf)
353 {
354 	struct scsi_disk *sdkp = to_scsi_disk(dev);
355 
356 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
357 }
358 
359 static ssize_t
360 sd_store_max_medium_access_timeouts(struct device *dev,
361 				    struct device_attribute *attr,
362 				    const char *buf, size_t count)
363 {
364 	struct scsi_disk *sdkp = to_scsi_disk(dev);
365 	int err;
366 
367 	if (!capable(CAP_SYS_ADMIN))
368 		return -EACCES;
369 
370 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
371 
372 	return err ? err : count;
373 }
374 
375 static struct device_attribute sd_disk_attrs[] = {
376 	__ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
377 	       sd_store_cache_type),
378 	__ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
379 	__ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
380 	       sd_store_allow_restart),
381 	__ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
382 	       sd_store_manage_start_stop),
383 	__ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
384 	__ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
385 	__ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
386 	__ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
387 	__ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
388 	       sd_store_provisioning_mode),
389 	__ATTR(max_medium_access_timeouts, S_IRUGO|S_IWUSR,
390 	       sd_show_max_medium_access_timeouts,
391 	       sd_store_max_medium_access_timeouts),
392 	__ATTR_NULL,
393 };
394 
395 static struct class sd_disk_class = {
396 	.name		= "scsi_disk",
397 	.owner		= THIS_MODULE,
398 	.dev_release	= scsi_disk_release,
399 	.dev_attrs	= sd_disk_attrs,
400 };
401 
402 static struct scsi_driver sd_template = {
403 	.owner			= THIS_MODULE,
404 	.gendrv = {
405 		.name		= "sd",
406 		.probe		= sd_probe,
407 		.remove		= sd_remove,
408 		.suspend	= sd_suspend,
409 		.resume		= sd_resume,
410 		.shutdown	= sd_shutdown,
411 	},
412 	.rescan			= sd_rescan,
413 	.done			= sd_done,
414 	.eh_action		= sd_eh_action,
415 };
416 
417 /*
418  * Device no to disk mapping:
419  *
420  *       major         disc2     disc  p1
421  *   |............|.............|....|....| <- dev_t
422  *    31        20 19          8 7  4 3  0
423  *
424  * Inside a major, we have 16k disks, however mapped non-
425  * contiguously. The first 16 disks are for major0, the next
426  * ones with major1, ... Disk 256 is for major0 again, disk 272
427  * for major1, ...
428  * As we stay compatible with our numbering scheme, we can reuse
429  * the well-know SCSI majors 8, 65--71, 136--143.
430  */
431 static int sd_major(int major_idx)
432 {
433 	switch (major_idx) {
434 	case 0:
435 		return SCSI_DISK0_MAJOR;
436 	case 1 ... 7:
437 		return SCSI_DISK1_MAJOR + major_idx - 1;
438 	case 8 ... 15:
439 		return SCSI_DISK8_MAJOR + major_idx - 8;
440 	default:
441 		BUG();
442 		return 0;	/* shut up gcc */
443 	}
444 }
445 
446 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
447 {
448 	struct scsi_disk *sdkp = NULL;
449 
450 	if (disk->private_data) {
451 		sdkp = scsi_disk(disk);
452 		if (scsi_device_get(sdkp->device) == 0)
453 			get_device(&sdkp->dev);
454 		else
455 			sdkp = NULL;
456 	}
457 	return sdkp;
458 }
459 
460 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
461 {
462 	struct scsi_disk *sdkp;
463 
464 	mutex_lock(&sd_ref_mutex);
465 	sdkp = __scsi_disk_get(disk);
466 	mutex_unlock(&sd_ref_mutex);
467 	return sdkp;
468 }
469 
470 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
471 {
472 	struct scsi_disk *sdkp;
473 
474 	mutex_lock(&sd_ref_mutex);
475 	sdkp = dev_get_drvdata(dev);
476 	if (sdkp)
477 		sdkp = __scsi_disk_get(sdkp->disk);
478 	mutex_unlock(&sd_ref_mutex);
479 	return sdkp;
480 }
481 
482 static void scsi_disk_put(struct scsi_disk *sdkp)
483 {
484 	struct scsi_device *sdev = sdkp->device;
485 
486 	mutex_lock(&sd_ref_mutex);
487 	put_device(&sdkp->dev);
488 	scsi_device_put(sdev);
489 	mutex_unlock(&sd_ref_mutex);
490 }
491 
492 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
493 {
494 	unsigned int prot_op = SCSI_PROT_NORMAL;
495 	unsigned int dix = scsi_prot_sg_count(scmd);
496 
497 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
498 		if (dif && dix)
499 			prot_op = SCSI_PROT_READ_PASS;
500 		else if (dif && !dix)
501 			prot_op = SCSI_PROT_READ_STRIP;
502 		else if (!dif && dix)
503 			prot_op = SCSI_PROT_READ_INSERT;
504 	} else {
505 		if (dif && dix)
506 			prot_op = SCSI_PROT_WRITE_PASS;
507 		else if (dif && !dix)
508 			prot_op = SCSI_PROT_WRITE_INSERT;
509 		else if (!dif && dix)
510 			prot_op = SCSI_PROT_WRITE_STRIP;
511 	}
512 
513 	scsi_set_prot_op(scmd, prot_op);
514 	scsi_set_prot_type(scmd, dif);
515 }
516 
517 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
518 {
519 	struct request_queue *q = sdkp->disk->queue;
520 	unsigned int logical_block_size = sdkp->device->sector_size;
521 	unsigned int max_blocks = 0;
522 
523 	q->limits.discard_zeroes_data = sdkp->lbprz;
524 	q->limits.discard_alignment = sdkp->unmap_alignment *
525 		logical_block_size;
526 	q->limits.discard_granularity =
527 		max(sdkp->physical_block_size,
528 		    sdkp->unmap_granularity * logical_block_size);
529 
530 	sdkp->provisioning_mode = mode;
531 
532 	switch (mode) {
533 
534 	case SD_LBP_DISABLE:
535 		q->limits.max_discard_sectors = 0;
536 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
537 		return;
538 
539 	case SD_LBP_UNMAP:
540 		max_blocks = min_not_zero(sdkp->max_unmap_blocks, 0xffffffff);
541 		break;
542 
543 	case SD_LBP_WS16:
544 		max_blocks = min_not_zero(sdkp->max_ws_blocks, 0xffffffff);
545 		break;
546 
547 	case SD_LBP_WS10:
548 		max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
549 		break;
550 
551 	case SD_LBP_ZERO:
552 		max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
553 		q->limits.discard_zeroes_data = 1;
554 		break;
555 	}
556 
557 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
558 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
559 }
560 
561 /**
562  * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
563  * @sdp: scsi device to operate one
564  * @rq: Request to prepare
565  *
566  * Will issue either UNMAP or WRITE SAME(16) depending on preference
567  * indicated by target device.
568  **/
569 static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
570 {
571 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
572 	struct bio *bio = rq->bio;
573 	sector_t sector = bio->bi_sector;
574 	unsigned int nr_sectors = bio_sectors(bio);
575 	unsigned int len;
576 	int ret;
577 	char *buf;
578 	struct page *page;
579 
580 	if (sdkp->device->sector_size == 4096) {
581 		sector >>= 3;
582 		nr_sectors >>= 3;
583 	}
584 
585 	rq->timeout = SD_TIMEOUT;
586 
587 	memset(rq->cmd, 0, rq->cmd_len);
588 
589 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
590 	if (!page)
591 		return BLKPREP_DEFER;
592 
593 	switch (sdkp->provisioning_mode) {
594 	case SD_LBP_UNMAP:
595 		buf = page_address(page);
596 
597 		rq->cmd_len = 10;
598 		rq->cmd[0] = UNMAP;
599 		rq->cmd[8] = 24;
600 
601 		put_unaligned_be16(6 + 16, &buf[0]);
602 		put_unaligned_be16(16, &buf[2]);
603 		put_unaligned_be64(sector, &buf[8]);
604 		put_unaligned_be32(nr_sectors, &buf[16]);
605 
606 		len = 24;
607 		break;
608 
609 	case SD_LBP_WS16:
610 		rq->cmd_len = 16;
611 		rq->cmd[0] = WRITE_SAME_16;
612 		rq->cmd[1] = 0x8; /* UNMAP */
613 		put_unaligned_be64(sector, &rq->cmd[2]);
614 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
615 
616 		len = sdkp->device->sector_size;
617 		break;
618 
619 	case SD_LBP_WS10:
620 	case SD_LBP_ZERO:
621 		rq->cmd_len = 10;
622 		rq->cmd[0] = WRITE_SAME;
623 		if (sdkp->provisioning_mode == SD_LBP_WS10)
624 			rq->cmd[1] = 0x8; /* UNMAP */
625 		put_unaligned_be32(sector, &rq->cmd[2]);
626 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
627 
628 		len = sdkp->device->sector_size;
629 		break;
630 
631 	default:
632 		ret = BLKPREP_KILL;
633 		goto out;
634 	}
635 
636 	blk_add_request_payload(rq, page, len);
637 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
638 	rq->buffer = page_address(page);
639 
640 out:
641 	if (ret != BLKPREP_OK) {
642 		__free_page(page);
643 		rq->buffer = NULL;
644 	}
645 	return ret;
646 }
647 
648 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
649 {
650 	rq->timeout = SD_FLUSH_TIMEOUT;
651 	rq->retries = SD_MAX_RETRIES;
652 	rq->cmd[0] = SYNCHRONIZE_CACHE;
653 	rq->cmd_len = 10;
654 
655 	return scsi_setup_blk_pc_cmnd(sdp, rq);
656 }
657 
658 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
659 {
660 	if (rq->cmd_flags & REQ_DISCARD) {
661 		free_page((unsigned long)rq->buffer);
662 		rq->buffer = NULL;
663 	}
664 }
665 
666 /**
667  *	sd_prep_fn - build a scsi (read or write) command from
668  *	information in the request structure.
669  *	@SCpnt: pointer to mid-level's per scsi command structure that
670  *	contains request and into which the scsi command is written
671  *
672  *	Returns 1 if successful and 0 if error (or cannot be done now).
673  **/
674 static int sd_prep_fn(struct request_queue *q, struct request *rq)
675 {
676 	struct scsi_cmnd *SCpnt;
677 	struct scsi_device *sdp = q->queuedata;
678 	struct gendisk *disk = rq->rq_disk;
679 	struct scsi_disk *sdkp;
680 	sector_t block = blk_rq_pos(rq);
681 	sector_t threshold;
682 	unsigned int this_count = blk_rq_sectors(rq);
683 	int ret, host_dif;
684 	unsigned char protect;
685 
686 	/*
687 	 * Discard request come in as REQ_TYPE_FS but we turn them into
688 	 * block PC requests to make life easier.
689 	 */
690 	if (rq->cmd_flags & REQ_DISCARD) {
691 		ret = scsi_setup_discard_cmnd(sdp, rq);
692 		goto out;
693 	} else if (rq->cmd_flags & REQ_FLUSH) {
694 		ret = scsi_setup_flush_cmnd(sdp, rq);
695 		goto out;
696 	} else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
697 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
698 		goto out;
699 	} else if (rq->cmd_type != REQ_TYPE_FS) {
700 		ret = BLKPREP_KILL;
701 		goto out;
702 	}
703 	ret = scsi_setup_fs_cmnd(sdp, rq);
704 	if (ret != BLKPREP_OK)
705 		goto out;
706 	SCpnt = rq->special;
707 	sdkp = scsi_disk(disk);
708 
709 	/* from here on until we're complete, any goto out
710 	 * is used for a killable error condition */
711 	ret = BLKPREP_KILL;
712 
713 	SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
714 					"sd_prep_fn: block=%llu, "
715 					"count=%d\n",
716 					(unsigned long long)block,
717 					this_count));
718 
719 	if (!sdp || !scsi_device_online(sdp) ||
720 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
721 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
722 						"Finishing %u sectors\n",
723 						blk_rq_sectors(rq)));
724 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
725 						"Retry with 0x%p\n", SCpnt));
726 		goto out;
727 	}
728 
729 	if (sdp->changed) {
730 		/*
731 		 * quietly refuse to do anything to a changed disc until
732 		 * the changed bit has been reset
733 		 */
734 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
735 		goto out;
736 	}
737 
738 	/*
739 	 * Some SD card readers can't handle multi-sector accesses which touch
740 	 * the last one or two hardware sectors.  Split accesses as needed.
741 	 */
742 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
743 		(sdp->sector_size / 512);
744 
745 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
746 		if (block < threshold) {
747 			/* Access up to the threshold but not beyond */
748 			this_count = threshold - block;
749 		} else {
750 			/* Access only a single hardware sector */
751 			this_count = sdp->sector_size / 512;
752 		}
753 	}
754 
755 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
756 					(unsigned long long)block));
757 
758 	/*
759 	 * If we have a 1K hardware sectorsize, prevent access to single
760 	 * 512 byte sectors.  In theory we could handle this - in fact
761 	 * the scsi cdrom driver must be able to handle this because
762 	 * we typically use 1K blocksizes, and cdroms typically have
763 	 * 2K hardware sectorsizes.  Of course, things are simpler
764 	 * with the cdrom, since it is read-only.  For performance
765 	 * reasons, the filesystems should be able to handle this
766 	 * and not force the scsi disk driver to use bounce buffers
767 	 * for this.
768 	 */
769 	if (sdp->sector_size == 1024) {
770 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
771 			scmd_printk(KERN_ERR, SCpnt,
772 				    "Bad block number requested\n");
773 			goto out;
774 		} else {
775 			block = block >> 1;
776 			this_count = this_count >> 1;
777 		}
778 	}
779 	if (sdp->sector_size == 2048) {
780 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
781 			scmd_printk(KERN_ERR, SCpnt,
782 				    "Bad block number requested\n");
783 			goto out;
784 		} else {
785 			block = block >> 2;
786 			this_count = this_count >> 2;
787 		}
788 	}
789 	if (sdp->sector_size == 4096) {
790 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
791 			scmd_printk(KERN_ERR, SCpnt,
792 				    "Bad block number requested\n");
793 			goto out;
794 		} else {
795 			block = block >> 3;
796 			this_count = this_count >> 3;
797 		}
798 	}
799 	if (rq_data_dir(rq) == WRITE) {
800 		if (!sdp->writeable) {
801 			goto out;
802 		}
803 		SCpnt->cmnd[0] = WRITE_6;
804 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
805 
806 		if (blk_integrity_rq(rq) &&
807 		    sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
808 			goto out;
809 
810 	} else if (rq_data_dir(rq) == READ) {
811 		SCpnt->cmnd[0] = READ_6;
812 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
813 	} else {
814 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
815 		goto out;
816 	}
817 
818 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
819 					"%s %d/%u 512 byte blocks.\n",
820 					(rq_data_dir(rq) == WRITE) ?
821 					"writing" : "reading", this_count,
822 					blk_rq_sectors(rq)));
823 
824 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
825 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
826 	if (host_dif)
827 		protect = 1 << 5;
828 	else
829 		protect = 0;
830 
831 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
832 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
833 
834 		if (unlikely(SCpnt->cmnd == NULL)) {
835 			ret = BLKPREP_DEFER;
836 			goto out;
837 		}
838 
839 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
840 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
841 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
842 		SCpnt->cmnd[7] = 0x18;
843 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
844 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
845 
846 		/* LBA */
847 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
848 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
849 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
850 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
851 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
852 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
853 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
854 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
855 
856 		/* Expected Indirect LBA */
857 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
858 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
859 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
860 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
861 
862 		/* Transfer length */
863 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
864 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
865 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
866 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
867 	} else if (block > 0xffffffff) {
868 		SCpnt->cmnd[0] += READ_16 - READ_6;
869 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
870 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
871 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
872 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
873 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
874 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
875 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
876 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
877 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
878 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
879 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
880 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
881 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
882 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
883 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
884 		   scsi_device_protection(SCpnt->device) ||
885 		   SCpnt->device->use_10_for_rw) {
886 		if (this_count > 0xffff)
887 			this_count = 0xffff;
888 
889 		SCpnt->cmnd[0] += READ_10 - READ_6;
890 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
891 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
892 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
893 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
894 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
895 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
896 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
897 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
898 	} else {
899 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
900 			/*
901 			 * This happens only if this drive failed
902 			 * 10byte rw command with ILLEGAL_REQUEST
903 			 * during operation and thus turned off
904 			 * use_10_for_rw.
905 			 */
906 			scmd_printk(KERN_ERR, SCpnt,
907 				    "FUA write on READ/WRITE(6) drive\n");
908 			goto out;
909 		}
910 
911 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
912 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
913 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
914 		SCpnt->cmnd[4] = (unsigned char) this_count;
915 		SCpnt->cmnd[5] = 0;
916 	}
917 	SCpnt->sdb.length = this_count * sdp->sector_size;
918 
919 	/* If DIF or DIX is enabled, tell HBA how to handle request */
920 	if (host_dif || scsi_prot_sg_count(SCpnt))
921 		sd_prot_op(SCpnt, host_dif);
922 
923 	/*
924 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
925 	 * host adapter, it's safe to assume that we can at least transfer
926 	 * this many bytes between each connect / disconnect.
927 	 */
928 	SCpnt->transfersize = sdp->sector_size;
929 	SCpnt->underflow = this_count << 9;
930 	SCpnt->allowed = SD_MAX_RETRIES;
931 
932 	/*
933 	 * This indicates that the command is ready from our end to be
934 	 * queued.
935 	 */
936 	ret = BLKPREP_OK;
937  out:
938 	return scsi_prep_return(q, rq, ret);
939 }
940 
941 /**
942  *	sd_open - open a scsi disk device
943  *	@inode: only i_rdev member may be used
944  *	@filp: only f_mode and f_flags may be used
945  *
946  *	Returns 0 if successful. Returns a negated errno value in case
947  *	of error.
948  *
949  *	Note: This can be called from a user context (e.g. fsck(1) )
950  *	or from within the kernel (e.g. as a result of a mount(1) ).
951  *	In the latter case @inode and @filp carry an abridged amount
952  *	of information as noted above.
953  *
954  *	Locking: called with bdev->bd_mutex held.
955  **/
956 static int sd_open(struct block_device *bdev, fmode_t mode)
957 {
958 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
959 	struct scsi_device *sdev;
960 	int retval;
961 
962 	if (!sdkp)
963 		return -ENXIO;
964 
965 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
966 
967 	sdev = sdkp->device;
968 
969 	retval = scsi_autopm_get_device(sdev);
970 	if (retval)
971 		goto error_autopm;
972 
973 	/*
974 	 * If the device is in error recovery, wait until it is done.
975 	 * If the device is offline, then disallow any access to it.
976 	 */
977 	retval = -ENXIO;
978 	if (!scsi_block_when_processing_errors(sdev))
979 		goto error_out;
980 
981 	if (sdev->removable || sdkp->write_prot)
982 		check_disk_change(bdev);
983 
984 	/*
985 	 * If the drive is empty, just let the open fail.
986 	 */
987 	retval = -ENOMEDIUM;
988 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
989 		goto error_out;
990 
991 	/*
992 	 * If the device has the write protect tab set, have the open fail
993 	 * if the user expects to be able to write to the thing.
994 	 */
995 	retval = -EROFS;
996 	if (sdkp->write_prot && (mode & FMODE_WRITE))
997 		goto error_out;
998 
999 	/*
1000 	 * It is possible that the disk changing stuff resulted in
1001 	 * the device being taken offline.  If this is the case,
1002 	 * report this to the user, and don't pretend that the
1003 	 * open actually succeeded.
1004 	 */
1005 	retval = -ENXIO;
1006 	if (!scsi_device_online(sdev))
1007 		goto error_out;
1008 
1009 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1010 		if (scsi_block_when_processing_errors(sdev))
1011 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1012 	}
1013 
1014 	return 0;
1015 
1016 error_out:
1017 	scsi_autopm_put_device(sdev);
1018 error_autopm:
1019 	scsi_disk_put(sdkp);
1020 	return retval;
1021 }
1022 
1023 /**
1024  *	sd_release - invoked when the (last) close(2) is called on this
1025  *	scsi disk.
1026  *	@inode: only i_rdev member may be used
1027  *	@filp: only f_mode and f_flags may be used
1028  *
1029  *	Returns 0.
1030  *
1031  *	Note: may block (uninterruptible) if error recovery is underway
1032  *	on this disk.
1033  *
1034  *	Locking: called with bdev->bd_mutex held.
1035  **/
1036 static int sd_release(struct gendisk *disk, fmode_t mode)
1037 {
1038 	struct scsi_disk *sdkp = scsi_disk(disk);
1039 	struct scsi_device *sdev = sdkp->device;
1040 
1041 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1042 
1043 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1044 		if (scsi_block_when_processing_errors(sdev))
1045 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1046 	}
1047 
1048 	/*
1049 	 * XXX and what if there are packets in flight and this close()
1050 	 * XXX is followed by a "rmmod sd_mod"?
1051 	 */
1052 
1053 	scsi_autopm_put_device(sdev);
1054 	scsi_disk_put(sdkp);
1055 	return 0;
1056 }
1057 
1058 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1059 {
1060 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1061 	struct scsi_device *sdp = sdkp->device;
1062 	struct Scsi_Host *host = sdp->host;
1063 	int diskinfo[4];
1064 
1065 	/* default to most commonly used values */
1066         diskinfo[0] = 0x40;	/* 1 << 6 */
1067        	diskinfo[1] = 0x20;	/* 1 << 5 */
1068        	diskinfo[2] = sdkp->capacity >> 11;
1069 
1070 	/* override with calculated, extended default, or driver values */
1071 	if (host->hostt->bios_param)
1072 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1073 	else
1074 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1075 
1076 	geo->heads = diskinfo[0];
1077 	geo->sectors = diskinfo[1];
1078 	geo->cylinders = diskinfo[2];
1079 	return 0;
1080 }
1081 
1082 /**
1083  *	sd_ioctl - process an ioctl
1084  *	@inode: only i_rdev/i_bdev members may be used
1085  *	@filp: only f_mode and f_flags may be used
1086  *	@cmd: ioctl command number
1087  *	@arg: this is third argument given to ioctl(2) system call.
1088  *	Often contains a pointer.
1089  *
1090  *	Returns 0 if successful (some ioctls return positive numbers on
1091  *	success as well). Returns a negated errno value in case of error.
1092  *
1093  *	Note: most ioctls are forward onto the block subsystem or further
1094  *	down in the scsi subsystem.
1095  **/
1096 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1097 		    unsigned int cmd, unsigned long arg)
1098 {
1099 	struct gendisk *disk = bdev->bd_disk;
1100 	struct scsi_disk *sdkp = scsi_disk(disk);
1101 	struct scsi_device *sdp = sdkp->device;
1102 	void __user *p = (void __user *)arg;
1103 	int error;
1104 
1105 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1106 				    "cmd=0x%x\n", disk->disk_name, cmd));
1107 
1108 	error = scsi_verify_blk_ioctl(bdev, cmd);
1109 	if (error < 0)
1110 		return error;
1111 
1112 	/*
1113 	 * If we are in the middle of error recovery, don't let anyone
1114 	 * else try and use this device.  Also, if error recovery fails, it
1115 	 * may try and take the device offline, in which case all further
1116 	 * access to the device is prohibited.
1117 	 */
1118 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1119 					(mode & FMODE_NDELAY) != 0);
1120 	if (!scsi_block_when_processing_errors(sdp) || !error)
1121 		goto out;
1122 
1123 	/*
1124 	 * Send SCSI addressing ioctls directly to mid level, send other
1125 	 * ioctls to block level and then onto mid level if they can't be
1126 	 * resolved.
1127 	 */
1128 	switch (cmd) {
1129 		case SCSI_IOCTL_GET_IDLUN:
1130 		case SCSI_IOCTL_GET_BUS_NUMBER:
1131 			error = scsi_ioctl(sdp, cmd, p);
1132 			break;
1133 		default:
1134 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1135 			if (error != -ENOTTY)
1136 				break;
1137 			error = scsi_ioctl(sdp, cmd, p);
1138 			break;
1139 	}
1140 out:
1141 	return error;
1142 }
1143 
1144 static void set_media_not_present(struct scsi_disk *sdkp)
1145 {
1146 	if (sdkp->media_present)
1147 		sdkp->device->changed = 1;
1148 
1149 	if (sdkp->device->removable) {
1150 		sdkp->media_present = 0;
1151 		sdkp->capacity = 0;
1152 	}
1153 }
1154 
1155 static int media_not_present(struct scsi_disk *sdkp,
1156 			     struct scsi_sense_hdr *sshdr)
1157 {
1158 	if (!scsi_sense_valid(sshdr))
1159 		return 0;
1160 
1161 	/* not invoked for commands that could return deferred errors */
1162 	switch (sshdr->sense_key) {
1163 	case UNIT_ATTENTION:
1164 	case NOT_READY:
1165 		/* medium not present */
1166 		if (sshdr->asc == 0x3A) {
1167 			set_media_not_present(sdkp);
1168 			return 1;
1169 		}
1170 	}
1171 	return 0;
1172 }
1173 
1174 /**
1175  *	sd_check_events - check media events
1176  *	@disk: kernel device descriptor
1177  *	@clearing: disk events currently being cleared
1178  *
1179  *	Returns mask of DISK_EVENT_*.
1180  *
1181  *	Note: this function is invoked from the block subsystem.
1182  **/
1183 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1184 {
1185 	struct scsi_disk *sdkp = scsi_disk(disk);
1186 	struct scsi_device *sdp = sdkp->device;
1187 	struct scsi_sense_hdr *sshdr = NULL;
1188 	int retval;
1189 
1190 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1191 
1192 	/*
1193 	 * If the device is offline, don't send any commands - just pretend as
1194 	 * if the command failed.  If the device ever comes back online, we
1195 	 * can deal with it then.  It is only because of unrecoverable errors
1196 	 * that we would ever take a device offline in the first place.
1197 	 */
1198 	if (!scsi_device_online(sdp)) {
1199 		set_media_not_present(sdkp);
1200 		goto out;
1201 	}
1202 
1203 	/*
1204 	 * Using TEST_UNIT_READY enables differentiation between drive with
1205 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1206 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1207 	 *
1208 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1209 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1210 	 * sd_revalidate() is called.
1211 	 */
1212 	retval = -ENODEV;
1213 
1214 	if (scsi_block_when_processing_errors(sdp)) {
1215 		retval = scsi_autopm_get_device(sdp);
1216 		if (retval)
1217 			goto out;
1218 
1219 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1220 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1221 					      sshdr);
1222 		scsi_autopm_put_device(sdp);
1223 	}
1224 
1225 	/* failed to execute TUR, assume media not present */
1226 	if (host_byte(retval)) {
1227 		set_media_not_present(sdkp);
1228 		goto out;
1229 	}
1230 
1231 	if (media_not_present(sdkp, sshdr))
1232 		goto out;
1233 
1234 	/*
1235 	 * For removable scsi disk we have to recognise the presence
1236 	 * of a disk in the drive.
1237 	 */
1238 	if (!sdkp->media_present)
1239 		sdp->changed = 1;
1240 	sdkp->media_present = 1;
1241 out:
1242 	/*
1243 	 * sdp->changed is set under the following conditions:
1244 	 *
1245 	 *	Medium present state has changed in either direction.
1246 	 *	Device has indicated UNIT_ATTENTION.
1247 	 */
1248 	kfree(sshdr);
1249 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1250 	sdp->changed = 0;
1251 	return retval;
1252 }
1253 
1254 static int sd_sync_cache(struct scsi_disk *sdkp)
1255 {
1256 	int retries, res;
1257 	struct scsi_device *sdp = sdkp->device;
1258 	struct scsi_sense_hdr sshdr;
1259 
1260 	if (!scsi_device_online(sdp))
1261 		return -ENODEV;
1262 
1263 
1264 	for (retries = 3; retries > 0; --retries) {
1265 		unsigned char cmd[10] = { 0 };
1266 
1267 		cmd[0] = SYNCHRONIZE_CACHE;
1268 		/*
1269 		 * Leave the rest of the command zero to indicate
1270 		 * flush everything.
1271 		 */
1272 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1273 				       SD_FLUSH_TIMEOUT, SD_MAX_RETRIES, NULL);
1274 		if (res == 0)
1275 			break;
1276 	}
1277 
1278 	if (res) {
1279 		sd_print_result(sdkp, res);
1280 		if (driver_byte(res) & DRIVER_SENSE)
1281 			sd_print_sense_hdr(sdkp, &sshdr);
1282 	}
1283 
1284 	if (res)
1285 		return -EIO;
1286 	return 0;
1287 }
1288 
1289 static void sd_rescan(struct device *dev)
1290 {
1291 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1292 
1293 	if (sdkp) {
1294 		revalidate_disk(sdkp->disk);
1295 		scsi_disk_put(sdkp);
1296 	}
1297 }
1298 
1299 
1300 #ifdef CONFIG_COMPAT
1301 /*
1302  * This gets directly called from VFS. When the ioctl
1303  * is not recognized we go back to the other translation paths.
1304  */
1305 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1306 			   unsigned int cmd, unsigned long arg)
1307 {
1308 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1309 	int ret;
1310 
1311 	ret = scsi_verify_blk_ioctl(bdev, cmd);
1312 	if (ret < 0)
1313 		return ret;
1314 
1315 	/*
1316 	 * If we are in the middle of error recovery, don't let anyone
1317 	 * else try and use this device.  Also, if error recovery fails, it
1318 	 * may try and take the device offline, in which case all further
1319 	 * access to the device is prohibited.
1320 	 */
1321 	if (!scsi_block_when_processing_errors(sdev))
1322 		return -ENODEV;
1323 
1324 	if (sdev->host->hostt->compat_ioctl) {
1325 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1326 
1327 		return ret;
1328 	}
1329 
1330 	/*
1331 	 * Let the static ioctl translation table take care of it.
1332 	 */
1333 	return -ENOIOCTLCMD;
1334 }
1335 #endif
1336 
1337 static const struct block_device_operations sd_fops = {
1338 	.owner			= THIS_MODULE,
1339 	.open			= sd_open,
1340 	.release		= sd_release,
1341 	.ioctl			= sd_ioctl,
1342 	.getgeo			= sd_getgeo,
1343 #ifdef CONFIG_COMPAT
1344 	.compat_ioctl		= sd_compat_ioctl,
1345 #endif
1346 	.check_events		= sd_check_events,
1347 	.revalidate_disk	= sd_revalidate_disk,
1348 	.unlock_native_capacity	= sd_unlock_native_capacity,
1349 };
1350 
1351 /**
1352  *	sd_eh_action - error handling callback
1353  *	@scmd:		sd-issued command that has failed
1354  *	@eh_cmnd:	The command that was sent during error handling
1355  *	@eh_cmnd_len:	Length of eh_cmnd in bytes
1356  *	@eh_disp:	The recovery disposition suggested by the midlayer
1357  *
1358  *	This function is called by the SCSI midlayer upon completion of
1359  *	an error handling command (TEST UNIT READY, START STOP UNIT,
1360  *	etc.) The command sent to the device by the error handler is
1361  *	stored in eh_cmnd. The result of sending the eh command is
1362  *	passed in eh_disp.
1363  **/
1364 static int sd_eh_action(struct scsi_cmnd *scmd, unsigned char *eh_cmnd,
1365 			int eh_cmnd_len, int eh_disp)
1366 {
1367 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1368 
1369 	if (!scsi_device_online(scmd->device) ||
1370 	    !scsi_medium_access_command(scmd))
1371 		return eh_disp;
1372 
1373 	/*
1374 	 * The device has timed out executing a medium access command.
1375 	 * However, the TEST UNIT READY command sent during error
1376 	 * handling completed successfully. Either the device is in the
1377 	 * process of recovering or has it suffered an internal failure
1378 	 * that prevents access to the storage medium.
1379 	 */
1380 	if (host_byte(scmd->result) == DID_TIME_OUT && eh_disp == SUCCESS &&
1381 	    eh_cmnd_len && eh_cmnd[0] == TEST_UNIT_READY)
1382 		sdkp->medium_access_timed_out++;
1383 
1384 	/*
1385 	 * If the device keeps failing read/write commands but TEST UNIT
1386 	 * READY always completes successfully we assume that medium
1387 	 * access is no longer possible and take the device offline.
1388 	 */
1389 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1390 		scmd_printk(KERN_ERR, scmd,
1391 			    "Medium access timeout failure. Offlining disk!\n");
1392 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1393 
1394 		return FAILED;
1395 	}
1396 
1397 	return eh_disp;
1398 }
1399 
1400 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1401 {
1402 	u64 start_lba = blk_rq_pos(scmd->request);
1403 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1404 	u64 bad_lba;
1405 	int info_valid;
1406 	/*
1407 	 * resid is optional but mostly filled in.  When it's unused,
1408 	 * its value is zero, so we assume the whole buffer transferred
1409 	 */
1410 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1411 	unsigned int good_bytes;
1412 
1413 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1414 		return 0;
1415 
1416 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1417 					     SCSI_SENSE_BUFFERSIZE,
1418 					     &bad_lba);
1419 	if (!info_valid)
1420 		return 0;
1421 
1422 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1423 		return 0;
1424 
1425 	if (scmd->device->sector_size < 512) {
1426 		/* only legitimate sector_size here is 256 */
1427 		start_lba <<= 1;
1428 		end_lba <<= 1;
1429 	} else {
1430 		/* be careful ... don't want any overflows */
1431 		u64 factor = scmd->device->sector_size / 512;
1432 		do_div(start_lba, factor);
1433 		do_div(end_lba, factor);
1434 	}
1435 
1436 	/* The bad lba was reported incorrectly, we have no idea where
1437 	 * the error is.
1438 	 */
1439 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1440 		return 0;
1441 
1442 	/* This computation should always be done in terms of
1443 	 * the resolution of the device's medium.
1444 	 */
1445 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1446 	return min(good_bytes, transferred);
1447 }
1448 
1449 /**
1450  *	sd_done - bottom half handler: called when the lower level
1451  *	driver has completed (successfully or otherwise) a scsi command.
1452  *	@SCpnt: mid-level's per command structure.
1453  *
1454  *	Note: potentially run from within an ISR. Must not block.
1455  **/
1456 static int sd_done(struct scsi_cmnd *SCpnt)
1457 {
1458 	int result = SCpnt->result;
1459 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1460 	struct scsi_sense_hdr sshdr;
1461 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1462 	int sense_valid = 0;
1463 	int sense_deferred = 0;
1464 	unsigned char op = SCpnt->cmnd[0];
1465 
1466 	if ((SCpnt->request->cmd_flags & REQ_DISCARD) && !result)
1467 		scsi_set_resid(SCpnt, 0);
1468 
1469 	if (result) {
1470 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1471 		if (sense_valid)
1472 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1473 	}
1474 #ifdef CONFIG_SCSI_LOGGING
1475 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1476 	if (sense_valid) {
1477 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1478 						   "sd_done: sb[respc,sk,asc,"
1479 						   "ascq]=%x,%x,%x,%x\n",
1480 						   sshdr.response_code,
1481 						   sshdr.sense_key, sshdr.asc,
1482 						   sshdr.ascq));
1483 	}
1484 #endif
1485 	if (driver_byte(result) != DRIVER_SENSE &&
1486 	    (!sense_valid || sense_deferred))
1487 		goto out;
1488 
1489 	sdkp->medium_access_timed_out = 0;
1490 
1491 	switch (sshdr.sense_key) {
1492 	case HARDWARE_ERROR:
1493 	case MEDIUM_ERROR:
1494 		good_bytes = sd_completed_bytes(SCpnt);
1495 		break;
1496 	case RECOVERED_ERROR:
1497 		good_bytes = scsi_bufflen(SCpnt);
1498 		break;
1499 	case NO_SENSE:
1500 		/* This indicates a false check condition, so ignore it.  An
1501 		 * unknown amount of data was transferred so treat it as an
1502 		 * error.
1503 		 */
1504 		scsi_print_sense("sd", SCpnt);
1505 		SCpnt->result = 0;
1506 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1507 		break;
1508 	case ABORTED_COMMAND:
1509 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1510 			good_bytes = sd_completed_bytes(SCpnt);
1511 		break;
1512 	case ILLEGAL_REQUEST:
1513 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1514 			good_bytes = sd_completed_bytes(SCpnt);
1515 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1516 		if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1517 		    (op == UNMAP || op == WRITE_SAME_16 || op == WRITE_SAME))
1518 			sd_config_discard(sdkp, SD_LBP_DISABLE);
1519 		break;
1520 	default:
1521 		break;
1522 	}
1523  out:
1524 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1525 		sd_dif_complete(SCpnt, good_bytes);
1526 
1527 	if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1528 	    == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1529 
1530 		/* We have to print a failed command here as the
1531 		 * extended CDB gets freed before scsi_io_completion()
1532 		 * is called.
1533 		 */
1534 		if (result)
1535 			scsi_print_command(SCpnt);
1536 
1537 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1538 		SCpnt->cmnd = NULL;
1539 		SCpnt->cmd_len = 0;
1540 	}
1541 
1542 	return good_bytes;
1543 }
1544 
1545 /*
1546  * spinup disk - called only in sd_revalidate_disk()
1547  */
1548 static void
1549 sd_spinup_disk(struct scsi_disk *sdkp)
1550 {
1551 	unsigned char cmd[10];
1552 	unsigned long spintime_expire = 0;
1553 	int retries, spintime;
1554 	unsigned int the_result;
1555 	struct scsi_sense_hdr sshdr;
1556 	int sense_valid = 0;
1557 
1558 	spintime = 0;
1559 
1560 	/* Spin up drives, as required.  Only do this at boot time */
1561 	/* Spinup needs to be done for module loads too. */
1562 	do {
1563 		retries = 0;
1564 
1565 		do {
1566 			cmd[0] = TEST_UNIT_READY;
1567 			memset((void *) &cmd[1], 0, 9);
1568 
1569 			the_result = scsi_execute_req(sdkp->device, cmd,
1570 						      DMA_NONE, NULL, 0,
1571 						      &sshdr, SD_TIMEOUT,
1572 						      SD_MAX_RETRIES, NULL);
1573 
1574 			/*
1575 			 * If the drive has indicated to us that it
1576 			 * doesn't have any media in it, don't bother
1577 			 * with any more polling.
1578 			 */
1579 			if (media_not_present(sdkp, &sshdr))
1580 				return;
1581 
1582 			if (the_result)
1583 				sense_valid = scsi_sense_valid(&sshdr);
1584 			retries++;
1585 		} while (retries < 3 &&
1586 			 (!scsi_status_is_good(the_result) ||
1587 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1588 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1589 
1590 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1591 			/* no sense, TUR either succeeded or failed
1592 			 * with a status error */
1593 			if(!spintime && !scsi_status_is_good(the_result)) {
1594 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1595 				sd_print_result(sdkp, the_result);
1596 			}
1597 			break;
1598 		}
1599 
1600 		/*
1601 		 * The device does not want the automatic start to be issued.
1602 		 */
1603 		if (sdkp->device->no_start_on_add)
1604 			break;
1605 
1606 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1607 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1608 				break;	/* manual intervention required */
1609 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1610 				break;	/* standby */
1611 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1612 				break;	/* unavailable */
1613 			/*
1614 			 * Issue command to spin up drive when not ready
1615 			 */
1616 			if (!spintime) {
1617 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1618 				cmd[0] = START_STOP;
1619 				cmd[1] = 1;	/* Return immediately */
1620 				memset((void *) &cmd[2], 0, 8);
1621 				cmd[4] = 1;	/* Start spin cycle */
1622 				if (sdkp->device->start_stop_pwr_cond)
1623 					cmd[4] |= 1 << 4;
1624 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1625 						 NULL, 0, &sshdr,
1626 						 SD_TIMEOUT, SD_MAX_RETRIES,
1627 						 NULL);
1628 				spintime_expire = jiffies + 100 * HZ;
1629 				spintime = 1;
1630 			}
1631 			/* Wait 1 second for next try */
1632 			msleep(1000);
1633 			printk(".");
1634 
1635 		/*
1636 		 * Wait for USB flash devices with slow firmware.
1637 		 * Yes, this sense key/ASC combination shouldn't
1638 		 * occur here.  It's characteristic of these devices.
1639 		 */
1640 		} else if (sense_valid &&
1641 				sshdr.sense_key == UNIT_ATTENTION &&
1642 				sshdr.asc == 0x28) {
1643 			if (!spintime) {
1644 				spintime_expire = jiffies + 5 * HZ;
1645 				spintime = 1;
1646 			}
1647 			/* Wait 1 second for next try */
1648 			msleep(1000);
1649 		} else {
1650 			/* we don't understand the sense code, so it's
1651 			 * probably pointless to loop */
1652 			if(!spintime) {
1653 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1654 				sd_print_sense_hdr(sdkp, &sshdr);
1655 			}
1656 			break;
1657 		}
1658 
1659 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1660 
1661 	if (spintime) {
1662 		if (scsi_status_is_good(the_result))
1663 			printk("ready\n");
1664 		else
1665 			printk("not responding...\n");
1666 	}
1667 }
1668 
1669 
1670 /*
1671  * Determine whether disk supports Data Integrity Field.
1672  */
1673 static void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1674 {
1675 	struct scsi_device *sdp = sdkp->device;
1676 	u8 type;
1677 
1678 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1679 		return;
1680 
1681 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1682 
1683 	if (type == sdkp->protection_type || !sdkp->first_scan)
1684 		return;
1685 
1686 	sdkp->protection_type = type;
1687 
1688 	if (type > SD_DIF_TYPE3_PROTECTION) {
1689 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported "	\
1690 			  "protection type %u. Disabling disk!\n", type);
1691 		sdkp->capacity = 0;
1692 		return;
1693 	}
1694 
1695 	if (scsi_host_dif_capable(sdp->host, type))
1696 		sd_printk(KERN_NOTICE, sdkp,
1697 			  "Enabling DIF Type %u protection\n", type);
1698 	else
1699 		sd_printk(KERN_NOTICE, sdkp,
1700 			  "Disabling DIF Type %u protection\n", type);
1701 }
1702 
1703 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1704 			struct scsi_sense_hdr *sshdr, int sense_valid,
1705 			int the_result)
1706 {
1707 	sd_print_result(sdkp, the_result);
1708 	if (driver_byte(the_result) & DRIVER_SENSE)
1709 		sd_print_sense_hdr(sdkp, sshdr);
1710 	else
1711 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1712 
1713 	/*
1714 	 * Set dirty bit for removable devices if not ready -
1715 	 * sometimes drives will not report this properly.
1716 	 */
1717 	if (sdp->removable &&
1718 	    sense_valid && sshdr->sense_key == NOT_READY)
1719 		set_media_not_present(sdkp);
1720 
1721 	/*
1722 	 * We used to set media_present to 0 here to indicate no media
1723 	 * in the drive, but some drives fail read capacity even with
1724 	 * media present, so we can't do that.
1725 	 */
1726 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1727 }
1728 
1729 #define RC16_LEN 32
1730 #if RC16_LEN > SD_BUF_SIZE
1731 #error RC16_LEN must not be more than SD_BUF_SIZE
1732 #endif
1733 
1734 #define READ_CAPACITY_RETRIES_ON_RESET	10
1735 
1736 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1737 						unsigned char *buffer)
1738 {
1739 	unsigned char cmd[16];
1740 	struct scsi_sense_hdr sshdr;
1741 	int sense_valid = 0;
1742 	int the_result;
1743 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1744 	unsigned int alignment;
1745 	unsigned long long lba;
1746 	unsigned sector_size;
1747 
1748 	if (sdp->no_read_capacity_16)
1749 		return -EINVAL;
1750 
1751 	do {
1752 		memset(cmd, 0, 16);
1753 		cmd[0] = SERVICE_ACTION_IN;
1754 		cmd[1] = SAI_READ_CAPACITY_16;
1755 		cmd[13] = RC16_LEN;
1756 		memset(buffer, 0, RC16_LEN);
1757 
1758 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1759 					buffer, RC16_LEN, &sshdr,
1760 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1761 
1762 		if (media_not_present(sdkp, &sshdr))
1763 			return -ENODEV;
1764 
1765 		if (the_result) {
1766 			sense_valid = scsi_sense_valid(&sshdr);
1767 			if (sense_valid &&
1768 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1769 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1770 			    sshdr.ascq == 0x00)
1771 				/* Invalid Command Operation Code or
1772 				 * Invalid Field in CDB, just retry
1773 				 * silently with RC10 */
1774 				return -EINVAL;
1775 			if (sense_valid &&
1776 			    sshdr.sense_key == UNIT_ATTENTION &&
1777 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1778 				/* Device reset might occur several times,
1779 				 * give it one more chance */
1780 				if (--reset_retries > 0)
1781 					continue;
1782 		}
1783 		retries--;
1784 
1785 	} while (the_result && retries);
1786 
1787 	if (the_result) {
1788 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1789 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1790 		return -EINVAL;
1791 	}
1792 
1793 	sector_size = get_unaligned_be32(&buffer[8]);
1794 	lba = get_unaligned_be64(&buffer[0]);
1795 
1796 	sd_read_protection_type(sdkp, buffer);
1797 
1798 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1799 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1800 			"kernel compiled with support for large block "
1801 			"devices.\n");
1802 		sdkp->capacity = 0;
1803 		return -EOVERFLOW;
1804 	}
1805 
1806 	/* Logical blocks per physical block exponent */
1807 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
1808 
1809 	/* Lowest aligned logical block */
1810 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1811 	blk_queue_alignment_offset(sdp->request_queue, alignment);
1812 	if (alignment && sdkp->first_scan)
1813 		sd_printk(KERN_NOTICE, sdkp,
1814 			  "physical block alignment offset: %u\n", alignment);
1815 
1816 	if (buffer[14] & 0x80) { /* LBPME */
1817 		sdkp->lbpme = 1;
1818 
1819 		if (buffer[14] & 0x40) /* LBPRZ */
1820 			sdkp->lbprz = 1;
1821 
1822 		sd_config_discard(sdkp, SD_LBP_WS16);
1823 	}
1824 
1825 	sdkp->capacity = lba + 1;
1826 	return sector_size;
1827 }
1828 
1829 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1830 						unsigned char *buffer)
1831 {
1832 	unsigned char cmd[16];
1833 	struct scsi_sense_hdr sshdr;
1834 	int sense_valid = 0;
1835 	int the_result;
1836 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1837 	sector_t lba;
1838 	unsigned sector_size;
1839 
1840 	do {
1841 		cmd[0] = READ_CAPACITY;
1842 		memset(&cmd[1], 0, 9);
1843 		memset(buffer, 0, 8);
1844 
1845 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1846 					buffer, 8, &sshdr,
1847 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1848 
1849 		if (media_not_present(sdkp, &sshdr))
1850 			return -ENODEV;
1851 
1852 		if (the_result) {
1853 			sense_valid = scsi_sense_valid(&sshdr);
1854 			if (sense_valid &&
1855 			    sshdr.sense_key == UNIT_ATTENTION &&
1856 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1857 				/* Device reset might occur several times,
1858 				 * give it one more chance */
1859 				if (--reset_retries > 0)
1860 					continue;
1861 		}
1862 		retries--;
1863 
1864 	} while (the_result && retries);
1865 
1866 	if (the_result) {
1867 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1868 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1869 		return -EINVAL;
1870 	}
1871 
1872 	sector_size = get_unaligned_be32(&buffer[4]);
1873 	lba = get_unaligned_be32(&buffer[0]);
1874 
1875 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
1876 		/* Some buggy (usb cardreader) devices return an lba of
1877 		   0xffffffff when the want to report a size of 0 (with
1878 		   which they really mean no media is present) */
1879 		sdkp->capacity = 0;
1880 		sdkp->physical_block_size = sector_size;
1881 		return sector_size;
1882 	}
1883 
1884 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1885 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1886 			"kernel compiled with support for large block "
1887 			"devices.\n");
1888 		sdkp->capacity = 0;
1889 		return -EOVERFLOW;
1890 	}
1891 
1892 	sdkp->capacity = lba + 1;
1893 	sdkp->physical_block_size = sector_size;
1894 	return sector_size;
1895 }
1896 
1897 static int sd_try_rc16_first(struct scsi_device *sdp)
1898 {
1899 	if (sdp->host->max_cmd_len < 16)
1900 		return 0;
1901 	if (sdp->scsi_level > SCSI_SPC_2)
1902 		return 1;
1903 	if (scsi_device_protection(sdp))
1904 		return 1;
1905 	return 0;
1906 }
1907 
1908 /*
1909  * read disk capacity
1910  */
1911 static void
1912 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1913 {
1914 	int sector_size;
1915 	struct scsi_device *sdp = sdkp->device;
1916 	sector_t old_capacity = sdkp->capacity;
1917 
1918 	if (sd_try_rc16_first(sdp)) {
1919 		sector_size = read_capacity_16(sdkp, sdp, buffer);
1920 		if (sector_size == -EOVERFLOW)
1921 			goto got_data;
1922 		if (sector_size == -ENODEV)
1923 			return;
1924 		if (sector_size < 0)
1925 			sector_size = read_capacity_10(sdkp, sdp, buffer);
1926 		if (sector_size < 0)
1927 			return;
1928 	} else {
1929 		sector_size = read_capacity_10(sdkp, sdp, buffer);
1930 		if (sector_size == -EOVERFLOW)
1931 			goto got_data;
1932 		if (sector_size < 0)
1933 			return;
1934 		if ((sizeof(sdkp->capacity) > 4) &&
1935 		    (sdkp->capacity > 0xffffffffULL)) {
1936 			int old_sector_size = sector_size;
1937 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1938 					"Trying to use READ CAPACITY(16).\n");
1939 			sector_size = read_capacity_16(sdkp, sdp, buffer);
1940 			if (sector_size < 0) {
1941 				sd_printk(KERN_NOTICE, sdkp,
1942 					"Using 0xffffffff as device size\n");
1943 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
1944 				sector_size = old_sector_size;
1945 				goto got_data;
1946 			}
1947 		}
1948 	}
1949 
1950 	/* Some devices are known to return the total number of blocks,
1951 	 * not the highest block number.  Some devices have versions
1952 	 * which do this and others which do not.  Some devices we might
1953 	 * suspect of doing this but we don't know for certain.
1954 	 *
1955 	 * If we know the reported capacity is wrong, decrement it.  If
1956 	 * we can only guess, then assume the number of blocks is even
1957 	 * (usually true but not always) and err on the side of lowering
1958 	 * the capacity.
1959 	 */
1960 	if (sdp->fix_capacity ||
1961 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1962 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1963 				"from its reported value: %llu\n",
1964 				(unsigned long long) sdkp->capacity);
1965 		--sdkp->capacity;
1966 	}
1967 
1968 got_data:
1969 	if (sector_size == 0) {
1970 		sector_size = 512;
1971 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
1972 			  "assuming 512.\n");
1973 	}
1974 
1975 	if (sector_size != 512 &&
1976 	    sector_size != 1024 &&
1977 	    sector_size != 2048 &&
1978 	    sector_size != 4096 &&
1979 	    sector_size != 256) {
1980 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
1981 			  sector_size);
1982 		/*
1983 		 * The user might want to re-format the drive with
1984 		 * a supported sectorsize.  Once this happens, it
1985 		 * would be relatively trivial to set the thing up.
1986 		 * For this reason, we leave the thing in the table.
1987 		 */
1988 		sdkp->capacity = 0;
1989 		/*
1990 		 * set a bogus sector size so the normal read/write
1991 		 * logic in the block layer will eventually refuse any
1992 		 * request on this device without tripping over power
1993 		 * of two sector size assumptions
1994 		 */
1995 		sector_size = 512;
1996 	}
1997 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
1998 
1999 	{
2000 		char cap_str_2[10], cap_str_10[10];
2001 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
2002 
2003 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
2004 				sizeof(cap_str_2));
2005 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
2006 				sizeof(cap_str_10));
2007 
2008 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2009 			sd_printk(KERN_NOTICE, sdkp,
2010 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2011 				  (unsigned long long)sdkp->capacity,
2012 				  sector_size, cap_str_10, cap_str_2);
2013 
2014 			if (sdkp->physical_block_size != sector_size)
2015 				sd_printk(KERN_NOTICE, sdkp,
2016 					  "%u-byte physical blocks\n",
2017 					  sdkp->physical_block_size);
2018 		}
2019 	}
2020 
2021 	/* Rescale capacity to 512-byte units */
2022 	if (sector_size == 4096)
2023 		sdkp->capacity <<= 3;
2024 	else if (sector_size == 2048)
2025 		sdkp->capacity <<= 2;
2026 	else if (sector_size == 1024)
2027 		sdkp->capacity <<= 1;
2028 	else if (sector_size == 256)
2029 		sdkp->capacity >>= 1;
2030 
2031 	blk_queue_physical_block_size(sdp->request_queue,
2032 				      sdkp->physical_block_size);
2033 	sdkp->device->sector_size = sector_size;
2034 }
2035 
2036 /* called with buffer of length 512 */
2037 static inline int
2038 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2039 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2040 		 struct scsi_sense_hdr *sshdr)
2041 {
2042 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2043 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2044 			       sshdr);
2045 }
2046 
2047 /*
2048  * read write protect setting, if possible - called only in sd_revalidate_disk()
2049  * called with buffer of length SD_BUF_SIZE
2050  */
2051 static void
2052 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2053 {
2054 	int res;
2055 	struct scsi_device *sdp = sdkp->device;
2056 	struct scsi_mode_data data;
2057 	int old_wp = sdkp->write_prot;
2058 
2059 	set_disk_ro(sdkp->disk, 0);
2060 	if (sdp->skip_ms_page_3f) {
2061 		sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2062 		return;
2063 	}
2064 
2065 	if (sdp->use_192_bytes_for_3f) {
2066 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2067 	} else {
2068 		/*
2069 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2070 		 * We have to start carefully: some devices hang if we ask
2071 		 * for more than is available.
2072 		 */
2073 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2074 
2075 		/*
2076 		 * Second attempt: ask for page 0 When only page 0 is
2077 		 * implemented, a request for page 3F may return Sense Key
2078 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2079 		 * CDB.
2080 		 */
2081 		if (!scsi_status_is_good(res))
2082 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2083 
2084 		/*
2085 		 * Third attempt: ask 255 bytes, as we did earlier.
2086 		 */
2087 		if (!scsi_status_is_good(res))
2088 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2089 					       &data, NULL);
2090 	}
2091 
2092 	if (!scsi_status_is_good(res)) {
2093 		sd_printk(KERN_WARNING, sdkp,
2094 			  "Test WP failed, assume Write Enabled\n");
2095 	} else {
2096 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2097 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2098 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2099 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2100 				  sdkp->write_prot ? "on" : "off");
2101 			sd_printk(KERN_DEBUG, sdkp,
2102 				  "Mode Sense: %02x %02x %02x %02x\n",
2103 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2104 		}
2105 	}
2106 }
2107 
2108 /*
2109  * sd_read_cache_type - called only from sd_revalidate_disk()
2110  * called with buffer of length SD_BUF_SIZE
2111  */
2112 static void
2113 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2114 {
2115 	int len = 0, res;
2116 	struct scsi_device *sdp = sdkp->device;
2117 
2118 	int dbd;
2119 	int modepage;
2120 	int first_len;
2121 	struct scsi_mode_data data;
2122 	struct scsi_sense_hdr sshdr;
2123 	int old_wce = sdkp->WCE;
2124 	int old_rcd = sdkp->RCD;
2125 	int old_dpofua = sdkp->DPOFUA;
2126 
2127 	first_len = 4;
2128 	if (sdp->skip_ms_page_8) {
2129 		if (sdp->type == TYPE_RBC)
2130 			goto defaults;
2131 		else {
2132 			if (sdp->skip_ms_page_3f)
2133 				goto defaults;
2134 			modepage = 0x3F;
2135 			if (sdp->use_192_bytes_for_3f)
2136 				first_len = 192;
2137 			dbd = 0;
2138 		}
2139 	} else if (sdp->type == TYPE_RBC) {
2140 		modepage = 6;
2141 		dbd = 8;
2142 	} else {
2143 		modepage = 8;
2144 		dbd = 0;
2145 	}
2146 
2147 	/* cautiously ask */
2148 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2149 			&data, &sshdr);
2150 
2151 	if (!scsi_status_is_good(res))
2152 		goto bad_sense;
2153 
2154 	if (!data.header_length) {
2155 		modepage = 6;
2156 		first_len = 0;
2157 		sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
2158 	}
2159 
2160 	/* that went OK, now ask for the proper length */
2161 	len = data.length;
2162 
2163 	/*
2164 	 * We're only interested in the first three bytes, actually.
2165 	 * But the data cache page is defined for the first 20.
2166 	 */
2167 	if (len < 3)
2168 		goto bad_sense;
2169 	else if (len > SD_BUF_SIZE) {
2170 		sd_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2171 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2172 		len = SD_BUF_SIZE;
2173 	}
2174 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2175 		len = 192;
2176 
2177 	/* Get the data */
2178 	if (len > first_len)
2179 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2180 				&data, &sshdr);
2181 
2182 	if (scsi_status_is_good(res)) {
2183 		int offset = data.header_length + data.block_descriptor_length;
2184 
2185 		while (offset < len) {
2186 			u8 page_code = buffer[offset] & 0x3F;
2187 			u8 spf       = buffer[offset] & 0x40;
2188 
2189 			if (page_code == 8 || page_code == 6) {
2190 				/* We're interested only in the first 3 bytes.
2191 				 */
2192 				if (len - offset <= 2) {
2193 					sd_printk(KERN_ERR, sdkp, "Incomplete "
2194 						  "mode parameter data\n");
2195 					goto defaults;
2196 				} else {
2197 					modepage = page_code;
2198 					goto Page_found;
2199 				}
2200 			} else {
2201 				/* Go to the next page */
2202 				if (spf && len - offset > 3)
2203 					offset += 4 + (buffer[offset+2] << 8) +
2204 						buffer[offset+3];
2205 				else if (!spf && len - offset > 1)
2206 					offset += 2 + buffer[offset+1];
2207 				else {
2208 					sd_printk(KERN_ERR, sdkp, "Incomplete "
2209 						  "mode parameter data\n");
2210 					goto defaults;
2211 				}
2212 			}
2213 		}
2214 
2215 		if (modepage == 0x3F) {
2216 			sd_printk(KERN_ERR, sdkp, "No Caching mode page "
2217 				  "present\n");
2218 			goto defaults;
2219 		} else if ((buffer[offset] & 0x3f) != modepage) {
2220 			sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
2221 			goto defaults;
2222 		}
2223 	Page_found:
2224 		if (modepage == 8) {
2225 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2226 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2227 		} else {
2228 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2229 			sdkp->RCD = 0;
2230 		}
2231 
2232 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2233 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2234 			sd_printk(KERN_NOTICE, sdkp,
2235 				  "Uses READ/WRITE(6), disabling FUA\n");
2236 			sdkp->DPOFUA = 0;
2237 		}
2238 
2239 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2240 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2241 			sd_printk(KERN_NOTICE, sdkp,
2242 				  "Write cache: %s, read cache: %s, %s\n",
2243 				  sdkp->WCE ? "enabled" : "disabled",
2244 				  sdkp->RCD ? "disabled" : "enabled",
2245 				  sdkp->DPOFUA ? "supports DPO and FUA"
2246 				  : "doesn't support DPO or FUA");
2247 
2248 		return;
2249 	}
2250 
2251 bad_sense:
2252 	if (scsi_sense_valid(&sshdr) &&
2253 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2254 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2255 		/* Invalid field in CDB */
2256 		sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2257 	else
2258 		sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
2259 
2260 defaults:
2261 	sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
2262 	sdkp->WCE = 0;
2263 	sdkp->RCD = 0;
2264 	sdkp->DPOFUA = 0;
2265 }
2266 
2267 /*
2268  * The ATO bit indicates whether the DIF application tag is available
2269  * for use by the operating system.
2270  */
2271 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2272 {
2273 	int res, offset;
2274 	struct scsi_device *sdp = sdkp->device;
2275 	struct scsi_mode_data data;
2276 	struct scsi_sense_hdr sshdr;
2277 
2278 	if (sdp->type != TYPE_DISK)
2279 		return;
2280 
2281 	if (sdkp->protection_type == 0)
2282 		return;
2283 
2284 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2285 			      SD_MAX_RETRIES, &data, &sshdr);
2286 
2287 	if (!scsi_status_is_good(res) || !data.header_length ||
2288 	    data.length < 6) {
2289 		sd_printk(KERN_WARNING, sdkp,
2290 			  "getting Control mode page failed, assume no ATO\n");
2291 
2292 		if (scsi_sense_valid(&sshdr))
2293 			sd_print_sense_hdr(sdkp, &sshdr);
2294 
2295 		return;
2296 	}
2297 
2298 	offset = data.header_length + data.block_descriptor_length;
2299 
2300 	if ((buffer[offset] & 0x3f) != 0x0a) {
2301 		sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2302 		return;
2303 	}
2304 
2305 	if ((buffer[offset + 5] & 0x80) == 0)
2306 		return;
2307 
2308 	sdkp->ATO = 1;
2309 
2310 	return;
2311 }
2312 
2313 /**
2314  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2315  * @disk: disk to query
2316  */
2317 static void sd_read_block_limits(struct scsi_disk *sdkp)
2318 {
2319 	unsigned int sector_sz = sdkp->device->sector_size;
2320 	const int vpd_len = 64;
2321 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2322 
2323 	if (!buffer ||
2324 	    /* Block Limits VPD */
2325 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2326 		goto out;
2327 
2328 	blk_queue_io_min(sdkp->disk->queue,
2329 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2330 	blk_queue_io_opt(sdkp->disk->queue,
2331 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2332 
2333 	if (buffer[3] == 0x3c) {
2334 		unsigned int lba_count, desc_count;
2335 
2336 		sdkp->max_ws_blocks =
2337 			(u32) min_not_zero(get_unaligned_be64(&buffer[36]),
2338 					   (u64)0xffffffff);
2339 
2340 		if (!sdkp->lbpme)
2341 			goto out;
2342 
2343 		lba_count = get_unaligned_be32(&buffer[20]);
2344 		desc_count = get_unaligned_be32(&buffer[24]);
2345 
2346 		if (lba_count && desc_count)
2347 			sdkp->max_unmap_blocks = lba_count;
2348 
2349 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2350 
2351 		if (buffer[32] & 0x80)
2352 			sdkp->unmap_alignment =
2353 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2354 
2355 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2356 
2357 			if (sdkp->max_unmap_blocks)
2358 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2359 			else
2360 				sd_config_discard(sdkp, SD_LBP_WS16);
2361 
2362 		} else {	/* LBP VPD page tells us what to use */
2363 
2364 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2365 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2366 			else if (sdkp->lbpws)
2367 				sd_config_discard(sdkp, SD_LBP_WS16);
2368 			else if (sdkp->lbpws10)
2369 				sd_config_discard(sdkp, SD_LBP_WS10);
2370 			else
2371 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2372 		}
2373 	}
2374 
2375  out:
2376 	kfree(buffer);
2377 }
2378 
2379 /**
2380  * sd_read_block_characteristics - Query block dev. characteristics
2381  * @disk: disk to query
2382  */
2383 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2384 {
2385 	unsigned char *buffer;
2386 	u16 rot;
2387 	const int vpd_len = 64;
2388 
2389 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2390 
2391 	if (!buffer ||
2392 	    /* Block Device Characteristics VPD */
2393 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2394 		goto out;
2395 
2396 	rot = get_unaligned_be16(&buffer[4]);
2397 
2398 	if (rot == 1)
2399 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2400 
2401  out:
2402 	kfree(buffer);
2403 }
2404 
2405 /**
2406  * sd_read_block_provisioning - Query provisioning VPD page
2407  * @disk: disk to query
2408  */
2409 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2410 {
2411 	unsigned char *buffer;
2412 	const int vpd_len = 8;
2413 
2414 	if (sdkp->lbpme == 0)
2415 		return;
2416 
2417 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2418 
2419 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2420 		goto out;
2421 
2422 	sdkp->lbpvpd	= 1;
2423 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2424 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2425 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2426 
2427  out:
2428 	kfree(buffer);
2429 }
2430 
2431 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2432 {
2433 	/*
2434 	 * Although VPD inquiries can go to SCSI-2 type devices,
2435 	 * some USB ones crash on receiving them, and the pages
2436 	 * we currently ask for are for SPC-3 and beyond
2437 	 */
2438 	if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2439 		return 1;
2440 	return 0;
2441 }
2442 
2443 /**
2444  *	sd_revalidate_disk - called the first time a new disk is seen,
2445  *	performs disk spin up, read_capacity, etc.
2446  *	@disk: struct gendisk we care about
2447  **/
2448 static int sd_revalidate_disk(struct gendisk *disk)
2449 {
2450 	struct scsi_disk *sdkp = scsi_disk(disk);
2451 	struct scsi_device *sdp = sdkp->device;
2452 	unsigned char *buffer;
2453 	unsigned flush = 0;
2454 
2455 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2456 				      "sd_revalidate_disk\n"));
2457 
2458 	/*
2459 	 * If the device is offline, don't try and read capacity or any
2460 	 * of the other niceties.
2461 	 */
2462 	if (!scsi_device_online(sdp))
2463 		goto out;
2464 
2465 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2466 	if (!buffer) {
2467 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2468 			  "allocation failure.\n");
2469 		goto out;
2470 	}
2471 
2472 	sd_spinup_disk(sdkp);
2473 
2474 	/*
2475 	 * Without media there is no reason to ask; moreover, some devices
2476 	 * react badly if we do.
2477 	 */
2478 	if (sdkp->media_present) {
2479 		sd_read_capacity(sdkp, buffer);
2480 
2481 		if (sd_try_extended_inquiry(sdp)) {
2482 			sd_read_block_provisioning(sdkp);
2483 			sd_read_block_limits(sdkp);
2484 			sd_read_block_characteristics(sdkp);
2485 		}
2486 
2487 		sd_read_write_protect_flag(sdkp, buffer);
2488 		sd_read_cache_type(sdkp, buffer);
2489 		sd_read_app_tag_own(sdkp, buffer);
2490 	}
2491 
2492 	sdkp->first_scan = 0;
2493 
2494 	/*
2495 	 * We now have all cache related info, determine how we deal
2496 	 * with flush requests.
2497 	 */
2498 	if (sdkp->WCE) {
2499 		flush |= REQ_FLUSH;
2500 		if (sdkp->DPOFUA)
2501 			flush |= REQ_FUA;
2502 	}
2503 
2504 	blk_queue_flush(sdkp->disk->queue, flush);
2505 
2506 	set_capacity(disk, sdkp->capacity);
2507 	kfree(buffer);
2508 
2509  out:
2510 	return 0;
2511 }
2512 
2513 /**
2514  *	sd_unlock_native_capacity - unlock native capacity
2515  *	@disk: struct gendisk to set capacity for
2516  *
2517  *	Block layer calls this function if it detects that partitions
2518  *	on @disk reach beyond the end of the device.  If the SCSI host
2519  *	implements ->unlock_native_capacity() method, it's invoked to
2520  *	give it a chance to adjust the device capacity.
2521  *
2522  *	CONTEXT:
2523  *	Defined by block layer.  Might sleep.
2524  */
2525 static void sd_unlock_native_capacity(struct gendisk *disk)
2526 {
2527 	struct scsi_device *sdev = scsi_disk(disk)->device;
2528 
2529 	if (sdev->host->hostt->unlock_native_capacity)
2530 		sdev->host->hostt->unlock_native_capacity(sdev);
2531 }
2532 
2533 /**
2534  *	sd_format_disk_name - format disk name
2535  *	@prefix: name prefix - ie. "sd" for SCSI disks
2536  *	@index: index of the disk to format name for
2537  *	@buf: output buffer
2538  *	@buflen: length of the output buffer
2539  *
2540  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2541  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2542  *	which is followed by sdaaa.
2543  *
2544  *	This is basically 26 base counting with one extra 'nil' entry
2545  *	at the beginning from the second digit on and can be
2546  *	determined using similar method as 26 base conversion with the
2547  *	index shifted -1 after each digit is computed.
2548  *
2549  *	CONTEXT:
2550  *	Don't care.
2551  *
2552  *	RETURNS:
2553  *	0 on success, -errno on failure.
2554  */
2555 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2556 {
2557 	const int base = 'z' - 'a' + 1;
2558 	char *begin = buf + strlen(prefix);
2559 	char *end = buf + buflen;
2560 	char *p;
2561 	int unit;
2562 
2563 	p = end - 1;
2564 	*p = '\0';
2565 	unit = base;
2566 	do {
2567 		if (p == begin)
2568 			return -EINVAL;
2569 		*--p = 'a' + (index % unit);
2570 		index = (index / unit) - 1;
2571 	} while (index >= 0);
2572 
2573 	memmove(begin, p, end - p);
2574 	memcpy(buf, prefix, strlen(prefix));
2575 
2576 	return 0;
2577 }
2578 
2579 /*
2580  * The asynchronous part of sd_probe
2581  */
2582 static void sd_probe_async(void *data, async_cookie_t cookie)
2583 {
2584 	struct scsi_disk *sdkp = data;
2585 	struct scsi_device *sdp;
2586 	struct gendisk *gd;
2587 	u32 index;
2588 	struct device *dev;
2589 
2590 	sdp = sdkp->device;
2591 	gd = sdkp->disk;
2592 	index = sdkp->index;
2593 	dev = &sdp->sdev_gendev;
2594 
2595 	gd->major = sd_major((index & 0xf0) >> 4);
2596 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2597 	gd->minors = SD_MINORS;
2598 
2599 	gd->fops = &sd_fops;
2600 	gd->private_data = &sdkp->driver;
2601 	gd->queue = sdkp->device->request_queue;
2602 
2603 	/* defaults, until the device tells us otherwise */
2604 	sdp->sector_size = 512;
2605 	sdkp->capacity = 0;
2606 	sdkp->media_present = 1;
2607 	sdkp->write_prot = 0;
2608 	sdkp->WCE = 0;
2609 	sdkp->RCD = 0;
2610 	sdkp->ATO = 0;
2611 	sdkp->first_scan = 1;
2612 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2613 
2614 	sd_revalidate_disk(gd);
2615 
2616 	blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2617 	blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2618 
2619 	gd->driverfs_dev = &sdp->sdev_gendev;
2620 	gd->flags = GENHD_FL_EXT_DEVT;
2621 	if (sdp->removable) {
2622 		gd->flags |= GENHD_FL_REMOVABLE;
2623 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2624 	}
2625 
2626 	add_disk(gd);
2627 	sd_dif_config_host(sdkp);
2628 
2629 	sd_revalidate_disk(gd);
2630 
2631 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2632 		  sdp->removable ? "removable " : "");
2633 	scsi_autopm_put_device(sdp);
2634 	put_device(&sdkp->dev);
2635 }
2636 
2637 /**
2638  *	sd_probe - called during driver initialization and whenever a
2639  *	new scsi device is attached to the system. It is called once
2640  *	for each scsi device (not just disks) present.
2641  *	@dev: pointer to device object
2642  *
2643  *	Returns 0 if successful (or not interested in this scsi device
2644  *	(e.g. scanner)); 1 when there is an error.
2645  *
2646  *	Note: this function is invoked from the scsi mid-level.
2647  *	This function sets up the mapping between a given
2648  *	<host,channel,id,lun> (found in sdp) and new device name
2649  *	(e.g. /dev/sda). More precisely it is the block device major
2650  *	and minor number that is chosen here.
2651  *
2652  *	Assume sd_probe is not re-entrant (for time being)
2653  *	Also think about sd_probe() and sd_remove() running coincidentally.
2654  **/
2655 static int sd_probe(struct device *dev)
2656 {
2657 	struct scsi_device *sdp = to_scsi_device(dev);
2658 	struct scsi_disk *sdkp;
2659 	struct gendisk *gd;
2660 	int index;
2661 	int error;
2662 
2663 	error = -ENODEV;
2664 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2665 		goto out;
2666 
2667 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2668 					"sd_probe\n"));
2669 
2670 	error = -ENOMEM;
2671 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2672 	if (!sdkp)
2673 		goto out;
2674 
2675 	gd = alloc_disk(SD_MINORS);
2676 	if (!gd)
2677 		goto out_free;
2678 
2679 	do {
2680 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2681 			goto out_put;
2682 
2683 		spin_lock(&sd_index_lock);
2684 		error = ida_get_new(&sd_index_ida, &index);
2685 		spin_unlock(&sd_index_lock);
2686 	} while (error == -EAGAIN);
2687 
2688 	if (error) {
2689 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2690 		goto out_put;
2691 	}
2692 
2693 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2694 	if (error) {
2695 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2696 		goto out_free_index;
2697 	}
2698 
2699 	sdkp->device = sdp;
2700 	sdkp->driver = &sd_template;
2701 	sdkp->disk = gd;
2702 	sdkp->index = index;
2703 	atomic_set(&sdkp->openers, 0);
2704 
2705 	if (!sdp->request_queue->rq_timeout) {
2706 		if (sdp->type != TYPE_MOD)
2707 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2708 		else
2709 			blk_queue_rq_timeout(sdp->request_queue,
2710 					     SD_MOD_TIMEOUT);
2711 	}
2712 
2713 	device_initialize(&sdkp->dev);
2714 	sdkp->dev.parent = dev;
2715 	sdkp->dev.class = &sd_disk_class;
2716 	dev_set_name(&sdkp->dev, dev_name(dev));
2717 
2718 	if (device_add(&sdkp->dev))
2719 		goto out_free_index;
2720 
2721 	get_device(dev);
2722 	dev_set_drvdata(dev, sdkp);
2723 
2724 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2725 	async_schedule(sd_probe_async, sdkp);
2726 
2727 	return 0;
2728 
2729  out_free_index:
2730 	spin_lock(&sd_index_lock);
2731 	ida_remove(&sd_index_ida, index);
2732 	spin_unlock(&sd_index_lock);
2733  out_put:
2734 	put_disk(gd);
2735  out_free:
2736 	kfree(sdkp);
2737  out:
2738 	return error;
2739 }
2740 
2741 /**
2742  *	sd_remove - called whenever a scsi disk (previously recognized by
2743  *	sd_probe) is detached from the system. It is called (potentially
2744  *	multiple times) during sd module unload.
2745  *	@sdp: pointer to mid level scsi device object
2746  *
2747  *	Note: this function is invoked from the scsi mid-level.
2748  *	This function potentially frees up a device name (e.g. /dev/sdc)
2749  *	that could be re-used by a subsequent sd_probe().
2750  *	This function is not called when the built-in sd driver is "exit-ed".
2751  **/
2752 static int sd_remove(struct device *dev)
2753 {
2754 	struct scsi_disk *sdkp;
2755 
2756 	sdkp = dev_get_drvdata(dev);
2757 	scsi_autopm_get_device(sdkp->device);
2758 
2759 	async_synchronize_full();
2760 	blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2761 	blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2762 	device_del(&sdkp->dev);
2763 	del_gendisk(sdkp->disk);
2764 	sd_shutdown(dev);
2765 
2766 	mutex_lock(&sd_ref_mutex);
2767 	dev_set_drvdata(dev, NULL);
2768 	put_device(&sdkp->dev);
2769 	mutex_unlock(&sd_ref_mutex);
2770 
2771 	return 0;
2772 }
2773 
2774 /**
2775  *	scsi_disk_release - Called to free the scsi_disk structure
2776  *	@dev: pointer to embedded class device
2777  *
2778  *	sd_ref_mutex must be held entering this routine.  Because it is
2779  *	called on last put, you should always use the scsi_disk_get()
2780  *	scsi_disk_put() helpers which manipulate the semaphore directly
2781  *	and never do a direct put_device.
2782  **/
2783 static void scsi_disk_release(struct device *dev)
2784 {
2785 	struct scsi_disk *sdkp = to_scsi_disk(dev);
2786 	struct gendisk *disk = sdkp->disk;
2787 
2788 	spin_lock(&sd_index_lock);
2789 	ida_remove(&sd_index_ida, sdkp->index);
2790 	spin_unlock(&sd_index_lock);
2791 
2792 	disk->private_data = NULL;
2793 	put_disk(disk);
2794 	put_device(&sdkp->device->sdev_gendev);
2795 
2796 	kfree(sdkp);
2797 }
2798 
2799 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2800 {
2801 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
2802 	struct scsi_sense_hdr sshdr;
2803 	struct scsi_device *sdp = sdkp->device;
2804 	int res;
2805 
2806 	if (start)
2807 		cmd[4] |= 1;	/* START */
2808 
2809 	if (sdp->start_stop_pwr_cond)
2810 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
2811 
2812 	if (!scsi_device_online(sdp))
2813 		return -ENODEV;
2814 
2815 	res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2816 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2817 	if (res) {
2818 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2819 		sd_print_result(sdkp, res);
2820 		if (driver_byte(res) & DRIVER_SENSE)
2821 			sd_print_sense_hdr(sdkp, &sshdr);
2822 	}
2823 
2824 	return res;
2825 }
2826 
2827 /*
2828  * Send a SYNCHRONIZE CACHE instruction down to the device through
2829  * the normal SCSI command structure.  Wait for the command to
2830  * complete.
2831  */
2832 static void sd_shutdown(struct device *dev)
2833 {
2834 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2835 
2836 	if (!sdkp)
2837 		return;         /* this can happen */
2838 
2839 	if (pm_runtime_suspended(dev))
2840 		goto exit;
2841 
2842 	if (sdkp->WCE) {
2843 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2844 		sd_sync_cache(sdkp);
2845 	}
2846 
2847 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2848 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2849 		sd_start_stop_device(sdkp, 0);
2850 	}
2851 
2852 exit:
2853 	scsi_disk_put(sdkp);
2854 }
2855 
2856 static int sd_suspend(struct device *dev, pm_message_t mesg)
2857 {
2858 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2859 	int ret = 0;
2860 
2861 	if (!sdkp)
2862 		return 0;	/* this can happen */
2863 
2864 	if (sdkp->WCE) {
2865 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2866 		ret = sd_sync_cache(sdkp);
2867 		if (ret)
2868 			goto done;
2869 	}
2870 
2871 	if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2872 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2873 		ret = sd_start_stop_device(sdkp, 0);
2874 	}
2875 
2876 done:
2877 	scsi_disk_put(sdkp);
2878 	return ret;
2879 }
2880 
2881 static int sd_resume(struct device *dev)
2882 {
2883 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2884 	int ret = 0;
2885 
2886 	if (!sdkp->device->manage_start_stop)
2887 		goto done;
2888 
2889 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2890 	ret = sd_start_stop_device(sdkp, 1);
2891 
2892 done:
2893 	scsi_disk_put(sdkp);
2894 	return ret;
2895 }
2896 
2897 /**
2898  *	init_sd - entry point for this driver (both when built in or when
2899  *	a module).
2900  *
2901  *	Note: this function registers this driver with the scsi mid-level.
2902  **/
2903 static int __init init_sd(void)
2904 {
2905 	int majors = 0, i, err;
2906 
2907 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2908 
2909 	for (i = 0; i < SD_MAJORS; i++)
2910 		if (register_blkdev(sd_major(i), "sd") == 0)
2911 			majors++;
2912 
2913 	if (!majors)
2914 		return -ENODEV;
2915 
2916 	err = class_register(&sd_disk_class);
2917 	if (err)
2918 		goto err_out;
2919 
2920 	err = scsi_register_driver(&sd_template.gendrv);
2921 	if (err)
2922 		goto err_out_class;
2923 
2924 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2925 					 0, 0, NULL);
2926 	if (!sd_cdb_cache) {
2927 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
2928 		goto err_out_class;
2929 	}
2930 
2931 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2932 	if (!sd_cdb_pool) {
2933 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
2934 		goto err_out_cache;
2935 	}
2936 
2937 	return 0;
2938 
2939 err_out_cache:
2940 	kmem_cache_destroy(sd_cdb_cache);
2941 
2942 err_out_class:
2943 	class_unregister(&sd_disk_class);
2944 err_out:
2945 	for (i = 0; i < SD_MAJORS; i++)
2946 		unregister_blkdev(sd_major(i), "sd");
2947 	return err;
2948 }
2949 
2950 /**
2951  *	exit_sd - exit point for this driver (when it is a module).
2952  *
2953  *	Note: this function unregisters this driver from the scsi mid-level.
2954  **/
2955 static void __exit exit_sd(void)
2956 {
2957 	int i;
2958 
2959 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
2960 
2961 	mempool_destroy(sd_cdb_pool);
2962 	kmem_cache_destroy(sd_cdb_cache);
2963 
2964 	scsi_unregister_driver(&sd_template.gendrv);
2965 	class_unregister(&sd_disk_class);
2966 
2967 	for (i = 0; i < SD_MAJORS; i++)
2968 		unregister_blkdev(sd_major(i), "sd");
2969 }
2970 
2971 module_init(init_sd);
2972 module_exit(exit_sd);
2973 
2974 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
2975 			       struct scsi_sense_hdr *sshdr)
2976 {
2977 	sd_printk(KERN_INFO, sdkp, " ");
2978 	scsi_show_sense_hdr(sshdr);
2979 	sd_printk(KERN_INFO, sdkp, " ");
2980 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
2981 }
2982 
2983 static void sd_print_result(struct scsi_disk *sdkp, int result)
2984 {
2985 	sd_printk(KERN_INFO, sdkp, " ");
2986 	scsi_show_result(result);
2987 }
2988 
2989