xref: /linux/drivers/scsi/sd.c (revision ba1dc7f273c73b93e0e1dd9707b239ed69eebd70)
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 #define SD_MINORS	16
102 
103 static void sd_config_discard(struct scsi_disk *, unsigned int);
104 static void sd_config_write_same(struct scsi_disk *);
105 static int  sd_revalidate_disk(struct gendisk *);
106 static void sd_unlock_native_capacity(struct gendisk *disk);
107 static int  sd_probe(struct device *);
108 static int  sd_remove(struct device *);
109 static void sd_shutdown(struct device *);
110 static int sd_suspend_system(struct device *);
111 static int sd_suspend_runtime(struct device *);
112 static int sd_resume(struct device *);
113 static void sd_rescan(struct device *);
114 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
115 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
116 static int sd_done(struct scsi_cmnd *);
117 static void sd_eh_reset(struct scsi_cmnd *);
118 static int sd_eh_action(struct scsi_cmnd *, int);
119 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
120 static void scsi_disk_release(struct device *cdev);
121 
122 static DEFINE_IDA(sd_index_ida);
123 
124 /* This semaphore is used to mediate the 0->1 reference get in the
125  * face of object destruction (i.e. we can't allow a get on an
126  * object after last put) */
127 static DEFINE_MUTEX(sd_ref_mutex);
128 
129 static struct kmem_cache *sd_cdb_cache;
130 static mempool_t *sd_cdb_pool;
131 static mempool_t *sd_page_pool;
132 static struct lock_class_key sd_bio_compl_lkclass;
133 
134 static const char *sd_cache_types[] = {
135 	"write through", "none", "write back",
136 	"write back, no read (daft)"
137 };
138 
139 static void sd_set_flush_flag(struct scsi_disk *sdkp)
140 {
141 	bool wc = false, fua = false;
142 
143 	if (sdkp->WCE) {
144 		wc = true;
145 		if (sdkp->DPOFUA)
146 			fua = true;
147 	}
148 
149 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
150 }
151 
152 static ssize_t
153 cache_type_store(struct device *dev, struct device_attribute *attr,
154 		 const char *buf, size_t count)
155 {
156 	int ct, rcd, wce, sp;
157 	struct scsi_disk *sdkp = to_scsi_disk(dev);
158 	struct scsi_device *sdp = sdkp->device;
159 	char buffer[64];
160 	char *buffer_data;
161 	struct scsi_mode_data data;
162 	struct scsi_sense_hdr sshdr;
163 	static const char temp[] = "temporary ";
164 	int len;
165 
166 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
167 		/* no cache control on RBC devices; theoretically they
168 		 * can do it, but there's probably so many exceptions
169 		 * it's not worth the risk */
170 		return -EINVAL;
171 
172 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
173 		buf += sizeof(temp) - 1;
174 		sdkp->cache_override = 1;
175 	} else {
176 		sdkp->cache_override = 0;
177 	}
178 
179 	ct = sysfs_match_string(sd_cache_types, buf);
180 	if (ct < 0)
181 		return -EINVAL;
182 
183 	rcd = ct & 0x01 ? 1 : 0;
184 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
185 
186 	if (sdkp->cache_override) {
187 		sdkp->WCE = wce;
188 		sdkp->RCD = rcd;
189 		sd_set_flush_flag(sdkp);
190 		return count;
191 	}
192 
193 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
194 			    sdkp->max_retries, &data, NULL))
195 		return -EINVAL;
196 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
197 		  data.block_descriptor_length);
198 	buffer_data = buffer + data.header_length +
199 		data.block_descriptor_length;
200 	buffer_data[2] &= ~0x05;
201 	buffer_data[2] |= wce << 2 | rcd;
202 	sp = buffer_data[0] & 0x80 ? 1 : 0;
203 	buffer_data[0] &= ~0x80;
204 
205 	/*
206 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
207 	 * received mode parameter buffer before doing MODE SELECT.
208 	 */
209 	data.device_specific = 0;
210 
211 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
212 			     sdkp->max_retries, &data, &sshdr)) {
213 		if (scsi_sense_valid(&sshdr))
214 			sd_print_sense_hdr(sdkp, &sshdr);
215 		return -EINVAL;
216 	}
217 	sd_revalidate_disk(sdkp->disk);
218 	return count;
219 }
220 
221 static ssize_t
222 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
223 		       char *buf)
224 {
225 	struct scsi_disk *sdkp = to_scsi_disk(dev);
226 	struct scsi_device *sdp = sdkp->device;
227 
228 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
229 }
230 
231 static ssize_t
232 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
233 			const char *buf, size_t count)
234 {
235 	struct scsi_disk *sdkp = to_scsi_disk(dev);
236 	struct scsi_device *sdp = sdkp->device;
237 	bool v;
238 
239 	if (!capable(CAP_SYS_ADMIN))
240 		return -EACCES;
241 
242 	if (kstrtobool(buf, &v))
243 		return -EINVAL;
244 
245 	sdp->manage_start_stop = v;
246 
247 	return count;
248 }
249 static DEVICE_ATTR_RW(manage_start_stop);
250 
251 static ssize_t
252 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
253 {
254 	struct scsi_disk *sdkp = to_scsi_disk(dev);
255 
256 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
257 }
258 
259 static ssize_t
260 allow_restart_store(struct device *dev, struct device_attribute *attr,
261 		    const char *buf, size_t count)
262 {
263 	bool v;
264 	struct scsi_disk *sdkp = to_scsi_disk(dev);
265 	struct scsi_device *sdp = sdkp->device;
266 
267 	if (!capable(CAP_SYS_ADMIN))
268 		return -EACCES;
269 
270 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
271 		return -EINVAL;
272 
273 	if (kstrtobool(buf, &v))
274 		return -EINVAL;
275 
276 	sdp->allow_restart = v;
277 
278 	return count;
279 }
280 static DEVICE_ATTR_RW(allow_restart);
281 
282 static ssize_t
283 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
284 {
285 	struct scsi_disk *sdkp = to_scsi_disk(dev);
286 	int ct = sdkp->RCD + 2*sdkp->WCE;
287 
288 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
289 }
290 static DEVICE_ATTR_RW(cache_type);
291 
292 static ssize_t
293 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
294 {
295 	struct scsi_disk *sdkp = to_scsi_disk(dev);
296 
297 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
298 }
299 static DEVICE_ATTR_RO(FUA);
300 
301 static ssize_t
302 protection_type_show(struct device *dev, struct device_attribute *attr,
303 		     char *buf)
304 {
305 	struct scsi_disk *sdkp = to_scsi_disk(dev);
306 
307 	return sprintf(buf, "%u\n", sdkp->protection_type);
308 }
309 
310 static ssize_t
311 protection_type_store(struct device *dev, struct device_attribute *attr,
312 		      const char *buf, size_t count)
313 {
314 	struct scsi_disk *sdkp = to_scsi_disk(dev);
315 	unsigned int val;
316 	int err;
317 
318 	if (!capable(CAP_SYS_ADMIN))
319 		return -EACCES;
320 
321 	err = kstrtouint(buf, 10, &val);
322 
323 	if (err)
324 		return err;
325 
326 	if (val <= T10_PI_TYPE3_PROTECTION)
327 		sdkp->protection_type = val;
328 
329 	return count;
330 }
331 static DEVICE_ATTR_RW(protection_type);
332 
333 static ssize_t
334 protection_mode_show(struct device *dev, struct device_attribute *attr,
335 		     char *buf)
336 {
337 	struct scsi_disk *sdkp = to_scsi_disk(dev);
338 	struct scsi_device *sdp = sdkp->device;
339 	unsigned int dif, dix;
340 
341 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
342 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
343 
344 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
345 		dif = 0;
346 		dix = 1;
347 	}
348 
349 	if (!dif && !dix)
350 		return sprintf(buf, "none\n");
351 
352 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
353 }
354 static DEVICE_ATTR_RO(protection_mode);
355 
356 static ssize_t
357 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
358 {
359 	struct scsi_disk *sdkp = to_scsi_disk(dev);
360 
361 	return sprintf(buf, "%u\n", sdkp->ATO);
362 }
363 static DEVICE_ATTR_RO(app_tag_own);
364 
365 static ssize_t
366 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
367 		       char *buf)
368 {
369 	struct scsi_disk *sdkp = to_scsi_disk(dev);
370 
371 	return sprintf(buf, "%u\n", sdkp->lbpme);
372 }
373 static DEVICE_ATTR_RO(thin_provisioning);
374 
375 /* sysfs_match_string() requires dense arrays */
376 static const char *lbp_mode[] = {
377 	[SD_LBP_FULL]		= "full",
378 	[SD_LBP_UNMAP]		= "unmap",
379 	[SD_LBP_WS16]		= "writesame_16",
380 	[SD_LBP_WS10]		= "writesame_10",
381 	[SD_LBP_ZERO]		= "writesame_zero",
382 	[SD_LBP_DISABLE]	= "disabled",
383 };
384 
385 static ssize_t
386 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
387 		       char *buf)
388 {
389 	struct scsi_disk *sdkp = to_scsi_disk(dev);
390 
391 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
392 }
393 
394 static ssize_t
395 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
396 			const char *buf, size_t count)
397 {
398 	struct scsi_disk *sdkp = to_scsi_disk(dev);
399 	struct scsi_device *sdp = sdkp->device;
400 	int mode;
401 
402 	if (!capable(CAP_SYS_ADMIN))
403 		return -EACCES;
404 
405 	if (sd_is_zoned(sdkp)) {
406 		sd_config_discard(sdkp, SD_LBP_DISABLE);
407 		return count;
408 	}
409 
410 	if (sdp->type != TYPE_DISK)
411 		return -EINVAL;
412 
413 	mode = sysfs_match_string(lbp_mode, buf);
414 	if (mode < 0)
415 		return -EINVAL;
416 
417 	sd_config_discard(sdkp, mode);
418 
419 	return count;
420 }
421 static DEVICE_ATTR_RW(provisioning_mode);
422 
423 /* sysfs_match_string() requires dense arrays */
424 static const char *zeroing_mode[] = {
425 	[SD_ZERO_WRITE]		= "write",
426 	[SD_ZERO_WS]		= "writesame",
427 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
428 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
429 };
430 
431 static ssize_t
432 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
433 		  char *buf)
434 {
435 	struct scsi_disk *sdkp = to_scsi_disk(dev);
436 
437 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
438 }
439 
440 static ssize_t
441 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
442 		   const char *buf, size_t count)
443 {
444 	struct scsi_disk *sdkp = to_scsi_disk(dev);
445 	int mode;
446 
447 	if (!capable(CAP_SYS_ADMIN))
448 		return -EACCES;
449 
450 	mode = sysfs_match_string(zeroing_mode, buf);
451 	if (mode < 0)
452 		return -EINVAL;
453 
454 	sdkp->zeroing_mode = mode;
455 
456 	return count;
457 }
458 static DEVICE_ATTR_RW(zeroing_mode);
459 
460 static ssize_t
461 max_medium_access_timeouts_show(struct device *dev,
462 				struct device_attribute *attr, char *buf)
463 {
464 	struct scsi_disk *sdkp = to_scsi_disk(dev);
465 
466 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
467 }
468 
469 static ssize_t
470 max_medium_access_timeouts_store(struct device *dev,
471 				 struct device_attribute *attr, const char *buf,
472 				 size_t count)
473 {
474 	struct scsi_disk *sdkp = to_scsi_disk(dev);
475 	int err;
476 
477 	if (!capable(CAP_SYS_ADMIN))
478 		return -EACCES;
479 
480 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
481 
482 	return err ? err : count;
483 }
484 static DEVICE_ATTR_RW(max_medium_access_timeouts);
485 
486 static ssize_t
487 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
488 			   char *buf)
489 {
490 	struct scsi_disk *sdkp = to_scsi_disk(dev);
491 
492 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
493 }
494 
495 static ssize_t
496 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
497 			    const char *buf, size_t count)
498 {
499 	struct scsi_disk *sdkp = to_scsi_disk(dev);
500 	struct scsi_device *sdp = sdkp->device;
501 	unsigned long max;
502 	int err;
503 
504 	if (!capable(CAP_SYS_ADMIN))
505 		return -EACCES;
506 
507 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
508 		return -EINVAL;
509 
510 	err = kstrtoul(buf, 10, &max);
511 
512 	if (err)
513 		return err;
514 
515 	if (max == 0)
516 		sdp->no_write_same = 1;
517 	else if (max <= SD_MAX_WS16_BLOCKS) {
518 		sdp->no_write_same = 0;
519 		sdkp->max_ws_blocks = max;
520 	}
521 
522 	sd_config_write_same(sdkp);
523 
524 	return count;
525 }
526 static DEVICE_ATTR_RW(max_write_same_blocks);
527 
528 static ssize_t
529 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
530 {
531 	struct scsi_disk *sdkp = to_scsi_disk(dev);
532 
533 	if (sdkp->device->type == TYPE_ZBC)
534 		return sprintf(buf, "host-managed\n");
535 	if (sdkp->zoned == 1)
536 		return sprintf(buf, "host-aware\n");
537 	if (sdkp->zoned == 2)
538 		return sprintf(buf, "drive-managed\n");
539 	return sprintf(buf, "none\n");
540 }
541 static DEVICE_ATTR_RO(zoned_cap);
542 
543 static ssize_t
544 max_retries_store(struct device *dev, struct device_attribute *attr,
545 		  const char *buf, size_t count)
546 {
547 	struct scsi_disk *sdkp = to_scsi_disk(dev);
548 	struct scsi_device *sdev = sdkp->device;
549 	int retries, err;
550 
551 	err = kstrtoint(buf, 10, &retries);
552 	if (err)
553 		return err;
554 
555 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
556 		sdkp->max_retries = retries;
557 		return count;
558 	}
559 
560 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
561 		    SD_MAX_RETRIES);
562 	return -EINVAL;
563 }
564 
565 static ssize_t
566 max_retries_show(struct device *dev, struct device_attribute *attr,
567 		 char *buf)
568 {
569 	struct scsi_disk *sdkp = to_scsi_disk(dev);
570 
571 	return sprintf(buf, "%d\n", sdkp->max_retries);
572 }
573 
574 static DEVICE_ATTR_RW(max_retries);
575 
576 static struct attribute *sd_disk_attrs[] = {
577 	&dev_attr_cache_type.attr,
578 	&dev_attr_FUA.attr,
579 	&dev_attr_allow_restart.attr,
580 	&dev_attr_manage_start_stop.attr,
581 	&dev_attr_protection_type.attr,
582 	&dev_attr_protection_mode.attr,
583 	&dev_attr_app_tag_own.attr,
584 	&dev_attr_thin_provisioning.attr,
585 	&dev_attr_provisioning_mode.attr,
586 	&dev_attr_zeroing_mode.attr,
587 	&dev_attr_max_write_same_blocks.attr,
588 	&dev_attr_max_medium_access_timeouts.attr,
589 	&dev_attr_zoned_cap.attr,
590 	&dev_attr_max_retries.attr,
591 	NULL,
592 };
593 ATTRIBUTE_GROUPS(sd_disk);
594 
595 static struct class sd_disk_class = {
596 	.name		= "scsi_disk",
597 	.owner		= THIS_MODULE,
598 	.dev_release	= scsi_disk_release,
599 	.dev_groups	= sd_disk_groups,
600 };
601 
602 static const struct dev_pm_ops sd_pm_ops = {
603 	.suspend		= sd_suspend_system,
604 	.resume			= sd_resume,
605 	.poweroff		= sd_suspend_system,
606 	.restore		= sd_resume,
607 	.runtime_suspend	= sd_suspend_runtime,
608 	.runtime_resume		= sd_resume,
609 };
610 
611 static struct scsi_driver sd_template = {
612 	.gendrv = {
613 		.name		= "sd",
614 		.owner		= THIS_MODULE,
615 		.probe		= sd_probe,
616 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
617 		.remove		= sd_remove,
618 		.shutdown	= sd_shutdown,
619 		.pm		= &sd_pm_ops,
620 	},
621 	.rescan			= sd_rescan,
622 	.init_command		= sd_init_command,
623 	.uninit_command		= sd_uninit_command,
624 	.done			= sd_done,
625 	.eh_action		= sd_eh_action,
626 	.eh_reset		= sd_eh_reset,
627 };
628 
629 /*
630  * Don't request a new module, as that could deadlock in multipath
631  * environment.
632  */
633 static void sd_default_probe(dev_t devt)
634 {
635 }
636 
637 /*
638  * Device no to disk mapping:
639  *
640  *       major         disc2     disc  p1
641  *   |............|.............|....|....| <- dev_t
642  *    31        20 19          8 7  4 3  0
643  *
644  * Inside a major, we have 16k disks, however mapped non-
645  * contiguously. The first 16 disks are for major0, the next
646  * ones with major1, ... Disk 256 is for major0 again, disk 272
647  * for major1, ...
648  * As we stay compatible with our numbering scheme, we can reuse
649  * the well-know SCSI majors 8, 65--71, 136--143.
650  */
651 static int sd_major(int major_idx)
652 {
653 	switch (major_idx) {
654 	case 0:
655 		return SCSI_DISK0_MAJOR;
656 	case 1 ... 7:
657 		return SCSI_DISK1_MAJOR + major_idx - 1;
658 	case 8 ... 15:
659 		return SCSI_DISK8_MAJOR + major_idx - 8;
660 	default:
661 		BUG();
662 		return 0;	/* shut up gcc */
663 	}
664 }
665 
666 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
667 {
668 	struct scsi_disk *sdkp = NULL;
669 
670 	mutex_lock(&sd_ref_mutex);
671 
672 	if (disk->private_data) {
673 		sdkp = scsi_disk(disk);
674 		if (scsi_device_get(sdkp->device) == 0)
675 			get_device(&sdkp->dev);
676 		else
677 			sdkp = NULL;
678 	}
679 	mutex_unlock(&sd_ref_mutex);
680 	return sdkp;
681 }
682 
683 static void scsi_disk_put(struct scsi_disk *sdkp)
684 {
685 	struct scsi_device *sdev = sdkp->device;
686 
687 	mutex_lock(&sd_ref_mutex);
688 	put_device(&sdkp->dev);
689 	scsi_device_put(sdev);
690 	mutex_unlock(&sd_ref_mutex);
691 }
692 
693 #ifdef CONFIG_BLK_SED_OPAL
694 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
695 		size_t len, bool send)
696 {
697 	struct scsi_disk *sdkp = data;
698 	struct scsi_device *sdev = sdkp->device;
699 	u8 cdb[12] = { 0, };
700 	int ret;
701 
702 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
703 	cdb[1] = secp;
704 	put_unaligned_be16(spsp, &cdb[2]);
705 	put_unaligned_be32(len, &cdb[6]);
706 
707 	ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
708 		buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
709 		RQF_PM, NULL);
710 	return ret <= 0 ? ret : -EIO;
711 }
712 #endif /* CONFIG_BLK_SED_OPAL */
713 
714 /*
715  * Look up the DIX operation based on whether the command is read or
716  * write and whether dix and dif are enabled.
717  */
718 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
719 {
720 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
721 	static const unsigned int ops[] = {	/* wrt dix dif */
722 		SCSI_PROT_NORMAL,		/*  0	0   0  */
723 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
724 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
725 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
726 		SCSI_PROT_NORMAL,		/*  1	0   0  */
727 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
728 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
729 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
730 	};
731 
732 	return ops[write << 2 | dix << 1 | dif];
733 }
734 
735 /*
736  * Returns a mask of the protection flags that are valid for a given DIX
737  * operation.
738  */
739 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
740 {
741 	static const unsigned int flag_mask[] = {
742 		[SCSI_PROT_NORMAL]		= 0,
743 
744 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
745 						  SCSI_PROT_GUARD_CHECK |
746 						  SCSI_PROT_REF_CHECK |
747 						  SCSI_PROT_REF_INCREMENT,
748 
749 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
750 						  SCSI_PROT_IP_CHECKSUM,
751 
752 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
753 						  SCSI_PROT_GUARD_CHECK |
754 						  SCSI_PROT_REF_CHECK |
755 						  SCSI_PROT_REF_INCREMENT |
756 						  SCSI_PROT_IP_CHECKSUM,
757 
758 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
759 						  SCSI_PROT_REF_INCREMENT,
760 
761 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
762 						  SCSI_PROT_REF_CHECK |
763 						  SCSI_PROT_REF_INCREMENT |
764 						  SCSI_PROT_IP_CHECKSUM,
765 
766 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
767 						  SCSI_PROT_GUARD_CHECK |
768 						  SCSI_PROT_REF_CHECK |
769 						  SCSI_PROT_REF_INCREMENT |
770 						  SCSI_PROT_IP_CHECKSUM,
771 	};
772 
773 	return flag_mask[prot_op];
774 }
775 
776 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
777 					   unsigned int dix, unsigned int dif)
778 {
779 	struct bio *bio = scmd->request->bio;
780 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
781 	unsigned int protect = 0;
782 
783 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
784 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
785 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
786 
787 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
788 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
789 	}
790 
791 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
792 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
793 
794 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
795 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
796 	}
797 
798 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
799 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
800 
801 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
802 			protect = 3 << 5;	/* Disable target PI checking */
803 		else
804 			protect = 1 << 5;	/* Enable target PI checking */
805 	}
806 
807 	scsi_set_prot_op(scmd, prot_op);
808 	scsi_set_prot_type(scmd, dif);
809 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
810 
811 	return protect;
812 }
813 
814 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
815 {
816 	struct request_queue *q = sdkp->disk->queue;
817 	unsigned int logical_block_size = sdkp->device->sector_size;
818 	unsigned int max_blocks = 0;
819 
820 	q->limits.discard_alignment =
821 		sdkp->unmap_alignment * logical_block_size;
822 	q->limits.discard_granularity =
823 		max(sdkp->physical_block_size,
824 		    sdkp->unmap_granularity * logical_block_size);
825 	sdkp->provisioning_mode = mode;
826 
827 	switch (mode) {
828 
829 	case SD_LBP_FULL:
830 	case SD_LBP_DISABLE:
831 		blk_queue_max_discard_sectors(q, 0);
832 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
833 		return;
834 
835 	case SD_LBP_UNMAP:
836 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
837 					  (u32)SD_MAX_WS16_BLOCKS);
838 		break;
839 
840 	case SD_LBP_WS16:
841 		if (sdkp->device->unmap_limit_for_ws)
842 			max_blocks = sdkp->max_unmap_blocks;
843 		else
844 			max_blocks = sdkp->max_ws_blocks;
845 
846 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
847 		break;
848 
849 	case SD_LBP_WS10:
850 		if (sdkp->device->unmap_limit_for_ws)
851 			max_blocks = sdkp->max_unmap_blocks;
852 		else
853 			max_blocks = sdkp->max_ws_blocks;
854 
855 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
856 		break;
857 
858 	case SD_LBP_ZERO:
859 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
860 					  (u32)SD_MAX_WS10_BLOCKS);
861 		break;
862 	}
863 
864 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
865 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
866 }
867 
868 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
869 {
870 	struct scsi_device *sdp = cmd->device;
871 	struct request *rq = cmd->request;
872 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
873 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
874 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
875 	unsigned int data_len = 24;
876 	char *buf;
877 
878 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
879 	if (!rq->special_vec.bv_page)
880 		return BLK_STS_RESOURCE;
881 	clear_highpage(rq->special_vec.bv_page);
882 	rq->special_vec.bv_offset = 0;
883 	rq->special_vec.bv_len = data_len;
884 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
885 
886 	cmd->cmd_len = 10;
887 	cmd->cmnd[0] = UNMAP;
888 	cmd->cmnd[8] = 24;
889 
890 	buf = bvec_virt(&rq->special_vec);
891 	put_unaligned_be16(6 + 16, &buf[0]);
892 	put_unaligned_be16(16, &buf[2]);
893 	put_unaligned_be64(lba, &buf[8]);
894 	put_unaligned_be32(nr_blocks, &buf[16]);
895 
896 	cmd->allowed = sdkp->max_retries;
897 	cmd->transfersize = data_len;
898 	rq->timeout = SD_TIMEOUT;
899 
900 	return scsi_alloc_sgtables(cmd);
901 }
902 
903 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
904 		bool unmap)
905 {
906 	struct scsi_device *sdp = cmd->device;
907 	struct request *rq = cmd->request;
908 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
909 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
910 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
911 	u32 data_len = sdp->sector_size;
912 
913 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
914 	if (!rq->special_vec.bv_page)
915 		return BLK_STS_RESOURCE;
916 	clear_highpage(rq->special_vec.bv_page);
917 	rq->special_vec.bv_offset = 0;
918 	rq->special_vec.bv_len = data_len;
919 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
920 
921 	cmd->cmd_len = 16;
922 	cmd->cmnd[0] = WRITE_SAME_16;
923 	if (unmap)
924 		cmd->cmnd[1] = 0x8; /* UNMAP */
925 	put_unaligned_be64(lba, &cmd->cmnd[2]);
926 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
927 
928 	cmd->allowed = sdkp->max_retries;
929 	cmd->transfersize = data_len;
930 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
931 
932 	return scsi_alloc_sgtables(cmd);
933 }
934 
935 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
936 		bool unmap)
937 {
938 	struct scsi_device *sdp = cmd->device;
939 	struct request *rq = cmd->request;
940 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
941 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
942 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
943 	u32 data_len = sdp->sector_size;
944 
945 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
946 	if (!rq->special_vec.bv_page)
947 		return BLK_STS_RESOURCE;
948 	clear_highpage(rq->special_vec.bv_page);
949 	rq->special_vec.bv_offset = 0;
950 	rq->special_vec.bv_len = data_len;
951 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
952 
953 	cmd->cmd_len = 10;
954 	cmd->cmnd[0] = WRITE_SAME;
955 	if (unmap)
956 		cmd->cmnd[1] = 0x8; /* UNMAP */
957 	put_unaligned_be32(lba, &cmd->cmnd[2]);
958 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
959 
960 	cmd->allowed = sdkp->max_retries;
961 	cmd->transfersize = data_len;
962 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
963 
964 	return scsi_alloc_sgtables(cmd);
965 }
966 
967 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
968 {
969 	struct request *rq = cmd->request;
970 	struct scsi_device *sdp = cmd->device;
971 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
972 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
973 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
974 
975 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
976 		switch (sdkp->zeroing_mode) {
977 		case SD_ZERO_WS16_UNMAP:
978 			return sd_setup_write_same16_cmnd(cmd, true);
979 		case SD_ZERO_WS10_UNMAP:
980 			return sd_setup_write_same10_cmnd(cmd, true);
981 		}
982 	}
983 
984 	if (sdp->no_write_same) {
985 		rq->rq_flags |= RQF_QUIET;
986 		return BLK_STS_TARGET;
987 	}
988 
989 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
990 		return sd_setup_write_same16_cmnd(cmd, false);
991 
992 	return sd_setup_write_same10_cmnd(cmd, false);
993 }
994 
995 static void sd_config_write_same(struct scsi_disk *sdkp)
996 {
997 	struct request_queue *q = sdkp->disk->queue;
998 	unsigned int logical_block_size = sdkp->device->sector_size;
999 
1000 	if (sdkp->device->no_write_same) {
1001 		sdkp->max_ws_blocks = 0;
1002 		goto out;
1003 	}
1004 
1005 	/* Some devices can not handle block counts above 0xffff despite
1006 	 * supporting WRITE SAME(16). Consequently we default to 64k
1007 	 * blocks per I/O unless the device explicitly advertises a
1008 	 * bigger limit.
1009 	 */
1010 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1011 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1012 						   (u32)SD_MAX_WS16_BLOCKS);
1013 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1014 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1015 						   (u32)SD_MAX_WS10_BLOCKS);
1016 	else {
1017 		sdkp->device->no_write_same = 1;
1018 		sdkp->max_ws_blocks = 0;
1019 	}
1020 
1021 	if (sdkp->lbprz && sdkp->lbpws)
1022 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1023 	else if (sdkp->lbprz && sdkp->lbpws10)
1024 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1025 	else if (sdkp->max_ws_blocks)
1026 		sdkp->zeroing_mode = SD_ZERO_WS;
1027 	else
1028 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1029 
1030 	if (sdkp->max_ws_blocks &&
1031 	    sdkp->physical_block_size > logical_block_size) {
1032 		/*
1033 		 * Reporting a maximum number of blocks that is not aligned
1034 		 * on the device physical size would cause a large write same
1035 		 * request to be split into physically unaligned chunks by
1036 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1037 		 * even if the caller of these functions took care to align the
1038 		 * large request. So make sure the maximum reported is aligned
1039 		 * to the device physical block size. This is only an optional
1040 		 * optimization for regular disks, but this is mandatory to
1041 		 * avoid failure of large write same requests directed at
1042 		 * sequential write required zones of host-managed ZBC disks.
1043 		 */
1044 		sdkp->max_ws_blocks =
1045 			round_down(sdkp->max_ws_blocks,
1046 				   bytes_to_logical(sdkp->device,
1047 						    sdkp->physical_block_size));
1048 	}
1049 
1050 out:
1051 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1052 					 (logical_block_size >> 9));
1053 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1054 					 (logical_block_size >> 9));
1055 }
1056 
1057 /**
1058  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1059  * @cmd: command to prepare
1060  *
1061  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1062  * the preference indicated by the target device.
1063  **/
1064 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1065 {
1066 	struct request *rq = cmd->request;
1067 	struct scsi_device *sdp = cmd->device;
1068 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1069 	struct bio *bio = rq->bio;
1070 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1071 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1072 	blk_status_t ret;
1073 
1074 	if (sdkp->device->no_write_same)
1075 		return BLK_STS_TARGET;
1076 
1077 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1078 
1079 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1080 
1081 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1082 		cmd->cmd_len = 16;
1083 		cmd->cmnd[0] = WRITE_SAME_16;
1084 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1085 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1086 	} else {
1087 		cmd->cmd_len = 10;
1088 		cmd->cmnd[0] = WRITE_SAME;
1089 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1090 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1091 	}
1092 
1093 	cmd->transfersize = sdp->sector_size;
1094 	cmd->allowed = sdkp->max_retries;
1095 
1096 	/*
1097 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1098 	 * different from the amount of data actually written to the target.
1099 	 *
1100 	 * We set up __data_len to the amount of data transferred via the
1101 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1102 	 * to transfer a single sector of data first, but then reset it to
1103 	 * the amount of data to be written right after so that the I/O path
1104 	 * knows how much to actually write.
1105 	 */
1106 	rq->__data_len = sdp->sector_size;
1107 	ret = scsi_alloc_sgtables(cmd);
1108 	rq->__data_len = blk_rq_bytes(rq);
1109 
1110 	return ret;
1111 }
1112 
1113 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1114 {
1115 	struct request *rq = cmd->request;
1116 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1117 
1118 	/* flush requests don't perform I/O, zero the S/G table */
1119 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1120 
1121 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1122 	cmd->cmd_len = 10;
1123 	cmd->transfersize = 0;
1124 	cmd->allowed = sdkp->max_retries;
1125 
1126 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1127 	return BLK_STS_OK;
1128 }
1129 
1130 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1131 				       sector_t lba, unsigned int nr_blocks,
1132 				       unsigned char flags)
1133 {
1134 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1135 	if (unlikely(cmd->cmnd == NULL))
1136 		return BLK_STS_RESOURCE;
1137 
1138 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1139 	memset(cmd->cmnd, 0, cmd->cmd_len);
1140 
1141 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1142 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1143 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1144 	cmd->cmnd[10] = flags;
1145 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1146 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1147 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1148 
1149 	return BLK_STS_OK;
1150 }
1151 
1152 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1153 				       sector_t lba, unsigned int nr_blocks,
1154 				       unsigned char flags)
1155 {
1156 	cmd->cmd_len  = 16;
1157 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1158 	cmd->cmnd[1]  = flags;
1159 	cmd->cmnd[14] = 0;
1160 	cmd->cmnd[15] = 0;
1161 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1162 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1163 
1164 	return BLK_STS_OK;
1165 }
1166 
1167 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1168 				       sector_t lba, unsigned int nr_blocks,
1169 				       unsigned char flags)
1170 {
1171 	cmd->cmd_len = 10;
1172 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1173 	cmd->cmnd[1] = flags;
1174 	cmd->cmnd[6] = 0;
1175 	cmd->cmnd[9] = 0;
1176 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1177 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1178 
1179 	return BLK_STS_OK;
1180 }
1181 
1182 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1183 				      sector_t lba, unsigned int nr_blocks,
1184 				      unsigned char flags)
1185 {
1186 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1187 	if (WARN_ON_ONCE(nr_blocks == 0))
1188 		return BLK_STS_IOERR;
1189 
1190 	if (unlikely(flags & 0x8)) {
1191 		/*
1192 		 * This happens only if this drive failed 10byte rw
1193 		 * command with ILLEGAL_REQUEST during operation and
1194 		 * thus turned off use_10_for_rw.
1195 		 */
1196 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1197 		return BLK_STS_IOERR;
1198 	}
1199 
1200 	cmd->cmd_len = 6;
1201 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1202 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1203 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1204 	cmd->cmnd[3] = lba & 0xff;
1205 	cmd->cmnd[4] = nr_blocks;
1206 	cmd->cmnd[5] = 0;
1207 
1208 	return BLK_STS_OK;
1209 }
1210 
1211 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1212 {
1213 	struct request *rq = cmd->request;
1214 	struct scsi_device *sdp = cmd->device;
1215 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1216 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1217 	sector_t threshold;
1218 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1219 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1220 	bool write = rq_data_dir(rq) == WRITE;
1221 	unsigned char protect, fua;
1222 	blk_status_t ret;
1223 	unsigned int dif;
1224 	bool dix;
1225 
1226 	ret = scsi_alloc_sgtables(cmd);
1227 	if (ret != BLK_STS_OK)
1228 		return ret;
1229 
1230 	ret = BLK_STS_IOERR;
1231 	if (!scsi_device_online(sdp) || sdp->changed) {
1232 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1233 		goto fail;
1234 	}
1235 
1236 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1237 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1238 		goto fail;
1239 	}
1240 
1241 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1242 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1243 		goto fail;
1244 	}
1245 
1246 	/*
1247 	 * Some SD card readers can't handle accesses which touch the
1248 	 * last one or two logical blocks. Split accesses as needed.
1249 	 */
1250 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1251 
1252 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1253 		if (lba < threshold) {
1254 			/* Access up to the threshold but not beyond */
1255 			nr_blocks = threshold - lba;
1256 		} else {
1257 			/* Access only a single logical block */
1258 			nr_blocks = 1;
1259 		}
1260 	}
1261 
1262 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1263 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1264 		if (ret)
1265 			goto fail;
1266 	}
1267 
1268 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1269 	dix = scsi_prot_sg_count(cmd);
1270 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1271 
1272 	if (dif || dix)
1273 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1274 	else
1275 		protect = 0;
1276 
1277 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1278 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1279 					 protect | fua);
1280 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1281 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1282 					 protect | fua);
1283 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1284 		   sdp->use_10_for_rw || protect) {
1285 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1286 					 protect | fua);
1287 	} else {
1288 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1289 					protect | fua);
1290 	}
1291 
1292 	if (unlikely(ret != BLK_STS_OK))
1293 		goto fail;
1294 
1295 	/*
1296 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1297 	 * host adapter, it's safe to assume that we can at least transfer
1298 	 * this many bytes between each connect / disconnect.
1299 	 */
1300 	cmd->transfersize = sdp->sector_size;
1301 	cmd->underflow = nr_blocks << 9;
1302 	cmd->allowed = sdkp->max_retries;
1303 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1304 
1305 	SCSI_LOG_HLQUEUE(1,
1306 			 scmd_printk(KERN_INFO, cmd,
1307 				     "%s: block=%llu, count=%d\n", __func__,
1308 				     (unsigned long long)blk_rq_pos(rq),
1309 				     blk_rq_sectors(rq)));
1310 	SCSI_LOG_HLQUEUE(2,
1311 			 scmd_printk(KERN_INFO, cmd,
1312 				     "%s %d/%u 512 byte blocks.\n",
1313 				     write ? "writing" : "reading", nr_blocks,
1314 				     blk_rq_sectors(rq)));
1315 
1316 	/*
1317 	 * This indicates that the command is ready from our end to be queued.
1318 	 */
1319 	return BLK_STS_OK;
1320 fail:
1321 	scsi_free_sgtables(cmd);
1322 	return ret;
1323 }
1324 
1325 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1326 {
1327 	struct request *rq = cmd->request;
1328 
1329 	switch (req_op(rq)) {
1330 	case REQ_OP_DISCARD:
1331 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1332 		case SD_LBP_UNMAP:
1333 			return sd_setup_unmap_cmnd(cmd);
1334 		case SD_LBP_WS16:
1335 			return sd_setup_write_same16_cmnd(cmd, true);
1336 		case SD_LBP_WS10:
1337 			return sd_setup_write_same10_cmnd(cmd, true);
1338 		case SD_LBP_ZERO:
1339 			return sd_setup_write_same10_cmnd(cmd, false);
1340 		default:
1341 			return BLK_STS_TARGET;
1342 		}
1343 	case REQ_OP_WRITE_ZEROES:
1344 		return sd_setup_write_zeroes_cmnd(cmd);
1345 	case REQ_OP_WRITE_SAME:
1346 		return sd_setup_write_same_cmnd(cmd);
1347 	case REQ_OP_FLUSH:
1348 		return sd_setup_flush_cmnd(cmd);
1349 	case REQ_OP_READ:
1350 	case REQ_OP_WRITE:
1351 	case REQ_OP_ZONE_APPEND:
1352 		return sd_setup_read_write_cmnd(cmd);
1353 	case REQ_OP_ZONE_RESET:
1354 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1355 						   false);
1356 	case REQ_OP_ZONE_RESET_ALL:
1357 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358 						   true);
1359 	case REQ_OP_ZONE_OPEN:
1360 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1361 	case REQ_OP_ZONE_CLOSE:
1362 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1363 	case REQ_OP_ZONE_FINISH:
1364 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1365 	default:
1366 		WARN_ON_ONCE(1);
1367 		return BLK_STS_NOTSUPP;
1368 	}
1369 }
1370 
1371 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1372 {
1373 	struct request *rq = SCpnt->request;
1374 	u8 *cmnd;
1375 
1376 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1377 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1378 
1379 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1380 		cmnd = SCpnt->cmnd;
1381 		SCpnt->cmnd = NULL;
1382 		SCpnt->cmd_len = 0;
1383 		mempool_free(cmnd, sd_cdb_pool);
1384 	}
1385 }
1386 
1387 static bool sd_need_revalidate(struct block_device *bdev,
1388 		struct scsi_disk *sdkp)
1389 {
1390 	if (sdkp->device->removable || sdkp->write_prot) {
1391 		if (bdev_check_media_change(bdev))
1392 			return true;
1393 	}
1394 
1395 	/*
1396 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1397 	 * nothing to do with partitions, BLKRRPART is used to force a full
1398 	 * revalidate after things like a format for historical reasons.
1399 	 */
1400 	return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1401 }
1402 
1403 /**
1404  *	sd_open - open a scsi disk device
1405  *	@bdev: Block device of the scsi disk to open
1406  *	@mode: FMODE_* mask
1407  *
1408  *	Returns 0 if successful. Returns a negated errno value in case
1409  *	of error.
1410  *
1411  *	Note: This can be called from a user context (e.g. fsck(1) )
1412  *	or from within the kernel (e.g. as a result of a mount(1) ).
1413  *	In the latter case @inode and @filp carry an abridged amount
1414  *	of information as noted above.
1415  *
1416  *	Locking: called with bdev->bd_disk->open_mutex held.
1417  **/
1418 static int sd_open(struct block_device *bdev, fmode_t mode)
1419 {
1420 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1421 	struct scsi_device *sdev;
1422 	int retval;
1423 
1424 	if (!sdkp)
1425 		return -ENXIO;
1426 
1427 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1428 
1429 	sdev = sdkp->device;
1430 
1431 	/*
1432 	 * If the device is in error recovery, wait until it is done.
1433 	 * If the device is offline, then disallow any access to it.
1434 	 */
1435 	retval = -ENXIO;
1436 	if (!scsi_block_when_processing_errors(sdev))
1437 		goto error_out;
1438 
1439 	if (sd_need_revalidate(bdev, sdkp))
1440 		sd_revalidate_disk(bdev->bd_disk);
1441 
1442 	/*
1443 	 * If the drive is empty, just let the open fail.
1444 	 */
1445 	retval = -ENOMEDIUM;
1446 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1447 		goto error_out;
1448 
1449 	/*
1450 	 * If the device has the write protect tab set, have the open fail
1451 	 * if the user expects to be able to write to the thing.
1452 	 */
1453 	retval = -EROFS;
1454 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1455 		goto error_out;
1456 
1457 	/*
1458 	 * It is possible that the disk changing stuff resulted in
1459 	 * the device being taken offline.  If this is the case,
1460 	 * report this to the user, and don't pretend that the
1461 	 * open actually succeeded.
1462 	 */
1463 	retval = -ENXIO;
1464 	if (!scsi_device_online(sdev))
1465 		goto error_out;
1466 
1467 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1468 		if (scsi_block_when_processing_errors(sdev))
1469 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1470 	}
1471 
1472 	return 0;
1473 
1474 error_out:
1475 	scsi_disk_put(sdkp);
1476 	return retval;
1477 }
1478 
1479 /**
1480  *	sd_release - invoked when the (last) close(2) is called on this
1481  *	scsi disk.
1482  *	@disk: disk to release
1483  *	@mode: FMODE_* mask
1484  *
1485  *	Returns 0.
1486  *
1487  *	Note: may block (uninterruptible) if error recovery is underway
1488  *	on this disk.
1489  *
1490  *	Locking: called with bdev->bd_disk->open_mutex held.
1491  **/
1492 static void sd_release(struct gendisk *disk, fmode_t mode)
1493 {
1494 	struct scsi_disk *sdkp = scsi_disk(disk);
1495 	struct scsi_device *sdev = sdkp->device;
1496 
1497 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1498 
1499 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1500 		if (scsi_block_when_processing_errors(sdev))
1501 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1502 	}
1503 
1504 	scsi_disk_put(sdkp);
1505 }
1506 
1507 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1508 {
1509 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1510 	struct scsi_device *sdp = sdkp->device;
1511 	struct Scsi_Host *host = sdp->host;
1512 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1513 	int diskinfo[4];
1514 
1515 	/* default to most commonly used values */
1516 	diskinfo[0] = 0x40;	/* 1 << 6 */
1517 	diskinfo[1] = 0x20;	/* 1 << 5 */
1518 	diskinfo[2] = capacity >> 11;
1519 
1520 	/* override with calculated, extended default, or driver values */
1521 	if (host->hostt->bios_param)
1522 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1523 	else
1524 		scsicam_bios_param(bdev, capacity, diskinfo);
1525 
1526 	geo->heads = diskinfo[0];
1527 	geo->sectors = diskinfo[1];
1528 	geo->cylinders = diskinfo[2];
1529 	return 0;
1530 }
1531 
1532 /**
1533  *	sd_ioctl_common - process an ioctl
1534  *	@bdev: target block device
1535  *	@mode: FMODE_* mask
1536  *	@cmd: ioctl command number
1537  *	@p: this is third argument given to ioctl(2) system call.
1538  *	Often contains a pointer.
1539  *
1540  *	Returns 0 if successful (some ioctls return positive numbers on
1541  *	success as well). Returns a negated errno value in case of error.
1542  *
1543  *	Note: most ioctls are forward onto the block subsystem or further
1544  *	down in the scsi subsystem.
1545  **/
1546 static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1547 			   unsigned int cmd, void __user *p)
1548 {
1549 	struct gendisk *disk = bdev->bd_disk;
1550 	struct scsi_disk *sdkp = scsi_disk(disk);
1551 	struct scsi_device *sdp = sdkp->device;
1552 	int error;
1553 
1554 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1555 				    "cmd=0x%x\n", disk->disk_name, cmd));
1556 
1557 	error = scsi_verify_blk_ioctl(bdev, cmd);
1558 	if (error < 0)
1559 		return error;
1560 
1561 	/*
1562 	 * If we are in the middle of error recovery, don't let anyone
1563 	 * else try and use this device.  Also, if error recovery fails, it
1564 	 * may try and take the device offline, in which case all further
1565 	 * access to the device is prohibited.
1566 	 */
1567 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1568 			(mode & FMODE_NDELAY) != 0);
1569 	if (error)
1570 		goto out;
1571 
1572 	if (is_sed_ioctl(cmd))
1573 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1574 
1575 	/*
1576 	 * Send SCSI addressing ioctls directly to mid level, send other
1577 	 * ioctls to block level and then onto mid level if they can't be
1578 	 * resolved.
1579 	 */
1580 	switch (cmd) {
1581 		case SCSI_IOCTL_GET_IDLUN:
1582 		case SCSI_IOCTL_GET_BUS_NUMBER:
1583 			error = scsi_ioctl(sdp, cmd, p);
1584 			break;
1585 		default:
1586 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1587 			break;
1588 	}
1589 out:
1590 	return error;
1591 }
1592 
1593 static void set_media_not_present(struct scsi_disk *sdkp)
1594 {
1595 	if (sdkp->media_present)
1596 		sdkp->device->changed = 1;
1597 
1598 	if (sdkp->device->removable) {
1599 		sdkp->media_present = 0;
1600 		sdkp->capacity = 0;
1601 	}
1602 }
1603 
1604 static int media_not_present(struct scsi_disk *sdkp,
1605 			     struct scsi_sense_hdr *sshdr)
1606 {
1607 	if (!scsi_sense_valid(sshdr))
1608 		return 0;
1609 
1610 	/* not invoked for commands that could return deferred errors */
1611 	switch (sshdr->sense_key) {
1612 	case UNIT_ATTENTION:
1613 	case NOT_READY:
1614 		/* medium not present */
1615 		if (sshdr->asc == 0x3A) {
1616 			set_media_not_present(sdkp);
1617 			return 1;
1618 		}
1619 	}
1620 	return 0;
1621 }
1622 
1623 /**
1624  *	sd_check_events - check media events
1625  *	@disk: kernel device descriptor
1626  *	@clearing: disk events currently being cleared
1627  *
1628  *	Returns mask of DISK_EVENT_*.
1629  *
1630  *	Note: this function is invoked from the block subsystem.
1631  **/
1632 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1633 {
1634 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1635 	struct scsi_device *sdp;
1636 	int retval;
1637 	bool disk_changed;
1638 
1639 	if (!sdkp)
1640 		return 0;
1641 
1642 	sdp = sdkp->device;
1643 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1644 
1645 	/*
1646 	 * If the device is offline, don't send any commands - just pretend as
1647 	 * if the command failed.  If the device ever comes back online, we
1648 	 * can deal with it then.  It is only because of unrecoverable errors
1649 	 * that we would ever take a device offline in the first place.
1650 	 */
1651 	if (!scsi_device_online(sdp)) {
1652 		set_media_not_present(sdkp);
1653 		goto out;
1654 	}
1655 
1656 	/*
1657 	 * Using TEST_UNIT_READY enables differentiation between drive with
1658 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1659 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1660 	 *
1661 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1662 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1663 	 * sd_revalidate() is called.
1664 	 */
1665 	if (scsi_block_when_processing_errors(sdp)) {
1666 		struct scsi_sense_hdr sshdr = { 0, };
1667 
1668 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1669 					      &sshdr);
1670 
1671 		/* failed to execute TUR, assume media not present */
1672 		if (retval < 0 || host_byte(retval)) {
1673 			set_media_not_present(sdkp);
1674 			goto out;
1675 		}
1676 
1677 		if (media_not_present(sdkp, &sshdr))
1678 			goto out;
1679 	}
1680 
1681 	/*
1682 	 * For removable scsi disk we have to recognise the presence
1683 	 * of a disk in the drive.
1684 	 */
1685 	if (!sdkp->media_present)
1686 		sdp->changed = 1;
1687 	sdkp->media_present = 1;
1688 out:
1689 	/*
1690 	 * sdp->changed is set under the following conditions:
1691 	 *
1692 	 *	Medium present state has changed in either direction.
1693 	 *	Device has indicated UNIT_ATTENTION.
1694 	 */
1695 	disk_changed = sdp->changed;
1696 	sdp->changed = 0;
1697 	scsi_disk_put(sdkp);
1698 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1699 }
1700 
1701 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1702 {
1703 	int retries, res;
1704 	struct scsi_device *sdp = sdkp->device;
1705 	const int timeout = sdp->request_queue->rq_timeout
1706 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1707 	struct scsi_sense_hdr my_sshdr;
1708 
1709 	if (!scsi_device_online(sdp))
1710 		return -ENODEV;
1711 
1712 	/* caller might not be interested in sense, but we need it */
1713 	if (!sshdr)
1714 		sshdr = &my_sshdr;
1715 
1716 	for (retries = 3; retries > 0; --retries) {
1717 		unsigned char cmd[10] = { 0 };
1718 
1719 		cmd[0] = SYNCHRONIZE_CACHE;
1720 		/*
1721 		 * Leave the rest of the command zero to indicate
1722 		 * flush everything.
1723 		 */
1724 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1725 				timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1726 		if (res == 0)
1727 			break;
1728 	}
1729 
1730 	if (res) {
1731 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1732 
1733 		if (res < 0)
1734 			return res;
1735 
1736 		if (scsi_status_is_check_condition(res) &&
1737 		    scsi_sense_valid(sshdr)) {
1738 			sd_print_sense_hdr(sdkp, sshdr);
1739 
1740 			/* we need to evaluate the error return  */
1741 			if (sshdr->asc == 0x3a ||	/* medium not present */
1742 			    sshdr->asc == 0x20 ||	/* invalid command */
1743 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1744 				/* this is no error here */
1745 				return 0;
1746 		}
1747 
1748 		switch (host_byte(res)) {
1749 		/* ignore errors due to racing a disconnection */
1750 		case DID_BAD_TARGET:
1751 		case DID_NO_CONNECT:
1752 			return 0;
1753 		/* signal the upper layer it might try again */
1754 		case DID_BUS_BUSY:
1755 		case DID_IMM_RETRY:
1756 		case DID_REQUEUE:
1757 		case DID_SOFT_ERROR:
1758 			return -EBUSY;
1759 		default:
1760 			return -EIO;
1761 		}
1762 	}
1763 	return 0;
1764 }
1765 
1766 static void sd_rescan(struct device *dev)
1767 {
1768 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1769 
1770 	sd_revalidate_disk(sdkp->disk);
1771 }
1772 
1773 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1774 		    unsigned int cmd, unsigned long arg)
1775 {
1776 	void __user *p = (void __user *)arg;
1777 	int ret;
1778 
1779 	ret = sd_ioctl_common(bdev, mode, cmd, p);
1780 	if (ret != -ENOTTY)
1781 		return ret;
1782 
1783 	return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1784 }
1785 
1786 #ifdef CONFIG_COMPAT
1787 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1788 			   unsigned int cmd, unsigned long arg)
1789 {
1790 	void __user *p = compat_ptr(arg);
1791 	int ret;
1792 
1793 	ret = sd_ioctl_common(bdev, mode, cmd, p);
1794 	if (ret != -ENOTTY)
1795 		return ret;
1796 
1797 	return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1798 }
1799 #endif
1800 
1801 static char sd_pr_type(enum pr_type type)
1802 {
1803 	switch (type) {
1804 	case PR_WRITE_EXCLUSIVE:
1805 		return 0x01;
1806 	case PR_EXCLUSIVE_ACCESS:
1807 		return 0x03;
1808 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1809 		return 0x05;
1810 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1811 		return 0x06;
1812 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1813 		return 0x07;
1814 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1815 		return 0x08;
1816 	default:
1817 		return 0;
1818 	}
1819 };
1820 
1821 static int sd_pr_command(struct block_device *bdev, u8 sa,
1822 		u64 key, u64 sa_key, u8 type, u8 flags)
1823 {
1824 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1825 	struct scsi_device *sdev = sdkp->device;
1826 	struct scsi_sense_hdr sshdr;
1827 	int result;
1828 	u8 cmd[16] = { 0, };
1829 	u8 data[24] = { 0, };
1830 
1831 	cmd[0] = PERSISTENT_RESERVE_OUT;
1832 	cmd[1] = sa;
1833 	cmd[2] = type;
1834 	put_unaligned_be32(sizeof(data), &cmd[5]);
1835 
1836 	put_unaligned_be64(key, &data[0]);
1837 	put_unaligned_be64(sa_key, &data[8]);
1838 	data[20] = flags;
1839 
1840 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1841 			&sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1842 
1843 	if (scsi_status_is_check_condition(result) &&
1844 	    scsi_sense_valid(&sshdr)) {
1845 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1846 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1847 	}
1848 
1849 	return result;
1850 }
1851 
1852 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1853 		u32 flags)
1854 {
1855 	if (flags & ~PR_FL_IGNORE_KEY)
1856 		return -EOPNOTSUPP;
1857 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1858 			old_key, new_key, 0,
1859 			(1 << 0) /* APTPL */);
1860 }
1861 
1862 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1863 		u32 flags)
1864 {
1865 	if (flags)
1866 		return -EOPNOTSUPP;
1867 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1868 }
1869 
1870 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1871 {
1872 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1873 }
1874 
1875 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1876 		enum pr_type type, bool abort)
1877 {
1878 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1879 			     sd_pr_type(type), 0);
1880 }
1881 
1882 static int sd_pr_clear(struct block_device *bdev, u64 key)
1883 {
1884 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1885 }
1886 
1887 static const struct pr_ops sd_pr_ops = {
1888 	.pr_register	= sd_pr_register,
1889 	.pr_reserve	= sd_pr_reserve,
1890 	.pr_release	= sd_pr_release,
1891 	.pr_preempt	= sd_pr_preempt,
1892 	.pr_clear	= sd_pr_clear,
1893 };
1894 
1895 static const struct block_device_operations sd_fops = {
1896 	.owner			= THIS_MODULE,
1897 	.open			= sd_open,
1898 	.release		= sd_release,
1899 	.ioctl			= sd_ioctl,
1900 	.getgeo			= sd_getgeo,
1901 #ifdef CONFIG_COMPAT
1902 	.compat_ioctl		= sd_compat_ioctl,
1903 #endif
1904 	.check_events		= sd_check_events,
1905 	.unlock_native_capacity	= sd_unlock_native_capacity,
1906 	.report_zones		= sd_zbc_report_zones,
1907 	.pr_ops			= &sd_pr_ops,
1908 };
1909 
1910 /**
1911  *	sd_eh_reset - reset error handling callback
1912  *	@scmd:		sd-issued command that has failed
1913  *
1914  *	This function is called by the SCSI midlayer before starting
1915  *	SCSI EH. When counting medium access failures we have to be
1916  *	careful to register it only only once per device and SCSI EH run;
1917  *	there might be several timed out commands which will cause the
1918  *	'max_medium_access_timeouts' counter to trigger after the first
1919  *	SCSI EH run already and set the device to offline.
1920  *	So this function resets the internal counter before starting SCSI EH.
1921  **/
1922 static void sd_eh_reset(struct scsi_cmnd *scmd)
1923 {
1924 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1925 
1926 	/* New SCSI EH run, reset gate variable */
1927 	sdkp->ignore_medium_access_errors = false;
1928 }
1929 
1930 /**
1931  *	sd_eh_action - error handling callback
1932  *	@scmd:		sd-issued command that has failed
1933  *	@eh_disp:	The recovery disposition suggested by the midlayer
1934  *
1935  *	This function is called by the SCSI midlayer upon completion of an
1936  *	error test command (currently TEST UNIT READY). The result of sending
1937  *	the eh command is passed in eh_disp.  We're looking for devices that
1938  *	fail medium access commands but are OK with non access commands like
1939  *	test unit ready (so wrongly see the device as having a successful
1940  *	recovery)
1941  **/
1942 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1943 {
1944 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1945 	struct scsi_device *sdev = scmd->device;
1946 
1947 	if (!scsi_device_online(sdev) ||
1948 	    !scsi_medium_access_command(scmd) ||
1949 	    host_byte(scmd->result) != DID_TIME_OUT ||
1950 	    eh_disp != SUCCESS)
1951 		return eh_disp;
1952 
1953 	/*
1954 	 * The device has timed out executing a medium access command.
1955 	 * However, the TEST UNIT READY command sent during error
1956 	 * handling completed successfully. Either the device is in the
1957 	 * process of recovering or has it suffered an internal failure
1958 	 * that prevents access to the storage medium.
1959 	 */
1960 	if (!sdkp->ignore_medium_access_errors) {
1961 		sdkp->medium_access_timed_out++;
1962 		sdkp->ignore_medium_access_errors = true;
1963 	}
1964 
1965 	/*
1966 	 * If the device keeps failing read/write commands but TEST UNIT
1967 	 * READY always completes successfully we assume that medium
1968 	 * access is no longer possible and take the device offline.
1969 	 */
1970 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1971 		scmd_printk(KERN_ERR, scmd,
1972 			    "Medium access timeout failure. Offlining disk!\n");
1973 		mutex_lock(&sdev->state_mutex);
1974 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1975 		mutex_unlock(&sdev->state_mutex);
1976 
1977 		return SUCCESS;
1978 	}
1979 
1980 	return eh_disp;
1981 }
1982 
1983 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1984 {
1985 	struct request *req = scmd->request;
1986 	struct scsi_device *sdev = scmd->device;
1987 	unsigned int transferred, good_bytes;
1988 	u64 start_lba, end_lba, bad_lba;
1989 
1990 	/*
1991 	 * Some commands have a payload smaller than the device logical
1992 	 * block size (e.g. INQUIRY on a 4K disk).
1993 	 */
1994 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1995 		return 0;
1996 
1997 	/* Check if we have a 'bad_lba' information */
1998 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1999 				     SCSI_SENSE_BUFFERSIZE,
2000 				     &bad_lba))
2001 		return 0;
2002 
2003 	/*
2004 	 * If the bad lba was reported incorrectly, we have no idea where
2005 	 * the error is.
2006 	 */
2007 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2008 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2009 	if (bad_lba < start_lba || bad_lba >= end_lba)
2010 		return 0;
2011 
2012 	/*
2013 	 * resid is optional but mostly filled in.  When it's unused,
2014 	 * its value is zero, so we assume the whole buffer transferred
2015 	 */
2016 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2017 
2018 	/* This computation should always be done in terms of the
2019 	 * resolution of the device's medium.
2020 	 */
2021 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2022 
2023 	return min(good_bytes, transferred);
2024 }
2025 
2026 /**
2027  *	sd_done - bottom half handler: called when the lower level
2028  *	driver has completed (successfully or otherwise) a scsi command.
2029  *	@SCpnt: mid-level's per command structure.
2030  *
2031  *	Note: potentially run from within an ISR. Must not block.
2032  **/
2033 static int sd_done(struct scsi_cmnd *SCpnt)
2034 {
2035 	int result = SCpnt->result;
2036 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2037 	unsigned int sector_size = SCpnt->device->sector_size;
2038 	unsigned int resid;
2039 	struct scsi_sense_hdr sshdr;
2040 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
2041 	struct request *req = SCpnt->request;
2042 	int sense_valid = 0;
2043 	int sense_deferred = 0;
2044 
2045 	switch (req_op(req)) {
2046 	case REQ_OP_DISCARD:
2047 	case REQ_OP_WRITE_ZEROES:
2048 	case REQ_OP_WRITE_SAME:
2049 	case REQ_OP_ZONE_RESET:
2050 	case REQ_OP_ZONE_RESET_ALL:
2051 	case REQ_OP_ZONE_OPEN:
2052 	case REQ_OP_ZONE_CLOSE:
2053 	case REQ_OP_ZONE_FINISH:
2054 		if (!result) {
2055 			good_bytes = blk_rq_bytes(req);
2056 			scsi_set_resid(SCpnt, 0);
2057 		} else {
2058 			good_bytes = 0;
2059 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2060 		}
2061 		break;
2062 	default:
2063 		/*
2064 		 * In case of bogus fw or device, we could end up having
2065 		 * an unaligned partial completion. Check this here and force
2066 		 * alignment.
2067 		 */
2068 		resid = scsi_get_resid(SCpnt);
2069 		if (resid & (sector_size - 1)) {
2070 			sd_printk(KERN_INFO, sdkp,
2071 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2072 				resid, sector_size);
2073 			scsi_print_command(SCpnt);
2074 			resid = min(scsi_bufflen(SCpnt),
2075 				    round_up(resid, sector_size));
2076 			scsi_set_resid(SCpnt, resid);
2077 		}
2078 	}
2079 
2080 	if (result) {
2081 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2082 		if (sense_valid)
2083 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2084 	}
2085 	sdkp->medium_access_timed_out = 0;
2086 
2087 	if (!scsi_status_is_check_condition(result) &&
2088 	    (!sense_valid || sense_deferred))
2089 		goto out;
2090 
2091 	switch (sshdr.sense_key) {
2092 	case HARDWARE_ERROR:
2093 	case MEDIUM_ERROR:
2094 		good_bytes = sd_completed_bytes(SCpnt);
2095 		break;
2096 	case RECOVERED_ERROR:
2097 		good_bytes = scsi_bufflen(SCpnt);
2098 		break;
2099 	case NO_SENSE:
2100 		/* This indicates a false check condition, so ignore it.  An
2101 		 * unknown amount of data was transferred so treat it as an
2102 		 * error.
2103 		 */
2104 		SCpnt->result = 0;
2105 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2106 		break;
2107 	case ABORTED_COMMAND:
2108 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2109 			good_bytes = sd_completed_bytes(SCpnt);
2110 		break;
2111 	case ILLEGAL_REQUEST:
2112 		switch (sshdr.asc) {
2113 		case 0x10:	/* DIX: Host detected corruption */
2114 			good_bytes = sd_completed_bytes(SCpnt);
2115 			break;
2116 		case 0x20:	/* INVALID COMMAND OPCODE */
2117 		case 0x24:	/* INVALID FIELD IN CDB */
2118 			switch (SCpnt->cmnd[0]) {
2119 			case UNMAP:
2120 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2121 				break;
2122 			case WRITE_SAME_16:
2123 			case WRITE_SAME:
2124 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2125 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2126 				} else {
2127 					sdkp->device->no_write_same = 1;
2128 					sd_config_write_same(sdkp);
2129 					req->rq_flags |= RQF_QUIET;
2130 				}
2131 				break;
2132 			}
2133 		}
2134 		break;
2135 	default:
2136 		break;
2137 	}
2138 
2139  out:
2140 	if (sd_is_zoned(sdkp))
2141 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2142 
2143 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2144 					   "sd_done: completed %d of %d bytes\n",
2145 					   good_bytes, scsi_bufflen(SCpnt)));
2146 
2147 	return good_bytes;
2148 }
2149 
2150 /*
2151  * spinup disk - called only in sd_revalidate_disk()
2152  */
2153 static void
2154 sd_spinup_disk(struct scsi_disk *sdkp)
2155 {
2156 	unsigned char cmd[10];
2157 	unsigned long spintime_expire = 0;
2158 	int retries, spintime;
2159 	unsigned int the_result;
2160 	struct scsi_sense_hdr sshdr;
2161 	int sense_valid = 0;
2162 
2163 	spintime = 0;
2164 
2165 	/* Spin up drives, as required.  Only do this at boot time */
2166 	/* Spinup needs to be done for module loads too. */
2167 	do {
2168 		retries = 0;
2169 
2170 		do {
2171 			cmd[0] = TEST_UNIT_READY;
2172 			memset((void *) &cmd[1], 0, 9);
2173 
2174 			the_result = scsi_execute_req(sdkp->device, cmd,
2175 						      DMA_NONE, NULL, 0,
2176 						      &sshdr, SD_TIMEOUT,
2177 						      sdkp->max_retries, NULL);
2178 
2179 			/*
2180 			 * If the drive has indicated to us that it
2181 			 * doesn't have any media in it, don't bother
2182 			 * with any more polling.
2183 			 */
2184 			if (media_not_present(sdkp, &sshdr))
2185 				return;
2186 
2187 			if (the_result)
2188 				sense_valid = scsi_sense_valid(&sshdr);
2189 			retries++;
2190 		} while (retries < 3 &&
2191 			 (!scsi_status_is_good(the_result) ||
2192 			  (scsi_status_is_check_condition(the_result) &&
2193 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2194 
2195 		if (!scsi_status_is_check_condition(the_result)) {
2196 			/* no sense, TUR either succeeded or failed
2197 			 * with a status error */
2198 			if(!spintime && !scsi_status_is_good(the_result)) {
2199 				sd_print_result(sdkp, "Test Unit Ready failed",
2200 						the_result);
2201 			}
2202 			break;
2203 		}
2204 
2205 		/*
2206 		 * The device does not want the automatic start to be issued.
2207 		 */
2208 		if (sdkp->device->no_start_on_add)
2209 			break;
2210 
2211 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2212 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2213 				break;	/* manual intervention required */
2214 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2215 				break;	/* standby */
2216 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2217 				break;	/* unavailable */
2218 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2219 				break;	/* sanitize in progress */
2220 			/*
2221 			 * Issue command to spin up drive when not ready
2222 			 */
2223 			if (!spintime) {
2224 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2225 				cmd[0] = START_STOP;
2226 				cmd[1] = 1;	/* Return immediately */
2227 				memset((void *) &cmd[2], 0, 8);
2228 				cmd[4] = 1;	/* Start spin cycle */
2229 				if (sdkp->device->start_stop_pwr_cond)
2230 					cmd[4] |= 1 << 4;
2231 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2232 						 NULL, 0, &sshdr,
2233 						 SD_TIMEOUT, sdkp->max_retries,
2234 						 NULL);
2235 				spintime_expire = jiffies + 100 * HZ;
2236 				spintime = 1;
2237 			}
2238 			/* Wait 1 second for next try */
2239 			msleep(1000);
2240 			printk(KERN_CONT ".");
2241 
2242 		/*
2243 		 * Wait for USB flash devices with slow firmware.
2244 		 * Yes, this sense key/ASC combination shouldn't
2245 		 * occur here.  It's characteristic of these devices.
2246 		 */
2247 		} else if (sense_valid &&
2248 				sshdr.sense_key == UNIT_ATTENTION &&
2249 				sshdr.asc == 0x28) {
2250 			if (!spintime) {
2251 				spintime_expire = jiffies + 5 * HZ;
2252 				spintime = 1;
2253 			}
2254 			/* Wait 1 second for next try */
2255 			msleep(1000);
2256 		} else {
2257 			/* we don't understand the sense code, so it's
2258 			 * probably pointless to loop */
2259 			if(!spintime) {
2260 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2261 				sd_print_sense_hdr(sdkp, &sshdr);
2262 			}
2263 			break;
2264 		}
2265 
2266 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2267 
2268 	if (spintime) {
2269 		if (scsi_status_is_good(the_result))
2270 			printk(KERN_CONT "ready\n");
2271 		else
2272 			printk(KERN_CONT "not responding...\n");
2273 	}
2274 }
2275 
2276 /*
2277  * Determine whether disk supports Data Integrity Field.
2278  */
2279 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2280 {
2281 	struct scsi_device *sdp = sdkp->device;
2282 	u8 type;
2283 	int ret = 0;
2284 
2285 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2286 		sdkp->protection_type = 0;
2287 		return ret;
2288 	}
2289 
2290 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2291 
2292 	if (type > T10_PI_TYPE3_PROTECTION)
2293 		ret = -ENODEV;
2294 	else if (scsi_host_dif_capable(sdp->host, type))
2295 		ret = 1;
2296 
2297 	if (sdkp->first_scan || type != sdkp->protection_type)
2298 		switch (ret) {
2299 		case -ENODEV:
2300 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2301 				  " protection type %u. Disabling disk!\n",
2302 				  type);
2303 			break;
2304 		case 1:
2305 			sd_printk(KERN_NOTICE, sdkp,
2306 				  "Enabling DIF Type %u protection\n", type);
2307 			break;
2308 		case 0:
2309 			sd_printk(KERN_NOTICE, sdkp,
2310 				  "Disabling DIF Type %u protection\n", type);
2311 			break;
2312 		}
2313 
2314 	sdkp->protection_type = type;
2315 
2316 	return ret;
2317 }
2318 
2319 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2320 			struct scsi_sense_hdr *sshdr, int sense_valid,
2321 			int the_result)
2322 {
2323 	if (sense_valid)
2324 		sd_print_sense_hdr(sdkp, sshdr);
2325 	else
2326 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2327 
2328 	/*
2329 	 * Set dirty bit for removable devices if not ready -
2330 	 * sometimes drives will not report this properly.
2331 	 */
2332 	if (sdp->removable &&
2333 	    sense_valid && sshdr->sense_key == NOT_READY)
2334 		set_media_not_present(sdkp);
2335 
2336 	/*
2337 	 * We used to set media_present to 0 here to indicate no media
2338 	 * in the drive, but some drives fail read capacity even with
2339 	 * media present, so we can't do that.
2340 	 */
2341 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2342 }
2343 
2344 #define RC16_LEN 32
2345 #if RC16_LEN > SD_BUF_SIZE
2346 #error RC16_LEN must not be more than SD_BUF_SIZE
2347 #endif
2348 
2349 #define READ_CAPACITY_RETRIES_ON_RESET	10
2350 
2351 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2352 						unsigned char *buffer)
2353 {
2354 	unsigned char cmd[16];
2355 	struct scsi_sense_hdr sshdr;
2356 	int sense_valid = 0;
2357 	int the_result;
2358 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2359 	unsigned int alignment;
2360 	unsigned long long lba;
2361 	unsigned sector_size;
2362 
2363 	if (sdp->no_read_capacity_16)
2364 		return -EINVAL;
2365 
2366 	do {
2367 		memset(cmd, 0, 16);
2368 		cmd[0] = SERVICE_ACTION_IN_16;
2369 		cmd[1] = SAI_READ_CAPACITY_16;
2370 		cmd[13] = RC16_LEN;
2371 		memset(buffer, 0, RC16_LEN);
2372 
2373 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2374 					buffer, RC16_LEN, &sshdr,
2375 					SD_TIMEOUT, sdkp->max_retries, NULL);
2376 
2377 		if (media_not_present(sdkp, &sshdr))
2378 			return -ENODEV;
2379 
2380 		if (the_result > 0) {
2381 			sense_valid = scsi_sense_valid(&sshdr);
2382 			if (sense_valid &&
2383 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2384 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2385 			    sshdr.ascq == 0x00)
2386 				/* Invalid Command Operation Code or
2387 				 * Invalid Field in CDB, just retry
2388 				 * silently with RC10 */
2389 				return -EINVAL;
2390 			if (sense_valid &&
2391 			    sshdr.sense_key == UNIT_ATTENTION &&
2392 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2393 				/* Device reset might occur several times,
2394 				 * give it one more chance */
2395 				if (--reset_retries > 0)
2396 					continue;
2397 		}
2398 		retries--;
2399 
2400 	} while (the_result && retries);
2401 
2402 	if (the_result) {
2403 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2404 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2405 		return -EINVAL;
2406 	}
2407 
2408 	sector_size = get_unaligned_be32(&buffer[8]);
2409 	lba = get_unaligned_be64(&buffer[0]);
2410 
2411 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2412 		sdkp->capacity = 0;
2413 		return -ENODEV;
2414 	}
2415 
2416 	/* Logical blocks per physical block exponent */
2417 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2418 
2419 	/* RC basis */
2420 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2421 
2422 	/* Lowest aligned logical block */
2423 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2424 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2425 	if (alignment && sdkp->first_scan)
2426 		sd_printk(KERN_NOTICE, sdkp,
2427 			  "physical block alignment offset: %u\n", alignment);
2428 
2429 	if (buffer[14] & 0x80) { /* LBPME */
2430 		sdkp->lbpme = 1;
2431 
2432 		if (buffer[14] & 0x40) /* LBPRZ */
2433 			sdkp->lbprz = 1;
2434 
2435 		sd_config_discard(sdkp, SD_LBP_WS16);
2436 	}
2437 
2438 	sdkp->capacity = lba + 1;
2439 	return sector_size;
2440 }
2441 
2442 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2443 						unsigned char *buffer)
2444 {
2445 	unsigned char cmd[16];
2446 	struct scsi_sense_hdr sshdr;
2447 	int sense_valid = 0;
2448 	int the_result;
2449 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2450 	sector_t lba;
2451 	unsigned sector_size;
2452 
2453 	do {
2454 		cmd[0] = READ_CAPACITY;
2455 		memset(&cmd[1], 0, 9);
2456 		memset(buffer, 0, 8);
2457 
2458 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2459 					buffer, 8, &sshdr,
2460 					SD_TIMEOUT, sdkp->max_retries, NULL);
2461 
2462 		if (media_not_present(sdkp, &sshdr))
2463 			return -ENODEV;
2464 
2465 		if (the_result > 0) {
2466 			sense_valid = scsi_sense_valid(&sshdr);
2467 			if (sense_valid &&
2468 			    sshdr.sense_key == UNIT_ATTENTION &&
2469 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2470 				/* Device reset might occur several times,
2471 				 * give it one more chance */
2472 				if (--reset_retries > 0)
2473 					continue;
2474 		}
2475 		retries--;
2476 
2477 	} while (the_result && retries);
2478 
2479 	if (the_result) {
2480 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2481 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2482 		return -EINVAL;
2483 	}
2484 
2485 	sector_size = get_unaligned_be32(&buffer[4]);
2486 	lba = get_unaligned_be32(&buffer[0]);
2487 
2488 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2489 		/* Some buggy (usb cardreader) devices return an lba of
2490 		   0xffffffff when the want to report a size of 0 (with
2491 		   which they really mean no media is present) */
2492 		sdkp->capacity = 0;
2493 		sdkp->physical_block_size = sector_size;
2494 		return sector_size;
2495 	}
2496 
2497 	sdkp->capacity = lba + 1;
2498 	sdkp->physical_block_size = sector_size;
2499 	return sector_size;
2500 }
2501 
2502 static int sd_try_rc16_first(struct scsi_device *sdp)
2503 {
2504 	if (sdp->host->max_cmd_len < 16)
2505 		return 0;
2506 	if (sdp->try_rc_10_first)
2507 		return 0;
2508 	if (sdp->scsi_level > SCSI_SPC_2)
2509 		return 1;
2510 	if (scsi_device_protection(sdp))
2511 		return 1;
2512 	return 0;
2513 }
2514 
2515 /*
2516  * read disk capacity
2517  */
2518 static void
2519 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2520 {
2521 	int sector_size;
2522 	struct scsi_device *sdp = sdkp->device;
2523 
2524 	if (sd_try_rc16_first(sdp)) {
2525 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2526 		if (sector_size == -EOVERFLOW)
2527 			goto got_data;
2528 		if (sector_size == -ENODEV)
2529 			return;
2530 		if (sector_size < 0)
2531 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2532 		if (sector_size < 0)
2533 			return;
2534 	} else {
2535 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2536 		if (sector_size == -EOVERFLOW)
2537 			goto got_data;
2538 		if (sector_size < 0)
2539 			return;
2540 		if ((sizeof(sdkp->capacity) > 4) &&
2541 		    (sdkp->capacity > 0xffffffffULL)) {
2542 			int old_sector_size = sector_size;
2543 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2544 					"Trying to use READ CAPACITY(16).\n");
2545 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2546 			if (sector_size < 0) {
2547 				sd_printk(KERN_NOTICE, sdkp,
2548 					"Using 0xffffffff as device size\n");
2549 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2550 				sector_size = old_sector_size;
2551 				goto got_data;
2552 			}
2553 			/* Remember that READ CAPACITY(16) succeeded */
2554 			sdp->try_rc_10_first = 0;
2555 		}
2556 	}
2557 
2558 	/* Some devices are known to return the total number of blocks,
2559 	 * not the highest block number.  Some devices have versions
2560 	 * which do this and others which do not.  Some devices we might
2561 	 * suspect of doing this but we don't know for certain.
2562 	 *
2563 	 * If we know the reported capacity is wrong, decrement it.  If
2564 	 * we can only guess, then assume the number of blocks is even
2565 	 * (usually true but not always) and err on the side of lowering
2566 	 * the capacity.
2567 	 */
2568 	if (sdp->fix_capacity ||
2569 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2570 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2571 				"from its reported value: %llu\n",
2572 				(unsigned long long) sdkp->capacity);
2573 		--sdkp->capacity;
2574 	}
2575 
2576 got_data:
2577 	if (sector_size == 0) {
2578 		sector_size = 512;
2579 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2580 			  "assuming 512.\n");
2581 	}
2582 
2583 	if (sector_size != 512 &&
2584 	    sector_size != 1024 &&
2585 	    sector_size != 2048 &&
2586 	    sector_size != 4096) {
2587 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2588 			  sector_size);
2589 		/*
2590 		 * The user might want to re-format the drive with
2591 		 * a supported sectorsize.  Once this happens, it
2592 		 * would be relatively trivial to set the thing up.
2593 		 * For this reason, we leave the thing in the table.
2594 		 */
2595 		sdkp->capacity = 0;
2596 		/*
2597 		 * set a bogus sector size so the normal read/write
2598 		 * logic in the block layer will eventually refuse any
2599 		 * request on this device without tripping over power
2600 		 * of two sector size assumptions
2601 		 */
2602 		sector_size = 512;
2603 	}
2604 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2605 	blk_queue_physical_block_size(sdp->request_queue,
2606 				      sdkp->physical_block_size);
2607 	sdkp->device->sector_size = sector_size;
2608 
2609 	if (sdkp->capacity > 0xffffffff)
2610 		sdp->use_16_for_rw = 1;
2611 
2612 }
2613 
2614 /*
2615  * Print disk capacity
2616  */
2617 static void
2618 sd_print_capacity(struct scsi_disk *sdkp,
2619 		  sector_t old_capacity)
2620 {
2621 	int sector_size = sdkp->device->sector_size;
2622 	char cap_str_2[10], cap_str_10[10];
2623 
2624 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2625 		return;
2626 
2627 	string_get_size(sdkp->capacity, sector_size,
2628 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2629 	string_get_size(sdkp->capacity, sector_size,
2630 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2631 
2632 	sd_printk(KERN_NOTICE, sdkp,
2633 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2634 		  (unsigned long long)sdkp->capacity,
2635 		  sector_size, cap_str_10, cap_str_2);
2636 
2637 	if (sdkp->physical_block_size != sector_size)
2638 		sd_printk(KERN_NOTICE, sdkp,
2639 			  "%u-byte physical blocks\n",
2640 			  sdkp->physical_block_size);
2641 }
2642 
2643 /* called with buffer of length 512 */
2644 static inline int
2645 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2646 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2647 		 struct scsi_sense_hdr *sshdr)
2648 {
2649 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2650 			       SD_TIMEOUT, sdkp->max_retries, data,
2651 			       sshdr);
2652 }
2653 
2654 /*
2655  * read write protect setting, if possible - called only in sd_revalidate_disk()
2656  * called with buffer of length SD_BUF_SIZE
2657  */
2658 static void
2659 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2660 {
2661 	int res;
2662 	struct scsi_device *sdp = sdkp->device;
2663 	struct scsi_mode_data data;
2664 	int old_wp = sdkp->write_prot;
2665 
2666 	set_disk_ro(sdkp->disk, 0);
2667 	if (sdp->skip_ms_page_3f) {
2668 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2669 		return;
2670 	}
2671 
2672 	if (sdp->use_192_bytes_for_3f) {
2673 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2674 	} else {
2675 		/*
2676 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2677 		 * We have to start carefully: some devices hang if we ask
2678 		 * for more than is available.
2679 		 */
2680 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2681 
2682 		/*
2683 		 * Second attempt: ask for page 0 When only page 0 is
2684 		 * implemented, a request for page 3F may return Sense Key
2685 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2686 		 * CDB.
2687 		 */
2688 		if (res < 0)
2689 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2690 
2691 		/*
2692 		 * Third attempt: ask 255 bytes, as we did earlier.
2693 		 */
2694 		if (res < 0)
2695 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2696 					       &data, NULL);
2697 	}
2698 
2699 	if (res < 0) {
2700 		sd_first_printk(KERN_WARNING, sdkp,
2701 			  "Test WP failed, assume Write Enabled\n");
2702 	} else {
2703 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2704 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2705 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2706 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2707 				  sdkp->write_prot ? "on" : "off");
2708 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2709 		}
2710 	}
2711 }
2712 
2713 /*
2714  * sd_read_cache_type - called only from sd_revalidate_disk()
2715  * called with buffer of length SD_BUF_SIZE
2716  */
2717 static void
2718 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2719 {
2720 	int len = 0, res;
2721 	struct scsi_device *sdp = sdkp->device;
2722 
2723 	int dbd;
2724 	int modepage;
2725 	int first_len;
2726 	struct scsi_mode_data data;
2727 	struct scsi_sense_hdr sshdr;
2728 	int old_wce = sdkp->WCE;
2729 	int old_rcd = sdkp->RCD;
2730 	int old_dpofua = sdkp->DPOFUA;
2731 
2732 
2733 	if (sdkp->cache_override)
2734 		return;
2735 
2736 	first_len = 4;
2737 	if (sdp->skip_ms_page_8) {
2738 		if (sdp->type == TYPE_RBC)
2739 			goto defaults;
2740 		else {
2741 			if (sdp->skip_ms_page_3f)
2742 				goto defaults;
2743 			modepage = 0x3F;
2744 			if (sdp->use_192_bytes_for_3f)
2745 				first_len = 192;
2746 			dbd = 0;
2747 		}
2748 	} else if (sdp->type == TYPE_RBC) {
2749 		modepage = 6;
2750 		dbd = 8;
2751 	} else {
2752 		modepage = 8;
2753 		dbd = 0;
2754 	}
2755 
2756 	/* cautiously ask */
2757 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2758 			&data, &sshdr);
2759 
2760 	if (res < 0)
2761 		goto bad_sense;
2762 
2763 	if (!data.header_length) {
2764 		modepage = 6;
2765 		first_len = 0;
2766 		sd_first_printk(KERN_ERR, sdkp,
2767 				"Missing header in MODE_SENSE response\n");
2768 	}
2769 
2770 	/* that went OK, now ask for the proper length */
2771 	len = data.length;
2772 
2773 	/*
2774 	 * We're only interested in the first three bytes, actually.
2775 	 * But the data cache page is defined for the first 20.
2776 	 */
2777 	if (len < 3)
2778 		goto bad_sense;
2779 	else if (len > SD_BUF_SIZE) {
2780 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2781 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2782 		len = SD_BUF_SIZE;
2783 	}
2784 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2785 		len = 192;
2786 
2787 	/* Get the data */
2788 	if (len > first_len)
2789 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2790 				&data, &sshdr);
2791 
2792 	if (!res) {
2793 		int offset = data.header_length + data.block_descriptor_length;
2794 
2795 		while (offset < len) {
2796 			u8 page_code = buffer[offset] & 0x3F;
2797 			u8 spf       = buffer[offset] & 0x40;
2798 
2799 			if (page_code == 8 || page_code == 6) {
2800 				/* We're interested only in the first 3 bytes.
2801 				 */
2802 				if (len - offset <= 2) {
2803 					sd_first_printk(KERN_ERR, sdkp,
2804 						"Incomplete mode parameter "
2805 							"data\n");
2806 					goto defaults;
2807 				} else {
2808 					modepage = page_code;
2809 					goto Page_found;
2810 				}
2811 			} else {
2812 				/* Go to the next page */
2813 				if (spf && len - offset > 3)
2814 					offset += 4 + (buffer[offset+2] << 8) +
2815 						buffer[offset+3];
2816 				else if (!spf && len - offset > 1)
2817 					offset += 2 + buffer[offset+1];
2818 				else {
2819 					sd_first_printk(KERN_ERR, sdkp,
2820 							"Incomplete mode "
2821 							"parameter data\n");
2822 					goto defaults;
2823 				}
2824 			}
2825 		}
2826 
2827 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2828 		goto defaults;
2829 
2830 	Page_found:
2831 		if (modepage == 8) {
2832 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2833 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2834 		} else {
2835 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2836 			sdkp->RCD = 0;
2837 		}
2838 
2839 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2840 		if (sdp->broken_fua) {
2841 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2842 			sdkp->DPOFUA = 0;
2843 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2844 			   !sdkp->device->use_16_for_rw) {
2845 			sd_first_printk(KERN_NOTICE, sdkp,
2846 				  "Uses READ/WRITE(6), disabling FUA\n");
2847 			sdkp->DPOFUA = 0;
2848 		}
2849 
2850 		/* No cache flush allowed for write protected devices */
2851 		if (sdkp->WCE && sdkp->write_prot)
2852 			sdkp->WCE = 0;
2853 
2854 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2855 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2856 			sd_printk(KERN_NOTICE, sdkp,
2857 				  "Write cache: %s, read cache: %s, %s\n",
2858 				  sdkp->WCE ? "enabled" : "disabled",
2859 				  sdkp->RCD ? "disabled" : "enabled",
2860 				  sdkp->DPOFUA ? "supports DPO and FUA"
2861 				  : "doesn't support DPO or FUA");
2862 
2863 		return;
2864 	}
2865 
2866 bad_sense:
2867 	if (scsi_sense_valid(&sshdr) &&
2868 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2869 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2870 		/* Invalid field in CDB */
2871 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2872 	else
2873 		sd_first_printk(KERN_ERR, sdkp,
2874 				"Asking for cache data failed\n");
2875 
2876 defaults:
2877 	if (sdp->wce_default_on) {
2878 		sd_first_printk(KERN_NOTICE, sdkp,
2879 				"Assuming drive cache: write back\n");
2880 		sdkp->WCE = 1;
2881 	} else {
2882 		sd_first_printk(KERN_ERR, sdkp,
2883 				"Assuming drive cache: write through\n");
2884 		sdkp->WCE = 0;
2885 	}
2886 	sdkp->RCD = 0;
2887 	sdkp->DPOFUA = 0;
2888 }
2889 
2890 /*
2891  * The ATO bit indicates whether the DIF application tag is available
2892  * for use by the operating system.
2893  */
2894 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2895 {
2896 	int res, offset;
2897 	struct scsi_device *sdp = sdkp->device;
2898 	struct scsi_mode_data data;
2899 	struct scsi_sense_hdr sshdr;
2900 
2901 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2902 		return;
2903 
2904 	if (sdkp->protection_type == 0)
2905 		return;
2906 
2907 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2908 			      sdkp->max_retries, &data, &sshdr);
2909 
2910 	if (res < 0 || !data.header_length ||
2911 	    data.length < 6) {
2912 		sd_first_printk(KERN_WARNING, sdkp,
2913 			  "getting Control mode page failed, assume no ATO\n");
2914 
2915 		if (scsi_sense_valid(&sshdr))
2916 			sd_print_sense_hdr(sdkp, &sshdr);
2917 
2918 		return;
2919 	}
2920 
2921 	offset = data.header_length + data.block_descriptor_length;
2922 
2923 	if ((buffer[offset] & 0x3f) != 0x0a) {
2924 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2925 		return;
2926 	}
2927 
2928 	if ((buffer[offset + 5] & 0x80) == 0)
2929 		return;
2930 
2931 	sdkp->ATO = 1;
2932 
2933 	return;
2934 }
2935 
2936 /**
2937  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2938  * @sdkp: disk to query
2939  */
2940 static void sd_read_block_limits(struct scsi_disk *sdkp)
2941 {
2942 	unsigned int sector_sz = sdkp->device->sector_size;
2943 	const int vpd_len = 64;
2944 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2945 
2946 	if (!buffer ||
2947 	    /* Block Limits VPD */
2948 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2949 		goto out;
2950 
2951 	blk_queue_io_min(sdkp->disk->queue,
2952 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2953 
2954 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2955 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2956 
2957 	if (buffer[3] == 0x3c) {
2958 		unsigned int lba_count, desc_count;
2959 
2960 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2961 
2962 		if (!sdkp->lbpme)
2963 			goto out;
2964 
2965 		lba_count = get_unaligned_be32(&buffer[20]);
2966 		desc_count = get_unaligned_be32(&buffer[24]);
2967 
2968 		if (lba_count && desc_count)
2969 			sdkp->max_unmap_blocks = lba_count;
2970 
2971 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2972 
2973 		if (buffer[32] & 0x80)
2974 			sdkp->unmap_alignment =
2975 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2976 
2977 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2978 
2979 			if (sdkp->max_unmap_blocks)
2980 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2981 			else
2982 				sd_config_discard(sdkp, SD_LBP_WS16);
2983 
2984 		} else {	/* LBP VPD page tells us what to use */
2985 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2986 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2987 			else if (sdkp->lbpws)
2988 				sd_config_discard(sdkp, SD_LBP_WS16);
2989 			else if (sdkp->lbpws10)
2990 				sd_config_discard(sdkp, SD_LBP_WS10);
2991 			else
2992 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2993 		}
2994 	}
2995 
2996  out:
2997 	kfree(buffer);
2998 }
2999 
3000 /**
3001  * sd_read_block_characteristics - Query block dev. characteristics
3002  * @sdkp: disk to query
3003  */
3004 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3005 {
3006 	struct request_queue *q = sdkp->disk->queue;
3007 	unsigned char *buffer;
3008 	u16 rot;
3009 	const int vpd_len = 64;
3010 
3011 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3012 
3013 	if (!buffer ||
3014 	    /* Block Device Characteristics VPD */
3015 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
3016 		goto out;
3017 
3018 	rot = get_unaligned_be16(&buffer[4]);
3019 
3020 	if (rot == 1) {
3021 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3022 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3023 	}
3024 
3025 	if (sdkp->device->type == TYPE_ZBC) {
3026 		/* Host-managed */
3027 		blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3028 	} else {
3029 		sdkp->zoned = (buffer[8] >> 4) & 3;
3030 		if (sdkp->zoned == 1) {
3031 			/* Host-aware */
3032 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3033 		} else {
3034 			/* Regular disk or drive managed disk */
3035 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3036 		}
3037 	}
3038 
3039 	if (!sdkp->first_scan)
3040 		goto out;
3041 
3042 	if (blk_queue_is_zoned(q)) {
3043 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3044 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3045 	} else {
3046 		if (sdkp->zoned == 1)
3047 			sd_printk(KERN_NOTICE, sdkp,
3048 				  "Host-aware SMR disk used as regular disk\n");
3049 		else if (sdkp->zoned == 2)
3050 			sd_printk(KERN_NOTICE, sdkp,
3051 				  "Drive-managed SMR disk\n");
3052 	}
3053 
3054  out:
3055 	kfree(buffer);
3056 }
3057 
3058 /**
3059  * sd_read_block_provisioning - Query provisioning VPD page
3060  * @sdkp: disk to query
3061  */
3062 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3063 {
3064 	unsigned char *buffer;
3065 	const int vpd_len = 8;
3066 
3067 	if (sdkp->lbpme == 0)
3068 		return;
3069 
3070 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3071 
3072 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3073 		goto out;
3074 
3075 	sdkp->lbpvpd	= 1;
3076 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
3077 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
3078 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
3079 
3080  out:
3081 	kfree(buffer);
3082 }
3083 
3084 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3085 {
3086 	struct scsi_device *sdev = sdkp->device;
3087 
3088 	if (sdev->host->no_write_same) {
3089 		sdev->no_write_same = 1;
3090 
3091 		return;
3092 	}
3093 
3094 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3095 		/* too large values might cause issues with arcmsr */
3096 		int vpd_buf_len = 64;
3097 
3098 		sdev->no_report_opcodes = 1;
3099 
3100 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3101 		 * CODES is unsupported and the device has an ATA
3102 		 * Information VPD page (SAT).
3103 		 */
3104 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3105 			sdev->no_write_same = 1;
3106 	}
3107 
3108 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3109 		sdkp->ws16 = 1;
3110 
3111 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3112 		sdkp->ws10 = 1;
3113 }
3114 
3115 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3116 {
3117 	struct scsi_device *sdev = sdkp->device;
3118 
3119 	if (!sdev->security_supported)
3120 		return;
3121 
3122 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3123 			SECURITY_PROTOCOL_IN) == 1 &&
3124 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3125 			SECURITY_PROTOCOL_OUT) == 1)
3126 		sdkp->security = 1;
3127 }
3128 
3129 /*
3130  * Determine the device's preferred I/O size for reads and writes
3131  * unless the reported value is unreasonably small, large, not a
3132  * multiple of the physical block size, or simply garbage.
3133  */
3134 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3135 				      unsigned int dev_max)
3136 {
3137 	struct scsi_device *sdp = sdkp->device;
3138 	unsigned int opt_xfer_bytes =
3139 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3140 
3141 	if (sdkp->opt_xfer_blocks == 0)
3142 		return false;
3143 
3144 	if (sdkp->opt_xfer_blocks > dev_max) {
3145 		sd_first_printk(KERN_WARNING, sdkp,
3146 				"Optimal transfer size %u logical blocks " \
3147 				"> dev_max (%u logical blocks)\n",
3148 				sdkp->opt_xfer_blocks, dev_max);
3149 		return false;
3150 	}
3151 
3152 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3153 		sd_first_printk(KERN_WARNING, sdkp,
3154 				"Optimal transfer size %u logical blocks " \
3155 				"> sd driver limit (%u logical blocks)\n",
3156 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3157 		return false;
3158 	}
3159 
3160 	if (opt_xfer_bytes < PAGE_SIZE) {
3161 		sd_first_printk(KERN_WARNING, sdkp,
3162 				"Optimal transfer size %u bytes < " \
3163 				"PAGE_SIZE (%u bytes)\n",
3164 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3165 		return false;
3166 	}
3167 
3168 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3169 		sd_first_printk(KERN_WARNING, sdkp,
3170 				"Optimal transfer size %u bytes not a " \
3171 				"multiple of physical block size (%u bytes)\n",
3172 				opt_xfer_bytes, sdkp->physical_block_size);
3173 		return false;
3174 	}
3175 
3176 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3177 			opt_xfer_bytes);
3178 	return true;
3179 }
3180 
3181 /**
3182  *	sd_revalidate_disk - called the first time a new disk is seen,
3183  *	performs disk spin up, read_capacity, etc.
3184  *	@disk: struct gendisk we care about
3185  **/
3186 static int sd_revalidate_disk(struct gendisk *disk)
3187 {
3188 	struct scsi_disk *sdkp = scsi_disk(disk);
3189 	struct scsi_device *sdp = sdkp->device;
3190 	struct request_queue *q = sdkp->disk->queue;
3191 	sector_t old_capacity = sdkp->capacity;
3192 	unsigned char *buffer;
3193 	unsigned int dev_max, rw_max;
3194 
3195 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3196 				      "sd_revalidate_disk\n"));
3197 
3198 	/*
3199 	 * If the device is offline, don't try and read capacity or any
3200 	 * of the other niceties.
3201 	 */
3202 	if (!scsi_device_online(sdp))
3203 		goto out;
3204 
3205 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3206 	if (!buffer) {
3207 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3208 			  "allocation failure.\n");
3209 		goto out;
3210 	}
3211 
3212 	sd_spinup_disk(sdkp);
3213 
3214 	/*
3215 	 * Without media there is no reason to ask; moreover, some devices
3216 	 * react badly if we do.
3217 	 */
3218 	if (sdkp->media_present) {
3219 		sd_read_capacity(sdkp, buffer);
3220 
3221 		/*
3222 		 * set the default to rotational.  All non-rotational devices
3223 		 * support the block characteristics VPD page, which will
3224 		 * cause this to be updated correctly and any device which
3225 		 * doesn't support it should be treated as rotational.
3226 		 */
3227 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3228 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3229 
3230 		if (scsi_device_supports_vpd(sdp)) {
3231 			sd_read_block_provisioning(sdkp);
3232 			sd_read_block_limits(sdkp);
3233 			sd_read_block_characteristics(sdkp);
3234 			sd_zbc_read_zones(sdkp, buffer);
3235 		}
3236 
3237 		sd_print_capacity(sdkp, old_capacity);
3238 
3239 		sd_read_write_protect_flag(sdkp, buffer);
3240 		sd_read_cache_type(sdkp, buffer);
3241 		sd_read_app_tag_own(sdkp, buffer);
3242 		sd_read_write_same(sdkp, buffer);
3243 		sd_read_security(sdkp, buffer);
3244 	}
3245 
3246 	/*
3247 	 * We now have all cache related info, determine how we deal
3248 	 * with flush requests.
3249 	 */
3250 	sd_set_flush_flag(sdkp);
3251 
3252 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3253 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3254 
3255 	/* Some devices report a maximum block count for READ/WRITE requests. */
3256 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3257 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3258 
3259 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3260 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3261 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3262 	} else {
3263 		q->limits.io_opt = 0;
3264 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3265 				      (sector_t)BLK_DEF_MAX_SECTORS);
3266 	}
3267 
3268 	/* Do not exceed controller limit */
3269 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3270 
3271 	/*
3272 	 * Only update max_sectors if previously unset or if the current value
3273 	 * exceeds the capabilities of the hardware.
3274 	 */
3275 	if (sdkp->first_scan ||
3276 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3277 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3278 		q->limits.max_sectors = rw_max;
3279 
3280 	sdkp->first_scan = 0;
3281 
3282 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3283 	sd_config_write_same(sdkp);
3284 	kfree(buffer);
3285 
3286 	/*
3287 	 * For a zoned drive, revalidating the zones can be done only once
3288 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3289 	 * capacity to 0.
3290 	 */
3291 	if (sd_zbc_revalidate_zones(sdkp))
3292 		set_capacity_and_notify(disk, 0);
3293 
3294  out:
3295 	return 0;
3296 }
3297 
3298 /**
3299  *	sd_unlock_native_capacity - unlock native capacity
3300  *	@disk: struct gendisk to set capacity for
3301  *
3302  *	Block layer calls this function if it detects that partitions
3303  *	on @disk reach beyond the end of the device.  If the SCSI host
3304  *	implements ->unlock_native_capacity() method, it's invoked to
3305  *	give it a chance to adjust the device capacity.
3306  *
3307  *	CONTEXT:
3308  *	Defined by block layer.  Might sleep.
3309  */
3310 static void sd_unlock_native_capacity(struct gendisk *disk)
3311 {
3312 	struct scsi_device *sdev = scsi_disk(disk)->device;
3313 
3314 	if (sdev->host->hostt->unlock_native_capacity)
3315 		sdev->host->hostt->unlock_native_capacity(sdev);
3316 }
3317 
3318 /**
3319  *	sd_format_disk_name - format disk name
3320  *	@prefix: name prefix - ie. "sd" for SCSI disks
3321  *	@index: index of the disk to format name for
3322  *	@buf: output buffer
3323  *	@buflen: length of the output buffer
3324  *
3325  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3326  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3327  *	which is followed by sdaaa.
3328  *
3329  *	This is basically 26 base counting with one extra 'nil' entry
3330  *	at the beginning from the second digit on and can be
3331  *	determined using similar method as 26 base conversion with the
3332  *	index shifted -1 after each digit is computed.
3333  *
3334  *	CONTEXT:
3335  *	Don't care.
3336  *
3337  *	RETURNS:
3338  *	0 on success, -errno on failure.
3339  */
3340 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3341 {
3342 	const int base = 'z' - 'a' + 1;
3343 	char *begin = buf + strlen(prefix);
3344 	char *end = buf + buflen;
3345 	char *p;
3346 	int unit;
3347 
3348 	p = end - 1;
3349 	*p = '\0';
3350 	unit = base;
3351 	do {
3352 		if (p == begin)
3353 			return -EINVAL;
3354 		*--p = 'a' + (index % unit);
3355 		index = (index / unit) - 1;
3356 	} while (index >= 0);
3357 
3358 	memmove(begin, p, end - p);
3359 	memcpy(buf, prefix, strlen(prefix));
3360 
3361 	return 0;
3362 }
3363 
3364 /**
3365  *	sd_probe - called during driver initialization and whenever a
3366  *	new scsi device is attached to the system. It is called once
3367  *	for each scsi device (not just disks) present.
3368  *	@dev: pointer to device object
3369  *
3370  *	Returns 0 if successful (or not interested in this scsi device
3371  *	(e.g. scanner)); 1 when there is an error.
3372  *
3373  *	Note: this function is invoked from the scsi mid-level.
3374  *	This function sets up the mapping between a given
3375  *	<host,channel,id,lun> (found in sdp) and new device name
3376  *	(e.g. /dev/sda). More precisely it is the block device major
3377  *	and minor number that is chosen here.
3378  *
3379  *	Assume sd_probe is not re-entrant (for time being)
3380  *	Also think about sd_probe() and sd_remove() running coincidentally.
3381  **/
3382 static int sd_probe(struct device *dev)
3383 {
3384 	struct scsi_device *sdp = to_scsi_device(dev);
3385 	struct scsi_disk *sdkp;
3386 	struct gendisk *gd;
3387 	int index;
3388 	int error;
3389 
3390 	scsi_autopm_get_device(sdp);
3391 	error = -ENODEV;
3392 	if (sdp->type != TYPE_DISK &&
3393 	    sdp->type != TYPE_ZBC &&
3394 	    sdp->type != TYPE_MOD &&
3395 	    sdp->type != TYPE_RBC)
3396 		goto out;
3397 
3398 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3399 		sdev_printk(KERN_WARNING, sdp,
3400 			    "Unsupported ZBC host-managed device.\n");
3401 		goto out;
3402 	}
3403 
3404 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3405 					"sd_probe\n"));
3406 
3407 	error = -ENOMEM;
3408 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3409 	if (!sdkp)
3410 		goto out;
3411 
3412 	gd = __alloc_disk_node(sdp->request_queue, NUMA_NO_NODE,
3413 			       &sd_bio_compl_lkclass);
3414 	if (!gd)
3415 		goto out_free;
3416 
3417 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3418 	if (index < 0) {
3419 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3420 		goto out_put;
3421 	}
3422 
3423 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3424 	if (error) {
3425 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3426 		goto out_free_index;
3427 	}
3428 
3429 	sdkp->device = sdp;
3430 	sdkp->driver = &sd_template;
3431 	sdkp->disk = gd;
3432 	sdkp->index = index;
3433 	sdkp->max_retries = SD_MAX_RETRIES;
3434 	atomic_set(&sdkp->openers, 0);
3435 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3436 
3437 	if (!sdp->request_queue->rq_timeout) {
3438 		if (sdp->type != TYPE_MOD)
3439 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3440 		else
3441 			blk_queue_rq_timeout(sdp->request_queue,
3442 					     SD_MOD_TIMEOUT);
3443 	}
3444 
3445 	device_initialize(&sdkp->dev);
3446 	sdkp->dev.parent = dev;
3447 	sdkp->dev.class = &sd_disk_class;
3448 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3449 
3450 	error = device_add(&sdkp->dev);
3451 	if (error)
3452 		goto out_free_index;
3453 
3454 	get_device(dev);
3455 	dev_set_drvdata(dev, sdkp);
3456 
3457 	gd->major = sd_major((index & 0xf0) >> 4);
3458 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3459 	gd->minors = SD_MINORS;
3460 
3461 	gd->fops = &sd_fops;
3462 	gd->private_data = &sdkp->driver;
3463 
3464 	/* defaults, until the device tells us otherwise */
3465 	sdp->sector_size = 512;
3466 	sdkp->capacity = 0;
3467 	sdkp->media_present = 1;
3468 	sdkp->write_prot = 0;
3469 	sdkp->cache_override = 0;
3470 	sdkp->WCE = 0;
3471 	sdkp->RCD = 0;
3472 	sdkp->ATO = 0;
3473 	sdkp->first_scan = 1;
3474 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3475 
3476 	sd_revalidate_disk(gd);
3477 
3478 	gd->flags = GENHD_FL_EXT_DEVT;
3479 	if (sdp->removable) {
3480 		gd->flags |= GENHD_FL_REMOVABLE;
3481 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3482 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3483 	}
3484 
3485 	blk_pm_runtime_init(sdp->request_queue, dev);
3486 	if (sdp->rpm_autosuspend) {
3487 		pm_runtime_set_autosuspend_delay(dev,
3488 			sdp->host->hostt->rpm_autosuspend_delay);
3489 	}
3490 	device_add_disk(dev, gd, NULL);
3491 	if (sdkp->capacity)
3492 		sd_dif_config_host(sdkp);
3493 
3494 	sd_revalidate_disk(gd);
3495 
3496 	if (sdkp->security) {
3497 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3498 		if (sdkp->opal_dev)
3499 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3500 	}
3501 
3502 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3503 		  sdp->removable ? "removable " : "");
3504 	scsi_autopm_put_device(sdp);
3505 
3506 	return 0;
3507 
3508  out_free_index:
3509 	ida_free(&sd_index_ida, index);
3510  out_put:
3511 	put_disk(gd);
3512  out_free:
3513 	sd_zbc_release_disk(sdkp);
3514 	kfree(sdkp);
3515  out:
3516 	scsi_autopm_put_device(sdp);
3517 	return error;
3518 }
3519 
3520 /**
3521  *	sd_remove - called whenever a scsi disk (previously recognized by
3522  *	sd_probe) is detached from the system. It is called (potentially
3523  *	multiple times) during sd module unload.
3524  *	@dev: pointer to device object
3525  *
3526  *	Note: this function is invoked from the scsi mid-level.
3527  *	This function potentially frees up a device name (e.g. /dev/sdc)
3528  *	that could be re-used by a subsequent sd_probe().
3529  *	This function is not called when the built-in sd driver is "exit-ed".
3530  **/
3531 static int sd_remove(struct device *dev)
3532 {
3533 	struct scsi_disk *sdkp;
3534 
3535 	sdkp = dev_get_drvdata(dev);
3536 	scsi_autopm_get_device(sdkp->device);
3537 
3538 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3539 	device_del(&sdkp->dev);
3540 	del_gendisk(sdkp->disk);
3541 	sd_shutdown(dev);
3542 
3543 	free_opal_dev(sdkp->opal_dev);
3544 
3545 	mutex_lock(&sd_ref_mutex);
3546 	dev_set_drvdata(dev, NULL);
3547 	put_device(&sdkp->dev);
3548 	mutex_unlock(&sd_ref_mutex);
3549 
3550 	return 0;
3551 }
3552 
3553 /**
3554  *	scsi_disk_release - Called to free the scsi_disk structure
3555  *	@dev: pointer to embedded class device
3556  *
3557  *	sd_ref_mutex must be held entering this routine.  Because it is
3558  *	called on last put, you should always use the scsi_disk_get()
3559  *	scsi_disk_put() helpers which manipulate the semaphore directly
3560  *	and never do a direct put_device.
3561  **/
3562 static void scsi_disk_release(struct device *dev)
3563 {
3564 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3565 	struct gendisk *disk = sdkp->disk;
3566 	struct request_queue *q = disk->queue;
3567 
3568 	ida_free(&sd_index_ida, sdkp->index);
3569 
3570 	/*
3571 	 * Wait until all requests that are in progress have completed.
3572 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3573 	 * due to clearing the disk->private_data pointer. Wait from inside
3574 	 * scsi_disk_release() instead of from sd_release() to avoid that
3575 	 * freezing and unfreezing the request queue affects user space I/O
3576 	 * in case multiple processes open a /dev/sd... node concurrently.
3577 	 */
3578 	blk_mq_freeze_queue(q);
3579 	blk_mq_unfreeze_queue(q);
3580 
3581 	disk->private_data = NULL;
3582 	put_disk(disk);
3583 	put_device(&sdkp->device->sdev_gendev);
3584 
3585 	sd_zbc_release_disk(sdkp);
3586 
3587 	kfree(sdkp);
3588 }
3589 
3590 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3591 {
3592 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3593 	struct scsi_sense_hdr sshdr;
3594 	struct scsi_device *sdp = sdkp->device;
3595 	int res;
3596 
3597 	if (start)
3598 		cmd[4] |= 1;	/* START */
3599 
3600 	if (sdp->start_stop_pwr_cond)
3601 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3602 
3603 	if (!scsi_device_online(sdp))
3604 		return -ENODEV;
3605 
3606 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3607 			SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3608 	if (res) {
3609 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3610 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3611 			sd_print_sense_hdr(sdkp, &sshdr);
3612 			/* 0x3a is medium not present */
3613 			if (sshdr.asc == 0x3a)
3614 				res = 0;
3615 		}
3616 	}
3617 
3618 	/* SCSI error codes must not go to the generic layer */
3619 	if (res)
3620 		return -EIO;
3621 
3622 	return 0;
3623 }
3624 
3625 /*
3626  * Send a SYNCHRONIZE CACHE instruction down to the device through
3627  * the normal SCSI command structure.  Wait for the command to
3628  * complete.
3629  */
3630 static void sd_shutdown(struct device *dev)
3631 {
3632 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3633 
3634 	if (!sdkp)
3635 		return;         /* this can happen */
3636 
3637 	if (pm_runtime_suspended(dev))
3638 		return;
3639 
3640 	if (sdkp->WCE && sdkp->media_present) {
3641 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3642 		sd_sync_cache(sdkp, NULL);
3643 	}
3644 
3645 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3646 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3647 		sd_start_stop_device(sdkp, 0);
3648 	}
3649 }
3650 
3651 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3652 {
3653 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3654 	struct scsi_sense_hdr sshdr;
3655 	int ret = 0;
3656 
3657 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3658 		return 0;
3659 
3660 	if (sdkp->WCE && sdkp->media_present) {
3661 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3662 		ret = sd_sync_cache(sdkp, &sshdr);
3663 
3664 		if (ret) {
3665 			/* ignore OFFLINE device */
3666 			if (ret == -ENODEV)
3667 				return 0;
3668 
3669 			if (!scsi_sense_valid(&sshdr) ||
3670 			    sshdr.sense_key != ILLEGAL_REQUEST)
3671 				return ret;
3672 
3673 			/*
3674 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3675 			 * doesn't support sync. There's not much to do and
3676 			 * suspend shouldn't fail.
3677 			 */
3678 			ret = 0;
3679 		}
3680 	}
3681 
3682 	if (sdkp->device->manage_start_stop) {
3683 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3684 		/* an error is not worth aborting a system sleep */
3685 		ret = sd_start_stop_device(sdkp, 0);
3686 		if (ignore_stop_errors)
3687 			ret = 0;
3688 	}
3689 
3690 	return ret;
3691 }
3692 
3693 static int sd_suspend_system(struct device *dev)
3694 {
3695 	return sd_suspend_common(dev, true);
3696 }
3697 
3698 static int sd_suspend_runtime(struct device *dev)
3699 {
3700 	return sd_suspend_common(dev, false);
3701 }
3702 
3703 static int sd_resume(struct device *dev)
3704 {
3705 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3706 	int ret;
3707 
3708 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3709 		return 0;
3710 
3711 	if (!sdkp->device->manage_start_stop)
3712 		return 0;
3713 
3714 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3715 	ret = sd_start_stop_device(sdkp, 1);
3716 	if (!ret)
3717 		opal_unlock_from_suspend(sdkp->opal_dev);
3718 	return ret;
3719 }
3720 
3721 /**
3722  *	init_sd - entry point for this driver (both when built in or when
3723  *	a module).
3724  *
3725  *	Note: this function registers this driver with the scsi mid-level.
3726  **/
3727 static int __init init_sd(void)
3728 {
3729 	int majors = 0, i, err;
3730 
3731 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3732 
3733 	for (i = 0; i < SD_MAJORS; i++) {
3734 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3735 			continue;
3736 		majors++;
3737 	}
3738 
3739 	if (!majors)
3740 		return -ENODEV;
3741 
3742 	err = class_register(&sd_disk_class);
3743 	if (err)
3744 		goto err_out;
3745 
3746 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3747 					 0, 0, NULL);
3748 	if (!sd_cdb_cache) {
3749 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3750 		err = -ENOMEM;
3751 		goto err_out_class;
3752 	}
3753 
3754 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3755 	if (!sd_cdb_pool) {
3756 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3757 		err = -ENOMEM;
3758 		goto err_out_cache;
3759 	}
3760 
3761 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3762 	if (!sd_page_pool) {
3763 		printk(KERN_ERR "sd: can't init discard page pool\n");
3764 		err = -ENOMEM;
3765 		goto err_out_ppool;
3766 	}
3767 
3768 	err = scsi_register_driver(&sd_template.gendrv);
3769 	if (err)
3770 		goto err_out_driver;
3771 
3772 	return 0;
3773 
3774 err_out_driver:
3775 	mempool_destroy(sd_page_pool);
3776 
3777 err_out_ppool:
3778 	mempool_destroy(sd_cdb_pool);
3779 
3780 err_out_cache:
3781 	kmem_cache_destroy(sd_cdb_cache);
3782 
3783 err_out_class:
3784 	class_unregister(&sd_disk_class);
3785 err_out:
3786 	for (i = 0; i < SD_MAJORS; i++)
3787 		unregister_blkdev(sd_major(i), "sd");
3788 	return err;
3789 }
3790 
3791 /**
3792  *	exit_sd - exit point for this driver (when it is a module).
3793  *
3794  *	Note: this function unregisters this driver from the scsi mid-level.
3795  **/
3796 static void __exit exit_sd(void)
3797 {
3798 	int i;
3799 
3800 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3801 
3802 	scsi_unregister_driver(&sd_template.gendrv);
3803 	mempool_destroy(sd_cdb_pool);
3804 	mempool_destroy(sd_page_pool);
3805 	kmem_cache_destroy(sd_cdb_cache);
3806 
3807 	class_unregister(&sd_disk_class);
3808 
3809 	for (i = 0; i < SD_MAJORS; i++)
3810 		unregister_blkdev(sd_major(i), "sd");
3811 }
3812 
3813 module_init(init_sd);
3814 module_exit(exit_sd);
3815 
3816 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3817 {
3818 	scsi_print_sense_hdr(sdkp->device,
3819 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3820 }
3821 
3822 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3823 {
3824 	const char *hb_string = scsi_hostbyte_string(result);
3825 
3826 	if (hb_string)
3827 		sd_printk(KERN_INFO, sdkp,
3828 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3829 			  hb_string ? hb_string : "invalid",
3830 			  "DRIVER_OK");
3831 	else
3832 		sd_printk(KERN_INFO, sdkp,
3833 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3834 			  msg, host_byte(result), "DRIVER_OK");
3835 }
3836