xref: /linux/drivers/scsi/sd.c (revision a6021aa24f6417416d93318bbfa022ab229c33c8)
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/bio-integrity.h>
37 #include <linux/module.h>
38 #include <linux/fs.h>
39 #include <linux/kernel.h>
40 #include <linux/mm.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/rw_hint.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 <linux/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_devinfo.h>
67 #include <scsi/scsi_driver.h>
68 #include <scsi/scsi_eh.h>
69 #include <scsi/scsi_host.h>
70 #include <scsi/scsi_ioctl.h>
71 #include <scsi/scsicam.h>
72 #include <scsi/scsi_common.h>
73 
74 #include "sd.h"
75 #include "scsi_priv.h"
76 #include "scsi_logging.h"
77 
78 MODULE_AUTHOR("Eric Youngdale");
79 MODULE_DESCRIPTION("SCSI disk (sd) driver");
80 MODULE_LICENSE("GPL");
81 
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
96 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
97 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
100 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
101 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
102 
103 #define SD_MINORS	16
104 
105 static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
106 		unsigned int mode);
107 static void sd_config_write_same(struct scsi_disk *sdkp,
108 		struct queue_limits *lim);
109 static int  sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static void sd_shutdown(struct device *);
112 static void scsi_disk_release(struct device *cdev);
113 
114 static DEFINE_IDA(sd_index_ida);
115 
116 static mempool_t *sd_page_pool;
117 static struct lock_class_key sd_bio_compl_lkclass;
118 
119 static const char *sd_cache_types[] = {
120 	"write through", "none", "write back",
121 	"write back, no read (daft)"
122 };
123 
124 static void sd_set_flush_flag(struct scsi_disk *sdkp,
125 		struct queue_limits *lim)
126 {
127 	if (sdkp->WCE) {
128 		lim->features |= BLK_FEAT_WRITE_CACHE;
129 		if (sdkp->DPOFUA)
130 			lim->features |= BLK_FEAT_FUA;
131 		else
132 			lim->features &= ~BLK_FEAT_FUA;
133 	} else {
134 		lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA);
135 	}
136 }
137 
138 static ssize_t
139 cache_type_store(struct device *dev, struct device_attribute *attr,
140 		 const char *buf, size_t count)
141 {
142 	int ct, rcd, wce, sp;
143 	struct scsi_disk *sdkp = to_scsi_disk(dev);
144 	struct scsi_device *sdp = sdkp->device;
145 	char buffer[64];
146 	char *buffer_data;
147 	struct scsi_mode_data data;
148 	struct scsi_sense_hdr sshdr;
149 	static const char temp[] = "temporary ";
150 	int len, ret;
151 
152 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
153 		/* no cache control on RBC devices; theoretically they
154 		 * can do it, but there's probably so many exceptions
155 		 * it's not worth the risk */
156 		return -EINVAL;
157 
158 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
159 		buf += sizeof(temp) - 1;
160 		sdkp->cache_override = 1;
161 	} else {
162 		sdkp->cache_override = 0;
163 	}
164 
165 	ct = sysfs_match_string(sd_cache_types, buf);
166 	if (ct < 0)
167 		return -EINVAL;
168 
169 	rcd = ct & 0x01 ? 1 : 0;
170 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
171 
172 	if (sdkp->cache_override) {
173 		struct queue_limits lim;
174 
175 		sdkp->WCE = wce;
176 		sdkp->RCD = rcd;
177 
178 		lim = queue_limits_start_update(sdkp->disk->queue);
179 		sd_set_flush_flag(sdkp, &lim);
180 		blk_mq_freeze_queue(sdkp->disk->queue);
181 		ret = queue_limits_commit_update(sdkp->disk->queue, &lim);
182 		blk_mq_unfreeze_queue(sdkp->disk->queue);
183 		if (ret)
184 			return ret;
185 		return count;
186 	}
187 
188 	if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
189 			    sdkp->max_retries, &data, NULL))
190 		return -EINVAL;
191 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
192 		  data.block_descriptor_length);
193 	buffer_data = buffer + data.header_length +
194 		data.block_descriptor_length;
195 	buffer_data[2] &= ~0x05;
196 	buffer_data[2] |= wce << 2 | rcd;
197 	sp = buffer_data[0] & 0x80 ? 1 : 0;
198 	buffer_data[0] &= ~0x80;
199 
200 	/*
201 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
202 	 * received mode parameter buffer before doing MODE SELECT.
203 	 */
204 	data.device_specific = 0;
205 
206 	ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
207 			       sdkp->max_retries, &data, &sshdr);
208 	if (ret) {
209 		if (ret > 0 && scsi_sense_valid(&sshdr))
210 			sd_print_sense_hdr(sdkp, &sshdr);
211 		return -EINVAL;
212 	}
213 	sd_revalidate_disk(sdkp->disk);
214 	return count;
215 }
216 
217 static ssize_t
218 manage_start_stop_show(struct device *dev,
219 		       struct device_attribute *attr, char *buf)
220 {
221 	struct scsi_disk *sdkp = to_scsi_disk(dev);
222 	struct scsi_device *sdp = sdkp->device;
223 
224 	return sysfs_emit(buf, "%u\n",
225 			  sdp->manage_system_start_stop &&
226 			  sdp->manage_runtime_start_stop &&
227 			  sdp->manage_shutdown);
228 }
229 static DEVICE_ATTR_RO(manage_start_stop);
230 
231 static ssize_t
232 manage_system_start_stop_show(struct device *dev,
233 			      struct device_attribute *attr, char *buf)
234 {
235 	struct scsi_disk *sdkp = to_scsi_disk(dev);
236 	struct scsi_device *sdp = sdkp->device;
237 
238 	return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
239 }
240 
241 static ssize_t
242 manage_system_start_stop_store(struct device *dev,
243 			       struct device_attribute *attr,
244 			       const char *buf, size_t count)
245 {
246 	struct scsi_disk *sdkp = to_scsi_disk(dev);
247 	struct scsi_device *sdp = sdkp->device;
248 	bool v;
249 
250 	if (!capable(CAP_SYS_ADMIN))
251 		return -EACCES;
252 
253 	if (kstrtobool(buf, &v))
254 		return -EINVAL;
255 
256 	sdp->manage_system_start_stop = v;
257 
258 	return count;
259 }
260 static DEVICE_ATTR_RW(manage_system_start_stop);
261 
262 static ssize_t
263 manage_runtime_start_stop_show(struct device *dev,
264 			       struct device_attribute *attr, char *buf)
265 {
266 	struct scsi_disk *sdkp = to_scsi_disk(dev);
267 	struct scsi_device *sdp = sdkp->device;
268 
269 	return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
270 }
271 
272 static ssize_t
273 manage_runtime_start_stop_store(struct device *dev,
274 				struct device_attribute *attr,
275 				const char *buf, size_t count)
276 {
277 	struct scsi_disk *sdkp = to_scsi_disk(dev);
278 	struct scsi_device *sdp = sdkp->device;
279 	bool v;
280 
281 	if (!capable(CAP_SYS_ADMIN))
282 		return -EACCES;
283 
284 	if (kstrtobool(buf, &v))
285 		return -EINVAL;
286 
287 	sdp->manage_runtime_start_stop = v;
288 
289 	return count;
290 }
291 static DEVICE_ATTR_RW(manage_runtime_start_stop);
292 
293 static ssize_t manage_shutdown_show(struct device *dev,
294 				    struct device_attribute *attr, char *buf)
295 {
296 	struct scsi_disk *sdkp = to_scsi_disk(dev);
297 	struct scsi_device *sdp = sdkp->device;
298 
299 	return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
300 }
301 
302 static ssize_t manage_shutdown_store(struct device *dev,
303 				     struct device_attribute *attr,
304 				     const char *buf, size_t count)
305 {
306 	struct scsi_disk *sdkp = to_scsi_disk(dev);
307 	struct scsi_device *sdp = sdkp->device;
308 	bool v;
309 
310 	if (!capable(CAP_SYS_ADMIN))
311 		return -EACCES;
312 
313 	if (kstrtobool(buf, &v))
314 		return -EINVAL;
315 
316 	sdp->manage_shutdown = v;
317 
318 	return count;
319 }
320 static DEVICE_ATTR_RW(manage_shutdown);
321 
322 static ssize_t
323 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
324 {
325 	struct scsi_disk *sdkp = to_scsi_disk(dev);
326 
327 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
328 }
329 
330 static ssize_t
331 allow_restart_store(struct device *dev, struct device_attribute *attr,
332 		    const char *buf, size_t count)
333 {
334 	bool v;
335 	struct scsi_disk *sdkp = to_scsi_disk(dev);
336 	struct scsi_device *sdp = sdkp->device;
337 
338 	if (!capable(CAP_SYS_ADMIN))
339 		return -EACCES;
340 
341 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
342 		return -EINVAL;
343 
344 	if (kstrtobool(buf, &v))
345 		return -EINVAL;
346 
347 	sdp->allow_restart = v;
348 
349 	return count;
350 }
351 static DEVICE_ATTR_RW(allow_restart);
352 
353 static ssize_t
354 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
355 {
356 	struct scsi_disk *sdkp = to_scsi_disk(dev);
357 	int ct = sdkp->RCD + 2*sdkp->WCE;
358 
359 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
360 }
361 static DEVICE_ATTR_RW(cache_type);
362 
363 static ssize_t
364 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
365 {
366 	struct scsi_disk *sdkp = to_scsi_disk(dev);
367 
368 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
369 }
370 static DEVICE_ATTR_RO(FUA);
371 
372 static ssize_t
373 protection_type_show(struct device *dev, struct device_attribute *attr,
374 		     char *buf)
375 {
376 	struct scsi_disk *sdkp = to_scsi_disk(dev);
377 
378 	return sprintf(buf, "%u\n", sdkp->protection_type);
379 }
380 
381 static ssize_t
382 protection_type_store(struct device *dev, struct device_attribute *attr,
383 		      const char *buf, size_t count)
384 {
385 	struct scsi_disk *sdkp = to_scsi_disk(dev);
386 	unsigned int val;
387 	int err;
388 
389 	if (!capable(CAP_SYS_ADMIN))
390 		return -EACCES;
391 
392 	err = kstrtouint(buf, 10, &val);
393 
394 	if (err)
395 		return err;
396 
397 	if (val <= T10_PI_TYPE3_PROTECTION)
398 		sdkp->protection_type = val;
399 
400 	return count;
401 }
402 static DEVICE_ATTR_RW(protection_type);
403 
404 static ssize_t
405 protection_mode_show(struct device *dev, struct device_attribute *attr,
406 		     char *buf)
407 {
408 	struct scsi_disk *sdkp = to_scsi_disk(dev);
409 	struct scsi_device *sdp = sdkp->device;
410 	unsigned int dif, dix;
411 
412 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
413 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
414 
415 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
416 		dif = 0;
417 		dix = 1;
418 	}
419 
420 	if (!dif && !dix)
421 		return sprintf(buf, "none\n");
422 
423 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
424 }
425 static DEVICE_ATTR_RO(protection_mode);
426 
427 static ssize_t
428 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
429 {
430 	struct scsi_disk *sdkp = to_scsi_disk(dev);
431 
432 	return sprintf(buf, "%u\n", sdkp->ATO);
433 }
434 static DEVICE_ATTR_RO(app_tag_own);
435 
436 static ssize_t
437 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
438 		       char *buf)
439 {
440 	struct scsi_disk *sdkp = to_scsi_disk(dev);
441 
442 	return sprintf(buf, "%u\n", sdkp->lbpme);
443 }
444 static DEVICE_ATTR_RO(thin_provisioning);
445 
446 /* sysfs_match_string() requires dense arrays */
447 static const char *lbp_mode[] = {
448 	[SD_LBP_FULL]		= "full",
449 	[SD_LBP_UNMAP]		= "unmap",
450 	[SD_LBP_WS16]		= "writesame_16",
451 	[SD_LBP_WS10]		= "writesame_10",
452 	[SD_LBP_ZERO]		= "writesame_zero",
453 	[SD_LBP_DISABLE]	= "disabled",
454 };
455 
456 static ssize_t
457 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
458 		       char *buf)
459 {
460 	struct scsi_disk *sdkp = to_scsi_disk(dev);
461 
462 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
463 }
464 
465 static ssize_t
466 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
467 			const char *buf, size_t count)
468 {
469 	struct scsi_disk *sdkp = to_scsi_disk(dev);
470 	struct scsi_device *sdp = sdkp->device;
471 	struct queue_limits lim;
472 	int mode, err;
473 
474 	if (!capable(CAP_SYS_ADMIN))
475 		return -EACCES;
476 
477 	if (sdp->type != TYPE_DISK)
478 		return -EINVAL;
479 
480 	mode = sysfs_match_string(lbp_mode, buf);
481 	if (mode < 0)
482 		return -EINVAL;
483 
484 	lim = queue_limits_start_update(sdkp->disk->queue);
485 	sd_config_discard(sdkp, &lim, mode);
486 	blk_mq_freeze_queue(sdkp->disk->queue);
487 	err = queue_limits_commit_update(sdkp->disk->queue, &lim);
488 	blk_mq_unfreeze_queue(sdkp->disk->queue);
489 	if (err)
490 		return err;
491 	return count;
492 }
493 static DEVICE_ATTR_RW(provisioning_mode);
494 
495 /* sysfs_match_string() requires dense arrays */
496 static const char *zeroing_mode[] = {
497 	[SD_ZERO_WRITE]		= "write",
498 	[SD_ZERO_WS]		= "writesame",
499 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
500 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
501 };
502 
503 static ssize_t
504 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
505 		  char *buf)
506 {
507 	struct scsi_disk *sdkp = to_scsi_disk(dev);
508 
509 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
510 }
511 
512 static ssize_t
513 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
514 		   const char *buf, size_t count)
515 {
516 	struct scsi_disk *sdkp = to_scsi_disk(dev);
517 	int mode;
518 
519 	if (!capable(CAP_SYS_ADMIN))
520 		return -EACCES;
521 
522 	mode = sysfs_match_string(zeroing_mode, buf);
523 	if (mode < 0)
524 		return -EINVAL;
525 
526 	sdkp->zeroing_mode = mode;
527 
528 	return count;
529 }
530 static DEVICE_ATTR_RW(zeroing_mode);
531 
532 static ssize_t
533 max_medium_access_timeouts_show(struct device *dev,
534 				struct device_attribute *attr, char *buf)
535 {
536 	struct scsi_disk *sdkp = to_scsi_disk(dev);
537 
538 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
539 }
540 
541 static ssize_t
542 max_medium_access_timeouts_store(struct device *dev,
543 				 struct device_attribute *attr, const char *buf,
544 				 size_t count)
545 {
546 	struct scsi_disk *sdkp = to_scsi_disk(dev);
547 	int err;
548 
549 	if (!capable(CAP_SYS_ADMIN))
550 		return -EACCES;
551 
552 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
553 
554 	return err ? err : count;
555 }
556 static DEVICE_ATTR_RW(max_medium_access_timeouts);
557 
558 static ssize_t
559 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
560 			   char *buf)
561 {
562 	struct scsi_disk *sdkp = to_scsi_disk(dev);
563 
564 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
565 }
566 
567 static ssize_t
568 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
569 			    const char *buf, size_t count)
570 {
571 	struct scsi_disk *sdkp = to_scsi_disk(dev);
572 	struct scsi_device *sdp = sdkp->device;
573 	struct queue_limits lim;
574 	unsigned long max;
575 	int err;
576 
577 	if (!capable(CAP_SYS_ADMIN))
578 		return -EACCES;
579 
580 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
581 		return -EINVAL;
582 
583 	err = kstrtoul(buf, 10, &max);
584 
585 	if (err)
586 		return err;
587 
588 	if (max == 0)
589 		sdp->no_write_same = 1;
590 	else if (max <= SD_MAX_WS16_BLOCKS) {
591 		sdp->no_write_same = 0;
592 		sdkp->max_ws_blocks = max;
593 	}
594 
595 	lim = queue_limits_start_update(sdkp->disk->queue);
596 	sd_config_write_same(sdkp, &lim);
597 	blk_mq_freeze_queue(sdkp->disk->queue);
598 	err = queue_limits_commit_update(sdkp->disk->queue, &lim);
599 	blk_mq_unfreeze_queue(sdkp->disk->queue);
600 	if (err)
601 		return err;
602 	return count;
603 }
604 static DEVICE_ATTR_RW(max_write_same_blocks);
605 
606 static ssize_t
607 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
608 {
609 	struct scsi_disk *sdkp = to_scsi_disk(dev);
610 
611 	if (sdkp->device->type == TYPE_ZBC)
612 		return sprintf(buf, "host-managed\n");
613 	if (sdkp->zoned == 1)
614 		return sprintf(buf, "host-aware\n");
615 	if (sdkp->zoned == 2)
616 		return sprintf(buf, "drive-managed\n");
617 	return sprintf(buf, "none\n");
618 }
619 static DEVICE_ATTR_RO(zoned_cap);
620 
621 static ssize_t
622 max_retries_store(struct device *dev, struct device_attribute *attr,
623 		  const char *buf, size_t count)
624 {
625 	struct scsi_disk *sdkp = to_scsi_disk(dev);
626 	struct scsi_device *sdev = sdkp->device;
627 	int retries, err;
628 
629 	err = kstrtoint(buf, 10, &retries);
630 	if (err)
631 		return err;
632 
633 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
634 		sdkp->max_retries = retries;
635 		return count;
636 	}
637 
638 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
639 		    SD_MAX_RETRIES);
640 	return -EINVAL;
641 }
642 
643 static ssize_t
644 max_retries_show(struct device *dev, struct device_attribute *attr,
645 		 char *buf)
646 {
647 	struct scsi_disk *sdkp = to_scsi_disk(dev);
648 
649 	return sprintf(buf, "%d\n", sdkp->max_retries);
650 }
651 
652 static DEVICE_ATTR_RW(max_retries);
653 
654 static struct attribute *sd_disk_attrs[] = {
655 	&dev_attr_cache_type.attr,
656 	&dev_attr_FUA.attr,
657 	&dev_attr_allow_restart.attr,
658 	&dev_attr_manage_start_stop.attr,
659 	&dev_attr_manage_system_start_stop.attr,
660 	&dev_attr_manage_runtime_start_stop.attr,
661 	&dev_attr_manage_shutdown.attr,
662 	&dev_attr_protection_type.attr,
663 	&dev_attr_protection_mode.attr,
664 	&dev_attr_app_tag_own.attr,
665 	&dev_attr_thin_provisioning.attr,
666 	&dev_attr_provisioning_mode.attr,
667 	&dev_attr_zeroing_mode.attr,
668 	&dev_attr_max_write_same_blocks.attr,
669 	&dev_attr_max_medium_access_timeouts.attr,
670 	&dev_attr_zoned_cap.attr,
671 	&dev_attr_max_retries.attr,
672 	NULL,
673 };
674 ATTRIBUTE_GROUPS(sd_disk);
675 
676 static struct class sd_disk_class = {
677 	.name		= "scsi_disk",
678 	.dev_release	= scsi_disk_release,
679 	.dev_groups	= sd_disk_groups,
680 };
681 
682 /*
683  * Don't request a new module, as that could deadlock in multipath
684  * environment.
685  */
686 static void sd_default_probe(dev_t devt)
687 {
688 }
689 
690 /*
691  * Device no to disk mapping:
692  *
693  *       major         disc2     disc  p1
694  *   |............|.............|....|....| <- dev_t
695  *    31        20 19          8 7  4 3  0
696  *
697  * Inside a major, we have 16k disks, however mapped non-
698  * contiguously. The first 16 disks are for major0, the next
699  * ones with major1, ... Disk 256 is for major0 again, disk 272
700  * for major1, ...
701  * As we stay compatible with our numbering scheme, we can reuse
702  * the well-know SCSI majors 8, 65--71, 136--143.
703  */
704 static int sd_major(int major_idx)
705 {
706 	switch (major_idx) {
707 	case 0:
708 		return SCSI_DISK0_MAJOR;
709 	case 1 ... 7:
710 		return SCSI_DISK1_MAJOR + major_idx - 1;
711 	case 8 ... 15:
712 		return SCSI_DISK8_MAJOR + major_idx - 8;
713 	default:
714 		BUG();
715 		return 0;	/* shut up gcc */
716 	}
717 }
718 
719 #ifdef CONFIG_BLK_SED_OPAL
720 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
721 		size_t len, bool send)
722 {
723 	struct scsi_disk *sdkp = data;
724 	struct scsi_device *sdev = sdkp->device;
725 	u8 cdb[12] = { 0, };
726 	const struct scsi_exec_args exec_args = {
727 		.req_flags = BLK_MQ_REQ_PM,
728 	};
729 	int ret;
730 
731 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
732 	cdb[1] = secp;
733 	put_unaligned_be16(spsp, &cdb[2]);
734 	put_unaligned_be32(len, &cdb[6]);
735 
736 	ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
737 			       buffer, len, SD_TIMEOUT, sdkp->max_retries,
738 			       &exec_args);
739 	return ret <= 0 ? ret : -EIO;
740 }
741 #endif /* CONFIG_BLK_SED_OPAL */
742 
743 /*
744  * Look up the DIX operation based on whether the command is read or
745  * write and whether dix and dif are enabled.
746  */
747 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
748 {
749 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
750 	static const unsigned int ops[] = {	/* wrt dix dif */
751 		SCSI_PROT_NORMAL,		/*  0	0   0  */
752 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
753 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
754 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
755 		SCSI_PROT_NORMAL,		/*  1	0   0  */
756 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
757 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
758 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
759 	};
760 
761 	return ops[write << 2 | dix << 1 | dif];
762 }
763 
764 /*
765  * Returns a mask of the protection flags that are valid for a given DIX
766  * operation.
767  */
768 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
769 {
770 	static const unsigned int flag_mask[] = {
771 		[SCSI_PROT_NORMAL]		= 0,
772 
773 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
774 						  SCSI_PROT_GUARD_CHECK |
775 						  SCSI_PROT_REF_CHECK |
776 						  SCSI_PROT_REF_INCREMENT,
777 
778 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
779 						  SCSI_PROT_IP_CHECKSUM,
780 
781 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
782 						  SCSI_PROT_GUARD_CHECK |
783 						  SCSI_PROT_REF_CHECK |
784 						  SCSI_PROT_REF_INCREMENT |
785 						  SCSI_PROT_IP_CHECKSUM,
786 
787 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
788 						  SCSI_PROT_REF_INCREMENT,
789 
790 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
791 						  SCSI_PROT_REF_CHECK |
792 						  SCSI_PROT_REF_INCREMENT |
793 						  SCSI_PROT_IP_CHECKSUM,
794 
795 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
796 						  SCSI_PROT_GUARD_CHECK |
797 						  SCSI_PROT_REF_CHECK |
798 						  SCSI_PROT_REF_INCREMENT |
799 						  SCSI_PROT_IP_CHECKSUM,
800 	};
801 
802 	return flag_mask[prot_op];
803 }
804 
805 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
806 					   unsigned int dix, unsigned int dif)
807 {
808 	struct request *rq = scsi_cmd_to_rq(scmd);
809 	struct bio *bio = rq->bio;
810 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
811 	unsigned int protect = 0;
812 
813 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
814 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
815 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
816 
817 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
818 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
819 	}
820 
821 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
822 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
823 
824 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
825 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
826 	}
827 
828 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
829 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
830 
831 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
832 			protect = 3 << 5;	/* Disable target PI checking */
833 		else
834 			protect = 1 << 5;	/* Enable target PI checking */
835 	}
836 
837 	scsi_set_prot_op(scmd, prot_op);
838 	scsi_set_prot_type(scmd, dif);
839 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
840 
841 	return protect;
842 }
843 
844 static void sd_disable_discard(struct scsi_disk *sdkp)
845 {
846 	sdkp->provisioning_mode = SD_LBP_DISABLE;
847 	blk_queue_disable_discard(sdkp->disk->queue);
848 }
849 
850 static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
851 		unsigned int mode)
852 {
853 	unsigned int logical_block_size = sdkp->device->sector_size;
854 	unsigned int max_blocks = 0;
855 
856 	lim->discard_alignment = sdkp->unmap_alignment * logical_block_size;
857 	lim->discard_granularity = max(sdkp->physical_block_size,
858 			sdkp->unmap_granularity * logical_block_size);
859 	sdkp->provisioning_mode = mode;
860 
861 	switch (mode) {
862 
863 	case SD_LBP_FULL:
864 	case SD_LBP_DISABLE:
865 		break;
866 
867 	case SD_LBP_UNMAP:
868 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
869 					  (u32)SD_MAX_WS16_BLOCKS);
870 		break;
871 
872 	case SD_LBP_WS16:
873 		if (sdkp->device->unmap_limit_for_ws)
874 			max_blocks = sdkp->max_unmap_blocks;
875 		else
876 			max_blocks = sdkp->max_ws_blocks;
877 
878 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
879 		break;
880 
881 	case SD_LBP_WS10:
882 		if (sdkp->device->unmap_limit_for_ws)
883 			max_blocks = sdkp->max_unmap_blocks;
884 		else
885 			max_blocks = sdkp->max_ws_blocks;
886 
887 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
888 		break;
889 
890 	case SD_LBP_ZERO:
891 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
892 					  (u32)SD_MAX_WS10_BLOCKS);
893 		break;
894 	}
895 
896 	lim->max_hw_discard_sectors = max_blocks *
897 		(logical_block_size >> SECTOR_SHIFT);
898 }
899 
900 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
901 {
902 	struct page *page;
903 
904 	page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
905 	if (!page)
906 		return NULL;
907 	clear_highpage(page);
908 	bvec_set_page(&rq->special_vec, page, data_len, 0);
909 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
910 	return bvec_virt(&rq->special_vec);
911 }
912 
913 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
914 {
915 	struct scsi_device *sdp = cmd->device;
916 	struct request *rq = scsi_cmd_to_rq(cmd);
917 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
918 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
919 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
920 	unsigned int data_len = 24;
921 	char *buf;
922 
923 	buf = sd_set_special_bvec(rq, data_len);
924 	if (!buf)
925 		return BLK_STS_RESOURCE;
926 
927 	cmd->cmd_len = 10;
928 	cmd->cmnd[0] = UNMAP;
929 	cmd->cmnd[8] = 24;
930 
931 	put_unaligned_be16(6 + 16, &buf[0]);
932 	put_unaligned_be16(16, &buf[2]);
933 	put_unaligned_be64(lba, &buf[8]);
934 	put_unaligned_be32(nr_blocks, &buf[16]);
935 
936 	cmd->allowed = sdkp->max_retries;
937 	cmd->transfersize = data_len;
938 	rq->timeout = SD_TIMEOUT;
939 
940 	return scsi_alloc_sgtables(cmd);
941 }
942 
943 static void sd_config_atomic(struct scsi_disk *sdkp, struct queue_limits *lim)
944 {
945 	unsigned int logical_block_size = sdkp->device->sector_size,
946 		physical_block_size_sectors, max_atomic, unit_min, unit_max;
947 
948 	if ((!sdkp->max_atomic && !sdkp->max_atomic_with_boundary) ||
949 	    sdkp->protection_type == T10_PI_TYPE2_PROTECTION)
950 		return;
951 
952 	physical_block_size_sectors = sdkp->physical_block_size /
953 					sdkp->device->sector_size;
954 
955 	unit_min = rounddown_pow_of_two(sdkp->atomic_granularity ?
956 					sdkp->atomic_granularity :
957 					physical_block_size_sectors);
958 
959 	/*
960 	 * Only use atomic boundary when we have the odd scenario of
961 	 * sdkp->max_atomic == 0, which the spec does permit.
962 	 */
963 	if (sdkp->max_atomic) {
964 		max_atomic = sdkp->max_atomic;
965 		unit_max = rounddown_pow_of_two(sdkp->max_atomic);
966 		sdkp->use_atomic_write_boundary = 0;
967 	} else {
968 		max_atomic = sdkp->max_atomic_with_boundary;
969 		unit_max = rounddown_pow_of_two(sdkp->max_atomic_boundary);
970 		sdkp->use_atomic_write_boundary = 1;
971 	}
972 
973 	/*
974 	 * Ensure compliance with granularity and alignment. For now, keep it
975 	 * simple and just don't support atomic writes for values mismatched
976 	 * with max_{boundary}atomic, physical block size, and
977 	 * atomic_granularity itself.
978 	 *
979 	 * We're really being distrustful by checking unit_max also...
980 	 */
981 	if (sdkp->atomic_granularity > 1) {
982 		if (unit_min > 1 && unit_min % sdkp->atomic_granularity)
983 			return;
984 		if (unit_max > 1 && unit_max % sdkp->atomic_granularity)
985 			return;
986 	}
987 
988 	if (sdkp->atomic_alignment > 1) {
989 		if (unit_min > 1 && unit_min % sdkp->atomic_alignment)
990 			return;
991 		if (unit_max > 1 && unit_max % sdkp->atomic_alignment)
992 			return;
993 	}
994 
995 	lim->atomic_write_hw_max = max_atomic * logical_block_size;
996 	lim->atomic_write_hw_boundary = 0;
997 	lim->atomic_write_hw_unit_min = unit_min * logical_block_size;
998 	lim->atomic_write_hw_unit_max = unit_max * logical_block_size;
999 }
1000 
1001 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
1002 		bool unmap)
1003 {
1004 	struct scsi_device *sdp = cmd->device;
1005 	struct request *rq = scsi_cmd_to_rq(cmd);
1006 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1007 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1008 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1009 	u32 data_len = sdp->sector_size;
1010 
1011 	if (!sd_set_special_bvec(rq, data_len))
1012 		return BLK_STS_RESOURCE;
1013 
1014 	cmd->cmd_len = 16;
1015 	cmd->cmnd[0] = WRITE_SAME_16;
1016 	if (unmap)
1017 		cmd->cmnd[1] = 0x8; /* UNMAP */
1018 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1019 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1020 
1021 	cmd->allowed = sdkp->max_retries;
1022 	cmd->transfersize = data_len;
1023 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1024 
1025 	return scsi_alloc_sgtables(cmd);
1026 }
1027 
1028 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
1029 		bool unmap)
1030 {
1031 	struct scsi_device *sdp = cmd->device;
1032 	struct request *rq = scsi_cmd_to_rq(cmd);
1033 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1034 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1035 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1036 	u32 data_len = sdp->sector_size;
1037 
1038 	if (!sd_set_special_bvec(rq, data_len))
1039 		return BLK_STS_RESOURCE;
1040 
1041 	cmd->cmd_len = 10;
1042 	cmd->cmnd[0] = WRITE_SAME;
1043 	if (unmap)
1044 		cmd->cmnd[1] = 0x8; /* UNMAP */
1045 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1046 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1047 
1048 	cmd->allowed = sdkp->max_retries;
1049 	cmd->transfersize = data_len;
1050 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1051 
1052 	return scsi_alloc_sgtables(cmd);
1053 }
1054 
1055 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
1056 {
1057 	struct request *rq = scsi_cmd_to_rq(cmd);
1058 	struct scsi_device *sdp = cmd->device;
1059 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1060 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1061 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1062 
1063 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
1064 		switch (sdkp->zeroing_mode) {
1065 		case SD_ZERO_WS16_UNMAP:
1066 			return sd_setup_write_same16_cmnd(cmd, true);
1067 		case SD_ZERO_WS10_UNMAP:
1068 			return sd_setup_write_same10_cmnd(cmd, true);
1069 		}
1070 	}
1071 
1072 	if (sdp->no_write_same) {
1073 		rq->rq_flags |= RQF_QUIET;
1074 		return BLK_STS_TARGET;
1075 	}
1076 
1077 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
1078 		return sd_setup_write_same16_cmnd(cmd, false);
1079 
1080 	return sd_setup_write_same10_cmnd(cmd, false);
1081 }
1082 
1083 static void sd_disable_write_same(struct scsi_disk *sdkp)
1084 {
1085 	sdkp->device->no_write_same = 1;
1086 	sdkp->max_ws_blocks = 0;
1087 	blk_queue_disable_write_zeroes(sdkp->disk->queue);
1088 }
1089 
1090 static void sd_config_write_same(struct scsi_disk *sdkp,
1091 		struct queue_limits *lim)
1092 {
1093 	unsigned int logical_block_size = sdkp->device->sector_size;
1094 
1095 	if (sdkp->device->no_write_same) {
1096 		sdkp->max_ws_blocks = 0;
1097 		goto out;
1098 	}
1099 
1100 	/* Some devices can not handle block counts above 0xffff despite
1101 	 * supporting WRITE SAME(16). Consequently we default to 64k
1102 	 * blocks per I/O unless the device explicitly advertises a
1103 	 * bigger limit.
1104 	 */
1105 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1106 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1107 						   (u32)SD_MAX_WS16_BLOCKS);
1108 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1109 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1110 						   (u32)SD_MAX_WS10_BLOCKS);
1111 	else {
1112 		sdkp->device->no_write_same = 1;
1113 		sdkp->max_ws_blocks = 0;
1114 	}
1115 
1116 	if (sdkp->lbprz && sdkp->lbpws)
1117 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1118 	else if (sdkp->lbprz && sdkp->lbpws10)
1119 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1120 	else if (sdkp->max_ws_blocks)
1121 		sdkp->zeroing_mode = SD_ZERO_WS;
1122 	else
1123 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1124 
1125 	if (sdkp->max_ws_blocks &&
1126 	    sdkp->physical_block_size > logical_block_size) {
1127 		/*
1128 		 * Reporting a maximum number of blocks that is not aligned
1129 		 * on the device physical size would cause a large write same
1130 		 * request to be split into physically unaligned chunks by
1131 		 * __blkdev_issue_write_zeroes() even if the caller of this
1132 		 * functions took care to align the large request. So make sure
1133 		 * the maximum reported is aligned to the device physical block
1134 		 * size. This is only an optional optimization for regular
1135 		 * disks, but this is mandatory to avoid failure of large write
1136 		 * same requests directed at sequential write required zones of
1137 		 * host-managed ZBC disks.
1138 		 */
1139 		sdkp->max_ws_blocks =
1140 			round_down(sdkp->max_ws_blocks,
1141 				   bytes_to_logical(sdkp->device,
1142 						    sdkp->physical_block_size));
1143 	}
1144 
1145 out:
1146 	lim->max_write_zeroes_sectors =
1147 		sdkp->max_ws_blocks * (logical_block_size >> SECTOR_SHIFT);
1148 }
1149 
1150 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1151 {
1152 	struct request *rq = scsi_cmd_to_rq(cmd);
1153 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1154 
1155 	/* flush requests don't perform I/O, zero the S/G table */
1156 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1157 
1158 	if (cmd->device->use_16_for_sync) {
1159 		cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1160 		cmd->cmd_len = 16;
1161 	} else {
1162 		cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1163 		cmd->cmd_len = 10;
1164 	}
1165 	cmd->transfersize = 0;
1166 	cmd->allowed = sdkp->max_retries;
1167 
1168 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1169 	return BLK_STS_OK;
1170 }
1171 
1172 /**
1173  * sd_group_number() - Compute the GROUP NUMBER field
1174  * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1175  *	field.
1176  *
1177  * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1178  * 0: no relative lifetime.
1179  * 1: shortest relative lifetime.
1180  * 2: second shortest relative lifetime.
1181  * 3 - 0x3d: intermediate relative lifetimes.
1182  * 0x3e: second longest relative lifetime.
1183  * 0x3f: longest relative lifetime.
1184  */
1185 static u8 sd_group_number(struct scsi_cmnd *cmd)
1186 {
1187 	const struct request *rq = scsi_cmd_to_rq(cmd);
1188 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1189 
1190 	if (!sdkp->rscs)
1191 		return 0;
1192 
1193 	return min3((u32)rq->write_hint, (u32)sdkp->permanent_stream_count,
1194 		    0x3fu);
1195 }
1196 
1197 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1198 				       sector_t lba, unsigned int nr_blocks,
1199 				       unsigned char flags, unsigned int dld)
1200 {
1201 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1202 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1203 	cmd->cmnd[6]  = sd_group_number(cmd);
1204 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1205 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1206 	cmd->cmnd[10] = flags;
1207 	cmd->cmnd[11] = dld & 0x07;
1208 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1209 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1210 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1211 
1212 	return BLK_STS_OK;
1213 }
1214 
1215 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1216 				       sector_t lba, unsigned int nr_blocks,
1217 				       unsigned char flags, unsigned int dld)
1218 {
1219 	cmd->cmd_len  = 16;
1220 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1221 	cmd->cmnd[1]  = flags | ((dld >> 2) & 0x01);
1222 	cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1223 	cmd->cmnd[15] = 0;
1224 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1225 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1226 
1227 	return BLK_STS_OK;
1228 }
1229 
1230 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1231 				       sector_t lba, unsigned int nr_blocks,
1232 				       unsigned char flags)
1233 {
1234 	cmd->cmd_len = 10;
1235 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1236 	cmd->cmnd[1] = flags;
1237 	cmd->cmnd[6] = sd_group_number(cmd);
1238 	cmd->cmnd[9] = 0;
1239 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1240 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1241 
1242 	return BLK_STS_OK;
1243 }
1244 
1245 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1246 				      sector_t lba, unsigned int nr_blocks,
1247 				      unsigned char flags)
1248 {
1249 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1250 	if (WARN_ON_ONCE(nr_blocks == 0))
1251 		return BLK_STS_IOERR;
1252 
1253 	if (unlikely(flags & 0x8)) {
1254 		/*
1255 		 * This happens only if this drive failed 10byte rw
1256 		 * command with ILLEGAL_REQUEST during operation and
1257 		 * thus turned off use_10_for_rw.
1258 		 */
1259 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1260 		return BLK_STS_IOERR;
1261 	}
1262 
1263 	cmd->cmd_len = 6;
1264 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1265 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1266 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1267 	cmd->cmnd[3] = lba & 0xff;
1268 	cmd->cmnd[4] = nr_blocks;
1269 	cmd->cmnd[5] = 0;
1270 
1271 	return BLK_STS_OK;
1272 }
1273 
1274 /*
1275  * Check if a command has a duration limit set. If it does, and the target
1276  * device supports CDL and the feature is enabled, return the limit
1277  * descriptor index to use. Return 0 (no limit) otherwise.
1278  */
1279 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1280 {
1281 	struct scsi_device *sdp = sdkp->device;
1282 	int hint;
1283 
1284 	if (!sdp->cdl_supported || !sdp->cdl_enable)
1285 		return 0;
1286 
1287 	/*
1288 	 * Use "no limit" if the request ioprio does not specify a duration
1289 	 * limit hint.
1290 	 */
1291 	hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1292 	if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1293 	    hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1294 		return 0;
1295 
1296 	return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1297 }
1298 
1299 static blk_status_t sd_setup_atomic_cmnd(struct scsi_cmnd *cmd,
1300 					sector_t lba, unsigned int nr_blocks,
1301 					bool boundary, unsigned char flags)
1302 {
1303 	cmd->cmd_len  = 16;
1304 	cmd->cmnd[0]  = WRITE_ATOMIC_16;
1305 	cmd->cmnd[1]  = flags;
1306 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1307 	put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1308 	if (boundary)
1309 		put_unaligned_be16(nr_blocks, &cmd->cmnd[10]);
1310 	else
1311 		put_unaligned_be16(0, &cmd->cmnd[10]);
1312 	put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1313 	cmd->cmnd[14] = 0;
1314 	cmd->cmnd[15] = 0;
1315 
1316 	return BLK_STS_OK;
1317 }
1318 
1319 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1320 {
1321 	struct request *rq = scsi_cmd_to_rq(cmd);
1322 	struct scsi_device *sdp = cmd->device;
1323 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1324 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1325 	sector_t threshold;
1326 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1327 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1328 	bool write = rq_data_dir(rq) == WRITE;
1329 	unsigned char protect, fua;
1330 	unsigned int dld;
1331 	blk_status_t ret;
1332 	unsigned int dif;
1333 	bool dix;
1334 
1335 	ret = scsi_alloc_sgtables(cmd);
1336 	if (ret != BLK_STS_OK)
1337 		return ret;
1338 
1339 	ret = BLK_STS_IOERR;
1340 	if (!scsi_device_online(sdp) || sdp->changed) {
1341 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1342 		goto fail;
1343 	}
1344 
1345 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1346 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1347 		goto fail;
1348 	}
1349 
1350 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1351 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1352 		goto fail;
1353 	}
1354 
1355 	/*
1356 	 * Some SD card readers can't handle accesses which touch the
1357 	 * last one or two logical blocks. Split accesses as needed.
1358 	 */
1359 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1360 
1361 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1362 		if (lba < threshold) {
1363 			/* Access up to the threshold but not beyond */
1364 			nr_blocks = threshold - lba;
1365 		} else {
1366 			/* Access only a single logical block */
1367 			nr_blocks = 1;
1368 		}
1369 	}
1370 
1371 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1372 	dix = scsi_prot_sg_count(cmd);
1373 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1374 	dld = sd_cdl_dld(sdkp, cmd);
1375 
1376 	if (dif || dix)
1377 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1378 	else
1379 		protect = 0;
1380 
1381 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1382 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1383 					 protect | fua, dld);
1384 	} else if (rq->cmd_flags & REQ_ATOMIC) {
1385 		ret = sd_setup_atomic_cmnd(cmd, lba, nr_blocks,
1386 				sdkp->use_atomic_write_boundary,
1387 				protect | fua);
1388 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1389 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1390 					 protect | fua, dld);
1391 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1392 		   sdp->use_10_for_rw || protect || rq->write_hint) {
1393 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1394 					 protect | fua);
1395 	} else {
1396 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1397 					protect | fua);
1398 	}
1399 
1400 	if (unlikely(ret != BLK_STS_OK))
1401 		goto fail;
1402 
1403 	/*
1404 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1405 	 * host adapter, it's safe to assume that we can at least transfer
1406 	 * this many bytes between each connect / disconnect.
1407 	 */
1408 	cmd->transfersize = sdp->sector_size;
1409 	cmd->underflow = nr_blocks << 9;
1410 	cmd->allowed = sdkp->max_retries;
1411 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1412 
1413 	SCSI_LOG_HLQUEUE(1,
1414 			 scmd_printk(KERN_INFO, cmd,
1415 				     "%s: block=%llu, count=%d\n", __func__,
1416 				     (unsigned long long)blk_rq_pos(rq),
1417 				     blk_rq_sectors(rq)));
1418 	SCSI_LOG_HLQUEUE(2,
1419 			 scmd_printk(KERN_INFO, cmd,
1420 				     "%s %d/%u 512 byte blocks.\n",
1421 				     write ? "writing" : "reading", nr_blocks,
1422 				     blk_rq_sectors(rq)));
1423 
1424 	/*
1425 	 * This indicates that the command is ready from our end to be queued.
1426 	 */
1427 	return BLK_STS_OK;
1428 fail:
1429 	scsi_free_sgtables(cmd);
1430 	return ret;
1431 }
1432 
1433 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1434 {
1435 	struct request *rq = scsi_cmd_to_rq(cmd);
1436 
1437 	switch (req_op(rq)) {
1438 	case REQ_OP_DISCARD:
1439 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1440 		case SD_LBP_UNMAP:
1441 			return sd_setup_unmap_cmnd(cmd);
1442 		case SD_LBP_WS16:
1443 			return sd_setup_write_same16_cmnd(cmd, true);
1444 		case SD_LBP_WS10:
1445 			return sd_setup_write_same10_cmnd(cmd, true);
1446 		case SD_LBP_ZERO:
1447 			return sd_setup_write_same10_cmnd(cmd, false);
1448 		default:
1449 			return BLK_STS_TARGET;
1450 		}
1451 	case REQ_OP_WRITE_ZEROES:
1452 		return sd_setup_write_zeroes_cmnd(cmd);
1453 	case REQ_OP_FLUSH:
1454 		return sd_setup_flush_cmnd(cmd);
1455 	case REQ_OP_READ:
1456 	case REQ_OP_WRITE:
1457 		return sd_setup_read_write_cmnd(cmd);
1458 	case REQ_OP_ZONE_RESET:
1459 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1460 						   false);
1461 	case REQ_OP_ZONE_RESET_ALL:
1462 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1463 						   true);
1464 	case REQ_OP_ZONE_OPEN:
1465 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1466 	case REQ_OP_ZONE_CLOSE:
1467 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1468 	case REQ_OP_ZONE_FINISH:
1469 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1470 	default:
1471 		WARN_ON_ONCE(1);
1472 		return BLK_STS_NOTSUPP;
1473 	}
1474 }
1475 
1476 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1477 {
1478 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1479 
1480 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1481 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1482 }
1483 
1484 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1485 {
1486 	if (sdkp->device->removable || sdkp->write_prot) {
1487 		if (disk_check_media_change(disk))
1488 			return true;
1489 	}
1490 
1491 	/*
1492 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1493 	 * nothing to do with partitions, BLKRRPART is used to force a full
1494 	 * revalidate after things like a format for historical reasons.
1495 	 */
1496 	return test_bit(GD_NEED_PART_SCAN, &disk->state);
1497 }
1498 
1499 /**
1500  *	sd_open - open a scsi disk device
1501  *	@disk: disk to open
1502  *	@mode: open mode
1503  *
1504  *	Returns 0 if successful. Returns a negated errno value in case
1505  *	of error.
1506  *
1507  *	Note: This can be called from a user context (e.g. fsck(1) )
1508  *	or from within the kernel (e.g. as a result of a mount(1) ).
1509  *	In the latter case @inode and @filp carry an abridged amount
1510  *	of information as noted above.
1511  *
1512  *	Locking: called with disk->open_mutex held.
1513  **/
1514 static int sd_open(struct gendisk *disk, blk_mode_t mode)
1515 {
1516 	struct scsi_disk *sdkp = scsi_disk(disk);
1517 	struct scsi_device *sdev = sdkp->device;
1518 	int retval;
1519 
1520 	if (scsi_device_get(sdev))
1521 		return -ENXIO;
1522 
1523 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1524 
1525 	/*
1526 	 * If the device is in error recovery, wait until it is done.
1527 	 * If the device is offline, then disallow any access to it.
1528 	 */
1529 	retval = -ENXIO;
1530 	if (!scsi_block_when_processing_errors(sdev))
1531 		goto error_out;
1532 
1533 	if (sd_need_revalidate(disk, sdkp))
1534 		sd_revalidate_disk(disk);
1535 
1536 	/*
1537 	 * If the drive is empty, just let the open fail.
1538 	 */
1539 	retval = -ENOMEDIUM;
1540 	if (sdev->removable && !sdkp->media_present &&
1541 	    !(mode & BLK_OPEN_NDELAY))
1542 		goto error_out;
1543 
1544 	/*
1545 	 * If the device has the write protect tab set, have the open fail
1546 	 * if the user expects to be able to write to the thing.
1547 	 */
1548 	retval = -EROFS;
1549 	if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1550 		goto error_out;
1551 
1552 	/*
1553 	 * It is possible that the disk changing stuff resulted in
1554 	 * the device being taken offline.  If this is the case,
1555 	 * report this to the user, and don't pretend that the
1556 	 * open actually succeeded.
1557 	 */
1558 	retval = -ENXIO;
1559 	if (!scsi_device_online(sdev))
1560 		goto error_out;
1561 
1562 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1563 		if (scsi_block_when_processing_errors(sdev))
1564 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1565 	}
1566 
1567 	return 0;
1568 
1569 error_out:
1570 	scsi_device_put(sdev);
1571 	return retval;
1572 }
1573 
1574 /**
1575  *	sd_release - invoked when the (last) close(2) is called on this
1576  *	scsi disk.
1577  *	@disk: disk to release
1578  *
1579  *	Returns 0.
1580  *
1581  *	Note: may block (uninterruptible) if error recovery is underway
1582  *	on this disk.
1583  *
1584  *	Locking: called with disk->open_mutex held.
1585  **/
1586 static void sd_release(struct gendisk *disk)
1587 {
1588 	struct scsi_disk *sdkp = scsi_disk(disk);
1589 	struct scsi_device *sdev = sdkp->device;
1590 
1591 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1592 
1593 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1594 		if (scsi_block_when_processing_errors(sdev))
1595 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1596 	}
1597 
1598 	scsi_device_put(sdev);
1599 }
1600 
1601 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1602 {
1603 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1604 	struct scsi_device *sdp = sdkp->device;
1605 	struct Scsi_Host *host = sdp->host;
1606 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1607 	int diskinfo[4];
1608 
1609 	/* default to most commonly used values */
1610 	diskinfo[0] = 0x40;	/* 1 << 6 */
1611 	diskinfo[1] = 0x20;	/* 1 << 5 */
1612 	diskinfo[2] = capacity >> 11;
1613 
1614 	/* override with calculated, extended default, or driver values */
1615 	if (host->hostt->bios_param)
1616 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1617 	else
1618 		scsicam_bios_param(bdev, capacity, diskinfo);
1619 
1620 	geo->heads = diskinfo[0];
1621 	geo->sectors = diskinfo[1];
1622 	geo->cylinders = diskinfo[2];
1623 	return 0;
1624 }
1625 
1626 /**
1627  *	sd_ioctl - process an ioctl
1628  *	@bdev: target block device
1629  *	@mode: open mode
1630  *	@cmd: ioctl command number
1631  *	@arg: this is third argument given to ioctl(2) system call.
1632  *	Often contains a pointer.
1633  *
1634  *	Returns 0 if successful (some ioctls return positive numbers on
1635  *	success as well). Returns a negated errno value in case of error.
1636  *
1637  *	Note: most ioctls are forward onto the block subsystem or further
1638  *	down in the scsi subsystem.
1639  **/
1640 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1641 		    unsigned int cmd, unsigned long arg)
1642 {
1643 	struct gendisk *disk = bdev->bd_disk;
1644 	struct scsi_disk *sdkp = scsi_disk(disk);
1645 	struct scsi_device *sdp = sdkp->device;
1646 	void __user *p = (void __user *)arg;
1647 	int error;
1648 
1649 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1650 				    "cmd=0x%x\n", disk->disk_name, cmd));
1651 
1652 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1653 		return -ENOIOCTLCMD;
1654 
1655 	/*
1656 	 * If we are in the middle of error recovery, don't let anyone
1657 	 * else try and use this device.  Also, if error recovery fails, it
1658 	 * may try and take the device offline, in which case all further
1659 	 * access to the device is prohibited.
1660 	 */
1661 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1662 			(mode & BLK_OPEN_NDELAY));
1663 	if (error)
1664 		return error;
1665 
1666 	if (is_sed_ioctl(cmd))
1667 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1668 	return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1669 }
1670 
1671 static void set_media_not_present(struct scsi_disk *sdkp)
1672 {
1673 	if (sdkp->media_present)
1674 		sdkp->device->changed = 1;
1675 
1676 	if (sdkp->device->removable) {
1677 		sdkp->media_present = 0;
1678 		sdkp->capacity = 0;
1679 	}
1680 }
1681 
1682 static int media_not_present(struct scsi_disk *sdkp,
1683 			     struct scsi_sense_hdr *sshdr)
1684 {
1685 	if (!scsi_sense_valid(sshdr))
1686 		return 0;
1687 
1688 	/* not invoked for commands that could return deferred errors */
1689 	switch (sshdr->sense_key) {
1690 	case UNIT_ATTENTION:
1691 	case NOT_READY:
1692 		/* medium not present */
1693 		if (sshdr->asc == 0x3A) {
1694 			set_media_not_present(sdkp);
1695 			return 1;
1696 		}
1697 	}
1698 	return 0;
1699 }
1700 
1701 /**
1702  *	sd_check_events - check media events
1703  *	@disk: kernel device descriptor
1704  *	@clearing: disk events currently being cleared
1705  *
1706  *	Returns mask of DISK_EVENT_*.
1707  *
1708  *	Note: this function is invoked from the block subsystem.
1709  **/
1710 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1711 {
1712 	struct scsi_disk *sdkp = disk->private_data;
1713 	struct scsi_device *sdp;
1714 	int retval;
1715 	bool disk_changed;
1716 
1717 	if (!sdkp)
1718 		return 0;
1719 
1720 	sdp = sdkp->device;
1721 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1722 
1723 	/*
1724 	 * If the device is offline, don't send any commands - just pretend as
1725 	 * if the command failed.  If the device ever comes back online, we
1726 	 * can deal with it then.  It is only because of unrecoverable errors
1727 	 * that we would ever take a device offline in the first place.
1728 	 */
1729 	if (!scsi_device_online(sdp)) {
1730 		set_media_not_present(sdkp);
1731 		goto out;
1732 	}
1733 
1734 	/*
1735 	 * Using TEST_UNIT_READY enables differentiation between drive with
1736 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1737 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1738 	 *
1739 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1740 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1741 	 * sd_revalidate() is called.
1742 	 */
1743 	if (scsi_block_when_processing_errors(sdp)) {
1744 		struct scsi_sense_hdr sshdr = { 0, };
1745 
1746 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1747 					      &sshdr);
1748 
1749 		/* failed to execute TUR, assume media not present */
1750 		if (retval < 0 || host_byte(retval)) {
1751 			set_media_not_present(sdkp);
1752 			goto out;
1753 		}
1754 
1755 		if (media_not_present(sdkp, &sshdr))
1756 			goto out;
1757 	}
1758 
1759 	/*
1760 	 * For removable scsi disk we have to recognise the presence
1761 	 * of a disk in the drive.
1762 	 */
1763 	if (!sdkp->media_present)
1764 		sdp->changed = 1;
1765 	sdkp->media_present = 1;
1766 out:
1767 	/*
1768 	 * sdp->changed is set under the following conditions:
1769 	 *
1770 	 *	Medium present state has changed in either direction.
1771 	 *	Device has indicated UNIT_ATTENTION.
1772 	 */
1773 	disk_changed = sdp->changed;
1774 	sdp->changed = 0;
1775 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1776 }
1777 
1778 static int sd_sync_cache(struct scsi_disk *sdkp)
1779 {
1780 	int res;
1781 	struct scsi_device *sdp = sdkp->device;
1782 	const int timeout = sdp->request_queue->rq_timeout
1783 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1784 	/* Leave the rest of the command zero to indicate flush everything. */
1785 	const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1786 				SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1787 	struct scsi_sense_hdr sshdr;
1788 	struct scsi_failure failure_defs[] = {
1789 		{
1790 			.allowed = 3,
1791 			.result = SCMD_FAILURE_RESULT_ANY,
1792 		},
1793 		{}
1794 	};
1795 	struct scsi_failures failures = {
1796 		.failure_definitions = failure_defs,
1797 	};
1798 	const struct scsi_exec_args exec_args = {
1799 		.req_flags = BLK_MQ_REQ_PM,
1800 		.sshdr = &sshdr,
1801 		.failures = &failures,
1802 	};
1803 
1804 	if (!scsi_device_online(sdp))
1805 		return -ENODEV;
1806 
1807 	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1808 			       sdkp->max_retries, &exec_args);
1809 	if (res) {
1810 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1811 
1812 		if (res < 0)
1813 			return res;
1814 
1815 		if (scsi_status_is_check_condition(res) &&
1816 		    scsi_sense_valid(&sshdr)) {
1817 			sd_print_sense_hdr(sdkp, &sshdr);
1818 
1819 			/* we need to evaluate the error return  */
1820 			if (sshdr.asc == 0x3a ||	/* medium not present */
1821 			    sshdr.asc == 0x20 ||	/* invalid command */
1822 			    (sshdr.asc == 0x74 && sshdr.ascq == 0x71))	/* drive is password locked */
1823 				/* this is no error here */
1824 				return 0;
1825 
1826 			/*
1827 			 * If a format is in progress or if the drive does not
1828 			 * support sync, there is not much we can do because
1829 			 * this is called during shutdown or suspend so just
1830 			 * return success so those operations can proceed.
1831 			 */
1832 			if ((sshdr.asc == 0x04 && sshdr.ascq == 0x04) ||
1833 			    sshdr.sense_key == ILLEGAL_REQUEST)
1834 				return 0;
1835 		}
1836 
1837 		switch (host_byte(res)) {
1838 		/* ignore errors due to racing a disconnection */
1839 		case DID_BAD_TARGET:
1840 		case DID_NO_CONNECT:
1841 			return 0;
1842 		/* signal the upper layer it might try again */
1843 		case DID_BUS_BUSY:
1844 		case DID_IMM_RETRY:
1845 		case DID_REQUEUE:
1846 		case DID_SOFT_ERROR:
1847 			return -EBUSY;
1848 		default:
1849 			return -EIO;
1850 		}
1851 	}
1852 	return 0;
1853 }
1854 
1855 static void sd_rescan(struct device *dev)
1856 {
1857 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1858 
1859 	sd_revalidate_disk(sdkp->disk);
1860 }
1861 
1862 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1863 		enum blk_unique_id type)
1864 {
1865 	struct scsi_device *sdev = scsi_disk(disk)->device;
1866 	const struct scsi_vpd *vpd;
1867 	const unsigned char *d;
1868 	int ret = -ENXIO, len;
1869 
1870 	rcu_read_lock();
1871 	vpd = rcu_dereference(sdev->vpd_pg83);
1872 	if (!vpd)
1873 		goto out_unlock;
1874 
1875 	ret = -EINVAL;
1876 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1877 		/* we only care about designators with LU association */
1878 		if (((d[1] >> 4) & 0x3) != 0x00)
1879 			continue;
1880 		if ((d[1] & 0xf) != type)
1881 			continue;
1882 
1883 		/*
1884 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1885 		 * keep looking as one with more entropy might still show up.
1886 		 */
1887 		len = d[3];
1888 		if (len != 8 && len != 12 && len != 16)
1889 			continue;
1890 		ret = len;
1891 		memcpy(id, d + 4, len);
1892 		if (len == 16)
1893 			break;
1894 	}
1895 out_unlock:
1896 	rcu_read_unlock();
1897 	return ret;
1898 }
1899 
1900 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1901 {
1902 	switch (host_byte(result)) {
1903 	case DID_TRANSPORT_MARGINAL:
1904 	case DID_TRANSPORT_DISRUPTED:
1905 	case DID_BUS_BUSY:
1906 		return PR_STS_RETRY_PATH_FAILURE;
1907 	case DID_NO_CONNECT:
1908 		return PR_STS_PATH_FAILED;
1909 	case DID_TRANSPORT_FAILFAST:
1910 		return PR_STS_PATH_FAST_FAILED;
1911 	}
1912 
1913 	switch (status_byte(result)) {
1914 	case SAM_STAT_RESERVATION_CONFLICT:
1915 		return PR_STS_RESERVATION_CONFLICT;
1916 	case SAM_STAT_CHECK_CONDITION:
1917 		if (!scsi_sense_valid(sshdr))
1918 			return PR_STS_IOERR;
1919 
1920 		if (sshdr->sense_key == ILLEGAL_REQUEST &&
1921 		    (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1922 			return -EINVAL;
1923 
1924 		fallthrough;
1925 	default:
1926 		return PR_STS_IOERR;
1927 	}
1928 }
1929 
1930 static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1931 			    unsigned char *data, int data_len)
1932 {
1933 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1934 	struct scsi_device *sdev = sdkp->device;
1935 	struct scsi_sense_hdr sshdr;
1936 	u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1937 	struct scsi_failure failure_defs[] = {
1938 		{
1939 			.sense = UNIT_ATTENTION,
1940 			.asc = SCMD_FAILURE_ASC_ANY,
1941 			.ascq = SCMD_FAILURE_ASCQ_ANY,
1942 			.allowed = 5,
1943 			.result = SAM_STAT_CHECK_CONDITION,
1944 		},
1945 		{}
1946 	};
1947 	struct scsi_failures failures = {
1948 		.failure_definitions = failure_defs,
1949 	};
1950 	const struct scsi_exec_args exec_args = {
1951 		.sshdr = &sshdr,
1952 		.failures = &failures,
1953 	};
1954 	int result;
1955 
1956 	put_unaligned_be16(data_len, &cmd[7]);
1957 
1958 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1959 				  SD_TIMEOUT, sdkp->max_retries, &exec_args);
1960 	if (scsi_status_is_check_condition(result) &&
1961 	    scsi_sense_valid(&sshdr)) {
1962 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1963 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1964 	}
1965 
1966 	if (result <= 0)
1967 		return result;
1968 
1969 	return sd_scsi_to_pr_err(&sshdr, result);
1970 }
1971 
1972 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1973 {
1974 	int result, i, data_offset, num_copy_keys;
1975 	u32 num_keys = keys_info->num_keys;
1976 	int data_len = num_keys * 8 + 8;
1977 	u8 *data;
1978 
1979 	data = kzalloc(data_len, GFP_KERNEL);
1980 	if (!data)
1981 		return -ENOMEM;
1982 
1983 	result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1984 	if (result)
1985 		goto free_data;
1986 
1987 	keys_info->generation = get_unaligned_be32(&data[0]);
1988 	keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1989 
1990 	data_offset = 8;
1991 	num_copy_keys = min(num_keys, keys_info->num_keys);
1992 
1993 	for (i = 0; i < num_copy_keys; i++) {
1994 		keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1995 		data_offset += 8;
1996 	}
1997 
1998 free_data:
1999 	kfree(data);
2000 	return result;
2001 }
2002 
2003 static int sd_pr_read_reservation(struct block_device *bdev,
2004 				  struct pr_held_reservation *rsv)
2005 {
2006 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2007 	struct scsi_device *sdev = sdkp->device;
2008 	u8 data[24] = { };
2009 	int result, len;
2010 
2011 	result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
2012 	if (result)
2013 		return result;
2014 
2015 	len = get_unaligned_be32(&data[4]);
2016 	if (!len)
2017 		return 0;
2018 
2019 	/* Make sure we have at least the key and type */
2020 	if (len < 14) {
2021 		sdev_printk(KERN_INFO, sdev,
2022 			    "READ RESERVATION failed due to short return buffer of %d bytes\n",
2023 			    len);
2024 		return -EINVAL;
2025 	}
2026 
2027 	rsv->generation = get_unaligned_be32(&data[0]);
2028 	rsv->key = get_unaligned_be64(&data[8]);
2029 	rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
2030 	return 0;
2031 }
2032 
2033 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
2034 			     u64 sa_key, enum scsi_pr_type type, u8 flags)
2035 {
2036 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2037 	struct scsi_device *sdev = sdkp->device;
2038 	struct scsi_sense_hdr sshdr;
2039 	struct scsi_failure failure_defs[] = {
2040 		{
2041 			.sense = UNIT_ATTENTION,
2042 			.asc = SCMD_FAILURE_ASC_ANY,
2043 			.ascq = SCMD_FAILURE_ASCQ_ANY,
2044 			.allowed = 5,
2045 			.result = SAM_STAT_CHECK_CONDITION,
2046 		},
2047 		{}
2048 	};
2049 	struct scsi_failures failures = {
2050 		.failure_definitions = failure_defs,
2051 	};
2052 	const struct scsi_exec_args exec_args = {
2053 		.sshdr = &sshdr,
2054 		.failures = &failures,
2055 	};
2056 	int result;
2057 	u8 cmd[16] = { 0, };
2058 	u8 data[24] = { 0, };
2059 
2060 	cmd[0] = PERSISTENT_RESERVE_OUT;
2061 	cmd[1] = sa;
2062 	cmd[2] = type;
2063 	put_unaligned_be32(sizeof(data), &cmd[5]);
2064 
2065 	put_unaligned_be64(key, &data[0]);
2066 	put_unaligned_be64(sa_key, &data[8]);
2067 	data[20] = flags;
2068 
2069 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
2070 				  sizeof(data), SD_TIMEOUT, sdkp->max_retries,
2071 				  &exec_args);
2072 
2073 	if (scsi_status_is_check_condition(result) &&
2074 	    scsi_sense_valid(&sshdr)) {
2075 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
2076 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
2077 	}
2078 
2079 	if (result <= 0)
2080 		return result;
2081 
2082 	return sd_scsi_to_pr_err(&sshdr, result);
2083 }
2084 
2085 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2086 		u32 flags)
2087 {
2088 	if (flags & ~PR_FL_IGNORE_KEY)
2089 		return -EOPNOTSUPP;
2090 	return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
2091 			old_key, new_key, 0,
2092 			(1 << 0) /* APTPL */);
2093 }
2094 
2095 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2096 		u32 flags)
2097 {
2098 	if (flags)
2099 		return -EOPNOTSUPP;
2100 	return sd_pr_out_command(bdev, 0x01, key, 0,
2101 				 block_pr_type_to_scsi(type), 0);
2102 }
2103 
2104 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2105 {
2106 	return sd_pr_out_command(bdev, 0x02, key, 0,
2107 				 block_pr_type_to_scsi(type), 0);
2108 }
2109 
2110 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2111 		enum pr_type type, bool abort)
2112 {
2113 	return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2114 				 block_pr_type_to_scsi(type), 0);
2115 }
2116 
2117 static int sd_pr_clear(struct block_device *bdev, u64 key)
2118 {
2119 	return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2120 }
2121 
2122 static const struct pr_ops sd_pr_ops = {
2123 	.pr_register	= sd_pr_register,
2124 	.pr_reserve	= sd_pr_reserve,
2125 	.pr_release	= sd_pr_release,
2126 	.pr_preempt	= sd_pr_preempt,
2127 	.pr_clear	= sd_pr_clear,
2128 	.pr_read_keys	= sd_pr_read_keys,
2129 	.pr_read_reservation = sd_pr_read_reservation,
2130 };
2131 
2132 static void scsi_disk_free_disk(struct gendisk *disk)
2133 {
2134 	struct scsi_disk *sdkp = scsi_disk(disk);
2135 
2136 	put_device(&sdkp->disk_dev);
2137 }
2138 
2139 static const struct block_device_operations sd_fops = {
2140 	.owner			= THIS_MODULE,
2141 	.open			= sd_open,
2142 	.release		= sd_release,
2143 	.ioctl			= sd_ioctl,
2144 	.getgeo			= sd_getgeo,
2145 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
2146 	.check_events		= sd_check_events,
2147 	.unlock_native_capacity	= sd_unlock_native_capacity,
2148 	.report_zones		= sd_zbc_report_zones,
2149 	.get_unique_id		= sd_get_unique_id,
2150 	.free_disk		= scsi_disk_free_disk,
2151 	.pr_ops			= &sd_pr_ops,
2152 };
2153 
2154 /**
2155  *	sd_eh_reset - reset error handling callback
2156  *	@scmd:		sd-issued command that has failed
2157  *
2158  *	This function is called by the SCSI midlayer before starting
2159  *	SCSI EH. When counting medium access failures we have to be
2160  *	careful to register it only only once per device and SCSI EH run;
2161  *	there might be several timed out commands which will cause the
2162  *	'max_medium_access_timeouts' counter to trigger after the first
2163  *	SCSI EH run already and set the device to offline.
2164  *	So this function resets the internal counter before starting SCSI EH.
2165  **/
2166 static void sd_eh_reset(struct scsi_cmnd *scmd)
2167 {
2168 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2169 
2170 	/* New SCSI EH run, reset gate variable */
2171 	sdkp->ignore_medium_access_errors = false;
2172 }
2173 
2174 /**
2175  *	sd_eh_action - error handling callback
2176  *	@scmd:		sd-issued command that has failed
2177  *	@eh_disp:	The recovery disposition suggested by the midlayer
2178  *
2179  *	This function is called by the SCSI midlayer upon completion of an
2180  *	error test command (currently TEST UNIT READY). The result of sending
2181  *	the eh command is passed in eh_disp.  We're looking for devices that
2182  *	fail medium access commands but are OK with non access commands like
2183  *	test unit ready (so wrongly see the device as having a successful
2184  *	recovery)
2185  **/
2186 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2187 {
2188 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2189 	struct scsi_device *sdev = scmd->device;
2190 
2191 	if (!scsi_device_online(sdev) ||
2192 	    !scsi_medium_access_command(scmd) ||
2193 	    host_byte(scmd->result) != DID_TIME_OUT ||
2194 	    eh_disp != SUCCESS)
2195 		return eh_disp;
2196 
2197 	/*
2198 	 * The device has timed out executing a medium access command.
2199 	 * However, the TEST UNIT READY command sent during error
2200 	 * handling completed successfully. Either the device is in the
2201 	 * process of recovering or has it suffered an internal failure
2202 	 * that prevents access to the storage medium.
2203 	 */
2204 	if (!sdkp->ignore_medium_access_errors) {
2205 		sdkp->medium_access_timed_out++;
2206 		sdkp->ignore_medium_access_errors = true;
2207 	}
2208 
2209 	/*
2210 	 * If the device keeps failing read/write commands but TEST UNIT
2211 	 * READY always completes successfully we assume that medium
2212 	 * access is no longer possible and take the device offline.
2213 	 */
2214 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2215 		scmd_printk(KERN_ERR, scmd,
2216 			    "Medium access timeout failure. Offlining disk!\n");
2217 		mutex_lock(&sdev->state_mutex);
2218 		scsi_device_set_state(sdev, SDEV_OFFLINE);
2219 		mutex_unlock(&sdev->state_mutex);
2220 
2221 		return SUCCESS;
2222 	}
2223 
2224 	return eh_disp;
2225 }
2226 
2227 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2228 {
2229 	struct request *req = scsi_cmd_to_rq(scmd);
2230 	struct scsi_device *sdev = scmd->device;
2231 	unsigned int transferred, good_bytes;
2232 	u64 start_lba, end_lba, bad_lba;
2233 
2234 	/*
2235 	 * Some commands have a payload smaller than the device logical
2236 	 * block size (e.g. INQUIRY on a 4K disk).
2237 	 */
2238 	if (scsi_bufflen(scmd) <= sdev->sector_size)
2239 		return 0;
2240 
2241 	/* Check if we have a 'bad_lba' information */
2242 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2243 				     SCSI_SENSE_BUFFERSIZE,
2244 				     &bad_lba))
2245 		return 0;
2246 
2247 	/*
2248 	 * If the bad lba was reported incorrectly, we have no idea where
2249 	 * the error is.
2250 	 */
2251 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2252 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2253 	if (bad_lba < start_lba || bad_lba >= end_lba)
2254 		return 0;
2255 
2256 	/*
2257 	 * resid is optional but mostly filled in.  When it's unused,
2258 	 * its value is zero, so we assume the whole buffer transferred
2259 	 */
2260 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2261 
2262 	/* This computation should always be done in terms of the
2263 	 * resolution of the device's medium.
2264 	 */
2265 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2266 
2267 	return min(good_bytes, transferred);
2268 }
2269 
2270 /**
2271  *	sd_done - bottom half handler: called when the lower level
2272  *	driver has completed (successfully or otherwise) a scsi command.
2273  *	@SCpnt: mid-level's per command structure.
2274  *
2275  *	Note: potentially run from within an ISR. Must not block.
2276  **/
2277 static int sd_done(struct scsi_cmnd *SCpnt)
2278 {
2279 	int result = SCpnt->result;
2280 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2281 	unsigned int sector_size = SCpnt->device->sector_size;
2282 	unsigned int resid;
2283 	struct scsi_sense_hdr sshdr;
2284 	struct request *req = scsi_cmd_to_rq(SCpnt);
2285 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2286 	int sense_valid = 0;
2287 	int sense_deferred = 0;
2288 
2289 	switch (req_op(req)) {
2290 	case REQ_OP_DISCARD:
2291 	case REQ_OP_WRITE_ZEROES:
2292 	case REQ_OP_ZONE_RESET:
2293 	case REQ_OP_ZONE_RESET_ALL:
2294 	case REQ_OP_ZONE_OPEN:
2295 	case REQ_OP_ZONE_CLOSE:
2296 	case REQ_OP_ZONE_FINISH:
2297 		if (!result) {
2298 			good_bytes = blk_rq_bytes(req);
2299 			scsi_set_resid(SCpnt, 0);
2300 		} else {
2301 			good_bytes = 0;
2302 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2303 		}
2304 		break;
2305 	default:
2306 		/*
2307 		 * In case of bogus fw or device, we could end up having
2308 		 * an unaligned partial completion. Check this here and force
2309 		 * alignment.
2310 		 */
2311 		resid = scsi_get_resid(SCpnt);
2312 		if (resid & (sector_size - 1)) {
2313 			sd_printk(KERN_INFO, sdkp,
2314 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2315 				resid, sector_size);
2316 			scsi_print_command(SCpnt);
2317 			resid = min(scsi_bufflen(SCpnt),
2318 				    round_up(resid, sector_size));
2319 			scsi_set_resid(SCpnt, resid);
2320 		}
2321 	}
2322 
2323 	if (result) {
2324 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2325 		if (sense_valid)
2326 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2327 	}
2328 	sdkp->medium_access_timed_out = 0;
2329 
2330 	if (!scsi_status_is_check_condition(result) &&
2331 	    (!sense_valid || sense_deferred))
2332 		goto out;
2333 
2334 	switch (sshdr.sense_key) {
2335 	case HARDWARE_ERROR:
2336 	case MEDIUM_ERROR:
2337 		good_bytes = sd_completed_bytes(SCpnt);
2338 		break;
2339 	case RECOVERED_ERROR:
2340 		good_bytes = scsi_bufflen(SCpnt);
2341 		break;
2342 	case NO_SENSE:
2343 		/* This indicates a false check condition, so ignore it.  An
2344 		 * unknown amount of data was transferred so treat it as an
2345 		 * error.
2346 		 */
2347 		SCpnt->result = 0;
2348 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2349 		break;
2350 	case ABORTED_COMMAND:
2351 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2352 			good_bytes = sd_completed_bytes(SCpnt);
2353 		break;
2354 	case ILLEGAL_REQUEST:
2355 		switch (sshdr.asc) {
2356 		case 0x10:	/* DIX: Host detected corruption */
2357 			good_bytes = sd_completed_bytes(SCpnt);
2358 			break;
2359 		case 0x20:	/* INVALID COMMAND OPCODE */
2360 		case 0x24:	/* INVALID FIELD IN CDB */
2361 			switch (SCpnt->cmnd[0]) {
2362 			case UNMAP:
2363 				sd_disable_discard(sdkp);
2364 				break;
2365 			case WRITE_SAME_16:
2366 			case WRITE_SAME:
2367 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2368 					sd_disable_discard(sdkp);
2369 				} else {
2370 					sd_disable_write_same(sdkp);
2371 					req->rq_flags |= RQF_QUIET;
2372 				}
2373 				break;
2374 			}
2375 		}
2376 		break;
2377 	default:
2378 		break;
2379 	}
2380 
2381  out:
2382 	if (sdkp->device->type == TYPE_ZBC)
2383 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2384 
2385 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2386 					   "sd_done: completed %d of %d bytes\n",
2387 					   good_bytes, scsi_bufflen(SCpnt)));
2388 
2389 	return good_bytes;
2390 }
2391 
2392 /*
2393  * spinup disk - called only in sd_revalidate_disk()
2394  */
2395 static void
2396 sd_spinup_disk(struct scsi_disk *sdkp)
2397 {
2398 	static const u8 cmd[10] = { TEST_UNIT_READY };
2399 	unsigned long spintime_expire = 0;
2400 	int spintime, sense_valid = 0;
2401 	unsigned int the_result;
2402 	struct scsi_sense_hdr sshdr;
2403 	struct scsi_failure failure_defs[] = {
2404 		/* Do not retry Medium Not Present */
2405 		{
2406 			.sense = UNIT_ATTENTION,
2407 			.asc = 0x3A,
2408 			.ascq = SCMD_FAILURE_ASCQ_ANY,
2409 			.result = SAM_STAT_CHECK_CONDITION,
2410 		},
2411 		{
2412 			.sense = NOT_READY,
2413 			.asc = 0x3A,
2414 			.ascq = SCMD_FAILURE_ASCQ_ANY,
2415 			.result = SAM_STAT_CHECK_CONDITION,
2416 		},
2417 		/* Retry when scsi_status_is_good would return false 3 times */
2418 		{
2419 			.result = SCMD_FAILURE_STAT_ANY,
2420 			.allowed = 3,
2421 		},
2422 		{}
2423 	};
2424 	struct scsi_failures failures = {
2425 		.failure_definitions = failure_defs,
2426 	};
2427 	const struct scsi_exec_args exec_args = {
2428 		.sshdr = &sshdr,
2429 		.failures = &failures,
2430 	};
2431 
2432 	spintime = 0;
2433 
2434 	/* Spin up drives, as required.  Only do this at boot time */
2435 	/* Spinup needs to be done for module loads too. */
2436 	do {
2437 		bool media_was_present = sdkp->media_present;
2438 
2439 		scsi_failures_reset_retries(&failures);
2440 
2441 		the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2442 					      NULL, 0, SD_TIMEOUT,
2443 					      sdkp->max_retries, &exec_args);
2444 
2445 
2446 		if (the_result > 0) {
2447 			/*
2448 			 * If the drive has indicated to us that it doesn't
2449 			 * have any media in it, don't bother with any more
2450 			 * polling.
2451 			 */
2452 			if (media_not_present(sdkp, &sshdr)) {
2453 				if (media_was_present)
2454 					sd_printk(KERN_NOTICE, sdkp,
2455 						  "Media removed, stopped polling\n");
2456 				return;
2457 			}
2458 			sense_valid = scsi_sense_valid(&sshdr);
2459 		}
2460 
2461 		if (!scsi_status_is_check_condition(the_result)) {
2462 			/* no sense, TUR either succeeded or failed
2463 			 * with a status error */
2464 			if(!spintime && !scsi_status_is_good(the_result)) {
2465 				sd_print_result(sdkp, "Test Unit Ready failed",
2466 						the_result);
2467 			}
2468 			break;
2469 		}
2470 
2471 		/*
2472 		 * The device does not want the automatic start to be issued.
2473 		 */
2474 		if (sdkp->device->no_start_on_add)
2475 			break;
2476 
2477 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2478 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2479 				break;	/* manual intervention required */
2480 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2481 				break;	/* standby */
2482 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2483 				break;	/* unavailable */
2484 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2485 				break;	/* sanitize in progress */
2486 			if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2487 				break;	/* depopulation in progress */
2488 			if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2489 				break;	/* depopulation restoration in progress */
2490 			/*
2491 			 * Issue command to spin up drive when not ready
2492 			 */
2493 			if (!spintime) {
2494 				/* Return immediately and start spin cycle */
2495 				const u8 start_cmd[10] = {
2496 					[0] = START_STOP,
2497 					[1] = 1,
2498 					[4] = sdkp->device->start_stop_pwr_cond ?
2499 						0x11 : 1,
2500 				};
2501 
2502 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2503 				scsi_execute_cmd(sdkp->device, start_cmd,
2504 						 REQ_OP_DRV_IN, NULL, 0,
2505 						 SD_TIMEOUT, sdkp->max_retries,
2506 						 &exec_args);
2507 				spintime_expire = jiffies + 100 * HZ;
2508 				spintime = 1;
2509 			}
2510 			/* Wait 1 second for next try */
2511 			msleep(1000);
2512 			printk(KERN_CONT ".");
2513 
2514 		/*
2515 		 * Wait for USB flash devices with slow firmware.
2516 		 * Yes, this sense key/ASC combination shouldn't
2517 		 * occur here.  It's characteristic of these devices.
2518 		 */
2519 		} else if (sense_valid &&
2520 				sshdr.sense_key == UNIT_ATTENTION &&
2521 				sshdr.asc == 0x28) {
2522 			if (!spintime) {
2523 				spintime_expire = jiffies + 5 * HZ;
2524 				spintime = 1;
2525 			}
2526 			/* Wait 1 second for next try */
2527 			msleep(1000);
2528 		} else {
2529 			/* we don't understand the sense code, so it's
2530 			 * probably pointless to loop */
2531 			if(!spintime) {
2532 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2533 				sd_print_sense_hdr(sdkp, &sshdr);
2534 			}
2535 			break;
2536 		}
2537 
2538 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2539 
2540 	if (spintime) {
2541 		if (scsi_status_is_good(the_result))
2542 			printk(KERN_CONT "ready\n");
2543 		else
2544 			printk(KERN_CONT "not responding...\n");
2545 	}
2546 }
2547 
2548 /*
2549  * Determine whether disk supports Data Integrity Field.
2550  */
2551 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2552 {
2553 	struct scsi_device *sdp = sdkp->device;
2554 	u8 type;
2555 
2556 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2557 		sdkp->protection_type = 0;
2558 		return 0;
2559 	}
2560 
2561 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2562 
2563 	if (type > T10_PI_TYPE3_PROTECTION) {
2564 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported"	\
2565 			  " protection type %u. Disabling disk!\n",
2566 			  type);
2567 		sdkp->protection_type = 0;
2568 		return -ENODEV;
2569 	}
2570 
2571 	sdkp->protection_type = type;
2572 
2573 	return 0;
2574 }
2575 
2576 static void sd_config_protection(struct scsi_disk *sdkp,
2577 		struct queue_limits *lim)
2578 {
2579 	struct scsi_device *sdp = sdkp->device;
2580 
2581 	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
2582 		sd_dif_config_host(sdkp, lim);
2583 
2584 	if (!sdkp->protection_type)
2585 		return;
2586 
2587 	if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2588 		sd_first_printk(KERN_NOTICE, sdkp,
2589 				"Disabling DIF Type %u protection\n",
2590 				sdkp->protection_type);
2591 		sdkp->protection_type = 0;
2592 	}
2593 
2594 	sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2595 			sdkp->protection_type);
2596 }
2597 
2598 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2599 			struct scsi_sense_hdr *sshdr, int sense_valid,
2600 			int the_result)
2601 {
2602 	if (sense_valid)
2603 		sd_print_sense_hdr(sdkp, sshdr);
2604 	else
2605 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2606 
2607 	/*
2608 	 * Set dirty bit for removable devices if not ready -
2609 	 * sometimes drives will not report this properly.
2610 	 */
2611 	if (sdp->removable &&
2612 	    sense_valid && sshdr->sense_key == NOT_READY)
2613 		set_media_not_present(sdkp);
2614 
2615 	/*
2616 	 * We used to set media_present to 0 here to indicate no media
2617 	 * in the drive, but some drives fail read capacity even with
2618 	 * media present, so we can't do that.
2619 	 */
2620 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2621 }
2622 
2623 #define RC16_LEN 32
2624 #if RC16_LEN > SD_BUF_SIZE
2625 #error RC16_LEN must not be more than SD_BUF_SIZE
2626 #endif
2627 
2628 #define READ_CAPACITY_RETRIES_ON_RESET	10
2629 
2630 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2631 		struct queue_limits *lim, unsigned char *buffer)
2632 {
2633 	unsigned char cmd[16];
2634 	struct scsi_sense_hdr sshdr;
2635 	const struct scsi_exec_args exec_args = {
2636 		.sshdr = &sshdr,
2637 	};
2638 	int sense_valid = 0;
2639 	int the_result;
2640 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2641 	unsigned int alignment;
2642 	unsigned long long lba;
2643 	unsigned sector_size;
2644 
2645 	if (sdp->no_read_capacity_16)
2646 		return -EINVAL;
2647 
2648 	do {
2649 		memset(cmd, 0, 16);
2650 		cmd[0] = SERVICE_ACTION_IN_16;
2651 		cmd[1] = SAI_READ_CAPACITY_16;
2652 		cmd[13] = RC16_LEN;
2653 		memset(buffer, 0, RC16_LEN);
2654 
2655 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2656 					      buffer, RC16_LEN, SD_TIMEOUT,
2657 					      sdkp->max_retries, &exec_args);
2658 		if (the_result > 0) {
2659 			if (media_not_present(sdkp, &sshdr))
2660 				return -ENODEV;
2661 
2662 			sense_valid = scsi_sense_valid(&sshdr);
2663 			if (sense_valid &&
2664 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2665 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2666 			    sshdr.ascq == 0x00)
2667 				/* Invalid Command Operation Code or
2668 				 * Invalid Field in CDB, just retry
2669 				 * silently with RC10 */
2670 				return -EINVAL;
2671 			if (sense_valid &&
2672 			    sshdr.sense_key == UNIT_ATTENTION &&
2673 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2674 				/* Device reset might occur several times,
2675 				 * give it one more chance */
2676 				if (--reset_retries > 0)
2677 					continue;
2678 		}
2679 		retries--;
2680 
2681 	} while (the_result && retries);
2682 
2683 	if (the_result) {
2684 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2685 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2686 		return -EINVAL;
2687 	}
2688 
2689 	sector_size = get_unaligned_be32(&buffer[8]);
2690 	lba = get_unaligned_be64(&buffer[0]);
2691 
2692 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2693 		sdkp->capacity = 0;
2694 		return -ENODEV;
2695 	}
2696 
2697 	/* Logical blocks per physical block exponent */
2698 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2699 
2700 	/* RC basis */
2701 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2702 
2703 	/* Lowest aligned logical block */
2704 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2705 	lim->alignment_offset = alignment;
2706 	if (alignment && sdkp->first_scan)
2707 		sd_printk(KERN_NOTICE, sdkp,
2708 			  "physical block alignment offset: %u\n", alignment);
2709 
2710 	if (buffer[14] & 0x80) { /* LBPME */
2711 		sdkp->lbpme = 1;
2712 
2713 		if (buffer[14] & 0x40) /* LBPRZ */
2714 			sdkp->lbprz = 1;
2715 	}
2716 
2717 	sdkp->capacity = lba + 1;
2718 	return sector_size;
2719 }
2720 
2721 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2722 						unsigned char *buffer)
2723 {
2724 	static const u8 cmd[10] = { READ_CAPACITY };
2725 	struct scsi_sense_hdr sshdr;
2726 	struct scsi_failure failure_defs[] = {
2727 		/* Do not retry Medium Not Present */
2728 		{
2729 			.sense = UNIT_ATTENTION,
2730 			.asc = 0x3A,
2731 			.result = SAM_STAT_CHECK_CONDITION,
2732 		},
2733 		{
2734 			.sense = NOT_READY,
2735 			.asc = 0x3A,
2736 			.result = SAM_STAT_CHECK_CONDITION,
2737 		},
2738 		 /* Device reset might occur several times so retry a lot */
2739 		{
2740 			.sense = UNIT_ATTENTION,
2741 			.asc = 0x29,
2742 			.allowed = READ_CAPACITY_RETRIES_ON_RESET,
2743 			.result = SAM_STAT_CHECK_CONDITION,
2744 		},
2745 		/* Any other error not listed above retry 3 times */
2746 		{
2747 			.result = SCMD_FAILURE_RESULT_ANY,
2748 			.allowed = 3,
2749 		},
2750 		{}
2751 	};
2752 	struct scsi_failures failures = {
2753 		.failure_definitions = failure_defs,
2754 	};
2755 	const struct scsi_exec_args exec_args = {
2756 		.sshdr = &sshdr,
2757 		.failures = &failures,
2758 	};
2759 	int sense_valid = 0;
2760 	int the_result;
2761 	sector_t lba;
2762 	unsigned sector_size;
2763 
2764 	memset(buffer, 0, 8);
2765 
2766 	the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2767 				      8, SD_TIMEOUT, sdkp->max_retries,
2768 				      &exec_args);
2769 
2770 	if (the_result > 0) {
2771 		sense_valid = scsi_sense_valid(&sshdr);
2772 
2773 		if (media_not_present(sdkp, &sshdr))
2774 			return -ENODEV;
2775 	}
2776 
2777 	if (the_result) {
2778 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2779 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2780 		return -EINVAL;
2781 	}
2782 
2783 	sector_size = get_unaligned_be32(&buffer[4]);
2784 	lba = get_unaligned_be32(&buffer[0]);
2785 
2786 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2787 		/* Some buggy (usb cardreader) devices return an lba of
2788 		   0xffffffff when the want to report a size of 0 (with
2789 		   which they really mean no media is present) */
2790 		sdkp->capacity = 0;
2791 		sdkp->physical_block_size = sector_size;
2792 		return sector_size;
2793 	}
2794 
2795 	sdkp->capacity = lba + 1;
2796 	sdkp->physical_block_size = sector_size;
2797 	return sector_size;
2798 }
2799 
2800 static int sd_try_rc16_first(struct scsi_device *sdp)
2801 {
2802 	if (sdp->host->max_cmd_len < 16)
2803 		return 0;
2804 	if (sdp->try_rc_10_first)
2805 		return 0;
2806 	if (sdp->scsi_level > SCSI_SPC_2)
2807 		return 1;
2808 	if (scsi_device_protection(sdp))
2809 		return 1;
2810 	return 0;
2811 }
2812 
2813 /*
2814  * read disk capacity
2815  */
2816 static void
2817 sd_read_capacity(struct scsi_disk *sdkp, struct queue_limits *lim,
2818 		unsigned char *buffer)
2819 {
2820 	int sector_size;
2821 	struct scsi_device *sdp = sdkp->device;
2822 
2823 	if (sd_try_rc16_first(sdp)) {
2824 		sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2825 		if (sector_size == -EOVERFLOW)
2826 			goto got_data;
2827 		if (sector_size == -ENODEV)
2828 			return;
2829 		if (sector_size < 0)
2830 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2831 		if (sector_size < 0)
2832 			return;
2833 	} else {
2834 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2835 		if (sector_size == -EOVERFLOW)
2836 			goto got_data;
2837 		if (sector_size < 0)
2838 			return;
2839 		if ((sizeof(sdkp->capacity) > 4) &&
2840 		    (sdkp->capacity > 0xffffffffULL)) {
2841 			int old_sector_size = sector_size;
2842 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2843 					"Trying to use READ CAPACITY(16).\n");
2844 			sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2845 			if (sector_size < 0) {
2846 				sd_printk(KERN_NOTICE, sdkp,
2847 					"Using 0xffffffff as device size\n");
2848 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2849 				sector_size = old_sector_size;
2850 				goto got_data;
2851 			}
2852 			/* Remember that READ CAPACITY(16) succeeded */
2853 			sdp->try_rc_10_first = 0;
2854 		}
2855 	}
2856 
2857 	/* Some devices are known to return the total number of blocks,
2858 	 * not the highest block number.  Some devices have versions
2859 	 * which do this and others which do not.  Some devices we might
2860 	 * suspect of doing this but we don't know for certain.
2861 	 *
2862 	 * If we know the reported capacity is wrong, decrement it.  If
2863 	 * we can only guess, then assume the number of blocks is even
2864 	 * (usually true but not always) and err on the side of lowering
2865 	 * the capacity.
2866 	 */
2867 	if (sdp->fix_capacity ||
2868 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2869 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2870 				"from its reported value: %llu\n",
2871 				(unsigned long long) sdkp->capacity);
2872 		--sdkp->capacity;
2873 	}
2874 
2875 got_data:
2876 	if (sector_size == 0) {
2877 		sector_size = 512;
2878 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2879 			  "assuming 512.\n");
2880 	}
2881 
2882 	if (sector_size != 512 &&
2883 	    sector_size != 1024 &&
2884 	    sector_size != 2048 &&
2885 	    sector_size != 4096) {
2886 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2887 			  sector_size);
2888 		/*
2889 		 * The user might want to re-format the drive with
2890 		 * a supported sectorsize.  Once this happens, it
2891 		 * would be relatively trivial to set the thing up.
2892 		 * For this reason, we leave the thing in the table.
2893 		 */
2894 		sdkp->capacity = 0;
2895 		/*
2896 		 * set a bogus sector size so the normal read/write
2897 		 * logic in the block layer will eventually refuse any
2898 		 * request on this device without tripping over power
2899 		 * of two sector size assumptions
2900 		 */
2901 		sector_size = 512;
2902 	}
2903 	lim->logical_block_size = sector_size;
2904 	lim->physical_block_size = sdkp->physical_block_size;
2905 	sdkp->device->sector_size = sector_size;
2906 
2907 	if (sdkp->capacity > 0xffffffff)
2908 		sdp->use_16_for_rw = 1;
2909 
2910 }
2911 
2912 /*
2913  * Print disk capacity
2914  */
2915 static void
2916 sd_print_capacity(struct scsi_disk *sdkp,
2917 		  sector_t old_capacity)
2918 {
2919 	int sector_size = sdkp->device->sector_size;
2920 	char cap_str_2[10], cap_str_10[10];
2921 
2922 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2923 		return;
2924 
2925 	string_get_size(sdkp->capacity, sector_size,
2926 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2927 	string_get_size(sdkp->capacity, sector_size,
2928 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2929 
2930 	sd_printk(KERN_NOTICE, sdkp,
2931 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2932 		  (unsigned long long)sdkp->capacity,
2933 		  sector_size, cap_str_10, cap_str_2);
2934 
2935 	if (sdkp->physical_block_size != sector_size)
2936 		sd_printk(KERN_NOTICE, sdkp,
2937 			  "%u-byte physical blocks\n",
2938 			  sdkp->physical_block_size);
2939 }
2940 
2941 /* called with buffer of length 512 */
2942 static inline int
2943 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2944 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2945 		 struct scsi_sense_hdr *sshdr)
2946 {
2947 	/*
2948 	 * If we must use MODE SENSE(10), make sure that the buffer length
2949 	 * is at least 8 bytes so that the mode sense header fits.
2950 	 */
2951 	if (sdkp->device->use_10_for_ms && len < 8)
2952 		len = 8;
2953 
2954 	return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2955 			       SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2956 }
2957 
2958 /*
2959  * read write protect setting, if possible - called only in sd_revalidate_disk()
2960  * called with buffer of length SD_BUF_SIZE
2961  */
2962 static void
2963 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2964 {
2965 	int res;
2966 	struct scsi_device *sdp = sdkp->device;
2967 	struct scsi_mode_data data;
2968 	int old_wp = sdkp->write_prot;
2969 
2970 	set_disk_ro(sdkp->disk, 0);
2971 	if (sdp->skip_ms_page_3f) {
2972 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2973 		return;
2974 	}
2975 
2976 	if (sdp->use_192_bytes_for_3f) {
2977 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2978 	} else {
2979 		/*
2980 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2981 		 * We have to start carefully: some devices hang if we ask
2982 		 * for more than is available.
2983 		 */
2984 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2985 
2986 		/*
2987 		 * Second attempt: ask for page 0 When only page 0 is
2988 		 * implemented, a request for page 3F may return Sense Key
2989 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2990 		 * CDB.
2991 		 */
2992 		if (res < 0)
2993 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2994 
2995 		/*
2996 		 * Third attempt: ask 255 bytes, as we did earlier.
2997 		 */
2998 		if (res < 0)
2999 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
3000 					       &data, NULL);
3001 	}
3002 
3003 	if (res < 0) {
3004 		sd_first_printk(KERN_WARNING, sdkp,
3005 			  "Test WP failed, assume Write Enabled\n");
3006 	} else {
3007 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
3008 		set_disk_ro(sdkp->disk, sdkp->write_prot);
3009 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
3010 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
3011 				  sdkp->write_prot ? "on" : "off");
3012 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
3013 		}
3014 	}
3015 }
3016 
3017 /*
3018  * sd_read_cache_type - called only from sd_revalidate_disk()
3019  * called with buffer of length SD_BUF_SIZE
3020  */
3021 static void
3022 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
3023 {
3024 	int len = 0, res;
3025 	struct scsi_device *sdp = sdkp->device;
3026 
3027 	int dbd;
3028 	int modepage;
3029 	int first_len;
3030 	struct scsi_mode_data data;
3031 	struct scsi_sense_hdr sshdr;
3032 	int old_wce = sdkp->WCE;
3033 	int old_rcd = sdkp->RCD;
3034 	int old_dpofua = sdkp->DPOFUA;
3035 
3036 
3037 	if (sdkp->cache_override)
3038 		return;
3039 
3040 	first_len = 4;
3041 	if (sdp->skip_ms_page_8) {
3042 		if (sdp->type == TYPE_RBC)
3043 			goto defaults;
3044 		else {
3045 			if (sdp->skip_ms_page_3f)
3046 				goto defaults;
3047 			modepage = 0x3F;
3048 			if (sdp->use_192_bytes_for_3f)
3049 				first_len = 192;
3050 			dbd = 0;
3051 		}
3052 	} else if (sdp->type == TYPE_RBC) {
3053 		modepage = 6;
3054 		dbd = 8;
3055 	} else {
3056 		modepage = 8;
3057 		dbd = 0;
3058 	}
3059 
3060 	/* cautiously ask */
3061 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
3062 			&data, &sshdr);
3063 
3064 	if (res < 0)
3065 		goto bad_sense;
3066 
3067 	if (!data.header_length) {
3068 		modepage = 6;
3069 		first_len = 0;
3070 		sd_first_printk(KERN_ERR, sdkp,
3071 				"Missing header in MODE_SENSE response\n");
3072 	}
3073 
3074 	/* that went OK, now ask for the proper length */
3075 	len = data.length;
3076 
3077 	/*
3078 	 * We're only interested in the first three bytes, actually.
3079 	 * But the data cache page is defined for the first 20.
3080 	 */
3081 	if (len < 3)
3082 		goto bad_sense;
3083 	else if (len > SD_BUF_SIZE) {
3084 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
3085 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
3086 		len = SD_BUF_SIZE;
3087 	}
3088 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
3089 		len = 192;
3090 
3091 	/* Get the data */
3092 	if (len > first_len)
3093 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
3094 				&data, &sshdr);
3095 
3096 	if (!res) {
3097 		int offset = data.header_length + data.block_descriptor_length;
3098 
3099 		while (offset < len) {
3100 			u8 page_code = buffer[offset] & 0x3F;
3101 			u8 spf       = buffer[offset] & 0x40;
3102 
3103 			if (page_code == 8 || page_code == 6) {
3104 				/* We're interested only in the first 3 bytes.
3105 				 */
3106 				if (len - offset <= 2) {
3107 					sd_first_printk(KERN_ERR, sdkp,
3108 						"Incomplete mode parameter "
3109 							"data\n");
3110 					goto defaults;
3111 				} else {
3112 					modepage = page_code;
3113 					goto Page_found;
3114 				}
3115 			} else {
3116 				/* Go to the next page */
3117 				if (spf && len - offset > 3)
3118 					offset += 4 + (buffer[offset+2] << 8) +
3119 						buffer[offset+3];
3120 				else if (!spf && len - offset > 1)
3121 					offset += 2 + buffer[offset+1];
3122 				else {
3123 					sd_first_printk(KERN_ERR, sdkp,
3124 							"Incomplete mode "
3125 							"parameter data\n");
3126 					goto defaults;
3127 				}
3128 			}
3129 		}
3130 
3131 		sd_first_printk(KERN_WARNING, sdkp,
3132 				"No Caching mode page found\n");
3133 		goto defaults;
3134 
3135 	Page_found:
3136 		if (modepage == 8) {
3137 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3138 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3139 		} else {
3140 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3141 			sdkp->RCD = 0;
3142 		}
3143 
3144 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3145 		if (sdp->broken_fua) {
3146 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3147 			sdkp->DPOFUA = 0;
3148 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3149 			   !sdkp->device->use_16_for_rw) {
3150 			sd_first_printk(KERN_NOTICE, sdkp,
3151 				  "Uses READ/WRITE(6), disabling FUA\n");
3152 			sdkp->DPOFUA = 0;
3153 		}
3154 
3155 		/* No cache flush allowed for write protected devices */
3156 		if (sdkp->WCE && sdkp->write_prot)
3157 			sdkp->WCE = 0;
3158 
3159 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
3160 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3161 			sd_printk(KERN_NOTICE, sdkp,
3162 				  "Write cache: %s, read cache: %s, %s\n",
3163 				  sdkp->WCE ? "enabled" : "disabled",
3164 				  sdkp->RCD ? "disabled" : "enabled",
3165 				  sdkp->DPOFUA ? "supports DPO and FUA"
3166 				  : "doesn't support DPO or FUA");
3167 
3168 		return;
3169 	}
3170 
3171 bad_sense:
3172 	if (res == -EIO && scsi_sense_valid(&sshdr) &&
3173 	    sshdr.sense_key == ILLEGAL_REQUEST &&
3174 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3175 		/* Invalid field in CDB */
3176 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3177 	else
3178 		sd_first_printk(KERN_ERR, sdkp,
3179 				"Asking for cache data failed\n");
3180 
3181 defaults:
3182 	if (sdp->wce_default_on) {
3183 		sd_first_printk(KERN_NOTICE, sdkp,
3184 				"Assuming drive cache: write back\n");
3185 		sdkp->WCE = 1;
3186 	} else {
3187 		sd_first_printk(KERN_WARNING, sdkp,
3188 				"Assuming drive cache: write through\n");
3189 		sdkp->WCE = 0;
3190 	}
3191 	sdkp->RCD = 0;
3192 	sdkp->DPOFUA = 0;
3193 }
3194 
3195 static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3196 {
3197 	u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3198 	struct {
3199 		struct scsi_stream_status_header h;
3200 		struct scsi_stream_status s;
3201 	} buf;
3202 	struct scsi_device *sdev = sdkp->device;
3203 	struct scsi_sense_hdr sshdr;
3204 	const struct scsi_exec_args exec_args = {
3205 		.sshdr = &sshdr,
3206 	};
3207 	int res;
3208 
3209 	put_unaligned_be16(stream_id, &cdb[4]);
3210 	put_unaligned_be32(sizeof(buf), &cdb[10]);
3211 
3212 	res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3213 			       SD_TIMEOUT, sdkp->max_retries, &exec_args);
3214 	if (res < 0)
3215 		return false;
3216 	if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3217 		sd_print_sense_hdr(sdkp, &sshdr);
3218 	if (res)
3219 		return false;
3220 	if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3221 		return false;
3222 	return buf.h.stream_status[0].perm;
3223 }
3224 
3225 static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3226 {
3227 	struct scsi_device *sdp = sdkp->device;
3228 	const struct scsi_io_group_descriptor *desc, *start, *end;
3229 	u16 permanent_stream_count_old;
3230 	struct scsi_sense_hdr sshdr;
3231 	struct scsi_mode_data data;
3232 	int res;
3233 
3234 	if (sdp->sdev_bflags & BLIST_SKIP_IO_HINTS)
3235 		return;
3236 
3237 	res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3238 			      /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3239 			      sdkp->max_retries, &data, &sshdr);
3240 	if (res < 0)
3241 		return;
3242 	start = (void *)buffer + data.header_length + 16;
3243 	end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3244 					  sizeof(*end));
3245 	/*
3246 	 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3247 	 * should assign the lowest numbered stream identifiers to permanent
3248 	 * streams.
3249 	 */
3250 	for (desc = start; desc < end; desc++)
3251 		if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3252 			break;
3253 	permanent_stream_count_old = sdkp->permanent_stream_count;
3254 	sdkp->permanent_stream_count = desc - start;
3255 	if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3256 		sd_printk(KERN_INFO, sdkp,
3257 			  "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3258 			  sdkp->permanent_stream_count);
3259 	else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3260 		sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3261 			  sdkp->permanent_stream_count);
3262 }
3263 
3264 /*
3265  * The ATO bit indicates whether the DIF application tag is available
3266  * for use by the operating system.
3267  */
3268 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3269 {
3270 	int res, offset;
3271 	struct scsi_device *sdp = sdkp->device;
3272 	struct scsi_mode_data data;
3273 	struct scsi_sense_hdr sshdr;
3274 
3275 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3276 		return;
3277 
3278 	if (sdkp->protection_type == 0)
3279 		return;
3280 
3281 	res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3282 			      sdkp->max_retries, &data, &sshdr);
3283 
3284 	if (res < 0 || !data.header_length ||
3285 	    data.length < 6) {
3286 		sd_first_printk(KERN_WARNING, sdkp,
3287 			  "getting Control mode page failed, assume no ATO\n");
3288 
3289 		if (res == -EIO && scsi_sense_valid(&sshdr))
3290 			sd_print_sense_hdr(sdkp, &sshdr);
3291 
3292 		return;
3293 	}
3294 
3295 	offset = data.header_length + data.block_descriptor_length;
3296 
3297 	if ((buffer[offset] & 0x3f) != 0x0a) {
3298 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3299 		return;
3300 	}
3301 
3302 	if ((buffer[offset + 5] & 0x80) == 0)
3303 		return;
3304 
3305 	sdkp->ATO = 1;
3306 
3307 	return;
3308 }
3309 
3310 static unsigned int sd_discard_mode(struct scsi_disk *sdkp)
3311 {
3312 	if (!sdkp->lbpme)
3313 		return SD_LBP_FULL;
3314 
3315 	if (!sdkp->lbpvpd) {
3316 		/* LBP VPD page not provided */
3317 		if (sdkp->max_unmap_blocks)
3318 			return SD_LBP_UNMAP;
3319 		return SD_LBP_WS16;
3320 	}
3321 
3322 	/* LBP VPD page tells us what to use */
3323 	if (sdkp->lbpu && sdkp->max_unmap_blocks)
3324 		return SD_LBP_UNMAP;
3325 	if (sdkp->lbpws)
3326 		return SD_LBP_WS16;
3327 	if (sdkp->lbpws10)
3328 		return SD_LBP_WS10;
3329 	return SD_LBP_DISABLE;
3330 }
3331 
3332 /*
3333  * Query disk device for preferred I/O sizes.
3334  */
3335 static void sd_read_block_limits(struct scsi_disk *sdkp,
3336 		struct queue_limits *lim)
3337 {
3338 	struct scsi_vpd *vpd;
3339 
3340 	rcu_read_lock();
3341 
3342 	vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3343 	if (!vpd || vpd->len < 16)
3344 		goto out;
3345 
3346 	sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3347 	sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3348 	sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3349 
3350 	if (vpd->len >= 64) {
3351 		unsigned int lba_count, desc_count;
3352 
3353 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3354 
3355 		if (!sdkp->lbpme)
3356 			goto config_atomic;
3357 
3358 		lba_count = get_unaligned_be32(&vpd->data[20]);
3359 		desc_count = get_unaligned_be32(&vpd->data[24]);
3360 
3361 		if (lba_count && desc_count)
3362 			sdkp->max_unmap_blocks = lba_count;
3363 
3364 		sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3365 
3366 		if (vpd->data[32] & 0x80)
3367 			sdkp->unmap_alignment =
3368 				get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3369 
3370 config_atomic:
3371 		sdkp->max_atomic = get_unaligned_be32(&vpd->data[44]);
3372 		sdkp->atomic_alignment = get_unaligned_be32(&vpd->data[48]);
3373 		sdkp->atomic_granularity = get_unaligned_be32(&vpd->data[52]);
3374 		sdkp->max_atomic_with_boundary = get_unaligned_be32(&vpd->data[56]);
3375 		sdkp->max_atomic_boundary = get_unaligned_be32(&vpd->data[60]);
3376 
3377 		sd_config_atomic(sdkp, lim);
3378 	}
3379 
3380  out:
3381 	rcu_read_unlock();
3382 }
3383 
3384 /* Parse the Block Limits Extension VPD page (0xb7) */
3385 static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3386 {
3387 	struct scsi_vpd *vpd;
3388 
3389 	rcu_read_lock();
3390 	vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3391 	if (vpd && vpd->len >= 2)
3392 		sdkp->rscs = vpd->data[5] & 1;
3393 	rcu_read_unlock();
3394 }
3395 
3396 /* Query block device characteristics */
3397 static void sd_read_block_characteristics(struct scsi_disk *sdkp,
3398 		struct queue_limits *lim)
3399 {
3400 	struct scsi_vpd *vpd;
3401 	u16 rot;
3402 
3403 	rcu_read_lock();
3404 	vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3405 
3406 	if (!vpd || vpd->len <= 8) {
3407 		rcu_read_unlock();
3408 	        return;
3409 	}
3410 
3411 	rot = get_unaligned_be16(&vpd->data[4]);
3412 	sdkp->zoned = (vpd->data[8] >> 4) & 3;
3413 	rcu_read_unlock();
3414 
3415 	if (rot == 1)
3416 		lim->features &= ~(BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3417 
3418 	if (!sdkp->first_scan)
3419 		return;
3420 
3421 	if (sdkp->device->type == TYPE_ZBC)
3422 		sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3423 	else if (sdkp->zoned == 1)
3424 		sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3425 	else if (sdkp->zoned == 2)
3426 		sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3427 }
3428 
3429 /**
3430  * sd_read_block_provisioning - Query provisioning VPD page
3431  * @sdkp: disk to query
3432  */
3433 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3434 {
3435 	struct scsi_vpd *vpd;
3436 
3437 	if (sdkp->lbpme == 0)
3438 		return;
3439 
3440 	rcu_read_lock();
3441 	vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3442 
3443 	if (!vpd || vpd->len < 8) {
3444 		rcu_read_unlock();
3445 		return;
3446 	}
3447 
3448 	sdkp->lbpvpd	= 1;
3449 	sdkp->lbpu	= (vpd->data[5] >> 7) & 1; /* UNMAP */
3450 	sdkp->lbpws	= (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3451 	sdkp->lbpws10	= (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3452 	rcu_read_unlock();
3453 }
3454 
3455 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3456 {
3457 	struct scsi_device *sdev = sdkp->device;
3458 
3459 	if (sdev->host->no_write_same) {
3460 		sdev->no_write_same = 1;
3461 
3462 		return;
3463 	}
3464 
3465 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3466 		struct scsi_vpd *vpd;
3467 
3468 		sdev->no_report_opcodes = 1;
3469 
3470 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3471 		 * CODES is unsupported and the device has an ATA
3472 		 * Information VPD page (SAT).
3473 		 */
3474 		rcu_read_lock();
3475 		vpd = rcu_dereference(sdev->vpd_pg89);
3476 		if (vpd)
3477 			sdev->no_write_same = 1;
3478 		rcu_read_unlock();
3479 	}
3480 
3481 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3482 		sdkp->ws16 = 1;
3483 
3484 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3485 		sdkp->ws10 = 1;
3486 }
3487 
3488 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3489 {
3490 	struct scsi_device *sdev = sdkp->device;
3491 
3492 	if (!sdev->security_supported)
3493 		return;
3494 
3495 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3496 			SECURITY_PROTOCOL_IN, 0) == 1 &&
3497 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3498 			SECURITY_PROTOCOL_OUT, 0) == 1)
3499 		sdkp->security = 1;
3500 }
3501 
3502 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3503 {
3504 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3505 }
3506 
3507 /**
3508  * sd_read_cpr - Query concurrent positioning ranges
3509  * @sdkp:	disk to query
3510  */
3511 static void sd_read_cpr(struct scsi_disk *sdkp)
3512 {
3513 	struct blk_independent_access_ranges *iars = NULL;
3514 	unsigned char *buffer = NULL;
3515 	unsigned int nr_cpr = 0;
3516 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3517 	u8 *desc;
3518 
3519 	/*
3520 	 * We need to have the capacity set first for the block layer to be
3521 	 * able to check the ranges.
3522 	 */
3523 	if (sdkp->first_scan)
3524 		return;
3525 
3526 	if (!sdkp->capacity)
3527 		goto out;
3528 
3529 	/*
3530 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3531 	 * leading to a maximum page size of 64 + 256*32 bytes.
3532 	 */
3533 	buf_len = 64 + 256*32;
3534 	buffer = kmalloc(buf_len, GFP_KERNEL);
3535 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3536 		goto out;
3537 
3538 	/* We must have at least a 64B header and one 32B range descriptor */
3539 	vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3540 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3541 		sd_printk(KERN_ERR, sdkp,
3542 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3543 		goto out;
3544 	}
3545 
3546 	nr_cpr = (vpd_len - 64) / 32;
3547 	if (nr_cpr == 1) {
3548 		nr_cpr = 0;
3549 		goto out;
3550 	}
3551 
3552 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3553 	if (!iars) {
3554 		nr_cpr = 0;
3555 		goto out;
3556 	}
3557 
3558 	desc = &buffer[64];
3559 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3560 		if (desc[0] != i) {
3561 			sd_printk(KERN_ERR, sdkp,
3562 				"Invalid Concurrent Positioning Range number\n");
3563 			nr_cpr = 0;
3564 			break;
3565 		}
3566 
3567 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3568 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3569 	}
3570 
3571 out:
3572 	disk_set_independent_access_ranges(sdkp->disk, iars);
3573 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3574 		sd_printk(KERN_NOTICE, sdkp,
3575 			  "%u concurrent positioning ranges\n", nr_cpr);
3576 		sdkp->nr_actuators = nr_cpr;
3577 	}
3578 
3579 	kfree(buffer);
3580 }
3581 
3582 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3583 {
3584 	struct scsi_device *sdp = sdkp->device;
3585 	unsigned int min_xfer_bytes =
3586 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3587 
3588 	if (sdkp->min_xfer_blocks == 0)
3589 		return false;
3590 
3591 	if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3592 		sd_first_printk(KERN_WARNING, sdkp,
3593 				"Preferred minimum I/O size %u bytes not a " \
3594 				"multiple of physical block size (%u bytes)\n",
3595 				min_xfer_bytes, sdkp->physical_block_size);
3596 		sdkp->min_xfer_blocks = 0;
3597 		return false;
3598 	}
3599 
3600 	sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3601 			min_xfer_bytes);
3602 	return true;
3603 }
3604 
3605 /*
3606  * Determine the device's preferred I/O size for reads and writes
3607  * unless the reported value is unreasonably small, large, not a
3608  * multiple of the physical block size, or simply garbage.
3609  */
3610 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3611 				      unsigned int dev_max)
3612 {
3613 	struct scsi_device *sdp = sdkp->device;
3614 	unsigned int opt_xfer_bytes =
3615 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3616 	unsigned int min_xfer_bytes =
3617 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3618 
3619 	if (sdkp->opt_xfer_blocks == 0)
3620 		return false;
3621 
3622 	if (sdkp->opt_xfer_blocks > dev_max) {
3623 		sd_first_printk(KERN_WARNING, sdkp,
3624 				"Optimal transfer size %u logical blocks " \
3625 				"> dev_max (%u logical blocks)\n",
3626 				sdkp->opt_xfer_blocks, dev_max);
3627 		return false;
3628 	}
3629 
3630 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3631 		sd_first_printk(KERN_WARNING, sdkp,
3632 				"Optimal transfer size %u logical blocks " \
3633 				"> sd driver limit (%u logical blocks)\n",
3634 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3635 		return false;
3636 	}
3637 
3638 	if (opt_xfer_bytes < PAGE_SIZE) {
3639 		sd_first_printk(KERN_WARNING, sdkp,
3640 				"Optimal transfer size %u bytes < " \
3641 				"PAGE_SIZE (%u bytes)\n",
3642 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3643 		return false;
3644 	}
3645 
3646 	if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3647 		sd_first_printk(KERN_WARNING, sdkp,
3648 				"Optimal transfer size %u bytes not a " \
3649 				"multiple of preferred minimum block " \
3650 				"size (%u bytes)\n",
3651 				opt_xfer_bytes, min_xfer_bytes);
3652 		return false;
3653 	}
3654 
3655 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3656 		sd_first_printk(KERN_WARNING, sdkp,
3657 				"Optimal transfer size %u bytes not a " \
3658 				"multiple of physical block size (%u bytes)\n",
3659 				opt_xfer_bytes, sdkp->physical_block_size);
3660 		return false;
3661 	}
3662 
3663 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3664 			opt_xfer_bytes);
3665 	return true;
3666 }
3667 
3668 static void sd_read_block_zero(struct scsi_disk *sdkp)
3669 {
3670 	struct scsi_device *sdev = sdkp->device;
3671 	unsigned int buf_len = sdev->sector_size;
3672 	u8 *buffer, cmd[16] = { };
3673 
3674 	buffer = kmalloc(buf_len, GFP_KERNEL);
3675 	if (!buffer)
3676 		return;
3677 
3678 	if (sdev->use_16_for_rw) {
3679 		cmd[0] = READ_16;
3680 		put_unaligned_be64(0, &cmd[2]); /* Logical block address 0 */
3681 		put_unaligned_be32(1, &cmd[10]);/* Transfer 1 logical block */
3682 	} else {
3683 		cmd[0] = READ_10;
3684 		put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3685 		put_unaligned_be16(1, &cmd[7]);	/* Transfer 1 logical block */
3686 	}
3687 
3688 	scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3689 			 SD_TIMEOUT, sdkp->max_retries, NULL);
3690 	kfree(buffer);
3691 }
3692 
3693 /**
3694  *	sd_revalidate_disk - called the first time a new disk is seen,
3695  *	performs disk spin up, read_capacity, etc.
3696  *	@disk: struct gendisk we care about
3697  **/
3698 static int sd_revalidate_disk(struct gendisk *disk)
3699 {
3700 	struct scsi_disk *sdkp = scsi_disk(disk);
3701 	struct scsi_device *sdp = sdkp->device;
3702 	sector_t old_capacity = sdkp->capacity;
3703 	struct queue_limits lim;
3704 	unsigned char *buffer;
3705 	unsigned int dev_max;
3706 	int err;
3707 
3708 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3709 				      "sd_revalidate_disk\n"));
3710 
3711 	/*
3712 	 * If the device is offline, don't try and read capacity or any
3713 	 * of the other niceties.
3714 	 */
3715 	if (!scsi_device_online(sdp))
3716 		goto out;
3717 
3718 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3719 	if (!buffer) {
3720 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3721 			  "allocation failure.\n");
3722 		goto out;
3723 	}
3724 
3725 	sd_spinup_disk(sdkp);
3726 
3727 	lim = queue_limits_start_update(sdkp->disk->queue);
3728 
3729 	/*
3730 	 * Without media there is no reason to ask; moreover, some devices
3731 	 * react badly if we do.
3732 	 */
3733 	if (sdkp->media_present) {
3734 		sd_read_capacity(sdkp, &lim, buffer);
3735 		/*
3736 		 * Some USB/UAS devices return generic values for mode pages
3737 		 * until the media has been accessed. Trigger a READ operation
3738 		 * to force the device to populate mode pages.
3739 		 */
3740 		if (sdp->read_before_ms)
3741 			sd_read_block_zero(sdkp);
3742 		/*
3743 		 * set the default to rotational.  All non-rotational devices
3744 		 * support the block characteristics VPD page, which will
3745 		 * cause this to be updated correctly and any device which
3746 		 * doesn't support it should be treated as rotational.
3747 		 */
3748 		lim.features |= (BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3749 
3750 		if (scsi_device_supports_vpd(sdp)) {
3751 			sd_read_block_provisioning(sdkp);
3752 			sd_read_block_limits(sdkp, &lim);
3753 			sd_read_block_limits_ext(sdkp);
3754 			sd_read_block_characteristics(sdkp, &lim);
3755 			sd_zbc_read_zones(sdkp, &lim, buffer);
3756 		}
3757 
3758 		sd_config_discard(sdkp, &lim, sd_discard_mode(sdkp));
3759 
3760 		sd_print_capacity(sdkp, old_capacity);
3761 
3762 		sd_read_write_protect_flag(sdkp, buffer);
3763 		sd_read_cache_type(sdkp, buffer);
3764 		sd_read_io_hints(sdkp, buffer);
3765 		sd_read_app_tag_own(sdkp, buffer);
3766 		sd_read_write_same(sdkp, buffer);
3767 		sd_read_security(sdkp, buffer);
3768 		sd_config_protection(sdkp, &lim);
3769 	}
3770 
3771 	/*
3772 	 * We now have all cache related info, determine how we deal
3773 	 * with flush requests.
3774 	 */
3775 	sd_set_flush_flag(sdkp, &lim);
3776 
3777 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3778 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3779 
3780 	/* Some devices report a maximum block count for READ/WRITE requests. */
3781 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3782 	lim.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3783 
3784 	if (sd_validate_min_xfer_size(sdkp))
3785 		lim.io_min = logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3786 	else
3787 		lim.io_min = 0;
3788 
3789 	/*
3790 	 * Limit default to SCSI host optimal sector limit if set. There may be
3791 	 * an impact on performance for when the size of a request exceeds this
3792 	 * host limit.
3793 	 */
3794 	lim.io_opt = sdp->host->opt_sectors << SECTOR_SHIFT;
3795 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3796 		lim.io_opt = min_not_zero(lim.io_opt,
3797 				logical_to_bytes(sdp, sdkp->opt_xfer_blocks));
3798 	}
3799 
3800 	sdkp->first_scan = 0;
3801 
3802 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3803 	sd_config_write_same(sdkp, &lim);
3804 	kfree(buffer);
3805 
3806 	blk_mq_freeze_queue(sdkp->disk->queue);
3807 	err = queue_limits_commit_update(sdkp->disk->queue, &lim);
3808 	blk_mq_unfreeze_queue(sdkp->disk->queue);
3809 	if (err)
3810 		return err;
3811 
3812 	/*
3813 	 * Query concurrent positioning ranges after
3814 	 * queue_limits_commit_update() unlocked q->limits_lock to avoid
3815 	 * deadlock with q->sysfs_dir_lock and q->sysfs_lock.
3816 	 */
3817 	if (sdkp->media_present && scsi_device_supports_vpd(sdp))
3818 		sd_read_cpr(sdkp);
3819 
3820 	/*
3821 	 * For a zoned drive, revalidating the zones can be done only once
3822 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3823 	 * capacity to 0.
3824 	 */
3825 	if (sd_zbc_revalidate_zones(sdkp))
3826 		set_capacity_and_notify(disk, 0);
3827 
3828  out:
3829 	return 0;
3830 }
3831 
3832 /**
3833  *	sd_unlock_native_capacity - unlock native capacity
3834  *	@disk: struct gendisk to set capacity for
3835  *
3836  *	Block layer calls this function if it detects that partitions
3837  *	on @disk reach beyond the end of the device.  If the SCSI host
3838  *	implements ->unlock_native_capacity() method, it's invoked to
3839  *	give it a chance to adjust the device capacity.
3840  *
3841  *	CONTEXT:
3842  *	Defined by block layer.  Might sleep.
3843  */
3844 static void sd_unlock_native_capacity(struct gendisk *disk)
3845 {
3846 	struct scsi_device *sdev = scsi_disk(disk)->device;
3847 
3848 	if (sdev->host->hostt->unlock_native_capacity)
3849 		sdev->host->hostt->unlock_native_capacity(sdev);
3850 }
3851 
3852 /**
3853  *	sd_format_disk_name - format disk name
3854  *	@prefix: name prefix - ie. "sd" for SCSI disks
3855  *	@index: index of the disk to format name for
3856  *	@buf: output buffer
3857  *	@buflen: length of the output buffer
3858  *
3859  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3860  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3861  *	which is followed by sdaaa.
3862  *
3863  *	This is basically 26 base counting with one extra 'nil' entry
3864  *	at the beginning from the second digit on and can be
3865  *	determined using similar method as 26 base conversion with the
3866  *	index shifted -1 after each digit is computed.
3867  *
3868  *	CONTEXT:
3869  *	Don't care.
3870  *
3871  *	RETURNS:
3872  *	0 on success, -errno on failure.
3873  */
3874 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3875 {
3876 	const int base = 'z' - 'a' + 1;
3877 	char *begin = buf + strlen(prefix);
3878 	char *end = buf + buflen;
3879 	char *p;
3880 	int unit;
3881 
3882 	p = end - 1;
3883 	*p = '\0';
3884 	unit = base;
3885 	do {
3886 		if (p == begin)
3887 			return -EINVAL;
3888 		*--p = 'a' + (index % unit);
3889 		index = (index / unit) - 1;
3890 	} while (index >= 0);
3891 
3892 	memmove(begin, p, end - p);
3893 	memcpy(buf, prefix, strlen(prefix));
3894 
3895 	return 0;
3896 }
3897 
3898 /**
3899  *	sd_probe - called during driver initialization and whenever a
3900  *	new scsi device is attached to the system. It is called once
3901  *	for each scsi device (not just disks) present.
3902  *	@dev: pointer to device object
3903  *
3904  *	Returns 0 if successful (or not interested in this scsi device
3905  *	(e.g. scanner)); 1 when there is an error.
3906  *
3907  *	Note: this function is invoked from the scsi mid-level.
3908  *	This function sets up the mapping between a given
3909  *	<host,channel,id,lun> (found in sdp) and new device name
3910  *	(e.g. /dev/sda). More precisely it is the block device major
3911  *	and minor number that is chosen here.
3912  *
3913  *	Assume sd_probe is not re-entrant (for time being)
3914  *	Also think about sd_probe() and sd_remove() running coincidentally.
3915  **/
3916 static int sd_probe(struct device *dev)
3917 {
3918 	struct scsi_device *sdp = to_scsi_device(dev);
3919 	struct scsi_disk *sdkp;
3920 	struct gendisk *gd;
3921 	int index;
3922 	int error;
3923 
3924 	scsi_autopm_get_device(sdp);
3925 	error = -ENODEV;
3926 	if (sdp->type != TYPE_DISK &&
3927 	    sdp->type != TYPE_ZBC &&
3928 	    sdp->type != TYPE_MOD &&
3929 	    sdp->type != TYPE_RBC)
3930 		goto out;
3931 
3932 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3933 		sdev_printk(KERN_WARNING, sdp,
3934 			    "Unsupported ZBC host-managed device.\n");
3935 		goto out;
3936 	}
3937 
3938 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3939 					"sd_probe\n"));
3940 
3941 	error = -ENOMEM;
3942 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3943 	if (!sdkp)
3944 		goto out;
3945 
3946 	gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3947 					 &sd_bio_compl_lkclass);
3948 	if (!gd)
3949 		goto out_free;
3950 
3951 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3952 	if (index < 0) {
3953 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3954 		goto out_put;
3955 	}
3956 
3957 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3958 	if (error) {
3959 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3960 		goto out_free_index;
3961 	}
3962 
3963 	sdkp->device = sdp;
3964 	sdkp->disk = gd;
3965 	sdkp->index = index;
3966 	sdkp->max_retries = SD_MAX_RETRIES;
3967 	atomic_set(&sdkp->openers, 0);
3968 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3969 
3970 	if (!sdp->request_queue->rq_timeout) {
3971 		if (sdp->type != TYPE_MOD)
3972 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3973 		else
3974 			blk_queue_rq_timeout(sdp->request_queue,
3975 					     SD_MOD_TIMEOUT);
3976 	}
3977 
3978 	device_initialize(&sdkp->disk_dev);
3979 	sdkp->disk_dev.parent = get_device(dev);
3980 	sdkp->disk_dev.class = &sd_disk_class;
3981 	dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3982 
3983 	error = device_add(&sdkp->disk_dev);
3984 	if (error) {
3985 		put_device(&sdkp->disk_dev);
3986 		goto out;
3987 	}
3988 
3989 	dev_set_drvdata(dev, sdkp);
3990 
3991 	gd->major = sd_major((index & 0xf0) >> 4);
3992 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3993 	gd->minors = SD_MINORS;
3994 
3995 	gd->fops = &sd_fops;
3996 	gd->private_data = sdkp;
3997 
3998 	/* defaults, until the device tells us otherwise */
3999 	sdp->sector_size = 512;
4000 	sdkp->capacity = 0;
4001 	sdkp->media_present = 1;
4002 	sdkp->write_prot = 0;
4003 	sdkp->cache_override = 0;
4004 	sdkp->WCE = 0;
4005 	sdkp->RCD = 0;
4006 	sdkp->ATO = 0;
4007 	sdkp->first_scan = 1;
4008 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
4009 
4010 	sd_revalidate_disk(gd);
4011 
4012 	if (sdp->removable) {
4013 		gd->flags |= GENHD_FL_REMOVABLE;
4014 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
4015 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
4016 	}
4017 
4018 	blk_pm_runtime_init(sdp->request_queue, dev);
4019 	if (sdp->rpm_autosuspend) {
4020 		pm_runtime_set_autosuspend_delay(dev,
4021 			sdp->host->rpm_autosuspend_delay);
4022 	}
4023 
4024 	error = device_add_disk(dev, gd, NULL);
4025 	if (error) {
4026 		device_unregister(&sdkp->disk_dev);
4027 		put_disk(gd);
4028 		goto out;
4029 	}
4030 
4031 	if (sdkp->security) {
4032 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
4033 		if (sdkp->opal_dev)
4034 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
4035 	}
4036 
4037 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
4038 		  sdp->removable ? "removable " : "");
4039 	scsi_autopm_put_device(sdp);
4040 
4041 	return 0;
4042 
4043  out_free_index:
4044 	ida_free(&sd_index_ida, index);
4045  out_put:
4046 	put_disk(gd);
4047  out_free:
4048 	kfree(sdkp);
4049  out:
4050 	scsi_autopm_put_device(sdp);
4051 	return error;
4052 }
4053 
4054 /**
4055  *	sd_remove - called whenever a scsi disk (previously recognized by
4056  *	sd_probe) is detached from the system. It is called (potentially
4057  *	multiple times) during sd module unload.
4058  *	@dev: pointer to device object
4059  *
4060  *	Note: this function is invoked from the scsi mid-level.
4061  *	This function potentially frees up a device name (e.g. /dev/sdc)
4062  *	that could be re-used by a subsequent sd_probe().
4063  *	This function is not called when the built-in sd driver is "exit-ed".
4064  **/
4065 static int sd_remove(struct device *dev)
4066 {
4067 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4068 
4069 	scsi_autopm_get_device(sdkp->device);
4070 
4071 	device_del(&sdkp->disk_dev);
4072 	del_gendisk(sdkp->disk);
4073 	if (!sdkp->suspended)
4074 		sd_shutdown(dev);
4075 
4076 	put_disk(sdkp->disk);
4077 	return 0;
4078 }
4079 
4080 static void scsi_disk_release(struct device *dev)
4081 {
4082 	struct scsi_disk *sdkp = to_scsi_disk(dev);
4083 
4084 	ida_free(&sd_index_ida, sdkp->index);
4085 	put_device(&sdkp->device->sdev_gendev);
4086 	free_opal_dev(sdkp->opal_dev);
4087 
4088 	kfree(sdkp);
4089 }
4090 
4091 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
4092 {
4093 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
4094 	struct scsi_sense_hdr sshdr;
4095 	struct scsi_failure failure_defs[] = {
4096 		{
4097 			/* Power on, reset, or bus device reset occurred */
4098 			.sense = UNIT_ATTENTION,
4099 			.asc = 0x29,
4100 			.ascq = 0,
4101 			.result = SAM_STAT_CHECK_CONDITION,
4102 		},
4103 		{
4104 			/* Power on occurred */
4105 			.sense = UNIT_ATTENTION,
4106 			.asc = 0x29,
4107 			.ascq = 1,
4108 			.result = SAM_STAT_CHECK_CONDITION,
4109 		},
4110 		{
4111 			/* SCSI bus reset */
4112 			.sense = UNIT_ATTENTION,
4113 			.asc = 0x29,
4114 			.ascq = 2,
4115 			.result = SAM_STAT_CHECK_CONDITION,
4116 		},
4117 		{}
4118 	};
4119 	struct scsi_failures failures = {
4120 		.total_allowed = 3,
4121 		.failure_definitions = failure_defs,
4122 	};
4123 	const struct scsi_exec_args exec_args = {
4124 		.sshdr = &sshdr,
4125 		.req_flags = BLK_MQ_REQ_PM,
4126 		.failures = &failures,
4127 	};
4128 	struct scsi_device *sdp = sdkp->device;
4129 	int res;
4130 
4131 	if (start)
4132 		cmd[4] |= 1;	/* START */
4133 
4134 	if (sdp->start_stop_pwr_cond)
4135 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
4136 
4137 	if (!scsi_device_online(sdp))
4138 		return -ENODEV;
4139 
4140 	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4141 			       sdkp->max_retries, &exec_args);
4142 	if (res) {
4143 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
4144 		if (res > 0 && scsi_sense_valid(&sshdr)) {
4145 			sd_print_sense_hdr(sdkp, &sshdr);
4146 			/* 0x3a is medium not present */
4147 			if (sshdr.asc == 0x3a)
4148 				res = 0;
4149 		}
4150 	}
4151 
4152 	/* SCSI error codes must not go to the generic layer */
4153 	if (res)
4154 		return -EIO;
4155 
4156 	return 0;
4157 }
4158 
4159 /*
4160  * Send a SYNCHRONIZE CACHE instruction down to the device through
4161  * the normal SCSI command structure.  Wait for the command to
4162  * complete.
4163  */
4164 static void sd_shutdown(struct device *dev)
4165 {
4166 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4167 
4168 	if (!sdkp)
4169 		return;         /* this can happen */
4170 
4171 	if (pm_runtime_suspended(dev))
4172 		return;
4173 
4174 	if (sdkp->WCE && sdkp->media_present) {
4175 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4176 		sd_sync_cache(sdkp);
4177 	}
4178 
4179 	if ((system_state != SYSTEM_RESTART &&
4180 	     sdkp->device->manage_system_start_stop) ||
4181 	    (system_state == SYSTEM_POWER_OFF &&
4182 	     sdkp->device->manage_shutdown)) {
4183 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4184 		sd_start_stop_device(sdkp, 0);
4185 	}
4186 }
4187 
4188 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4189 {
4190 	return (sdev->manage_system_start_stop && !runtime) ||
4191 		(sdev->manage_runtime_start_stop && runtime);
4192 }
4193 
4194 static int sd_suspend_common(struct device *dev, bool runtime)
4195 {
4196 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4197 	int ret = 0;
4198 
4199 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
4200 		return 0;
4201 
4202 	if (sdkp->WCE && sdkp->media_present) {
4203 		if (!sdkp->device->silence_suspend)
4204 			sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4205 		ret = sd_sync_cache(sdkp);
4206 		/* ignore OFFLINE device */
4207 		if (ret == -ENODEV)
4208 			return 0;
4209 
4210 		if (ret)
4211 			return ret;
4212 	}
4213 
4214 	if (sd_do_start_stop(sdkp->device, runtime)) {
4215 		if (!sdkp->device->silence_suspend)
4216 			sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4217 		/* an error is not worth aborting a system sleep */
4218 		ret = sd_start_stop_device(sdkp, 0);
4219 		if (!runtime)
4220 			ret = 0;
4221 	}
4222 
4223 	if (!ret)
4224 		sdkp->suspended = true;
4225 
4226 	return ret;
4227 }
4228 
4229 static int sd_suspend_system(struct device *dev)
4230 {
4231 	if (pm_runtime_suspended(dev))
4232 		return 0;
4233 
4234 	return sd_suspend_common(dev, false);
4235 }
4236 
4237 static int sd_suspend_runtime(struct device *dev)
4238 {
4239 	return sd_suspend_common(dev, true);
4240 }
4241 
4242 static int sd_resume(struct device *dev)
4243 {
4244 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4245 
4246 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4247 
4248 	if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4249 		sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4250 		return -EIO;
4251 	}
4252 
4253 	return 0;
4254 }
4255 
4256 static int sd_resume_common(struct device *dev, bool runtime)
4257 {
4258 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4259 	int ret;
4260 
4261 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
4262 		return 0;
4263 
4264 	if (!sd_do_start_stop(sdkp->device, runtime)) {
4265 		sdkp->suspended = false;
4266 		return 0;
4267 	}
4268 
4269 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4270 	ret = sd_start_stop_device(sdkp, 1);
4271 	if (!ret) {
4272 		sd_resume(dev);
4273 		sdkp->suspended = false;
4274 	}
4275 
4276 	return ret;
4277 }
4278 
4279 static int sd_resume_system(struct device *dev)
4280 {
4281 	if (pm_runtime_suspended(dev)) {
4282 		struct scsi_disk *sdkp = dev_get_drvdata(dev);
4283 		struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4284 
4285 		if (sdp && sdp->force_runtime_start_on_system_start)
4286 			pm_request_resume(dev);
4287 
4288 		return 0;
4289 	}
4290 
4291 	return sd_resume_common(dev, false);
4292 }
4293 
4294 static int sd_resume_runtime(struct device *dev)
4295 {
4296 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4297 	struct scsi_device *sdp;
4298 
4299 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
4300 		return 0;
4301 
4302 	sdp = sdkp->device;
4303 
4304 	if (sdp->ignore_media_change) {
4305 		/* clear the device's sense data */
4306 		static const u8 cmd[10] = { REQUEST_SENSE };
4307 		const struct scsi_exec_args exec_args = {
4308 			.req_flags = BLK_MQ_REQ_PM,
4309 		};
4310 
4311 		if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4312 				     sdp->request_queue->rq_timeout, 1,
4313 				     &exec_args))
4314 			sd_printk(KERN_NOTICE, sdkp,
4315 				  "Failed to clear sense data\n");
4316 	}
4317 
4318 	return sd_resume_common(dev, true);
4319 }
4320 
4321 static const struct dev_pm_ops sd_pm_ops = {
4322 	.suspend		= sd_suspend_system,
4323 	.resume			= sd_resume_system,
4324 	.poweroff		= sd_suspend_system,
4325 	.restore		= sd_resume_system,
4326 	.runtime_suspend	= sd_suspend_runtime,
4327 	.runtime_resume		= sd_resume_runtime,
4328 };
4329 
4330 static struct scsi_driver sd_template = {
4331 	.gendrv = {
4332 		.name		= "sd",
4333 		.probe		= sd_probe,
4334 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
4335 		.remove		= sd_remove,
4336 		.shutdown	= sd_shutdown,
4337 		.pm		= &sd_pm_ops,
4338 	},
4339 	.rescan			= sd_rescan,
4340 	.resume			= sd_resume,
4341 	.init_command		= sd_init_command,
4342 	.uninit_command		= sd_uninit_command,
4343 	.done			= sd_done,
4344 	.eh_action		= sd_eh_action,
4345 	.eh_reset		= sd_eh_reset,
4346 };
4347 
4348 /**
4349  *	init_sd - entry point for this driver (both when built in or when
4350  *	a module).
4351  *
4352  *	Note: this function registers this driver with the scsi mid-level.
4353  **/
4354 static int __init init_sd(void)
4355 {
4356 	int majors = 0, i, err;
4357 
4358 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4359 
4360 	for (i = 0; i < SD_MAJORS; i++) {
4361 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4362 			continue;
4363 		majors++;
4364 	}
4365 
4366 	if (!majors)
4367 		return -ENODEV;
4368 
4369 	err = class_register(&sd_disk_class);
4370 	if (err)
4371 		goto err_out;
4372 
4373 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4374 	if (!sd_page_pool) {
4375 		printk(KERN_ERR "sd: can't init discard page pool\n");
4376 		err = -ENOMEM;
4377 		goto err_out_class;
4378 	}
4379 
4380 	err = scsi_register_driver(&sd_template.gendrv);
4381 	if (err)
4382 		goto err_out_driver;
4383 
4384 	return 0;
4385 
4386 err_out_driver:
4387 	mempool_destroy(sd_page_pool);
4388 err_out_class:
4389 	class_unregister(&sd_disk_class);
4390 err_out:
4391 	for (i = 0; i < SD_MAJORS; i++)
4392 		unregister_blkdev(sd_major(i), "sd");
4393 	return err;
4394 }
4395 
4396 /**
4397  *	exit_sd - exit point for this driver (when it is a module).
4398  *
4399  *	Note: this function unregisters this driver from the scsi mid-level.
4400  **/
4401 static void __exit exit_sd(void)
4402 {
4403 	int i;
4404 
4405 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4406 
4407 	scsi_unregister_driver(&sd_template.gendrv);
4408 	mempool_destroy(sd_page_pool);
4409 
4410 	class_unregister(&sd_disk_class);
4411 
4412 	for (i = 0; i < SD_MAJORS; i++)
4413 		unregister_blkdev(sd_major(i), "sd");
4414 }
4415 
4416 module_init(init_sd);
4417 module_exit(exit_sd);
4418 
4419 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4420 {
4421 	scsi_print_sense_hdr(sdkp->device,
4422 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4423 }
4424 
4425 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4426 {
4427 	const char *hb_string = scsi_hostbyte_string(result);
4428 
4429 	if (hb_string)
4430 		sd_printk(KERN_INFO, sdkp,
4431 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4432 			  hb_string ? hb_string : "invalid",
4433 			  "DRIVER_OK");
4434 	else
4435 		sd_printk(KERN_INFO, sdkp,
4436 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4437 			  msg, host_byte(result), "DRIVER_OK");
4438 }
4439