xref: /openbmc/linux/drivers/scsi/sd.c (revision 78560d41)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *      sd.c Copyright (C) 1992 Drew Eckhardt
4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5  *
6  *      Linux scsi disk driver
7  *              Initial versions: Drew Eckhardt
8  *              Subsequent revisions: Eric Youngdale
9  *	Modification history:
10  *       - Drew Eckhardt <drew@colorado.edu> original
11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12  *         outstanding request, and other enhancements.
13  *         Support loadable low-level scsi drivers.
14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15  *         eight major numbers.
16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18  *	   sd_init and cleanups.
19  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
20  *	   not being read in sd_open. Fix problem where removable media
21  *	   could be ejected after sd_open.
22  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25  *	   Support 32k/1M disks.
26  *
27  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
28  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
32  *	Note: when the logging level is set by the user, it must be greater
33  *	than the level indicated above to trigger output.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
57 #include <linux/pr.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
61 
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
71 
72 #include "sd.h"
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
75 
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
79 
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100 
101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
102 #define SD_MINORS	16
103 #else
104 #define SD_MINORS	0
105 #endif
106 
107 static void sd_config_discard(struct scsi_disk *, unsigned int);
108 static void sd_config_write_same(struct scsi_disk *);
109 static int  sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static int  sd_probe(struct device *);
112 static int  sd_remove(struct device *);
113 static void sd_shutdown(struct device *);
114 static int sd_suspend_system(struct device *);
115 static int sd_suspend_runtime(struct device *);
116 static int sd_resume(struct device *);
117 static void sd_rescan(struct device *);
118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120 static int sd_done(struct scsi_cmnd *);
121 static void sd_eh_reset(struct scsi_cmnd *);
122 static int sd_eh_action(struct scsi_cmnd *, int);
123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124 static void scsi_disk_release(struct device *cdev);
125 
126 static DEFINE_IDA(sd_index_ida);
127 
128 /* This semaphore is used to mediate the 0->1 reference get in the
129  * face of object destruction (i.e. we can't allow a get on an
130  * object after last put) */
131 static DEFINE_MUTEX(sd_ref_mutex);
132 
133 static struct kmem_cache *sd_cdb_cache;
134 static mempool_t *sd_cdb_pool;
135 static mempool_t *sd_page_pool;
136 
137 static const char *sd_cache_types[] = {
138 	"write through", "none", "write back",
139 	"write back, no read (daft)"
140 };
141 
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
143 {
144 	bool wc = false, fua = false;
145 
146 	if (sdkp->WCE) {
147 		wc = true;
148 		if (sdkp->DPOFUA)
149 			fua = true;
150 	}
151 
152 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
153 }
154 
155 static ssize_t
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 		 const char *buf, size_t count)
158 {
159 	int ct, rcd, wce, sp;
160 	struct scsi_disk *sdkp = to_scsi_disk(dev);
161 	struct scsi_device *sdp = sdkp->device;
162 	char buffer[64];
163 	char *buffer_data;
164 	struct scsi_mode_data data;
165 	struct scsi_sense_hdr sshdr;
166 	static const char temp[] = "temporary ";
167 	int len;
168 
169 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 		/* no cache control on RBC devices; theoretically they
171 		 * can do it, but there's probably so many exceptions
172 		 * it's not worth the risk */
173 		return -EINVAL;
174 
175 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 		buf += sizeof(temp) - 1;
177 		sdkp->cache_override = 1;
178 	} else {
179 		sdkp->cache_override = 0;
180 	}
181 
182 	ct = sysfs_match_string(sd_cache_types, buf);
183 	if (ct < 0)
184 		return -EINVAL;
185 
186 	rcd = ct & 0x01 ? 1 : 0;
187 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
188 
189 	if (sdkp->cache_override) {
190 		sdkp->WCE = wce;
191 		sdkp->RCD = rcd;
192 		sd_set_flush_flag(sdkp);
193 		return count;
194 	}
195 
196 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 			    SD_MAX_RETRIES, &data, NULL))
198 		return -EINVAL;
199 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 		  data.block_descriptor_length);
201 	buffer_data = buffer + data.header_length +
202 		data.block_descriptor_length;
203 	buffer_data[2] &= ~0x05;
204 	buffer_data[2] |= wce << 2 | rcd;
205 	sp = buffer_data[0] & 0x80 ? 1 : 0;
206 	buffer_data[0] &= ~0x80;
207 
208 	/*
209 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
210 	 * received mode parameter buffer before doing MODE SELECT.
211 	 */
212 	data.device_specific = 0;
213 
214 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
215 			     SD_MAX_RETRIES, &data, &sshdr)) {
216 		if (scsi_sense_valid(&sshdr))
217 			sd_print_sense_hdr(sdkp, &sshdr);
218 		return -EINVAL;
219 	}
220 	revalidate_disk(sdkp->disk);
221 	return count;
222 }
223 
224 static ssize_t
225 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
226 		       char *buf)
227 {
228 	struct scsi_disk *sdkp = to_scsi_disk(dev);
229 	struct scsi_device *sdp = sdkp->device;
230 
231 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
232 }
233 
234 static ssize_t
235 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
236 			const char *buf, size_t count)
237 {
238 	struct scsi_disk *sdkp = to_scsi_disk(dev);
239 	struct scsi_device *sdp = sdkp->device;
240 	bool v;
241 
242 	if (!capable(CAP_SYS_ADMIN))
243 		return -EACCES;
244 
245 	if (kstrtobool(buf, &v))
246 		return -EINVAL;
247 
248 	sdp->manage_start_stop = v;
249 
250 	return count;
251 }
252 static DEVICE_ATTR_RW(manage_start_stop);
253 
254 static ssize_t
255 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
256 {
257 	struct scsi_disk *sdkp = to_scsi_disk(dev);
258 
259 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
260 }
261 
262 static ssize_t
263 allow_restart_store(struct device *dev, struct device_attribute *attr,
264 		    const char *buf, size_t count)
265 {
266 	bool v;
267 	struct scsi_disk *sdkp = to_scsi_disk(dev);
268 	struct scsi_device *sdp = sdkp->device;
269 
270 	if (!capable(CAP_SYS_ADMIN))
271 		return -EACCES;
272 
273 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
274 		return -EINVAL;
275 
276 	if (kstrtobool(buf, &v))
277 		return -EINVAL;
278 
279 	sdp->allow_restart = v;
280 
281 	return count;
282 }
283 static DEVICE_ATTR_RW(allow_restart);
284 
285 static ssize_t
286 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
287 {
288 	struct scsi_disk *sdkp = to_scsi_disk(dev);
289 	int ct = sdkp->RCD + 2*sdkp->WCE;
290 
291 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
292 }
293 static DEVICE_ATTR_RW(cache_type);
294 
295 static ssize_t
296 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
297 {
298 	struct scsi_disk *sdkp = to_scsi_disk(dev);
299 
300 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
301 }
302 static DEVICE_ATTR_RO(FUA);
303 
304 static ssize_t
305 protection_type_show(struct device *dev, struct device_attribute *attr,
306 		     char *buf)
307 {
308 	struct scsi_disk *sdkp = to_scsi_disk(dev);
309 
310 	return sprintf(buf, "%u\n", sdkp->protection_type);
311 }
312 
313 static ssize_t
314 protection_type_store(struct device *dev, struct device_attribute *attr,
315 		      const char *buf, size_t count)
316 {
317 	struct scsi_disk *sdkp = to_scsi_disk(dev);
318 	unsigned int val;
319 	int err;
320 
321 	if (!capable(CAP_SYS_ADMIN))
322 		return -EACCES;
323 
324 	err = kstrtouint(buf, 10, &val);
325 
326 	if (err)
327 		return err;
328 
329 	if (val <= T10_PI_TYPE3_PROTECTION)
330 		sdkp->protection_type = val;
331 
332 	return count;
333 }
334 static DEVICE_ATTR_RW(protection_type);
335 
336 static ssize_t
337 protection_mode_show(struct device *dev, struct device_attribute *attr,
338 		     char *buf)
339 {
340 	struct scsi_disk *sdkp = to_scsi_disk(dev);
341 	struct scsi_device *sdp = sdkp->device;
342 	unsigned int dif, dix;
343 
344 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
345 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
346 
347 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
348 		dif = 0;
349 		dix = 1;
350 	}
351 
352 	if (!dif && !dix)
353 		return sprintf(buf, "none\n");
354 
355 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
356 }
357 static DEVICE_ATTR_RO(protection_mode);
358 
359 static ssize_t
360 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
361 {
362 	struct scsi_disk *sdkp = to_scsi_disk(dev);
363 
364 	return sprintf(buf, "%u\n", sdkp->ATO);
365 }
366 static DEVICE_ATTR_RO(app_tag_own);
367 
368 static ssize_t
369 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
370 		       char *buf)
371 {
372 	struct scsi_disk *sdkp = to_scsi_disk(dev);
373 
374 	return sprintf(buf, "%u\n", sdkp->lbpme);
375 }
376 static DEVICE_ATTR_RO(thin_provisioning);
377 
378 /* sysfs_match_string() requires dense arrays */
379 static const char *lbp_mode[] = {
380 	[SD_LBP_FULL]		= "full",
381 	[SD_LBP_UNMAP]		= "unmap",
382 	[SD_LBP_WS16]		= "writesame_16",
383 	[SD_LBP_WS10]		= "writesame_10",
384 	[SD_LBP_ZERO]		= "writesame_zero",
385 	[SD_LBP_DISABLE]	= "disabled",
386 };
387 
388 static ssize_t
389 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
390 		       char *buf)
391 {
392 	struct scsi_disk *sdkp = to_scsi_disk(dev);
393 
394 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
395 }
396 
397 static ssize_t
398 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
399 			const char *buf, size_t count)
400 {
401 	struct scsi_disk *sdkp = to_scsi_disk(dev);
402 	struct scsi_device *sdp = sdkp->device;
403 	int mode;
404 
405 	if (!capable(CAP_SYS_ADMIN))
406 		return -EACCES;
407 
408 	if (sd_is_zoned(sdkp)) {
409 		sd_config_discard(sdkp, SD_LBP_DISABLE);
410 		return count;
411 	}
412 
413 	if (sdp->type != TYPE_DISK)
414 		return -EINVAL;
415 
416 	mode = sysfs_match_string(lbp_mode, buf);
417 	if (mode < 0)
418 		return -EINVAL;
419 
420 	sd_config_discard(sdkp, mode);
421 
422 	return count;
423 }
424 static DEVICE_ATTR_RW(provisioning_mode);
425 
426 /* sysfs_match_string() requires dense arrays */
427 static const char *zeroing_mode[] = {
428 	[SD_ZERO_WRITE]		= "write",
429 	[SD_ZERO_WS]		= "writesame",
430 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
431 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
432 };
433 
434 static ssize_t
435 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
436 		  char *buf)
437 {
438 	struct scsi_disk *sdkp = to_scsi_disk(dev);
439 
440 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
441 }
442 
443 static ssize_t
444 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
445 		   const char *buf, size_t count)
446 {
447 	struct scsi_disk *sdkp = to_scsi_disk(dev);
448 	int mode;
449 
450 	if (!capable(CAP_SYS_ADMIN))
451 		return -EACCES;
452 
453 	mode = sysfs_match_string(zeroing_mode, buf);
454 	if (mode < 0)
455 		return -EINVAL;
456 
457 	sdkp->zeroing_mode = mode;
458 
459 	return count;
460 }
461 static DEVICE_ATTR_RW(zeroing_mode);
462 
463 static ssize_t
464 max_medium_access_timeouts_show(struct device *dev,
465 				struct device_attribute *attr, char *buf)
466 {
467 	struct scsi_disk *sdkp = to_scsi_disk(dev);
468 
469 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
470 }
471 
472 static ssize_t
473 max_medium_access_timeouts_store(struct device *dev,
474 				 struct device_attribute *attr, const char *buf,
475 				 size_t count)
476 {
477 	struct scsi_disk *sdkp = to_scsi_disk(dev);
478 	int err;
479 
480 	if (!capable(CAP_SYS_ADMIN))
481 		return -EACCES;
482 
483 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
484 
485 	return err ? err : count;
486 }
487 static DEVICE_ATTR_RW(max_medium_access_timeouts);
488 
489 static ssize_t
490 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
491 			   char *buf)
492 {
493 	struct scsi_disk *sdkp = to_scsi_disk(dev);
494 
495 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
496 }
497 
498 static ssize_t
499 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
500 			    const char *buf, size_t count)
501 {
502 	struct scsi_disk *sdkp = to_scsi_disk(dev);
503 	struct scsi_device *sdp = sdkp->device;
504 	unsigned long max;
505 	int err;
506 
507 	if (!capable(CAP_SYS_ADMIN))
508 		return -EACCES;
509 
510 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
511 		return -EINVAL;
512 
513 	err = kstrtoul(buf, 10, &max);
514 
515 	if (err)
516 		return err;
517 
518 	if (max == 0)
519 		sdp->no_write_same = 1;
520 	else if (max <= SD_MAX_WS16_BLOCKS) {
521 		sdp->no_write_same = 0;
522 		sdkp->max_ws_blocks = max;
523 	}
524 
525 	sd_config_write_same(sdkp);
526 
527 	return count;
528 }
529 static DEVICE_ATTR_RW(max_write_same_blocks);
530 
531 static struct attribute *sd_disk_attrs[] = {
532 	&dev_attr_cache_type.attr,
533 	&dev_attr_FUA.attr,
534 	&dev_attr_allow_restart.attr,
535 	&dev_attr_manage_start_stop.attr,
536 	&dev_attr_protection_type.attr,
537 	&dev_attr_protection_mode.attr,
538 	&dev_attr_app_tag_own.attr,
539 	&dev_attr_thin_provisioning.attr,
540 	&dev_attr_provisioning_mode.attr,
541 	&dev_attr_zeroing_mode.attr,
542 	&dev_attr_max_write_same_blocks.attr,
543 	&dev_attr_max_medium_access_timeouts.attr,
544 	NULL,
545 };
546 ATTRIBUTE_GROUPS(sd_disk);
547 
548 static struct class sd_disk_class = {
549 	.name		= "scsi_disk",
550 	.owner		= THIS_MODULE,
551 	.dev_release	= scsi_disk_release,
552 	.dev_groups	= sd_disk_groups,
553 };
554 
555 static const struct dev_pm_ops sd_pm_ops = {
556 	.suspend		= sd_suspend_system,
557 	.resume			= sd_resume,
558 	.poweroff		= sd_suspend_system,
559 	.restore		= sd_resume,
560 	.runtime_suspend	= sd_suspend_runtime,
561 	.runtime_resume		= sd_resume,
562 };
563 
564 static struct scsi_driver sd_template = {
565 	.gendrv = {
566 		.name		= "sd",
567 		.owner		= THIS_MODULE,
568 		.probe		= sd_probe,
569 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
570 		.remove		= sd_remove,
571 		.shutdown	= sd_shutdown,
572 		.pm		= &sd_pm_ops,
573 	},
574 	.rescan			= sd_rescan,
575 	.init_command		= sd_init_command,
576 	.uninit_command		= sd_uninit_command,
577 	.done			= sd_done,
578 	.eh_action		= sd_eh_action,
579 	.eh_reset		= sd_eh_reset,
580 };
581 
582 /*
583  * Dummy kobj_map->probe function.
584  * The default ->probe function will call modprobe, which is
585  * pointless as this module is already loaded.
586  */
587 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
588 {
589 	return NULL;
590 }
591 
592 /*
593  * Device no to disk mapping:
594  *
595  *       major         disc2     disc  p1
596  *   |............|.............|....|....| <- dev_t
597  *    31        20 19          8 7  4 3  0
598  *
599  * Inside a major, we have 16k disks, however mapped non-
600  * contiguously. The first 16 disks are for major0, the next
601  * ones with major1, ... Disk 256 is for major0 again, disk 272
602  * for major1, ...
603  * As we stay compatible with our numbering scheme, we can reuse
604  * the well-know SCSI majors 8, 65--71, 136--143.
605  */
606 static int sd_major(int major_idx)
607 {
608 	switch (major_idx) {
609 	case 0:
610 		return SCSI_DISK0_MAJOR;
611 	case 1 ... 7:
612 		return SCSI_DISK1_MAJOR + major_idx - 1;
613 	case 8 ... 15:
614 		return SCSI_DISK8_MAJOR + major_idx - 8;
615 	default:
616 		BUG();
617 		return 0;	/* shut up gcc */
618 	}
619 }
620 
621 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
622 {
623 	struct scsi_disk *sdkp = NULL;
624 
625 	mutex_lock(&sd_ref_mutex);
626 
627 	if (disk->private_data) {
628 		sdkp = scsi_disk(disk);
629 		if (scsi_device_get(sdkp->device) == 0)
630 			get_device(&sdkp->dev);
631 		else
632 			sdkp = NULL;
633 	}
634 	mutex_unlock(&sd_ref_mutex);
635 	return sdkp;
636 }
637 
638 static void scsi_disk_put(struct scsi_disk *sdkp)
639 {
640 	struct scsi_device *sdev = sdkp->device;
641 
642 	mutex_lock(&sd_ref_mutex);
643 	put_device(&sdkp->dev);
644 	scsi_device_put(sdev);
645 	mutex_unlock(&sd_ref_mutex);
646 }
647 
648 #ifdef CONFIG_BLK_SED_OPAL
649 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
650 		size_t len, bool send)
651 {
652 	struct scsi_device *sdev = data;
653 	u8 cdb[12] = { 0, };
654 	int ret;
655 
656 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
657 	cdb[1] = secp;
658 	put_unaligned_be16(spsp, &cdb[2]);
659 	put_unaligned_be32(len, &cdb[6]);
660 
661 	ret = scsi_execute_req(sdev, cdb,
662 			send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
663 			buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
664 	return ret <= 0 ? ret : -EIO;
665 }
666 #endif /* CONFIG_BLK_SED_OPAL */
667 
668 /*
669  * Look up the DIX operation based on whether the command is read or
670  * write and whether dix and dif are enabled.
671  */
672 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
673 {
674 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
675 	static const unsigned int ops[] = {	/* wrt dix dif */
676 		SCSI_PROT_NORMAL,		/*  0	0   0  */
677 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
678 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
679 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
680 		SCSI_PROT_NORMAL,		/*  1	0   0  */
681 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
682 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
683 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
684 	};
685 
686 	return ops[write << 2 | dix << 1 | dif];
687 }
688 
689 /*
690  * Returns a mask of the protection flags that are valid for a given DIX
691  * operation.
692  */
693 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
694 {
695 	static const unsigned int flag_mask[] = {
696 		[SCSI_PROT_NORMAL]		= 0,
697 
698 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
699 						  SCSI_PROT_GUARD_CHECK |
700 						  SCSI_PROT_REF_CHECK |
701 						  SCSI_PROT_REF_INCREMENT,
702 
703 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
704 						  SCSI_PROT_IP_CHECKSUM,
705 
706 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
707 						  SCSI_PROT_GUARD_CHECK |
708 						  SCSI_PROT_REF_CHECK |
709 						  SCSI_PROT_REF_INCREMENT |
710 						  SCSI_PROT_IP_CHECKSUM,
711 
712 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
713 						  SCSI_PROT_REF_INCREMENT,
714 
715 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
716 						  SCSI_PROT_REF_CHECK |
717 						  SCSI_PROT_REF_INCREMENT |
718 						  SCSI_PROT_IP_CHECKSUM,
719 
720 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
721 						  SCSI_PROT_GUARD_CHECK |
722 						  SCSI_PROT_REF_CHECK |
723 						  SCSI_PROT_REF_INCREMENT |
724 						  SCSI_PROT_IP_CHECKSUM,
725 	};
726 
727 	return flag_mask[prot_op];
728 }
729 
730 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
731 					   unsigned int dix, unsigned int dif)
732 {
733 	struct bio *bio = scmd->request->bio;
734 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
735 	unsigned int protect = 0;
736 
737 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
738 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
739 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
740 
741 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
742 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
743 	}
744 
745 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
746 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
747 
748 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
749 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
750 	}
751 
752 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
753 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
754 
755 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
756 			protect = 3 << 5;	/* Disable target PI checking */
757 		else
758 			protect = 1 << 5;	/* Enable target PI checking */
759 	}
760 
761 	scsi_set_prot_op(scmd, prot_op);
762 	scsi_set_prot_type(scmd, dif);
763 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
764 
765 	return protect;
766 }
767 
768 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
769 {
770 	struct request_queue *q = sdkp->disk->queue;
771 	unsigned int logical_block_size = sdkp->device->sector_size;
772 	unsigned int max_blocks = 0;
773 
774 	q->limits.discard_alignment =
775 		sdkp->unmap_alignment * logical_block_size;
776 	q->limits.discard_granularity =
777 		max(sdkp->physical_block_size,
778 		    sdkp->unmap_granularity * logical_block_size);
779 	sdkp->provisioning_mode = mode;
780 
781 	switch (mode) {
782 
783 	case SD_LBP_FULL:
784 	case SD_LBP_DISABLE:
785 		blk_queue_max_discard_sectors(q, 0);
786 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
787 		return;
788 
789 	case SD_LBP_UNMAP:
790 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
791 					  (u32)SD_MAX_WS16_BLOCKS);
792 		break;
793 
794 	case SD_LBP_WS16:
795 		if (sdkp->device->unmap_limit_for_ws)
796 			max_blocks = sdkp->max_unmap_blocks;
797 		else
798 			max_blocks = sdkp->max_ws_blocks;
799 
800 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
801 		break;
802 
803 	case SD_LBP_WS10:
804 		if (sdkp->device->unmap_limit_for_ws)
805 			max_blocks = sdkp->max_unmap_blocks;
806 		else
807 			max_blocks = sdkp->max_ws_blocks;
808 
809 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
810 		break;
811 
812 	case SD_LBP_ZERO:
813 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
814 					  (u32)SD_MAX_WS10_BLOCKS);
815 		break;
816 	}
817 
818 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
819 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
820 }
821 
822 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
823 {
824 	struct scsi_device *sdp = cmd->device;
825 	struct request *rq = cmd->request;
826 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
827 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
828 	unsigned int data_len = 24;
829 	char *buf;
830 
831 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
832 	if (!rq->special_vec.bv_page)
833 		return BLK_STS_RESOURCE;
834 	clear_highpage(rq->special_vec.bv_page);
835 	rq->special_vec.bv_offset = 0;
836 	rq->special_vec.bv_len = data_len;
837 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
838 
839 	cmd->cmd_len = 10;
840 	cmd->cmnd[0] = UNMAP;
841 	cmd->cmnd[8] = 24;
842 
843 	buf = page_address(rq->special_vec.bv_page);
844 	put_unaligned_be16(6 + 16, &buf[0]);
845 	put_unaligned_be16(16, &buf[2]);
846 	put_unaligned_be64(lba, &buf[8]);
847 	put_unaligned_be32(nr_blocks, &buf[16]);
848 
849 	cmd->allowed = SD_MAX_RETRIES;
850 	cmd->transfersize = data_len;
851 	rq->timeout = SD_TIMEOUT;
852 
853 	return scsi_init_io(cmd);
854 }
855 
856 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
857 		bool unmap)
858 {
859 	struct scsi_device *sdp = cmd->device;
860 	struct request *rq = cmd->request;
861 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
862 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
863 	u32 data_len = sdp->sector_size;
864 
865 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
866 	if (!rq->special_vec.bv_page)
867 		return BLK_STS_RESOURCE;
868 	clear_highpage(rq->special_vec.bv_page);
869 	rq->special_vec.bv_offset = 0;
870 	rq->special_vec.bv_len = data_len;
871 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
872 
873 	cmd->cmd_len = 16;
874 	cmd->cmnd[0] = WRITE_SAME_16;
875 	if (unmap)
876 		cmd->cmnd[1] = 0x8; /* UNMAP */
877 	put_unaligned_be64(lba, &cmd->cmnd[2]);
878 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
879 
880 	cmd->allowed = SD_MAX_RETRIES;
881 	cmd->transfersize = data_len;
882 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
883 
884 	return scsi_init_io(cmd);
885 }
886 
887 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
888 		bool unmap)
889 {
890 	struct scsi_device *sdp = cmd->device;
891 	struct request *rq = cmd->request;
892 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
893 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
894 	u32 data_len = sdp->sector_size;
895 
896 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
897 	if (!rq->special_vec.bv_page)
898 		return BLK_STS_RESOURCE;
899 	clear_highpage(rq->special_vec.bv_page);
900 	rq->special_vec.bv_offset = 0;
901 	rq->special_vec.bv_len = data_len;
902 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
903 
904 	cmd->cmd_len = 10;
905 	cmd->cmnd[0] = WRITE_SAME;
906 	if (unmap)
907 		cmd->cmnd[1] = 0x8; /* UNMAP */
908 	put_unaligned_be32(lba, &cmd->cmnd[2]);
909 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
910 
911 	cmd->allowed = SD_MAX_RETRIES;
912 	cmd->transfersize = data_len;
913 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
914 
915 	return scsi_init_io(cmd);
916 }
917 
918 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
919 {
920 	struct request *rq = cmd->request;
921 	struct scsi_device *sdp = cmd->device;
922 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
923 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
924 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
925 
926 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
927 		switch (sdkp->zeroing_mode) {
928 		case SD_ZERO_WS16_UNMAP:
929 			return sd_setup_write_same16_cmnd(cmd, true);
930 		case SD_ZERO_WS10_UNMAP:
931 			return sd_setup_write_same10_cmnd(cmd, true);
932 		}
933 	}
934 
935 	if (sdp->no_write_same)
936 		return BLK_STS_TARGET;
937 
938 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
939 		return sd_setup_write_same16_cmnd(cmd, false);
940 
941 	return sd_setup_write_same10_cmnd(cmd, false);
942 }
943 
944 static void sd_config_write_same(struct scsi_disk *sdkp)
945 {
946 	struct request_queue *q = sdkp->disk->queue;
947 	unsigned int logical_block_size = sdkp->device->sector_size;
948 
949 	if (sdkp->device->no_write_same) {
950 		sdkp->max_ws_blocks = 0;
951 		goto out;
952 	}
953 
954 	/* Some devices can not handle block counts above 0xffff despite
955 	 * supporting WRITE SAME(16). Consequently we default to 64k
956 	 * blocks per I/O unless the device explicitly advertises a
957 	 * bigger limit.
958 	 */
959 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
960 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
961 						   (u32)SD_MAX_WS16_BLOCKS);
962 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
963 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
964 						   (u32)SD_MAX_WS10_BLOCKS);
965 	else {
966 		sdkp->device->no_write_same = 1;
967 		sdkp->max_ws_blocks = 0;
968 	}
969 
970 	if (sdkp->lbprz && sdkp->lbpws)
971 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
972 	else if (sdkp->lbprz && sdkp->lbpws10)
973 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
974 	else if (sdkp->max_ws_blocks)
975 		sdkp->zeroing_mode = SD_ZERO_WS;
976 	else
977 		sdkp->zeroing_mode = SD_ZERO_WRITE;
978 
979 	if (sdkp->max_ws_blocks &&
980 	    sdkp->physical_block_size > logical_block_size) {
981 		/*
982 		 * Reporting a maximum number of blocks that is not aligned
983 		 * on the device physical size would cause a large write same
984 		 * request to be split into physically unaligned chunks by
985 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
986 		 * even if the caller of these functions took care to align the
987 		 * large request. So make sure the maximum reported is aligned
988 		 * to the device physical block size. This is only an optional
989 		 * optimization for regular disks, but this is mandatory to
990 		 * avoid failure of large write same requests directed at
991 		 * sequential write required zones of host-managed ZBC disks.
992 		 */
993 		sdkp->max_ws_blocks =
994 			round_down(sdkp->max_ws_blocks,
995 				   bytes_to_logical(sdkp->device,
996 						    sdkp->physical_block_size));
997 	}
998 
999 out:
1000 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1001 					 (logical_block_size >> 9));
1002 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1003 					 (logical_block_size >> 9));
1004 }
1005 
1006 /**
1007  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1008  * @cmd: command to prepare
1009  *
1010  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1011  * the preference indicated by the target device.
1012  **/
1013 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1014 {
1015 	struct request *rq = cmd->request;
1016 	struct scsi_device *sdp = cmd->device;
1017 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1018 	struct bio *bio = rq->bio;
1019 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1020 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1021 	blk_status_t ret;
1022 
1023 	if (sdkp->device->no_write_same)
1024 		return BLK_STS_TARGET;
1025 
1026 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1027 
1028 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1029 
1030 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1031 		cmd->cmd_len = 16;
1032 		cmd->cmnd[0] = WRITE_SAME_16;
1033 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1034 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1035 	} else {
1036 		cmd->cmd_len = 10;
1037 		cmd->cmnd[0] = WRITE_SAME;
1038 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1039 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1040 	}
1041 
1042 	cmd->transfersize = sdp->sector_size;
1043 	cmd->allowed = SD_MAX_RETRIES;
1044 
1045 	/*
1046 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1047 	 * different from the amount of data actually written to the target.
1048 	 *
1049 	 * We set up __data_len to the amount of data transferred via the
1050 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1051 	 * to transfer a single sector of data first, but then reset it to
1052 	 * the amount of data to be written right after so that the I/O path
1053 	 * knows how much to actually write.
1054 	 */
1055 	rq->__data_len = sdp->sector_size;
1056 	ret = scsi_init_io(cmd);
1057 	rq->__data_len = blk_rq_bytes(rq);
1058 
1059 	return ret;
1060 }
1061 
1062 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1063 {
1064 	struct request *rq = cmd->request;
1065 
1066 	/* flush requests don't perform I/O, zero the S/G table */
1067 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1068 
1069 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1070 	cmd->cmd_len = 10;
1071 	cmd->transfersize = 0;
1072 	cmd->allowed = SD_MAX_RETRIES;
1073 
1074 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1075 	return BLK_STS_OK;
1076 }
1077 
1078 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1079 				       sector_t lba, unsigned int nr_blocks,
1080 				       unsigned char flags)
1081 {
1082 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1083 	if (unlikely(cmd->cmnd == NULL))
1084 		return BLK_STS_RESOURCE;
1085 
1086 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1087 	memset(cmd->cmnd, 0, cmd->cmd_len);
1088 
1089 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1090 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1091 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1092 	cmd->cmnd[10] = flags;
1093 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1094 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1095 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1096 
1097 	return BLK_STS_OK;
1098 }
1099 
1100 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1101 				       sector_t lba, unsigned int nr_blocks,
1102 				       unsigned char flags)
1103 {
1104 	cmd->cmd_len  = 16;
1105 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1106 	cmd->cmnd[1]  = flags;
1107 	cmd->cmnd[14] = 0;
1108 	cmd->cmnd[15] = 0;
1109 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1110 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1111 
1112 	return BLK_STS_OK;
1113 }
1114 
1115 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1116 				       sector_t lba, unsigned int nr_blocks,
1117 				       unsigned char flags)
1118 {
1119 	cmd->cmd_len = 10;
1120 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1121 	cmd->cmnd[1] = flags;
1122 	cmd->cmnd[6] = 0;
1123 	cmd->cmnd[9] = 0;
1124 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1125 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1126 
1127 	return BLK_STS_OK;
1128 }
1129 
1130 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1131 				      sector_t lba, unsigned int nr_blocks,
1132 				      unsigned char flags)
1133 {
1134 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1135 	if (WARN_ON_ONCE(nr_blocks == 0))
1136 		return BLK_STS_IOERR;
1137 
1138 	if (unlikely(flags & 0x8)) {
1139 		/*
1140 		 * This happens only if this drive failed 10byte rw
1141 		 * command with ILLEGAL_REQUEST during operation and
1142 		 * thus turned off use_10_for_rw.
1143 		 */
1144 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1145 		return BLK_STS_IOERR;
1146 	}
1147 
1148 	cmd->cmd_len = 6;
1149 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1150 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1151 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1152 	cmd->cmnd[3] = lba & 0xff;
1153 	cmd->cmnd[4] = nr_blocks;
1154 	cmd->cmnd[5] = 0;
1155 
1156 	return BLK_STS_OK;
1157 }
1158 
1159 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1160 {
1161 	struct request *rq = cmd->request;
1162 	struct scsi_device *sdp = cmd->device;
1163 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1164 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1165 	sector_t threshold;
1166 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1167 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1168 	bool write = rq_data_dir(rq) == WRITE;
1169 	unsigned char protect, fua;
1170 	blk_status_t ret;
1171 	unsigned int dif;
1172 	bool dix;
1173 
1174 	ret = scsi_init_io(cmd);
1175 	if (ret != BLK_STS_OK)
1176 		return ret;
1177 
1178 	if (!scsi_device_online(sdp) || sdp->changed) {
1179 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1180 		return BLK_STS_IOERR;
1181 	}
1182 
1183 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1184 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1185 		return BLK_STS_IOERR;
1186 	}
1187 
1188 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1189 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1190 		return BLK_STS_IOERR;
1191 	}
1192 
1193 	/*
1194 	 * Some SD card readers can't handle accesses which touch the
1195 	 * last one or two logical blocks. Split accesses as needed.
1196 	 */
1197 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1198 
1199 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1200 		if (lba < threshold) {
1201 			/* Access up to the threshold but not beyond */
1202 			nr_blocks = threshold - lba;
1203 		} else {
1204 			/* Access only a single logical block */
1205 			nr_blocks = 1;
1206 		}
1207 	}
1208 
1209 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1210 	dix = scsi_prot_sg_count(cmd);
1211 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1212 
1213 	if (dif || dix)
1214 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1215 	else
1216 		protect = 0;
1217 
1218 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1219 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1220 					 protect | fua);
1221 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1222 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1223 					 protect | fua);
1224 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1225 		   sdp->use_10_for_rw || protect) {
1226 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1227 					 protect | fua);
1228 	} else {
1229 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1230 					protect | fua);
1231 	}
1232 
1233 	if (unlikely(ret != BLK_STS_OK))
1234 		return ret;
1235 
1236 	/*
1237 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1238 	 * host adapter, it's safe to assume that we can at least transfer
1239 	 * this many bytes between each connect / disconnect.
1240 	 */
1241 	cmd->transfersize = sdp->sector_size;
1242 	cmd->underflow = nr_blocks << 9;
1243 	cmd->allowed = SD_MAX_RETRIES;
1244 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1245 
1246 	SCSI_LOG_HLQUEUE(1,
1247 			 scmd_printk(KERN_INFO, cmd,
1248 				     "%s: block=%llu, count=%d\n", __func__,
1249 				     (unsigned long long)blk_rq_pos(rq),
1250 				     blk_rq_sectors(rq)));
1251 	SCSI_LOG_HLQUEUE(2,
1252 			 scmd_printk(KERN_INFO, cmd,
1253 				     "%s %d/%u 512 byte blocks.\n",
1254 				     write ? "writing" : "reading", nr_blocks,
1255 				     blk_rq_sectors(rq)));
1256 
1257 	/*
1258 	 * This indicates that the command is ready from our end to be
1259 	 * queued.
1260 	 */
1261 	return BLK_STS_OK;
1262 }
1263 
1264 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1265 {
1266 	struct request *rq = cmd->request;
1267 
1268 	switch (req_op(rq)) {
1269 	case REQ_OP_DISCARD:
1270 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1271 		case SD_LBP_UNMAP:
1272 			return sd_setup_unmap_cmnd(cmd);
1273 		case SD_LBP_WS16:
1274 			return sd_setup_write_same16_cmnd(cmd, true);
1275 		case SD_LBP_WS10:
1276 			return sd_setup_write_same10_cmnd(cmd, true);
1277 		case SD_LBP_ZERO:
1278 			return sd_setup_write_same10_cmnd(cmd, false);
1279 		default:
1280 			return BLK_STS_TARGET;
1281 		}
1282 	case REQ_OP_WRITE_ZEROES:
1283 		return sd_setup_write_zeroes_cmnd(cmd);
1284 	case REQ_OP_WRITE_SAME:
1285 		return sd_setup_write_same_cmnd(cmd);
1286 	case REQ_OP_FLUSH:
1287 		return sd_setup_flush_cmnd(cmd);
1288 	case REQ_OP_READ:
1289 	case REQ_OP_WRITE:
1290 		return sd_setup_read_write_cmnd(cmd);
1291 	case REQ_OP_ZONE_RESET:
1292 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1293 						   false);
1294 	case REQ_OP_ZONE_RESET_ALL:
1295 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1296 						   true);
1297 	case REQ_OP_ZONE_OPEN:
1298 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1299 	case REQ_OP_ZONE_CLOSE:
1300 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1301 	case REQ_OP_ZONE_FINISH:
1302 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1303 	default:
1304 		WARN_ON_ONCE(1);
1305 		return BLK_STS_NOTSUPP;
1306 	}
1307 }
1308 
1309 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1310 {
1311 	struct request *rq = SCpnt->request;
1312 	u8 *cmnd;
1313 
1314 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1315 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1316 
1317 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1318 		cmnd = SCpnt->cmnd;
1319 		SCpnt->cmnd = NULL;
1320 		SCpnt->cmd_len = 0;
1321 		mempool_free(cmnd, sd_cdb_pool);
1322 	}
1323 }
1324 
1325 /**
1326  *	sd_open - open a scsi disk device
1327  *	@bdev: Block device of the scsi disk to open
1328  *	@mode: FMODE_* mask
1329  *
1330  *	Returns 0 if successful. Returns a negated errno value in case
1331  *	of error.
1332  *
1333  *	Note: This can be called from a user context (e.g. fsck(1) )
1334  *	or from within the kernel (e.g. as a result of a mount(1) ).
1335  *	In the latter case @inode and @filp carry an abridged amount
1336  *	of information as noted above.
1337  *
1338  *	Locking: called with bdev->bd_mutex held.
1339  **/
1340 static int sd_open(struct block_device *bdev, fmode_t mode)
1341 {
1342 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1343 	struct scsi_device *sdev;
1344 	int retval;
1345 
1346 	if (!sdkp)
1347 		return -ENXIO;
1348 
1349 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1350 
1351 	sdev = sdkp->device;
1352 
1353 	/*
1354 	 * If the device is in error recovery, wait until it is done.
1355 	 * If the device is offline, then disallow any access to it.
1356 	 */
1357 	retval = -ENXIO;
1358 	if (!scsi_block_when_processing_errors(sdev))
1359 		goto error_out;
1360 
1361 	if (sdev->removable || sdkp->write_prot)
1362 		check_disk_change(bdev);
1363 
1364 	/*
1365 	 * If the drive is empty, just let the open fail.
1366 	 */
1367 	retval = -ENOMEDIUM;
1368 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1369 		goto error_out;
1370 
1371 	/*
1372 	 * If the device has the write protect tab set, have the open fail
1373 	 * if the user expects to be able to write to the thing.
1374 	 */
1375 	retval = -EROFS;
1376 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1377 		goto error_out;
1378 
1379 	/*
1380 	 * It is possible that the disk changing stuff resulted in
1381 	 * the device being taken offline.  If this is the case,
1382 	 * report this to the user, and don't pretend that the
1383 	 * open actually succeeded.
1384 	 */
1385 	retval = -ENXIO;
1386 	if (!scsi_device_online(sdev))
1387 		goto error_out;
1388 
1389 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1390 		if (scsi_block_when_processing_errors(sdev))
1391 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1392 	}
1393 
1394 	return 0;
1395 
1396 error_out:
1397 	scsi_disk_put(sdkp);
1398 	return retval;
1399 }
1400 
1401 /**
1402  *	sd_release - invoked when the (last) close(2) is called on this
1403  *	scsi disk.
1404  *	@disk: disk to release
1405  *	@mode: FMODE_* mask
1406  *
1407  *	Returns 0.
1408  *
1409  *	Note: may block (uninterruptible) if error recovery is underway
1410  *	on this disk.
1411  *
1412  *	Locking: called with bdev->bd_mutex held.
1413  **/
1414 static void sd_release(struct gendisk *disk, fmode_t mode)
1415 {
1416 	struct scsi_disk *sdkp = scsi_disk(disk);
1417 	struct scsi_device *sdev = sdkp->device;
1418 
1419 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1420 
1421 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1422 		if (scsi_block_when_processing_errors(sdev))
1423 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1424 	}
1425 
1426 	scsi_disk_put(sdkp);
1427 }
1428 
1429 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1430 {
1431 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1432 	struct scsi_device *sdp = sdkp->device;
1433 	struct Scsi_Host *host = sdp->host;
1434 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1435 	int diskinfo[4];
1436 
1437 	/* default to most commonly used values */
1438 	diskinfo[0] = 0x40;	/* 1 << 6 */
1439 	diskinfo[1] = 0x20;	/* 1 << 5 */
1440 	diskinfo[2] = capacity >> 11;
1441 
1442 	/* override with calculated, extended default, or driver values */
1443 	if (host->hostt->bios_param)
1444 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1445 	else
1446 		scsicam_bios_param(bdev, capacity, diskinfo);
1447 
1448 	geo->heads = diskinfo[0];
1449 	geo->sectors = diskinfo[1];
1450 	geo->cylinders = diskinfo[2];
1451 	return 0;
1452 }
1453 
1454 /**
1455  *	sd_ioctl - process an ioctl
1456  *	@bdev: target block device
1457  *	@mode: FMODE_* mask
1458  *	@cmd: ioctl command number
1459  *	@arg: this is third argument given to ioctl(2) system call.
1460  *	Often contains a pointer.
1461  *
1462  *	Returns 0 if successful (some ioctls return positive numbers on
1463  *	success as well). Returns a negated errno value in case of error.
1464  *
1465  *	Note: most ioctls are forward onto the block subsystem or further
1466  *	down in the scsi subsystem.
1467  **/
1468 static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1469 			   unsigned int cmd, void __user *p)
1470 {
1471 	struct gendisk *disk = bdev->bd_disk;
1472 	struct scsi_disk *sdkp = scsi_disk(disk);
1473 	struct scsi_device *sdp = sdkp->device;
1474 	int error;
1475 
1476 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1477 				    "cmd=0x%x\n", disk->disk_name, cmd));
1478 
1479 	error = scsi_verify_blk_ioctl(bdev, cmd);
1480 	if (error < 0)
1481 		return error;
1482 
1483 	/*
1484 	 * If we are in the middle of error recovery, don't let anyone
1485 	 * else try and use this device.  Also, if error recovery fails, it
1486 	 * may try and take the device offline, in which case all further
1487 	 * access to the device is prohibited.
1488 	 */
1489 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1490 			(mode & FMODE_NDELAY) != 0);
1491 	if (error)
1492 		goto out;
1493 
1494 	if (is_sed_ioctl(cmd))
1495 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1496 
1497 	/*
1498 	 * Send SCSI addressing ioctls directly to mid level, send other
1499 	 * ioctls to block level and then onto mid level if they can't be
1500 	 * resolved.
1501 	 */
1502 	switch (cmd) {
1503 		case SCSI_IOCTL_GET_IDLUN:
1504 		case SCSI_IOCTL_GET_BUS_NUMBER:
1505 			error = scsi_ioctl(sdp, cmd, p);
1506 			break;
1507 		default:
1508 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1509 			break;
1510 	}
1511 out:
1512 	return error;
1513 }
1514 
1515 static void set_media_not_present(struct scsi_disk *sdkp)
1516 {
1517 	if (sdkp->media_present)
1518 		sdkp->device->changed = 1;
1519 
1520 	if (sdkp->device->removable) {
1521 		sdkp->media_present = 0;
1522 		sdkp->capacity = 0;
1523 	}
1524 }
1525 
1526 static int media_not_present(struct scsi_disk *sdkp,
1527 			     struct scsi_sense_hdr *sshdr)
1528 {
1529 	if (!scsi_sense_valid(sshdr))
1530 		return 0;
1531 
1532 	/* not invoked for commands that could return deferred errors */
1533 	switch (sshdr->sense_key) {
1534 	case UNIT_ATTENTION:
1535 	case NOT_READY:
1536 		/* medium not present */
1537 		if (sshdr->asc == 0x3A) {
1538 			set_media_not_present(sdkp);
1539 			return 1;
1540 		}
1541 	}
1542 	return 0;
1543 }
1544 
1545 /**
1546  *	sd_check_events - check media events
1547  *	@disk: kernel device descriptor
1548  *	@clearing: disk events currently being cleared
1549  *
1550  *	Returns mask of DISK_EVENT_*.
1551  *
1552  *	Note: this function is invoked from the block subsystem.
1553  **/
1554 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1555 {
1556 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1557 	struct scsi_device *sdp;
1558 	int retval;
1559 
1560 	if (!sdkp)
1561 		return 0;
1562 
1563 	sdp = sdkp->device;
1564 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1565 
1566 	/*
1567 	 * If the device is offline, don't send any commands - just pretend as
1568 	 * if the command failed.  If the device ever comes back online, we
1569 	 * can deal with it then.  It is only because of unrecoverable errors
1570 	 * that we would ever take a device offline in the first place.
1571 	 */
1572 	if (!scsi_device_online(sdp)) {
1573 		set_media_not_present(sdkp);
1574 		goto out;
1575 	}
1576 
1577 	/*
1578 	 * Using TEST_UNIT_READY enables differentiation between drive with
1579 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1580 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1581 	 *
1582 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1583 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1584 	 * sd_revalidate() is called.
1585 	 */
1586 	if (scsi_block_when_processing_errors(sdp)) {
1587 		struct scsi_sense_hdr sshdr = { 0, };
1588 
1589 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1590 					      &sshdr);
1591 
1592 		/* failed to execute TUR, assume media not present */
1593 		if (host_byte(retval)) {
1594 			set_media_not_present(sdkp);
1595 			goto out;
1596 		}
1597 
1598 		if (media_not_present(sdkp, &sshdr))
1599 			goto out;
1600 	}
1601 
1602 	/*
1603 	 * For removable scsi disk we have to recognise the presence
1604 	 * of a disk in the drive.
1605 	 */
1606 	if (!sdkp->media_present)
1607 		sdp->changed = 1;
1608 	sdkp->media_present = 1;
1609 out:
1610 	/*
1611 	 * sdp->changed is set under the following conditions:
1612 	 *
1613 	 *	Medium present state has changed in either direction.
1614 	 *	Device has indicated UNIT_ATTENTION.
1615 	 */
1616 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1617 	sdp->changed = 0;
1618 	scsi_disk_put(sdkp);
1619 	return retval;
1620 }
1621 
1622 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1623 {
1624 	int retries, res;
1625 	struct scsi_device *sdp = sdkp->device;
1626 	const int timeout = sdp->request_queue->rq_timeout
1627 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1628 	struct scsi_sense_hdr my_sshdr;
1629 
1630 	if (!scsi_device_online(sdp))
1631 		return -ENODEV;
1632 
1633 	/* caller might not be interested in sense, but we need it */
1634 	if (!sshdr)
1635 		sshdr = &my_sshdr;
1636 
1637 	for (retries = 3; retries > 0; --retries) {
1638 		unsigned char cmd[10] = { 0 };
1639 
1640 		cmd[0] = SYNCHRONIZE_CACHE;
1641 		/*
1642 		 * Leave the rest of the command zero to indicate
1643 		 * flush everything.
1644 		 */
1645 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1646 				timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1647 		if (res == 0)
1648 			break;
1649 	}
1650 
1651 	if (res) {
1652 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1653 
1654 		if (driver_byte(res) == DRIVER_SENSE)
1655 			sd_print_sense_hdr(sdkp, sshdr);
1656 
1657 		/* we need to evaluate the error return  */
1658 		if (scsi_sense_valid(sshdr) &&
1659 			(sshdr->asc == 0x3a ||	/* medium not present */
1660 			 sshdr->asc == 0x20 ||	/* invalid command */
1661 			 (sshdr->asc == 0x74 && sshdr->ascq == 0x71)))	/* drive is password locked */
1662 				/* this is no error here */
1663 				return 0;
1664 
1665 		switch (host_byte(res)) {
1666 		/* ignore errors due to racing a disconnection */
1667 		case DID_BAD_TARGET:
1668 		case DID_NO_CONNECT:
1669 			return 0;
1670 		/* signal the upper layer it might try again */
1671 		case DID_BUS_BUSY:
1672 		case DID_IMM_RETRY:
1673 		case DID_REQUEUE:
1674 		case DID_SOFT_ERROR:
1675 			return -EBUSY;
1676 		default:
1677 			return -EIO;
1678 		}
1679 	}
1680 	return 0;
1681 }
1682 
1683 static void sd_rescan(struct device *dev)
1684 {
1685 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1686 
1687 	revalidate_disk(sdkp->disk);
1688 }
1689 
1690 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1691 		    unsigned int cmd, unsigned long arg)
1692 {
1693 	void __user *p = (void __user *)arg;
1694 	int ret;
1695 
1696 	ret = sd_ioctl_common(bdev, mode, cmd, p);
1697 	if (ret != -ENOTTY)
1698 		return ret;
1699 
1700 	return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1701 }
1702 
1703 #ifdef CONFIG_COMPAT
1704 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1705 			   unsigned int cmd, unsigned long arg)
1706 {
1707 	void __user *p = compat_ptr(arg);
1708 	int ret;
1709 
1710 	ret = sd_ioctl_common(bdev, mode, cmd, p);
1711 	if (ret != -ENOTTY)
1712 		return ret;
1713 
1714 	return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1715 }
1716 #endif
1717 
1718 static char sd_pr_type(enum pr_type type)
1719 {
1720 	switch (type) {
1721 	case PR_WRITE_EXCLUSIVE:
1722 		return 0x01;
1723 	case PR_EXCLUSIVE_ACCESS:
1724 		return 0x03;
1725 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1726 		return 0x05;
1727 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1728 		return 0x06;
1729 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1730 		return 0x07;
1731 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1732 		return 0x08;
1733 	default:
1734 		return 0;
1735 	}
1736 };
1737 
1738 static int sd_pr_command(struct block_device *bdev, u8 sa,
1739 		u64 key, u64 sa_key, u8 type, u8 flags)
1740 {
1741 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1742 	struct scsi_sense_hdr sshdr;
1743 	int result;
1744 	u8 cmd[16] = { 0, };
1745 	u8 data[24] = { 0, };
1746 
1747 	cmd[0] = PERSISTENT_RESERVE_OUT;
1748 	cmd[1] = sa;
1749 	cmd[2] = type;
1750 	put_unaligned_be32(sizeof(data), &cmd[5]);
1751 
1752 	put_unaligned_be64(key, &data[0]);
1753 	put_unaligned_be64(sa_key, &data[8]);
1754 	data[20] = flags;
1755 
1756 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1757 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1758 
1759 	if (driver_byte(result) == DRIVER_SENSE &&
1760 	    scsi_sense_valid(&sshdr)) {
1761 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1762 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1763 	}
1764 
1765 	return result;
1766 }
1767 
1768 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1769 		u32 flags)
1770 {
1771 	if (flags & ~PR_FL_IGNORE_KEY)
1772 		return -EOPNOTSUPP;
1773 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1774 			old_key, new_key, 0,
1775 			(1 << 0) /* APTPL */);
1776 }
1777 
1778 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1779 		u32 flags)
1780 {
1781 	if (flags)
1782 		return -EOPNOTSUPP;
1783 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1784 }
1785 
1786 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1787 {
1788 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1789 }
1790 
1791 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1792 		enum pr_type type, bool abort)
1793 {
1794 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1795 			     sd_pr_type(type), 0);
1796 }
1797 
1798 static int sd_pr_clear(struct block_device *bdev, u64 key)
1799 {
1800 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1801 }
1802 
1803 static const struct pr_ops sd_pr_ops = {
1804 	.pr_register	= sd_pr_register,
1805 	.pr_reserve	= sd_pr_reserve,
1806 	.pr_release	= sd_pr_release,
1807 	.pr_preempt	= sd_pr_preempt,
1808 	.pr_clear	= sd_pr_clear,
1809 };
1810 
1811 static const struct block_device_operations sd_fops = {
1812 	.owner			= THIS_MODULE,
1813 	.open			= sd_open,
1814 	.release		= sd_release,
1815 	.ioctl			= sd_ioctl,
1816 	.getgeo			= sd_getgeo,
1817 #ifdef CONFIG_COMPAT
1818 	.compat_ioctl		= sd_compat_ioctl,
1819 #endif
1820 	.check_events		= sd_check_events,
1821 	.revalidate_disk	= sd_revalidate_disk,
1822 	.unlock_native_capacity	= sd_unlock_native_capacity,
1823 	.report_zones		= sd_zbc_report_zones,
1824 	.pr_ops			= &sd_pr_ops,
1825 };
1826 
1827 /**
1828  *	sd_eh_reset - reset error handling callback
1829  *	@scmd:		sd-issued command that has failed
1830  *
1831  *	This function is called by the SCSI midlayer before starting
1832  *	SCSI EH. When counting medium access failures we have to be
1833  *	careful to register it only only once per device and SCSI EH run;
1834  *	there might be several timed out commands which will cause the
1835  *	'max_medium_access_timeouts' counter to trigger after the first
1836  *	SCSI EH run already and set the device to offline.
1837  *	So this function resets the internal counter before starting SCSI EH.
1838  **/
1839 static void sd_eh_reset(struct scsi_cmnd *scmd)
1840 {
1841 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1842 
1843 	/* New SCSI EH run, reset gate variable */
1844 	sdkp->ignore_medium_access_errors = false;
1845 }
1846 
1847 /**
1848  *	sd_eh_action - error handling callback
1849  *	@scmd:		sd-issued command that has failed
1850  *	@eh_disp:	The recovery disposition suggested by the midlayer
1851  *
1852  *	This function is called by the SCSI midlayer upon completion of an
1853  *	error test command (currently TEST UNIT READY). The result of sending
1854  *	the eh command is passed in eh_disp.  We're looking for devices that
1855  *	fail medium access commands but are OK with non access commands like
1856  *	test unit ready (so wrongly see the device as having a successful
1857  *	recovery)
1858  **/
1859 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1860 {
1861 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1862 	struct scsi_device *sdev = scmd->device;
1863 
1864 	if (!scsi_device_online(sdev) ||
1865 	    !scsi_medium_access_command(scmd) ||
1866 	    host_byte(scmd->result) != DID_TIME_OUT ||
1867 	    eh_disp != SUCCESS)
1868 		return eh_disp;
1869 
1870 	/*
1871 	 * The device has timed out executing a medium access command.
1872 	 * However, the TEST UNIT READY command sent during error
1873 	 * handling completed successfully. Either the device is in the
1874 	 * process of recovering or has it suffered an internal failure
1875 	 * that prevents access to the storage medium.
1876 	 */
1877 	if (!sdkp->ignore_medium_access_errors) {
1878 		sdkp->medium_access_timed_out++;
1879 		sdkp->ignore_medium_access_errors = true;
1880 	}
1881 
1882 	/*
1883 	 * If the device keeps failing read/write commands but TEST UNIT
1884 	 * READY always completes successfully we assume that medium
1885 	 * access is no longer possible and take the device offline.
1886 	 */
1887 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1888 		scmd_printk(KERN_ERR, scmd,
1889 			    "Medium access timeout failure. Offlining disk!\n");
1890 		mutex_lock(&sdev->state_mutex);
1891 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1892 		mutex_unlock(&sdev->state_mutex);
1893 
1894 		return SUCCESS;
1895 	}
1896 
1897 	return eh_disp;
1898 }
1899 
1900 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1901 {
1902 	struct request *req = scmd->request;
1903 	struct scsi_device *sdev = scmd->device;
1904 	unsigned int transferred, good_bytes;
1905 	u64 start_lba, end_lba, bad_lba;
1906 
1907 	/*
1908 	 * Some commands have a payload smaller than the device logical
1909 	 * block size (e.g. INQUIRY on a 4K disk).
1910 	 */
1911 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1912 		return 0;
1913 
1914 	/* Check if we have a 'bad_lba' information */
1915 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1916 				     SCSI_SENSE_BUFFERSIZE,
1917 				     &bad_lba))
1918 		return 0;
1919 
1920 	/*
1921 	 * If the bad lba was reported incorrectly, we have no idea where
1922 	 * the error is.
1923 	 */
1924 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1925 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1926 	if (bad_lba < start_lba || bad_lba >= end_lba)
1927 		return 0;
1928 
1929 	/*
1930 	 * resid is optional but mostly filled in.  When it's unused,
1931 	 * its value is zero, so we assume the whole buffer transferred
1932 	 */
1933 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1934 
1935 	/* This computation should always be done in terms of the
1936 	 * resolution of the device's medium.
1937 	 */
1938 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1939 
1940 	return min(good_bytes, transferred);
1941 }
1942 
1943 /**
1944  *	sd_done - bottom half handler: called when the lower level
1945  *	driver has completed (successfully or otherwise) a scsi command.
1946  *	@SCpnt: mid-level's per command structure.
1947  *
1948  *	Note: potentially run from within an ISR. Must not block.
1949  **/
1950 static int sd_done(struct scsi_cmnd *SCpnt)
1951 {
1952 	int result = SCpnt->result;
1953 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1954 	unsigned int sector_size = SCpnt->device->sector_size;
1955 	unsigned int resid;
1956 	struct scsi_sense_hdr sshdr;
1957 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1958 	struct request *req = SCpnt->request;
1959 	int sense_valid = 0;
1960 	int sense_deferred = 0;
1961 
1962 	switch (req_op(req)) {
1963 	case REQ_OP_DISCARD:
1964 	case REQ_OP_WRITE_ZEROES:
1965 	case REQ_OP_WRITE_SAME:
1966 	case REQ_OP_ZONE_RESET:
1967 	case REQ_OP_ZONE_RESET_ALL:
1968 	case REQ_OP_ZONE_OPEN:
1969 	case REQ_OP_ZONE_CLOSE:
1970 	case REQ_OP_ZONE_FINISH:
1971 		if (!result) {
1972 			good_bytes = blk_rq_bytes(req);
1973 			scsi_set_resid(SCpnt, 0);
1974 		} else {
1975 			good_bytes = 0;
1976 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1977 		}
1978 		break;
1979 	default:
1980 		/*
1981 		 * In case of bogus fw or device, we could end up having
1982 		 * an unaligned partial completion. Check this here and force
1983 		 * alignment.
1984 		 */
1985 		resid = scsi_get_resid(SCpnt);
1986 		if (resid & (sector_size - 1)) {
1987 			sd_printk(KERN_INFO, sdkp,
1988 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1989 				resid, sector_size);
1990 			scsi_print_command(SCpnt);
1991 			resid = min(scsi_bufflen(SCpnt),
1992 				    round_up(resid, sector_size));
1993 			scsi_set_resid(SCpnt, resid);
1994 		}
1995 	}
1996 
1997 	if (result) {
1998 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1999 		if (sense_valid)
2000 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2001 	}
2002 	sdkp->medium_access_timed_out = 0;
2003 
2004 	if (driver_byte(result) != DRIVER_SENSE &&
2005 	    (!sense_valid || sense_deferred))
2006 		goto out;
2007 
2008 	switch (sshdr.sense_key) {
2009 	case HARDWARE_ERROR:
2010 	case MEDIUM_ERROR:
2011 		good_bytes = sd_completed_bytes(SCpnt);
2012 		break;
2013 	case RECOVERED_ERROR:
2014 		good_bytes = scsi_bufflen(SCpnt);
2015 		break;
2016 	case NO_SENSE:
2017 		/* This indicates a false check condition, so ignore it.  An
2018 		 * unknown amount of data was transferred so treat it as an
2019 		 * error.
2020 		 */
2021 		SCpnt->result = 0;
2022 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2023 		break;
2024 	case ABORTED_COMMAND:
2025 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2026 			good_bytes = sd_completed_bytes(SCpnt);
2027 		break;
2028 	case ILLEGAL_REQUEST:
2029 		switch (sshdr.asc) {
2030 		case 0x10:	/* DIX: Host detected corruption */
2031 			good_bytes = sd_completed_bytes(SCpnt);
2032 			break;
2033 		case 0x20:	/* INVALID COMMAND OPCODE */
2034 		case 0x24:	/* INVALID FIELD IN CDB */
2035 			switch (SCpnt->cmnd[0]) {
2036 			case UNMAP:
2037 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2038 				break;
2039 			case WRITE_SAME_16:
2040 			case WRITE_SAME:
2041 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2042 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2043 				} else {
2044 					sdkp->device->no_write_same = 1;
2045 					sd_config_write_same(sdkp);
2046 					req->rq_flags |= RQF_QUIET;
2047 				}
2048 				break;
2049 			}
2050 		}
2051 		break;
2052 	default:
2053 		break;
2054 	}
2055 
2056  out:
2057 	if (sd_is_zoned(sdkp))
2058 		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2059 
2060 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2061 					   "sd_done: completed %d of %d bytes\n",
2062 					   good_bytes, scsi_bufflen(SCpnt)));
2063 
2064 	return good_bytes;
2065 }
2066 
2067 /*
2068  * spinup disk - called only in sd_revalidate_disk()
2069  */
2070 static void
2071 sd_spinup_disk(struct scsi_disk *sdkp)
2072 {
2073 	unsigned char cmd[10];
2074 	unsigned long spintime_expire = 0;
2075 	int retries, spintime;
2076 	unsigned int the_result;
2077 	struct scsi_sense_hdr sshdr;
2078 	int sense_valid = 0;
2079 
2080 	spintime = 0;
2081 
2082 	/* Spin up drives, as required.  Only do this at boot time */
2083 	/* Spinup needs to be done for module loads too. */
2084 	do {
2085 		retries = 0;
2086 
2087 		do {
2088 			cmd[0] = TEST_UNIT_READY;
2089 			memset((void *) &cmd[1], 0, 9);
2090 
2091 			the_result = scsi_execute_req(sdkp->device, cmd,
2092 						      DMA_NONE, NULL, 0,
2093 						      &sshdr, SD_TIMEOUT,
2094 						      SD_MAX_RETRIES, NULL);
2095 
2096 			/*
2097 			 * If the drive has indicated to us that it
2098 			 * doesn't have any media in it, don't bother
2099 			 * with any more polling.
2100 			 */
2101 			if (media_not_present(sdkp, &sshdr))
2102 				return;
2103 
2104 			if (the_result)
2105 				sense_valid = scsi_sense_valid(&sshdr);
2106 			retries++;
2107 		} while (retries < 3 &&
2108 			 (!scsi_status_is_good(the_result) ||
2109 			  ((driver_byte(the_result) == DRIVER_SENSE) &&
2110 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2111 
2112 		if (driver_byte(the_result) != DRIVER_SENSE) {
2113 			/* no sense, TUR either succeeded or failed
2114 			 * with a status error */
2115 			if(!spintime && !scsi_status_is_good(the_result)) {
2116 				sd_print_result(sdkp, "Test Unit Ready failed",
2117 						the_result);
2118 			}
2119 			break;
2120 		}
2121 
2122 		/*
2123 		 * The device does not want the automatic start to be issued.
2124 		 */
2125 		if (sdkp->device->no_start_on_add)
2126 			break;
2127 
2128 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2129 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2130 				break;	/* manual intervention required */
2131 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2132 				break;	/* standby */
2133 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2134 				break;	/* unavailable */
2135 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2136 				break;	/* sanitize in progress */
2137 			/*
2138 			 * Issue command to spin up drive when not ready
2139 			 */
2140 			if (!spintime) {
2141 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2142 				cmd[0] = START_STOP;
2143 				cmd[1] = 1;	/* Return immediately */
2144 				memset((void *) &cmd[2], 0, 8);
2145 				cmd[4] = 1;	/* Start spin cycle */
2146 				if (sdkp->device->start_stop_pwr_cond)
2147 					cmd[4] |= 1 << 4;
2148 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2149 						 NULL, 0, &sshdr,
2150 						 SD_TIMEOUT, SD_MAX_RETRIES,
2151 						 NULL);
2152 				spintime_expire = jiffies + 100 * HZ;
2153 				spintime = 1;
2154 			}
2155 			/* Wait 1 second for next try */
2156 			msleep(1000);
2157 			printk(KERN_CONT ".");
2158 
2159 		/*
2160 		 * Wait for USB flash devices with slow firmware.
2161 		 * Yes, this sense key/ASC combination shouldn't
2162 		 * occur here.  It's characteristic of these devices.
2163 		 */
2164 		} else if (sense_valid &&
2165 				sshdr.sense_key == UNIT_ATTENTION &&
2166 				sshdr.asc == 0x28) {
2167 			if (!spintime) {
2168 				spintime_expire = jiffies + 5 * HZ;
2169 				spintime = 1;
2170 			}
2171 			/* Wait 1 second for next try */
2172 			msleep(1000);
2173 		} else {
2174 			/* we don't understand the sense code, so it's
2175 			 * probably pointless to loop */
2176 			if(!spintime) {
2177 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2178 				sd_print_sense_hdr(sdkp, &sshdr);
2179 			}
2180 			break;
2181 		}
2182 
2183 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2184 
2185 	if (spintime) {
2186 		if (scsi_status_is_good(the_result))
2187 			printk(KERN_CONT "ready\n");
2188 		else
2189 			printk(KERN_CONT "not responding...\n");
2190 	}
2191 }
2192 
2193 /*
2194  * Determine whether disk supports Data Integrity Field.
2195  */
2196 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2197 {
2198 	struct scsi_device *sdp = sdkp->device;
2199 	u8 type;
2200 	int ret = 0;
2201 
2202 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2203 		sdkp->protection_type = 0;
2204 		return ret;
2205 	}
2206 
2207 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2208 
2209 	if (type > T10_PI_TYPE3_PROTECTION)
2210 		ret = -ENODEV;
2211 	else if (scsi_host_dif_capable(sdp->host, type))
2212 		ret = 1;
2213 
2214 	if (sdkp->first_scan || type != sdkp->protection_type)
2215 		switch (ret) {
2216 		case -ENODEV:
2217 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2218 				  " protection type %u. Disabling disk!\n",
2219 				  type);
2220 			break;
2221 		case 1:
2222 			sd_printk(KERN_NOTICE, sdkp,
2223 				  "Enabling DIF Type %u protection\n", type);
2224 			break;
2225 		case 0:
2226 			sd_printk(KERN_NOTICE, sdkp,
2227 				  "Disabling DIF Type %u protection\n", type);
2228 			break;
2229 		}
2230 
2231 	sdkp->protection_type = type;
2232 
2233 	return ret;
2234 }
2235 
2236 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2237 			struct scsi_sense_hdr *sshdr, int sense_valid,
2238 			int the_result)
2239 {
2240 	if (driver_byte(the_result) == DRIVER_SENSE)
2241 		sd_print_sense_hdr(sdkp, sshdr);
2242 	else
2243 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2244 
2245 	/*
2246 	 * Set dirty bit for removable devices if not ready -
2247 	 * sometimes drives will not report this properly.
2248 	 */
2249 	if (sdp->removable &&
2250 	    sense_valid && sshdr->sense_key == NOT_READY)
2251 		set_media_not_present(sdkp);
2252 
2253 	/*
2254 	 * We used to set media_present to 0 here to indicate no media
2255 	 * in the drive, but some drives fail read capacity even with
2256 	 * media present, so we can't do that.
2257 	 */
2258 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2259 }
2260 
2261 #define RC16_LEN 32
2262 #if RC16_LEN > SD_BUF_SIZE
2263 #error RC16_LEN must not be more than SD_BUF_SIZE
2264 #endif
2265 
2266 #define READ_CAPACITY_RETRIES_ON_RESET	10
2267 
2268 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2269 						unsigned char *buffer)
2270 {
2271 	unsigned char cmd[16];
2272 	struct scsi_sense_hdr sshdr;
2273 	int sense_valid = 0;
2274 	int the_result;
2275 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2276 	unsigned int alignment;
2277 	unsigned long long lba;
2278 	unsigned sector_size;
2279 
2280 	if (sdp->no_read_capacity_16)
2281 		return -EINVAL;
2282 
2283 	do {
2284 		memset(cmd, 0, 16);
2285 		cmd[0] = SERVICE_ACTION_IN_16;
2286 		cmd[1] = SAI_READ_CAPACITY_16;
2287 		cmd[13] = RC16_LEN;
2288 		memset(buffer, 0, RC16_LEN);
2289 
2290 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2291 					buffer, RC16_LEN, &sshdr,
2292 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2293 
2294 		if (media_not_present(sdkp, &sshdr))
2295 			return -ENODEV;
2296 
2297 		if (the_result) {
2298 			sense_valid = scsi_sense_valid(&sshdr);
2299 			if (sense_valid &&
2300 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2301 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2302 			    sshdr.ascq == 0x00)
2303 				/* Invalid Command Operation Code or
2304 				 * Invalid Field in CDB, just retry
2305 				 * silently with RC10 */
2306 				return -EINVAL;
2307 			if (sense_valid &&
2308 			    sshdr.sense_key == UNIT_ATTENTION &&
2309 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2310 				/* Device reset might occur several times,
2311 				 * give it one more chance */
2312 				if (--reset_retries > 0)
2313 					continue;
2314 		}
2315 		retries--;
2316 
2317 	} while (the_result && retries);
2318 
2319 	if (the_result) {
2320 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2321 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2322 		return -EINVAL;
2323 	}
2324 
2325 	sector_size = get_unaligned_be32(&buffer[8]);
2326 	lba = get_unaligned_be64(&buffer[0]);
2327 
2328 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2329 		sdkp->capacity = 0;
2330 		return -ENODEV;
2331 	}
2332 
2333 	/* Logical blocks per physical block exponent */
2334 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2335 
2336 	/* RC basis */
2337 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2338 
2339 	/* Lowest aligned logical block */
2340 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2341 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2342 	if (alignment && sdkp->first_scan)
2343 		sd_printk(KERN_NOTICE, sdkp,
2344 			  "physical block alignment offset: %u\n", alignment);
2345 
2346 	if (buffer[14] & 0x80) { /* LBPME */
2347 		sdkp->lbpme = 1;
2348 
2349 		if (buffer[14] & 0x40) /* LBPRZ */
2350 			sdkp->lbprz = 1;
2351 
2352 		sd_config_discard(sdkp, SD_LBP_WS16);
2353 	}
2354 
2355 	sdkp->capacity = lba + 1;
2356 	return sector_size;
2357 }
2358 
2359 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2360 						unsigned char *buffer)
2361 {
2362 	unsigned char cmd[16];
2363 	struct scsi_sense_hdr sshdr;
2364 	int sense_valid = 0;
2365 	int the_result;
2366 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2367 	sector_t lba;
2368 	unsigned sector_size;
2369 
2370 	do {
2371 		cmd[0] = READ_CAPACITY;
2372 		memset(&cmd[1], 0, 9);
2373 		memset(buffer, 0, 8);
2374 
2375 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2376 					buffer, 8, &sshdr,
2377 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2378 
2379 		if (media_not_present(sdkp, &sshdr))
2380 			return -ENODEV;
2381 
2382 		if (the_result) {
2383 			sense_valid = scsi_sense_valid(&sshdr);
2384 			if (sense_valid &&
2385 			    sshdr.sense_key == UNIT_ATTENTION &&
2386 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2387 				/* Device reset might occur several times,
2388 				 * give it one more chance */
2389 				if (--reset_retries > 0)
2390 					continue;
2391 		}
2392 		retries--;
2393 
2394 	} while (the_result && retries);
2395 
2396 	if (the_result) {
2397 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2398 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2399 		return -EINVAL;
2400 	}
2401 
2402 	sector_size = get_unaligned_be32(&buffer[4]);
2403 	lba = get_unaligned_be32(&buffer[0]);
2404 
2405 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2406 		/* Some buggy (usb cardreader) devices return an lba of
2407 		   0xffffffff when the want to report a size of 0 (with
2408 		   which they really mean no media is present) */
2409 		sdkp->capacity = 0;
2410 		sdkp->physical_block_size = sector_size;
2411 		return sector_size;
2412 	}
2413 
2414 	sdkp->capacity = lba + 1;
2415 	sdkp->physical_block_size = sector_size;
2416 	return sector_size;
2417 }
2418 
2419 static int sd_try_rc16_first(struct scsi_device *sdp)
2420 {
2421 	if (sdp->host->max_cmd_len < 16)
2422 		return 0;
2423 	if (sdp->try_rc_10_first)
2424 		return 0;
2425 	if (sdp->scsi_level > SCSI_SPC_2)
2426 		return 1;
2427 	if (scsi_device_protection(sdp))
2428 		return 1;
2429 	return 0;
2430 }
2431 
2432 /*
2433  * read disk capacity
2434  */
2435 static void
2436 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2437 {
2438 	int sector_size;
2439 	struct scsi_device *sdp = sdkp->device;
2440 
2441 	if (sd_try_rc16_first(sdp)) {
2442 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2443 		if (sector_size == -EOVERFLOW)
2444 			goto got_data;
2445 		if (sector_size == -ENODEV)
2446 			return;
2447 		if (sector_size < 0)
2448 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2449 		if (sector_size < 0)
2450 			return;
2451 	} else {
2452 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2453 		if (sector_size == -EOVERFLOW)
2454 			goto got_data;
2455 		if (sector_size < 0)
2456 			return;
2457 		if ((sizeof(sdkp->capacity) > 4) &&
2458 		    (sdkp->capacity > 0xffffffffULL)) {
2459 			int old_sector_size = sector_size;
2460 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2461 					"Trying to use READ CAPACITY(16).\n");
2462 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2463 			if (sector_size < 0) {
2464 				sd_printk(KERN_NOTICE, sdkp,
2465 					"Using 0xffffffff as device size\n");
2466 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2467 				sector_size = old_sector_size;
2468 				goto got_data;
2469 			}
2470 			/* Remember that READ CAPACITY(16) succeeded */
2471 			sdp->try_rc_10_first = 0;
2472 		}
2473 	}
2474 
2475 	/* Some devices are known to return the total number of blocks,
2476 	 * not the highest block number.  Some devices have versions
2477 	 * which do this and others which do not.  Some devices we might
2478 	 * suspect of doing this but we don't know for certain.
2479 	 *
2480 	 * If we know the reported capacity is wrong, decrement it.  If
2481 	 * we can only guess, then assume the number of blocks is even
2482 	 * (usually true but not always) and err on the side of lowering
2483 	 * the capacity.
2484 	 */
2485 	if (sdp->fix_capacity ||
2486 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2487 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2488 				"from its reported value: %llu\n",
2489 				(unsigned long long) sdkp->capacity);
2490 		--sdkp->capacity;
2491 	}
2492 
2493 got_data:
2494 	if (sector_size == 0) {
2495 		sector_size = 512;
2496 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2497 			  "assuming 512.\n");
2498 	}
2499 
2500 	if (sector_size != 512 &&
2501 	    sector_size != 1024 &&
2502 	    sector_size != 2048 &&
2503 	    sector_size != 4096) {
2504 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2505 			  sector_size);
2506 		/*
2507 		 * The user might want to re-format the drive with
2508 		 * a supported sectorsize.  Once this happens, it
2509 		 * would be relatively trivial to set the thing up.
2510 		 * For this reason, we leave the thing in the table.
2511 		 */
2512 		sdkp->capacity = 0;
2513 		/*
2514 		 * set a bogus sector size so the normal read/write
2515 		 * logic in the block layer will eventually refuse any
2516 		 * request on this device without tripping over power
2517 		 * of two sector size assumptions
2518 		 */
2519 		sector_size = 512;
2520 	}
2521 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2522 	blk_queue_physical_block_size(sdp->request_queue,
2523 				      sdkp->physical_block_size);
2524 	sdkp->device->sector_size = sector_size;
2525 
2526 	if (sdkp->capacity > 0xffffffff)
2527 		sdp->use_16_for_rw = 1;
2528 
2529 }
2530 
2531 /*
2532  * Print disk capacity
2533  */
2534 static void
2535 sd_print_capacity(struct scsi_disk *sdkp,
2536 		  sector_t old_capacity)
2537 {
2538 	int sector_size = sdkp->device->sector_size;
2539 	char cap_str_2[10], cap_str_10[10];
2540 
2541 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2542 		return;
2543 
2544 	string_get_size(sdkp->capacity, sector_size,
2545 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2546 	string_get_size(sdkp->capacity, sector_size,
2547 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2548 
2549 	sd_printk(KERN_NOTICE, sdkp,
2550 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2551 		  (unsigned long long)sdkp->capacity,
2552 		  sector_size, cap_str_10, cap_str_2);
2553 
2554 	if (sdkp->physical_block_size != sector_size)
2555 		sd_printk(KERN_NOTICE, sdkp,
2556 			  "%u-byte physical blocks\n",
2557 			  sdkp->physical_block_size);
2558 
2559 	sd_zbc_print_zones(sdkp);
2560 }
2561 
2562 /* called with buffer of length 512 */
2563 static inline int
2564 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2565 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2566 		 struct scsi_sense_hdr *sshdr)
2567 {
2568 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2569 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2570 			       sshdr);
2571 }
2572 
2573 /*
2574  * read write protect setting, if possible - called only in sd_revalidate_disk()
2575  * called with buffer of length SD_BUF_SIZE
2576  */
2577 static void
2578 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2579 {
2580 	int res;
2581 	struct scsi_device *sdp = sdkp->device;
2582 	struct scsi_mode_data data;
2583 	int old_wp = sdkp->write_prot;
2584 
2585 	set_disk_ro(sdkp->disk, 0);
2586 	if (sdp->skip_ms_page_3f) {
2587 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2588 		return;
2589 	}
2590 
2591 	if (sdp->use_192_bytes_for_3f) {
2592 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2593 	} else {
2594 		/*
2595 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2596 		 * We have to start carefully: some devices hang if we ask
2597 		 * for more than is available.
2598 		 */
2599 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2600 
2601 		/*
2602 		 * Second attempt: ask for page 0 When only page 0 is
2603 		 * implemented, a request for page 3F may return Sense Key
2604 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2605 		 * CDB.
2606 		 */
2607 		if (!scsi_status_is_good(res))
2608 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2609 
2610 		/*
2611 		 * Third attempt: ask 255 bytes, as we did earlier.
2612 		 */
2613 		if (!scsi_status_is_good(res))
2614 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2615 					       &data, NULL);
2616 	}
2617 
2618 	if (!scsi_status_is_good(res)) {
2619 		sd_first_printk(KERN_WARNING, sdkp,
2620 			  "Test WP failed, assume Write Enabled\n");
2621 	} else {
2622 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2623 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2624 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2625 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2626 				  sdkp->write_prot ? "on" : "off");
2627 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2628 		}
2629 	}
2630 }
2631 
2632 /*
2633  * sd_read_cache_type - called only from sd_revalidate_disk()
2634  * called with buffer of length SD_BUF_SIZE
2635  */
2636 static void
2637 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2638 {
2639 	int len = 0, res;
2640 	struct scsi_device *sdp = sdkp->device;
2641 
2642 	int dbd;
2643 	int modepage;
2644 	int first_len;
2645 	struct scsi_mode_data data;
2646 	struct scsi_sense_hdr sshdr;
2647 	int old_wce = sdkp->WCE;
2648 	int old_rcd = sdkp->RCD;
2649 	int old_dpofua = sdkp->DPOFUA;
2650 
2651 
2652 	if (sdkp->cache_override)
2653 		return;
2654 
2655 	first_len = 4;
2656 	if (sdp->skip_ms_page_8) {
2657 		if (sdp->type == TYPE_RBC)
2658 			goto defaults;
2659 		else {
2660 			if (sdp->skip_ms_page_3f)
2661 				goto defaults;
2662 			modepage = 0x3F;
2663 			if (sdp->use_192_bytes_for_3f)
2664 				first_len = 192;
2665 			dbd = 0;
2666 		}
2667 	} else if (sdp->type == TYPE_RBC) {
2668 		modepage = 6;
2669 		dbd = 8;
2670 	} else {
2671 		modepage = 8;
2672 		dbd = 0;
2673 	}
2674 
2675 	/* cautiously ask */
2676 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2677 			&data, &sshdr);
2678 
2679 	if (!scsi_status_is_good(res))
2680 		goto bad_sense;
2681 
2682 	if (!data.header_length) {
2683 		modepage = 6;
2684 		first_len = 0;
2685 		sd_first_printk(KERN_ERR, sdkp,
2686 				"Missing header in MODE_SENSE response\n");
2687 	}
2688 
2689 	/* that went OK, now ask for the proper length */
2690 	len = data.length;
2691 
2692 	/*
2693 	 * We're only interested in the first three bytes, actually.
2694 	 * But the data cache page is defined for the first 20.
2695 	 */
2696 	if (len < 3)
2697 		goto bad_sense;
2698 	else if (len > SD_BUF_SIZE) {
2699 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2700 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2701 		len = SD_BUF_SIZE;
2702 	}
2703 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2704 		len = 192;
2705 
2706 	/* Get the data */
2707 	if (len > first_len)
2708 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2709 				&data, &sshdr);
2710 
2711 	if (scsi_status_is_good(res)) {
2712 		int offset = data.header_length + data.block_descriptor_length;
2713 
2714 		while (offset < len) {
2715 			u8 page_code = buffer[offset] & 0x3F;
2716 			u8 spf       = buffer[offset] & 0x40;
2717 
2718 			if (page_code == 8 || page_code == 6) {
2719 				/* We're interested only in the first 3 bytes.
2720 				 */
2721 				if (len - offset <= 2) {
2722 					sd_first_printk(KERN_ERR, sdkp,
2723 						"Incomplete mode parameter "
2724 							"data\n");
2725 					goto defaults;
2726 				} else {
2727 					modepage = page_code;
2728 					goto Page_found;
2729 				}
2730 			} else {
2731 				/* Go to the next page */
2732 				if (spf && len - offset > 3)
2733 					offset += 4 + (buffer[offset+2] << 8) +
2734 						buffer[offset+3];
2735 				else if (!spf && len - offset > 1)
2736 					offset += 2 + buffer[offset+1];
2737 				else {
2738 					sd_first_printk(KERN_ERR, sdkp,
2739 							"Incomplete mode "
2740 							"parameter data\n");
2741 					goto defaults;
2742 				}
2743 			}
2744 		}
2745 
2746 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2747 		goto defaults;
2748 
2749 	Page_found:
2750 		if (modepage == 8) {
2751 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2752 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2753 		} else {
2754 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2755 			sdkp->RCD = 0;
2756 		}
2757 
2758 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2759 		if (sdp->broken_fua) {
2760 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2761 			sdkp->DPOFUA = 0;
2762 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2763 			   !sdkp->device->use_16_for_rw) {
2764 			sd_first_printk(KERN_NOTICE, sdkp,
2765 				  "Uses READ/WRITE(6), disabling FUA\n");
2766 			sdkp->DPOFUA = 0;
2767 		}
2768 
2769 		/* No cache flush allowed for write protected devices */
2770 		if (sdkp->WCE && sdkp->write_prot)
2771 			sdkp->WCE = 0;
2772 
2773 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2774 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2775 			sd_printk(KERN_NOTICE, sdkp,
2776 				  "Write cache: %s, read cache: %s, %s\n",
2777 				  sdkp->WCE ? "enabled" : "disabled",
2778 				  sdkp->RCD ? "disabled" : "enabled",
2779 				  sdkp->DPOFUA ? "supports DPO and FUA"
2780 				  : "doesn't support DPO or FUA");
2781 
2782 		return;
2783 	}
2784 
2785 bad_sense:
2786 	if (scsi_sense_valid(&sshdr) &&
2787 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2788 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2789 		/* Invalid field in CDB */
2790 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2791 	else
2792 		sd_first_printk(KERN_ERR, sdkp,
2793 				"Asking for cache data failed\n");
2794 
2795 defaults:
2796 	if (sdp->wce_default_on) {
2797 		sd_first_printk(KERN_NOTICE, sdkp,
2798 				"Assuming drive cache: write back\n");
2799 		sdkp->WCE = 1;
2800 	} else {
2801 		sd_first_printk(KERN_ERR, sdkp,
2802 				"Assuming drive cache: write through\n");
2803 		sdkp->WCE = 0;
2804 	}
2805 	sdkp->RCD = 0;
2806 	sdkp->DPOFUA = 0;
2807 }
2808 
2809 /*
2810  * The ATO bit indicates whether the DIF application tag is available
2811  * for use by the operating system.
2812  */
2813 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2814 {
2815 	int res, offset;
2816 	struct scsi_device *sdp = sdkp->device;
2817 	struct scsi_mode_data data;
2818 	struct scsi_sense_hdr sshdr;
2819 
2820 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2821 		return;
2822 
2823 	if (sdkp->protection_type == 0)
2824 		return;
2825 
2826 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2827 			      SD_MAX_RETRIES, &data, &sshdr);
2828 
2829 	if (!scsi_status_is_good(res) || !data.header_length ||
2830 	    data.length < 6) {
2831 		sd_first_printk(KERN_WARNING, sdkp,
2832 			  "getting Control mode page failed, assume no ATO\n");
2833 
2834 		if (scsi_sense_valid(&sshdr))
2835 			sd_print_sense_hdr(sdkp, &sshdr);
2836 
2837 		return;
2838 	}
2839 
2840 	offset = data.header_length + data.block_descriptor_length;
2841 
2842 	if ((buffer[offset] & 0x3f) != 0x0a) {
2843 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2844 		return;
2845 	}
2846 
2847 	if ((buffer[offset + 5] & 0x80) == 0)
2848 		return;
2849 
2850 	sdkp->ATO = 1;
2851 
2852 	return;
2853 }
2854 
2855 /**
2856  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2857  * @sdkp: disk to query
2858  */
2859 static void sd_read_block_limits(struct scsi_disk *sdkp)
2860 {
2861 	unsigned int sector_sz = sdkp->device->sector_size;
2862 	const int vpd_len = 64;
2863 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2864 
2865 	if (!buffer ||
2866 	    /* Block Limits VPD */
2867 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2868 		goto out;
2869 
2870 	blk_queue_io_min(sdkp->disk->queue,
2871 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2872 
2873 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2874 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2875 
2876 	if (buffer[3] == 0x3c) {
2877 		unsigned int lba_count, desc_count;
2878 
2879 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2880 
2881 		if (!sdkp->lbpme)
2882 			goto out;
2883 
2884 		lba_count = get_unaligned_be32(&buffer[20]);
2885 		desc_count = get_unaligned_be32(&buffer[24]);
2886 
2887 		if (lba_count && desc_count)
2888 			sdkp->max_unmap_blocks = lba_count;
2889 
2890 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2891 
2892 		if (buffer[32] & 0x80)
2893 			sdkp->unmap_alignment =
2894 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2895 
2896 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2897 
2898 			if (sdkp->max_unmap_blocks)
2899 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2900 			else
2901 				sd_config_discard(sdkp, SD_LBP_WS16);
2902 
2903 		} else {	/* LBP VPD page tells us what to use */
2904 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2905 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2906 			else if (sdkp->lbpws)
2907 				sd_config_discard(sdkp, SD_LBP_WS16);
2908 			else if (sdkp->lbpws10)
2909 				sd_config_discard(sdkp, SD_LBP_WS10);
2910 			else
2911 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2912 		}
2913 	}
2914 
2915  out:
2916 	kfree(buffer);
2917 }
2918 
2919 /**
2920  * sd_read_block_characteristics - Query block dev. characteristics
2921  * @sdkp: disk to query
2922  */
2923 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2924 {
2925 	struct request_queue *q = sdkp->disk->queue;
2926 	unsigned char *buffer;
2927 	u16 rot;
2928 	const int vpd_len = 64;
2929 
2930 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2931 
2932 	if (!buffer ||
2933 	    /* Block Device Characteristics VPD */
2934 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2935 		goto out;
2936 
2937 	rot = get_unaligned_be16(&buffer[4]);
2938 
2939 	if (rot == 1) {
2940 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2941 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2942 	}
2943 
2944 	if (sdkp->device->type == TYPE_ZBC) {
2945 		/* Host-managed */
2946 		q->limits.zoned = BLK_ZONED_HM;
2947 	} else {
2948 		sdkp->zoned = (buffer[8] >> 4) & 3;
2949 		if (sdkp->zoned == 1 && !disk_has_partitions(sdkp->disk)) {
2950 			/* Host-aware */
2951 			q->limits.zoned = BLK_ZONED_HA;
2952 		} else {
2953 			/*
2954 			 * Treat drive-managed devices and host-aware devices
2955 			 * with partitions as regular block devices.
2956 			 */
2957 			q->limits.zoned = BLK_ZONED_NONE;
2958 		}
2959 	}
2960 	if (blk_queue_is_zoned(q) && sdkp->first_scan)
2961 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2962 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2963 
2964  out:
2965 	kfree(buffer);
2966 }
2967 
2968 /**
2969  * sd_read_block_provisioning - Query provisioning VPD page
2970  * @sdkp: disk to query
2971  */
2972 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2973 {
2974 	unsigned char *buffer;
2975 	const int vpd_len = 8;
2976 
2977 	if (sdkp->lbpme == 0)
2978 		return;
2979 
2980 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2981 
2982 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2983 		goto out;
2984 
2985 	sdkp->lbpvpd	= 1;
2986 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2987 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2988 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2989 
2990  out:
2991 	kfree(buffer);
2992 }
2993 
2994 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2995 {
2996 	struct scsi_device *sdev = sdkp->device;
2997 
2998 	if (sdev->host->no_write_same) {
2999 		sdev->no_write_same = 1;
3000 
3001 		return;
3002 	}
3003 
3004 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3005 		/* too large values might cause issues with arcmsr */
3006 		int vpd_buf_len = 64;
3007 
3008 		sdev->no_report_opcodes = 1;
3009 
3010 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3011 		 * CODES is unsupported and the device has an ATA
3012 		 * Information VPD page (SAT).
3013 		 */
3014 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3015 			sdev->no_write_same = 1;
3016 	}
3017 
3018 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3019 		sdkp->ws16 = 1;
3020 
3021 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3022 		sdkp->ws10 = 1;
3023 }
3024 
3025 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3026 {
3027 	struct scsi_device *sdev = sdkp->device;
3028 
3029 	if (!sdev->security_supported)
3030 		return;
3031 
3032 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3033 			SECURITY_PROTOCOL_IN) == 1 &&
3034 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3035 			SECURITY_PROTOCOL_OUT) == 1)
3036 		sdkp->security = 1;
3037 }
3038 
3039 /*
3040  * Determine the device's preferred I/O size for reads and writes
3041  * unless the reported value is unreasonably small, large, not a
3042  * multiple of the physical block size, or simply garbage.
3043  */
3044 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3045 				      unsigned int dev_max)
3046 {
3047 	struct scsi_device *sdp = sdkp->device;
3048 	unsigned int opt_xfer_bytes =
3049 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3050 
3051 	if (sdkp->opt_xfer_blocks == 0)
3052 		return false;
3053 
3054 	if (sdkp->opt_xfer_blocks > dev_max) {
3055 		sd_first_printk(KERN_WARNING, sdkp,
3056 				"Optimal transfer size %u logical blocks " \
3057 				"> dev_max (%u logical blocks)\n",
3058 				sdkp->opt_xfer_blocks, dev_max);
3059 		return false;
3060 	}
3061 
3062 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3063 		sd_first_printk(KERN_WARNING, sdkp,
3064 				"Optimal transfer size %u logical blocks " \
3065 				"> sd driver limit (%u logical blocks)\n",
3066 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3067 		return false;
3068 	}
3069 
3070 	if (opt_xfer_bytes < PAGE_SIZE) {
3071 		sd_first_printk(KERN_WARNING, sdkp,
3072 				"Optimal transfer size %u bytes < " \
3073 				"PAGE_SIZE (%u bytes)\n",
3074 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3075 		return false;
3076 	}
3077 
3078 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3079 		sd_first_printk(KERN_WARNING, sdkp,
3080 				"Optimal transfer size %u bytes not a " \
3081 				"multiple of physical block size (%u bytes)\n",
3082 				opt_xfer_bytes, sdkp->physical_block_size);
3083 		return false;
3084 	}
3085 
3086 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3087 			opt_xfer_bytes);
3088 	return true;
3089 }
3090 
3091 /**
3092  *	sd_revalidate_disk - called the first time a new disk is seen,
3093  *	performs disk spin up, read_capacity, etc.
3094  *	@disk: struct gendisk we care about
3095  **/
3096 static int sd_revalidate_disk(struct gendisk *disk)
3097 {
3098 	struct scsi_disk *sdkp = scsi_disk(disk);
3099 	struct scsi_device *sdp = sdkp->device;
3100 	struct request_queue *q = sdkp->disk->queue;
3101 	sector_t old_capacity = sdkp->capacity;
3102 	unsigned char *buffer;
3103 	unsigned int dev_max, rw_max;
3104 
3105 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3106 				      "sd_revalidate_disk\n"));
3107 
3108 	/*
3109 	 * If the device is offline, don't try and read capacity or any
3110 	 * of the other niceties.
3111 	 */
3112 	if (!scsi_device_online(sdp))
3113 		goto out;
3114 
3115 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3116 	if (!buffer) {
3117 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3118 			  "allocation failure.\n");
3119 		goto out;
3120 	}
3121 
3122 	sd_spinup_disk(sdkp);
3123 
3124 	/*
3125 	 * Without media there is no reason to ask; moreover, some devices
3126 	 * react badly if we do.
3127 	 */
3128 	if (sdkp->media_present) {
3129 		sd_read_capacity(sdkp, buffer);
3130 
3131 		/*
3132 		 * set the default to rotational.  All non-rotational devices
3133 		 * support the block characteristics VPD page, which will
3134 		 * cause this to be updated correctly and any device which
3135 		 * doesn't support it should be treated as rotational.
3136 		 */
3137 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3138 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3139 
3140 		if (scsi_device_supports_vpd(sdp)) {
3141 			sd_read_block_provisioning(sdkp);
3142 			sd_read_block_limits(sdkp);
3143 			sd_read_block_characteristics(sdkp);
3144 			sd_zbc_read_zones(sdkp, buffer);
3145 		}
3146 
3147 		sd_print_capacity(sdkp, old_capacity);
3148 
3149 		sd_read_write_protect_flag(sdkp, buffer);
3150 		sd_read_cache_type(sdkp, buffer);
3151 		sd_read_app_tag_own(sdkp, buffer);
3152 		sd_read_write_same(sdkp, buffer);
3153 		sd_read_security(sdkp, buffer);
3154 	}
3155 
3156 	/*
3157 	 * We now have all cache related info, determine how we deal
3158 	 * with flush requests.
3159 	 */
3160 	sd_set_flush_flag(sdkp);
3161 
3162 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3163 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3164 
3165 	/* Some devices report a maximum block count for READ/WRITE requests. */
3166 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3167 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3168 
3169 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3170 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3171 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3172 	} else
3173 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3174 				      (sector_t)BLK_DEF_MAX_SECTORS);
3175 
3176 	/* Do not exceed controller limit */
3177 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3178 
3179 	/*
3180 	 * Only update max_sectors if previously unset or if the current value
3181 	 * exceeds the capabilities of the hardware.
3182 	 */
3183 	if (sdkp->first_scan ||
3184 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3185 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3186 		q->limits.max_sectors = rw_max;
3187 
3188 	sdkp->first_scan = 0;
3189 
3190 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3191 	sd_config_write_same(sdkp);
3192 	kfree(buffer);
3193 
3194  out:
3195 	return 0;
3196 }
3197 
3198 /**
3199  *	sd_unlock_native_capacity - unlock native capacity
3200  *	@disk: struct gendisk to set capacity for
3201  *
3202  *	Block layer calls this function if it detects that partitions
3203  *	on @disk reach beyond the end of the device.  If the SCSI host
3204  *	implements ->unlock_native_capacity() method, it's invoked to
3205  *	give it a chance to adjust the device capacity.
3206  *
3207  *	CONTEXT:
3208  *	Defined by block layer.  Might sleep.
3209  */
3210 static void sd_unlock_native_capacity(struct gendisk *disk)
3211 {
3212 	struct scsi_device *sdev = scsi_disk(disk)->device;
3213 
3214 	if (sdev->host->hostt->unlock_native_capacity)
3215 		sdev->host->hostt->unlock_native_capacity(sdev);
3216 }
3217 
3218 /**
3219  *	sd_format_disk_name - format disk name
3220  *	@prefix: name prefix - ie. "sd" for SCSI disks
3221  *	@index: index of the disk to format name for
3222  *	@buf: output buffer
3223  *	@buflen: length of the output buffer
3224  *
3225  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3226  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3227  *	which is followed by sdaaa.
3228  *
3229  *	This is basically 26 base counting with one extra 'nil' entry
3230  *	at the beginning from the second digit on and can be
3231  *	determined using similar method as 26 base conversion with the
3232  *	index shifted -1 after each digit is computed.
3233  *
3234  *	CONTEXT:
3235  *	Don't care.
3236  *
3237  *	RETURNS:
3238  *	0 on success, -errno on failure.
3239  */
3240 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3241 {
3242 	const int base = 'z' - 'a' + 1;
3243 	char *begin = buf + strlen(prefix);
3244 	char *end = buf + buflen;
3245 	char *p;
3246 	int unit;
3247 
3248 	p = end - 1;
3249 	*p = '\0';
3250 	unit = base;
3251 	do {
3252 		if (p == begin)
3253 			return -EINVAL;
3254 		*--p = 'a' + (index % unit);
3255 		index = (index / unit) - 1;
3256 	} while (index >= 0);
3257 
3258 	memmove(begin, p, end - p);
3259 	memcpy(buf, prefix, strlen(prefix));
3260 
3261 	return 0;
3262 }
3263 
3264 /**
3265  *	sd_probe - called during driver initialization and whenever a
3266  *	new scsi device is attached to the system. It is called once
3267  *	for each scsi device (not just disks) present.
3268  *	@dev: pointer to device object
3269  *
3270  *	Returns 0 if successful (or not interested in this scsi device
3271  *	(e.g. scanner)); 1 when there is an error.
3272  *
3273  *	Note: this function is invoked from the scsi mid-level.
3274  *	This function sets up the mapping between a given
3275  *	<host,channel,id,lun> (found in sdp) and new device name
3276  *	(e.g. /dev/sda). More precisely it is the block device major
3277  *	and minor number that is chosen here.
3278  *
3279  *	Assume sd_probe is not re-entrant (for time being)
3280  *	Also think about sd_probe() and sd_remove() running coincidentally.
3281  **/
3282 static int sd_probe(struct device *dev)
3283 {
3284 	struct scsi_device *sdp = to_scsi_device(dev);
3285 	struct scsi_disk *sdkp;
3286 	struct gendisk *gd;
3287 	int index;
3288 	int error;
3289 
3290 	scsi_autopm_get_device(sdp);
3291 	error = -ENODEV;
3292 	if (sdp->type != TYPE_DISK &&
3293 	    sdp->type != TYPE_ZBC &&
3294 	    sdp->type != TYPE_MOD &&
3295 	    sdp->type != TYPE_RBC)
3296 		goto out;
3297 
3298 #ifndef CONFIG_BLK_DEV_ZONED
3299 	if (sdp->type == TYPE_ZBC)
3300 		goto out;
3301 #endif
3302 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3303 					"sd_probe\n"));
3304 
3305 	error = -ENOMEM;
3306 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3307 	if (!sdkp)
3308 		goto out;
3309 
3310 	gd = alloc_disk(SD_MINORS);
3311 	if (!gd)
3312 		goto out_free;
3313 
3314 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3315 	if (index < 0) {
3316 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3317 		goto out_put;
3318 	}
3319 
3320 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3321 	if (error) {
3322 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3323 		goto out_free_index;
3324 	}
3325 
3326 	sdkp->device = sdp;
3327 	sdkp->driver = &sd_template;
3328 	sdkp->disk = gd;
3329 	sdkp->index = index;
3330 	atomic_set(&sdkp->openers, 0);
3331 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3332 
3333 	if (!sdp->request_queue->rq_timeout) {
3334 		if (sdp->type != TYPE_MOD)
3335 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3336 		else
3337 			blk_queue_rq_timeout(sdp->request_queue,
3338 					     SD_MOD_TIMEOUT);
3339 	}
3340 
3341 	device_initialize(&sdkp->dev);
3342 	sdkp->dev.parent = dev;
3343 	sdkp->dev.class = &sd_disk_class;
3344 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3345 
3346 	error = device_add(&sdkp->dev);
3347 	if (error)
3348 		goto out_free_index;
3349 
3350 	get_device(dev);
3351 	dev_set_drvdata(dev, sdkp);
3352 
3353 	gd->major = sd_major((index & 0xf0) >> 4);
3354 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3355 
3356 	gd->fops = &sd_fops;
3357 	gd->private_data = &sdkp->driver;
3358 	gd->queue = sdkp->device->request_queue;
3359 
3360 	/* defaults, until the device tells us otherwise */
3361 	sdp->sector_size = 512;
3362 	sdkp->capacity = 0;
3363 	sdkp->media_present = 1;
3364 	sdkp->write_prot = 0;
3365 	sdkp->cache_override = 0;
3366 	sdkp->WCE = 0;
3367 	sdkp->RCD = 0;
3368 	sdkp->ATO = 0;
3369 	sdkp->first_scan = 1;
3370 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3371 
3372 	sd_revalidate_disk(gd);
3373 
3374 	gd->flags = GENHD_FL_EXT_DEVT;
3375 	if (sdp->removable) {
3376 		gd->flags |= GENHD_FL_REMOVABLE;
3377 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3378 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3379 	}
3380 
3381 	blk_pm_runtime_init(sdp->request_queue, dev);
3382 	if (sdp->rpm_autosuspend) {
3383 		pm_runtime_set_autosuspend_delay(dev,
3384 			sdp->host->hostt->rpm_autosuspend_delay);
3385 	}
3386 	device_add_disk(dev, gd, NULL);
3387 	if (sdkp->capacity)
3388 		sd_dif_config_host(sdkp);
3389 
3390 	sd_revalidate_disk(gd);
3391 
3392 	if (sdkp->security) {
3393 		sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3394 		if (sdkp->opal_dev)
3395 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3396 	}
3397 
3398 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3399 		  sdp->removable ? "removable " : "");
3400 	scsi_autopm_put_device(sdp);
3401 
3402 	return 0;
3403 
3404  out_free_index:
3405 	ida_free(&sd_index_ida, index);
3406  out_put:
3407 	put_disk(gd);
3408  out_free:
3409 	kfree(sdkp);
3410  out:
3411 	scsi_autopm_put_device(sdp);
3412 	return error;
3413 }
3414 
3415 /**
3416  *	sd_remove - called whenever a scsi disk (previously recognized by
3417  *	sd_probe) is detached from the system. It is called (potentially
3418  *	multiple times) during sd module unload.
3419  *	@dev: pointer to device object
3420  *
3421  *	Note: this function is invoked from the scsi mid-level.
3422  *	This function potentially frees up a device name (e.g. /dev/sdc)
3423  *	that could be re-used by a subsequent sd_probe().
3424  *	This function is not called when the built-in sd driver is "exit-ed".
3425  **/
3426 static int sd_remove(struct device *dev)
3427 {
3428 	struct scsi_disk *sdkp;
3429 	dev_t devt;
3430 
3431 	sdkp = dev_get_drvdata(dev);
3432 	devt = disk_devt(sdkp->disk);
3433 	scsi_autopm_get_device(sdkp->device);
3434 
3435 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3436 	device_del(&sdkp->dev);
3437 	del_gendisk(sdkp->disk);
3438 	sd_shutdown(dev);
3439 
3440 	free_opal_dev(sdkp->opal_dev);
3441 
3442 	blk_register_region(devt, SD_MINORS, NULL,
3443 			    sd_default_probe, NULL, NULL);
3444 
3445 	mutex_lock(&sd_ref_mutex);
3446 	dev_set_drvdata(dev, NULL);
3447 	put_device(&sdkp->dev);
3448 	mutex_unlock(&sd_ref_mutex);
3449 
3450 	return 0;
3451 }
3452 
3453 /**
3454  *	scsi_disk_release - Called to free the scsi_disk structure
3455  *	@dev: pointer to embedded class device
3456  *
3457  *	sd_ref_mutex must be held entering this routine.  Because it is
3458  *	called on last put, you should always use the scsi_disk_get()
3459  *	scsi_disk_put() helpers which manipulate the semaphore directly
3460  *	and never do a direct put_device.
3461  **/
3462 static void scsi_disk_release(struct device *dev)
3463 {
3464 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3465 	struct gendisk *disk = sdkp->disk;
3466 	struct request_queue *q = disk->queue;
3467 
3468 	ida_free(&sd_index_ida, sdkp->index);
3469 
3470 	/*
3471 	 * Wait until all requests that are in progress have completed.
3472 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3473 	 * due to clearing the disk->private_data pointer. Wait from inside
3474 	 * scsi_disk_release() instead of from sd_release() to avoid that
3475 	 * freezing and unfreezing the request queue affects user space I/O
3476 	 * in case multiple processes open a /dev/sd... node concurrently.
3477 	 */
3478 	blk_mq_freeze_queue(q);
3479 	blk_mq_unfreeze_queue(q);
3480 
3481 	disk->private_data = NULL;
3482 	put_disk(disk);
3483 	put_device(&sdkp->device->sdev_gendev);
3484 
3485 	kfree(sdkp);
3486 }
3487 
3488 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3489 {
3490 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3491 	struct scsi_sense_hdr sshdr;
3492 	struct scsi_device *sdp = sdkp->device;
3493 	int res;
3494 
3495 	if (start)
3496 		cmd[4] |= 1;	/* START */
3497 
3498 	if (sdp->start_stop_pwr_cond)
3499 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3500 
3501 	if (!scsi_device_online(sdp))
3502 		return -ENODEV;
3503 
3504 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3505 			SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3506 	if (res) {
3507 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3508 		if (driver_byte(res) == DRIVER_SENSE)
3509 			sd_print_sense_hdr(sdkp, &sshdr);
3510 		if (scsi_sense_valid(&sshdr) &&
3511 			/* 0x3a is medium not present */
3512 			sshdr.asc == 0x3a)
3513 			res = 0;
3514 	}
3515 
3516 	/* SCSI error codes must not go to the generic layer */
3517 	if (res)
3518 		return -EIO;
3519 
3520 	return 0;
3521 }
3522 
3523 /*
3524  * Send a SYNCHRONIZE CACHE instruction down to the device through
3525  * the normal SCSI command structure.  Wait for the command to
3526  * complete.
3527  */
3528 static void sd_shutdown(struct device *dev)
3529 {
3530 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3531 
3532 	if (!sdkp)
3533 		return;         /* this can happen */
3534 
3535 	if (pm_runtime_suspended(dev))
3536 		return;
3537 
3538 	if (sdkp->WCE && sdkp->media_present) {
3539 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3540 		sd_sync_cache(sdkp, NULL);
3541 	}
3542 
3543 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3544 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3545 		sd_start_stop_device(sdkp, 0);
3546 	}
3547 }
3548 
3549 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3550 {
3551 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3552 	struct scsi_sense_hdr sshdr;
3553 	int ret = 0;
3554 
3555 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3556 		return 0;
3557 
3558 	if (sdkp->WCE && sdkp->media_present) {
3559 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3560 		ret = sd_sync_cache(sdkp, &sshdr);
3561 
3562 		if (ret) {
3563 			/* ignore OFFLINE device */
3564 			if (ret == -ENODEV)
3565 				return 0;
3566 
3567 			if (!scsi_sense_valid(&sshdr) ||
3568 			    sshdr.sense_key != ILLEGAL_REQUEST)
3569 				return ret;
3570 
3571 			/*
3572 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3573 			 * doesn't support sync. There's not much to do and
3574 			 * suspend shouldn't fail.
3575 			 */
3576 			ret = 0;
3577 		}
3578 	}
3579 
3580 	if (sdkp->device->manage_start_stop) {
3581 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3582 		/* an error is not worth aborting a system sleep */
3583 		ret = sd_start_stop_device(sdkp, 0);
3584 		if (ignore_stop_errors)
3585 			ret = 0;
3586 	}
3587 
3588 	return ret;
3589 }
3590 
3591 static int sd_suspend_system(struct device *dev)
3592 {
3593 	return sd_suspend_common(dev, true);
3594 }
3595 
3596 static int sd_suspend_runtime(struct device *dev)
3597 {
3598 	return sd_suspend_common(dev, false);
3599 }
3600 
3601 static int sd_resume(struct device *dev)
3602 {
3603 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3604 	int ret;
3605 
3606 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3607 		return 0;
3608 
3609 	if (!sdkp->device->manage_start_stop)
3610 		return 0;
3611 
3612 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3613 	ret = sd_start_stop_device(sdkp, 1);
3614 	if (!ret)
3615 		opal_unlock_from_suspend(sdkp->opal_dev);
3616 	return ret;
3617 }
3618 
3619 /**
3620  *	init_sd - entry point for this driver (both when built in or when
3621  *	a module).
3622  *
3623  *	Note: this function registers this driver with the scsi mid-level.
3624  **/
3625 static int __init init_sd(void)
3626 {
3627 	int majors = 0, i, err;
3628 
3629 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3630 
3631 	for (i = 0; i < SD_MAJORS; i++) {
3632 		if (register_blkdev(sd_major(i), "sd") != 0)
3633 			continue;
3634 		majors++;
3635 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3636 				    sd_default_probe, NULL, NULL);
3637 	}
3638 
3639 	if (!majors)
3640 		return -ENODEV;
3641 
3642 	err = class_register(&sd_disk_class);
3643 	if (err)
3644 		goto err_out;
3645 
3646 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3647 					 0, 0, NULL);
3648 	if (!sd_cdb_cache) {
3649 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3650 		err = -ENOMEM;
3651 		goto err_out_class;
3652 	}
3653 
3654 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3655 	if (!sd_cdb_pool) {
3656 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3657 		err = -ENOMEM;
3658 		goto err_out_cache;
3659 	}
3660 
3661 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3662 	if (!sd_page_pool) {
3663 		printk(KERN_ERR "sd: can't init discard page pool\n");
3664 		err = -ENOMEM;
3665 		goto err_out_ppool;
3666 	}
3667 
3668 	err = scsi_register_driver(&sd_template.gendrv);
3669 	if (err)
3670 		goto err_out_driver;
3671 
3672 	return 0;
3673 
3674 err_out_driver:
3675 	mempool_destroy(sd_page_pool);
3676 
3677 err_out_ppool:
3678 	mempool_destroy(sd_cdb_pool);
3679 
3680 err_out_cache:
3681 	kmem_cache_destroy(sd_cdb_cache);
3682 
3683 err_out_class:
3684 	class_unregister(&sd_disk_class);
3685 err_out:
3686 	for (i = 0; i < SD_MAJORS; i++)
3687 		unregister_blkdev(sd_major(i), "sd");
3688 	return err;
3689 }
3690 
3691 /**
3692  *	exit_sd - exit point for this driver (when it is a module).
3693  *
3694  *	Note: this function unregisters this driver from the scsi mid-level.
3695  **/
3696 static void __exit exit_sd(void)
3697 {
3698 	int i;
3699 
3700 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3701 
3702 	scsi_unregister_driver(&sd_template.gendrv);
3703 	mempool_destroy(sd_cdb_pool);
3704 	mempool_destroy(sd_page_pool);
3705 	kmem_cache_destroy(sd_cdb_cache);
3706 
3707 	class_unregister(&sd_disk_class);
3708 
3709 	for (i = 0; i < SD_MAJORS; i++) {
3710 		blk_unregister_region(sd_major(i), SD_MINORS);
3711 		unregister_blkdev(sd_major(i), "sd");
3712 	}
3713 }
3714 
3715 module_init(init_sd);
3716 module_exit(exit_sd);
3717 
3718 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3719 {
3720 	scsi_print_sense_hdr(sdkp->device,
3721 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3722 }
3723 
3724 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3725 {
3726 	const char *hb_string = scsi_hostbyte_string(result);
3727 	const char *db_string = scsi_driverbyte_string(result);
3728 
3729 	if (hb_string || db_string)
3730 		sd_printk(KERN_INFO, sdkp,
3731 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3732 			  hb_string ? hb_string : "invalid",
3733 			  db_string ? db_string : "invalid");
3734 	else
3735 		sd_printk(KERN_INFO, sdkp,
3736 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3737 			  msg, host_byte(result), driver_byte(result));
3738 }
3739