xref: /openbmc/linux/drivers/scsi/sd.c (revision b593bce5)
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 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
126 static void sd_print_result(const struct scsi_disk *, const char *, int);
127 
128 static DEFINE_IDA(sd_index_ida);
129 
130 /* This semaphore is used to mediate the 0->1 reference get in the
131  * face of object destruction (i.e. we can't allow a get on an
132  * object after last put) */
133 static DEFINE_MUTEX(sd_ref_mutex);
134 
135 static struct kmem_cache *sd_cdb_cache;
136 static mempool_t *sd_cdb_pool;
137 static mempool_t *sd_page_pool;
138 
139 static const char *sd_cache_types[] = {
140 	"write through", "none", "write back",
141 	"write back, no read (daft)"
142 };
143 
144 static void sd_set_flush_flag(struct scsi_disk *sdkp)
145 {
146 	bool wc = false, fua = false;
147 
148 	if (sdkp->WCE) {
149 		wc = true;
150 		if (sdkp->DPOFUA)
151 			fua = true;
152 	}
153 
154 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
155 }
156 
157 static ssize_t
158 cache_type_store(struct device *dev, struct device_attribute *attr,
159 		 const char *buf, size_t count)
160 {
161 	int ct, rcd, wce, sp;
162 	struct scsi_disk *sdkp = to_scsi_disk(dev);
163 	struct scsi_device *sdp = sdkp->device;
164 	char buffer[64];
165 	char *buffer_data;
166 	struct scsi_mode_data data;
167 	struct scsi_sense_hdr sshdr;
168 	static const char temp[] = "temporary ";
169 	int len;
170 
171 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
172 		/* no cache control on RBC devices; theoretically they
173 		 * can do it, but there's probably so many exceptions
174 		 * it's not worth the risk */
175 		return -EINVAL;
176 
177 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
178 		buf += sizeof(temp) - 1;
179 		sdkp->cache_override = 1;
180 	} else {
181 		sdkp->cache_override = 0;
182 	}
183 
184 	ct = sysfs_match_string(sd_cache_types, buf);
185 	if (ct < 0)
186 		return -EINVAL;
187 
188 	rcd = ct & 0x01 ? 1 : 0;
189 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 
191 	if (sdkp->cache_override) {
192 		sdkp->WCE = wce;
193 		sdkp->RCD = rcd;
194 		sd_set_flush_flag(sdkp);
195 		return count;
196 	}
197 
198 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 			    SD_MAX_RETRIES, &data, NULL))
200 		return -EINVAL;
201 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 		  data.block_descriptor_length);
203 	buffer_data = buffer + data.header_length +
204 		data.block_descriptor_length;
205 	buffer_data[2] &= ~0x05;
206 	buffer_data[2] |= wce << 2 | rcd;
207 	sp = buffer_data[0] & 0x80 ? 1 : 0;
208 	buffer_data[0] &= ~0x80;
209 
210 	/*
211 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
212 	 * received mode parameter buffer before doing MODE SELECT.
213 	 */
214 	data.device_specific = 0;
215 
216 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
217 			     SD_MAX_RETRIES, &data, &sshdr)) {
218 		if (scsi_sense_valid(&sshdr))
219 			sd_print_sense_hdr(sdkp, &sshdr);
220 		return -EINVAL;
221 	}
222 	revalidate_disk(sdkp->disk);
223 	return count;
224 }
225 
226 static ssize_t
227 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
228 		       char *buf)
229 {
230 	struct scsi_disk *sdkp = to_scsi_disk(dev);
231 	struct scsi_device *sdp = sdkp->device;
232 
233 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
234 }
235 
236 static ssize_t
237 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
238 			const char *buf, size_t count)
239 {
240 	struct scsi_disk *sdkp = to_scsi_disk(dev);
241 	struct scsi_device *sdp = sdkp->device;
242 	bool v;
243 
244 	if (!capable(CAP_SYS_ADMIN))
245 		return -EACCES;
246 
247 	if (kstrtobool(buf, &v))
248 		return -EINVAL;
249 
250 	sdp->manage_start_stop = v;
251 
252 	return count;
253 }
254 static DEVICE_ATTR_RW(manage_start_stop);
255 
256 static ssize_t
257 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
258 {
259 	struct scsi_disk *sdkp = to_scsi_disk(dev);
260 
261 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
262 }
263 
264 static ssize_t
265 allow_restart_store(struct device *dev, struct device_attribute *attr,
266 		    const char *buf, size_t count)
267 {
268 	bool v;
269 	struct scsi_disk *sdkp = to_scsi_disk(dev);
270 	struct scsi_device *sdp = sdkp->device;
271 
272 	if (!capable(CAP_SYS_ADMIN))
273 		return -EACCES;
274 
275 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
276 		return -EINVAL;
277 
278 	if (kstrtobool(buf, &v))
279 		return -EINVAL;
280 
281 	sdp->allow_restart = v;
282 
283 	return count;
284 }
285 static DEVICE_ATTR_RW(allow_restart);
286 
287 static ssize_t
288 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
289 {
290 	struct scsi_disk *sdkp = to_scsi_disk(dev);
291 	int ct = sdkp->RCD + 2*sdkp->WCE;
292 
293 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
294 }
295 static DEVICE_ATTR_RW(cache_type);
296 
297 static ssize_t
298 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
299 {
300 	struct scsi_disk *sdkp = to_scsi_disk(dev);
301 
302 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
303 }
304 static DEVICE_ATTR_RO(FUA);
305 
306 static ssize_t
307 protection_type_show(struct device *dev, struct device_attribute *attr,
308 		     char *buf)
309 {
310 	struct scsi_disk *sdkp = to_scsi_disk(dev);
311 
312 	return sprintf(buf, "%u\n", sdkp->protection_type);
313 }
314 
315 static ssize_t
316 protection_type_store(struct device *dev, struct device_attribute *attr,
317 		      const char *buf, size_t count)
318 {
319 	struct scsi_disk *sdkp = to_scsi_disk(dev);
320 	unsigned int val;
321 	int err;
322 
323 	if (!capable(CAP_SYS_ADMIN))
324 		return -EACCES;
325 
326 	err = kstrtouint(buf, 10, &val);
327 
328 	if (err)
329 		return err;
330 
331 	if (val <= T10_PI_TYPE3_PROTECTION)
332 		sdkp->protection_type = val;
333 
334 	return count;
335 }
336 static DEVICE_ATTR_RW(protection_type);
337 
338 static ssize_t
339 protection_mode_show(struct device *dev, struct device_attribute *attr,
340 		     char *buf)
341 {
342 	struct scsi_disk *sdkp = to_scsi_disk(dev);
343 	struct scsi_device *sdp = sdkp->device;
344 	unsigned int dif, dix;
345 
346 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
347 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
348 
349 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
350 		dif = 0;
351 		dix = 1;
352 	}
353 
354 	if (!dif && !dix)
355 		return sprintf(buf, "none\n");
356 
357 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
358 }
359 static DEVICE_ATTR_RO(protection_mode);
360 
361 static ssize_t
362 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
363 {
364 	struct scsi_disk *sdkp = to_scsi_disk(dev);
365 
366 	return sprintf(buf, "%u\n", sdkp->ATO);
367 }
368 static DEVICE_ATTR_RO(app_tag_own);
369 
370 static ssize_t
371 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
372 		       char *buf)
373 {
374 	struct scsi_disk *sdkp = to_scsi_disk(dev);
375 
376 	return sprintf(buf, "%u\n", sdkp->lbpme);
377 }
378 static DEVICE_ATTR_RO(thin_provisioning);
379 
380 /* sysfs_match_string() requires dense arrays */
381 static const char *lbp_mode[] = {
382 	[SD_LBP_FULL]		= "full",
383 	[SD_LBP_UNMAP]		= "unmap",
384 	[SD_LBP_WS16]		= "writesame_16",
385 	[SD_LBP_WS10]		= "writesame_10",
386 	[SD_LBP_ZERO]		= "writesame_zero",
387 	[SD_LBP_DISABLE]	= "disabled",
388 };
389 
390 static ssize_t
391 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
392 		       char *buf)
393 {
394 	struct scsi_disk *sdkp = to_scsi_disk(dev);
395 
396 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
397 }
398 
399 static ssize_t
400 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
401 			const char *buf, size_t count)
402 {
403 	struct scsi_disk *sdkp = to_scsi_disk(dev);
404 	struct scsi_device *sdp = sdkp->device;
405 	int mode;
406 
407 	if (!capable(CAP_SYS_ADMIN))
408 		return -EACCES;
409 
410 	if (sd_is_zoned(sdkp)) {
411 		sd_config_discard(sdkp, SD_LBP_DISABLE);
412 		return count;
413 	}
414 
415 	if (sdp->type != TYPE_DISK)
416 		return -EINVAL;
417 
418 	mode = sysfs_match_string(lbp_mode, buf);
419 	if (mode < 0)
420 		return -EINVAL;
421 
422 	sd_config_discard(sdkp, mode);
423 
424 	return count;
425 }
426 static DEVICE_ATTR_RW(provisioning_mode);
427 
428 /* sysfs_match_string() requires dense arrays */
429 static const char *zeroing_mode[] = {
430 	[SD_ZERO_WRITE]		= "write",
431 	[SD_ZERO_WS]		= "writesame",
432 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
433 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
434 };
435 
436 static ssize_t
437 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
438 		  char *buf)
439 {
440 	struct scsi_disk *sdkp = to_scsi_disk(dev);
441 
442 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
443 }
444 
445 static ssize_t
446 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
447 		   const char *buf, size_t count)
448 {
449 	struct scsi_disk *sdkp = to_scsi_disk(dev);
450 	int mode;
451 
452 	if (!capable(CAP_SYS_ADMIN))
453 		return -EACCES;
454 
455 	mode = sysfs_match_string(zeroing_mode, buf);
456 	if (mode < 0)
457 		return -EINVAL;
458 
459 	sdkp->zeroing_mode = mode;
460 
461 	return count;
462 }
463 static DEVICE_ATTR_RW(zeroing_mode);
464 
465 static ssize_t
466 max_medium_access_timeouts_show(struct device *dev,
467 				struct device_attribute *attr, char *buf)
468 {
469 	struct scsi_disk *sdkp = to_scsi_disk(dev);
470 
471 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
472 }
473 
474 static ssize_t
475 max_medium_access_timeouts_store(struct device *dev,
476 				 struct device_attribute *attr, const char *buf,
477 				 size_t count)
478 {
479 	struct scsi_disk *sdkp = to_scsi_disk(dev);
480 	int err;
481 
482 	if (!capable(CAP_SYS_ADMIN))
483 		return -EACCES;
484 
485 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
486 
487 	return err ? err : count;
488 }
489 static DEVICE_ATTR_RW(max_medium_access_timeouts);
490 
491 static ssize_t
492 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
493 			   char *buf)
494 {
495 	struct scsi_disk *sdkp = to_scsi_disk(dev);
496 
497 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
498 }
499 
500 static ssize_t
501 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
502 			    const char *buf, size_t count)
503 {
504 	struct scsi_disk *sdkp = to_scsi_disk(dev);
505 	struct scsi_device *sdp = sdkp->device;
506 	unsigned long max;
507 	int err;
508 
509 	if (!capable(CAP_SYS_ADMIN))
510 		return -EACCES;
511 
512 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
513 		return -EINVAL;
514 
515 	err = kstrtoul(buf, 10, &max);
516 
517 	if (err)
518 		return err;
519 
520 	if (max == 0)
521 		sdp->no_write_same = 1;
522 	else if (max <= SD_MAX_WS16_BLOCKS) {
523 		sdp->no_write_same = 0;
524 		sdkp->max_ws_blocks = max;
525 	}
526 
527 	sd_config_write_same(sdkp);
528 
529 	return count;
530 }
531 static DEVICE_ATTR_RW(max_write_same_blocks);
532 
533 static struct attribute *sd_disk_attrs[] = {
534 	&dev_attr_cache_type.attr,
535 	&dev_attr_FUA.attr,
536 	&dev_attr_allow_restart.attr,
537 	&dev_attr_manage_start_stop.attr,
538 	&dev_attr_protection_type.attr,
539 	&dev_attr_protection_mode.attr,
540 	&dev_attr_app_tag_own.attr,
541 	&dev_attr_thin_provisioning.attr,
542 	&dev_attr_provisioning_mode.attr,
543 	&dev_attr_zeroing_mode.attr,
544 	&dev_attr_max_write_same_blocks.attr,
545 	&dev_attr_max_medium_access_timeouts.attr,
546 	NULL,
547 };
548 ATTRIBUTE_GROUPS(sd_disk);
549 
550 static struct class sd_disk_class = {
551 	.name		= "scsi_disk",
552 	.owner		= THIS_MODULE,
553 	.dev_release	= scsi_disk_release,
554 	.dev_groups	= sd_disk_groups,
555 };
556 
557 static const struct dev_pm_ops sd_pm_ops = {
558 	.suspend		= sd_suspend_system,
559 	.resume			= sd_resume,
560 	.poweroff		= sd_suspend_system,
561 	.restore		= sd_resume,
562 	.runtime_suspend	= sd_suspend_runtime,
563 	.runtime_resume		= sd_resume,
564 };
565 
566 static struct scsi_driver sd_template = {
567 	.gendrv = {
568 		.name		= "sd",
569 		.owner		= THIS_MODULE,
570 		.probe		= sd_probe,
571 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
572 		.remove		= sd_remove,
573 		.shutdown	= sd_shutdown,
574 		.pm		= &sd_pm_ops,
575 	},
576 	.rescan			= sd_rescan,
577 	.init_command		= sd_init_command,
578 	.uninit_command		= sd_uninit_command,
579 	.done			= sd_done,
580 	.eh_action		= sd_eh_action,
581 	.eh_reset		= sd_eh_reset,
582 };
583 
584 /*
585  * Dummy kobj_map->probe function.
586  * The default ->probe function will call modprobe, which is
587  * pointless as this module is already loaded.
588  */
589 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
590 {
591 	return NULL;
592 }
593 
594 /*
595  * Device no to disk mapping:
596  *
597  *       major         disc2     disc  p1
598  *   |............|.............|....|....| <- dev_t
599  *    31        20 19          8 7  4 3  0
600  *
601  * Inside a major, we have 16k disks, however mapped non-
602  * contiguously. The first 16 disks are for major0, the next
603  * ones with major1, ... Disk 256 is for major0 again, disk 272
604  * for major1, ...
605  * As we stay compatible with our numbering scheme, we can reuse
606  * the well-know SCSI majors 8, 65--71, 136--143.
607  */
608 static int sd_major(int major_idx)
609 {
610 	switch (major_idx) {
611 	case 0:
612 		return SCSI_DISK0_MAJOR;
613 	case 1 ... 7:
614 		return SCSI_DISK1_MAJOR + major_idx - 1;
615 	case 8 ... 15:
616 		return SCSI_DISK8_MAJOR + major_idx - 8;
617 	default:
618 		BUG();
619 		return 0;	/* shut up gcc */
620 	}
621 }
622 
623 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
624 {
625 	struct scsi_disk *sdkp = NULL;
626 
627 	mutex_lock(&sd_ref_mutex);
628 
629 	if (disk->private_data) {
630 		sdkp = scsi_disk(disk);
631 		if (scsi_device_get(sdkp->device) == 0)
632 			get_device(&sdkp->dev);
633 		else
634 			sdkp = NULL;
635 	}
636 	mutex_unlock(&sd_ref_mutex);
637 	return sdkp;
638 }
639 
640 static void scsi_disk_put(struct scsi_disk *sdkp)
641 {
642 	struct scsi_device *sdev = sdkp->device;
643 
644 	mutex_lock(&sd_ref_mutex);
645 	put_device(&sdkp->dev);
646 	scsi_device_put(sdev);
647 	mutex_unlock(&sd_ref_mutex);
648 }
649 
650 #ifdef CONFIG_BLK_SED_OPAL
651 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
652 		size_t len, bool send)
653 {
654 	struct scsi_device *sdev = data;
655 	u8 cdb[12] = { 0, };
656 	int ret;
657 
658 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
659 	cdb[1] = secp;
660 	put_unaligned_be16(spsp, &cdb[2]);
661 	put_unaligned_be32(len, &cdb[6]);
662 
663 	ret = scsi_execute_req(sdev, cdb,
664 			send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
665 			buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
666 	return ret <= 0 ? ret : -EIO;
667 }
668 #endif /* CONFIG_BLK_SED_OPAL */
669 
670 /*
671  * Look up the DIX operation based on whether the command is read or
672  * write and whether dix and dif are enabled.
673  */
674 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
675 {
676 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
677 	static const unsigned int ops[] = {	/* wrt dix dif */
678 		SCSI_PROT_NORMAL,		/*  0	0   0  */
679 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
680 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
681 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
682 		SCSI_PROT_NORMAL,		/*  1	0   0  */
683 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
684 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
685 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
686 	};
687 
688 	return ops[write << 2 | dix << 1 | dif];
689 }
690 
691 /*
692  * Returns a mask of the protection flags that are valid for a given DIX
693  * operation.
694  */
695 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
696 {
697 	static const unsigned int flag_mask[] = {
698 		[SCSI_PROT_NORMAL]		= 0,
699 
700 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
701 						  SCSI_PROT_GUARD_CHECK |
702 						  SCSI_PROT_REF_CHECK |
703 						  SCSI_PROT_REF_INCREMENT,
704 
705 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
706 						  SCSI_PROT_IP_CHECKSUM,
707 
708 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
709 						  SCSI_PROT_GUARD_CHECK |
710 						  SCSI_PROT_REF_CHECK |
711 						  SCSI_PROT_REF_INCREMENT |
712 						  SCSI_PROT_IP_CHECKSUM,
713 
714 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
715 						  SCSI_PROT_REF_INCREMENT,
716 
717 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
718 						  SCSI_PROT_REF_CHECK |
719 						  SCSI_PROT_REF_INCREMENT |
720 						  SCSI_PROT_IP_CHECKSUM,
721 
722 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
723 						  SCSI_PROT_GUARD_CHECK |
724 						  SCSI_PROT_REF_CHECK |
725 						  SCSI_PROT_REF_INCREMENT |
726 						  SCSI_PROT_IP_CHECKSUM,
727 	};
728 
729 	return flag_mask[prot_op];
730 }
731 
732 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
733 					   unsigned int dix, unsigned int dif)
734 {
735 	struct bio *bio = scmd->request->bio;
736 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
737 	unsigned int protect = 0;
738 
739 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
740 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
741 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
742 
743 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
744 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
745 	}
746 
747 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
748 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
749 
750 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
751 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
752 	}
753 
754 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
755 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
756 
757 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
758 			protect = 3 << 5;	/* Disable target PI checking */
759 		else
760 			protect = 1 << 5;	/* Enable target PI checking */
761 	}
762 
763 	scsi_set_prot_op(scmd, prot_op);
764 	scsi_set_prot_type(scmd, dif);
765 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
766 
767 	return protect;
768 }
769 
770 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
771 {
772 	struct request_queue *q = sdkp->disk->queue;
773 	unsigned int logical_block_size = sdkp->device->sector_size;
774 	unsigned int max_blocks = 0;
775 
776 	q->limits.discard_alignment =
777 		sdkp->unmap_alignment * logical_block_size;
778 	q->limits.discard_granularity =
779 		max(sdkp->physical_block_size,
780 		    sdkp->unmap_granularity * logical_block_size);
781 	sdkp->provisioning_mode = mode;
782 
783 	switch (mode) {
784 
785 	case SD_LBP_FULL:
786 	case SD_LBP_DISABLE:
787 		blk_queue_max_discard_sectors(q, 0);
788 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
789 		return;
790 
791 	case SD_LBP_UNMAP:
792 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
793 					  (u32)SD_MAX_WS16_BLOCKS);
794 		break;
795 
796 	case SD_LBP_WS16:
797 		if (sdkp->device->unmap_limit_for_ws)
798 			max_blocks = sdkp->max_unmap_blocks;
799 		else
800 			max_blocks = sdkp->max_ws_blocks;
801 
802 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
803 		break;
804 
805 	case SD_LBP_WS10:
806 		if (sdkp->device->unmap_limit_for_ws)
807 			max_blocks = sdkp->max_unmap_blocks;
808 		else
809 			max_blocks = sdkp->max_ws_blocks;
810 
811 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
812 		break;
813 
814 	case SD_LBP_ZERO:
815 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
816 					  (u32)SD_MAX_WS10_BLOCKS);
817 		break;
818 	}
819 
820 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
821 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
822 }
823 
824 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
825 {
826 	struct scsi_device *sdp = cmd->device;
827 	struct request *rq = cmd->request;
828 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
829 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
830 	unsigned int data_len = 24;
831 	char *buf;
832 
833 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
834 	if (!rq->special_vec.bv_page)
835 		return BLK_STS_RESOURCE;
836 	clear_highpage(rq->special_vec.bv_page);
837 	rq->special_vec.bv_offset = 0;
838 	rq->special_vec.bv_len = data_len;
839 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
840 
841 	cmd->cmd_len = 10;
842 	cmd->cmnd[0] = UNMAP;
843 	cmd->cmnd[8] = 24;
844 
845 	buf = page_address(rq->special_vec.bv_page);
846 	put_unaligned_be16(6 + 16, &buf[0]);
847 	put_unaligned_be16(16, &buf[2]);
848 	put_unaligned_be64(lba, &buf[8]);
849 	put_unaligned_be32(nr_blocks, &buf[16]);
850 
851 	cmd->allowed = SD_MAX_RETRIES;
852 	cmd->transfersize = data_len;
853 	rq->timeout = SD_TIMEOUT;
854 
855 	return scsi_init_io(cmd);
856 }
857 
858 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
859 		bool unmap)
860 {
861 	struct scsi_device *sdp = cmd->device;
862 	struct request *rq = cmd->request;
863 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
864 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
865 	u32 data_len = sdp->sector_size;
866 
867 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
868 	if (!rq->special_vec.bv_page)
869 		return BLK_STS_RESOURCE;
870 	clear_highpage(rq->special_vec.bv_page);
871 	rq->special_vec.bv_offset = 0;
872 	rq->special_vec.bv_len = data_len;
873 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
874 
875 	cmd->cmd_len = 16;
876 	cmd->cmnd[0] = WRITE_SAME_16;
877 	if (unmap)
878 		cmd->cmnd[1] = 0x8; /* UNMAP */
879 	put_unaligned_be64(lba, &cmd->cmnd[2]);
880 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
881 
882 	cmd->allowed = SD_MAX_RETRIES;
883 	cmd->transfersize = data_len;
884 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
885 
886 	return scsi_init_io(cmd);
887 }
888 
889 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
890 		bool unmap)
891 {
892 	struct scsi_device *sdp = cmd->device;
893 	struct request *rq = cmd->request;
894 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
895 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
896 	u32 data_len = sdp->sector_size;
897 
898 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
899 	if (!rq->special_vec.bv_page)
900 		return BLK_STS_RESOURCE;
901 	clear_highpage(rq->special_vec.bv_page);
902 	rq->special_vec.bv_offset = 0;
903 	rq->special_vec.bv_len = data_len;
904 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905 
906 	cmd->cmd_len = 10;
907 	cmd->cmnd[0] = WRITE_SAME;
908 	if (unmap)
909 		cmd->cmnd[1] = 0x8; /* UNMAP */
910 	put_unaligned_be32(lba, &cmd->cmnd[2]);
911 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
912 
913 	cmd->allowed = SD_MAX_RETRIES;
914 	cmd->transfersize = data_len;
915 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
916 
917 	return scsi_init_io(cmd);
918 }
919 
920 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
921 {
922 	struct request *rq = cmd->request;
923 	struct scsi_device *sdp = cmd->device;
924 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
925 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
926 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
927 
928 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
929 		switch (sdkp->zeroing_mode) {
930 		case SD_ZERO_WS16_UNMAP:
931 			return sd_setup_write_same16_cmnd(cmd, true);
932 		case SD_ZERO_WS10_UNMAP:
933 			return sd_setup_write_same10_cmnd(cmd, true);
934 		}
935 	}
936 
937 	if (sdp->no_write_same)
938 		return BLK_STS_TARGET;
939 
940 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
941 		return sd_setup_write_same16_cmnd(cmd, false);
942 
943 	return sd_setup_write_same10_cmnd(cmd, false);
944 }
945 
946 static void sd_config_write_same(struct scsi_disk *sdkp)
947 {
948 	struct request_queue *q = sdkp->disk->queue;
949 	unsigned int logical_block_size = sdkp->device->sector_size;
950 
951 	if (sdkp->device->no_write_same) {
952 		sdkp->max_ws_blocks = 0;
953 		goto out;
954 	}
955 
956 	/* Some devices can not handle block counts above 0xffff despite
957 	 * supporting WRITE SAME(16). Consequently we default to 64k
958 	 * blocks per I/O unless the device explicitly advertises a
959 	 * bigger limit.
960 	 */
961 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
962 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
963 						   (u32)SD_MAX_WS16_BLOCKS);
964 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
965 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
966 						   (u32)SD_MAX_WS10_BLOCKS);
967 	else {
968 		sdkp->device->no_write_same = 1;
969 		sdkp->max_ws_blocks = 0;
970 	}
971 
972 	if (sdkp->lbprz && sdkp->lbpws)
973 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
974 	else if (sdkp->lbprz && sdkp->lbpws10)
975 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
976 	else if (sdkp->max_ws_blocks)
977 		sdkp->zeroing_mode = SD_ZERO_WS;
978 	else
979 		sdkp->zeroing_mode = SD_ZERO_WRITE;
980 
981 	if (sdkp->max_ws_blocks &&
982 	    sdkp->physical_block_size > logical_block_size) {
983 		/*
984 		 * Reporting a maximum number of blocks that is not aligned
985 		 * on the device physical size would cause a large write same
986 		 * request to be split into physically unaligned chunks by
987 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
988 		 * even if the caller of these functions took care to align the
989 		 * large request. So make sure the maximum reported is aligned
990 		 * to the device physical block size. This is only an optional
991 		 * optimization for regular disks, but this is mandatory to
992 		 * avoid failure of large write same requests directed at
993 		 * sequential write required zones of host-managed ZBC disks.
994 		 */
995 		sdkp->max_ws_blocks =
996 			round_down(sdkp->max_ws_blocks,
997 				   bytes_to_logical(sdkp->device,
998 						    sdkp->physical_block_size));
999 	}
1000 
1001 out:
1002 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1003 					 (logical_block_size >> 9));
1004 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1005 					 (logical_block_size >> 9));
1006 }
1007 
1008 /**
1009  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1010  * @cmd: command to prepare
1011  *
1012  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1013  * the preference indicated by the target device.
1014  **/
1015 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1016 {
1017 	struct request *rq = cmd->request;
1018 	struct scsi_device *sdp = cmd->device;
1019 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1020 	struct bio *bio = rq->bio;
1021 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1022 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023 	blk_status_t ret;
1024 
1025 	if (sdkp->device->no_write_same)
1026 		return BLK_STS_TARGET;
1027 
1028 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1029 
1030 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1031 
1032 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1033 		cmd->cmd_len = 16;
1034 		cmd->cmnd[0] = WRITE_SAME_16;
1035 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1036 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037 	} else {
1038 		cmd->cmd_len = 10;
1039 		cmd->cmnd[0] = WRITE_SAME;
1040 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1041 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042 	}
1043 
1044 	cmd->transfersize = sdp->sector_size;
1045 	cmd->allowed = SD_MAX_RETRIES;
1046 
1047 	/*
1048 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1049 	 * different from the amount of data actually written to the target.
1050 	 *
1051 	 * We set up __data_len to the amount of data transferred via the
1052 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1053 	 * to transfer a single sector of data first, but then reset it to
1054 	 * the amount of data to be written right after so that the I/O path
1055 	 * knows how much to actually write.
1056 	 */
1057 	rq->__data_len = sdp->sector_size;
1058 	ret = scsi_init_io(cmd);
1059 	rq->__data_len = blk_rq_bytes(rq);
1060 
1061 	return ret;
1062 }
1063 
1064 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1065 {
1066 	struct request *rq = cmd->request;
1067 
1068 	/* flush requests don't perform I/O, zero the S/G table */
1069 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070 
1071 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1072 	cmd->cmd_len = 10;
1073 	cmd->transfersize = 0;
1074 	cmd->allowed = SD_MAX_RETRIES;
1075 
1076 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1077 	return BLK_STS_OK;
1078 }
1079 
1080 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1081 				       sector_t lba, unsigned int nr_blocks,
1082 				       unsigned char flags)
1083 {
1084 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1085 	if (unlikely(cmd->cmnd == NULL))
1086 		return BLK_STS_RESOURCE;
1087 
1088 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1089 	memset(cmd->cmnd, 0, cmd->cmd_len);
1090 
1091 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1092 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1093 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1094 	cmd->cmnd[10] = flags;
1095 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1096 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1097 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1098 
1099 	return BLK_STS_OK;
1100 }
1101 
1102 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1103 				       sector_t lba, unsigned int nr_blocks,
1104 				       unsigned char flags)
1105 {
1106 	cmd->cmd_len  = 16;
1107 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1108 	cmd->cmnd[1]  = flags;
1109 	cmd->cmnd[14] = 0;
1110 	cmd->cmnd[15] = 0;
1111 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1112 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1113 
1114 	return BLK_STS_OK;
1115 }
1116 
1117 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1118 				       sector_t lba, unsigned int nr_blocks,
1119 				       unsigned char flags)
1120 {
1121 	cmd->cmd_len = 10;
1122 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1123 	cmd->cmnd[1] = flags;
1124 	cmd->cmnd[6] = 0;
1125 	cmd->cmnd[9] = 0;
1126 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1127 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1128 
1129 	return BLK_STS_OK;
1130 }
1131 
1132 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1133 				      sector_t lba, unsigned int nr_blocks,
1134 				      unsigned char flags)
1135 {
1136 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1137 	if (WARN_ON_ONCE(nr_blocks == 0))
1138 		return BLK_STS_IOERR;
1139 
1140 	if (unlikely(flags & 0x8)) {
1141 		/*
1142 		 * This happens only if this drive failed 10byte rw
1143 		 * command with ILLEGAL_REQUEST during operation and
1144 		 * thus turned off use_10_for_rw.
1145 		 */
1146 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1147 		return BLK_STS_IOERR;
1148 	}
1149 
1150 	cmd->cmd_len = 6;
1151 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1152 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1153 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1154 	cmd->cmnd[3] = lba & 0xff;
1155 	cmd->cmnd[4] = nr_blocks;
1156 	cmd->cmnd[5] = 0;
1157 
1158 	return BLK_STS_OK;
1159 }
1160 
1161 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1162 {
1163 	struct request *rq = cmd->request;
1164 	struct scsi_device *sdp = cmd->device;
1165 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1166 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1167 	sector_t threshold;
1168 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1169 	bool dif, dix;
1170 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1171 	bool write = rq_data_dir(rq) == WRITE;
1172 	unsigned char protect, fua;
1173 	blk_status_t ret;
1174 
1175 	ret = scsi_init_io(cmd);
1176 	if (ret != BLK_STS_OK)
1177 		return ret;
1178 
1179 	if (!scsi_device_online(sdp) || sdp->changed) {
1180 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1181 		return BLK_STS_IOERR;
1182 	}
1183 
1184 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1185 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1186 		return BLK_STS_IOERR;
1187 	}
1188 
1189 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1190 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1191 		return BLK_STS_IOERR;
1192 	}
1193 
1194 	/*
1195 	 * Some SD card readers can't handle accesses which touch the
1196 	 * last one or two logical blocks. Split accesses as needed.
1197 	 */
1198 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1199 
1200 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1201 		if (lba < threshold) {
1202 			/* Access up to the threshold but not beyond */
1203 			nr_blocks = threshold - lba;
1204 		} else {
1205 			/* Access only a single logical block */
1206 			nr_blocks = 1;
1207 		}
1208 	}
1209 
1210 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1211 	dix = scsi_prot_sg_count(cmd);
1212 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1213 
1214 	if (write && dix)
1215 		t10_pi_prepare(cmd->request, sdkp->protection_type);
1216 
1217 	if (dif || dix)
1218 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1219 	else
1220 		protect = 0;
1221 
1222 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1223 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1224 					 protect | fua);
1225 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1226 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1227 					 protect | fua);
1228 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1229 		   sdp->use_10_for_rw || protect) {
1230 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1231 					 protect | fua);
1232 	} else {
1233 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1234 					protect | fua);
1235 	}
1236 
1237 	if (unlikely(ret != BLK_STS_OK))
1238 		return ret;
1239 
1240 	/*
1241 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1242 	 * host adapter, it's safe to assume that we can at least transfer
1243 	 * this many bytes between each connect / disconnect.
1244 	 */
1245 	cmd->transfersize = sdp->sector_size;
1246 	cmd->underflow = nr_blocks << 9;
1247 	cmd->allowed = SD_MAX_RETRIES;
1248 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1249 
1250 	SCSI_LOG_HLQUEUE(1,
1251 			 scmd_printk(KERN_INFO, cmd,
1252 				     "%s: block=%llu, count=%d\n", __func__,
1253 				     (unsigned long long)blk_rq_pos(rq),
1254 				     blk_rq_sectors(rq)));
1255 	SCSI_LOG_HLQUEUE(2,
1256 			 scmd_printk(KERN_INFO, cmd,
1257 				     "%s %d/%u 512 byte blocks.\n",
1258 				     write ? "writing" : "reading", nr_blocks,
1259 				     blk_rq_sectors(rq)));
1260 
1261 	/*
1262 	 * This indicates that the command is ready from our end to be
1263 	 * queued.
1264 	 */
1265 	return BLK_STS_OK;
1266 }
1267 
1268 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1269 {
1270 	struct request *rq = cmd->request;
1271 
1272 	switch (req_op(rq)) {
1273 	case REQ_OP_DISCARD:
1274 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1275 		case SD_LBP_UNMAP:
1276 			return sd_setup_unmap_cmnd(cmd);
1277 		case SD_LBP_WS16:
1278 			return sd_setup_write_same16_cmnd(cmd, true);
1279 		case SD_LBP_WS10:
1280 			return sd_setup_write_same10_cmnd(cmd, true);
1281 		case SD_LBP_ZERO:
1282 			return sd_setup_write_same10_cmnd(cmd, false);
1283 		default:
1284 			return BLK_STS_TARGET;
1285 		}
1286 	case REQ_OP_WRITE_ZEROES:
1287 		return sd_setup_write_zeroes_cmnd(cmd);
1288 	case REQ_OP_WRITE_SAME:
1289 		return sd_setup_write_same_cmnd(cmd);
1290 	case REQ_OP_FLUSH:
1291 		return sd_setup_flush_cmnd(cmd);
1292 	case REQ_OP_READ:
1293 	case REQ_OP_WRITE:
1294 		return sd_setup_read_write_cmnd(cmd);
1295 	case REQ_OP_ZONE_RESET:
1296 		return sd_zbc_setup_reset_cmnd(cmd);
1297 	default:
1298 		WARN_ON_ONCE(1);
1299 		return BLK_STS_NOTSUPP;
1300 	}
1301 }
1302 
1303 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1304 {
1305 	struct request *rq = SCpnt->request;
1306 	u8 *cmnd;
1307 
1308 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1309 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1310 
1311 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1312 		cmnd = SCpnt->cmnd;
1313 		SCpnt->cmnd = NULL;
1314 		SCpnt->cmd_len = 0;
1315 		mempool_free(cmnd, sd_cdb_pool);
1316 	}
1317 }
1318 
1319 /**
1320  *	sd_open - open a scsi disk device
1321  *	@bdev: Block device of the scsi disk to open
1322  *	@mode: FMODE_* mask
1323  *
1324  *	Returns 0 if successful. Returns a negated errno value in case
1325  *	of error.
1326  *
1327  *	Note: This can be called from a user context (e.g. fsck(1) )
1328  *	or from within the kernel (e.g. as a result of a mount(1) ).
1329  *	In the latter case @inode and @filp carry an abridged amount
1330  *	of information as noted above.
1331  *
1332  *	Locking: called with bdev->bd_mutex held.
1333  **/
1334 static int sd_open(struct block_device *bdev, fmode_t mode)
1335 {
1336 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1337 	struct scsi_device *sdev;
1338 	int retval;
1339 
1340 	if (!sdkp)
1341 		return -ENXIO;
1342 
1343 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1344 
1345 	sdev = sdkp->device;
1346 
1347 	/*
1348 	 * If the device is in error recovery, wait until it is done.
1349 	 * If the device is offline, then disallow any access to it.
1350 	 */
1351 	retval = -ENXIO;
1352 	if (!scsi_block_when_processing_errors(sdev))
1353 		goto error_out;
1354 
1355 	if (sdev->removable || sdkp->write_prot)
1356 		check_disk_change(bdev);
1357 
1358 	/*
1359 	 * If the drive is empty, just let the open fail.
1360 	 */
1361 	retval = -ENOMEDIUM;
1362 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1363 		goto error_out;
1364 
1365 	/*
1366 	 * If the device has the write protect tab set, have the open fail
1367 	 * if the user expects to be able to write to the thing.
1368 	 */
1369 	retval = -EROFS;
1370 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1371 		goto error_out;
1372 
1373 	/*
1374 	 * It is possible that the disk changing stuff resulted in
1375 	 * the device being taken offline.  If this is the case,
1376 	 * report this to the user, and don't pretend that the
1377 	 * open actually succeeded.
1378 	 */
1379 	retval = -ENXIO;
1380 	if (!scsi_device_online(sdev))
1381 		goto error_out;
1382 
1383 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1384 		if (scsi_block_when_processing_errors(sdev))
1385 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1386 	}
1387 
1388 	return 0;
1389 
1390 error_out:
1391 	scsi_disk_put(sdkp);
1392 	return retval;
1393 }
1394 
1395 /**
1396  *	sd_release - invoked when the (last) close(2) is called on this
1397  *	scsi disk.
1398  *	@disk: disk to release
1399  *	@mode: FMODE_* mask
1400  *
1401  *	Returns 0.
1402  *
1403  *	Note: may block (uninterruptible) if error recovery is underway
1404  *	on this disk.
1405  *
1406  *	Locking: called with bdev->bd_mutex held.
1407  **/
1408 static void sd_release(struct gendisk *disk, fmode_t mode)
1409 {
1410 	struct scsi_disk *sdkp = scsi_disk(disk);
1411 	struct scsi_device *sdev = sdkp->device;
1412 
1413 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1414 
1415 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1416 		if (scsi_block_when_processing_errors(sdev))
1417 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1418 	}
1419 
1420 	scsi_disk_put(sdkp);
1421 }
1422 
1423 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1424 {
1425 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1426 	struct scsi_device *sdp = sdkp->device;
1427 	struct Scsi_Host *host = sdp->host;
1428 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1429 	int diskinfo[4];
1430 
1431 	/* default to most commonly used values */
1432 	diskinfo[0] = 0x40;	/* 1 << 6 */
1433 	diskinfo[1] = 0x20;	/* 1 << 5 */
1434 	diskinfo[2] = capacity >> 11;
1435 
1436 	/* override with calculated, extended default, or driver values */
1437 	if (host->hostt->bios_param)
1438 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1439 	else
1440 		scsicam_bios_param(bdev, capacity, diskinfo);
1441 
1442 	geo->heads = diskinfo[0];
1443 	geo->sectors = diskinfo[1];
1444 	geo->cylinders = diskinfo[2];
1445 	return 0;
1446 }
1447 
1448 /**
1449  *	sd_ioctl - process an ioctl
1450  *	@bdev: target block device
1451  *	@mode: FMODE_* mask
1452  *	@cmd: ioctl command number
1453  *	@arg: this is third argument given to ioctl(2) system call.
1454  *	Often contains a pointer.
1455  *
1456  *	Returns 0 if successful (some ioctls return positive numbers on
1457  *	success as well). Returns a negated errno value in case of error.
1458  *
1459  *	Note: most ioctls are forward onto the block subsystem or further
1460  *	down in the scsi subsystem.
1461  **/
1462 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1463 		    unsigned int cmd, unsigned long arg)
1464 {
1465 	struct gendisk *disk = bdev->bd_disk;
1466 	struct scsi_disk *sdkp = scsi_disk(disk);
1467 	struct scsi_device *sdp = sdkp->device;
1468 	void __user *p = (void __user *)arg;
1469 	int error;
1470 
1471 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1472 				    "cmd=0x%x\n", disk->disk_name, cmd));
1473 
1474 	error = scsi_verify_blk_ioctl(bdev, cmd);
1475 	if (error < 0)
1476 		return error;
1477 
1478 	/*
1479 	 * If we are in the middle of error recovery, don't let anyone
1480 	 * else try and use this device.  Also, if error recovery fails, it
1481 	 * may try and take the device offline, in which case all further
1482 	 * access to the device is prohibited.
1483 	 */
1484 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1485 			(mode & FMODE_NDELAY) != 0);
1486 	if (error)
1487 		goto out;
1488 
1489 	if (is_sed_ioctl(cmd))
1490 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1491 
1492 	/*
1493 	 * Send SCSI addressing ioctls directly to mid level, send other
1494 	 * ioctls to block level and then onto mid level if they can't be
1495 	 * resolved.
1496 	 */
1497 	switch (cmd) {
1498 		case SCSI_IOCTL_GET_IDLUN:
1499 		case SCSI_IOCTL_GET_BUS_NUMBER:
1500 			error = scsi_ioctl(sdp, cmd, p);
1501 			break;
1502 		default:
1503 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1504 			if (error != -ENOTTY)
1505 				break;
1506 			error = scsi_ioctl(sdp, cmd, p);
1507 			break;
1508 	}
1509 out:
1510 	return error;
1511 }
1512 
1513 static void set_media_not_present(struct scsi_disk *sdkp)
1514 {
1515 	if (sdkp->media_present)
1516 		sdkp->device->changed = 1;
1517 
1518 	if (sdkp->device->removable) {
1519 		sdkp->media_present = 0;
1520 		sdkp->capacity = 0;
1521 	}
1522 }
1523 
1524 static int media_not_present(struct scsi_disk *sdkp,
1525 			     struct scsi_sense_hdr *sshdr)
1526 {
1527 	if (!scsi_sense_valid(sshdr))
1528 		return 0;
1529 
1530 	/* not invoked for commands that could return deferred errors */
1531 	switch (sshdr->sense_key) {
1532 	case UNIT_ATTENTION:
1533 	case NOT_READY:
1534 		/* medium not present */
1535 		if (sshdr->asc == 0x3A) {
1536 			set_media_not_present(sdkp);
1537 			return 1;
1538 		}
1539 	}
1540 	return 0;
1541 }
1542 
1543 /**
1544  *	sd_check_events - check media events
1545  *	@disk: kernel device descriptor
1546  *	@clearing: disk events currently being cleared
1547  *
1548  *	Returns mask of DISK_EVENT_*.
1549  *
1550  *	Note: this function is invoked from the block subsystem.
1551  **/
1552 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1553 {
1554 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1555 	struct scsi_device *sdp;
1556 	int retval;
1557 
1558 	if (!sdkp)
1559 		return 0;
1560 
1561 	sdp = sdkp->device;
1562 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1563 
1564 	/*
1565 	 * If the device is offline, don't send any commands - just pretend as
1566 	 * if the command failed.  If the device ever comes back online, we
1567 	 * can deal with it then.  It is only because of unrecoverable errors
1568 	 * that we would ever take a device offline in the first place.
1569 	 */
1570 	if (!scsi_device_online(sdp)) {
1571 		set_media_not_present(sdkp);
1572 		goto out;
1573 	}
1574 
1575 	/*
1576 	 * Using TEST_UNIT_READY enables differentiation between drive with
1577 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1578 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1579 	 *
1580 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1581 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1582 	 * sd_revalidate() is called.
1583 	 */
1584 	if (scsi_block_when_processing_errors(sdp)) {
1585 		struct scsi_sense_hdr sshdr = { 0, };
1586 
1587 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1588 					      &sshdr);
1589 
1590 		/* failed to execute TUR, assume media not present */
1591 		if (host_byte(retval)) {
1592 			set_media_not_present(sdkp);
1593 			goto out;
1594 		}
1595 
1596 		if (media_not_present(sdkp, &sshdr))
1597 			goto out;
1598 	}
1599 
1600 	/*
1601 	 * For removable scsi disk we have to recognise the presence
1602 	 * of a disk in the drive.
1603 	 */
1604 	if (!sdkp->media_present)
1605 		sdp->changed = 1;
1606 	sdkp->media_present = 1;
1607 out:
1608 	/*
1609 	 * sdp->changed is set under the following conditions:
1610 	 *
1611 	 *	Medium present state has changed in either direction.
1612 	 *	Device has indicated UNIT_ATTENTION.
1613 	 */
1614 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1615 	sdp->changed = 0;
1616 	scsi_disk_put(sdkp);
1617 	return retval;
1618 }
1619 
1620 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1621 {
1622 	int retries, res;
1623 	struct scsi_device *sdp = sdkp->device;
1624 	const int timeout = sdp->request_queue->rq_timeout
1625 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1626 	struct scsi_sense_hdr my_sshdr;
1627 
1628 	if (!scsi_device_online(sdp))
1629 		return -ENODEV;
1630 
1631 	/* caller might not be interested in sense, but we need it */
1632 	if (!sshdr)
1633 		sshdr = &my_sshdr;
1634 
1635 	for (retries = 3; retries > 0; --retries) {
1636 		unsigned char cmd[10] = { 0 };
1637 
1638 		cmd[0] = SYNCHRONIZE_CACHE;
1639 		/*
1640 		 * Leave the rest of the command zero to indicate
1641 		 * flush everything.
1642 		 */
1643 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1644 				timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1645 		if (res == 0)
1646 			break;
1647 	}
1648 
1649 	if (res) {
1650 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1651 
1652 		if (driver_byte(res) == DRIVER_SENSE)
1653 			sd_print_sense_hdr(sdkp, sshdr);
1654 
1655 		/* we need to evaluate the error return  */
1656 		if (scsi_sense_valid(sshdr) &&
1657 			(sshdr->asc == 0x3a ||	/* medium not present */
1658 			 sshdr->asc == 0x20))	/* invalid command */
1659 				/* this is no error here */
1660 				return 0;
1661 
1662 		switch (host_byte(res)) {
1663 		/* ignore errors due to racing a disconnection */
1664 		case DID_BAD_TARGET:
1665 		case DID_NO_CONNECT:
1666 			return 0;
1667 		/* signal the upper layer it might try again */
1668 		case DID_BUS_BUSY:
1669 		case DID_IMM_RETRY:
1670 		case DID_REQUEUE:
1671 		case DID_SOFT_ERROR:
1672 			return -EBUSY;
1673 		default:
1674 			return -EIO;
1675 		}
1676 	}
1677 	return 0;
1678 }
1679 
1680 static void sd_rescan(struct device *dev)
1681 {
1682 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1683 
1684 	revalidate_disk(sdkp->disk);
1685 }
1686 
1687 
1688 #ifdef CONFIG_COMPAT
1689 /*
1690  * This gets directly called from VFS. When the ioctl
1691  * is not recognized we go back to the other translation paths.
1692  */
1693 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1694 			   unsigned int cmd, unsigned long arg)
1695 {
1696 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1697 	int error;
1698 
1699 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1700 			(mode & FMODE_NDELAY) != 0);
1701 	if (error)
1702 		return error;
1703 
1704 	/*
1705 	 * Let the static ioctl translation table take care of it.
1706 	 */
1707 	if (!sdev->host->hostt->compat_ioctl)
1708 		return -ENOIOCTLCMD;
1709 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1710 }
1711 #endif
1712 
1713 static char sd_pr_type(enum pr_type type)
1714 {
1715 	switch (type) {
1716 	case PR_WRITE_EXCLUSIVE:
1717 		return 0x01;
1718 	case PR_EXCLUSIVE_ACCESS:
1719 		return 0x03;
1720 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1721 		return 0x05;
1722 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1723 		return 0x06;
1724 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1725 		return 0x07;
1726 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1727 		return 0x08;
1728 	default:
1729 		return 0;
1730 	}
1731 };
1732 
1733 static int sd_pr_command(struct block_device *bdev, u8 sa,
1734 		u64 key, u64 sa_key, u8 type, u8 flags)
1735 {
1736 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1737 	struct scsi_sense_hdr sshdr;
1738 	int result;
1739 	u8 cmd[16] = { 0, };
1740 	u8 data[24] = { 0, };
1741 
1742 	cmd[0] = PERSISTENT_RESERVE_OUT;
1743 	cmd[1] = sa;
1744 	cmd[2] = type;
1745 	put_unaligned_be32(sizeof(data), &cmd[5]);
1746 
1747 	put_unaligned_be64(key, &data[0]);
1748 	put_unaligned_be64(sa_key, &data[8]);
1749 	data[20] = flags;
1750 
1751 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1752 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1753 
1754 	if (driver_byte(result) == DRIVER_SENSE &&
1755 	    scsi_sense_valid(&sshdr)) {
1756 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1757 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1758 	}
1759 
1760 	return result;
1761 }
1762 
1763 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1764 		u32 flags)
1765 {
1766 	if (flags & ~PR_FL_IGNORE_KEY)
1767 		return -EOPNOTSUPP;
1768 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1769 			old_key, new_key, 0,
1770 			(1 << 0) /* APTPL */);
1771 }
1772 
1773 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1774 		u32 flags)
1775 {
1776 	if (flags)
1777 		return -EOPNOTSUPP;
1778 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1779 }
1780 
1781 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1782 {
1783 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1784 }
1785 
1786 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1787 		enum pr_type type, bool abort)
1788 {
1789 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1790 			     sd_pr_type(type), 0);
1791 }
1792 
1793 static int sd_pr_clear(struct block_device *bdev, u64 key)
1794 {
1795 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1796 }
1797 
1798 static const struct pr_ops sd_pr_ops = {
1799 	.pr_register	= sd_pr_register,
1800 	.pr_reserve	= sd_pr_reserve,
1801 	.pr_release	= sd_pr_release,
1802 	.pr_preempt	= sd_pr_preempt,
1803 	.pr_clear	= sd_pr_clear,
1804 };
1805 
1806 static const struct block_device_operations sd_fops = {
1807 	.owner			= THIS_MODULE,
1808 	.open			= sd_open,
1809 	.release		= sd_release,
1810 	.ioctl			= sd_ioctl,
1811 	.getgeo			= sd_getgeo,
1812 #ifdef CONFIG_COMPAT
1813 	.compat_ioctl		= sd_compat_ioctl,
1814 #endif
1815 	.check_events		= sd_check_events,
1816 	.revalidate_disk	= sd_revalidate_disk,
1817 	.unlock_native_capacity	= sd_unlock_native_capacity,
1818 	.report_zones		= sd_zbc_report_zones,
1819 	.pr_ops			= &sd_pr_ops,
1820 };
1821 
1822 /**
1823  *	sd_eh_reset - reset error handling callback
1824  *	@scmd:		sd-issued command that has failed
1825  *
1826  *	This function is called by the SCSI midlayer before starting
1827  *	SCSI EH. When counting medium access failures we have to be
1828  *	careful to register it only only once per device and SCSI EH run;
1829  *	there might be several timed out commands which will cause the
1830  *	'max_medium_access_timeouts' counter to trigger after the first
1831  *	SCSI EH run already and set the device to offline.
1832  *	So this function resets the internal counter before starting SCSI EH.
1833  **/
1834 static void sd_eh_reset(struct scsi_cmnd *scmd)
1835 {
1836 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1837 
1838 	/* New SCSI EH run, reset gate variable */
1839 	sdkp->ignore_medium_access_errors = false;
1840 }
1841 
1842 /**
1843  *	sd_eh_action - error handling callback
1844  *	@scmd:		sd-issued command that has failed
1845  *	@eh_disp:	The recovery disposition suggested by the midlayer
1846  *
1847  *	This function is called by the SCSI midlayer upon completion of an
1848  *	error test command (currently TEST UNIT READY). The result of sending
1849  *	the eh command is passed in eh_disp.  We're looking for devices that
1850  *	fail medium access commands but are OK with non access commands like
1851  *	test unit ready (so wrongly see the device as having a successful
1852  *	recovery)
1853  **/
1854 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1855 {
1856 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1857 	struct scsi_device *sdev = scmd->device;
1858 
1859 	if (!scsi_device_online(sdev) ||
1860 	    !scsi_medium_access_command(scmd) ||
1861 	    host_byte(scmd->result) != DID_TIME_OUT ||
1862 	    eh_disp != SUCCESS)
1863 		return eh_disp;
1864 
1865 	/*
1866 	 * The device has timed out executing a medium access command.
1867 	 * However, the TEST UNIT READY command sent during error
1868 	 * handling completed successfully. Either the device is in the
1869 	 * process of recovering or has it suffered an internal failure
1870 	 * that prevents access to the storage medium.
1871 	 */
1872 	if (!sdkp->ignore_medium_access_errors) {
1873 		sdkp->medium_access_timed_out++;
1874 		sdkp->ignore_medium_access_errors = true;
1875 	}
1876 
1877 	/*
1878 	 * If the device keeps failing read/write commands but TEST UNIT
1879 	 * READY always completes successfully we assume that medium
1880 	 * access is no longer possible and take the device offline.
1881 	 */
1882 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1883 		scmd_printk(KERN_ERR, scmd,
1884 			    "Medium access timeout failure. Offlining disk!\n");
1885 		mutex_lock(&sdev->state_mutex);
1886 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1887 		mutex_unlock(&sdev->state_mutex);
1888 
1889 		return SUCCESS;
1890 	}
1891 
1892 	return eh_disp;
1893 }
1894 
1895 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1896 {
1897 	struct request *req = scmd->request;
1898 	struct scsi_device *sdev = scmd->device;
1899 	unsigned int transferred, good_bytes;
1900 	u64 start_lba, end_lba, bad_lba;
1901 
1902 	/*
1903 	 * Some commands have a payload smaller than the device logical
1904 	 * block size (e.g. INQUIRY on a 4K disk).
1905 	 */
1906 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1907 		return 0;
1908 
1909 	/* Check if we have a 'bad_lba' information */
1910 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1911 				     SCSI_SENSE_BUFFERSIZE,
1912 				     &bad_lba))
1913 		return 0;
1914 
1915 	/*
1916 	 * If the bad lba was reported incorrectly, we have no idea where
1917 	 * the error is.
1918 	 */
1919 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1920 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1921 	if (bad_lba < start_lba || bad_lba >= end_lba)
1922 		return 0;
1923 
1924 	/*
1925 	 * resid is optional but mostly filled in.  When it's unused,
1926 	 * its value is zero, so we assume the whole buffer transferred
1927 	 */
1928 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1929 
1930 	/* This computation should always be done in terms of the
1931 	 * resolution of the device's medium.
1932 	 */
1933 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1934 
1935 	return min(good_bytes, transferred);
1936 }
1937 
1938 /**
1939  *	sd_done - bottom half handler: called when the lower level
1940  *	driver has completed (successfully or otherwise) a scsi command.
1941  *	@SCpnt: mid-level's per command structure.
1942  *
1943  *	Note: potentially run from within an ISR. Must not block.
1944  **/
1945 static int sd_done(struct scsi_cmnd *SCpnt)
1946 {
1947 	int result = SCpnt->result;
1948 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1949 	unsigned int sector_size = SCpnt->device->sector_size;
1950 	unsigned int resid;
1951 	struct scsi_sense_hdr sshdr;
1952 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1953 	struct request *req = SCpnt->request;
1954 	int sense_valid = 0;
1955 	int sense_deferred = 0;
1956 
1957 	switch (req_op(req)) {
1958 	case REQ_OP_DISCARD:
1959 	case REQ_OP_WRITE_ZEROES:
1960 	case REQ_OP_WRITE_SAME:
1961 	case REQ_OP_ZONE_RESET:
1962 		if (!result) {
1963 			good_bytes = blk_rq_bytes(req);
1964 			scsi_set_resid(SCpnt, 0);
1965 		} else {
1966 			good_bytes = 0;
1967 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1968 		}
1969 		break;
1970 	default:
1971 		/*
1972 		 * In case of bogus fw or device, we could end up having
1973 		 * an unaligned partial completion. Check this here and force
1974 		 * alignment.
1975 		 */
1976 		resid = scsi_get_resid(SCpnt);
1977 		if (resid & (sector_size - 1)) {
1978 			sd_printk(KERN_INFO, sdkp,
1979 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1980 				resid, sector_size);
1981 			resid = min(scsi_bufflen(SCpnt),
1982 				    round_up(resid, sector_size));
1983 			scsi_set_resid(SCpnt, resid);
1984 		}
1985 	}
1986 
1987 	if (result) {
1988 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1989 		if (sense_valid)
1990 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1991 	}
1992 	sdkp->medium_access_timed_out = 0;
1993 
1994 	if (driver_byte(result) != DRIVER_SENSE &&
1995 	    (!sense_valid || sense_deferred))
1996 		goto out;
1997 
1998 	switch (sshdr.sense_key) {
1999 	case HARDWARE_ERROR:
2000 	case MEDIUM_ERROR:
2001 		good_bytes = sd_completed_bytes(SCpnt);
2002 		break;
2003 	case RECOVERED_ERROR:
2004 		good_bytes = scsi_bufflen(SCpnt);
2005 		break;
2006 	case NO_SENSE:
2007 		/* This indicates a false check condition, so ignore it.  An
2008 		 * unknown amount of data was transferred so treat it as an
2009 		 * error.
2010 		 */
2011 		SCpnt->result = 0;
2012 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2013 		break;
2014 	case ABORTED_COMMAND:
2015 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2016 			good_bytes = sd_completed_bytes(SCpnt);
2017 		break;
2018 	case ILLEGAL_REQUEST:
2019 		switch (sshdr.asc) {
2020 		case 0x10:	/* DIX: Host detected corruption */
2021 			good_bytes = sd_completed_bytes(SCpnt);
2022 			break;
2023 		case 0x20:	/* INVALID COMMAND OPCODE */
2024 		case 0x24:	/* INVALID FIELD IN CDB */
2025 			switch (SCpnt->cmnd[0]) {
2026 			case UNMAP:
2027 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2028 				break;
2029 			case WRITE_SAME_16:
2030 			case WRITE_SAME:
2031 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2032 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2033 				} else {
2034 					sdkp->device->no_write_same = 1;
2035 					sd_config_write_same(sdkp);
2036 					req->rq_flags |= RQF_QUIET;
2037 				}
2038 				break;
2039 			}
2040 		}
2041 		break;
2042 	default:
2043 		break;
2044 	}
2045 
2046  out:
2047 	if (sd_is_zoned(sdkp))
2048 		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2049 
2050 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2051 					   "sd_done: completed %d of %d bytes\n",
2052 					   good_bytes, scsi_bufflen(SCpnt)));
2053 
2054 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2055 	    good_bytes)
2056 		t10_pi_complete(SCpnt->request, sdkp->protection_type,
2057 				good_bytes / scsi_prot_interval(SCpnt));
2058 
2059 	return good_bytes;
2060 }
2061 
2062 /*
2063  * spinup disk - called only in sd_revalidate_disk()
2064  */
2065 static void
2066 sd_spinup_disk(struct scsi_disk *sdkp)
2067 {
2068 	unsigned char cmd[10];
2069 	unsigned long spintime_expire = 0;
2070 	int retries, spintime;
2071 	unsigned int the_result;
2072 	struct scsi_sense_hdr sshdr;
2073 	int sense_valid = 0;
2074 
2075 	spintime = 0;
2076 
2077 	/* Spin up drives, as required.  Only do this at boot time */
2078 	/* Spinup needs to be done for module loads too. */
2079 	do {
2080 		retries = 0;
2081 
2082 		do {
2083 			cmd[0] = TEST_UNIT_READY;
2084 			memset((void *) &cmd[1], 0, 9);
2085 
2086 			the_result = scsi_execute_req(sdkp->device, cmd,
2087 						      DMA_NONE, NULL, 0,
2088 						      &sshdr, SD_TIMEOUT,
2089 						      SD_MAX_RETRIES, NULL);
2090 
2091 			/*
2092 			 * If the drive has indicated to us that it
2093 			 * doesn't have any media in it, don't bother
2094 			 * with any more polling.
2095 			 */
2096 			if (media_not_present(sdkp, &sshdr))
2097 				return;
2098 
2099 			if (the_result)
2100 				sense_valid = scsi_sense_valid(&sshdr);
2101 			retries++;
2102 		} while (retries < 3 &&
2103 			 (!scsi_status_is_good(the_result) ||
2104 			  ((driver_byte(the_result) == DRIVER_SENSE) &&
2105 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2106 
2107 		if (driver_byte(the_result) != DRIVER_SENSE) {
2108 			/* no sense, TUR either succeeded or failed
2109 			 * with a status error */
2110 			if(!spintime && !scsi_status_is_good(the_result)) {
2111 				sd_print_result(sdkp, "Test Unit Ready failed",
2112 						the_result);
2113 			}
2114 			break;
2115 		}
2116 
2117 		/*
2118 		 * The device does not want the automatic start to be issued.
2119 		 */
2120 		if (sdkp->device->no_start_on_add)
2121 			break;
2122 
2123 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2124 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2125 				break;	/* manual intervention required */
2126 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2127 				break;	/* standby */
2128 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2129 				break;	/* unavailable */
2130 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2131 				break;	/* sanitize in progress */
2132 			/*
2133 			 * Issue command to spin up drive when not ready
2134 			 */
2135 			if (!spintime) {
2136 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2137 				cmd[0] = START_STOP;
2138 				cmd[1] = 1;	/* Return immediately */
2139 				memset((void *) &cmd[2], 0, 8);
2140 				cmd[4] = 1;	/* Start spin cycle */
2141 				if (sdkp->device->start_stop_pwr_cond)
2142 					cmd[4] |= 1 << 4;
2143 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2144 						 NULL, 0, &sshdr,
2145 						 SD_TIMEOUT, SD_MAX_RETRIES,
2146 						 NULL);
2147 				spintime_expire = jiffies + 100 * HZ;
2148 				spintime = 1;
2149 			}
2150 			/* Wait 1 second for next try */
2151 			msleep(1000);
2152 			printk(KERN_CONT ".");
2153 
2154 		/*
2155 		 * Wait for USB flash devices with slow firmware.
2156 		 * Yes, this sense key/ASC combination shouldn't
2157 		 * occur here.  It's characteristic of these devices.
2158 		 */
2159 		} else if (sense_valid &&
2160 				sshdr.sense_key == UNIT_ATTENTION &&
2161 				sshdr.asc == 0x28) {
2162 			if (!spintime) {
2163 				spintime_expire = jiffies + 5 * HZ;
2164 				spintime = 1;
2165 			}
2166 			/* Wait 1 second for next try */
2167 			msleep(1000);
2168 		} else {
2169 			/* we don't understand the sense code, so it's
2170 			 * probably pointless to loop */
2171 			if(!spintime) {
2172 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2173 				sd_print_sense_hdr(sdkp, &sshdr);
2174 			}
2175 			break;
2176 		}
2177 
2178 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2179 
2180 	if (spintime) {
2181 		if (scsi_status_is_good(the_result))
2182 			printk(KERN_CONT "ready\n");
2183 		else
2184 			printk(KERN_CONT "not responding...\n");
2185 	}
2186 }
2187 
2188 /*
2189  * Determine whether disk supports Data Integrity Field.
2190  */
2191 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2192 {
2193 	struct scsi_device *sdp = sdkp->device;
2194 	u8 type;
2195 	int ret = 0;
2196 
2197 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2198 		return ret;
2199 
2200 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2201 
2202 	if (type > T10_PI_TYPE3_PROTECTION)
2203 		ret = -ENODEV;
2204 	else if (scsi_host_dif_capable(sdp->host, type))
2205 		ret = 1;
2206 
2207 	if (sdkp->first_scan || type != sdkp->protection_type)
2208 		switch (ret) {
2209 		case -ENODEV:
2210 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2211 				  " protection type %u. Disabling disk!\n",
2212 				  type);
2213 			break;
2214 		case 1:
2215 			sd_printk(KERN_NOTICE, sdkp,
2216 				  "Enabling DIF Type %u protection\n", type);
2217 			break;
2218 		case 0:
2219 			sd_printk(KERN_NOTICE, sdkp,
2220 				  "Disabling DIF Type %u protection\n", type);
2221 			break;
2222 		}
2223 
2224 	sdkp->protection_type = type;
2225 
2226 	return ret;
2227 }
2228 
2229 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2230 			struct scsi_sense_hdr *sshdr, int sense_valid,
2231 			int the_result)
2232 {
2233 	if (driver_byte(the_result) == DRIVER_SENSE)
2234 		sd_print_sense_hdr(sdkp, sshdr);
2235 	else
2236 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2237 
2238 	/*
2239 	 * Set dirty bit for removable devices if not ready -
2240 	 * sometimes drives will not report this properly.
2241 	 */
2242 	if (sdp->removable &&
2243 	    sense_valid && sshdr->sense_key == NOT_READY)
2244 		set_media_not_present(sdkp);
2245 
2246 	/*
2247 	 * We used to set media_present to 0 here to indicate no media
2248 	 * in the drive, but some drives fail read capacity even with
2249 	 * media present, so we can't do that.
2250 	 */
2251 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2252 }
2253 
2254 #define RC16_LEN 32
2255 #if RC16_LEN > SD_BUF_SIZE
2256 #error RC16_LEN must not be more than SD_BUF_SIZE
2257 #endif
2258 
2259 #define READ_CAPACITY_RETRIES_ON_RESET	10
2260 
2261 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2262 						unsigned char *buffer)
2263 {
2264 	unsigned char cmd[16];
2265 	struct scsi_sense_hdr sshdr;
2266 	int sense_valid = 0;
2267 	int the_result;
2268 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2269 	unsigned int alignment;
2270 	unsigned long long lba;
2271 	unsigned sector_size;
2272 
2273 	if (sdp->no_read_capacity_16)
2274 		return -EINVAL;
2275 
2276 	do {
2277 		memset(cmd, 0, 16);
2278 		cmd[0] = SERVICE_ACTION_IN_16;
2279 		cmd[1] = SAI_READ_CAPACITY_16;
2280 		cmd[13] = RC16_LEN;
2281 		memset(buffer, 0, RC16_LEN);
2282 
2283 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2284 					buffer, RC16_LEN, &sshdr,
2285 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2286 
2287 		if (media_not_present(sdkp, &sshdr))
2288 			return -ENODEV;
2289 
2290 		if (the_result) {
2291 			sense_valid = scsi_sense_valid(&sshdr);
2292 			if (sense_valid &&
2293 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2294 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2295 			    sshdr.ascq == 0x00)
2296 				/* Invalid Command Operation Code or
2297 				 * Invalid Field in CDB, just retry
2298 				 * silently with RC10 */
2299 				return -EINVAL;
2300 			if (sense_valid &&
2301 			    sshdr.sense_key == UNIT_ATTENTION &&
2302 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2303 				/* Device reset might occur several times,
2304 				 * give it one more chance */
2305 				if (--reset_retries > 0)
2306 					continue;
2307 		}
2308 		retries--;
2309 
2310 	} while (the_result && retries);
2311 
2312 	if (the_result) {
2313 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2314 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2315 		return -EINVAL;
2316 	}
2317 
2318 	sector_size = get_unaligned_be32(&buffer[8]);
2319 	lba = get_unaligned_be64(&buffer[0]);
2320 
2321 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2322 		sdkp->capacity = 0;
2323 		return -ENODEV;
2324 	}
2325 
2326 	/* Logical blocks per physical block exponent */
2327 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2328 
2329 	/* RC basis */
2330 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2331 
2332 	/* Lowest aligned logical block */
2333 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2334 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2335 	if (alignment && sdkp->first_scan)
2336 		sd_printk(KERN_NOTICE, sdkp,
2337 			  "physical block alignment offset: %u\n", alignment);
2338 
2339 	if (buffer[14] & 0x80) { /* LBPME */
2340 		sdkp->lbpme = 1;
2341 
2342 		if (buffer[14] & 0x40) /* LBPRZ */
2343 			sdkp->lbprz = 1;
2344 
2345 		sd_config_discard(sdkp, SD_LBP_WS16);
2346 	}
2347 
2348 	sdkp->capacity = lba + 1;
2349 	return sector_size;
2350 }
2351 
2352 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2353 						unsigned char *buffer)
2354 {
2355 	unsigned char cmd[16];
2356 	struct scsi_sense_hdr sshdr;
2357 	int sense_valid = 0;
2358 	int the_result;
2359 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2360 	sector_t lba;
2361 	unsigned sector_size;
2362 
2363 	do {
2364 		cmd[0] = READ_CAPACITY;
2365 		memset(&cmd[1], 0, 9);
2366 		memset(buffer, 0, 8);
2367 
2368 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2369 					buffer, 8, &sshdr,
2370 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2371 
2372 		if (media_not_present(sdkp, &sshdr))
2373 			return -ENODEV;
2374 
2375 		if (the_result) {
2376 			sense_valid = scsi_sense_valid(&sshdr);
2377 			if (sense_valid &&
2378 			    sshdr.sense_key == UNIT_ATTENTION &&
2379 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2380 				/* Device reset might occur several times,
2381 				 * give it one more chance */
2382 				if (--reset_retries > 0)
2383 					continue;
2384 		}
2385 		retries--;
2386 
2387 	} while (the_result && retries);
2388 
2389 	if (the_result) {
2390 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2391 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2392 		return -EINVAL;
2393 	}
2394 
2395 	sector_size = get_unaligned_be32(&buffer[4]);
2396 	lba = get_unaligned_be32(&buffer[0]);
2397 
2398 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2399 		/* Some buggy (usb cardreader) devices return an lba of
2400 		   0xffffffff when the want to report a size of 0 (with
2401 		   which they really mean no media is present) */
2402 		sdkp->capacity = 0;
2403 		sdkp->physical_block_size = sector_size;
2404 		return sector_size;
2405 	}
2406 
2407 	sdkp->capacity = lba + 1;
2408 	sdkp->physical_block_size = sector_size;
2409 	return sector_size;
2410 }
2411 
2412 static int sd_try_rc16_first(struct scsi_device *sdp)
2413 {
2414 	if (sdp->host->max_cmd_len < 16)
2415 		return 0;
2416 	if (sdp->try_rc_10_first)
2417 		return 0;
2418 	if (sdp->scsi_level > SCSI_SPC_2)
2419 		return 1;
2420 	if (scsi_device_protection(sdp))
2421 		return 1;
2422 	return 0;
2423 }
2424 
2425 /*
2426  * read disk capacity
2427  */
2428 static void
2429 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2430 {
2431 	int sector_size;
2432 	struct scsi_device *sdp = sdkp->device;
2433 
2434 	if (sd_try_rc16_first(sdp)) {
2435 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2436 		if (sector_size == -EOVERFLOW)
2437 			goto got_data;
2438 		if (sector_size == -ENODEV)
2439 			return;
2440 		if (sector_size < 0)
2441 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2442 		if (sector_size < 0)
2443 			return;
2444 	} else {
2445 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2446 		if (sector_size == -EOVERFLOW)
2447 			goto got_data;
2448 		if (sector_size < 0)
2449 			return;
2450 		if ((sizeof(sdkp->capacity) > 4) &&
2451 		    (sdkp->capacity > 0xffffffffULL)) {
2452 			int old_sector_size = sector_size;
2453 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2454 					"Trying to use READ CAPACITY(16).\n");
2455 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2456 			if (sector_size < 0) {
2457 				sd_printk(KERN_NOTICE, sdkp,
2458 					"Using 0xffffffff as device size\n");
2459 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2460 				sector_size = old_sector_size;
2461 				goto got_data;
2462 			}
2463 			/* Remember that READ CAPACITY(16) succeeded */
2464 			sdp->try_rc_10_first = 0;
2465 		}
2466 	}
2467 
2468 	/* Some devices are known to return the total number of blocks,
2469 	 * not the highest block number.  Some devices have versions
2470 	 * which do this and others which do not.  Some devices we might
2471 	 * suspect of doing this but we don't know for certain.
2472 	 *
2473 	 * If we know the reported capacity is wrong, decrement it.  If
2474 	 * we can only guess, then assume the number of blocks is even
2475 	 * (usually true but not always) and err on the side of lowering
2476 	 * the capacity.
2477 	 */
2478 	if (sdp->fix_capacity ||
2479 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2480 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2481 				"from its reported value: %llu\n",
2482 				(unsigned long long) sdkp->capacity);
2483 		--sdkp->capacity;
2484 	}
2485 
2486 got_data:
2487 	if (sector_size == 0) {
2488 		sector_size = 512;
2489 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2490 			  "assuming 512.\n");
2491 	}
2492 
2493 	if (sector_size != 512 &&
2494 	    sector_size != 1024 &&
2495 	    sector_size != 2048 &&
2496 	    sector_size != 4096) {
2497 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2498 			  sector_size);
2499 		/*
2500 		 * The user might want to re-format the drive with
2501 		 * a supported sectorsize.  Once this happens, it
2502 		 * would be relatively trivial to set the thing up.
2503 		 * For this reason, we leave the thing in the table.
2504 		 */
2505 		sdkp->capacity = 0;
2506 		/*
2507 		 * set a bogus sector size so the normal read/write
2508 		 * logic in the block layer will eventually refuse any
2509 		 * request on this device without tripping over power
2510 		 * of two sector size assumptions
2511 		 */
2512 		sector_size = 512;
2513 	}
2514 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2515 	blk_queue_physical_block_size(sdp->request_queue,
2516 				      sdkp->physical_block_size);
2517 	sdkp->device->sector_size = sector_size;
2518 
2519 	if (sdkp->capacity > 0xffffffff)
2520 		sdp->use_16_for_rw = 1;
2521 
2522 }
2523 
2524 /*
2525  * Print disk capacity
2526  */
2527 static void
2528 sd_print_capacity(struct scsi_disk *sdkp,
2529 		  sector_t old_capacity)
2530 {
2531 	int sector_size = sdkp->device->sector_size;
2532 	char cap_str_2[10], cap_str_10[10];
2533 
2534 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2535 		return;
2536 
2537 	string_get_size(sdkp->capacity, sector_size,
2538 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2539 	string_get_size(sdkp->capacity, sector_size,
2540 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2541 
2542 	sd_printk(KERN_NOTICE, sdkp,
2543 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2544 		  (unsigned long long)sdkp->capacity,
2545 		  sector_size, cap_str_10, cap_str_2);
2546 
2547 	if (sdkp->physical_block_size != sector_size)
2548 		sd_printk(KERN_NOTICE, sdkp,
2549 			  "%u-byte physical blocks\n",
2550 			  sdkp->physical_block_size);
2551 
2552 	sd_zbc_print_zones(sdkp);
2553 }
2554 
2555 /* called with buffer of length 512 */
2556 static inline int
2557 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2558 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2559 		 struct scsi_sense_hdr *sshdr)
2560 {
2561 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2562 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2563 			       sshdr);
2564 }
2565 
2566 /*
2567  * read write protect setting, if possible - called only in sd_revalidate_disk()
2568  * called with buffer of length SD_BUF_SIZE
2569  */
2570 static void
2571 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2572 {
2573 	int res;
2574 	struct scsi_device *sdp = sdkp->device;
2575 	struct scsi_mode_data data;
2576 	int old_wp = sdkp->write_prot;
2577 
2578 	set_disk_ro(sdkp->disk, 0);
2579 	if (sdp->skip_ms_page_3f) {
2580 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2581 		return;
2582 	}
2583 
2584 	if (sdp->use_192_bytes_for_3f) {
2585 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2586 	} else {
2587 		/*
2588 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2589 		 * We have to start carefully: some devices hang if we ask
2590 		 * for more than is available.
2591 		 */
2592 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2593 
2594 		/*
2595 		 * Second attempt: ask for page 0 When only page 0 is
2596 		 * implemented, a request for page 3F may return Sense Key
2597 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2598 		 * CDB.
2599 		 */
2600 		if (!scsi_status_is_good(res))
2601 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2602 
2603 		/*
2604 		 * Third attempt: ask 255 bytes, as we did earlier.
2605 		 */
2606 		if (!scsi_status_is_good(res))
2607 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2608 					       &data, NULL);
2609 	}
2610 
2611 	if (!scsi_status_is_good(res)) {
2612 		sd_first_printk(KERN_WARNING, sdkp,
2613 			  "Test WP failed, assume Write Enabled\n");
2614 	} else {
2615 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2616 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2617 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2618 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2619 				  sdkp->write_prot ? "on" : "off");
2620 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2621 		}
2622 	}
2623 }
2624 
2625 /*
2626  * sd_read_cache_type - called only from sd_revalidate_disk()
2627  * called with buffer of length SD_BUF_SIZE
2628  */
2629 static void
2630 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2631 {
2632 	int len = 0, res;
2633 	struct scsi_device *sdp = sdkp->device;
2634 
2635 	int dbd;
2636 	int modepage;
2637 	int first_len;
2638 	struct scsi_mode_data data;
2639 	struct scsi_sense_hdr sshdr;
2640 	int old_wce = sdkp->WCE;
2641 	int old_rcd = sdkp->RCD;
2642 	int old_dpofua = sdkp->DPOFUA;
2643 
2644 
2645 	if (sdkp->cache_override)
2646 		return;
2647 
2648 	first_len = 4;
2649 	if (sdp->skip_ms_page_8) {
2650 		if (sdp->type == TYPE_RBC)
2651 			goto defaults;
2652 		else {
2653 			if (sdp->skip_ms_page_3f)
2654 				goto defaults;
2655 			modepage = 0x3F;
2656 			if (sdp->use_192_bytes_for_3f)
2657 				first_len = 192;
2658 			dbd = 0;
2659 		}
2660 	} else if (sdp->type == TYPE_RBC) {
2661 		modepage = 6;
2662 		dbd = 8;
2663 	} else {
2664 		modepage = 8;
2665 		dbd = 0;
2666 	}
2667 
2668 	/* cautiously ask */
2669 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2670 			&data, &sshdr);
2671 
2672 	if (!scsi_status_is_good(res))
2673 		goto bad_sense;
2674 
2675 	if (!data.header_length) {
2676 		modepage = 6;
2677 		first_len = 0;
2678 		sd_first_printk(KERN_ERR, sdkp,
2679 				"Missing header in MODE_SENSE response\n");
2680 	}
2681 
2682 	/* that went OK, now ask for the proper length */
2683 	len = data.length;
2684 
2685 	/*
2686 	 * We're only interested in the first three bytes, actually.
2687 	 * But the data cache page is defined for the first 20.
2688 	 */
2689 	if (len < 3)
2690 		goto bad_sense;
2691 	else if (len > SD_BUF_SIZE) {
2692 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2693 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2694 		len = SD_BUF_SIZE;
2695 	}
2696 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2697 		len = 192;
2698 
2699 	/* Get the data */
2700 	if (len > first_len)
2701 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2702 				&data, &sshdr);
2703 
2704 	if (scsi_status_is_good(res)) {
2705 		int offset = data.header_length + data.block_descriptor_length;
2706 
2707 		while (offset < len) {
2708 			u8 page_code = buffer[offset] & 0x3F;
2709 			u8 spf       = buffer[offset] & 0x40;
2710 
2711 			if (page_code == 8 || page_code == 6) {
2712 				/* We're interested only in the first 3 bytes.
2713 				 */
2714 				if (len - offset <= 2) {
2715 					sd_first_printk(KERN_ERR, sdkp,
2716 						"Incomplete mode parameter "
2717 							"data\n");
2718 					goto defaults;
2719 				} else {
2720 					modepage = page_code;
2721 					goto Page_found;
2722 				}
2723 			} else {
2724 				/* Go to the next page */
2725 				if (spf && len - offset > 3)
2726 					offset += 4 + (buffer[offset+2] << 8) +
2727 						buffer[offset+3];
2728 				else if (!spf && len - offset > 1)
2729 					offset += 2 + buffer[offset+1];
2730 				else {
2731 					sd_first_printk(KERN_ERR, sdkp,
2732 							"Incomplete mode "
2733 							"parameter data\n");
2734 					goto defaults;
2735 				}
2736 			}
2737 		}
2738 
2739 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2740 		goto defaults;
2741 
2742 	Page_found:
2743 		if (modepage == 8) {
2744 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2745 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2746 		} else {
2747 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2748 			sdkp->RCD = 0;
2749 		}
2750 
2751 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2752 		if (sdp->broken_fua) {
2753 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2754 			sdkp->DPOFUA = 0;
2755 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2756 			   !sdkp->device->use_16_for_rw) {
2757 			sd_first_printk(KERN_NOTICE, sdkp,
2758 				  "Uses READ/WRITE(6), disabling FUA\n");
2759 			sdkp->DPOFUA = 0;
2760 		}
2761 
2762 		/* No cache flush allowed for write protected devices */
2763 		if (sdkp->WCE && sdkp->write_prot)
2764 			sdkp->WCE = 0;
2765 
2766 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2767 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2768 			sd_printk(KERN_NOTICE, sdkp,
2769 				  "Write cache: %s, read cache: %s, %s\n",
2770 				  sdkp->WCE ? "enabled" : "disabled",
2771 				  sdkp->RCD ? "disabled" : "enabled",
2772 				  sdkp->DPOFUA ? "supports DPO and FUA"
2773 				  : "doesn't support DPO or FUA");
2774 
2775 		return;
2776 	}
2777 
2778 bad_sense:
2779 	if (scsi_sense_valid(&sshdr) &&
2780 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2781 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2782 		/* Invalid field in CDB */
2783 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2784 	else
2785 		sd_first_printk(KERN_ERR, sdkp,
2786 				"Asking for cache data failed\n");
2787 
2788 defaults:
2789 	if (sdp->wce_default_on) {
2790 		sd_first_printk(KERN_NOTICE, sdkp,
2791 				"Assuming drive cache: write back\n");
2792 		sdkp->WCE = 1;
2793 	} else {
2794 		sd_first_printk(KERN_ERR, sdkp,
2795 				"Assuming drive cache: write through\n");
2796 		sdkp->WCE = 0;
2797 	}
2798 	sdkp->RCD = 0;
2799 	sdkp->DPOFUA = 0;
2800 }
2801 
2802 /*
2803  * The ATO bit indicates whether the DIF application tag is available
2804  * for use by the operating system.
2805  */
2806 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2807 {
2808 	int res, offset;
2809 	struct scsi_device *sdp = sdkp->device;
2810 	struct scsi_mode_data data;
2811 	struct scsi_sense_hdr sshdr;
2812 
2813 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2814 		return;
2815 
2816 	if (sdkp->protection_type == 0)
2817 		return;
2818 
2819 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2820 			      SD_MAX_RETRIES, &data, &sshdr);
2821 
2822 	if (!scsi_status_is_good(res) || !data.header_length ||
2823 	    data.length < 6) {
2824 		sd_first_printk(KERN_WARNING, sdkp,
2825 			  "getting Control mode page failed, assume no ATO\n");
2826 
2827 		if (scsi_sense_valid(&sshdr))
2828 			sd_print_sense_hdr(sdkp, &sshdr);
2829 
2830 		return;
2831 	}
2832 
2833 	offset = data.header_length + data.block_descriptor_length;
2834 
2835 	if ((buffer[offset] & 0x3f) != 0x0a) {
2836 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2837 		return;
2838 	}
2839 
2840 	if ((buffer[offset + 5] & 0x80) == 0)
2841 		return;
2842 
2843 	sdkp->ATO = 1;
2844 
2845 	return;
2846 }
2847 
2848 /**
2849  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2850  * @sdkp: disk to query
2851  */
2852 static void sd_read_block_limits(struct scsi_disk *sdkp)
2853 {
2854 	unsigned int sector_sz = sdkp->device->sector_size;
2855 	const int vpd_len = 64;
2856 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2857 
2858 	if (!buffer ||
2859 	    /* Block Limits VPD */
2860 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2861 		goto out;
2862 
2863 	blk_queue_io_min(sdkp->disk->queue,
2864 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2865 
2866 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2867 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2868 
2869 	if (buffer[3] == 0x3c) {
2870 		unsigned int lba_count, desc_count;
2871 
2872 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2873 
2874 		if (!sdkp->lbpme)
2875 			goto out;
2876 
2877 		lba_count = get_unaligned_be32(&buffer[20]);
2878 		desc_count = get_unaligned_be32(&buffer[24]);
2879 
2880 		if (lba_count && desc_count)
2881 			sdkp->max_unmap_blocks = lba_count;
2882 
2883 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2884 
2885 		if (buffer[32] & 0x80)
2886 			sdkp->unmap_alignment =
2887 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2888 
2889 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2890 
2891 			if (sdkp->max_unmap_blocks)
2892 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2893 			else
2894 				sd_config_discard(sdkp, SD_LBP_WS16);
2895 
2896 		} else {	/* LBP VPD page tells us what to use */
2897 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2898 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2899 			else if (sdkp->lbpws)
2900 				sd_config_discard(sdkp, SD_LBP_WS16);
2901 			else if (sdkp->lbpws10)
2902 				sd_config_discard(sdkp, SD_LBP_WS10);
2903 			else
2904 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2905 		}
2906 	}
2907 
2908  out:
2909 	kfree(buffer);
2910 }
2911 
2912 /**
2913  * sd_read_block_characteristics - Query block dev. characteristics
2914  * @sdkp: disk to query
2915  */
2916 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2917 {
2918 	struct request_queue *q = sdkp->disk->queue;
2919 	unsigned char *buffer;
2920 	u16 rot;
2921 	const int vpd_len = 64;
2922 
2923 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2924 
2925 	if (!buffer ||
2926 	    /* Block Device Characteristics VPD */
2927 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2928 		goto out;
2929 
2930 	rot = get_unaligned_be16(&buffer[4]);
2931 
2932 	if (rot == 1) {
2933 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2934 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2935 	}
2936 
2937 	if (sdkp->device->type == TYPE_ZBC) {
2938 		/* Host-managed */
2939 		q->limits.zoned = BLK_ZONED_HM;
2940 	} else {
2941 		sdkp->zoned = (buffer[8] >> 4) & 3;
2942 		if (sdkp->zoned == 1)
2943 			/* Host-aware */
2944 			q->limits.zoned = BLK_ZONED_HA;
2945 		else
2946 			/*
2947 			 * Treat drive-managed devices as
2948 			 * regular block devices.
2949 			 */
2950 			q->limits.zoned = BLK_ZONED_NONE;
2951 	}
2952 	if (blk_queue_is_zoned(q) && sdkp->first_scan)
2953 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2954 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2955 
2956  out:
2957 	kfree(buffer);
2958 }
2959 
2960 /**
2961  * sd_read_block_provisioning - Query provisioning VPD page
2962  * @sdkp: disk to query
2963  */
2964 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2965 {
2966 	unsigned char *buffer;
2967 	const int vpd_len = 8;
2968 
2969 	if (sdkp->lbpme == 0)
2970 		return;
2971 
2972 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2973 
2974 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2975 		goto out;
2976 
2977 	sdkp->lbpvpd	= 1;
2978 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2979 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2980 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2981 
2982  out:
2983 	kfree(buffer);
2984 }
2985 
2986 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2987 {
2988 	struct scsi_device *sdev = sdkp->device;
2989 
2990 	if (sdev->host->no_write_same) {
2991 		sdev->no_write_same = 1;
2992 
2993 		return;
2994 	}
2995 
2996 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2997 		/* too large values might cause issues with arcmsr */
2998 		int vpd_buf_len = 64;
2999 
3000 		sdev->no_report_opcodes = 1;
3001 
3002 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3003 		 * CODES is unsupported and the device has an ATA
3004 		 * Information VPD page (SAT).
3005 		 */
3006 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3007 			sdev->no_write_same = 1;
3008 	}
3009 
3010 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3011 		sdkp->ws16 = 1;
3012 
3013 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3014 		sdkp->ws10 = 1;
3015 }
3016 
3017 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3018 {
3019 	struct scsi_device *sdev = sdkp->device;
3020 
3021 	if (!sdev->security_supported)
3022 		return;
3023 
3024 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3025 			SECURITY_PROTOCOL_IN) == 1 &&
3026 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3027 			SECURITY_PROTOCOL_OUT) == 1)
3028 		sdkp->security = 1;
3029 }
3030 
3031 /*
3032  * Determine the device's preferred I/O size for reads and writes
3033  * unless the reported value is unreasonably small, large, not a
3034  * multiple of the physical block size, or simply garbage.
3035  */
3036 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3037 				      unsigned int dev_max)
3038 {
3039 	struct scsi_device *sdp = sdkp->device;
3040 	unsigned int opt_xfer_bytes =
3041 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3042 
3043 	if (sdkp->opt_xfer_blocks == 0)
3044 		return false;
3045 
3046 	if (sdkp->opt_xfer_blocks > dev_max) {
3047 		sd_first_printk(KERN_WARNING, sdkp,
3048 				"Optimal transfer size %u logical blocks " \
3049 				"> dev_max (%u logical blocks)\n",
3050 				sdkp->opt_xfer_blocks, dev_max);
3051 		return false;
3052 	}
3053 
3054 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3055 		sd_first_printk(KERN_WARNING, sdkp,
3056 				"Optimal transfer size %u logical blocks " \
3057 				"> sd driver limit (%u logical blocks)\n",
3058 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3059 		return false;
3060 	}
3061 
3062 	if (opt_xfer_bytes < PAGE_SIZE) {
3063 		sd_first_printk(KERN_WARNING, sdkp,
3064 				"Optimal transfer size %u bytes < " \
3065 				"PAGE_SIZE (%u bytes)\n",
3066 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3067 		return false;
3068 	}
3069 
3070 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3071 		sd_first_printk(KERN_WARNING, sdkp,
3072 				"Optimal transfer size %u bytes not a " \
3073 				"multiple of physical block size (%u bytes)\n",
3074 				opt_xfer_bytes, sdkp->physical_block_size);
3075 		return false;
3076 	}
3077 
3078 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3079 			opt_xfer_bytes);
3080 	return true;
3081 }
3082 
3083 /**
3084  *	sd_revalidate_disk - called the first time a new disk is seen,
3085  *	performs disk spin up, read_capacity, etc.
3086  *	@disk: struct gendisk we care about
3087  **/
3088 static int sd_revalidate_disk(struct gendisk *disk)
3089 {
3090 	struct scsi_disk *sdkp = scsi_disk(disk);
3091 	struct scsi_device *sdp = sdkp->device;
3092 	struct request_queue *q = sdkp->disk->queue;
3093 	sector_t old_capacity = sdkp->capacity;
3094 	unsigned char *buffer;
3095 	unsigned int dev_max, rw_max;
3096 
3097 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3098 				      "sd_revalidate_disk\n"));
3099 
3100 	/*
3101 	 * If the device is offline, don't try and read capacity or any
3102 	 * of the other niceties.
3103 	 */
3104 	if (!scsi_device_online(sdp))
3105 		goto out;
3106 
3107 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3108 	if (!buffer) {
3109 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3110 			  "allocation failure.\n");
3111 		goto out;
3112 	}
3113 
3114 	sd_spinup_disk(sdkp);
3115 
3116 	/*
3117 	 * Without media there is no reason to ask; moreover, some devices
3118 	 * react badly if we do.
3119 	 */
3120 	if (sdkp->media_present) {
3121 		sd_read_capacity(sdkp, buffer);
3122 
3123 		/*
3124 		 * set the default to rotational.  All non-rotational devices
3125 		 * support the block characteristics VPD page, which will
3126 		 * cause this to be updated correctly and any device which
3127 		 * doesn't support it should be treated as rotational.
3128 		 */
3129 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3130 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3131 
3132 		if (scsi_device_supports_vpd(sdp)) {
3133 			sd_read_block_provisioning(sdkp);
3134 			sd_read_block_limits(sdkp);
3135 			sd_read_block_characteristics(sdkp);
3136 			sd_zbc_read_zones(sdkp, buffer);
3137 		}
3138 
3139 		sd_print_capacity(sdkp, old_capacity);
3140 
3141 		sd_read_write_protect_flag(sdkp, buffer);
3142 		sd_read_cache_type(sdkp, buffer);
3143 		sd_read_app_tag_own(sdkp, buffer);
3144 		sd_read_write_same(sdkp, buffer);
3145 		sd_read_security(sdkp, buffer);
3146 	}
3147 
3148 	/*
3149 	 * We now have all cache related info, determine how we deal
3150 	 * with flush requests.
3151 	 */
3152 	sd_set_flush_flag(sdkp);
3153 
3154 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3155 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3156 
3157 	/* Some devices report a maximum block count for READ/WRITE requests. */
3158 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3159 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3160 
3161 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3162 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3163 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3164 	} else
3165 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3166 				      (sector_t)BLK_DEF_MAX_SECTORS);
3167 
3168 	/* Do not exceed controller limit */
3169 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3170 
3171 	/*
3172 	 * Only update max_sectors if previously unset or if the current value
3173 	 * exceeds the capabilities of the hardware.
3174 	 */
3175 	if (sdkp->first_scan ||
3176 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3177 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3178 		q->limits.max_sectors = rw_max;
3179 
3180 	sdkp->first_scan = 0;
3181 
3182 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3183 	sd_config_write_same(sdkp);
3184 	kfree(buffer);
3185 
3186  out:
3187 	return 0;
3188 }
3189 
3190 /**
3191  *	sd_unlock_native_capacity - unlock native capacity
3192  *	@disk: struct gendisk to set capacity for
3193  *
3194  *	Block layer calls this function if it detects that partitions
3195  *	on @disk reach beyond the end of the device.  If the SCSI host
3196  *	implements ->unlock_native_capacity() method, it's invoked to
3197  *	give it a chance to adjust the device capacity.
3198  *
3199  *	CONTEXT:
3200  *	Defined by block layer.  Might sleep.
3201  */
3202 static void sd_unlock_native_capacity(struct gendisk *disk)
3203 {
3204 	struct scsi_device *sdev = scsi_disk(disk)->device;
3205 
3206 	if (sdev->host->hostt->unlock_native_capacity)
3207 		sdev->host->hostt->unlock_native_capacity(sdev);
3208 }
3209 
3210 /**
3211  *	sd_format_disk_name - format disk name
3212  *	@prefix: name prefix - ie. "sd" for SCSI disks
3213  *	@index: index of the disk to format name for
3214  *	@buf: output buffer
3215  *	@buflen: length of the output buffer
3216  *
3217  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3218  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3219  *	which is followed by sdaaa.
3220  *
3221  *	This is basically 26 base counting with one extra 'nil' entry
3222  *	at the beginning from the second digit on and can be
3223  *	determined using similar method as 26 base conversion with the
3224  *	index shifted -1 after each digit is computed.
3225  *
3226  *	CONTEXT:
3227  *	Don't care.
3228  *
3229  *	RETURNS:
3230  *	0 on success, -errno on failure.
3231  */
3232 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3233 {
3234 	const int base = 'z' - 'a' + 1;
3235 	char *begin = buf + strlen(prefix);
3236 	char *end = buf + buflen;
3237 	char *p;
3238 	int unit;
3239 
3240 	p = end - 1;
3241 	*p = '\0';
3242 	unit = base;
3243 	do {
3244 		if (p == begin)
3245 			return -EINVAL;
3246 		*--p = 'a' + (index % unit);
3247 		index = (index / unit) - 1;
3248 	} while (index >= 0);
3249 
3250 	memmove(begin, p, end - p);
3251 	memcpy(buf, prefix, strlen(prefix));
3252 
3253 	return 0;
3254 }
3255 
3256 /**
3257  *	sd_probe - called during driver initialization and whenever a
3258  *	new scsi device is attached to the system. It is called once
3259  *	for each scsi device (not just disks) present.
3260  *	@dev: pointer to device object
3261  *
3262  *	Returns 0 if successful (or not interested in this scsi device
3263  *	(e.g. scanner)); 1 when there is an error.
3264  *
3265  *	Note: this function is invoked from the scsi mid-level.
3266  *	This function sets up the mapping between a given
3267  *	<host,channel,id,lun> (found in sdp) and new device name
3268  *	(e.g. /dev/sda). More precisely it is the block device major
3269  *	and minor number that is chosen here.
3270  *
3271  *	Assume sd_probe is not re-entrant (for time being)
3272  *	Also think about sd_probe() and sd_remove() running coincidentally.
3273  **/
3274 static int sd_probe(struct device *dev)
3275 {
3276 	struct scsi_device *sdp = to_scsi_device(dev);
3277 	struct scsi_disk *sdkp;
3278 	struct gendisk *gd;
3279 	int index;
3280 	int error;
3281 
3282 	scsi_autopm_get_device(sdp);
3283 	error = -ENODEV;
3284 	if (sdp->type != TYPE_DISK &&
3285 	    sdp->type != TYPE_ZBC &&
3286 	    sdp->type != TYPE_MOD &&
3287 	    sdp->type != TYPE_RBC)
3288 		goto out;
3289 
3290 #ifndef CONFIG_BLK_DEV_ZONED
3291 	if (sdp->type == TYPE_ZBC)
3292 		goto out;
3293 #endif
3294 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3295 					"sd_probe\n"));
3296 
3297 	error = -ENOMEM;
3298 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3299 	if (!sdkp)
3300 		goto out;
3301 
3302 	gd = alloc_disk(SD_MINORS);
3303 	if (!gd)
3304 		goto out_free;
3305 
3306 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3307 	if (index < 0) {
3308 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3309 		goto out_put;
3310 	}
3311 
3312 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3313 	if (error) {
3314 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3315 		goto out_free_index;
3316 	}
3317 
3318 	sdkp->device = sdp;
3319 	sdkp->driver = &sd_template;
3320 	sdkp->disk = gd;
3321 	sdkp->index = index;
3322 	atomic_set(&sdkp->openers, 0);
3323 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3324 
3325 	if (!sdp->request_queue->rq_timeout) {
3326 		if (sdp->type != TYPE_MOD)
3327 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3328 		else
3329 			blk_queue_rq_timeout(sdp->request_queue,
3330 					     SD_MOD_TIMEOUT);
3331 	}
3332 
3333 	device_initialize(&sdkp->dev);
3334 	sdkp->dev.parent = dev;
3335 	sdkp->dev.class = &sd_disk_class;
3336 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3337 
3338 	error = device_add(&sdkp->dev);
3339 	if (error)
3340 		goto out_free_index;
3341 
3342 	get_device(dev);
3343 	dev_set_drvdata(dev, sdkp);
3344 
3345 	gd->major = sd_major((index & 0xf0) >> 4);
3346 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3347 
3348 	gd->fops = &sd_fops;
3349 	gd->private_data = &sdkp->driver;
3350 	gd->queue = sdkp->device->request_queue;
3351 
3352 	/* defaults, until the device tells us otherwise */
3353 	sdp->sector_size = 512;
3354 	sdkp->capacity = 0;
3355 	sdkp->media_present = 1;
3356 	sdkp->write_prot = 0;
3357 	sdkp->cache_override = 0;
3358 	sdkp->WCE = 0;
3359 	sdkp->RCD = 0;
3360 	sdkp->ATO = 0;
3361 	sdkp->first_scan = 1;
3362 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3363 
3364 	sd_revalidate_disk(gd);
3365 
3366 	gd->flags = GENHD_FL_EXT_DEVT;
3367 	if (sdp->removable) {
3368 		gd->flags |= GENHD_FL_REMOVABLE;
3369 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3370 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3371 	}
3372 
3373 	blk_pm_runtime_init(sdp->request_queue, dev);
3374 	device_add_disk(dev, gd, NULL);
3375 	if (sdkp->capacity)
3376 		sd_dif_config_host(sdkp);
3377 
3378 	sd_revalidate_disk(gd);
3379 
3380 	if (sdkp->security) {
3381 		sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3382 		if (sdkp->opal_dev)
3383 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3384 	}
3385 
3386 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3387 		  sdp->removable ? "removable " : "");
3388 	scsi_autopm_put_device(sdp);
3389 
3390 	return 0;
3391 
3392  out_free_index:
3393 	ida_free(&sd_index_ida, index);
3394  out_put:
3395 	put_disk(gd);
3396  out_free:
3397 	kfree(sdkp);
3398  out:
3399 	scsi_autopm_put_device(sdp);
3400 	return error;
3401 }
3402 
3403 /**
3404  *	sd_remove - called whenever a scsi disk (previously recognized by
3405  *	sd_probe) is detached from the system. It is called (potentially
3406  *	multiple times) during sd module unload.
3407  *	@dev: pointer to device object
3408  *
3409  *	Note: this function is invoked from the scsi mid-level.
3410  *	This function potentially frees up a device name (e.g. /dev/sdc)
3411  *	that could be re-used by a subsequent sd_probe().
3412  *	This function is not called when the built-in sd driver is "exit-ed".
3413  **/
3414 static int sd_remove(struct device *dev)
3415 {
3416 	struct scsi_disk *sdkp;
3417 	dev_t devt;
3418 
3419 	sdkp = dev_get_drvdata(dev);
3420 	devt = disk_devt(sdkp->disk);
3421 	scsi_autopm_get_device(sdkp->device);
3422 
3423 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3424 	device_del(&sdkp->dev);
3425 	del_gendisk(sdkp->disk);
3426 	sd_shutdown(dev);
3427 
3428 	free_opal_dev(sdkp->opal_dev);
3429 
3430 	blk_register_region(devt, SD_MINORS, NULL,
3431 			    sd_default_probe, NULL, NULL);
3432 
3433 	mutex_lock(&sd_ref_mutex);
3434 	dev_set_drvdata(dev, NULL);
3435 	put_device(&sdkp->dev);
3436 	mutex_unlock(&sd_ref_mutex);
3437 
3438 	return 0;
3439 }
3440 
3441 /**
3442  *	scsi_disk_release - Called to free the scsi_disk structure
3443  *	@dev: pointer to embedded class device
3444  *
3445  *	sd_ref_mutex must be held entering this routine.  Because it is
3446  *	called on last put, you should always use the scsi_disk_get()
3447  *	scsi_disk_put() helpers which manipulate the semaphore directly
3448  *	and never do a direct put_device.
3449  **/
3450 static void scsi_disk_release(struct device *dev)
3451 {
3452 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3453 	struct gendisk *disk = sdkp->disk;
3454 	struct request_queue *q = disk->queue;
3455 
3456 	ida_free(&sd_index_ida, sdkp->index);
3457 
3458 	/*
3459 	 * Wait until all requests that are in progress have completed.
3460 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3461 	 * due to clearing the disk->private_data pointer. Wait from inside
3462 	 * scsi_disk_release() instead of from sd_release() to avoid that
3463 	 * freezing and unfreezing the request queue affects user space I/O
3464 	 * in case multiple processes open a /dev/sd... node concurrently.
3465 	 */
3466 	blk_mq_freeze_queue(q);
3467 	blk_mq_unfreeze_queue(q);
3468 
3469 	disk->private_data = NULL;
3470 	put_disk(disk);
3471 	put_device(&sdkp->device->sdev_gendev);
3472 
3473 	kfree(sdkp);
3474 }
3475 
3476 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3477 {
3478 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3479 	struct scsi_sense_hdr sshdr;
3480 	struct scsi_device *sdp = sdkp->device;
3481 	int res;
3482 
3483 	if (start)
3484 		cmd[4] |= 1;	/* START */
3485 
3486 	if (sdp->start_stop_pwr_cond)
3487 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3488 
3489 	if (!scsi_device_online(sdp))
3490 		return -ENODEV;
3491 
3492 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3493 			SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3494 	if (res) {
3495 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3496 		if (driver_byte(res) == DRIVER_SENSE)
3497 			sd_print_sense_hdr(sdkp, &sshdr);
3498 		if (scsi_sense_valid(&sshdr) &&
3499 			/* 0x3a is medium not present */
3500 			sshdr.asc == 0x3a)
3501 			res = 0;
3502 	}
3503 
3504 	/* SCSI error codes must not go to the generic layer */
3505 	if (res)
3506 		return -EIO;
3507 
3508 	return 0;
3509 }
3510 
3511 /*
3512  * Send a SYNCHRONIZE CACHE instruction down to the device through
3513  * the normal SCSI command structure.  Wait for the command to
3514  * complete.
3515  */
3516 static void sd_shutdown(struct device *dev)
3517 {
3518 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3519 
3520 	if (!sdkp)
3521 		return;         /* this can happen */
3522 
3523 	if (pm_runtime_suspended(dev))
3524 		return;
3525 
3526 	if (sdkp->WCE && sdkp->media_present) {
3527 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3528 		sd_sync_cache(sdkp, NULL);
3529 	}
3530 
3531 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3532 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3533 		sd_start_stop_device(sdkp, 0);
3534 	}
3535 }
3536 
3537 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3538 {
3539 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3540 	struct scsi_sense_hdr sshdr;
3541 	int ret = 0;
3542 
3543 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3544 		return 0;
3545 
3546 	if (sdkp->WCE && sdkp->media_present) {
3547 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3548 		ret = sd_sync_cache(sdkp, &sshdr);
3549 
3550 		if (ret) {
3551 			/* ignore OFFLINE device */
3552 			if (ret == -ENODEV)
3553 				return 0;
3554 
3555 			if (!scsi_sense_valid(&sshdr) ||
3556 			    sshdr.sense_key != ILLEGAL_REQUEST)
3557 				return ret;
3558 
3559 			/*
3560 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3561 			 * doesn't support sync. There's not much to do and
3562 			 * suspend shouldn't fail.
3563 			 */
3564 			ret = 0;
3565 		}
3566 	}
3567 
3568 	if (sdkp->device->manage_start_stop) {
3569 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3570 		/* an error is not worth aborting a system sleep */
3571 		ret = sd_start_stop_device(sdkp, 0);
3572 		if (ignore_stop_errors)
3573 			ret = 0;
3574 	}
3575 
3576 	return ret;
3577 }
3578 
3579 static int sd_suspend_system(struct device *dev)
3580 {
3581 	return sd_suspend_common(dev, true);
3582 }
3583 
3584 static int sd_suspend_runtime(struct device *dev)
3585 {
3586 	return sd_suspend_common(dev, false);
3587 }
3588 
3589 static int sd_resume(struct device *dev)
3590 {
3591 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3592 	int ret;
3593 
3594 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3595 		return 0;
3596 
3597 	if (!sdkp->device->manage_start_stop)
3598 		return 0;
3599 
3600 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3601 	ret = sd_start_stop_device(sdkp, 1);
3602 	if (!ret)
3603 		opal_unlock_from_suspend(sdkp->opal_dev);
3604 	return ret;
3605 }
3606 
3607 /**
3608  *	init_sd - entry point for this driver (both when built in or when
3609  *	a module).
3610  *
3611  *	Note: this function registers this driver with the scsi mid-level.
3612  **/
3613 static int __init init_sd(void)
3614 {
3615 	int majors = 0, i, err;
3616 
3617 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3618 
3619 	for (i = 0; i < SD_MAJORS; i++) {
3620 		if (register_blkdev(sd_major(i), "sd") != 0)
3621 			continue;
3622 		majors++;
3623 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3624 				    sd_default_probe, NULL, NULL);
3625 	}
3626 
3627 	if (!majors)
3628 		return -ENODEV;
3629 
3630 	err = class_register(&sd_disk_class);
3631 	if (err)
3632 		goto err_out;
3633 
3634 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3635 					 0, 0, NULL);
3636 	if (!sd_cdb_cache) {
3637 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3638 		err = -ENOMEM;
3639 		goto err_out_class;
3640 	}
3641 
3642 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3643 	if (!sd_cdb_pool) {
3644 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3645 		err = -ENOMEM;
3646 		goto err_out_cache;
3647 	}
3648 
3649 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3650 	if (!sd_page_pool) {
3651 		printk(KERN_ERR "sd: can't init discard page pool\n");
3652 		err = -ENOMEM;
3653 		goto err_out_ppool;
3654 	}
3655 
3656 	err = scsi_register_driver(&sd_template.gendrv);
3657 	if (err)
3658 		goto err_out_driver;
3659 
3660 	return 0;
3661 
3662 err_out_driver:
3663 	mempool_destroy(sd_page_pool);
3664 
3665 err_out_ppool:
3666 	mempool_destroy(sd_cdb_pool);
3667 
3668 err_out_cache:
3669 	kmem_cache_destroy(sd_cdb_cache);
3670 
3671 err_out_class:
3672 	class_unregister(&sd_disk_class);
3673 err_out:
3674 	for (i = 0; i < SD_MAJORS; i++)
3675 		unregister_blkdev(sd_major(i), "sd");
3676 	return err;
3677 }
3678 
3679 /**
3680  *	exit_sd - exit point for this driver (when it is a module).
3681  *
3682  *	Note: this function unregisters this driver from the scsi mid-level.
3683  **/
3684 static void __exit exit_sd(void)
3685 {
3686 	int i;
3687 
3688 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3689 
3690 	scsi_unregister_driver(&sd_template.gendrv);
3691 	mempool_destroy(sd_cdb_pool);
3692 	mempool_destroy(sd_page_pool);
3693 	kmem_cache_destroy(sd_cdb_cache);
3694 
3695 	class_unregister(&sd_disk_class);
3696 
3697 	for (i = 0; i < SD_MAJORS; i++) {
3698 		blk_unregister_region(sd_major(i), SD_MINORS);
3699 		unregister_blkdev(sd_major(i), "sd");
3700 	}
3701 }
3702 
3703 module_init(init_sd);
3704 module_exit(exit_sd);
3705 
3706 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3707 			       struct scsi_sense_hdr *sshdr)
3708 {
3709 	scsi_print_sense_hdr(sdkp->device,
3710 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3711 }
3712 
3713 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3714 			    int result)
3715 {
3716 	const char *hb_string = scsi_hostbyte_string(result);
3717 	const char *db_string = scsi_driverbyte_string(result);
3718 
3719 	if (hb_string || db_string)
3720 		sd_printk(KERN_INFO, sdkp,
3721 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3722 			  hb_string ? hb_string : "invalid",
3723 			  db_string ? db_string : "invalid");
3724 	else
3725 		sd_printk(KERN_INFO, sdkp,
3726 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3727 			  msg, host_byte(result), driver_byte(result));
3728 }
3729 
3730