xref: /linux/drivers/scsi/sd.c (revision 4c9ca2fd7c402c8eab6e949899af04bc3c249b2b)
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/sed-opal.h>
54 #include <linux/pm_runtime.h>
55 #include <linux/pr.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
59 
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
69 
70 #include "sd.h"
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
73 
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
77 
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
98 
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
100 #define SD_MINORS	16
101 #else
102 #define SD_MINORS	0
103 #endif
104 
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int  sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int  sd_probe(struct device *);
110 static int  sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
125 
126 static DEFINE_IDA(sd_index_ida);
127 
128 /* This semaphore is used to mediate the 0->1 reference get in the
129  * face of object destruction (i.e. we can't allow a get on an
130  * object after last put) */
131 static DEFINE_MUTEX(sd_ref_mutex);
132 
133 static struct kmem_cache *sd_cdb_cache;
134 static mempool_t *sd_cdb_pool;
135 
136 static const char *sd_cache_types[] = {
137 	"write through", "none", "write back",
138 	"write back, no read (daft)"
139 };
140 
141 static void sd_set_flush_flag(struct scsi_disk *sdkp)
142 {
143 	bool wc = false, fua = false;
144 
145 	if (sdkp->WCE) {
146 		wc = true;
147 		if (sdkp->DPOFUA)
148 			fua = true;
149 	}
150 
151 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
152 }
153 
154 static ssize_t
155 cache_type_store(struct device *dev, struct device_attribute *attr,
156 		 const char *buf, size_t count)
157 {
158 	int ct, rcd, wce, sp;
159 	struct scsi_disk *sdkp = to_scsi_disk(dev);
160 	struct scsi_device *sdp = sdkp->device;
161 	char buffer[64];
162 	char *buffer_data;
163 	struct scsi_mode_data data;
164 	struct scsi_sense_hdr sshdr;
165 	static const char temp[] = "temporary ";
166 	int len;
167 
168 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
169 		/* no cache control on RBC devices; theoretically they
170 		 * can do it, but there's probably so many exceptions
171 		 * it's not worth the risk */
172 		return -EINVAL;
173 
174 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
175 		buf += sizeof(temp) - 1;
176 		sdkp->cache_override = 1;
177 	} else {
178 		sdkp->cache_override = 0;
179 	}
180 
181 	ct = sysfs_match_string(sd_cache_types, buf);
182 	if (ct < 0)
183 		return -EINVAL;
184 
185 	rcd = ct & 0x01 ? 1 : 0;
186 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
187 
188 	if (sdkp->cache_override) {
189 		sdkp->WCE = wce;
190 		sdkp->RCD = rcd;
191 		sd_set_flush_flag(sdkp);
192 		return count;
193 	}
194 
195 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
196 			    SD_MAX_RETRIES, &data, NULL))
197 		return -EINVAL;
198 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
199 		  data.block_descriptor_length);
200 	buffer_data = buffer + data.header_length +
201 		data.block_descriptor_length;
202 	buffer_data[2] &= ~0x05;
203 	buffer_data[2] |= wce << 2 | rcd;
204 	sp = buffer_data[0] & 0x80 ? 1 : 0;
205 	buffer_data[0] &= ~0x80;
206 
207 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
208 			     SD_MAX_RETRIES, &data, &sshdr)) {
209 		if (scsi_sense_valid(&sshdr))
210 			sd_print_sense_hdr(sdkp, &sshdr);
211 		return -EINVAL;
212 	}
213 	revalidate_disk(sdkp->disk);
214 	return count;
215 }
216 
217 static ssize_t
218 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
219 		       char *buf)
220 {
221 	struct scsi_disk *sdkp = to_scsi_disk(dev);
222 	struct scsi_device *sdp = sdkp->device;
223 
224 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
225 }
226 
227 static ssize_t
228 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
229 			const char *buf, size_t count)
230 {
231 	struct scsi_disk *sdkp = to_scsi_disk(dev);
232 	struct scsi_device *sdp = sdkp->device;
233 	bool v;
234 
235 	if (!capable(CAP_SYS_ADMIN))
236 		return -EACCES;
237 
238 	if (kstrtobool(buf, &v))
239 		return -EINVAL;
240 
241 	sdp->manage_start_stop = v;
242 
243 	return count;
244 }
245 static DEVICE_ATTR_RW(manage_start_stop);
246 
247 static ssize_t
248 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
249 {
250 	struct scsi_disk *sdkp = to_scsi_disk(dev);
251 
252 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
253 }
254 
255 static ssize_t
256 allow_restart_store(struct device *dev, struct device_attribute *attr,
257 		    const char *buf, size_t count)
258 {
259 	bool v;
260 	struct scsi_disk *sdkp = to_scsi_disk(dev);
261 	struct scsi_device *sdp = sdkp->device;
262 
263 	if (!capable(CAP_SYS_ADMIN))
264 		return -EACCES;
265 
266 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
267 		return -EINVAL;
268 
269 	if (kstrtobool(buf, &v))
270 		return -EINVAL;
271 
272 	sdp->allow_restart = v;
273 
274 	return count;
275 }
276 static DEVICE_ATTR_RW(allow_restart);
277 
278 static ssize_t
279 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
280 {
281 	struct scsi_disk *sdkp = to_scsi_disk(dev);
282 	int ct = sdkp->RCD + 2*sdkp->WCE;
283 
284 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
285 }
286 static DEVICE_ATTR_RW(cache_type);
287 
288 static ssize_t
289 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
290 {
291 	struct scsi_disk *sdkp = to_scsi_disk(dev);
292 
293 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
294 }
295 static DEVICE_ATTR_RO(FUA);
296 
297 static ssize_t
298 protection_type_show(struct device *dev, struct device_attribute *attr,
299 		     char *buf)
300 {
301 	struct scsi_disk *sdkp = to_scsi_disk(dev);
302 
303 	return sprintf(buf, "%u\n", sdkp->protection_type);
304 }
305 
306 static ssize_t
307 protection_type_store(struct device *dev, struct device_attribute *attr,
308 		      const char *buf, size_t count)
309 {
310 	struct scsi_disk *sdkp = to_scsi_disk(dev);
311 	unsigned int val;
312 	int err;
313 
314 	if (!capable(CAP_SYS_ADMIN))
315 		return -EACCES;
316 
317 	err = kstrtouint(buf, 10, &val);
318 
319 	if (err)
320 		return err;
321 
322 	if (val <= T10_PI_TYPE3_PROTECTION)
323 		sdkp->protection_type = val;
324 
325 	return count;
326 }
327 static DEVICE_ATTR_RW(protection_type);
328 
329 static ssize_t
330 protection_mode_show(struct device *dev, struct device_attribute *attr,
331 		     char *buf)
332 {
333 	struct scsi_disk *sdkp = to_scsi_disk(dev);
334 	struct scsi_device *sdp = sdkp->device;
335 	unsigned int dif, dix;
336 
337 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
338 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
339 
340 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
341 		dif = 0;
342 		dix = 1;
343 	}
344 
345 	if (!dif && !dix)
346 		return sprintf(buf, "none\n");
347 
348 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
349 }
350 static DEVICE_ATTR_RO(protection_mode);
351 
352 static ssize_t
353 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
354 {
355 	struct scsi_disk *sdkp = to_scsi_disk(dev);
356 
357 	return sprintf(buf, "%u\n", sdkp->ATO);
358 }
359 static DEVICE_ATTR_RO(app_tag_own);
360 
361 static ssize_t
362 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
363 		       char *buf)
364 {
365 	struct scsi_disk *sdkp = to_scsi_disk(dev);
366 
367 	return sprintf(buf, "%u\n", sdkp->lbpme);
368 }
369 static DEVICE_ATTR_RO(thin_provisioning);
370 
371 /* sysfs_match_string() requires dense arrays */
372 static const char *lbp_mode[] = {
373 	[SD_LBP_FULL]		= "full",
374 	[SD_LBP_UNMAP]		= "unmap",
375 	[SD_LBP_WS16]		= "writesame_16",
376 	[SD_LBP_WS10]		= "writesame_10",
377 	[SD_LBP_ZERO]		= "writesame_zero",
378 	[SD_LBP_DISABLE]	= "disabled",
379 };
380 
381 static ssize_t
382 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
383 		       char *buf)
384 {
385 	struct scsi_disk *sdkp = to_scsi_disk(dev);
386 
387 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
388 }
389 
390 static ssize_t
391 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
392 			const char *buf, size_t count)
393 {
394 	struct scsi_disk *sdkp = to_scsi_disk(dev);
395 	struct scsi_device *sdp = sdkp->device;
396 	int mode;
397 
398 	if (!capable(CAP_SYS_ADMIN))
399 		return -EACCES;
400 
401 	if (sd_is_zoned(sdkp)) {
402 		sd_config_discard(sdkp, SD_LBP_DISABLE);
403 		return count;
404 	}
405 
406 	if (sdp->type != TYPE_DISK)
407 		return -EINVAL;
408 
409 	mode = sysfs_match_string(lbp_mode, buf);
410 	if (mode < 0)
411 		return -EINVAL;
412 
413 	sd_config_discard(sdkp, mode);
414 
415 	return count;
416 }
417 static DEVICE_ATTR_RW(provisioning_mode);
418 
419 /* sysfs_match_string() requires dense arrays */
420 static const char *zeroing_mode[] = {
421 	[SD_ZERO_WRITE]		= "write",
422 	[SD_ZERO_WS]		= "writesame",
423 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
424 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
425 };
426 
427 static ssize_t
428 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
429 		  char *buf)
430 {
431 	struct scsi_disk *sdkp = to_scsi_disk(dev);
432 
433 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
434 }
435 
436 static ssize_t
437 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
438 		   const char *buf, size_t count)
439 {
440 	struct scsi_disk *sdkp = to_scsi_disk(dev);
441 	int mode;
442 
443 	if (!capable(CAP_SYS_ADMIN))
444 		return -EACCES;
445 
446 	mode = sysfs_match_string(zeroing_mode, buf);
447 	if (mode < 0)
448 		return -EINVAL;
449 
450 	sdkp->zeroing_mode = mode;
451 
452 	return count;
453 }
454 static DEVICE_ATTR_RW(zeroing_mode);
455 
456 static ssize_t
457 max_medium_access_timeouts_show(struct device *dev,
458 				struct device_attribute *attr, char *buf)
459 {
460 	struct scsi_disk *sdkp = to_scsi_disk(dev);
461 
462 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
463 }
464 
465 static ssize_t
466 max_medium_access_timeouts_store(struct device *dev,
467 				 struct device_attribute *attr, const char *buf,
468 				 size_t count)
469 {
470 	struct scsi_disk *sdkp = to_scsi_disk(dev);
471 	int err;
472 
473 	if (!capable(CAP_SYS_ADMIN))
474 		return -EACCES;
475 
476 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
477 
478 	return err ? err : count;
479 }
480 static DEVICE_ATTR_RW(max_medium_access_timeouts);
481 
482 static ssize_t
483 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
484 			   char *buf)
485 {
486 	struct scsi_disk *sdkp = to_scsi_disk(dev);
487 
488 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
489 }
490 
491 static ssize_t
492 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
493 			    const char *buf, size_t count)
494 {
495 	struct scsi_disk *sdkp = to_scsi_disk(dev);
496 	struct scsi_device *sdp = sdkp->device;
497 	unsigned long max;
498 	int err;
499 
500 	if (!capable(CAP_SYS_ADMIN))
501 		return -EACCES;
502 
503 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
504 		return -EINVAL;
505 
506 	err = kstrtoul(buf, 10, &max);
507 
508 	if (err)
509 		return err;
510 
511 	if (max == 0)
512 		sdp->no_write_same = 1;
513 	else if (max <= SD_MAX_WS16_BLOCKS) {
514 		sdp->no_write_same = 0;
515 		sdkp->max_ws_blocks = max;
516 	}
517 
518 	sd_config_write_same(sdkp);
519 
520 	return count;
521 }
522 static DEVICE_ATTR_RW(max_write_same_blocks);
523 
524 static struct attribute *sd_disk_attrs[] = {
525 	&dev_attr_cache_type.attr,
526 	&dev_attr_FUA.attr,
527 	&dev_attr_allow_restart.attr,
528 	&dev_attr_manage_start_stop.attr,
529 	&dev_attr_protection_type.attr,
530 	&dev_attr_protection_mode.attr,
531 	&dev_attr_app_tag_own.attr,
532 	&dev_attr_thin_provisioning.attr,
533 	&dev_attr_provisioning_mode.attr,
534 	&dev_attr_zeroing_mode.attr,
535 	&dev_attr_max_write_same_blocks.attr,
536 	&dev_attr_max_medium_access_timeouts.attr,
537 	NULL,
538 };
539 ATTRIBUTE_GROUPS(sd_disk);
540 
541 static struct class sd_disk_class = {
542 	.name		= "scsi_disk",
543 	.owner		= THIS_MODULE,
544 	.dev_release	= scsi_disk_release,
545 	.dev_groups	= sd_disk_groups,
546 };
547 
548 static const struct dev_pm_ops sd_pm_ops = {
549 	.suspend		= sd_suspend_system,
550 	.resume			= sd_resume,
551 	.poweroff		= sd_suspend_system,
552 	.restore		= sd_resume,
553 	.runtime_suspend	= sd_suspend_runtime,
554 	.runtime_resume		= sd_resume,
555 };
556 
557 static struct scsi_driver sd_template = {
558 	.gendrv = {
559 		.name		= "sd",
560 		.owner		= THIS_MODULE,
561 		.probe		= sd_probe,
562 		.remove		= sd_remove,
563 		.shutdown	= sd_shutdown,
564 		.pm		= &sd_pm_ops,
565 	},
566 	.rescan			= sd_rescan,
567 	.init_command		= sd_init_command,
568 	.uninit_command		= sd_uninit_command,
569 	.done			= sd_done,
570 	.eh_action		= sd_eh_action,
571 	.eh_reset		= sd_eh_reset,
572 };
573 
574 /*
575  * Dummy kobj_map->probe function.
576  * The default ->probe function will call modprobe, which is
577  * pointless as this module is already loaded.
578  */
579 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
580 {
581 	return NULL;
582 }
583 
584 /*
585  * Device no to disk mapping:
586  *
587  *       major         disc2     disc  p1
588  *   |............|.............|....|....| <- dev_t
589  *    31        20 19          8 7  4 3  0
590  *
591  * Inside a major, we have 16k disks, however mapped non-
592  * contiguously. The first 16 disks are for major0, the next
593  * ones with major1, ... Disk 256 is for major0 again, disk 272
594  * for major1, ...
595  * As we stay compatible with our numbering scheme, we can reuse
596  * the well-know SCSI majors 8, 65--71, 136--143.
597  */
598 static int sd_major(int major_idx)
599 {
600 	switch (major_idx) {
601 	case 0:
602 		return SCSI_DISK0_MAJOR;
603 	case 1 ... 7:
604 		return SCSI_DISK1_MAJOR + major_idx - 1;
605 	case 8 ... 15:
606 		return SCSI_DISK8_MAJOR + major_idx - 8;
607 	default:
608 		BUG();
609 		return 0;	/* shut up gcc */
610 	}
611 }
612 
613 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
614 {
615 	struct scsi_disk *sdkp = NULL;
616 
617 	mutex_lock(&sd_ref_mutex);
618 
619 	if (disk->private_data) {
620 		sdkp = scsi_disk(disk);
621 		if (scsi_device_get(sdkp->device) == 0)
622 			get_device(&sdkp->dev);
623 		else
624 			sdkp = NULL;
625 	}
626 	mutex_unlock(&sd_ref_mutex);
627 	return sdkp;
628 }
629 
630 static void scsi_disk_put(struct scsi_disk *sdkp)
631 {
632 	struct scsi_device *sdev = sdkp->device;
633 
634 	mutex_lock(&sd_ref_mutex);
635 	put_device(&sdkp->dev);
636 	scsi_device_put(sdev);
637 	mutex_unlock(&sd_ref_mutex);
638 }
639 
640 #ifdef CONFIG_BLK_SED_OPAL
641 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
642 		size_t len, bool send)
643 {
644 	struct scsi_device *sdev = data;
645 	u8 cdb[12] = { 0, };
646 	int ret;
647 
648 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
649 	cdb[1] = secp;
650 	put_unaligned_be16(spsp, &cdb[2]);
651 	put_unaligned_be32(len, &cdb[6]);
652 
653 	ret = scsi_execute_req(sdev, cdb,
654 			send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
655 			buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
656 	return ret <= 0 ? ret : -EIO;
657 }
658 #endif /* CONFIG_BLK_SED_OPAL */
659 
660 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
661 					   unsigned int dix, unsigned int dif)
662 {
663 	struct bio *bio = scmd->request->bio;
664 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
665 	unsigned int protect = 0;
666 
667 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
668 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
669 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
670 
671 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
672 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
673 	}
674 
675 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
676 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
677 
678 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
679 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
680 	}
681 
682 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
683 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
684 
685 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
686 			protect = 3 << 5;	/* Disable target PI checking */
687 		else
688 			protect = 1 << 5;	/* Enable target PI checking */
689 	}
690 
691 	scsi_set_prot_op(scmd, prot_op);
692 	scsi_set_prot_type(scmd, dif);
693 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
694 
695 	return protect;
696 }
697 
698 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
699 {
700 	struct request_queue *q = sdkp->disk->queue;
701 	unsigned int logical_block_size = sdkp->device->sector_size;
702 	unsigned int max_blocks = 0;
703 
704 	q->limits.discard_alignment =
705 		sdkp->unmap_alignment * logical_block_size;
706 	q->limits.discard_granularity =
707 		max(sdkp->physical_block_size,
708 		    sdkp->unmap_granularity * logical_block_size);
709 	sdkp->provisioning_mode = mode;
710 
711 	switch (mode) {
712 
713 	case SD_LBP_FULL:
714 	case SD_LBP_DISABLE:
715 		blk_queue_max_discard_sectors(q, 0);
716 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
717 		return;
718 
719 	case SD_LBP_UNMAP:
720 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
721 					  (u32)SD_MAX_WS16_BLOCKS);
722 		break;
723 
724 	case SD_LBP_WS16:
725 		if (sdkp->device->unmap_limit_for_ws)
726 			max_blocks = sdkp->max_unmap_blocks;
727 		else
728 			max_blocks = sdkp->max_ws_blocks;
729 
730 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
731 		break;
732 
733 	case SD_LBP_WS10:
734 		if (sdkp->device->unmap_limit_for_ws)
735 			max_blocks = sdkp->max_unmap_blocks;
736 		else
737 			max_blocks = sdkp->max_ws_blocks;
738 
739 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
740 		break;
741 
742 	case SD_LBP_ZERO:
743 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
744 					  (u32)SD_MAX_WS10_BLOCKS);
745 		break;
746 	}
747 
748 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
749 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
750 }
751 
752 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
753 {
754 	struct scsi_device *sdp = cmd->device;
755 	struct request *rq = cmd->request;
756 	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
757 	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
758 	unsigned int data_len = 24;
759 	char *buf;
760 
761 	rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
762 	if (!rq->special_vec.bv_page)
763 		return BLKPREP_DEFER;
764 	rq->special_vec.bv_offset = 0;
765 	rq->special_vec.bv_len = data_len;
766 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
767 
768 	cmd->cmd_len = 10;
769 	cmd->cmnd[0] = UNMAP;
770 	cmd->cmnd[8] = 24;
771 
772 	buf = page_address(rq->special_vec.bv_page);
773 	put_unaligned_be16(6 + 16, &buf[0]);
774 	put_unaligned_be16(16, &buf[2]);
775 	put_unaligned_be64(sector, &buf[8]);
776 	put_unaligned_be32(nr_sectors, &buf[16]);
777 
778 	cmd->allowed = SD_MAX_RETRIES;
779 	cmd->transfersize = data_len;
780 	rq->timeout = SD_TIMEOUT;
781 	scsi_req(rq)->resid_len = data_len;
782 
783 	return scsi_init_io(cmd);
784 }
785 
786 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
787 {
788 	struct scsi_device *sdp = cmd->device;
789 	struct request *rq = cmd->request;
790 	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
791 	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
792 	u32 data_len = sdp->sector_size;
793 
794 	rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
795 	if (!rq->special_vec.bv_page)
796 		return BLKPREP_DEFER;
797 	rq->special_vec.bv_offset = 0;
798 	rq->special_vec.bv_len = data_len;
799 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
800 
801 	cmd->cmd_len = 16;
802 	cmd->cmnd[0] = WRITE_SAME_16;
803 	if (unmap)
804 		cmd->cmnd[1] = 0x8; /* UNMAP */
805 	put_unaligned_be64(sector, &cmd->cmnd[2]);
806 	put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
807 
808 	cmd->allowed = SD_MAX_RETRIES;
809 	cmd->transfersize = data_len;
810 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
811 	scsi_req(rq)->resid_len = data_len;
812 
813 	return scsi_init_io(cmd);
814 }
815 
816 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
817 {
818 	struct scsi_device *sdp = cmd->device;
819 	struct request *rq = cmd->request;
820 	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
821 	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
822 	u32 data_len = sdp->sector_size;
823 
824 	rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
825 	if (!rq->special_vec.bv_page)
826 		return BLKPREP_DEFER;
827 	rq->special_vec.bv_offset = 0;
828 	rq->special_vec.bv_len = data_len;
829 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
830 
831 	cmd->cmd_len = 10;
832 	cmd->cmnd[0] = WRITE_SAME;
833 	if (unmap)
834 		cmd->cmnd[1] = 0x8; /* UNMAP */
835 	put_unaligned_be32(sector, &cmd->cmnd[2]);
836 	put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
837 
838 	cmd->allowed = SD_MAX_RETRIES;
839 	cmd->transfersize = data_len;
840 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
841 	scsi_req(rq)->resid_len = data_len;
842 
843 	return scsi_init_io(cmd);
844 }
845 
846 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
847 {
848 	struct request *rq = cmd->request;
849 	struct scsi_device *sdp = cmd->device;
850 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
851 	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
852 	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
853 
854 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
855 		switch (sdkp->zeroing_mode) {
856 		case SD_ZERO_WS16_UNMAP:
857 			return sd_setup_write_same16_cmnd(cmd, true);
858 		case SD_ZERO_WS10_UNMAP:
859 			return sd_setup_write_same10_cmnd(cmd, true);
860 		}
861 	}
862 
863 	if (sdp->no_write_same)
864 		return BLKPREP_INVALID;
865 
866 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
867 		return sd_setup_write_same16_cmnd(cmd, false);
868 
869 	return sd_setup_write_same10_cmnd(cmd, false);
870 }
871 
872 static void sd_config_write_same(struct scsi_disk *sdkp)
873 {
874 	struct request_queue *q = sdkp->disk->queue;
875 	unsigned int logical_block_size = sdkp->device->sector_size;
876 
877 	if (sdkp->device->no_write_same) {
878 		sdkp->max_ws_blocks = 0;
879 		goto out;
880 	}
881 
882 	/* Some devices can not handle block counts above 0xffff despite
883 	 * supporting WRITE SAME(16). Consequently we default to 64k
884 	 * blocks per I/O unless the device explicitly advertises a
885 	 * bigger limit.
886 	 */
887 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
888 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
889 						   (u32)SD_MAX_WS16_BLOCKS);
890 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
891 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
892 						   (u32)SD_MAX_WS10_BLOCKS);
893 	else {
894 		sdkp->device->no_write_same = 1;
895 		sdkp->max_ws_blocks = 0;
896 	}
897 
898 	if (sdkp->lbprz && sdkp->lbpws)
899 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
900 	else if (sdkp->lbprz && sdkp->lbpws10)
901 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
902 	else if (sdkp->max_ws_blocks)
903 		sdkp->zeroing_mode = SD_ZERO_WS;
904 	else
905 		sdkp->zeroing_mode = SD_ZERO_WRITE;
906 
907 	if (sdkp->max_ws_blocks &&
908 	    sdkp->physical_block_size > logical_block_size) {
909 		/*
910 		 * Reporting a maximum number of blocks that is not aligned
911 		 * on the device physical size would cause a large write same
912 		 * request to be split into physically unaligned chunks by
913 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
914 		 * even if the caller of these functions took care to align the
915 		 * large request. So make sure the maximum reported is aligned
916 		 * to the device physical block size. This is only an optional
917 		 * optimization for regular disks, but this is mandatory to
918 		 * avoid failure of large write same requests directed at
919 		 * sequential write required zones of host-managed ZBC disks.
920 		 */
921 		sdkp->max_ws_blocks =
922 			round_down(sdkp->max_ws_blocks,
923 				   bytes_to_logical(sdkp->device,
924 						    sdkp->physical_block_size));
925 	}
926 
927 out:
928 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
929 					 (logical_block_size >> 9));
930 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
931 					 (logical_block_size >> 9));
932 }
933 
934 /**
935  * sd_setup_write_same_cmnd - write the same data to multiple blocks
936  * @cmd: command to prepare
937  *
938  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
939  * the preference indicated by the target device.
940  **/
941 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
942 {
943 	struct request *rq = cmd->request;
944 	struct scsi_device *sdp = cmd->device;
945 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
946 	struct bio *bio = rq->bio;
947 	sector_t sector = blk_rq_pos(rq);
948 	unsigned int nr_sectors = blk_rq_sectors(rq);
949 	unsigned int nr_bytes = blk_rq_bytes(rq);
950 	int ret;
951 
952 	if (sdkp->device->no_write_same)
953 		return BLKPREP_INVALID;
954 
955 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
956 
957 	sector >>= ilog2(sdp->sector_size) - 9;
958 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
959 
960 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
961 
962 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
963 		cmd->cmd_len = 16;
964 		cmd->cmnd[0] = WRITE_SAME_16;
965 		put_unaligned_be64(sector, &cmd->cmnd[2]);
966 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
967 	} else {
968 		cmd->cmd_len = 10;
969 		cmd->cmnd[0] = WRITE_SAME;
970 		put_unaligned_be32(sector, &cmd->cmnd[2]);
971 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
972 	}
973 
974 	cmd->transfersize = sdp->sector_size;
975 	cmd->allowed = SD_MAX_RETRIES;
976 
977 	/*
978 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
979 	 * different from the amount of data actually written to the target.
980 	 *
981 	 * We set up __data_len to the amount of data transferred via the
982 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
983 	 * to transfer a single sector of data first, but then reset it to
984 	 * the amount of data to be written right after so that the I/O path
985 	 * knows how much to actually write.
986 	 */
987 	rq->__data_len = sdp->sector_size;
988 	ret = scsi_init_io(cmd);
989 	rq->__data_len = nr_bytes;
990 
991 	return ret;
992 }
993 
994 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
995 {
996 	struct request *rq = cmd->request;
997 
998 	/* flush requests don't perform I/O, zero the S/G table */
999 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1000 
1001 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1002 	cmd->cmd_len = 10;
1003 	cmd->transfersize = 0;
1004 	cmd->allowed = SD_MAX_RETRIES;
1005 
1006 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1007 	return BLKPREP_OK;
1008 }
1009 
1010 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1011 {
1012 	struct request *rq = SCpnt->request;
1013 	struct scsi_device *sdp = SCpnt->device;
1014 	struct gendisk *disk = rq->rq_disk;
1015 	struct scsi_disk *sdkp = scsi_disk(disk);
1016 	sector_t block = blk_rq_pos(rq);
1017 	sector_t threshold;
1018 	unsigned int this_count = blk_rq_sectors(rq);
1019 	unsigned int dif, dix;
1020 	int ret;
1021 	unsigned char protect;
1022 
1023 	ret = scsi_init_io(SCpnt);
1024 	if (ret != BLKPREP_OK)
1025 		return ret;
1026 	WARN_ON_ONCE(SCpnt != rq->special);
1027 
1028 	/* from here on until we're complete, any goto out
1029 	 * is used for a killable error condition */
1030 	ret = BLKPREP_KILL;
1031 
1032 	SCSI_LOG_HLQUEUE(1,
1033 		scmd_printk(KERN_INFO, SCpnt,
1034 			"%s: block=%llu, count=%d\n",
1035 			__func__, (unsigned long long)block, this_count));
1036 
1037 	if (!sdp || !scsi_device_online(sdp) ||
1038 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
1039 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1040 						"Finishing %u sectors\n",
1041 						blk_rq_sectors(rq)));
1042 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1043 						"Retry with 0x%p\n", SCpnt));
1044 		goto out;
1045 	}
1046 
1047 	if (sdp->changed) {
1048 		/*
1049 		 * quietly refuse to do anything to a changed disc until
1050 		 * the changed bit has been reset
1051 		 */
1052 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1053 		goto out;
1054 	}
1055 
1056 	/*
1057 	 * Some SD card readers can't handle multi-sector accesses which touch
1058 	 * the last one or two hardware sectors.  Split accesses as needed.
1059 	 */
1060 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1061 		(sdp->sector_size / 512);
1062 
1063 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1064 		if (block < threshold) {
1065 			/* Access up to the threshold but not beyond */
1066 			this_count = threshold - block;
1067 		} else {
1068 			/* Access only a single hardware sector */
1069 			this_count = sdp->sector_size / 512;
1070 		}
1071 	}
1072 
1073 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1074 					(unsigned long long)block));
1075 
1076 	/*
1077 	 * If we have a 1K hardware sectorsize, prevent access to single
1078 	 * 512 byte sectors.  In theory we could handle this - in fact
1079 	 * the scsi cdrom driver must be able to handle this because
1080 	 * we typically use 1K blocksizes, and cdroms typically have
1081 	 * 2K hardware sectorsizes.  Of course, things are simpler
1082 	 * with the cdrom, since it is read-only.  For performance
1083 	 * reasons, the filesystems should be able to handle this
1084 	 * and not force the scsi disk driver to use bounce buffers
1085 	 * for this.
1086 	 */
1087 	if (sdp->sector_size == 1024) {
1088 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1089 			scmd_printk(KERN_ERR, SCpnt,
1090 				    "Bad block number requested\n");
1091 			goto out;
1092 		} else {
1093 			block = block >> 1;
1094 			this_count = this_count >> 1;
1095 		}
1096 	}
1097 	if (sdp->sector_size == 2048) {
1098 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1099 			scmd_printk(KERN_ERR, SCpnt,
1100 				    "Bad block number requested\n");
1101 			goto out;
1102 		} else {
1103 			block = block >> 2;
1104 			this_count = this_count >> 2;
1105 		}
1106 	}
1107 	if (sdp->sector_size == 4096) {
1108 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1109 			scmd_printk(KERN_ERR, SCpnt,
1110 				    "Bad block number requested\n");
1111 			goto out;
1112 		} else {
1113 			block = block >> 3;
1114 			this_count = this_count >> 3;
1115 		}
1116 	}
1117 	if (rq_data_dir(rq) == WRITE) {
1118 		SCpnt->cmnd[0] = WRITE_6;
1119 
1120 		if (blk_integrity_rq(rq))
1121 			sd_dif_prepare(SCpnt);
1122 
1123 	} else if (rq_data_dir(rq) == READ) {
1124 		SCpnt->cmnd[0] = READ_6;
1125 	} else {
1126 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1127 		goto out;
1128 	}
1129 
1130 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1131 					"%s %d/%u 512 byte blocks.\n",
1132 					(rq_data_dir(rq) == WRITE) ?
1133 					"writing" : "reading", this_count,
1134 					blk_rq_sectors(rq)));
1135 
1136 	dix = scsi_prot_sg_count(SCpnt);
1137 	dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1138 
1139 	if (dif || dix)
1140 		protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1141 	else
1142 		protect = 0;
1143 
1144 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1145 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1146 
1147 		if (unlikely(SCpnt->cmnd == NULL)) {
1148 			ret = BLKPREP_DEFER;
1149 			goto out;
1150 		}
1151 
1152 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1153 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1154 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1155 		SCpnt->cmnd[7] = 0x18;
1156 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1157 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1158 
1159 		/* LBA */
1160 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1161 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1162 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1163 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1164 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1165 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1166 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1167 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1168 
1169 		/* Expected Indirect LBA */
1170 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1171 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1172 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1173 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1174 
1175 		/* Transfer length */
1176 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1177 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1178 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1179 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1180 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1181 		SCpnt->cmnd[0] += READ_16 - READ_6;
1182 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1183 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1184 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1185 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1186 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1187 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1188 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1189 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1190 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1191 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1192 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1193 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1194 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1195 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1196 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1197 		   scsi_device_protection(SCpnt->device) ||
1198 		   SCpnt->device->use_10_for_rw) {
1199 		SCpnt->cmnd[0] += READ_10 - READ_6;
1200 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1201 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1202 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1203 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1204 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1205 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1206 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1207 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1208 	} else {
1209 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1210 			/*
1211 			 * This happens only if this drive failed
1212 			 * 10byte rw command with ILLEGAL_REQUEST
1213 			 * during operation and thus turned off
1214 			 * use_10_for_rw.
1215 			 */
1216 			scmd_printk(KERN_ERR, SCpnt,
1217 				    "FUA write on READ/WRITE(6) drive\n");
1218 			goto out;
1219 		}
1220 
1221 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1222 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1223 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1224 		SCpnt->cmnd[4] = (unsigned char) this_count;
1225 		SCpnt->cmnd[5] = 0;
1226 	}
1227 	SCpnt->sdb.length = this_count * sdp->sector_size;
1228 
1229 	/*
1230 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1231 	 * host adapter, it's safe to assume that we can at least transfer
1232 	 * this many bytes between each connect / disconnect.
1233 	 */
1234 	SCpnt->transfersize = sdp->sector_size;
1235 	SCpnt->underflow = this_count << 9;
1236 	SCpnt->allowed = SD_MAX_RETRIES;
1237 
1238 	/*
1239 	 * This indicates that the command is ready from our end to be
1240 	 * queued.
1241 	 */
1242 	ret = BLKPREP_OK;
1243  out:
1244 	return ret;
1245 }
1246 
1247 static int sd_init_command(struct scsi_cmnd *cmd)
1248 {
1249 	struct request *rq = cmd->request;
1250 
1251 	switch (req_op(rq)) {
1252 	case REQ_OP_DISCARD:
1253 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1254 		case SD_LBP_UNMAP:
1255 			return sd_setup_unmap_cmnd(cmd);
1256 		case SD_LBP_WS16:
1257 			return sd_setup_write_same16_cmnd(cmd, true);
1258 		case SD_LBP_WS10:
1259 			return sd_setup_write_same10_cmnd(cmd, true);
1260 		case SD_LBP_ZERO:
1261 			return sd_setup_write_same10_cmnd(cmd, false);
1262 		default:
1263 			return BLKPREP_INVALID;
1264 		}
1265 	case REQ_OP_WRITE_ZEROES:
1266 		return sd_setup_write_zeroes_cmnd(cmd);
1267 	case REQ_OP_WRITE_SAME:
1268 		return sd_setup_write_same_cmnd(cmd);
1269 	case REQ_OP_FLUSH:
1270 		return sd_setup_flush_cmnd(cmd);
1271 	case REQ_OP_READ:
1272 	case REQ_OP_WRITE:
1273 		return sd_setup_read_write_cmnd(cmd);
1274 	case REQ_OP_ZONE_REPORT:
1275 		return sd_zbc_setup_report_cmnd(cmd);
1276 	case REQ_OP_ZONE_RESET:
1277 		return sd_zbc_setup_reset_cmnd(cmd);
1278 	default:
1279 		BUG();
1280 	}
1281 }
1282 
1283 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1284 {
1285 	struct request *rq = SCpnt->request;
1286 	u8 *cmnd;
1287 
1288 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1289 		__free_page(rq->special_vec.bv_page);
1290 
1291 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1292 		cmnd = SCpnt->cmnd;
1293 		SCpnt->cmnd = NULL;
1294 		SCpnt->cmd_len = 0;
1295 		mempool_free(cmnd, sd_cdb_pool);
1296 	}
1297 }
1298 
1299 /**
1300  *	sd_open - open a scsi disk device
1301  *	@bdev: Block device of the scsi disk to open
1302  *	@mode: FMODE_* mask
1303  *
1304  *	Returns 0 if successful. Returns a negated errno value in case
1305  *	of error.
1306  *
1307  *	Note: This can be called from a user context (e.g. fsck(1) )
1308  *	or from within the kernel (e.g. as a result of a mount(1) ).
1309  *	In the latter case @inode and @filp carry an abridged amount
1310  *	of information as noted above.
1311  *
1312  *	Locking: called with bdev->bd_mutex held.
1313  **/
1314 static int sd_open(struct block_device *bdev, fmode_t mode)
1315 {
1316 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1317 	struct scsi_device *sdev;
1318 	int retval;
1319 
1320 	if (!sdkp)
1321 		return -ENXIO;
1322 
1323 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1324 
1325 	sdev = sdkp->device;
1326 
1327 	/*
1328 	 * If the device is in error recovery, wait until it is done.
1329 	 * If the device is offline, then disallow any access to it.
1330 	 */
1331 	retval = -ENXIO;
1332 	if (!scsi_block_when_processing_errors(sdev))
1333 		goto error_out;
1334 
1335 	if (sdev->removable || sdkp->write_prot)
1336 		check_disk_change(bdev);
1337 
1338 	/*
1339 	 * If the drive is empty, just let the open fail.
1340 	 */
1341 	retval = -ENOMEDIUM;
1342 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1343 		goto error_out;
1344 
1345 	/*
1346 	 * If the device has the write protect tab set, have the open fail
1347 	 * if the user expects to be able to write to the thing.
1348 	 */
1349 	retval = -EROFS;
1350 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1351 		goto error_out;
1352 
1353 	/*
1354 	 * It is possible that the disk changing stuff resulted in
1355 	 * the device being taken offline.  If this is the case,
1356 	 * report this to the user, and don't pretend that the
1357 	 * open actually succeeded.
1358 	 */
1359 	retval = -ENXIO;
1360 	if (!scsi_device_online(sdev))
1361 		goto error_out;
1362 
1363 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1364 		if (scsi_block_when_processing_errors(sdev))
1365 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1366 	}
1367 
1368 	return 0;
1369 
1370 error_out:
1371 	scsi_disk_put(sdkp);
1372 	return retval;
1373 }
1374 
1375 /**
1376  *	sd_release - invoked when the (last) close(2) is called on this
1377  *	scsi disk.
1378  *	@disk: disk to release
1379  *	@mode: FMODE_* mask
1380  *
1381  *	Returns 0.
1382  *
1383  *	Note: may block (uninterruptible) if error recovery is underway
1384  *	on this disk.
1385  *
1386  *	Locking: called with bdev->bd_mutex held.
1387  **/
1388 static void sd_release(struct gendisk *disk, fmode_t mode)
1389 {
1390 	struct scsi_disk *sdkp = scsi_disk(disk);
1391 	struct scsi_device *sdev = sdkp->device;
1392 
1393 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1394 
1395 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1396 		if (scsi_block_when_processing_errors(sdev))
1397 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1398 	}
1399 
1400 	/*
1401 	 * XXX and what if there are packets in flight and this close()
1402 	 * XXX is followed by a "rmmod sd_mod"?
1403 	 */
1404 
1405 	scsi_disk_put(sdkp);
1406 }
1407 
1408 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1409 {
1410 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1411 	struct scsi_device *sdp = sdkp->device;
1412 	struct Scsi_Host *host = sdp->host;
1413 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1414 	int diskinfo[4];
1415 
1416 	/* default to most commonly used values */
1417 	diskinfo[0] = 0x40;	/* 1 << 6 */
1418 	diskinfo[1] = 0x20;	/* 1 << 5 */
1419 	diskinfo[2] = capacity >> 11;
1420 
1421 	/* override with calculated, extended default, or driver values */
1422 	if (host->hostt->bios_param)
1423 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1424 	else
1425 		scsicam_bios_param(bdev, capacity, diskinfo);
1426 
1427 	geo->heads = diskinfo[0];
1428 	geo->sectors = diskinfo[1];
1429 	geo->cylinders = diskinfo[2];
1430 	return 0;
1431 }
1432 
1433 /**
1434  *	sd_ioctl - process an ioctl
1435  *	@bdev: target block device
1436  *	@mode: FMODE_* mask
1437  *	@cmd: ioctl command number
1438  *	@arg: this is third argument given to ioctl(2) system call.
1439  *	Often contains a pointer.
1440  *
1441  *	Returns 0 if successful (some ioctls return positive numbers on
1442  *	success as well). Returns a negated errno value in case of error.
1443  *
1444  *	Note: most ioctls are forward onto the block subsystem or further
1445  *	down in the scsi subsystem.
1446  **/
1447 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1448 		    unsigned int cmd, unsigned long arg)
1449 {
1450 	struct gendisk *disk = bdev->bd_disk;
1451 	struct scsi_disk *sdkp = scsi_disk(disk);
1452 	struct scsi_device *sdp = sdkp->device;
1453 	void __user *p = (void __user *)arg;
1454 	int error;
1455 
1456 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1457 				    "cmd=0x%x\n", disk->disk_name, cmd));
1458 
1459 	error = scsi_verify_blk_ioctl(bdev, cmd);
1460 	if (error < 0)
1461 		return error;
1462 
1463 	/*
1464 	 * If we are in the middle of error recovery, don't let anyone
1465 	 * else try and use this device.  Also, if error recovery fails, it
1466 	 * may try and take the device offline, in which case all further
1467 	 * access to the device is prohibited.
1468 	 */
1469 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1470 			(mode & FMODE_NDELAY) != 0);
1471 	if (error)
1472 		goto out;
1473 
1474 	if (is_sed_ioctl(cmd))
1475 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1476 
1477 	/*
1478 	 * Send SCSI addressing ioctls directly to mid level, send other
1479 	 * ioctls to block level and then onto mid level if they can't be
1480 	 * resolved.
1481 	 */
1482 	switch (cmd) {
1483 		case SCSI_IOCTL_GET_IDLUN:
1484 		case SCSI_IOCTL_GET_BUS_NUMBER:
1485 			error = scsi_ioctl(sdp, cmd, p);
1486 			break;
1487 		default:
1488 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1489 			if (error != -ENOTTY)
1490 				break;
1491 			error = scsi_ioctl(sdp, cmd, p);
1492 			break;
1493 	}
1494 out:
1495 	return error;
1496 }
1497 
1498 static void set_media_not_present(struct scsi_disk *sdkp)
1499 {
1500 	if (sdkp->media_present)
1501 		sdkp->device->changed = 1;
1502 
1503 	if (sdkp->device->removable) {
1504 		sdkp->media_present = 0;
1505 		sdkp->capacity = 0;
1506 	}
1507 }
1508 
1509 static int media_not_present(struct scsi_disk *sdkp,
1510 			     struct scsi_sense_hdr *sshdr)
1511 {
1512 	if (!scsi_sense_valid(sshdr))
1513 		return 0;
1514 
1515 	/* not invoked for commands that could return deferred errors */
1516 	switch (sshdr->sense_key) {
1517 	case UNIT_ATTENTION:
1518 	case NOT_READY:
1519 		/* medium not present */
1520 		if (sshdr->asc == 0x3A) {
1521 			set_media_not_present(sdkp);
1522 			return 1;
1523 		}
1524 	}
1525 	return 0;
1526 }
1527 
1528 /**
1529  *	sd_check_events - check media events
1530  *	@disk: kernel device descriptor
1531  *	@clearing: disk events currently being cleared
1532  *
1533  *	Returns mask of DISK_EVENT_*.
1534  *
1535  *	Note: this function is invoked from the block subsystem.
1536  **/
1537 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1538 {
1539 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1540 	struct scsi_device *sdp;
1541 	int retval;
1542 
1543 	if (!sdkp)
1544 		return 0;
1545 
1546 	sdp = sdkp->device;
1547 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1548 
1549 	/*
1550 	 * If the device is offline, don't send any commands - just pretend as
1551 	 * if the command failed.  If the device ever comes back online, we
1552 	 * can deal with it then.  It is only because of unrecoverable errors
1553 	 * that we would ever take a device offline in the first place.
1554 	 */
1555 	if (!scsi_device_online(sdp)) {
1556 		set_media_not_present(sdkp);
1557 		goto out;
1558 	}
1559 
1560 	/*
1561 	 * Using TEST_UNIT_READY enables differentiation between drive with
1562 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1563 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1564 	 *
1565 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1566 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1567 	 * sd_revalidate() is called.
1568 	 */
1569 	if (scsi_block_when_processing_errors(sdp)) {
1570 		struct scsi_sense_hdr sshdr = { 0, };
1571 
1572 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1573 					      &sshdr);
1574 
1575 		/* failed to execute TUR, assume media not present */
1576 		if (host_byte(retval)) {
1577 			set_media_not_present(sdkp);
1578 			goto out;
1579 		}
1580 
1581 		if (media_not_present(sdkp, &sshdr))
1582 			goto out;
1583 	}
1584 
1585 	/*
1586 	 * For removable scsi disk we have to recognise the presence
1587 	 * of a disk in the drive.
1588 	 */
1589 	if (!sdkp->media_present)
1590 		sdp->changed = 1;
1591 	sdkp->media_present = 1;
1592 out:
1593 	/*
1594 	 * sdp->changed is set under the following conditions:
1595 	 *
1596 	 *	Medium present state has changed in either direction.
1597 	 *	Device has indicated UNIT_ATTENTION.
1598 	 */
1599 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1600 	sdp->changed = 0;
1601 	scsi_disk_put(sdkp);
1602 	return retval;
1603 }
1604 
1605 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1606 {
1607 	int retries, res;
1608 	struct scsi_device *sdp = sdkp->device;
1609 	const int timeout = sdp->request_queue->rq_timeout
1610 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1611 	struct scsi_sense_hdr my_sshdr;
1612 
1613 	if (!scsi_device_online(sdp))
1614 		return -ENODEV;
1615 
1616 	/* caller might not be interested in sense, but we need it */
1617 	if (!sshdr)
1618 		sshdr = &my_sshdr;
1619 
1620 	for (retries = 3; retries > 0; --retries) {
1621 		unsigned char cmd[10] = { 0 };
1622 
1623 		cmd[0] = SYNCHRONIZE_CACHE;
1624 		/*
1625 		 * Leave the rest of the command zero to indicate
1626 		 * flush everything.
1627 		 */
1628 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1629 				timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1630 		if (res == 0)
1631 			break;
1632 	}
1633 
1634 	if (res) {
1635 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1636 
1637 		if (driver_byte(res) & DRIVER_SENSE)
1638 			sd_print_sense_hdr(sdkp, sshdr);
1639 
1640 		/* we need to evaluate the error return  */
1641 		if (scsi_sense_valid(sshdr) &&
1642 			(sshdr->asc == 0x3a ||	/* medium not present */
1643 			 sshdr->asc == 0x20))	/* invalid command */
1644 				/* this is no error here */
1645 				return 0;
1646 
1647 		switch (host_byte(res)) {
1648 		/* ignore errors due to racing a disconnection */
1649 		case DID_BAD_TARGET:
1650 		case DID_NO_CONNECT:
1651 			return 0;
1652 		/* signal the upper layer it might try again */
1653 		case DID_BUS_BUSY:
1654 		case DID_IMM_RETRY:
1655 		case DID_REQUEUE:
1656 		case DID_SOFT_ERROR:
1657 			return -EBUSY;
1658 		default:
1659 			return -EIO;
1660 		}
1661 	}
1662 	return 0;
1663 }
1664 
1665 static void sd_rescan(struct device *dev)
1666 {
1667 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1668 
1669 	revalidate_disk(sdkp->disk);
1670 }
1671 
1672 
1673 #ifdef CONFIG_COMPAT
1674 /*
1675  * This gets directly called from VFS. When the ioctl
1676  * is not recognized we go back to the other translation paths.
1677  */
1678 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1679 			   unsigned int cmd, unsigned long arg)
1680 {
1681 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1682 	int error;
1683 
1684 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1685 			(mode & FMODE_NDELAY) != 0);
1686 	if (error)
1687 		return error;
1688 
1689 	/*
1690 	 * Let the static ioctl translation table take care of it.
1691 	 */
1692 	if (!sdev->host->hostt->compat_ioctl)
1693 		return -ENOIOCTLCMD;
1694 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1695 }
1696 #endif
1697 
1698 static char sd_pr_type(enum pr_type type)
1699 {
1700 	switch (type) {
1701 	case PR_WRITE_EXCLUSIVE:
1702 		return 0x01;
1703 	case PR_EXCLUSIVE_ACCESS:
1704 		return 0x03;
1705 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1706 		return 0x05;
1707 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1708 		return 0x06;
1709 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1710 		return 0x07;
1711 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1712 		return 0x08;
1713 	default:
1714 		return 0;
1715 	}
1716 };
1717 
1718 static int sd_pr_command(struct block_device *bdev, u8 sa,
1719 		u64 key, u64 sa_key, u8 type, u8 flags)
1720 {
1721 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1722 	struct scsi_sense_hdr sshdr;
1723 	int result;
1724 	u8 cmd[16] = { 0, };
1725 	u8 data[24] = { 0, };
1726 
1727 	cmd[0] = PERSISTENT_RESERVE_OUT;
1728 	cmd[1] = sa;
1729 	cmd[2] = type;
1730 	put_unaligned_be32(sizeof(data), &cmd[5]);
1731 
1732 	put_unaligned_be64(key, &data[0]);
1733 	put_unaligned_be64(sa_key, &data[8]);
1734 	data[20] = flags;
1735 
1736 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1737 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1738 
1739 	if ((driver_byte(result) & DRIVER_SENSE) &&
1740 	    (scsi_sense_valid(&sshdr))) {
1741 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1742 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1743 	}
1744 
1745 	return result;
1746 }
1747 
1748 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1749 		u32 flags)
1750 {
1751 	if (flags & ~PR_FL_IGNORE_KEY)
1752 		return -EOPNOTSUPP;
1753 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1754 			old_key, new_key, 0,
1755 			(1 << 0) /* APTPL */);
1756 }
1757 
1758 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1759 		u32 flags)
1760 {
1761 	if (flags)
1762 		return -EOPNOTSUPP;
1763 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1764 }
1765 
1766 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1767 {
1768 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1769 }
1770 
1771 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1772 		enum pr_type type, bool abort)
1773 {
1774 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1775 			     sd_pr_type(type), 0);
1776 }
1777 
1778 static int sd_pr_clear(struct block_device *bdev, u64 key)
1779 {
1780 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1781 }
1782 
1783 static const struct pr_ops sd_pr_ops = {
1784 	.pr_register	= sd_pr_register,
1785 	.pr_reserve	= sd_pr_reserve,
1786 	.pr_release	= sd_pr_release,
1787 	.pr_preempt	= sd_pr_preempt,
1788 	.pr_clear	= sd_pr_clear,
1789 };
1790 
1791 static const struct block_device_operations sd_fops = {
1792 	.owner			= THIS_MODULE,
1793 	.open			= sd_open,
1794 	.release		= sd_release,
1795 	.ioctl			= sd_ioctl,
1796 	.getgeo			= sd_getgeo,
1797 #ifdef CONFIG_COMPAT
1798 	.compat_ioctl		= sd_compat_ioctl,
1799 #endif
1800 	.check_events		= sd_check_events,
1801 	.revalidate_disk	= sd_revalidate_disk,
1802 	.unlock_native_capacity	= sd_unlock_native_capacity,
1803 	.pr_ops			= &sd_pr_ops,
1804 };
1805 
1806 /**
1807  *	sd_eh_reset - reset error handling callback
1808  *	@scmd:		sd-issued command that has failed
1809  *
1810  *	This function is called by the SCSI midlayer before starting
1811  *	SCSI EH. When counting medium access failures we have to be
1812  *	careful to register it only only once per device and SCSI EH run;
1813  *	there might be several timed out commands which will cause the
1814  *	'max_medium_access_timeouts' counter to trigger after the first
1815  *	SCSI EH run already and set the device to offline.
1816  *	So this function resets the internal counter before starting SCSI EH.
1817  **/
1818 static void sd_eh_reset(struct scsi_cmnd *scmd)
1819 {
1820 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1821 
1822 	/* New SCSI EH run, reset gate variable */
1823 	sdkp->ignore_medium_access_errors = false;
1824 }
1825 
1826 /**
1827  *	sd_eh_action - error handling callback
1828  *	@scmd:		sd-issued command that has failed
1829  *	@eh_disp:	The recovery disposition suggested by the midlayer
1830  *
1831  *	This function is called by the SCSI midlayer upon completion of an
1832  *	error test command (currently TEST UNIT READY). The result of sending
1833  *	the eh command is passed in eh_disp.  We're looking for devices that
1834  *	fail medium access commands but are OK with non access commands like
1835  *	test unit ready (so wrongly see the device as having a successful
1836  *	recovery)
1837  **/
1838 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1839 {
1840 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1841 	struct scsi_device *sdev = scmd->device;
1842 
1843 	if (!scsi_device_online(sdev) ||
1844 	    !scsi_medium_access_command(scmd) ||
1845 	    host_byte(scmd->result) != DID_TIME_OUT ||
1846 	    eh_disp != SUCCESS)
1847 		return eh_disp;
1848 
1849 	/*
1850 	 * The device has timed out executing a medium access command.
1851 	 * However, the TEST UNIT READY command sent during error
1852 	 * handling completed successfully. Either the device is in the
1853 	 * process of recovering or has it suffered an internal failure
1854 	 * that prevents access to the storage medium.
1855 	 */
1856 	if (!sdkp->ignore_medium_access_errors) {
1857 		sdkp->medium_access_timed_out++;
1858 		sdkp->ignore_medium_access_errors = true;
1859 	}
1860 
1861 	/*
1862 	 * If the device keeps failing read/write commands but TEST UNIT
1863 	 * READY always completes successfully we assume that medium
1864 	 * access is no longer possible and take the device offline.
1865 	 */
1866 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1867 		scmd_printk(KERN_ERR, scmd,
1868 			    "Medium access timeout failure. Offlining disk!\n");
1869 		mutex_lock(&sdev->state_mutex);
1870 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1871 		mutex_unlock(&sdev->state_mutex);
1872 
1873 		return SUCCESS;
1874 	}
1875 
1876 	return eh_disp;
1877 }
1878 
1879 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1880 {
1881 	struct request *req = scmd->request;
1882 	struct scsi_device *sdev = scmd->device;
1883 	unsigned int transferred, good_bytes;
1884 	u64 start_lba, end_lba, bad_lba;
1885 
1886 	/*
1887 	 * Some commands have a payload smaller than the device logical
1888 	 * block size (e.g. INQUIRY on a 4K disk).
1889 	 */
1890 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1891 		return 0;
1892 
1893 	/* Check if we have a 'bad_lba' information */
1894 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1895 				     SCSI_SENSE_BUFFERSIZE,
1896 				     &bad_lba))
1897 		return 0;
1898 
1899 	/*
1900 	 * If the bad lba was reported incorrectly, we have no idea where
1901 	 * the error is.
1902 	 */
1903 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1904 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1905 	if (bad_lba < start_lba || bad_lba >= end_lba)
1906 		return 0;
1907 
1908 	/*
1909 	 * resid is optional but mostly filled in.  When it's unused,
1910 	 * its value is zero, so we assume the whole buffer transferred
1911 	 */
1912 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1913 
1914 	/* This computation should always be done in terms of the
1915 	 * resolution of the device's medium.
1916 	 */
1917 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1918 
1919 	return min(good_bytes, transferred);
1920 }
1921 
1922 /**
1923  *	sd_done - bottom half handler: called when the lower level
1924  *	driver has completed (successfully or otherwise) a scsi command.
1925  *	@SCpnt: mid-level's per command structure.
1926  *
1927  *	Note: potentially run from within an ISR. Must not block.
1928  **/
1929 static int sd_done(struct scsi_cmnd *SCpnt)
1930 {
1931 	int result = SCpnt->result;
1932 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1933 	unsigned int sector_size = SCpnt->device->sector_size;
1934 	unsigned int resid;
1935 	struct scsi_sense_hdr sshdr;
1936 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1937 	struct request *req = SCpnt->request;
1938 	int sense_valid = 0;
1939 	int sense_deferred = 0;
1940 
1941 	switch (req_op(req)) {
1942 	case REQ_OP_DISCARD:
1943 	case REQ_OP_WRITE_ZEROES:
1944 	case REQ_OP_WRITE_SAME:
1945 	case REQ_OP_ZONE_RESET:
1946 		if (!result) {
1947 			good_bytes = blk_rq_bytes(req);
1948 			scsi_set_resid(SCpnt, 0);
1949 		} else {
1950 			good_bytes = 0;
1951 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1952 		}
1953 		break;
1954 	case REQ_OP_ZONE_REPORT:
1955 		if (!result) {
1956 			good_bytes = scsi_bufflen(SCpnt)
1957 				- scsi_get_resid(SCpnt);
1958 			scsi_set_resid(SCpnt, 0);
1959 		} else {
1960 			good_bytes = 0;
1961 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1962 		}
1963 		break;
1964 	default:
1965 		/*
1966 		 * In case of bogus fw or device, we could end up having
1967 		 * an unaligned partial completion. Check this here and force
1968 		 * alignment.
1969 		 */
1970 		resid = scsi_get_resid(SCpnt);
1971 		if (resid & (sector_size - 1)) {
1972 			sd_printk(KERN_INFO, sdkp,
1973 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1974 				resid, sector_size);
1975 			resid = min(scsi_bufflen(SCpnt),
1976 				    round_up(resid, sector_size));
1977 			scsi_set_resid(SCpnt, resid);
1978 		}
1979 	}
1980 
1981 	if (result) {
1982 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1983 		if (sense_valid)
1984 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1985 	}
1986 	sdkp->medium_access_timed_out = 0;
1987 
1988 	if (driver_byte(result) != DRIVER_SENSE &&
1989 	    (!sense_valid || sense_deferred))
1990 		goto out;
1991 
1992 	switch (sshdr.sense_key) {
1993 	case HARDWARE_ERROR:
1994 	case MEDIUM_ERROR:
1995 		good_bytes = sd_completed_bytes(SCpnt);
1996 		break;
1997 	case RECOVERED_ERROR:
1998 		good_bytes = scsi_bufflen(SCpnt);
1999 		break;
2000 	case NO_SENSE:
2001 		/* This indicates a false check condition, so ignore it.  An
2002 		 * unknown amount of data was transferred so treat it as an
2003 		 * error.
2004 		 */
2005 		SCpnt->result = 0;
2006 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2007 		break;
2008 	case ABORTED_COMMAND:
2009 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2010 			good_bytes = sd_completed_bytes(SCpnt);
2011 		break;
2012 	case ILLEGAL_REQUEST:
2013 		switch (sshdr.asc) {
2014 		case 0x10:	/* DIX: Host detected corruption */
2015 			good_bytes = sd_completed_bytes(SCpnt);
2016 			break;
2017 		case 0x20:	/* INVALID COMMAND OPCODE */
2018 		case 0x24:	/* INVALID FIELD IN CDB */
2019 			switch (SCpnt->cmnd[0]) {
2020 			case UNMAP:
2021 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2022 				break;
2023 			case WRITE_SAME_16:
2024 			case WRITE_SAME:
2025 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2026 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2027 				} else {
2028 					sdkp->device->no_write_same = 1;
2029 					sd_config_write_same(sdkp);
2030 					req->__data_len = blk_rq_bytes(req);
2031 					req->rq_flags |= RQF_QUIET;
2032 				}
2033 				break;
2034 			}
2035 		}
2036 		break;
2037 	default:
2038 		break;
2039 	}
2040 
2041  out:
2042 	if (sd_is_zoned(sdkp))
2043 		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2044 
2045 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2046 					   "sd_done: completed %d of %d bytes\n",
2047 					   good_bytes, scsi_bufflen(SCpnt)));
2048 
2049 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2050 		sd_dif_complete(SCpnt, good_bytes);
2051 
2052 	return good_bytes;
2053 }
2054 
2055 /*
2056  * spinup disk - called only in sd_revalidate_disk()
2057  */
2058 static void
2059 sd_spinup_disk(struct scsi_disk *sdkp)
2060 {
2061 	unsigned char cmd[10];
2062 	unsigned long spintime_expire = 0;
2063 	int retries, spintime;
2064 	unsigned int the_result;
2065 	struct scsi_sense_hdr sshdr;
2066 	int sense_valid = 0;
2067 
2068 	spintime = 0;
2069 
2070 	/* Spin up drives, as required.  Only do this at boot time */
2071 	/* Spinup needs to be done for module loads too. */
2072 	do {
2073 		retries = 0;
2074 
2075 		do {
2076 			cmd[0] = TEST_UNIT_READY;
2077 			memset((void *) &cmd[1], 0, 9);
2078 
2079 			the_result = scsi_execute_req(sdkp->device, cmd,
2080 						      DMA_NONE, NULL, 0,
2081 						      &sshdr, SD_TIMEOUT,
2082 						      SD_MAX_RETRIES, NULL);
2083 
2084 			/*
2085 			 * If the drive has indicated to us that it
2086 			 * doesn't have any media in it, don't bother
2087 			 * with any more polling.
2088 			 */
2089 			if (media_not_present(sdkp, &sshdr))
2090 				return;
2091 
2092 			if (the_result)
2093 				sense_valid = scsi_sense_valid(&sshdr);
2094 			retries++;
2095 		} while (retries < 3 &&
2096 			 (!scsi_status_is_good(the_result) ||
2097 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
2098 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2099 
2100 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2101 			/* no sense, TUR either succeeded or failed
2102 			 * with a status error */
2103 			if(!spintime && !scsi_status_is_good(the_result)) {
2104 				sd_print_result(sdkp, "Test Unit Ready failed",
2105 						the_result);
2106 			}
2107 			break;
2108 		}
2109 
2110 		/*
2111 		 * The device does not want the automatic start to be issued.
2112 		 */
2113 		if (sdkp->device->no_start_on_add)
2114 			break;
2115 
2116 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2117 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2118 				break;	/* manual intervention required */
2119 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2120 				break;	/* standby */
2121 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2122 				break;	/* unavailable */
2123 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2124 				break;	/* sanitize in progress */
2125 			/*
2126 			 * Issue command to spin up drive when not ready
2127 			 */
2128 			if (!spintime) {
2129 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2130 				cmd[0] = START_STOP;
2131 				cmd[1] = 1;	/* Return immediately */
2132 				memset((void *) &cmd[2], 0, 8);
2133 				cmd[4] = 1;	/* Start spin cycle */
2134 				if (sdkp->device->start_stop_pwr_cond)
2135 					cmd[4] |= 1 << 4;
2136 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2137 						 NULL, 0, &sshdr,
2138 						 SD_TIMEOUT, SD_MAX_RETRIES,
2139 						 NULL);
2140 				spintime_expire = jiffies + 100 * HZ;
2141 				spintime = 1;
2142 			}
2143 			/* Wait 1 second for next try */
2144 			msleep(1000);
2145 			printk(KERN_CONT ".");
2146 
2147 		/*
2148 		 * Wait for USB flash devices with slow firmware.
2149 		 * Yes, this sense key/ASC combination shouldn't
2150 		 * occur here.  It's characteristic of these devices.
2151 		 */
2152 		} else if (sense_valid &&
2153 				sshdr.sense_key == UNIT_ATTENTION &&
2154 				sshdr.asc == 0x28) {
2155 			if (!spintime) {
2156 				spintime_expire = jiffies + 5 * HZ;
2157 				spintime = 1;
2158 			}
2159 			/* Wait 1 second for next try */
2160 			msleep(1000);
2161 		} else {
2162 			/* we don't understand the sense code, so it's
2163 			 * probably pointless to loop */
2164 			if(!spintime) {
2165 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2166 				sd_print_sense_hdr(sdkp, &sshdr);
2167 			}
2168 			break;
2169 		}
2170 
2171 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2172 
2173 	if (spintime) {
2174 		if (scsi_status_is_good(the_result))
2175 			printk(KERN_CONT "ready\n");
2176 		else
2177 			printk(KERN_CONT "not responding...\n");
2178 	}
2179 }
2180 
2181 /*
2182  * Determine whether disk supports Data Integrity Field.
2183  */
2184 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2185 {
2186 	struct scsi_device *sdp = sdkp->device;
2187 	u8 type;
2188 	int ret = 0;
2189 
2190 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2191 		return ret;
2192 
2193 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2194 
2195 	if (type > T10_PI_TYPE3_PROTECTION)
2196 		ret = -ENODEV;
2197 	else if (scsi_host_dif_capable(sdp->host, type))
2198 		ret = 1;
2199 
2200 	if (sdkp->first_scan || type != sdkp->protection_type)
2201 		switch (ret) {
2202 		case -ENODEV:
2203 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2204 				  " protection type %u. Disabling disk!\n",
2205 				  type);
2206 			break;
2207 		case 1:
2208 			sd_printk(KERN_NOTICE, sdkp,
2209 				  "Enabling DIF Type %u protection\n", type);
2210 			break;
2211 		case 0:
2212 			sd_printk(KERN_NOTICE, sdkp,
2213 				  "Disabling DIF Type %u protection\n", type);
2214 			break;
2215 		}
2216 
2217 	sdkp->protection_type = type;
2218 
2219 	return ret;
2220 }
2221 
2222 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2223 			struct scsi_sense_hdr *sshdr, int sense_valid,
2224 			int the_result)
2225 {
2226 	if (driver_byte(the_result) & DRIVER_SENSE)
2227 		sd_print_sense_hdr(sdkp, sshdr);
2228 	else
2229 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2230 
2231 	/*
2232 	 * Set dirty bit for removable devices if not ready -
2233 	 * sometimes drives will not report this properly.
2234 	 */
2235 	if (sdp->removable &&
2236 	    sense_valid && sshdr->sense_key == NOT_READY)
2237 		set_media_not_present(sdkp);
2238 
2239 	/*
2240 	 * We used to set media_present to 0 here to indicate no media
2241 	 * in the drive, but some drives fail read capacity even with
2242 	 * media present, so we can't do that.
2243 	 */
2244 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2245 }
2246 
2247 #define RC16_LEN 32
2248 #if RC16_LEN > SD_BUF_SIZE
2249 #error RC16_LEN must not be more than SD_BUF_SIZE
2250 #endif
2251 
2252 #define READ_CAPACITY_RETRIES_ON_RESET	10
2253 
2254 /*
2255  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2256  * and the reported logical block size is bigger than 512 bytes. Note
2257  * that last_sector is a u64 and therefore logical_to_sectors() is not
2258  * applicable.
2259  */
2260 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2261 {
2262 	u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2263 
2264 	if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2265 		return false;
2266 
2267 	return true;
2268 }
2269 
2270 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2271 						unsigned char *buffer)
2272 {
2273 	unsigned char cmd[16];
2274 	struct scsi_sense_hdr sshdr;
2275 	int sense_valid = 0;
2276 	int the_result;
2277 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2278 	unsigned int alignment;
2279 	unsigned long long lba;
2280 	unsigned sector_size;
2281 
2282 	if (sdp->no_read_capacity_16)
2283 		return -EINVAL;
2284 
2285 	do {
2286 		memset(cmd, 0, 16);
2287 		cmd[0] = SERVICE_ACTION_IN_16;
2288 		cmd[1] = SAI_READ_CAPACITY_16;
2289 		cmd[13] = RC16_LEN;
2290 		memset(buffer, 0, RC16_LEN);
2291 
2292 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2293 					buffer, RC16_LEN, &sshdr,
2294 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2295 
2296 		if (media_not_present(sdkp, &sshdr))
2297 			return -ENODEV;
2298 
2299 		if (the_result) {
2300 			sense_valid = scsi_sense_valid(&sshdr);
2301 			if (sense_valid &&
2302 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2303 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2304 			    sshdr.ascq == 0x00)
2305 				/* Invalid Command Operation Code or
2306 				 * Invalid Field in CDB, just retry
2307 				 * silently with RC10 */
2308 				return -EINVAL;
2309 			if (sense_valid &&
2310 			    sshdr.sense_key == UNIT_ATTENTION &&
2311 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2312 				/* Device reset might occur several times,
2313 				 * give it one more chance */
2314 				if (--reset_retries > 0)
2315 					continue;
2316 		}
2317 		retries--;
2318 
2319 	} while (the_result && retries);
2320 
2321 	if (the_result) {
2322 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2323 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2324 		return -EINVAL;
2325 	}
2326 
2327 	sector_size = get_unaligned_be32(&buffer[8]);
2328 	lba = get_unaligned_be64(&buffer[0]);
2329 
2330 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2331 		sdkp->capacity = 0;
2332 		return -ENODEV;
2333 	}
2334 
2335 	if (!sd_addressable_capacity(lba, sector_size)) {
2336 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2337 			"kernel compiled with support for large block "
2338 			"devices.\n");
2339 		sdkp->capacity = 0;
2340 		return -EOVERFLOW;
2341 	}
2342 
2343 	/* Logical blocks per physical block exponent */
2344 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2345 
2346 	/* RC basis */
2347 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2348 
2349 	/* Lowest aligned logical block */
2350 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2351 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2352 	if (alignment && sdkp->first_scan)
2353 		sd_printk(KERN_NOTICE, sdkp,
2354 			  "physical block alignment offset: %u\n", alignment);
2355 
2356 	if (buffer[14] & 0x80) { /* LBPME */
2357 		sdkp->lbpme = 1;
2358 
2359 		if (buffer[14] & 0x40) /* LBPRZ */
2360 			sdkp->lbprz = 1;
2361 
2362 		sd_config_discard(sdkp, SD_LBP_WS16);
2363 	}
2364 
2365 	sdkp->capacity = lba + 1;
2366 	return sector_size;
2367 }
2368 
2369 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2370 						unsigned char *buffer)
2371 {
2372 	unsigned char cmd[16];
2373 	struct scsi_sense_hdr sshdr;
2374 	int sense_valid = 0;
2375 	int the_result;
2376 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2377 	sector_t lba;
2378 	unsigned sector_size;
2379 
2380 	do {
2381 		cmd[0] = READ_CAPACITY;
2382 		memset(&cmd[1], 0, 9);
2383 		memset(buffer, 0, 8);
2384 
2385 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2386 					buffer, 8, &sshdr,
2387 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2388 
2389 		if (media_not_present(sdkp, &sshdr))
2390 			return -ENODEV;
2391 
2392 		if (the_result) {
2393 			sense_valid = scsi_sense_valid(&sshdr);
2394 			if (sense_valid &&
2395 			    sshdr.sense_key == UNIT_ATTENTION &&
2396 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2397 				/* Device reset might occur several times,
2398 				 * give it one more chance */
2399 				if (--reset_retries > 0)
2400 					continue;
2401 		}
2402 		retries--;
2403 
2404 	} while (the_result && retries);
2405 
2406 	if (the_result) {
2407 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2408 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2409 		return -EINVAL;
2410 	}
2411 
2412 	sector_size = get_unaligned_be32(&buffer[4]);
2413 	lba = get_unaligned_be32(&buffer[0]);
2414 
2415 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2416 		/* Some buggy (usb cardreader) devices return an lba of
2417 		   0xffffffff when the want to report a size of 0 (with
2418 		   which they really mean no media is present) */
2419 		sdkp->capacity = 0;
2420 		sdkp->physical_block_size = sector_size;
2421 		return sector_size;
2422 	}
2423 
2424 	if (!sd_addressable_capacity(lba, sector_size)) {
2425 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2426 			"kernel compiled with support for large block "
2427 			"devices.\n");
2428 		sdkp->capacity = 0;
2429 		return -EOVERFLOW;
2430 	}
2431 
2432 	sdkp->capacity = lba + 1;
2433 	sdkp->physical_block_size = sector_size;
2434 	return sector_size;
2435 }
2436 
2437 static int sd_try_rc16_first(struct scsi_device *sdp)
2438 {
2439 	if (sdp->host->max_cmd_len < 16)
2440 		return 0;
2441 	if (sdp->try_rc_10_first)
2442 		return 0;
2443 	if (sdp->scsi_level > SCSI_SPC_2)
2444 		return 1;
2445 	if (scsi_device_protection(sdp))
2446 		return 1;
2447 	return 0;
2448 }
2449 
2450 /*
2451  * read disk capacity
2452  */
2453 static void
2454 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2455 {
2456 	int sector_size;
2457 	struct scsi_device *sdp = sdkp->device;
2458 
2459 	if (sd_try_rc16_first(sdp)) {
2460 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2461 		if (sector_size == -EOVERFLOW)
2462 			goto got_data;
2463 		if (sector_size == -ENODEV)
2464 			return;
2465 		if (sector_size < 0)
2466 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2467 		if (sector_size < 0)
2468 			return;
2469 	} else {
2470 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2471 		if (sector_size == -EOVERFLOW)
2472 			goto got_data;
2473 		if (sector_size < 0)
2474 			return;
2475 		if ((sizeof(sdkp->capacity) > 4) &&
2476 		    (sdkp->capacity > 0xffffffffULL)) {
2477 			int old_sector_size = sector_size;
2478 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2479 					"Trying to use READ CAPACITY(16).\n");
2480 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2481 			if (sector_size < 0) {
2482 				sd_printk(KERN_NOTICE, sdkp,
2483 					"Using 0xffffffff as device size\n");
2484 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2485 				sector_size = old_sector_size;
2486 				goto got_data;
2487 			}
2488 			/* Remember that READ CAPACITY(16) succeeded */
2489 			sdp->try_rc_10_first = 0;
2490 		}
2491 	}
2492 
2493 	/* Some devices are known to return the total number of blocks,
2494 	 * not the highest block number.  Some devices have versions
2495 	 * which do this and others which do not.  Some devices we might
2496 	 * suspect of doing this but we don't know for certain.
2497 	 *
2498 	 * If we know the reported capacity is wrong, decrement it.  If
2499 	 * we can only guess, then assume the number of blocks is even
2500 	 * (usually true but not always) and err on the side of lowering
2501 	 * the capacity.
2502 	 */
2503 	if (sdp->fix_capacity ||
2504 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2505 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2506 				"from its reported value: %llu\n",
2507 				(unsigned long long) sdkp->capacity);
2508 		--sdkp->capacity;
2509 	}
2510 
2511 got_data:
2512 	if (sector_size == 0) {
2513 		sector_size = 512;
2514 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2515 			  "assuming 512.\n");
2516 	}
2517 
2518 	if (sector_size != 512 &&
2519 	    sector_size != 1024 &&
2520 	    sector_size != 2048 &&
2521 	    sector_size != 4096) {
2522 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2523 			  sector_size);
2524 		/*
2525 		 * The user might want to re-format the drive with
2526 		 * a supported sectorsize.  Once this happens, it
2527 		 * would be relatively trivial to set the thing up.
2528 		 * For this reason, we leave the thing in the table.
2529 		 */
2530 		sdkp->capacity = 0;
2531 		/*
2532 		 * set a bogus sector size so the normal read/write
2533 		 * logic in the block layer will eventually refuse any
2534 		 * request on this device without tripping over power
2535 		 * of two sector size assumptions
2536 		 */
2537 		sector_size = 512;
2538 	}
2539 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2540 	blk_queue_physical_block_size(sdp->request_queue,
2541 				      sdkp->physical_block_size);
2542 	sdkp->device->sector_size = sector_size;
2543 
2544 	if (sdkp->capacity > 0xffffffff)
2545 		sdp->use_16_for_rw = 1;
2546 
2547 }
2548 
2549 /*
2550  * Print disk capacity
2551  */
2552 static void
2553 sd_print_capacity(struct scsi_disk *sdkp,
2554 		  sector_t old_capacity)
2555 {
2556 	int sector_size = sdkp->device->sector_size;
2557 	char cap_str_2[10], cap_str_10[10];
2558 
2559 	string_get_size(sdkp->capacity, sector_size,
2560 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2561 	string_get_size(sdkp->capacity, sector_size,
2562 			STRING_UNITS_10, cap_str_10,
2563 			sizeof(cap_str_10));
2564 
2565 	if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2566 		sd_printk(KERN_NOTICE, sdkp,
2567 			  "%llu %d-byte logical blocks: (%s/%s)\n",
2568 			  (unsigned long long)sdkp->capacity,
2569 			  sector_size, cap_str_10, cap_str_2);
2570 
2571 		if (sdkp->physical_block_size != sector_size)
2572 			sd_printk(KERN_NOTICE, sdkp,
2573 				  "%u-byte physical blocks\n",
2574 				  sdkp->physical_block_size);
2575 
2576 		sd_zbc_print_zones(sdkp);
2577 	}
2578 }
2579 
2580 /* called with buffer of length 512 */
2581 static inline int
2582 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2583 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2584 		 struct scsi_sense_hdr *sshdr)
2585 {
2586 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2587 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2588 			       sshdr);
2589 }
2590 
2591 /*
2592  * read write protect setting, if possible - called only in sd_revalidate_disk()
2593  * called with buffer of length SD_BUF_SIZE
2594  */
2595 static void
2596 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2597 {
2598 	int res;
2599 	struct scsi_device *sdp = sdkp->device;
2600 	struct scsi_mode_data data;
2601 	int disk_ro = get_disk_ro(sdkp->disk);
2602 	int old_wp = sdkp->write_prot;
2603 
2604 	set_disk_ro(sdkp->disk, 0);
2605 	if (sdp->skip_ms_page_3f) {
2606 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2607 		return;
2608 	}
2609 
2610 	if (sdp->use_192_bytes_for_3f) {
2611 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2612 	} else {
2613 		/*
2614 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2615 		 * We have to start carefully: some devices hang if we ask
2616 		 * for more than is available.
2617 		 */
2618 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2619 
2620 		/*
2621 		 * Second attempt: ask for page 0 When only page 0 is
2622 		 * implemented, a request for page 3F may return Sense Key
2623 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2624 		 * CDB.
2625 		 */
2626 		if (!scsi_status_is_good(res))
2627 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2628 
2629 		/*
2630 		 * Third attempt: ask 255 bytes, as we did earlier.
2631 		 */
2632 		if (!scsi_status_is_good(res))
2633 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2634 					       &data, NULL);
2635 	}
2636 
2637 	if (!scsi_status_is_good(res)) {
2638 		sd_first_printk(KERN_WARNING, sdkp,
2639 			  "Test WP failed, assume Write Enabled\n");
2640 	} else {
2641 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2642 		set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2643 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2644 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2645 				  sdkp->write_prot ? "on" : "off");
2646 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2647 		}
2648 	}
2649 }
2650 
2651 /*
2652  * sd_read_cache_type - called only from sd_revalidate_disk()
2653  * called with buffer of length SD_BUF_SIZE
2654  */
2655 static void
2656 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2657 {
2658 	int len = 0, res;
2659 	struct scsi_device *sdp = sdkp->device;
2660 
2661 	int dbd;
2662 	int modepage;
2663 	int first_len;
2664 	struct scsi_mode_data data;
2665 	struct scsi_sense_hdr sshdr;
2666 	int old_wce = sdkp->WCE;
2667 	int old_rcd = sdkp->RCD;
2668 	int old_dpofua = sdkp->DPOFUA;
2669 
2670 
2671 	if (sdkp->cache_override)
2672 		return;
2673 
2674 	first_len = 4;
2675 	if (sdp->skip_ms_page_8) {
2676 		if (sdp->type == TYPE_RBC)
2677 			goto defaults;
2678 		else {
2679 			if (sdp->skip_ms_page_3f)
2680 				goto defaults;
2681 			modepage = 0x3F;
2682 			if (sdp->use_192_bytes_for_3f)
2683 				first_len = 192;
2684 			dbd = 0;
2685 		}
2686 	} else if (sdp->type == TYPE_RBC) {
2687 		modepage = 6;
2688 		dbd = 8;
2689 	} else {
2690 		modepage = 8;
2691 		dbd = 0;
2692 	}
2693 
2694 	/* cautiously ask */
2695 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2696 			&data, &sshdr);
2697 
2698 	if (!scsi_status_is_good(res))
2699 		goto bad_sense;
2700 
2701 	if (!data.header_length) {
2702 		modepage = 6;
2703 		first_len = 0;
2704 		sd_first_printk(KERN_ERR, sdkp,
2705 				"Missing header in MODE_SENSE response\n");
2706 	}
2707 
2708 	/* that went OK, now ask for the proper length */
2709 	len = data.length;
2710 
2711 	/*
2712 	 * We're only interested in the first three bytes, actually.
2713 	 * But the data cache page is defined for the first 20.
2714 	 */
2715 	if (len < 3)
2716 		goto bad_sense;
2717 	else if (len > SD_BUF_SIZE) {
2718 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2719 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2720 		len = SD_BUF_SIZE;
2721 	}
2722 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2723 		len = 192;
2724 
2725 	/* Get the data */
2726 	if (len > first_len)
2727 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2728 				&data, &sshdr);
2729 
2730 	if (scsi_status_is_good(res)) {
2731 		int offset = data.header_length + data.block_descriptor_length;
2732 
2733 		while (offset < len) {
2734 			u8 page_code = buffer[offset] & 0x3F;
2735 			u8 spf       = buffer[offset] & 0x40;
2736 
2737 			if (page_code == 8 || page_code == 6) {
2738 				/* We're interested only in the first 3 bytes.
2739 				 */
2740 				if (len - offset <= 2) {
2741 					sd_first_printk(KERN_ERR, sdkp,
2742 						"Incomplete mode parameter "
2743 							"data\n");
2744 					goto defaults;
2745 				} else {
2746 					modepage = page_code;
2747 					goto Page_found;
2748 				}
2749 			} else {
2750 				/* Go to the next page */
2751 				if (spf && len - offset > 3)
2752 					offset += 4 + (buffer[offset+2] << 8) +
2753 						buffer[offset+3];
2754 				else if (!spf && len - offset > 1)
2755 					offset += 2 + buffer[offset+1];
2756 				else {
2757 					sd_first_printk(KERN_ERR, sdkp,
2758 							"Incomplete mode "
2759 							"parameter data\n");
2760 					goto defaults;
2761 				}
2762 			}
2763 		}
2764 
2765 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2766 		goto defaults;
2767 
2768 	Page_found:
2769 		if (modepage == 8) {
2770 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2771 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2772 		} else {
2773 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2774 			sdkp->RCD = 0;
2775 		}
2776 
2777 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2778 		if (sdp->broken_fua) {
2779 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2780 			sdkp->DPOFUA = 0;
2781 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2782 			   !sdkp->device->use_16_for_rw) {
2783 			sd_first_printk(KERN_NOTICE, sdkp,
2784 				  "Uses READ/WRITE(6), disabling FUA\n");
2785 			sdkp->DPOFUA = 0;
2786 		}
2787 
2788 		/* No cache flush allowed for write protected devices */
2789 		if (sdkp->WCE && sdkp->write_prot)
2790 			sdkp->WCE = 0;
2791 
2792 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2793 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2794 			sd_printk(KERN_NOTICE, sdkp,
2795 				  "Write cache: %s, read cache: %s, %s\n",
2796 				  sdkp->WCE ? "enabled" : "disabled",
2797 				  sdkp->RCD ? "disabled" : "enabled",
2798 				  sdkp->DPOFUA ? "supports DPO and FUA"
2799 				  : "doesn't support DPO or FUA");
2800 
2801 		return;
2802 	}
2803 
2804 bad_sense:
2805 	if (scsi_sense_valid(&sshdr) &&
2806 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2807 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2808 		/* Invalid field in CDB */
2809 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2810 	else
2811 		sd_first_printk(KERN_ERR, sdkp,
2812 				"Asking for cache data failed\n");
2813 
2814 defaults:
2815 	if (sdp->wce_default_on) {
2816 		sd_first_printk(KERN_NOTICE, sdkp,
2817 				"Assuming drive cache: write back\n");
2818 		sdkp->WCE = 1;
2819 	} else {
2820 		sd_first_printk(KERN_ERR, sdkp,
2821 				"Assuming drive cache: write through\n");
2822 		sdkp->WCE = 0;
2823 	}
2824 	sdkp->RCD = 0;
2825 	sdkp->DPOFUA = 0;
2826 }
2827 
2828 /*
2829  * The ATO bit indicates whether the DIF application tag is available
2830  * for use by the operating system.
2831  */
2832 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2833 {
2834 	int res, offset;
2835 	struct scsi_device *sdp = sdkp->device;
2836 	struct scsi_mode_data data;
2837 	struct scsi_sense_hdr sshdr;
2838 
2839 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2840 		return;
2841 
2842 	if (sdkp->protection_type == 0)
2843 		return;
2844 
2845 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2846 			      SD_MAX_RETRIES, &data, &sshdr);
2847 
2848 	if (!scsi_status_is_good(res) || !data.header_length ||
2849 	    data.length < 6) {
2850 		sd_first_printk(KERN_WARNING, sdkp,
2851 			  "getting Control mode page failed, assume no ATO\n");
2852 
2853 		if (scsi_sense_valid(&sshdr))
2854 			sd_print_sense_hdr(sdkp, &sshdr);
2855 
2856 		return;
2857 	}
2858 
2859 	offset = data.header_length + data.block_descriptor_length;
2860 
2861 	if ((buffer[offset] & 0x3f) != 0x0a) {
2862 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2863 		return;
2864 	}
2865 
2866 	if ((buffer[offset + 5] & 0x80) == 0)
2867 		return;
2868 
2869 	sdkp->ATO = 1;
2870 
2871 	return;
2872 }
2873 
2874 /**
2875  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2876  * @sdkp: disk to query
2877  */
2878 static void sd_read_block_limits(struct scsi_disk *sdkp)
2879 {
2880 	unsigned int sector_sz = sdkp->device->sector_size;
2881 	const int vpd_len = 64;
2882 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2883 
2884 	if (!buffer ||
2885 	    /* Block Limits VPD */
2886 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2887 		goto out;
2888 
2889 	blk_queue_io_min(sdkp->disk->queue,
2890 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2891 
2892 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2893 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2894 
2895 	if (buffer[3] == 0x3c) {
2896 		unsigned int lba_count, desc_count;
2897 
2898 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2899 
2900 		if (!sdkp->lbpme)
2901 			goto out;
2902 
2903 		lba_count = get_unaligned_be32(&buffer[20]);
2904 		desc_count = get_unaligned_be32(&buffer[24]);
2905 
2906 		if (lba_count && desc_count)
2907 			sdkp->max_unmap_blocks = lba_count;
2908 
2909 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2910 
2911 		if (buffer[32] & 0x80)
2912 			sdkp->unmap_alignment =
2913 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2914 
2915 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2916 
2917 			if (sdkp->max_unmap_blocks)
2918 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2919 			else
2920 				sd_config_discard(sdkp, SD_LBP_WS16);
2921 
2922 		} else {	/* LBP VPD page tells us what to use */
2923 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2924 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2925 			else if (sdkp->lbpws)
2926 				sd_config_discard(sdkp, SD_LBP_WS16);
2927 			else if (sdkp->lbpws10)
2928 				sd_config_discard(sdkp, SD_LBP_WS10);
2929 			else
2930 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2931 		}
2932 	}
2933 
2934  out:
2935 	kfree(buffer);
2936 }
2937 
2938 /**
2939  * sd_read_block_characteristics - Query block dev. characteristics
2940  * @sdkp: disk to query
2941  */
2942 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2943 {
2944 	struct request_queue *q = sdkp->disk->queue;
2945 	unsigned char *buffer;
2946 	u16 rot;
2947 	const int vpd_len = 64;
2948 
2949 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2950 
2951 	if (!buffer ||
2952 	    /* Block Device Characteristics VPD */
2953 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2954 		goto out;
2955 
2956 	rot = get_unaligned_be16(&buffer[4]);
2957 
2958 	if (rot == 1) {
2959 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2960 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2961 	}
2962 
2963 	if (sdkp->device->type == TYPE_ZBC) {
2964 		/* Host-managed */
2965 		q->limits.zoned = BLK_ZONED_HM;
2966 	} else {
2967 		sdkp->zoned = (buffer[8] >> 4) & 3;
2968 		if (sdkp->zoned == 1)
2969 			/* Host-aware */
2970 			q->limits.zoned = BLK_ZONED_HA;
2971 		else
2972 			/*
2973 			 * Treat drive-managed devices as
2974 			 * regular block devices.
2975 			 */
2976 			q->limits.zoned = BLK_ZONED_NONE;
2977 	}
2978 	if (blk_queue_is_zoned(q) && sdkp->first_scan)
2979 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2980 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2981 
2982  out:
2983 	kfree(buffer);
2984 }
2985 
2986 /**
2987  * sd_read_block_provisioning - Query provisioning VPD page
2988  * @sdkp: disk to query
2989  */
2990 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2991 {
2992 	unsigned char *buffer;
2993 	const int vpd_len = 8;
2994 
2995 	if (sdkp->lbpme == 0)
2996 		return;
2997 
2998 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2999 
3000 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3001 		goto out;
3002 
3003 	sdkp->lbpvpd	= 1;
3004 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
3005 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
3006 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
3007 
3008  out:
3009 	kfree(buffer);
3010 }
3011 
3012 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3013 {
3014 	struct scsi_device *sdev = sdkp->device;
3015 
3016 	if (sdev->host->no_write_same) {
3017 		sdev->no_write_same = 1;
3018 
3019 		return;
3020 	}
3021 
3022 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3023 		/* too large values might cause issues with arcmsr */
3024 		int vpd_buf_len = 64;
3025 
3026 		sdev->no_report_opcodes = 1;
3027 
3028 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3029 		 * CODES is unsupported and the device has an ATA
3030 		 * Information VPD page (SAT).
3031 		 */
3032 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3033 			sdev->no_write_same = 1;
3034 	}
3035 
3036 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3037 		sdkp->ws16 = 1;
3038 
3039 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3040 		sdkp->ws10 = 1;
3041 }
3042 
3043 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3044 {
3045 	struct scsi_device *sdev = sdkp->device;
3046 
3047 	if (!sdev->security_supported)
3048 		return;
3049 
3050 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3051 			SECURITY_PROTOCOL_IN) == 1 &&
3052 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3053 			SECURITY_PROTOCOL_OUT) == 1)
3054 		sdkp->security = 1;
3055 }
3056 
3057 /**
3058  *	sd_revalidate_disk - called the first time a new disk is seen,
3059  *	performs disk spin up, read_capacity, etc.
3060  *	@disk: struct gendisk we care about
3061  **/
3062 static int sd_revalidate_disk(struct gendisk *disk)
3063 {
3064 	struct scsi_disk *sdkp = scsi_disk(disk);
3065 	struct scsi_device *sdp = sdkp->device;
3066 	struct request_queue *q = sdkp->disk->queue;
3067 	sector_t old_capacity = sdkp->capacity;
3068 	unsigned char *buffer;
3069 	unsigned int dev_max, rw_max;
3070 
3071 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3072 				      "sd_revalidate_disk\n"));
3073 
3074 	/*
3075 	 * If the device is offline, don't try and read capacity or any
3076 	 * of the other niceties.
3077 	 */
3078 	if (!scsi_device_online(sdp))
3079 		goto out;
3080 
3081 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3082 	if (!buffer) {
3083 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3084 			  "allocation failure.\n");
3085 		goto out;
3086 	}
3087 
3088 	sd_spinup_disk(sdkp);
3089 
3090 	/*
3091 	 * Without media there is no reason to ask; moreover, some devices
3092 	 * react badly if we do.
3093 	 */
3094 	if (sdkp->media_present) {
3095 		sd_read_capacity(sdkp, buffer);
3096 
3097 		if (scsi_device_supports_vpd(sdp)) {
3098 			sd_read_block_provisioning(sdkp);
3099 			sd_read_block_limits(sdkp);
3100 			sd_read_block_characteristics(sdkp);
3101 			sd_zbc_read_zones(sdkp, buffer);
3102 		}
3103 
3104 		sd_print_capacity(sdkp, old_capacity);
3105 
3106 		sd_read_write_protect_flag(sdkp, buffer);
3107 		sd_read_cache_type(sdkp, buffer);
3108 		sd_read_app_tag_own(sdkp, buffer);
3109 		sd_read_write_same(sdkp, buffer);
3110 		sd_read_security(sdkp, buffer);
3111 	}
3112 
3113 	/*
3114 	 * We now have all cache related info, determine how we deal
3115 	 * with flush requests.
3116 	 */
3117 	sd_set_flush_flag(sdkp);
3118 
3119 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3120 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3121 
3122 	/* Some devices report a maximum block count for READ/WRITE requests. */
3123 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3124 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3125 
3126 	/*
3127 	 * Determine the device's preferred I/O size for reads and writes
3128 	 * unless the reported value is unreasonably small, large, or
3129 	 * garbage.
3130 	 */
3131 	if (sdkp->opt_xfer_blocks &&
3132 	    sdkp->opt_xfer_blocks <= dev_max &&
3133 	    sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3134 	    logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3135 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3136 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3137 	} else
3138 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3139 				      (sector_t)BLK_DEF_MAX_SECTORS);
3140 
3141 	/* Do not exceed controller limit */
3142 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3143 
3144 	/*
3145 	 * Only update max_sectors if previously unset or if the current value
3146 	 * exceeds the capabilities of the hardware.
3147 	 */
3148 	if (sdkp->first_scan ||
3149 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3150 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3151 		q->limits.max_sectors = rw_max;
3152 
3153 	sdkp->first_scan = 0;
3154 
3155 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3156 	sd_config_write_same(sdkp);
3157 	kfree(buffer);
3158 
3159  out:
3160 	return 0;
3161 }
3162 
3163 /**
3164  *	sd_unlock_native_capacity - unlock native capacity
3165  *	@disk: struct gendisk to set capacity for
3166  *
3167  *	Block layer calls this function if it detects that partitions
3168  *	on @disk reach beyond the end of the device.  If the SCSI host
3169  *	implements ->unlock_native_capacity() method, it's invoked to
3170  *	give it a chance to adjust the device capacity.
3171  *
3172  *	CONTEXT:
3173  *	Defined by block layer.  Might sleep.
3174  */
3175 static void sd_unlock_native_capacity(struct gendisk *disk)
3176 {
3177 	struct scsi_device *sdev = scsi_disk(disk)->device;
3178 
3179 	if (sdev->host->hostt->unlock_native_capacity)
3180 		sdev->host->hostt->unlock_native_capacity(sdev);
3181 }
3182 
3183 /**
3184  *	sd_format_disk_name - format disk name
3185  *	@prefix: name prefix - ie. "sd" for SCSI disks
3186  *	@index: index of the disk to format name for
3187  *	@buf: output buffer
3188  *	@buflen: length of the output buffer
3189  *
3190  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3191  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3192  *	which is followed by sdaaa.
3193  *
3194  *	This is basically 26 base counting with one extra 'nil' entry
3195  *	at the beginning from the second digit on and can be
3196  *	determined using similar method as 26 base conversion with the
3197  *	index shifted -1 after each digit is computed.
3198  *
3199  *	CONTEXT:
3200  *	Don't care.
3201  *
3202  *	RETURNS:
3203  *	0 on success, -errno on failure.
3204  */
3205 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3206 {
3207 	const int base = 'z' - 'a' + 1;
3208 	char *begin = buf + strlen(prefix);
3209 	char *end = buf + buflen;
3210 	char *p;
3211 	int unit;
3212 
3213 	p = end - 1;
3214 	*p = '\0';
3215 	unit = base;
3216 	do {
3217 		if (p == begin)
3218 			return -EINVAL;
3219 		*--p = 'a' + (index % unit);
3220 		index = (index / unit) - 1;
3221 	} while (index >= 0);
3222 
3223 	memmove(begin, p, end - p);
3224 	memcpy(buf, prefix, strlen(prefix));
3225 
3226 	return 0;
3227 }
3228 
3229 /*
3230  * The asynchronous part of sd_probe
3231  */
3232 static void sd_probe_async(void *data, async_cookie_t cookie)
3233 {
3234 	struct scsi_disk *sdkp = data;
3235 	struct scsi_device *sdp;
3236 	struct gendisk *gd;
3237 	u32 index;
3238 	struct device *dev;
3239 
3240 	sdp = sdkp->device;
3241 	gd = sdkp->disk;
3242 	index = sdkp->index;
3243 	dev = &sdp->sdev_gendev;
3244 
3245 	gd->major = sd_major((index & 0xf0) >> 4);
3246 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3247 
3248 	gd->fops = &sd_fops;
3249 	gd->private_data = &sdkp->driver;
3250 	gd->queue = sdkp->device->request_queue;
3251 
3252 	/* defaults, until the device tells us otherwise */
3253 	sdp->sector_size = 512;
3254 	sdkp->capacity = 0;
3255 	sdkp->media_present = 1;
3256 	sdkp->write_prot = 0;
3257 	sdkp->cache_override = 0;
3258 	sdkp->WCE = 0;
3259 	sdkp->RCD = 0;
3260 	sdkp->ATO = 0;
3261 	sdkp->first_scan = 1;
3262 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3263 
3264 	sd_revalidate_disk(gd);
3265 
3266 	gd->flags = GENHD_FL_EXT_DEVT;
3267 	if (sdp->removable) {
3268 		gd->flags |= GENHD_FL_REMOVABLE;
3269 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3270 	}
3271 
3272 	blk_pm_runtime_init(sdp->request_queue, dev);
3273 	device_add_disk(dev, gd);
3274 	if (sdkp->capacity)
3275 		sd_dif_config_host(sdkp);
3276 
3277 	sd_revalidate_disk(gd);
3278 
3279 	if (sdkp->security) {
3280 		sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3281 		if (sdkp->opal_dev)
3282 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3283 	}
3284 
3285 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3286 		  sdp->removable ? "removable " : "");
3287 	scsi_autopm_put_device(sdp);
3288 	put_device(&sdkp->dev);
3289 }
3290 
3291 /**
3292  *	sd_probe - called during driver initialization and whenever a
3293  *	new scsi device is attached to the system. It is called once
3294  *	for each scsi device (not just disks) present.
3295  *	@dev: pointer to device object
3296  *
3297  *	Returns 0 if successful (or not interested in this scsi device
3298  *	(e.g. scanner)); 1 when there is an error.
3299  *
3300  *	Note: this function is invoked from the scsi mid-level.
3301  *	This function sets up the mapping between a given
3302  *	<host,channel,id,lun> (found in sdp) and new device name
3303  *	(e.g. /dev/sda). More precisely it is the block device major
3304  *	and minor number that is chosen here.
3305  *
3306  *	Assume sd_probe is not re-entrant (for time being)
3307  *	Also think about sd_probe() and sd_remove() running coincidentally.
3308  **/
3309 static int sd_probe(struct device *dev)
3310 {
3311 	struct scsi_device *sdp = to_scsi_device(dev);
3312 	struct scsi_disk *sdkp;
3313 	struct gendisk *gd;
3314 	int index;
3315 	int error;
3316 
3317 	scsi_autopm_get_device(sdp);
3318 	error = -ENODEV;
3319 	if (sdp->type != TYPE_DISK &&
3320 	    sdp->type != TYPE_ZBC &&
3321 	    sdp->type != TYPE_MOD &&
3322 	    sdp->type != TYPE_RBC)
3323 		goto out;
3324 
3325 #ifndef CONFIG_BLK_DEV_ZONED
3326 	if (sdp->type == TYPE_ZBC)
3327 		goto out;
3328 #endif
3329 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3330 					"sd_probe\n"));
3331 
3332 	error = -ENOMEM;
3333 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3334 	if (!sdkp)
3335 		goto out;
3336 
3337 	gd = alloc_disk(SD_MINORS);
3338 	if (!gd)
3339 		goto out_free;
3340 
3341 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3342 	if (index < 0) {
3343 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3344 		goto out_put;
3345 	}
3346 
3347 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3348 	if (error) {
3349 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3350 		goto out_free_index;
3351 	}
3352 
3353 	sdkp->device = sdp;
3354 	sdkp->driver = &sd_template;
3355 	sdkp->disk = gd;
3356 	sdkp->index = index;
3357 	atomic_set(&sdkp->openers, 0);
3358 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3359 
3360 	if (!sdp->request_queue->rq_timeout) {
3361 		if (sdp->type != TYPE_MOD)
3362 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3363 		else
3364 			blk_queue_rq_timeout(sdp->request_queue,
3365 					     SD_MOD_TIMEOUT);
3366 	}
3367 
3368 	device_initialize(&sdkp->dev);
3369 	sdkp->dev.parent = dev;
3370 	sdkp->dev.class = &sd_disk_class;
3371 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3372 
3373 	error = device_add(&sdkp->dev);
3374 	if (error)
3375 		goto out_free_index;
3376 
3377 	get_device(dev);
3378 	dev_set_drvdata(dev, sdkp);
3379 
3380 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3381 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3382 
3383 	return 0;
3384 
3385  out_free_index:
3386 	ida_free(&sd_index_ida, index);
3387  out_put:
3388 	put_disk(gd);
3389  out_free:
3390 	kfree(sdkp);
3391  out:
3392 	scsi_autopm_put_device(sdp);
3393 	return error;
3394 }
3395 
3396 /**
3397  *	sd_remove - called whenever a scsi disk (previously recognized by
3398  *	sd_probe) is detached from the system. It is called (potentially
3399  *	multiple times) during sd module unload.
3400  *	@dev: pointer to device object
3401  *
3402  *	Note: this function is invoked from the scsi mid-level.
3403  *	This function potentially frees up a device name (e.g. /dev/sdc)
3404  *	that could be re-used by a subsequent sd_probe().
3405  *	This function is not called when the built-in sd driver is "exit-ed".
3406  **/
3407 static int sd_remove(struct device *dev)
3408 {
3409 	struct scsi_disk *sdkp;
3410 	dev_t devt;
3411 
3412 	sdkp = dev_get_drvdata(dev);
3413 	devt = disk_devt(sdkp->disk);
3414 	scsi_autopm_get_device(sdkp->device);
3415 
3416 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3417 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3418 	device_del(&sdkp->dev);
3419 	del_gendisk(sdkp->disk);
3420 	sd_shutdown(dev);
3421 
3422 	sd_zbc_remove(sdkp);
3423 
3424 	free_opal_dev(sdkp->opal_dev);
3425 
3426 	blk_register_region(devt, SD_MINORS, NULL,
3427 			    sd_default_probe, NULL, NULL);
3428 
3429 	mutex_lock(&sd_ref_mutex);
3430 	dev_set_drvdata(dev, NULL);
3431 	put_device(&sdkp->dev);
3432 	mutex_unlock(&sd_ref_mutex);
3433 
3434 	return 0;
3435 }
3436 
3437 /**
3438  *	scsi_disk_release - Called to free the scsi_disk structure
3439  *	@dev: pointer to embedded class device
3440  *
3441  *	sd_ref_mutex must be held entering this routine.  Because it is
3442  *	called on last put, you should always use the scsi_disk_get()
3443  *	scsi_disk_put() helpers which manipulate the semaphore directly
3444  *	and never do a direct put_device.
3445  **/
3446 static void scsi_disk_release(struct device *dev)
3447 {
3448 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3449 	struct gendisk *disk = sdkp->disk;
3450 
3451 	ida_free(&sd_index_ida, sdkp->index);
3452 
3453 	disk->private_data = NULL;
3454 	put_disk(disk);
3455 	put_device(&sdkp->device->sdev_gendev);
3456 
3457 	kfree(sdkp);
3458 }
3459 
3460 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3461 {
3462 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3463 	struct scsi_sense_hdr sshdr;
3464 	struct scsi_device *sdp = sdkp->device;
3465 	int res;
3466 
3467 	if (start)
3468 		cmd[4] |= 1;	/* START */
3469 
3470 	if (sdp->start_stop_pwr_cond)
3471 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3472 
3473 	if (!scsi_device_online(sdp))
3474 		return -ENODEV;
3475 
3476 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3477 			SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3478 	if (res) {
3479 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3480 		if (driver_byte(res) & DRIVER_SENSE)
3481 			sd_print_sense_hdr(sdkp, &sshdr);
3482 		if (scsi_sense_valid(&sshdr) &&
3483 			/* 0x3a is medium not present */
3484 			sshdr.asc == 0x3a)
3485 			res = 0;
3486 	}
3487 
3488 	/* SCSI error codes must not go to the generic layer */
3489 	if (res)
3490 		return -EIO;
3491 
3492 	return 0;
3493 }
3494 
3495 /*
3496  * Send a SYNCHRONIZE CACHE instruction down to the device through
3497  * the normal SCSI command structure.  Wait for the command to
3498  * complete.
3499  */
3500 static void sd_shutdown(struct device *dev)
3501 {
3502 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3503 
3504 	if (!sdkp)
3505 		return;         /* this can happen */
3506 
3507 	if (pm_runtime_suspended(dev))
3508 		return;
3509 
3510 	if (sdkp->WCE && sdkp->media_present) {
3511 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3512 		sd_sync_cache(sdkp, NULL);
3513 	}
3514 
3515 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3516 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3517 		sd_start_stop_device(sdkp, 0);
3518 	}
3519 }
3520 
3521 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3522 {
3523 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3524 	struct scsi_sense_hdr sshdr;
3525 	int ret = 0;
3526 
3527 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3528 		return 0;
3529 
3530 	if (sdkp->WCE && sdkp->media_present) {
3531 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3532 		ret = sd_sync_cache(sdkp, &sshdr);
3533 
3534 		if (ret) {
3535 			/* ignore OFFLINE device */
3536 			if (ret == -ENODEV)
3537 				return 0;
3538 
3539 			if (!scsi_sense_valid(&sshdr) ||
3540 			    sshdr.sense_key != ILLEGAL_REQUEST)
3541 				return ret;
3542 
3543 			/*
3544 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3545 			 * doesn't support sync. There's not much to do and
3546 			 * suspend shouldn't fail.
3547 			 */
3548 			ret = 0;
3549 		}
3550 	}
3551 
3552 	if (sdkp->device->manage_start_stop) {
3553 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3554 		/* an error is not worth aborting a system sleep */
3555 		ret = sd_start_stop_device(sdkp, 0);
3556 		if (ignore_stop_errors)
3557 			ret = 0;
3558 	}
3559 
3560 	return ret;
3561 }
3562 
3563 static int sd_suspend_system(struct device *dev)
3564 {
3565 	return sd_suspend_common(dev, true);
3566 }
3567 
3568 static int sd_suspend_runtime(struct device *dev)
3569 {
3570 	return sd_suspend_common(dev, false);
3571 }
3572 
3573 static int sd_resume(struct device *dev)
3574 {
3575 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3576 	int ret;
3577 
3578 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3579 		return 0;
3580 
3581 	if (!sdkp->device->manage_start_stop)
3582 		return 0;
3583 
3584 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3585 	ret = sd_start_stop_device(sdkp, 1);
3586 	if (!ret)
3587 		opal_unlock_from_suspend(sdkp->opal_dev);
3588 	return ret;
3589 }
3590 
3591 /**
3592  *	init_sd - entry point for this driver (both when built in or when
3593  *	a module).
3594  *
3595  *	Note: this function registers this driver with the scsi mid-level.
3596  **/
3597 static int __init init_sd(void)
3598 {
3599 	int majors = 0, i, err;
3600 
3601 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3602 
3603 	for (i = 0; i < SD_MAJORS; i++) {
3604 		if (register_blkdev(sd_major(i), "sd") != 0)
3605 			continue;
3606 		majors++;
3607 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3608 				    sd_default_probe, NULL, NULL);
3609 	}
3610 
3611 	if (!majors)
3612 		return -ENODEV;
3613 
3614 	err = class_register(&sd_disk_class);
3615 	if (err)
3616 		goto err_out;
3617 
3618 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3619 					 0, 0, NULL);
3620 	if (!sd_cdb_cache) {
3621 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3622 		err = -ENOMEM;
3623 		goto err_out_class;
3624 	}
3625 
3626 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3627 	if (!sd_cdb_pool) {
3628 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3629 		err = -ENOMEM;
3630 		goto err_out_cache;
3631 	}
3632 
3633 	err = scsi_register_driver(&sd_template.gendrv);
3634 	if (err)
3635 		goto err_out_driver;
3636 
3637 	return 0;
3638 
3639 err_out_driver:
3640 	mempool_destroy(sd_cdb_pool);
3641 
3642 err_out_cache:
3643 	kmem_cache_destroy(sd_cdb_cache);
3644 
3645 err_out_class:
3646 	class_unregister(&sd_disk_class);
3647 err_out:
3648 	for (i = 0; i < SD_MAJORS; i++)
3649 		unregister_blkdev(sd_major(i), "sd");
3650 	return err;
3651 }
3652 
3653 /**
3654  *	exit_sd - exit point for this driver (when it is a module).
3655  *
3656  *	Note: this function unregisters this driver from the scsi mid-level.
3657  **/
3658 static void __exit exit_sd(void)
3659 {
3660 	int i;
3661 
3662 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3663 
3664 	scsi_unregister_driver(&sd_template.gendrv);
3665 	mempool_destroy(sd_cdb_pool);
3666 	kmem_cache_destroy(sd_cdb_cache);
3667 
3668 	class_unregister(&sd_disk_class);
3669 
3670 	for (i = 0; i < SD_MAJORS; i++) {
3671 		blk_unregister_region(sd_major(i), SD_MINORS);
3672 		unregister_blkdev(sd_major(i), "sd");
3673 	}
3674 }
3675 
3676 module_init(init_sd);
3677 module_exit(exit_sd);
3678 
3679 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3680 			       struct scsi_sense_hdr *sshdr)
3681 {
3682 	scsi_print_sense_hdr(sdkp->device,
3683 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3684 }
3685 
3686 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3687 			    int result)
3688 {
3689 	const char *hb_string = scsi_hostbyte_string(result);
3690 	const char *db_string = scsi_driverbyte_string(result);
3691 
3692 	if (hb_string || db_string)
3693 		sd_printk(KERN_INFO, sdkp,
3694 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3695 			  hb_string ? hb_string : "invalid",
3696 			  db_string ? db_string : "invalid");
3697 	else
3698 		sd_printk(KERN_INFO, sdkp,
3699 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3700 			  msg, host_byte(result), driver_byte(result));
3701 }
3702 
3703