xref: /openbmc/linux/drivers/scsi/sd.c (revision 2cf1c348)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *      sd.c Copyright (C) 1992 Drew Eckhardt
4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5  *
6  *      Linux scsi disk driver
7  *              Initial versions: Drew Eckhardt
8  *              Subsequent revisions: Eric Youngdale
9  *	Modification history:
10  *       - Drew Eckhardt <drew@colorado.edu> original
11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12  *         outstanding request, and other enhancements.
13  *         Support loadable low-level scsi drivers.
14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15  *         eight major numbers.
16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18  *	   sd_init and cleanups.
19  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
20  *	   not being read in sd_open. Fix problem where removable media
21  *	   could be ejected after sd_open.
22  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25  *	   Support 32k/1M disks.
26  *
27  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
28  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
32  *	Note: when the logging level is set by the user, it must be greater
33  *	than the level indicated above to trigger output.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/major.h>
52 #include <linux/mutex.h>
53 #include <linux/string_helpers.h>
54 #include <linux/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 #define SD_MINORS	16
102 
103 static void sd_config_discard(struct scsi_disk *, unsigned int);
104 static void sd_config_write_same(struct scsi_disk *);
105 static int  sd_revalidate_disk(struct gendisk *);
106 static void sd_unlock_native_capacity(struct gendisk *disk);
107 static int  sd_probe(struct device *);
108 static int  sd_remove(struct device *);
109 static void sd_shutdown(struct device *);
110 static int sd_suspend_system(struct device *);
111 static int sd_suspend_runtime(struct device *);
112 static int sd_resume_system(struct device *);
113 static int sd_resume_runtime(struct device *);
114 static void sd_rescan(struct device *);
115 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
116 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117 static int sd_done(struct scsi_cmnd *);
118 static void sd_eh_reset(struct scsi_cmnd *);
119 static int sd_eh_action(struct scsi_cmnd *, int);
120 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
121 static void scsi_disk_release(struct device *cdev);
122 
123 static DEFINE_IDA(sd_index_ida);
124 
125 /* This semaphore is used to mediate the 0->1 reference get in the
126  * face of object destruction (i.e. we can't allow a get on an
127  * object after last put) */
128 static DEFINE_MUTEX(sd_ref_mutex);
129 
130 static struct kmem_cache *sd_cdb_cache;
131 static mempool_t *sd_cdb_pool;
132 static mempool_t *sd_page_pool;
133 static struct lock_class_key sd_bio_compl_lkclass;
134 
135 static const char *sd_cache_types[] = {
136 	"write through", "none", "write back",
137 	"write back, no read (daft)"
138 };
139 
140 static void sd_set_flush_flag(struct scsi_disk *sdkp)
141 {
142 	bool wc = false, fua = false;
143 
144 	if (sdkp->WCE) {
145 		wc = true;
146 		if (sdkp->DPOFUA)
147 			fua = true;
148 	}
149 
150 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
151 }
152 
153 static ssize_t
154 cache_type_store(struct device *dev, struct device_attribute *attr,
155 		 const char *buf, size_t count)
156 {
157 	int ct, rcd, wce, sp;
158 	struct scsi_disk *sdkp = to_scsi_disk(dev);
159 	struct scsi_device *sdp = sdkp->device;
160 	char buffer[64];
161 	char *buffer_data;
162 	struct scsi_mode_data data;
163 	struct scsi_sense_hdr sshdr;
164 	static const char temp[] = "temporary ";
165 	int len;
166 
167 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
168 		/* no cache control on RBC devices; theoretically they
169 		 * can do it, but there's probably so many exceptions
170 		 * it's not worth the risk */
171 		return -EINVAL;
172 
173 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
174 		buf += sizeof(temp) - 1;
175 		sdkp->cache_override = 1;
176 	} else {
177 		sdkp->cache_override = 0;
178 	}
179 
180 	ct = sysfs_match_string(sd_cache_types, buf);
181 	if (ct < 0)
182 		return -EINVAL;
183 
184 	rcd = ct & 0x01 ? 1 : 0;
185 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
186 
187 	if (sdkp->cache_override) {
188 		sdkp->WCE = wce;
189 		sdkp->RCD = rcd;
190 		sd_set_flush_flag(sdkp);
191 		return count;
192 	}
193 
194 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
195 			    sdkp->max_retries, &data, NULL))
196 		return -EINVAL;
197 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
198 		  data.block_descriptor_length);
199 	buffer_data = buffer + data.header_length +
200 		data.block_descriptor_length;
201 	buffer_data[2] &= ~0x05;
202 	buffer_data[2] |= wce << 2 | rcd;
203 	sp = buffer_data[0] & 0x80 ? 1 : 0;
204 	buffer_data[0] &= ~0x80;
205 
206 	/*
207 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
208 	 * received mode parameter buffer before doing MODE SELECT.
209 	 */
210 	data.device_specific = 0;
211 
212 	if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
213 			     sdkp->max_retries, &data, &sshdr)) {
214 		if (scsi_sense_valid(&sshdr))
215 			sd_print_sense_hdr(sdkp, &sshdr);
216 		return -EINVAL;
217 	}
218 	sd_revalidate_disk(sdkp->disk);
219 	return count;
220 }
221 
222 static ssize_t
223 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
224 		       char *buf)
225 {
226 	struct scsi_disk *sdkp = to_scsi_disk(dev);
227 	struct scsi_device *sdp = sdkp->device;
228 
229 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
230 }
231 
232 static ssize_t
233 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
234 			const char *buf, size_t count)
235 {
236 	struct scsi_disk *sdkp = to_scsi_disk(dev);
237 	struct scsi_device *sdp = sdkp->device;
238 	bool v;
239 
240 	if (!capable(CAP_SYS_ADMIN))
241 		return -EACCES;
242 
243 	if (kstrtobool(buf, &v))
244 		return -EINVAL;
245 
246 	sdp->manage_start_stop = v;
247 
248 	return count;
249 }
250 static DEVICE_ATTR_RW(manage_start_stop);
251 
252 static ssize_t
253 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
254 {
255 	struct scsi_disk *sdkp = to_scsi_disk(dev);
256 
257 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
258 }
259 
260 static ssize_t
261 allow_restart_store(struct device *dev, struct device_attribute *attr,
262 		    const char *buf, size_t count)
263 {
264 	bool v;
265 	struct scsi_disk *sdkp = to_scsi_disk(dev);
266 	struct scsi_device *sdp = sdkp->device;
267 
268 	if (!capable(CAP_SYS_ADMIN))
269 		return -EACCES;
270 
271 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
272 		return -EINVAL;
273 
274 	if (kstrtobool(buf, &v))
275 		return -EINVAL;
276 
277 	sdp->allow_restart = v;
278 
279 	return count;
280 }
281 static DEVICE_ATTR_RW(allow_restart);
282 
283 static ssize_t
284 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
285 {
286 	struct scsi_disk *sdkp = to_scsi_disk(dev);
287 	int ct = sdkp->RCD + 2*sdkp->WCE;
288 
289 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
290 }
291 static DEVICE_ATTR_RW(cache_type);
292 
293 static ssize_t
294 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
295 {
296 	struct scsi_disk *sdkp = to_scsi_disk(dev);
297 
298 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
299 }
300 static DEVICE_ATTR_RO(FUA);
301 
302 static ssize_t
303 protection_type_show(struct device *dev, struct device_attribute *attr,
304 		     char *buf)
305 {
306 	struct scsi_disk *sdkp = to_scsi_disk(dev);
307 
308 	return sprintf(buf, "%u\n", sdkp->protection_type);
309 }
310 
311 static ssize_t
312 protection_type_store(struct device *dev, struct device_attribute *attr,
313 		      const char *buf, size_t count)
314 {
315 	struct scsi_disk *sdkp = to_scsi_disk(dev);
316 	unsigned int val;
317 	int err;
318 
319 	if (!capable(CAP_SYS_ADMIN))
320 		return -EACCES;
321 
322 	err = kstrtouint(buf, 10, &val);
323 
324 	if (err)
325 		return err;
326 
327 	if (val <= T10_PI_TYPE3_PROTECTION)
328 		sdkp->protection_type = val;
329 
330 	return count;
331 }
332 static DEVICE_ATTR_RW(protection_type);
333 
334 static ssize_t
335 protection_mode_show(struct device *dev, struct device_attribute *attr,
336 		     char *buf)
337 {
338 	struct scsi_disk *sdkp = to_scsi_disk(dev);
339 	struct scsi_device *sdp = sdkp->device;
340 	unsigned int dif, dix;
341 
342 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
343 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
344 
345 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
346 		dif = 0;
347 		dix = 1;
348 	}
349 
350 	if (!dif && !dix)
351 		return sprintf(buf, "none\n");
352 
353 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
354 }
355 static DEVICE_ATTR_RO(protection_mode);
356 
357 static ssize_t
358 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
359 {
360 	struct scsi_disk *sdkp = to_scsi_disk(dev);
361 
362 	return sprintf(buf, "%u\n", sdkp->ATO);
363 }
364 static DEVICE_ATTR_RO(app_tag_own);
365 
366 static ssize_t
367 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
368 		       char *buf)
369 {
370 	struct scsi_disk *sdkp = to_scsi_disk(dev);
371 
372 	return sprintf(buf, "%u\n", sdkp->lbpme);
373 }
374 static DEVICE_ATTR_RO(thin_provisioning);
375 
376 /* sysfs_match_string() requires dense arrays */
377 static const char *lbp_mode[] = {
378 	[SD_LBP_FULL]		= "full",
379 	[SD_LBP_UNMAP]		= "unmap",
380 	[SD_LBP_WS16]		= "writesame_16",
381 	[SD_LBP_WS10]		= "writesame_10",
382 	[SD_LBP_ZERO]		= "writesame_zero",
383 	[SD_LBP_DISABLE]	= "disabled",
384 };
385 
386 static ssize_t
387 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
388 		       char *buf)
389 {
390 	struct scsi_disk *sdkp = to_scsi_disk(dev);
391 
392 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
393 }
394 
395 static ssize_t
396 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
397 			const char *buf, size_t count)
398 {
399 	struct scsi_disk *sdkp = to_scsi_disk(dev);
400 	struct scsi_device *sdp = sdkp->device;
401 	int mode;
402 
403 	if (!capable(CAP_SYS_ADMIN))
404 		return -EACCES;
405 
406 	if (sd_is_zoned(sdkp)) {
407 		sd_config_discard(sdkp, SD_LBP_DISABLE);
408 		return count;
409 	}
410 
411 	if (sdp->type != TYPE_DISK)
412 		return -EINVAL;
413 
414 	mode = sysfs_match_string(lbp_mode, buf);
415 	if (mode < 0)
416 		return -EINVAL;
417 
418 	sd_config_discard(sdkp, mode);
419 
420 	return count;
421 }
422 static DEVICE_ATTR_RW(provisioning_mode);
423 
424 /* sysfs_match_string() requires dense arrays */
425 static const char *zeroing_mode[] = {
426 	[SD_ZERO_WRITE]		= "write",
427 	[SD_ZERO_WS]		= "writesame",
428 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
429 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
430 };
431 
432 static ssize_t
433 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
434 		  char *buf)
435 {
436 	struct scsi_disk *sdkp = to_scsi_disk(dev);
437 
438 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
439 }
440 
441 static ssize_t
442 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
443 		   const char *buf, size_t count)
444 {
445 	struct scsi_disk *sdkp = to_scsi_disk(dev);
446 	int mode;
447 
448 	if (!capable(CAP_SYS_ADMIN))
449 		return -EACCES;
450 
451 	mode = sysfs_match_string(zeroing_mode, buf);
452 	if (mode < 0)
453 		return -EINVAL;
454 
455 	sdkp->zeroing_mode = mode;
456 
457 	return count;
458 }
459 static DEVICE_ATTR_RW(zeroing_mode);
460 
461 static ssize_t
462 max_medium_access_timeouts_show(struct device *dev,
463 				struct device_attribute *attr, char *buf)
464 {
465 	struct scsi_disk *sdkp = to_scsi_disk(dev);
466 
467 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
468 }
469 
470 static ssize_t
471 max_medium_access_timeouts_store(struct device *dev,
472 				 struct device_attribute *attr, const char *buf,
473 				 size_t count)
474 {
475 	struct scsi_disk *sdkp = to_scsi_disk(dev);
476 	int err;
477 
478 	if (!capable(CAP_SYS_ADMIN))
479 		return -EACCES;
480 
481 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
482 
483 	return err ? err : count;
484 }
485 static DEVICE_ATTR_RW(max_medium_access_timeouts);
486 
487 static ssize_t
488 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
489 			   char *buf)
490 {
491 	struct scsi_disk *sdkp = to_scsi_disk(dev);
492 
493 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
494 }
495 
496 static ssize_t
497 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
498 			    const char *buf, size_t count)
499 {
500 	struct scsi_disk *sdkp = to_scsi_disk(dev);
501 	struct scsi_device *sdp = sdkp->device;
502 	unsigned long max;
503 	int err;
504 
505 	if (!capable(CAP_SYS_ADMIN))
506 		return -EACCES;
507 
508 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
509 		return -EINVAL;
510 
511 	err = kstrtoul(buf, 10, &max);
512 
513 	if (err)
514 		return err;
515 
516 	if (max == 0)
517 		sdp->no_write_same = 1;
518 	else if (max <= SD_MAX_WS16_BLOCKS) {
519 		sdp->no_write_same = 0;
520 		sdkp->max_ws_blocks = max;
521 	}
522 
523 	sd_config_write_same(sdkp);
524 
525 	return count;
526 }
527 static DEVICE_ATTR_RW(max_write_same_blocks);
528 
529 static ssize_t
530 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
531 {
532 	struct scsi_disk *sdkp = to_scsi_disk(dev);
533 
534 	if (sdkp->device->type == TYPE_ZBC)
535 		return sprintf(buf, "host-managed\n");
536 	if (sdkp->zoned == 1)
537 		return sprintf(buf, "host-aware\n");
538 	if (sdkp->zoned == 2)
539 		return sprintf(buf, "drive-managed\n");
540 	return sprintf(buf, "none\n");
541 }
542 static DEVICE_ATTR_RO(zoned_cap);
543 
544 static ssize_t
545 max_retries_store(struct device *dev, struct device_attribute *attr,
546 		  const char *buf, size_t count)
547 {
548 	struct scsi_disk *sdkp = to_scsi_disk(dev);
549 	struct scsi_device *sdev = sdkp->device;
550 	int retries, err;
551 
552 	err = kstrtoint(buf, 10, &retries);
553 	if (err)
554 		return err;
555 
556 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
557 		sdkp->max_retries = retries;
558 		return count;
559 	}
560 
561 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
562 		    SD_MAX_RETRIES);
563 	return -EINVAL;
564 }
565 
566 static ssize_t
567 max_retries_show(struct device *dev, struct device_attribute *attr,
568 		 char *buf)
569 {
570 	struct scsi_disk *sdkp = to_scsi_disk(dev);
571 
572 	return sprintf(buf, "%d\n", sdkp->max_retries);
573 }
574 
575 static DEVICE_ATTR_RW(max_retries);
576 
577 static struct attribute *sd_disk_attrs[] = {
578 	&dev_attr_cache_type.attr,
579 	&dev_attr_FUA.attr,
580 	&dev_attr_allow_restart.attr,
581 	&dev_attr_manage_start_stop.attr,
582 	&dev_attr_protection_type.attr,
583 	&dev_attr_protection_mode.attr,
584 	&dev_attr_app_tag_own.attr,
585 	&dev_attr_thin_provisioning.attr,
586 	&dev_attr_provisioning_mode.attr,
587 	&dev_attr_zeroing_mode.attr,
588 	&dev_attr_max_write_same_blocks.attr,
589 	&dev_attr_max_medium_access_timeouts.attr,
590 	&dev_attr_zoned_cap.attr,
591 	&dev_attr_max_retries.attr,
592 	NULL,
593 };
594 ATTRIBUTE_GROUPS(sd_disk);
595 
596 static struct class sd_disk_class = {
597 	.name		= "scsi_disk",
598 	.owner		= THIS_MODULE,
599 	.dev_release	= scsi_disk_release,
600 	.dev_groups	= sd_disk_groups,
601 };
602 
603 static const struct dev_pm_ops sd_pm_ops = {
604 	.suspend		= sd_suspend_system,
605 	.resume			= sd_resume_system,
606 	.poweroff		= sd_suspend_system,
607 	.restore		= sd_resume_system,
608 	.runtime_suspend	= sd_suspend_runtime,
609 	.runtime_resume		= sd_resume_runtime,
610 };
611 
612 static struct scsi_driver sd_template = {
613 	.gendrv = {
614 		.name		= "sd",
615 		.owner		= THIS_MODULE,
616 		.probe		= sd_probe,
617 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
618 		.remove		= sd_remove,
619 		.shutdown	= sd_shutdown,
620 		.pm		= &sd_pm_ops,
621 	},
622 	.rescan			= sd_rescan,
623 	.init_command		= sd_init_command,
624 	.uninit_command		= sd_uninit_command,
625 	.done			= sd_done,
626 	.eh_action		= sd_eh_action,
627 	.eh_reset		= sd_eh_reset,
628 };
629 
630 /*
631  * Don't request a new module, as that could deadlock in multipath
632  * environment.
633  */
634 static void sd_default_probe(dev_t devt)
635 {
636 }
637 
638 /*
639  * Device no to disk mapping:
640  *
641  *       major         disc2     disc  p1
642  *   |............|.............|....|....| <- dev_t
643  *    31        20 19          8 7  4 3  0
644  *
645  * Inside a major, we have 16k disks, however mapped non-
646  * contiguously. The first 16 disks are for major0, the next
647  * ones with major1, ... Disk 256 is for major0 again, disk 272
648  * for major1, ...
649  * As we stay compatible with our numbering scheme, we can reuse
650  * the well-know SCSI majors 8, 65--71, 136--143.
651  */
652 static int sd_major(int major_idx)
653 {
654 	switch (major_idx) {
655 	case 0:
656 		return SCSI_DISK0_MAJOR;
657 	case 1 ... 7:
658 		return SCSI_DISK1_MAJOR + major_idx - 1;
659 	case 8 ... 15:
660 		return SCSI_DISK8_MAJOR + major_idx - 8;
661 	default:
662 		BUG();
663 		return 0;	/* shut up gcc */
664 	}
665 }
666 
667 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
668 {
669 	struct scsi_disk *sdkp = NULL;
670 
671 	mutex_lock(&sd_ref_mutex);
672 
673 	if (disk->private_data) {
674 		sdkp = scsi_disk(disk);
675 		if (scsi_device_get(sdkp->device) == 0)
676 			get_device(&sdkp->dev);
677 		else
678 			sdkp = NULL;
679 	}
680 	mutex_unlock(&sd_ref_mutex);
681 	return sdkp;
682 }
683 
684 static void scsi_disk_put(struct scsi_disk *sdkp)
685 {
686 	struct scsi_device *sdev = sdkp->device;
687 
688 	mutex_lock(&sd_ref_mutex);
689 	put_device(&sdkp->dev);
690 	scsi_device_put(sdev);
691 	mutex_unlock(&sd_ref_mutex);
692 }
693 
694 #ifdef CONFIG_BLK_SED_OPAL
695 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
696 		size_t len, bool send)
697 {
698 	struct scsi_disk *sdkp = data;
699 	struct scsi_device *sdev = sdkp->device;
700 	u8 cdb[12] = { 0, };
701 	int ret;
702 
703 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
704 	cdb[1] = secp;
705 	put_unaligned_be16(spsp, &cdb[2]);
706 	put_unaligned_be32(len, &cdb[6]);
707 
708 	ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
709 		buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
710 		RQF_PM, NULL);
711 	return ret <= 0 ? ret : -EIO;
712 }
713 #endif /* CONFIG_BLK_SED_OPAL */
714 
715 /*
716  * Look up the DIX operation based on whether the command is read or
717  * write and whether dix and dif are enabled.
718  */
719 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
720 {
721 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
722 	static const unsigned int ops[] = {	/* wrt dix dif */
723 		SCSI_PROT_NORMAL,		/*  0	0   0  */
724 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
725 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
726 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
727 		SCSI_PROT_NORMAL,		/*  1	0   0  */
728 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
729 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
730 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
731 	};
732 
733 	return ops[write << 2 | dix << 1 | dif];
734 }
735 
736 /*
737  * Returns a mask of the protection flags that are valid for a given DIX
738  * operation.
739  */
740 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
741 {
742 	static const unsigned int flag_mask[] = {
743 		[SCSI_PROT_NORMAL]		= 0,
744 
745 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
746 						  SCSI_PROT_GUARD_CHECK |
747 						  SCSI_PROT_REF_CHECK |
748 						  SCSI_PROT_REF_INCREMENT,
749 
750 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
751 						  SCSI_PROT_IP_CHECKSUM,
752 
753 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
754 						  SCSI_PROT_GUARD_CHECK |
755 						  SCSI_PROT_REF_CHECK |
756 						  SCSI_PROT_REF_INCREMENT |
757 						  SCSI_PROT_IP_CHECKSUM,
758 
759 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
760 						  SCSI_PROT_REF_INCREMENT,
761 
762 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
763 						  SCSI_PROT_REF_CHECK |
764 						  SCSI_PROT_REF_INCREMENT |
765 						  SCSI_PROT_IP_CHECKSUM,
766 
767 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
768 						  SCSI_PROT_GUARD_CHECK |
769 						  SCSI_PROT_REF_CHECK |
770 						  SCSI_PROT_REF_INCREMENT |
771 						  SCSI_PROT_IP_CHECKSUM,
772 	};
773 
774 	return flag_mask[prot_op];
775 }
776 
777 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
778 					   unsigned int dix, unsigned int dif)
779 {
780 	struct request *rq = scsi_cmd_to_rq(scmd);
781 	struct bio *bio = rq->bio;
782 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
783 	unsigned int protect = 0;
784 
785 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
786 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
787 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
788 
789 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
790 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
791 	}
792 
793 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
794 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
795 
796 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
797 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
798 	}
799 
800 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
801 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
802 
803 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
804 			protect = 3 << 5;	/* Disable target PI checking */
805 		else
806 			protect = 1 << 5;	/* Enable target PI checking */
807 	}
808 
809 	scsi_set_prot_op(scmd, prot_op);
810 	scsi_set_prot_type(scmd, dif);
811 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
812 
813 	return protect;
814 }
815 
816 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
817 {
818 	struct request_queue *q = sdkp->disk->queue;
819 	unsigned int logical_block_size = sdkp->device->sector_size;
820 	unsigned int max_blocks = 0;
821 
822 	q->limits.discard_alignment =
823 		sdkp->unmap_alignment * logical_block_size;
824 	q->limits.discard_granularity =
825 		max(sdkp->physical_block_size,
826 		    sdkp->unmap_granularity * logical_block_size);
827 	sdkp->provisioning_mode = mode;
828 
829 	switch (mode) {
830 
831 	case SD_LBP_FULL:
832 	case SD_LBP_DISABLE:
833 		blk_queue_max_discard_sectors(q, 0);
834 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
835 		return;
836 
837 	case SD_LBP_UNMAP:
838 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
839 					  (u32)SD_MAX_WS16_BLOCKS);
840 		break;
841 
842 	case SD_LBP_WS16:
843 		if (sdkp->device->unmap_limit_for_ws)
844 			max_blocks = sdkp->max_unmap_blocks;
845 		else
846 			max_blocks = sdkp->max_ws_blocks;
847 
848 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
849 		break;
850 
851 	case SD_LBP_WS10:
852 		if (sdkp->device->unmap_limit_for_ws)
853 			max_blocks = sdkp->max_unmap_blocks;
854 		else
855 			max_blocks = sdkp->max_ws_blocks;
856 
857 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
858 		break;
859 
860 	case SD_LBP_ZERO:
861 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
862 					  (u32)SD_MAX_WS10_BLOCKS);
863 		break;
864 	}
865 
866 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
867 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
868 }
869 
870 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
871 {
872 	struct scsi_device *sdp = cmd->device;
873 	struct request *rq = scsi_cmd_to_rq(cmd);
874 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
875 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
876 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
877 	unsigned int data_len = 24;
878 	char *buf;
879 
880 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
881 	if (!rq->special_vec.bv_page)
882 		return BLK_STS_RESOURCE;
883 	clear_highpage(rq->special_vec.bv_page);
884 	rq->special_vec.bv_offset = 0;
885 	rq->special_vec.bv_len = data_len;
886 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
887 
888 	cmd->cmd_len = 10;
889 	cmd->cmnd[0] = UNMAP;
890 	cmd->cmnd[8] = 24;
891 
892 	buf = bvec_virt(&rq->special_vec);
893 	put_unaligned_be16(6 + 16, &buf[0]);
894 	put_unaligned_be16(16, &buf[2]);
895 	put_unaligned_be64(lba, &buf[8]);
896 	put_unaligned_be32(nr_blocks, &buf[16]);
897 
898 	cmd->allowed = sdkp->max_retries;
899 	cmd->transfersize = data_len;
900 	rq->timeout = SD_TIMEOUT;
901 
902 	return scsi_alloc_sgtables(cmd);
903 }
904 
905 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
906 		bool unmap)
907 {
908 	struct scsi_device *sdp = cmd->device;
909 	struct request *rq = scsi_cmd_to_rq(cmd);
910 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
911 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
912 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
913 	u32 data_len = sdp->sector_size;
914 
915 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
916 	if (!rq->special_vec.bv_page)
917 		return BLK_STS_RESOURCE;
918 	clear_highpage(rq->special_vec.bv_page);
919 	rq->special_vec.bv_offset = 0;
920 	rq->special_vec.bv_len = data_len;
921 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
922 
923 	cmd->cmd_len = 16;
924 	cmd->cmnd[0] = WRITE_SAME_16;
925 	if (unmap)
926 		cmd->cmnd[1] = 0x8; /* UNMAP */
927 	put_unaligned_be64(lba, &cmd->cmnd[2]);
928 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
929 
930 	cmd->allowed = sdkp->max_retries;
931 	cmd->transfersize = data_len;
932 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
933 
934 	return scsi_alloc_sgtables(cmd);
935 }
936 
937 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
938 		bool unmap)
939 {
940 	struct scsi_device *sdp = cmd->device;
941 	struct request *rq = scsi_cmd_to_rq(cmd);
942 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
943 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
944 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
945 	u32 data_len = sdp->sector_size;
946 
947 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
948 	if (!rq->special_vec.bv_page)
949 		return BLK_STS_RESOURCE;
950 	clear_highpage(rq->special_vec.bv_page);
951 	rq->special_vec.bv_offset = 0;
952 	rq->special_vec.bv_len = data_len;
953 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
954 
955 	cmd->cmd_len = 10;
956 	cmd->cmnd[0] = WRITE_SAME;
957 	if (unmap)
958 		cmd->cmnd[1] = 0x8; /* UNMAP */
959 	put_unaligned_be32(lba, &cmd->cmnd[2]);
960 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
961 
962 	cmd->allowed = sdkp->max_retries;
963 	cmd->transfersize = data_len;
964 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
965 
966 	return scsi_alloc_sgtables(cmd);
967 }
968 
969 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
970 {
971 	struct request *rq = scsi_cmd_to_rq(cmd);
972 	struct scsi_device *sdp = cmd->device;
973 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
974 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
975 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
976 
977 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
978 		switch (sdkp->zeroing_mode) {
979 		case SD_ZERO_WS16_UNMAP:
980 			return sd_setup_write_same16_cmnd(cmd, true);
981 		case SD_ZERO_WS10_UNMAP:
982 			return sd_setup_write_same10_cmnd(cmd, true);
983 		}
984 	}
985 
986 	if (sdp->no_write_same) {
987 		rq->rq_flags |= RQF_QUIET;
988 		return BLK_STS_TARGET;
989 	}
990 
991 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
992 		return sd_setup_write_same16_cmnd(cmd, false);
993 
994 	return sd_setup_write_same10_cmnd(cmd, false);
995 }
996 
997 static void sd_config_write_same(struct scsi_disk *sdkp)
998 {
999 	struct request_queue *q = sdkp->disk->queue;
1000 	unsigned int logical_block_size = sdkp->device->sector_size;
1001 
1002 	if (sdkp->device->no_write_same) {
1003 		sdkp->max_ws_blocks = 0;
1004 		goto out;
1005 	}
1006 
1007 	/* Some devices can not handle block counts above 0xffff despite
1008 	 * supporting WRITE SAME(16). Consequently we default to 64k
1009 	 * blocks per I/O unless the device explicitly advertises a
1010 	 * bigger limit.
1011 	 */
1012 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1013 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1014 						   (u32)SD_MAX_WS16_BLOCKS);
1015 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1016 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1017 						   (u32)SD_MAX_WS10_BLOCKS);
1018 	else {
1019 		sdkp->device->no_write_same = 1;
1020 		sdkp->max_ws_blocks = 0;
1021 	}
1022 
1023 	if (sdkp->lbprz && sdkp->lbpws)
1024 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1025 	else if (sdkp->lbprz && sdkp->lbpws10)
1026 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1027 	else if (sdkp->max_ws_blocks)
1028 		sdkp->zeroing_mode = SD_ZERO_WS;
1029 	else
1030 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1031 
1032 	if (sdkp->max_ws_blocks &&
1033 	    sdkp->physical_block_size > logical_block_size) {
1034 		/*
1035 		 * Reporting a maximum number of blocks that is not aligned
1036 		 * on the device physical size would cause a large write same
1037 		 * request to be split into physically unaligned chunks by
1038 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1039 		 * even if the caller of these functions took care to align the
1040 		 * large request. So make sure the maximum reported is aligned
1041 		 * to the device physical block size. This is only an optional
1042 		 * optimization for regular disks, but this is mandatory to
1043 		 * avoid failure of large write same requests directed at
1044 		 * sequential write required zones of host-managed ZBC disks.
1045 		 */
1046 		sdkp->max_ws_blocks =
1047 			round_down(sdkp->max_ws_blocks,
1048 				   bytes_to_logical(sdkp->device,
1049 						    sdkp->physical_block_size));
1050 	}
1051 
1052 out:
1053 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1054 					 (logical_block_size >> 9));
1055 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1056 					 (logical_block_size >> 9));
1057 }
1058 
1059 /**
1060  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1061  * @cmd: command to prepare
1062  *
1063  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1064  * the preference indicated by the target device.
1065  **/
1066 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1067 {
1068 	struct request *rq = scsi_cmd_to_rq(cmd);
1069 	struct scsi_device *sdp = cmd->device;
1070 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1071 	struct bio *bio = rq->bio;
1072 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1073 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1074 	blk_status_t ret;
1075 
1076 	if (sdkp->device->no_write_same)
1077 		return BLK_STS_TARGET;
1078 
1079 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1080 
1081 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1082 
1083 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1084 		cmd->cmd_len = 16;
1085 		cmd->cmnd[0] = WRITE_SAME_16;
1086 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1087 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1088 	} else {
1089 		cmd->cmd_len = 10;
1090 		cmd->cmnd[0] = WRITE_SAME;
1091 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1092 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1093 	}
1094 
1095 	cmd->transfersize = sdp->sector_size;
1096 	cmd->allowed = sdkp->max_retries;
1097 
1098 	/*
1099 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1100 	 * different from the amount of data actually written to the target.
1101 	 *
1102 	 * We set up __data_len to the amount of data transferred via the
1103 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1104 	 * to transfer a single sector of data first, but then reset it to
1105 	 * the amount of data to be written right after so that the I/O path
1106 	 * knows how much to actually write.
1107 	 */
1108 	rq->__data_len = sdp->sector_size;
1109 	ret = scsi_alloc_sgtables(cmd);
1110 	rq->__data_len = blk_rq_bytes(rq);
1111 
1112 	return ret;
1113 }
1114 
1115 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1116 {
1117 	struct request *rq = scsi_cmd_to_rq(cmd);
1118 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1119 
1120 	/* flush requests don't perform I/O, zero the S/G table */
1121 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1122 
1123 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1124 	cmd->cmd_len = 10;
1125 	cmd->transfersize = 0;
1126 	cmd->allowed = sdkp->max_retries;
1127 
1128 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1129 	return BLK_STS_OK;
1130 }
1131 
1132 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1133 				       sector_t lba, unsigned int nr_blocks,
1134 				       unsigned char flags)
1135 {
1136 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1137 	if (unlikely(cmd->cmnd == NULL))
1138 		return BLK_STS_RESOURCE;
1139 
1140 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1141 	memset(cmd->cmnd, 0, cmd->cmd_len);
1142 
1143 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1144 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1145 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1146 	cmd->cmnd[10] = flags;
1147 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1148 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1149 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1150 
1151 	return BLK_STS_OK;
1152 }
1153 
1154 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1155 				       sector_t lba, unsigned int nr_blocks,
1156 				       unsigned char flags)
1157 {
1158 	cmd->cmd_len  = 16;
1159 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1160 	cmd->cmnd[1]  = flags;
1161 	cmd->cmnd[14] = 0;
1162 	cmd->cmnd[15] = 0;
1163 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1164 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1165 
1166 	return BLK_STS_OK;
1167 }
1168 
1169 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1170 				       sector_t lba, unsigned int nr_blocks,
1171 				       unsigned char flags)
1172 {
1173 	cmd->cmd_len = 10;
1174 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1175 	cmd->cmnd[1] = flags;
1176 	cmd->cmnd[6] = 0;
1177 	cmd->cmnd[9] = 0;
1178 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1179 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1180 
1181 	return BLK_STS_OK;
1182 }
1183 
1184 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1185 				      sector_t lba, unsigned int nr_blocks,
1186 				      unsigned char flags)
1187 {
1188 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1189 	if (WARN_ON_ONCE(nr_blocks == 0))
1190 		return BLK_STS_IOERR;
1191 
1192 	if (unlikely(flags & 0x8)) {
1193 		/*
1194 		 * This happens only if this drive failed 10byte rw
1195 		 * command with ILLEGAL_REQUEST during operation and
1196 		 * thus turned off use_10_for_rw.
1197 		 */
1198 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1199 		return BLK_STS_IOERR;
1200 	}
1201 
1202 	cmd->cmd_len = 6;
1203 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1204 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1205 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1206 	cmd->cmnd[3] = lba & 0xff;
1207 	cmd->cmnd[4] = nr_blocks;
1208 	cmd->cmnd[5] = 0;
1209 
1210 	return BLK_STS_OK;
1211 }
1212 
1213 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1214 {
1215 	struct request *rq = scsi_cmd_to_rq(cmd);
1216 	struct scsi_device *sdp = cmd->device;
1217 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1218 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1219 	sector_t threshold;
1220 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1221 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1222 	bool write = rq_data_dir(rq) == WRITE;
1223 	unsigned char protect, fua;
1224 	blk_status_t ret;
1225 	unsigned int dif;
1226 	bool dix;
1227 
1228 	ret = scsi_alloc_sgtables(cmd);
1229 	if (ret != BLK_STS_OK)
1230 		return ret;
1231 
1232 	ret = BLK_STS_IOERR;
1233 	if (!scsi_device_online(sdp) || sdp->changed) {
1234 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1235 		goto fail;
1236 	}
1237 
1238 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1239 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1240 		goto fail;
1241 	}
1242 
1243 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1244 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1245 		goto fail;
1246 	}
1247 
1248 	/*
1249 	 * Some SD card readers can't handle accesses which touch the
1250 	 * last one or two logical blocks. Split accesses as needed.
1251 	 */
1252 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1253 
1254 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1255 		if (lba < threshold) {
1256 			/* Access up to the threshold but not beyond */
1257 			nr_blocks = threshold - lba;
1258 		} else {
1259 			/* Access only a single logical block */
1260 			nr_blocks = 1;
1261 		}
1262 	}
1263 
1264 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1265 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1266 		if (ret)
1267 			goto fail;
1268 	}
1269 
1270 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1271 	dix = scsi_prot_sg_count(cmd);
1272 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1273 
1274 	if (dif || dix)
1275 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1276 	else
1277 		protect = 0;
1278 
1279 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1280 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1281 					 protect | fua);
1282 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1283 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1284 					 protect | fua);
1285 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1286 		   sdp->use_10_for_rw || protect) {
1287 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1288 					 protect | fua);
1289 	} else {
1290 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1291 					protect | fua);
1292 	}
1293 
1294 	if (unlikely(ret != BLK_STS_OK))
1295 		goto fail;
1296 
1297 	/*
1298 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1299 	 * host adapter, it's safe to assume that we can at least transfer
1300 	 * this many bytes between each connect / disconnect.
1301 	 */
1302 	cmd->transfersize = sdp->sector_size;
1303 	cmd->underflow = nr_blocks << 9;
1304 	cmd->allowed = sdkp->max_retries;
1305 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1306 
1307 	SCSI_LOG_HLQUEUE(1,
1308 			 scmd_printk(KERN_INFO, cmd,
1309 				     "%s: block=%llu, count=%d\n", __func__,
1310 				     (unsigned long long)blk_rq_pos(rq),
1311 				     blk_rq_sectors(rq)));
1312 	SCSI_LOG_HLQUEUE(2,
1313 			 scmd_printk(KERN_INFO, cmd,
1314 				     "%s %d/%u 512 byte blocks.\n",
1315 				     write ? "writing" : "reading", nr_blocks,
1316 				     blk_rq_sectors(rq)));
1317 
1318 	/*
1319 	 * This indicates that the command is ready from our end to be queued.
1320 	 */
1321 	return BLK_STS_OK;
1322 fail:
1323 	scsi_free_sgtables(cmd);
1324 	return ret;
1325 }
1326 
1327 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1328 {
1329 	struct request *rq = scsi_cmd_to_rq(cmd);
1330 
1331 	switch (req_op(rq)) {
1332 	case REQ_OP_DISCARD:
1333 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1334 		case SD_LBP_UNMAP:
1335 			return sd_setup_unmap_cmnd(cmd);
1336 		case SD_LBP_WS16:
1337 			return sd_setup_write_same16_cmnd(cmd, true);
1338 		case SD_LBP_WS10:
1339 			return sd_setup_write_same10_cmnd(cmd, true);
1340 		case SD_LBP_ZERO:
1341 			return sd_setup_write_same10_cmnd(cmd, false);
1342 		default:
1343 			return BLK_STS_TARGET;
1344 		}
1345 	case REQ_OP_WRITE_ZEROES:
1346 		return sd_setup_write_zeroes_cmnd(cmd);
1347 	case REQ_OP_WRITE_SAME:
1348 		return sd_setup_write_same_cmnd(cmd);
1349 	case REQ_OP_FLUSH:
1350 		return sd_setup_flush_cmnd(cmd);
1351 	case REQ_OP_READ:
1352 	case REQ_OP_WRITE:
1353 	case REQ_OP_ZONE_APPEND:
1354 		return sd_setup_read_write_cmnd(cmd);
1355 	case REQ_OP_ZONE_RESET:
1356 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1357 						   false);
1358 	case REQ_OP_ZONE_RESET_ALL:
1359 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1360 						   true);
1361 	case REQ_OP_ZONE_OPEN:
1362 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1363 	case REQ_OP_ZONE_CLOSE:
1364 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1365 	case REQ_OP_ZONE_FINISH:
1366 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1367 	default:
1368 		WARN_ON_ONCE(1);
1369 		return BLK_STS_NOTSUPP;
1370 	}
1371 }
1372 
1373 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1374 {
1375 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1376 	u8 *cmnd;
1377 
1378 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1379 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1380 
1381 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1382 		cmnd = SCpnt->cmnd;
1383 		SCpnt->cmnd = NULL;
1384 		SCpnt->cmd_len = 0;
1385 		mempool_free(cmnd, sd_cdb_pool);
1386 	}
1387 }
1388 
1389 static bool sd_need_revalidate(struct block_device *bdev,
1390 		struct scsi_disk *sdkp)
1391 {
1392 	if (sdkp->device->removable || sdkp->write_prot) {
1393 		if (bdev_check_media_change(bdev))
1394 			return true;
1395 	}
1396 
1397 	/*
1398 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1399 	 * nothing to do with partitions, BLKRRPART is used to force a full
1400 	 * revalidate after things like a format for historical reasons.
1401 	 */
1402 	return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1403 }
1404 
1405 /**
1406  *	sd_open - open a scsi disk device
1407  *	@bdev: Block device of the scsi disk to open
1408  *	@mode: FMODE_* mask
1409  *
1410  *	Returns 0 if successful. Returns a negated errno value in case
1411  *	of error.
1412  *
1413  *	Note: This can be called from a user context (e.g. fsck(1) )
1414  *	or from within the kernel (e.g. as a result of a mount(1) ).
1415  *	In the latter case @inode and @filp carry an abridged amount
1416  *	of information as noted above.
1417  *
1418  *	Locking: called with bdev->bd_disk->open_mutex held.
1419  **/
1420 static int sd_open(struct block_device *bdev, fmode_t mode)
1421 {
1422 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1423 	struct scsi_device *sdev;
1424 	int retval;
1425 
1426 	if (!sdkp)
1427 		return -ENXIO;
1428 
1429 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1430 
1431 	sdev = sdkp->device;
1432 
1433 	/*
1434 	 * If the device is in error recovery, wait until it is done.
1435 	 * If the device is offline, then disallow any access to it.
1436 	 */
1437 	retval = -ENXIO;
1438 	if (!scsi_block_when_processing_errors(sdev))
1439 		goto error_out;
1440 
1441 	if (sd_need_revalidate(bdev, sdkp))
1442 		sd_revalidate_disk(bdev->bd_disk);
1443 
1444 	/*
1445 	 * If the drive is empty, just let the open fail.
1446 	 */
1447 	retval = -ENOMEDIUM;
1448 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1449 		goto error_out;
1450 
1451 	/*
1452 	 * If the device has the write protect tab set, have the open fail
1453 	 * if the user expects to be able to write to the thing.
1454 	 */
1455 	retval = -EROFS;
1456 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1457 		goto error_out;
1458 
1459 	/*
1460 	 * It is possible that the disk changing stuff resulted in
1461 	 * the device being taken offline.  If this is the case,
1462 	 * report this to the user, and don't pretend that the
1463 	 * open actually succeeded.
1464 	 */
1465 	retval = -ENXIO;
1466 	if (!scsi_device_online(sdev))
1467 		goto error_out;
1468 
1469 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1470 		if (scsi_block_when_processing_errors(sdev))
1471 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1472 	}
1473 
1474 	return 0;
1475 
1476 error_out:
1477 	scsi_disk_put(sdkp);
1478 	return retval;
1479 }
1480 
1481 /**
1482  *	sd_release - invoked when the (last) close(2) is called on this
1483  *	scsi disk.
1484  *	@disk: disk to release
1485  *	@mode: FMODE_* mask
1486  *
1487  *	Returns 0.
1488  *
1489  *	Note: may block (uninterruptible) if error recovery is underway
1490  *	on this disk.
1491  *
1492  *	Locking: called with bdev->bd_disk->open_mutex held.
1493  **/
1494 static void sd_release(struct gendisk *disk, fmode_t mode)
1495 {
1496 	struct scsi_disk *sdkp = scsi_disk(disk);
1497 	struct scsi_device *sdev = sdkp->device;
1498 
1499 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1500 
1501 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1502 		if (scsi_block_when_processing_errors(sdev))
1503 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1504 	}
1505 
1506 	scsi_disk_put(sdkp);
1507 }
1508 
1509 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1510 {
1511 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1512 	struct scsi_device *sdp = sdkp->device;
1513 	struct Scsi_Host *host = sdp->host;
1514 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1515 	int diskinfo[4];
1516 
1517 	/* default to most commonly used values */
1518 	diskinfo[0] = 0x40;	/* 1 << 6 */
1519 	diskinfo[1] = 0x20;	/* 1 << 5 */
1520 	diskinfo[2] = capacity >> 11;
1521 
1522 	/* override with calculated, extended default, or driver values */
1523 	if (host->hostt->bios_param)
1524 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1525 	else
1526 		scsicam_bios_param(bdev, capacity, diskinfo);
1527 
1528 	geo->heads = diskinfo[0];
1529 	geo->sectors = diskinfo[1];
1530 	geo->cylinders = diskinfo[2];
1531 	return 0;
1532 }
1533 
1534 /**
1535  *	sd_ioctl - process an ioctl
1536  *	@bdev: target block device
1537  *	@mode: FMODE_* mask
1538  *	@cmd: ioctl command number
1539  *	@arg: this is third argument given to ioctl(2) system call.
1540  *	Often contains a pointer.
1541  *
1542  *	Returns 0 if successful (some ioctls return positive numbers on
1543  *	success as well). Returns a negated errno value in case of error.
1544  *
1545  *	Note: most ioctls are forward onto the block subsystem or further
1546  *	down in the scsi subsystem.
1547  **/
1548 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1549 		    unsigned int cmd, unsigned long arg)
1550 {
1551 	struct gendisk *disk = bdev->bd_disk;
1552 	struct scsi_disk *sdkp = scsi_disk(disk);
1553 	struct scsi_device *sdp = sdkp->device;
1554 	void __user *p = (void __user *)arg;
1555 	int error;
1556 
1557 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1558 				    "cmd=0x%x\n", disk->disk_name, cmd));
1559 
1560 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1561 		return -ENOIOCTLCMD;
1562 
1563 	/*
1564 	 * If we are in the middle of error recovery, don't let anyone
1565 	 * else try and use this device.  Also, if error recovery fails, it
1566 	 * may try and take the device offline, in which case all further
1567 	 * access to the device is prohibited.
1568 	 */
1569 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1570 			(mode & FMODE_NDELAY) != 0);
1571 	if (error)
1572 		return error;
1573 
1574 	if (is_sed_ioctl(cmd))
1575 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1576 	return scsi_ioctl(sdp, mode, cmd, p);
1577 }
1578 
1579 static void set_media_not_present(struct scsi_disk *sdkp)
1580 {
1581 	if (sdkp->media_present)
1582 		sdkp->device->changed = 1;
1583 
1584 	if (sdkp->device->removable) {
1585 		sdkp->media_present = 0;
1586 		sdkp->capacity = 0;
1587 	}
1588 }
1589 
1590 static int media_not_present(struct scsi_disk *sdkp,
1591 			     struct scsi_sense_hdr *sshdr)
1592 {
1593 	if (!scsi_sense_valid(sshdr))
1594 		return 0;
1595 
1596 	/* not invoked for commands that could return deferred errors */
1597 	switch (sshdr->sense_key) {
1598 	case UNIT_ATTENTION:
1599 	case NOT_READY:
1600 		/* medium not present */
1601 		if (sshdr->asc == 0x3A) {
1602 			set_media_not_present(sdkp);
1603 			return 1;
1604 		}
1605 	}
1606 	return 0;
1607 }
1608 
1609 /**
1610  *	sd_check_events - check media events
1611  *	@disk: kernel device descriptor
1612  *	@clearing: disk events currently being cleared
1613  *
1614  *	Returns mask of DISK_EVENT_*.
1615  *
1616  *	Note: this function is invoked from the block subsystem.
1617  **/
1618 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1619 {
1620 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1621 	struct scsi_device *sdp;
1622 	int retval;
1623 	bool disk_changed;
1624 
1625 	if (!sdkp)
1626 		return 0;
1627 
1628 	sdp = sdkp->device;
1629 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1630 
1631 	/*
1632 	 * If the device is offline, don't send any commands - just pretend as
1633 	 * if the command failed.  If the device ever comes back online, we
1634 	 * can deal with it then.  It is only because of unrecoverable errors
1635 	 * that we would ever take a device offline in the first place.
1636 	 */
1637 	if (!scsi_device_online(sdp)) {
1638 		set_media_not_present(sdkp);
1639 		goto out;
1640 	}
1641 
1642 	/*
1643 	 * Using TEST_UNIT_READY enables differentiation between drive with
1644 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1645 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1646 	 *
1647 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1648 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1649 	 * sd_revalidate() is called.
1650 	 */
1651 	if (scsi_block_when_processing_errors(sdp)) {
1652 		struct scsi_sense_hdr sshdr = { 0, };
1653 
1654 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1655 					      &sshdr);
1656 
1657 		/* failed to execute TUR, assume media not present */
1658 		if (retval < 0 || host_byte(retval)) {
1659 			set_media_not_present(sdkp);
1660 			goto out;
1661 		}
1662 
1663 		if (media_not_present(sdkp, &sshdr))
1664 			goto out;
1665 	}
1666 
1667 	/*
1668 	 * For removable scsi disk we have to recognise the presence
1669 	 * of a disk in the drive.
1670 	 */
1671 	if (!sdkp->media_present)
1672 		sdp->changed = 1;
1673 	sdkp->media_present = 1;
1674 out:
1675 	/*
1676 	 * sdp->changed is set under the following conditions:
1677 	 *
1678 	 *	Medium present state has changed in either direction.
1679 	 *	Device has indicated UNIT_ATTENTION.
1680 	 */
1681 	disk_changed = sdp->changed;
1682 	sdp->changed = 0;
1683 	scsi_disk_put(sdkp);
1684 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1685 }
1686 
1687 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1688 {
1689 	int retries, res;
1690 	struct scsi_device *sdp = sdkp->device;
1691 	const int timeout = sdp->request_queue->rq_timeout
1692 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1693 	struct scsi_sense_hdr my_sshdr;
1694 
1695 	if (!scsi_device_online(sdp))
1696 		return -ENODEV;
1697 
1698 	/* caller might not be interested in sense, but we need it */
1699 	if (!sshdr)
1700 		sshdr = &my_sshdr;
1701 
1702 	for (retries = 3; retries > 0; --retries) {
1703 		unsigned char cmd[10] = { 0 };
1704 
1705 		cmd[0] = SYNCHRONIZE_CACHE;
1706 		/*
1707 		 * Leave the rest of the command zero to indicate
1708 		 * flush everything.
1709 		 */
1710 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1711 				timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1712 		if (res == 0)
1713 			break;
1714 	}
1715 
1716 	if (res) {
1717 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1718 
1719 		if (res < 0)
1720 			return res;
1721 
1722 		if (scsi_status_is_check_condition(res) &&
1723 		    scsi_sense_valid(sshdr)) {
1724 			sd_print_sense_hdr(sdkp, sshdr);
1725 
1726 			/* we need to evaluate the error return  */
1727 			if (sshdr->asc == 0x3a ||	/* medium not present */
1728 			    sshdr->asc == 0x20 ||	/* invalid command */
1729 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1730 				/* this is no error here */
1731 				return 0;
1732 		}
1733 
1734 		switch (host_byte(res)) {
1735 		/* ignore errors due to racing a disconnection */
1736 		case DID_BAD_TARGET:
1737 		case DID_NO_CONNECT:
1738 			return 0;
1739 		/* signal the upper layer it might try again */
1740 		case DID_BUS_BUSY:
1741 		case DID_IMM_RETRY:
1742 		case DID_REQUEUE:
1743 		case DID_SOFT_ERROR:
1744 			return -EBUSY;
1745 		default:
1746 			return -EIO;
1747 		}
1748 	}
1749 	return 0;
1750 }
1751 
1752 static void sd_rescan(struct device *dev)
1753 {
1754 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1755 
1756 	sd_revalidate_disk(sdkp->disk);
1757 }
1758 
1759 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1760 		enum blk_unique_id type)
1761 {
1762 	struct scsi_device *sdev = scsi_disk(disk)->device;
1763 	const struct scsi_vpd *vpd;
1764 	const unsigned char *d;
1765 	int ret = -ENXIO, len;
1766 
1767 	rcu_read_lock();
1768 	vpd = rcu_dereference(sdev->vpd_pg83);
1769 	if (!vpd)
1770 		goto out_unlock;
1771 
1772 	ret = -EINVAL;
1773 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1774 		/* we only care about designators with LU association */
1775 		if (((d[1] >> 4) & 0x3) != 0x00)
1776 			continue;
1777 		if ((d[1] & 0xf) != type)
1778 			continue;
1779 
1780 		/*
1781 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1782 		 * keep looking as one with more entropy might still show up.
1783 		 */
1784 		len = d[3];
1785 		if (len != 8 && len != 12 && len != 16)
1786 			continue;
1787 		ret = len;
1788 		memcpy(id, d + 4, len);
1789 		if (len == 16)
1790 			break;
1791 	}
1792 out_unlock:
1793 	rcu_read_unlock();
1794 	return ret;
1795 }
1796 
1797 static char sd_pr_type(enum pr_type type)
1798 {
1799 	switch (type) {
1800 	case PR_WRITE_EXCLUSIVE:
1801 		return 0x01;
1802 	case PR_EXCLUSIVE_ACCESS:
1803 		return 0x03;
1804 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1805 		return 0x05;
1806 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1807 		return 0x06;
1808 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1809 		return 0x07;
1810 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1811 		return 0x08;
1812 	default:
1813 		return 0;
1814 	}
1815 };
1816 
1817 static int sd_pr_command(struct block_device *bdev, u8 sa,
1818 		u64 key, u64 sa_key, u8 type, u8 flags)
1819 {
1820 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1821 	struct scsi_device *sdev = sdkp->device;
1822 	struct scsi_sense_hdr sshdr;
1823 	int result;
1824 	u8 cmd[16] = { 0, };
1825 	u8 data[24] = { 0, };
1826 
1827 	cmd[0] = PERSISTENT_RESERVE_OUT;
1828 	cmd[1] = sa;
1829 	cmd[2] = type;
1830 	put_unaligned_be32(sizeof(data), &cmd[5]);
1831 
1832 	put_unaligned_be64(key, &data[0]);
1833 	put_unaligned_be64(sa_key, &data[8]);
1834 	data[20] = flags;
1835 
1836 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1837 			&sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1838 
1839 	if (scsi_status_is_check_condition(result) &&
1840 	    scsi_sense_valid(&sshdr)) {
1841 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1842 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1843 	}
1844 
1845 	return result;
1846 }
1847 
1848 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1849 		u32 flags)
1850 {
1851 	if (flags & ~PR_FL_IGNORE_KEY)
1852 		return -EOPNOTSUPP;
1853 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1854 			old_key, new_key, 0,
1855 			(1 << 0) /* APTPL */);
1856 }
1857 
1858 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1859 		u32 flags)
1860 {
1861 	if (flags)
1862 		return -EOPNOTSUPP;
1863 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1864 }
1865 
1866 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1867 {
1868 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1869 }
1870 
1871 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1872 		enum pr_type type, bool abort)
1873 {
1874 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1875 			     sd_pr_type(type), 0);
1876 }
1877 
1878 static int sd_pr_clear(struct block_device *bdev, u64 key)
1879 {
1880 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1881 }
1882 
1883 static const struct pr_ops sd_pr_ops = {
1884 	.pr_register	= sd_pr_register,
1885 	.pr_reserve	= sd_pr_reserve,
1886 	.pr_release	= sd_pr_release,
1887 	.pr_preempt	= sd_pr_preempt,
1888 	.pr_clear	= sd_pr_clear,
1889 };
1890 
1891 static const struct block_device_operations sd_fops = {
1892 	.owner			= THIS_MODULE,
1893 	.open			= sd_open,
1894 	.release		= sd_release,
1895 	.ioctl			= sd_ioctl,
1896 	.getgeo			= sd_getgeo,
1897 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1898 	.check_events		= sd_check_events,
1899 	.unlock_native_capacity	= sd_unlock_native_capacity,
1900 	.report_zones		= sd_zbc_report_zones,
1901 	.get_unique_id		= sd_get_unique_id,
1902 	.pr_ops			= &sd_pr_ops,
1903 };
1904 
1905 /**
1906  *	sd_eh_reset - reset error handling callback
1907  *	@scmd:		sd-issued command that has failed
1908  *
1909  *	This function is called by the SCSI midlayer before starting
1910  *	SCSI EH. When counting medium access failures we have to be
1911  *	careful to register it only only once per device and SCSI EH run;
1912  *	there might be several timed out commands which will cause the
1913  *	'max_medium_access_timeouts' counter to trigger after the first
1914  *	SCSI EH run already and set the device to offline.
1915  *	So this function resets the internal counter before starting SCSI EH.
1916  **/
1917 static void sd_eh_reset(struct scsi_cmnd *scmd)
1918 {
1919 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1920 
1921 	/* New SCSI EH run, reset gate variable */
1922 	sdkp->ignore_medium_access_errors = false;
1923 }
1924 
1925 /**
1926  *	sd_eh_action - error handling callback
1927  *	@scmd:		sd-issued command that has failed
1928  *	@eh_disp:	The recovery disposition suggested by the midlayer
1929  *
1930  *	This function is called by the SCSI midlayer upon completion of an
1931  *	error test command (currently TEST UNIT READY). The result of sending
1932  *	the eh command is passed in eh_disp.  We're looking for devices that
1933  *	fail medium access commands but are OK with non access commands like
1934  *	test unit ready (so wrongly see the device as having a successful
1935  *	recovery)
1936  **/
1937 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1938 {
1939 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1940 	struct scsi_device *sdev = scmd->device;
1941 
1942 	if (!scsi_device_online(sdev) ||
1943 	    !scsi_medium_access_command(scmd) ||
1944 	    host_byte(scmd->result) != DID_TIME_OUT ||
1945 	    eh_disp != SUCCESS)
1946 		return eh_disp;
1947 
1948 	/*
1949 	 * The device has timed out executing a medium access command.
1950 	 * However, the TEST UNIT READY command sent during error
1951 	 * handling completed successfully. Either the device is in the
1952 	 * process of recovering or has it suffered an internal failure
1953 	 * that prevents access to the storage medium.
1954 	 */
1955 	if (!sdkp->ignore_medium_access_errors) {
1956 		sdkp->medium_access_timed_out++;
1957 		sdkp->ignore_medium_access_errors = true;
1958 	}
1959 
1960 	/*
1961 	 * If the device keeps failing read/write commands but TEST UNIT
1962 	 * READY always completes successfully we assume that medium
1963 	 * access is no longer possible and take the device offline.
1964 	 */
1965 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1966 		scmd_printk(KERN_ERR, scmd,
1967 			    "Medium access timeout failure. Offlining disk!\n");
1968 		mutex_lock(&sdev->state_mutex);
1969 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1970 		mutex_unlock(&sdev->state_mutex);
1971 
1972 		return SUCCESS;
1973 	}
1974 
1975 	return eh_disp;
1976 }
1977 
1978 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1979 {
1980 	struct request *req = scsi_cmd_to_rq(scmd);
1981 	struct scsi_device *sdev = scmd->device;
1982 	unsigned int transferred, good_bytes;
1983 	u64 start_lba, end_lba, bad_lba;
1984 
1985 	/*
1986 	 * Some commands have a payload smaller than the device logical
1987 	 * block size (e.g. INQUIRY on a 4K disk).
1988 	 */
1989 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1990 		return 0;
1991 
1992 	/* Check if we have a 'bad_lba' information */
1993 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1994 				     SCSI_SENSE_BUFFERSIZE,
1995 				     &bad_lba))
1996 		return 0;
1997 
1998 	/*
1999 	 * If the bad lba was reported incorrectly, we have no idea where
2000 	 * the error is.
2001 	 */
2002 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2003 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2004 	if (bad_lba < start_lba || bad_lba >= end_lba)
2005 		return 0;
2006 
2007 	/*
2008 	 * resid is optional but mostly filled in.  When it's unused,
2009 	 * its value is zero, so we assume the whole buffer transferred
2010 	 */
2011 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2012 
2013 	/* This computation should always be done in terms of the
2014 	 * resolution of the device's medium.
2015 	 */
2016 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2017 
2018 	return min(good_bytes, transferred);
2019 }
2020 
2021 /**
2022  *	sd_done - bottom half handler: called when the lower level
2023  *	driver has completed (successfully or otherwise) a scsi command.
2024  *	@SCpnt: mid-level's per command structure.
2025  *
2026  *	Note: potentially run from within an ISR. Must not block.
2027  **/
2028 static int sd_done(struct scsi_cmnd *SCpnt)
2029 {
2030 	int result = SCpnt->result;
2031 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2032 	unsigned int sector_size = SCpnt->device->sector_size;
2033 	unsigned int resid;
2034 	struct scsi_sense_hdr sshdr;
2035 	struct request *req = scsi_cmd_to_rq(SCpnt);
2036 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2037 	int sense_valid = 0;
2038 	int sense_deferred = 0;
2039 
2040 	switch (req_op(req)) {
2041 	case REQ_OP_DISCARD:
2042 	case REQ_OP_WRITE_ZEROES:
2043 	case REQ_OP_WRITE_SAME:
2044 	case REQ_OP_ZONE_RESET:
2045 	case REQ_OP_ZONE_RESET_ALL:
2046 	case REQ_OP_ZONE_OPEN:
2047 	case REQ_OP_ZONE_CLOSE:
2048 	case REQ_OP_ZONE_FINISH:
2049 		if (!result) {
2050 			good_bytes = blk_rq_bytes(req);
2051 			scsi_set_resid(SCpnt, 0);
2052 		} else {
2053 			good_bytes = 0;
2054 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2055 		}
2056 		break;
2057 	default:
2058 		/*
2059 		 * In case of bogus fw or device, we could end up having
2060 		 * an unaligned partial completion. Check this here and force
2061 		 * alignment.
2062 		 */
2063 		resid = scsi_get_resid(SCpnt);
2064 		if (resid & (sector_size - 1)) {
2065 			sd_printk(KERN_INFO, sdkp,
2066 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2067 				resid, sector_size);
2068 			scsi_print_command(SCpnt);
2069 			resid = min(scsi_bufflen(SCpnt),
2070 				    round_up(resid, sector_size));
2071 			scsi_set_resid(SCpnt, resid);
2072 		}
2073 	}
2074 
2075 	if (result) {
2076 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2077 		if (sense_valid)
2078 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2079 	}
2080 	sdkp->medium_access_timed_out = 0;
2081 
2082 	if (!scsi_status_is_check_condition(result) &&
2083 	    (!sense_valid || sense_deferred))
2084 		goto out;
2085 
2086 	switch (sshdr.sense_key) {
2087 	case HARDWARE_ERROR:
2088 	case MEDIUM_ERROR:
2089 		good_bytes = sd_completed_bytes(SCpnt);
2090 		break;
2091 	case RECOVERED_ERROR:
2092 		good_bytes = scsi_bufflen(SCpnt);
2093 		break;
2094 	case NO_SENSE:
2095 		/* This indicates a false check condition, so ignore it.  An
2096 		 * unknown amount of data was transferred so treat it as an
2097 		 * error.
2098 		 */
2099 		SCpnt->result = 0;
2100 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2101 		break;
2102 	case ABORTED_COMMAND:
2103 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2104 			good_bytes = sd_completed_bytes(SCpnt);
2105 		break;
2106 	case ILLEGAL_REQUEST:
2107 		switch (sshdr.asc) {
2108 		case 0x10:	/* DIX: Host detected corruption */
2109 			good_bytes = sd_completed_bytes(SCpnt);
2110 			break;
2111 		case 0x20:	/* INVALID COMMAND OPCODE */
2112 		case 0x24:	/* INVALID FIELD IN CDB */
2113 			switch (SCpnt->cmnd[0]) {
2114 			case UNMAP:
2115 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2116 				break;
2117 			case WRITE_SAME_16:
2118 			case WRITE_SAME:
2119 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2120 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2121 				} else {
2122 					sdkp->device->no_write_same = 1;
2123 					sd_config_write_same(sdkp);
2124 					req->rq_flags |= RQF_QUIET;
2125 				}
2126 				break;
2127 			}
2128 		}
2129 		break;
2130 	default:
2131 		break;
2132 	}
2133 
2134  out:
2135 	if (sd_is_zoned(sdkp))
2136 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2137 
2138 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2139 					   "sd_done: completed %d of %d bytes\n",
2140 					   good_bytes, scsi_bufflen(SCpnt)));
2141 
2142 	return good_bytes;
2143 }
2144 
2145 /*
2146  * spinup disk - called only in sd_revalidate_disk()
2147  */
2148 static void
2149 sd_spinup_disk(struct scsi_disk *sdkp)
2150 {
2151 	unsigned char cmd[10];
2152 	unsigned long spintime_expire = 0;
2153 	int retries, spintime;
2154 	unsigned int the_result;
2155 	struct scsi_sense_hdr sshdr;
2156 	int sense_valid = 0;
2157 
2158 	spintime = 0;
2159 
2160 	/* Spin up drives, as required.  Only do this at boot time */
2161 	/* Spinup needs to be done for module loads too. */
2162 	do {
2163 		retries = 0;
2164 
2165 		do {
2166 			bool media_was_present = sdkp->media_present;
2167 
2168 			cmd[0] = TEST_UNIT_READY;
2169 			memset((void *) &cmd[1], 0, 9);
2170 
2171 			the_result = scsi_execute_req(sdkp->device, cmd,
2172 						      DMA_NONE, NULL, 0,
2173 						      &sshdr, SD_TIMEOUT,
2174 						      sdkp->max_retries, NULL);
2175 
2176 			/*
2177 			 * If the drive has indicated to us that it
2178 			 * doesn't have any media in it, don't bother
2179 			 * with any more polling.
2180 			 */
2181 			if (media_not_present(sdkp, &sshdr)) {
2182 				if (media_was_present)
2183 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2184 				return;
2185 			}
2186 
2187 			if (the_result)
2188 				sense_valid = scsi_sense_valid(&sshdr);
2189 			retries++;
2190 		} while (retries < 3 &&
2191 			 (!scsi_status_is_good(the_result) ||
2192 			  (scsi_status_is_check_condition(the_result) &&
2193 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2194 
2195 		if (!scsi_status_is_check_condition(the_result)) {
2196 			/* no sense, TUR either succeeded or failed
2197 			 * with a status error */
2198 			if(!spintime && !scsi_status_is_good(the_result)) {
2199 				sd_print_result(sdkp, "Test Unit Ready failed",
2200 						the_result);
2201 			}
2202 			break;
2203 		}
2204 
2205 		/*
2206 		 * The device does not want the automatic start to be issued.
2207 		 */
2208 		if (sdkp->device->no_start_on_add)
2209 			break;
2210 
2211 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2212 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2213 				break;	/* manual intervention required */
2214 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2215 				break;	/* standby */
2216 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2217 				break;	/* unavailable */
2218 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2219 				break;	/* sanitize in progress */
2220 			/*
2221 			 * Issue command to spin up drive when not ready
2222 			 */
2223 			if (!spintime) {
2224 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2225 				cmd[0] = START_STOP;
2226 				cmd[1] = 1;	/* Return immediately */
2227 				memset((void *) &cmd[2], 0, 8);
2228 				cmd[4] = 1;	/* Start spin cycle */
2229 				if (sdkp->device->start_stop_pwr_cond)
2230 					cmd[4] |= 1 << 4;
2231 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2232 						 NULL, 0, &sshdr,
2233 						 SD_TIMEOUT, sdkp->max_retries,
2234 						 NULL);
2235 				spintime_expire = jiffies + 100 * HZ;
2236 				spintime = 1;
2237 			}
2238 			/* Wait 1 second for next try */
2239 			msleep(1000);
2240 			printk(KERN_CONT ".");
2241 
2242 		/*
2243 		 * Wait for USB flash devices with slow firmware.
2244 		 * Yes, this sense key/ASC combination shouldn't
2245 		 * occur here.  It's characteristic of these devices.
2246 		 */
2247 		} else if (sense_valid &&
2248 				sshdr.sense_key == UNIT_ATTENTION &&
2249 				sshdr.asc == 0x28) {
2250 			if (!spintime) {
2251 				spintime_expire = jiffies + 5 * HZ;
2252 				spintime = 1;
2253 			}
2254 			/* Wait 1 second for next try */
2255 			msleep(1000);
2256 		} else {
2257 			/* we don't understand the sense code, so it's
2258 			 * probably pointless to loop */
2259 			if(!spintime) {
2260 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2261 				sd_print_sense_hdr(sdkp, &sshdr);
2262 			}
2263 			break;
2264 		}
2265 
2266 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2267 
2268 	if (spintime) {
2269 		if (scsi_status_is_good(the_result))
2270 			printk(KERN_CONT "ready\n");
2271 		else
2272 			printk(KERN_CONT "not responding...\n");
2273 	}
2274 }
2275 
2276 /*
2277  * Determine whether disk supports Data Integrity Field.
2278  */
2279 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2280 {
2281 	struct scsi_device *sdp = sdkp->device;
2282 	u8 type;
2283 	int ret = 0;
2284 
2285 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2286 		sdkp->protection_type = 0;
2287 		return ret;
2288 	}
2289 
2290 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2291 
2292 	if (type > T10_PI_TYPE3_PROTECTION)
2293 		ret = -ENODEV;
2294 	else if (scsi_host_dif_capable(sdp->host, type))
2295 		ret = 1;
2296 
2297 	if (sdkp->first_scan || type != sdkp->protection_type)
2298 		switch (ret) {
2299 		case -ENODEV:
2300 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2301 				  " protection type %u. Disabling disk!\n",
2302 				  type);
2303 			break;
2304 		case 1:
2305 			sd_printk(KERN_NOTICE, sdkp,
2306 				  "Enabling DIF Type %u protection\n", type);
2307 			break;
2308 		case 0:
2309 			sd_printk(KERN_NOTICE, sdkp,
2310 				  "Disabling DIF Type %u protection\n", type);
2311 			break;
2312 		}
2313 
2314 	sdkp->protection_type = type;
2315 
2316 	return ret;
2317 }
2318 
2319 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2320 			struct scsi_sense_hdr *sshdr, int sense_valid,
2321 			int the_result)
2322 {
2323 	if (sense_valid)
2324 		sd_print_sense_hdr(sdkp, sshdr);
2325 	else
2326 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2327 
2328 	/*
2329 	 * Set dirty bit for removable devices if not ready -
2330 	 * sometimes drives will not report this properly.
2331 	 */
2332 	if (sdp->removable &&
2333 	    sense_valid && sshdr->sense_key == NOT_READY)
2334 		set_media_not_present(sdkp);
2335 
2336 	/*
2337 	 * We used to set media_present to 0 here to indicate no media
2338 	 * in the drive, but some drives fail read capacity even with
2339 	 * media present, so we can't do that.
2340 	 */
2341 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2342 }
2343 
2344 #define RC16_LEN 32
2345 #if RC16_LEN > SD_BUF_SIZE
2346 #error RC16_LEN must not be more than SD_BUF_SIZE
2347 #endif
2348 
2349 #define READ_CAPACITY_RETRIES_ON_RESET	10
2350 
2351 static int read_capacity_16(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 	unsigned int alignment;
2360 	unsigned long long lba;
2361 	unsigned sector_size;
2362 
2363 	if (sdp->no_read_capacity_16)
2364 		return -EINVAL;
2365 
2366 	do {
2367 		memset(cmd, 0, 16);
2368 		cmd[0] = SERVICE_ACTION_IN_16;
2369 		cmd[1] = SAI_READ_CAPACITY_16;
2370 		cmd[13] = RC16_LEN;
2371 		memset(buffer, 0, RC16_LEN);
2372 
2373 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2374 					buffer, RC16_LEN, &sshdr,
2375 					SD_TIMEOUT, sdkp->max_retries, NULL);
2376 
2377 		if (media_not_present(sdkp, &sshdr))
2378 			return -ENODEV;
2379 
2380 		if (the_result > 0) {
2381 			sense_valid = scsi_sense_valid(&sshdr);
2382 			if (sense_valid &&
2383 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2384 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2385 			    sshdr.ascq == 0x00)
2386 				/* Invalid Command Operation Code or
2387 				 * Invalid Field in CDB, just retry
2388 				 * silently with RC10 */
2389 				return -EINVAL;
2390 			if (sense_valid &&
2391 			    sshdr.sense_key == UNIT_ATTENTION &&
2392 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2393 				/* Device reset might occur several times,
2394 				 * give it one more chance */
2395 				if (--reset_retries > 0)
2396 					continue;
2397 		}
2398 		retries--;
2399 
2400 	} while (the_result && retries);
2401 
2402 	if (the_result) {
2403 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2404 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2405 		return -EINVAL;
2406 	}
2407 
2408 	sector_size = get_unaligned_be32(&buffer[8]);
2409 	lba = get_unaligned_be64(&buffer[0]);
2410 
2411 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2412 		sdkp->capacity = 0;
2413 		return -ENODEV;
2414 	}
2415 
2416 	/* Logical blocks per physical block exponent */
2417 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2418 
2419 	/* RC basis */
2420 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2421 
2422 	/* Lowest aligned logical block */
2423 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2424 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2425 	if (alignment && sdkp->first_scan)
2426 		sd_printk(KERN_NOTICE, sdkp,
2427 			  "physical block alignment offset: %u\n", alignment);
2428 
2429 	if (buffer[14] & 0x80) { /* LBPME */
2430 		sdkp->lbpme = 1;
2431 
2432 		if (buffer[14] & 0x40) /* LBPRZ */
2433 			sdkp->lbprz = 1;
2434 
2435 		sd_config_discard(sdkp, SD_LBP_WS16);
2436 	}
2437 
2438 	sdkp->capacity = lba + 1;
2439 	return sector_size;
2440 }
2441 
2442 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2443 						unsigned char *buffer)
2444 {
2445 	unsigned char cmd[16];
2446 	struct scsi_sense_hdr sshdr;
2447 	int sense_valid = 0;
2448 	int the_result;
2449 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2450 	sector_t lba;
2451 	unsigned sector_size;
2452 
2453 	do {
2454 		cmd[0] = READ_CAPACITY;
2455 		memset(&cmd[1], 0, 9);
2456 		memset(buffer, 0, 8);
2457 
2458 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2459 					buffer, 8, &sshdr,
2460 					SD_TIMEOUT, sdkp->max_retries, NULL);
2461 
2462 		if (media_not_present(sdkp, &sshdr))
2463 			return -ENODEV;
2464 
2465 		if (the_result > 0) {
2466 			sense_valid = scsi_sense_valid(&sshdr);
2467 			if (sense_valid &&
2468 			    sshdr.sense_key == UNIT_ATTENTION &&
2469 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2470 				/* Device reset might occur several times,
2471 				 * give it one more chance */
2472 				if (--reset_retries > 0)
2473 					continue;
2474 		}
2475 		retries--;
2476 
2477 	} while (the_result && retries);
2478 
2479 	if (the_result) {
2480 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2481 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2482 		return -EINVAL;
2483 	}
2484 
2485 	sector_size = get_unaligned_be32(&buffer[4]);
2486 	lba = get_unaligned_be32(&buffer[0]);
2487 
2488 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2489 		/* Some buggy (usb cardreader) devices return an lba of
2490 		   0xffffffff when the want to report a size of 0 (with
2491 		   which they really mean no media is present) */
2492 		sdkp->capacity = 0;
2493 		sdkp->physical_block_size = sector_size;
2494 		return sector_size;
2495 	}
2496 
2497 	sdkp->capacity = lba + 1;
2498 	sdkp->physical_block_size = sector_size;
2499 	return sector_size;
2500 }
2501 
2502 static int sd_try_rc16_first(struct scsi_device *sdp)
2503 {
2504 	if (sdp->host->max_cmd_len < 16)
2505 		return 0;
2506 	if (sdp->try_rc_10_first)
2507 		return 0;
2508 	if (sdp->scsi_level > SCSI_SPC_2)
2509 		return 1;
2510 	if (scsi_device_protection(sdp))
2511 		return 1;
2512 	return 0;
2513 }
2514 
2515 /*
2516  * read disk capacity
2517  */
2518 static void
2519 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2520 {
2521 	int sector_size;
2522 	struct scsi_device *sdp = sdkp->device;
2523 
2524 	if (sd_try_rc16_first(sdp)) {
2525 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2526 		if (sector_size == -EOVERFLOW)
2527 			goto got_data;
2528 		if (sector_size == -ENODEV)
2529 			return;
2530 		if (sector_size < 0)
2531 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2532 		if (sector_size < 0)
2533 			return;
2534 	} else {
2535 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2536 		if (sector_size == -EOVERFLOW)
2537 			goto got_data;
2538 		if (sector_size < 0)
2539 			return;
2540 		if ((sizeof(sdkp->capacity) > 4) &&
2541 		    (sdkp->capacity > 0xffffffffULL)) {
2542 			int old_sector_size = sector_size;
2543 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2544 					"Trying to use READ CAPACITY(16).\n");
2545 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2546 			if (sector_size < 0) {
2547 				sd_printk(KERN_NOTICE, sdkp,
2548 					"Using 0xffffffff as device size\n");
2549 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2550 				sector_size = old_sector_size;
2551 				goto got_data;
2552 			}
2553 			/* Remember that READ CAPACITY(16) succeeded */
2554 			sdp->try_rc_10_first = 0;
2555 		}
2556 	}
2557 
2558 	/* Some devices are known to return the total number of blocks,
2559 	 * not the highest block number.  Some devices have versions
2560 	 * which do this and others which do not.  Some devices we might
2561 	 * suspect of doing this but we don't know for certain.
2562 	 *
2563 	 * If we know the reported capacity is wrong, decrement it.  If
2564 	 * we can only guess, then assume the number of blocks is even
2565 	 * (usually true but not always) and err on the side of lowering
2566 	 * the capacity.
2567 	 */
2568 	if (sdp->fix_capacity ||
2569 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2570 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2571 				"from its reported value: %llu\n",
2572 				(unsigned long long) sdkp->capacity);
2573 		--sdkp->capacity;
2574 	}
2575 
2576 got_data:
2577 	if (sector_size == 0) {
2578 		sector_size = 512;
2579 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2580 			  "assuming 512.\n");
2581 	}
2582 
2583 	if (sector_size != 512 &&
2584 	    sector_size != 1024 &&
2585 	    sector_size != 2048 &&
2586 	    sector_size != 4096) {
2587 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2588 			  sector_size);
2589 		/*
2590 		 * The user might want to re-format the drive with
2591 		 * a supported sectorsize.  Once this happens, it
2592 		 * would be relatively trivial to set the thing up.
2593 		 * For this reason, we leave the thing in the table.
2594 		 */
2595 		sdkp->capacity = 0;
2596 		/*
2597 		 * set a bogus sector size so the normal read/write
2598 		 * logic in the block layer will eventually refuse any
2599 		 * request on this device without tripping over power
2600 		 * of two sector size assumptions
2601 		 */
2602 		sector_size = 512;
2603 	}
2604 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2605 	blk_queue_physical_block_size(sdp->request_queue,
2606 				      sdkp->physical_block_size);
2607 	sdkp->device->sector_size = sector_size;
2608 
2609 	if (sdkp->capacity > 0xffffffff)
2610 		sdp->use_16_for_rw = 1;
2611 
2612 }
2613 
2614 /*
2615  * Print disk capacity
2616  */
2617 static void
2618 sd_print_capacity(struct scsi_disk *sdkp,
2619 		  sector_t old_capacity)
2620 {
2621 	int sector_size = sdkp->device->sector_size;
2622 	char cap_str_2[10], cap_str_10[10];
2623 
2624 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2625 		return;
2626 
2627 	string_get_size(sdkp->capacity, sector_size,
2628 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2629 	string_get_size(sdkp->capacity, sector_size,
2630 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2631 
2632 	sd_printk(KERN_NOTICE, sdkp,
2633 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2634 		  (unsigned long long)sdkp->capacity,
2635 		  sector_size, cap_str_10, cap_str_2);
2636 
2637 	if (sdkp->physical_block_size != sector_size)
2638 		sd_printk(KERN_NOTICE, sdkp,
2639 			  "%u-byte physical blocks\n",
2640 			  sdkp->physical_block_size);
2641 }
2642 
2643 /* called with buffer of length 512 */
2644 static inline int
2645 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2646 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2647 		 struct scsi_sense_hdr *sshdr)
2648 {
2649 	/*
2650 	 * If we must use MODE SENSE(10), make sure that the buffer length
2651 	 * is at least 8 bytes so that the mode sense header fits.
2652 	 */
2653 	if (sdkp->device->use_10_for_ms && len < 8)
2654 		len = 8;
2655 
2656 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2657 			       SD_TIMEOUT, sdkp->max_retries, data,
2658 			       sshdr);
2659 }
2660 
2661 /*
2662  * read write protect setting, if possible - called only in sd_revalidate_disk()
2663  * called with buffer of length SD_BUF_SIZE
2664  */
2665 static void
2666 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2667 {
2668 	int res;
2669 	struct scsi_device *sdp = sdkp->device;
2670 	struct scsi_mode_data data;
2671 	int old_wp = sdkp->write_prot;
2672 
2673 	set_disk_ro(sdkp->disk, 0);
2674 	if (sdp->skip_ms_page_3f) {
2675 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2676 		return;
2677 	}
2678 
2679 	if (sdp->use_192_bytes_for_3f) {
2680 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2681 	} else {
2682 		/*
2683 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2684 		 * We have to start carefully: some devices hang if we ask
2685 		 * for more than is available.
2686 		 */
2687 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2688 
2689 		/*
2690 		 * Second attempt: ask for page 0 When only page 0 is
2691 		 * implemented, a request for page 3F may return Sense Key
2692 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2693 		 * CDB.
2694 		 */
2695 		if (res < 0)
2696 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2697 
2698 		/*
2699 		 * Third attempt: ask 255 bytes, as we did earlier.
2700 		 */
2701 		if (res < 0)
2702 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2703 					       &data, NULL);
2704 	}
2705 
2706 	if (res < 0) {
2707 		sd_first_printk(KERN_WARNING, sdkp,
2708 			  "Test WP failed, assume Write Enabled\n");
2709 	} else {
2710 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2711 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2712 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2713 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2714 				  sdkp->write_prot ? "on" : "off");
2715 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2716 		}
2717 	}
2718 }
2719 
2720 /*
2721  * sd_read_cache_type - called only from sd_revalidate_disk()
2722  * called with buffer of length SD_BUF_SIZE
2723  */
2724 static void
2725 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2726 {
2727 	int len = 0, res;
2728 	struct scsi_device *sdp = sdkp->device;
2729 
2730 	int dbd;
2731 	int modepage;
2732 	int first_len;
2733 	struct scsi_mode_data data;
2734 	struct scsi_sense_hdr sshdr;
2735 	int old_wce = sdkp->WCE;
2736 	int old_rcd = sdkp->RCD;
2737 	int old_dpofua = sdkp->DPOFUA;
2738 
2739 
2740 	if (sdkp->cache_override)
2741 		return;
2742 
2743 	first_len = 4;
2744 	if (sdp->skip_ms_page_8) {
2745 		if (sdp->type == TYPE_RBC)
2746 			goto defaults;
2747 		else {
2748 			if (sdp->skip_ms_page_3f)
2749 				goto defaults;
2750 			modepage = 0x3F;
2751 			if (sdp->use_192_bytes_for_3f)
2752 				first_len = 192;
2753 			dbd = 0;
2754 		}
2755 	} else if (sdp->type == TYPE_RBC) {
2756 		modepage = 6;
2757 		dbd = 8;
2758 	} else {
2759 		modepage = 8;
2760 		dbd = 0;
2761 	}
2762 
2763 	/* cautiously ask */
2764 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2765 			&data, &sshdr);
2766 
2767 	if (res < 0)
2768 		goto bad_sense;
2769 
2770 	if (!data.header_length) {
2771 		modepage = 6;
2772 		first_len = 0;
2773 		sd_first_printk(KERN_ERR, sdkp,
2774 				"Missing header in MODE_SENSE response\n");
2775 	}
2776 
2777 	/* that went OK, now ask for the proper length */
2778 	len = data.length;
2779 
2780 	/*
2781 	 * We're only interested in the first three bytes, actually.
2782 	 * But the data cache page is defined for the first 20.
2783 	 */
2784 	if (len < 3)
2785 		goto bad_sense;
2786 	else if (len > SD_BUF_SIZE) {
2787 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2788 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2789 		len = SD_BUF_SIZE;
2790 	}
2791 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2792 		len = 192;
2793 
2794 	/* Get the data */
2795 	if (len > first_len)
2796 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2797 				&data, &sshdr);
2798 
2799 	if (!res) {
2800 		int offset = data.header_length + data.block_descriptor_length;
2801 
2802 		while (offset < len) {
2803 			u8 page_code = buffer[offset] & 0x3F;
2804 			u8 spf       = buffer[offset] & 0x40;
2805 
2806 			if (page_code == 8 || page_code == 6) {
2807 				/* We're interested only in the first 3 bytes.
2808 				 */
2809 				if (len - offset <= 2) {
2810 					sd_first_printk(KERN_ERR, sdkp,
2811 						"Incomplete mode parameter "
2812 							"data\n");
2813 					goto defaults;
2814 				} else {
2815 					modepage = page_code;
2816 					goto Page_found;
2817 				}
2818 			} else {
2819 				/* Go to the next page */
2820 				if (spf && len - offset > 3)
2821 					offset += 4 + (buffer[offset+2] << 8) +
2822 						buffer[offset+3];
2823 				else if (!spf && len - offset > 1)
2824 					offset += 2 + buffer[offset+1];
2825 				else {
2826 					sd_first_printk(KERN_ERR, sdkp,
2827 							"Incomplete mode "
2828 							"parameter data\n");
2829 					goto defaults;
2830 				}
2831 			}
2832 		}
2833 
2834 		sd_first_printk(KERN_WARNING, sdkp,
2835 				"No Caching mode page found\n");
2836 		goto defaults;
2837 
2838 	Page_found:
2839 		if (modepage == 8) {
2840 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2841 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2842 		} else {
2843 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2844 			sdkp->RCD = 0;
2845 		}
2846 
2847 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2848 		if (sdp->broken_fua) {
2849 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2850 			sdkp->DPOFUA = 0;
2851 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2852 			   !sdkp->device->use_16_for_rw) {
2853 			sd_first_printk(KERN_NOTICE, sdkp,
2854 				  "Uses READ/WRITE(6), disabling FUA\n");
2855 			sdkp->DPOFUA = 0;
2856 		}
2857 
2858 		/* No cache flush allowed for write protected devices */
2859 		if (sdkp->WCE && sdkp->write_prot)
2860 			sdkp->WCE = 0;
2861 
2862 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2863 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2864 			sd_printk(KERN_NOTICE, sdkp,
2865 				  "Write cache: %s, read cache: %s, %s\n",
2866 				  sdkp->WCE ? "enabled" : "disabled",
2867 				  sdkp->RCD ? "disabled" : "enabled",
2868 				  sdkp->DPOFUA ? "supports DPO and FUA"
2869 				  : "doesn't support DPO or FUA");
2870 
2871 		return;
2872 	}
2873 
2874 bad_sense:
2875 	if (scsi_sense_valid(&sshdr) &&
2876 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2877 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2878 		/* Invalid field in CDB */
2879 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2880 	else
2881 		sd_first_printk(KERN_ERR, sdkp,
2882 				"Asking for cache data failed\n");
2883 
2884 defaults:
2885 	if (sdp->wce_default_on) {
2886 		sd_first_printk(KERN_NOTICE, sdkp,
2887 				"Assuming drive cache: write back\n");
2888 		sdkp->WCE = 1;
2889 	} else {
2890 		sd_first_printk(KERN_WARNING, sdkp,
2891 				"Assuming drive cache: write through\n");
2892 		sdkp->WCE = 0;
2893 	}
2894 	sdkp->RCD = 0;
2895 	sdkp->DPOFUA = 0;
2896 }
2897 
2898 /*
2899  * The ATO bit indicates whether the DIF application tag is available
2900  * for use by the operating system.
2901  */
2902 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2903 {
2904 	int res, offset;
2905 	struct scsi_device *sdp = sdkp->device;
2906 	struct scsi_mode_data data;
2907 	struct scsi_sense_hdr sshdr;
2908 
2909 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2910 		return;
2911 
2912 	if (sdkp->protection_type == 0)
2913 		return;
2914 
2915 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2916 			      sdkp->max_retries, &data, &sshdr);
2917 
2918 	if (res < 0 || !data.header_length ||
2919 	    data.length < 6) {
2920 		sd_first_printk(KERN_WARNING, sdkp,
2921 			  "getting Control mode page failed, assume no ATO\n");
2922 
2923 		if (scsi_sense_valid(&sshdr))
2924 			sd_print_sense_hdr(sdkp, &sshdr);
2925 
2926 		return;
2927 	}
2928 
2929 	offset = data.header_length + data.block_descriptor_length;
2930 
2931 	if ((buffer[offset] & 0x3f) != 0x0a) {
2932 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2933 		return;
2934 	}
2935 
2936 	if ((buffer[offset + 5] & 0x80) == 0)
2937 		return;
2938 
2939 	sdkp->ATO = 1;
2940 
2941 	return;
2942 }
2943 
2944 /**
2945  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2946  * @sdkp: disk to query
2947  */
2948 static void sd_read_block_limits(struct scsi_disk *sdkp)
2949 {
2950 	unsigned int sector_sz = sdkp->device->sector_size;
2951 	const int vpd_len = 64;
2952 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2953 
2954 	if (!buffer ||
2955 	    /* Block Limits VPD */
2956 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2957 		goto out;
2958 
2959 	blk_queue_io_min(sdkp->disk->queue,
2960 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2961 
2962 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2963 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2964 
2965 	if (buffer[3] == 0x3c) {
2966 		unsigned int lba_count, desc_count;
2967 
2968 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2969 
2970 		if (!sdkp->lbpme)
2971 			goto out;
2972 
2973 		lba_count = get_unaligned_be32(&buffer[20]);
2974 		desc_count = get_unaligned_be32(&buffer[24]);
2975 
2976 		if (lba_count && desc_count)
2977 			sdkp->max_unmap_blocks = lba_count;
2978 
2979 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2980 
2981 		if (buffer[32] & 0x80)
2982 			sdkp->unmap_alignment =
2983 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2984 
2985 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2986 
2987 			if (sdkp->max_unmap_blocks)
2988 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2989 			else
2990 				sd_config_discard(sdkp, SD_LBP_WS16);
2991 
2992 		} else {	/* LBP VPD page tells us what to use */
2993 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2994 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2995 			else if (sdkp->lbpws)
2996 				sd_config_discard(sdkp, SD_LBP_WS16);
2997 			else if (sdkp->lbpws10)
2998 				sd_config_discard(sdkp, SD_LBP_WS10);
2999 			else
3000 				sd_config_discard(sdkp, SD_LBP_DISABLE);
3001 		}
3002 	}
3003 
3004  out:
3005 	kfree(buffer);
3006 }
3007 
3008 /**
3009  * sd_read_block_characteristics - Query block dev. characteristics
3010  * @sdkp: disk to query
3011  */
3012 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3013 {
3014 	struct request_queue *q = sdkp->disk->queue;
3015 	unsigned char *buffer;
3016 	u16 rot;
3017 	const int vpd_len = 64;
3018 
3019 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3020 
3021 	if (!buffer ||
3022 	    /* Block Device Characteristics VPD */
3023 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
3024 		goto out;
3025 
3026 	rot = get_unaligned_be16(&buffer[4]);
3027 
3028 	if (rot == 1) {
3029 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3030 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3031 	}
3032 
3033 	if (sdkp->device->type == TYPE_ZBC) {
3034 		/* Host-managed */
3035 		blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3036 	} else {
3037 		sdkp->zoned = (buffer[8] >> 4) & 3;
3038 		if (sdkp->zoned == 1) {
3039 			/* Host-aware */
3040 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3041 		} else {
3042 			/* Regular disk or drive managed disk */
3043 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3044 		}
3045 	}
3046 
3047 	if (!sdkp->first_scan)
3048 		goto out;
3049 
3050 	if (blk_queue_is_zoned(q)) {
3051 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3052 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3053 	} else {
3054 		if (sdkp->zoned == 1)
3055 			sd_printk(KERN_NOTICE, sdkp,
3056 				  "Host-aware SMR disk used as regular disk\n");
3057 		else if (sdkp->zoned == 2)
3058 			sd_printk(KERN_NOTICE, sdkp,
3059 				  "Drive-managed SMR disk\n");
3060 	}
3061 
3062  out:
3063 	kfree(buffer);
3064 }
3065 
3066 /**
3067  * sd_read_block_provisioning - Query provisioning VPD page
3068  * @sdkp: disk to query
3069  */
3070 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3071 {
3072 	unsigned char *buffer;
3073 	const int vpd_len = 8;
3074 
3075 	if (sdkp->lbpme == 0)
3076 		return;
3077 
3078 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3079 
3080 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3081 		goto out;
3082 
3083 	sdkp->lbpvpd	= 1;
3084 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
3085 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
3086 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
3087 
3088  out:
3089 	kfree(buffer);
3090 }
3091 
3092 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3093 {
3094 	struct scsi_device *sdev = sdkp->device;
3095 
3096 	if (sdev->host->no_write_same) {
3097 		sdev->no_write_same = 1;
3098 
3099 		return;
3100 	}
3101 
3102 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3103 		/* too large values might cause issues with arcmsr */
3104 		int vpd_buf_len = 64;
3105 
3106 		sdev->no_report_opcodes = 1;
3107 
3108 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3109 		 * CODES is unsupported and the device has an ATA
3110 		 * Information VPD page (SAT).
3111 		 */
3112 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3113 			sdev->no_write_same = 1;
3114 	}
3115 
3116 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3117 		sdkp->ws16 = 1;
3118 
3119 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3120 		sdkp->ws10 = 1;
3121 }
3122 
3123 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3124 {
3125 	struct scsi_device *sdev = sdkp->device;
3126 
3127 	if (!sdev->security_supported)
3128 		return;
3129 
3130 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3131 			SECURITY_PROTOCOL_IN) == 1 &&
3132 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3133 			SECURITY_PROTOCOL_OUT) == 1)
3134 		sdkp->security = 1;
3135 }
3136 
3137 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3138 {
3139 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3140 }
3141 
3142 /**
3143  * sd_read_cpr - Query concurrent positioning ranges
3144  * @sdkp:	disk to query
3145  */
3146 static void sd_read_cpr(struct scsi_disk *sdkp)
3147 {
3148 	struct blk_independent_access_ranges *iars = NULL;
3149 	unsigned char *buffer = NULL;
3150 	unsigned int nr_cpr = 0;
3151 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3152 	u8 *desc;
3153 
3154 	/*
3155 	 * We need to have the capacity set first for the block layer to be
3156 	 * able to check the ranges.
3157 	 */
3158 	if (sdkp->first_scan)
3159 		return;
3160 
3161 	if (!sdkp->capacity)
3162 		goto out;
3163 
3164 	/*
3165 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3166 	 * leading to a maximum page size of 64 + 256*32 bytes.
3167 	 */
3168 	buf_len = 64 + 256*32;
3169 	buffer = kmalloc(buf_len, GFP_KERNEL);
3170 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3171 		goto out;
3172 
3173 	/* We must have at least a 64B header and one 32B range descriptor */
3174 	vpd_len = get_unaligned_be16(&buffer[2]) + 3;
3175 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3176 		sd_printk(KERN_ERR, sdkp,
3177 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3178 		goto out;
3179 	}
3180 
3181 	nr_cpr = (vpd_len - 64) / 32;
3182 	if (nr_cpr == 1) {
3183 		nr_cpr = 0;
3184 		goto out;
3185 	}
3186 
3187 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3188 	if (!iars) {
3189 		nr_cpr = 0;
3190 		goto out;
3191 	}
3192 
3193 	desc = &buffer[64];
3194 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3195 		if (desc[0] != i) {
3196 			sd_printk(KERN_ERR, sdkp,
3197 				"Invalid Concurrent Positioning Range number\n");
3198 			nr_cpr = 0;
3199 			break;
3200 		}
3201 
3202 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3203 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3204 	}
3205 
3206 out:
3207 	disk_set_independent_access_ranges(sdkp->disk, iars);
3208 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3209 		sd_printk(KERN_NOTICE, sdkp,
3210 			  "%u concurrent positioning ranges\n", nr_cpr);
3211 		sdkp->nr_actuators = nr_cpr;
3212 	}
3213 
3214 	kfree(buffer);
3215 }
3216 
3217 /*
3218  * Determine the device's preferred I/O size for reads and writes
3219  * unless the reported value is unreasonably small, large, not a
3220  * multiple of the physical block size, or simply garbage.
3221  */
3222 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3223 				      unsigned int dev_max)
3224 {
3225 	struct scsi_device *sdp = sdkp->device;
3226 	unsigned int opt_xfer_bytes =
3227 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3228 
3229 	if (sdkp->opt_xfer_blocks == 0)
3230 		return false;
3231 
3232 	if (sdkp->opt_xfer_blocks > dev_max) {
3233 		sd_first_printk(KERN_WARNING, sdkp,
3234 				"Optimal transfer size %u logical blocks " \
3235 				"> dev_max (%u logical blocks)\n",
3236 				sdkp->opt_xfer_blocks, dev_max);
3237 		return false;
3238 	}
3239 
3240 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3241 		sd_first_printk(KERN_WARNING, sdkp,
3242 				"Optimal transfer size %u logical blocks " \
3243 				"> sd driver limit (%u logical blocks)\n",
3244 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3245 		return false;
3246 	}
3247 
3248 	if (opt_xfer_bytes < PAGE_SIZE) {
3249 		sd_first_printk(KERN_WARNING, sdkp,
3250 				"Optimal transfer size %u bytes < " \
3251 				"PAGE_SIZE (%u bytes)\n",
3252 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3253 		return false;
3254 	}
3255 
3256 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3257 		sd_first_printk(KERN_WARNING, sdkp,
3258 				"Optimal transfer size %u bytes not a " \
3259 				"multiple of physical block size (%u bytes)\n",
3260 				opt_xfer_bytes, sdkp->physical_block_size);
3261 		return false;
3262 	}
3263 
3264 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3265 			opt_xfer_bytes);
3266 	return true;
3267 }
3268 
3269 /**
3270  *	sd_revalidate_disk - called the first time a new disk is seen,
3271  *	performs disk spin up, read_capacity, etc.
3272  *	@disk: struct gendisk we care about
3273  **/
3274 static int sd_revalidate_disk(struct gendisk *disk)
3275 {
3276 	struct scsi_disk *sdkp = scsi_disk(disk);
3277 	struct scsi_device *sdp = sdkp->device;
3278 	struct request_queue *q = sdkp->disk->queue;
3279 	sector_t old_capacity = sdkp->capacity;
3280 	unsigned char *buffer;
3281 	unsigned int dev_max, rw_max;
3282 
3283 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3284 				      "sd_revalidate_disk\n"));
3285 
3286 	/*
3287 	 * If the device is offline, don't try and read capacity or any
3288 	 * of the other niceties.
3289 	 */
3290 	if (!scsi_device_online(sdp))
3291 		goto out;
3292 
3293 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3294 	if (!buffer) {
3295 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3296 			  "allocation failure.\n");
3297 		goto out;
3298 	}
3299 
3300 	sd_spinup_disk(sdkp);
3301 
3302 	/*
3303 	 * Without media there is no reason to ask; moreover, some devices
3304 	 * react badly if we do.
3305 	 */
3306 	if (sdkp->media_present) {
3307 		sd_read_capacity(sdkp, buffer);
3308 
3309 		/*
3310 		 * set the default to rotational.  All non-rotational devices
3311 		 * support the block characteristics VPD page, which will
3312 		 * cause this to be updated correctly and any device which
3313 		 * doesn't support it should be treated as rotational.
3314 		 */
3315 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3316 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3317 
3318 		if (scsi_device_supports_vpd(sdp)) {
3319 			sd_read_block_provisioning(sdkp);
3320 			sd_read_block_limits(sdkp);
3321 			sd_read_block_characteristics(sdkp);
3322 			sd_zbc_read_zones(sdkp, buffer);
3323 		}
3324 
3325 		sd_print_capacity(sdkp, old_capacity);
3326 
3327 		sd_read_write_protect_flag(sdkp, buffer);
3328 		sd_read_cache_type(sdkp, buffer);
3329 		sd_read_app_tag_own(sdkp, buffer);
3330 		sd_read_write_same(sdkp, buffer);
3331 		sd_read_security(sdkp, buffer);
3332 		sd_read_cpr(sdkp);
3333 	}
3334 
3335 	/*
3336 	 * We now have all cache related info, determine how we deal
3337 	 * with flush requests.
3338 	 */
3339 	sd_set_flush_flag(sdkp);
3340 
3341 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3342 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3343 
3344 	/* Some devices report a maximum block count for READ/WRITE requests. */
3345 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3346 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3347 
3348 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3349 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3350 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3351 	} else {
3352 		q->limits.io_opt = 0;
3353 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3354 				      (sector_t)BLK_DEF_MAX_SECTORS);
3355 	}
3356 
3357 	/* Do not exceed controller limit */
3358 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3359 
3360 	/*
3361 	 * Only update max_sectors if previously unset or if the current value
3362 	 * exceeds the capabilities of the hardware.
3363 	 */
3364 	if (sdkp->first_scan ||
3365 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3366 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3367 		q->limits.max_sectors = rw_max;
3368 
3369 	sdkp->first_scan = 0;
3370 
3371 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3372 	sd_config_write_same(sdkp);
3373 	kfree(buffer);
3374 
3375 	/*
3376 	 * For a zoned drive, revalidating the zones can be done only once
3377 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3378 	 * capacity to 0.
3379 	 */
3380 	if (sd_zbc_revalidate_zones(sdkp))
3381 		set_capacity_and_notify(disk, 0);
3382 
3383  out:
3384 	return 0;
3385 }
3386 
3387 /**
3388  *	sd_unlock_native_capacity - unlock native capacity
3389  *	@disk: struct gendisk to set capacity for
3390  *
3391  *	Block layer calls this function if it detects that partitions
3392  *	on @disk reach beyond the end of the device.  If the SCSI host
3393  *	implements ->unlock_native_capacity() method, it's invoked to
3394  *	give it a chance to adjust the device capacity.
3395  *
3396  *	CONTEXT:
3397  *	Defined by block layer.  Might sleep.
3398  */
3399 static void sd_unlock_native_capacity(struct gendisk *disk)
3400 {
3401 	struct scsi_device *sdev = scsi_disk(disk)->device;
3402 
3403 	if (sdev->host->hostt->unlock_native_capacity)
3404 		sdev->host->hostt->unlock_native_capacity(sdev);
3405 }
3406 
3407 /**
3408  *	sd_format_disk_name - format disk name
3409  *	@prefix: name prefix - ie. "sd" for SCSI disks
3410  *	@index: index of the disk to format name for
3411  *	@buf: output buffer
3412  *	@buflen: length of the output buffer
3413  *
3414  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3415  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3416  *	which is followed by sdaaa.
3417  *
3418  *	This is basically 26 base counting with one extra 'nil' entry
3419  *	at the beginning from the second digit on and can be
3420  *	determined using similar method as 26 base conversion with the
3421  *	index shifted -1 after each digit is computed.
3422  *
3423  *	CONTEXT:
3424  *	Don't care.
3425  *
3426  *	RETURNS:
3427  *	0 on success, -errno on failure.
3428  */
3429 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3430 {
3431 	const int base = 'z' - 'a' + 1;
3432 	char *begin = buf + strlen(prefix);
3433 	char *end = buf + buflen;
3434 	char *p;
3435 	int unit;
3436 
3437 	p = end - 1;
3438 	*p = '\0';
3439 	unit = base;
3440 	do {
3441 		if (p == begin)
3442 			return -EINVAL;
3443 		*--p = 'a' + (index % unit);
3444 		index = (index / unit) - 1;
3445 	} while (index >= 0);
3446 
3447 	memmove(begin, p, end - p);
3448 	memcpy(buf, prefix, strlen(prefix));
3449 
3450 	return 0;
3451 }
3452 
3453 /**
3454  *	sd_probe - called during driver initialization and whenever a
3455  *	new scsi device is attached to the system. It is called once
3456  *	for each scsi device (not just disks) present.
3457  *	@dev: pointer to device object
3458  *
3459  *	Returns 0 if successful (or not interested in this scsi device
3460  *	(e.g. scanner)); 1 when there is an error.
3461  *
3462  *	Note: this function is invoked from the scsi mid-level.
3463  *	This function sets up the mapping between a given
3464  *	<host,channel,id,lun> (found in sdp) and new device name
3465  *	(e.g. /dev/sda). More precisely it is the block device major
3466  *	and minor number that is chosen here.
3467  *
3468  *	Assume sd_probe is not re-entrant (for time being)
3469  *	Also think about sd_probe() and sd_remove() running coincidentally.
3470  **/
3471 static int sd_probe(struct device *dev)
3472 {
3473 	struct scsi_device *sdp = to_scsi_device(dev);
3474 	struct scsi_disk *sdkp;
3475 	struct gendisk *gd;
3476 	int index;
3477 	int error;
3478 
3479 	scsi_autopm_get_device(sdp);
3480 	error = -ENODEV;
3481 	if (sdp->type != TYPE_DISK &&
3482 	    sdp->type != TYPE_ZBC &&
3483 	    sdp->type != TYPE_MOD &&
3484 	    sdp->type != TYPE_RBC)
3485 		goto out;
3486 
3487 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3488 		sdev_printk(KERN_WARNING, sdp,
3489 			    "Unsupported ZBC host-managed device.\n");
3490 		goto out;
3491 	}
3492 
3493 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3494 					"sd_probe\n"));
3495 
3496 	error = -ENOMEM;
3497 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3498 	if (!sdkp)
3499 		goto out;
3500 
3501 	gd = __alloc_disk_node(sdp->request_queue, NUMA_NO_NODE,
3502 			       &sd_bio_compl_lkclass);
3503 	if (!gd)
3504 		goto out_free;
3505 
3506 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3507 	if (index < 0) {
3508 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3509 		goto out_put;
3510 	}
3511 
3512 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3513 	if (error) {
3514 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3515 		goto out_free_index;
3516 	}
3517 
3518 	sdkp->device = sdp;
3519 	sdkp->driver = &sd_template;
3520 	sdkp->disk = gd;
3521 	sdkp->index = index;
3522 	sdkp->max_retries = SD_MAX_RETRIES;
3523 	atomic_set(&sdkp->openers, 0);
3524 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3525 
3526 	if (!sdp->request_queue->rq_timeout) {
3527 		if (sdp->type != TYPE_MOD)
3528 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3529 		else
3530 			blk_queue_rq_timeout(sdp->request_queue,
3531 					     SD_MOD_TIMEOUT);
3532 	}
3533 
3534 	device_initialize(&sdkp->dev);
3535 	sdkp->dev.parent = get_device(dev);
3536 	sdkp->dev.class = &sd_disk_class;
3537 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3538 
3539 	error = device_add(&sdkp->dev);
3540 	if (error) {
3541 		put_device(&sdkp->dev);
3542 		goto out;
3543 	}
3544 
3545 	dev_set_drvdata(dev, sdkp);
3546 
3547 	gd->major = sd_major((index & 0xf0) >> 4);
3548 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3549 	gd->minors = SD_MINORS;
3550 
3551 	gd->fops = &sd_fops;
3552 	gd->private_data = &sdkp->driver;
3553 
3554 	/* defaults, until the device tells us otherwise */
3555 	sdp->sector_size = 512;
3556 	sdkp->capacity = 0;
3557 	sdkp->media_present = 1;
3558 	sdkp->write_prot = 0;
3559 	sdkp->cache_override = 0;
3560 	sdkp->WCE = 0;
3561 	sdkp->RCD = 0;
3562 	sdkp->ATO = 0;
3563 	sdkp->first_scan = 1;
3564 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3565 
3566 	sd_revalidate_disk(gd);
3567 
3568 	if (sdp->removable) {
3569 		gd->flags |= GENHD_FL_REMOVABLE;
3570 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3571 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3572 	}
3573 
3574 	blk_pm_runtime_init(sdp->request_queue, dev);
3575 	if (sdp->rpm_autosuspend) {
3576 		pm_runtime_set_autosuspend_delay(dev,
3577 			sdp->host->hostt->rpm_autosuspend_delay);
3578 	}
3579 
3580 	error = device_add_disk(dev, gd, NULL);
3581 	if (error) {
3582 		put_device(&sdkp->dev);
3583 		goto out;
3584 	}
3585 
3586 	if (sdkp->capacity)
3587 		sd_dif_config_host(sdkp);
3588 
3589 	sd_revalidate_disk(gd);
3590 
3591 	if (sdkp->security) {
3592 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3593 		if (sdkp->opal_dev)
3594 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3595 	}
3596 
3597 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3598 		  sdp->removable ? "removable " : "");
3599 	scsi_autopm_put_device(sdp);
3600 
3601 	return 0;
3602 
3603  out_free_index:
3604 	ida_free(&sd_index_ida, index);
3605  out_put:
3606 	put_disk(gd);
3607  out_free:
3608 	sd_zbc_release_disk(sdkp);
3609 	kfree(sdkp);
3610  out:
3611 	scsi_autopm_put_device(sdp);
3612 	return error;
3613 }
3614 
3615 /**
3616  *	sd_remove - called whenever a scsi disk (previously recognized by
3617  *	sd_probe) is detached from the system. It is called (potentially
3618  *	multiple times) during sd module unload.
3619  *	@dev: pointer to device object
3620  *
3621  *	Note: this function is invoked from the scsi mid-level.
3622  *	This function potentially frees up a device name (e.g. /dev/sdc)
3623  *	that could be re-used by a subsequent sd_probe().
3624  *	This function is not called when the built-in sd driver is "exit-ed".
3625  **/
3626 static int sd_remove(struct device *dev)
3627 {
3628 	struct scsi_disk *sdkp;
3629 
3630 	sdkp = dev_get_drvdata(dev);
3631 	scsi_autopm_get_device(sdkp->device);
3632 
3633 	device_del(&sdkp->dev);
3634 	del_gendisk(sdkp->disk);
3635 	sd_shutdown(dev);
3636 
3637 	free_opal_dev(sdkp->opal_dev);
3638 
3639 	mutex_lock(&sd_ref_mutex);
3640 	dev_set_drvdata(dev, NULL);
3641 	put_device(&sdkp->dev);
3642 	mutex_unlock(&sd_ref_mutex);
3643 
3644 	return 0;
3645 }
3646 
3647 /**
3648  *	scsi_disk_release - Called to free the scsi_disk structure
3649  *	@dev: pointer to embedded class device
3650  *
3651  *	sd_ref_mutex must be held entering this routine.  Because it is
3652  *	called on last put, you should always use the scsi_disk_get()
3653  *	scsi_disk_put() helpers which manipulate the semaphore directly
3654  *	and never do a direct put_device.
3655  **/
3656 static void scsi_disk_release(struct device *dev)
3657 {
3658 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3659 	struct gendisk *disk = sdkp->disk;
3660 	struct request_queue *q = disk->queue;
3661 
3662 	ida_free(&sd_index_ida, sdkp->index);
3663 
3664 	/*
3665 	 * Wait until all requests that are in progress have completed.
3666 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3667 	 * due to clearing the disk->private_data pointer. Wait from inside
3668 	 * scsi_disk_release() instead of from sd_release() to avoid that
3669 	 * freezing and unfreezing the request queue affects user space I/O
3670 	 * in case multiple processes open a /dev/sd... node concurrently.
3671 	 */
3672 	blk_mq_freeze_queue(q);
3673 	blk_mq_unfreeze_queue(q);
3674 
3675 	disk->private_data = NULL;
3676 	put_disk(disk);
3677 	put_device(&sdkp->device->sdev_gendev);
3678 
3679 	sd_zbc_release_disk(sdkp);
3680 
3681 	kfree(sdkp);
3682 }
3683 
3684 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3685 {
3686 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3687 	struct scsi_sense_hdr sshdr;
3688 	struct scsi_device *sdp = sdkp->device;
3689 	int res;
3690 
3691 	if (start)
3692 		cmd[4] |= 1;	/* START */
3693 
3694 	if (sdp->start_stop_pwr_cond)
3695 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3696 
3697 	if (!scsi_device_online(sdp))
3698 		return -ENODEV;
3699 
3700 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3701 			SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3702 	if (res) {
3703 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3704 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3705 			sd_print_sense_hdr(sdkp, &sshdr);
3706 			/* 0x3a is medium not present */
3707 			if (sshdr.asc == 0x3a)
3708 				res = 0;
3709 		}
3710 	}
3711 
3712 	/* SCSI error codes must not go to the generic layer */
3713 	if (res)
3714 		return -EIO;
3715 
3716 	return 0;
3717 }
3718 
3719 /*
3720  * Send a SYNCHRONIZE CACHE instruction down to the device through
3721  * the normal SCSI command structure.  Wait for the command to
3722  * complete.
3723  */
3724 static void sd_shutdown(struct device *dev)
3725 {
3726 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3727 
3728 	if (!sdkp)
3729 		return;         /* this can happen */
3730 
3731 	if (pm_runtime_suspended(dev))
3732 		return;
3733 
3734 	if (sdkp->WCE && sdkp->media_present) {
3735 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3736 		sd_sync_cache(sdkp, NULL);
3737 	}
3738 
3739 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3740 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3741 		sd_start_stop_device(sdkp, 0);
3742 	}
3743 }
3744 
3745 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3746 {
3747 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3748 	struct scsi_sense_hdr sshdr;
3749 	int ret = 0;
3750 
3751 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3752 		return 0;
3753 
3754 	if (sdkp->WCE && sdkp->media_present) {
3755 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3756 		ret = sd_sync_cache(sdkp, &sshdr);
3757 
3758 		if (ret) {
3759 			/* ignore OFFLINE device */
3760 			if (ret == -ENODEV)
3761 				return 0;
3762 
3763 			if (!scsi_sense_valid(&sshdr) ||
3764 			    sshdr.sense_key != ILLEGAL_REQUEST)
3765 				return ret;
3766 
3767 			/*
3768 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3769 			 * doesn't support sync. There's not much to do and
3770 			 * suspend shouldn't fail.
3771 			 */
3772 			ret = 0;
3773 		}
3774 	}
3775 
3776 	if (sdkp->device->manage_start_stop) {
3777 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3778 		/* an error is not worth aborting a system sleep */
3779 		ret = sd_start_stop_device(sdkp, 0);
3780 		if (ignore_stop_errors)
3781 			ret = 0;
3782 	}
3783 
3784 	return ret;
3785 }
3786 
3787 static int sd_suspend_system(struct device *dev)
3788 {
3789 	if (pm_runtime_suspended(dev))
3790 		return 0;
3791 
3792 	return sd_suspend_common(dev, true);
3793 }
3794 
3795 static int sd_suspend_runtime(struct device *dev)
3796 {
3797 	return sd_suspend_common(dev, false);
3798 }
3799 
3800 static int sd_resume(struct device *dev)
3801 {
3802 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3803 	int ret;
3804 
3805 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3806 		return 0;
3807 
3808 	if (!sdkp->device->manage_start_stop)
3809 		return 0;
3810 
3811 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3812 	ret = sd_start_stop_device(sdkp, 1);
3813 	if (!ret)
3814 		opal_unlock_from_suspend(sdkp->opal_dev);
3815 	return ret;
3816 }
3817 
3818 static int sd_resume_system(struct device *dev)
3819 {
3820 	if (pm_runtime_suspended(dev))
3821 		return 0;
3822 
3823 	return sd_resume(dev);
3824 }
3825 
3826 static int sd_resume_runtime(struct device *dev)
3827 {
3828 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3829 	struct scsi_device *sdp;
3830 
3831 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3832 		return 0;
3833 
3834 	sdp = sdkp->device;
3835 
3836 	if (sdp->ignore_media_change) {
3837 		/* clear the device's sense data */
3838 		static const u8 cmd[10] = { REQUEST_SENSE };
3839 
3840 		if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3841 				 NULL, sdp->request_queue->rq_timeout, 1, 0,
3842 				 RQF_PM, NULL))
3843 			sd_printk(KERN_NOTICE, sdkp,
3844 				  "Failed to clear sense data\n");
3845 	}
3846 
3847 	return sd_resume(dev);
3848 }
3849 
3850 /**
3851  *	init_sd - entry point for this driver (both when built in or when
3852  *	a module).
3853  *
3854  *	Note: this function registers this driver with the scsi mid-level.
3855  **/
3856 static int __init init_sd(void)
3857 {
3858 	int majors = 0, i, err;
3859 
3860 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3861 
3862 	for (i = 0; i < SD_MAJORS; i++) {
3863 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3864 			continue;
3865 		majors++;
3866 	}
3867 
3868 	if (!majors)
3869 		return -ENODEV;
3870 
3871 	err = class_register(&sd_disk_class);
3872 	if (err)
3873 		goto err_out;
3874 
3875 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3876 					 0, 0, NULL);
3877 	if (!sd_cdb_cache) {
3878 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3879 		err = -ENOMEM;
3880 		goto err_out_class;
3881 	}
3882 
3883 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3884 	if (!sd_cdb_pool) {
3885 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3886 		err = -ENOMEM;
3887 		goto err_out_cache;
3888 	}
3889 
3890 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3891 	if (!sd_page_pool) {
3892 		printk(KERN_ERR "sd: can't init discard page pool\n");
3893 		err = -ENOMEM;
3894 		goto err_out_ppool;
3895 	}
3896 
3897 	err = scsi_register_driver(&sd_template.gendrv);
3898 	if (err)
3899 		goto err_out_driver;
3900 
3901 	return 0;
3902 
3903 err_out_driver:
3904 	mempool_destroy(sd_page_pool);
3905 
3906 err_out_ppool:
3907 	mempool_destroy(sd_cdb_pool);
3908 
3909 err_out_cache:
3910 	kmem_cache_destroy(sd_cdb_cache);
3911 
3912 err_out_class:
3913 	class_unregister(&sd_disk_class);
3914 err_out:
3915 	for (i = 0; i < SD_MAJORS; i++)
3916 		unregister_blkdev(sd_major(i), "sd");
3917 	return err;
3918 }
3919 
3920 /**
3921  *	exit_sd - exit point for this driver (when it is a module).
3922  *
3923  *	Note: this function unregisters this driver from the scsi mid-level.
3924  **/
3925 static void __exit exit_sd(void)
3926 {
3927 	int i;
3928 
3929 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3930 
3931 	scsi_unregister_driver(&sd_template.gendrv);
3932 	mempool_destroy(sd_cdb_pool);
3933 	mempool_destroy(sd_page_pool);
3934 	kmem_cache_destroy(sd_cdb_cache);
3935 
3936 	class_unregister(&sd_disk_class);
3937 
3938 	for (i = 0; i < SD_MAJORS; i++)
3939 		unregister_blkdev(sd_major(i), "sd");
3940 }
3941 
3942 module_init(init_sd);
3943 module_exit(exit_sd);
3944 
3945 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3946 {
3947 	scsi_print_sense_hdr(sdkp->device,
3948 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3949 }
3950 
3951 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3952 {
3953 	const char *hb_string = scsi_hostbyte_string(result);
3954 
3955 	if (hb_string)
3956 		sd_printk(KERN_INFO, sdkp,
3957 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3958 			  hb_string ? hb_string : "invalid",
3959 			  "DRIVER_OK");
3960 	else
3961 		sd_printk(KERN_INFO, sdkp,
3962 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3963 			  msg, host_byte(result), "DRIVER_OK");
3964 }
3965