xref: /openbmc/linux/drivers/scsi/sd.c (revision 67559900)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *      sd.c Copyright (C) 1992 Drew Eckhardt
4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5  *
6  *      Linux scsi disk driver
7  *              Initial versions: Drew Eckhardt
8  *              Subsequent revisions: Eric Youngdale
9  *	Modification history:
10  *       - Drew Eckhardt <drew@colorado.edu> original
11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12  *         outstanding request, and other enhancements.
13  *         Support loadable low-level scsi drivers.
14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15  *         eight major numbers.
16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18  *	   sd_init and cleanups.
19  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
20  *	   not being read in sd_open. Fix problem where removable media
21  *	   could be ejected after sd_open.
22  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25  *	   Support 32k/1M disks.
26  *
27  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
28  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
32  *	Note: when the logging level is set by the user, it must be greater
33  *	than the level indicated above to trigger output.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/major.h>
52 #include <linux/mutex.h>
53 #include <linux/string_helpers.h>
54 #include <linux/async.h>
55 #include <linux/slab.h>
56 #include <linux/sed-opal.h>
57 #include <linux/pm_runtime.h>
58 #include <linux/pr.h>
59 #include <linux/t10-pi.h>
60 #include <linux/uaccess.h>
61 #include <asm/unaligned.h>
62 
63 #include <scsi/scsi.h>
64 #include <scsi/scsi_cmnd.h>
65 #include <scsi/scsi_dbg.h>
66 #include <scsi/scsi_device.h>
67 #include <scsi/scsi_driver.h>
68 #include <scsi/scsi_eh.h>
69 #include <scsi/scsi_host.h>
70 #include <scsi/scsi_ioctl.h>
71 #include <scsi/scsicam.h>
72 
73 #include "sd.h"
74 #include "scsi_priv.h"
75 #include "scsi_logging.h"
76 
77 MODULE_AUTHOR("Eric Youngdale");
78 MODULE_DESCRIPTION("SCSI disk (sd) driver");
79 MODULE_LICENSE("GPL");
80 
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
96 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
100 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
101 
102 #define SD_MINORS	16
103 
104 static void sd_config_discard(struct scsi_disk *, unsigned int);
105 static void sd_config_write_same(struct scsi_disk *);
106 static int  sd_revalidate_disk(struct gendisk *);
107 static void sd_unlock_native_capacity(struct gendisk *disk);
108 static int  sd_probe(struct device *);
109 static int  sd_remove(struct device *);
110 static void sd_shutdown(struct device *);
111 static int sd_suspend_system(struct device *);
112 static int sd_suspend_runtime(struct device *);
113 static int sd_resume_system(struct device *);
114 static int sd_resume_runtime(struct device *);
115 static void sd_rescan(struct device *);
116 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 
124 static DEFINE_IDA(sd_index_ida);
125 
126 /* This semaphore is used to mediate the 0->1 reference get in the
127  * face of object destruction (i.e. we can't allow a get on an
128  * object after last put) */
129 static DEFINE_MUTEX(sd_ref_mutex);
130 
131 static struct kmem_cache *sd_cdb_cache;
132 static mempool_t *sd_cdb_pool;
133 static mempool_t *sd_page_pool;
134 static struct lock_class_key sd_bio_compl_lkclass;
135 
136 static const char *sd_cache_types[] = {
137 	"write through", "none", "write back",
138 	"write back, no read (daft)"
139 };
140 
141 static void sd_set_flush_flag(struct scsi_disk *sdkp)
142 {
143 	bool wc = false, fua = false;
144 
145 	if (sdkp->WCE) {
146 		wc = true;
147 		if (sdkp->DPOFUA)
148 			fua = true;
149 	}
150 
151 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
152 }
153 
154 static ssize_t
155 cache_type_store(struct device *dev, struct device_attribute *attr,
156 		 const char *buf, size_t count)
157 {
158 	int ct, rcd, wce, sp;
159 	struct scsi_disk *sdkp = to_scsi_disk(dev);
160 	struct scsi_device *sdp = sdkp->device;
161 	char buffer[64];
162 	char *buffer_data;
163 	struct scsi_mode_data data;
164 	struct scsi_sense_hdr sshdr;
165 	static const char temp[] = "temporary ";
166 	int len;
167 
168 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
169 		/* no cache control on RBC devices; theoretically they
170 		 * can do it, but there's probably so many exceptions
171 		 * it's not worth the risk */
172 		return -EINVAL;
173 
174 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
175 		buf += sizeof(temp) - 1;
176 		sdkp->cache_override = 1;
177 	} else {
178 		sdkp->cache_override = 0;
179 	}
180 
181 	ct = sysfs_match_string(sd_cache_types, buf);
182 	if (ct < 0)
183 		return -EINVAL;
184 
185 	rcd = ct & 0x01 ? 1 : 0;
186 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
187 
188 	if (sdkp->cache_override) {
189 		sdkp->WCE = wce;
190 		sdkp->RCD = rcd;
191 		sd_set_flush_flag(sdkp);
192 		return count;
193 	}
194 
195 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
196 			    sdkp->max_retries, &data, NULL))
197 		return -EINVAL;
198 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
199 		  data.block_descriptor_length);
200 	buffer_data = buffer + data.header_length +
201 		data.block_descriptor_length;
202 	buffer_data[2] &= ~0x05;
203 	buffer_data[2] |= wce << 2 | rcd;
204 	sp = buffer_data[0] & 0x80 ? 1 : 0;
205 	buffer_data[0] &= ~0x80;
206 
207 	/*
208 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
209 	 * received mode parameter buffer before doing MODE SELECT.
210 	 */
211 	data.device_specific = 0;
212 
213 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
214 			     sdkp->max_retries, &data, &sshdr)) {
215 		if (scsi_sense_valid(&sshdr))
216 			sd_print_sense_hdr(sdkp, &sshdr);
217 		return -EINVAL;
218 	}
219 	sd_revalidate_disk(sdkp->disk);
220 	return count;
221 }
222 
223 static ssize_t
224 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
225 		       char *buf)
226 {
227 	struct scsi_disk *sdkp = to_scsi_disk(dev);
228 	struct scsi_device *sdp = sdkp->device;
229 
230 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
231 }
232 
233 static ssize_t
234 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
235 			const char *buf, size_t count)
236 {
237 	struct scsi_disk *sdkp = to_scsi_disk(dev);
238 	struct scsi_device *sdp = sdkp->device;
239 	bool v;
240 
241 	if (!capable(CAP_SYS_ADMIN))
242 		return -EACCES;
243 
244 	if (kstrtobool(buf, &v))
245 		return -EINVAL;
246 
247 	sdp->manage_start_stop = v;
248 
249 	return count;
250 }
251 static DEVICE_ATTR_RW(manage_start_stop);
252 
253 static ssize_t
254 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
255 {
256 	struct scsi_disk *sdkp = to_scsi_disk(dev);
257 
258 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
259 }
260 
261 static ssize_t
262 allow_restart_store(struct device *dev, struct device_attribute *attr,
263 		    const char *buf, size_t count)
264 {
265 	bool v;
266 	struct scsi_disk *sdkp = to_scsi_disk(dev);
267 	struct scsi_device *sdp = sdkp->device;
268 
269 	if (!capable(CAP_SYS_ADMIN))
270 		return -EACCES;
271 
272 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
273 		return -EINVAL;
274 
275 	if (kstrtobool(buf, &v))
276 		return -EINVAL;
277 
278 	sdp->allow_restart = v;
279 
280 	return count;
281 }
282 static DEVICE_ATTR_RW(allow_restart);
283 
284 static ssize_t
285 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
286 {
287 	struct scsi_disk *sdkp = to_scsi_disk(dev);
288 	int ct = sdkp->RCD + 2*sdkp->WCE;
289 
290 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
291 }
292 static DEVICE_ATTR_RW(cache_type);
293 
294 static ssize_t
295 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
296 {
297 	struct scsi_disk *sdkp = to_scsi_disk(dev);
298 
299 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
300 }
301 static DEVICE_ATTR_RO(FUA);
302 
303 static ssize_t
304 protection_type_show(struct device *dev, struct device_attribute *attr,
305 		     char *buf)
306 {
307 	struct scsi_disk *sdkp = to_scsi_disk(dev);
308 
309 	return sprintf(buf, "%u\n", sdkp->protection_type);
310 }
311 
312 static ssize_t
313 protection_type_store(struct device *dev, struct device_attribute *attr,
314 		      const char *buf, size_t count)
315 {
316 	struct scsi_disk *sdkp = to_scsi_disk(dev);
317 	unsigned int val;
318 	int err;
319 
320 	if (!capable(CAP_SYS_ADMIN))
321 		return -EACCES;
322 
323 	err = kstrtouint(buf, 10, &val);
324 
325 	if (err)
326 		return err;
327 
328 	if (val <= T10_PI_TYPE3_PROTECTION)
329 		sdkp->protection_type = val;
330 
331 	return count;
332 }
333 static DEVICE_ATTR_RW(protection_type);
334 
335 static ssize_t
336 protection_mode_show(struct device *dev, struct device_attribute *attr,
337 		     char *buf)
338 {
339 	struct scsi_disk *sdkp = to_scsi_disk(dev);
340 	struct scsi_device *sdp = sdkp->device;
341 	unsigned int dif, dix;
342 
343 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
344 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
345 
346 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
347 		dif = 0;
348 		dix = 1;
349 	}
350 
351 	if (!dif && !dix)
352 		return sprintf(buf, "none\n");
353 
354 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
355 }
356 static DEVICE_ATTR_RO(protection_mode);
357 
358 static ssize_t
359 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
360 {
361 	struct scsi_disk *sdkp = to_scsi_disk(dev);
362 
363 	return sprintf(buf, "%u\n", sdkp->ATO);
364 }
365 static DEVICE_ATTR_RO(app_tag_own);
366 
367 static ssize_t
368 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
369 		       char *buf)
370 {
371 	struct scsi_disk *sdkp = to_scsi_disk(dev);
372 
373 	return sprintf(buf, "%u\n", sdkp->lbpme);
374 }
375 static DEVICE_ATTR_RO(thin_provisioning);
376 
377 /* sysfs_match_string() requires dense arrays */
378 static const char *lbp_mode[] = {
379 	[SD_LBP_FULL]		= "full",
380 	[SD_LBP_UNMAP]		= "unmap",
381 	[SD_LBP_WS16]		= "writesame_16",
382 	[SD_LBP_WS10]		= "writesame_10",
383 	[SD_LBP_ZERO]		= "writesame_zero",
384 	[SD_LBP_DISABLE]	= "disabled",
385 };
386 
387 static ssize_t
388 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
389 		       char *buf)
390 {
391 	struct scsi_disk *sdkp = to_scsi_disk(dev);
392 
393 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
394 }
395 
396 static ssize_t
397 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
398 			const char *buf, size_t count)
399 {
400 	struct scsi_disk *sdkp = to_scsi_disk(dev);
401 	struct scsi_device *sdp = sdkp->device;
402 	int mode;
403 
404 	if (!capable(CAP_SYS_ADMIN))
405 		return -EACCES;
406 
407 	if (sd_is_zoned(sdkp)) {
408 		sd_config_discard(sdkp, SD_LBP_DISABLE);
409 		return count;
410 	}
411 
412 	if (sdp->type != TYPE_DISK)
413 		return -EINVAL;
414 
415 	mode = sysfs_match_string(lbp_mode, buf);
416 	if (mode < 0)
417 		return -EINVAL;
418 
419 	sd_config_discard(sdkp, mode);
420 
421 	return count;
422 }
423 static DEVICE_ATTR_RW(provisioning_mode);
424 
425 /* sysfs_match_string() requires dense arrays */
426 static const char *zeroing_mode[] = {
427 	[SD_ZERO_WRITE]		= "write",
428 	[SD_ZERO_WS]		= "writesame",
429 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
430 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
431 };
432 
433 static ssize_t
434 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
435 		  char *buf)
436 {
437 	struct scsi_disk *sdkp = to_scsi_disk(dev);
438 
439 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
440 }
441 
442 static ssize_t
443 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
444 		   const char *buf, size_t count)
445 {
446 	struct scsi_disk *sdkp = to_scsi_disk(dev);
447 	int mode;
448 
449 	if (!capable(CAP_SYS_ADMIN))
450 		return -EACCES;
451 
452 	mode = sysfs_match_string(zeroing_mode, buf);
453 	if (mode < 0)
454 		return -EINVAL;
455 
456 	sdkp->zeroing_mode = mode;
457 
458 	return count;
459 }
460 static DEVICE_ATTR_RW(zeroing_mode);
461 
462 static ssize_t
463 max_medium_access_timeouts_show(struct device *dev,
464 				struct device_attribute *attr, char *buf)
465 {
466 	struct scsi_disk *sdkp = to_scsi_disk(dev);
467 
468 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
469 }
470 
471 static ssize_t
472 max_medium_access_timeouts_store(struct device *dev,
473 				 struct device_attribute *attr, const char *buf,
474 				 size_t count)
475 {
476 	struct scsi_disk *sdkp = to_scsi_disk(dev);
477 	int err;
478 
479 	if (!capable(CAP_SYS_ADMIN))
480 		return -EACCES;
481 
482 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
483 
484 	return err ? err : count;
485 }
486 static DEVICE_ATTR_RW(max_medium_access_timeouts);
487 
488 static ssize_t
489 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
490 			   char *buf)
491 {
492 	struct scsi_disk *sdkp = to_scsi_disk(dev);
493 
494 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
495 }
496 
497 static ssize_t
498 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
499 			    const char *buf, size_t count)
500 {
501 	struct scsi_disk *sdkp = to_scsi_disk(dev);
502 	struct scsi_device *sdp = sdkp->device;
503 	unsigned long max;
504 	int err;
505 
506 	if (!capable(CAP_SYS_ADMIN))
507 		return -EACCES;
508 
509 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
510 		return -EINVAL;
511 
512 	err = kstrtoul(buf, 10, &max);
513 
514 	if (err)
515 		return err;
516 
517 	if (max == 0)
518 		sdp->no_write_same = 1;
519 	else if (max <= SD_MAX_WS16_BLOCKS) {
520 		sdp->no_write_same = 0;
521 		sdkp->max_ws_blocks = max;
522 	}
523 
524 	sd_config_write_same(sdkp);
525 
526 	return count;
527 }
528 static DEVICE_ATTR_RW(max_write_same_blocks);
529 
530 static ssize_t
531 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
532 {
533 	struct scsi_disk *sdkp = to_scsi_disk(dev);
534 
535 	if (sdkp->device->type == TYPE_ZBC)
536 		return sprintf(buf, "host-managed\n");
537 	if (sdkp->zoned == 1)
538 		return sprintf(buf, "host-aware\n");
539 	if (sdkp->zoned == 2)
540 		return sprintf(buf, "drive-managed\n");
541 	return sprintf(buf, "none\n");
542 }
543 static DEVICE_ATTR_RO(zoned_cap);
544 
545 static ssize_t
546 max_retries_store(struct device *dev, struct device_attribute *attr,
547 		  const char *buf, size_t count)
548 {
549 	struct scsi_disk *sdkp = to_scsi_disk(dev);
550 	struct scsi_device *sdev = sdkp->device;
551 	int retries, err;
552 
553 	err = kstrtoint(buf, 10, &retries);
554 	if (err)
555 		return err;
556 
557 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
558 		sdkp->max_retries = retries;
559 		return count;
560 	}
561 
562 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
563 		    SD_MAX_RETRIES);
564 	return -EINVAL;
565 }
566 
567 static ssize_t
568 max_retries_show(struct device *dev, struct device_attribute *attr,
569 		 char *buf)
570 {
571 	struct scsi_disk *sdkp = to_scsi_disk(dev);
572 
573 	return sprintf(buf, "%d\n", sdkp->max_retries);
574 }
575 
576 static DEVICE_ATTR_RW(max_retries);
577 
578 static struct attribute *sd_disk_attrs[] = {
579 	&dev_attr_cache_type.attr,
580 	&dev_attr_FUA.attr,
581 	&dev_attr_allow_restart.attr,
582 	&dev_attr_manage_start_stop.attr,
583 	&dev_attr_protection_type.attr,
584 	&dev_attr_protection_mode.attr,
585 	&dev_attr_app_tag_own.attr,
586 	&dev_attr_thin_provisioning.attr,
587 	&dev_attr_provisioning_mode.attr,
588 	&dev_attr_zeroing_mode.attr,
589 	&dev_attr_max_write_same_blocks.attr,
590 	&dev_attr_max_medium_access_timeouts.attr,
591 	&dev_attr_zoned_cap.attr,
592 	&dev_attr_max_retries.attr,
593 	NULL,
594 };
595 ATTRIBUTE_GROUPS(sd_disk);
596 
597 static struct class sd_disk_class = {
598 	.name		= "scsi_disk",
599 	.owner		= THIS_MODULE,
600 	.dev_release	= scsi_disk_release,
601 	.dev_groups	= sd_disk_groups,
602 };
603 
604 static const struct dev_pm_ops sd_pm_ops = {
605 	.suspend		= sd_suspend_system,
606 	.resume			= sd_resume_system,
607 	.poweroff		= sd_suspend_system,
608 	.restore		= sd_resume_system,
609 	.runtime_suspend	= sd_suspend_runtime,
610 	.runtime_resume		= sd_resume_runtime,
611 };
612 
613 static struct scsi_driver sd_template = {
614 	.gendrv = {
615 		.name		= "sd",
616 		.owner		= THIS_MODULE,
617 		.probe		= sd_probe,
618 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
619 		.remove		= sd_remove,
620 		.shutdown	= sd_shutdown,
621 		.pm		= &sd_pm_ops,
622 	},
623 	.rescan			= sd_rescan,
624 	.init_command		= sd_init_command,
625 	.uninit_command		= sd_uninit_command,
626 	.done			= sd_done,
627 	.eh_action		= sd_eh_action,
628 	.eh_reset		= sd_eh_reset,
629 };
630 
631 /*
632  * Don't request a new module, as that could deadlock in multipath
633  * environment.
634  */
635 static void sd_default_probe(dev_t devt)
636 {
637 }
638 
639 /*
640  * Device no to disk mapping:
641  *
642  *       major         disc2     disc  p1
643  *   |............|.............|....|....| <- dev_t
644  *    31        20 19          8 7  4 3  0
645  *
646  * Inside a major, we have 16k disks, however mapped non-
647  * contiguously. The first 16 disks are for major0, the next
648  * ones with major1, ... Disk 256 is for major0 again, disk 272
649  * for major1, ...
650  * As we stay compatible with our numbering scheme, we can reuse
651  * the well-know SCSI majors 8, 65--71, 136--143.
652  */
653 static int sd_major(int major_idx)
654 {
655 	switch (major_idx) {
656 	case 0:
657 		return SCSI_DISK0_MAJOR;
658 	case 1 ... 7:
659 		return SCSI_DISK1_MAJOR + major_idx - 1;
660 	case 8 ... 15:
661 		return SCSI_DISK8_MAJOR + major_idx - 8;
662 	default:
663 		BUG();
664 		return 0;	/* shut up gcc */
665 	}
666 }
667 
668 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
669 {
670 	struct scsi_disk *sdkp = NULL;
671 
672 	mutex_lock(&sd_ref_mutex);
673 
674 	if (disk->private_data) {
675 		sdkp = scsi_disk(disk);
676 		if (scsi_device_get(sdkp->device) == 0)
677 			get_device(&sdkp->dev);
678 		else
679 			sdkp = NULL;
680 	}
681 	mutex_unlock(&sd_ref_mutex);
682 	return sdkp;
683 }
684 
685 static void scsi_disk_put(struct scsi_disk *sdkp)
686 {
687 	struct scsi_device *sdev = sdkp->device;
688 
689 	mutex_lock(&sd_ref_mutex);
690 	put_device(&sdkp->dev);
691 	scsi_device_put(sdev);
692 	mutex_unlock(&sd_ref_mutex);
693 }
694 
695 #ifdef CONFIG_BLK_SED_OPAL
696 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
697 		size_t len, bool send)
698 {
699 	struct scsi_disk *sdkp = data;
700 	struct scsi_device *sdev = sdkp->device;
701 	u8 cdb[12] = { 0, };
702 	int ret;
703 
704 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
705 	cdb[1] = secp;
706 	put_unaligned_be16(spsp, &cdb[2]);
707 	put_unaligned_be32(len, &cdb[6]);
708 
709 	ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
710 		buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
711 		RQF_PM, NULL);
712 	return ret <= 0 ? ret : -EIO;
713 }
714 #endif /* CONFIG_BLK_SED_OPAL */
715 
716 /*
717  * Look up the DIX operation based on whether the command is read or
718  * write and whether dix and dif are enabled.
719  */
720 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
721 {
722 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
723 	static const unsigned int ops[] = {	/* wrt dix dif */
724 		SCSI_PROT_NORMAL,		/*  0	0   0  */
725 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
726 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
727 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
728 		SCSI_PROT_NORMAL,		/*  1	0   0  */
729 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
730 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
731 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
732 	};
733 
734 	return ops[write << 2 | dix << 1 | dif];
735 }
736 
737 /*
738  * Returns a mask of the protection flags that are valid for a given DIX
739  * operation.
740  */
741 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
742 {
743 	static const unsigned int flag_mask[] = {
744 		[SCSI_PROT_NORMAL]		= 0,
745 
746 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
747 						  SCSI_PROT_GUARD_CHECK |
748 						  SCSI_PROT_REF_CHECK |
749 						  SCSI_PROT_REF_INCREMENT,
750 
751 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
752 						  SCSI_PROT_IP_CHECKSUM,
753 
754 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
755 						  SCSI_PROT_GUARD_CHECK |
756 						  SCSI_PROT_REF_CHECK |
757 						  SCSI_PROT_REF_INCREMENT |
758 						  SCSI_PROT_IP_CHECKSUM,
759 
760 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
761 						  SCSI_PROT_REF_INCREMENT,
762 
763 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
764 						  SCSI_PROT_REF_CHECK |
765 						  SCSI_PROT_REF_INCREMENT |
766 						  SCSI_PROT_IP_CHECKSUM,
767 
768 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
769 						  SCSI_PROT_GUARD_CHECK |
770 						  SCSI_PROT_REF_CHECK |
771 						  SCSI_PROT_REF_INCREMENT |
772 						  SCSI_PROT_IP_CHECKSUM,
773 	};
774 
775 	return flag_mask[prot_op];
776 }
777 
778 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
779 					   unsigned int dix, unsigned int dif)
780 {
781 	struct request *rq = scsi_cmd_to_rq(scmd);
782 	struct bio *bio = rq->bio;
783 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
784 	unsigned int protect = 0;
785 
786 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
787 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
788 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
789 
790 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
791 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
792 	}
793 
794 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
795 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
796 
797 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
798 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
799 	}
800 
801 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
802 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
803 
804 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
805 			protect = 3 << 5;	/* Disable target PI checking */
806 		else
807 			protect = 1 << 5;	/* Enable target PI checking */
808 	}
809 
810 	scsi_set_prot_op(scmd, prot_op);
811 	scsi_set_prot_type(scmd, dif);
812 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
813 
814 	return protect;
815 }
816 
817 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
818 {
819 	struct request_queue *q = sdkp->disk->queue;
820 	unsigned int logical_block_size = sdkp->device->sector_size;
821 	unsigned int max_blocks = 0;
822 
823 	q->limits.discard_alignment =
824 		sdkp->unmap_alignment * logical_block_size;
825 	q->limits.discard_granularity =
826 		max(sdkp->physical_block_size,
827 		    sdkp->unmap_granularity * logical_block_size);
828 	sdkp->provisioning_mode = mode;
829 
830 	switch (mode) {
831 
832 	case SD_LBP_FULL:
833 	case SD_LBP_DISABLE:
834 		blk_queue_max_discard_sectors(q, 0);
835 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
836 		return;
837 
838 	case SD_LBP_UNMAP:
839 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
840 					  (u32)SD_MAX_WS16_BLOCKS);
841 		break;
842 
843 	case SD_LBP_WS16:
844 		if (sdkp->device->unmap_limit_for_ws)
845 			max_blocks = sdkp->max_unmap_blocks;
846 		else
847 			max_blocks = sdkp->max_ws_blocks;
848 
849 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
850 		break;
851 
852 	case SD_LBP_WS10:
853 		if (sdkp->device->unmap_limit_for_ws)
854 			max_blocks = sdkp->max_unmap_blocks;
855 		else
856 			max_blocks = sdkp->max_ws_blocks;
857 
858 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
859 		break;
860 
861 	case SD_LBP_ZERO:
862 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
863 					  (u32)SD_MAX_WS10_BLOCKS);
864 		break;
865 	}
866 
867 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
868 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
869 }
870 
871 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
872 {
873 	struct scsi_device *sdp = cmd->device;
874 	struct request *rq = scsi_cmd_to_rq(cmd);
875 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
876 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
877 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
878 	unsigned int data_len = 24;
879 	char *buf;
880 
881 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 	if (!rq->special_vec.bv_page)
883 		return BLK_STS_RESOURCE;
884 	clear_highpage(rq->special_vec.bv_page);
885 	rq->special_vec.bv_offset = 0;
886 	rq->special_vec.bv_len = data_len;
887 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
888 
889 	cmd->cmd_len = 10;
890 	cmd->cmnd[0] = UNMAP;
891 	cmd->cmnd[8] = 24;
892 
893 	buf = bvec_virt(&rq->special_vec);
894 	put_unaligned_be16(6 + 16, &buf[0]);
895 	put_unaligned_be16(16, &buf[2]);
896 	put_unaligned_be64(lba, &buf[8]);
897 	put_unaligned_be32(nr_blocks, &buf[16]);
898 
899 	cmd->allowed = sdkp->max_retries;
900 	cmd->transfersize = data_len;
901 	rq->timeout = SD_TIMEOUT;
902 
903 	return scsi_alloc_sgtables(cmd);
904 }
905 
906 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
907 		bool unmap)
908 {
909 	struct scsi_device *sdp = cmd->device;
910 	struct request *rq = scsi_cmd_to_rq(cmd);
911 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
912 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
913 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
914 	u32 data_len = sdp->sector_size;
915 
916 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
917 	if (!rq->special_vec.bv_page)
918 		return BLK_STS_RESOURCE;
919 	clear_highpage(rq->special_vec.bv_page);
920 	rq->special_vec.bv_offset = 0;
921 	rq->special_vec.bv_len = data_len;
922 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
923 
924 	cmd->cmd_len = 16;
925 	cmd->cmnd[0] = WRITE_SAME_16;
926 	if (unmap)
927 		cmd->cmnd[1] = 0x8; /* UNMAP */
928 	put_unaligned_be64(lba, &cmd->cmnd[2]);
929 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
930 
931 	cmd->allowed = sdkp->max_retries;
932 	cmd->transfersize = data_len;
933 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
934 
935 	return scsi_alloc_sgtables(cmd);
936 }
937 
938 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
939 		bool unmap)
940 {
941 	struct scsi_device *sdp = cmd->device;
942 	struct request *rq = scsi_cmd_to_rq(cmd);
943 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
944 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
945 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
946 	u32 data_len = sdp->sector_size;
947 
948 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
949 	if (!rq->special_vec.bv_page)
950 		return BLK_STS_RESOURCE;
951 	clear_highpage(rq->special_vec.bv_page);
952 	rq->special_vec.bv_offset = 0;
953 	rq->special_vec.bv_len = data_len;
954 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
955 
956 	cmd->cmd_len = 10;
957 	cmd->cmnd[0] = WRITE_SAME;
958 	if (unmap)
959 		cmd->cmnd[1] = 0x8; /* UNMAP */
960 	put_unaligned_be32(lba, &cmd->cmnd[2]);
961 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
962 
963 	cmd->allowed = sdkp->max_retries;
964 	cmd->transfersize = data_len;
965 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
966 
967 	return scsi_alloc_sgtables(cmd);
968 }
969 
970 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
971 {
972 	struct request *rq = scsi_cmd_to_rq(cmd);
973 	struct scsi_device *sdp = cmd->device;
974 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
975 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
976 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
977 
978 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
979 		switch (sdkp->zeroing_mode) {
980 		case SD_ZERO_WS16_UNMAP:
981 			return sd_setup_write_same16_cmnd(cmd, true);
982 		case SD_ZERO_WS10_UNMAP:
983 			return sd_setup_write_same10_cmnd(cmd, true);
984 		}
985 	}
986 
987 	if (sdp->no_write_same) {
988 		rq->rq_flags |= RQF_QUIET;
989 		return BLK_STS_TARGET;
990 	}
991 
992 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
993 		return sd_setup_write_same16_cmnd(cmd, false);
994 
995 	return sd_setup_write_same10_cmnd(cmd, false);
996 }
997 
998 static void sd_config_write_same(struct scsi_disk *sdkp)
999 {
1000 	struct request_queue *q = sdkp->disk->queue;
1001 	unsigned int logical_block_size = sdkp->device->sector_size;
1002 
1003 	if (sdkp->device->no_write_same) {
1004 		sdkp->max_ws_blocks = 0;
1005 		goto out;
1006 	}
1007 
1008 	/* Some devices can not handle block counts above 0xffff despite
1009 	 * supporting WRITE SAME(16). Consequently we default to 64k
1010 	 * blocks per I/O unless the device explicitly advertises a
1011 	 * bigger limit.
1012 	 */
1013 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1014 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1015 						   (u32)SD_MAX_WS16_BLOCKS);
1016 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1017 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1018 						   (u32)SD_MAX_WS10_BLOCKS);
1019 	else {
1020 		sdkp->device->no_write_same = 1;
1021 		sdkp->max_ws_blocks = 0;
1022 	}
1023 
1024 	if (sdkp->lbprz && sdkp->lbpws)
1025 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1026 	else if (sdkp->lbprz && sdkp->lbpws10)
1027 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1028 	else if (sdkp->max_ws_blocks)
1029 		sdkp->zeroing_mode = SD_ZERO_WS;
1030 	else
1031 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1032 
1033 	if (sdkp->max_ws_blocks &&
1034 	    sdkp->physical_block_size > logical_block_size) {
1035 		/*
1036 		 * Reporting a maximum number of blocks that is not aligned
1037 		 * on the device physical size would cause a large write same
1038 		 * request to be split into physically unaligned chunks by
1039 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1040 		 * even if the caller of these functions took care to align the
1041 		 * large request. So make sure the maximum reported is aligned
1042 		 * to the device physical block size. This is only an optional
1043 		 * optimization for regular disks, but this is mandatory to
1044 		 * avoid failure of large write same requests directed at
1045 		 * sequential write required zones of host-managed ZBC disks.
1046 		 */
1047 		sdkp->max_ws_blocks =
1048 			round_down(sdkp->max_ws_blocks,
1049 				   bytes_to_logical(sdkp->device,
1050 						    sdkp->physical_block_size));
1051 	}
1052 
1053 out:
1054 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1055 					 (logical_block_size >> 9));
1056 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1057 					 (logical_block_size >> 9));
1058 }
1059 
1060 /**
1061  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1062  * @cmd: command to prepare
1063  *
1064  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1065  * the preference indicated by the target device.
1066  **/
1067 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1068 {
1069 	struct request *rq = scsi_cmd_to_rq(cmd);
1070 	struct scsi_device *sdp = cmd->device;
1071 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1072 	struct bio *bio = rq->bio;
1073 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1074 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1075 	blk_status_t ret;
1076 
1077 	if (sdkp->device->no_write_same)
1078 		return BLK_STS_TARGET;
1079 
1080 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1081 
1082 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1083 
1084 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1085 		cmd->cmd_len = 16;
1086 		cmd->cmnd[0] = WRITE_SAME_16;
1087 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1088 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1089 	} else {
1090 		cmd->cmd_len = 10;
1091 		cmd->cmnd[0] = WRITE_SAME;
1092 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1093 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1094 	}
1095 
1096 	cmd->transfersize = sdp->sector_size;
1097 	cmd->allowed = sdkp->max_retries;
1098 
1099 	/*
1100 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1101 	 * different from the amount of data actually written to the target.
1102 	 *
1103 	 * We set up __data_len to the amount of data transferred via the
1104 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1105 	 * to transfer a single sector of data first, but then reset it to
1106 	 * the amount of data to be written right after so that the I/O path
1107 	 * knows how much to actually write.
1108 	 */
1109 	rq->__data_len = sdp->sector_size;
1110 	ret = scsi_alloc_sgtables(cmd);
1111 	rq->__data_len = blk_rq_bytes(rq);
1112 
1113 	return ret;
1114 }
1115 
1116 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1117 {
1118 	struct request *rq = scsi_cmd_to_rq(cmd);
1119 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1120 
1121 	/* flush requests don't perform I/O, zero the S/G table */
1122 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1123 
1124 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1125 	cmd->cmd_len = 10;
1126 	cmd->transfersize = 0;
1127 	cmd->allowed = sdkp->max_retries;
1128 
1129 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1130 	return BLK_STS_OK;
1131 }
1132 
1133 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1134 				       sector_t lba, unsigned int nr_blocks,
1135 				       unsigned char flags)
1136 {
1137 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1138 	if (unlikely(cmd->cmnd == NULL))
1139 		return BLK_STS_RESOURCE;
1140 
1141 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1142 	memset(cmd->cmnd, 0, cmd->cmd_len);
1143 
1144 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1145 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1146 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1147 	cmd->cmnd[10] = flags;
1148 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1149 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1150 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1151 
1152 	return BLK_STS_OK;
1153 }
1154 
1155 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1156 				       sector_t lba, unsigned int nr_blocks,
1157 				       unsigned char flags)
1158 {
1159 	cmd->cmd_len  = 16;
1160 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1161 	cmd->cmnd[1]  = flags;
1162 	cmd->cmnd[14] = 0;
1163 	cmd->cmnd[15] = 0;
1164 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1165 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1166 
1167 	return BLK_STS_OK;
1168 }
1169 
1170 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1171 				       sector_t lba, unsigned int nr_blocks,
1172 				       unsigned char flags)
1173 {
1174 	cmd->cmd_len = 10;
1175 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1176 	cmd->cmnd[1] = flags;
1177 	cmd->cmnd[6] = 0;
1178 	cmd->cmnd[9] = 0;
1179 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1180 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1181 
1182 	return BLK_STS_OK;
1183 }
1184 
1185 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1186 				      sector_t lba, unsigned int nr_blocks,
1187 				      unsigned char flags)
1188 {
1189 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1190 	if (WARN_ON_ONCE(nr_blocks == 0))
1191 		return BLK_STS_IOERR;
1192 
1193 	if (unlikely(flags & 0x8)) {
1194 		/*
1195 		 * This happens only if this drive failed 10byte rw
1196 		 * command with ILLEGAL_REQUEST during operation and
1197 		 * thus turned off use_10_for_rw.
1198 		 */
1199 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1200 		return BLK_STS_IOERR;
1201 	}
1202 
1203 	cmd->cmd_len = 6;
1204 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1205 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1206 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1207 	cmd->cmnd[3] = lba & 0xff;
1208 	cmd->cmnd[4] = nr_blocks;
1209 	cmd->cmnd[5] = 0;
1210 
1211 	return BLK_STS_OK;
1212 }
1213 
1214 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1215 {
1216 	struct request *rq = scsi_cmd_to_rq(cmd);
1217 	struct scsi_device *sdp = cmd->device;
1218 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1219 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1220 	sector_t threshold;
1221 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1222 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1223 	bool write = rq_data_dir(rq) == WRITE;
1224 	unsigned char protect, fua;
1225 	blk_status_t ret;
1226 	unsigned int dif;
1227 	bool dix;
1228 
1229 	ret = scsi_alloc_sgtables(cmd);
1230 	if (ret != BLK_STS_OK)
1231 		return ret;
1232 
1233 	ret = BLK_STS_IOERR;
1234 	if (!scsi_device_online(sdp) || sdp->changed) {
1235 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1236 		goto fail;
1237 	}
1238 
1239 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1240 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1241 		goto fail;
1242 	}
1243 
1244 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1245 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1246 		goto fail;
1247 	}
1248 
1249 	/*
1250 	 * Some SD card readers can't handle accesses which touch the
1251 	 * last one or two logical blocks. Split accesses as needed.
1252 	 */
1253 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1254 
1255 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1256 		if (lba < threshold) {
1257 			/* Access up to the threshold but not beyond */
1258 			nr_blocks = threshold - lba;
1259 		} else {
1260 			/* Access only a single logical block */
1261 			nr_blocks = 1;
1262 		}
1263 	}
1264 
1265 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1266 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1267 		if (ret)
1268 			goto fail;
1269 	}
1270 
1271 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1272 	dix = scsi_prot_sg_count(cmd);
1273 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1274 
1275 	if (dif || dix)
1276 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1277 	else
1278 		protect = 0;
1279 
1280 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1281 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1282 					 protect | fua);
1283 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1284 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1285 					 protect | fua);
1286 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1287 		   sdp->use_10_for_rw || protect) {
1288 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1289 					 protect | fua);
1290 	} else {
1291 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1292 					protect | fua);
1293 	}
1294 
1295 	if (unlikely(ret != BLK_STS_OK))
1296 		goto fail;
1297 
1298 	/*
1299 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1300 	 * host adapter, it's safe to assume that we can at least transfer
1301 	 * this many bytes between each connect / disconnect.
1302 	 */
1303 	cmd->transfersize = sdp->sector_size;
1304 	cmd->underflow = nr_blocks << 9;
1305 	cmd->allowed = sdkp->max_retries;
1306 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1307 
1308 	SCSI_LOG_HLQUEUE(1,
1309 			 scmd_printk(KERN_INFO, cmd,
1310 				     "%s: block=%llu, count=%d\n", __func__,
1311 				     (unsigned long long)blk_rq_pos(rq),
1312 				     blk_rq_sectors(rq)));
1313 	SCSI_LOG_HLQUEUE(2,
1314 			 scmd_printk(KERN_INFO, cmd,
1315 				     "%s %d/%u 512 byte blocks.\n",
1316 				     write ? "writing" : "reading", nr_blocks,
1317 				     blk_rq_sectors(rq)));
1318 
1319 	/*
1320 	 * This indicates that the command is ready from our end to be queued.
1321 	 */
1322 	return BLK_STS_OK;
1323 fail:
1324 	scsi_free_sgtables(cmd);
1325 	return ret;
1326 }
1327 
1328 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1329 {
1330 	struct request *rq = scsi_cmd_to_rq(cmd);
1331 
1332 	switch (req_op(rq)) {
1333 	case REQ_OP_DISCARD:
1334 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1335 		case SD_LBP_UNMAP:
1336 			return sd_setup_unmap_cmnd(cmd);
1337 		case SD_LBP_WS16:
1338 			return sd_setup_write_same16_cmnd(cmd, true);
1339 		case SD_LBP_WS10:
1340 			return sd_setup_write_same10_cmnd(cmd, true);
1341 		case SD_LBP_ZERO:
1342 			return sd_setup_write_same10_cmnd(cmd, false);
1343 		default:
1344 			return BLK_STS_TARGET;
1345 		}
1346 	case REQ_OP_WRITE_ZEROES:
1347 		return sd_setup_write_zeroes_cmnd(cmd);
1348 	case REQ_OP_WRITE_SAME:
1349 		return sd_setup_write_same_cmnd(cmd);
1350 	case REQ_OP_FLUSH:
1351 		return sd_setup_flush_cmnd(cmd);
1352 	case REQ_OP_READ:
1353 	case REQ_OP_WRITE:
1354 	case REQ_OP_ZONE_APPEND:
1355 		return sd_setup_read_write_cmnd(cmd);
1356 	case REQ_OP_ZONE_RESET:
1357 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358 						   false);
1359 	case REQ_OP_ZONE_RESET_ALL:
1360 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1361 						   true);
1362 	case REQ_OP_ZONE_OPEN:
1363 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1364 	case REQ_OP_ZONE_CLOSE:
1365 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1366 	case REQ_OP_ZONE_FINISH:
1367 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1368 	default:
1369 		WARN_ON_ONCE(1);
1370 		return BLK_STS_NOTSUPP;
1371 	}
1372 }
1373 
1374 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1375 {
1376 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1377 	u8 *cmnd;
1378 
1379 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1380 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1381 
1382 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1383 		cmnd = SCpnt->cmnd;
1384 		SCpnt->cmnd = NULL;
1385 		SCpnt->cmd_len = 0;
1386 		mempool_free(cmnd, sd_cdb_pool);
1387 	}
1388 }
1389 
1390 static bool sd_need_revalidate(struct block_device *bdev,
1391 		struct scsi_disk *sdkp)
1392 {
1393 	if (sdkp->device->removable || sdkp->write_prot) {
1394 		if (bdev_check_media_change(bdev))
1395 			return true;
1396 	}
1397 
1398 	/*
1399 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1400 	 * nothing to do with partitions, BLKRRPART is used to force a full
1401 	 * revalidate after things like a format for historical reasons.
1402 	 */
1403 	return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1404 }
1405 
1406 /**
1407  *	sd_open - open a scsi disk device
1408  *	@bdev: Block device of the scsi disk to open
1409  *	@mode: FMODE_* mask
1410  *
1411  *	Returns 0 if successful. Returns a negated errno value in case
1412  *	of error.
1413  *
1414  *	Note: This can be called from a user context (e.g. fsck(1) )
1415  *	or from within the kernel (e.g. as a result of a mount(1) ).
1416  *	In the latter case @inode and @filp carry an abridged amount
1417  *	of information as noted above.
1418  *
1419  *	Locking: called with bdev->bd_disk->open_mutex held.
1420  **/
1421 static int sd_open(struct block_device *bdev, fmode_t mode)
1422 {
1423 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1424 	struct scsi_device *sdev;
1425 	int retval;
1426 
1427 	if (!sdkp)
1428 		return -ENXIO;
1429 
1430 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1431 
1432 	sdev = sdkp->device;
1433 
1434 	/*
1435 	 * If the device is in error recovery, wait until it is done.
1436 	 * If the device is offline, then disallow any access to it.
1437 	 */
1438 	retval = -ENXIO;
1439 	if (!scsi_block_when_processing_errors(sdev))
1440 		goto error_out;
1441 
1442 	if (sd_need_revalidate(bdev, sdkp))
1443 		sd_revalidate_disk(bdev->bd_disk);
1444 
1445 	/*
1446 	 * If the drive is empty, just let the open fail.
1447 	 */
1448 	retval = -ENOMEDIUM;
1449 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1450 		goto error_out;
1451 
1452 	/*
1453 	 * If the device has the write protect tab set, have the open fail
1454 	 * if the user expects to be able to write to the thing.
1455 	 */
1456 	retval = -EROFS;
1457 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1458 		goto error_out;
1459 
1460 	/*
1461 	 * It is possible that the disk changing stuff resulted in
1462 	 * the device being taken offline.  If this is the case,
1463 	 * report this to the user, and don't pretend that the
1464 	 * open actually succeeded.
1465 	 */
1466 	retval = -ENXIO;
1467 	if (!scsi_device_online(sdev))
1468 		goto error_out;
1469 
1470 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1471 		if (scsi_block_when_processing_errors(sdev))
1472 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1473 	}
1474 
1475 	return 0;
1476 
1477 error_out:
1478 	scsi_disk_put(sdkp);
1479 	return retval;
1480 }
1481 
1482 /**
1483  *	sd_release - invoked when the (last) close(2) is called on this
1484  *	scsi disk.
1485  *	@disk: disk to release
1486  *	@mode: FMODE_* mask
1487  *
1488  *	Returns 0.
1489  *
1490  *	Note: may block (uninterruptible) if error recovery is underway
1491  *	on this disk.
1492  *
1493  *	Locking: called with bdev->bd_disk->open_mutex held.
1494  **/
1495 static void sd_release(struct gendisk *disk, fmode_t mode)
1496 {
1497 	struct scsi_disk *sdkp = scsi_disk(disk);
1498 	struct scsi_device *sdev = sdkp->device;
1499 
1500 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1501 
1502 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1503 		if (scsi_block_when_processing_errors(sdev))
1504 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1505 	}
1506 
1507 	scsi_disk_put(sdkp);
1508 }
1509 
1510 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1511 {
1512 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1513 	struct scsi_device *sdp = sdkp->device;
1514 	struct Scsi_Host *host = sdp->host;
1515 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1516 	int diskinfo[4];
1517 
1518 	/* default to most commonly used values */
1519 	diskinfo[0] = 0x40;	/* 1 << 6 */
1520 	diskinfo[1] = 0x20;	/* 1 << 5 */
1521 	diskinfo[2] = capacity >> 11;
1522 
1523 	/* override with calculated, extended default, or driver values */
1524 	if (host->hostt->bios_param)
1525 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1526 	else
1527 		scsicam_bios_param(bdev, capacity, diskinfo);
1528 
1529 	geo->heads = diskinfo[0];
1530 	geo->sectors = diskinfo[1];
1531 	geo->cylinders = diskinfo[2];
1532 	return 0;
1533 }
1534 
1535 /**
1536  *	sd_ioctl - process an ioctl
1537  *	@bdev: target block device
1538  *	@mode: FMODE_* mask
1539  *	@cmd: ioctl command number
1540  *	@arg: this is third argument given to ioctl(2) system call.
1541  *	Often contains a pointer.
1542  *
1543  *	Returns 0 if successful (some ioctls return positive numbers on
1544  *	success as well). Returns a negated errno value in case of error.
1545  *
1546  *	Note: most ioctls are forward onto the block subsystem or further
1547  *	down in the scsi subsystem.
1548  **/
1549 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1550 		    unsigned int cmd, unsigned long arg)
1551 {
1552 	struct gendisk *disk = bdev->bd_disk;
1553 	struct scsi_disk *sdkp = scsi_disk(disk);
1554 	struct scsi_device *sdp = sdkp->device;
1555 	void __user *p = (void __user *)arg;
1556 	int error;
1557 
1558 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1559 				    "cmd=0x%x\n", disk->disk_name, cmd));
1560 
1561 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1562 		return -ENOIOCTLCMD;
1563 
1564 	/*
1565 	 * If we are in the middle of error recovery, don't let anyone
1566 	 * else try and use this device.  Also, if error recovery fails, it
1567 	 * may try and take the device offline, in which case all further
1568 	 * access to the device is prohibited.
1569 	 */
1570 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1571 			(mode & FMODE_NDELAY) != 0);
1572 	if (error)
1573 		return error;
1574 
1575 	if (is_sed_ioctl(cmd))
1576 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1577 	return scsi_ioctl(sdp, disk, mode, cmd, p);
1578 }
1579 
1580 static void set_media_not_present(struct scsi_disk *sdkp)
1581 {
1582 	if (sdkp->media_present)
1583 		sdkp->device->changed = 1;
1584 
1585 	if (sdkp->device->removable) {
1586 		sdkp->media_present = 0;
1587 		sdkp->capacity = 0;
1588 	}
1589 }
1590 
1591 static int media_not_present(struct scsi_disk *sdkp,
1592 			     struct scsi_sense_hdr *sshdr)
1593 {
1594 	if (!scsi_sense_valid(sshdr))
1595 		return 0;
1596 
1597 	/* not invoked for commands that could return deferred errors */
1598 	switch (sshdr->sense_key) {
1599 	case UNIT_ATTENTION:
1600 	case NOT_READY:
1601 		/* medium not present */
1602 		if (sshdr->asc == 0x3A) {
1603 			set_media_not_present(sdkp);
1604 			return 1;
1605 		}
1606 	}
1607 	return 0;
1608 }
1609 
1610 /**
1611  *	sd_check_events - check media events
1612  *	@disk: kernel device descriptor
1613  *	@clearing: disk events currently being cleared
1614  *
1615  *	Returns mask of DISK_EVENT_*.
1616  *
1617  *	Note: this function is invoked from the block subsystem.
1618  **/
1619 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1620 {
1621 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1622 	struct scsi_device *sdp;
1623 	int retval;
1624 	bool disk_changed;
1625 
1626 	if (!sdkp)
1627 		return 0;
1628 
1629 	sdp = sdkp->device;
1630 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1631 
1632 	/*
1633 	 * If the device is offline, don't send any commands - just pretend as
1634 	 * if the command failed.  If the device ever comes back online, we
1635 	 * can deal with it then.  It is only because of unrecoverable errors
1636 	 * that we would ever take a device offline in the first place.
1637 	 */
1638 	if (!scsi_device_online(sdp)) {
1639 		set_media_not_present(sdkp);
1640 		goto out;
1641 	}
1642 
1643 	/*
1644 	 * Using TEST_UNIT_READY enables differentiation between drive with
1645 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1646 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1647 	 *
1648 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1649 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1650 	 * sd_revalidate() is called.
1651 	 */
1652 	if (scsi_block_when_processing_errors(sdp)) {
1653 		struct scsi_sense_hdr sshdr = { 0, };
1654 
1655 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1656 					      &sshdr);
1657 
1658 		/* failed to execute TUR, assume media not present */
1659 		if (retval < 0 || host_byte(retval)) {
1660 			set_media_not_present(sdkp);
1661 			goto out;
1662 		}
1663 
1664 		if (media_not_present(sdkp, &sshdr))
1665 			goto out;
1666 	}
1667 
1668 	/*
1669 	 * For removable scsi disk we have to recognise the presence
1670 	 * of a disk in the drive.
1671 	 */
1672 	if (!sdkp->media_present)
1673 		sdp->changed = 1;
1674 	sdkp->media_present = 1;
1675 out:
1676 	/*
1677 	 * sdp->changed is set under the following conditions:
1678 	 *
1679 	 *	Medium present state has changed in either direction.
1680 	 *	Device has indicated UNIT_ATTENTION.
1681 	 */
1682 	disk_changed = sdp->changed;
1683 	sdp->changed = 0;
1684 	scsi_disk_put(sdkp);
1685 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1686 }
1687 
1688 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1689 {
1690 	int retries, res;
1691 	struct scsi_device *sdp = sdkp->device;
1692 	const int timeout = sdp->request_queue->rq_timeout
1693 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1694 	struct scsi_sense_hdr my_sshdr;
1695 
1696 	if (!scsi_device_online(sdp))
1697 		return -ENODEV;
1698 
1699 	/* caller might not be interested in sense, but we need it */
1700 	if (!sshdr)
1701 		sshdr = &my_sshdr;
1702 
1703 	for (retries = 3; retries > 0; --retries) {
1704 		unsigned char cmd[10] = { 0 };
1705 
1706 		cmd[0] = SYNCHRONIZE_CACHE;
1707 		/*
1708 		 * Leave the rest of the command zero to indicate
1709 		 * flush everything.
1710 		 */
1711 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1712 				timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1713 		if (res == 0)
1714 			break;
1715 	}
1716 
1717 	if (res) {
1718 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1719 
1720 		if (res < 0)
1721 			return res;
1722 
1723 		if (scsi_status_is_check_condition(res) &&
1724 		    scsi_sense_valid(sshdr)) {
1725 			sd_print_sense_hdr(sdkp, sshdr);
1726 
1727 			/* we need to evaluate the error return  */
1728 			if (sshdr->asc == 0x3a ||	/* medium not present */
1729 			    sshdr->asc == 0x20 ||	/* invalid command */
1730 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1731 				/* this is no error here */
1732 				return 0;
1733 		}
1734 
1735 		switch (host_byte(res)) {
1736 		/* ignore errors due to racing a disconnection */
1737 		case DID_BAD_TARGET:
1738 		case DID_NO_CONNECT:
1739 			return 0;
1740 		/* signal the upper layer it might try again */
1741 		case DID_BUS_BUSY:
1742 		case DID_IMM_RETRY:
1743 		case DID_REQUEUE:
1744 		case DID_SOFT_ERROR:
1745 			return -EBUSY;
1746 		default:
1747 			return -EIO;
1748 		}
1749 	}
1750 	return 0;
1751 }
1752 
1753 static void sd_rescan(struct device *dev)
1754 {
1755 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1756 
1757 	sd_revalidate_disk(sdkp->disk);
1758 }
1759 
1760 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1761 		enum blk_unique_id type)
1762 {
1763 	struct scsi_device *sdev = scsi_disk(disk)->device;
1764 	const struct scsi_vpd *vpd;
1765 	const unsigned char *d;
1766 	int ret = -ENXIO, len;
1767 
1768 	rcu_read_lock();
1769 	vpd = rcu_dereference(sdev->vpd_pg83);
1770 	if (!vpd)
1771 		goto out_unlock;
1772 
1773 	ret = -EINVAL;
1774 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1775 		/* we only care about designators with LU association */
1776 		if (((d[1] >> 4) & 0x3) != 0x00)
1777 			continue;
1778 		if ((d[1] & 0xf) != type)
1779 			continue;
1780 
1781 		/*
1782 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1783 		 * keep looking as one with more entropy might still show up.
1784 		 */
1785 		len = d[3];
1786 		if (len != 8 && len != 12 && len != 16)
1787 			continue;
1788 		ret = len;
1789 		memcpy(id, d + 4, len);
1790 		if (len == 16)
1791 			break;
1792 	}
1793 out_unlock:
1794 	rcu_read_unlock();
1795 	return ret;
1796 }
1797 
1798 static char sd_pr_type(enum pr_type type)
1799 {
1800 	switch (type) {
1801 	case PR_WRITE_EXCLUSIVE:
1802 		return 0x01;
1803 	case PR_EXCLUSIVE_ACCESS:
1804 		return 0x03;
1805 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1806 		return 0x05;
1807 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1808 		return 0x06;
1809 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1810 		return 0x07;
1811 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1812 		return 0x08;
1813 	default:
1814 		return 0;
1815 	}
1816 };
1817 
1818 static int sd_pr_command(struct block_device *bdev, u8 sa,
1819 		u64 key, u64 sa_key, u8 type, u8 flags)
1820 {
1821 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1822 	struct scsi_device *sdev = sdkp->device;
1823 	struct scsi_sense_hdr sshdr;
1824 	int result;
1825 	u8 cmd[16] = { 0, };
1826 	u8 data[24] = { 0, };
1827 
1828 	cmd[0] = PERSISTENT_RESERVE_OUT;
1829 	cmd[1] = sa;
1830 	cmd[2] = type;
1831 	put_unaligned_be32(sizeof(data), &cmd[5]);
1832 
1833 	put_unaligned_be64(key, &data[0]);
1834 	put_unaligned_be64(sa_key, &data[8]);
1835 	data[20] = flags;
1836 
1837 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1838 			&sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1839 
1840 	if (scsi_status_is_check_condition(result) &&
1841 	    scsi_sense_valid(&sshdr)) {
1842 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1843 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1844 	}
1845 
1846 	return result;
1847 }
1848 
1849 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1850 		u32 flags)
1851 {
1852 	if (flags & ~PR_FL_IGNORE_KEY)
1853 		return -EOPNOTSUPP;
1854 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1855 			old_key, new_key, 0,
1856 			(1 << 0) /* APTPL */);
1857 }
1858 
1859 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1860 		u32 flags)
1861 {
1862 	if (flags)
1863 		return -EOPNOTSUPP;
1864 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1865 }
1866 
1867 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1868 {
1869 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1870 }
1871 
1872 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1873 		enum pr_type type, bool abort)
1874 {
1875 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1876 			     sd_pr_type(type), 0);
1877 }
1878 
1879 static int sd_pr_clear(struct block_device *bdev, u64 key)
1880 {
1881 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1882 }
1883 
1884 static const struct pr_ops sd_pr_ops = {
1885 	.pr_register	= sd_pr_register,
1886 	.pr_reserve	= sd_pr_reserve,
1887 	.pr_release	= sd_pr_release,
1888 	.pr_preempt	= sd_pr_preempt,
1889 	.pr_clear	= sd_pr_clear,
1890 };
1891 
1892 static const struct block_device_operations sd_fops = {
1893 	.owner			= THIS_MODULE,
1894 	.open			= sd_open,
1895 	.release		= sd_release,
1896 	.ioctl			= sd_ioctl,
1897 	.getgeo			= sd_getgeo,
1898 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1899 	.check_events		= sd_check_events,
1900 	.unlock_native_capacity	= sd_unlock_native_capacity,
1901 	.report_zones		= sd_zbc_report_zones,
1902 	.get_unique_id		= sd_get_unique_id,
1903 	.pr_ops			= &sd_pr_ops,
1904 };
1905 
1906 /**
1907  *	sd_eh_reset - reset error handling callback
1908  *	@scmd:		sd-issued command that has failed
1909  *
1910  *	This function is called by the SCSI midlayer before starting
1911  *	SCSI EH. When counting medium access failures we have to be
1912  *	careful to register it only only once per device and SCSI EH run;
1913  *	there might be several timed out commands which will cause the
1914  *	'max_medium_access_timeouts' counter to trigger after the first
1915  *	SCSI EH run already and set the device to offline.
1916  *	So this function resets the internal counter before starting SCSI EH.
1917  **/
1918 static void sd_eh_reset(struct scsi_cmnd *scmd)
1919 {
1920 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1921 
1922 	/* New SCSI EH run, reset gate variable */
1923 	sdkp->ignore_medium_access_errors = false;
1924 }
1925 
1926 /**
1927  *	sd_eh_action - error handling callback
1928  *	@scmd:		sd-issued command that has failed
1929  *	@eh_disp:	The recovery disposition suggested by the midlayer
1930  *
1931  *	This function is called by the SCSI midlayer upon completion of an
1932  *	error test command (currently TEST UNIT READY). The result of sending
1933  *	the eh command is passed in eh_disp.  We're looking for devices that
1934  *	fail medium access commands but are OK with non access commands like
1935  *	test unit ready (so wrongly see the device as having a successful
1936  *	recovery)
1937  **/
1938 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1939 {
1940 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1941 	struct scsi_device *sdev = scmd->device;
1942 
1943 	if (!scsi_device_online(sdev) ||
1944 	    !scsi_medium_access_command(scmd) ||
1945 	    host_byte(scmd->result) != DID_TIME_OUT ||
1946 	    eh_disp != SUCCESS)
1947 		return eh_disp;
1948 
1949 	/*
1950 	 * The device has timed out executing a medium access command.
1951 	 * However, the TEST UNIT READY command sent during error
1952 	 * handling completed successfully. Either the device is in the
1953 	 * process of recovering or has it suffered an internal failure
1954 	 * that prevents access to the storage medium.
1955 	 */
1956 	if (!sdkp->ignore_medium_access_errors) {
1957 		sdkp->medium_access_timed_out++;
1958 		sdkp->ignore_medium_access_errors = true;
1959 	}
1960 
1961 	/*
1962 	 * If the device keeps failing read/write commands but TEST UNIT
1963 	 * READY always completes successfully we assume that medium
1964 	 * access is no longer possible and take the device offline.
1965 	 */
1966 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1967 		scmd_printk(KERN_ERR, scmd,
1968 			    "Medium access timeout failure. Offlining disk!\n");
1969 		mutex_lock(&sdev->state_mutex);
1970 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1971 		mutex_unlock(&sdev->state_mutex);
1972 
1973 		return SUCCESS;
1974 	}
1975 
1976 	return eh_disp;
1977 }
1978 
1979 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1980 {
1981 	struct request *req = scsi_cmd_to_rq(scmd);
1982 	struct scsi_device *sdev = scmd->device;
1983 	unsigned int transferred, good_bytes;
1984 	u64 start_lba, end_lba, bad_lba;
1985 
1986 	/*
1987 	 * Some commands have a payload smaller than the device logical
1988 	 * block size (e.g. INQUIRY on a 4K disk).
1989 	 */
1990 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1991 		return 0;
1992 
1993 	/* Check if we have a 'bad_lba' information */
1994 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1995 				     SCSI_SENSE_BUFFERSIZE,
1996 				     &bad_lba))
1997 		return 0;
1998 
1999 	/*
2000 	 * If the bad lba was reported incorrectly, we have no idea where
2001 	 * the error is.
2002 	 */
2003 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2004 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2005 	if (bad_lba < start_lba || bad_lba >= end_lba)
2006 		return 0;
2007 
2008 	/*
2009 	 * resid is optional but mostly filled in.  When it's unused,
2010 	 * its value is zero, so we assume the whole buffer transferred
2011 	 */
2012 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2013 
2014 	/* This computation should always be done in terms of the
2015 	 * resolution of the device's medium.
2016 	 */
2017 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2018 
2019 	return min(good_bytes, transferred);
2020 }
2021 
2022 /**
2023  *	sd_done - bottom half handler: called when the lower level
2024  *	driver has completed (successfully or otherwise) a scsi command.
2025  *	@SCpnt: mid-level's per command structure.
2026  *
2027  *	Note: potentially run from within an ISR. Must not block.
2028  **/
2029 static int sd_done(struct scsi_cmnd *SCpnt)
2030 {
2031 	int result = SCpnt->result;
2032 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2033 	unsigned int sector_size = SCpnt->device->sector_size;
2034 	unsigned int resid;
2035 	struct scsi_sense_hdr sshdr;
2036 	struct request *req = scsi_cmd_to_rq(SCpnt);
2037 	struct scsi_disk *sdkp = scsi_disk(req->rq_disk);
2038 	int sense_valid = 0;
2039 	int sense_deferred = 0;
2040 
2041 	switch (req_op(req)) {
2042 	case REQ_OP_DISCARD:
2043 	case REQ_OP_WRITE_ZEROES:
2044 	case REQ_OP_WRITE_SAME:
2045 	case REQ_OP_ZONE_RESET:
2046 	case REQ_OP_ZONE_RESET_ALL:
2047 	case REQ_OP_ZONE_OPEN:
2048 	case REQ_OP_ZONE_CLOSE:
2049 	case REQ_OP_ZONE_FINISH:
2050 		if (!result) {
2051 			good_bytes = blk_rq_bytes(req);
2052 			scsi_set_resid(SCpnt, 0);
2053 		} else {
2054 			good_bytes = 0;
2055 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2056 		}
2057 		break;
2058 	default:
2059 		/*
2060 		 * In case of bogus fw or device, we could end up having
2061 		 * an unaligned partial completion. Check this here and force
2062 		 * alignment.
2063 		 */
2064 		resid = scsi_get_resid(SCpnt);
2065 		if (resid & (sector_size - 1)) {
2066 			sd_printk(KERN_INFO, sdkp,
2067 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2068 				resid, sector_size);
2069 			scsi_print_command(SCpnt);
2070 			resid = min(scsi_bufflen(SCpnt),
2071 				    round_up(resid, sector_size));
2072 			scsi_set_resid(SCpnt, resid);
2073 		}
2074 	}
2075 
2076 	if (result) {
2077 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2078 		if (sense_valid)
2079 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2080 	}
2081 	sdkp->medium_access_timed_out = 0;
2082 
2083 	if (!scsi_status_is_check_condition(result) &&
2084 	    (!sense_valid || sense_deferred))
2085 		goto out;
2086 
2087 	switch (sshdr.sense_key) {
2088 	case HARDWARE_ERROR:
2089 	case MEDIUM_ERROR:
2090 		good_bytes = sd_completed_bytes(SCpnt);
2091 		break;
2092 	case RECOVERED_ERROR:
2093 		good_bytes = scsi_bufflen(SCpnt);
2094 		break;
2095 	case NO_SENSE:
2096 		/* This indicates a false check condition, so ignore it.  An
2097 		 * unknown amount of data was transferred so treat it as an
2098 		 * error.
2099 		 */
2100 		SCpnt->result = 0;
2101 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2102 		break;
2103 	case ABORTED_COMMAND:
2104 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2105 			good_bytes = sd_completed_bytes(SCpnt);
2106 		break;
2107 	case ILLEGAL_REQUEST:
2108 		switch (sshdr.asc) {
2109 		case 0x10:	/* DIX: Host detected corruption */
2110 			good_bytes = sd_completed_bytes(SCpnt);
2111 			break;
2112 		case 0x20:	/* INVALID COMMAND OPCODE */
2113 		case 0x24:	/* INVALID FIELD IN CDB */
2114 			switch (SCpnt->cmnd[0]) {
2115 			case UNMAP:
2116 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2117 				break;
2118 			case WRITE_SAME_16:
2119 			case WRITE_SAME:
2120 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2121 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2122 				} else {
2123 					sdkp->device->no_write_same = 1;
2124 					sd_config_write_same(sdkp);
2125 					req->rq_flags |= RQF_QUIET;
2126 				}
2127 				break;
2128 			}
2129 		}
2130 		break;
2131 	default:
2132 		break;
2133 	}
2134 
2135  out:
2136 	if (sd_is_zoned(sdkp))
2137 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2138 
2139 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2140 					   "sd_done: completed %d of %d bytes\n",
2141 					   good_bytes, scsi_bufflen(SCpnt)));
2142 
2143 	return good_bytes;
2144 }
2145 
2146 /*
2147  * spinup disk - called only in sd_revalidate_disk()
2148  */
2149 static void
2150 sd_spinup_disk(struct scsi_disk *sdkp)
2151 {
2152 	unsigned char cmd[10];
2153 	unsigned long spintime_expire = 0;
2154 	int retries, spintime;
2155 	unsigned int the_result;
2156 	struct scsi_sense_hdr sshdr;
2157 	int sense_valid = 0;
2158 
2159 	spintime = 0;
2160 
2161 	/* Spin up drives, as required.  Only do this at boot time */
2162 	/* Spinup needs to be done for module loads too. */
2163 	do {
2164 		retries = 0;
2165 
2166 		do {
2167 			bool media_was_present = sdkp->media_present;
2168 
2169 			cmd[0] = TEST_UNIT_READY;
2170 			memset((void *) &cmd[1], 0, 9);
2171 
2172 			the_result = scsi_execute_req(sdkp->device, cmd,
2173 						      DMA_NONE, NULL, 0,
2174 						      &sshdr, SD_TIMEOUT,
2175 						      sdkp->max_retries, NULL);
2176 
2177 			/*
2178 			 * If the drive has indicated to us that it
2179 			 * doesn't have any media in it, don't bother
2180 			 * with any more polling.
2181 			 */
2182 			if (media_not_present(sdkp, &sshdr)) {
2183 				if (media_was_present)
2184 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2185 				return;
2186 			}
2187 
2188 			if (the_result)
2189 				sense_valid = scsi_sense_valid(&sshdr);
2190 			retries++;
2191 		} while (retries < 3 &&
2192 			 (!scsi_status_is_good(the_result) ||
2193 			  (scsi_status_is_check_condition(the_result) &&
2194 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2195 
2196 		if (!scsi_status_is_check_condition(the_result)) {
2197 			/* no sense, TUR either succeeded or failed
2198 			 * with a status error */
2199 			if(!spintime && !scsi_status_is_good(the_result)) {
2200 				sd_print_result(sdkp, "Test Unit Ready failed",
2201 						the_result);
2202 			}
2203 			break;
2204 		}
2205 
2206 		/*
2207 		 * The device does not want the automatic start to be issued.
2208 		 */
2209 		if (sdkp->device->no_start_on_add)
2210 			break;
2211 
2212 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2213 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2214 				break;	/* manual intervention required */
2215 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2216 				break;	/* standby */
2217 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2218 				break;	/* unavailable */
2219 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2220 				break;	/* sanitize in progress */
2221 			/*
2222 			 * Issue command to spin up drive when not ready
2223 			 */
2224 			if (!spintime) {
2225 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2226 				cmd[0] = START_STOP;
2227 				cmd[1] = 1;	/* Return immediately */
2228 				memset((void *) &cmd[2], 0, 8);
2229 				cmd[4] = 1;	/* Start spin cycle */
2230 				if (sdkp->device->start_stop_pwr_cond)
2231 					cmd[4] |= 1 << 4;
2232 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2233 						 NULL, 0, &sshdr,
2234 						 SD_TIMEOUT, sdkp->max_retries,
2235 						 NULL);
2236 				spintime_expire = jiffies + 100 * HZ;
2237 				spintime = 1;
2238 			}
2239 			/* Wait 1 second for next try */
2240 			msleep(1000);
2241 			printk(KERN_CONT ".");
2242 
2243 		/*
2244 		 * Wait for USB flash devices with slow firmware.
2245 		 * Yes, this sense key/ASC combination shouldn't
2246 		 * occur here.  It's characteristic of these devices.
2247 		 */
2248 		} else if (sense_valid &&
2249 				sshdr.sense_key == UNIT_ATTENTION &&
2250 				sshdr.asc == 0x28) {
2251 			if (!spintime) {
2252 				spintime_expire = jiffies + 5 * HZ;
2253 				spintime = 1;
2254 			}
2255 			/* Wait 1 second for next try */
2256 			msleep(1000);
2257 		} else {
2258 			/* we don't understand the sense code, so it's
2259 			 * probably pointless to loop */
2260 			if(!spintime) {
2261 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2262 				sd_print_sense_hdr(sdkp, &sshdr);
2263 			}
2264 			break;
2265 		}
2266 
2267 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2268 
2269 	if (spintime) {
2270 		if (scsi_status_is_good(the_result))
2271 			printk(KERN_CONT "ready\n");
2272 		else
2273 			printk(KERN_CONT "not responding...\n");
2274 	}
2275 }
2276 
2277 /*
2278  * Determine whether disk supports Data Integrity Field.
2279  */
2280 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2281 {
2282 	struct scsi_device *sdp = sdkp->device;
2283 	u8 type;
2284 	int ret = 0;
2285 
2286 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2287 		sdkp->protection_type = 0;
2288 		return ret;
2289 	}
2290 
2291 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2292 
2293 	if (type > T10_PI_TYPE3_PROTECTION)
2294 		ret = -ENODEV;
2295 	else if (scsi_host_dif_capable(sdp->host, type))
2296 		ret = 1;
2297 
2298 	if (sdkp->first_scan || type != sdkp->protection_type)
2299 		switch (ret) {
2300 		case -ENODEV:
2301 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2302 				  " protection type %u. Disabling disk!\n",
2303 				  type);
2304 			break;
2305 		case 1:
2306 			sd_printk(KERN_NOTICE, sdkp,
2307 				  "Enabling DIF Type %u protection\n", type);
2308 			break;
2309 		case 0:
2310 			sd_printk(KERN_NOTICE, sdkp,
2311 				  "Disabling DIF Type %u protection\n", type);
2312 			break;
2313 		}
2314 
2315 	sdkp->protection_type = type;
2316 
2317 	return ret;
2318 }
2319 
2320 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2321 			struct scsi_sense_hdr *sshdr, int sense_valid,
2322 			int the_result)
2323 {
2324 	if (sense_valid)
2325 		sd_print_sense_hdr(sdkp, sshdr);
2326 	else
2327 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2328 
2329 	/*
2330 	 * Set dirty bit for removable devices if not ready -
2331 	 * sometimes drives will not report this properly.
2332 	 */
2333 	if (sdp->removable &&
2334 	    sense_valid && sshdr->sense_key == NOT_READY)
2335 		set_media_not_present(sdkp);
2336 
2337 	/*
2338 	 * We used to set media_present to 0 here to indicate no media
2339 	 * in the drive, but some drives fail read capacity even with
2340 	 * media present, so we can't do that.
2341 	 */
2342 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2343 }
2344 
2345 #define RC16_LEN 32
2346 #if RC16_LEN > SD_BUF_SIZE
2347 #error RC16_LEN must not be more than SD_BUF_SIZE
2348 #endif
2349 
2350 #define READ_CAPACITY_RETRIES_ON_RESET	10
2351 
2352 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2353 						unsigned char *buffer)
2354 {
2355 	unsigned char cmd[16];
2356 	struct scsi_sense_hdr sshdr;
2357 	int sense_valid = 0;
2358 	int the_result;
2359 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2360 	unsigned int alignment;
2361 	unsigned long long lba;
2362 	unsigned sector_size;
2363 
2364 	if (sdp->no_read_capacity_16)
2365 		return -EINVAL;
2366 
2367 	do {
2368 		memset(cmd, 0, 16);
2369 		cmd[0] = SERVICE_ACTION_IN_16;
2370 		cmd[1] = SAI_READ_CAPACITY_16;
2371 		cmd[13] = RC16_LEN;
2372 		memset(buffer, 0, RC16_LEN);
2373 
2374 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2375 					buffer, RC16_LEN, &sshdr,
2376 					SD_TIMEOUT, sdkp->max_retries, NULL);
2377 
2378 		if (media_not_present(sdkp, &sshdr))
2379 			return -ENODEV;
2380 
2381 		if (the_result > 0) {
2382 			sense_valid = scsi_sense_valid(&sshdr);
2383 			if (sense_valid &&
2384 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2385 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2386 			    sshdr.ascq == 0x00)
2387 				/* Invalid Command Operation Code or
2388 				 * Invalid Field in CDB, just retry
2389 				 * silently with RC10 */
2390 				return -EINVAL;
2391 			if (sense_valid &&
2392 			    sshdr.sense_key == UNIT_ATTENTION &&
2393 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2394 				/* Device reset might occur several times,
2395 				 * give it one more chance */
2396 				if (--reset_retries > 0)
2397 					continue;
2398 		}
2399 		retries--;
2400 
2401 	} while (the_result && retries);
2402 
2403 	if (the_result) {
2404 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2405 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2406 		return -EINVAL;
2407 	}
2408 
2409 	sector_size = get_unaligned_be32(&buffer[8]);
2410 	lba = get_unaligned_be64(&buffer[0]);
2411 
2412 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2413 		sdkp->capacity = 0;
2414 		return -ENODEV;
2415 	}
2416 
2417 	/* Logical blocks per physical block exponent */
2418 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2419 
2420 	/* RC basis */
2421 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2422 
2423 	/* Lowest aligned logical block */
2424 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2425 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2426 	if (alignment && sdkp->first_scan)
2427 		sd_printk(KERN_NOTICE, sdkp,
2428 			  "physical block alignment offset: %u\n", alignment);
2429 
2430 	if (buffer[14] & 0x80) { /* LBPME */
2431 		sdkp->lbpme = 1;
2432 
2433 		if (buffer[14] & 0x40) /* LBPRZ */
2434 			sdkp->lbprz = 1;
2435 
2436 		sd_config_discard(sdkp, SD_LBP_WS16);
2437 	}
2438 
2439 	sdkp->capacity = lba + 1;
2440 	return sector_size;
2441 }
2442 
2443 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2444 						unsigned char *buffer)
2445 {
2446 	unsigned char cmd[16];
2447 	struct scsi_sense_hdr sshdr;
2448 	int sense_valid = 0;
2449 	int the_result;
2450 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2451 	sector_t lba;
2452 	unsigned sector_size;
2453 
2454 	do {
2455 		cmd[0] = READ_CAPACITY;
2456 		memset(&cmd[1], 0, 9);
2457 		memset(buffer, 0, 8);
2458 
2459 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2460 					buffer, 8, &sshdr,
2461 					SD_TIMEOUT, sdkp->max_retries, NULL);
2462 
2463 		if (media_not_present(sdkp, &sshdr))
2464 			return -ENODEV;
2465 
2466 		if (the_result > 0) {
2467 			sense_valid = scsi_sense_valid(&sshdr);
2468 			if (sense_valid &&
2469 			    sshdr.sense_key == UNIT_ATTENTION &&
2470 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2471 				/* Device reset might occur several times,
2472 				 * give it one more chance */
2473 				if (--reset_retries > 0)
2474 					continue;
2475 		}
2476 		retries--;
2477 
2478 	} while (the_result && retries);
2479 
2480 	if (the_result) {
2481 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2482 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2483 		return -EINVAL;
2484 	}
2485 
2486 	sector_size = get_unaligned_be32(&buffer[4]);
2487 	lba = get_unaligned_be32(&buffer[0]);
2488 
2489 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2490 		/* Some buggy (usb cardreader) devices return an lba of
2491 		   0xffffffff when the want to report a size of 0 (with
2492 		   which they really mean no media is present) */
2493 		sdkp->capacity = 0;
2494 		sdkp->physical_block_size = sector_size;
2495 		return sector_size;
2496 	}
2497 
2498 	sdkp->capacity = lba + 1;
2499 	sdkp->physical_block_size = sector_size;
2500 	return sector_size;
2501 }
2502 
2503 static int sd_try_rc16_first(struct scsi_device *sdp)
2504 {
2505 	if (sdp->host->max_cmd_len < 16)
2506 		return 0;
2507 	if (sdp->try_rc_10_first)
2508 		return 0;
2509 	if (sdp->scsi_level > SCSI_SPC_2)
2510 		return 1;
2511 	if (scsi_device_protection(sdp))
2512 		return 1;
2513 	return 0;
2514 }
2515 
2516 /*
2517  * read disk capacity
2518  */
2519 static void
2520 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2521 {
2522 	int sector_size;
2523 	struct scsi_device *sdp = sdkp->device;
2524 
2525 	if (sd_try_rc16_first(sdp)) {
2526 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2527 		if (sector_size == -EOVERFLOW)
2528 			goto got_data;
2529 		if (sector_size == -ENODEV)
2530 			return;
2531 		if (sector_size < 0)
2532 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2533 		if (sector_size < 0)
2534 			return;
2535 	} else {
2536 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2537 		if (sector_size == -EOVERFLOW)
2538 			goto got_data;
2539 		if (sector_size < 0)
2540 			return;
2541 		if ((sizeof(sdkp->capacity) > 4) &&
2542 		    (sdkp->capacity > 0xffffffffULL)) {
2543 			int old_sector_size = sector_size;
2544 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2545 					"Trying to use READ CAPACITY(16).\n");
2546 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2547 			if (sector_size < 0) {
2548 				sd_printk(KERN_NOTICE, sdkp,
2549 					"Using 0xffffffff as device size\n");
2550 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2551 				sector_size = old_sector_size;
2552 				goto got_data;
2553 			}
2554 			/* Remember that READ CAPACITY(16) succeeded */
2555 			sdp->try_rc_10_first = 0;
2556 		}
2557 	}
2558 
2559 	/* Some devices are known to return the total number of blocks,
2560 	 * not the highest block number.  Some devices have versions
2561 	 * which do this and others which do not.  Some devices we might
2562 	 * suspect of doing this but we don't know for certain.
2563 	 *
2564 	 * If we know the reported capacity is wrong, decrement it.  If
2565 	 * we can only guess, then assume the number of blocks is even
2566 	 * (usually true but not always) and err on the side of lowering
2567 	 * the capacity.
2568 	 */
2569 	if (sdp->fix_capacity ||
2570 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2571 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2572 				"from its reported value: %llu\n",
2573 				(unsigned long long) sdkp->capacity);
2574 		--sdkp->capacity;
2575 	}
2576 
2577 got_data:
2578 	if (sector_size == 0) {
2579 		sector_size = 512;
2580 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2581 			  "assuming 512.\n");
2582 	}
2583 
2584 	if (sector_size != 512 &&
2585 	    sector_size != 1024 &&
2586 	    sector_size != 2048 &&
2587 	    sector_size != 4096) {
2588 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2589 			  sector_size);
2590 		/*
2591 		 * The user might want to re-format the drive with
2592 		 * a supported sectorsize.  Once this happens, it
2593 		 * would be relatively trivial to set the thing up.
2594 		 * For this reason, we leave the thing in the table.
2595 		 */
2596 		sdkp->capacity = 0;
2597 		/*
2598 		 * set a bogus sector size so the normal read/write
2599 		 * logic in the block layer will eventually refuse any
2600 		 * request on this device without tripping over power
2601 		 * of two sector size assumptions
2602 		 */
2603 		sector_size = 512;
2604 	}
2605 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2606 	blk_queue_physical_block_size(sdp->request_queue,
2607 				      sdkp->physical_block_size);
2608 	sdkp->device->sector_size = sector_size;
2609 
2610 	if (sdkp->capacity > 0xffffffff)
2611 		sdp->use_16_for_rw = 1;
2612 
2613 }
2614 
2615 /*
2616  * Print disk capacity
2617  */
2618 static void
2619 sd_print_capacity(struct scsi_disk *sdkp,
2620 		  sector_t old_capacity)
2621 {
2622 	int sector_size = sdkp->device->sector_size;
2623 	char cap_str_2[10], cap_str_10[10];
2624 
2625 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2626 		return;
2627 
2628 	string_get_size(sdkp->capacity, sector_size,
2629 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2630 	string_get_size(sdkp->capacity, sector_size,
2631 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2632 
2633 	sd_printk(KERN_NOTICE, sdkp,
2634 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2635 		  (unsigned long long)sdkp->capacity,
2636 		  sector_size, cap_str_10, cap_str_2);
2637 
2638 	if (sdkp->physical_block_size != sector_size)
2639 		sd_printk(KERN_NOTICE, sdkp,
2640 			  "%u-byte physical blocks\n",
2641 			  sdkp->physical_block_size);
2642 }
2643 
2644 /* called with buffer of length 512 */
2645 static inline int
2646 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2647 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2648 		 struct scsi_sense_hdr *sshdr)
2649 {
2650 	/*
2651 	 * If we must use MODE SENSE(10), make sure that the buffer length
2652 	 * is at least 8 bytes so that the mode sense header fits.
2653 	 */
2654 	if (sdkp->device->use_10_for_ms && len < 8)
2655 		len = 8;
2656 
2657 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2658 			       SD_TIMEOUT, sdkp->max_retries, data,
2659 			       sshdr);
2660 }
2661 
2662 /*
2663  * read write protect setting, if possible - called only in sd_revalidate_disk()
2664  * called with buffer of length SD_BUF_SIZE
2665  */
2666 static void
2667 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2668 {
2669 	int res;
2670 	struct scsi_device *sdp = sdkp->device;
2671 	struct scsi_mode_data data;
2672 	int old_wp = sdkp->write_prot;
2673 
2674 	set_disk_ro(sdkp->disk, 0);
2675 	if (sdp->skip_ms_page_3f) {
2676 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2677 		return;
2678 	}
2679 
2680 	if (sdp->use_192_bytes_for_3f) {
2681 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2682 	} else {
2683 		/*
2684 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2685 		 * We have to start carefully: some devices hang if we ask
2686 		 * for more than is available.
2687 		 */
2688 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2689 
2690 		/*
2691 		 * Second attempt: ask for page 0 When only page 0 is
2692 		 * implemented, a request for page 3F may return Sense Key
2693 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2694 		 * CDB.
2695 		 */
2696 		if (res < 0)
2697 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2698 
2699 		/*
2700 		 * Third attempt: ask 255 bytes, as we did earlier.
2701 		 */
2702 		if (res < 0)
2703 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2704 					       &data, NULL);
2705 	}
2706 
2707 	if (res < 0) {
2708 		sd_first_printk(KERN_WARNING, sdkp,
2709 			  "Test WP failed, assume Write Enabled\n");
2710 	} else {
2711 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2712 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2713 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2714 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2715 				  sdkp->write_prot ? "on" : "off");
2716 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2717 		}
2718 	}
2719 }
2720 
2721 /*
2722  * sd_read_cache_type - called only from sd_revalidate_disk()
2723  * called with buffer of length SD_BUF_SIZE
2724  */
2725 static void
2726 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2727 {
2728 	int len = 0, res;
2729 	struct scsi_device *sdp = sdkp->device;
2730 
2731 	int dbd;
2732 	int modepage;
2733 	int first_len;
2734 	struct scsi_mode_data data;
2735 	struct scsi_sense_hdr sshdr;
2736 	int old_wce = sdkp->WCE;
2737 	int old_rcd = sdkp->RCD;
2738 	int old_dpofua = sdkp->DPOFUA;
2739 
2740 
2741 	if (sdkp->cache_override)
2742 		return;
2743 
2744 	first_len = 4;
2745 	if (sdp->skip_ms_page_8) {
2746 		if (sdp->type == TYPE_RBC)
2747 			goto defaults;
2748 		else {
2749 			if (sdp->skip_ms_page_3f)
2750 				goto defaults;
2751 			modepage = 0x3F;
2752 			if (sdp->use_192_bytes_for_3f)
2753 				first_len = 192;
2754 			dbd = 0;
2755 		}
2756 	} else if (sdp->type == TYPE_RBC) {
2757 		modepage = 6;
2758 		dbd = 8;
2759 	} else {
2760 		modepage = 8;
2761 		dbd = 0;
2762 	}
2763 
2764 	/* cautiously ask */
2765 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2766 			&data, &sshdr);
2767 
2768 	if (res < 0)
2769 		goto bad_sense;
2770 
2771 	if (!data.header_length) {
2772 		modepage = 6;
2773 		first_len = 0;
2774 		sd_first_printk(KERN_ERR, sdkp,
2775 				"Missing header in MODE_SENSE response\n");
2776 	}
2777 
2778 	/* that went OK, now ask for the proper length */
2779 	len = data.length;
2780 
2781 	/*
2782 	 * We're only interested in the first three bytes, actually.
2783 	 * But the data cache page is defined for the first 20.
2784 	 */
2785 	if (len < 3)
2786 		goto bad_sense;
2787 	else if (len > SD_BUF_SIZE) {
2788 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2789 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2790 		len = SD_BUF_SIZE;
2791 	}
2792 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2793 		len = 192;
2794 
2795 	/* Get the data */
2796 	if (len > first_len)
2797 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2798 				&data, &sshdr);
2799 
2800 	if (!res) {
2801 		int offset = data.header_length + data.block_descriptor_length;
2802 
2803 		while (offset < len) {
2804 			u8 page_code = buffer[offset] & 0x3F;
2805 			u8 spf       = buffer[offset] & 0x40;
2806 
2807 			if (page_code == 8 || page_code == 6) {
2808 				/* We're interested only in the first 3 bytes.
2809 				 */
2810 				if (len - offset <= 2) {
2811 					sd_first_printk(KERN_ERR, sdkp,
2812 						"Incomplete mode parameter "
2813 							"data\n");
2814 					goto defaults;
2815 				} else {
2816 					modepage = page_code;
2817 					goto Page_found;
2818 				}
2819 			} else {
2820 				/* Go to the next page */
2821 				if (spf && len - offset > 3)
2822 					offset += 4 + (buffer[offset+2] << 8) +
2823 						buffer[offset+3];
2824 				else if (!spf && len - offset > 1)
2825 					offset += 2 + buffer[offset+1];
2826 				else {
2827 					sd_first_printk(KERN_ERR, sdkp,
2828 							"Incomplete mode "
2829 							"parameter data\n");
2830 					goto defaults;
2831 				}
2832 			}
2833 		}
2834 
2835 		sd_first_printk(KERN_WARNING, sdkp,
2836 				"No Caching mode page found\n");
2837 		goto defaults;
2838 
2839 	Page_found:
2840 		if (modepage == 8) {
2841 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2842 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2843 		} else {
2844 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2845 			sdkp->RCD = 0;
2846 		}
2847 
2848 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2849 		if (sdp->broken_fua) {
2850 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2851 			sdkp->DPOFUA = 0;
2852 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2853 			   !sdkp->device->use_16_for_rw) {
2854 			sd_first_printk(KERN_NOTICE, sdkp,
2855 				  "Uses READ/WRITE(6), disabling FUA\n");
2856 			sdkp->DPOFUA = 0;
2857 		}
2858 
2859 		/* No cache flush allowed for write protected devices */
2860 		if (sdkp->WCE && sdkp->write_prot)
2861 			sdkp->WCE = 0;
2862 
2863 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2864 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2865 			sd_printk(KERN_NOTICE, sdkp,
2866 				  "Write cache: %s, read cache: %s, %s\n",
2867 				  sdkp->WCE ? "enabled" : "disabled",
2868 				  sdkp->RCD ? "disabled" : "enabled",
2869 				  sdkp->DPOFUA ? "supports DPO and FUA"
2870 				  : "doesn't support DPO or FUA");
2871 
2872 		return;
2873 	}
2874 
2875 bad_sense:
2876 	if (scsi_sense_valid(&sshdr) &&
2877 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2878 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2879 		/* Invalid field in CDB */
2880 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2881 	else
2882 		sd_first_printk(KERN_ERR, sdkp,
2883 				"Asking for cache data failed\n");
2884 
2885 defaults:
2886 	if (sdp->wce_default_on) {
2887 		sd_first_printk(KERN_NOTICE, sdkp,
2888 				"Assuming drive cache: write back\n");
2889 		sdkp->WCE = 1;
2890 	} else {
2891 		sd_first_printk(KERN_WARNING, sdkp,
2892 				"Assuming drive cache: write through\n");
2893 		sdkp->WCE = 0;
2894 	}
2895 	sdkp->RCD = 0;
2896 	sdkp->DPOFUA = 0;
2897 }
2898 
2899 /*
2900  * The ATO bit indicates whether the DIF application tag is available
2901  * for use by the operating system.
2902  */
2903 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2904 {
2905 	int res, offset;
2906 	struct scsi_device *sdp = sdkp->device;
2907 	struct scsi_mode_data data;
2908 	struct scsi_sense_hdr sshdr;
2909 
2910 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2911 		return;
2912 
2913 	if (sdkp->protection_type == 0)
2914 		return;
2915 
2916 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2917 			      sdkp->max_retries, &data, &sshdr);
2918 
2919 	if (res < 0 || !data.header_length ||
2920 	    data.length < 6) {
2921 		sd_first_printk(KERN_WARNING, sdkp,
2922 			  "getting Control mode page failed, assume no ATO\n");
2923 
2924 		if (scsi_sense_valid(&sshdr))
2925 			sd_print_sense_hdr(sdkp, &sshdr);
2926 
2927 		return;
2928 	}
2929 
2930 	offset = data.header_length + data.block_descriptor_length;
2931 
2932 	if ((buffer[offset] & 0x3f) != 0x0a) {
2933 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2934 		return;
2935 	}
2936 
2937 	if ((buffer[offset + 5] & 0x80) == 0)
2938 		return;
2939 
2940 	sdkp->ATO = 1;
2941 
2942 	return;
2943 }
2944 
2945 /**
2946  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2947  * @sdkp: disk to query
2948  */
2949 static void sd_read_block_limits(struct scsi_disk *sdkp)
2950 {
2951 	unsigned int sector_sz = sdkp->device->sector_size;
2952 	const int vpd_len = 64;
2953 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2954 
2955 	if (!buffer ||
2956 	    /* Block Limits VPD */
2957 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2958 		goto out;
2959 
2960 	blk_queue_io_min(sdkp->disk->queue,
2961 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2962 
2963 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2964 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2965 
2966 	if (buffer[3] == 0x3c) {
2967 		unsigned int lba_count, desc_count;
2968 
2969 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2970 
2971 		if (!sdkp->lbpme)
2972 			goto out;
2973 
2974 		lba_count = get_unaligned_be32(&buffer[20]);
2975 		desc_count = get_unaligned_be32(&buffer[24]);
2976 
2977 		if (lba_count && desc_count)
2978 			sdkp->max_unmap_blocks = lba_count;
2979 
2980 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2981 
2982 		if (buffer[32] & 0x80)
2983 			sdkp->unmap_alignment =
2984 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2985 
2986 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2987 
2988 			if (sdkp->max_unmap_blocks)
2989 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2990 			else
2991 				sd_config_discard(sdkp, SD_LBP_WS16);
2992 
2993 		} else {	/* LBP VPD page tells us what to use */
2994 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2995 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2996 			else if (sdkp->lbpws)
2997 				sd_config_discard(sdkp, SD_LBP_WS16);
2998 			else if (sdkp->lbpws10)
2999 				sd_config_discard(sdkp, SD_LBP_WS10);
3000 			else
3001 				sd_config_discard(sdkp, SD_LBP_DISABLE);
3002 		}
3003 	}
3004 
3005  out:
3006 	kfree(buffer);
3007 }
3008 
3009 /**
3010  * sd_read_block_characteristics - Query block dev. characteristics
3011  * @sdkp: disk to query
3012  */
3013 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3014 {
3015 	struct request_queue *q = sdkp->disk->queue;
3016 	unsigned char *buffer;
3017 	u16 rot;
3018 	const int vpd_len = 64;
3019 
3020 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3021 
3022 	if (!buffer ||
3023 	    /* Block Device Characteristics VPD */
3024 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
3025 		goto out;
3026 
3027 	rot = get_unaligned_be16(&buffer[4]);
3028 
3029 	if (rot == 1) {
3030 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3031 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3032 	}
3033 
3034 	if (sdkp->device->type == TYPE_ZBC) {
3035 		/* Host-managed */
3036 		blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3037 	} else {
3038 		sdkp->zoned = (buffer[8] >> 4) & 3;
3039 		if (sdkp->zoned == 1) {
3040 			/* Host-aware */
3041 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3042 		} else {
3043 			/* Regular disk or drive managed disk */
3044 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3045 		}
3046 	}
3047 
3048 	if (!sdkp->first_scan)
3049 		goto out;
3050 
3051 	if (blk_queue_is_zoned(q)) {
3052 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3053 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3054 	} else {
3055 		if (sdkp->zoned == 1)
3056 			sd_printk(KERN_NOTICE, sdkp,
3057 				  "Host-aware SMR disk used as regular disk\n");
3058 		else if (sdkp->zoned == 2)
3059 			sd_printk(KERN_NOTICE, sdkp,
3060 				  "Drive-managed SMR disk\n");
3061 	}
3062 
3063  out:
3064 	kfree(buffer);
3065 }
3066 
3067 /**
3068  * sd_read_block_provisioning - Query provisioning VPD page
3069  * @sdkp: disk to query
3070  */
3071 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3072 {
3073 	unsigned char *buffer;
3074 	const int vpd_len = 8;
3075 
3076 	if (sdkp->lbpme == 0)
3077 		return;
3078 
3079 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3080 
3081 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3082 		goto out;
3083 
3084 	sdkp->lbpvpd	= 1;
3085 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
3086 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
3087 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
3088 
3089  out:
3090 	kfree(buffer);
3091 }
3092 
3093 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3094 {
3095 	struct scsi_device *sdev = sdkp->device;
3096 
3097 	if (sdev->host->no_write_same) {
3098 		sdev->no_write_same = 1;
3099 
3100 		return;
3101 	}
3102 
3103 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3104 		/* too large values might cause issues with arcmsr */
3105 		int vpd_buf_len = 64;
3106 
3107 		sdev->no_report_opcodes = 1;
3108 
3109 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3110 		 * CODES is unsupported and the device has an ATA
3111 		 * Information VPD page (SAT).
3112 		 */
3113 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3114 			sdev->no_write_same = 1;
3115 	}
3116 
3117 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3118 		sdkp->ws16 = 1;
3119 
3120 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3121 		sdkp->ws10 = 1;
3122 }
3123 
3124 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3125 {
3126 	struct scsi_device *sdev = sdkp->device;
3127 
3128 	if (!sdev->security_supported)
3129 		return;
3130 
3131 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3132 			SECURITY_PROTOCOL_IN) == 1 &&
3133 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3134 			SECURITY_PROTOCOL_OUT) == 1)
3135 		sdkp->security = 1;
3136 }
3137 
3138 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3139 {
3140 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3141 }
3142 
3143 /**
3144  * sd_read_cpr - Query concurrent positioning ranges
3145  * @sdkp:	disk to query
3146  */
3147 static void sd_read_cpr(struct scsi_disk *sdkp)
3148 {
3149 	struct blk_independent_access_ranges *iars = NULL;
3150 	unsigned char *buffer = NULL;
3151 	unsigned int nr_cpr = 0;
3152 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3153 	u8 *desc;
3154 
3155 	/*
3156 	 * We need to have the capacity set first for the block layer to be
3157 	 * able to check the ranges.
3158 	 */
3159 	if (sdkp->first_scan)
3160 		return;
3161 
3162 	if (!sdkp->capacity)
3163 		goto out;
3164 
3165 	/*
3166 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3167 	 * leading to a maximum page size of 64 + 256*32 bytes.
3168 	 */
3169 	buf_len = 64 + 256*32;
3170 	buffer = kmalloc(buf_len, GFP_KERNEL);
3171 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3172 		goto out;
3173 
3174 	/* We must have at least a 64B header and one 32B range descriptor */
3175 	vpd_len = get_unaligned_be16(&buffer[2]) + 3;
3176 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3177 		sd_printk(KERN_ERR, sdkp,
3178 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3179 		goto out;
3180 	}
3181 
3182 	nr_cpr = (vpd_len - 64) / 32;
3183 	if (nr_cpr == 1) {
3184 		nr_cpr = 0;
3185 		goto out;
3186 	}
3187 
3188 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3189 	if (!iars) {
3190 		nr_cpr = 0;
3191 		goto out;
3192 	}
3193 
3194 	desc = &buffer[64];
3195 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3196 		if (desc[0] != i) {
3197 			sd_printk(KERN_ERR, sdkp,
3198 				"Invalid Concurrent Positioning Range number\n");
3199 			nr_cpr = 0;
3200 			break;
3201 		}
3202 
3203 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3204 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3205 	}
3206 
3207 out:
3208 	disk_set_independent_access_ranges(sdkp->disk, iars);
3209 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3210 		sd_printk(KERN_NOTICE, sdkp,
3211 			  "%u concurrent positioning ranges\n", nr_cpr);
3212 		sdkp->nr_actuators = nr_cpr;
3213 	}
3214 
3215 	kfree(buffer);
3216 }
3217 
3218 /*
3219  * Determine the device's preferred I/O size for reads and writes
3220  * unless the reported value is unreasonably small, large, not a
3221  * multiple of the physical block size, or simply garbage.
3222  */
3223 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3224 				      unsigned int dev_max)
3225 {
3226 	struct scsi_device *sdp = sdkp->device;
3227 	unsigned int opt_xfer_bytes =
3228 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3229 
3230 	if (sdkp->opt_xfer_blocks == 0)
3231 		return false;
3232 
3233 	if (sdkp->opt_xfer_blocks > dev_max) {
3234 		sd_first_printk(KERN_WARNING, sdkp,
3235 				"Optimal transfer size %u logical blocks " \
3236 				"> dev_max (%u logical blocks)\n",
3237 				sdkp->opt_xfer_blocks, dev_max);
3238 		return false;
3239 	}
3240 
3241 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3242 		sd_first_printk(KERN_WARNING, sdkp,
3243 				"Optimal transfer size %u logical blocks " \
3244 				"> sd driver limit (%u logical blocks)\n",
3245 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3246 		return false;
3247 	}
3248 
3249 	if (opt_xfer_bytes < PAGE_SIZE) {
3250 		sd_first_printk(KERN_WARNING, sdkp,
3251 				"Optimal transfer size %u bytes < " \
3252 				"PAGE_SIZE (%u bytes)\n",
3253 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3254 		return false;
3255 	}
3256 
3257 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3258 		sd_first_printk(KERN_WARNING, sdkp,
3259 				"Optimal transfer size %u bytes not a " \
3260 				"multiple of physical block size (%u bytes)\n",
3261 				opt_xfer_bytes, sdkp->physical_block_size);
3262 		return false;
3263 	}
3264 
3265 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3266 			opt_xfer_bytes);
3267 	return true;
3268 }
3269 
3270 /**
3271  *	sd_revalidate_disk - called the first time a new disk is seen,
3272  *	performs disk spin up, read_capacity, etc.
3273  *	@disk: struct gendisk we care about
3274  **/
3275 static int sd_revalidate_disk(struct gendisk *disk)
3276 {
3277 	struct scsi_disk *sdkp = scsi_disk(disk);
3278 	struct scsi_device *sdp = sdkp->device;
3279 	struct request_queue *q = sdkp->disk->queue;
3280 	sector_t old_capacity = sdkp->capacity;
3281 	unsigned char *buffer;
3282 	unsigned int dev_max, rw_max;
3283 
3284 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3285 				      "sd_revalidate_disk\n"));
3286 
3287 	/*
3288 	 * If the device is offline, don't try and read capacity or any
3289 	 * of the other niceties.
3290 	 */
3291 	if (!scsi_device_online(sdp))
3292 		goto out;
3293 
3294 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3295 	if (!buffer) {
3296 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3297 			  "allocation failure.\n");
3298 		goto out;
3299 	}
3300 
3301 	sd_spinup_disk(sdkp);
3302 
3303 	/*
3304 	 * Without media there is no reason to ask; moreover, some devices
3305 	 * react badly if we do.
3306 	 */
3307 	if (sdkp->media_present) {
3308 		sd_read_capacity(sdkp, buffer);
3309 
3310 		/*
3311 		 * set the default to rotational.  All non-rotational devices
3312 		 * support the block characteristics VPD page, which will
3313 		 * cause this to be updated correctly and any device which
3314 		 * doesn't support it should be treated as rotational.
3315 		 */
3316 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3317 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3318 
3319 		if (scsi_device_supports_vpd(sdp)) {
3320 			sd_read_block_provisioning(sdkp);
3321 			sd_read_block_limits(sdkp);
3322 			sd_read_block_characteristics(sdkp);
3323 			sd_zbc_read_zones(sdkp, buffer);
3324 		}
3325 
3326 		sd_print_capacity(sdkp, old_capacity);
3327 
3328 		sd_read_write_protect_flag(sdkp, buffer);
3329 		sd_read_cache_type(sdkp, buffer);
3330 		sd_read_app_tag_own(sdkp, buffer);
3331 		sd_read_write_same(sdkp, buffer);
3332 		sd_read_security(sdkp, buffer);
3333 		sd_read_cpr(sdkp);
3334 	}
3335 
3336 	/*
3337 	 * We now have all cache related info, determine how we deal
3338 	 * with flush requests.
3339 	 */
3340 	sd_set_flush_flag(sdkp);
3341 
3342 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3343 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3344 
3345 	/* Some devices report a maximum block count for READ/WRITE requests. */
3346 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3347 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
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 	/* Do not exceed controller limit */
3359 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3360 
3361 	/*
3362 	 * Only update max_sectors if previously unset or if the current value
3363 	 * exceeds the capabilities of the hardware.
3364 	 */
3365 	if (sdkp->first_scan ||
3366 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3367 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3368 		q->limits.max_sectors = rw_max;
3369 
3370 	sdkp->first_scan = 0;
3371 
3372 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3373 	sd_config_write_same(sdkp);
3374 	kfree(buffer);
3375 
3376 	/*
3377 	 * For a zoned drive, revalidating the zones can be done only once
3378 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3379 	 * capacity to 0.
3380 	 */
3381 	if (sd_zbc_revalidate_zones(sdkp))
3382 		set_capacity_and_notify(disk, 0);
3383 
3384  out:
3385 	return 0;
3386 }
3387 
3388 /**
3389  *	sd_unlock_native_capacity - unlock native capacity
3390  *	@disk: struct gendisk to set capacity for
3391  *
3392  *	Block layer calls this function if it detects that partitions
3393  *	on @disk reach beyond the end of the device.  If the SCSI host
3394  *	implements ->unlock_native_capacity() method, it's invoked to
3395  *	give it a chance to adjust the device capacity.
3396  *
3397  *	CONTEXT:
3398  *	Defined by block layer.  Might sleep.
3399  */
3400 static void sd_unlock_native_capacity(struct gendisk *disk)
3401 {
3402 	struct scsi_device *sdev = scsi_disk(disk)->device;
3403 
3404 	if (sdev->host->hostt->unlock_native_capacity)
3405 		sdev->host->hostt->unlock_native_capacity(sdev);
3406 }
3407 
3408 /**
3409  *	sd_format_disk_name - format disk name
3410  *	@prefix: name prefix - ie. "sd" for SCSI disks
3411  *	@index: index of the disk to format name for
3412  *	@buf: output buffer
3413  *	@buflen: length of the output buffer
3414  *
3415  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3416  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3417  *	which is followed by sdaaa.
3418  *
3419  *	This is basically 26 base counting with one extra 'nil' entry
3420  *	at the beginning from the second digit on and can be
3421  *	determined using similar method as 26 base conversion with the
3422  *	index shifted -1 after each digit is computed.
3423  *
3424  *	CONTEXT:
3425  *	Don't care.
3426  *
3427  *	RETURNS:
3428  *	0 on success, -errno on failure.
3429  */
3430 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3431 {
3432 	const int base = 'z' - 'a' + 1;
3433 	char *begin = buf + strlen(prefix);
3434 	char *end = buf + buflen;
3435 	char *p;
3436 	int unit;
3437 
3438 	p = end - 1;
3439 	*p = '\0';
3440 	unit = base;
3441 	do {
3442 		if (p == begin)
3443 			return -EINVAL;
3444 		*--p = 'a' + (index % unit);
3445 		index = (index / unit) - 1;
3446 	} while (index >= 0);
3447 
3448 	memmove(begin, p, end - p);
3449 	memcpy(buf, prefix, strlen(prefix));
3450 
3451 	return 0;
3452 }
3453 
3454 /**
3455  *	sd_probe - called during driver initialization and whenever a
3456  *	new scsi device is attached to the system. It is called once
3457  *	for each scsi device (not just disks) present.
3458  *	@dev: pointer to device object
3459  *
3460  *	Returns 0 if successful (or not interested in this scsi device
3461  *	(e.g. scanner)); 1 when there is an error.
3462  *
3463  *	Note: this function is invoked from the scsi mid-level.
3464  *	This function sets up the mapping between a given
3465  *	<host,channel,id,lun> (found in sdp) and new device name
3466  *	(e.g. /dev/sda). More precisely it is the block device major
3467  *	and minor number that is chosen here.
3468  *
3469  *	Assume sd_probe is not re-entrant (for time being)
3470  *	Also think about sd_probe() and sd_remove() running coincidentally.
3471  **/
3472 static int sd_probe(struct device *dev)
3473 {
3474 	struct scsi_device *sdp = to_scsi_device(dev);
3475 	struct scsi_disk *sdkp;
3476 	struct gendisk *gd;
3477 	int index;
3478 	int error;
3479 
3480 	scsi_autopm_get_device(sdp);
3481 	error = -ENODEV;
3482 	if (sdp->type != TYPE_DISK &&
3483 	    sdp->type != TYPE_ZBC &&
3484 	    sdp->type != TYPE_MOD &&
3485 	    sdp->type != TYPE_RBC)
3486 		goto out;
3487 
3488 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3489 		sdev_printk(KERN_WARNING, sdp,
3490 			    "Unsupported ZBC host-managed device.\n");
3491 		goto out;
3492 	}
3493 
3494 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3495 					"sd_probe\n"));
3496 
3497 	error = -ENOMEM;
3498 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3499 	if (!sdkp)
3500 		goto out;
3501 
3502 	gd = __alloc_disk_node(sdp->request_queue, NUMA_NO_NODE,
3503 			       &sd_bio_compl_lkclass);
3504 	if (!gd)
3505 		goto out_free;
3506 
3507 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3508 	if (index < 0) {
3509 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3510 		goto out_put;
3511 	}
3512 
3513 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3514 	if (error) {
3515 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3516 		goto out_free_index;
3517 	}
3518 
3519 	sdkp->device = sdp;
3520 	sdkp->driver = &sd_template;
3521 	sdkp->disk = gd;
3522 	sdkp->index = index;
3523 	sdkp->max_retries = SD_MAX_RETRIES;
3524 	atomic_set(&sdkp->openers, 0);
3525 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3526 
3527 	if (!sdp->request_queue->rq_timeout) {
3528 		if (sdp->type != TYPE_MOD)
3529 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3530 		else
3531 			blk_queue_rq_timeout(sdp->request_queue,
3532 					     SD_MOD_TIMEOUT);
3533 	}
3534 
3535 	device_initialize(&sdkp->dev);
3536 	sdkp->dev.parent = get_device(dev);
3537 	sdkp->dev.class = &sd_disk_class;
3538 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3539 
3540 	error = device_add(&sdkp->dev);
3541 	if (error) {
3542 		put_device(&sdkp->dev);
3543 		goto out;
3544 	}
3545 
3546 	dev_set_drvdata(dev, sdkp);
3547 
3548 	gd->major = sd_major((index & 0xf0) >> 4);
3549 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3550 	gd->minors = SD_MINORS;
3551 
3552 	gd->fops = &sd_fops;
3553 	gd->private_data = &sdkp->driver;
3554 
3555 	/* defaults, until the device tells us otherwise */
3556 	sdp->sector_size = 512;
3557 	sdkp->capacity = 0;
3558 	sdkp->media_present = 1;
3559 	sdkp->write_prot = 0;
3560 	sdkp->cache_override = 0;
3561 	sdkp->WCE = 0;
3562 	sdkp->RCD = 0;
3563 	sdkp->ATO = 0;
3564 	sdkp->first_scan = 1;
3565 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3566 
3567 	sd_revalidate_disk(gd);
3568 
3569 	gd->flags = GENHD_FL_EXT_DEVT;
3570 	if (sdp->removable) {
3571 		gd->flags |= GENHD_FL_REMOVABLE;
3572 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3573 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3574 	}
3575 
3576 	blk_pm_runtime_init(sdp->request_queue, dev);
3577 	if (sdp->rpm_autosuspend) {
3578 		pm_runtime_set_autosuspend_delay(dev,
3579 			sdp->host->hostt->rpm_autosuspend_delay);
3580 	}
3581 
3582 	error = device_add_disk(dev, gd, NULL);
3583 	if (error) {
3584 		put_device(&sdkp->dev);
3585 		goto out;
3586 	}
3587 
3588 	if (sdkp->capacity)
3589 		sd_dif_config_host(sdkp);
3590 
3591 	sd_revalidate_disk(gd);
3592 
3593 	if (sdkp->security) {
3594 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3595 		if (sdkp->opal_dev)
3596 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3597 	}
3598 
3599 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3600 		  sdp->removable ? "removable " : "");
3601 	scsi_autopm_put_device(sdp);
3602 
3603 	return 0;
3604 
3605  out_free_index:
3606 	ida_free(&sd_index_ida, index);
3607  out_put:
3608 	put_disk(gd);
3609  out_free:
3610 	sd_zbc_release_disk(sdkp);
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;
3631 
3632 	sdkp = dev_get_drvdata(dev);
3633 	scsi_autopm_get_device(sdkp->device);
3634 
3635 	device_del(&sdkp->dev);
3636 	del_gendisk(sdkp->disk);
3637 	sd_shutdown(dev);
3638 
3639 	free_opal_dev(sdkp->opal_dev);
3640 
3641 	mutex_lock(&sd_ref_mutex);
3642 	dev_set_drvdata(dev, NULL);
3643 	put_device(&sdkp->dev);
3644 	mutex_unlock(&sd_ref_mutex);
3645 
3646 	return 0;
3647 }
3648 
3649 /**
3650  *	scsi_disk_release - Called to free the scsi_disk structure
3651  *	@dev: pointer to embedded class device
3652  *
3653  *	sd_ref_mutex must be held entering this routine.  Because it is
3654  *	called on last put, you should always use the scsi_disk_get()
3655  *	scsi_disk_put() helpers which manipulate the semaphore directly
3656  *	and never do a direct put_device.
3657  **/
3658 static void scsi_disk_release(struct device *dev)
3659 {
3660 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3661 	struct gendisk *disk = sdkp->disk;
3662 	struct request_queue *q = disk->queue;
3663 
3664 	ida_free(&sd_index_ida, sdkp->index);
3665 
3666 	/*
3667 	 * Wait until all requests that are in progress have completed.
3668 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3669 	 * due to clearing the disk->private_data pointer. Wait from inside
3670 	 * scsi_disk_release() instead of from sd_release() to avoid that
3671 	 * freezing and unfreezing the request queue affects user space I/O
3672 	 * in case multiple processes open a /dev/sd... node concurrently.
3673 	 */
3674 	blk_mq_freeze_queue(q);
3675 	blk_mq_unfreeze_queue(q);
3676 
3677 	disk->private_data = NULL;
3678 	put_disk(disk);
3679 	put_device(&sdkp->device->sdev_gendev);
3680 
3681 	sd_zbc_release_disk(sdkp);
3682 
3683 	kfree(sdkp);
3684 }
3685 
3686 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3687 {
3688 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3689 	struct scsi_sense_hdr sshdr;
3690 	struct scsi_device *sdp = sdkp->device;
3691 	int res;
3692 
3693 	if (start)
3694 		cmd[4] |= 1;	/* START */
3695 
3696 	if (sdp->start_stop_pwr_cond)
3697 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3698 
3699 	if (!scsi_device_online(sdp))
3700 		return -ENODEV;
3701 
3702 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3703 			SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3704 	if (res) {
3705 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3706 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3707 			sd_print_sense_hdr(sdkp, &sshdr);
3708 			/* 0x3a is medium not present */
3709 			if (sshdr.asc == 0x3a)
3710 				res = 0;
3711 		}
3712 	}
3713 
3714 	/* SCSI error codes must not go to the generic layer */
3715 	if (res)
3716 		return -EIO;
3717 
3718 	return 0;
3719 }
3720 
3721 /*
3722  * Send a SYNCHRONIZE CACHE instruction down to the device through
3723  * the normal SCSI command structure.  Wait for the command to
3724  * complete.
3725  */
3726 static void sd_shutdown(struct device *dev)
3727 {
3728 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3729 
3730 	if (!sdkp)
3731 		return;         /* this can happen */
3732 
3733 	if (pm_runtime_suspended(dev))
3734 		return;
3735 
3736 	if (sdkp->WCE && sdkp->media_present) {
3737 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3738 		sd_sync_cache(sdkp, NULL);
3739 	}
3740 
3741 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3742 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3743 		sd_start_stop_device(sdkp, 0);
3744 	}
3745 }
3746 
3747 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3748 {
3749 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3750 	struct scsi_sense_hdr sshdr;
3751 	int ret = 0;
3752 
3753 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3754 		return 0;
3755 
3756 	if (sdkp->WCE && sdkp->media_present) {
3757 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3758 		ret = sd_sync_cache(sdkp, &sshdr);
3759 
3760 		if (ret) {
3761 			/* ignore OFFLINE device */
3762 			if (ret == -ENODEV)
3763 				return 0;
3764 
3765 			if (!scsi_sense_valid(&sshdr) ||
3766 			    sshdr.sense_key != ILLEGAL_REQUEST)
3767 				return ret;
3768 
3769 			/*
3770 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3771 			 * doesn't support sync. There's not much to do and
3772 			 * suspend shouldn't fail.
3773 			 */
3774 			ret = 0;
3775 		}
3776 	}
3777 
3778 	if (sdkp->device->manage_start_stop) {
3779 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3780 		/* an error is not worth aborting a system sleep */
3781 		ret = sd_start_stop_device(sdkp, 0);
3782 		if (ignore_stop_errors)
3783 			ret = 0;
3784 	}
3785 
3786 	return ret;
3787 }
3788 
3789 static int sd_suspend_system(struct device *dev)
3790 {
3791 	if (pm_runtime_suspended(dev))
3792 		return 0;
3793 
3794 	return sd_suspend_common(dev, true);
3795 }
3796 
3797 static int sd_suspend_runtime(struct device *dev)
3798 {
3799 	return sd_suspend_common(dev, false);
3800 }
3801 
3802 static int sd_resume(struct device *dev)
3803 {
3804 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3805 	int ret;
3806 
3807 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3808 		return 0;
3809 
3810 	if (!sdkp->device->manage_start_stop)
3811 		return 0;
3812 
3813 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3814 	ret = sd_start_stop_device(sdkp, 1);
3815 	if (!ret)
3816 		opal_unlock_from_suspend(sdkp->opal_dev);
3817 	return ret;
3818 }
3819 
3820 static int sd_resume_system(struct device *dev)
3821 {
3822 	if (pm_runtime_suspended(dev))
3823 		return 0;
3824 
3825 	return sd_resume(dev);
3826 }
3827 
3828 static int sd_resume_runtime(struct device *dev)
3829 {
3830 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3831 	struct scsi_device *sdp;
3832 
3833 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3834 		return 0;
3835 
3836 	sdp = sdkp->device;
3837 
3838 	if (sdp->ignore_media_change) {
3839 		/* clear the device's sense data */
3840 		static const u8 cmd[10] = { REQUEST_SENSE };
3841 
3842 		if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3843 				 NULL, sdp->request_queue->rq_timeout, 1, 0,
3844 				 RQF_PM, NULL))
3845 			sd_printk(KERN_NOTICE, sdkp,
3846 				  "Failed to clear sense data\n");
3847 	}
3848 
3849 	return sd_resume(dev);
3850 }
3851 
3852 /**
3853  *	init_sd - entry point for this driver (both when built in or when
3854  *	a module).
3855  *
3856  *	Note: this function registers this driver with the scsi mid-level.
3857  **/
3858 static int __init init_sd(void)
3859 {
3860 	int majors = 0, i, err;
3861 
3862 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3863 
3864 	for (i = 0; i < SD_MAJORS; i++) {
3865 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3866 			continue;
3867 		majors++;
3868 	}
3869 
3870 	if (!majors)
3871 		return -ENODEV;
3872 
3873 	err = class_register(&sd_disk_class);
3874 	if (err)
3875 		goto err_out;
3876 
3877 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3878 					 0, 0, NULL);
3879 	if (!sd_cdb_cache) {
3880 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3881 		err = -ENOMEM;
3882 		goto err_out_class;
3883 	}
3884 
3885 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3886 	if (!sd_cdb_pool) {
3887 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3888 		err = -ENOMEM;
3889 		goto err_out_cache;
3890 	}
3891 
3892 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3893 	if (!sd_page_pool) {
3894 		printk(KERN_ERR "sd: can't init discard page pool\n");
3895 		err = -ENOMEM;
3896 		goto err_out_ppool;
3897 	}
3898 
3899 	err = scsi_register_driver(&sd_template.gendrv);
3900 	if (err)
3901 		goto err_out_driver;
3902 
3903 	return 0;
3904 
3905 err_out_driver:
3906 	mempool_destroy(sd_page_pool);
3907 
3908 err_out_ppool:
3909 	mempool_destroy(sd_cdb_pool);
3910 
3911 err_out_cache:
3912 	kmem_cache_destroy(sd_cdb_cache);
3913 
3914 err_out_class:
3915 	class_unregister(&sd_disk_class);
3916 err_out:
3917 	for (i = 0; i < SD_MAJORS; i++)
3918 		unregister_blkdev(sd_major(i), "sd");
3919 	return err;
3920 }
3921 
3922 /**
3923  *	exit_sd - exit point for this driver (when it is a module).
3924  *
3925  *	Note: this function unregisters this driver from the scsi mid-level.
3926  **/
3927 static void __exit exit_sd(void)
3928 {
3929 	int i;
3930 
3931 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3932 
3933 	scsi_unregister_driver(&sd_template.gendrv);
3934 	mempool_destroy(sd_cdb_pool);
3935 	mempool_destroy(sd_page_pool);
3936 	kmem_cache_destroy(sd_cdb_cache);
3937 
3938 	class_unregister(&sd_disk_class);
3939 
3940 	for (i = 0; i < SD_MAJORS; i++)
3941 		unregister_blkdev(sd_major(i), "sd");
3942 }
3943 
3944 module_init(init_sd);
3945 module_exit(exit_sd);
3946 
3947 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3948 {
3949 	scsi_print_sense_hdr(sdkp->device,
3950 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3951 }
3952 
3953 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3954 {
3955 	const char *hb_string = scsi_hostbyte_string(result);
3956 
3957 	if (hb_string)
3958 		sd_printk(KERN_INFO, sdkp,
3959 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3960 			  hb_string ? hb_string : "invalid",
3961 			  "DRIVER_OK");
3962 	else
3963 		sd_printk(KERN_INFO, sdkp,
3964 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3965 			  msg, host_byte(result), "DRIVER_OK");
3966 }
3967