xref: /openbmc/linux/drivers/scsi/sd.c (revision 22d55f02)
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 		.remove		= sd_remove,
572 		.shutdown	= sd_shutdown,
573 		.pm		= &sd_pm_ops,
574 	},
575 	.rescan			= sd_rescan,
576 	.init_command		= sd_init_command,
577 	.uninit_command		= sd_uninit_command,
578 	.done			= sd_done,
579 	.eh_action		= sd_eh_action,
580 	.eh_reset		= sd_eh_reset,
581 };
582 
583 /*
584  * Dummy kobj_map->probe function.
585  * The default ->probe function will call modprobe, which is
586  * pointless as this module is already loaded.
587  */
588 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
589 {
590 	return NULL;
591 }
592 
593 /*
594  * Device no to disk mapping:
595  *
596  *       major         disc2     disc  p1
597  *   |............|.............|....|....| <- dev_t
598  *    31        20 19          8 7  4 3  0
599  *
600  * Inside a major, we have 16k disks, however mapped non-
601  * contiguously. The first 16 disks are for major0, the next
602  * ones with major1, ... Disk 256 is for major0 again, disk 272
603  * for major1, ...
604  * As we stay compatible with our numbering scheme, we can reuse
605  * the well-know SCSI majors 8, 65--71, 136--143.
606  */
607 static int sd_major(int major_idx)
608 {
609 	switch (major_idx) {
610 	case 0:
611 		return SCSI_DISK0_MAJOR;
612 	case 1 ... 7:
613 		return SCSI_DISK1_MAJOR + major_idx - 1;
614 	case 8 ... 15:
615 		return SCSI_DISK8_MAJOR + major_idx - 8;
616 	default:
617 		BUG();
618 		return 0;	/* shut up gcc */
619 	}
620 }
621 
622 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
623 {
624 	struct scsi_disk *sdkp = NULL;
625 
626 	mutex_lock(&sd_ref_mutex);
627 
628 	if (disk->private_data) {
629 		sdkp = scsi_disk(disk);
630 		if (scsi_device_get(sdkp->device) == 0)
631 			get_device(&sdkp->dev);
632 		else
633 			sdkp = NULL;
634 	}
635 	mutex_unlock(&sd_ref_mutex);
636 	return sdkp;
637 }
638 
639 static void scsi_disk_put(struct scsi_disk *sdkp)
640 {
641 	struct scsi_device *sdev = sdkp->device;
642 
643 	mutex_lock(&sd_ref_mutex);
644 	put_device(&sdkp->dev);
645 	scsi_device_put(sdev);
646 	mutex_unlock(&sd_ref_mutex);
647 }
648 
649 #ifdef CONFIG_BLK_SED_OPAL
650 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
651 		size_t len, bool send)
652 {
653 	struct scsi_device *sdev = data;
654 	u8 cdb[12] = { 0, };
655 	int ret;
656 
657 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
658 	cdb[1] = secp;
659 	put_unaligned_be16(spsp, &cdb[2]);
660 	put_unaligned_be32(len, &cdb[6]);
661 
662 	ret = scsi_execute_req(sdev, cdb,
663 			send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
664 			buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
665 	return ret <= 0 ? ret : -EIO;
666 }
667 #endif /* CONFIG_BLK_SED_OPAL */
668 
669 /*
670  * Look up the DIX operation based on whether the command is read or
671  * write and whether dix and dif are enabled.
672  */
673 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
674 {
675 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
676 	static const unsigned int ops[] = {	/* wrt dix dif */
677 		SCSI_PROT_NORMAL,		/*  0	0   0  */
678 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
679 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
680 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
681 		SCSI_PROT_NORMAL,		/*  1	0   0  */
682 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
683 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
684 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
685 	};
686 
687 	return ops[write << 2 | dix << 1 | dif];
688 }
689 
690 /*
691  * Returns a mask of the protection flags that are valid for a given DIX
692  * operation.
693  */
694 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
695 {
696 	static const unsigned int flag_mask[] = {
697 		[SCSI_PROT_NORMAL]		= 0,
698 
699 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
700 						  SCSI_PROT_GUARD_CHECK |
701 						  SCSI_PROT_REF_CHECK |
702 						  SCSI_PROT_REF_INCREMENT,
703 
704 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
705 						  SCSI_PROT_IP_CHECKSUM,
706 
707 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
708 						  SCSI_PROT_GUARD_CHECK |
709 						  SCSI_PROT_REF_CHECK |
710 						  SCSI_PROT_REF_INCREMENT |
711 						  SCSI_PROT_IP_CHECKSUM,
712 
713 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
714 						  SCSI_PROT_REF_INCREMENT,
715 
716 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
717 						  SCSI_PROT_REF_CHECK |
718 						  SCSI_PROT_REF_INCREMENT |
719 						  SCSI_PROT_IP_CHECKSUM,
720 
721 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
722 						  SCSI_PROT_GUARD_CHECK |
723 						  SCSI_PROT_REF_CHECK |
724 						  SCSI_PROT_REF_INCREMENT |
725 						  SCSI_PROT_IP_CHECKSUM,
726 	};
727 
728 	return flag_mask[prot_op];
729 }
730 
731 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
732 					   unsigned int dix, unsigned int dif)
733 {
734 	struct bio *bio = scmd->request->bio;
735 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
736 	unsigned int protect = 0;
737 
738 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
739 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
740 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
741 
742 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
743 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
744 	}
745 
746 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
747 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
748 
749 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
750 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
751 	}
752 
753 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
754 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
755 
756 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
757 			protect = 3 << 5;	/* Disable target PI checking */
758 		else
759 			protect = 1 << 5;	/* Enable target PI checking */
760 	}
761 
762 	scsi_set_prot_op(scmd, prot_op);
763 	scsi_set_prot_type(scmd, dif);
764 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
765 
766 	return protect;
767 }
768 
769 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
770 {
771 	struct request_queue *q = sdkp->disk->queue;
772 	unsigned int logical_block_size = sdkp->device->sector_size;
773 	unsigned int max_blocks = 0;
774 
775 	q->limits.discard_alignment =
776 		sdkp->unmap_alignment * logical_block_size;
777 	q->limits.discard_granularity =
778 		max(sdkp->physical_block_size,
779 		    sdkp->unmap_granularity * logical_block_size);
780 	sdkp->provisioning_mode = mode;
781 
782 	switch (mode) {
783 
784 	case SD_LBP_FULL:
785 	case SD_LBP_DISABLE:
786 		blk_queue_max_discard_sectors(q, 0);
787 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
788 		return;
789 
790 	case SD_LBP_UNMAP:
791 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
792 					  (u32)SD_MAX_WS16_BLOCKS);
793 		break;
794 
795 	case SD_LBP_WS16:
796 		if (sdkp->device->unmap_limit_for_ws)
797 			max_blocks = sdkp->max_unmap_blocks;
798 		else
799 			max_blocks = sdkp->max_ws_blocks;
800 
801 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
802 		break;
803 
804 	case SD_LBP_WS10:
805 		if (sdkp->device->unmap_limit_for_ws)
806 			max_blocks = sdkp->max_unmap_blocks;
807 		else
808 			max_blocks = sdkp->max_ws_blocks;
809 
810 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
811 		break;
812 
813 	case SD_LBP_ZERO:
814 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
815 					  (u32)SD_MAX_WS10_BLOCKS);
816 		break;
817 	}
818 
819 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
820 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
821 }
822 
823 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
824 {
825 	struct scsi_device *sdp = cmd->device;
826 	struct request *rq = cmd->request;
827 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
828 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
829 	unsigned int data_len = 24;
830 	char *buf;
831 
832 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
833 	if (!rq->special_vec.bv_page)
834 		return BLK_STS_RESOURCE;
835 	clear_highpage(rq->special_vec.bv_page);
836 	rq->special_vec.bv_offset = 0;
837 	rq->special_vec.bv_len = data_len;
838 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
839 
840 	cmd->cmd_len = 10;
841 	cmd->cmnd[0] = UNMAP;
842 	cmd->cmnd[8] = 24;
843 
844 	buf = page_address(rq->special_vec.bv_page);
845 	put_unaligned_be16(6 + 16, &buf[0]);
846 	put_unaligned_be16(16, &buf[2]);
847 	put_unaligned_be64(lba, &buf[8]);
848 	put_unaligned_be32(nr_blocks, &buf[16]);
849 
850 	cmd->allowed = SD_MAX_RETRIES;
851 	cmd->transfersize = data_len;
852 	rq->timeout = SD_TIMEOUT;
853 
854 	return scsi_init_io(cmd);
855 }
856 
857 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
858 		bool unmap)
859 {
860 	struct scsi_device *sdp = cmd->device;
861 	struct request *rq = cmd->request;
862 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
863 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
864 	u32 data_len = sdp->sector_size;
865 
866 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
867 	if (!rq->special_vec.bv_page)
868 		return BLK_STS_RESOURCE;
869 	clear_highpage(rq->special_vec.bv_page);
870 	rq->special_vec.bv_offset = 0;
871 	rq->special_vec.bv_len = data_len;
872 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
873 
874 	cmd->cmd_len = 16;
875 	cmd->cmnd[0] = WRITE_SAME_16;
876 	if (unmap)
877 		cmd->cmnd[1] = 0x8; /* UNMAP */
878 	put_unaligned_be64(lba, &cmd->cmnd[2]);
879 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
880 
881 	cmd->allowed = SD_MAX_RETRIES;
882 	cmd->transfersize = data_len;
883 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
884 
885 	return scsi_init_io(cmd);
886 }
887 
888 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
889 		bool unmap)
890 {
891 	struct scsi_device *sdp = cmd->device;
892 	struct request *rq = cmd->request;
893 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
894 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
895 	u32 data_len = sdp->sector_size;
896 
897 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
898 	if (!rq->special_vec.bv_page)
899 		return BLK_STS_RESOURCE;
900 	clear_highpage(rq->special_vec.bv_page);
901 	rq->special_vec.bv_offset = 0;
902 	rq->special_vec.bv_len = data_len;
903 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
904 
905 	cmd->cmd_len = 10;
906 	cmd->cmnd[0] = WRITE_SAME;
907 	if (unmap)
908 		cmd->cmnd[1] = 0x8; /* UNMAP */
909 	put_unaligned_be32(lba, &cmd->cmnd[2]);
910 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
911 
912 	cmd->allowed = SD_MAX_RETRIES;
913 	cmd->transfersize = data_len;
914 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
915 
916 	return scsi_init_io(cmd);
917 }
918 
919 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
920 {
921 	struct request *rq = cmd->request;
922 	struct scsi_device *sdp = cmd->device;
923 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
924 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
925 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
926 
927 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
928 		switch (sdkp->zeroing_mode) {
929 		case SD_ZERO_WS16_UNMAP:
930 			return sd_setup_write_same16_cmnd(cmd, true);
931 		case SD_ZERO_WS10_UNMAP:
932 			return sd_setup_write_same10_cmnd(cmd, true);
933 		}
934 	}
935 
936 	if (sdp->no_write_same)
937 		return BLK_STS_TARGET;
938 
939 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
940 		return sd_setup_write_same16_cmnd(cmd, false);
941 
942 	return sd_setup_write_same10_cmnd(cmd, false);
943 }
944 
945 static void sd_config_write_same(struct scsi_disk *sdkp)
946 {
947 	struct request_queue *q = sdkp->disk->queue;
948 	unsigned int logical_block_size = sdkp->device->sector_size;
949 
950 	if (sdkp->device->no_write_same) {
951 		sdkp->max_ws_blocks = 0;
952 		goto out;
953 	}
954 
955 	/* Some devices can not handle block counts above 0xffff despite
956 	 * supporting WRITE SAME(16). Consequently we default to 64k
957 	 * blocks per I/O unless the device explicitly advertises a
958 	 * bigger limit.
959 	 */
960 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
961 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
962 						   (u32)SD_MAX_WS16_BLOCKS);
963 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
964 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
965 						   (u32)SD_MAX_WS10_BLOCKS);
966 	else {
967 		sdkp->device->no_write_same = 1;
968 		sdkp->max_ws_blocks = 0;
969 	}
970 
971 	if (sdkp->lbprz && sdkp->lbpws)
972 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
973 	else if (sdkp->lbprz && sdkp->lbpws10)
974 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
975 	else if (sdkp->max_ws_blocks)
976 		sdkp->zeroing_mode = SD_ZERO_WS;
977 	else
978 		sdkp->zeroing_mode = SD_ZERO_WRITE;
979 
980 	if (sdkp->max_ws_blocks &&
981 	    sdkp->physical_block_size > logical_block_size) {
982 		/*
983 		 * Reporting a maximum number of blocks that is not aligned
984 		 * on the device physical size would cause a large write same
985 		 * request to be split into physically unaligned chunks by
986 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
987 		 * even if the caller of these functions took care to align the
988 		 * large request. So make sure the maximum reported is aligned
989 		 * to the device physical block size. This is only an optional
990 		 * optimization for regular disks, but this is mandatory to
991 		 * avoid failure of large write same requests directed at
992 		 * sequential write required zones of host-managed ZBC disks.
993 		 */
994 		sdkp->max_ws_blocks =
995 			round_down(sdkp->max_ws_blocks,
996 				   bytes_to_logical(sdkp->device,
997 						    sdkp->physical_block_size));
998 	}
999 
1000 out:
1001 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1002 					 (logical_block_size >> 9));
1003 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1004 					 (logical_block_size >> 9));
1005 }
1006 
1007 /**
1008  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1009  * @cmd: command to prepare
1010  *
1011  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1012  * the preference indicated by the target device.
1013  **/
1014 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1015 {
1016 	struct request *rq = cmd->request;
1017 	struct scsi_device *sdp = cmd->device;
1018 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1019 	struct bio *bio = rq->bio;
1020 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1021 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1022 	blk_status_t ret;
1023 
1024 	if (sdkp->device->no_write_same)
1025 		return BLK_STS_TARGET;
1026 
1027 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1028 
1029 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1030 
1031 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1032 		cmd->cmd_len = 16;
1033 		cmd->cmnd[0] = WRITE_SAME_16;
1034 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1035 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1036 	} else {
1037 		cmd->cmd_len = 10;
1038 		cmd->cmnd[0] = WRITE_SAME;
1039 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1040 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1041 	}
1042 
1043 	cmd->transfersize = sdp->sector_size;
1044 	cmd->allowed = SD_MAX_RETRIES;
1045 
1046 	/*
1047 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1048 	 * different from the amount of data actually written to the target.
1049 	 *
1050 	 * We set up __data_len to the amount of data transferred via the
1051 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1052 	 * to transfer a single sector of data first, but then reset it to
1053 	 * the amount of data to be written right after so that the I/O path
1054 	 * knows how much to actually write.
1055 	 */
1056 	rq->__data_len = sdp->sector_size;
1057 	ret = scsi_init_io(cmd);
1058 	rq->__data_len = blk_rq_bytes(rq);
1059 
1060 	return ret;
1061 }
1062 
1063 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1064 {
1065 	struct request *rq = cmd->request;
1066 
1067 	/* flush requests don't perform I/O, zero the S/G table */
1068 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1069 
1070 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1071 	cmd->cmd_len = 10;
1072 	cmd->transfersize = 0;
1073 	cmd->allowed = SD_MAX_RETRIES;
1074 
1075 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1076 	return BLK_STS_OK;
1077 }
1078 
1079 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1080 				       sector_t lba, unsigned int nr_blocks,
1081 				       unsigned char flags)
1082 {
1083 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1084 	if (unlikely(cmd->cmnd == NULL))
1085 		return BLK_STS_RESOURCE;
1086 
1087 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1088 	memset(cmd->cmnd, 0, cmd->cmd_len);
1089 
1090 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1091 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1092 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1093 	cmd->cmnd[10] = flags;
1094 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1095 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1096 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1097 
1098 	return BLK_STS_OK;
1099 }
1100 
1101 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1102 				       sector_t lba, unsigned int nr_blocks,
1103 				       unsigned char flags)
1104 {
1105 	cmd->cmd_len  = 16;
1106 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1107 	cmd->cmnd[1]  = flags;
1108 	cmd->cmnd[14] = 0;
1109 	cmd->cmnd[15] = 0;
1110 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1111 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1112 
1113 	return BLK_STS_OK;
1114 }
1115 
1116 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1117 				       sector_t lba, unsigned int nr_blocks,
1118 				       unsigned char flags)
1119 {
1120 	cmd->cmd_len = 10;
1121 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1122 	cmd->cmnd[1] = flags;
1123 	cmd->cmnd[6] = 0;
1124 	cmd->cmnd[9] = 0;
1125 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1126 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1127 
1128 	return BLK_STS_OK;
1129 }
1130 
1131 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1132 				      sector_t lba, unsigned int nr_blocks,
1133 				      unsigned char flags)
1134 {
1135 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1136 	if (WARN_ON_ONCE(nr_blocks == 0))
1137 		return BLK_STS_IOERR;
1138 
1139 	if (unlikely(flags & 0x8)) {
1140 		/*
1141 		 * This happens only if this drive failed 10byte rw
1142 		 * command with ILLEGAL_REQUEST during operation and
1143 		 * thus turned off use_10_for_rw.
1144 		 */
1145 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1146 		return BLK_STS_IOERR;
1147 	}
1148 
1149 	cmd->cmd_len = 6;
1150 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1151 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1152 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1153 	cmd->cmnd[3] = lba & 0xff;
1154 	cmd->cmnd[4] = nr_blocks;
1155 	cmd->cmnd[5] = 0;
1156 
1157 	return BLK_STS_OK;
1158 }
1159 
1160 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1161 {
1162 	struct request *rq = cmd->request;
1163 	struct scsi_device *sdp = cmd->device;
1164 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1165 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1166 	sector_t threshold;
1167 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1168 	bool dif, dix;
1169 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170 	bool write = rq_data_dir(rq) == WRITE;
1171 	unsigned char protect, fua;
1172 	blk_status_t ret;
1173 
1174 	ret = scsi_init_io(cmd);
1175 	if (ret != BLK_STS_OK)
1176 		return ret;
1177 
1178 	if (!scsi_device_online(sdp) || sdp->changed) {
1179 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1180 		return BLK_STS_IOERR;
1181 	}
1182 
1183 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1184 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1185 		return BLK_STS_IOERR;
1186 	}
1187 
1188 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1189 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1190 		return BLK_STS_IOERR;
1191 	}
1192 
1193 	/*
1194 	 * Some SD card readers can't handle accesses which touch the
1195 	 * last one or two logical blocks. Split accesses as needed.
1196 	 */
1197 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1198 
1199 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1200 		if (lba < threshold) {
1201 			/* Access up to the threshold but not beyond */
1202 			nr_blocks = threshold - lba;
1203 		} else {
1204 			/* Access only a single logical block */
1205 			nr_blocks = 1;
1206 		}
1207 	}
1208 
1209 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1210 	dix = scsi_prot_sg_count(cmd);
1211 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1212 
1213 	if (write && dix)
1214 		t10_pi_prepare(cmd->request, sdkp->protection_type);
1215 
1216 	if (dif || dix)
1217 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1218 	else
1219 		protect = 0;
1220 
1221 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1222 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1223 					 protect | fua);
1224 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1225 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1226 					 protect | fua);
1227 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1228 		   sdp->use_10_for_rw || protect) {
1229 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1230 					 protect | fua);
1231 	} else {
1232 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1233 					protect | fua);
1234 	}
1235 
1236 	if (unlikely(ret != BLK_STS_OK))
1237 		return ret;
1238 
1239 	/*
1240 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1241 	 * host adapter, it's safe to assume that we can at least transfer
1242 	 * this many bytes between each connect / disconnect.
1243 	 */
1244 	cmd->transfersize = sdp->sector_size;
1245 	cmd->underflow = nr_blocks << 9;
1246 	cmd->allowed = SD_MAX_RETRIES;
1247 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1248 
1249 	SCSI_LOG_HLQUEUE(1,
1250 			 scmd_printk(KERN_INFO, cmd,
1251 				     "%s: block=%llu, count=%d\n", __func__,
1252 				     (unsigned long long)blk_rq_pos(rq),
1253 				     blk_rq_sectors(rq)));
1254 	SCSI_LOG_HLQUEUE(2,
1255 			 scmd_printk(KERN_INFO, cmd,
1256 				     "%s %d/%u 512 byte blocks.\n",
1257 				     write ? "writing" : "reading", nr_blocks,
1258 				     blk_rq_sectors(rq)));
1259 
1260 	/*
1261 	 * This indicates that the command is ready from our end to be
1262 	 * queued.
1263 	 */
1264 	return BLK_STS_OK;
1265 }
1266 
1267 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1268 {
1269 	struct request *rq = cmd->request;
1270 
1271 	switch (req_op(rq)) {
1272 	case REQ_OP_DISCARD:
1273 		switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1274 		case SD_LBP_UNMAP:
1275 			return sd_setup_unmap_cmnd(cmd);
1276 		case SD_LBP_WS16:
1277 			return sd_setup_write_same16_cmnd(cmd, true);
1278 		case SD_LBP_WS10:
1279 			return sd_setup_write_same10_cmnd(cmd, true);
1280 		case SD_LBP_ZERO:
1281 			return sd_setup_write_same10_cmnd(cmd, false);
1282 		default:
1283 			return BLK_STS_TARGET;
1284 		}
1285 	case REQ_OP_WRITE_ZEROES:
1286 		return sd_setup_write_zeroes_cmnd(cmd);
1287 	case REQ_OP_WRITE_SAME:
1288 		return sd_setup_write_same_cmnd(cmd);
1289 	case REQ_OP_FLUSH:
1290 		return sd_setup_flush_cmnd(cmd);
1291 	case REQ_OP_READ:
1292 	case REQ_OP_WRITE:
1293 		return sd_setup_read_write_cmnd(cmd);
1294 	case REQ_OP_ZONE_RESET:
1295 		return sd_zbc_setup_reset_cmnd(cmd);
1296 	default:
1297 		WARN_ON_ONCE(1);
1298 		return BLK_STS_NOTSUPP;
1299 	}
1300 }
1301 
1302 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1303 {
1304 	struct request *rq = SCpnt->request;
1305 	u8 *cmnd;
1306 
1307 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1308 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1309 
1310 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1311 		cmnd = SCpnt->cmnd;
1312 		SCpnt->cmnd = NULL;
1313 		SCpnt->cmd_len = 0;
1314 		mempool_free(cmnd, sd_cdb_pool);
1315 	}
1316 }
1317 
1318 /**
1319  *	sd_open - open a scsi disk device
1320  *	@bdev: Block device of the scsi disk to open
1321  *	@mode: FMODE_* mask
1322  *
1323  *	Returns 0 if successful. Returns a negated errno value in case
1324  *	of error.
1325  *
1326  *	Note: This can be called from a user context (e.g. fsck(1) )
1327  *	or from within the kernel (e.g. as a result of a mount(1) ).
1328  *	In the latter case @inode and @filp carry an abridged amount
1329  *	of information as noted above.
1330  *
1331  *	Locking: called with bdev->bd_mutex held.
1332  **/
1333 static int sd_open(struct block_device *bdev, fmode_t mode)
1334 {
1335 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1336 	struct scsi_device *sdev;
1337 	int retval;
1338 
1339 	if (!sdkp)
1340 		return -ENXIO;
1341 
1342 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1343 
1344 	sdev = sdkp->device;
1345 
1346 	/*
1347 	 * If the device is in error recovery, wait until it is done.
1348 	 * If the device is offline, then disallow any access to it.
1349 	 */
1350 	retval = -ENXIO;
1351 	if (!scsi_block_when_processing_errors(sdev))
1352 		goto error_out;
1353 
1354 	if (sdev->removable || sdkp->write_prot)
1355 		check_disk_change(bdev);
1356 
1357 	/*
1358 	 * If the drive is empty, just let the open fail.
1359 	 */
1360 	retval = -ENOMEDIUM;
1361 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1362 		goto error_out;
1363 
1364 	/*
1365 	 * If the device has the write protect tab set, have the open fail
1366 	 * if the user expects to be able to write to the thing.
1367 	 */
1368 	retval = -EROFS;
1369 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1370 		goto error_out;
1371 
1372 	/*
1373 	 * It is possible that the disk changing stuff resulted in
1374 	 * the device being taken offline.  If this is the case,
1375 	 * report this to the user, and don't pretend that the
1376 	 * open actually succeeded.
1377 	 */
1378 	retval = -ENXIO;
1379 	if (!scsi_device_online(sdev))
1380 		goto error_out;
1381 
1382 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1383 		if (scsi_block_when_processing_errors(sdev))
1384 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1385 	}
1386 
1387 	return 0;
1388 
1389 error_out:
1390 	scsi_disk_put(sdkp);
1391 	return retval;
1392 }
1393 
1394 /**
1395  *	sd_release - invoked when the (last) close(2) is called on this
1396  *	scsi disk.
1397  *	@disk: disk to release
1398  *	@mode: FMODE_* mask
1399  *
1400  *	Returns 0.
1401  *
1402  *	Note: may block (uninterruptible) if error recovery is underway
1403  *	on this disk.
1404  *
1405  *	Locking: called with bdev->bd_mutex held.
1406  **/
1407 static void sd_release(struct gendisk *disk, fmode_t mode)
1408 {
1409 	struct scsi_disk *sdkp = scsi_disk(disk);
1410 	struct scsi_device *sdev = sdkp->device;
1411 
1412 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1413 
1414 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1415 		if (scsi_block_when_processing_errors(sdev))
1416 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1417 	}
1418 
1419 	scsi_disk_put(sdkp);
1420 }
1421 
1422 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1423 {
1424 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1425 	struct scsi_device *sdp = sdkp->device;
1426 	struct Scsi_Host *host = sdp->host;
1427 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1428 	int diskinfo[4];
1429 
1430 	/* default to most commonly used values */
1431 	diskinfo[0] = 0x40;	/* 1 << 6 */
1432 	diskinfo[1] = 0x20;	/* 1 << 5 */
1433 	diskinfo[2] = capacity >> 11;
1434 
1435 	/* override with calculated, extended default, or driver values */
1436 	if (host->hostt->bios_param)
1437 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1438 	else
1439 		scsicam_bios_param(bdev, capacity, diskinfo);
1440 
1441 	geo->heads = diskinfo[0];
1442 	geo->sectors = diskinfo[1];
1443 	geo->cylinders = diskinfo[2];
1444 	return 0;
1445 }
1446 
1447 /**
1448  *	sd_ioctl - process an ioctl
1449  *	@bdev: target block device
1450  *	@mode: FMODE_* mask
1451  *	@cmd: ioctl command number
1452  *	@arg: this is third argument given to ioctl(2) system call.
1453  *	Often contains a pointer.
1454  *
1455  *	Returns 0 if successful (some ioctls return positive numbers on
1456  *	success as well). Returns a negated errno value in case of error.
1457  *
1458  *	Note: most ioctls are forward onto the block subsystem or further
1459  *	down in the scsi subsystem.
1460  **/
1461 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1462 		    unsigned int cmd, unsigned long arg)
1463 {
1464 	struct gendisk *disk = bdev->bd_disk;
1465 	struct scsi_disk *sdkp = scsi_disk(disk);
1466 	struct scsi_device *sdp = sdkp->device;
1467 	void __user *p = (void __user *)arg;
1468 	int error;
1469 
1470 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1471 				    "cmd=0x%x\n", disk->disk_name, cmd));
1472 
1473 	error = scsi_verify_blk_ioctl(bdev, cmd);
1474 	if (error < 0)
1475 		return error;
1476 
1477 	/*
1478 	 * If we are in the middle of error recovery, don't let anyone
1479 	 * else try and use this device.  Also, if error recovery fails, it
1480 	 * may try and take the device offline, in which case all further
1481 	 * access to the device is prohibited.
1482 	 */
1483 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1484 			(mode & FMODE_NDELAY) != 0);
1485 	if (error)
1486 		goto out;
1487 
1488 	if (is_sed_ioctl(cmd))
1489 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1490 
1491 	/*
1492 	 * Send SCSI addressing ioctls directly to mid level, send other
1493 	 * ioctls to block level and then onto mid level if they can't be
1494 	 * resolved.
1495 	 */
1496 	switch (cmd) {
1497 		case SCSI_IOCTL_GET_IDLUN:
1498 		case SCSI_IOCTL_GET_BUS_NUMBER:
1499 			error = scsi_ioctl(sdp, cmd, p);
1500 			break;
1501 		default:
1502 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1503 			if (error != -ENOTTY)
1504 				break;
1505 			error = scsi_ioctl(sdp, cmd, p);
1506 			break;
1507 	}
1508 out:
1509 	return error;
1510 }
1511 
1512 static void set_media_not_present(struct scsi_disk *sdkp)
1513 {
1514 	if (sdkp->media_present)
1515 		sdkp->device->changed = 1;
1516 
1517 	if (sdkp->device->removable) {
1518 		sdkp->media_present = 0;
1519 		sdkp->capacity = 0;
1520 	}
1521 }
1522 
1523 static int media_not_present(struct scsi_disk *sdkp,
1524 			     struct scsi_sense_hdr *sshdr)
1525 {
1526 	if (!scsi_sense_valid(sshdr))
1527 		return 0;
1528 
1529 	/* not invoked for commands that could return deferred errors */
1530 	switch (sshdr->sense_key) {
1531 	case UNIT_ATTENTION:
1532 	case NOT_READY:
1533 		/* medium not present */
1534 		if (sshdr->asc == 0x3A) {
1535 			set_media_not_present(sdkp);
1536 			return 1;
1537 		}
1538 	}
1539 	return 0;
1540 }
1541 
1542 /**
1543  *	sd_check_events - check media events
1544  *	@disk: kernel device descriptor
1545  *	@clearing: disk events currently being cleared
1546  *
1547  *	Returns mask of DISK_EVENT_*.
1548  *
1549  *	Note: this function is invoked from the block subsystem.
1550  **/
1551 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1552 {
1553 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1554 	struct scsi_device *sdp;
1555 	int retval;
1556 
1557 	if (!sdkp)
1558 		return 0;
1559 
1560 	sdp = sdkp->device;
1561 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1562 
1563 	/*
1564 	 * If the device is offline, don't send any commands - just pretend as
1565 	 * if the command failed.  If the device ever comes back online, we
1566 	 * can deal with it then.  It is only because of unrecoverable errors
1567 	 * that we would ever take a device offline in the first place.
1568 	 */
1569 	if (!scsi_device_online(sdp)) {
1570 		set_media_not_present(sdkp);
1571 		goto out;
1572 	}
1573 
1574 	/*
1575 	 * Using TEST_UNIT_READY enables differentiation between drive with
1576 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1577 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1578 	 *
1579 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1580 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1581 	 * sd_revalidate() is called.
1582 	 */
1583 	if (scsi_block_when_processing_errors(sdp)) {
1584 		struct scsi_sense_hdr sshdr = { 0, };
1585 
1586 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1587 					      &sshdr);
1588 
1589 		/* failed to execute TUR, assume media not present */
1590 		if (host_byte(retval)) {
1591 			set_media_not_present(sdkp);
1592 			goto out;
1593 		}
1594 
1595 		if (media_not_present(sdkp, &sshdr))
1596 			goto out;
1597 	}
1598 
1599 	/*
1600 	 * For removable scsi disk we have to recognise the presence
1601 	 * of a disk in the drive.
1602 	 */
1603 	if (!sdkp->media_present)
1604 		sdp->changed = 1;
1605 	sdkp->media_present = 1;
1606 out:
1607 	/*
1608 	 * sdp->changed is set under the following conditions:
1609 	 *
1610 	 *	Medium present state has changed in either direction.
1611 	 *	Device has indicated UNIT_ATTENTION.
1612 	 */
1613 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1614 	sdp->changed = 0;
1615 	scsi_disk_put(sdkp);
1616 	return retval;
1617 }
1618 
1619 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1620 {
1621 	int retries, res;
1622 	struct scsi_device *sdp = sdkp->device;
1623 	const int timeout = sdp->request_queue->rq_timeout
1624 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1625 	struct scsi_sense_hdr my_sshdr;
1626 
1627 	if (!scsi_device_online(sdp))
1628 		return -ENODEV;
1629 
1630 	/* caller might not be interested in sense, but we need it */
1631 	if (!sshdr)
1632 		sshdr = &my_sshdr;
1633 
1634 	for (retries = 3; retries > 0; --retries) {
1635 		unsigned char cmd[10] = { 0 };
1636 
1637 		cmd[0] = SYNCHRONIZE_CACHE;
1638 		/*
1639 		 * Leave the rest of the command zero to indicate
1640 		 * flush everything.
1641 		 */
1642 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1643 				timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1644 		if (res == 0)
1645 			break;
1646 	}
1647 
1648 	if (res) {
1649 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1650 
1651 		if (driver_byte(res) == DRIVER_SENSE)
1652 			sd_print_sense_hdr(sdkp, sshdr);
1653 
1654 		/* we need to evaluate the error return  */
1655 		if (scsi_sense_valid(sshdr) &&
1656 			(sshdr->asc == 0x3a ||	/* medium not present */
1657 			 sshdr->asc == 0x20))	/* invalid command */
1658 				/* this is no error here */
1659 				return 0;
1660 
1661 		switch (host_byte(res)) {
1662 		/* ignore errors due to racing a disconnection */
1663 		case DID_BAD_TARGET:
1664 		case DID_NO_CONNECT:
1665 			return 0;
1666 		/* signal the upper layer it might try again */
1667 		case DID_BUS_BUSY:
1668 		case DID_IMM_RETRY:
1669 		case DID_REQUEUE:
1670 		case DID_SOFT_ERROR:
1671 			return -EBUSY;
1672 		default:
1673 			return -EIO;
1674 		}
1675 	}
1676 	return 0;
1677 }
1678 
1679 static void sd_rescan(struct device *dev)
1680 {
1681 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1682 
1683 	revalidate_disk(sdkp->disk);
1684 }
1685 
1686 
1687 #ifdef CONFIG_COMPAT
1688 /*
1689  * This gets directly called from VFS. When the ioctl
1690  * is not recognized we go back to the other translation paths.
1691  */
1692 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1693 			   unsigned int cmd, unsigned long arg)
1694 {
1695 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1696 	int error;
1697 
1698 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1699 			(mode & FMODE_NDELAY) != 0);
1700 	if (error)
1701 		return error;
1702 
1703 	/*
1704 	 * Let the static ioctl translation table take care of it.
1705 	 */
1706 	if (!sdev->host->hostt->compat_ioctl)
1707 		return -ENOIOCTLCMD;
1708 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1709 }
1710 #endif
1711 
1712 static char sd_pr_type(enum pr_type type)
1713 {
1714 	switch (type) {
1715 	case PR_WRITE_EXCLUSIVE:
1716 		return 0x01;
1717 	case PR_EXCLUSIVE_ACCESS:
1718 		return 0x03;
1719 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1720 		return 0x05;
1721 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1722 		return 0x06;
1723 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1724 		return 0x07;
1725 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1726 		return 0x08;
1727 	default:
1728 		return 0;
1729 	}
1730 };
1731 
1732 static int sd_pr_command(struct block_device *bdev, u8 sa,
1733 		u64 key, u64 sa_key, u8 type, u8 flags)
1734 {
1735 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1736 	struct scsi_sense_hdr sshdr;
1737 	int result;
1738 	u8 cmd[16] = { 0, };
1739 	u8 data[24] = { 0, };
1740 
1741 	cmd[0] = PERSISTENT_RESERVE_OUT;
1742 	cmd[1] = sa;
1743 	cmd[2] = type;
1744 	put_unaligned_be32(sizeof(data), &cmd[5]);
1745 
1746 	put_unaligned_be64(key, &data[0]);
1747 	put_unaligned_be64(sa_key, &data[8]);
1748 	data[20] = flags;
1749 
1750 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1751 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1752 
1753 	if (driver_byte(result) == DRIVER_SENSE &&
1754 	    scsi_sense_valid(&sshdr)) {
1755 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1756 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1757 	}
1758 
1759 	return result;
1760 }
1761 
1762 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1763 		u32 flags)
1764 {
1765 	if (flags & ~PR_FL_IGNORE_KEY)
1766 		return -EOPNOTSUPP;
1767 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1768 			old_key, new_key, 0,
1769 			(1 << 0) /* APTPL */);
1770 }
1771 
1772 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1773 		u32 flags)
1774 {
1775 	if (flags)
1776 		return -EOPNOTSUPP;
1777 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1778 }
1779 
1780 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1781 {
1782 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1783 }
1784 
1785 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1786 		enum pr_type type, bool abort)
1787 {
1788 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1789 			     sd_pr_type(type), 0);
1790 }
1791 
1792 static int sd_pr_clear(struct block_device *bdev, u64 key)
1793 {
1794 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1795 }
1796 
1797 static const struct pr_ops sd_pr_ops = {
1798 	.pr_register	= sd_pr_register,
1799 	.pr_reserve	= sd_pr_reserve,
1800 	.pr_release	= sd_pr_release,
1801 	.pr_preempt	= sd_pr_preempt,
1802 	.pr_clear	= sd_pr_clear,
1803 };
1804 
1805 static const struct block_device_operations sd_fops = {
1806 	.owner			= THIS_MODULE,
1807 	.open			= sd_open,
1808 	.release		= sd_release,
1809 	.ioctl			= sd_ioctl,
1810 	.getgeo			= sd_getgeo,
1811 #ifdef CONFIG_COMPAT
1812 	.compat_ioctl		= sd_compat_ioctl,
1813 #endif
1814 	.check_events		= sd_check_events,
1815 	.revalidate_disk	= sd_revalidate_disk,
1816 	.unlock_native_capacity	= sd_unlock_native_capacity,
1817 	.report_zones		= sd_zbc_report_zones,
1818 	.pr_ops			= &sd_pr_ops,
1819 };
1820 
1821 /**
1822  *	sd_eh_reset - reset error handling callback
1823  *	@scmd:		sd-issued command that has failed
1824  *
1825  *	This function is called by the SCSI midlayer before starting
1826  *	SCSI EH. When counting medium access failures we have to be
1827  *	careful to register it only only once per device and SCSI EH run;
1828  *	there might be several timed out commands which will cause the
1829  *	'max_medium_access_timeouts' counter to trigger after the first
1830  *	SCSI EH run already and set the device to offline.
1831  *	So this function resets the internal counter before starting SCSI EH.
1832  **/
1833 static void sd_eh_reset(struct scsi_cmnd *scmd)
1834 {
1835 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1836 
1837 	/* New SCSI EH run, reset gate variable */
1838 	sdkp->ignore_medium_access_errors = false;
1839 }
1840 
1841 /**
1842  *	sd_eh_action - error handling callback
1843  *	@scmd:		sd-issued command that has failed
1844  *	@eh_disp:	The recovery disposition suggested by the midlayer
1845  *
1846  *	This function is called by the SCSI midlayer upon completion of an
1847  *	error test command (currently TEST UNIT READY). The result of sending
1848  *	the eh command is passed in eh_disp.  We're looking for devices that
1849  *	fail medium access commands but are OK with non access commands like
1850  *	test unit ready (so wrongly see the device as having a successful
1851  *	recovery)
1852  **/
1853 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1854 {
1855 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1856 	struct scsi_device *sdev = scmd->device;
1857 
1858 	if (!scsi_device_online(sdev) ||
1859 	    !scsi_medium_access_command(scmd) ||
1860 	    host_byte(scmd->result) != DID_TIME_OUT ||
1861 	    eh_disp != SUCCESS)
1862 		return eh_disp;
1863 
1864 	/*
1865 	 * The device has timed out executing a medium access command.
1866 	 * However, the TEST UNIT READY command sent during error
1867 	 * handling completed successfully. Either the device is in the
1868 	 * process of recovering or has it suffered an internal failure
1869 	 * that prevents access to the storage medium.
1870 	 */
1871 	if (!sdkp->ignore_medium_access_errors) {
1872 		sdkp->medium_access_timed_out++;
1873 		sdkp->ignore_medium_access_errors = true;
1874 	}
1875 
1876 	/*
1877 	 * If the device keeps failing read/write commands but TEST UNIT
1878 	 * READY always completes successfully we assume that medium
1879 	 * access is no longer possible and take the device offline.
1880 	 */
1881 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1882 		scmd_printk(KERN_ERR, scmd,
1883 			    "Medium access timeout failure. Offlining disk!\n");
1884 		mutex_lock(&sdev->state_mutex);
1885 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1886 		mutex_unlock(&sdev->state_mutex);
1887 
1888 		return SUCCESS;
1889 	}
1890 
1891 	return eh_disp;
1892 }
1893 
1894 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1895 {
1896 	struct request *req = scmd->request;
1897 	struct scsi_device *sdev = scmd->device;
1898 	unsigned int transferred, good_bytes;
1899 	u64 start_lba, end_lba, bad_lba;
1900 
1901 	/*
1902 	 * Some commands have a payload smaller than the device logical
1903 	 * block size (e.g. INQUIRY on a 4K disk).
1904 	 */
1905 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1906 		return 0;
1907 
1908 	/* Check if we have a 'bad_lba' information */
1909 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1910 				     SCSI_SENSE_BUFFERSIZE,
1911 				     &bad_lba))
1912 		return 0;
1913 
1914 	/*
1915 	 * If the bad lba was reported incorrectly, we have no idea where
1916 	 * the error is.
1917 	 */
1918 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1919 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1920 	if (bad_lba < start_lba || bad_lba >= end_lba)
1921 		return 0;
1922 
1923 	/*
1924 	 * resid is optional but mostly filled in.  When it's unused,
1925 	 * its value is zero, so we assume the whole buffer transferred
1926 	 */
1927 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1928 
1929 	/* This computation should always be done in terms of the
1930 	 * resolution of the device's medium.
1931 	 */
1932 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1933 
1934 	return min(good_bytes, transferred);
1935 }
1936 
1937 /**
1938  *	sd_done - bottom half handler: called when the lower level
1939  *	driver has completed (successfully or otherwise) a scsi command.
1940  *	@SCpnt: mid-level's per command structure.
1941  *
1942  *	Note: potentially run from within an ISR. Must not block.
1943  **/
1944 static int sd_done(struct scsi_cmnd *SCpnt)
1945 {
1946 	int result = SCpnt->result;
1947 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1948 	unsigned int sector_size = SCpnt->device->sector_size;
1949 	unsigned int resid;
1950 	struct scsi_sense_hdr sshdr;
1951 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1952 	struct request *req = SCpnt->request;
1953 	int sense_valid = 0;
1954 	int sense_deferred = 0;
1955 
1956 	switch (req_op(req)) {
1957 	case REQ_OP_DISCARD:
1958 	case REQ_OP_WRITE_ZEROES:
1959 	case REQ_OP_WRITE_SAME:
1960 	case REQ_OP_ZONE_RESET:
1961 		if (!result) {
1962 			good_bytes = blk_rq_bytes(req);
1963 			scsi_set_resid(SCpnt, 0);
1964 		} else {
1965 			good_bytes = 0;
1966 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1967 		}
1968 		break;
1969 	default:
1970 		/*
1971 		 * In case of bogus fw or device, we could end up having
1972 		 * an unaligned partial completion. Check this here and force
1973 		 * alignment.
1974 		 */
1975 		resid = scsi_get_resid(SCpnt);
1976 		if (resid & (sector_size - 1)) {
1977 			sd_printk(KERN_INFO, sdkp,
1978 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1979 				resid, sector_size);
1980 			resid = min(scsi_bufflen(SCpnt),
1981 				    round_up(resid, sector_size));
1982 			scsi_set_resid(SCpnt, resid);
1983 		}
1984 	}
1985 
1986 	if (result) {
1987 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1988 		if (sense_valid)
1989 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1990 	}
1991 	sdkp->medium_access_timed_out = 0;
1992 
1993 	if (driver_byte(result) != DRIVER_SENSE &&
1994 	    (!sense_valid || sense_deferred))
1995 		goto out;
1996 
1997 	switch (sshdr.sense_key) {
1998 	case HARDWARE_ERROR:
1999 	case MEDIUM_ERROR:
2000 		good_bytes = sd_completed_bytes(SCpnt);
2001 		break;
2002 	case RECOVERED_ERROR:
2003 		good_bytes = scsi_bufflen(SCpnt);
2004 		break;
2005 	case NO_SENSE:
2006 		/* This indicates a false check condition, so ignore it.  An
2007 		 * unknown amount of data was transferred so treat it as an
2008 		 * error.
2009 		 */
2010 		SCpnt->result = 0;
2011 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2012 		break;
2013 	case ABORTED_COMMAND:
2014 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2015 			good_bytes = sd_completed_bytes(SCpnt);
2016 		break;
2017 	case ILLEGAL_REQUEST:
2018 		switch (sshdr.asc) {
2019 		case 0x10:	/* DIX: Host detected corruption */
2020 			good_bytes = sd_completed_bytes(SCpnt);
2021 			break;
2022 		case 0x20:	/* INVALID COMMAND OPCODE */
2023 		case 0x24:	/* INVALID FIELD IN CDB */
2024 			switch (SCpnt->cmnd[0]) {
2025 			case UNMAP:
2026 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2027 				break;
2028 			case WRITE_SAME_16:
2029 			case WRITE_SAME:
2030 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2031 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2032 				} else {
2033 					sdkp->device->no_write_same = 1;
2034 					sd_config_write_same(sdkp);
2035 					req->rq_flags |= RQF_QUIET;
2036 				}
2037 				break;
2038 			}
2039 		}
2040 		break;
2041 	default:
2042 		break;
2043 	}
2044 
2045  out:
2046 	if (sd_is_zoned(sdkp))
2047 		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2048 
2049 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2050 					   "sd_done: completed %d of %d bytes\n",
2051 					   good_bytes, scsi_bufflen(SCpnt)));
2052 
2053 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2054 	    good_bytes)
2055 		t10_pi_complete(SCpnt->request, sdkp->protection_type,
2056 				good_bytes / scsi_prot_interval(SCpnt));
2057 
2058 	return good_bytes;
2059 }
2060 
2061 /*
2062  * spinup disk - called only in sd_revalidate_disk()
2063  */
2064 static void
2065 sd_spinup_disk(struct scsi_disk *sdkp)
2066 {
2067 	unsigned char cmd[10];
2068 	unsigned long spintime_expire = 0;
2069 	int retries, spintime;
2070 	unsigned int the_result;
2071 	struct scsi_sense_hdr sshdr;
2072 	int sense_valid = 0;
2073 
2074 	spintime = 0;
2075 
2076 	/* Spin up drives, as required.  Only do this at boot time */
2077 	/* Spinup needs to be done for module loads too. */
2078 	do {
2079 		retries = 0;
2080 
2081 		do {
2082 			cmd[0] = TEST_UNIT_READY;
2083 			memset((void *) &cmd[1], 0, 9);
2084 
2085 			the_result = scsi_execute_req(sdkp->device, cmd,
2086 						      DMA_NONE, NULL, 0,
2087 						      &sshdr, SD_TIMEOUT,
2088 						      SD_MAX_RETRIES, NULL);
2089 
2090 			/*
2091 			 * If the drive has indicated to us that it
2092 			 * doesn't have any media in it, don't bother
2093 			 * with any more polling.
2094 			 */
2095 			if (media_not_present(sdkp, &sshdr))
2096 				return;
2097 
2098 			if (the_result)
2099 				sense_valid = scsi_sense_valid(&sshdr);
2100 			retries++;
2101 		} while (retries < 3 &&
2102 			 (!scsi_status_is_good(the_result) ||
2103 			  ((driver_byte(the_result) == DRIVER_SENSE) &&
2104 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2105 
2106 		if (driver_byte(the_result) != DRIVER_SENSE) {
2107 			/* no sense, TUR either succeeded or failed
2108 			 * with a status error */
2109 			if(!spintime && !scsi_status_is_good(the_result)) {
2110 				sd_print_result(sdkp, "Test Unit Ready failed",
2111 						the_result);
2112 			}
2113 			break;
2114 		}
2115 
2116 		/*
2117 		 * The device does not want the automatic start to be issued.
2118 		 */
2119 		if (sdkp->device->no_start_on_add)
2120 			break;
2121 
2122 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2123 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2124 				break;	/* manual intervention required */
2125 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2126 				break;	/* standby */
2127 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2128 				break;	/* unavailable */
2129 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2130 				break;	/* sanitize in progress */
2131 			/*
2132 			 * Issue command to spin up drive when not ready
2133 			 */
2134 			if (!spintime) {
2135 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2136 				cmd[0] = START_STOP;
2137 				cmd[1] = 1;	/* Return immediately */
2138 				memset((void *) &cmd[2], 0, 8);
2139 				cmd[4] = 1;	/* Start spin cycle */
2140 				if (sdkp->device->start_stop_pwr_cond)
2141 					cmd[4] |= 1 << 4;
2142 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2143 						 NULL, 0, &sshdr,
2144 						 SD_TIMEOUT, SD_MAX_RETRIES,
2145 						 NULL);
2146 				spintime_expire = jiffies + 100 * HZ;
2147 				spintime = 1;
2148 			}
2149 			/* Wait 1 second for next try */
2150 			msleep(1000);
2151 			printk(KERN_CONT ".");
2152 
2153 		/*
2154 		 * Wait for USB flash devices with slow firmware.
2155 		 * Yes, this sense key/ASC combination shouldn't
2156 		 * occur here.  It's characteristic of these devices.
2157 		 */
2158 		} else if (sense_valid &&
2159 				sshdr.sense_key == UNIT_ATTENTION &&
2160 				sshdr.asc == 0x28) {
2161 			if (!spintime) {
2162 				spintime_expire = jiffies + 5 * HZ;
2163 				spintime = 1;
2164 			}
2165 			/* Wait 1 second for next try */
2166 			msleep(1000);
2167 		} else {
2168 			/* we don't understand the sense code, so it's
2169 			 * probably pointless to loop */
2170 			if(!spintime) {
2171 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2172 				sd_print_sense_hdr(sdkp, &sshdr);
2173 			}
2174 			break;
2175 		}
2176 
2177 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2178 
2179 	if (spintime) {
2180 		if (scsi_status_is_good(the_result))
2181 			printk(KERN_CONT "ready\n");
2182 		else
2183 			printk(KERN_CONT "not responding...\n");
2184 	}
2185 }
2186 
2187 /*
2188  * Determine whether disk supports Data Integrity Field.
2189  */
2190 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2191 {
2192 	struct scsi_device *sdp = sdkp->device;
2193 	u8 type;
2194 	int ret = 0;
2195 
2196 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2197 		return ret;
2198 
2199 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2200 
2201 	if (type > T10_PI_TYPE3_PROTECTION)
2202 		ret = -ENODEV;
2203 	else if (scsi_host_dif_capable(sdp->host, type))
2204 		ret = 1;
2205 
2206 	if (sdkp->first_scan || type != sdkp->protection_type)
2207 		switch (ret) {
2208 		case -ENODEV:
2209 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2210 				  " protection type %u. Disabling disk!\n",
2211 				  type);
2212 			break;
2213 		case 1:
2214 			sd_printk(KERN_NOTICE, sdkp,
2215 				  "Enabling DIF Type %u protection\n", type);
2216 			break;
2217 		case 0:
2218 			sd_printk(KERN_NOTICE, sdkp,
2219 				  "Disabling DIF Type %u protection\n", type);
2220 			break;
2221 		}
2222 
2223 	sdkp->protection_type = type;
2224 
2225 	return ret;
2226 }
2227 
2228 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2229 			struct scsi_sense_hdr *sshdr, int sense_valid,
2230 			int the_result)
2231 {
2232 	if (driver_byte(the_result) == DRIVER_SENSE)
2233 		sd_print_sense_hdr(sdkp, sshdr);
2234 	else
2235 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2236 
2237 	/*
2238 	 * Set dirty bit for removable devices if not ready -
2239 	 * sometimes drives will not report this properly.
2240 	 */
2241 	if (sdp->removable &&
2242 	    sense_valid && sshdr->sense_key == NOT_READY)
2243 		set_media_not_present(sdkp);
2244 
2245 	/*
2246 	 * We used to set media_present to 0 here to indicate no media
2247 	 * in the drive, but some drives fail read capacity even with
2248 	 * media present, so we can't do that.
2249 	 */
2250 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2251 }
2252 
2253 #define RC16_LEN 32
2254 #if RC16_LEN > SD_BUF_SIZE
2255 #error RC16_LEN must not be more than SD_BUF_SIZE
2256 #endif
2257 
2258 #define READ_CAPACITY_RETRIES_ON_RESET	10
2259 
2260 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2261 						unsigned char *buffer)
2262 {
2263 	unsigned char cmd[16];
2264 	struct scsi_sense_hdr sshdr;
2265 	int sense_valid = 0;
2266 	int the_result;
2267 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2268 	unsigned int alignment;
2269 	unsigned long long lba;
2270 	unsigned sector_size;
2271 
2272 	if (sdp->no_read_capacity_16)
2273 		return -EINVAL;
2274 
2275 	do {
2276 		memset(cmd, 0, 16);
2277 		cmd[0] = SERVICE_ACTION_IN_16;
2278 		cmd[1] = SAI_READ_CAPACITY_16;
2279 		cmd[13] = RC16_LEN;
2280 		memset(buffer, 0, RC16_LEN);
2281 
2282 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2283 					buffer, RC16_LEN, &sshdr,
2284 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2285 
2286 		if (media_not_present(sdkp, &sshdr))
2287 			return -ENODEV;
2288 
2289 		if (the_result) {
2290 			sense_valid = scsi_sense_valid(&sshdr);
2291 			if (sense_valid &&
2292 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2293 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2294 			    sshdr.ascq == 0x00)
2295 				/* Invalid Command Operation Code or
2296 				 * Invalid Field in CDB, just retry
2297 				 * silently with RC10 */
2298 				return -EINVAL;
2299 			if (sense_valid &&
2300 			    sshdr.sense_key == UNIT_ATTENTION &&
2301 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2302 				/* Device reset might occur several times,
2303 				 * give it one more chance */
2304 				if (--reset_retries > 0)
2305 					continue;
2306 		}
2307 		retries--;
2308 
2309 	} while (the_result && retries);
2310 
2311 	if (the_result) {
2312 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2313 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2314 		return -EINVAL;
2315 	}
2316 
2317 	sector_size = get_unaligned_be32(&buffer[8]);
2318 	lba = get_unaligned_be64(&buffer[0]);
2319 
2320 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2321 		sdkp->capacity = 0;
2322 		return -ENODEV;
2323 	}
2324 
2325 	/* Logical blocks per physical block exponent */
2326 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2327 
2328 	/* RC basis */
2329 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2330 
2331 	/* Lowest aligned logical block */
2332 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2333 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2334 	if (alignment && sdkp->first_scan)
2335 		sd_printk(KERN_NOTICE, sdkp,
2336 			  "physical block alignment offset: %u\n", alignment);
2337 
2338 	if (buffer[14] & 0x80) { /* LBPME */
2339 		sdkp->lbpme = 1;
2340 
2341 		if (buffer[14] & 0x40) /* LBPRZ */
2342 			sdkp->lbprz = 1;
2343 
2344 		sd_config_discard(sdkp, SD_LBP_WS16);
2345 	}
2346 
2347 	sdkp->capacity = lba + 1;
2348 	return sector_size;
2349 }
2350 
2351 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2352 						unsigned char *buffer)
2353 {
2354 	unsigned char cmd[16];
2355 	struct scsi_sense_hdr sshdr;
2356 	int sense_valid = 0;
2357 	int the_result;
2358 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2359 	sector_t lba;
2360 	unsigned sector_size;
2361 
2362 	do {
2363 		cmd[0] = READ_CAPACITY;
2364 		memset(&cmd[1], 0, 9);
2365 		memset(buffer, 0, 8);
2366 
2367 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2368 					buffer, 8, &sshdr,
2369 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2370 
2371 		if (media_not_present(sdkp, &sshdr))
2372 			return -ENODEV;
2373 
2374 		if (the_result) {
2375 			sense_valid = scsi_sense_valid(&sshdr);
2376 			if (sense_valid &&
2377 			    sshdr.sense_key == UNIT_ATTENTION &&
2378 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2379 				/* Device reset might occur several times,
2380 				 * give it one more chance */
2381 				if (--reset_retries > 0)
2382 					continue;
2383 		}
2384 		retries--;
2385 
2386 	} while (the_result && retries);
2387 
2388 	if (the_result) {
2389 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2390 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2391 		return -EINVAL;
2392 	}
2393 
2394 	sector_size = get_unaligned_be32(&buffer[4]);
2395 	lba = get_unaligned_be32(&buffer[0]);
2396 
2397 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2398 		/* Some buggy (usb cardreader) devices return an lba of
2399 		   0xffffffff when the want to report a size of 0 (with
2400 		   which they really mean no media is present) */
2401 		sdkp->capacity = 0;
2402 		sdkp->physical_block_size = sector_size;
2403 		return sector_size;
2404 	}
2405 
2406 	sdkp->capacity = lba + 1;
2407 	sdkp->physical_block_size = sector_size;
2408 	return sector_size;
2409 }
2410 
2411 static int sd_try_rc16_first(struct scsi_device *sdp)
2412 {
2413 	if (sdp->host->max_cmd_len < 16)
2414 		return 0;
2415 	if (sdp->try_rc_10_first)
2416 		return 0;
2417 	if (sdp->scsi_level > SCSI_SPC_2)
2418 		return 1;
2419 	if (scsi_device_protection(sdp))
2420 		return 1;
2421 	return 0;
2422 }
2423 
2424 /*
2425  * read disk capacity
2426  */
2427 static void
2428 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2429 {
2430 	int sector_size;
2431 	struct scsi_device *sdp = sdkp->device;
2432 
2433 	if (sd_try_rc16_first(sdp)) {
2434 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2435 		if (sector_size == -EOVERFLOW)
2436 			goto got_data;
2437 		if (sector_size == -ENODEV)
2438 			return;
2439 		if (sector_size < 0)
2440 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2441 		if (sector_size < 0)
2442 			return;
2443 	} else {
2444 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2445 		if (sector_size == -EOVERFLOW)
2446 			goto got_data;
2447 		if (sector_size < 0)
2448 			return;
2449 		if ((sizeof(sdkp->capacity) > 4) &&
2450 		    (sdkp->capacity > 0xffffffffULL)) {
2451 			int old_sector_size = sector_size;
2452 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2453 					"Trying to use READ CAPACITY(16).\n");
2454 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2455 			if (sector_size < 0) {
2456 				sd_printk(KERN_NOTICE, sdkp,
2457 					"Using 0xffffffff as device size\n");
2458 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2459 				sector_size = old_sector_size;
2460 				goto got_data;
2461 			}
2462 			/* Remember that READ CAPACITY(16) succeeded */
2463 			sdp->try_rc_10_first = 0;
2464 		}
2465 	}
2466 
2467 	/* Some devices are known to return the total number of blocks,
2468 	 * not the highest block number.  Some devices have versions
2469 	 * which do this and others which do not.  Some devices we might
2470 	 * suspect of doing this but we don't know for certain.
2471 	 *
2472 	 * If we know the reported capacity is wrong, decrement it.  If
2473 	 * we can only guess, then assume the number of blocks is even
2474 	 * (usually true but not always) and err on the side of lowering
2475 	 * the capacity.
2476 	 */
2477 	if (sdp->fix_capacity ||
2478 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2479 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2480 				"from its reported value: %llu\n",
2481 				(unsigned long long) sdkp->capacity);
2482 		--sdkp->capacity;
2483 	}
2484 
2485 got_data:
2486 	if (sector_size == 0) {
2487 		sector_size = 512;
2488 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2489 			  "assuming 512.\n");
2490 	}
2491 
2492 	if (sector_size != 512 &&
2493 	    sector_size != 1024 &&
2494 	    sector_size != 2048 &&
2495 	    sector_size != 4096) {
2496 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2497 			  sector_size);
2498 		/*
2499 		 * The user might want to re-format the drive with
2500 		 * a supported sectorsize.  Once this happens, it
2501 		 * would be relatively trivial to set the thing up.
2502 		 * For this reason, we leave the thing in the table.
2503 		 */
2504 		sdkp->capacity = 0;
2505 		/*
2506 		 * set a bogus sector size so the normal read/write
2507 		 * logic in the block layer will eventually refuse any
2508 		 * request on this device without tripping over power
2509 		 * of two sector size assumptions
2510 		 */
2511 		sector_size = 512;
2512 	}
2513 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2514 	blk_queue_physical_block_size(sdp->request_queue,
2515 				      sdkp->physical_block_size);
2516 	sdkp->device->sector_size = sector_size;
2517 
2518 	if (sdkp->capacity > 0xffffffff)
2519 		sdp->use_16_for_rw = 1;
2520 
2521 }
2522 
2523 /*
2524  * Print disk capacity
2525  */
2526 static void
2527 sd_print_capacity(struct scsi_disk *sdkp,
2528 		  sector_t old_capacity)
2529 {
2530 	int sector_size = sdkp->device->sector_size;
2531 	char cap_str_2[10], cap_str_10[10];
2532 
2533 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2534 		return;
2535 
2536 	string_get_size(sdkp->capacity, sector_size,
2537 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2538 	string_get_size(sdkp->capacity, sector_size,
2539 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2540 
2541 	sd_printk(KERN_NOTICE, sdkp,
2542 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2543 		  (unsigned long long)sdkp->capacity,
2544 		  sector_size, cap_str_10, cap_str_2);
2545 
2546 	if (sdkp->physical_block_size != sector_size)
2547 		sd_printk(KERN_NOTICE, sdkp,
2548 			  "%u-byte physical blocks\n",
2549 			  sdkp->physical_block_size);
2550 
2551 	sd_zbc_print_zones(sdkp);
2552 }
2553 
2554 /* called with buffer of length 512 */
2555 static inline int
2556 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2557 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2558 		 struct scsi_sense_hdr *sshdr)
2559 {
2560 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2561 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2562 			       sshdr);
2563 }
2564 
2565 /*
2566  * read write protect setting, if possible - called only in sd_revalidate_disk()
2567  * called with buffer of length SD_BUF_SIZE
2568  */
2569 static void
2570 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2571 {
2572 	int res;
2573 	struct scsi_device *sdp = sdkp->device;
2574 	struct scsi_mode_data data;
2575 	int old_wp = sdkp->write_prot;
2576 
2577 	set_disk_ro(sdkp->disk, 0);
2578 	if (sdp->skip_ms_page_3f) {
2579 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2580 		return;
2581 	}
2582 
2583 	if (sdp->use_192_bytes_for_3f) {
2584 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2585 	} else {
2586 		/*
2587 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2588 		 * We have to start carefully: some devices hang if we ask
2589 		 * for more than is available.
2590 		 */
2591 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2592 
2593 		/*
2594 		 * Second attempt: ask for page 0 When only page 0 is
2595 		 * implemented, a request for page 3F may return Sense Key
2596 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2597 		 * CDB.
2598 		 */
2599 		if (!scsi_status_is_good(res))
2600 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2601 
2602 		/*
2603 		 * Third attempt: ask 255 bytes, as we did earlier.
2604 		 */
2605 		if (!scsi_status_is_good(res))
2606 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2607 					       &data, NULL);
2608 	}
2609 
2610 	if (!scsi_status_is_good(res)) {
2611 		sd_first_printk(KERN_WARNING, sdkp,
2612 			  "Test WP failed, assume Write Enabled\n");
2613 	} else {
2614 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2615 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2616 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2617 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2618 				  sdkp->write_prot ? "on" : "off");
2619 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2620 		}
2621 	}
2622 }
2623 
2624 /*
2625  * sd_read_cache_type - called only from sd_revalidate_disk()
2626  * called with buffer of length SD_BUF_SIZE
2627  */
2628 static void
2629 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2630 {
2631 	int len = 0, res;
2632 	struct scsi_device *sdp = sdkp->device;
2633 
2634 	int dbd;
2635 	int modepage;
2636 	int first_len;
2637 	struct scsi_mode_data data;
2638 	struct scsi_sense_hdr sshdr;
2639 	int old_wce = sdkp->WCE;
2640 	int old_rcd = sdkp->RCD;
2641 	int old_dpofua = sdkp->DPOFUA;
2642 
2643 
2644 	if (sdkp->cache_override)
2645 		return;
2646 
2647 	first_len = 4;
2648 	if (sdp->skip_ms_page_8) {
2649 		if (sdp->type == TYPE_RBC)
2650 			goto defaults;
2651 		else {
2652 			if (sdp->skip_ms_page_3f)
2653 				goto defaults;
2654 			modepage = 0x3F;
2655 			if (sdp->use_192_bytes_for_3f)
2656 				first_len = 192;
2657 			dbd = 0;
2658 		}
2659 	} else if (sdp->type == TYPE_RBC) {
2660 		modepage = 6;
2661 		dbd = 8;
2662 	} else {
2663 		modepage = 8;
2664 		dbd = 0;
2665 	}
2666 
2667 	/* cautiously ask */
2668 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2669 			&data, &sshdr);
2670 
2671 	if (!scsi_status_is_good(res))
2672 		goto bad_sense;
2673 
2674 	if (!data.header_length) {
2675 		modepage = 6;
2676 		first_len = 0;
2677 		sd_first_printk(KERN_ERR, sdkp,
2678 				"Missing header in MODE_SENSE response\n");
2679 	}
2680 
2681 	/* that went OK, now ask for the proper length */
2682 	len = data.length;
2683 
2684 	/*
2685 	 * We're only interested in the first three bytes, actually.
2686 	 * But the data cache page is defined for the first 20.
2687 	 */
2688 	if (len < 3)
2689 		goto bad_sense;
2690 	else if (len > SD_BUF_SIZE) {
2691 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2692 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2693 		len = SD_BUF_SIZE;
2694 	}
2695 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2696 		len = 192;
2697 
2698 	/* Get the data */
2699 	if (len > first_len)
2700 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2701 				&data, &sshdr);
2702 
2703 	if (scsi_status_is_good(res)) {
2704 		int offset = data.header_length + data.block_descriptor_length;
2705 
2706 		while (offset < len) {
2707 			u8 page_code = buffer[offset] & 0x3F;
2708 			u8 spf       = buffer[offset] & 0x40;
2709 
2710 			if (page_code == 8 || page_code == 6) {
2711 				/* We're interested only in the first 3 bytes.
2712 				 */
2713 				if (len - offset <= 2) {
2714 					sd_first_printk(KERN_ERR, sdkp,
2715 						"Incomplete mode parameter "
2716 							"data\n");
2717 					goto defaults;
2718 				} else {
2719 					modepage = page_code;
2720 					goto Page_found;
2721 				}
2722 			} else {
2723 				/* Go to the next page */
2724 				if (spf && len - offset > 3)
2725 					offset += 4 + (buffer[offset+2] << 8) +
2726 						buffer[offset+3];
2727 				else if (!spf && len - offset > 1)
2728 					offset += 2 + buffer[offset+1];
2729 				else {
2730 					sd_first_printk(KERN_ERR, sdkp,
2731 							"Incomplete mode "
2732 							"parameter data\n");
2733 					goto defaults;
2734 				}
2735 			}
2736 		}
2737 
2738 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2739 		goto defaults;
2740 
2741 	Page_found:
2742 		if (modepage == 8) {
2743 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2744 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2745 		} else {
2746 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2747 			sdkp->RCD = 0;
2748 		}
2749 
2750 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2751 		if (sdp->broken_fua) {
2752 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2753 			sdkp->DPOFUA = 0;
2754 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2755 			   !sdkp->device->use_16_for_rw) {
2756 			sd_first_printk(KERN_NOTICE, sdkp,
2757 				  "Uses READ/WRITE(6), disabling FUA\n");
2758 			sdkp->DPOFUA = 0;
2759 		}
2760 
2761 		/* No cache flush allowed for write protected devices */
2762 		if (sdkp->WCE && sdkp->write_prot)
2763 			sdkp->WCE = 0;
2764 
2765 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2766 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2767 			sd_printk(KERN_NOTICE, sdkp,
2768 				  "Write cache: %s, read cache: %s, %s\n",
2769 				  sdkp->WCE ? "enabled" : "disabled",
2770 				  sdkp->RCD ? "disabled" : "enabled",
2771 				  sdkp->DPOFUA ? "supports DPO and FUA"
2772 				  : "doesn't support DPO or FUA");
2773 
2774 		return;
2775 	}
2776 
2777 bad_sense:
2778 	if (scsi_sense_valid(&sshdr) &&
2779 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2780 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2781 		/* Invalid field in CDB */
2782 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2783 	else
2784 		sd_first_printk(KERN_ERR, sdkp,
2785 				"Asking for cache data failed\n");
2786 
2787 defaults:
2788 	if (sdp->wce_default_on) {
2789 		sd_first_printk(KERN_NOTICE, sdkp,
2790 				"Assuming drive cache: write back\n");
2791 		sdkp->WCE = 1;
2792 	} else {
2793 		sd_first_printk(KERN_ERR, sdkp,
2794 				"Assuming drive cache: write through\n");
2795 		sdkp->WCE = 0;
2796 	}
2797 	sdkp->RCD = 0;
2798 	sdkp->DPOFUA = 0;
2799 }
2800 
2801 /*
2802  * The ATO bit indicates whether the DIF application tag is available
2803  * for use by the operating system.
2804  */
2805 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2806 {
2807 	int res, offset;
2808 	struct scsi_device *sdp = sdkp->device;
2809 	struct scsi_mode_data data;
2810 	struct scsi_sense_hdr sshdr;
2811 
2812 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2813 		return;
2814 
2815 	if (sdkp->protection_type == 0)
2816 		return;
2817 
2818 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2819 			      SD_MAX_RETRIES, &data, &sshdr);
2820 
2821 	if (!scsi_status_is_good(res) || !data.header_length ||
2822 	    data.length < 6) {
2823 		sd_first_printk(KERN_WARNING, sdkp,
2824 			  "getting Control mode page failed, assume no ATO\n");
2825 
2826 		if (scsi_sense_valid(&sshdr))
2827 			sd_print_sense_hdr(sdkp, &sshdr);
2828 
2829 		return;
2830 	}
2831 
2832 	offset = data.header_length + data.block_descriptor_length;
2833 
2834 	if ((buffer[offset] & 0x3f) != 0x0a) {
2835 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2836 		return;
2837 	}
2838 
2839 	if ((buffer[offset + 5] & 0x80) == 0)
2840 		return;
2841 
2842 	sdkp->ATO = 1;
2843 
2844 	return;
2845 }
2846 
2847 /**
2848  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2849  * @sdkp: disk to query
2850  */
2851 static void sd_read_block_limits(struct scsi_disk *sdkp)
2852 {
2853 	unsigned int sector_sz = sdkp->device->sector_size;
2854 	const int vpd_len = 64;
2855 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2856 
2857 	if (!buffer ||
2858 	    /* Block Limits VPD */
2859 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2860 		goto out;
2861 
2862 	blk_queue_io_min(sdkp->disk->queue,
2863 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2864 
2865 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2866 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2867 
2868 	if (buffer[3] == 0x3c) {
2869 		unsigned int lba_count, desc_count;
2870 
2871 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2872 
2873 		if (!sdkp->lbpme)
2874 			goto out;
2875 
2876 		lba_count = get_unaligned_be32(&buffer[20]);
2877 		desc_count = get_unaligned_be32(&buffer[24]);
2878 
2879 		if (lba_count && desc_count)
2880 			sdkp->max_unmap_blocks = lba_count;
2881 
2882 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2883 
2884 		if (buffer[32] & 0x80)
2885 			sdkp->unmap_alignment =
2886 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2887 
2888 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2889 
2890 			if (sdkp->max_unmap_blocks)
2891 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2892 			else
2893 				sd_config_discard(sdkp, SD_LBP_WS16);
2894 
2895 		} else {	/* LBP VPD page tells us what to use */
2896 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2897 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2898 			else if (sdkp->lbpws)
2899 				sd_config_discard(sdkp, SD_LBP_WS16);
2900 			else if (sdkp->lbpws10)
2901 				sd_config_discard(sdkp, SD_LBP_WS10);
2902 			else
2903 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2904 		}
2905 	}
2906 
2907  out:
2908 	kfree(buffer);
2909 }
2910 
2911 /**
2912  * sd_read_block_characteristics - Query block dev. characteristics
2913  * @sdkp: disk to query
2914  */
2915 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2916 {
2917 	struct request_queue *q = sdkp->disk->queue;
2918 	unsigned char *buffer;
2919 	u16 rot;
2920 	const int vpd_len = 64;
2921 
2922 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2923 
2924 	if (!buffer ||
2925 	    /* Block Device Characteristics VPD */
2926 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2927 		goto out;
2928 
2929 	rot = get_unaligned_be16(&buffer[4]);
2930 
2931 	if (rot == 1) {
2932 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2933 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2934 	}
2935 
2936 	if (sdkp->device->type == TYPE_ZBC) {
2937 		/* Host-managed */
2938 		q->limits.zoned = BLK_ZONED_HM;
2939 	} else {
2940 		sdkp->zoned = (buffer[8] >> 4) & 3;
2941 		if (sdkp->zoned == 1)
2942 			/* Host-aware */
2943 			q->limits.zoned = BLK_ZONED_HA;
2944 		else
2945 			/*
2946 			 * Treat drive-managed devices as
2947 			 * regular block devices.
2948 			 */
2949 			q->limits.zoned = BLK_ZONED_NONE;
2950 	}
2951 	if (blk_queue_is_zoned(q) && sdkp->first_scan)
2952 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2953 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2954 
2955  out:
2956 	kfree(buffer);
2957 }
2958 
2959 /**
2960  * sd_read_block_provisioning - Query provisioning VPD page
2961  * @sdkp: disk to query
2962  */
2963 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2964 {
2965 	unsigned char *buffer;
2966 	const int vpd_len = 8;
2967 
2968 	if (sdkp->lbpme == 0)
2969 		return;
2970 
2971 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2972 
2973 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2974 		goto out;
2975 
2976 	sdkp->lbpvpd	= 1;
2977 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2978 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2979 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2980 
2981  out:
2982 	kfree(buffer);
2983 }
2984 
2985 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2986 {
2987 	struct scsi_device *sdev = sdkp->device;
2988 
2989 	if (sdev->host->no_write_same) {
2990 		sdev->no_write_same = 1;
2991 
2992 		return;
2993 	}
2994 
2995 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2996 		/* too large values might cause issues with arcmsr */
2997 		int vpd_buf_len = 64;
2998 
2999 		sdev->no_report_opcodes = 1;
3000 
3001 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3002 		 * CODES is unsupported and the device has an ATA
3003 		 * Information VPD page (SAT).
3004 		 */
3005 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3006 			sdev->no_write_same = 1;
3007 	}
3008 
3009 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3010 		sdkp->ws16 = 1;
3011 
3012 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3013 		sdkp->ws10 = 1;
3014 }
3015 
3016 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3017 {
3018 	struct scsi_device *sdev = sdkp->device;
3019 
3020 	if (!sdev->security_supported)
3021 		return;
3022 
3023 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3024 			SECURITY_PROTOCOL_IN) == 1 &&
3025 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3026 			SECURITY_PROTOCOL_OUT) == 1)
3027 		sdkp->security = 1;
3028 }
3029 
3030 /*
3031  * Determine the device's preferred I/O size for reads and writes
3032  * unless the reported value is unreasonably small, large, not a
3033  * multiple of the physical block size, or simply garbage.
3034  */
3035 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3036 				      unsigned int dev_max)
3037 {
3038 	struct scsi_device *sdp = sdkp->device;
3039 	unsigned int opt_xfer_bytes =
3040 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3041 
3042 	if (sdkp->opt_xfer_blocks == 0)
3043 		return false;
3044 
3045 	if (sdkp->opt_xfer_blocks > dev_max) {
3046 		sd_first_printk(KERN_WARNING, sdkp,
3047 				"Optimal transfer size %u logical blocks " \
3048 				"> dev_max (%u logical blocks)\n",
3049 				sdkp->opt_xfer_blocks, dev_max);
3050 		return false;
3051 	}
3052 
3053 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3054 		sd_first_printk(KERN_WARNING, sdkp,
3055 				"Optimal transfer size %u logical blocks " \
3056 				"> sd driver limit (%u logical blocks)\n",
3057 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3058 		return false;
3059 	}
3060 
3061 	if (opt_xfer_bytes < PAGE_SIZE) {
3062 		sd_first_printk(KERN_WARNING, sdkp,
3063 				"Optimal transfer size %u bytes < " \
3064 				"PAGE_SIZE (%u bytes)\n",
3065 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3066 		return false;
3067 	}
3068 
3069 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3070 		sd_first_printk(KERN_WARNING, sdkp,
3071 				"Optimal transfer size %u bytes not a " \
3072 				"multiple of physical block size (%u bytes)\n",
3073 				opt_xfer_bytes, sdkp->physical_block_size);
3074 		return false;
3075 	}
3076 
3077 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3078 			opt_xfer_bytes);
3079 	return true;
3080 }
3081 
3082 /**
3083  *	sd_revalidate_disk - called the first time a new disk is seen,
3084  *	performs disk spin up, read_capacity, etc.
3085  *	@disk: struct gendisk we care about
3086  **/
3087 static int sd_revalidate_disk(struct gendisk *disk)
3088 {
3089 	struct scsi_disk *sdkp = scsi_disk(disk);
3090 	struct scsi_device *sdp = sdkp->device;
3091 	struct request_queue *q = sdkp->disk->queue;
3092 	sector_t old_capacity = sdkp->capacity;
3093 	unsigned char *buffer;
3094 	unsigned int dev_max, rw_max;
3095 
3096 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3097 				      "sd_revalidate_disk\n"));
3098 
3099 	/*
3100 	 * If the device is offline, don't try and read capacity or any
3101 	 * of the other niceties.
3102 	 */
3103 	if (!scsi_device_online(sdp))
3104 		goto out;
3105 
3106 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3107 	if (!buffer) {
3108 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3109 			  "allocation failure.\n");
3110 		goto out;
3111 	}
3112 
3113 	sd_spinup_disk(sdkp);
3114 
3115 	/*
3116 	 * Without media there is no reason to ask; moreover, some devices
3117 	 * react badly if we do.
3118 	 */
3119 	if (sdkp->media_present) {
3120 		sd_read_capacity(sdkp, buffer);
3121 
3122 		/*
3123 		 * set the default to rotational.  All non-rotational devices
3124 		 * support the block characteristics VPD page, which will
3125 		 * cause this to be updated correctly and any device which
3126 		 * doesn't support it should be treated as rotational.
3127 		 */
3128 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3129 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3130 
3131 		if (scsi_device_supports_vpd(sdp)) {
3132 			sd_read_block_provisioning(sdkp);
3133 			sd_read_block_limits(sdkp);
3134 			sd_read_block_characteristics(sdkp);
3135 			sd_zbc_read_zones(sdkp, buffer);
3136 		}
3137 
3138 		sd_print_capacity(sdkp, old_capacity);
3139 
3140 		sd_read_write_protect_flag(sdkp, buffer);
3141 		sd_read_cache_type(sdkp, buffer);
3142 		sd_read_app_tag_own(sdkp, buffer);
3143 		sd_read_write_same(sdkp, buffer);
3144 		sd_read_security(sdkp, buffer);
3145 	}
3146 
3147 	/*
3148 	 * We now have all cache related info, determine how we deal
3149 	 * with flush requests.
3150 	 */
3151 	sd_set_flush_flag(sdkp);
3152 
3153 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3154 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3155 
3156 	/* Some devices report a maximum block count for READ/WRITE requests. */
3157 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3158 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3159 
3160 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3161 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3162 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3163 	} else
3164 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3165 				      (sector_t)BLK_DEF_MAX_SECTORS);
3166 
3167 	/* Do not exceed controller limit */
3168 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3169 
3170 	/*
3171 	 * Only update max_sectors if previously unset or if the current value
3172 	 * exceeds the capabilities of the hardware.
3173 	 */
3174 	if (sdkp->first_scan ||
3175 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3176 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3177 		q->limits.max_sectors = rw_max;
3178 
3179 	sdkp->first_scan = 0;
3180 
3181 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3182 	sd_config_write_same(sdkp);
3183 	kfree(buffer);
3184 
3185  out:
3186 	return 0;
3187 }
3188 
3189 /**
3190  *	sd_unlock_native_capacity - unlock native capacity
3191  *	@disk: struct gendisk to set capacity for
3192  *
3193  *	Block layer calls this function if it detects that partitions
3194  *	on @disk reach beyond the end of the device.  If the SCSI host
3195  *	implements ->unlock_native_capacity() method, it's invoked to
3196  *	give it a chance to adjust the device capacity.
3197  *
3198  *	CONTEXT:
3199  *	Defined by block layer.  Might sleep.
3200  */
3201 static void sd_unlock_native_capacity(struct gendisk *disk)
3202 {
3203 	struct scsi_device *sdev = scsi_disk(disk)->device;
3204 
3205 	if (sdev->host->hostt->unlock_native_capacity)
3206 		sdev->host->hostt->unlock_native_capacity(sdev);
3207 }
3208 
3209 /**
3210  *	sd_format_disk_name - format disk name
3211  *	@prefix: name prefix - ie. "sd" for SCSI disks
3212  *	@index: index of the disk to format name for
3213  *	@buf: output buffer
3214  *	@buflen: length of the output buffer
3215  *
3216  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3217  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3218  *	which is followed by sdaaa.
3219  *
3220  *	This is basically 26 base counting with one extra 'nil' entry
3221  *	at the beginning from the second digit on and can be
3222  *	determined using similar method as 26 base conversion with the
3223  *	index shifted -1 after each digit is computed.
3224  *
3225  *	CONTEXT:
3226  *	Don't care.
3227  *
3228  *	RETURNS:
3229  *	0 on success, -errno on failure.
3230  */
3231 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3232 {
3233 	const int base = 'z' - 'a' + 1;
3234 	char *begin = buf + strlen(prefix);
3235 	char *end = buf + buflen;
3236 	char *p;
3237 	int unit;
3238 
3239 	p = end - 1;
3240 	*p = '\0';
3241 	unit = base;
3242 	do {
3243 		if (p == begin)
3244 			return -EINVAL;
3245 		*--p = 'a' + (index % unit);
3246 		index = (index / unit) - 1;
3247 	} while (index >= 0);
3248 
3249 	memmove(begin, p, end - p);
3250 	memcpy(buf, prefix, strlen(prefix));
3251 
3252 	return 0;
3253 }
3254 
3255 /*
3256  * The asynchronous part of sd_probe
3257  */
3258 static void sd_probe_async(void *data, async_cookie_t cookie)
3259 {
3260 	struct scsi_disk *sdkp = data;
3261 	struct scsi_device *sdp;
3262 	struct gendisk *gd;
3263 	u32 index;
3264 	struct device *dev;
3265 
3266 	sdp = sdkp->device;
3267 	gd = sdkp->disk;
3268 	index = sdkp->index;
3269 	dev = &sdp->sdev_gendev;
3270 
3271 	gd->major = sd_major((index & 0xf0) >> 4);
3272 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3273 
3274 	gd->fops = &sd_fops;
3275 	gd->private_data = &sdkp->driver;
3276 	gd->queue = sdkp->device->request_queue;
3277 
3278 	/* defaults, until the device tells us otherwise */
3279 	sdp->sector_size = 512;
3280 	sdkp->capacity = 0;
3281 	sdkp->media_present = 1;
3282 	sdkp->write_prot = 0;
3283 	sdkp->cache_override = 0;
3284 	sdkp->WCE = 0;
3285 	sdkp->RCD = 0;
3286 	sdkp->ATO = 0;
3287 	sdkp->first_scan = 1;
3288 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3289 
3290 	sd_revalidate_disk(gd);
3291 
3292 	gd->flags = GENHD_FL_EXT_DEVT;
3293 	if (sdp->removable) {
3294 		gd->flags |= GENHD_FL_REMOVABLE;
3295 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3296 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3297 	}
3298 
3299 	blk_pm_runtime_init(sdp->request_queue, dev);
3300 	device_add_disk(dev, gd, NULL);
3301 	if (sdkp->capacity)
3302 		sd_dif_config_host(sdkp);
3303 
3304 	sd_revalidate_disk(gd);
3305 
3306 	if (sdkp->security) {
3307 		sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3308 		if (sdkp->opal_dev)
3309 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3310 	}
3311 
3312 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3313 		  sdp->removable ? "removable " : "");
3314 	scsi_autopm_put_device(sdp);
3315 	put_device(&sdkp->dev);
3316 }
3317 
3318 /**
3319  *	sd_probe - called during driver initialization and whenever a
3320  *	new scsi device is attached to the system. It is called once
3321  *	for each scsi device (not just disks) present.
3322  *	@dev: pointer to device object
3323  *
3324  *	Returns 0 if successful (or not interested in this scsi device
3325  *	(e.g. scanner)); 1 when there is an error.
3326  *
3327  *	Note: this function is invoked from the scsi mid-level.
3328  *	This function sets up the mapping between a given
3329  *	<host,channel,id,lun> (found in sdp) and new device name
3330  *	(e.g. /dev/sda). More precisely it is the block device major
3331  *	and minor number that is chosen here.
3332  *
3333  *	Assume sd_probe is not re-entrant (for time being)
3334  *	Also think about sd_probe() and sd_remove() running coincidentally.
3335  **/
3336 static int sd_probe(struct device *dev)
3337 {
3338 	struct scsi_device *sdp = to_scsi_device(dev);
3339 	struct scsi_disk *sdkp;
3340 	struct gendisk *gd;
3341 	int index;
3342 	int error;
3343 
3344 	scsi_autopm_get_device(sdp);
3345 	error = -ENODEV;
3346 	if (sdp->type != TYPE_DISK &&
3347 	    sdp->type != TYPE_ZBC &&
3348 	    sdp->type != TYPE_MOD &&
3349 	    sdp->type != TYPE_RBC)
3350 		goto out;
3351 
3352 #ifndef CONFIG_BLK_DEV_ZONED
3353 	if (sdp->type == TYPE_ZBC)
3354 		goto out;
3355 #endif
3356 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3357 					"sd_probe\n"));
3358 
3359 	error = -ENOMEM;
3360 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3361 	if (!sdkp)
3362 		goto out;
3363 
3364 	gd = alloc_disk(SD_MINORS);
3365 	if (!gd)
3366 		goto out_free;
3367 
3368 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3369 	if (index < 0) {
3370 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3371 		goto out_put;
3372 	}
3373 
3374 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3375 	if (error) {
3376 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3377 		goto out_free_index;
3378 	}
3379 
3380 	sdkp->device = sdp;
3381 	sdkp->driver = &sd_template;
3382 	sdkp->disk = gd;
3383 	sdkp->index = index;
3384 	atomic_set(&sdkp->openers, 0);
3385 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3386 
3387 	if (!sdp->request_queue->rq_timeout) {
3388 		if (sdp->type != TYPE_MOD)
3389 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3390 		else
3391 			blk_queue_rq_timeout(sdp->request_queue,
3392 					     SD_MOD_TIMEOUT);
3393 	}
3394 
3395 	device_initialize(&sdkp->dev);
3396 	sdkp->dev.parent = dev;
3397 	sdkp->dev.class = &sd_disk_class;
3398 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3399 
3400 	error = device_add(&sdkp->dev);
3401 	if (error)
3402 		goto out_free_index;
3403 
3404 	get_device(dev);
3405 	dev_set_drvdata(dev, sdkp);
3406 
3407 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3408 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3409 
3410 	return 0;
3411 
3412  out_free_index:
3413 	ida_free(&sd_index_ida, index);
3414  out_put:
3415 	put_disk(gd);
3416  out_free:
3417 	kfree(sdkp);
3418  out:
3419 	scsi_autopm_put_device(sdp);
3420 	return error;
3421 }
3422 
3423 /**
3424  *	sd_remove - called whenever a scsi disk (previously recognized by
3425  *	sd_probe) is detached from the system. It is called (potentially
3426  *	multiple times) during sd module unload.
3427  *	@dev: pointer to device object
3428  *
3429  *	Note: this function is invoked from the scsi mid-level.
3430  *	This function potentially frees up a device name (e.g. /dev/sdc)
3431  *	that could be re-used by a subsequent sd_probe().
3432  *	This function is not called when the built-in sd driver is "exit-ed".
3433  **/
3434 static int sd_remove(struct device *dev)
3435 {
3436 	struct scsi_disk *sdkp;
3437 	dev_t devt;
3438 
3439 	sdkp = dev_get_drvdata(dev);
3440 	devt = disk_devt(sdkp->disk);
3441 	scsi_autopm_get_device(sdkp->device);
3442 
3443 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3444 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3445 	device_del(&sdkp->dev);
3446 	del_gendisk(sdkp->disk);
3447 	sd_shutdown(dev);
3448 
3449 	free_opal_dev(sdkp->opal_dev);
3450 
3451 	blk_register_region(devt, SD_MINORS, NULL,
3452 			    sd_default_probe, NULL, NULL);
3453 
3454 	mutex_lock(&sd_ref_mutex);
3455 	dev_set_drvdata(dev, NULL);
3456 	put_device(&sdkp->dev);
3457 	mutex_unlock(&sd_ref_mutex);
3458 
3459 	return 0;
3460 }
3461 
3462 /**
3463  *	scsi_disk_release - Called to free the scsi_disk structure
3464  *	@dev: pointer to embedded class device
3465  *
3466  *	sd_ref_mutex must be held entering this routine.  Because it is
3467  *	called on last put, you should always use the scsi_disk_get()
3468  *	scsi_disk_put() helpers which manipulate the semaphore directly
3469  *	and never do a direct put_device.
3470  **/
3471 static void scsi_disk_release(struct device *dev)
3472 {
3473 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3474 	struct gendisk *disk = sdkp->disk;
3475 	struct request_queue *q = disk->queue;
3476 
3477 	ida_free(&sd_index_ida, sdkp->index);
3478 
3479 	/*
3480 	 * Wait until all requests that are in progress have completed.
3481 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3482 	 * due to clearing the disk->private_data pointer. Wait from inside
3483 	 * scsi_disk_release() instead of from sd_release() to avoid that
3484 	 * freezing and unfreezing the request queue affects user space I/O
3485 	 * in case multiple processes open a /dev/sd... node concurrently.
3486 	 */
3487 	blk_mq_freeze_queue(q);
3488 	blk_mq_unfreeze_queue(q);
3489 
3490 	disk->private_data = NULL;
3491 	put_disk(disk);
3492 	put_device(&sdkp->device->sdev_gendev);
3493 
3494 	kfree(sdkp);
3495 }
3496 
3497 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3498 {
3499 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3500 	struct scsi_sense_hdr sshdr;
3501 	struct scsi_device *sdp = sdkp->device;
3502 	int res;
3503 
3504 	if (start)
3505 		cmd[4] |= 1;	/* START */
3506 
3507 	if (sdp->start_stop_pwr_cond)
3508 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3509 
3510 	if (!scsi_device_online(sdp))
3511 		return -ENODEV;
3512 
3513 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3514 			SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3515 	if (res) {
3516 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3517 		if (driver_byte(res) == DRIVER_SENSE)
3518 			sd_print_sense_hdr(sdkp, &sshdr);
3519 		if (scsi_sense_valid(&sshdr) &&
3520 			/* 0x3a is medium not present */
3521 			sshdr.asc == 0x3a)
3522 			res = 0;
3523 	}
3524 
3525 	/* SCSI error codes must not go to the generic layer */
3526 	if (res)
3527 		return -EIO;
3528 
3529 	return 0;
3530 }
3531 
3532 /*
3533  * Send a SYNCHRONIZE CACHE instruction down to the device through
3534  * the normal SCSI command structure.  Wait for the command to
3535  * complete.
3536  */
3537 static void sd_shutdown(struct device *dev)
3538 {
3539 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3540 
3541 	if (!sdkp)
3542 		return;         /* this can happen */
3543 
3544 	if (pm_runtime_suspended(dev))
3545 		return;
3546 
3547 	if (sdkp->WCE && sdkp->media_present) {
3548 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3549 		sd_sync_cache(sdkp, NULL);
3550 	}
3551 
3552 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3553 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3554 		sd_start_stop_device(sdkp, 0);
3555 	}
3556 }
3557 
3558 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3559 {
3560 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3561 	struct scsi_sense_hdr sshdr;
3562 	int ret = 0;
3563 
3564 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3565 		return 0;
3566 
3567 	if (sdkp->WCE && sdkp->media_present) {
3568 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3569 		ret = sd_sync_cache(sdkp, &sshdr);
3570 
3571 		if (ret) {
3572 			/* ignore OFFLINE device */
3573 			if (ret == -ENODEV)
3574 				return 0;
3575 
3576 			if (!scsi_sense_valid(&sshdr) ||
3577 			    sshdr.sense_key != ILLEGAL_REQUEST)
3578 				return ret;
3579 
3580 			/*
3581 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3582 			 * doesn't support sync. There's not much to do and
3583 			 * suspend shouldn't fail.
3584 			 */
3585 			ret = 0;
3586 		}
3587 	}
3588 
3589 	if (sdkp->device->manage_start_stop) {
3590 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3591 		/* an error is not worth aborting a system sleep */
3592 		ret = sd_start_stop_device(sdkp, 0);
3593 		if (ignore_stop_errors)
3594 			ret = 0;
3595 	}
3596 
3597 	return ret;
3598 }
3599 
3600 static int sd_suspend_system(struct device *dev)
3601 {
3602 	return sd_suspend_common(dev, true);
3603 }
3604 
3605 static int sd_suspend_runtime(struct device *dev)
3606 {
3607 	return sd_suspend_common(dev, false);
3608 }
3609 
3610 static int sd_resume(struct device *dev)
3611 {
3612 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3613 	int ret;
3614 
3615 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3616 		return 0;
3617 
3618 	if (!sdkp->device->manage_start_stop)
3619 		return 0;
3620 
3621 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3622 	ret = sd_start_stop_device(sdkp, 1);
3623 	if (!ret)
3624 		opal_unlock_from_suspend(sdkp->opal_dev);
3625 	return ret;
3626 }
3627 
3628 /**
3629  *	init_sd - entry point for this driver (both when built in or when
3630  *	a module).
3631  *
3632  *	Note: this function registers this driver with the scsi mid-level.
3633  **/
3634 static int __init init_sd(void)
3635 {
3636 	int majors = 0, i, err;
3637 
3638 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3639 
3640 	for (i = 0; i < SD_MAJORS; i++) {
3641 		if (register_blkdev(sd_major(i), "sd") != 0)
3642 			continue;
3643 		majors++;
3644 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3645 				    sd_default_probe, NULL, NULL);
3646 	}
3647 
3648 	if (!majors)
3649 		return -ENODEV;
3650 
3651 	err = class_register(&sd_disk_class);
3652 	if (err)
3653 		goto err_out;
3654 
3655 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3656 					 0, 0, NULL);
3657 	if (!sd_cdb_cache) {
3658 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3659 		err = -ENOMEM;
3660 		goto err_out_class;
3661 	}
3662 
3663 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3664 	if (!sd_cdb_pool) {
3665 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3666 		err = -ENOMEM;
3667 		goto err_out_cache;
3668 	}
3669 
3670 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3671 	if (!sd_page_pool) {
3672 		printk(KERN_ERR "sd: can't init discard page pool\n");
3673 		err = -ENOMEM;
3674 		goto err_out_ppool;
3675 	}
3676 
3677 	err = scsi_register_driver(&sd_template.gendrv);
3678 	if (err)
3679 		goto err_out_driver;
3680 
3681 	return 0;
3682 
3683 err_out_driver:
3684 	mempool_destroy(sd_page_pool);
3685 
3686 err_out_ppool:
3687 	mempool_destroy(sd_cdb_pool);
3688 
3689 err_out_cache:
3690 	kmem_cache_destroy(sd_cdb_cache);
3691 
3692 err_out_class:
3693 	class_unregister(&sd_disk_class);
3694 err_out:
3695 	for (i = 0; i < SD_MAJORS; i++)
3696 		unregister_blkdev(sd_major(i), "sd");
3697 	return err;
3698 }
3699 
3700 /**
3701  *	exit_sd - exit point for this driver (when it is a module).
3702  *
3703  *	Note: this function unregisters this driver from the scsi mid-level.
3704  **/
3705 static void __exit exit_sd(void)
3706 {
3707 	int i;
3708 
3709 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3710 
3711 	scsi_unregister_driver(&sd_template.gendrv);
3712 	mempool_destroy(sd_cdb_pool);
3713 	mempool_destroy(sd_page_pool);
3714 	kmem_cache_destroy(sd_cdb_cache);
3715 
3716 	class_unregister(&sd_disk_class);
3717 
3718 	for (i = 0; i < SD_MAJORS; i++) {
3719 		blk_unregister_region(sd_major(i), SD_MINORS);
3720 		unregister_blkdev(sd_major(i), "sd");
3721 	}
3722 }
3723 
3724 module_init(init_sd);
3725 module_exit(exit_sd);
3726 
3727 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3728 			       struct scsi_sense_hdr *sshdr)
3729 {
3730 	scsi_print_sense_hdr(sdkp->device,
3731 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3732 }
3733 
3734 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3735 			    int result)
3736 {
3737 	const char *hb_string = scsi_hostbyte_string(result);
3738 	const char *db_string = scsi_driverbyte_string(result);
3739 
3740 	if (hb_string || db_string)
3741 		sd_printk(KERN_INFO, sdkp,
3742 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3743 			  hb_string ? hb_string : "invalid",
3744 			  db_string ? db_string : "invalid");
3745 	else
3746 		sd_printk(KERN_INFO, sdkp,
3747 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3748 			  msg, host_byte(result), driver_byte(result));
3749 }
3750 
3751