xref: /openbmc/linux/drivers/scsi/sd.c (revision 745b7908)
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 mempool_t *sd_page_pool;
125 static struct lock_class_key sd_bio_compl_lkclass;
126 
127 static const char *sd_cache_types[] = {
128 	"write through", "none", "write back",
129 	"write back, no read (daft)"
130 };
131 
132 static void sd_set_flush_flag(struct scsi_disk *sdkp)
133 {
134 	bool wc = false, fua = false;
135 
136 	if (sdkp->WCE) {
137 		wc = true;
138 		if (sdkp->DPOFUA)
139 			fua = true;
140 	}
141 
142 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
143 }
144 
145 static ssize_t
146 cache_type_store(struct device *dev, struct device_attribute *attr,
147 		 const char *buf, size_t count)
148 {
149 	int ct, rcd, wce, sp;
150 	struct scsi_disk *sdkp = to_scsi_disk(dev);
151 	struct scsi_device *sdp = sdkp->device;
152 	char buffer[64];
153 	char *buffer_data;
154 	struct scsi_mode_data data;
155 	struct scsi_sense_hdr sshdr;
156 	static const char temp[] = "temporary ";
157 	int len;
158 
159 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
160 		/* no cache control on RBC devices; theoretically they
161 		 * can do it, but there's probably so many exceptions
162 		 * it's not worth the risk */
163 		return -EINVAL;
164 
165 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
166 		buf += sizeof(temp) - 1;
167 		sdkp->cache_override = 1;
168 	} else {
169 		sdkp->cache_override = 0;
170 	}
171 
172 	ct = sysfs_match_string(sd_cache_types, buf);
173 	if (ct < 0)
174 		return -EINVAL;
175 
176 	rcd = ct & 0x01 ? 1 : 0;
177 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
178 
179 	if (sdkp->cache_override) {
180 		sdkp->WCE = wce;
181 		sdkp->RCD = rcd;
182 		sd_set_flush_flag(sdkp);
183 		return count;
184 	}
185 
186 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
187 			    sdkp->max_retries, &data, NULL))
188 		return -EINVAL;
189 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
190 		  data.block_descriptor_length);
191 	buffer_data = buffer + data.header_length +
192 		data.block_descriptor_length;
193 	buffer_data[2] &= ~0x05;
194 	buffer_data[2] |= wce << 2 | rcd;
195 	sp = buffer_data[0] & 0x80 ? 1 : 0;
196 	buffer_data[0] &= ~0x80;
197 
198 	/*
199 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
200 	 * received mode parameter buffer before doing MODE SELECT.
201 	 */
202 	data.device_specific = 0;
203 
204 	if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
205 			     sdkp->max_retries, &data, &sshdr)) {
206 		if (scsi_sense_valid(&sshdr))
207 			sd_print_sense_hdr(sdkp, &sshdr);
208 		return -EINVAL;
209 	}
210 	sd_revalidate_disk(sdkp->disk);
211 	return count;
212 }
213 
214 static ssize_t
215 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
216 		       char *buf)
217 {
218 	struct scsi_disk *sdkp = to_scsi_disk(dev);
219 	struct scsi_device *sdp = sdkp->device;
220 
221 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
222 }
223 
224 static ssize_t
225 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
226 			const char *buf, size_t count)
227 {
228 	struct scsi_disk *sdkp = to_scsi_disk(dev);
229 	struct scsi_device *sdp = sdkp->device;
230 	bool v;
231 
232 	if (!capable(CAP_SYS_ADMIN))
233 		return -EACCES;
234 
235 	if (kstrtobool(buf, &v))
236 		return -EINVAL;
237 
238 	sdp->manage_start_stop = v;
239 
240 	return count;
241 }
242 static DEVICE_ATTR_RW(manage_start_stop);
243 
244 static ssize_t
245 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
246 {
247 	struct scsi_disk *sdkp = to_scsi_disk(dev);
248 
249 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
250 }
251 
252 static ssize_t
253 allow_restart_store(struct device *dev, struct device_attribute *attr,
254 		    const char *buf, size_t count)
255 {
256 	bool v;
257 	struct scsi_disk *sdkp = to_scsi_disk(dev);
258 	struct scsi_device *sdp = sdkp->device;
259 
260 	if (!capable(CAP_SYS_ADMIN))
261 		return -EACCES;
262 
263 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
264 		return -EINVAL;
265 
266 	if (kstrtobool(buf, &v))
267 		return -EINVAL;
268 
269 	sdp->allow_restart = v;
270 
271 	return count;
272 }
273 static DEVICE_ATTR_RW(allow_restart);
274 
275 static ssize_t
276 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
277 {
278 	struct scsi_disk *sdkp = to_scsi_disk(dev);
279 	int ct = sdkp->RCD + 2*sdkp->WCE;
280 
281 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
282 }
283 static DEVICE_ATTR_RW(cache_type);
284 
285 static ssize_t
286 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
287 {
288 	struct scsi_disk *sdkp = to_scsi_disk(dev);
289 
290 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
291 }
292 static DEVICE_ATTR_RO(FUA);
293 
294 static ssize_t
295 protection_type_show(struct device *dev, struct device_attribute *attr,
296 		     char *buf)
297 {
298 	struct scsi_disk *sdkp = to_scsi_disk(dev);
299 
300 	return sprintf(buf, "%u\n", sdkp->protection_type);
301 }
302 
303 static ssize_t
304 protection_type_store(struct device *dev, struct device_attribute *attr,
305 		      const char *buf, size_t count)
306 {
307 	struct scsi_disk *sdkp = to_scsi_disk(dev);
308 	unsigned int val;
309 	int err;
310 
311 	if (!capable(CAP_SYS_ADMIN))
312 		return -EACCES;
313 
314 	err = kstrtouint(buf, 10, &val);
315 
316 	if (err)
317 		return err;
318 
319 	if (val <= T10_PI_TYPE3_PROTECTION)
320 		sdkp->protection_type = val;
321 
322 	return count;
323 }
324 static DEVICE_ATTR_RW(protection_type);
325 
326 static ssize_t
327 protection_mode_show(struct device *dev, struct device_attribute *attr,
328 		     char *buf)
329 {
330 	struct scsi_disk *sdkp = to_scsi_disk(dev);
331 	struct scsi_device *sdp = sdkp->device;
332 	unsigned int dif, dix;
333 
334 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
335 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
336 
337 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
338 		dif = 0;
339 		dix = 1;
340 	}
341 
342 	if (!dif && !dix)
343 		return sprintf(buf, "none\n");
344 
345 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
346 }
347 static DEVICE_ATTR_RO(protection_mode);
348 
349 static ssize_t
350 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
351 {
352 	struct scsi_disk *sdkp = to_scsi_disk(dev);
353 
354 	return sprintf(buf, "%u\n", sdkp->ATO);
355 }
356 static DEVICE_ATTR_RO(app_tag_own);
357 
358 static ssize_t
359 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
360 		       char *buf)
361 {
362 	struct scsi_disk *sdkp = to_scsi_disk(dev);
363 
364 	return sprintf(buf, "%u\n", sdkp->lbpme);
365 }
366 static DEVICE_ATTR_RO(thin_provisioning);
367 
368 /* sysfs_match_string() requires dense arrays */
369 static const char *lbp_mode[] = {
370 	[SD_LBP_FULL]		= "full",
371 	[SD_LBP_UNMAP]		= "unmap",
372 	[SD_LBP_WS16]		= "writesame_16",
373 	[SD_LBP_WS10]		= "writesame_10",
374 	[SD_LBP_ZERO]		= "writesame_zero",
375 	[SD_LBP_DISABLE]	= "disabled",
376 };
377 
378 static ssize_t
379 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
380 		       char *buf)
381 {
382 	struct scsi_disk *sdkp = to_scsi_disk(dev);
383 
384 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
385 }
386 
387 static ssize_t
388 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
389 			const char *buf, size_t count)
390 {
391 	struct scsi_disk *sdkp = to_scsi_disk(dev);
392 	struct scsi_device *sdp = sdkp->device;
393 	int mode;
394 
395 	if (!capable(CAP_SYS_ADMIN))
396 		return -EACCES;
397 
398 	if (sd_is_zoned(sdkp)) {
399 		sd_config_discard(sdkp, SD_LBP_DISABLE);
400 		return count;
401 	}
402 
403 	if (sdp->type != TYPE_DISK)
404 		return -EINVAL;
405 
406 	mode = sysfs_match_string(lbp_mode, buf);
407 	if (mode < 0)
408 		return -EINVAL;
409 
410 	sd_config_discard(sdkp, mode);
411 
412 	return count;
413 }
414 static DEVICE_ATTR_RW(provisioning_mode);
415 
416 /* sysfs_match_string() requires dense arrays */
417 static const char *zeroing_mode[] = {
418 	[SD_ZERO_WRITE]		= "write",
419 	[SD_ZERO_WS]		= "writesame",
420 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
421 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
422 };
423 
424 static ssize_t
425 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
426 		  char *buf)
427 {
428 	struct scsi_disk *sdkp = to_scsi_disk(dev);
429 
430 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
431 }
432 
433 static ssize_t
434 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
435 		   const char *buf, size_t count)
436 {
437 	struct scsi_disk *sdkp = to_scsi_disk(dev);
438 	int mode;
439 
440 	if (!capable(CAP_SYS_ADMIN))
441 		return -EACCES;
442 
443 	mode = sysfs_match_string(zeroing_mode, buf);
444 	if (mode < 0)
445 		return -EINVAL;
446 
447 	sdkp->zeroing_mode = mode;
448 
449 	return count;
450 }
451 static DEVICE_ATTR_RW(zeroing_mode);
452 
453 static ssize_t
454 max_medium_access_timeouts_show(struct device *dev,
455 				struct device_attribute *attr, char *buf)
456 {
457 	struct scsi_disk *sdkp = to_scsi_disk(dev);
458 
459 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
460 }
461 
462 static ssize_t
463 max_medium_access_timeouts_store(struct device *dev,
464 				 struct device_attribute *attr, const char *buf,
465 				 size_t count)
466 {
467 	struct scsi_disk *sdkp = to_scsi_disk(dev);
468 	int err;
469 
470 	if (!capable(CAP_SYS_ADMIN))
471 		return -EACCES;
472 
473 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
474 
475 	return err ? err : count;
476 }
477 static DEVICE_ATTR_RW(max_medium_access_timeouts);
478 
479 static ssize_t
480 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
481 			   char *buf)
482 {
483 	struct scsi_disk *sdkp = to_scsi_disk(dev);
484 
485 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
486 }
487 
488 static ssize_t
489 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
490 			    const char *buf, size_t count)
491 {
492 	struct scsi_disk *sdkp = to_scsi_disk(dev);
493 	struct scsi_device *sdp = sdkp->device;
494 	unsigned long max;
495 	int err;
496 
497 	if (!capable(CAP_SYS_ADMIN))
498 		return -EACCES;
499 
500 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
501 		return -EINVAL;
502 
503 	err = kstrtoul(buf, 10, &max);
504 
505 	if (err)
506 		return err;
507 
508 	if (max == 0)
509 		sdp->no_write_same = 1;
510 	else if (max <= SD_MAX_WS16_BLOCKS) {
511 		sdp->no_write_same = 0;
512 		sdkp->max_ws_blocks = max;
513 	}
514 
515 	sd_config_write_same(sdkp);
516 
517 	return count;
518 }
519 static DEVICE_ATTR_RW(max_write_same_blocks);
520 
521 static ssize_t
522 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
523 {
524 	struct scsi_disk *sdkp = to_scsi_disk(dev);
525 
526 	if (sdkp->device->type == TYPE_ZBC)
527 		return sprintf(buf, "host-managed\n");
528 	if (sdkp->zoned == 1)
529 		return sprintf(buf, "host-aware\n");
530 	if (sdkp->zoned == 2)
531 		return sprintf(buf, "drive-managed\n");
532 	return sprintf(buf, "none\n");
533 }
534 static DEVICE_ATTR_RO(zoned_cap);
535 
536 static ssize_t
537 max_retries_store(struct device *dev, struct device_attribute *attr,
538 		  const char *buf, size_t count)
539 {
540 	struct scsi_disk *sdkp = to_scsi_disk(dev);
541 	struct scsi_device *sdev = sdkp->device;
542 	int retries, err;
543 
544 	err = kstrtoint(buf, 10, &retries);
545 	if (err)
546 		return err;
547 
548 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
549 		sdkp->max_retries = retries;
550 		return count;
551 	}
552 
553 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
554 		    SD_MAX_RETRIES);
555 	return -EINVAL;
556 }
557 
558 static ssize_t
559 max_retries_show(struct device *dev, struct device_attribute *attr,
560 		 char *buf)
561 {
562 	struct scsi_disk *sdkp = to_scsi_disk(dev);
563 
564 	return sprintf(buf, "%d\n", sdkp->max_retries);
565 }
566 
567 static DEVICE_ATTR_RW(max_retries);
568 
569 static struct attribute *sd_disk_attrs[] = {
570 	&dev_attr_cache_type.attr,
571 	&dev_attr_FUA.attr,
572 	&dev_attr_allow_restart.attr,
573 	&dev_attr_manage_start_stop.attr,
574 	&dev_attr_protection_type.attr,
575 	&dev_attr_protection_mode.attr,
576 	&dev_attr_app_tag_own.attr,
577 	&dev_attr_thin_provisioning.attr,
578 	&dev_attr_provisioning_mode.attr,
579 	&dev_attr_zeroing_mode.attr,
580 	&dev_attr_max_write_same_blocks.attr,
581 	&dev_attr_max_medium_access_timeouts.attr,
582 	&dev_attr_zoned_cap.attr,
583 	&dev_attr_max_retries.attr,
584 	NULL,
585 };
586 ATTRIBUTE_GROUPS(sd_disk);
587 
588 static struct class sd_disk_class = {
589 	.name		= "scsi_disk",
590 	.dev_release	= scsi_disk_release,
591 	.dev_groups	= sd_disk_groups,
592 };
593 
594 static const struct dev_pm_ops sd_pm_ops = {
595 	.suspend		= sd_suspend_system,
596 	.resume			= sd_resume_system,
597 	.poweroff		= sd_suspend_system,
598 	.restore		= sd_resume_system,
599 	.runtime_suspend	= sd_suspend_runtime,
600 	.runtime_resume		= sd_resume_runtime,
601 };
602 
603 static struct scsi_driver sd_template = {
604 	.gendrv = {
605 		.name		= "sd",
606 		.owner		= THIS_MODULE,
607 		.probe		= sd_probe,
608 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
609 		.remove		= sd_remove,
610 		.shutdown	= sd_shutdown,
611 		.pm		= &sd_pm_ops,
612 	},
613 	.rescan			= sd_rescan,
614 	.init_command		= sd_init_command,
615 	.uninit_command		= sd_uninit_command,
616 	.done			= sd_done,
617 	.eh_action		= sd_eh_action,
618 	.eh_reset		= sd_eh_reset,
619 };
620 
621 /*
622  * Don't request a new module, as that could deadlock in multipath
623  * environment.
624  */
625 static void sd_default_probe(dev_t devt)
626 {
627 }
628 
629 /*
630  * Device no to disk mapping:
631  *
632  *       major         disc2     disc  p1
633  *   |............|.............|....|....| <- dev_t
634  *    31        20 19          8 7  4 3  0
635  *
636  * Inside a major, we have 16k disks, however mapped non-
637  * contiguously. The first 16 disks are for major0, the next
638  * ones with major1, ... Disk 256 is for major0 again, disk 272
639  * for major1, ...
640  * As we stay compatible with our numbering scheme, we can reuse
641  * the well-know SCSI majors 8, 65--71, 136--143.
642  */
643 static int sd_major(int major_idx)
644 {
645 	switch (major_idx) {
646 	case 0:
647 		return SCSI_DISK0_MAJOR;
648 	case 1 ... 7:
649 		return SCSI_DISK1_MAJOR + major_idx - 1;
650 	case 8 ... 15:
651 		return SCSI_DISK8_MAJOR + major_idx - 8;
652 	default:
653 		BUG();
654 		return 0;	/* shut up gcc */
655 	}
656 }
657 
658 #ifdef CONFIG_BLK_SED_OPAL
659 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
660 		size_t len, bool send)
661 {
662 	struct scsi_disk *sdkp = data;
663 	struct scsi_device *sdev = sdkp->device;
664 	u8 cdb[12] = { 0, };
665 	const struct scsi_exec_args exec_args = {
666 		.req_flags = BLK_MQ_REQ_PM,
667 	};
668 	int ret;
669 
670 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
671 	cdb[1] = secp;
672 	put_unaligned_be16(spsp, &cdb[2]);
673 	put_unaligned_be32(len, &cdb[6]);
674 
675 	ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
676 			       buffer, len, SD_TIMEOUT, sdkp->max_retries,
677 			       &exec_args);
678 	return ret <= 0 ? ret : -EIO;
679 }
680 #endif /* CONFIG_BLK_SED_OPAL */
681 
682 /*
683  * Look up the DIX operation based on whether the command is read or
684  * write and whether dix and dif are enabled.
685  */
686 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
687 {
688 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
689 	static const unsigned int ops[] = {	/* wrt dix dif */
690 		SCSI_PROT_NORMAL,		/*  0	0   0  */
691 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
692 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
693 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
694 		SCSI_PROT_NORMAL,		/*  1	0   0  */
695 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
696 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
697 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
698 	};
699 
700 	return ops[write << 2 | dix << 1 | dif];
701 }
702 
703 /*
704  * Returns a mask of the protection flags that are valid for a given DIX
705  * operation.
706  */
707 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
708 {
709 	static const unsigned int flag_mask[] = {
710 		[SCSI_PROT_NORMAL]		= 0,
711 
712 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
713 						  SCSI_PROT_GUARD_CHECK |
714 						  SCSI_PROT_REF_CHECK |
715 						  SCSI_PROT_REF_INCREMENT,
716 
717 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
718 						  SCSI_PROT_IP_CHECKSUM,
719 
720 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
721 						  SCSI_PROT_GUARD_CHECK |
722 						  SCSI_PROT_REF_CHECK |
723 						  SCSI_PROT_REF_INCREMENT |
724 						  SCSI_PROT_IP_CHECKSUM,
725 
726 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
727 						  SCSI_PROT_REF_INCREMENT,
728 
729 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
730 						  SCSI_PROT_REF_CHECK |
731 						  SCSI_PROT_REF_INCREMENT |
732 						  SCSI_PROT_IP_CHECKSUM,
733 
734 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
735 						  SCSI_PROT_GUARD_CHECK |
736 						  SCSI_PROT_REF_CHECK |
737 						  SCSI_PROT_REF_INCREMENT |
738 						  SCSI_PROT_IP_CHECKSUM,
739 	};
740 
741 	return flag_mask[prot_op];
742 }
743 
744 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
745 					   unsigned int dix, unsigned int dif)
746 {
747 	struct request *rq = scsi_cmd_to_rq(scmd);
748 	struct bio *bio = rq->bio;
749 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
750 	unsigned int protect = 0;
751 
752 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
753 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
754 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
755 
756 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
757 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
758 	}
759 
760 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
761 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
762 
763 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
764 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
765 	}
766 
767 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
768 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
769 
770 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
771 			protect = 3 << 5;	/* Disable target PI checking */
772 		else
773 			protect = 1 << 5;	/* Enable target PI checking */
774 	}
775 
776 	scsi_set_prot_op(scmd, prot_op);
777 	scsi_set_prot_type(scmd, dif);
778 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
779 
780 	return protect;
781 }
782 
783 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
784 {
785 	struct request_queue *q = sdkp->disk->queue;
786 	unsigned int logical_block_size = sdkp->device->sector_size;
787 	unsigned int max_blocks = 0;
788 
789 	q->limits.discard_alignment =
790 		sdkp->unmap_alignment * logical_block_size;
791 	q->limits.discard_granularity =
792 		max(sdkp->physical_block_size,
793 		    sdkp->unmap_granularity * logical_block_size);
794 	sdkp->provisioning_mode = mode;
795 
796 	switch (mode) {
797 
798 	case SD_LBP_FULL:
799 	case SD_LBP_DISABLE:
800 		blk_queue_max_discard_sectors(q, 0);
801 		return;
802 
803 	case SD_LBP_UNMAP:
804 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
805 					  (u32)SD_MAX_WS16_BLOCKS);
806 		break;
807 
808 	case SD_LBP_WS16:
809 		if (sdkp->device->unmap_limit_for_ws)
810 			max_blocks = sdkp->max_unmap_blocks;
811 		else
812 			max_blocks = sdkp->max_ws_blocks;
813 
814 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
815 		break;
816 
817 	case SD_LBP_WS10:
818 		if (sdkp->device->unmap_limit_for_ws)
819 			max_blocks = sdkp->max_unmap_blocks;
820 		else
821 			max_blocks = sdkp->max_ws_blocks;
822 
823 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
824 		break;
825 
826 	case SD_LBP_ZERO:
827 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
828 					  (u32)SD_MAX_WS10_BLOCKS);
829 		break;
830 	}
831 
832 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
833 }
834 
835 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
836 {
837 	struct page *page;
838 
839 	page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
840 	if (!page)
841 		return NULL;
842 	clear_highpage(page);
843 	bvec_set_page(&rq->special_vec, page, data_len, 0);
844 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
845 	return bvec_virt(&rq->special_vec);
846 }
847 
848 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
849 {
850 	struct scsi_device *sdp = cmd->device;
851 	struct request *rq = scsi_cmd_to_rq(cmd);
852 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
853 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
854 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
855 	unsigned int data_len = 24;
856 	char *buf;
857 
858 	buf = sd_set_special_bvec(rq, data_len);
859 	if (!buf)
860 		return BLK_STS_RESOURCE;
861 
862 	cmd->cmd_len = 10;
863 	cmd->cmnd[0] = UNMAP;
864 	cmd->cmnd[8] = 24;
865 
866 	put_unaligned_be16(6 + 16, &buf[0]);
867 	put_unaligned_be16(16, &buf[2]);
868 	put_unaligned_be64(lba, &buf[8]);
869 	put_unaligned_be32(nr_blocks, &buf[16]);
870 
871 	cmd->allowed = sdkp->max_retries;
872 	cmd->transfersize = data_len;
873 	rq->timeout = SD_TIMEOUT;
874 
875 	return scsi_alloc_sgtables(cmd);
876 }
877 
878 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
879 		bool unmap)
880 {
881 	struct scsi_device *sdp = cmd->device;
882 	struct request *rq = scsi_cmd_to_rq(cmd);
883 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
884 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
885 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
886 	u32 data_len = sdp->sector_size;
887 
888 	if (!sd_set_special_bvec(rq, data_len))
889 		return BLK_STS_RESOURCE;
890 
891 	cmd->cmd_len = 16;
892 	cmd->cmnd[0] = WRITE_SAME_16;
893 	if (unmap)
894 		cmd->cmnd[1] = 0x8; /* UNMAP */
895 	put_unaligned_be64(lba, &cmd->cmnd[2]);
896 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
897 
898 	cmd->allowed = sdkp->max_retries;
899 	cmd->transfersize = data_len;
900 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
901 
902 	return scsi_alloc_sgtables(cmd);
903 }
904 
905 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
906 		bool unmap)
907 {
908 	struct scsi_device *sdp = cmd->device;
909 	struct request *rq = scsi_cmd_to_rq(cmd);
910 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
911 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
912 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
913 	u32 data_len = sdp->sector_size;
914 
915 	if (!sd_set_special_bvec(rq, data_len))
916 		return BLK_STS_RESOURCE;
917 
918 	cmd->cmd_len = 10;
919 	cmd->cmnd[0] = WRITE_SAME;
920 	if (unmap)
921 		cmd->cmnd[1] = 0x8; /* UNMAP */
922 	put_unaligned_be32(lba, &cmd->cmnd[2]);
923 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
924 
925 	cmd->allowed = sdkp->max_retries;
926 	cmd->transfersize = data_len;
927 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
928 
929 	return scsi_alloc_sgtables(cmd);
930 }
931 
932 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
933 {
934 	struct request *rq = scsi_cmd_to_rq(cmd);
935 	struct scsi_device *sdp = cmd->device;
936 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
937 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
938 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
939 
940 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
941 		switch (sdkp->zeroing_mode) {
942 		case SD_ZERO_WS16_UNMAP:
943 			return sd_setup_write_same16_cmnd(cmd, true);
944 		case SD_ZERO_WS10_UNMAP:
945 			return sd_setup_write_same10_cmnd(cmd, true);
946 		}
947 	}
948 
949 	if (sdp->no_write_same) {
950 		rq->rq_flags |= RQF_QUIET;
951 		return BLK_STS_TARGET;
952 	}
953 
954 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
955 		return sd_setup_write_same16_cmnd(cmd, false);
956 
957 	return sd_setup_write_same10_cmnd(cmd, false);
958 }
959 
960 static void sd_config_write_same(struct scsi_disk *sdkp)
961 {
962 	struct request_queue *q = sdkp->disk->queue;
963 	unsigned int logical_block_size = sdkp->device->sector_size;
964 
965 	if (sdkp->device->no_write_same) {
966 		sdkp->max_ws_blocks = 0;
967 		goto out;
968 	}
969 
970 	/* Some devices can not handle block counts above 0xffff despite
971 	 * supporting WRITE SAME(16). Consequently we default to 64k
972 	 * blocks per I/O unless the device explicitly advertises a
973 	 * bigger limit.
974 	 */
975 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
976 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
977 						   (u32)SD_MAX_WS16_BLOCKS);
978 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
979 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
980 						   (u32)SD_MAX_WS10_BLOCKS);
981 	else {
982 		sdkp->device->no_write_same = 1;
983 		sdkp->max_ws_blocks = 0;
984 	}
985 
986 	if (sdkp->lbprz && sdkp->lbpws)
987 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
988 	else if (sdkp->lbprz && sdkp->lbpws10)
989 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
990 	else if (sdkp->max_ws_blocks)
991 		sdkp->zeroing_mode = SD_ZERO_WS;
992 	else
993 		sdkp->zeroing_mode = SD_ZERO_WRITE;
994 
995 	if (sdkp->max_ws_blocks &&
996 	    sdkp->physical_block_size > logical_block_size) {
997 		/*
998 		 * Reporting a maximum number of blocks that is not aligned
999 		 * on the device physical size would cause a large write same
1000 		 * request to be split into physically unaligned chunks by
1001 		 * __blkdev_issue_write_zeroes() even if the caller of this
1002 		 * functions took care to align the large request. So make sure
1003 		 * the maximum reported is aligned to the device physical block
1004 		 * size. This is only an optional optimization for regular
1005 		 * disks, but this is mandatory to avoid failure of large write
1006 		 * same requests directed at sequential write required zones of
1007 		 * host-managed ZBC disks.
1008 		 */
1009 		sdkp->max_ws_blocks =
1010 			round_down(sdkp->max_ws_blocks,
1011 				   bytes_to_logical(sdkp->device,
1012 						    sdkp->physical_block_size));
1013 	}
1014 
1015 out:
1016 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1017 					 (logical_block_size >> 9));
1018 }
1019 
1020 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1021 {
1022 	struct request *rq = scsi_cmd_to_rq(cmd);
1023 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1024 
1025 	/* flush requests don't perform I/O, zero the S/G table */
1026 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1027 
1028 	if (cmd->device->use_16_for_sync) {
1029 		cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1030 		cmd->cmd_len = 16;
1031 	} else {
1032 		cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1033 		cmd->cmd_len = 10;
1034 	}
1035 	cmd->transfersize = 0;
1036 	cmd->allowed = sdkp->max_retries;
1037 
1038 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1039 	return BLK_STS_OK;
1040 }
1041 
1042 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1043 				       sector_t lba, unsigned int nr_blocks,
1044 				       unsigned char flags)
1045 {
1046 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1047 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1048 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1049 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1050 	cmd->cmnd[10] = flags;
1051 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1052 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1053 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1054 
1055 	return BLK_STS_OK;
1056 }
1057 
1058 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1059 				       sector_t lba, unsigned int nr_blocks,
1060 				       unsigned char flags)
1061 {
1062 	cmd->cmd_len  = 16;
1063 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1064 	cmd->cmnd[1]  = flags;
1065 	cmd->cmnd[14] = 0;
1066 	cmd->cmnd[15] = 0;
1067 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1068 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1069 
1070 	return BLK_STS_OK;
1071 }
1072 
1073 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1074 				       sector_t lba, unsigned int nr_blocks,
1075 				       unsigned char flags)
1076 {
1077 	cmd->cmd_len = 10;
1078 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1079 	cmd->cmnd[1] = flags;
1080 	cmd->cmnd[6] = 0;
1081 	cmd->cmnd[9] = 0;
1082 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1083 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1084 
1085 	return BLK_STS_OK;
1086 }
1087 
1088 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1089 				      sector_t lba, unsigned int nr_blocks,
1090 				      unsigned char flags)
1091 {
1092 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1093 	if (WARN_ON_ONCE(nr_blocks == 0))
1094 		return BLK_STS_IOERR;
1095 
1096 	if (unlikely(flags & 0x8)) {
1097 		/*
1098 		 * This happens only if this drive failed 10byte rw
1099 		 * command with ILLEGAL_REQUEST during operation and
1100 		 * thus turned off use_10_for_rw.
1101 		 */
1102 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1103 		return BLK_STS_IOERR;
1104 	}
1105 
1106 	cmd->cmd_len = 6;
1107 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1108 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1109 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1110 	cmd->cmnd[3] = lba & 0xff;
1111 	cmd->cmnd[4] = nr_blocks;
1112 	cmd->cmnd[5] = 0;
1113 
1114 	return BLK_STS_OK;
1115 }
1116 
1117 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1118 {
1119 	struct request *rq = scsi_cmd_to_rq(cmd);
1120 	struct scsi_device *sdp = cmd->device;
1121 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1122 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1123 	sector_t threshold;
1124 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1125 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1126 	bool write = rq_data_dir(rq) == WRITE;
1127 	unsigned char protect, fua;
1128 	blk_status_t ret;
1129 	unsigned int dif;
1130 	bool dix;
1131 
1132 	ret = scsi_alloc_sgtables(cmd);
1133 	if (ret != BLK_STS_OK)
1134 		return ret;
1135 
1136 	ret = BLK_STS_IOERR;
1137 	if (!scsi_device_online(sdp) || sdp->changed) {
1138 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1139 		goto fail;
1140 	}
1141 
1142 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1143 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1144 		goto fail;
1145 	}
1146 
1147 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1148 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1149 		goto fail;
1150 	}
1151 
1152 	/*
1153 	 * Some SD card readers can't handle accesses which touch the
1154 	 * last one or two logical blocks. Split accesses as needed.
1155 	 */
1156 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1157 
1158 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1159 		if (lba < threshold) {
1160 			/* Access up to the threshold but not beyond */
1161 			nr_blocks = threshold - lba;
1162 		} else {
1163 			/* Access only a single logical block */
1164 			nr_blocks = 1;
1165 		}
1166 	}
1167 
1168 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1169 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1170 		if (ret)
1171 			goto fail;
1172 	}
1173 
1174 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1175 	dix = scsi_prot_sg_count(cmd);
1176 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1177 
1178 	if (dif || dix)
1179 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1180 	else
1181 		protect = 0;
1182 
1183 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1184 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1185 					 protect | fua);
1186 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1187 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1188 					 protect | fua);
1189 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1190 		   sdp->use_10_for_rw || protect) {
1191 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1192 					 protect | fua);
1193 	} else {
1194 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1195 					protect | fua);
1196 	}
1197 
1198 	if (unlikely(ret != BLK_STS_OK))
1199 		goto fail;
1200 
1201 	/*
1202 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1203 	 * host adapter, it's safe to assume that we can at least transfer
1204 	 * this many bytes between each connect / disconnect.
1205 	 */
1206 	cmd->transfersize = sdp->sector_size;
1207 	cmd->underflow = nr_blocks << 9;
1208 	cmd->allowed = sdkp->max_retries;
1209 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1210 
1211 	SCSI_LOG_HLQUEUE(1,
1212 			 scmd_printk(KERN_INFO, cmd,
1213 				     "%s: block=%llu, count=%d\n", __func__,
1214 				     (unsigned long long)blk_rq_pos(rq),
1215 				     blk_rq_sectors(rq)));
1216 	SCSI_LOG_HLQUEUE(2,
1217 			 scmd_printk(KERN_INFO, cmd,
1218 				     "%s %d/%u 512 byte blocks.\n",
1219 				     write ? "writing" : "reading", nr_blocks,
1220 				     blk_rq_sectors(rq)));
1221 
1222 	/*
1223 	 * This indicates that the command is ready from our end to be queued.
1224 	 */
1225 	return BLK_STS_OK;
1226 fail:
1227 	scsi_free_sgtables(cmd);
1228 	return ret;
1229 }
1230 
1231 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1232 {
1233 	struct request *rq = scsi_cmd_to_rq(cmd);
1234 
1235 	switch (req_op(rq)) {
1236 	case REQ_OP_DISCARD:
1237 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1238 		case SD_LBP_UNMAP:
1239 			return sd_setup_unmap_cmnd(cmd);
1240 		case SD_LBP_WS16:
1241 			return sd_setup_write_same16_cmnd(cmd, true);
1242 		case SD_LBP_WS10:
1243 			return sd_setup_write_same10_cmnd(cmd, true);
1244 		case SD_LBP_ZERO:
1245 			return sd_setup_write_same10_cmnd(cmd, false);
1246 		default:
1247 			return BLK_STS_TARGET;
1248 		}
1249 	case REQ_OP_WRITE_ZEROES:
1250 		return sd_setup_write_zeroes_cmnd(cmd);
1251 	case REQ_OP_FLUSH:
1252 		return sd_setup_flush_cmnd(cmd);
1253 	case REQ_OP_READ:
1254 	case REQ_OP_WRITE:
1255 	case REQ_OP_ZONE_APPEND:
1256 		return sd_setup_read_write_cmnd(cmd);
1257 	case REQ_OP_ZONE_RESET:
1258 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1259 						   false);
1260 	case REQ_OP_ZONE_RESET_ALL:
1261 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1262 						   true);
1263 	case REQ_OP_ZONE_OPEN:
1264 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1265 	case REQ_OP_ZONE_CLOSE:
1266 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1267 	case REQ_OP_ZONE_FINISH:
1268 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1269 	default:
1270 		WARN_ON_ONCE(1);
1271 		return BLK_STS_NOTSUPP;
1272 	}
1273 }
1274 
1275 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1276 {
1277 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1278 
1279 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1280 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1281 }
1282 
1283 static bool sd_need_revalidate(struct block_device *bdev,
1284 		struct scsi_disk *sdkp)
1285 {
1286 	if (sdkp->device->removable || sdkp->write_prot) {
1287 		if (bdev_check_media_change(bdev))
1288 			return true;
1289 	}
1290 
1291 	/*
1292 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1293 	 * nothing to do with partitions, BLKRRPART is used to force a full
1294 	 * revalidate after things like a format for historical reasons.
1295 	 */
1296 	return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1297 }
1298 
1299 /**
1300  *	sd_open - open a scsi disk device
1301  *	@bdev: Block device of the scsi disk to open
1302  *	@mode: FMODE_* mask
1303  *
1304  *	Returns 0 if successful. Returns a negated errno value in case
1305  *	of error.
1306  *
1307  *	Note: This can be called from a user context (e.g. fsck(1) )
1308  *	or from within the kernel (e.g. as a result of a mount(1) ).
1309  *	In the latter case @inode and @filp carry an abridged amount
1310  *	of information as noted above.
1311  *
1312  *	Locking: called with bdev->bd_disk->open_mutex held.
1313  **/
1314 static int sd_open(struct block_device *bdev, fmode_t mode)
1315 {
1316 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1317 	struct scsi_device *sdev = sdkp->device;
1318 	int retval;
1319 
1320 	if (scsi_device_get(sdev))
1321 		return -ENXIO;
1322 
1323 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1324 
1325 	/*
1326 	 * If the device is in error recovery, wait until it is done.
1327 	 * If the device is offline, then disallow any access to it.
1328 	 */
1329 	retval = -ENXIO;
1330 	if (!scsi_block_when_processing_errors(sdev))
1331 		goto error_out;
1332 
1333 	if (sd_need_revalidate(bdev, sdkp))
1334 		sd_revalidate_disk(bdev->bd_disk);
1335 
1336 	/*
1337 	 * If the drive is empty, just let the open fail.
1338 	 */
1339 	retval = -ENOMEDIUM;
1340 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1341 		goto error_out;
1342 
1343 	/*
1344 	 * If the device has the write protect tab set, have the open fail
1345 	 * if the user expects to be able to write to the thing.
1346 	 */
1347 	retval = -EROFS;
1348 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1349 		goto error_out;
1350 
1351 	/*
1352 	 * It is possible that the disk changing stuff resulted in
1353 	 * the device being taken offline.  If this is the case,
1354 	 * report this to the user, and don't pretend that the
1355 	 * open actually succeeded.
1356 	 */
1357 	retval = -ENXIO;
1358 	if (!scsi_device_online(sdev))
1359 		goto error_out;
1360 
1361 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1362 		if (scsi_block_when_processing_errors(sdev))
1363 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1364 	}
1365 
1366 	return 0;
1367 
1368 error_out:
1369 	scsi_device_put(sdev);
1370 	return retval;
1371 }
1372 
1373 /**
1374  *	sd_release - invoked when the (last) close(2) is called on this
1375  *	scsi disk.
1376  *	@disk: disk to release
1377  *	@mode: FMODE_* mask
1378  *
1379  *	Returns 0.
1380  *
1381  *	Note: may block (uninterruptible) if error recovery is underway
1382  *	on this disk.
1383  *
1384  *	Locking: called with bdev->bd_disk->open_mutex held.
1385  **/
1386 static void sd_release(struct gendisk *disk, fmode_t mode)
1387 {
1388 	struct scsi_disk *sdkp = scsi_disk(disk);
1389 	struct scsi_device *sdev = sdkp->device;
1390 
1391 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1392 
1393 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1394 		if (scsi_block_when_processing_errors(sdev))
1395 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1396 	}
1397 
1398 	scsi_device_put(sdev);
1399 }
1400 
1401 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1402 {
1403 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1404 	struct scsi_device *sdp = sdkp->device;
1405 	struct Scsi_Host *host = sdp->host;
1406 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1407 	int diskinfo[4];
1408 
1409 	/* default to most commonly used values */
1410 	diskinfo[0] = 0x40;	/* 1 << 6 */
1411 	diskinfo[1] = 0x20;	/* 1 << 5 */
1412 	diskinfo[2] = capacity >> 11;
1413 
1414 	/* override with calculated, extended default, or driver values */
1415 	if (host->hostt->bios_param)
1416 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1417 	else
1418 		scsicam_bios_param(bdev, capacity, diskinfo);
1419 
1420 	geo->heads = diskinfo[0];
1421 	geo->sectors = diskinfo[1];
1422 	geo->cylinders = diskinfo[2];
1423 	return 0;
1424 }
1425 
1426 /**
1427  *	sd_ioctl - process an ioctl
1428  *	@bdev: target block device
1429  *	@mode: FMODE_* mask
1430  *	@cmd: ioctl command number
1431  *	@arg: this is third argument given to ioctl(2) system call.
1432  *	Often contains a pointer.
1433  *
1434  *	Returns 0 if successful (some ioctls return positive numbers on
1435  *	success as well). Returns a negated errno value in case of error.
1436  *
1437  *	Note: most ioctls are forward onto the block subsystem or further
1438  *	down in the scsi subsystem.
1439  **/
1440 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1441 		    unsigned int cmd, unsigned long arg)
1442 {
1443 	struct gendisk *disk = bdev->bd_disk;
1444 	struct scsi_disk *sdkp = scsi_disk(disk);
1445 	struct scsi_device *sdp = sdkp->device;
1446 	void __user *p = (void __user *)arg;
1447 	int error;
1448 
1449 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1450 				    "cmd=0x%x\n", disk->disk_name, cmd));
1451 
1452 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1453 		return -ENOIOCTLCMD;
1454 
1455 	/*
1456 	 * If we are in the middle of error recovery, don't let anyone
1457 	 * else try and use this device.  Also, if error recovery fails, it
1458 	 * may try and take the device offline, in which case all further
1459 	 * access to the device is prohibited.
1460 	 */
1461 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1462 			(mode & FMODE_NDELAY) != 0);
1463 	if (error)
1464 		return error;
1465 
1466 	if (is_sed_ioctl(cmd))
1467 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1468 	return scsi_ioctl(sdp, mode, cmd, p);
1469 }
1470 
1471 static void set_media_not_present(struct scsi_disk *sdkp)
1472 {
1473 	if (sdkp->media_present)
1474 		sdkp->device->changed = 1;
1475 
1476 	if (sdkp->device->removable) {
1477 		sdkp->media_present = 0;
1478 		sdkp->capacity = 0;
1479 	}
1480 }
1481 
1482 static int media_not_present(struct scsi_disk *sdkp,
1483 			     struct scsi_sense_hdr *sshdr)
1484 {
1485 	if (!scsi_sense_valid(sshdr))
1486 		return 0;
1487 
1488 	/* not invoked for commands that could return deferred errors */
1489 	switch (sshdr->sense_key) {
1490 	case UNIT_ATTENTION:
1491 	case NOT_READY:
1492 		/* medium not present */
1493 		if (sshdr->asc == 0x3A) {
1494 			set_media_not_present(sdkp);
1495 			return 1;
1496 		}
1497 	}
1498 	return 0;
1499 }
1500 
1501 /**
1502  *	sd_check_events - check media events
1503  *	@disk: kernel device descriptor
1504  *	@clearing: disk events currently being cleared
1505  *
1506  *	Returns mask of DISK_EVENT_*.
1507  *
1508  *	Note: this function is invoked from the block subsystem.
1509  **/
1510 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1511 {
1512 	struct scsi_disk *sdkp = disk->private_data;
1513 	struct scsi_device *sdp;
1514 	int retval;
1515 	bool disk_changed;
1516 
1517 	if (!sdkp)
1518 		return 0;
1519 
1520 	sdp = sdkp->device;
1521 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1522 
1523 	/*
1524 	 * If the device is offline, don't send any commands - just pretend as
1525 	 * if the command failed.  If the device ever comes back online, we
1526 	 * can deal with it then.  It is only because of unrecoverable errors
1527 	 * that we would ever take a device offline in the first place.
1528 	 */
1529 	if (!scsi_device_online(sdp)) {
1530 		set_media_not_present(sdkp);
1531 		goto out;
1532 	}
1533 
1534 	/*
1535 	 * Using TEST_UNIT_READY enables differentiation between drive with
1536 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1537 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1538 	 *
1539 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1540 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1541 	 * sd_revalidate() is called.
1542 	 */
1543 	if (scsi_block_when_processing_errors(sdp)) {
1544 		struct scsi_sense_hdr sshdr = { 0, };
1545 
1546 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1547 					      &sshdr);
1548 
1549 		/* failed to execute TUR, assume media not present */
1550 		if (retval < 0 || host_byte(retval)) {
1551 			set_media_not_present(sdkp);
1552 			goto out;
1553 		}
1554 
1555 		if (media_not_present(sdkp, &sshdr))
1556 			goto out;
1557 	}
1558 
1559 	/*
1560 	 * For removable scsi disk we have to recognise the presence
1561 	 * of a disk in the drive.
1562 	 */
1563 	if (!sdkp->media_present)
1564 		sdp->changed = 1;
1565 	sdkp->media_present = 1;
1566 out:
1567 	/*
1568 	 * sdp->changed is set under the following conditions:
1569 	 *
1570 	 *	Medium present state has changed in either direction.
1571 	 *	Device has indicated UNIT_ATTENTION.
1572 	 */
1573 	disk_changed = sdp->changed;
1574 	sdp->changed = 0;
1575 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1576 }
1577 
1578 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1579 {
1580 	int retries, res;
1581 	struct scsi_device *sdp = sdkp->device;
1582 	const int timeout = sdp->request_queue->rq_timeout
1583 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1584 	struct scsi_sense_hdr my_sshdr;
1585 	const struct scsi_exec_args exec_args = {
1586 		.req_flags = BLK_MQ_REQ_PM,
1587 		/* caller might not be interested in sense, but we need it */
1588 		.sshdr = sshdr ? : &my_sshdr,
1589 	};
1590 
1591 	if (!scsi_device_online(sdp))
1592 		return -ENODEV;
1593 
1594 	sshdr = exec_args.sshdr;
1595 
1596 	for (retries = 3; retries > 0; --retries) {
1597 		unsigned char cmd[16] = { 0 };
1598 
1599 		if (sdp->use_16_for_sync)
1600 			cmd[0] = SYNCHRONIZE_CACHE_16;
1601 		else
1602 			cmd[0] = SYNCHRONIZE_CACHE;
1603 		/*
1604 		 * Leave the rest of the command zero to indicate
1605 		 * flush everything.
1606 		 */
1607 		res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1608 				       timeout, sdkp->max_retries, &exec_args);
1609 		if (res == 0)
1610 			break;
1611 	}
1612 
1613 	if (res) {
1614 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1615 
1616 		if (res < 0)
1617 			return res;
1618 
1619 		if (scsi_status_is_check_condition(res) &&
1620 		    scsi_sense_valid(sshdr)) {
1621 			sd_print_sense_hdr(sdkp, sshdr);
1622 
1623 			/* we need to evaluate the error return  */
1624 			if (sshdr->asc == 0x3a ||	/* medium not present */
1625 			    sshdr->asc == 0x20 ||	/* invalid command */
1626 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1627 				/* this is no error here */
1628 				return 0;
1629 		}
1630 
1631 		switch (host_byte(res)) {
1632 		/* ignore errors due to racing a disconnection */
1633 		case DID_BAD_TARGET:
1634 		case DID_NO_CONNECT:
1635 			return 0;
1636 		/* signal the upper layer it might try again */
1637 		case DID_BUS_BUSY:
1638 		case DID_IMM_RETRY:
1639 		case DID_REQUEUE:
1640 		case DID_SOFT_ERROR:
1641 			return -EBUSY;
1642 		default:
1643 			return -EIO;
1644 		}
1645 	}
1646 	return 0;
1647 }
1648 
1649 static void sd_rescan(struct device *dev)
1650 {
1651 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1652 
1653 	sd_revalidate_disk(sdkp->disk);
1654 }
1655 
1656 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1657 		enum blk_unique_id type)
1658 {
1659 	struct scsi_device *sdev = scsi_disk(disk)->device;
1660 	const struct scsi_vpd *vpd;
1661 	const unsigned char *d;
1662 	int ret = -ENXIO, len;
1663 
1664 	rcu_read_lock();
1665 	vpd = rcu_dereference(sdev->vpd_pg83);
1666 	if (!vpd)
1667 		goto out_unlock;
1668 
1669 	ret = -EINVAL;
1670 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1671 		/* we only care about designators with LU association */
1672 		if (((d[1] >> 4) & 0x3) != 0x00)
1673 			continue;
1674 		if ((d[1] & 0xf) != type)
1675 			continue;
1676 
1677 		/*
1678 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1679 		 * keep looking as one with more entropy might still show up.
1680 		 */
1681 		len = d[3];
1682 		if (len != 8 && len != 12 && len != 16)
1683 			continue;
1684 		ret = len;
1685 		memcpy(id, d + 4, len);
1686 		if (len == 16)
1687 			break;
1688 	}
1689 out_unlock:
1690 	rcu_read_unlock();
1691 	return ret;
1692 }
1693 
1694 static char sd_pr_type(enum pr_type type)
1695 {
1696 	switch (type) {
1697 	case PR_WRITE_EXCLUSIVE:
1698 		return 0x01;
1699 	case PR_EXCLUSIVE_ACCESS:
1700 		return 0x03;
1701 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1702 		return 0x05;
1703 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1704 		return 0x06;
1705 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1706 		return 0x07;
1707 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1708 		return 0x08;
1709 	default:
1710 		return 0;
1711 	}
1712 };
1713 
1714 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1715 {
1716 	switch (host_byte(result)) {
1717 	case DID_TRANSPORT_MARGINAL:
1718 	case DID_TRANSPORT_DISRUPTED:
1719 	case DID_BUS_BUSY:
1720 		return PR_STS_RETRY_PATH_FAILURE;
1721 	case DID_NO_CONNECT:
1722 		return PR_STS_PATH_FAILED;
1723 	case DID_TRANSPORT_FAILFAST:
1724 		return PR_STS_PATH_FAST_FAILED;
1725 	}
1726 
1727 	switch (status_byte(result)) {
1728 	case SAM_STAT_RESERVATION_CONFLICT:
1729 		return PR_STS_RESERVATION_CONFLICT;
1730 	case SAM_STAT_CHECK_CONDITION:
1731 		if (!scsi_sense_valid(sshdr))
1732 			return PR_STS_IOERR;
1733 
1734 		if (sshdr->sense_key == ILLEGAL_REQUEST &&
1735 		    (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1736 			return -EINVAL;
1737 
1738 		fallthrough;
1739 	default:
1740 		return PR_STS_IOERR;
1741 	}
1742 }
1743 
1744 static int sd_pr_command(struct block_device *bdev, u8 sa,
1745 		u64 key, u64 sa_key, u8 type, u8 flags)
1746 {
1747 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1748 	struct scsi_device *sdev = sdkp->device;
1749 	struct scsi_sense_hdr sshdr;
1750 	const struct scsi_exec_args exec_args = {
1751 		.sshdr = &sshdr,
1752 	};
1753 	int result;
1754 	u8 cmd[16] = { 0, };
1755 	u8 data[24] = { 0, };
1756 
1757 	cmd[0] = PERSISTENT_RESERVE_OUT;
1758 	cmd[1] = sa;
1759 	cmd[2] = type;
1760 	put_unaligned_be32(sizeof(data), &cmd[5]);
1761 
1762 	put_unaligned_be64(key, &data[0]);
1763 	put_unaligned_be64(sa_key, &data[8]);
1764 	data[20] = flags;
1765 
1766 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1767 				  sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1768 				  &exec_args);
1769 
1770 	if (scsi_status_is_check_condition(result) &&
1771 	    scsi_sense_valid(&sshdr)) {
1772 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1773 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1774 	}
1775 
1776 	if (result <= 0)
1777 		return result;
1778 
1779 	return sd_scsi_to_pr_err(&sshdr, result);
1780 }
1781 
1782 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1783 		u32 flags)
1784 {
1785 	if (flags & ~PR_FL_IGNORE_KEY)
1786 		return -EOPNOTSUPP;
1787 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1788 			old_key, new_key, 0,
1789 			(1 << 0) /* APTPL */);
1790 }
1791 
1792 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1793 		u32 flags)
1794 {
1795 	if (flags)
1796 		return -EOPNOTSUPP;
1797 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1798 }
1799 
1800 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1801 {
1802 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1803 }
1804 
1805 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1806 		enum pr_type type, bool abort)
1807 {
1808 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1809 			     sd_pr_type(type), 0);
1810 }
1811 
1812 static int sd_pr_clear(struct block_device *bdev, u64 key)
1813 {
1814 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1815 }
1816 
1817 static const struct pr_ops sd_pr_ops = {
1818 	.pr_register	= sd_pr_register,
1819 	.pr_reserve	= sd_pr_reserve,
1820 	.pr_release	= sd_pr_release,
1821 	.pr_preempt	= sd_pr_preempt,
1822 	.pr_clear	= sd_pr_clear,
1823 };
1824 
1825 static void scsi_disk_free_disk(struct gendisk *disk)
1826 {
1827 	struct scsi_disk *sdkp = scsi_disk(disk);
1828 
1829 	put_device(&sdkp->disk_dev);
1830 }
1831 
1832 static const struct block_device_operations sd_fops = {
1833 	.owner			= THIS_MODULE,
1834 	.open			= sd_open,
1835 	.release		= sd_release,
1836 	.ioctl			= sd_ioctl,
1837 	.getgeo			= sd_getgeo,
1838 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1839 	.check_events		= sd_check_events,
1840 	.unlock_native_capacity	= sd_unlock_native_capacity,
1841 	.report_zones		= sd_zbc_report_zones,
1842 	.get_unique_id		= sd_get_unique_id,
1843 	.free_disk		= scsi_disk_free_disk,
1844 	.pr_ops			= &sd_pr_ops,
1845 };
1846 
1847 /**
1848  *	sd_eh_reset - reset error handling callback
1849  *	@scmd:		sd-issued command that has failed
1850  *
1851  *	This function is called by the SCSI midlayer before starting
1852  *	SCSI EH. When counting medium access failures we have to be
1853  *	careful to register it only only once per device and SCSI EH run;
1854  *	there might be several timed out commands which will cause the
1855  *	'max_medium_access_timeouts' counter to trigger after the first
1856  *	SCSI EH run already and set the device to offline.
1857  *	So this function resets the internal counter before starting SCSI EH.
1858  **/
1859 static void sd_eh_reset(struct scsi_cmnd *scmd)
1860 {
1861 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1862 
1863 	/* New SCSI EH run, reset gate variable */
1864 	sdkp->ignore_medium_access_errors = false;
1865 }
1866 
1867 /**
1868  *	sd_eh_action - error handling callback
1869  *	@scmd:		sd-issued command that has failed
1870  *	@eh_disp:	The recovery disposition suggested by the midlayer
1871  *
1872  *	This function is called by the SCSI midlayer upon completion of an
1873  *	error test command (currently TEST UNIT READY). The result of sending
1874  *	the eh command is passed in eh_disp.  We're looking for devices that
1875  *	fail medium access commands but are OK with non access commands like
1876  *	test unit ready (so wrongly see the device as having a successful
1877  *	recovery)
1878  **/
1879 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1880 {
1881 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1882 	struct scsi_device *sdev = scmd->device;
1883 
1884 	if (!scsi_device_online(sdev) ||
1885 	    !scsi_medium_access_command(scmd) ||
1886 	    host_byte(scmd->result) != DID_TIME_OUT ||
1887 	    eh_disp != SUCCESS)
1888 		return eh_disp;
1889 
1890 	/*
1891 	 * The device has timed out executing a medium access command.
1892 	 * However, the TEST UNIT READY command sent during error
1893 	 * handling completed successfully. Either the device is in the
1894 	 * process of recovering or has it suffered an internal failure
1895 	 * that prevents access to the storage medium.
1896 	 */
1897 	if (!sdkp->ignore_medium_access_errors) {
1898 		sdkp->medium_access_timed_out++;
1899 		sdkp->ignore_medium_access_errors = true;
1900 	}
1901 
1902 	/*
1903 	 * If the device keeps failing read/write commands but TEST UNIT
1904 	 * READY always completes successfully we assume that medium
1905 	 * access is no longer possible and take the device offline.
1906 	 */
1907 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1908 		scmd_printk(KERN_ERR, scmd,
1909 			    "Medium access timeout failure. Offlining disk!\n");
1910 		mutex_lock(&sdev->state_mutex);
1911 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1912 		mutex_unlock(&sdev->state_mutex);
1913 
1914 		return SUCCESS;
1915 	}
1916 
1917 	return eh_disp;
1918 }
1919 
1920 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1921 {
1922 	struct request *req = scsi_cmd_to_rq(scmd);
1923 	struct scsi_device *sdev = scmd->device;
1924 	unsigned int transferred, good_bytes;
1925 	u64 start_lba, end_lba, bad_lba;
1926 
1927 	/*
1928 	 * Some commands have a payload smaller than the device logical
1929 	 * block size (e.g. INQUIRY on a 4K disk).
1930 	 */
1931 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1932 		return 0;
1933 
1934 	/* Check if we have a 'bad_lba' information */
1935 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1936 				     SCSI_SENSE_BUFFERSIZE,
1937 				     &bad_lba))
1938 		return 0;
1939 
1940 	/*
1941 	 * If the bad lba was reported incorrectly, we have no idea where
1942 	 * the error is.
1943 	 */
1944 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1945 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1946 	if (bad_lba < start_lba || bad_lba >= end_lba)
1947 		return 0;
1948 
1949 	/*
1950 	 * resid is optional but mostly filled in.  When it's unused,
1951 	 * its value is zero, so we assume the whole buffer transferred
1952 	 */
1953 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1954 
1955 	/* This computation should always be done in terms of the
1956 	 * resolution of the device's medium.
1957 	 */
1958 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1959 
1960 	return min(good_bytes, transferred);
1961 }
1962 
1963 /**
1964  *	sd_done - bottom half handler: called when the lower level
1965  *	driver has completed (successfully or otherwise) a scsi command.
1966  *	@SCpnt: mid-level's per command structure.
1967  *
1968  *	Note: potentially run from within an ISR. Must not block.
1969  **/
1970 static int sd_done(struct scsi_cmnd *SCpnt)
1971 {
1972 	int result = SCpnt->result;
1973 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1974 	unsigned int sector_size = SCpnt->device->sector_size;
1975 	unsigned int resid;
1976 	struct scsi_sense_hdr sshdr;
1977 	struct request *req = scsi_cmd_to_rq(SCpnt);
1978 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
1979 	int sense_valid = 0;
1980 	int sense_deferred = 0;
1981 
1982 	switch (req_op(req)) {
1983 	case REQ_OP_DISCARD:
1984 	case REQ_OP_WRITE_ZEROES:
1985 	case REQ_OP_ZONE_RESET:
1986 	case REQ_OP_ZONE_RESET_ALL:
1987 	case REQ_OP_ZONE_OPEN:
1988 	case REQ_OP_ZONE_CLOSE:
1989 	case REQ_OP_ZONE_FINISH:
1990 		if (!result) {
1991 			good_bytes = blk_rq_bytes(req);
1992 			scsi_set_resid(SCpnt, 0);
1993 		} else {
1994 			good_bytes = 0;
1995 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1996 		}
1997 		break;
1998 	default:
1999 		/*
2000 		 * In case of bogus fw or device, we could end up having
2001 		 * an unaligned partial completion. Check this here and force
2002 		 * alignment.
2003 		 */
2004 		resid = scsi_get_resid(SCpnt);
2005 		if (resid & (sector_size - 1)) {
2006 			sd_printk(KERN_INFO, sdkp,
2007 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2008 				resid, sector_size);
2009 			scsi_print_command(SCpnt);
2010 			resid = min(scsi_bufflen(SCpnt),
2011 				    round_up(resid, sector_size));
2012 			scsi_set_resid(SCpnt, resid);
2013 		}
2014 	}
2015 
2016 	if (result) {
2017 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2018 		if (sense_valid)
2019 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2020 	}
2021 	sdkp->medium_access_timed_out = 0;
2022 
2023 	if (!scsi_status_is_check_condition(result) &&
2024 	    (!sense_valid || sense_deferred))
2025 		goto out;
2026 
2027 	switch (sshdr.sense_key) {
2028 	case HARDWARE_ERROR:
2029 	case MEDIUM_ERROR:
2030 		good_bytes = sd_completed_bytes(SCpnt);
2031 		break;
2032 	case RECOVERED_ERROR:
2033 		good_bytes = scsi_bufflen(SCpnt);
2034 		break;
2035 	case NO_SENSE:
2036 		/* This indicates a false check condition, so ignore it.  An
2037 		 * unknown amount of data was transferred so treat it as an
2038 		 * error.
2039 		 */
2040 		SCpnt->result = 0;
2041 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2042 		break;
2043 	case ABORTED_COMMAND:
2044 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2045 			good_bytes = sd_completed_bytes(SCpnt);
2046 		break;
2047 	case ILLEGAL_REQUEST:
2048 		switch (sshdr.asc) {
2049 		case 0x10:	/* DIX: Host detected corruption */
2050 			good_bytes = sd_completed_bytes(SCpnt);
2051 			break;
2052 		case 0x20:	/* INVALID COMMAND OPCODE */
2053 		case 0x24:	/* INVALID FIELD IN CDB */
2054 			switch (SCpnt->cmnd[0]) {
2055 			case UNMAP:
2056 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2057 				break;
2058 			case WRITE_SAME_16:
2059 			case WRITE_SAME:
2060 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2061 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2062 				} else {
2063 					sdkp->device->no_write_same = 1;
2064 					sd_config_write_same(sdkp);
2065 					req->rq_flags |= RQF_QUIET;
2066 				}
2067 				break;
2068 			}
2069 		}
2070 		break;
2071 	default:
2072 		break;
2073 	}
2074 
2075  out:
2076 	if (sd_is_zoned(sdkp))
2077 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2078 
2079 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2080 					   "sd_done: completed %d of %d bytes\n",
2081 					   good_bytes, scsi_bufflen(SCpnt)));
2082 
2083 	return good_bytes;
2084 }
2085 
2086 /*
2087  * spinup disk - called only in sd_revalidate_disk()
2088  */
2089 static void
2090 sd_spinup_disk(struct scsi_disk *sdkp)
2091 {
2092 	unsigned char cmd[10];
2093 	unsigned long spintime_expire = 0;
2094 	int retries, spintime;
2095 	unsigned int the_result;
2096 	struct scsi_sense_hdr sshdr;
2097 	const struct scsi_exec_args exec_args = {
2098 		.sshdr = &sshdr,
2099 	};
2100 	int sense_valid = 0;
2101 
2102 	spintime = 0;
2103 
2104 	/* Spin up drives, as required.  Only do this at boot time */
2105 	/* Spinup needs to be done for module loads too. */
2106 	do {
2107 		retries = 0;
2108 
2109 		do {
2110 			bool media_was_present = sdkp->media_present;
2111 
2112 			cmd[0] = TEST_UNIT_READY;
2113 			memset((void *) &cmd[1], 0, 9);
2114 
2115 			the_result = scsi_execute_cmd(sdkp->device, cmd,
2116 						      REQ_OP_DRV_IN, NULL, 0,
2117 						      SD_TIMEOUT,
2118 						      sdkp->max_retries,
2119 						      &exec_args);
2120 
2121 			/*
2122 			 * If the drive has indicated to us that it
2123 			 * doesn't have any media in it, don't bother
2124 			 * with any more polling.
2125 			 */
2126 			if (media_not_present(sdkp, &sshdr)) {
2127 				if (media_was_present)
2128 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2129 				return;
2130 			}
2131 
2132 			if (the_result)
2133 				sense_valid = scsi_sense_valid(&sshdr);
2134 			retries++;
2135 		} while (retries < 3 &&
2136 			 (!scsi_status_is_good(the_result) ||
2137 			  (scsi_status_is_check_condition(the_result) &&
2138 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2139 
2140 		if (!scsi_status_is_check_condition(the_result)) {
2141 			/* no sense, TUR either succeeded or failed
2142 			 * with a status error */
2143 			if(!spintime && !scsi_status_is_good(the_result)) {
2144 				sd_print_result(sdkp, "Test Unit Ready failed",
2145 						the_result);
2146 			}
2147 			break;
2148 		}
2149 
2150 		/*
2151 		 * The device does not want the automatic start to be issued.
2152 		 */
2153 		if (sdkp->device->no_start_on_add)
2154 			break;
2155 
2156 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2157 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2158 				break;	/* manual intervention required */
2159 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2160 				break;	/* standby */
2161 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2162 				break;	/* unavailable */
2163 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2164 				break;	/* sanitize in progress */
2165 			/*
2166 			 * Issue command to spin up drive when not ready
2167 			 */
2168 			if (!spintime) {
2169 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2170 				cmd[0] = START_STOP;
2171 				cmd[1] = 1;	/* Return immediately */
2172 				memset((void *) &cmd[2], 0, 8);
2173 				cmd[4] = 1;	/* Start spin cycle */
2174 				if (sdkp->device->start_stop_pwr_cond)
2175 					cmd[4] |= 1 << 4;
2176 				scsi_execute_cmd(sdkp->device, cmd,
2177 						 REQ_OP_DRV_IN, NULL, 0,
2178 						 SD_TIMEOUT, sdkp->max_retries,
2179 						 &exec_args);
2180 				spintime_expire = jiffies + 100 * HZ;
2181 				spintime = 1;
2182 			}
2183 			/* Wait 1 second for next try */
2184 			msleep(1000);
2185 			printk(KERN_CONT ".");
2186 
2187 		/*
2188 		 * Wait for USB flash devices with slow firmware.
2189 		 * Yes, this sense key/ASC combination shouldn't
2190 		 * occur here.  It's characteristic of these devices.
2191 		 */
2192 		} else if (sense_valid &&
2193 				sshdr.sense_key == UNIT_ATTENTION &&
2194 				sshdr.asc == 0x28) {
2195 			if (!spintime) {
2196 				spintime_expire = jiffies + 5 * HZ;
2197 				spintime = 1;
2198 			}
2199 			/* Wait 1 second for next try */
2200 			msleep(1000);
2201 		} else {
2202 			/* we don't understand the sense code, so it's
2203 			 * probably pointless to loop */
2204 			if(!spintime) {
2205 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2206 				sd_print_sense_hdr(sdkp, &sshdr);
2207 			}
2208 			break;
2209 		}
2210 
2211 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2212 
2213 	if (spintime) {
2214 		if (scsi_status_is_good(the_result))
2215 			printk(KERN_CONT "ready\n");
2216 		else
2217 			printk(KERN_CONT "not responding...\n");
2218 	}
2219 }
2220 
2221 /*
2222  * Determine whether disk supports Data Integrity Field.
2223  */
2224 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2225 {
2226 	struct scsi_device *sdp = sdkp->device;
2227 	u8 type;
2228 
2229 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2230 		sdkp->protection_type = 0;
2231 		return 0;
2232 	}
2233 
2234 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2235 
2236 	if (type > T10_PI_TYPE3_PROTECTION) {
2237 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported"	\
2238 			  " protection type %u. Disabling disk!\n",
2239 			  type);
2240 		sdkp->protection_type = 0;
2241 		return -ENODEV;
2242 	}
2243 
2244 	sdkp->protection_type = type;
2245 
2246 	return 0;
2247 }
2248 
2249 static void sd_config_protection(struct scsi_disk *sdkp)
2250 {
2251 	struct scsi_device *sdp = sdkp->device;
2252 
2253 	sd_dif_config_host(sdkp);
2254 
2255 	if (!sdkp->protection_type)
2256 		return;
2257 
2258 	if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2259 		sd_first_printk(KERN_NOTICE, sdkp,
2260 				"Disabling DIF Type %u protection\n",
2261 				sdkp->protection_type);
2262 		sdkp->protection_type = 0;
2263 	}
2264 
2265 	sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2266 			sdkp->protection_type);
2267 }
2268 
2269 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2270 			struct scsi_sense_hdr *sshdr, int sense_valid,
2271 			int the_result)
2272 {
2273 	if (sense_valid)
2274 		sd_print_sense_hdr(sdkp, sshdr);
2275 	else
2276 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2277 
2278 	/*
2279 	 * Set dirty bit for removable devices if not ready -
2280 	 * sometimes drives will not report this properly.
2281 	 */
2282 	if (sdp->removable &&
2283 	    sense_valid && sshdr->sense_key == NOT_READY)
2284 		set_media_not_present(sdkp);
2285 
2286 	/*
2287 	 * We used to set media_present to 0 here to indicate no media
2288 	 * in the drive, but some drives fail read capacity even with
2289 	 * media present, so we can't do that.
2290 	 */
2291 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2292 }
2293 
2294 #define RC16_LEN 32
2295 #if RC16_LEN > SD_BUF_SIZE
2296 #error RC16_LEN must not be more than SD_BUF_SIZE
2297 #endif
2298 
2299 #define READ_CAPACITY_RETRIES_ON_RESET	10
2300 
2301 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2302 						unsigned char *buffer)
2303 {
2304 	unsigned char cmd[16];
2305 	struct scsi_sense_hdr sshdr;
2306 	const struct scsi_exec_args exec_args = {
2307 		.sshdr = &sshdr,
2308 	};
2309 	int sense_valid = 0;
2310 	int the_result;
2311 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2312 	unsigned int alignment;
2313 	unsigned long long lba;
2314 	unsigned sector_size;
2315 
2316 	if (sdp->no_read_capacity_16)
2317 		return -EINVAL;
2318 
2319 	do {
2320 		memset(cmd, 0, 16);
2321 		cmd[0] = SERVICE_ACTION_IN_16;
2322 		cmd[1] = SAI_READ_CAPACITY_16;
2323 		cmd[13] = RC16_LEN;
2324 		memset(buffer, 0, RC16_LEN);
2325 
2326 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2327 					      buffer, RC16_LEN, SD_TIMEOUT,
2328 					      sdkp->max_retries, &exec_args);
2329 
2330 		if (media_not_present(sdkp, &sshdr))
2331 			return -ENODEV;
2332 
2333 		if (the_result > 0) {
2334 			sense_valid = scsi_sense_valid(&sshdr);
2335 			if (sense_valid &&
2336 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2337 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2338 			    sshdr.ascq == 0x00)
2339 				/* Invalid Command Operation Code or
2340 				 * Invalid Field in CDB, just retry
2341 				 * silently with RC10 */
2342 				return -EINVAL;
2343 			if (sense_valid &&
2344 			    sshdr.sense_key == UNIT_ATTENTION &&
2345 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2346 				/* Device reset might occur several times,
2347 				 * give it one more chance */
2348 				if (--reset_retries > 0)
2349 					continue;
2350 		}
2351 		retries--;
2352 
2353 	} while (the_result && retries);
2354 
2355 	if (the_result) {
2356 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2357 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2358 		return -EINVAL;
2359 	}
2360 
2361 	sector_size = get_unaligned_be32(&buffer[8]);
2362 	lba = get_unaligned_be64(&buffer[0]);
2363 
2364 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2365 		sdkp->capacity = 0;
2366 		return -ENODEV;
2367 	}
2368 
2369 	/* Logical blocks per physical block exponent */
2370 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2371 
2372 	/* RC basis */
2373 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2374 
2375 	/* Lowest aligned logical block */
2376 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2377 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2378 	if (alignment && sdkp->first_scan)
2379 		sd_printk(KERN_NOTICE, sdkp,
2380 			  "physical block alignment offset: %u\n", alignment);
2381 
2382 	if (buffer[14] & 0x80) { /* LBPME */
2383 		sdkp->lbpme = 1;
2384 
2385 		if (buffer[14] & 0x40) /* LBPRZ */
2386 			sdkp->lbprz = 1;
2387 
2388 		sd_config_discard(sdkp, SD_LBP_WS16);
2389 	}
2390 
2391 	sdkp->capacity = lba + 1;
2392 	return sector_size;
2393 }
2394 
2395 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2396 						unsigned char *buffer)
2397 {
2398 	unsigned char cmd[16];
2399 	struct scsi_sense_hdr sshdr;
2400 	const struct scsi_exec_args exec_args = {
2401 		.sshdr = &sshdr,
2402 	};
2403 	int sense_valid = 0;
2404 	int the_result;
2405 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2406 	sector_t lba;
2407 	unsigned sector_size;
2408 
2409 	do {
2410 		cmd[0] = READ_CAPACITY;
2411 		memset(&cmd[1], 0, 9);
2412 		memset(buffer, 0, 8);
2413 
2414 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2415 					      8, SD_TIMEOUT, sdkp->max_retries,
2416 					      &exec_args);
2417 
2418 		if (media_not_present(sdkp, &sshdr))
2419 			return -ENODEV;
2420 
2421 		if (the_result > 0) {
2422 			sense_valid = scsi_sense_valid(&sshdr);
2423 			if (sense_valid &&
2424 			    sshdr.sense_key == UNIT_ATTENTION &&
2425 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2426 				/* Device reset might occur several times,
2427 				 * give it one more chance */
2428 				if (--reset_retries > 0)
2429 					continue;
2430 		}
2431 		retries--;
2432 
2433 	} while (the_result && retries);
2434 
2435 	if (the_result) {
2436 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2437 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2438 		return -EINVAL;
2439 	}
2440 
2441 	sector_size = get_unaligned_be32(&buffer[4]);
2442 	lba = get_unaligned_be32(&buffer[0]);
2443 
2444 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2445 		/* Some buggy (usb cardreader) devices return an lba of
2446 		   0xffffffff when the want to report a size of 0 (with
2447 		   which they really mean no media is present) */
2448 		sdkp->capacity = 0;
2449 		sdkp->physical_block_size = sector_size;
2450 		return sector_size;
2451 	}
2452 
2453 	sdkp->capacity = lba + 1;
2454 	sdkp->physical_block_size = sector_size;
2455 	return sector_size;
2456 }
2457 
2458 static int sd_try_rc16_first(struct scsi_device *sdp)
2459 {
2460 	if (sdp->host->max_cmd_len < 16)
2461 		return 0;
2462 	if (sdp->try_rc_10_first)
2463 		return 0;
2464 	if (sdp->scsi_level > SCSI_SPC_2)
2465 		return 1;
2466 	if (scsi_device_protection(sdp))
2467 		return 1;
2468 	return 0;
2469 }
2470 
2471 /*
2472  * read disk capacity
2473  */
2474 static void
2475 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2476 {
2477 	int sector_size;
2478 	struct scsi_device *sdp = sdkp->device;
2479 
2480 	if (sd_try_rc16_first(sdp)) {
2481 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2482 		if (sector_size == -EOVERFLOW)
2483 			goto got_data;
2484 		if (sector_size == -ENODEV)
2485 			return;
2486 		if (sector_size < 0)
2487 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2488 		if (sector_size < 0)
2489 			return;
2490 	} else {
2491 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2492 		if (sector_size == -EOVERFLOW)
2493 			goto got_data;
2494 		if (sector_size < 0)
2495 			return;
2496 		if ((sizeof(sdkp->capacity) > 4) &&
2497 		    (sdkp->capacity > 0xffffffffULL)) {
2498 			int old_sector_size = sector_size;
2499 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2500 					"Trying to use READ CAPACITY(16).\n");
2501 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2502 			if (sector_size < 0) {
2503 				sd_printk(KERN_NOTICE, sdkp,
2504 					"Using 0xffffffff as device size\n");
2505 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2506 				sector_size = old_sector_size;
2507 				goto got_data;
2508 			}
2509 			/* Remember that READ CAPACITY(16) succeeded */
2510 			sdp->try_rc_10_first = 0;
2511 		}
2512 	}
2513 
2514 	/* Some devices are known to return the total number of blocks,
2515 	 * not the highest block number.  Some devices have versions
2516 	 * which do this and others which do not.  Some devices we might
2517 	 * suspect of doing this but we don't know for certain.
2518 	 *
2519 	 * If we know the reported capacity is wrong, decrement it.  If
2520 	 * we can only guess, then assume the number of blocks is even
2521 	 * (usually true but not always) and err on the side of lowering
2522 	 * the capacity.
2523 	 */
2524 	if (sdp->fix_capacity ||
2525 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2526 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2527 				"from its reported value: %llu\n",
2528 				(unsigned long long) sdkp->capacity);
2529 		--sdkp->capacity;
2530 	}
2531 
2532 got_data:
2533 	if (sector_size == 0) {
2534 		sector_size = 512;
2535 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2536 			  "assuming 512.\n");
2537 	}
2538 
2539 	if (sector_size != 512 &&
2540 	    sector_size != 1024 &&
2541 	    sector_size != 2048 &&
2542 	    sector_size != 4096) {
2543 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2544 			  sector_size);
2545 		/*
2546 		 * The user might want to re-format the drive with
2547 		 * a supported sectorsize.  Once this happens, it
2548 		 * would be relatively trivial to set the thing up.
2549 		 * For this reason, we leave the thing in the table.
2550 		 */
2551 		sdkp->capacity = 0;
2552 		/*
2553 		 * set a bogus sector size so the normal read/write
2554 		 * logic in the block layer will eventually refuse any
2555 		 * request on this device without tripping over power
2556 		 * of two sector size assumptions
2557 		 */
2558 		sector_size = 512;
2559 	}
2560 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2561 	blk_queue_physical_block_size(sdp->request_queue,
2562 				      sdkp->physical_block_size);
2563 	sdkp->device->sector_size = sector_size;
2564 
2565 	if (sdkp->capacity > 0xffffffff)
2566 		sdp->use_16_for_rw = 1;
2567 
2568 }
2569 
2570 /*
2571  * Print disk capacity
2572  */
2573 static void
2574 sd_print_capacity(struct scsi_disk *sdkp,
2575 		  sector_t old_capacity)
2576 {
2577 	int sector_size = sdkp->device->sector_size;
2578 	char cap_str_2[10], cap_str_10[10];
2579 
2580 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2581 		return;
2582 
2583 	string_get_size(sdkp->capacity, sector_size,
2584 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2585 	string_get_size(sdkp->capacity, sector_size,
2586 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2587 
2588 	sd_printk(KERN_NOTICE, sdkp,
2589 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2590 		  (unsigned long long)sdkp->capacity,
2591 		  sector_size, cap_str_10, cap_str_2);
2592 
2593 	if (sdkp->physical_block_size != sector_size)
2594 		sd_printk(KERN_NOTICE, sdkp,
2595 			  "%u-byte physical blocks\n",
2596 			  sdkp->physical_block_size);
2597 }
2598 
2599 /* called with buffer of length 512 */
2600 static inline int
2601 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2602 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2603 		 struct scsi_sense_hdr *sshdr)
2604 {
2605 	/*
2606 	 * If we must use MODE SENSE(10), make sure that the buffer length
2607 	 * is at least 8 bytes so that the mode sense header fits.
2608 	 */
2609 	if (sdkp->device->use_10_for_ms && len < 8)
2610 		len = 8;
2611 
2612 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2613 			       SD_TIMEOUT, sdkp->max_retries, data,
2614 			       sshdr);
2615 }
2616 
2617 /*
2618  * read write protect setting, if possible - called only in sd_revalidate_disk()
2619  * called with buffer of length SD_BUF_SIZE
2620  */
2621 static void
2622 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2623 {
2624 	int res;
2625 	struct scsi_device *sdp = sdkp->device;
2626 	struct scsi_mode_data data;
2627 	int old_wp = sdkp->write_prot;
2628 
2629 	set_disk_ro(sdkp->disk, 0);
2630 	if (sdp->skip_ms_page_3f) {
2631 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2632 		return;
2633 	}
2634 
2635 	if (sdp->use_192_bytes_for_3f) {
2636 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2637 	} else {
2638 		/*
2639 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2640 		 * We have to start carefully: some devices hang if we ask
2641 		 * for more than is available.
2642 		 */
2643 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2644 
2645 		/*
2646 		 * Second attempt: ask for page 0 When only page 0 is
2647 		 * implemented, a request for page 3F may return Sense Key
2648 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2649 		 * CDB.
2650 		 */
2651 		if (res < 0)
2652 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2653 
2654 		/*
2655 		 * Third attempt: ask 255 bytes, as we did earlier.
2656 		 */
2657 		if (res < 0)
2658 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2659 					       &data, NULL);
2660 	}
2661 
2662 	if (res < 0) {
2663 		sd_first_printk(KERN_WARNING, sdkp,
2664 			  "Test WP failed, assume Write Enabled\n");
2665 	} else {
2666 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2667 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2668 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2669 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2670 				  sdkp->write_prot ? "on" : "off");
2671 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2672 		}
2673 	}
2674 }
2675 
2676 /*
2677  * sd_read_cache_type - called only from sd_revalidate_disk()
2678  * called with buffer of length SD_BUF_SIZE
2679  */
2680 static void
2681 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2682 {
2683 	int len = 0, res;
2684 	struct scsi_device *sdp = sdkp->device;
2685 
2686 	int dbd;
2687 	int modepage;
2688 	int first_len;
2689 	struct scsi_mode_data data;
2690 	struct scsi_sense_hdr sshdr;
2691 	int old_wce = sdkp->WCE;
2692 	int old_rcd = sdkp->RCD;
2693 	int old_dpofua = sdkp->DPOFUA;
2694 
2695 
2696 	if (sdkp->cache_override)
2697 		return;
2698 
2699 	first_len = 4;
2700 	if (sdp->skip_ms_page_8) {
2701 		if (sdp->type == TYPE_RBC)
2702 			goto defaults;
2703 		else {
2704 			if (sdp->skip_ms_page_3f)
2705 				goto defaults;
2706 			modepage = 0x3F;
2707 			if (sdp->use_192_bytes_for_3f)
2708 				first_len = 192;
2709 			dbd = 0;
2710 		}
2711 	} else if (sdp->type == TYPE_RBC) {
2712 		modepage = 6;
2713 		dbd = 8;
2714 	} else {
2715 		modepage = 8;
2716 		dbd = 0;
2717 	}
2718 
2719 	/* cautiously ask */
2720 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2721 			&data, &sshdr);
2722 
2723 	if (res < 0)
2724 		goto bad_sense;
2725 
2726 	if (!data.header_length) {
2727 		modepage = 6;
2728 		first_len = 0;
2729 		sd_first_printk(KERN_ERR, sdkp,
2730 				"Missing header in MODE_SENSE response\n");
2731 	}
2732 
2733 	/* that went OK, now ask for the proper length */
2734 	len = data.length;
2735 
2736 	/*
2737 	 * We're only interested in the first three bytes, actually.
2738 	 * But the data cache page is defined for the first 20.
2739 	 */
2740 	if (len < 3)
2741 		goto bad_sense;
2742 	else if (len > SD_BUF_SIZE) {
2743 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2744 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2745 		len = SD_BUF_SIZE;
2746 	}
2747 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2748 		len = 192;
2749 
2750 	/* Get the data */
2751 	if (len > first_len)
2752 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2753 				&data, &sshdr);
2754 
2755 	if (!res) {
2756 		int offset = data.header_length + data.block_descriptor_length;
2757 
2758 		while (offset < len) {
2759 			u8 page_code = buffer[offset] & 0x3F;
2760 			u8 spf       = buffer[offset] & 0x40;
2761 
2762 			if (page_code == 8 || page_code == 6) {
2763 				/* We're interested only in the first 3 bytes.
2764 				 */
2765 				if (len - offset <= 2) {
2766 					sd_first_printk(KERN_ERR, sdkp,
2767 						"Incomplete mode parameter "
2768 							"data\n");
2769 					goto defaults;
2770 				} else {
2771 					modepage = page_code;
2772 					goto Page_found;
2773 				}
2774 			} else {
2775 				/* Go to the next page */
2776 				if (spf && len - offset > 3)
2777 					offset += 4 + (buffer[offset+2] << 8) +
2778 						buffer[offset+3];
2779 				else if (!spf && len - offset > 1)
2780 					offset += 2 + buffer[offset+1];
2781 				else {
2782 					sd_first_printk(KERN_ERR, sdkp,
2783 							"Incomplete mode "
2784 							"parameter data\n");
2785 					goto defaults;
2786 				}
2787 			}
2788 		}
2789 
2790 		sd_first_printk(KERN_WARNING, sdkp,
2791 				"No Caching mode page found\n");
2792 		goto defaults;
2793 
2794 	Page_found:
2795 		if (modepage == 8) {
2796 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2797 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2798 		} else {
2799 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2800 			sdkp->RCD = 0;
2801 		}
2802 
2803 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2804 		if (sdp->broken_fua) {
2805 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2806 			sdkp->DPOFUA = 0;
2807 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2808 			   !sdkp->device->use_16_for_rw) {
2809 			sd_first_printk(KERN_NOTICE, sdkp,
2810 				  "Uses READ/WRITE(6), disabling FUA\n");
2811 			sdkp->DPOFUA = 0;
2812 		}
2813 
2814 		/* No cache flush allowed for write protected devices */
2815 		if (sdkp->WCE && sdkp->write_prot)
2816 			sdkp->WCE = 0;
2817 
2818 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2819 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2820 			sd_printk(KERN_NOTICE, sdkp,
2821 				  "Write cache: %s, read cache: %s, %s\n",
2822 				  sdkp->WCE ? "enabled" : "disabled",
2823 				  sdkp->RCD ? "disabled" : "enabled",
2824 				  sdkp->DPOFUA ? "supports DPO and FUA"
2825 				  : "doesn't support DPO or FUA");
2826 
2827 		return;
2828 	}
2829 
2830 bad_sense:
2831 	if (scsi_sense_valid(&sshdr) &&
2832 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2833 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2834 		/* Invalid field in CDB */
2835 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2836 	else
2837 		sd_first_printk(KERN_ERR, sdkp,
2838 				"Asking for cache data failed\n");
2839 
2840 defaults:
2841 	if (sdp->wce_default_on) {
2842 		sd_first_printk(KERN_NOTICE, sdkp,
2843 				"Assuming drive cache: write back\n");
2844 		sdkp->WCE = 1;
2845 	} else {
2846 		sd_first_printk(KERN_WARNING, sdkp,
2847 				"Assuming drive cache: write through\n");
2848 		sdkp->WCE = 0;
2849 	}
2850 	sdkp->RCD = 0;
2851 	sdkp->DPOFUA = 0;
2852 }
2853 
2854 /*
2855  * The ATO bit indicates whether the DIF application tag is available
2856  * for use by the operating system.
2857  */
2858 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2859 {
2860 	int res, offset;
2861 	struct scsi_device *sdp = sdkp->device;
2862 	struct scsi_mode_data data;
2863 	struct scsi_sense_hdr sshdr;
2864 
2865 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2866 		return;
2867 
2868 	if (sdkp->protection_type == 0)
2869 		return;
2870 
2871 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2872 			      sdkp->max_retries, &data, &sshdr);
2873 
2874 	if (res < 0 || !data.header_length ||
2875 	    data.length < 6) {
2876 		sd_first_printk(KERN_WARNING, sdkp,
2877 			  "getting Control mode page failed, assume no ATO\n");
2878 
2879 		if (scsi_sense_valid(&sshdr))
2880 			sd_print_sense_hdr(sdkp, &sshdr);
2881 
2882 		return;
2883 	}
2884 
2885 	offset = data.header_length + data.block_descriptor_length;
2886 
2887 	if ((buffer[offset] & 0x3f) != 0x0a) {
2888 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2889 		return;
2890 	}
2891 
2892 	if ((buffer[offset + 5] & 0x80) == 0)
2893 		return;
2894 
2895 	sdkp->ATO = 1;
2896 
2897 	return;
2898 }
2899 
2900 /**
2901  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2902  * @sdkp: disk to query
2903  */
2904 static void sd_read_block_limits(struct scsi_disk *sdkp)
2905 {
2906 	struct scsi_vpd *vpd;
2907 
2908 	rcu_read_lock();
2909 
2910 	vpd = rcu_dereference(sdkp->device->vpd_pgb0);
2911 	if (!vpd || vpd->len < 16)
2912 		goto out;
2913 
2914 	sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
2915 	sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
2916 	sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
2917 
2918 	if (vpd->len >= 64) {
2919 		unsigned int lba_count, desc_count;
2920 
2921 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
2922 
2923 		if (!sdkp->lbpme)
2924 			goto out;
2925 
2926 		lba_count = get_unaligned_be32(&vpd->data[20]);
2927 		desc_count = get_unaligned_be32(&vpd->data[24]);
2928 
2929 		if (lba_count && desc_count)
2930 			sdkp->max_unmap_blocks = lba_count;
2931 
2932 		sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
2933 
2934 		if (vpd->data[32] & 0x80)
2935 			sdkp->unmap_alignment =
2936 				get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
2937 
2938 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2939 
2940 			if (sdkp->max_unmap_blocks)
2941 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2942 			else
2943 				sd_config_discard(sdkp, SD_LBP_WS16);
2944 
2945 		} else {	/* LBP VPD page tells us what to use */
2946 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2947 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2948 			else if (sdkp->lbpws)
2949 				sd_config_discard(sdkp, SD_LBP_WS16);
2950 			else if (sdkp->lbpws10)
2951 				sd_config_discard(sdkp, SD_LBP_WS10);
2952 			else
2953 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2954 		}
2955 	}
2956 
2957  out:
2958 	rcu_read_unlock();
2959 }
2960 
2961 /**
2962  * sd_read_block_characteristics - Query block dev. characteristics
2963  * @sdkp: disk to query
2964  */
2965 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2966 {
2967 	struct request_queue *q = sdkp->disk->queue;
2968 	struct scsi_vpd *vpd;
2969 	u16 rot;
2970 	u8 zoned;
2971 
2972 	rcu_read_lock();
2973 	vpd = rcu_dereference(sdkp->device->vpd_pgb1);
2974 
2975 	if (!vpd || vpd->len < 8) {
2976 		rcu_read_unlock();
2977 	        return;
2978 	}
2979 
2980 	rot = get_unaligned_be16(&vpd->data[4]);
2981 	zoned = (vpd->data[8] >> 4) & 3;
2982 	rcu_read_unlock();
2983 
2984 	if (rot == 1) {
2985 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2986 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2987 	}
2988 
2989 	if (sdkp->device->type == TYPE_ZBC) {
2990 		/*
2991 		 * Host-managed: Per ZBC and ZAC specifications, writes in
2992 		 * sequential write required zones of host-managed devices must
2993 		 * be aligned to the device physical block size.
2994 		 */
2995 		disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
2996 		blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
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_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3855 	if (!sd_page_pool) {
3856 		printk(KERN_ERR "sd: can't init discard page pool\n");
3857 		err = -ENOMEM;
3858 		goto err_out_class;
3859 	}
3860 
3861 	err = scsi_register_driver(&sd_template.gendrv);
3862 	if (err)
3863 		goto err_out_driver;
3864 
3865 	return 0;
3866 
3867 err_out_driver:
3868 	mempool_destroy(sd_page_pool);
3869 err_out_class:
3870 	class_unregister(&sd_disk_class);
3871 err_out:
3872 	for (i = 0; i < SD_MAJORS; i++)
3873 		unregister_blkdev(sd_major(i), "sd");
3874 	return err;
3875 }
3876 
3877 /**
3878  *	exit_sd - exit point for this driver (when it is a module).
3879  *
3880  *	Note: this function unregisters this driver from the scsi mid-level.
3881  **/
3882 static void __exit exit_sd(void)
3883 {
3884 	int i;
3885 
3886 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3887 
3888 	scsi_unregister_driver(&sd_template.gendrv);
3889 	mempool_destroy(sd_page_pool);
3890 
3891 	class_unregister(&sd_disk_class);
3892 
3893 	for (i = 0; i < SD_MAJORS; i++)
3894 		unregister_blkdev(sd_major(i), "sd");
3895 }
3896 
3897 module_init(init_sd);
3898 module_exit(exit_sd);
3899 
3900 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3901 {
3902 	scsi_print_sense_hdr(sdkp->device,
3903 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3904 }
3905 
3906 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3907 {
3908 	const char *hb_string = scsi_hostbyte_string(result);
3909 
3910 	if (hb_string)
3911 		sd_printk(KERN_INFO, sdkp,
3912 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3913 			  hb_string ? hb_string : "invalid",
3914 			  "DRIVER_OK");
3915 	else
3916 		sd_printk(KERN_INFO, sdkp,
3917 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3918 			  msg, host_byte(result), "DRIVER_OK");
3919 }
3920