xref: /openbmc/linux/drivers/scsi/sd.c (revision 9e6bfd42b14b45737cae8bc84c759f1874949b8b)
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/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/major.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
56 #include <linux/pr.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
60 
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
70 
71 #include "sd.h"
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
74 
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
78 
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 
100 #define SD_MINORS	16
101 
102 static void sd_config_discard(struct scsi_disk *, unsigned int);
103 static void sd_config_write_same(struct scsi_disk *);
104 static int  sd_revalidate_disk(struct gendisk *);
105 static void sd_unlock_native_capacity(struct gendisk *disk);
106 static int  sd_probe(struct device *);
107 static int  sd_remove(struct device *);
108 static void sd_shutdown(struct device *);
109 static int sd_suspend_system(struct device *);
110 static int sd_suspend_runtime(struct device *);
111 static int sd_resume_system(struct device *);
112 static int sd_resume_runtime(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 static struct kmem_cache *sd_cdb_cache;
125 static mempool_t *sd_page_pool;
126 static struct lock_class_key sd_bio_compl_lkclass;
127 
128 static const char *sd_cache_types[] = {
129 	"write through", "none", "write back",
130 	"write back, no read (daft)"
131 };
132 
133 static void sd_set_flush_flag(struct scsi_disk *sdkp)
134 {
135 	bool wc = false, fua = false;
136 
137 	if (sdkp->WCE) {
138 		wc = true;
139 		if (sdkp->DPOFUA)
140 			fua = true;
141 	}
142 
143 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
144 }
145 
146 static ssize_t
147 cache_type_store(struct device *dev, struct device_attribute *attr,
148 		 const char *buf, size_t count)
149 {
150 	int ct, rcd, wce, sp;
151 	struct scsi_disk *sdkp = to_scsi_disk(dev);
152 	struct scsi_device *sdp = sdkp->device;
153 	char buffer[64];
154 	char *buffer_data;
155 	struct scsi_mode_data data;
156 	struct scsi_sense_hdr sshdr;
157 	static const char temp[] = "temporary ";
158 	int len;
159 
160 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
161 		/* no cache control on RBC devices; theoretically they
162 		 * can do it, but there's probably so many exceptions
163 		 * it's not worth the risk */
164 		return -EINVAL;
165 
166 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
167 		buf += sizeof(temp) - 1;
168 		sdkp->cache_override = 1;
169 	} else {
170 		sdkp->cache_override = 0;
171 	}
172 
173 	ct = sysfs_match_string(sd_cache_types, buf);
174 	if (ct < 0)
175 		return -EINVAL;
176 
177 	rcd = ct & 0x01 ? 1 : 0;
178 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
179 
180 	if (sdkp->cache_override) {
181 		sdkp->WCE = wce;
182 		sdkp->RCD = rcd;
183 		sd_set_flush_flag(sdkp);
184 		return count;
185 	}
186 
187 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
188 			    sdkp->max_retries, &data, NULL))
189 		return -EINVAL;
190 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
191 		  data.block_descriptor_length);
192 	buffer_data = buffer + data.header_length +
193 		data.block_descriptor_length;
194 	buffer_data[2] &= ~0x05;
195 	buffer_data[2] |= wce << 2 | rcd;
196 	sp = buffer_data[0] & 0x80 ? 1 : 0;
197 	buffer_data[0] &= ~0x80;
198 
199 	/*
200 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
201 	 * received mode parameter buffer before doing MODE SELECT.
202 	 */
203 	data.device_specific = 0;
204 
205 	if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
206 			     sdkp->max_retries, &data, &sshdr)) {
207 		if (scsi_sense_valid(&sshdr))
208 			sd_print_sense_hdr(sdkp, &sshdr);
209 		return -EINVAL;
210 	}
211 	sd_revalidate_disk(sdkp->disk);
212 	return count;
213 }
214 
215 static ssize_t
216 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
217 		       char *buf)
218 {
219 	struct scsi_disk *sdkp = to_scsi_disk(dev);
220 	struct scsi_device *sdp = sdkp->device;
221 
222 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
223 }
224 
225 static ssize_t
226 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
227 			const char *buf, size_t count)
228 {
229 	struct scsi_disk *sdkp = to_scsi_disk(dev);
230 	struct scsi_device *sdp = sdkp->device;
231 	bool v;
232 
233 	if (!capable(CAP_SYS_ADMIN))
234 		return -EACCES;
235 
236 	if (kstrtobool(buf, &v))
237 		return -EINVAL;
238 
239 	sdp->manage_start_stop = v;
240 
241 	return count;
242 }
243 static DEVICE_ATTR_RW(manage_start_stop);
244 
245 static ssize_t
246 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
247 {
248 	struct scsi_disk *sdkp = to_scsi_disk(dev);
249 
250 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
251 }
252 
253 static ssize_t
254 allow_restart_store(struct device *dev, struct device_attribute *attr,
255 		    const char *buf, size_t count)
256 {
257 	bool v;
258 	struct scsi_disk *sdkp = to_scsi_disk(dev);
259 	struct scsi_device *sdp = sdkp->device;
260 
261 	if (!capable(CAP_SYS_ADMIN))
262 		return -EACCES;
263 
264 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
265 		return -EINVAL;
266 
267 	if (kstrtobool(buf, &v))
268 		return -EINVAL;
269 
270 	sdp->allow_restart = v;
271 
272 	return count;
273 }
274 static DEVICE_ATTR_RW(allow_restart);
275 
276 static ssize_t
277 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
278 {
279 	struct scsi_disk *sdkp = to_scsi_disk(dev);
280 	int ct = sdkp->RCD + 2*sdkp->WCE;
281 
282 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
283 }
284 static DEVICE_ATTR_RW(cache_type);
285 
286 static ssize_t
287 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
288 {
289 	struct scsi_disk *sdkp = to_scsi_disk(dev);
290 
291 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
292 }
293 static DEVICE_ATTR_RO(FUA);
294 
295 static ssize_t
296 protection_type_show(struct device *dev, struct device_attribute *attr,
297 		     char *buf)
298 {
299 	struct scsi_disk *sdkp = to_scsi_disk(dev);
300 
301 	return sprintf(buf, "%u\n", sdkp->protection_type);
302 }
303 
304 static ssize_t
305 protection_type_store(struct device *dev, struct device_attribute *attr,
306 		      const char *buf, size_t count)
307 {
308 	struct scsi_disk *sdkp = to_scsi_disk(dev);
309 	unsigned int val;
310 	int err;
311 
312 	if (!capable(CAP_SYS_ADMIN))
313 		return -EACCES;
314 
315 	err = kstrtouint(buf, 10, &val);
316 
317 	if (err)
318 		return err;
319 
320 	if (val <= T10_PI_TYPE3_PROTECTION)
321 		sdkp->protection_type = val;
322 
323 	return count;
324 }
325 static DEVICE_ATTR_RW(protection_type);
326 
327 static ssize_t
328 protection_mode_show(struct device *dev, struct device_attribute *attr,
329 		     char *buf)
330 {
331 	struct scsi_disk *sdkp = to_scsi_disk(dev);
332 	struct scsi_device *sdp = sdkp->device;
333 	unsigned int dif, dix;
334 
335 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
336 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
337 
338 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
339 		dif = 0;
340 		dix = 1;
341 	}
342 
343 	if (!dif && !dix)
344 		return sprintf(buf, "none\n");
345 
346 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
347 }
348 static DEVICE_ATTR_RO(protection_mode);
349 
350 static ssize_t
351 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
352 {
353 	struct scsi_disk *sdkp = to_scsi_disk(dev);
354 
355 	return sprintf(buf, "%u\n", sdkp->ATO);
356 }
357 static DEVICE_ATTR_RO(app_tag_own);
358 
359 static ssize_t
360 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
361 		       char *buf)
362 {
363 	struct scsi_disk *sdkp = to_scsi_disk(dev);
364 
365 	return sprintf(buf, "%u\n", sdkp->lbpme);
366 }
367 static DEVICE_ATTR_RO(thin_provisioning);
368 
369 /* sysfs_match_string() requires dense arrays */
370 static const char *lbp_mode[] = {
371 	[SD_LBP_FULL]		= "full",
372 	[SD_LBP_UNMAP]		= "unmap",
373 	[SD_LBP_WS16]		= "writesame_16",
374 	[SD_LBP_WS10]		= "writesame_10",
375 	[SD_LBP_ZERO]		= "writesame_zero",
376 	[SD_LBP_DISABLE]	= "disabled",
377 };
378 
379 static ssize_t
380 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
381 		       char *buf)
382 {
383 	struct scsi_disk *sdkp = to_scsi_disk(dev);
384 
385 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
386 }
387 
388 static ssize_t
389 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
390 			const char *buf, size_t count)
391 {
392 	struct scsi_disk *sdkp = to_scsi_disk(dev);
393 	struct scsi_device *sdp = sdkp->device;
394 	int mode;
395 
396 	if (!capable(CAP_SYS_ADMIN))
397 		return -EACCES;
398 
399 	if (sd_is_zoned(sdkp)) {
400 		sd_config_discard(sdkp, SD_LBP_DISABLE);
401 		return count;
402 	}
403 
404 	if (sdp->type != TYPE_DISK)
405 		return -EINVAL;
406 
407 	mode = sysfs_match_string(lbp_mode, buf);
408 	if (mode < 0)
409 		return -EINVAL;
410 
411 	sd_config_discard(sdkp, mode);
412 
413 	return count;
414 }
415 static DEVICE_ATTR_RW(provisioning_mode);
416 
417 /* sysfs_match_string() requires dense arrays */
418 static const char *zeroing_mode[] = {
419 	[SD_ZERO_WRITE]		= "write",
420 	[SD_ZERO_WS]		= "writesame",
421 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
422 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
423 };
424 
425 static ssize_t
426 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
427 		  char *buf)
428 {
429 	struct scsi_disk *sdkp = to_scsi_disk(dev);
430 
431 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
432 }
433 
434 static ssize_t
435 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
436 		   const char *buf, size_t count)
437 {
438 	struct scsi_disk *sdkp = to_scsi_disk(dev);
439 	int mode;
440 
441 	if (!capable(CAP_SYS_ADMIN))
442 		return -EACCES;
443 
444 	mode = sysfs_match_string(zeroing_mode, buf);
445 	if (mode < 0)
446 		return -EINVAL;
447 
448 	sdkp->zeroing_mode = mode;
449 
450 	return count;
451 }
452 static DEVICE_ATTR_RW(zeroing_mode);
453 
454 static ssize_t
455 max_medium_access_timeouts_show(struct device *dev,
456 				struct device_attribute *attr, char *buf)
457 {
458 	struct scsi_disk *sdkp = to_scsi_disk(dev);
459 
460 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
461 }
462 
463 static ssize_t
464 max_medium_access_timeouts_store(struct device *dev,
465 				 struct device_attribute *attr, const char *buf,
466 				 size_t count)
467 {
468 	struct scsi_disk *sdkp = to_scsi_disk(dev);
469 	int err;
470 
471 	if (!capable(CAP_SYS_ADMIN))
472 		return -EACCES;
473 
474 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
475 
476 	return err ? err : count;
477 }
478 static DEVICE_ATTR_RW(max_medium_access_timeouts);
479 
480 static ssize_t
481 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
482 			   char *buf)
483 {
484 	struct scsi_disk *sdkp = to_scsi_disk(dev);
485 
486 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
487 }
488 
489 static ssize_t
490 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
491 			    const char *buf, size_t count)
492 {
493 	struct scsi_disk *sdkp = to_scsi_disk(dev);
494 	struct scsi_device *sdp = sdkp->device;
495 	unsigned long max;
496 	int err;
497 
498 	if (!capable(CAP_SYS_ADMIN))
499 		return -EACCES;
500 
501 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
502 		return -EINVAL;
503 
504 	err = kstrtoul(buf, 10, &max);
505 
506 	if (err)
507 		return err;
508 
509 	if (max == 0)
510 		sdp->no_write_same = 1;
511 	else if (max <= SD_MAX_WS16_BLOCKS) {
512 		sdp->no_write_same = 0;
513 		sdkp->max_ws_blocks = max;
514 	}
515 
516 	sd_config_write_same(sdkp);
517 
518 	return count;
519 }
520 static DEVICE_ATTR_RW(max_write_same_blocks);
521 
522 static ssize_t
523 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
524 {
525 	struct scsi_disk *sdkp = to_scsi_disk(dev);
526 
527 	if (sdkp->device->type == TYPE_ZBC)
528 		return sprintf(buf, "host-managed\n");
529 	if (sdkp->zoned == 1)
530 		return sprintf(buf, "host-aware\n");
531 	if (sdkp->zoned == 2)
532 		return sprintf(buf, "drive-managed\n");
533 	return sprintf(buf, "none\n");
534 }
535 static DEVICE_ATTR_RO(zoned_cap);
536 
537 static ssize_t
538 max_retries_store(struct device *dev, struct device_attribute *attr,
539 		  const char *buf, size_t count)
540 {
541 	struct scsi_disk *sdkp = to_scsi_disk(dev);
542 	struct scsi_device *sdev = sdkp->device;
543 	int retries, err;
544 
545 	err = kstrtoint(buf, 10, &retries);
546 	if (err)
547 		return err;
548 
549 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
550 		sdkp->max_retries = retries;
551 		return count;
552 	}
553 
554 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
555 		    SD_MAX_RETRIES);
556 	return -EINVAL;
557 }
558 
559 static ssize_t
560 max_retries_show(struct device *dev, struct device_attribute *attr,
561 		 char *buf)
562 {
563 	struct scsi_disk *sdkp = to_scsi_disk(dev);
564 
565 	return sprintf(buf, "%d\n", sdkp->max_retries);
566 }
567 
568 static DEVICE_ATTR_RW(max_retries);
569 
570 static struct attribute *sd_disk_attrs[] = {
571 	&dev_attr_cache_type.attr,
572 	&dev_attr_FUA.attr,
573 	&dev_attr_allow_restart.attr,
574 	&dev_attr_manage_start_stop.attr,
575 	&dev_attr_protection_type.attr,
576 	&dev_attr_protection_mode.attr,
577 	&dev_attr_app_tag_own.attr,
578 	&dev_attr_thin_provisioning.attr,
579 	&dev_attr_provisioning_mode.attr,
580 	&dev_attr_zeroing_mode.attr,
581 	&dev_attr_max_write_same_blocks.attr,
582 	&dev_attr_max_medium_access_timeouts.attr,
583 	&dev_attr_zoned_cap.attr,
584 	&dev_attr_max_retries.attr,
585 	NULL,
586 };
587 ATTRIBUTE_GROUPS(sd_disk);
588 
589 static struct class sd_disk_class = {
590 	.name		= "scsi_disk",
591 	.owner		= THIS_MODULE,
592 	.dev_release	= scsi_disk_release,
593 	.dev_groups	= sd_disk_groups,
594 };
595 
596 static const struct dev_pm_ops sd_pm_ops = {
597 	.suspend		= sd_suspend_system,
598 	.resume			= sd_resume_system,
599 	.poweroff		= sd_suspend_system,
600 	.restore		= sd_resume_system,
601 	.runtime_suspend	= sd_suspend_runtime,
602 	.runtime_resume		= sd_resume_runtime,
603 };
604 
605 static struct scsi_driver sd_template = {
606 	.gendrv = {
607 		.name		= "sd",
608 		.owner		= THIS_MODULE,
609 		.probe		= sd_probe,
610 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
611 		.remove		= sd_remove,
612 		.shutdown	= sd_shutdown,
613 		.pm		= &sd_pm_ops,
614 	},
615 	.rescan			= sd_rescan,
616 	.init_command		= sd_init_command,
617 	.uninit_command		= sd_uninit_command,
618 	.done			= sd_done,
619 	.eh_action		= sd_eh_action,
620 	.eh_reset		= sd_eh_reset,
621 };
622 
623 /*
624  * Don't request a new module, as that could deadlock in multipath
625  * environment.
626  */
627 static void sd_default_probe(dev_t devt)
628 {
629 }
630 
631 /*
632  * Device no to disk mapping:
633  *
634  *       major         disc2     disc  p1
635  *   |............|.............|....|....| <- dev_t
636  *    31        20 19          8 7  4 3  0
637  *
638  * Inside a major, we have 16k disks, however mapped non-
639  * contiguously. The first 16 disks are for major0, the next
640  * ones with major1, ... Disk 256 is for major0 again, disk 272
641  * for major1, ...
642  * As we stay compatible with our numbering scheme, we can reuse
643  * the well-know SCSI majors 8, 65--71, 136--143.
644  */
645 static int sd_major(int major_idx)
646 {
647 	switch (major_idx) {
648 	case 0:
649 		return SCSI_DISK0_MAJOR;
650 	case 1 ... 7:
651 		return SCSI_DISK1_MAJOR + major_idx - 1;
652 	case 8 ... 15:
653 		return SCSI_DISK8_MAJOR + major_idx - 8;
654 	default:
655 		BUG();
656 		return 0;	/* shut up gcc */
657 	}
658 }
659 
660 #ifdef CONFIG_BLK_SED_OPAL
661 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
662 		size_t len, bool send)
663 {
664 	struct scsi_disk *sdkp = data;
665 	struct scsi_device *sdev = sdkp->device;
666 	u8 cdb[12] = { 0, };
667 	const struct scsi_exec_args exec_args = {
668 		.req_flags = BLK_MQ_REQ_PM,
669 	};
670 	int ret;
671 
672 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
673 	cdb[1] = secp;
674 	put_unaligned_be16(spsp, &cdb[2]);
675 	put_unaligned_be32(len, &cdb[6]);
676 
677 	ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
678 			       buffer, len, SD_TIMEOUT, sdkp->max_retries,
679 			       &exec_args);
680 	return ret <= 0 ? ret : -EIO;
681 }
682 #endif /* CONFIG_BLK_SED_OPAL */
683 
684 /*
685  * Look up the DIX operation based on whether the command is read or
686  * write and whether dix and dif are enabled.
687  */
688 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
689 {
690 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
691 	static const unsigned int ops[] = {	/* wrt dix dif */
692 		SCSI_PROT_NORMAL,		/*  0	0   0  */
693 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
694 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
695 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
696 		SCSI_PROT_NORMAL,		/*  1	0   0  */
697 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
698 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
699 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
700 	};
701 
702 	return ops[write << 2 | dix << 1 | dif];
703 }
704 
705 /*
706  * Returns a mask of the protection flags that are valid for a given DIX
707  * operation.
708  */
709 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
710 {
711 	static const unsigned int flag_mask[] = {
712 		[SCSI_PROT_NORMAL]		= 0,
713 
714 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
715 						  SCSI_PROT_GUARD_CHECK |
716 						  SCSI_PROT_REF_CHECK |
717 						  SCSI_PROT_REF_INCREMENT,
718 
719 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
720 						  SCSI_PROT_IP_CHECKSUM,
721 
722 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
723 						  SCSI_PROT_GUARD_CHECK |
724 						  SCSI_PROT_REF_CHECK |
725 						  SCSI_PROT_REF_INCREMENT |
726 						  SCSI_PROT_IP_CHECKSUM,
727 
728 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
729 						  SCSI_PROT_REF_INCREMENT,
730 
731 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
732 						  SCSI_PROT_REF_CHECK |
733 						  SCSI_PROT_REF_INCREMENT |
734 						  SCSI_PROT_IP_CHECKSUM,
735 
736 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
737 						  SCSI_PROT_GUARD_CHECK |
738 						  SCSI_PROT_REF_CHECK |
739 						  SCSI_PROT_REF_INCREMENT |
740 						  SCSI_PROT_IP_CHECKSUM,
741 	};
742 
743 	return flag_mask[prot_op];
744 }
745 
746 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
747 					   unsigned int dix, unsigned int dif)
748 {
749 	struct request *rq = scsi_cmd_to_rq(scmd);
750 	struct bio *bio = rq->bio;
751 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
752 	unsigned int protect = 0;
753 
754 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
755 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
756 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
757 
758 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
759 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
760 	}
761 
762 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
763 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
764 
765 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
766 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
767 	}
768 
769 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
770 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
771 
772 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
773 			protect = 3 << 5;	/* Disable target PI checking */
774 		else
775 			protect = 1 << 5;	/* Enable target PI checking */
776 	}
777 
778 	scsi_set_prot_op(scmd, prot_op);
779 	scsi_set_prot_type(scmd, dif);
780 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
781 
782 	return protect;
783 }
784 
785 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
786 {
787 	struct request_queue *q = sdkp->disk->queue;
788 	unsigned int logical_block_size = sdkp->device->sector_size;
789 	unsigned int max_blocks = 0;
790 
791 	q->limits.discard_alignment =
792 		sdkp->unmap_alignment * logical_block_size;
793 	q->limits.discard_granularity =
794 		max(sdkp->physical_block_size,
795 		    sdkp->unmap_granularity * logical_block_size);
796 	sdkp->provisioning_mode = mode;
797 
798 	switch (mode) {
799 
800 	case SD_LBP_FULL:
801 	case SD_LBP_DISABLE:
802 		blk_queue_max_discard_sectors(q, 0);
803 		return;
804 
805 	case SD_LBP_UNMAP:
806 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
807 					  (u32)SD_MAX_WS16_BLOCKS);
808 		break;
809 
810 	case SD_LBP_WS16:
811 		if (sdkp->device->unmap_limit_for_ws)
812 			max_blocks = sdkp->max_unmap_blocks;
813 		else
814 			max_blocks = sdkp->max_ws_blocks;
815 
816 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
817 		break;
818 
819 	case SD_LBP_WS10:
820 		if (sdkp->device->unmap_limit_for_ws)
821 			max_blocks = sdkp->max_unmap_blocks;
822 		else
823 			max_blocks = sdkp->max_ws_blocks;
824 
825 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
826 		break;
827 
828 	case SD_LBP_ZERO:
829 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
830 					  (u32)SD_MAX_WS10_BLOCKS);
831 		break;
832 	}
833 
834 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
835 }
836 
837 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
838 {
839 	struct page *page;
840 
841 	page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
842 	if (!page)
843 		return NULL;
844 	clear_highpage(page);
845 	bvec_set_page(&rq->special_vec, page, data_len, 0);
846 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
847 	return bvec_virt(&rq->special_vec);
848 }
849 
850 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
851 {
852 	struct scsi_device *sdp = cmd->device;
853 	struct request *rq = scsi_cmd_to_rq(cmd);
854 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
855 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
856 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
857 	unsigned int data_len = 24;
858 	char *buf;
859 
860 	buf = sd_set_special_bvec(rq, data_len);
861 	if (!buf)
862 		return BLK_STS_RESOURCE;
863 
864 	cmd->cmd_len = 10;
865 	cmd->cmnd[0] = UNMAP;
866 	cmd->cmnd[8] = 24;
867 
868 	put_unaligned_be16(6 + 16, &buf[0]);
869 	put_unaligned_be16(16, &buf[2]);
870 	put_unaligned_be64(lba, &buf[8]);
871 	put_unaligned_be32(nr_blocks, &buf[16]);
872 
873 	cmd->allowed = sdkp->max_retries;
874 	cmd->transfersize = data_len;
875 	rq->timeout = SD_TIMEOUT;
876 
877 	return scsi_alloc_sgtables(cmd);
878 }
879 
880 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
881 		bool unmap)
882 {
883 	struct scsi_device *sdp = cmd->device;
884 	struct request *rq = scsi_cmd_to_rq(cmd);
885 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
886 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
887 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
888 	u32 data_len = sdp->sector_size;
889 
890 	if (!sd_set_special_bvec(rq, data_len))
891 		return BLK_STS_RESOURCE;
892 
893 	cmd->cmd_len = 16;
894 	cmd->cmnd[0] = WRITE_SAME_16;
895 	if (unmap)
896 		cmd->cmnd[1] = 0x8; /* UNMAP */
897 	put_unaligned_be64(lba, &cmd->cmnd[2]);
898 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
899 
900 	cmd->allowed = sdkp->max_retries;
901 	cmd->transfersize = data_len;
902 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
903 
904 	return scsi_alloc_sgtables(cmd);
905 }
906 
907 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
908 		bool unmap)
909 {
910 	struct scsi_device *sdp = cmd->device;
911 	struct request *rq = scsi_cmd_to_rq(cmd);
912 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
913 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
914 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
915 	u32 data_len = sdp->sector_size;
916 
917 	if (!sd_set_special_bvec(rq, data_len))
918 		return BLK_STS_RESOURCE;
919 
920 	cmd->cmd_len = 10;
921 	cmd->cmnd[0] = WRITE_SAME;
922 	if (unmap)
923 		cmd->cmnd[1] = 0x8; /* UNMAP */
924 	put_unaligned_be32(lba, &cmd->cmnd[2]);
925 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
926 
927 	cmd->allowed = sdkp->max_retries;
928 	cmd->transfersize = data_len;
929 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
930 
931 	return scsi_alloc_sgtables(cmd);
932 }
933 
934 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
935 {
936 	struct request *rq = scsi_cmd_to_rq(cmd);
937 	struct scsi_device *sdp = cmd->device;
938 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
939 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
940 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
941 
942 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
943 		switch (sdkp->zeroing_mode) {
944 		case SD_ZERO_WS16_UNMAP:
945 			return sd_setup_write_same16_cmnd(cmd, true);
946 		case SD_ZERO_WS10_UNMAP:
947 			return sd_setup_write_same10_cmnd(cmd, true);
948 		}
949 	}
950 
951 	if (sdp->no_write_same) {
952 		rq->rq_flags |= RQF_QUIET;
953 		return BLK_STS_TARGET;
954 	}
955 
956 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
957 		return sd_setup_write_same16_cmnd(cmd, false);
958 
959 	return sd_setup_write_same10_cmnd(cmd, false);
960 }
961 
962 static void sd_config_write_same(struct scsi_disk *sdkp)
963 {
964 	struct request_queue *q = sdkp->disk->queue;
965 	unsigned int logical_block_size = sdkp->device->sector_size;
966 
967 	if (sdkp->device->no_write_same) {
968 		sdkp->max_ws_blocks = 0;
969 		goto out;
970 	}
971 
972 	/* Some devices can not handle block counts above 0xffff despite
973 	 * supporting WRITE SAME(16). Consequently we default to 64k
974 	 * blocks per I/O unless the device explicitly advertises a
975 	 * bigger limit.
976 	 */
977 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
978 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
979 						   (u32)SD_MAX_WS16_BLOCKS);
980 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
981 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
982 						   (u32)SD_MAX_WS10_BLOCKS);
983 	else {
984 		sdkp->device->no_write_same = 1;
985 		sdkp->max_ws_blocks = 0;
986 	}
987 
988 	if (sdkp->lbprz && sdkp->lbpws)
989 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
990 	else if (sdkp->lbprz && sdkp->lbpws10)
991 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
992 	else if (sdkp->max_ws_blocks)
993 		sdkp->zeroing_mode = SD_ZERO_WS;
994 	else
995 		sdkp->zeroing_mode = SD_ZERO_WRITE;
996 
997 	if (sdkp->max_ws_blocks &&
998 	    sdkp->physical_block_size > logical_block_size) {
999 		/*
1000 		 * Reporting a maximum number of blocks that is not aligned
1001 		 * on the device physical size would cause a large write same
1002 		 * request to be split into physically unaligned chunks by
1003 		 * __blkdev_issue_write_zeroes() even if the caller of this
1004 		 * functions took care to align the large request. So make sure
1005 		 * the maximum reported is aligned to the device physical block
1006 		 * size. This is only an optional optimization for regular
1007 		 * disks, but this is mandatory to avoid failure of large write
1008 		 * same requests directed at sequential write required zones of
1009 		 * host-managed ZBC disks.
1010 		 */
1011 		sdkp->max_ws_blocks =
1012 			round_down(sdkp->max_ws_blocks,
1013 				   bytes_to_logical(sdkp->device,
1014 						    sdkp->physical_block_size));
1015 	}
1016 
1017 out:
1018 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1019 					 (logical_block_size >> 9));
1020 }
1021 
1022 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1023 {
1024 	struct request *rq = scsi_cmd_to_rq(cmd);
1025 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1026 
1027 	/* flush requests don't perform I/O, zero the S/G table */
1028 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1029 
1030 	if (cmd->device->use_16_for_sync) {
1031 		cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1032 		cmd->cmd_len = 16;
1033 	} else {
1034 		cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1035 		cmd->cmd_len = 10;
1036 	}
1037 	cmd->transfersize = 0;
1038 	cmd->allowed = sdkp->max_retries;
1039 
1040 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1041 	return BLK_STS_OK;
1042 }
1043 
1044 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1045 				       sector_t lba, unsigned int nr_blocks,
1046 				       unsigned char flags)
1047 {
1048 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1049 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1050 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1051 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1052 	cmd->cmnd[10] = flags;
1053 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1054 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1055 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1056 
1057 	return BLK_STS_OK;
1058 }
1059 
1060 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1061 				       sector_t lba, unsigned int nr_blocks,
1062 				       unsigned char flags)
1063 {
1064 	cmd->cmd_len  = 16;
1065 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1066 	cmd->cmnd[1]  = flags;
1067 	cmd->cmnd[14] = 0;
1068 	cmd->cmnd[15] = 0;
1069 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1070 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1071 
1072 	return BLK_STS_OK;
1073 }
1074 
1075 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1076 				       sector_t lba, unsigned int nr_blocks,
1077 				       unsigned char flags)
1078 {
1079 	cmd->cmd_len = 10;
1080 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1081 	cmd->cmnd[1] = flags;
1082 	cmd->cmnd[6] = 0;
1083 	cmd->cmnd[9] = 0;
1084 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1085 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1086 
1087 	return BLK_STS_OK;
1088 }
1089 
1090 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1091 				      sector_t lba, unsigned int nr_blocks,
1092 				      unsigned char flags)
1093 {
1094 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1095 	if (WARN_ON_ONCE(nr_blocks == 0))
1096 		return BLK_STS_IOERR;
1097 
1098 	if (unlikely(flags & 0x8)) {
1099 		/*
1100 		 * This happens only if this drive failed 10byte rw
1101 		 * command with ILLEGAL_REQUEST during operation and
1102 		 * thus turned off use_10_for_rw.
1103 		 */
1104 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1105 		return BLK_STS_IOERR;
1106 	}
1107 
1108 	cmd->cmd_len = 6;
1109 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1110 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1111 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1112 	cmd->cmnd[3] = lba & 0xff;
1113 	cmd->cmnd[4] = nr_blocks;
1114 	cmd->cmnd[5] = 0;
1115 
1116 	return BLK_STS_OK;
1117 }
1118 
1119 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1120 {
1121 	struct request *rq = scsi_cmd_to_rq(cmd);
1122 	struct scsi_device *sdp = cmd->device;
1123 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1124 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1125 	sector_t threshold;
1126 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1127 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1128 	bool write = rq_data_dir(rq) == WRITE;
1129 	unsigned char protect, fua;
1130 	blk_status_t ret;
1131 	unsigned int dif;
1132 	bool dix;
1133 
1134 	ret = scsi_alloc_sgtables(cmd);
1135 	if (ret != BLK_STS_OK)
1136 		return ret;
1137 
1138 	ret = BLK_STS_IOERR;
1139 	if (!scsi_device_online(sdp) || sdp->changed) {
1140 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1141 		goto fail;
1142 	}
1143 
1144 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1145 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1146 		goto fail;
1147 	}
1148 
1149 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1150 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1151 		goto fail;
1152 	}
1153 
1154 	/*
1155 	 * Some SD card readers can't handle accesses which touch the
1156 	 * last one or two logical blocks. Split accesses as needed.
1157 	 */
1158 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1159 
1160 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1161 		if (lba < threshold) {
1162 			/* Access up to the threshold but not beyond */
1163 			nr_blocks = threshold - lba;
1164 		} else {
1165 			/* Access only a single logical block */
1166 			nr_blocks = 1;
1167 		}
1168 	}
1169 
1170 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1171 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1172 		if (ret)
1173 			goto fail;
1174 	}
1175 
1176 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1177 	dix = scsi_prot_sg_count(cmd);
1178 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1179 
1180 	if (dif || dix)
1181 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1182 	else
1183 		protect = 0;
1184 
1185 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1186 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1187 					 protect | fua);
1188 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1189 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1190 					 protect | fua);
1191 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1192 		   sdp->use_10_for_rw || protect) {
1193 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1194 					 protect | fua);
1195 	} else {
1196 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1197 					protect | fua);
1198 	}
1199 
1200 	if (unlikely(ret != BLK_STS_OK))
1201 		goto fail;
1202 
1203 	/*
1204 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1205 	 * host adapter, it's safe to assume that we can at least transfer
1206 	 * this many bytes between each connect / disconnect.
1207 	 */
1208 	cmd->transfersize = sdp->sector_size;
1209 	cmd->underflow = nr_blocks << 9;
1210 	cmd->allowed = sdkp->max_retries;
1211 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1212 
1213 	SCSI_LOG_HLQUEUE(1,
1214 			 scmd_printk(KERN_INFO, cmd,
1215 				     "%s: block=%llu, count=%d\n", __func__,
1216 				     (unsigned long long)blk_rq_pos(rq),
1217 				     blk_rq_sectors(rq)));
1218 	SCSI_LOG_HLQUEUE(2,
1219 			 scmd_printk(KERN_INFO, cmd,
1220 				     "%s %d/%u 512 byte blocks.\n",
1221 				     write ? "writing" : "reading", nr_blocks,
1222 				     blk_rq_sectors(rq)));
1223 
1224 	/*
1225 	 * This indicates that the command is ready from our end to be queued.
1226 	 */
1227 	return BLK_STS_OK;
1228 fail:
1229 	scsi_free_sgtables(cmd);
1230 	return ret;
1231 }
1232 
1233 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1234 {
1235 	struct request *rq = scsi_cmd_to_rq(cmd);
1236 
1237 	switch (req_op(rq)) {
1238 	case REQ_OP_DISCARD:
1239 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1240 		case SD_LBP_UNMAP:
1241 			return sd_setup_unmap_cmnd(cmd);
1242 		case SD_LBP_WS16:
1243 			return sd_setup_write_same16_cmnd(cmd, true);
1244 		case SD_LBP_WS10:
1245 			return sd_setup_write_same10_cmnd(cmd, true);
1246 		case SD_LBP_ZERO:
1247 			return sd_setup_write_same10_cmnd(cmd, false);
1248 		default:
1249 			return BLK_STS_TARGET;
1250 		}
1251 	case REQ_OP_WRITE_ZEROES:
1252 		return sd_setup_write_zeroes_cmnd(cmd);
1253 	case REQ_OP_FLUSH:
1254 		return sd_setup_flush_cmnd(cmd);
1255 	case REQ_OP_READ:
1256 	case REQ_OP_WRITE:
1257 	case REQ_OP_ZONE_APPEND:
1258 		return sd_setup_read_write_cmnd(cmd);
1259 	case REQ_OP_ZONE_RESET:
1260 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1261 						   false);
1262 	case REQ_OP_ZONE_RESET_ALL:
1263 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1264 						   true);
1265 	case REQ_OP_ZONE_OPEN:
1266 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1267 	case REQ_OP_ZONE_CLOSE:
1268 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1269 	case REQ_OP_ZONE_FINISH:
1270 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1271 	default:
1272 		WARN_ON_ONCE(1);
1273 		return BLK_STS_NOTSUPP;
1274 	}
1275 }
1276 
1277 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1278 {
1279 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1280 
1281 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1282 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1283 }
1284 
1285 static bool sd_need_revalidate(struct block_device *bdev,
1286 		struct scsi_disk *sdkp)
1287 {
1288 	if (sdkp->device->removable || sdkp->write_prot) {
1289 		if (bdev_check_media_change(bdev))
1290 			return true;
1291 	}
1292 
1293 	/*
1294 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1295 	 * nothing to do with partitions, BLKRRPART is used to force a full
1296 	 * revalidate after things like a format for historical reasons.
1297 	 */
1298 	return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1299 }
1300 
1301 /**
1302  *	sd_open - open a scsi disk device
1303  *	@bdev: Block device of the scsi disk to open
1304  *	@mode: FMODE_* mask
1305  *
1306  *	Returns 0 if successful. Returns a negated errno value in case
1307  *	of error.
1308  *
1309  *	Note: This can be called from a user context (e.g. fsck(1) )
1310  *	or from within the kernel (e.g. as a result of a mount(1) ).
1311  *	In the latter case @inode and @filp carry an abridged amount
1312  *	of information as noted above.
1313  *
1314  *	Locking: called with bdev->bd_disk->open_mutex held.
1315  **/
1316 static int sd_open(struct block_device *bdev, fmode_t mode)
1317 {
1318 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1319 	struct scsi_device *sdev = sdkp->device;
1320 	int retval;
1321 
1322 	if (scsi_device_get(sdev))
1323 		return -ENXIO;
1324 
1325 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1326 
1327 	/*
1328 	 * If the device is in error recovery, wait until it is done.
1329 	 * If the device is offline, then disallow any access to it.
1330 	 */
1331 	retval = -ENXIO;
1332 	if (!scsi_block_when_processing_errors(sdev))
1333 		goto error_out;
1334 
1335 	if (sd_need_revalidate(bdev, sdkp))
1336 		sd_revalidate_disk(bdev->bd_disk);
1337 
1338 	/*
1339 	 * If the drive is empty, just let the open fail.
1340 	 */
1341 	retval = -ENOMEDIUM;
1342 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1343 		goto error_out;
1344 
1345 	/*
1346 	 * If the device has the write protect tab set, have the open fail
1347 	 * if the user expects to be able to write to the thing.
1348 	 */
1349 	retval = -EROFS;
1350 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1351 		goto error_out;
1352 
1353 	/*
1354 	 * It is possible that the disk changing stuff resulted in
1355 	 * the device being taken offline.  If this is the case,
1356 	 * report this to the user, and don't pretend that the
1357 	 * open actually succeeded.
1358 	 */
1359 	retval = -ENXIO;
1360 	if (!scsi_device_online(sdev))
1361 		goto error_out;
1362 
1363 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1364 		if (scsi_block_when_processing_errors(sdev))
1365 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1366 	}
1367 
1368 	return 0;
1369 
1370 error_out:
1371 	scsi_device_put(sdev);
1372 	return retval;
1373 }
1374 
1375 /**
1376  *	sd_release - invoked when the (last) close(2) is called on this
1377  *	scsi disk.
1378  *	@disk: disk to release
1379  *	@mode: FMODE_* mask
1380  *
1381  *	Returns 0.
1382  *
1383  *	Note: may block (uninterruptible) if error recovery is underway
1384  *	on this disk.
1385  *
1386  *	Locking: called with bdev->bd_disk->open_mutex held.
1387  **/
1388 static void sd_release(struct gendisk *disk, fmode_t mode)
1389 {
1390 	struct scsi_disk *sdkp = scsi_disk(disk);
1391 	struct scsi_device *sdev = sdkp->device;
1392 
1393 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1394 
1395 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1396 		if (scsi_block_when_processing_errors(sdev))
1397 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1398 	}
1399 
1400 	scsi_device_put(sdev);
1401 }
1402 
1403 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1404 {
1405 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1406 	struct scsi_device *sdp = sdkp->device;
1407 	struct Scsi_Host *host = sdp->host;
1408 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1409 	int diskinfo[4];
1410 
1411 	/* default to most commonly used values */
1412 	diskinfo[0] = 0x40;	/* 1 << 6 */
1413 	diskinfo[1] = 0x20;	/* 1 << 5 */
1414 	diskinfo[2] = capacity >> 11;
1415 
1416 	/* override with calculated, extended default, or driver values */
1417 	if (host->hostt->bios_param)
1418 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1419 	else
1420 		scsicam_bios_param(bdev, capacity, diskinfo);
1421 
1422 	geo->heads = diskinfo[0];
1423 	geo->sectors = diskinfo[1];
1424 	geo->cylinders = diskinfo[2];
1425 	return 0;
1426 }
1427 
1428 /**
1429  *	sd_ioctl - process an ioctl
1430  *	@bdev: target block device
1431  *	@mode: FMODE_* mask
1432  *	@cmd: ioctl command number
1433  *	@arg: this is third argument given to ioctl(2) system call.
1434  *	Often contains a pointer.
1435  *
1436  *	Returns 0 if successful (some ioctls return positive numbers on
1437  *	success as well). Returns a negated errno value in case of error.
1438  *
1439  *	Note: most ioctls are forward onto the block subsystem or further
1440  *	down in the scsi subsystem.
1441  **/
1442 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1443 		    unsigned int cmd, unsigned long arg)
1444 {
1445 	struct gendisk *disk = bdev->bd_disk;
1446 	struct scsi_disk *sdkp = scsi_disk(disk);
1447 	struct scsi_device *sdp = sdkp->device;
1448 	void __user *p = (void __user *)arg;
1449 	int error;
1450 
1451 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1452 				    "cmd=0x%x\n", disk->disk_name, cmd));
1453 
1454 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1455 		return -ENOIOCTLCMD;
1456 
1457 	/*
1458 	 * If we are in the middle of error recovery, don't let anyone
1459 	 * else try and use this device.  Also, if error recovery fails, it
1460 	 * may try and take the device offline, in which case all further
1461 	 * access to the device is prohibited.
1462 	 */
1463 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1464 			(mode & FMODE_NDELAY) != 0);
1465 	if (error)
1466 		return error;
1467 
1468 	if (is_sed_ioctl(cmd))
1469 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1470 	return scsi_ioctl(sdp, mode, cmd, p);
1471 }
1472 
1473 static void set_media_not_present(struct scsi_disk *sdkp)
1474 {
1475 	if (sdkp->media_present)
1476 		sdkp->device->changed = 1;
1477 
1478 	if (sdkp->device->removable) {
1479 		sdkp->media_present = 0;
1480 		sdkp->capacity = 0;
1481 	}
1482 }
1483 
1484 static int media_not_present(struct scsi_disk *sdkp,
1485 			     struct scsi_sense_hdr *sshdr)
1486 {
1487 	if (!scsi_sense_valid(sshdr))
1488 		return 0;
1489 
1490 	/* not invoked for commands that could return deferred errors */
1491 	switch (sshdr->sense_key) {
1492 	case UNIT_ATTENTION:
1493 	case NOT_READY:
1494 		/* medium not present */
1495 		if (sshdr->asc == 0x3A) {
1496 			set_media_not_present(sdkp);
1497 			return 1;
1498 		}
1499 	}
1500 	return 0;
1501 }
1502 
1503 /**
1504  *	sd_check_events - check media events
1505  *	@disk: kernel device descriptor
1506  *	@clearing: disk events currently being cleared
1507  *
1508  *	Returns mask of DISK_EVENT_*.
1509  *
1510  *	Note: this function is invoked from the block subsystem.
1511  **/
1512 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1513 {
1514 	struct scsi_disk *sdkp = disk->private_data;
1515 	struct scsi_device *sdp;
1516 	int retval;
1517 	bool disk_changed;
1518 
1519 	if (!sdkp)
1520 		return 0;
1521 
1522 	sdp = sdkp->device;
1523 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1524 
1525 	/*
1526 	 * If the device is offline, don't send any commands - just pretend as
1527 	 * if the command failed.  If the device ever comes back online, we
1528 	 * can deal with it then.  It is only because of unrecoverable errors
1529 	 * that we would ever take a device offline in the first place.
1530 	 */
1531 	if (!scsi_device_online(sdp)) {
1532 		set_media_not_present(sdkp);
1533 		goto out;
1534 	}
1535 
1536 	/*
1537 	 * Using TEST_UNIT_READY enables differentiation between drive with
1538 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1539 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1540 	 *
1541 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1542 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1543 	 * sd_revalidate() is called.
1544 	 */
1545 	if (scsi_block_when_processing_errors(sdp)) {
1546 		struct scsi_sense_hdr sshdr = { 0, };
1547 
1548 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1549 					      &sshdr);
1550 
1551 		/* failed to execute TUR, assume media not present */
1552 		if (retval < 0 || host_byte(retval)) {
1553 			set_media_not_present(sdkp);
1554 			goto out;
1555 		}
1556 
1557 		if (media_not_present(sdkp, &sshdr))
1558 			goto out;
1559 	}
1560 
1561 	/*
1562 	 * For removable scsi disk we have to recognise the presence
1563 	 * of a disk in the drive.
1564 	 */
1565 	if (!sdkp->media_present)
1566 		sdp->changed = 1;
1567 	sdkp->media_present = 1;
1568 out:
1569 	/*
1570 	 * sdp->changed is set under the following conditions:
1571 	 *
1572 	 *	Medium present state has changed in either direction.
1573 	 *	Device has indicated UNIT_ATTENTION.
1574 	 */
1575 	disk_changed = sdp->changed;
1576 	sdp->changed = 0;
1577 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1578 }
1579 
1580 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1581 {
1582 	int retries, res;
1583 	struct scsi_device *sdp = sdkp->device;
1584 	const int timeout = sdp->request_queue->rq_timeout
1585 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1586 	struct scsi_sense_hdr my_sshdr;
1587 	const struct scsi_exec_args exec_args = {
1588 		.req_flags = BLK_MQ_REQ_PM,
1589 		/* caller might not be interested in sense, but we need it */
1590 		.sshdr = sshdr ? : &my_sshdr,
1591 	};
1592 
1593 	if (!scsi_device_online(sdp))
1594 		return -ENODEV;
1595 
1596 	sshdr = exec_args.sshdr;
1597 
1598 	for (retries = 3; retries > 0; --retries) {
1599 		unsigned char cmd[16] = { 0 };
1600 
1601 		if (sdp->use_16_for_sync)
1602 			cmd[0] = SYNCHRONIZE_CACHE_16;
1603 		else
1604 			cmd[0] = SYNCHRONIZE_CACHE;
1605 		/*
1606 		 * Leave the rest of the command zero to indicate
1607 		 * flush everything.
1608 		 */
1609 		res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1610 				       timeout, sdkp->max_retries, &exec_args);
1611 		if (res == 0)
1612 			break;
1613 	}
1614 
1615 	if (res) {
1616 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1617 
1618 		if (res < 0)
1619 			return res;
1620 
1621 		if (scsi_status_is_check_condition(res) &&
1622 		    scsi_sense_valid(sshdr)) {
1623 			sd_print_sense_hdr(sdkp, sshdr);
1624 
1625 			/* we need to evaluate the error return  */
1626 			if (sshdr->asc == 0x3a ||	/* medium not present */
1627 			    sshdr->asc == 0x20 ||	/* invalid command */
1628 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1629 				/* this is no error here */
1630 				return 0;
1631 		}
1632 
1633 		switch (host_byte(res)) {
1634 		/* ignore errors due to racing a disconnection */
1635 		case DID_BAD_TARGET:
1636 		case DID_NO_CONNECT:
1637 			return 0;
1638 		/* signal the upper layer it might try again */
1639 		case DID_BUS_BUSY:
1640 		case DID_IMM_RETRY:
1641 		case DID_REQUEUE:
1642 		case DID_SOFT_ERROR:
1643 			return -EBUSY;
1644 		default:
1645 			return -EIO;
1646 		}
1647 	}
1648 	return 0;
1649 }
1650 
1651 static void sd_rescan(struct device *dev)
1652 {
1653 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1654 
1655 	sd_revalidate_disk(sdkp->disk);
1656 }
1657 
1658 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1659 		enum blk_unique_id type)
1660 {
1661 	struct scsi_device *sdev = scsi_disk(disk)->device;
1662 	const struct scsi_vpd *vpd;
1663 	const unsigned char *d;
1664 	int ret = -ENXIO, len;
1665 
1666 	rcu_read_lock();
1667 	vpd = rcu_dereference(sdev->vpd_pg83);
1668 	if (!vpd)
1669 		goto out_unlock;
1670 
1671 	ret = -EINVAL;
1672 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1673 		/* we only care about designators with LU association */
1674 		if (((d[1] >> 4) & 0x3) != 0x00)
1675 			continue;
1676 		if ((d[1] & 0xf) != type)
1677 			continue;
1678 
1679 		/*
1680 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1681 		 * keep looking as one with more entropy might still show up.
1682 		 */
1683 		len = d[3];
1684 		if (len != 8 && len != 12 && len != 16)
1685 			continue;
1686 		ret = len;
1687 		memcpy(id, d + 4, len);
1688 		if (len == 16)
1689 			break;
1690 	}
1691 out_unlock:
1692 	rcu_read_unlock();
1693 	return ret;
1694 }
1695 
1696 static char sd_pr_type(enum pr_type type)
1697 {
1698 	switch (type) {
1699 	case PR_WRITE_EXCLUSIVE:
1700 		return 0x01;
1701 	case PR_EXCLUSIVE_ACCESS:
1702 		return 0x03;
1703 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1704 		return 0x05;
1705 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1706 		return 0x06;
1707 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1708 		return 0x07;
1709 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1710 		return 0x08;
1711 	default:
1712 		return 0;
1713 	}
1714 };
1715 
1716 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1717 {
1718 	switch (host_byte(result)) {
1719 	case DID_TRANSPORT_MARGINAL:
1720 	case DID_TRANSPORT_DISRUPTED:
1721 	case DID_BUS_BUSY:
1722 		return PR_STS_RETRY_PATH_FAILURE;
1723 	case DID_NO_CONNECT:
1724 		return PR_STS_PATH_FAILED;
1725 	case DID_TRANSPORT_FAILFAST:
1726 		return PR_STS_PATH_FAST_FAILED;
1727 	}
1728 
1729 	switch (status_byte(result)) {
1730 	case SAM_STAT_RESERVATION_CONFLICT:
1731 		return PR_STS_RESERVATION_CONFLICT;
1732 	case SAM_STAT_CHECK_CONDITION:
1733 		if (!scsi_sense_valid(sshdr))
1734 			return PR_STS_IOERR;
1735 
1736 		if (sshdr->sense_key == ILLEGAL_REQUEST &&
1737 		    (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1738 			return -EINVAL;
1739 
1740 		fallthrough;
1741 	default:
1742 		return PR_STS_IOERR;
1743 	}
1744 }
1745 
1746 static int sd_pr_command(struct block_device *bdev, u8 sa,
1747 		u64 key, u64 sa_key, u8 type, u8 flags)
1748 {
1749 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1750 	struct scsi_device *sdev = sdkp->device;
1751 	struct scsi_sense_hdr sshdr;
1752 	const struct scsi_exec_args exec_args = {
1753 		.sshdr = &sshdr,
1754 	};
1755 	int result;
1756 	u8 cmd[16] = { 0, };
1757 	u8 data[24] = { 0, };
1758 
1759 	cmd[0] = PERSISTENT_RESERVE_OUT;
1760 	cmd[1] = sa;
1761 	cmd[2] = type;
1762 	put_unaligned_be32(sizeof(data), &cmd[5]);
1763 
1764 	put_unaligned_be64(key, &data[0]);
1765 	put_unaligned_be64(sa_key, &data[8]);
1766 	data[20] = flags;
1767 
1768 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1769 				  sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1770 				  &exec_args);
1771 
1772 	if (scsi_status_is_check_condition(result) &&
1773 	    scsi_sense_valid(&sshdr)) {
1774 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1775 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1776 	}
1777 
1778 	if (result <= 0)
1779 		return result;
1780 
1781 	return sd_scsi_to_pr_err(&sshdr, result);
1782 }
1783 
1784 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1785 		u32 flags)
1786 {
1787 	if (flags & ~PR_FL_IGNORE_KEY)
1788 		return -EOPNOTSUPP;
1789 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1790 			old_key, new_key, 0,
1791 			(1 << 0) /* APTPL */);
1792 }
1793 
1794 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1795 		u32 flags)
1796 {
1797 	if (flags)
1798 		return -EOPNOTSUPP;
1799 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1800 }
1801 
1802 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1803 {
1804 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1805 }
1806 
1807 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1808 		enum pr_type type, bool abort)
1809 {
1810 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1811 			     sd_pr_type(type), 0);
1812 }
1813 
1814 static int sd_pr_clear(struct block_device *bdev, u64 key)
1815 {
1816 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1817 }
1818 
1819 static const struct pr_ops sd_pr_ops = {
1820 	.pr_register	= sd_pr_register,
1821 	.pr_reserve	= sd_pr_reserve,
1822 	.pr_release	= sd_pr_release,
1823 	.pr_preempt	= sd_pr_preempt,
1824 	.pr_clear	= sd_pr_clear,
1825 };
1826 
1827 static void scsi_disk_free_disk(struct gendisk *disk)
1828 {
1829 	struct scsi_disk *sdkp = scsi_disk(disk);
1830 
1831 	put_device(&sdkp->disk_dev);
1832 }
1833 
1834 static const struct block_device_operations sd_fops = {
1835 	.owner			= THIS_MODULE,
1836 	.open			= sd_open,
1837 	.release		= sd_release,
1838 	.ioctl			= sd_ioctl,
1839 	.getgeo			= sd_getgeo,
1840 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1841 	.check_events		= sd_check_events,
1842 	.unlock_native_capacity	= sd_unlock_native_capacity,
1843 	.report_zones		= sd_zbc_report_zones,
1844 	.get_unique_id		= sd_get_unique_id,
1845 	.free_disk		= scsi_disk_free_disk,
1846 	.pr_ops			= &sd_pr_ops,
1847 };
1848 
1849 /**
1850  *	sd_eh_reset - reset error handling callback
1851  *	@scmd:		sd-issued command that has failed
1852  *
1853  *	This function is called by the SCSI midlayer before starting
1854  *	SCSI EH. When counting medium access failures we have to be
1855  *	careful to register it only only once per device and SCSI EH run;
1856  *	there might be several timed out commands which will cause the
1857  *	'max_medium_access_timeouts' counter to trigger after the first
1858  *	SCSI EH run already and set the device to offline.
1859  *	So this function resets the internal counter before starting SCSI EH.
1860  **/
1861 static void sd_eh_reset(struct scsi_cmnd *scmd)
1862 {
1863 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1864 
1865 	/* New SCSI EH run, reset gate variable */
1866 	sdkp->ignore_medium_access_errors = false;
1867 }
1868 
1869 /**
1870  *	sd_eh_action - error handling callback
1871  *	@scmd:		sd-issued command that has failed
1872  *	@eh_disp:	The recovery disposition suggested by the midlayer
1873  *
1874  *	This function is called by the SCSI midlayer upon completion of an
1875  *	error test command (currently TEST UNIT READY). The result of sending
1876  *	the eh command is passed in eh_disp.  We're looking for devices that
1877  *	fail medium access commands but are OK with non access commands like
1878  *	test unit ready (so wrongly see the device as having a successful
1879  *	recovery)
1880  **/
1881 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1882 {
1883 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1884 	struct scsi_device *sdev = scmd->device;
1885 
1886 	if (!scsi_device_online(sdev) ||
1887 	    !scsi_medium_access_command(scmd) ||
1888 	    host_byte(scmd->result) != DID_TIME_OUT ||
1889 	    eh_disp != SUCCESS)
1890 		return eh_disp;
1891 
1892 	/*
1893 	 * The device has timed out executing a medium access command.
1894 	 * However, the TEST UNIT READY command sent during error
1895 	 * handling completed successfully. Either the device is in the
1896 	 * process of recovering or has it suffered an internal failure
1897 	 * that prevents access to the storage medium.
1898 	 */
1899 	if (!sdkp->ignore_medium_access_errors) {
1900 		sdkp->medium_access_timed_out++;
1901 		sdkp->ignore_medium_access_errors = true;
1902 	}
1903 
1904 	/*
1905 	 * If the device keeps failing read/write commands but TEST UNIT
1906 	 * READY always completes successfully we assume that medium
1907 	 * access is no longer possible and take the device offline.
1908 	 */
1909 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1910 		scmd_printk(KERN_ERR, scmd,
1911 			    "Medium access timeout failure. Offlining disk!\n");
1912 		mutex_lock(&sdev->state_mutex);
1913 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1914 		mutex_unlock(&sdev->state_mutex);
1915 
1916 		return SUCCESS;
1917 	}
1918 
1919 	return eh_disp;
1920 }
1921 
1922 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1923 {
1924 	struct request *req = scsi_cmd_to_rq(scmd);
1925 	struct scsi_device *sdev = scmd->device;
1926 	unsigned int transferred, good_bytes;
1927 	u64 start_lba, end_lba, bad_lba;
1928 
1929 	/*
1930 	 * Some commands have a payload smaller than the device logical
1931 	 * block size (e.g. INQUIRY on a 4K disk).
1932 	 */
1933 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1934 		return 0;
1935 
1936 	/* Check if we have a 'bad_lba' information */
1937 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1938 				     SCSI_SENSE_BUFFERSIZE,
1939 				     &bad_lba))
1940 		return 0;
1941 
1942 	/*
1943 	 * If the bad lba was reported incorrectly, we have no idea where
1944 	 * the error is.
1945 	 */
1946 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1947 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1948 	if (bad_lba < start_lba || bad_lba >= end_lba)
1949 		return 0;
1950 
1951 	/*
1952 	 * resid is optional but mostly filled in.  When it's unused,
1953 	 * its value is zero, so we assume the whole buffer transferred
1954 	 */
1955 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1956 
1957 	/* This computation should always be done in terms of the
1958 	 * resolution of the device's medium.
1959 	 */
1960 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1961 
1962 	return min(good_bytes, transferred);
1963 }
1964 
1965 /**
1966  *	sd_done - bottom half handler: called when the lower level
1967  *	driver has completed (successfully or otherwise) a scsi command.
1968  *	@SCpnt: mid-level's per command structure.
1969  *
1970  *	Note: potentially run from within an ISR. Must not block.
1971  **/
1972 static int sd_done(struct scsi_cmnd *SCpnt)
1973 {
1974 	int result = SCpnt->result;
1975 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1976 	unsigned int sector_size = SCpnt->device->sector_size;
1977 	unsigned int resid;
1978 	struct scsi_sense_hdr sshdr;
1979 	struct request *req = scsi_cmd_to_rq(SCpnt);
1980 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
1981 	int sense_valid = 0;
1982 	int sense_deferred = 0;
1983 
1984 	switch (req_op(req)) {
1985 	case REQ_OP_DISCARD:
1986 	case REQ_OP_WRITE_ZEROES:
1987 	case REQ_OP_ZONE_RESET:
1988 	case REQ_OP_ZONE_RESET_ALL:
1989 	case REQ_OP_ZONE_OPEN:
1990 	case REQ_OP_ZONE_CLOSE:
1991 	case REQ_OP_ZONE_FINISH:
1992 		if (!result) {
1993 			good_bytes = blk_rq_bytes(req);
1994 			scsi_set_resid(SCpnt, 0);
1995 		} else {
1996 			good_bytes = 0;
1997 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1998 		}
1999 		break;
2000 	default:
2001 		/*
2002 		 * In case of bogus fw or device, we could end up having
2003 		 * an unaligned partial completion. Check this here and force
2004 		 * alignment.
2005 		 */
2006 		resid = scsi_get_resid(SCpnt);
2007 		if (resid & (sector_size - 1)) {
2008 			sd_printk(KERN_INFO, sdkp,
2009 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2010 				resid, sector_size);
2011 			scsi_print_command(SCpnt);
2012 			resid = min(scsi_bufflen(SCpnt),
2013 				    round_up(resid, sector_size));
2014 			scsi_set_resid(SCpnt, resid);
2015 		}
2016 	}
2017 
2018 	if (result) {
2019 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2020 		if (sense_valid)
2021 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2022 	}
2023 	sdkp->medium_access_timed_out = 0;
2024 
2025 	if (!scsi_status_is_check_condition(result) &&
2026 	    (!sense_valid || sense_deferred))
2027 		goto out;
2028 
2029 	switch (sshdr.sense_key) {
2030 	case HARDWARE_ERROR:
2031 	case MEDIUM_ERROR:
2032 		good_bytes = sd_completed_bytes(SCpnt);
2033 		break;
2034 	case RECOVERED_ERROR:
2035 		good_bytes = scsi_bufflen(SCpnt);
2036 		break;
2037 	case NO_SENSE:
2038 		/* This indicates a false check condition, so ignore it.  An
2039 		 * unknown amount of data was transferred so treat it as an
2040 		 * error.
2041 		 */
2042 		SCpnt->result = 0;
2043 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2044 		break;
2045 	case ABORTED_COMMAND:
2046 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2047 			good_bytes = sd_completed_bytes(SCpnt);
2048 		break;
2049 	case ILLEGAL_REQUEST:
2050 		switch (sshdr.asc) {
2051 		case 0x10:	/* DIX: Host detected corruption */
2052 			good_bytes = sd_completed_bytes(SCpnt);
2053 			break;
2054 		case 0x20:	/* INVALID COMMAND OPCODE */
2055 		case 0x24:	/* INVALID FIELD IN CDB */
2056 			switch (SCpnt->cmnd[0]) {
2057 			case UNMAP:
2058 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2059 				break;
2060 			case WRITE_SAME_16:
2061 			case WRITE_SAME:
2062 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2063 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2064 				} else {
2065 					sdkp->device->no_write_same = 1;
2066 					sd_config_write_same(sdkp);
2067 					req->rq_flags |= RQF_QUIET;
2068 				}
2069 				break;
2070 			}
2071 		}
2072 		break;
2073 	default:
2074 		break;
2075 	}
2076 
2077  out:
2078 	if (sd_is_zoned(sdkp))
2079 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2080 
2081 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2082 					   "sd_done: completed %d of %d bytes\n",
2083 					   good_bytes, scsi_bufflen(SCpnt)));
2084 
2085 	return good_bytes;
2086 }
2087 
2088 /*
2089  * spinup disk - called only in sd_revalidate_disk()
2090  */
2091 static void
2092 sd_spinup_disk(struct scsi_disk *sdkp)
2093 {
2094 	unsigned char cmd[10];
2095 	unsigned long spintime_expire = 0;
2096 	int retries, spintime;
2097 	unsigned int the_result;
2098 	struct scsi_sense_hdr sshdr;
2099 	const struct scsi_exec_args exec_args = {
2100 		.sshdr = &sshdr,
2101 	};
2102 	int sense_valid = 0;
2103 
2104 	spintime = 0;
2105 
2106 	/* Spin up drives, as required.  Only do this at boot time */
2107 	/* Spinup needs to be done for module loads too. */
2108 	do {
2109 		retries = 0;
2110 
2111 		do {
2112 			bool media_was_present = sdkp->media_present;
2113 
2114 			cmd[0] = TEST_UNIT_READY;
2115 			memset((void *) &cmd[1], 0, 9);
2116 
2117 			the_result = scsi_execute_cmd(sdkp->device, cmd,
2118 						      REQ_OP_DRV_IN, NULL, 0,
2119 						      SD_TIMEOUT,
2120 						      sdkp->max_retries,
2121 						      &exec_args);
2122 
2123 			/*
2124 			 * If the drive has indicated to us that it
2125 			 * doesn't have any media in it, don't bother
2126 			 * with any more polling.
2127 			 */
2128 			if (media_not_present(sdkp, &sshdr)) {
2129 				if (media_was_present)
2130 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2131 				return;
2132 			}
2133 
2134 			if (the_result)
2135 				sense_valid = scsi_sense_valid(&sshdr);
2136 			retries++;
2137 		} while (retries < 3 &&
2138 			 (!scsi_status_is_good(the_result) ||
2139 			  (scsi_status_is_check_condition(the_result) &&
2140 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2141 
2142 		if (!scsi_status_is_check_condition(the_result)) {
2143 			/* no sense, TUR either succeeded or failed
2144 			 * with a status error */
2145 			if(!spintime && !scsi_status_is_good(the_result)) {
2146 				sd_print_result(sdkp, "Test Unit Ready failed",
2147 						the_result);
2148 			}
2149 			break;
2150 		}
2151 
2152 		/*
2153 		 * The device does not want the automatic start to be issued.
2154 		 */
2155 		if (sdkp->device->no_start_on_add)
2156 			break;
2157 
2158 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2159 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2160 				break;	/* manual intervention required */
2161 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2162 				break;	/* standby */
2163 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2164 				break;	/* unavailable */
2165 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2166 				break;	/* sanitize in progress */
2167 			/*
2168 			 * Issue command to spin up drive when not ready
2169 			 */
2170 			if (!spintime) {
2171 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2172 				cmd[0] = START_STOP;
2173 				cmd[1] = 1;	/* Return immediately */
2174 				memset((void *) &cmd[2], 0, 8);
2175 				cmd[4] = 1;	/* Start spin cycle */
2176 				if (sdkp->device->start_stop_pwr_cond)
2177 					cmd[4] |= 1 << 4;
2178 				scsi_execute_cmd(sdkp->device, cmd,
2179 						 REQ_OP_DRV_IN, NULL, 0,
2180 						 SD_TIMEOUT, sdkp->max_retries,
2181 						 &exec_args);
2182 				spintime_expire = jiffies + 100 * HZ;
2183 				spintime = 1;
2184 			}
2185 			/* Wait 1 second for next try */
2186 			msleep(1000);
2187 			printk(KERN_CONT ".");
2188 
2189 		/*
2190 		 * Wait for USB flash devices with slow firmware.
2191 		 * Yes, this sense key/ASC combination shouldn't
2192 		 * occur here.  It's characteristic of these devices.
2193 		 */
2194 		} else if (sense_valid &&
2195 				sshdr.sense_key == UNIT_ATTENTION &&
2196 				sshdr.asc == 0x28) {
2197 			if (!spintime) {
2198 				spintime_expire = jiffies + 5 * HZ;
2199 				spintime = 1;
2200 			}
2201 			/* Wait 1 second for next try */
2202 			msleep(1000);
2203 		} else {
2204 			/* we don't understand the sense code, so it's
2205 			 * probably pointless to loop */
2206 			if(!spintime) {
2207 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2208 				sd_print_sense_hdr(sdkp, &sshdr);
2209 			}
2210 			break;
2211 		}
2212 
2213 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2214 
2215 	if (spintime) {
2216 		if (scsi_status_is_good(the_result))
2217 			printk(KERN_CONT "ready\n");
2218 		else
2219 			printk(KERN_CONT "not responding...\n");
2220 	}
2221 }
2222 
2223 /*
2224  * Determine whether disk supports Data Integrity Field.
2225  */
2226 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2227 {
2228 	struct scsi_device *sdp = sdkp->device;
2229 	u8 type;
2230 
2231 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2232 		sdkp->protection_type = 0;
2233 		return 0;
2234 	}
2235 
2236 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2237 
2238 	if (type > T10_PI_TYPE3_PROTECTION) {
2239 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported"	\
2240 			  " protection type %u. Disabling disk!\n",
2241 			  type);
2242 		sdkp->protection_type = 0;
2243 		return -ENODEV;
2244 	}
2245 
2246 	sdkp->protection_type = type;
2247 
2248 	return 0;
2249 }
2250 
2251 static void sd_config_protection(struct scsi_disk *sdkp)
2252 {
2253 	struct scsi_device *sdp = sdkp->device;
2254 
2255 	if (!sdkp->first_scan)
2256 		return;
2257 
2258 	sd_dif_config_host(sdkp);
2259 
2260 	if (!sdkp->protection_type)
2261 		return;
2262 
2263 	if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2264 		sd_printk(KERN_NOTICE, sdkp,
2265 			  "Disabling DIF Type %u protection\n",
2266 			  sdkp->protection_type);
2267 		sdkp->protection_type = 0;
2268 	}
2269 
2270 	sd_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2271 		  sdkp->protection_type);
2272 }
2273 
2274 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2275 			struct scsi_sense_hdr *sshdr, int sense_valid,
2276 			int the_result)
2277 {
2278 	if (sense_valid)
2279 		sd_print_sense_hdr(sdkp, sshdr);
2280 	else
2281 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2282 
2283 	/*
2284 	 * Set dirty bit for removable devices if not ready -
2285 	 * sometimes drives will not report this properly.
2286 	 */
2287 	if (sdp->removable &&
2288 	    sense_valid && sshdr->sense_key == NOT_READY)
2289 		set_media_not_present(sdkp);
2290 
2291 	/*
2292 	 * We used to set media_present to 0 here to indicate no media
2293 	 * in the drive, but some drives fail read capacity even with
2294 	 * media present, so we can't do that.
2295 	 */
2296 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2297 }
2298 
2299 #define RC16_LEN 32
2300 #if RC16_LEN > SD_BUF_SIZE
2301 #error RC16_LEN must not be more than SD_BUF_SIZE
2302 #endif
2303 
2304 #define READ_CAPACITY_RETRIES_ON_RESET	10
2305 
2306 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2307 						unsigned char *buffer)
2308 {
2309 	unsigned char cmd[16];
2310 	struct scsi_sense_hdr sshdr;
2311 	const struct scsi_exec_args exec_args = {
2312 		.sshdr = &sshdr,
2313 	};
2314 	int sense_valid = 0;
2315 	int the_result;
2316 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2317 	unsigned int alignment;
2318 	unsigned long long lba;
2319 	unsigned sector_size;
2320 
2321 	if (sdp->no_read_capacity_16)
2322 		return -EINVAL;
2323 
2324 	do {
2325 		memset(cmd, 0, 16);
2326 		cmd[0] = SERVICE_ACTION_IN_16;
2327 		cmd[1] = SAI_READ_CAPACITY_16;
2328 		cmd[13] = RC16_LEN;
2329 		memset(buffer, 0, RC16_LEN);
2330 
2331 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2332 					      buffer, RC16_LEN, SD_TIMEOUT,
2333 					      sdkp->max_retries, &exec_args);
2334 
2335 		if (media_not_present(sdkp, &sshdr))
2336 			return -ENODEV;
2337 
2338 		if (the_result > 0) {
2339 			sense_valid = scsi_sense_valid(&sshdr);
2340 			if (sense_valid &&
2341 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2342 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2343 			    sshdr.ascq == 0x00)
2344 				/* Invalid Command Operation Code or
2345 				 * Invalid Field in CDB, just retry
2346 				 * silently with RC10 */
2347 				return -EINVAL;
2348 			if (sense_valid &&
2349 			    sshdr.sense_key == UNIT_ATTENTION &&
2350 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2351 				/* Device reset might occur several times,
2352 				 * give it one more chance */
2353 				if (--reset_retries > 0)
2354 					continue;
2355 		}
2356 		retries--;
2357 
2358 	} while (the_result && retries);
2359 
2360 	if (the_result) {
2361 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2362 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2363 		return -EINVAL;
2364 	}
2365 
2366 	sector_size = get_unaligned_be32(&buffer[8]);
2367 	lba = get_unaligned_be64(&buffer[0]);
2368 
2369 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2370 		sdkp->capacity = 0;
2371 		return -ENODEV;
2372 	}
2373 
2374 	/* Logical blocks per physical block exponent */
2375 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2376 
2377 	/* RC basis */
2378 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2379 
2380 	/* Lowest aligned logical block */
2381 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2382 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2383 	if (alignment && sdkp->first_scan)
2384 		sd_printk(KERN_NOTICE, sdkp,
2385 			  "physical block alignment offset: %u\n", alignment);
2386 
2387 	if (buffer[14] & 0x80) { /* LBPME */
2388 		sdkp->lbpme = 1;
2389 
2390 		if (buffer[14] & 0x40) /* LBPRZ */
2391 			sdkp->lbprz = 1;
2392 
2393 		sd_config_discard(sdkp, SD_LBP_WS16);
2394 	}
2395 
2396 	sdkp->capacity = lba + 1;
2397 	return sector_size;
2398 }
2399 
2400 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2401 						unsigned char *buffer)
2402 {
2403 	unsigned char cmd[16];
2404 	struct scsi_sense_hdr sshdr;
2405 	const struct scsi_exec_args exec_args = {
2406 		.sshdr = &sshdr,
2407 	};
2408 	int sense_valid = 0;
2409 	int the_result;
2410 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2411 	sector_t lba;
2412 	unsigned sector_size;
2413 
2414 	do {
2415 		cmd[0] = READ_CAPACITY;
2416 		memset(&cmd[1], 0, 9);
2417 		memset(buffer, 0, 8);
2418 
2419 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2420 					      8, SD_TIMEOUT, sdkp->max_retries,
2421 					      &exec_args);
2422 
2423 		if (media_not_present(sdkp, &sshdr))
2424 			return -ENODEV;
2425 
2426 		if (the_result > 0) {
2427 			sense_valid = scsi_sense_valid(&sshdr);
2428 			if (sense_valid &&
2429 			    sshdr.sense_key == UNIT_ATTENTION &&
2430 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2431 				/* Device reset might occur several times,
2432 				 * give it one more chance */
2433 				if (--reset_retries > 0)
2434 					continue;
2435 		}
2436 		retries--;
2437 
2438 	} while (the_result && retries);
2439 
2440 	if (the_result) {
2441 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2442 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2443 		return -EINVAL;
2444 	}
2445 
2446 	sector_size = get_unaligned_be32(&buffer[4]);
2447 	lba = get_unaligned_be32(&buffer[0]);
2448 
2449 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2450 		/* Some buggy (usb cardreader) devices return an lba of
2451 		   0xffffffff when the want to report a size of 0 (with
2452 		   which they really mean no media is present) */
2453 		sdkp->capacity = 0;
2454 		sdkp->physical_block_size = sector_size;
2455 		return sector_size;
2456 	}
2457 
2458 	sdkp->capacity = lba + 1;
2459 	sdkp->physical_block_size = sector_size;
2460 	return sector_size;
2461 }
2462 
2463 static int sd_try_rc16_first(struct scsi_device *sdp)
2464 {
2465 	if (sdp->host->max_cmd_len < 16)
2466 		return 0;
2467 	if (sdp->try_rc_10_first)
2468 		return 0;
2469 	if (sdp->scsi_level > SCSI_SPC_2)
2470 		return 1;
2471 	if (scsi_device_protection(sdp))
2472 		return 1;
2473 	return 0;
2474 }
2475 
2476 /*
2477  * read disk capacity
2478  */
2479 static void
2480 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2481 {
2482 	int sector_size;
2483 	struct scsi_device *sdp = sdkp->device;
2484 
2485 	if (sd_try_rc16_first(sdp)) {
2486 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2487 		if (sector_size == -EOVERFLOW)
2488 			goto got_data;
2489 		if (sector_size == -ENODEV)
2490 			return;
2491 		if (sector_size < 0)
2492 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2493 		if (sector_size < 0)
2494 			return;
2495 	} else {
2496 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2497 		if (sector_size == -EOVERFLOW)
2498 			goto got_data;
2499 		if (sector_size < 0)
2500 			return;
2501 		if ((sizeof(sdkp->capacity) > 4) &&
2502 		    (sdkp->capacity > 0xffffffffULL)) {
2503 			int old_sector_size = sector_size;
2504 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2505 					"Trying to use READ CAPACITY(16).\n");
2506 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2507 			if (sector_size < 0) {
2508 				sd_printk(KERN_NOTICE, sdkp,
2509 					"Using 0xffffffff as device size\n");
2510 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2511 				sector_size = old_sector_size;
2512 				goto got_data;
2513 			}
2514 			/* Remember that READ CAPACITY(16) succeeded */
2515 			sdp->try_rc_10_first = 0;
2516 		}
2517 	}
2518 
2519 	/* Some devices are known to return the total number of blocks,
2520 	 * not the highest block number.  Some devices have versions
2521 	 * which do this and others which do not.  Some devices we might
2522 	 * suspect of doing this but we don't know for certain.
2523 	 *
2524 	 * If we know the reported capacity is wrong, decrement it.  If
2525 	 * we can only guess, then assume the number of blocks is even
2526 	 * (usually true but not always) and err on the side of lowering
2527 	 * the capacity.
2528 	 */
2529 	if (sdp->fix_capacity ||
2530 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2531 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2532 				"from its reported value: %llu\n",
2533 				(unsigned long long) sdkp->capacity);
2534 		--sdkp->capacity;
2535 	}
2536 
2537 got_data:
2538 	if (sector_size == 0) {
2539 		sector_size = 512;
2540 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2541 			  "assuming 512.\n");
2542 	}
2543 
2544 	if (sector_size != 512 &&
2545 	    sector_size != 1024 &&
2546 	    sector_size != 2048 &&
2547 	    sector_size != 4096) {
2548 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2549 			  sector_size);
2550 		/*
2551 		 * The user might want to re-format the drive with
2552 		 * a supported sectorsize.  Once this happens, it
2553 		 * would be relatively trivial to set the thing up.
2554 		 * For this reason, we leave the thing in the table.
2555 		 */
2556 		sdkp->capacity = 0;
2557 		/*
2558 		 * set a bogus sector size so the normal read/write
2559 		 * logic in the block layer will eventually refuse any
2560 		 * request on this device without tripping over power
2561 		 * of two sector size assumptions
2562 		 */
2563 		sector_size = 512;
2564 	}
2565 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2566 	blk_queue_physical_block_size(sdp->request_queue,
2567 				      sdkp->physical_block_size);
2568 	sdkp->device->sector_size = sector_size;
2569 
2570 	if (sdkp->capacity > 0xffffffff)
2571 		sdp->use_16_for_rw = 1;
2572 
2573 }
2574 
2575 /*
2576  * Print disk capacity
2577  */
2578 static void
2579 sd_print_capacity(struct scsi_disk *sdkp,
2580 		  sector_t old_capacity)
2581 {
2582 	int sector_size = sdkp->device->sector_size;
2583 	char cap_str_2[10], cap_str_10[10];
2584 
2585 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2586 		return;
2587 
2588 	string_get_size(sdkp->capacity, sector_size,
2589 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2590 	string_get_size(sdkp->capacity, sector_size,
2591 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2592 
2593 	sd_printk(KERN_NOTICE, sdkp,
2594 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2595 		  (unsigned long long)sdkp->capacity,
2596 		  sector_size, cap_str_10, cap_str_2);
2597 
2598 	if (sdkp->physical_block_size != sector_size)
2599 		sd_printk(KERN_NOTICE, sdkp,
2600 			  "%u-byte physical blocks\n",
2601 			  sdkp->physical_block_size);
2602 }
2603 
2604 /* called with buffer of length 512 */
2605 static inline int
2606 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2607 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2608 		 struct scsi_sense_hdr *sshdr)
2609 {
2610 	/*
2611 	 * If we must use MODE SENSE(10), make sure that the buffer length
2612 	 * is at least 8 bytes so that the mode sense header fits.
2613 	 */
2614 	if (sdkp->device->use_10_for_ms && len < 8)
2615 		len = 8;
2616 
2617 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2618 			       SD_TIMEOUT, sdkp->max_retries, data,
2619 			       sshdr);
2620 }
2621 
2622 /*
2623  * read write protect setting, if possible - called only in sd_revalidate_disk()
2624  * called with buffer of length SD_BUF_SIZE
2625  */
2626 static void
2627 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2628 {
2629 	int res;
2630 	struct scsi_device *sdp = sdkp->device;
2631 	struct scsi_mode_data data;
2632 	int old_wp = sdkp->write_prot;
2633 
2634 	set_disk_ro(sdkp->disk, 0);
2635 	if (sdp->skip_ms_page_3f) {
2636 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2637 		return;
2638 	}
2639 
2640 	if (sdp->use_192_bytes_for_3f) {
2641 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2642 	} else {
2643 		/*
2644 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2645 		 * We have to start carefully: some devices hang if we ask
2646 		 * for more than is available.
2647 		 */
2648 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2649 
2650 		/*
2651 		 * Second attempt: ask for page 0 When only page 0 is
2652 		 * implemented, a request for page 3F may return Sense Key
2653 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2654 		 * CDB.
2655 		 */
2656 		if (res < 0)
2657 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2658 
2659 		/*
2660 		 * Third attempt: ask 255 bytes, as we did earlier.
2661 		 */
2662 		if (res < 0)
2663 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2664 					       &data, NULL);
2665 	}
2666 
2667 	if (res < 0) {
2668 		sd_first_printk(KERN_WARNING, sdkp,
2669 			  "Test WP failed, assume Write Enabled\n");
2670 	} else {
2671 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2672 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2673 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2674 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2675 				  sdkp->write_prot ? "on" : "off");
2676 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2677 		}
2678 	}
2679 }
2680 
2681 /*
2682  * sd_read_cache_type - called only from sd_revalidate_disk()
2683  * called with buffer of length SD_BUF_SIZE
2684  */
2685 static void
2686 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2687 {
2688 	int len = 0, res;
2689 	struct scsi_device *sdp = sdkp->device;
2690 
2691 	int dbd;
2692 	int modepage;
2693 	int first_len;
2694 	struct scsi_mode_data data;
2695 	struct scsi_sense_hdr sshdr;
2696 	int old_wce = sdkp->WCE;
2697 	int old_rcd = sdkp->RCD;
2698 	int old_dpofua = sdkp->DPOFUA;
2699 
2700 
2701 	if (sdkp->cache_override)
2702 		return;
2703 
2704 	first_len = 4;
2705 	if (sdp->skip_ms_page_8) {
2706 		if (sdp->type == TYPE_RBC)
2707 			goto defaults;
2708 		else {
2709 			if (sdp->skip_ms_page_3f)
2710 				goto defaults;
2711 			modepage = 0x3F;
2712 			if (sdp->use_192_bytes_for_3f)
2713 				first_len = 192;
2714 			dbd = 0;
2715 		}
2716 	} else if (sdp->type == TYPE_RBC) {
2717 		modepage = 6;
2718 		dbd = 8;
2719 	} else {
2720 		modepage = 8;
2721 		dbd = 0;
2722 	}
2723 
2724 	/* cautiously ask */
2725 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2726 			&data, &sshdr);
2727 
2728 	if (res < 0)
2729 		goto bad_sense;
2730 
2731 	if (!data.header_length) {
2732 		modepage = 6;
2733 		first_len = 0;
2734 		sd_first_printk(KERN_ERR, sdkp,
2735 				"Missing header in MODE_SENSE response\n");
2736 	}
2737 
2738 	/* that went OK, now ask for the proper length */
2739 	len = data.length;
2740 
2741 	/*
2742 	 * We're only interested in the first three bytes, actually.
2743 	 * But the data cache page is defined for the first 20.
2744 	 */
2745 	if (len < 3)
2746 		goto bad_sense;
2747 	else if (len > SD_BUF_SIZE) {
2748 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2749 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2750 		len = SD_BUF_SIZE;
2751 	}
2752 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2753 		len = 192;
2754 
2755 	/* Get the data */
2756 	if (len > first_len)
2757 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2758 				&data, &sshdr);
2759 
2760 	if (!res) {
2761 		int offset = data.header_length + data.block_descriptor_length;
2762 
2763 		while (offset < len) {
2764 			u8 page_code = buffer[offset] & 0x3F;
2765 			u8 spf       = buffer[offset] & 0x40;
2766 
2767 			if (page_code == 8 || page_code == 6) {
2768 				/* We're interested only in the first 3 bytes.
2769 				 */
2770 				if (len - offset <= 2) {
2771 					sd_first_printk(KERN_ERR, sdkp,
2772 						"Incomplete mode parameter "
2773 							"data\n");
2774 					goto defaults;
2775 				} else {
2776 					modepage = page_code;
2777 					goto Page_found;
2778 				}
2779 			} else {
2780 				/* Go to the next page */
2781 				if (spf && len - offset > 3)
2782 					offset += 4 + (buffer[offset+2] << 8) +
2783 						buffer[offset+3];
2784 				else if (!spf && len - offset > 1)
2785 					offset += 2 + buffer[offset+1];
2786 				else {
2787 					sd_first_printk(KERN_ERR, sdkp,
2788 							"Incomplete mode "
2789 							"parameter data\n");
2790 					goto defaults;
2791 				}
2792 			}
2793 		}
2794 
2795 		sd_first_printk(KERN_WARNING, sdkp,
2796 				"No Caching mode page found\n");
2797 		goto defaults;
2798 
2799 	Page_found:
2800 		if (modepage == 8) {
2801 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2802 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2803 		} else {
2804 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2805 			sdkp->RCD = 0;
2806 		}
2807 
2808 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2809 		if (sdp->broken_fua) {
2810 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2811 			sdkp->DPOFUA = 0;
2812 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2813 			   !sdkp->device->use_16_for_rw) {
2814 			sd_first_printk(KERN_NOTICE, sdkp,
2815 				  "Uses READ/WRITE(6), disabling FUA\n");
2816 			sdkp->DPOFUA = 0;
2817 		}
2818 
2819 		/* No cache flush allowed for write protected devices */
2820 		if (sdkp->WCE && sdkp->write_prot)
2821 			sdkp->WCE = 0;
2822 
2823 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2824 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2825 			sd_printk(KERN_NOTICE, sdkp,
2826 				  "Write cache: %s, read cache: %s, %s\n",
2827 				  sdkp->WCE ? "enabled" : "disabled",
2828 				  sdkp->RCD ? "disabled" : "enabled",
2829 				  sdkp->DPOFUA ? "supports DPO and FUA"
2830 				  : "doesn't support DPO or FUA");
2831 
2832 		return;
2833 	}
2834 
2835 bad_sense:
2836 	if (scsi_sense_valid(&sshdr) &&
2837 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2838 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2839 		/* Invalid field in CDB */
2840 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2841 	else
2842 		sd_first_printk(KERN_ERR, sdkp,
2843 				"Asking for cache data failed\n");
2844 
2845 defaults:
2846 	if (sdp->wce_default_on) {
2847 		sd_first_printk(KERN_NOTICE, sdkp,
2848 				"Assuming drive cache: write back\n");
2849 		sdkp->WCE = 1;
2850 	} else {
2851 		sd_first_printk(KERN_WARNING, sdkp,
2852 				"Assuming drive cache: write through\n");
2853 		sdkp->WCE = 0;
2854 	}
2855 	sdkp->RCD = 0;
2856 	sdkp->DPOFUA = 0;
2857 }
2858 
2859 /*
2860  * The ATO bit indicates whether the DIF application tag is available
2861  * for use by the operating system.
2862  */
2863 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2864 {
2865 	int res, offset;
2866 	struct scsi_device *sdp = sdkp->device;
2867 	struct scsi_mode_data data;
2868 	struct scsi_sense_hdr sshdr;
2869 
2870 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2871 		return;
2872 
2873 	if (sdkp->protection_type == 0)
2874 		return;
2875 
2876 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2877 			      sdkp->max_retries, &data, &sshdr);
2878 
2879 	if (res < 0 || !data.header_length ||
2880 	    data.length < 6) {
2881 		sd_first_printk(KERN_WARNING, sdkp,
2882 			  "getting Control mode page failed, assume no ATO\n");
2883 
2884 		if (scsi_sense_valid(&sshdr))
2885 			sd_print_sense_hdr(sdkp, &sshdr);
2886 
2887 		return;
2888 	}
2889 
2890 	offset = data.header_length + data.block_descriptor_length;
2891 
2892 	if ((buffer[offset] & 0x3f) != 0x0a) {
2893 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2894 		return;
2895 	}
2896 
2897 	if ((buffer[offset + 5] & 0x80) == 0)
2898 		return;
2899 
2900 	sdkp->ATO = 1;
2901 
2902 	return;
2903 }
2904 
2905 /**
2906  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2907  * @sdkp: disk to query
2908  */
2909 static void sd_read_block_limits(struct scsi_disk *sdkp)
2910 {
2911 	struct scsi_vpd *vpd;
2912 
2913 	rcu_read_lock();
2914 
2915 	vpd = rcu_dereference(sdkp->device->vpd_pgb0);
2916 	if (!vpd || vpd->len < 16)
2917 		goto out;
2918 
2919 	sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
2920 	sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
2921 	sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
2922 
2923 	if (vpd->len >= 64) {
2924 		unsigned int lba_count, desc_count;
2925 
2926 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
2927 
2928 		if (!sdkp->lbpme)
2929 			goto out;
2930 
2931 		lba_count = get_unaligned_be32(&vpd->data[20]);
2932 		desc_count = get_unaligned_be32(&vpd->data[24]);
2933 
2934 		if (lba_count && desc_count)
2935 			sdkp->max_unmap_blocks = lba_count;
2936 
2937 		sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
2938 
2939 		if (vpd->data[32] & 0x80)
2940 			sdkp->unmap_alignment =
2941 				get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
2942 
2943 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2944 
2945 			if (sdkp->max_unmap_blocks)
2946 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2947 			else
2948 				sd_config_discard(sdkp, SD_LBP_WS16);
2949 
2950 		} else {	/* LBP VPD page tells us what to use */
2951 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2952 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2953 			else if (sdkp->lbpws)
2954 				sd_config_discard(sdkp, SD_LBP_WS16);
2955 			else if (sdkp->lbpws10)
2956 				sd_config_discard(sdkp, SD_LBP_WS10);
2957 			else
2958 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2959 		}
2960 	}
2961 
2962  out:
2963 	rcu_read_unlock();
2964 }
2965 
2966 /**
2967  * sd_read_block_characteristics - Query block dev. characteristics
2968  * @sdkp: disk to query
2969  */
2970 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2971 {
2972 	struct request_queue *q = sdkp->disk->queue;
2973 	struct scsi_vpd *vpd;
2974 	u16 rot;
2975 	u8 zoned;
2976 
2977 	rcu_read_lock();
2978 	vpd = rcu_dereference(sdkp->device->vpd_pgb1);
2979 
2980 	if (!vpd || vpd->len < 8) {
2981 		rcu_read_unlock();
2982 	        return;
2983 	}
2984 
2985 	rot = get_unaligned_be16(&vpd->data[4]);
2986 	zoned = (vpd->data[8] >> 4) & 3;
2987 	rcu_read_unlock();
2988 
2989 	if (rot == 1) {
2990 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2991 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2992 	}
2993 
2994 	if (sdkp->device->type == TYPE_ZBC) {
2995 		/* Host-managed */
2996 		disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
2997 	} else {
2998 		sdkp->zoned = zoned;
2999 		if (sdkp->zoned == 1) {
3000 			/* Host-aware */
3001 			disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
3002 		} else {
3003 			/* Regular disk or drive managed disk */
3004 			disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3005 		}
3006 	}
3007 
3008 	if (!sdkp->first_scan)
3009 		return;
3010 
3011 	if (blk_queue_is_zoned(q)) {
3012 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3013 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3014 	} else {
3015 		if (sdkp->zoned == 1)
3016 			sd_printk(KERN_NOTICE, sdkp,
3017 				  "Host-aware SMR disk used as regular disk\n");
3018 		else if (sdkp->zoned == 2)
3019 			sd_printk(KERN_NOTICE, sdkp,
3020 				  "Drive-managed SMR disk\n");
3021 	}
3022 }
3023 
3024 /**
3025  * sd_read_block_provisioning - Query provisioning VPD page
3026  * @sdkp: disk to query
3027  */
3028 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3029 {
3030 	struct scsi_vpd *vpd;
3031 
3032 	if (sdkp->lbpme == 0)
3033 		return;
3034 
3035 	rcu_read_lock();
3036 	vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3037 
3038 	if (!vpd || vpd->len < 8) {
3039 		rcu_read_unlock();
3040 		return;
3041 	}
3042 
3043 	sdkp->lbpvpd	= 1;
3044 	sdkp->lbpu	= (vpd->data[5] >> 7) & 1; /* UNMAP */
3045 	sdkp->lbpws	= (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3046 	sdkp->lbpws10	= (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3047 	rcu_read_unlock();
3048 }
3049 
3050 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3051 {
3052 	struct scsi_device *sdev = sdkp->device;
3053 
3054 	if (sdev->host->no_write_same) {
3055 		sdev->no_write_same = 1;
3056 
3057 		return;
3058 	}
3059 
3060 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3061 		struct scsi_vpd *vpd;
3062 
3063 		sdev->no_report_opcodes = 1;
3064 
3065 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3066 		 * CODES is unsupported and the device has an ATA
3067 		 * Information VPD page (SAT).
3068 		 */
3069 		rcu_read_lock();
3070 		vpd = rcu_dereference(sdev->vpd_pg89);
3071 		if (vpd)
3072 			sdev->no_write_same = 1;
3073 		rcu_read_unlock();
3074 	}
3075 
3076 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3077 		sdkp->ws16 = 1;
3078 
3079 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3080 		sdkp->ws10 = 1;
3081 }
3082 
3083 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3084 {
3085 	struct scsi_device *sdev = sdkp->device;
3086 
3087 	if (!sdev->security_supported)
3088 		return;
3089 
3090 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3091 			SECURITY_PROTOCOL_IN) == 1 &&
3092 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3093 			SECURITY_PROTOCOL_OUT) == 1)
3094 		sdkp->security = 1;
3095 }
3096 
3097 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3098 {
3099 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3100 }
3101 
3102 /**
3103  * sd_read_cpr - Query concurrent positioning ranges
3104  * @sdkp:	disk to query
3105  */
3106 static void sd_read_cpr(struct scsi_disk *sdkp)
3107 {
3108 	struct blk_independent_access_ranges *iars = NULL;
3109 	unsigned char *buffer = NULL;
3110 	unsigned int nr_cpr = 0;
3111 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3112 	u8 *desc;
3113 
3114 	/*
3115 	 * We need to have the capacity set first for the block layer to be
3116 	 * able to check the ranges.
3117 	 */
3118 	if (sdkp->first_scan)
3119 		return;
3120 
3121 	if (!sdkp->capacity)
3122 		goto out;
3123 
3124 	/*
3125 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3126 	 * leading to a maximum page size of 64 + 256*32 bytes.
3127 	 */
3128 	buf_len = 64 + 256*32;
3129 	buffer = kmalloc(buf_len, GFP_KERNEL);
3130 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3131 		goto out;
3132 
3133 	/* We must have at least a 64B header and one 32B range descriptor */
3134 	vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3135 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3136 		sd_printk(KERN_ERR, sdkp,
3137 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3138 		goto out;
3139 	}
3140 
3141 	nr_cpr = (vpd_len - 64) / 32;
3142 	if (nr_cpr == 1) {
3143 		nr_cpr = 0;
3144 		goto out;
3145 	}
3146 
3147 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3148 	if (!iars) {
3149 		nr_cpr = 0;
3150 		goto out;
3151 	}
3152 
3153 	desc = &buffer[64];
3154 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3155 		if (desc[0] != i) {
3156 			sd_printk(KERN_ERR, sdkp,
3157 				"Invalid Concurrent Positioning Range number\n");
3158 			nr_cpr = 0;
3159 			break;
3160 		}
3161 
3162 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3163 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3164 	}
3165 
3166 out:
3167 	disk_set_independent_access_ranges(sdkp->disk, iars);
3168 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3169 		sd_printk(KERN_NOTICE, sdkp,
3170 			  "%u concurrent positioning ranges\n", nr_cpr);
3171 		sdkp->nr_actuators = nr_cpr;
3172 	}
3173 
3174 	kfree(buffer);
3175 }
3176 
3177 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3178 {
3179 	struct scsi_device *sdp = sdkp->device;
3180 	unsigned int min_xfer_bytes =
3181 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3182 
3183 	if (sdkp->min_xfer_blocks == 0)
3184 		return false;
3185 
3186 	if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3187 		sd_first_printk(KERN_WARNING, sdkp,
3188 				"Preferred minimum I/O size %u bytes not a " \
3189 				"multiple of physical block size (%u bytes)\n",
3190 				min_xfer_bytes, sdkp->physical_block_size);
3191 		sdkp->min_xfer_blocks = 0;
3192 		return false;
3193 	}
3194 
3195 	sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3196 			min_xfer_bytes);
3197 	return true;
3198 }
3199 
3200 /*
3201  * Determine the device's preferred I/O size for reads and writes
3202  * unless the reported value is unreasonably small, large, not a
3203  * multiple of the physical block size, or simply garbage.
3204  */
3205 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3206 				      unsigned int dev_max)
3207 {
3208 	struct scsi_device *sdp = sdkp->device;
3209 	unsigned int opt_xfer_bytes =
3210 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3211 	unsigned int min_xfer_bytes =
3212 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3213 
3214 	if (sdkp->opt_xfer_blocks == 0)
3215 		return false;
3216 
3217 	if (sdkp->opt_xfer_blocks > dev_max) {
3218 		sd_first_printk(KERN_WARNING, sdkp,
3219 				"Optimal transfer size %u logical blocks " \
3220 				"> dev_max (%u logical blocks)\n",
3221 				sdkp->opt_xfer_blocks, dev_max);
3222 		return false;
3223 	}
3224 
3225 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3226 		sd_first_printk(KERN_WARNING, sdkp,
3227 				"Optimal transfer size %u logical blocks " \
3228 				"> sd driver limit (%u logical blocks)\n",
3229 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3230 		return false;
3231 	}
3232 
3233 	if (opt_xfer_bytes < PAGE_SIZE) {
3234 		sd_first_printk(KERN_WARNING, sdkp,
3235 				"Optimal transfer size %u bytes < " \
3236 				"PAGE_SIZE (%u bytes)\n",
3237 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3238 		return false;
3239 	}
3240 
3241 	if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3242 		sd_first_printk(KERN_WARNING, sdkp,
3243 				"Optimal transfer size %u bytes not a " \
3244 				"multiple of preferred minimum block " \
3245 				"size (%u bytes)\n",
3246 				opt_xfer_bytes, min_xfer_bytes);
3247 		return false;
3248 	}
3249 
3250 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3251 		sd_first_printk(KERN_WARNING, sdkp,
3252 				"Optimal transfer size %u bytes not a " \
3253 				"multiple of physical block size (%u bytes)\n",
3254 				opt_xfer_bytes, sdkp->physical_block_size);
3255 		return false;
3256 	}
3257 
3258 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3259 			opt_xfer_bytes);
3260 	return true;
3261 }
3262 
3263 /**
3264  *	sd_revalidate_disk - called the first time a new disk is seen,
3265  *	performs disk spin up, read_capacity, etc.
3266  *	@disk: struct gendisk we care about
3267  **/
3268 static int sd_revalidate_disk(struct gendisk *disk)
3269 {
3270 	struct scsi_disk *sdkp = scsi_disk(disk);
3271 	struct scsi_device *sdp = sdkp->device;
3272 	struct request_queue *q = sdkp->disk->queue;
3273 	sector_t old_capacity = sdkp->capacity;
3274 	unsigned char *buffer;
3275 	unsigned int dev_max, rw_max;
3276 
3277 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3278 				      "sd_revalidate_disk\n"));
3279 
3280 	/*
3281 	 * If the device is offline, don't try and read capacity or any
3282 	 * of the other niceties.
3283 	 */
3284 	if (!scsi_device_online(sdp))
3285 		goto out;
3286 
3287 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3288 	if (!buffer) {
3289 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3290 			  "allocation failure.\n");
3291 		goto out;
3292 	}
3293 
3294 	sd_spinup_disk(sdkp);
3295 
3296 	/*
3297 	 * Without media there is no reason to ask; moreover, some devices
3298 	 * react badly if we do.
3299 	 */
3300 	if (sdkp->media_present) {
3301 		sd_read_capacity(sdkp, buffer);
3302 
3303 		/*
3304 		 * set the default to rotational.  All non-rotational devices
3305 		 * support the block characteristics VPD page, which will
3306 		 * cause this to be updated correctly and any device which
3307 		 * doesn't support it should be treated as rotational.
3308 		 */
3309 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3310 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3311 
3312 		if (scsi_device_supports_vpd(sdp)) {
3313 			sd_read_block_provisioning(sdkp);
3314 			sd_read_block_limits(sdkp);
3315 			sd_read_block_characteristics(sdkp);
3316 			sd_zbc_read_zones(sdkp, buffer);
3317 			sd_read_cpr(sdkp);
3318 		}
3319 
3320 		sd_print_capacity(sdkp, old_capacity);
3321 
3322 		sd_read_write_protect_flag(sdkp, buffer);
3323 		sd_read_cache_type(sdkp, buffer);
3324 		sd_read_app_tag_own(sdkp, buffer);
3325 		sd_read_write_same(sdkp, buffer);
3326 		sd_read_security(sdkp, buffer);
3327 		sd_config_protection(sdkp);
3328 	}
3329 
3330 	/*
3331 	 * We now have all cache related info, determine how we deal
3332 	 * with flush requests.
3333 	 */
3334 	sd_set_flush_flag(sdkp);
3335 
3336 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3337 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3338 
3339 	/* Some devices report a maximum block count for READ/WRITE requests. */
3340 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3341 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3342 
3343 	if (sd_validate_min_xfer_size(sdkp))
3344 		blk_queue_io_min(sdkp->disk->queue,
3345 				 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3346 	else
3347 		blk_queue_io_min(sdkp->disk->queue, 0);
3348 
3349 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3350 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3351 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3352 	} else {
3353 		q->limits.io_opt = 0;
3354 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3355 				      (sector_t)BLK_DEF_MAX_SECTORS);
3356 	}
3357 
3358 	/*
3359 	 * Limit default to SCSI host optimal sector limit if set. There may be
3360 	 * an impact on performance for when the size of a request exceeds this
3361 	 * host limit.
3362 	 */
3363 	rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3364 
3365 	/* Do not exceed controller limit */
3366 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3367 
3368 	/*
3369 	 * Only update max_sectors if previously unset or if the current value
3370 	 * exceeds the capabilities of the hardware.
3371 	 */
3372 	if (sdkp->first_scan ||
3373 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3374 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3375 		q->limits.max_sectors = rw_max;
3376 
3377 	sdkp->first_scan = 0;
3378 
3379 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3380 	sd_config_write_same(sdkp);
3381 	kfree(buffer);
3382 
3383 	/*
3384 	 * For a zoned drive, revalidating the zones can be done only once
3385 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3386 	 * capacity to 0.
3387 	 */
3388 	if (sd_zbc_revalidate_zones(sdkp))
3389 		set_capacity_and_notify(disk, 0);
3390 
3391  out:
3392 	return 0;
3393 }
3394 
3395 /**
3396  *	sd_unlock_native_capacity - unlock native capacity
3397  *	@disk: struct gendisk to set capacity for
3398  *
3399  *	Block layer calls this function if it detects that partitions
3400  *	on @disk reach beyond the end of the device.  If the SCSI host
3401  *	implements ->unlock_native_capacity() method, it's invoked to
3402  *	give it a chance to adjust the device capacity.
3403  *
3404  *	CONTEXT:
3405  *	Defined by block layer.  Might sleep.
3406  */
3407 static void sd_unlock_native_capacity(struct gendisk *disk)
3408 {
3409 	struct scsi_device *sdev = scsi_disk(disk)->device;
3410 
3411 	if (sdev->host->hostt->unlock_native_capacity)
3412 		sdev->host->hostt->unlock_native_capacity(sdev);
3413 }
3414 
3415 /**
3416  *	sd_format_disk_name - format disk name
3417  *	@prefix: name prefix - ie. "sd" for SCSI disks
3418  *	@index: index of the disk to format name for
3419  *	@buf: output buffer
3420  *	@buflen: length of the output buffer
3421  *
3422  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3423  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3424  *	which is followed by sdaaa.
3425  *
3426  *	This is basically 26 base counting with one extra 'nil' entry
3427  *	at the beginning from the second digit on and can be
3428  *	determined using similar method as 26 base conversion with the
3429  *	index shifted -1 after each digit is computed.
3430  *
3431  *	CONTEXT:
3432  *	Don't care.
3433  *
3434  *	RETURNS:
3435  *	0 on success, -errno on failure.
3436  */
3437 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3438 {
3439 	const int base = 'z' - 'a' + 1;
3440 	char *begin = buf + strlen(prefix);
3441 	char *end = buf + buflen;
3442 	char *p;
3443 	int unit;
3444 
3445 	p = end - 1;
3446 	*p = '\0';
3447 	unit = base;
3448 	do {
3449 		if (p == begin)
3450 			return -EINVAL;
3451 		*--p = 'a' + (index % unit);
3452 		index = (index / unit) - 1;
3453 	} while (index >= 0);
3454 
3455 	memmove(begin, p, end - p);
3456 	memcpy(buf, prefix, strlen(prefix));
3457 
3458 	return 0;
3459 }
3460 
3461 /**
3462  *	sd_probe - called during driver initialization and whenever a
3463  *	new scsi device is attached to the system. It is called once
3464  *	for each scsi device (not just disks) present.
3465  *	@dev: pointer to device object
3466  *
3467  *	Returns 0 if successful (or not interested in this scsi device
3468  *	(e.g. scanner)); 1 when there is an error.
3469  *
3470  *	Note: this function is invoked from the scsi mid-level.
3471  *	This function sets up the mapping between a given
3472  *	<host,channel,id,lun> (found in sdp) and new device name
3473  *	(e.g. /dev/sda). More precisely it is the block device major
3474  *	and minor number that is chosen here.
3475  *
3476  *	Assume sd_probe is not re-entrant (for time being)
3477  *	Also think about sd_probe() and sd_remove() running coincidentally.
3478  **/
3479 static int sd_probe(struct device *dev)
3480 {
3481 	struct scsi_device *sdp = to_scsi_device(dev);
3482 	struct scsi_disk *sdkp;
3483 	struct gendisk *gd;
3484 	int index;
3485 	int error;
3486 
3487 	scsi_autopm_get_device(sdp);
3488 	error = -ENODEV;
3489 	if (sdp->type != TYPE_DISK &&
3490 	    sdp->type != TYPE_ZBC &&
3491 	    sdp->type != TYPE_MOD &&
3492 	    sdp->type != TYPE_RBC)
3493 		goto out;
3494 
3495 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3496 		sdev_printk(KERN_WARNING, sdp,
3497 			    "Unsupported ZBC host-managed device.\n");
3498 		goto out;
3499 	}
3500 
3501 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3502 					"sd_probe\n"));
3503 
3504 	error = -ENOMEM;
3505 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3506 	if (!sdkp)
3507 		goto out;
3508 
3509 	gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3510 					 &sd_bio_compl_lkclass);
3511 	if (!gd)
3512 		goto out_free;
3513 
3514 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3515 	if (index < 0) {
3516 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3517 		goto out_put;
3518 	}
3519 
3520 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3521 	if (error) {
3522 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3523 		goto out_free_index;
3524 	}
3525 
3526 	sdkp->device = sdp;
3527 	sdkp->disk = gd;
3528 	sdkp->index = index;
3529 	sdkp->max_retries = SD_MAX_RETRIES;
3530 	atomic_set(&sdkp->openers, 0);
3531 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3532 
3533 	if (!sdp->request_queue->rq_timeout) {
3534 		if (sdp->type != TYPE_MOD)
3535 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3536 		else
3537 			blk_queue_rq_timeout(sdp->request_queue,
3538 					     SD_MOD_TIMEOUT);
3539 	}
3540 
3541 	device_initialize(&sdkp->disk_dev);
3542 	sdkp->disk_dev.parent = get_device(dev);
3543 	sdkp->disk_dev.class = &sd_disk_class;
3544 	dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3545 
3546 	error = device_add(&sdkp->disk_dev);
3547 	if (error) {
3548 		put_device(&sdkp->disk_dev);
3549 		goto out;
3550 	}
3551 
3552 	dev_set_drvdata(dev, sdkp);
3553 
3554 	gd->major = sd_major((index & 0xf0) >> 4);
3555 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3556 	gd->minors = SD_MINORS;
3557 
3558 	gd->fops = &sd_fops;
3559 	gd->private_data = sdkp;
3560 
3561 	/* defaults, until the device tells us otherwise */
3562 	sdp->sector_size = 512;
3563 	sdkp->capacity = 0;
3564 	sdkp->media_present = 1;
3565 	sdkp->write_prot = 0;
3566 	sdkp->cache_override = 0;
3567 	sdkp->WCE = 0;
3568 	sdkp->RCD = 0;
3569 	sdkp->ATO = 0;
3570 	sdkp->first_scan = 1;
3571 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3572 
3573 	sd_revalidate_disk(gd);
3574 
3575 	if (sdp->removable) {
3576 		gd->flags |= GENHD_FL_REMOVABLE;
3577 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3578 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3579 	}
3580 
3581 	blk_pm_runtime_init(sdp->request_queue, dev);
3582 	if (sdp->rpm_autosuspend) {
3583 		pm_runtime_set_autosuspend_delay(dev,
3584 			sdp->host->hostt->rpm_autosuspend_delay);
3585 	}
3586 
3587 	error = device_add_disk(dev, gd, NULL);
3588 	if (error) {
3589 		put_device(&sdkp->disk_dev);
3590 		put_disk(gd);
3591 		goto out;
3592 	}
3593 
3594 	if (sdkp->security) {
3595 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3596 		if (sdkp->opal_dev)
3597 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3598 	}
3599 
3600 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3601 		  sdp->removable ? "removable " : "");
3602 	scsi_autopm_put_device(sdp);
3603 
3604 	return 0;
3605 
3606  out_free_index:
3607 	ida_free(&sd_index_ida, index);
3608  out_put:
3609 	put_disk(gd);
3610  out_free:
3611 	kfree(sdkp);
3612  out:
3613 	scsi_autopm_put_device(sdp);
3614 	return error;
3615 }
3616 
3617 /**
3618  *	sd_remove - called whenever a scsi disk (previously recognized by
3619  *	sd_probe) is detached from the system. It is called (potentially
3620  *	multiple times) during sd module unload.
3621  *	@dev: pointer to device object
3622  *
3623  *	Note: this function is invoked from the scsi mid-level.
3624  *	This function potentially frees up a device name (e.g. /dev/sdc)
3625  *	that could be re-used by a subsequent sd_probe().
3626  *	This function is not called when the built-in sd driver is "exit-ed".
3627  **/
3628 static int sd_remove(struct device *dev)
3629 {
3630 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3631 
3632 	scsi_autopm_get_device(sdkp->device);
3633 
3634 	device_del(&sdkp->disk_dev);
3635 	del_gendisk(sdkp->disk);
3636 	sd_shutdown(dev);
3637 
3638 	put_disk(sdkp->disk);
3639 	return 0;
3640 }
3641 
3642 static void scsi_disk_release(struct device *dev)
3643 {
3644 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3645 
3646 	ida_free(&sd_index_ida, sdkp->index);
3647 	sd_zbc_free_zone_info(sdkp);
3648 	put_device(&sdkp->device->sdev_gendev);
3649 	free_opal_dev(sdkp->opal_dev);
3650 
3651 	kfree(sdkp);
3652 }
3653 
3654 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3655 {
3656 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3657 	struct scsi_sense_hdr sshdr;
3658 	const struct scsi_exec_args exec_args = {
3659 		.sshdr = &sshdr,
3660 		.req_flags = BLK_MQ_REQ_PM,
3661 	};
3662 	struct scsi_device *sdp = sdkp->device;
3663 	int res;
3664 
3665 	if (start)
3666 		cmd[4] |= 1;	/* START */
3667 
3668 	if (sdp->start_stop_pwr_cond)
3669 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3670 
3671 	if (!scsi_device_online(sdp))
3672 		return -ENODEV;
3673 
3674 	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
3675 			       sdkp->max_retries, &exec_args);
3676 	if (res) {
3677 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3678 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3679 			sd_print_sense_hdr(sdkp, &sshdr);
3680 			/* 0x3a is medium not present */
3681 			if (sshdr.asc == 0x3a)
3682 				res = 0;
3683 		}
3684 	}
3685 
3686 	/* SCSI error codes must not go to the generic layer */
3687 	if (res)
3688 		return -EIO;
3689 
3690 	return 0;
3691 }
3692 
3693 /*
3694  * Send a SYNCHRONIZE CACHE instruction down to the device through
3695  * the normal SCSI command structure.  Wait for the command to
3696  * complete.
3697  */
3698 static void sd_shutdown(struct device *dev)
3699 {
3700 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3701 
3702 	if (!sdkp)
3703 		return;         /* this can happen */
3704 
3705 	if (pm_runtime_suspended(dev))
3706 		return;
3707 
3708 	if (sdkp->WCE && sdkp->media_present) {
3709 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3710 		sd_sync_cache(sdkp, NULL);
3711 	}
3712 
3713 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3714 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3715 		sd_start_stop_device(sdkp, 0);
3716 	}
3717 }
3718 
3719 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3720 {
3721 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3722 	struct scsi_sense_hdr sshdr;
3723 	int ret = 0;
3724 
3725 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3726 		return 0;
3727 
3728 	if (sdkp->WCE && sdkp->media_present) {
3729 		if (!sdkp->device->silence_suspend)
3730 			sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3731 		ret = sd_sync_cache(sdkp, &sshdr);
3732 
3733 		if (ret) {
3734 			/* ignore OFFLINE device */
3735 			if (ret == -ENODEV)
3736 				return 0;
3737 
3738 			if (!scsi_sense_valid(&sshdr) ||
3739 			    sshdr.sense_key != ILLEGAL_REQUEST)
3740 				return ret;
3741 
3742 			/*
3743 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3744 			 * doesn't support sync. There's not much to do and
3745 			 * suspend shouldn't fail.
3746 			 */
3747 			ret = 0;
3748 		}
3749 	}
3750 
3751 	if (sdkp->device->manage_start_stop) {
3752 		if (!sdkp->device->silence_suspend)
3753 			sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3754 		/* an error is not worth aborting a system sleep */
3755 		ret = sd_start_stop_device(sdkp, 0);
3756 		if (ignore_stop_errors)
3757 			ret = 0;
3758 	}
3759 
3760 	return ret;
3761 }
3762 
3763 static int sd_suspend_system(struct device *dev)
3764 {
3765 	if (pm_runtime_suspended(dev))
3766 		return 0;
3767 
3768 	return sd_suspend_common(dev, true);
3769 }
3770 
3771 static int sd_suspend_runtime(struct device *dev)
3772 {
3773 	return sd_suspend_common(dev, false);
3774 }
3775 
3776 static int sd_resume(struct device *dev)
3777 {
3778 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3779 	int ret;
3780 
3781 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3782 		return 0;
3783 
3784 	if (!sdkp->device->manage_start_stop)
3785 		return 0;
3786 
3787 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3788 	ret = sd_start_stop_device(sdkp, 1);
3789 	if (!ret)
3790 		opal_unlock_from_suspend(sdkp->opal_dev);
3791 	return ret;
3792 }
3793 
3794 static int sd_resume_system(struct device *dev)
3795 {
3796 	if (pm_runtime_suspended(dev))
3797 		return 0;
3798 
3799 	return sd_resume(dev);
3800 }
3801 
3802 static int sd_resume_runtime(struct device *dev)
3803 {
3804 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3805 	struct scsi_device *sdp;
3806 
3807 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3808 		return 0;
3809 
3810 	sdp = sdkp->device;
3811 
3812 	if (sdp->ignore_media_change) {
3813 		/* clear the device's sense data */
3814 		static const u8 cmd[10] = { REQUEST_SENSE };
3815 		const struct scsi_exec_args exec_args = {
3816 			.req_flags = BLK_MQ_REQ_PM,
3817 		};
3818 
3819 		if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
3820 				     sdp->request_queue->rq_timeout, 1,
3821 				     &exec_args))
3822 			sd_printk(KERN_NOTICE, sdkp,
3823 				  "Failed to clear sense data\n");
3824 	}
3825 
3826 	return sd_resume(dev);
3827 }
3828 
3829 /**
3830  *	init_sd - entry point for this driver (both when built in or when
3831  *	a module).
3832  *
3833  *	Note: this function registers this driver with the scsi mid-level.
3834  **/
3835 static int __init init_sd(void)
3836 {
3837 	int majors = 0, i, err;
3838 
3839 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3840 
3841 	for (i = 0; i < SD_MAJORS; i++) {
3842 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3843 			continue;
3844 		majors++;
3845 	}
3846 
3847 	if (!majors)
3848 		return -ENODEV;
3849 
3850 	err = class_register(&sd_disk_class);
3851 	if (err)
3852 		goto err_out;
3853 
3854 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3855 					 0, 0, NULL);
3856 	if (!sd_cdb_cache) {
3857 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3858 		err = -ENOMEM;
3859 		goto err_out_class;
3860 	}
3861 
3862 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3863 	if (!sd_page_pool) {
3864 		printk(KERN_ERR "sd: can't init discard page pool\n");
3865 		err = -ENOMEM;
3866 		goto err_out_cache;
3867 	}
3868 
3869 	err = scsi_register_driver(&sd_template.gendrv);
3870 	if (err)
3871 		goto err_out_driver;
3872 
3873 	return 0;
3874 
3875 err_out_driver:
3876 	mempool_destroy(sd_page_pool);
3877 
3878 err_out_cache:
3879 	kmem_cache_destroy(sd_cdb_cache);
3880 
3881 err_out_class:
3882 	class_unregister(&sd_disk_class);
3883 err_out:
3884 	for (i = 0; i < SD_MAJORS; i++)
3885 		unregister_blkdev(sd_major(i), "sd");
3886 	return err;
3887 }
3888 
3889 /**
3890  *	exit_sd - exit point for this driver (when it is a module).
3891  *
3892  *	Note: this function unregisters this driver from the scsi mid-level.
3893  **/
3894 static void __exit exit_sd(void)
3895 {
3896 	int i;
3897 
3898 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3899 
3900 	scsi_unregister_driver(&sd_template.gendrv);
3901 	mempool_destroy(sd_page_pool);
3902 	kmem_cache_destroy(sd_cdb_cache);
3903 
3904 	class_unregister(&sd_disk_class);
3905 
3906 	for (i = 0; i < SD_MAJORS; i++)
3907 		unregister_blkdev(sd_major(i), "sd");
3908 }
3909 
3910 module_init(init_sd);
3911 module_exit(exit_sd);
3912 
3913 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3914 {
3915 	scsi_print_sense_hdr(sdkp->device,
3916 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3917 }
3918 
3919 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3920 {
3921 	const char *hb_string = scsi_hostbyte_string(result);
3922 
3923 	if (hb_string)
3924 		sd_printk(KERN_INFO, sdkp,
3925 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3926 			  hb_string ? hb_string : "invalid",
3927 			  "DRIVER_OK");
3928 	else
3929 		sd_printk(KERN_INFO, sdkp,
3930 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3931 			  msg, host_byte(result), "DRIVER_OK");
3932 }
3933