xref: /openbmc/linux/drivers/scsi/sd.c (revision a6c76bb0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *      sd.c Copyright (C) 1992 Drew Eckhardt
4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5  *
6  *      Linux scsi disk driver
7  *              Initial versions: Drew Eckhardt
8  *              Subsequent revisions: Eric Youngdale
9  *	Modification history:
10  *       - Drew Eckhardt <drew@colorado.edu> original
11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12  *         outstanding request, and other enhancements.
13  *         Support loadable low-level scsi drivers.
14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15  *         eight major numbers.
16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18  *	   sd_init and cleanups.
19  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
20  *	   not being read in sd_open. Fix problem where removable media
21  *	   could be ejected after sd_open.
22  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25  *	   Support 32k/1M disks.
26  *
27  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
28  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
32  *	Note: when the logging level is set by the user, it must be greater
33  *	than the level indicated above to trigger output.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
57 #include <linux/pr.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
61 
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
71 
72 #include "sd.h"
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
75 
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
79 
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100 
101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
102 #define SD_MINORS	16
103 #else
104 #define SD_MINORS	0
105 #endif
106 
107 static void sd_config_discard(struct scsi_disk *, unsigned int);
108 static void sd_config_write_same(struct scsi_disk *);
109 static int  sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static int  sd_probe(struct device *);
112 static int  sd_remove(struct device *);
113 static void sd_shutdown(struct device *);
114 static int sd_suspend_system(struct device *);
115 static int sd_suspend_runtime(struct device *);
116 static int sd_resume(struct device *);
117 static void sd_rescan(struct device *);
118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120 static int sd_done(struct scsi_cmnd *);
121 static void sd_eh_reset(struct scsi_cmnd *);
122 static int sd_eh_action(struct scsi_cmnd *, int);
123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124 static void scsi_disk_release(struct device *cdev);
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 static mempool_t *sd_page_pool;
136 
137 static const char *sd_cache_types[] = {
138 	"write through", "none", "write back",
139 	"write back, no read (daft)"
140 };
141 
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
143 {
144 	bool wc = false, fua = false;
145 
146 	if (sdkp->WCE) {
147 		wc = true;
148 		if (sdkp->DPOFUA)
149 			fua = true;
150 	}
151 
152 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
153 }
154 
155 static ssize_t
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 		 const char *buf, size_t count)
158 {
159 	int ct, rcd, wce, sp;
160 	struct scsi_disk *sdkp = to_scsi_disk(dev);
161 	struct scsi_device *sdp = sdkp->device;
162 	char buffer[64];
163 	char *buffer_data;
164 	struct scsi_mode_data data;
165 	struct scsi_sense_hdr sshdr;
166 	static const char temp[] = "temporary ";
167 	int len;
168 
169 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 		/* no cache control on RBC devices; theoretically they
171 		 * can do it, but there's probably so many exceptions
172 		 * it's not worth the risk */
173 		return -EINVAL;
174 
175 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 		buf += sizeof(temp) - 1;
177 		sdkp->cache_override = 1;
178 	} else {
179 		sdkp->cache_override = 0;
180 	}
181 
182 	ct = sysfs_match_string(sd_cache_types, buf);
183 	if (ct < 0)
184 		return -EINVAL;
185 
186 	rcd = ct & 0x01 ? 1 : 0;
187 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
188 
189 	if (sdkp->cache_override) {
190 		sdkp->WCE = wce;
191 		sdkp->RCD = rcd;
192 		sd_set_flush_flag(sdkp);
193 		return count;
194 	}
195 
196 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 			    sdkp->max_retries, &data, NULL))
198 		return -EINVAL;
199 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 		  data.block_descriptor_length);
201 	buffer_data = buffer + data.header_length +
202 		data.block_descriptor_length;
203 	buffer_data[2] &= ~0x05;
204 	buffer_data[2] |= wce << 2 | rcd;
205 	sp = buffer_data[0] & 0x80 ? 1 : 0;
206 	buffer_data[0] &= ~0x80;
207 
208 	/*
209 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
210 	 * received mode parameter buffer before doing MODE SELECT.
211 	 */
212 	data.device_specific = 0;
213 
214 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
215 			     sdkp->max_retries, &data, &sshdr)) {
216 		if (scsi_sense_valid(&sshdr))
217 			sd_print_sense_hdr(sdkp, &sshdr);
218 		return -EINVAL;
219 	}
220 	sd_revalidate_disk(sdkp->disk);
221 	return count;
222 }
223 
224 static ssize_t
225 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
226 		       char *buf)
227 {
228 	struct scsi_disk *sdkp = to_scsi_disk(dev);
229 	struct scsi_device *sdp = sdkp->device;
230 
231 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
232 }
233 
234 static ssize_t
235 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
236 			const char *buf, size_t count)
237 {
238 	struct scsi_disk *sdkp = to_scsi_disk(dev);
239 	struct scsi_device *sdp = sdkp->device;
240 	bool v;
241 
242 	if (!capable(CAP_SYS_ADMIN))
243 		return -EACCES;
244 
245 	if (kstrtobool(buf, &v))
246 		return -EINVAL;
247 
248 	sdp->manage_start_stop = v;
249 
250 	return count;
251 }
252 static DEVICE_ATTR_RW(manage_start_stop);
253 
254 static ssize_t
255 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
256 {
257 	struct scsi_disk *sdkp = to_scsi_disk(dev);
258 
259 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
260 }
261 
262 static ssize_t
263 allow_restart_store(struct device *dev, struct device_attribute *attr,
264 		    const char *buf, size_t count)
265 {
266 	bool v;
267 	struct scsi_disk *sdkp = to_scsi_disk(dev);
268 	struct scsi_device *sdp = sdkp->device;
269 
270 	if (!capable(CAP_SYS_ADMIN))
271 		return -EACCES;
272 
273 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
274 		return -EINVAL;
275 
276 	if (kstrtobool(buf, &v))
277 		return -EINVAL;
278 
279 	sdp->allow_restart = v;
280 
281 	return count;
282 }
283 static DEVICE_ATTR_RW(allow_restart);
284 
285 static ssize_t
286 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
287 {
288 	struct scsi_disk *sdkp = to_scsi_disk(dev);
289 	int ct = sdkp->RCD + 2*sdkp->WCE;
290 
291 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
292 }
293 static DEVICE_ATTR_RW(cache_type);
294 
295 static ssize_t
296 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
297 {
298 	struct scsi_disk *sdkp = to_scsi_disk(dev);
299 
300 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
301 }
302 static DEVICE_ATTR_RO(FUA);
303 
304 static ssize_t
305 protection_type_show(struct device *dev, struct device_attribute *attr,
306 		     char *buf)
307 {
308 	struct scsi_disk *sdkp = to_scsi_disk(dev);
309 
310 	return sprintf(buf, "%u\n", sdkp->protection_type);
311 }
312 
313 static ssize_t
314 protection_type_store(struct device *dev, struct device_attribute *attr,
315 		      const char *buf, size_t count)
316 {
317 	struct scsi_disk *sdkp = to_scsi_disk(dev);
318 	unsigned int val;
319 	int err;
320 
321 	if (!capable(CAP_SYS_ADMIN))
322 		return -EACCES;
323 
324 	err = kstrtouint(buf, 10, &val);
325 
326 	if (err)
327 		return err;
328 
329 	if (val <= T10_PI_TYPE3_PROTECTION)
330 		sdkp->protection_type = val;
331 
332 	return count;
333 }
334 static DEVICE_ATTR_RW(protection_type);
335 
336 static ssize_t
337 protection_mode_show(struct device *dev, struct device_attribute *attr,
338 		     char *buf)
339 {
340 	struct scsi_disk *sdkp = to_scsi_disk(dev);
341 	struct scsi_device *sdp = sdkp->device;
342 	unsigned int dif, dix;
343 
344 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
345 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
346 
347 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
348 		dif = 0;
349 		dix = 1;
350 	}
351 
352 	if (!dif && !dix)
353 		return sprintf(buf, "none\n");
354 
355 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
356 }
357 static DEVICE_ATTR_RO(protection_mode);
358 
359 static ssize_t
360 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
361 {
362 	struct scsi_disk *sdkp = to_scsi_disk(dev);
363 
364 	return sprintf(buf, "%u\n", sdkp->ATO);
365 }
366 static DEVICE_ATTR_RO(app_tag_own);
367 
368 static ssize_t
369 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
370 		       char *buf)
371 {
372 	struct scsi_disk *sdkp = to_scsi_disk(dev);
373 
374 	return sprintf(buf, "%u\n", sdkp->lbpme);
375 }
376 static DEVICE_ATTR_RO(thin_provisioning);
377 
378 /* sysfs_match_string() requires dense arrays */
379 static const char *lbp_mode[] = {
380 	[SD_LBP_FULL]		= "full",
381 	[SD_LBP_UNMAP]		= "unmap",
382 	[SD_LBP_WS16]		= "writesame_16",
383 	[SD_LBP_WS10]		= "writesame_10",
384 	[SD_LBP_ZERO]		= "writesame_zero",
385 	[SD_LBP_DISABLE]	= "disabled",
386 };
387 
388 static ssize_t
389 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
390 		       char *buf)
391 {
392 	struct scsi_disk *sdkp = to_scsi_disk(dev);
393 
394 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
395 }
396 
397 static ssize_t
398 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
399 			const char *buf, size_t count)
400 {
401 	struct scsi_disk *sdkp = to_scsi_disk(dev);
402 	struct scsi_device *sdp = sdkp->device;
403 	int mode;
404 
405 	if (!capable(CAP_SYS_ADMIN))
406 		return -EACCES;
407 
408 	if (sd_is_zoned(sdkp)) {
409 		sd_config_discard(sdkp, SD_LBP_DISABLE);
410 		return count;
411 	}
412 
413 	if (sdp->type != TYPE_DISK)
414 		return -EINVAL;
415 
416 	mode = sysfs_match_string(lbp_mode, buf);
417 	if (mode < 0)
418 		return -EINVAL;
419 
420 	sd_config_discard(sdkp, mode);
421 
422 	return count;
423 }
424 static DEVICE_ATTR_RW(provisioning_mode);
425 
426 /* sysfs_match_string() requires dense arrays */
427 static const char *zeroing_mode[] = {
428 	[SD_ZERO_WRITE]		= "write",
429 	[SD_ZERO_WS]		= "writesame",
430 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
431 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
432 };
433 
434 static ssize_t
435 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
436 		  char *buf)
437 {
438 	struct scsi_disk *sdkp = to_scsi_disk(dev);
439 
440 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
441 }
442 
443 static ssize_t
444 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
445 		   const char *buf, size_t count)
446 {
447 	struct scsi_disk *sdkp = to_scsi_disk(dev);
448 	int mode;
449 
450 	if (!capable(CAP_SYS_ADMIN))
451 		return -EACCES;
452 
453 	mode = sysfs_match_string(zeroing_mode, buf);
454 	if (mode < 0)
455 		return -EINVAL;
456 
457 	sdkp->zeroing_mode = mode;
458 
459 	return count;
460 }
461 static DEVICE_ATTR_RW(zeroing_mode);
462 
463 static ssize_t
464 max_medium_access_timeouts_show(struct device *dev,
465 				struct device_attribute *attr, char *buf)
466 {
467 	struct scsi_disk *sdkp = to_scsi_disk(dev);
468 
469 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
470 }
471 
472 static ssize_t
473 max_medium_access_timeouts_store(struct device *dev,
474 				 struct device_attribute *attr, const char *buf,
475 				 size_t count)
476 {
477 	struct scsi_disk *sdkp = to_scsi_disk(dev);
478 	int err;
479 
480 	if (!capable(CAP_SYS_ADMIN))
481 		return -EACCES;
482 
483 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
484 
485 	return err ? err : count;
486 }
487 static DEVICE_ATTR_RW(max_medium_access_timeouts);
488 
489 static ssize_t
490 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
491 			   char *buf)
492 {
493 	struct scsi_disk *sdkp = to_scsi_disk(dev);
494 
495 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
496 }
497 
498 static ssize_t
499 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
500 			    const char *buf, size_t count)
501 {
502 	struct scsi_disk *sdkp = to_scsi_disk(dev);
503 	struct scsi_device *sdp = sdkp->device;
504 	unsigned long max;
505 	int err;
506 
507 	if (!capable(CAP_SYS_ADMIN))
508 		return -EACCES;
509 
510 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
511 		return -EINVAL;
512 
513 	err = kstrtoul(buf, 10, &max);
514 
515 	if (err)
516 		return err;
517 
518 	if (max == 0)
519 		sdp->no_write_same = 1;
520 	else if (max <= SD_MAX_WS16_BLOCKS) {
521 		sdp->no_write_same = 0;
522 		sdkp->max_ws_blocks = max;
523 	}
524 
525 	sd_config_write_same(sdkp);
526 
527 	return count;
528 }
529 static DEVICE_ATTR_RW(max_write_same_blocks);
530 
531 static ssize_t
532 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
533 {
534 	struct scsi_disk *sdkp = to_scsi_disk(dev);
535 
536 	if (sdkp->device->type == TYPE_ZBC)
537 		return sprintf(buf, "host-managed\n");
538 	if (sdkp->zoned == 1)
539 		return sprintf(buf, "host-aware\n");
540 	if (sdkp->zoned == 2)
541 		return sprintf(buf, "drive-managed\n");
542 	return sprintf(buf, "none\n");
543 }
544 static DEVICE_ATTR_RO(zoned_cap);
545 
546 static ssize_t
547 max_retries_store(struct device *dev, struct device_attribute *attr,
548 		  const char *buf, size_t count)
549 {
550 	struct scsi_disk *sdkp = to_scsi_disk(dev);
551 	struct scsi_device *sdev = sdkp->device;
552 	int retries, err;
553 
554 	err = kstrtoint(buf, 10, &retries);
555 	if (err)
556 		return err;
557 
558 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
559 		sdkp->max_retries = retries;
560 		return count;
561 	}
562 
563 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
564 		    SD_MAX_RETRIES);
565 	return -EINVAL;
566 }
567 
568 static ssize_t
569 max_retries_show(struct device *dev, struct device_attribute *attr,
570 		 char *buf)
571 {
572 	struct scsi_disk *sdkp = to_scsi_disk(dev);
573 
574 	return sprintf(buf, "%d\n", sdkp->max_retries);
575 }
576 
577 static DEVICE_ATTR_RW(max_retries);
578 
579 static struct attribute *sd_disk_attrs[] = {
580 	&dev_attr_cache_type.attr,
581 	&dev_attr_FUA.attr,
582 	&dev_attr_allow_restart.attr,
583 	&dev_attr_manage_start_stop.attr,
584 	&dev_attr_protection_type.attr,
585 	&dev_attr_protection_mode.attr,
586 	&dev_attr_app_tag_own.attr,
587 	&dev_attr_thin_provisioning.attr,
588 	&dev_attr_provisioning_mode.attr,
589 	&dev_attr_zeroing_mode.attr,
590 	&dev_attr_max_write_same_blocks.attr,
591 	&dev_attr_max_medium_access_timeouts.attr,
592 	&dev_attr_zoned_cap.attr,
593 	&dev_attr_max_retries.attr,
594 	NULL,
595 };
596 ATTRIBUTE_GROUPS(sd_disk);
597 
598 static struct class sd_disk_class = {
599 	.name		= "scsi_disk",
600 	.owner		= THIS_MODULE,
601 	.dev_release	= scsi_disk_release,
602 	.dev_groups	= sd_disk_groups,
603 };
604 
605 static const struct dev_pm_ops sd_pm_ops = {
606 	.suspend		= sd_suspend_system,
607 	.resume			= sd_resume,
608 	.poweroff		= sd_suspend_system,
609 	.restore		= sd_resume,
610 	.runtime_suspend	= sd_suspend_runtime,
611 	.runtime_resume		= sd_resume,
612 };
613 
614 static struct scsi_driver sd_template = {
615 	.gendrv = {
616 		.name		= "sd",
617 		.owner		= THIS_MODULE,
618 		.probe		= sd_probe,
619 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
620 		.remove		= sd_remove,
621 		.shutdown	= sd_shutdown,
622 		.pm		= &sd_pm_ops,
623 	},
624 	.rescan			= sd_rescan,
625 	.init_command		= sd_init_command,
626 	.uninit_command		= sd_uninit_command,
627 	.done			= sd_done,
628 	.eh_action		= sd_eh_action,
629 	.eh_reset		= sd_eh_reset,
630 };
631 
632 /*
633  * Don't request a new module, as that could deadlock in multipath
634  * environment.
635  */
636 static void sd_default_probe(dev_t devt)
637 {
638 }
639 
640 /*
641  * Device no to disk mapping:
642  *
643  *       major         disc2     disc  p1
644  *   |............|.............|....|....| <- dev_t
645  *    31        20 19          8 7  4 3  0
646  *
647  * Inside a major, we have 16k disks, however mapped non-
648  * contiguously. The first 16 disks are for major0, the next
649  * ones with major1, ... Disk 256 is for major0 again, disk 272
650  * for major1, ...
651  * As we stay compatible with our numbering scheme, we can reuse
652  * the well-know SCSI majors 8, 65--71, 136--143.
653  */
654 static int sd_major(int major_idx)
655 {
656 	switch (major_idx) {
657 	case 0:
658 		return SCSI_DISK0_MAJOR;
659 	case 1 ... 7:
660 		return SCSI_DISK1_MAJOR + major_idx - 1;
661 	case 8 ... 15:
662 		return SCSI_DISK8_MAJOR + major_idx - 8;
663 	default:
664 		BUG();
665 		return 0;	/* shut up gcc */
666 	}
667 }
668 
669 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
670 {
671 	struct scsi_disk *sdkp = NULL;
672 
673 	mutex_lock(&sd_ref_mutex);
674 
675 	if (disk->private_data) {
676 		sdkp = scsi_disk(disk);
677 		if (scsi_device_get(sdkp->device) == 0)
678 			get_device(&sdkp->dev);
679 		else
680 			sdkp = NULL;
681 	}
682 	mutex_unlock(&sd_ref_mutex);
683 	return sdkp;
684 }
685 
686 static void scsi_disk_put(struct scsi_disk *sdkp)
687 {
688 	struct scsi_device *sdev = sdkp->device;
689 
690 	mutex_lock(&sd_ref_mutex);
691 	put_device(&sdkp->dev);
692 	scsi_device_put(sdev);
693 	mutex_unlock(&sd_ref_mutex);
694 }
695 
696 #ifdef CONFIG_BLK_SED_OPAL
697 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
698 		size_t len, bool send)
699 {
700 	struct scsi_disk *sdkp = data;
701 	struct scsi_device *sdev = sdkp->device;
702 	u8 cdb[12] = { 0, };
703 	int ret;
704 
705 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
706 	cdb[1] = secp;
707 	put_unaligned_be16(spsp, &cdb[2]);
708 	put_unaligned_be32(len, &cdb[6]);
709 
710 	ret = scsi_execute_req(sdev, cdb,
711 			send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
712 			buffer, len, NULL, SD_TIMEOUT, sdkp->max_retries, NULL);
713 	return ret <= 0 ? ret : -EIO;
714 }
715 #endif /* CONFIG_BLK_SED_OPAL */
716 
717 /*
718  * Look up the DIX operation based on whether the command is read or
719  * write and whether dix and dif are enabled.
720  */
721 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
722 {
723 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
724 	static const unsigned int ops[] = {	/* wrt dix dif */
725 		SCSI_PROT_NORMAL,		/*  0	0   0  */
726 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
727 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
728 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
729 		SCSI_PROT_NORMAL,		/*  1	0   0  */
730 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
731 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
732 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
733 	};
734 
735 	return ops[write << 2 | dix << 1 | dif];
736 }
737 
738 /*
739  * Returns a mask of the protection flags that are valid for a given DIX
740  * operation.
741  */
742 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
743 {
744 	static const unsigned int flag_mask[] = {
745 		[SCSI_PROT_NORMAL]		= 0,
746 
747 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
748 						  SCSI_PROT_GUARD_CHECK |
749 						  SCSI_PROT_REF_CHECK |
750 						  SCSI_PROT_REF_INCREMENT,
751 
752 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
753 						  SCSI_PROT_IP_CHECKSUM,
754 
755 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
756 						  SCSI_PROT_GUARD_CHECK |
757 						  SCSI_PROT_REF_CHECK |
758 						  SCSI_PROT_REF_INCREMENT |
759 						  SCSI_PROT_IP_CHECKSUM,
760 
761 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
762 						  SCSI_PROT_REF_INCREMENT,
763 
764 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
765 						  SCSI_PROT_REF_CHECK |
766 						  SCSI_PROT_REF_INCREMENT |
767 						  SCSI_PROT_IP_CHECKSUM,
768 
769 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
770 						  SCSI_PROT_GUARD_CHECK |
771 						  SCSI_PROT_REF_CHECK |
772 						  SCSI_PROT_REF_INCREMENT |
773 						  SCSI_PROT_IP_CHECKSUM,
774 	};
775 
776 	return flag_mask[prot_op];
777 }
778 
779 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
780 					   unsigned int dix, unsigned int dif)
781 {
782 	struct bio *bio = scmd->request->bio;
783 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
784 	unsigned int protect = 0;
785 
786 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
787 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
788 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
789 
790 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
791 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
792 	}
793 
794 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
795 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
796 
797 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
798 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
799 	}
800 
801 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
802 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
803 
804 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
805 			protect = 3 << 5;	/* Disable target PI checking */
806 		else
807 			protect = 1 << 5;	/* Enable target PI checking */
808 	}
809 
810 	scsi_set_prot_op(scmd, prot_op);
811 	scsi_set_prot_type(scmd, dif);
812 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
813 
814 	return protect;
815 }
816 
817 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
818 {
819 	struct request_queue *q = sdkp->disk->queue;
820 	unsigned int logical_block_size = sdkp->device->sector_size;
821 	unsigned int max_blocks = 0;
822 
823 	q->limits.discard_alignment =
824 		sdkp->unmap_alignment * logical_block_size;
825 	q->limits.discard_granularity =
826 		max(sdkp->physical_block_size,
827 		    sdkp->unmap_granularity * logical_block_size);
828 	sdkp->provisioning_mode = mode;
829 
830 	switch (mode) {
831 
832 	case SD_LBP_FULL:
833 	case SD_LBP_DISABLE:
834 		blk_queue_max_discard_sectors(q, 0);
835 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
836 		return;
837 
838 	case SD_LBP_UNMAP:
839 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
840 					  (u32)SD_MAX_WS16_BLOCKS);
841 		break;
842 
843 	case SD_LBP_WS16:
844 		if (sdkp->device->unmap_limit_for_ws)
845 			max_blocks = sdkp->max_unmap_blocks;
846 		else
847 			max_blocks = sdkp->max_ws_blocks;
848 
849 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
850 		break;
851 
852 	case SD_LBP_WS10:
853 		if (sdkp->device->unmap_limit_for_ws)
854 			max_blocks = sdkp->max_unmap_blocks;
855 		else
856 			max_blocks = sdkp->max_ws_blocks;
857 
858 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
859 		break;
860 
861 	case SD_LBP_ZERO:
862 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
863 					  (u32)SD_MAX_WS10_BLOCKS);
864 		break;
865 	}
866 
867 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
868 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
869 }
870 
871 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
872 {
873 	struct scsi_device *sdp = cmd->device;
874 	struct request *rq = cmd->request;
875 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
876 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
877 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
878 	unsigned int data_len = 24;
879 	char *buf;
880 
881 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 	if (!rq->special_vec.bv_page)
883 		return BLK_STS_RESOURCE;
884 	clear_highpage(rq->special_vec.bv_page);
885 	rq->special_vec.bv_offset = 0;
886 	rq->special_vec.bv_len = data_len;
887 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
888 
889 	cmd->cmd_len = 10;
890 	cmd->cmnd[0] = UNMAP;
891 	cmd->cmnd[8] = 24;
892 
893 	buf = page_address(rq->special_vec.bv_page);
894 	put_unaligned_be16(6 + 16, &buf[0]);
895 	put_unaligned_be16(16, &buf[2]);
896 	put_unaligned_be64(lba, &buf[8]);
897 	put_unaligned_be32(nr_blocks, &buf[16]);
898 
899 	cmd->allowed = sdkp->max_retries;
900 	cmd->transfersize = data_len;
901 	rq->timeout = SD_TIMEOUT;
902 
903 	return scsi_alloc_sgtables(cmd);
904 }
905 
906 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
907 		bool unmap)
908 {
909 	struct scsi_device *sdp = cmd->device;
910 	struct request *rq = cmd->request;
911 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
912 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
913 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
914 	u32 data_len = sdp->sector_size;
915 
916 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
917 	if (!rq->special_vec.bv_page)
918 		return BLK_STS_RESOURCE;
919 	clear_highpage(rq->special_vec.bv_page);
920 	rq->special_vec.bv_offset = 0;
921 	rq->special_vec.bv_len = data_len;
922 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
923 
924 	cmd->cmd_len = 16;
925 	cmd->cmnd[0] = WRITE_SAME_16;
926 	if (unmap)
927 		cmd->cmnd[1] = 0x8; /* UNMAP */
928 	put_unaligned_be64(lba, &cmd->cmnd[2]);
929 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
930 
931 	cmd->allowed = sdkp->max_retries;
932 	cmd->transfersize = data_len;
933 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
934 
935 	return scsi_alloc_sgtables(cmd);
936 }
937 
938 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
939 		bool unmap)
940 {
941 	struct scsi_device *sdp = cmd->device;
942 	struct request *rq = cmd->request;
943 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
944 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
945 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
946 	u32 data_len = sdp->sector_size;
947 
948 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
949 	if (!rq->special_vec.bv_page)
950 		return BLK_STS_RESOURCE;
951 	clear_highpage(rq->special_vec.bv_page);
952 	rq->special_vec.bv_offset = 0;
953 	rq->special_vec.bv_len = data_len;
954 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
955 
956 	cmd->cmd_len = 10;
957 	cmd->cmnd[0] = WRITE_SAME;
958 	if (unmap)
959 		cmd->cmnd[1] = 0x8; /* UNMAP */
960 	put_unaligned_be32(lba, &cmd->cmnd[2]);
961 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
962 
963 	cmd->allowed = sdkp->max_retries;
964 	cmd->transfersize = data_len;
965 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
966 
967 	return scsi_alloc_sgtables(cmd);
968 }
969 
970 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
971 {
972 	struct request *rq = cmd->request;
973 	struct scsi_device *sdp = cmd->device;
974 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
975 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
976 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
977 
978 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
979 		switch (sdkp->zeroing_mode) {
980 		case SD_ZERO_WS16_UNMAP:
981 			return sd_setup_write_same16_cmnd(cmd, true);
982 		case SD_ZERO_WS10_UNMAP:
983 			return sd_setup_write_same10_cmnd(cmd, true);
984 		}
985 	}
986 
987 	if (sdp->no_write_same) {
988 		rq->rq_flags |= RQF_QUIET;
989 		return BLK_STS_TARGET;
990 	}
991 
992 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
993 		return sd_setup_write_same16_cmnd(cmd, false);
994 
995 	return sd_setup_write_same10_cmnd(cmd, false);
996 }
997 
998 static void sd_config_write_same(struct scsi_disk *sdkp)
999 {
1000 	struct request_queue *q = sdkp->disk->queue;
1001 	unsigned int logical_block_size = sdkp->device->sector_size;
1002 
1003 	if (sdkp->device->no_write_same) {
1004 		sdkp->max_ws_blocks = 0;
1005 		goto out;
1006 	}
1007 
1008 	/* Some devices can not handle block counts above 0xffff despite
1009 	 * supporting WRITE SAME(16). Consequently we default to 64k
1010 	 * blocks per I/O unless the device explicitly advertises a
1011 	 * bigger limit.
1012 	 */
1013 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1014 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1015 						   (u32)SD_MAX_WS16_BLOCKS);
1016 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1017 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1018 						   (u32)SD_MAX_WS10_BLOCKS);
1019 	else {
1020 		sdkp->device->no_write_same = 1;
1021 		sdkp->max_ws_blocks = 0;
1022 	}
1023 
1024 	if (sdkp->lbprz && sdkp->lbpws)
1025 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1026 	else if (sdkp->lbprz && sdkp->lbpws10)
1027 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1028 	else if (sdkp->max_ws_blocks)
1029 		sdkp->zeroing_mode = SD_ZERO_WS;
1030 	else
1031 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1032 
1033 	if (sdkp->max_ws_blocks &&
1034 	    sdkp->physical_block_size > logical_block_size) {
1035 		/*
1036 		 * Reporting a maximum number of blocks that is not aligned
1037 		 * on the device physical size would cause a large write same
1038 		 * request to be split into physically unaligned chunks by
1039 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1040 		 * even if the caller of these functions took care to align the
1041 		 * large request. So make sure the maximum reported is aligned
1042 		 * to the device physical block size. This is only an optional
1043 		 * optimization for regular disks, but this is mandatory to
1044 		 * avoid failure of large write same requests directed at
1045 		 * sequential write required zones of host-managed ZBC disks.
1046 		 */
1047 		sdkp->max_ws_blocks =
1048 			round_down(sdkp->max_ws_blocks,
1049 				   bytes_to_logical(sdkp->device,
1050 						    sdkp->physical_block_size));
1051 	}
1052 
1053 out:
1054 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1055 					 (logical_block_size >> 9));
1056 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1057 					 (logical_block_size >> 9));
1058 }
1059 
1060 /**
1061  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1062  * @cmd: command to prepare
1063  *
1064  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1065  * the preference indicated by the target device.
1066  **/
1067 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1068 {
1069 	struct request *rq = cmd->request;
1070 	struct scsi_device *sdp = cmd->device;
1071 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1072 	struct bio *bio = rq->bio;
1073 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1074 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1075 	blk_status_t ret;
1076 
1077 	if (sdkp->device->no_write_same)
1078 		return BLK_STS_TARGET;
1079 
1080 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1081 
1082 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1083 
1084 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1085 		cmd->cmd_len = 16;
1086 		cmd->cmnd[0] = WRITE_SAME_16;
1087 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1088 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1089 	} else {
1090 		cmd->cmd_len = 10;
1091 		cmd->cmnd[0] = WRITE_SAME;
1092 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1093 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1094 	}
1095 
1096 	cmd->transfersize = sdp->sector_size;
1097 	cmd->allowed = sdkp->max_retries;
1098 
1099 	/*
1100 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1101 	 * different from the amount of data actually written to the target.
1102 	 *
1103 	 * We set up __data_len to the amount of data transferred via the
1104 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1105 	 * to transfer a single sector of data first, but then reset it to
1106 	 * the amount of data to be written right after so that the I/O path
1107 	 * knows how much to actually write.
1108 	 */
1109 	rq->__data_len = sdp->sector_size;
1110 	ret = scsi_alloc_sgtables(cmd);
1111 	rq->__data_len = blk_rq_bytes(rq);
1112 
1113 	return ret;
1114 }
1115 
1116 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1117 {
1118 	struct request *rq = cmd->request;
1119 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1120 
1121 	/* flush requests don't perform I/O, zero the S/G table */
1122 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1123 
1124 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1125 	cmd->cmd_len = 10;
1126 	cmd->transfersize = 0;
1127 	cmd->allowed = sdkp->max_retries;
1128 
1129 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1130 	return BLK_STS_OK;
1131 }
1132 
1133 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1134 				       sector_t lba, unsigned int nr_blocks,
1135 				       unsigned char flags)
1136 {
1137 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1138 	if (unlikely(cmd->cmnd == NULL))
1139 		return BLK_STS_RESOURCE;
1140 
1141 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1142 	memset(cmd->cmnd, 0, cmd->cmd_len);
1143 
1144 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1145 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1146 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1147 	cmd->cmnd[10] = flags;
1148 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1149 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1150 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1151 
1152 	return BLK_STS_OK;
1153 }
1154 
1155 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1156 				       sector_t lba, unsigned int nr_blocks,
1157 				       unsigned char flags)
1158 {
1159 	cmd->cmd_len  = 16;
1160 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1161 	cmd->cmnd[1]  = flags;
1162 	cmd->cmnd[14] = 0;
1163 	cmd->cmnd[15] = 0;
1164 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1165 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1166 
1167 	return BLK_STS_OK;
1168 }
1169 
1170 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1171 				       sector_t lba, unsigned int nr_blocks,
1172 				       unsigned char flags)
1173 {
1174 	cmd->cmd_len = 10;
1175 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1176 	cmd->cmnd[1] = flags;
1177 	cmd->cmnd[6] = 0;
1178 	cmd->cmnd[9] = 0;
1179 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1180 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1181 
1182 	return BLK_STS_OK;
1183 }
1184 
1185 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1186 				      sector_t lba, unsigned int nr_blocks,
1187 				      unsigned char flags)
1188 {
1189 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1190 	if (WARN_ON_ONCE(nr_blocks == 0))
1191 		return BLK_STS_IOERR;
1192 
1193 	if (unlikely(flags & 0x8)) {
1194 		/*
1195 		 * This happens only if this drive failed 10byte rw
1196 		 * command with ILLEGAL_REQUEST during operation and
1197 		 * thus turned off use_10_for_rw.
1198 		 */
1199 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1200 		return BLK_STS_IOERR;
1201 	}
1202 
1203 	cmd->cmd_len = 6;
1204 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1205 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1206 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1207 	cmd->cmnd[3] = lba & 0xff;
1208 	cmd->cmnd[4] = nr_blocks;
1209 	cmd->cmnd[5] = 0;
1210 
1211 	return BLK_STS_OK;
1212 }
1213 
1214 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1215 {
1216 	struct request *rq = cmd->request;
1217 	struct scsi_device *sdp = cmd->device;
1218 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1219 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1220 	sector_t threshold;
1221 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1222 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1223 	bool write = rq_data_dir(rq) == WRITE;
1224 	unsigned char protect, fua;
1225 	blk_status_t ret;
1226 	unsigned int dif;
1227 	bool dix;
1228 
1229 	ret = scsi_alloc_sgtables(cmd);
1230 	if (ret != BLK_STS_OK)
1231 		return ret;
1232 
1233 	ret = BLK_STS_IOERR;
1234 	if (!scsi_device_online(sdp) || sdp->changed) {
1235 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1236 		goto fail;
1237 	}
1238 
1239 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1240 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1241 		goto fail;
1242 	}
1243 
1244 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1245 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1246 		goto fail;
1247 	}
1248 
1249 	/*
1250 	 * Some SD card readers can't handle accesses which touch the
1251 	 * last one or two logical blocks. Split accesses as needed.
1252 	 */
1253 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1254 
1255 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1256 		if (lba < threshold) {
1257 			/* Access up to the threshold but not beyond */
1258 			nr_blocks = threshold - lba;
1259 		} else {
1260 			/* Access only a single logical block */
1261 			nr_blocks = 1;
1262 		}
1263 	}
1264 
1265 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1266 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1267 		if (ret)
1268 			goto fail;
1269 	}
1270 
1271 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1272 	dix = scsi_prot_sg_count(cmd);
1273 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1274 
1275 	if (dif || dix)
1276 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1277 	else
1278 		protect = 0;
1279 
1280 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1281 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1282 					 protect | fua);
1283 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1284 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1285 					 protect | fua);
1286 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1287 		   sdp->use_10_for_rw || protect) {
1288 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1289 					 protect | fua);
1290 	} else {
1291 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1292 					protect | fua);
1293 	}
1294 
1295 	if (unlikely(ret != BLK_STS_OK))
1296 		goto fail;
1297 
1298 	/*
1299 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1300 	 * host adapter, it's safe to assume that we can at least transfer
1301 	 * this many bytes between each connect / disconnect.
1302 	 */
1303 	cmd->transfersize = sdp->sector_size;
1304 	cmd->underflow = nr_blocks << 9;
1305 	cmd->allowed = sdkp->max_retries;
1306 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1307 
1308 	SCSI_LOG_HLQUEUE(1,
1309 			 scmd_printk(KERN_INFO, cmd,
1310 				     "%s: block=%llu, count=%d\n", __func__,
1311 				     (unsigned long long)blk_rq_pos(rq),
1312 				     blk_rq_sectors(rq)));
1313 	SCSI_LOG_HLQUEUE(2,
1314 			 scmd_printk(KERN_INFO, cmd,
1315 				     "%s %d/%u 512 byte blocks.\n",
1316 				     write ? "writing" : "reading", nr_blocks,
1317 				     blk_rq_sectors(rq)));
1318 
1319 	/*
1320 	 * This indicates that the command is ready from our end to be queued.
1321 	 */
1322 	return BLK_STS_OK;
1323 fail:
1324 	scsi_free_sgtables(cmd);
1325 	return ret;
1326 }
1327 
1328 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1329 {
1330 	struct request *rq = cmd->request;
1331 
1332 	switch (req_op(rq)) {
1333 	case REQ_OP_DISCARD:
1334 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1335 		case SD_LBP_UNMAP:
1336 			return sd_setup_unmap_cmnd(cmd);
1337 		case SD_LBP_WS16:
1338 			return sd_setup_write_same16_cmnd(cmd, true);
1339 		case SD_LBP_WS10:
1340 			return sd_setup_write_same10_cmnd(cmd, true);
1341 		case SD_LBP_ZERO:
1342 			return sd_setup_write_same10_cmnd(cmd, false);
1343 		default:
1344 			return BLK_STS_TARGET;
1345 		}
1346 	case REQ_OP_WRITE_ZEROES:
1347 		return sd_setup_write_zeroes_cmnd(cmd);
1348 	case REQ_OP_WRITE_SAME:
1349 		return sd_setup_write_same_cmnd(cmd);
1350 	case REQ_OP_FLUSH:
1351 		return sd_setup_flush_cmnd(cmd);
1352 	case REQ_OP_READ:
1353 	case REQ_OP_WRITE:
1354 	case REQ_OP_ZONE_APPEND:
1355 		return sd_setup_read_write_cmnd(cmd);
1356 	case REQ_OP_ZONE_RESET:
1357 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358 						   false);
1359 	case REQ_OP_ZONE_RESET_ALL:
1360 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1361 						   true);
1362 	case REQ_OP_ZONE_OPEN:
1363 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1364 	case REQ_OP_ZONE_CLOSE:
1365 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1366 	case REQ_OP_ZONE_FINISH:
1367 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1368 	default:
1369 		WARN_ON_ONCE(1);
1370 		return BLK_STS_NOTSUPP;
1371 	}
1372 }
1373 
1374 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1375 {
1376 	struct request *rq = SCpnt->request;
1377 	u8 *cmnd;
1378 
1379 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1380 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1381 
1382 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1383 		cmnd = SCpnt->cmnd;
1384 		SCpnt->cmnd = NULL;
1385 		SCpnt->cmd_len = 0;
1386 		mempool_free(cmnd, sd_cdb_pool);
1387 	}
1388 }
1389 
1390 /**
1391  *	sd_open - open a scsi disk device
1392  *	@bdev: Block device of the scsi disk to open
1393  *	@mode: FMODE_* mask
1394  *
1395  *	Returns 0 if successful. Returns a negated errno value in case
1396  *	of error.
1397  *
1398  *	Note: This can be called from a user context (e.g. fsck(1) )
1399  *	or from within the kernel (e.g. as a result of a mount(1) ).
1400  *	In the latter case @inode and @filp carry an abridged amount
1401  *	of information as noted above.
1402  *
1403  *	Locking: called with bdev->bd_mutex held.
1404  **/
1405 static int sd_open(struct block_device *bdev, fmode_t mode)
1406 {
1407 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1408 	struct scsi_device *sdev;
1409 	int retval;
1410 
1411 	if (!sdkp)
1412 		return -ENXIO;
1413 
1414 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1415 
1416 	sdev = sdkp->device;
1417 
1418 	/*
1419 	 * If the device is in error recovery, wait until it is done.
1420 	 * If the device is offline, then disallow any access to it.
1421 	 */
1422 	retval = -ENXIO;
1423 	if (!scsi_block_when_processing_errors(sdev))
1424 		goto error_out;
1425 
1426 	if (sdev->removable || sdkp->write_prot) {
1427 		if (bdev_check_media_change(bdev))
1428 			sd_revalidate_disk(bdev->bd_disk);
1429 	}
1430 
1431 	/*
1432 	 * If the drive is empty, just let the open fail.
1433 	 */
1434 	retval = -ENOMEDIUM;
1435 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1436 		goto error_out;
1437 
1438 	/*
1439 	 * If the device has the write protect tab set, have the open fail
1440 	 * if the user expects to be able to write to the thing.
1441 	 */
1442 	retval = -EROFS;
1443 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1444 		goto error_out;
1445 
1446 	/*
1447 	 * It is possible that the disk changing stuff resulted in
1448 	 * the device being taken offline.  If this is the case,
1449 	 * report this to the user, and don't pretend that the
1450 	 * open actually succeeded.
1451 	 */
1452 	retval = -ENXIO;
1453 	if (!scsi_device_online(sdev))
1454 		goto error_out;
1455 
1456 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1457 		if (scsi_block_when_processing_errors(sdev))
1458 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1459 	}
1460 
1461 	return 0;
1462 
1463 error_out:
1464 	scsi_disk_put(sdkp);
1465 	return retval;
1466 }
1467 
1468 /**
1469  *	sd_release - invoked when the (last) close(2) is called on this
1470  *	scsi disk.
1471  *	@disk: disk to release
1472  *	@mode: FMODE_* mask
1473  *
1474  *	Returns 0.
1475  *
1476  *	Note: may block (uninterruptible) if error recovery is underway
1477  *	on this disk.
1478  *
1479  *	Locking: called with bdev->bd_mutex held.
1480  **/
1481 static void sd_release(struct gendisk *disk, fmode_t mode)
1482 {
1483 	struct scsi_disk *sdkp = scsi_disk(disk);
1484 	struct scsi_device *sdev = sdkp->device;
1485 
1486 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1487 
1488 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1489 		if (scsi_block_when_processing_errors(sdev))
1490 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1491 	}
1492 
1493 	scsi_disk_put(sdkp);
1494 }
1495 
1496 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1497 {
1498 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1499 	struct scsi_device *sdp = sdkp->device;
1500 	struct Scsi_Host *host = sdp->host;
1501 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1502 	int diskinfo[4];
1503 
1504 	/* default to most commonly used values */
1505 	diskinfo[0] = 0x40;	/* 1 << 6 */
1506 	diskinfo[1] = 0x20;	/* 1 << 5 */
1507 	diskinfo[2] = capacity >> 11;
1508 
1509 	/* override with calculated, extended default, or driver values */
1510 	if (host->hostt->bios_param)
1511 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1512 	else
1513 		scsicam_bios_param(bdev, capacity, diskinfo);
1514 
1515 	geo->heads = diskinfo[0];
1516 	geo->sectors = diskinfo[1];
1517 	geo->cylinders = diskinfo[2];
1518 	return 0;
1519 }
1520 
1521 /**
1522  *	sd_ioctl - process an ioctl
1523  *	@bdev: target block device
1524  *	@mode: FMODE_* mask
1525  *	@cmd: ioctl command number
1526  *	@p: this is third argument given to ioctl(2) system call.
1527  *	Often contains a pointer.
1528  *
1529  *	Returns 0 if successful (some ioctls return positive numbers on
1530  *	success as well). Returns a negated errno value in case of error.
1531  *
1532  *	Note: most ioctls are forward onto the block subsystem or further
1533  *	down in the scsi subsystem.
1534  **/
1535 static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1536 			   unsigned int cmd, void __user *p)
1537 {
1538 	struct gendisk *disk = bdev->bd_disk;
1539 	struct scsi_disk *sdkp = scsi_disk(disk);
1540 	struct scsi_device *sdp = sdkp->device;
1541 	int error;
1542 
1543 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1544 				    "cmd=0x%x\n", disk->disk_name, cmd));
1545 
1546 	error = scsi_verify_blk_ioctl(bdev, cmd);
1547 	if (error < 0)
1548 		return error;
1549 
1550 	/*
1551 	 * If we are in the middle of error recovery, don't let anyone
1552 	 * else try and use this device.  Also, if error recovery fails, it
1553 	 * may try and take the device offline, in which case all further
1554 	 * access to the device is prohibited.
1555 	 */
1556 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1557 			(mode & FMODE_NDELAY) != 0);
1558 	if (error)
1559 		goto out;
1560 
1561 	if (is_sed_ioctl(cmd))
1562 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1563 
1564 	/*
1565 	 * Send SCSI addressing ioctls directly to mid level, send other
1566 	 * ioctls to block level and then onto mid level if they can't be
1567 	 * resolved.
1568 	 */
1569 	switch (cmd) {
1570 		case SCSI_IOCTL_GET_IDLUN:
1571 		case SCSI_IOCTL_GET_BUS_NUMBER:
1572 			error = scsi_ioctl(sdp, cmd, p);
1573 			break;
1574 		default:
1575 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1576 			break;
1577 	}
1578 out:
1579 	return error;
1580 }
1581 
1582 static void set_media_not_present(struct scsi_disk *sdkp)
1583 {
1584 	if (sdkp->media_present)
1585 		sdkp->device->changed = 1;
1586 
1587 	if (sdkp->device->removable) {
1588 		sdkp->media_present = 0;
1589 		sdkp->capacity = 0;
1590 	}
1591 }
1592 
1593 static int media_not_present(struct scsi_disk *sdkp,
1594 			     struct scsi_sense_hdr *sshdr)
1595 {
1596 	if (!scsi_sense_valid(sshdr))
1597 		return 0;
1598 
1599 	/* not invoked for commands that could return deferred errors */
1600 	switch (sshdr->sense_key) {
1601 	case UNIT_ATTENTION:
1602 	case NOT_READY:
1603 		/* medium not present */
1604 		if (sshdr->asc == 0x3A) {
1605 			set_media_not_present(sdkp);
1606 			return 1;
1607 		}
1608 	}
1609 	return 0;
1610 }
1611 
1612 /**
1613  *	sd_check_events - check media events
1614  *	@disk: kernel device descriptor
1615  *	@clearing: disk events currently being cleared
1616  *
1617  *	Returns mask of DISK_EVENT_*.
1618  *
1619  *	Note: this function is invoked from the block subsystem.
1620  **/
1621 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1622 {
1623 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1624 	struct scsi_device *sdp;
1625 	int retval;
1626 
1627 	if (!sdkp)
1628 		return 0;
1629 
1630 	sdp = sdkp->device;
1631 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1632 
1633 	/*
1634 	 * If the device is offline, don't send any commands - just pretend as
1635 	 * if the command failed.  If the device ever comes back online, we
1636 	 * can deal with it then.  It is only because of unrecoverable errors
1637 	 * that we would ever take a device offline in the first place.
1638 	 */
1639 	if (!scsi_device_online(sdp)) {
1640 		set_media_not_present(sdkp);
1641 		goto out;
1642 	}
1643 
1644 	/*
1645 	 * Using TEST_UNIT_READY enables differentiation between drive with
1646 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1647 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1648 	 *
1649 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1650 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1651 	 * sd_revalidate() is called.
1652 	 */
1653 	if (scsi_block_when_processing_errors(sdp)) {
1654 		struct scsi_sense_hdr sshdr = { 0, };
1655 
1656 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1657 					      &sshdr);
1658 
1659 		/* failed to execute TUR, assume media not present */
1660 		if (host_byte(retval)) {
1661 			set_media_not_present(sdkp);
1662 			goto out;
1663 		}
1664 
1665 		if (media_not_present(sdkp, &sshdr))
1666 			goto out;
1667 	}
1668 
1669 	/*
1670 	 * For removable scsi disk we have to recognise the presence
1671 	 * of a disk in the drive.
1672 	 */
1673 	if (!sdkp->media_present)
1674 		sdp->changed = 1;
1675 	sdkp->media_present = 1;
1676 out:
1677 	/*
1678 	 * sdp->changed is set under the following conditions:
1679 	 *
1680 	 *	Medium present state has changed in either direction.
1681 	 *	Device has indicated UNIT_ATTENTION.
1682 	 */
1683 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1684 	sdp->changed = 0;
1685 	scsi_disk_put(sdkp);
1686 	return retval;
1687 }
1688 
1689 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1690 {
1691 	int retries, res;
1692 	struct scsi_device *sdp = sdkp->device;
1693 	const int timeout = sdp->request_queue->rq_timeout
1694 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1695 	struct scsi_sense_hdr my_sshdr;
1696 
1697 	if (!scsi_device_online(sdp))
1698 		return -ENODEV;
1699 
1700 	/* caller might not be interested in sense, but we need it */
1701 	if (!sshdr)
1702 		sshdr = &my_sshdr;
1703 
1704 	for (retries = 3; retries > 0; --retries) {
1705 		unsigned char cmd[10] = { 0 };
1706 
1707 		cmd[0] = SYNCHRONIZE_CACHE;
1708 		/*
1709 		 * Leave the rest of the command zero to indicate
1710 		 * flush everything.
1711 		 */
1712 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1713 				timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1714 		if (res == 0)
1715 			break;
1716 	}
1717 
1718 	if (res) {
1719 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1720 
1721 		if (driver_byte(res) == DRIVER_SENSE)
1722 			sd_print_sense_hdr(sdkp, sshdr);
1723 
1724 		/* we need to evaluate the error return  */
1725 		if (scsi_sense_valid(sshdr) &&
1726 			(sshdr->asc == 0x3a ||	/* medium not present */
1727 			 sshdr->asc == 0x20 ||	/* invalid command */
1728 			 (sshdr->asc == 0x74 && sshdr->ascq == 0x71)))	/* drive is password locked */
1729 				/* this is no error here */
1730 				return 0;
1731 
1732 		switch (host_byte(res)) {
1733 		/* ignore errors due to racing a disconnection */
1734 		case DID_BAD_TARGET:
1735 		case DID_NO_CONNECT:
1736 			return 0;
1737 		/* signal the upper layer it might try again */
1738 		case DID_BUS_BUSY:
1739 		case DID_IMM_RETRY:
1740 		case DID_REQUEUE:
1741 		case DID_SOFT_ERROR:
1742 			return -EBUSY;
1743 		default:
1744 			return -EIO;
1745 		}
1746 	}
1747 	return 0;
1748 }
1749 
1750 static void sd_rescan(struct device *dev)
1751 {
1752 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1753 
1754 	sd_revalidate_disk(sdkp->disk);
1755 }
1756 
1757 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1758 		    unsigned int cmd, unsigned long arg)
1759 {
1760 	void __user *p = (void __user *)arg;
1761 	int ret;
1762 
1763 	ret = sd_ioctl_common(bdev, mode, cmd, p);
1764 	if (ret != -ENOTTY)
1765 		return ret;
1766 
1767 	return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1768 }
1769 
1770 #ifdef CONFIG_COMPAT
1771 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1772 			   unsigned int cmd, unsigned long arg)
1773 {
1774 	void __user *p = compat_ptr(arg);
1775 	int ret;
1776 
1777 	ret = sd_ioctl_common(bdev, mode, cmd, p);
1778 	if (ret != -ENOTTY)
1779 		return ret;
1780 
1781 	return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1782 }
1783 #endif
1784 
1785 static char sd_pr_type(enum pr_type type)
1786 {
1787 	switch (type) {
1788 	case PR_WRITE_EXCLUSIVE:
1789 		return 0x01;
1790 	case PR_EXCLUSIVE_ACCESS:
1791 		return 0x03;
1792 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1793 		return 0x05;
1794 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1795 		return 0x06;
1796 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1797 		return 0x07;
1798 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1799 		return 0x08;
1800 	default:
1801 		return 0;
1802 	}
1803 };
1804 
1805 static int sd_pr_command(struct block_device *bdev, u8 sa,
1806 		u64 key, u64 sa_key, u8 type, u8 flags)
1807 {
1808 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1809 	struct scsi_device *sdev = sdkp->device;
1810 	struct scsi_sense_hdr sshdr;
1811 	int result;
1812 	u8 cmd[16] = { 0, };
1813 	u8 data[24] = { 0, };
1814 
1815 	cmd[0] = PERSISTENT_RESERVE_OUT;
1816 	cmd[1] = sa;
1817 	cmd[2] = type;
1818 	put_unaligned_be32(sizeof(data), &cmd[5]);
1819 
1820 	put_unaligned_be64(key, &data[0]);
1821 	put_unaligned_be64(sa_key, &data[8]);
1822 	data[20] = flags;
1823 
1824 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1825 			&sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1826 
1827 	if (driver_byte(result) == DRIVER_SENSE &&
1828 	    scsi_sense_valid(&sshdr)) {
1829 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1830 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1831 	}
1832 
1833 	return result;
1834 }
1835 
1836 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1837 		u32 flags)
1838 {
1839 	if (flags & ~PR_FL_IGNORE_KEY)
1840 		return -EOPNOTSUPP;
1841 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1842 			old_key, new_key, 0,
1843 			(1 << 0) /* APTPL */);
1844 }
1845 
1846 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1847 		u32 flags)
1848 {
1849 	if (flags)
1850 		return -EOPNOTSUPP;
1851 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1852 }
1853 
1854 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1855 {
1856 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1857 }
1858 
1859 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1860 		enum pr_type type, bool abort)
1861 {
1862 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1863 			     sd_pr_type(type), 0);
1864 }
1865 
1866 static int sd_pr_clear(struct block_device *bdev, u64 key)
1867 {
1868 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1869 }
1870 
1871 static const struct pr_ops sd_pr_ops = {
1872 	.pr_register	= sd_pr_register,
1873 	.pr_reserve	= sd_pr_reserve,
1874 	.pr_release	= sd_pr_release,
1875 	.pr_preempt	= sd_pr_preempt,
1876 	.pr_clear	= sd_pr_clear,
1877 };
1878 
1879 static const struct block_device_operations sd_fops = {
1880 	.owner			= THIS_MODULE,
1881 	.open			= sd_open,
1882 	.release		= sd_release,
1883 	.ioctl			= sd_ioctl,
1884 	.getgeo			= sd_getgeo,
1885 #ifdef CONFIG_COMPAT
1886 	.compat_ioctl		= sd_compat_ioctl,
1887 #endif
1888 	.check_events		= sd_check_events,
1889 	.unlock_native_capacity	= sd_unlock_native_capacity,
1890 	.report_zones		= sd_zbc_report_zones,
1891 	.pr_ops			= &sd_pr_ops,
1892 };
1893 
1894 /**
1895  *	sd_eh_reset - reset error handling callback
1896  *	@scmd:		sd-issued command that has failed
1897  *
1898  *	This function is called by the SCSI midlayer before starting
1899  *	SCSI EH. When counting medium access failures we have to be
1900  *	careful to register it only only once per device and SCSI EH run;
1901  *	there might be several timed out commands which will cause the
1902  *	'max_medium_access_timeouts' counter to trigger after the first
1903  *	SCSI EH run already and set the device to offline.
1904  *	So this function resets the internal counter before starting SCSI EH.
1905  **/
1906 static void sd_eh_reset(struct scsi_cmnd *scmd)
1907 {
1908 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1909 
1910 	/* New SCSI EH run, reset gate variable */
1911 	sdkp->ignore_medium_access_errors = false;
1912 }
1913 
1914 /**
1915  *	sd_eh_action - error handling callback
1916  *	@scmd:		sd-issued command that has failed
1917  *	@eh_disp:	The recovery disposition suggested by the midlayer
1918  *
1919  *	This function is called by the SCSI midlayer upon completion of an
1920  *	error test command (currently TEST UNIT READY). The result of sending
1921  *	the eh command is passed in eh_disp.  We're looking for devices that
1922  *	fail medium access commands but are OK with non access commands like
1923  *	test unit ready (so wrongly see the device as having a successful
1924  *	recovery)
1925  **/
1926 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1927 {
1928 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1929 	struct scsi_device *sdev = scmd->device;
1930 
1931 	if (!scsi_device_online(sdev) ||
1932 	    !scsi_medium_access_command(scmd) ||
1933 	    host_byte(scmd->result) != DID_TIME_OUT ||
1934 	    eh_disp != SUCCESS)
1935 		return eh_disp;
1936 
1937 	/*
1938 	 * The device has timed out executing a medium access command.
1939 	 * However, the TEST UNIT READY command sent during error
1940 	 * handling completed successfully. Either the device is in the
1941 	 * process of recovering or has it suffered an internal failure
1942 	 * that prevents access to the storage medium.
1943 	 */
1944 	if (!sdkp->ignore_medium_access_errors) {
1945 		sdkp->medium_access_timed_out++;
1946 		sdkp->ignore_medium_access_errors = true;
1947 	}
1948 
1949 	/*
1950 	 * If the device keeps failing read/write commands but TEST UNIT
1951 	 * READY always completes successfully we assume that medium
1952 	 * access is no longer possible and take the device offline.
1953 	 */
1954 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1955 		scmd_printk(KERN_ERR, scmd,
1956 			    "Medium access timeout failure. Offlining disk!\n");
1957 		mutex_lock(&sdev->state_mutex);
1958 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1959 		mutex_unlock(&sdev->state_mutex);
1960 
1961 		return SUCCESS;
1962 	}
1963 
1964 	return eh_disp;
1965 }
1966 
1967 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1968 {
1969 	struct request *req = scmd->request;
1970 	struct scsi_device *sdev = scmd->device;
1971 	unsigned int transferred, good_bytes;
1972 	u64 start_lba, end_lba, bad_lba;
1973 
1974 	/*
1975 	 * Some commands have a payload smaller than the device logical
1976 	 * block size (e.g. INQUIRY on a 4K disk).
1977 	 */
1978 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1979 		return 0;
1980 
1981 	/* Check if we have a 'bad_lba' information */
1982 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1983 				     SCSI_SENSE_BUFFERSIZE,
1984 				     &bad_lba))
1985 		return 0;
1986 
1987 	/*
1988 	 * If the bad lba was reported incorrectly, we have no idea where
1989 	 * the error is.
1990 	 */
1991 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1992 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1993 	if (bad_lba < start_lba || bad_lba >= end_lba)
1994 		return 0;
1995 
1996 	/*
1997 	 * resid is optional but mostly filled in.  When it's unused,
1998 	 * its value is zero, so we assume the whole buffer transferred
1999 	 */
2000 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2001 
2002 	/* This computation should always be done in terms of the
2003 	 * resolution of the device's medium.
2004 	 */
2005 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2006 
2007 	return min(good_bytes, transferred);
2008 }
2009 
2010 /**
2011  *	sd_done - bottom half handler: called when the lower level
2012  *	driver has completed (successfully or otherwise) a scsi command.
2013  *	@SCpnt: mid-level's per command structure.
2014  *
2015  *	Note: potentially run from within an ISR. Must not block.
2016  **/
2017 static int sd_done(struct scsi_cmnd *SCpnt)
2018 {
2019 	int result = SCpnt->result;
2020 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2021 	unsigned int sector_size = SCpnt->device->sector_size;
2022 	unsigned int resid;
2023 	struct scsi_sense_hdr sshdr;
2024 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
2025 	struct request *req = SCpnt->request;
2026 	int sense_valid = 0;
2027 	int sense_deferred = 0;
2028 
2029 	switch (req_op(req)) {
2030 	case REQ_OP_DISCARD:
2031 	case REQ_OP_WRITE_ZEROES:
2032 	case REQ_OP_WRITE_SAME:
2033 	case REQ_OP_ZONE_RESET:
2034 	case REQ_OP_ZONE_RESET_ALL:
2035 	case REQ_OP_ZONE_OPEN:
2036 	case REQ_OP_ZONE_CLOSE:
2037 	case REQ_OP_ZONE_FINISH:
2038 		if (!result) {
2039 			good_bytes = blk_rq_bytes(req);
2040 			scsi_set_resid(SCpnt, 0);
2041 		} else {
2042 			good_bytes = 0;
2043 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2044 		}
2045 		break;
2046 	default:
2047 		/*
2048 		 * In case of bogus fw or device, we could end up having
2049 		 * an unaligned partial completion. Check this here and force
2050 		 * alignment.
2051 		 */
2052 		resid = scsi_get_resid(SCpnt);
2053 		if (resid & (sector_size - 1)) {
2054 			sd_printk(KERN_INFO, sdkp,
2055 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2056 				resid, sector_size);
2057 			scsi_print_command(SCpnt);
2058 			resid = min(scsi_bufflen(SCpnt),
2059 				    round_up(resid, sector_size));
2060 			scsi_set_resid(SCpnt, resid);
2061 		}
2062 	}
2063 
2064 	if (result) {
2065 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2066 		if (sense_valid)
2067 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2068 	}
2069 	sdkp->medium_access_timed_out = 0;
2070 
2071 	if (driver_byte(result) != DRIVER_SENSE &&
2072 	    (!sense_valid || sense_deferred))
2073 		goto out;
2074 
2075 	switch (sshdr.sense_key) {
2076 	case HARDWARE_ERROR:
2077 	case MEDIUM_ERROR:
2078 		good_bytes = sd_completed_bytes(SCpnt);
2079 		break;
2080 	case RECOVERED_ERROR:
2081 		good_bytes = scsi_bufflen(SCpnt);
2082 		break;
2083 	case NO_SENSE:
2084 		/* This indicates a false check condition, so ignore it.  An
2085 		 * unknown amount of data was transferred so treat it as an
2086 		 * error.
2087 		 */
2088 		SCpnt->result = 0;
2089 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2090 		break;
2091 	case ABORTED_COMMAND:
2092 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2093 			good_bytes = sd_completed_bytes(SCpnt);
2094 		break;
2095 	case ILLEGAL_REQUEST:
2096 		switch (sshdr.asc) {
2097 		case 0x10:	/* DIX: Host detected corruption */
2098 			good_bytes = sd_completed_bytes(SCpnt);
2099 			break;
2100 		case 0x20:	/* INVALID COMMAND OPCODE */
2101 		case 0x24:	/* INVALID FIELD IN CDB */
2102 			switch (SCpnt->cmnd[0]) {
2103 			case UNMAP:
2104 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2105 				break;
2106 			case WRITE_SAME_16:
2107 			case WRITE_SAME:
2108 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2109 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2110 				} else {
2111 					sdkp->device->no_write_same = 1;
2112 					sd_config_write_same(sdkp);
2113 					req->rq_flags |= RQF_QUIET;
2114 				}
2115 				break;
2116 			}
2117 		}
2118 		break;
2119 	default:
2120 		break;
2121 	}
2122 
2123  out:
2124 	if (sd_is_zoned(sdkp))
2125 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2126 
2127 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2128 					   "sd_done: completed %d of %d bytes\n",
2129 					   good_bytes, scsi_bufflen(SCpnt)));
2130 
2131 	return good_bytes;
2132 }
2133 
2134 /*
2135  * spinup disk - called only in sd_revalidate_disk()
2136  */
2137 static void
2138 sd_spinup_disk(struct scsi_disk *sdkp)
2139 {
2140 	unsigned char cmd[10];
2141 	unsigned long spintime_expire = 0;
2142 	int retries, spintime;
2143 	unsigned int the_result;
2144 	struct scsi_sense_hdr sshdr;
2145 	int sense_valid = 0;
2146 
2147 	spintime = 0;
2148 
2149 	/* Spin up drives, as required.  Only do this at boot time */
2150 	/* Spinup needs to be done for module loads too. */
2151 	do {
2152 		retries = 0;
2153 
2154 		do {
2155 			cmd[0] = TEST_UNIT_READY;
2156 			memset((void *) &cmd[1], 0, 9);
2157 
2158 			the_result = scsi_execute_req(sdkp->device, cmd,
2159 						      DMA_NONE, NULL, 0,
2160 						      &sshdr, SD_TIMEOUT,
2161 						      sdkp->max_retries, NULL);
2162 
2163 			/*
2164 			 * If the drive has indicated to us that it
2165 			 * doesn't have any media in it, don't bother
2166 			 * with any more polling.
2167 			 */
2168 			if (media_not_present(sdkp, &sshdr))
2169 				return;
2170 
2171 			if (the_result)
2172 				sense_valid = scsi_sense_valid(&sshdr);
2173 			retries++;
2174 		} while (retries < 3 &&
2175 			 (!scsi_status_is_good(the_result) ||
2176 			  ((driver_byte(the_result) == DRIVER_SENSE) &&
2177 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2178 
2179 		if (driver_byte(the_result) != DRIVER_SENSE) {
2180 			/* no sense, TUR either succeeded or failed
2181 			 * with a status error */
2182 			if(!spintime && !scsi_status_is_good(the_result)) {
2183 				sd_print_result(sdkp, "Test Unit Ready failed",
2184 						the_result);
2185 			}
2186 			break;
2187 		}
2188 
2189 		/*
2190 		 * The device does not want the automatic start to be issued.
2191 		 */
2192 		if (sdkp->device->no_start_on_add)
2193 			break;
2194 
2195 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2196 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2197 				break;	/* manual intervention required */
2198 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2199 				break;	/* standby */
2200 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2201 				break;	/* unavailable */
2202 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2203 				break;	/* sanitize in progress */
2204 			/*
2205 			 * Issue command to spin up drive when not ready
2206 			 */
2207 			if (!spintime) {
2208 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2209 				cmd[0] = START_STOP;
2210 				cmd[1] = 1;	/* Return immediately */
2211 				memset((void *) &cmd[2], 0, 8);
2212 				cmd[4] = 1;	/* Start spin cycle */
2213 				if (sdkp->device->start_stop_pwr_cond)
2214 					cmd[4] |= 1 << 4;
2215 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2216 						 NULL, 0, &sshdr,
2217 						 SD_TIMEOUT, sdkp->max_retries,
2218 						 NULL);
2219 				spintime_expire = jiffies + 100 * HZ;
2220 				spintime = 1;
2221 			}
2222 			/* Wait 1 second for next try */
2223 			msleep(1000);
2224 			printk(KERN_CONT ".");
2225 
2226 		/*
2227 		 * Wait for USB flash devices with slow firmware.
2228 		 * Yes, this sense key/ASC combination shouldn't
2229 		 * occur here.  It's characteristic of these devices.
2230 		 */
2231 		} else if (sense_valid &&
2232 				sshdr.sense_key == UNIT_ATTENTION &&
2233 				sshdr.asc == 0x28) {
2234 			if (!spintime) {
2235 				spintime_expire = jiffies + 5 * HZ;
2236 				spintime = 1;
2237 			}
2238 			/* Wait 1 second for next try */
2239 			msleep(1000);
2240 		} else {
2241 			/* we don't understand the sense code, so it's
2242 			 * probably pointless to loop */
2243 			if(!spintime) {
2244 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2245 				sd_print_sense_hdr(sdkp, &sshdr);
2246 			}
2247 			break;
2248 		}
2249 
2250 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2251 
2252 	if (spintime) {
2253 		if (scsi_status_is_good(the_result))
2254 			printk(KERN_CONT "ready\n");
2255 		else
2256 			printk(KERN_CONT "not responding...\n");
2257 	}
2258 }
2259 
2260 /*
2261  * Determine whether disk supports Data Integrity Field.
2262  */
2263 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2264 {
2265 	struct scsi_device *sdp = sdkp->device;
2266 	u8 type;
2267 	int ret = 0;
2268 
2269 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2270 		sdkp->protection_type = 0;
2271 		return ret;
2272 	}
2273 
2274 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2275 
2276 	if (type > T10_PI_TYPE3_PROTECTION)
2277 		ret = -ENODEV;
2278 	else if (scsi_host_dif_capable(sdp->host, type))
2279 		ret = 1;
2280 
2281 	if (sdkp->first_scan || type != sdkp->protection_type)
2282 		switch (ret) {
2283 		case -ENODEV:
2284 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2285 				  " protection type %u. Disabling disk!\n",
2286 				  type);
2287 			break;
2288 		case 1:
2289 			sd_printk(KERN_NOTICE, sdkp,
2290 				  "Enabling DIF Type %u protection\n", type);
2291 			break;
2292 		case 0:
2293 			sd_printk(KERN_NOTICE, sdkp,
2294 				  "Disabling DIF Type %u protection\n", type);
2295 			break;
2296 		}
2297 
2298 	sdkp->protection_type = type;
2299 
2300 	return ret;
2301 }
2302 
2303 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2304 			struct scsi_sense_hdr *sshdr, int sense_valid,
2305 			int the_result)
2306 {
2307 	if (driver_byte(the_result) == DRIVER_SENSE)
2308 		sd_print_sense_hdr(sdkp, sshdr);
2309 	else
2310 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2311 
2312 	/*
2313 	 * Set dirty bit for removable devices if not ready -
2314 	 * sometimes drives will not report this properly.
2315 	 */
2316 	if (sdp->removable &&
2317 	    sense_valid && sshdr->sense_key == NOT_READY)
2318 		set_media_not_present(sdkp);
2319 
2320 	/*
2321 	 * We used to set media_present to 0 here to indicate no media
2322 	 * in the drive, but some drives fail read capacity even with
2323 	 * media present, so we can't do that.
2324 	 */
2325 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2326 }
2327 
2328 #define RC16_LEN 32
2329 #if RC16_LEN > SD_BUF_SIZE
2330 #error RC16_LEN must not be more than SD_BUF_SIZE
2331 #endif
2332 
2333 #define READ_CAPACITY_RETRIES_ON_RESET	10
2334 
2335 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2336 						unsigned char *buffer)
2337 {
2338 	unsigned char cmd[16];
2339 	struct scsi_sense_hdr sshdr;
2340 	int sense_valid = 0;
2341 	int the_result;
2342 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2343 	unsigned int alignment;
2344 	unsigned long long lba;
2345 	unsigned sector_size;
2346 
2347 	if (sdp->no_read_capacity_16)
2348 		return -EINVAL;
2349 
2350 	do {
2351 		memset(cmd, 0, 16);
2352 		cmd[0] = SERVICE_ACTION_IN_16;
2353 		cmd[1] = SAI_READ_CAPACITY_16;
2354 		cmd[13] = RC16_LEN;
2355 		memset(buffer, 0, RC16_LEN);
2356 
2357 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2358 					buffer, RC16_LEN, &sshdr,
2359 					SD_TIMEOUT, sdkp->max_retries, NULL);
2360 
2361 		if (media_not_present(sdkp, &sshdr))
2362 			return -ENODEV;
2363 
2364 		if (the_result) {
2365 			sense_valid = scsi_sense_valid(&sshdr);
2366 			if (sense_valid &&
2367 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2368 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2369 			    sshdr.ascq == 0x00)
2370 				/* Invalid Command Operation Code or
2371 				 * Invalid Field in CDB, just retry
2372 				 * silently with RC10 */
2373 				return -EINVAL;
2374 			if (sense_valid &&
2375 			    sshdr.sense_key == UNIT_ATTENTION &&
2376 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2377 				/* Device reset might occur several times,
2378 				 * give it one more chance */
2379 				if (--reset_retries > 0)
2380 					continue;
2381 		}
2382 		retries--;
2383 
2384 	} while (the_result && retries);
2385 
2386 	if (the_result) {
2387 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2388 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2389 		return -EINVAL;
2390 	}
2391 
2392 	sector_size = get_unaligned_be32(&buffer[8]);
2393 	lba = get_unaligned_be64(&buffer[0]);
2394 
2395 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2396 		sdkp->capacity = 0;
2397 		return -ENODEV;
2398 	}
2399 
2400 	/* Logical blocks per physical block exponent */
2401 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2402 
2403 	/* RC basis */
2404 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2405 
2406 	/* Lowest aligned logical block */
2407 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2408 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2409 	if (alignment && sdkp->first_scan)
2410 		sd_printk(KERN_NOTICE, sdkp,
2411 			  "physical block alignment offset: %u\n", alignment);
2412 
2413 	if (buffer[14] & 0x80) { /* LBPME */
2414 		sdkp->lbpme = 1;
2415 
2416 		if (buffer[14] & 0x40) /* LBPRZ */
2417 			sdkp->lbprz = 1;
2418 
2419 		sd_config_discard(sdkp, SD_LBP_WS16);
2420 	}
2421 
2422 	sdkp->capacity = lba + 1;
2423 	return sector_size;
2424 }
2425 
2426 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2427 						unsigned char *buffer)
2428 {
2429 	unsigned char cmd[16];
2430 	struct scsi_sense_hdr sshdr;
2431 	int sense_valid = 0;
2432 	int the_result;
2433 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2434 	sector_t lba;
2435 	unsigned sector_size;
2436 
2437 	do {
2438 		cmd[0] = READ_CAPACITY;
2439 		memset(&cmd[1], 0, 9);
2440 		memset(buffer, 0, 8);
2441 
2442 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2443 					buffer, 8, &sshdr,
2444 					SD_TIMEOUT, sdkp->max_retries, NULL);
2445 
2446 		if (media_not_present(sdkp, &sshdr))
2447 			return -ENODEV;
2448 
2449 		if (the_result) {
2450 			sense_valid = scsi_sense_valid(&sshdr);
2451 			if (sense_valid &&
2452 			    sshdr.sense_key == UNIT_ATTENTION &&
2453 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2454 				/* Device reset might occur several times,
2455 				 * give it one more chance */
2456 				if (--reset_retries > 0)
2457 					continue;
2458 		}
2459 		retries--;
2460 
2461 	} while (the_result && retries);
2462 
2463 	if (the_result) {
2464 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2465 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2466 		return -EINVAL;
2467 	}
2468 
2469 	sector_size = get_unaligned_be32(&buffer[4]);
2470 	lba = get_unaligned_be32(&buffer[0]);
2471 
2472 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2473 		/* Some buggy (usb cardreader) devices return an lba of
2474 		   0xffffffff when the want to report a size of 0 (with
2475 		   which they really mean no media is present) */
2476 		sdkp->capacity = 0;
2477 		sdkp->physical_block_size = sector_size;
2478 		return sector_size;
2479 	}
2480 
2481 	sdkp->capacity = lba + 1;
2482 	sdkp->physical_block_size = sector_size;
2483 	return sector_size;
2484 }
2485 
2486 static int sd_try_rc16_first(struct scsi_device *sdp)
2487 {
2488 	if (sdp->host->max_cmd_len < 16)
2489 		return 0;
2490 	if (sdp->try_rc_10_first)
2491 		return 0;
2492 	if (sdp->scsi_level > SCSI_SPC_2)
2493 		return 1;
2494 	if (scsi_device_protection(sdp))
2495 		return 1;
2496 	return 0;
2497 }
2498 
2499 /*
2500  * read disk capacity
2501  */
2502 static void
2503 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2504 {
2505 	int sector_size;
2506 	struct scsi_device *sdp = sdkp->device;
2507 
2508 	if (sd_try_rc16_first(sdp)) {
2509 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2510 		if (sector_size == -EOVERFLOW)
2511 			goto got_data;
2512 		if (sector_size == -ENODEV)
2513 			return;
2514 		if (sector_size < 0)
2515 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2516 		if (sector_size < 0)
2517 			return;
2518 	} else {
2519 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2520 		if (sector_size == -EOVERFLOW)
2521 			goto got_data;
2522 		if (sector_size < 0)
2523 			return;
2524 		if ((sizeof(sdkp->capacity) > 4) &&
2525 		    (sdkp->capacity > 0xffffffffULL)) {
2526 			int old_sector_size = sector_size;
2527 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2528 					"Trying to use READ CAPACITY(16).\n");
2529 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2530 			if (sector_size < 0) {
2531 				sd_printk(KERN_NOTICE, sdkp,
2532 					"Using 0xffffffff as device size\n");
2533 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2534 				sector_size = old_sector_size;
2535 				goto got_data;
2536 			}
2537 			/* Remember that READ CAPACITY(16) succeeded */
2538 			sdp->try_rc_10_first = 0;
2539 		}
2540 	}
2541 
2542 	/* Some devices are known to return the total number of blocks,
2543 	 * not the highest block number.  Some devices have versions
2544 	 * which do this and others which do not.  Some devices we might
2545 	 * suspect of doing this but we don't know for certain.
2546 	 *
2547 	 * If we know the reported capacity is wrong, decrement it.  If
2548 	 * we can only guess, then assume the number of blocks is even
2549 	 * (usually true but not always) and err on the side of lowering
2550 	 * the capacity.
2551 	 */
2552 	if (sdp->fix_capacity ||
2553 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2554 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2555 				"from its reported value: %llu\n",
2556 				(unsigned long long) sdkp->capacity);
2557 		--sdkp->capacity;
2558 	}
2559 
2560 got_data:
2561 	if (sector_size == 0) {
2562 		sector_size = 512;
2563 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2564 			  "assuming 512.\n");
2565 	}
2566 
2567 	if (sector_size != 512 &&
2568 	    sector_size != 1024 &&
2569 	    sector_size != 2048 &&
2570 	    sector_size != 4096) {
2571 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2572 			  sector_size);
2573 		/*
2574 		 * The user might want to re-format the drive with
2575 		 * a supported sectorsize.  Once this happens, it
2576 		 * would be relatively trivial to set the thing up.
2577 		 * For this reason, we leave the thing in the table.
2578 		 */
2579 		sdkp->capacity = 0;
2580 		/*
2581 		 * set a bogus sector size so the normal read/write
2582 		 * logic in the block layer will eventually refuse any
2583 		 * request on this device without tripping over power
2584 		 * of two sector size assumptions
2585 		 */
2586 		sector_size = 512;
2587 	}
2588 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2589 	blk_queue_physical_block_size(sdp->request_queue,
2590 				      sdkp->physical_block_size);
2591 	sdkp->device->sector_size = sector_size;
2592 
2593 	if (sdkp->capacity > 0xffffffff)
2594 		sdp->use_16_for_rw = 1;
2595 
2596 }
2597 
2598 /*
2599  * Print disk capacity
2600  */
2601 static void
2602 sd_print_capacity(struct scsi_disk *sdkp,
2603 		  sector_t old_capacity)
2604 {
2605 	int sector_size = sdkp->device->sector_size;
2606 	char cap_str_2[10], cap_str_10[10];
2607 
2608 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2609 		return;
2610 
2611 	string_get_size(sdkp->capacity, sector_size,
2612 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2613 	string_get_size(sdkp->capacity, sector_size,
2614 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2615 
2616 	sd_printk(KERN_NOTICE, sdkp,
2617 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2618 		  (unsigned long long)sdkp->capacity,
2619 		  sector_size, cap_str_10, cap_str_2);
2620 
2621 	if (sdkp->physical_block_size != sector_size)
2622 		sd_printk(KERN_NOTICE, sdkp,
2623 			  "%u-byte physical blocks\n",
2624 			  sdkp->physical_block_size);
2625 }
2626 
2627 /* called with buffer of length 512 */
2628 static inline int
2629 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2630 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2631 		 struct scsi_sense_hdr *sshdr)
2632 {
2633 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2634 			       SD_TIMEOUT, sdkp->max_retries, data,
2635 			       sshdr);
2636 }
2637 
2638 /*
2639  * read write protect setting, if possible - called only in sd_revalidate_disk()
2640  * called with buffer of length SD_BUF_SIZE
2641  */
2642 static void
2643 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2644 {
2645 	int res;
2646 	struct scsi_device *sdp = sdkp->device;
2647 	struct scsi_mode_data data;
2648 	int old_wp = sdkp->write_prot;
2649 
2650 	set_disk_ro(sdkp->disk, 0);
2651 	if (sdp->skip_ms_page_3f) {
2652 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2653 		return;
2654 	}
2655 
2656 	if (sdp->use_192_bytes_for_3f) {
2657 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2658 	} else {
2659 		/*
2660 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2661 		 * We have to start carefully: some devices hang if we ask
2662 		 * for more than is available.
2663 		 */
2664 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2665 
2666 		/*
2667 		 * Second attempt: ask for page 0 When only page 0 is
2668 		 * implemented, a request for page 3F may return Sense Key
2669 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2670 		 * CDB.
2671 		 */
2672 		if (!scsi_status_is_good(res))
2673 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2674 
2675 		/*
2676 		 * Third attempt: ask 255 bytes, as we did earlier.
2677 		 */
2678 		if (!scsi_status_is_good(res))
2679 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2680 					       &data, NULL);
2681 	}
2682 
2683 	if (!scsi_status_is_good(res)) {
2684 		sd_first_printk(KERN_WARNING, sdkp,
2685 			  "Test WP failed, assume Write Enabled\n");
2686 	} else {
2687 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2688 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2689 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2690 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2691 				  sdkp->write_prot ? "on" : "off");
2692 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2693 		}
2694 	}
2695 }
2696 
2697 /*
2698  * sd_read_cache_type - called only from sd_revalidate_disk()
2699  * called with buffer of length SD_BUF_SIZE
2700  */
2701 static void
2702 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2703 {
2704 	int len = 0, res;
2705 	struct scsi_device *sdp = sdkp->device;
2706 
2707 	int dbd;
2708 	int modepage;
2709 	int first_len;
2710 	struct scsi_mode_data data;
2711 	struct scsi_sense_hdr sshdr;
2712 	int old_wce = sdkp->WCE;
2713 	int old_rcd = sdkp->RCD;
2714 	int old_dpofua = sdkp->DPOFUA;
2715 
2716 
2717 	if (sdkp->cache_override)
2718 		return;
2719 
2720 	first_len = 4;
2721 	if (sdp->skip_ms_page_8) {
2722 		if (sdp->type == TYPE_RBC)
2723 			goto defaults;
2724 		else {
2725 			if (sdp->skip_ms_page_3f)
2726 				goto defaults;
2727 			modepage = 0x3F;
2728 			if (sdp->use_192_bytes_for_3f)
2729 				first_len = 192;
2730 			dbd = 0;
2731 		}
2732 	} else if (sdp->type == TYPE_RBC) {
2733 		modepage = 6;
2734 		dbd = 8;
2735 	} else {
2736 		modepage = 8;
2737 		dbd = 0;
2738 	}
2739 
2740 	/* cautiously ask */
2741 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2742 			&data, &sshdr);
2743 
2744 	if (!scsi_status_is_good(res))
2745 		goto bad_sense;
2746 
2747 	if (!data.header_length) {
2748 		modepage = 6;
2749 		first_len = 0;
2750 		sd_first_printk(KERN_ERR, sdkp,
2751 				"Missing header in MODE_SENSE response\n");
2752 	}
2753 
2754 	/* that went OK, now ask for the proper length */
2755 	len = data.length;
2756 
2757 	/*
2758 	 * We're only interested in the first three bytes, actually.
2759 	 * But the data cache page is defined for the first 20.
2760 	 */
2761 	if (len < 3)
2762 		goto bad_sense;
2763 	else if (len > SD_BUF_SIZE) {
2764 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2765 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2766 		len = SD_BUF_SIZE;
2767 	}
2768 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2769 		len = 192;
2770 
2771 	/* Get the data */
2772 	if (len > first_len)
2773 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2774 				&data, &sshdr);
2775 
2776 	if (scsi_status_is_good(res)) {
2777 		int offset = data.header_length + data.block_descriptor_length;
2778 
2779 		while (offset < len) {
2780 			u8 page_code = buffer[offset] & 0x3F;
2781 			u8 spf       = buffer[offset] & 0x40;
2782 
2783 			if (page_code == 8 || page_code == 6) {
2784 				/* We're interested only in the first 3 bytes.
2785 				 */
2786 				if (len - offset <= 2) {
2787 					sd_first_printk(KERN_ERR, sdkp,
2788 						"Incomplete mode parameter "
2789 							"data\n");
2790 					goto defaults;
2791 				} else {
2792 					modepage = page_code;
2793 					goto Page_found;
2794 				}
2795 			} else {
2796 				/* Go to the next page */
2797 				if (spf && len - offset > 3)
2798 					offset += 4 + (buffer[offset+2] << 8) +
2799 						buffer[offset+3];
2800 				else if (!spf && len - offset > 1)
2801 					offset += 2 + buffer[offset+1];
2802 				else {
2803 					sd_first_printk(KERN_ERR, sdkp,
2804 							"Incomplete mode "
2805 							"parameter data\n");
2806 					goto defaults;
2807 				}
2808 			}
2809 		}
2810 
2811 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2812 		goto defaults;
2813 
2814 	Page_found:
2815 		if (modepage == 8) {
2816 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2817 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2818 		} else {
2819 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2820 			sdkp->RCD = 0;
2821 		}
2822 
2823 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2824 		if (sdp->broken_fua) {
2825 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2826 			sdkp->DPOFUA = 0;
2827 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2828 			   !sdkp->device->use_16_for_rw) {
2829 			sd_first_printk(KERN_NOTICE, sdkp,
2830 				  "Uses READ/WRITE(6), disabling FUA\n");
2831 			sdkp->DPOFUA = 0;
2832 		}
2833 
2834 		/* No cache flush allowed for write protected devices */
2835 		if (sdkp->WCE && sdkp->write_prot)
2836 			sdkp->WCE = 0;
2837 
2838 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2839 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2840 			sd_printk(KERN_NOTICE, sdkp,
2841 				  "Write cache: %s, read cache: %s, %s\n",
2842 				  sdkp->WCE ? "enabled" : "disabled",
2843 				  sdkp->RCD ? "disabled" : "enabled",
2844 				  sdkp->DPOFUA ? "supports DPO and FUA"
2845 				  : "doesn't support DPO or FUA");
2846 
2847 		return;
2848 	}
2849 
2850 bad_sense:
2851 	if (scsi_sense_valid(&sshdr) &&
2852 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2853 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2854 		/* Invalid field in CDB */
2855 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2856 	else
2857 		sd_first_printk(KERN_ERR, sdkp,
2858 				"Asking for cache data failed\n");
2859 
2860 defaults:
2861 	if (sdp->wce_default_on) {
2862 		sd_first_printk(KERN_NOTICE, sdkp,
2863 				"Assuming drive cache: write back\n");
2864 		sdkp->WCE = 1;
2865 	} else {
2866 		sd_first_printk(KERN_ERR, sdkp,
2867 				"Assuming drive cache: write through\n");
2868 		sdkp->WCE = 0;
2869 	}
2870 	sdkp->RCD = 0;
2871 	sdkp->DPOFUA = 0;
2872 }
2873 
2874 /*
2875  * The ATO bit indicates whether the DIF application tag is available
2876  * for use by the operating system.
2877  */
2878 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2879 {
2880 	int res, offset;
2881 	struct scsi_device *sdp = sdkp->device;
2882 	struct scsi_mode_data data;
2883 	struct scsi_sense_hdr sshdr;
2884 
2885 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2886 		return;
2887 
2888 	if (sdkp->protection_type == 0)
2889 		return;
2890 
2891 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2892 			      sdkp->max_retries, &data, &sshdr);
2893 
2894 	if (!scsi_status_is_good(res) || !data.header_length ||
2895 	    data.length < 6) {
2896 		sd_first_printk(KERN_WARNING, sdkp,
2897 			  "getting Control mode page failed, assume no ATO\n");
2898 
2899 		if (scsi_sense_valid(&sshdr))
2900 			sd_print_sense_hdr(sdkp, &sshdr);
2901 
2902 		return;
2903 	}
2904 
2905 	offset = data.header_length + data.block_descriptor_length;
2906 
2907 	if ((buffer[offset] & 0x3f) != 0x0a) {
2908 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2909 		return;
2910 	}
2911 
2912 	if ((buffer[offset + 5] & 0x80) == 0)
2913 		return;
2914 
2915 	sdkp->ATO = 1;
2916 
2917 	return;
2918 }
2919 
2920 /**
2921  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2922  * @sdkp: disk to query
2923  */
2924 static void sd_read_block_limits(struct scsi_disk *sdkp)
2925 {
2926 	unsigned int sector_sz = sdkp->device->sector_size;
2927 	const int vpd_len = 64;
2928 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2929 
2930 	if (!buffer ||
2931 	    /* Block Limits VPD */
2932 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2933 		goto out;
2934 
2935 	blk_queue_io_min(sdkp->disk->queue,
2936 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2937 
2938 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2939 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2940 
2941 	if (buffer[3] == 0x3c) {
2942 		unsigned int lba_count, desc_count;
2943 
2944 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2945 
2946 		if (!sdkp->lbpme)
2947 			goto out;
2948 
2949 		lba_count = get_unaligned_be32(&buffer[20]);
2950 		desc_count = get_unaligned_be32(&buffer[24]);
2951 
2952 		if (lba_count && desc_count)
2953 			sdkp->max_unmap_blocks = lba_count;
2954 
2955 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2956 
2957 		if (buffer[32] & 0x80)
2958 			sdkp->unmap_alignment =
2959 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2960 
2961 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2962 
2963 			if (sdkp->max_unmap_blocks)
2964 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2965 			else
2966 				sd_config_discard(sdkp, SD_LBP_WS16);
2967 
2968 		} else {	/* LBP VPD page tells us what to use */
2969 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2970 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2971 			else if (sdkp->lbpws)
2972 				sd_config_discard(sdkp, SD_LBP_WS16);
2973 			else if (sdkp->lbpws10)
2974 				sd_config_discard(sdkp, SD_LBP_WS10);
2975 			else
2976 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2977 		}
2978 	}
2979 
2980  out:
2981 	kfree(buffer);
2982 }
2983 
2984 /**
2985  * sd_read_block_characteristics - Query block dev. characteristics
2986  * @sdkp: disk to query
2987  */
2988 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2989 {
2990 	struct request_queue *q = sdkp->disk->queue;
2991 	unsigned char *buffer;
2992 	u16 rot;
2993 	const int vpd_len = 64;
2994 
2995 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2996 
2997 	if (!buffer ||
2998 	    /* Block Device Characteristics VPD */
2999 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
3000 		goto out;
3001 
3002 	rot = get_unaligned_be16(&buffer[4]);
3003 
3004 	if (rot == 1) {
3005 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3006 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3007 	}
3008 
3009 	if (sdkp->device->type == TYPE_ZBC) {
3010 		/* Host-managed */
3011 		blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3012 	} else {
3013 		sdkp->zoned = (buffer[8] >> 4) & 3;
3014 		if (sdkp->zoned == 1) {
3015 			/* Host-aware */
3016 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3017 		} else {
3018 			/* Regular disk or drive managed disk */
3019 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3020 		}
3021 	}
3022 
3023 	if (!sdkp->first_scan)
3024 		goto out;
3025 
3026 	if (blk_queue_is_zoned(q)) {
3027 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3028 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3029 	} else {
3030 		if (sdkp->zoned == 1)
3031 			sd_printk(KERN_NOTICE, sdkp,
3032 				  "Host-aware SMR disk used as regular disk\n");
3033 		else if (sdkp->zoned == 2)
3034 			sd_printk(KERN_NOTICE, sdkp,
3035 				  "Drive-managed SMR disk\n");
3036 	}
3037 
3038  out:
3039 	kfree(buffer);
3040 }
3041 
3042 /**
3043  * sd_read_block_provisioning - Query provisioning VPD page
3044  * @sdkp: disk to query
3045  */
3046 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3047 {
3048 	unsigned char *buffer;
3049 	const int vpd_len = 8;
3050 
3051 	if (sdkp->lbpme == 0)
3052 		return;
3053 
3054 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3055 
3056 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3057 		goto out;
3058 
3059 	sdkp->lbpvpd	= 1;
3060 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
3061 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
3062 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
3063 
3064  out:
3065 	kfree(buffer);
3066 }
3067 
3068 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3069 {
3070 	struct scsi_device *sdev = sdkp->device;
3071 
3072 	if (sdev->host->no_write_same) {
3073 		sdev->no_write_same = 1;
3074 
3075 		return;
3076 	}
3077 
3078 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3079 		/* too large values might cause issues with arcmsr */
3080 		int vpd_buf_len = 64;
3081 
3082 		sdev->no_report_opcodes = 1;
3083 
3084 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3085 		 * CODES is unsupported and the device has an ATA
3086 		 * Information VPD page (SAT).
3087 		 */
3088 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3089 			sdev->no_write_same = 1;
3090 	}
3091 
3092 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3093 		sdkp->ws16 = 1;
3094 
3095 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3096 		sdkp->ws10 = 1;
3097 }
3098 
3099 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3100 {
3101 	struct scsi_device *sdev = sdkp->device;
3102 
3103 	if (!sdev->security_supported)
3104 		return;
3105 
3106 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3107 			SECURITY_PROTOCOL_IN) == 1 &&
3108 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3109 			SECURITY_PROTOCOL_OUT) == 1)
3110 		sdkp->security = 1;
3111 }
3112 
3113 /*
3114  * Determine the device's preferred I/O size for reads and writes
3115  * unless the reported value is unreasonably small, large, not a
3116  * multiple of the physical block size, or simply garbage.
3117  */
3118 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3119 				      unsigned int dev_max)
3120 {
3121 	struct scsi_device *sdp = sdkp->device;
3122 	unsigned int opt_xfer_bytes =
3123 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3124 
3125 	if (sdkp->opt_xfer_blocks == 0)
3126 		return false;
3127 
3128 	if (sdkp->opt_xfer_blocks > dev_max) {
3129 		sd_first_printk(KERN_WARNING, sdkp,
3130 				"Optimal transfer size %u logical blocks " \
3131 				"> dev_max (%u logical blocks)\n",
3132 				sdkp->opt_xfer_blocks, dev_max);
3133 		return false;
3134 	}
3135 
3136 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3137 		sd_first_printk(KERN_WARNING, sdkp,
3138 				"Optimal transfer size %u logical blocks " \
3139 				"> sd driver limit (%u logical blocks)\n",
3140 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3141 		return false;
3142 	}
3143 
3144 	if (opt_xfer_bytes < PAGE_SIZE) {
3145 		sd_first_printk(KERN_WARNING, sdkp,
3146 				"Optimal transfer size %u bytes < " \
3147 				"PAGE_SIZE (%u bytes)\n",
3148 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3149 		return false;
3150 	}
3151 
3152 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3153 		sd_first_printk(KERN_WARNING, sdkp,
3154 				"Optimal transfer size %u bytes not a " \
3155 				"multiple of physical block size (%u bytes)\n",
3156 				opt_xfer_bytes, sdkp->physical_block_size);
3157 		return false;
3158 	}
3159 
3160 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3161 			opt_xfer_bytes);
3162 	return true;
3163 }
3164 
3165 /**
3166  *	sd_revalidate_disk - called the first time a new disk is seen,
3167  *	performs disk spin up, read_capacity, etc.
3168  *	@disk: struct gendisk we care about
3169  **/
3170 static int sd_revalidate_disk(struct gendisk *disk)
3171 {
3172 	struct scsi_disk *sdkp = scsi_disk(disk);
3173 	struct scsi_device *sdp = sdkp->device;
3174 	struct request_queue *q = sdkp->disk->queue;
3175 	sector_t old_capacity = sdkp->capacity;
3176 	unsigned char *buffer;
3177 	unsigned int dev_max, rw_max;
3178 
3179 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3180 				      "sd_revalidate_disk\n"));
3181 
3182 	/*
3183 	 * If the device is offline, don't try and read capacity or any
3184 	 * of the other niceties.
3185 	 */
3186 	if (!scsi_device_online(sdp))
3187 		goto out;
3188 
3189 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3190 	if (!buffer) {
3191 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3192 			  "allocation failure.\n");
3193 		goto out;
3194 	}
3195 
3196 	sd_spinup_disk(sdkp);
3197 
3198 	/*
3199 	 * Without media there is no reason to ask; moreover, some devices
3200 	 * react badly if we do.
3201 	 */
3202 	if (sdkp->media_present) {
3203 		sd_read_capacity(sdkp, buffer);
3204 
3205 		/*
3206 		 * set the default to rotational.  All non-rotational devices
3207 		 * support the block characteristics VPD page, which will
3208 		 * cause this to be updated correctly and any device which
3209 		 * doesn't support it should be treated as rotational.
3210 		 */
3211 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3212 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3213 
3214 		if (scsi_device_supports_vpd(sdp)) {
3215 			sd_read_block_provisioning(sdkp);
3216 			sd_read_block_limits(sdkp);
3217 			sd_read_block_characteristics(sdkp);
3218 			sd_zbc_read_zones(sdkp, buffer);
3219 		}
3220 
3221 		sd_print_capacity(sdkp, old_capacity);
3222 
3223 		sd_read_write_protect_flag(sdkp, buffer);
3224 		sd_read_cache_type(sdkp, buffer);
3225 		sd_read_app_tag_own(sdkp, buffer);
3226 		sd_read_write_same(sdkp, buffer);
3227 		sd_read_security(sdkp, buffer);
3228 	}
3229 
3230 	/*
3231 	 * We now have all cache related info, determine how we deal
3232 	 * with flush requests.
3233 	 */
3234 	sd_set_flush_flag(sdkp);
3235 
3236 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3237 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3238 
3239 	/* Some devices report a maximum block count for READ/WRITE requests. */
3240 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3241 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3242 
3243 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3244 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3245 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3246 	} else {
3247 		q->limits.io_opt = 0;
3248 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3249 				      (sector_t)BLK_DEF_MAX_SECTORS);
3250 	}
3251 
3252 	/* Do not exceed controller limit */
3253 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3254 
3255 	/*
3256 	 * Only update max_sectors if previously unset or if the current value
3257 	 * exceeds the capabilities of the hardware.
3258 	 */
3259 	if (sdkp->first_scan ||
3260 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3261 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3262 		q->limits.max_sectors = rw_max;
3263 
3264 	sdkp->first_scan = 0;
3265 
3266 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3267 	sd_config_write_same(sdkp);
3268 	kfree(buffer);
3269 
3270 	/*
3271 	 * For a zoned drive, revalidating the zones can be done only once
3272 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3273 	 * capacity to 0.
3274 	 */
3275 	if (sd_zbc_revalidate_zones(sdkp))
3276 		set_capacity_and_notify(disk, 0);
3277 
3278  out:
3279 	return 0;
3280 }
3281 
3282 /**
3283  *	sd_unlock_native_capacity - unlock native capacity
3284  *	@disk: struct gendisk to set capacity for
3285  *
3286  *	Block layer calls this function if it detects that partitions
3287  *	on @disk reach beyond the end of the device.  If the SCSI host
3288  *	implements ->unlock_native_capacity() method, it's invoked to
3289  *	give it a chance to adjust the device capacity.
3290  *
3291  *	CONTEXT:
3292  *	Defined by block layer.  Might sleep.
3293  */
3294 static void sd_unlock_native_capacity(struct gendisk *disk)
3295 {
3296 	struct scsi_device *sdev = scsi_disk(disk)->device;
3297 
3298 	if (sdev->host->hostt->unlock_native_capacity)
3299 		sdev->host->hostt->unlock_native_capacity(sdev);
3300 }
3301 
3302 /**
3303  *	sd_format_disk_name - format disk name
3304  *	@prefix: name prefix - ie. "sd" for SCSI disks
3305  *	@index: index of the disk to format name for
3306  *	@buf: output buffer
3307  *	@buflen: length of the output buffer
3308  *
3309  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3310  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3311  *	which is followed by sdaaa.
3312  *
3313  *	This is basically 26 base counting with one extra 'nil' entry
3314  *	at the beginning from the second digit on and can be
3315  *	determined using similar method as 26 base conversion with the
3316  *	index shifted -1 after each digit is computed.
3317  *
3318  *	CONTEXT:
3319  *	Don't care.
3320  *
3321  *	RETURNS:
3322  *	0 on success, -errno on failure.
3323  */
3324 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3325 {
3326 	const int base = 'z' - 'a' + 1;
3327 	char *begin = buf + strlen(prefix);
3328 	char *end = buf + buflen;
3329 	char *p;
3330 	int unit;
3331 
3332 	p = end - 1;
3333 	*p = '\0';
3334 	unit = base;
3335 	do {
3336 		if (p == begin)
3337 			return -EINVAL;
3338 		*--p = 'a' + (index % unit);
3339 		index = (index / unit) - 1;
3340 	} while (index >= 0);
3341 
3342 	memmove(begin, p, end - p);
3343 	memcpy(buf, prefix, strlen(prefix));
3344 
3345 	return 0;
3346 }
3347 
3348 /**
3349  *	sd_probe - called during driver initialization and whenever a
3350  *	new scsi device is attached to the system. It is called once
3351  *	for each scsi device (not just disks) present.
3352  *	@dev: pointer to device object
3353  *
3354  *	Returns 0 if successful (or not interested in this scsi device
3355  *	(e.g. scanner)); 1 when there is an error.
3356  *
3357  *	Note: this function is invoked from the scsi mid-level.
3358  *	This function sets up the mapping between a given
3359  *	<host,channel,id,lun> (found in sdp) and new device name
3360  *	(e.g. /dev/sda). More precisely it is the block device major
3361  *	and minor number that is chosen here.
3362  *
3363  *	Assume sd_probe is not re-entrant (for time being)
3364  *	Also think about sd_probe() and sd_remove() running coincidentally.
3365  **/
3366 static int sd_probe(struct device *dev)
3367 {
3368 	struct scsi_device *sdp = to_scsi_device(dev);
3369 	struct scsi_disk *sdkp;
3370 	struct gendisk *gd;
3371 	int index;
3372 	int error;
3373 
3374 	scsi_autopm_get_device(sdp);
3375 	error = -ENODEV;
3376 	if (sdp->type != TYPE_DISK &&
3377 	    sdp->type != TYPE_ZBC &&
3378 	    sdp->type != TYPE_MOD &&
3379 	    sdp->type != TYPE_RBC)
3380 		goto out;
3381 
3382 #ifndef CONFIG_BLK_DEV_ZONED
3383 	if (sdp->type == TYPE_ZBC)
3384 		goto out;
3385 #endif
3386 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3387 					"sd_probe\n"));
3388 
3389 	error = -ENOMEM;
3390 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3391 	if (!sdkp)
3392 		goto out;
3393 
3394 	gd = alloc_disk(SD_MINORS);
3395 	if (!gd)
3396 		goto out_free;
3397 
3398 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3399 	if (index < 0) {
3400 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3401 		goto out_put;
3402 	}
3403 
3404 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3405 	if (error) {
3406 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3407 		goto out_free_index;
3408 	}
3409 
3410 	sdkp->device = sdp;
3411 	sdkp->driver = &sd_template;
3412 	sdkp->disk = gd;
3413 	sdkp->index = index;
3414 	sdkp->max_retries = SD_MAX_RETRIES;
3415 	atomic_set(&sdkp->openers, 0);
3416 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3417 
3418 	if (!sdp->request_queue->rq_timeout) {
3419 		if (sdp->type != TYPE_MOD)
3420 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3421 		else
3422 			blk_queue_rq_timeout(sdp->request_queue,
3423 					     SD_MOD_TIMEOUT);
3424 	}
3425 
3426 	device_initialize(&sdkp->dev);
3427 	sdkp->dev.parent = dev;
3428 	sdkp->dev.class = &sd_disk_class;
3429 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3430 
3431 	error = device_add(&sdkp->dev);
3432 	if (error)
3433 		goto out_free_index;
3434 
3435 	get_device(dev);
3436 	dev_set_drvdata(dev, sdkp);
3437 
3438 	gd->major = sd_major((index & 0xf0) >> 4);
3439 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3440 
3441 	gd->fops = &sd_fops;
3442 	gd->private_data = &sdkp->driver;
3443 	gd->queue = sdkp->device->request_queue;
3444 
3445 	/* defaults, until the device tells us otherwise */
3446 	sdp->sector_size = 512;
3447 	sdkp->capacity = 0;
3448 	sdkp->media_present = 1;
3449 	sdkp->write_prot = 0;
3450 	sdkp->cache_override = 0;
3451 	sdkp->WCE = 0;
3452 	sdkp->RCD = 0;
3453 	sdkp->ATO = 0;
3454 	sdkp->first_scan = 1;
3455 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3456 
3457 	sd_revalidate_disk(gd);
3458 
3459 	gd->flags = GENHD_FL_EXT_DEVT;
3460 	if (sdp->removable) {
3461 		gd->flags |= GENHD_FL_REMOVABLE;
3462 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3463 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3464 	}
3465 
3466 	blk_pm_runtime_init(sdp->request_queue, dev);
3467 	if (sdp->rpm_autosuspend) {
3468 		pm_runtime_set_autosuspend_delay(dev,
3469 			sdp->host->hostt->rpm_autosuspend_delay);
3470 	}
3471 	device_add_disk(dev, gd, NULL);
3472 	if (sdkp->capacity)
3473 		sd_dif_config_host(sdkp);
3474 
3475 	sd_revalidate_disk(gd);
3476 
3477 	if (sdkp->security) {
3478 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3479 		if (sdkp->opal_dev)
3480 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3481 	}
3482 
3483 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3484 		  sdp->removable ? "removable " : "");
3485 	scsi_autopm_put_device(sdp);
3486 
3487 	return 0;
3488 
3489  out_free_index:
3490 	ida_free(&sd_index_ida, index);
3491  out_put:
3492 	put_disk(gd);
3493  out_free:
3494 	sd_zbc_release_disk(sdkp);
3495 	kfree(sdkp);
3496  out:
3497 	scsi_autopm_put_device(sdp);
3498 	return error;
3499 }
3500 
3501 /**
3502  *	sd_remove - called whenever a scsi disk (previously recognized by
3503  *	sd_probe) is detached from the system. It is called (potentially
3504  *	multiple times) during sd module unload.
3505  *	@dev: pointer to device object
3506  *
3507  *	Note: this function is invoked from the scsi mid-level.
3508  *	This function potentially frees up a device name (e.g. /dev/sdc)
3509  *	that could be re-used by a subsequent sd_probe().
3510  *	This function is not called when the built-in sd driver is "exit-ed".
3511  **/
3512 static int sd_remove(struct device *dev)
3513 {
3514 	struct scsi_disk *sdkp;
3515 
3516 	sdkp = dev_get_drvdata(dev);
3517 	scsi_autopm_get_device(sdkp->device);
3518 
3519 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3520 	device_del(&sdkp->dev);
3521 	del_gendisk(sdkp->disk);
3522 	sd_shutdown(dev);
3523 
3524 	free_opal_dev(sdkp->opal_dev);
3525 
3526 	mutex_lock(&sd_ref_mutex);
3527 	dev_set_drvdata(dev, NULL);
3528 	put_device(&sdkp->dev);
3529 	mutex_unlock(&sd_ref_mutex);
3530 
3531 	return 0;
3532 }
3533 
3534 /**
3535  *	scsi_disk_release - Called to free the scsi_disk structure
3536  *	@dev: pointer to embedded class device
3537  *
3538  *	sd_ref_mutex must be held entering this routine.  Because it is
3539  *	called on last put, you should always use the scsi_disk_get()
3540  *	scsi_disk_put() helpers which manipulate the semaphore directly
3541  *	and never do a direct put_device.
3542  **/
3543 static void scsi_disk_release(struct device *dev)
3544 {
3545 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3546 	struct gendisk *disk = sdkp->disk;
3547 	struct request_queue *q = disk->queue;
3548 
3549 	ida_free(&sd_index_ida, sdkp->index);
3550 
3551 	/*
3552 	 * Wait until all requests that are in progress have completed.
3553 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3554 	 * due to clearing the disk->private_data pointer. Wait from inside
3555 	 * scsi_disk_release() instead of from sd_release() to avoid that
3556 	 * freezing and unfreezing the request queue affects user space I/O
3557 	 * in case multiple processes open a /dev/sd... node concurrently.
3558 	 */
3559 	blk_mq_freeze_queue(q);
3560 	blk_mq_unfreeze_queue(q);
3561 
3562 	disk->private_data = NULL;
3563 	put_disk(disk);
3564 	put_device(&sdkp->device->sdev_gendev);
3565 
3566 	sd_zbc_release_disk(sdkp);
3567 
3568 	kfree(sdkp);
3569 }
3570 
3571 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3572 {
3573 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3574 	struct scsi_sense_hdr sshdr;
3575 	struct scsi_device *sdp = sdkp->device;
3576 	int res;
3577 
3578 	if (start)
3579 		cmd[4] |= 1;	/* START */
3580 
3581 	if (sdp->start_stop_pwr_cond)
3582 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3583 
3584 	if (!scsi_device_online(sdp))
3585 		return -ENODEV;
3586 
3587 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3588 			SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3589 	if (res) {
3590 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3591 		if (driver_byte(res) == DRIVER_SENSE)
3592 			sd_print_sense_hdr(sdkp, &sshdr);
3593 		if (scsi_sense_valid(&sshdr) &&
3594 			/* 0x3a is medium not present */
3595 			sshdr.asc == 0x3a)
3596 			res = 0;
3597 	}
3598 
3599 	/* SCSI error codes must not go to the generic layer */
3600 	if (res)
3601 		return -EIO;
3602 
3603 	return 0;
3604 }
3605 
3606 /*
3607  * Send a SYNCHRONIZE CACHE instruction down to the device through
3608  * the normal SCSI command structure.  Wait for the command to
3609  * complete.
3610  */
3611 static void sd_shutdown(struct device *dev)
3612 {
3613 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3614 
3615 	if (!sdkp)
3616 		return;         /* this can happen */
3617 
3618 	if (pm_runtime_suspended(dev))
3619 		return;
3620 
3621 	if (sdkp->WCE && sdkp->media_present) {
3622 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3623 		sd_sync_cache(sdkp, NULL);
3624 	}
3625 
3626 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3627 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3628 		sd_start_stop_device(sdkp, 0);
3629 	}
3630 }
3631 
3632 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3633 {
3634 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3635 	struct scsi_sense_hdr sshdr;
3636 	int ret = 0;
3637 
3638 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3639 		return 0;
3640 
3641 	if (sdkp->WCE && sdkp->media_present) {
3642 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3643 		ret = sd_sync_cache(sdkp, &sshdr);
3644 
3645 		if (ret) {
3646 			/* ignore OFFLINE device */
3647 			if (ret == -ENODEV)
3648 				return 0;
3649 
3650 			if (!scsi_sense_valid(&sshdr) ||
3651 			    sshdr.sense_key != ILLEGAL_REQUEST)
3652 				return ret;
3653 
3654 			/*
3655 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3656 			 * doesn't support sync. There's not much to do and
3657 			 * suspend shouldn't fail.
3658 			 */
3659 			ret = 0;
3660 		}
3661 	}
3662 
3663 	if (sdkp->device->manage_start_stop) {
3664 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3665 		/* an error is not worth aborting a system sleep */
3666 		ret = sd_start_stop_device(sdkp, 0);
3667 		if (ignore_stop_errors)
3668 			ret = 0;
3669 	}
3670 
3671 	return ret;
3672 }
3673 
3674 static int sd_suspend_system(struct device *dev)
3675 {
3676 	return sd_suspend_common(dev, true);
3677 }
3678 
3679 static int sd_suspend_runtime(struct device *dev)
3680 {
3681 	return sd_suspend_common(dev, false);
3682 }
3683 
3684 static int sd_resume(struct device *dev)
3685 {
3686 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3687 	int ret;
3688 
3689 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3690 		return 0;
3691 
3692 	if (!sdkp->device->manage_start_stop)
3693 		return 0;
3694 
3695 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3696 	ret = sd_start_stop_device(sdkp, 1);
3697 	if (!ret)
3698 		opal_unlock_from_suspend(sdkp->opal_dev);
3699 	return ret;
3700 }
3701 
3702 /**
3703  *	init_sd - entry point for this driver (both when built in or when
3704  *	a module).
3705  *
3706  *	Note: this function registers this driver with the scsi mid-level.
3707  **/
3708 static int __init init_sd(void)
3709 {
3710 	int majors = 0, i, err;
3711 
3712 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3713 
3714 	for (i = 0; i < SD_MAJORS; i++) {
3715 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3716 			continue;
3717 		majors++;
3718 	}
3719 
3720 	if (!majors)
3721 		return -ENODEV;
3722 
3723 	err = class_register(&sd_disk_class);
3724 	if (err)
3725 		goto err_out;
3726 
3727 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3728 					 0, 0, NULL);
3729 	if (!sd_cdb_cache) {
3730 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3731 		err = -ENOMEM;
3732 		goto err_out_class;
3733 	}
3734 
3735 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3736 	if (!sd_cdb_pool) {
3737 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3738 		err = -ENOMEM;
3739 		goto err_out_cache;
3740 	}
3741 
3742 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3743 	if (!sd_page_pool) {
3744 		printk(KERN_ERR "sd: can't init discard page pool\n");
3745 		err = -ENOMEM;
3746 		goto err_out_ppool;
3747 	}
3748 
3749 	err = scsi_register_driver(&sd_template.gendrv);
3750 	if (err)
3751 		goto err_out_driver;
3752 
3753 	return 0;
3754 
3755 err_out_driver:
3756 	mempool_destroy(sd_page_pool);
3757 
3758 err_out_ppool:
3759 	mempool_destroy(sd_cdb_pool);
3760 
3761 err_out_cache:
3762 	kmem_cache_destroy(sd_cdb_cache);
3763 
3764 err_out_class:
3765 	class_unregister(&sd_disk_class);
3766 err_out:
3767 	for (i = 0; i < SD_MAJORS; i++)
3768 		unregister_blkdev(sd_major(i), "sd");
3769 	return err;
3770 }
3771 
3772 /**
3773  *	exit_sd - exit point for this driver (when it is a module).
3774  *
3775  *	Note: this function unregisters this driver from the scsi mid-level.
3776  **/
3777 static void __exit exit_sd(void)
3778 {
3779 	int i;
3780 
3781 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3782 
3783 	scsi_unregister_driver(&sd_template.gendrv);
3784 	mempool_destroy(sd_cdb_pool);
3785 	mempool_destroy(sd_page_pool);
3786 	kmem_cache_destroy(sd_cdb_cache);
3787 
3788 	class_unregister(&sd_disk_class);
3789 
3790 	for (i = 0; i < SD_MAJORS; i++)
3791 		unregister_blkdev(sd_major(i), "sd");
3792 }
3793 
3794 module_init(init_sd);
3795 module_exit(exit_sd);
3796 
3797 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3798 {
3799 	scsi_print_sense_hdr(sdkp->device,
3800 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3801 }
3802 
3803 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3804 {
3805 	const char *hb_string = scsi_hostbyte_string(result);
3806 	const char *db_string = scsi_driverbyte_string(result);
3807 
3808 	if (hb_string || db_string)
3809 		sd_printk(KERN_INFO, sdkp,
3810 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3811 			  hb_string ? hb_string : "invalid",
3812 			  db_string ? db_string : "invalid");
3813 	else
3814 		sd_printk(KERN_INFO, sdkp,
3815 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3816 			  msg, host_byte(result), driver_byte(result));
3817 }
3818