xref: /openbmc/linux/drivers/scsi/sd.c (revision 568c7c4c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *      sd.c Copyright (C) 1992 Drew Eckhardt
4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5  *
6  *      Linux scsi disk driver
7  *              Initial versions: Drew Eckhardt
8  *              Subsequent revisions: Eric Youngdale
9  *	Modification history:
10  *       - Drew Eckhardt <drew@colorado.edu> original
11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12  *         outstanding request, and other enhancements.
13  *         Support loadable low-level scsi drivers.
14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15  *         eight major numbers.
16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18  *	   sd_init and cleanups.
19  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
20  *	   not being read in sd_open. Fix problem where removable media
21  *	   could be ejected after sd_open.
22  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25  *	   Support 32k/1M disks.
26  *
27  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
28  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
32  *	Note: when the logging level is set by the user, it must be greater
33  *	than the level indicated above to trigger output.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/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/major.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
56 #include <linux/pr.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
60 
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
70 #include <scsi/scsi_common.h>
71 
72 #include "sd.h"
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
75 
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
79 
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100 
101 #define SD_MINORS	16
102 
103 static void sd_config_discard(struct scsi_disk *, unsigned int);
104 static void sd_config_write_same(struct scsi_disk *);
105 static int  sd_revalidate_disk(struct gendisk *);
106 static void sd_unlock_native_capacity(struct gendisk *disk);
107 static void sd_shutdown(struct device *);
108 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
109 static void scsi_disk_release(struct device *cdev);
110 
111 static DEFINE_IDA(sd_index_ida);
112 
113 static mempool_t *sd_page_pool;
114 static struct lock_class_key sd_bio_compl_lkclass;
115 
116 static const char *sd_cache_types[] = {
117 	"write through", "none", "write back",
118 	"write back, no read (daft)"
119 };
120 
121 static void sd_set_flush_flag(struct scsi_disk *sdkp)
122 {
123 	bool wc = false, fua = false;
124 
125 	if (sdkp->WCE) {
126 		wc = true;
127 		if (sdkp->DPOFUA)
128 			fua = true;
129 	}
130 
131 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
132 }
133 
134 static ssize_t
135 cache_type_store(struct device *dev, struct device_attribute *attr,
136 		 const char *buf, size_t count)
137 {
138 	int ct, rcd, wce, sp;
139 	struct scsi_disk *sdkp = to_scsi_disk(dev);
140 	struct scsi_device *sdp = sdkp->device;
141 	char buffer[64];
142 	char *buffer_data;
143 	struct scsi_mode_data data;
144 	struct scsi_sense_hdr sshdr;
145 	static const char temp[] = "temporary ";
146 	int len;
147 
148 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
149 		/* no cache control on RBC devices; theoretically they
150 		 * can do it, but there's probably so many exceptions
151 		 * it's not worth the risk */
152 		return -EINVAL;
153 
154 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
155 		buf += sizeof(temp) - 1;
156 		sdkp->cache_override = 1;
157 	} else {
158 		sdkp->cache_override = 0;
159 	}
160 
161 	ct = sysfs_match_string(sd_cache_types, buf);
162 	if (ct < 0)
163 		return -EINVAL;
164 
165 	rcd = ct & 0x01 ? 1 : 0;
166 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
167 
168 	if (sdkp->cache_override) {
169 		sdkp->WCE = wce;
170 		sdkp->RCD = rcd;
171 		sd_set_flush_flag(sdkp);
172 		return count;
173 	}
174 
175 	if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
176 			    sdkp->max_retries, &data, NULL))
177 		return -EINVAL;
178 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
179 		  data.block_descriptor_length);
180 	buffer_data = buffer + data.header_length +
181 		data.block_descriptor_length;
182 	buffer_data[2] &= ~0x05;
183 	buffer_data[2] |= wce << 2 | rcd;
184 	sp = buffer_data[0] & 0x80 ? 1 : 0;
185 	buffer_data[0] &= ~0x80;
186 
187 	/*
188 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
189 	 * received mode parameter buffer before doing MODE SELECT.
190 	 */
191 	data.device_specific = 0;
192 
193 	if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
194 			     sdkp->max_retries, &data, &sshdr)) {
195 		if (scsi_sense_valid(&sshdr))
196 			sd_print_sense_hdr(sdkp, &sshdr);
197 		return -EINVAL;
198 	}
199 	sd_revalidate_disk(sdkp->disk);
200 	return count;
201 }
202 
203 static ssize_t
204 manage_start_stop_show(struct device *dev,
205 		       struct device_attribute *attr, char *buf)
206 {
207 	struct scsi_disk *sdkp = to_scsi_disk(dev);
208 	struct scsi_device *sdp = sdkp->device;
209 
210 	return sysfs_emit(buf, "%u\n",
211 			  sdp->manage_system_start_stop &&
212 			  sdp->manage_runtime_start_stop &&
213 			  sdp->manage_shutdown);
214 }
215 static DEVICE_ATTR_RO(manage_start_stop);
216 
217 static ssize_t
218 manage_system_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", sdp->manage_system_start_stop);
225 }
226 
227 static ssize_t
228 manage_system_start_stop_store(struct device *dev,
229 			       struct device_attribute *attr,
230 			       const char *buf, size_t count)
231 {
232 	struct scsi_disk *sdkp = to_scsi_disk(dev);
233 	struct scsi_device *sdp = sdkp->device;
234 	bool v;
235 
236 	if (!capable(CAP_SYS_ADMIN))
237 		return -EACCES;
238 
239 	if (kstrtobool(buf, &v))
240 		return -EINVAL;
241 
242 	sdp->manage_system_start_stop = v;
243 
244 	return count;
245 }
246 static DEVICE_ATTR_RW(manage_system_start_stop);
247 
248 static ssize_t
249 manage_runtime_start_stop_show(struct device *dev,
250 			       struct device_attribute *attr, char *buf)
251 {
252 	struct scsi_disk *sdkp = to_scsi_disk(dev);
253 	struct scsi_device *sdp = sdkp->device;
254 
255 	return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
256 }
257 
258 static ssize_t
259 manage_runtime_start_stop_store(struct device *dev,
260 				struct device_attribute *attr,
261 				const char *buf, size_t count)
262 {
263 	struct scsi_disk *sdkp = to_scsi_disk(dev);
264 	struct scsi_device *sdp = sdkp->device;
265 	bool v;
266 
267 	if (!capable(CAP_SYS_ADMIN))
268 		return -EACCES;
269 
270 	if (kstrtobool(buf, &v))
271 		return -EINVAL;
272 
273 	sdp->manage_runtime_start_stop = v;
274 
275 	return count;
276 }
277 static DEVICE_ATTR_RW(manage_runtime_start_stop);
278 
279 static ssize_t manage_shutdown_show(struct device *dev,
280 				    struct device_attribute *attr, char *buf)
281 {
282 	struct scsi_disk *sdkp = to_scsi_disk(dev);
283 	struct scsi_device *sdp = sdkp->device;
284 
285 	return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
286 }
287 
288 static ssize_t manage_shutdown_store(struct device *dev,
289 				     struct device_attribute *attr,
290 				     const char *buf, size_t count)
291 {
292 	struct scsi_disk *sdkp = to_scsi_disk(dev);
293 	struct scsi_device *sdp = sdkp->device;
294 	bool v;
295 
296 	if (!capable(CAP_SYS_ADMIN))
297 		return -EACCES;
298 
299 	if (kstrtobool(buf, &v))
300 		return -EINVAL;
301 
302 	sdp->manage_shutdown = v;
303 
304 	return count;
305 }
306 static DEVICE_ATTR_RW(manage_shutdown);
307 
308 static ssize_t
309 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
310 {
311 	struct scsi_disk *sdkp = to_scsi_disk(dev);
312 
313 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
314 }
315 
316 static ssize_t
317 allow_restart_store(struct device *dev, struct device_attribute *attr,
318 		    const char *buf, size_t count)
319 {
320 	bool v;
321 	struct scsi_disk *sdkp = to_scsi_disk(dev);
322 	struct scsi_device *sdp = sdkp->device;
323 
324 	if (!capable(CAP_SYS_ADMIN))
325 		return -EACCES;
326 
327 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
328 		return -EINVAL;
329 
330 	if (kstrtobool(buf, &v))
331 		return -EINVAL;
332 
333 	sdp->allow_restart = v;
334 
335 	return count;
336 }
337 static DEVICE_ATTR_RW(allow_restart);
338 
339 static ssize_t
340 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
341 {
342 	struct scsi_disk *sdkp = to_scsi_disk(dev);
343 	int ct = sdkp->RCD + 2*sdkp->WCE;
344 
345 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
346 }
347 static DEVICE_ATTR_RW(cache_type);
348 
349 static ssize_t
350 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
351 {
352 	struct scsi_disk *sdkp = to_scsi_disk(dev);
353 
354 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
355 }
356 static DEVICE_ATTR_RO(FUA);
357 
358 static ssize_t
359 protection_type_show(struct device *dev, struct device_attribute *attr,
360 		     char *buf)
361 {
362 	struct scsi_disk *sdkp = to_scsi_disk(dev);
363 
364 	return sprintf(buf, "%u\n", sdkp->protection_type);
365 }
366 
367 static ssize_t
368 protection_type_store(struct device *dev, struct device_attribute *attr,
369 		      const char *buf, size_t count)
370 {
371 	struct scsi_disk *sdkp = to_scsi_disk(dev);
372 	unsigned int val;
373 	int err;
374 
375 	if (!capable(CAP_SYS_ADMIN))
376 		return -EACCES;
377 
378 	err = kstrtouint(buf, 10, &val);
379 
380 	if (err)
381 		return err;
382 
383 	if (val <= T10_PI_TYPE3_PROTECTION)
384 		sdkp->protection_type = val;
385 
386 	return count;
387 }
388 static DEVICE_ATTR_RW(protection_type);
389 
390 static ssize_t
391 protection_mode_show(struct device *dev, struct device_attribute *attr,
392 		     char *buf)
393 {
394 	struct scsi_disk *sdkp = to_scsi_disk(dev);
395 	struct scsi_device *sdp = sdkp->device;
396 	unsigned int dif, dix;
397 
398 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
399 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
400 
401 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
402 		dif = 0;
403 		dix = 1;
404 	}
405 
406 	if (!dif && !dix)
407 		return sprintf(buf, "none\n");
408 
409 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
410 }
411 static DEVICE_ATTR_RO(protection_mode);
412 
413 static ssize_t
414 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
415 {
416 	struct scsi_disk *sdkp = to_scsi_disk(dev);
417 
418 	return sprintf(buf, "%u\n", sdkp->ATO);
419 }
420 static DEVICE_ATTR_RO(app_tag_own);
421 
422 static ssize_t
423 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
424 		       char *buf)
425 {
426 	struct scsi_disk *sdkp = to_scsi_disk(dev);
427 
428 	return sprintf(buf, "%u\n", sdkp->lbpme);
429 }
430 static DEVICE_ATTR_RO(thin_provisioning);
431 
432 /* sysfs_match_string() requires dense arrays */
433 static const char *lbp_mode[] = {
434 	[SD_LBP_FULL]		= "full",
435 	[SD_LBP_UNMAP]		= "unmap",
436 	[SD_LBP_WS16]		= "writesame_16",
437 	[SD_LBP_WS10]		= "writesame_10",
438 	[SD_LBP_ZERO]		= "writesame_zero",
439 	[SD_LBP_DISABLE]	= "disabled",
440 };
441 
442 static ssize_t
443 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
444 		       char *buf)
445 {
446 	struct scsi_disk *sdkp = to_scsi_disk(dev);
447 
448 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
449 }
450 
451 static ssize_t
452 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
453 			const char *buf, size_t count)
454 {
455 	struct scsi_disk *sdkp = to_scsi_disk(dev);
456 	struct scsi_device *sdp = sdkp->device;
457 	int mode;
458 
459 	if (!capable(CAP_SYS_ADMIN))
460 		return -EACCES;
461 
462 	if (sd_is_zoned(sdkp)) {
463 		sd_config_discard(sdkp, SD_LBP_DISABLE);
464 		return count;
465 	}
466 
467 	if (sdp->type != TYPE_DISK)
468 		return -EINVAL;
469 
470 	mode = sysfs_match_string(lbp_mode, buf);
471 	if (mode < 0)
472 		return -EINVAL;
473 
474 	sd_config_discard(sdkp, mode);
475 
476 	return count;
477 }
478 static DEVICE_ATTR_RW(provisioning_mode);
479 
480 /* sysfs_match_string() requires dense arrays */
481 static const char *zeroing_mode[] = {
482 	[SD_ZERO_WRITE]		= "write",
483 	[SD_ZERO_WS]		= "writesame",
484 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
485 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
486 };
487 
488 static ssize_t
489 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
490 		  char *buf)
491 {
492 	struct scsi_disk *sdkp = to_scsi_disk(dev);
493 
494 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
495 }
496 
497 static ssize_t
498 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
499 		   const char *buf, size_t count)
500 {
501 	struct scsi_disk *sdkp = to_scsi_disk(dev);
502 	int mode;
503 
504 	if (!capable(CAP_SYS_ADMIN))
505 		return -EACCES;
506 
507 	mode = sysfs_match_string(zeroing_mode, buf);
508 	if (mode < 0)
509 		return -EINVAL;
510 
511 	sdkp->zeroing_mode = mode;
512 
513 	return count;
514 }
515 static DEVICE_ATTR_RW(zeroing_mode);
516 
517 static ssize_t
518 max_medium_access_timeouts_show(struct device *dev,
519 				struct device_attribute *attr, char *buf)
520 {
521 	struct scsi_disk *sdkp = to_scsi_disk(dev);
522 
523 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
524 }
525 
526 static ssize_t
527 max_medium_access_timeouts_store(struct device *dev,
528 				 struct device_attribute *attr, const char *buf,
529 				 size_t count)
530 {
531 	struct scsi_disk *sdkp = to_scsi_disk(dev);
532 	int err;
533 
534 	if (!capable(CAP_SYS_ADMIN))
535 		return -EACCES;
536 
537 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
538 
539 	return err ? err : count;
540 }
541 static DEVICE_ATTR_RW(max_medium_access_timeouts);
542 
543 static ssize_t
544 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
545 			   char *buf)
546 {
547 	struct scsi_disk *sdkp = to_scsi_disk(dev);
548 
549 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
550 }
551 
552 static ssize_t
553 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
554 			    const char *buf, size_t count)
555 {
556 	struct scsi_disk *sdkp = to_scsi_disk(dev);
557 	struct scsi_device *sdp = sdkp->device;
558 	unsigned long max;
559 	int err;
560 
561 	if (!capable(CAP_SYS_ADMIN))
562 		return -EACCES;
563 
564 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
565 		return -EINVAL;
566 
567 	err = kstrtoul(buf, 10, &max);
568 
569 	if (err)
570 		return err;
571 
572 	if (max == 0)
573 		sdp->no_write_same = 1;
574 	else if (max <= SD_MAX_WS16_BLOCKS) {
575 		sdp->no_write_same = 0;
576 		sdkp->max_ws_blocks = max;
577 	}
578 
579 	sd_config_write_same(sdkp);
580 
581 	return count;
582 }
583 static DEVICE_ATTR_RW(max_write_same_blocks);
584 
585 static ssize_t
586 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
587 {
588 	struct scsi_disk *sdkp = to_scsi_disk(dev);
589 
590 	if (sdkp->device->type == TYPE_ZBC)
591 		return sprintf(buf, "host-managed\n");
592 	if (sdkp->zoned == 1)
593 		return sprintf(buf, "host-aware\n");
594 	if (sdkp->zoned == 2)
595 		return sprintf(buf, "drive-managed\n");
596 	return sprintf(buf, "none\n");
597 }
598 static DEVICE_ATTR_RO(zoned_cap);
599 
600 static ssize_t
601 max_retries_store(struct device *dev, struct device_attribute *attr,
602 		  const char *buf, size_t count)
603 {
604 	struct scsi_disk *sdkp = to_scsi_disk(dev);
605 	struct scsi_device *sdev = sdkp->device;
606 	int retries, err;
607 
608 	err = kstrtoint(buf, 10, &retries);
609 	if (err)
610 		return err;
611 
612 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
613 		sdkp->max_retries = retries;
614 		return count;
615 	}
616 
617 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
618 		    SD_MAX_RETRIES);
619 	return -EINVAL;
620 }
621 
622 static ssize_t
623 max_retries_show(struct device *dev, struct device_attribute *attr,
624 		 char *buf)
625 {
626 	struct scsi_disk *sdkp = to_scsi_disk(dev);
627 
628 	return sprintf(buf, "%d\n", sdkp->max_retries);
629 }
630 
631 static DEVICE_ATTR_RW(max_retries);
632 
633 static struct attribute *sd_disk_attrs[] = {
634 	&dev_attr_cache_type.attr,
635 	&dev_attr_FUA.attr,
636 	&dev_attr_allow_restart.attr,
637 	&dev_attr_manage_start_stop.attr,
638 	&dev_attr_manage_system_start_stop.attr,
639 	&dev_attr_manage_runtime_start_stop.attr,
640 	&dev_attr_manage_shutdown.attr,
641 	&dev_attr_protection_type.attr,
642 	&dev_attr_protection_mode.attr,
643 	&dev_attr_app_tag_own.attr,
644 	&dev_attr_thin_provisioning.attr,
645 	&dev_attr_provisioning_mode.attr,
646 	&dev_attr_zeroing_mode.attr,
647 	&dev_attr_max_write_same_blocks.attr,
648 	&dev_attr_max_medium_access_timeouts.attr,
649 	&dev_attr_zoned_cap.attr,
650 	&dev_attr_max_retries.attr,
651 	NULL,
652 };
653 ATTRIBUTE_GROUPS(sd_disk);
654 
655 static struct class sd_disk_class = {
656 	.name		= "scsi_disk",
657 	.dev_release	= scsi_disk_release,
658 	.dev_groups	= sd_disk_groups,
659 };
660 
661 /*
662  * Don't request a new module, as that could deadlock in multipath
663  * environment.
664  */
665 static void sd_default_probe(dev_t devt)
666 {
667 }
668 
669 /*
670  * Device no to disk mapping:
671  *
672  *       major         disc2     disc  p1
673  *   |............|.............|....|....| <- dev_t
674  *    31        20 19          8 7  4 3  0
675  *
676  * Inside a major, we have 16k disks, however mapped non-
677  * contiguously. The first 16 disks are for major0, the next
678  * ones with major1, ... Disk 256 is for major0 again, disk 272
679  * for major1, ...
680  * As we stay compatible with our numbering scheme, we can reuse
681  * the well-know SCSI majors 8, 65--71, 136--143.
682  */
683 static int sd_major(int major_idx)
684 {
685 	switch (major_idx) {
686 	case 0:
687 		return SCSI_DISK0_MAJOR;
688 	case 1 ... 7:
689 		return SCSI_DISK1_MAJOR + major_idx - 1;
690 	case 8 ... 15:
691 		return SCSI_DISK8_MAJOR + major_idx - 8;
692 	default:
693 		BUG();
694 		return 0;	/* shut up gcc */
695 	}
696 }
697 
698 #ifdef CONFIG_BLK_SED_OPAL
699 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
700 		size_t len, bool send)
701 {
702 	struct scsi_disk *sdkp = data;
703 	struct scsi_device *sdev = sdkp->device;
704 	u8 cdb[12] = { 0, };
705 	const struct scsi_exec_args exec_args = {
706 		.req_flags = BLK_MQ_REQ_PM,
707 	};
708 	int ret;
709 
710 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
711 	cdb[1] = secp;
712 	put_unaligned_be16(spsp, &cdb[2]);
713 	put_unaligned_be32(len, &cdb[6]);
714 
715 	ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
716 			       buffer, len, SD_TIMEOUT, sdkp->max_retries,
717 			       &exec_args);
718 	return ret <= 0 ? ret : -EIO;
719 }
720 #endif /* CONFIG_BLK_SED_OPAL */
721 
722 /*
723  * Look up the DIX operation based on whether the command is read or
724  * write and whether dix and dif are enabled.
725  */
726 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
727 {
728 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
729 	static const unsigned int ops[] = {	/* wrt dix dif */
730 		SCSI_PROT_NORMAL,		/*  0	0   0  */
731 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
732 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
733 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
734 		SCSI_PROT_NORMAL,		/*  1	0   0  */
735 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
736 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
737 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
738 	};
739 
740 	return ops[write << 2 | dix << 1 | dif];
741 }
742 
743 /*
744  * Returns a mask of the protection flags that are valid for a given DIX
745  * operation.
746  */
747 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
748 {
749 	static const unsigned int flag_mask[] = {
750 		[SCSI_PROT_NORMAL]		= 0,
751 
752 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
753 						  SCSI_PROT_GUARD_CHECK |
754 						  SCSI_PROT_REF_CHECK |
755 						  SCSI_PROT_REF_INCREMENT,
756 
757 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
758 						  SCSI_PROT_IP_CHECKSUM,
759 
760 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
761 						  SCSI_PROT_GUARD_CHECK |
762 						  SCSI_PROT_REF_CHECK |
763 						  SCSI_PROT_REF_INCREMENT |
764 						  SCSI_PROT_IP_CHECKSUM,
765 
766 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
767 						  SCSI_PROT_REF_INCREMENT,
768 
769 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
770 						  SCSI_PROT_REF_CHECK |
771 						  SCSI_PROT_REF_INCREMENT |
772 						  SCSI_PROT_IP_CHECKSUM,
773 
774 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
775 						  SCSI_PROT_GUARD_CHECK |
776 						  SCSI_PROT_REF_CHECK |
777 						  SCSI_PROT_REF_INCREMENT |
778 						  SCSI_PROT_IP_CHECKSUM,
779 	};
780 
781 	return flag_mask[prot_op];
782 }
783 
784 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
785 					   unsigned int dix, unsigned int dif)
786 {
787 	struct request *rq = scsi_cmd_to_rq(scmd);
788 	struct bio *bio = rq->bio;
789 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
790 	unsigned int protect = 0;
791 
792 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
793 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
794 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
795 
796 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
797 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
798 	}
799 
800 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
801 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
802 
803 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
804 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
805 	}
806 
807 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
808 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
809 
810 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
811 			protect = 3 << 5;	/* Disable target PI checking */
812 		else
813 			protect = 1 << 5;	/* Enable target PI checking */
814 	}
815 
816 	scsi_set_prot_op(scmd, prot_op);
817 	scsi_set_prot_type(scmd, dif);
818 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
819 
820 	return protect;
821 }
822 
823 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
824 {
825 	struct request_queue *q = sdkp->disk->queue;
826 	unsigned int logical_block_size = sdkp->device->sector_size;
827 	unsigned int max_blocks = 0;
828 
829 	q->limits.discard_alignment =
830 		sdkp->unmap_alignment * logical_block_size;
831 	q->limits.discard_granularity =
832 		max(sdkp->physical_block_size,
833 		    sdkp->unmap_granularity * logical_block_size);
834 	sdkp->provisioning_mode = mode;
835 
836 	switch (mode) {
837 
838 	case SD_LBP_FULL:
839 	case SD_LBP_DISABLE:
840 		blk_queue_max_discard_sectors(q, 0);
841 		return;
842 
843 	case SD_LBP_UNMAP:
844 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
845 					  (u32)SD_MAX_WS16_BLOCKS);
846 		break;
847 
848 	case SD_LBP_WS16:
849 		if (sdkp->device->unmap_limit_for_ws)
850 			max_blocks = sdkp->max_unmap_blocks;
851 		else
852 			max_blocks = sdkp->max_ws_blocks;
853 
854 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
855 		break;
856 
857 	case SD_LBP_WS10:
858 		if (sdkp->device->unmap_limit_for_ws)
859 			max_blocks = sdkp->max_unmap_blocks;
860 		else
861 			max_blocks = sdkp->max_ws_blocks;
862 
863 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
864 		break;
865 
866 	case SD_LBP_ZERO:
867 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
868 					  (u32)SD_MAX_WS10_BLOCKS);
869 		break;
870 	}
871 
872 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
873 }
874 
875 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
876 {
877 	struct page *page;
878 
879 	page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
880 	if (!page)
881 		return NULL;
882 	clear_highpage(page);
883 	bvec_set_page(&rq->special_vec, page, data_len, 0);
884 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
885 	return bvec_virt(&rq->special_vec);
886 }
887 
888 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
889 {
890 	struct scsi_device *sdp = cmd->device;
891 	struct request *rq = scsi_cmd_to_rq(cmd);
892 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
893 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
894 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
895 	unsigned int data_len = 24;
896 	char *buf;
897 
898 	buf = sd_set_special_bvec(rq, data_len);
899 	if (!buf)
900 		return BLK_STS_RESOURCE;
901 
902 	cmd->cmd_len = 10;
903 	cmd->cmnd[0] = UNMAP;
904 	cmd->cmnd[8] = 24;
905 
906 	put_unaligned_be16(6 + 16, &buf[0]);
907 	put_unaligned_be16(16, &buf[2]);
908 	put_unaligned_be64(lba, &buf[8]);
909 	put_unaligned_be32(nr_blocks, &buf[16]);
910 
911 	cmd->allowed = sdkp->max_retries;
912 	cmd->transfersize = data_len;
913 	rq->timeout = SD_TIMEOUT;
914 
915 	return scsi_alloc_sgtables(cmd);
916 }
917 
918 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
919 		bool unmap)
920 {
921 	struct scsi_device *sdp = cmd->device;
922 	struct request *rq = scsi_cmd_to_rq(cmd);
923 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
924 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
925 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
926 	u32 data_len = sdp->sector_size;
927 
928 	if (!sd_set_special_bvec(rq, data_len))
929 		return BLK_STS_RESOURCE;
930 
931 	cmd->cmd_len = 16;
932 	cmd->cmnd[0] = WRITE_SAME_16;
933 	if (unmap)
934 		cmd->cmnd[1] = 0x8; /* UNMAP */
935 	put_unaligned_be64(lba, &cmd->cmnd[2]);
936 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
937 
938 	cmd->allowed = sdkp->max_retries;
939 	cmd->transfersize = data_len;
940 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
941 
942 	return scsi_alloc_sgtables(cmd);
943 }
944 
945 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
946 		bool unmap)
947 {
948 	struct scsi_device *sdp = cmd->device;
949 	struct request *rq = scsi_cmd_to_rq(cmd);
950 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
951 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
952 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
953 	u32 data_len = sdp->sector_size;
954 
955 	if (!sd_set_special_bvec(rq, data_len))
956 		return BLK_STS_RESOURCE;
957 
958 	cmd->cmd_len = 10;
959 	cmd->cmnd[0] = WRITE_SAME;
960 	if (unmap)
961 		cmd->cmnd[1] = 0x8; /* UNMAP */
962 	put_unaligned_be32(lba, &cmd->cmnd[2]);
963 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
964 
965 	cmd->allowed = sdkp->max_retries;
966 	cmd->transfersize = data_len;
967 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
968 
969 	return scsi_alloc_sgtables(cmd);
970 }
971 
972 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
973 {
974 	struct request *rq = scsi_cmd_to_rq(cmd);
975 	struct scsi_device *sdp = cmd->device;
976 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
977 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
978 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
979 
980 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
981 		switch (sdkp->zeroing_mode) {
982 		case SD_ZERO_WS16_UNMAP:
983 			return sd_setup_write_same16_cmnd(cmd, true);
984 		case SD_ZERO_WS10_UNMAP:
985 			return sd_setup_write_same10_cmnd(cmd, true);
986 		}
987 	}
988 
989 	if (sdp->no_write_same) {
990 		rq->rq_flags |= RQF_QUIET;
991 		return BLK_STS_TARGET;
992 	}
993 
994 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
995 		return sd_setup_write_same16_cmnd(cmd, false);
996 
997 	return sd_setup_write_same10_cmnd(cmd, false);
998 }
999 
1000 static void sd_config_write_same(struct scsi_disk *sdkp)
1001 {
1002 	struct request_queue *q = sdkp->disk->queue;
1003 	unsigned int logical_block_size = sdkp->device->sector_size;
1004 
1005 	if (sdkp->device->no_write_same) {
1006 		sdkp->max_ws_blocks = 0;
1007 		goto out;
1008 	}
1009 
1010 	/* Some devices can not handle block counts above 0xffff despite
1011 	 * supporting WRITE SAME(16). Consequently we default to 64k
1012 	 * blocks per I/O unless the device explicitly advertises a
1013 	 * bigger limit.
1014 	 */
1015 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1016 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1017 						   (u32)SD_MAX_WS16_BLOCKS);
1018 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1019 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1020 						   (u32)SD_MAX_WS10_BLOCKS);
1021 	else {
1022 		sdkp->device->no_write_same = 1;
1023 		sdkp->max_ws_blocks = 0;
1024 	}
1025 
1026 	if (sdkp->lbprz && sdkp->lbpws)
1027 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1028 	else if (sdkp->lbprz && sdkp->lbpws10)
1029 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1030 	else if (sdkp->max_ws_blocks)
1031 		sdkp->zeroing_mode = SD_ZERO_WS;
1032 	else
1033 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1034 
1035 	if (sdkp->max_ws_blocks &&
1036 	    sdkp->physical_block_size > logical_block_size) {
1037 		/*
1038 		 * Reporting a maximum number of blocks that is not aligned
1039 		 * on the device physical size would cause a large write same
1040 		 * request to be split into physically unaligned chunks by
1041 		 * __blkdev_issue_write_zeroes() even if the caller of this
1042 		 * functions took care to align the large request. So make sure
1043 		 * the maximum reported is aligned to the device physical block
1044 		 * size. This is only an optional optimization for regular
1045 		 * disks, but this is mandatory to avoid failure of large write
1046 		 * same requests directed at sequential write required zones of
1047 		 * host-managed ZBC disks.
1048 		 */
1049 		sdkp->max_ws_blocks =
1050 			round_down(sdkp->max_ws_blocks,
1051 				   bytes_to_logical(sdkp->device,
1052 						    sdkp->physical_block_size));
1053 	}
1054 
1055 out:
1056 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1057 					 (logical_block_size >> 9));
1058 }
1059 
1060 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1061 {
1062 	struct request *rq = scsi_cmd_to_rq(cmd);
1063 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1064 
1065 	/* flush requests don't perform I/O, zero the S/G table */
1066 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1067 
1068 	if (cmd->device->use_16_for_sync) {
1069 		cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1070 		cmd->cmd_len = 16;
1071 	} else {
1072 		cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1073 		cmd->cmd_len = 10;
1074 	}
1075 	cmd->transfersize = 0;
1076 	cmd->allowed = sdkp->max_retries;
1077 
1078 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1079 	return BLK_STS_OK;
1080 }
1081 
1082 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1083 				       sector_t lba, unsigned int nr_blocks,
1084 				       unsigned char flags, unsigned int dld)
1085 {
1086 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1087 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1088 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1089 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1090 	cmd->cmnd[10] = flags;
1091 	cmd->cmnd[11] = dld & 0x07;
1092 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1093 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1094 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1095 
1096 	return BLK_STS_OK;
1097 }
1098 
1099 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1100 				       sector_t lba, unsigned int nr_blocks,
1101 				       unsigned char flags, unsigned int dld)
1102 {
1103 	cmd->cmd_len  = 16;
1104 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1105 	cmd->cmnd[1]  = flags | ((dld >> 2) & 0x01);
1106 	cmd->cmnd[14] = (dld & 0x03) << 6;
1107 	cmd->cmnd[15] = 0;
1108 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1109 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1110 
1111 	return BLK_STS_OK;
1112 }
1113 
1114 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1115 				       sector_t lba, unsigned int nr_blocks,
1116 				       unsigned char flags)
1117 {
1118 	cmd->cmd_len = 10;
1119 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1120 	cmd->cmnd[1] = flags;
1121 	cmd->cmnd[6] = 0;
1122 	cmd->cmnd[9] = 0;
1123 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1124 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1125 
1126 	return BLK_STS_OK;
1127 }
1128 
1129 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1130 				      sector_t lba, unsigned int nr_blocks,
1131 				      unsigned char flags)
1132 {
1133 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1134 	if (WARN_ON_ONCE(nr_blocks == 0))
1135 		return BLK_STS_IOERR;
1136 
1137 	if (unlikely(flags & 0x8)) {
1138 		/*
1139 		 * This happens only if this drive failed 10byte rw
1140 		 * command with ILLEGAL_REQUEST during operation and
1141 		 * thus turned off use_10_for_rw.
1142 		 */
1143 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1144 		return BLK_STS_IOERR;
1145 	}
1146 
1147 	cmd->cmd_len = 6;
1148 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1149 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1150 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1151 	cmd->cmnd[3] = lba & 0xff;
1152 	cmd->cmnd[4] = nr_blocks;
1153 	cmd->cmnd[5] = 0;
1154 
1155 	return BLK_STS_OK;
1156 }
1157 
1158 /*
1159  * Check if a command has a duration limit set. If it does, and the target
1160  * device supports CDL and the feature is enabled, return the limit
1161  * descriptor index to use. Return 0 (no limit) otherwise.
1162  */
1163 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1164 {
1165 	struct scsi_device *sdp = sdkp->device;
1166 	int hint;
1167 
1168 	if (!sdp->cdl_supported || !sdp->cdl_enable)
1169 		return 0;
1170 
1171 	/*
1172 	 * Use "no limit" if the request ioprio does not specify a duration
1173 	 * limit hint.
1174 	 */
1175 	hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1176 	if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1177 	    hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1178 		return 0;
1179 
1180 	return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1181 }
1182 
1183 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1184 {
1185 	struct request *rq = scsi_cmd_to_rq(cmd);
1186 	struct scsi_device *sdp = cmd->device;
1187 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1188 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1189 	sector_t threshold;
1190 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1191 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1192 	bool write = rq_data_dir(rq) == WRITE;
1193 	unsigned char protect, fua;
1194 	unsigned int dld;
1195 	blk_status_t ret;
1196 	unsigned int dif;
1197 	bool dix;
1198 
1199 	ret = scsi_alloc_sgtables(cmd);
1200 	if (ret != BLK_STS_OK)
1201 		return ret;
1202 
1203 	ret = BLK_STS_IOERR;
1204 	if (!scsi_device_online(sdp) || sdp->changed) {
1205 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1206 		goto fail;
1207 	}
1208 
1209 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1210 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1211 		goto fail;
1212 	}
1213 
1214 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1215 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1216 		goto fail;
1217 	}
1218 
1219 	/*
1220 	 * Some SD card readers can't handle accesses which touch the
1221 	 * last one or two logical blocks. Split accesses as needed.
1222 	 */
1223 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1224 
1225 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1226 		if (lba < threshold) {
1227 			/* Access up to the threshold but not beyond */
1228 			nr_blocks = threshold - lba;
1229 		} else {
1230 			/* Access only a single logical block */
1231 			nr_blocks = 1;
1232 		}
1233 	}
1234 
1235 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1236 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1237 		if (ret)
1238 			goto fail;
1239 	}
1240 
1241 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1242 	dix = scsi_prot_sg_count(cmd);
1243 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1244 	dld = sd_cdl_dld(sdkp, cmd);
1245 
1246 	if (dif || dix)
1247 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1248 	else
1249 		protect = 0;
1250 
1251 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1252 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1253 					 protect | fua, dld);
1254 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1255 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1256 					 protect | fua, dld);
1257 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1258 		   sdp->use_10_for_rw || protect) {
1259 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1260 					 protect | fua);
1261 	} else {
1262 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1263 					protect | fua);
1264 	}
1265 
1266 	if (unlikely(ret != BLK_STS_OK))
1267 		goto fail;
1268 
1269 	/*
1270 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1271 	 * host adapter, it's safe to assume that we can at least transfer
1272 	 * this many bytes between each connect / disconnect.
1273 	 */
1274 	cmd->transfersize = sdp->sector_size;
1275 	cmd->underflow = nr_blocks << 9;
1276 	cmd->allowed = sdkp->max_retries;
1277 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1278 
1279 	SCSI_LOG_HLQUEUE(1,
1280 			 scmd_printk(KERN_INFO, cmd,
1281 				     "%s: block=%llu, count=%d\n", __func__,
1282 				     (unsigned long long)blk_rq_pos(rq),
1283 				     blk_rq_sectors(rq)));
1284 	SCSI_LOG_HLQUEUE(2,
1285 			 scmd_printk(KERN_INFO, cmd,
1286 				     "%s %d/%u 512 byte blocks.\n",
1287 				     write ? "writing" : "reading", nr_blocks,
1288 				     blk_rq_sectors(rq)));
1289 
1290 	/*
1291 	 * This indicates that the command is ready from our end to be queued.
1292 	 */
1293 	return BLK_STS_OK;
1294 fail:
1295 	scsi_free_sgtables(cmd);
1296 	return ret;
1297 }
1298 
1299 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1300 {
1301 	struct request *rq = scsi_cmd_to_rq(cmd);
1302 
1303 	switch (req_op(rq)) {
1304 	case REQ_OP_DISCARD:
1305 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1306 		case SD_LBP_UNMAP:
1307 			return sd_setup_unmap_cmnd(cmd);
1308 		case SD_LBP_WS16:
1309 			return sd_setup_write_same16_cmnd(cmd, true);
1310 		case SD_LBP_WS10:
1311 			return sd_setup_write_same10_cmnd(cmd, true);
1312 		case SD_LBP_ZERO:
1313 			return sd_setup_write_same10_cmnd(cmd, false);
1314 		default:
1315 			return BLK_STS_TARGET;
1316 		}
1317 	case REQ_OP_WRITE_ZEROES:
1318 		return sd_setup_write_zeroes_cmnd(cmd);
1319 	case REQ_OP_FLUSH:
1320 		return sd_setup_flush_cmnd(cmd);
1321 	case REQ_OP_READ:
1322 	case REQ_OP_WRITE:
1323 	case REQ_OP_ZONE_APPEND:
1324 		return sd_setup_read_write_cmnd(cmd);
1325 	case REQ_OP_ZONE_RESET:
1326 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1327 						   false);
1328 	case REQ_OP_ZONE_RESET_ALL:
1329 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1330 						   true);
1331 	case REQ_OP_ZONE_OPEN:
1332 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1333 	case REQ_OP_ZONE_CLOSE:
1334 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1335 	case REQ_OP_ZONE_FINISH:
1336 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1337 	default:
1338 		WARN_ON_ONCE(1);
1339 		return BLK_STS_NOTSUPP;
1340 	}
1341 }
1342 
1343 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1344 {
1345 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1346 
1347 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1348 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1349 }
1350 
1351 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1352 {
1353 	if (sdkp->device->removable || sdkp->write_prot) {
1354 		if (disk_check_media_change(disk))
1355 			return true;
1356 	}
1357 
1358 	/*
1359 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1360 	 * nothing to do with partitions, BLKRRPART is used to force a full
1361 	 * revalidate after things like a format for historical reasons.
1362 	 */
1363 	return test_bit(GD_NEED_PART_SCAN, &disk->state);
1364 }
1365 
1366 /**
1367  *	sd_open - open a scsi disk device
1368  *	@disk: disk to open
1369  *	@mode: open mode
1370  *
1371  *	Returns 0 if successful. Returns a negated errno value in case
1372  *	of error.
1373  *
1374  *	Note: This can be called from a user context (e.g. fsck(1) )
1375  *	or from within the kernel (e.g. as a result of a mount(1) ).
1376  *	In the latter case @inode and @filp carry an abridged amount
1377  *	of information as noted above.
1378  *
1379  *	Locking: called with disk->open_mutex held.
1380  **/
1381 static int sd_open(struct gendisk *disk, blk_mode_t mode)
1382 {
1383 	struct scsi_disk *sdkp = scsi_disk(disk);
1384 	struct scsi_device *sdev = sdkp->device;
1385 	int retval;
1386 
1387 	if (scsi_device_get(sdev))
1388 		return -ENXIO;
1389 
1390 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1391 
1392 	/*
1393 	 * If the device is in error recovery, wait until it is done.
1394 	 * If the device is offline, then disallow any access to it.
1395 	 */
1396 	retval = -ENXIO;
1397 	if (!scsi_block_when_processing_errors(sdev))
1398 		goto error_out;
1399 
1400 	if (sd_need_revalidate(disk, sdkp))
1401 		sd_revalidate_disk(disk);
1402 
1403 	/*
1404 	 * If the drive is empty, just let the open fail.
1405 	 */
1406 	retval = -ENOMEDIUM;
1407 	if (sdev->removable && !sdkp->media_present &&
1408 	    !(mode & BLK_OPEN_NDELAY))
1409 		goto error_out;
1410 
1411 	/*
1412 	 * If the device has the write protect tab set, have the open fail
1413 	 * if the user expects to be able to write to the thing.
1414 	 */
1415 	retval = -EROFS;
1416 	if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1417 		goto error_out;
1418 
1419 	/*
1420 	 * It is possible that the disk changing stuff resulted in
1421 	 * the device being taken offline.  If this is the case,
1422 	 * report this to the user, and don't pretend that the
1423 	 * open actually succeeded.
1424 	 */
1425 	retval = -ENXIO;
1426 	if (!scsi_device_online(sdev))
1427 		goto error_out;
1428 
1429 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1430 		if (scsi_block_when_processing_errors(sdev))
1431 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1432 	}
1433 
1434 	return 0;
1435 
1436 error_out:
1437 	scsi_device_put(sdev);
1438 	return retval;
1439 }
1440 
1441 /**
1442  *	sd_release - invoked when the (last) close(2) is called on this
1443  *	scsi disk.
1444  *	@disk: disk to release
1445  *
1446  *	Returns 0.
1447  *
1448  *	Note: may block (uninterruptible) if error recovery is underway
1449  *	on this disk.
1450  *
1451  *	Locking: called with disk->open_mutex held.
1452  **/
1453 static void sd_release(struct gendisk *disk)
1454 {
1455 	struct scsi_disk *sdkp = scsi_disk(disk);
1456 	struct scsi_device *sdev = sdkp->device;
1457 
1458 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1459 
1460 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1461 		if (scsi_block_when_processing_errors(sdev))
1462 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1463 	}
1464 
1465 	scsi_device_put(sdev);
1466 }
1467 
1468 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1469 {
1470 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1471 	struct scsi_device *sdp = sdkp->device;
1472 	struct Scsi_Host *host = sdp->host;
1473 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1474 	int diskinfo[4];
1475 
1476 	/* default to most commonly used values */
1477 	diskinfo[0] = 0x40;	/* 1 << 6 */
1478 	diskinfo[1] = 0x20;	/* 1 << 5 */
1479 	diskinfo[2] = capacity >> 11;
1480 
1481 	/* override with calculated, extended default, or driver values */
1482 	if (host->hostt->bios_param)
1483 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1484 	else
1485 		scsicam_bios_param(bdev, capacity, diskinfo);
1486 
1487 	geo->heads = diskinfo[0];
1488 	geo->sectors = diskinfo[1];
1489 	geo->cylinders = diskinfo[2];
1490 	return 0;
1491 }
1492 
1493 /**
1494  *	sd_ioctl - process an ioctl
1495  *	@bdev: target block device
1496  *	@mode: open mode
1497  *	@cmd: ioctl command number
1498  *	@arg: this is third argument given to ioctl(2) system call.
1499  *	Often contains a pointer.
1500  *
1501  *	Returns 0 if successful (some ioctls return positive numbers on
1502  *	success as well). Returns a negated errno value in case of error.
1503  *
1504  *	Note: most ioctls are forward onto the block subsystem or further
1505  *	down in the scsi subsystem.
1506  **/
1507 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1508 		    unsigned int cmd, unsigned long arg)
1509 {
1510 	struct gendisk *disk = bdev->bd_disk;
1511 	struct scsi_disk *sdkp = scsi_disk(disk);
1512 	struct scsi_device *sdp = sdkp->device;
1513 	void __user *p = (void __user *)arg;
1514 	int error;
1515 
1516 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1517 				    "cmd=0x%x\n", disk->disk_name, cmd));
1518 
1519 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1520 		return -ENOIOCTLCMD;
1521 
1522 	/*
1523 	 * If we are in the middle of error recovery, don't let anyone
1524 	 * else try and use this device.  Also, if error recovery fails, it
1525 	 * may try and take the device offline, in which case all further
1526 	 * access to the device is prohibited.
1527 	 */
1528 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1529 			(mode & BLK_OPEN_NDELAY));
1530 	if (error)
1531 		return error;
1532 
1533 	if (is_sed_ioctl(cmd))
1534 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1535 	return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1536 }
1537 
1538 static void set_media_not_present(struct scsi_disk *sdkp)
1539 {
1540 	if (sdkp->media_present)
1541 		sdkp->device->changed = 1;
1542 
1543 	if (sdkp->device->removable) {
1544 		sdkp->media_present = 0;
1545 		sdkp->capacity = 0;
1546 	}
1547 }
1548 
1549 static int media_not_present(struct scsi_disk *sdkp,
1550 			     struct scsi_sense_hdr *sshdr)
1551 {
1552 	if (!scsi_sense_valid(sshdr))
1553 		return 0;
1554 
1555 	/* not invoked for commands that could return deferred errors */
1556 	switch (sshdr->sense_key) {
1557 	case UNIT_ATTENTION:
1558 	case NOT_READY:
1559 		/* medium not present */
1560 		if (sshdr->asc == 0x3A) {
1561 			set_media_not_present(sdkp);
1562 			return 1;
1563 		}
1564 	}
1565 	return 0;
1566 }
1567 
1568 /**
1569  *	sd_check_events - check media events
1570  *	@disk: kernel device descriptor
1571  *	@clearing: disk events currently being cleared
1572  *
1573  *	Returns mask of DISK_EVENT_*.
1574  *
1575  *	Note: this function is invoked from the block subsystem.
1576  **/
1577 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1578 {
1579 	struct scsi_disk *sdkp = disk->private_data;
1580 	struct scsi_device *sdp;
1581 	int retval;
1582 	bool disk_changed;
1583 
1584 	if (!sdkp)
1585 		return 0;
1586 
1587 	sdp = sdkp->device;
1588 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1589 
1590 	/*
1591 	 * If the device is offline, don't send any commands - just pretend as
1592 	 * if the command failed.  If the device ever comes back online, we
1593 	 * can deal with it then.  It is only because of unrecoverable errors
1594 	 * that we would ever take a device offline in the first place.
1595 	 */
1596 	if (!scsi_device_online(sdp)) {
1597 		set_media_not_present(sdkp);
1598 		goto out;
1599 	}
1600 
1601 	/*
1602 	 * Using TEST_UNIT_READY enables differentiation between drive with
1603 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1604 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1605 	 *
1606 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1607 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1608 	 * sd_revalidate() is called.
1609 	 */
1610 	if (scsi_block_when_processing_errors(sdp)) {
1611 		struct scsi_sense_hdr sshdr = { 0, };
1612 
1613 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1614 					      &sshdr);
1615 
1616 		/* failed to execute TUR, assume media not present */
1617 		if (retval < 0 || host_byte(retval)) {
1618 			set_media_not_present(sdkp);
1619 			goto out;
1620 		}
1621 
1622 		if (media_not_present(sdkp, &sshdr))
1623 			goto out;
1624 	}
1625 
1626 	/*
1627 	 * For removable scsi disk we have to recognise the presence
1628 	 * of a disk in the drive.
1629 	 */
1630 	if (!sdkp->media_present)
1631 		sdp->changed = 1;
1632 	sdkp->media_present = 1;
1633 out:
1634 	/*
1635 	 * sdp->changed is set under the following conditions:
1636 	 *
1637 	 *	Medium present state has changed in either direction.
1638 	 *	Device has indicated UNIT_ATTENTION.
1639 	 */
1640 	disk_changed = sdp->changed;
1641 	sdp->changed = 0;
1642 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1643 }
1644 
1645 static int sd_sync_cache(struct scsi_disk *sdkp)
1646 {
1647 	int retries, res;
1648 	struct scsi_device *sdp = sdkp->device;
1649 	const int timeout = sdp->request_queue->rq_timeout
1650 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1651 	struct scsi_sense_hdr sshdr;
1652 	const struct scsi_exec_args exec_args = {
1653 		.req_flags = BLK_MQ_REQ_PM,
1654 		.sshdr = &sshdr,
1655 	};
1656 
1657 	if (!scsi_device_online(sdp))
1658 		return -ENODEV;
1659 
1660 	for (retries = 3; retries > 0; --retries) {
1661 		unsigned char cmd[16] = { 0 };
1662 
1663 		if (sdp->use_16_for_sync)
1664 			cmd[0] = SYNCHRONIZE_CACHE_16;
1665 		else
1666 			cmd[0] = SYNCHRONIZE_CACHE;
1667 		/*
1668 		 * Leave the rest of the command zero to indicate
1669 		 * flush everything.
1670 		 */
1671 		res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1672 				       timeout, sdkp->max_retries, &exec_args);
1673 		if (res == 0)
1674 			break;
1675 	}
1676 
1677 	if (res) {
1678 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1679 
1680 		if (res < 0)
1681 			return res;
1682 
1683 		if (scsi_status_is_check_condition(res) &&
1684 		    scsi_sense_valid(&sshdr)) {
1685 			sd_print_sense_hdr(sdkp, &sshdr);
1686 
1687 			/* we need to evaluate the error return  */
1688 			if (sshdr.asc == 0x3a ||	/* medium not present */
1689 			    sshdr.asc == 0x20 ||	/* invalid command */
1690 			    (sshdr.asc == 0x74 && sshdr.ascq == 0x71))	/* drive is password locked */
1691 				/* this is no error here */
1692 				return 0;
1693 
1694 			/*
1695 			 * If a format is in progress or if the drive does not
1696 			 * support sync, there is not much we can do because
1697 			 * this is called during shutdown or suspend so just
1698 			 * return success so those operations can proceed.
1699 			 */
1700 			if ((sshdr.asc == 0x04 && sshdr.ascq == 0x04) ||
1701 			    sshdr.sense_key == ILLEGAL_REQUEST)
1702 				return 0;
1703 		}
1704 
1705 		switch (host_byte(res)) {
1706 		/* ignore errors due to racing a disconnection */
1707 		case DID_BAD_TARGET:
1708 		case DID_NO_CONNECT:
1709 			return 0;
1710 		/* signal the upper layer it might try again */
1711 		case DID_BUS_BUSY:
1712 		case DID_IMM_RETRY:
1713 		case DID_REQUEUE:
1714 		case DID_SOFT_ERROR:
1715 			return -EBUSY;
1716 		default:
1717 			return -EIO;
1718 		}
1719 	}
1720 	return 0;
1721 }
1722 
1723 static void sd_rescan(struct device *dev)
1724 {
1725 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1726 
1727 	sd_revalidate_disk(sdkp->disk);
1728 }
1729 
1730 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1731 		enum blk_unique_id type)
1732 {
1733 	struct scsi_device *sdev = scsi_disk(disk)->device;
1734 	const struct scsi_vpd *vpd;
1735 	const unsigned char *d;
1736 	int ret = -ENXIO, len;
1737 
1738 	rcu_read_lock();
1739 	vpd = rcu_dereference(sdev->vpd_pg83);
1740 	if (!vpd)
1741 		goto out_unlock;
1742 
1743 	ret = -EINVAL;
1744 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1745 		/* we only care about designators with LU association */
1746 		if (((d[1] >> 4) & 0x3) != 0x00)
1747 			continue;
1748 		if ((d[1] & 0xf) != type)
1749 			continue;
1750 
1751 		/*
1752 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1753 		 * keep looking as one with more entropy might still show up.
1754 		 */
1755 		len = d[3];
1756 		if (len != 8 && len != 12 && len != 16)
1757 			continue;
1758 		ret = len;
1759 		memcpy(id, d + 4, len);
1760 		if (len == 16)
1761 			break;
1762 	}
1763 out_unlock:
1764 	rcu_read_unlock();
1765 	return ret;
1766 }
1767 
1768 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1769 {
1770 	switch (host_byte(result)) {
1771 	case DID_TRANSPORT_MARGINAL:
1772 	case DID_TRANSPORT_DISRUPTED:
1773 	case DID_BUS_BUSY:
1774 		return PR_STS_RETRY_PATH_FAILURE;
1775 	case DID_NO_CONNECT:
1776 		return PR_STS_PATH_FAILED;
1777 	case DID_TRANSPORT_FAILFAST:
1778 		return PR_STS_PATH_FAST_FAILED;
1779 	}
1780 
1781 	switch (status_byte(result)) {
1782 	case SAM_STAT_RESERVATION_CONFLICT:
1783 		return PR_STS_RESERVATION_CONFLICT;
1784 	case SAM_STAT_CHECK_CONDITION:
1785 		if (!scsi_sense_valid(sshdr))
1786 			return PR_STS_IOERR;
1787 
1788 		if (sshdr->sense_key == ILLEGAL_REQUEST &&
1789 		    (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1790 			return -EINVAL;
1791 
1792 		fallthrough;
1793 	default:
1794 		return PR_STS_IOERR;
1795 	}
1796 }
1797 
1798 static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1799 			    unsigned char *data, int data_len)
1800 {
1801 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1802 	struct scsi_device *sdev = sdkp->device;
1803 	struct scsi_sense_hdr sshdr;
1804 	u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1805 	const struct scsi_exec_args exec_args = {
1806 		.sshdr = &sshdr,
1807 	};
1808 	int result;
1809 
1810 	put_unaligned_be16(data_len, &cmd[7]);
1811 
1812 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1813 				  SD_TIMEOUT, sdkp->max_retries, &exec_args);
1814 	if (scsi_status_is_check_condition(result) &&
1815 	    scsi_sense_valid(&sshdr)) {
1816 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1817 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1818 	}
1819 
1820 	if (result <= 0)
1821 		return result;
1822 
1823 	return sd_scsi_to_pr_err(&sshdr, result);
1824 }
1825 
1826 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1827 {
1828 	int result, i, data_offset, num_copy_keys;
1829 	u32 num_keys = keys_info->num_keys;
1830 	int data_len = num_keys * 8 + 8;
1831 	u8 *data;
1832 
1833 	data = kzalloc(data_len, GFP_KERNEL);
1834 	if (!data)
1835 		return -ENOMEM;
1836 
1837 	result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1838 	if (result)
1839 		goto free_data;
1840 
1841 	keys_info->generation = get_unaligned_be32(&data[0]);
1842 	keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1843 
1844 	data_offset = 8;
1845 	num_copy_keys = min(num_keys, keys_info->num_keys);
1846 
1847 	for (i = 0; i < num_copy_keys; i++) {
1848 		keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1849 		data_offset += 8;
1850 	}
1851 
1852 free_data:
1853 	kfree(data);
1854 	return result;
1855 }
1856 
1857 static int sd_pr_read_reservation(struct block_device *bdev,
1858 				  struct pr_held_reservation *rsv)
1859 {
1860 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1861 	struct scsi_device *sdev = sdkp->device;
1862 	u8 data[24] = { };
1863 	int result, len;
1864 
1865 	result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1866 	if (result)
1867 		return result;
1868 
1869 	len = get_unaligned_be32(&data[4]);
1870 	if (!len)
1871 		return 0;
1872 
1873 	/* Make sure we have at least the key and type */
1874 	if (len < 14) {
1875 		sdev_printk(KERN_INFO, sdev,
1876 			    "READ RESERVATION failed due to short return buffer of %d bytes\n",
1877 			    len);
1878 		return -EINVAL;
1879 	}
1880 
1881 	rsv->generation = get_unaligned_be32(&data[0]);
1882 	rsv->key = get_unaligned_be64(&data[8]);
1883 	rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1884 	return 0;
1885 }
1886 
1887 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1888 			     u64 sa_key, enum scsi_pr_type type, u8 flags)
1889 {
1890 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1891 	struct scsi_device *sdev = sdkp->device;
1892 	struct scsi_sense_hdr sshdr;
1893 	const struct scsi_exec_args exec_args = {
1894 		.sshdr = &sshdr,
1895 	};
1896 	int result;
1897 	u8 cmd[16] = { 0, };
1898 	u8 data[24] = { 0, };
1899 
1900 	cmd[0] = PERSISTENT_RESERVE_OUT;
1901 	cmd[1] = sa;
1902 	cmd[2] = type;
1903 	put_unaligned_be32(sizeof(data), &cmd[5]);
1904 
1905 	put_unaligned_be64(key, &data[0]);
1906 	put_unaligned_be64(sa_key, &data[8]);
1907 	data[20] = flags;
1908 
1909 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1910 				  sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1911 				  &exec_args);
1912 
1913 	if (scsi_status_is_check_condition(result) &&
1914 	    scsi_sense_valid(&sshdr)) {
1915 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1916 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1917 	}
1918 
1919 	if (result <= 0)
1920 		return result;
1921 
1922 	return sd_scsi_to_pr_err(&sshdr, result);
1923 }
1924 
1925 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1926 		u32 flags)
1927 {
1928 	if (flags & ~PR_FL_IGNORE_KEY)
1929 		return -EOPNOTSUPP;
1930 	return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1931 			old_key, new_key, 0,
1932 			(1 << 0) /* APTPL */);
1933 }
1934 
1935 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1936 		u32 flags)
1937 {
1938 	if (flags)
1939 		return -EOPNOTSUPP;
1940 	return sd_pr_out_command(bdev, 0x01, key, 0,
1941 				 block_pr_type_to_scsi(type), 0);
1942 }
1943 
1944 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1945 {
1946 	return sd_pr_out_command(bdev, 0x02, key, 0,
1947 				 block_pr_type_to_scsi(type), 0);
1948 }
1949 
1950 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1951 		enum pr_type type, bool abort)
1952 {
1953 	return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1954 				 block_pr_type_to_scsi(type), 0);
1955 }
1956 
1957 static int sd_pr_clear(struct block_device *bdev, u64 key)
1958 {
1959 	return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
1960 }
1961 
1962 static const struct pr_ops sd_pr_ops = {
1963 	.pr_register	= sd_pr_register,
1964 	.pr_reserve	= sd_pr_reserve,
1965 	.pr_release	= sd_pr_release,
1966 	.pr_preempt	= sd_pr_preempt,
1967 	.pr_clear	= sd_pr_clear,
1968 	.pr_read_keys	= sd_pr_read_keys,
1969 	.pr_read_reservation = sd_pr_read_reservation,
1970 };
1971 
1972 static void scsi_disk_free_disk(struct gendisk *disk)
1973 {
1974 	struct scsi_disk *sdkp = scsi_disk(disk);
1975 
1976 	put_device(&sdkp->disk_dev);
1977 }
1978 
1979 static const struct block_device_operations sd_fops = {
1980 	.owner			= THIS_MODULE,
1981 	.open			= sd_open,
1982 	.release		= sd_release,
1983 	.ioctl			= sd_ioctl,
1984 	.getgeo			= sd_getgeo,
1985 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1986 	.check_events		= sd_check_events,
1987 	.unlock_native_capacity	= sd_unlock_native_capacity,
1988 	.report_zones		= sd_zbc_report_zones,
1989 	.get_unique_id		= sd_get_unique_id,
1990 	.free_disk		= scsi_disk_free_disk,
1991 	.pr_ops			= &sd_pr_ops,
1992 };
1993 
1994 /**
1995  *	sd_eh_reset - reset error handling callback
1996  *	@scmd:		sd-issued command that has failed
1997  *
1998  *	This function is called by the SCSI midlayer before starting
1999  *	SCSI EH. When counting medium access failures we have to be
2000  *	careful to register it only only once per device and SCSI EH run;
2001  *	there might be several timed out commands which will cause the
2002  *	'max_medium_access_timeouts' counter to trigger after the first
2003  *	SCSI EH run already and set the device to offline.
2004  *	So this function resets the internal counter before starting SCSI EH.
2005  **/
2006 static void sd_eh_reset(struct scsi_cmnd *scmd)
2007 {
2008 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2009 
2010 	/* New SCSI EH run, reset gate variable */
2011 	sdkp->ignore_medium_access_errors = false;
2012 }
2013 
2014 /**
2015  *	sd_eh_action - error handling callback
2016  *	@scmd:		sd-issued command that has failed
2017  *	@eh_disp:	The recovery disposition suggested by the midlayer
2018  *
2019  *	This function is called by the SCSI midlayer upon completion of an
2020  *	error test command (currently TEST UNIT READY). The result of sending
2021  *	the eh command is passed in eh_disp.  We're looking for devices that
2022  *	fail medium access commands but are OK with non access commands like
2023  *	test unit ready (so wrongly see the device as having a successful
2024  *	recovery)
2025  **/
2026 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2027 {
2028 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2029 	struct scsi_device *sdev = scmd->device;
2030 
2031 	if (!scsi_device_online(sdev) ||
2032 	    !scsi_medium_access_command(scmd) ||
2033 	    host_byte(scmd->result) != DID_TIME_OUT ||
2034 	    eh_disp != SUCCESS)
2035 		return eh_disp;
2036 
2037 	/*
2038 	 * The device has timed out executing a medium access command.
2039 	 * However, the TEST UNIT READY command sent during error
2040 	 * handling completed successfully. Either the device is in the
2041 	 * process of recovering or has it suffered an internal failure
2042 	 * that prevents access to the storage medium.
2043 	 */
2044 	if (!sdkp->ignore_medium_access_errors) {
2045 		sdkp->medium_access_timed_out++;
2046 		sdkp->ignore_medium_access_errors = true;
2047 	}
2048 
2049 	/*
2050 	 * If the device keeps failing read/write commands but TEST UNIT
2051 	 * READY always completes successfully we assume that medium
2052 	 * access is no longer possible and take the device offline.
2053 	 */
2054 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2055 		scmd_printk(KERN_ERR, scmd,
2056 			    "Medium access timeout failure. Offlining disk!\n");
2057 		mutex_lock(&sdev->state_mutex);
2058 		scsi_device_set_state(sdev, SDEV_OFFLINE);
2059 		mutex_unlock(&sdev->state_mutex);
2060 
2061 		return SUCCESS;
2062 	}
2063 
2064 	return eh_disp;
2065 }
2066 
2067 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2068 {
2069 	struct request *req = scsi_cmd_to_rq(scmd);
2070 	struct scsi_device *sdev = scmd->device;
2071 	unsigned int transferred, good_bytes;
2072 	u64 start_lba, end_lba, bad_lba;
2073 
2074 	/*
2075 	 * Some commands have a payload smaller than the device logical
2076 	 * block size (e.g. INQUIRY on a 4K disk).
2077 	 */
2078 	if (scsi_bufflen(scmd) <= sdev->sector_size)
2079 		return 0;
2080 
2081 	/* Check if we have a 'bad_lba' information */
2082 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2083 				     SCSI_SENSE_BUFFERSIZE,
2084 				     &bad_lba))
2085 		return 0;
2086 
2087 	/*
2088 	 * If the bad lba was reported incorrectly, we have no idea where
2089 	 * the error is.
2090 	 */
2091 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2092 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2093 	if (bad_lba < start_lba || bad_lba >= end_lba)
2094 		return 0;
2095 
2096 	/*
2097 	 * resid is optional but mostly filled in.  When it's unused,
2098 	 * its value is zero, so we assume the whole buffer transferred
2099 	 */
2100 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2101 
2102 	/* This computation should always be done in terms of the
2103 	 * resolution of the device's medium.
2104 	 */
2105 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2106 
2107 	return min(good_bytes, transferred);
2108 }
2109 
2110 /**
2111  *	sd_done - bottom half handler: called when the lower level
2112  *	driver has completed (successfully or otherwise) a scsi command.
2113  *	@SCpnt: mid-level's per command structure.
2114  *
2115  *	Note: potentially run from within an ISR. Must not block.
2116  **/
2117 static int sd_done(struct scsi_cmnd *SCpnt)
2118 {
2119 	int result = SCpnt->result;
2120 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2121 	unsigned int sector_size = SCpnt->device->sector_size;
2122 	unsigned int resid;
2123 	struct scsi_sense_hdr sshdr;
2124 	struct request *req = scsi_cmd_to_rq(SCpnt);
2125 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2126 	int sense_valid = 0;
2127 	int sense_deferred = 0;
2128 
2129 	switch (req_op(req)) {
2130 	case REQ_OP_DISCARD:
2131 	case REQ_OP_WRITE_ZEROES:
2132 	case REQ_OP_ZONE_RESET:
2133 	case REQ_OP_ZONE_RESET_ALL:
2134 	case REQ_OP_ZONE_OPEN:
2135 	case REQ_OP_ZONE_CLOSE:
2136 	case REQ_OP_ZONE_FINISH:
2137 		if (!result) {
2138 			good_bytes = blk_rq_bytes(req);
2139 			scsi_set_resid(SCpnt, 0);
2140 		} else {
2141 			good_bytes = 0;
2142 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2143 		}
2144 		break;
2145 	default:
2146 		/*
2147 		 * In case of bogus fw or device, we could end up having
2148 		 * an unaligned partial completion. Check this here and force
2149 		 * alignment.
2150 		 */
2151 		resid = scsi_get_resid(SCpnt);
2152 		if (resid & (sector_size - 1)) {
2153 			sd_printk(KERN_INFO, sdkp,
2154 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2155 				resid, sector_size);
2156 			scsi_print_command(SCpnt);
2157 			resid = min(scsi_bufflen(SCpnt),
2158 				    round_up(resid, sector_size));
2159 			scsi_set_resid(SCpnt, resid);
2160 		}
2161 	}
2162 
2163 	if (result) {
2164 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2165 		if (sense_valid)
2166 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2167 	}
2168 	sdkp->medium_access_timed_out = 0;
2169 
2170 	if (!scsi_status_is_check_condition(result) &&
2171 	    (!sense_valid || sense_deferred))
2172 		goto out;
2173 
2174 	switch (sshdr.sense_key) {
2175 	case HARDWARE_ERROR:
2176 	case MEDIUM_ERROR:
2177 		good_bytes = sd_completed_bytes(SCpnt);
2178 		break;
2179 	case RECOVERED_ERROR:
2180 		good_bytes = scsi_bufflen(SCpnt);
2181 		break;
2182 	case NO_SENSE:
2183 		/* This indicates a false check condition, so ignore it.  An
2184 		 * unknown amount of data was transferred so treat it as an
2185 		 * error.
2186 		 */
2187 		SCpnt->result = 0;
2188 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2189 		break;
2190 	case ABORTED_COMMAND:
2191 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2192 			good_bytes = sd_completed_bytes(SCpnt);
2193 		break;
2194 	case ILLEGAL_REQUEST:
2195 		switch (sshdr.asc) {
2196 		case 0x10:	/* DIX: Host detected corruption */
2197 			good_bytes = sd_completed_bytes(SCpnt);
2198 			break;
2199 		case 0x20:	/* INVALID COMMAND OPCODE */
2200 		case 0x24:	/* INVALID FIELD IN CDB */
2201 			switch (SCpnt->cmnd[0]) {
2202 			case UNMAP:
2203 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2204 				break;
2205 			case WRITE_SAME_16:
2206 			case WRITE_SAME:
2207 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2208 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2209 				} else {
2210 					sdkp->device->no_write_same = 1;
2211 					sd_config_write_same(sdkp);
2212 					req->rq_flags |= RQF_QUIET;
2213 				}
2214 				break;
2215 			}
2216 		}
2217 		break;
2218 	default:
2219 		break;
2220 	}
2221 
2222  out:
2223 	if (sd_is_zoned(sdkp))
2224 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2225 
2226 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2227 					   "sd_done: completed %d of %d bytes\n",
2228 					   good_bytes, scsi_bufflen(SCpnt)));
2229 
2230 	return good_bytes;
2231 }
2232 
2233 /*
2234  * spinup disk - called only in sd_revalidate_disk()
2235  */
2236 static void
2237 sd_spinup_disk(struct scsi_disk *sdkp)
2238 {
2239 	unsigned char cmd[10];
2240 	unsigned long spintime_expire = 0;
2241 	int retries, spintime;
2242 	unsigned int the_result;
2243 	struct scsi_sense_hdr sshdr;
2244 	const struct scsi_exec_args exec_args = {
2245 		.sshdr = &sshdr,
2246 	};
2247 	int sense_valid = 0;
2248 
2249 	spintime = 0;
2250 
2251 	/* Spin up drives, as required.  Only do this at boot time */
2252 	/* Spinup needs to be done for module loads too. */
2253 	do {
2254 		retries = 0;
2255 
2256 		do {
2257 			bool media_was_present = sdkp->media_present;
2258 
2259 			cmd[0] = TEST_UNIT_READY;
2260 			memset((void *) &cmd[1], 0, 9);
2261 
2262 			the_result = scsi_execute_cmd(sdkp->device, cmd,
2263 						      REQ_OP_DRV_IN, NULL, 0,
2264 						      SD_TIMEOUT,
2265 						      sdkp->max_retries,
2266 						      &exec_args);
2267 
2268 			/*
2269 			 * If the drive has indicated to us that it
2270 			 * doesn't have any media in it, don't bother
2271 			 * with any more polling.
2272 			 */
2273 			if (media_not_present(sdkp, &sshdr)) {
2274 				if (media_was_present)
2275 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2276 				return;
2277 			}
2278 
2279 			if (the_result)
2280 				sense_valid = scsi_sense_valid(&sshdr);
2281 			retries++;
2282 		} while (retries < 3 &&
2283 			 (!scsi_status_is_good(the_result) ||
2284 			  (scsi_status_is_check_condition(the_result) &&
2285 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2286 
2287 		if (!scsi_status_is_check_condition(the_result)) {
2288 			/* no sense, TUR either succeeded or failed
2289 			 * with a status error */
2290 			if(!spintime && !scsi_status_is_good(the_result)) {
2291 				sd_print_result(sdkp, "Test Unit Ready failed",
2292 						the_result);
2293 			}
2294 			break;
2295 		}
2296 
2297 		/*
2298 		 * The device does not want the automatic start to be issued.
2299 		 */
2300 		if (sdkp->device->no_start_on_add)
2301 			break;
2302 
2303 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2304 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2305 				break;	/* manual intervention required */
2306 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2307 				break;	/* standby */
2308 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2309 				break;	/* unavailable */
2310 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2311 				break;	/* sanitize in progress */
2312 			/*
2313 			 * Issue command to spin up drive when not ready
2314 			 */
2315 			if (!spintime) {
2316 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2317 				cmd[0] = START_STOP;
2318 				cmd[1] = 1;	/* Return immediately */
2319 				memset((void *) &cmd[2], 0, 8);
2320 				cmd[4] = 1;	/* Start spin cycle */
2321 				if (sdkp->device->start_stop_pwr_cond)
2322 					cmd[4] |= 1 << 4;
2323 				scsi_execute_cmd(sdkp->device, cmd,
2324 						 REQ_OP_DRV_IN, NULL, 0,
2325 						 SD_TIMEOUT, sdkp->max_retries,
2326 						 &exec_args);
2327 				spintime_expire = jiffies + 100 * HZ;
2328 				spintime = 1;
2329 			}
2330 			/* Wait 1 second for next try */
2331 			msleep(1000);
2332 			printk(KERN_CONT ".");
2333 
2334 		/*
2335 		 * Wait for USB flash devices with slow firmware.
2336 		 * Yes, this sense key/ASC combination shouldn't
2337 		 * occur here.  It's characteristic of these devices.
2338 		 */
2339 		} else if (sense_valid &&
2340 				sshdr.sense_key == UNIT_ATTENTION &&
2341 				sshdr.asc == 0x28) {
2342 			if (!spintime) {
2343 				spintime_expire = jiffies + 5 * HZ;
2344 				spintime = 1;
2345 			}
2346 			/* Wait 1 second for next try */
2347 			msleep(1000);
2348 		} else {
2349 			/* we don't understand the sense code, so it's
2350 			 * probably pointless to loop */
2351 			if(!spintime) {
2352 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2353 				sd_print_sense_hdr(sdkp, &sshdr);
2354 			}
2355 			break;
2356 		}
2357 
2358 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2359 
2360 	if (spintime) {
2361 		if (scsi_status_is_good(the_result))
2362 			printk(KERN_CONT "ready\n");
2363 		else
2364 			printk(KERN_CONT "not responding...\n");
2365 	}
2366 }
2367 
2368 /*
2369  * Determine whether disk supports Data Integrity Field.
2370  */
2371 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2372 {
2373 	struct scsi_device *sdp = sdkp->device;
2374 	u8 type;
2375 
2376 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2377 		sdkp->protection_type = 0;
2378 		return 0;
2379 	}
2380 
2381 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2382 
2383 	if (type > T10_PI_TYPE3_PROTECTION) {
2384 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported"	\
2385 			  " protection type %u. Disabling disk!\n",
2386 			  type);
2387 		sdkp->protection_type = 0;
2388 		return -ENODEV;
2389 	}
2390 
2391 	sdkp->protection_type = type;
2392 
2393 	return 0;
2394 }
2395 
2396 static void sd_config_protection(struct scsi_disk *sdkp)
2397 {
2398 	struct scsi_device *sdp = sdkp->device;
2399 
2400 	sd_dif_config_host(sdkp);
2401 
2402 	if (!sdkp->protection_type)
2403 		return;
2404 
2405 	if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2406 		sd_first_printk(KERN_NOTICE, sdkp,
2407 				"Disabling DIF Type %u protection\n",
2408 				sdkp->protection_type);
2409 		sdkp->protection_type = 0;
2410 	}
2411 
2412 	sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2413 			sdkp->protection_type);
2414 }
2415 
2416 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2417 			struct scsi_sense_hdr *sshdr, int sense_valid,
2418 			int the_result)
2419 {
2420 	if (sense_valid)
2421 		sd_print_sense_hdr(sdkp, sshdr);
2422 	else
2423 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2424 
2425 	/*
2426 	 * Set dirty bit for removable devices if not ready -
2427 	 * sometimes drives will not report this properly.
2428 	 */
2429 	if (sdp->removable &&
2430 	    sense_valid && sshdr->sense_key == NOT_READY)
2431 		set_media_not_present(sdkp);
2432 
2433 	/*
2434 	 * We used to set media_present to 0 here to indicate no media
2435 	 * in the drive, but some drives fail read capacity even with
2436 	 * media present, so we can't do that.
2437 	 */
2438 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2439 }
2440 
2441 #define RC16_LEN 32
2442 #if RC16_LEN > SD_BUF_SIZE
2443 #error RC16_LEN must not be more than SD_BUF_SIZE
2444 #endif
2445 
2446 #define READ_CAPACITY_RETRIES_ON_RESET	10
2447 
2448 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2449 						unsigned char *buffer)
2450 {
2451 	unsigned char cmd[16];
2452 	struct scsi_sense_hdr sshdr;
2453 	const struct scsi_exec_args exec_args = {
2454 		.sshdr = &sshdr,
2455 	};
2456 	int sense_valid = 0;
2457 	int the_result;
2458 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2459 	unsigned int alignment;
2460 	unsigned long long lba;
2461 	unsigned sector_size;
2462 
2463 	if (sdp->no_read_capacity_16)
2464 		return -EINVAL;
2465 
2466 	do {
2467 		memset(cmd, 0, 16);
2468 		cmd[0] = SERVICE_ACTION_IN_16;
2469 		cmd[1] = SAI_READ_CAPACITY_16;
2470 		cmd[13] = RC16_LEN;
2471 		memset(buffer, 0, RC16_LEN);
2472 
2473 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2474 					      buffer, RC16_LEN, SD_TIMEOUT,
2475 					      sdkp->max_retries, &exec_args);
2476 
2477 		if (media_not_present(sdkp, &sshdr))
2478 			return -ENODEV;
2479 
2480 		if (the_result > 0) {
2481 			sense_valid = scsi_sense_valid(&sshdr);
2482 			if (sense_valid &&
2483 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2484 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2485 			    sshdr.ascq == 0x00)
2486 				/* Invalid Command Operation Code or
2487 				 * Invalid Field in CDB, just retry
2488 				 * silently with RC10 */
2489 				return -EINVAL;
2490 			if (sense_valid &&
2491 			    sshdr.sense_key == UNIT_ATTENTION &&
2492 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2493 				/* Device reset might occur several times,
2494 				 * give it one more chance */
2495 				if (--reset_retries > 0)
2496 					continue;
2497 		}
2498 		retries--;
2499 
2500 	} while (the_result && retries);
2501 
2502 	if (the_result) {
2503 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2504 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2505 		return -EINVAL;
2506 	}
2507 
2508 	sector_size = get_unaligned_be32(&buffer[8]);
2509 	lba = get_unaligned_be64(&buffer[0]);
2510 
2511 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2512 		sdkp->capacity = 0;
2513 		return -ENODEV;
2514 	}
2515 
2516 	/* Logical blocks per physical block exponent */
2517 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2518 
2519 	/* RC basis */
2520 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2521 
2522 	/* Lowest aligned logical block */
2523 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2524 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2525 	if (alignment && sdkp->first_scan)
2526 		sd_printk(KERN_NOTICE, sdkp,
2527 			  "physical block alignment offset: %u\n", alignment);
2528 
2529 	if (buffer[14] & 0x80) { /* LBPME */
2530 		sdkp->lbpme = 1;
2531 
2532 		if (buffer[14] & 0x40) /* LBPRZ */
2533 			sdkp->lbprz = 1;
2534 
2535 		sd_config_discard(sdkp, SD_LBP_WS16);
2536 	}
2537 
2538 	sdkp->capacity = lba + 1;
2539 	return sector_size;
2540 }
2541 
2542 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2543 						unsigned char *buffer)
2544 {
2545 	unsigned char cmd[16];
2546 	struct scsi_sense_hdr sshdr;
2547 	const struct scsi_exec_args exec_args = {
2548 		.sshdr = &sshdr,
2549 	};
2550 	int sense_valid = 0;
2551 	int the_result;
2552 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2553 	sector_t lba;
2554 	unsigned sector_size;
2555 
2556 	do {
2557 		cmd[0] = READ_CAPACITY;
2558 		memset(&cmd[1], 0, 9);
2559 		memset(buffer, 0, 8);
2560 
2561 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2562 					      8, SD_TIMEOUT, sdkp->max_retries,
2563 					      &exec_args);
2564 
2565 		if (media_not_present(sdkp, &sshdr))
2566 			return -ENODEV;
2567 
2568 		if (the_result > 0) {
2569 			sense_valid = scsi_sense_valid(&sshdr);
2570 			if (sense_valid &&
2571 			    sshdr.sense_key == UNIT_ATTENTION &&
2572 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2573 				/* Device reset might occur several times,
2574 				 * give it one more chance */
2575 				if (--reset_retries > 0)
2576 					continue;
2577 		}
2578 		retries--;
2579 
2580 	} while (the_result && retries);
2581 
2582 	if (the_result) {
2583 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2584 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2585 		return -EINVAL;
2586 	}
2587 
2588 	sector_size = get_unaligned_be32(&buffer[4]);
2589 	lba = get_unaligned_be32(&buffer[0]);
2590 
2591 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2592 		/* Some buggy (usb cardreader) devices return an lba of
2593 		   0xffffffff when the want to report a size of 0 (with
2594 		   which they really mean no media is present) */
2595 		sdkp->capacity = 0;
2596 		sdkp->physical_block_size = sector_size;
2597 		return sector_size;
2598 	}
2599 
2600 	sdkp->capacity = lba + 1;
2601 	sdkp->physical_block_size = sector_size;
2602 	return sector_size;
2603 }
2604 
2605 static int sd_try_rc16_first(struct scsi_device *sdp)
2606 {
2607 	if (sdp->host->max_cmd_len < 16)
2608 		return 0;
2609 	if (sdp->try_rc_10_first)
2610 		return 0;
2611 	if (sdp->scsi_level > SCSI_SPC_2)
2612 		return 1;
2613 	if (scsi_device_protection(sdp))
2614 		return 1;
2615 	return 0;
2616 }
2617 
2618 /*
2619  * read disk capacity
2620  */
2621 static void
2622 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2623 {
2624 	int sector_size;
2625 	struct scsi_device *sdp = sdkp->device;
2626 
2627 	if (sd_try_rc16_first(sdp)) {
2628 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2629 		if (sector_size == -EOVERFLOW)
2630 			goto got_data;
2631 		if (sector_size == -ENODEV)
2632 			return;
2633 		if (sector_size < 0)
2634 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2635 		if (sector_size < 0)
2636 			return;
2637 	} else {
2638 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2639 		if (sector_size == -EOVERFLOW)
2640 			goto got_data;
2641 		if (sector_size < 0)
2642 			return;
2643 		if ((sizeof(sdkp->capacity) > 4) &&
2644 		    (sdkp->capacity > 0xffffffffULL)) {
2645 			int old_sector_size = sector_size;
2646 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2647 					"Trying to use READ CAPACITY(16).\n");
2648 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2649 			if (sector_size < 0) {
2650 				sd_printk(KERN_NOTICE, sdkp,
2651 					"Using 0xffffffff as device size\n");
2652 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2653 				sector_size = old_sector_size;
2654 				goto got_data;
2655 			}
2656 			/* Remember that READ CAPACITY(16) succeeded */
2657 			sdp->try_rc_10_first = 0;
2658 		}
2659 	}
2660 
2661 	/* Some devices are known to return the total number of blocks,
2662 	 * not the highest block number.  Some devices have versions
2663 	 * which do this and others which do not.  Some devices we might
2664 	 * suspect of doing this but we don't know for certain.
2665 	 *
2666 	 * If we know the reported capacity is wrong, decrement it.  If
2667 	 * we can only guess, then assume the number of blocks is even
2668 	 * (usually true but not always) and err on the side of lowering
2669 	 * the capacity.
2670 	 */
2671 	if (sdp->fix_capacity ||
2672 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2673 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2674 				"from its reported value: %llu\n",
2675 				(unsigned long long) sdkp->capacity);
2676 		--sdkp->capacity;
2677 	}
2678 
2679 got_data:
2680 	if (sector_size == 0) {
2681 		sector_size = 512;
2682 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2683 			  "assuming 512.\n");
2684 	}
2685 
2686 	if (sector_size != 512 &&
2687 	    sector_size != 1024 &&
2688 	    sector_size != 2048 &&
2689 	    sector_size != 4096) {
2690 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2691 			  sector_size);
2692 		/*
2693 		 * The user might want to re-format the drive with
2694 		 * a supported sectorsize.  Once this happens, it
2695 		 * would be relatively trivial to set the thing up.
2696 		 * For this reason, we leave the thing in the table.
2697 		 */
2698 		sdkp->capacity = 0;
2699 		/*
2700 		 * set a bogus sector size so the normal read/write
2701 		 * logic in the block layer will eventually refuse any
2702 		 * request on this device without tripping over power
2703 		 * of two sector size assumptions
2704 		 */
2705 		sector_size = 512;
2706 	}
2707 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2708 	blk_queue_physical_block_size(sdp->request_queue,
2709 				      sdkp->physical_block_size);
2710 	sdkp->device->sector_size = sector_size;
2711 
2712 	if (sdkp->capacity > 0xffffffff)
2713 		sdp->use_16_for_rw = 1;
2714 
2715 }
2716 
2717 /*
2718  * Print disk capacity
2719  */
2720 static void
2721 sd_print_capacity(struct scsi_disk *sdkp,
2722 		  sector_t old_capacity)
2723 {
2724 	int sector_size = sdkp->device->sector_size;
2725 	char cap_str_2[10], cap_str_10[10];
2726 
2727 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2728 		return;
2729 
2730 	string_get_size(sdkp->capacity, sector_size,
2731 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2732 	string_get_size(sdkp->capacity, sector_size,
2733 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2734 
2735 	sd_printk(KERN_NOTICE, sdkp,
2736 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2737 		  (unsigned long long)sdkp->capacity,
2738 		  sector_size, cap_str_10, cap_str_2);
2739 
2740 	if (sdkp->physical_block_size != sector_size)
2741 		sd_printk(KERN_NOTICE, sdkp,
2742 			  "%u-byte physical blocks\n",
2743 			  sdkp->physical_block_size);
2744 }
2745 
2746 /* called with buffer of length 512 */
2747 static inline int
2748 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2749 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2750 		 struct scsi_sense_hdr *sshdr)
2751 {
2752 	/*
2753 	 * If we must use MODE SENSE(10), make sure that the buffer length
2754 	 * is at least 8 bytes so that the mode sense header fits.
2755 	 */
2756 	if (sdkp->device->use_10_for_ms && len < 8)
2757 		len = 8;
2758 
2759 	return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2760 			       SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2761 }
2762 
2763 /*
2764  * read write protect setting, if possible - called only in sd_revalidate_disk()
2765  * called with buffer of length SD_BUF_SIZE
2766  */
2767 static void
2768 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2769 {
2770 	int res;
2771 	struct scsi_device *sdp = sdkp->device;
2772 	struct scsi_mode_data data;
2773 	int old_wp = sdkp->write_prot;
2774 
2775 	set_disk_ro(sdkp->disk, 0);
2776 	if (sdp->skip_ms_page_3f) {
2777 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2778 		return;
2779 	}
2780 
2781 	if (sdp->use_192_bytes_for_3f) {
2782 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2783 	} else {
2784 		/*
2785 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2786 		 * We have to start carefully: some devices hang if we ask
2787 		 * for more than is available.
2788 		 */
2789 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2790 
2791 		/*
2792 		 * Second attempt: ask for page 0 When only page 0 is
2793 		 * implemented, a request for page 3F may return Sense Key
2794 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2795 		 * CDB.
2796 		 */
2797 		if (res < 0)
2798 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2799 
2800 		/*
2801 		 * Third attempt: ask 255 bytes, as we did earlier.
2802 		 */
2803 		if (res < 0)
2804 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2805 					       &data, NULL);
2806 	}
2807 
2808 	if (res < 0) {
2809 		sd_first_printk(KERN_WARNING, sdkp,
2810 			  "Test WP failed, assume Write Enabled\n");
2811 	} else {
2812 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2813 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2814 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2815 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2816 				  sdkp->write_prot ? "on" : "off");
2817 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2818 		}
2819 	}
2820 }
2821 
2822 /*
2823  * sd_read_cache_type - called only from sd_revalidate_disk()
2824  * called with buffer of length SD_BUF_SIZE
2825  */
2826 static void
2827 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2828 {
2829 	int len = 0, res;
2830 	struct scsi_device *sdp = sdkp->device;
2831 
2832 	int dbd;
2833 	int modepage;
2834 	int first_len;
2835 	struct scsi_mode_data data;
2836 	struct scsi_sense_hdr sshdr;
2837 	int old_wce = sdkp->WCE;
2838 	int old_rcd = sdkp->RCD;
2839 	int old_dpofua = sdkp->DPOFUA;
2840 
2841 
2842 	if (sdkp->cache_override)
2843 		return;
2844 
2845 	first_len = 4;
2846 	if (sdp->skip_ms_page_8) {
2847 		if (sdp->type == TYPE_RBC)
2848 			goto defaults;
2849 		else {
2850 			if (sdp->skip_ms_page_3f)
2851 				goto defaults;
2852 			modepage = 0x3F;
2853 			if (sdp->use_192_bytes_for_3f)
2854 				first_len = 192;
2855 			dbd = 0;
2856 		}
2857 	} else if (sdp->type == TYPE_RBC) {
2858 		modepage = 6;
2859 		dbd = 8;
2860 	} else {
2861 		modepage = 8;
2862 		dbd = 0;
2863 	}
2864 
2865 	/* cautiously ask */
2866 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2867 			&data, &sshdr);
2868 
2869 	if (res < 0)
2870 		goto bad_sense;
2871 
2872 	if (!data.header_length) {
2873 		modepage = 6;
2874 		first_len = 0;
2875 		sd_first_printk(KERN_ERR, sdkp,
2876 				"Missing header in MODE_SENSE response\n");
2877 	}
2878 
2879 	/* that went OK, now ask for the proper length */
2880 	len = data.length;
2881 
2882 	/*
2883 	 * We're only interested in the first three bytes, actually.
2884 	 * But the data cache page is defined for the first 20.
2885 	 */
2886 	if (len < 3)
2887 		goto bad_sense;
2888 	else if (len > SD_BUF_SIZE) {
2889 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2890 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2891 		len = SD_BUF_SIZE;
2892 	}
2893 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2894 		len = 192;
2895 
2896 	/* Get the data */
2897 	if (len > first_len)
2898 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2899 				&data, &sshdr);
2900 
2901 	if (!res) {
2902 		int offset = data.header_length + data.block_descriptor_length;
2903 
2904 		while (offset < len) {
2905 			u8 page_code = buffer[offset] & 0x3F;
2906 			u8 spf       = buffer[offset] & 0x40;
2907 
2908 			if (page_code == 8 || page_code == 6) {
2909 				/* We're interested only in the first 3 bytes.
2910 				 */
2911 				if (len - offset <= 2) {
2912 					sd_first_printk(KERN_ERR, sdkp,
2913 						"Incomplete mode parameter "
2914 							"data\n");
2915 					goto defaults;
2916 				} else {
2917 					modepage = page_code;
2918 					goto Page_found;
2919 				}
2920 			} else {
2921 				/* Go to the next page */
2922 				if (spf && len - offset > 3)
2923 					offset += 4 + (buffer[offset+2] << 8) +
2924 						buffer[offset+3];
2925 				else if (!spf && len - offset > 1)
2926 					offset += 2 + buffer[offset+1];
2927 				else {
2928 					sd_first_printk(KERN_ERR, sdkp,
2929 							"Incomplete mode "
2930 							"parameter data\n");
2931 					goto defaults;
2932 				}
2933 			}
2934 		}
2935 
2936 		sd_first_printk(KERN_WARNING, sdkp,
2937 				"No Caching mode page found\n");
2938 		goto defaults;
2939 
2940 	Page_found:
2941 		if (modepage == 8) {
2942 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2943 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2944 		} else {
2945 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2946 			sdkp->RCD = 0;
2947 		}
2948 
2949 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2950 		if (sdp->broken_fua) {
2951 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2952 			sdkp->DPOFUA = 0;
2953 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2954 			   !sdkp->device->use_16_for_rw) {
2955 			sd_first_printk(KERN_NOTICE, sdkp,
2956 				  "Uses READ/WRITE(6), disabling FUA\n");
2957 			sdkp->DPOFUA = 0;
2958 		}
2959 
2960 		/* No cache flush allowed for write protected devices */
2961 		if (sdkp->WCE && sdkp->write_prot)
2962 			sdkp->WCE = 0;
2963 
2964 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2965 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2966 			sd_printk(KERN_NOTICE, sdkp,
2967 				  "Write cache: %s, read cache: %s, %s\n",
2968 				  sdkp->WCE ? "enabled" : "disabled",
2969 				  sdkp->RCD ? "disabled" : "enabled",
2970 				  sdkp->DPOFUA ? "supports DPO and FUA"
2971 				  : "doesn't support DPO or FUA");
2972 
2973 		return;
2974 	}
2975 
2976 bad_sense:
2977 	if (scsi_sense_valid(&sshdr) &&
2978 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2979 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2980 		/* Invalid field in CDB */
2981 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2982 	else
2983 		sd_first_printk(KERN_ERR, sdkp,
2984 				"Asking for cache data failed\n");
2985 
2986 defaults:
2987 	if (sdp->wce_default_on) {
2988 		sd_first_printk(KERN_NOTICE, sdkp,
2989 				"Assuming drive cache: write back\n");
2990 		sdkp->WCE = 1;
2991 	} else {
2992 		sd_first_printk(KERN_WARNING, sdkp,
2993 				"Assuming drive cache: write through\n");
2994 		sdkp->WCE = 0;
2995 	}
2996 	sdkp->RCD = 0;
2997 	sdkp->DPOFUA = 0;
2998 }
2999 
3000 /*
3001  * The ATO bit indicates whether the DIF application tag is available
3002  * for use by the operating system.
3003  */
3004 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3005 {
3006 	int res, offset;
3007 	struct scsi_device *sdp = sdkp->device;
3008 	struct scsi_mode_data data;
3009 	struct scsi_sense_hdr sshdr;
3010 
3011 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3012 		return;
3013 
3014 	if (sdkp->protection_type == 0)
3015 		return;
3016 
3017 	res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3018 			      sdkp->max_retries, &data, &sshdr);
3019 
3020 	if (res < 0 || !data.header_length ||
3021 	    data.length < 6) {
3022 		sd_first_printk(KERN_WARNING, sdkp,
3023 			  "getting Control mode page failed, assume no ATO\n");
3024 
3025 		if (scsi_sense_valid(&sshdr))
3026 			sd_print_sense_hdr(sdkp, &sshdr);
3027 
3028 		return;
3029 	}
3030 
3031 	offset = data.header_length + data.block_descriptor_length;
3032 
3033 	if ((buffer[offset] & 0x3f) != 0x0a) {
3034 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3035 		return;
3036 	}
3037 
3038 	if ((buffer[offset + 5] & 0x80) == 0)
3039 		return;
3040 
3041 	sdkp->ATO = 1;
3042 
3043 	return;
3044 }
3045 
3046 /**
3047  * sd_read_block_limits - Query disk device for preferred I/O sizes.
3048  * @sdkp: disk to query
3049  */
3050 static void sd_read_block_limits(struct scsi_disk *sdkp)
3051 {
3052 	struct scsi_vpd *vpd;
3053 
3054 	rcu_read_lock();
3055 
3056 	vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3057 	if (!vpd || vpd->len < 16)
3058 		goto out;
3059 
3060 	sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3061 	sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3062 	sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3063 
3064 	if (vpd->len >= 64) {
3065 		unsigned int lba_count, desc_count;
3066 
3067 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3068 
3069 		if (!sdkp->lbpme)
3070 			goto out;
3071 
3072 		lba_count = get_unaligned_be32(&vpd->data[20]);
3073 		desc_count = get_unaligned_be32(&vpd->data[24]);
3074 
3075 		if (lba_count && desc_count)
3076 			sdkp->max_unmap_blocks = lba_count;
3077 
3078 		sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3079 
3080 		if (vpd->data[32] & 0x80)
3081 			sdkp->unmap_alignment =
3082 				get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3083 
3084 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3085 
3086 			if (sdkp->max_unmap_blocks)
3087 				sd_config_discard(sdkp, SD_LBP_UNMAP);
3088 			else
3089 				sd_config_discard(sdkp, SD_LBP_WS16);
3090 
3091 		} else {	/* LBP VPD page tells us what to use */
3092 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
3093 				sd_config_discard(sdkp, SD_LBP_UNMAP);
3094 			else if (sdkp->lbpws)
3095 				sd_config_discard(sdkp, SD_LBP_WS16);
3096 			else if (sdkp->lbpws10)
3097 				sd_config_discard(sdkp, SD_LBP_WS10);
3098 			else
3099 				sd_config_discard(sdkp, SD_LBP_DISABLE);
3100 		}
3101 	}
3102 
3103  out:
3104 	rcu_read_unlock();
3105 }
3106 
3107 /**
3108  * sd_read_block_characteristics - Query block dev. characteristics
3109  * @sdkp: disk to query
3110  */
3111 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3112 {
3113 	struct request_queue *q = sdkp->disk->queue;
3114 	struct scsi_vpd *vpd;
3115 	u16 rot;
3116 	u8 zoned;
3117 
3118 	rcu_read_lock();
3119 	vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3120 
3121 	if (!vpd || vpd->len <= 8) {
3122 		rcu_read_unlock();
3123 	        return;
3124 	}
3125 
3126 	rot = get_unaligned_be16(&vpd->data[4]);
3127 	zoned = (vpd->data[8] >> 4) & 3;
3128 	rcu_read_unlock();
3129 
3130 	if (rot == 1) {
3131 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3132 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3133 	}
3134 
3135 	if (sdkp->device->type == TYPE_ZBC) {
3136 		/*
3137 		 * Host-managed: Per ZBC and ZAC specifications, writes in
3138 		 * sequential write required zones of host-managed devices must
3139 		 * be aligned to the device physical block size.
3140 		 */
3141 		disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
3142 		blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3143 	} else {
3144 		sdkp->zoned = zoned;
3145 		if (sdkp->zoned == 1) {
3146 			/* Host-aware */
3147 			disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
3148 		} else {
3149 			/* Regular disk or drive managed disk */
3150 			disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3151 		}
3152 	}
3153 
3154 	if (!sdkp->first_scan)
3155 		return;
3156 
3157 	if (blk_queue_is_zoned(q)) {
3158 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3159 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3160 	} else {
3161 		if (sdkp->zoned == 1)
3162 			sd_printk(KERN_NOTICE, sdkp,
3163 				  "Host-aware SMR disk used as regular disk\n");
3164 		else if (sdkp->zoned == 2)
3165 			sd_printk(KERN_NOTICE, sdkp,
3166 				  "Drive-managed SMR disk\n");
3167 	}
3168 }
3169 
3170 /**
3171  * sd_read_block_provisioning - Query provisioning VPD page
3172  * @sdkp: disk to query
3173  */
3174 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3175 {
3176 	struct scsi_vpd *vpd;
3177 
3178 	if (sdkp->lbpme == 0)
3179 		return;
3180 
3181 	rcu_read_lock();
3182 	vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3183 
3184 	if (!vpd || vpd->len < 8) {
3185 		rcu_read_unlock();
3186 		return;
3187 	}
3188 
3189 	sdkp->lbpvpd	= 1;
3190 	sdkp->lbpu	= (vpd->data[5] >> 7) & 1; /* UNMAP */
3191 	sdkp->lbpws	= (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3192 	sdkp->lbpws10	= (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3193 	rcu_read_unlock();
3194 }
3195 
3196 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3197 {
3198 	struct scsi_device *sdev = sdkp->device;
3199 
3200 	if (sdev->host->no_write_same) {
3201 		sdev->no_write_same = 1;
3202 
3203 		return;
3204 	}
3205 
3206 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3207 		struct scsi_vpd *vpd;
3208 
3209 		sdev->no_report_opcodes = 1;
3210 
3211 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3212 		 * CODES is unsupported and the device has an ATA
3213 		 * Information VPD page (SAT).
3214 		 */
3215 		rcu_read_lock();
3216 		vpd = rcu_dereference(sdev->vpd_pg89);
3217 		if (vpd)
3218 			sdev->no_write_same = 1;
3219 		rcu_read_unlock();
3220 	}
3221 
3222 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3223 		sdkp->ws16 = 1;
3224 
3225 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3226 		sdkp->ws10 = 1;
3227 }
3228 
3229 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3230 {
3231 	struct scsi_device *sdev = sdkp->device;
3232 
3233 	if (!sdev->security_supported)
3234 		return;
3235 
3236 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3237 			SECURITY_PROTOCOL_IN, 0) == 1 &&
3238 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3239 			SECURITY_PROTOCOL_OUT, 0) == 1)
3240 		sdkp->security = 1;
3241 }
3242 
3243 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3244 {
3245 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3246 }
3247 
3248 /**
3249  * sd_read_cpr - Query concurrent positioning ranges
3250  * @sdkp:	disk to query
3251  */
3252 static void sd_read_cpr(struct scsi_disk *sdkp)
3253 {
3254 	struct blk_independent_access_ranges *iars = NULL;
3255 	unsigned char *buffer = NULL;
3256 	unsigned int nr_cpr = 0;
3257 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3258 	u8 *desc;
3259 
3260 	/*
3261 	 * We need to have the capacity set first for the block layer to be
3262 	 * able to check the ranges.
3263 	 */
3264 	if (sdkp->first_scan)
3265 		return;
3266 
3267 	if (!sdkp->capacity)
3268 		goto out;
3269 
3270 	/*
3271 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3272 	 * leading to a maximum page size of 64 + 256*32 bytes.
3273 	 */
3274 	buf_len = 64 + 256*32;
3275 	buffer = kmalloc(buf_len, GFP_KERNEL);
3276 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3277 		goto out;
3278 
3279 	/* We must have at least a 64B header and one 32B range descriptor */
3280 	vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3281 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3282 		sd_printk(KERN_ERR, sdkp,
3283 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3284 		goto out;
3285 	}
3286 
3287 	nr_cpr = (vpd_len - 64) / 32;
3288 	if (nr_cpr == 1) {
3289 		nr_cpr = 0;
3290 		goto out;
3291 	}
3292 
3293 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3294 	if (!iars) {
3295 		nr_cpr = 0;
3296 		goto out;
3297 	}
3298 
3299 	desc = &buffer[64];
3300 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3301 		if (desc[0] != i) {
3302 			sd_printk(KERN_ERR, sdkp,
3303 				"Invalid Concurrent Positioning Range number\n");
3304 			nr_cpr = 0;
3305 			break;
3306 		}
3307 
3308 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3309 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3310 	}
3311 
3312 out:
3313 	disk_set_independent_access_ranges(sdkp->disk, iars);
3314 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3315 		sd_printk(KERN_NOTICE, sdkp,
3316 			  "%u concurrent positioning ranges\n", nr_cpr);
3317 		sdkp->nr_actuators = nr_cpr;
3318 	}
3319 
3320 	kfree(buffer);
3321 }
3322 
3323 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3324 {
3325 	struct scsi_device *sdp = sdkp->device;
3326 	unsigned int min_xfer_bytes =
3327 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3328 
3329 	if (sdkp->min_xfer_blocks == 0)
3330 		return false;
3331 
3332 	if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3333 		sd_first_printk(KERN_WARNING, sdkp,
3334 				"Preferred minimum I/O size %u bytes not a " \
3335 				"multiple of physical block size (%u bytes)\n",
3336 				min_xfer_bytes, sdkp->physical_block_size);
3337 		sdkp->min_xfer_blocks = 0;
3338 		return false;
3339 	}
3340 
3341 	sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3342 			min_xfer_bytes);
3343 	return true;
3344 }
3345 
3346 /*
3347  * Determine the device's preferred I/O size for reads and writes
3348  * unless the reported value is unreasonably small, large, not a
3349  * multiple of the physical block size, or simply garbage.
3350  */
3351 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3352 				      unsigned int dev_max)
3353 {
3354 	struct scsi_device *sdp = sdkp->device;
3355 	unsigned int opt_xfer_bytes =
3356 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3357 	unsigned int min_xfer_bytes =
3358 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3359 
3360 	if (sdkp->opt_xfer_blocks == 0)
3361 		return false;
3362 
3363 	if (sdkp->opt_xfer_blocks > dev_max) {
3364 		sd_first_printk(KERN_WARNING, sdkp,
3365 				"Optimal transfer size %u logical blocks " \
3366 				"> dev_max (%u logical blocks)\n",
3367 				sdkp->opt_xfer_blocks, dev_max);
3368 		return false;
3369 	}
3370 
3371 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3372 		sd_first_printk(KERN_WARNING, sdkp,
3373 				"Optimal transfer size %u logical blocks " \
3374 				"> sd driver limit (%u logical blocks)\n",
3375 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3376 		return false;
3377 	}
3378 
3379 	if (opt_xfer_bytes < PAGE_SIZE) {
3380 		sd_first_printk(KERN_WARNING, sdkp,
3381 				"Optimal transfer size %u bytes < " \
3382 				"PAGE_SIZE (%u bytes)\n",
3383 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3384 		return false;
3385 	}
3386 
3387 	if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3388 		sd_first_printk(KERN_WARNING, sdkp,
3389 				"Optimal transfer size %u bytes not a " \
3390 				"multiple of preferred minimum block " \
3391 				"size (%u bytes)\n",
3392 				opt_xfer_bytes, min_xfer_bytes);
3393 		return false;
3394 	}
3395 
3396 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3397 		sd_first_printk(KERN_WARNING, sdkp,
3398 				"Optimal transfer size %u bytes not a " \
3399 				"multiple of physical block size (%u bytes)\n",
3400 				opt_xfer_bytes, sdkp->physical_block_size);
3401 		return false;
3402 	}
3403 
3404 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3405 			opt_xfer_bytes);
3406 	return true;
3407 }
3408 
3409 static void sd_read_block_zero(struct scsi_disk *sdkp)
3410 {
3411 	struct scsi_device *sdev = sdkp->device;
3412 	unsigned int buf_len = sdev->sector_size;
3413 	u8 *buffer, cmd[16] = { };
3414 
3415 	buffer = kmalloc(buf_len, GFP_KERNEL);
3416 	if (!buffer)
3417 		return;
3418 
3419 	if (sdev->use_16_for_rw) {
3420 		cmd[0] = READ_16;
3421 		put_unaligned_be64(0, &cmd[2]); /* Logical block address 0 */
3422 		put_unaligned_be32(1, &cmd[10]);/* Transfer 1 logical block */
3423 	} else {
3424 		cmd[0] = READ_10;
3425 		put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3426 		put_unaligned_be16(1, &cmd[7]);	/* Transfer 1 logical block */
3427 	}
3428 
3429 	scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3430 			 SD_TIMEOUT, sdkp->max_retries, NULL);
3431 	kfree(buffer);
3432 }
3433 
3434 /**
3435  *	sd_revalidate_disk - called the first time a new disk is seen,
3436  *	performs disk spin up, read_capacity, etc.
3437  *	@disk: struct gendisk we care about
3438  **/
3439 static int sd_revalidate_disk(struct gendisk *disk)
3440 {
3441 	struct scsi_disk *sdkp = scsi_disk(disk);
3442 	struct scsi_device *sdp = sdkp->device;
3443 	struct request_queue *q = sdkp->disk->queue;
3444 	sector_t old_capacity = sdkp->capacity;
3445 	unsigned char *buffer;
3446 	unsigned int dev_max, rw_max;
3447 
3448 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3449 				      "sd_revalidate_disk\n"));
3450 
3451 	/*
3452 	 * If the device is offline, don't try and read capacity or any
3453 	 * of the other niceties.
3454 	 */
3455 	if (!scsi_device_online(sdp))
3456 		goto out;
3457 
3458 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3459 	if (!buffer) {
3460 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3461 			  "allocation failure.\n");
3462 		goto out;
3463 	}
3464 
3465 	sd_spinup_disk(sdkp);
3466 
3467 	/*
3468 	 * Without media there is no reason to ask; moreover, some devices
3469 	 * react badly if we do.
3470 	 */
3471 	if (sdkp->media_present) {
3472 		sd_read_capacity(sdkp, buffer);
3473 		/*
3474 		 * Some USB/UAS devices return generic values for mode pages
3475 		 * until the media has been accessed. Trigger a READ operation
3476 		 * to force the device to populate mode pages.
3477 		 */
3478 		if (sdp->read_before_ms)
3479 			sd_read_block_zero(sdkp);
3480 		/*
3481 		 * set the default to rotational.  All non-rotational devices
3482 		 * support the block characteristics VPD page, which will
3483 		 * cause this to be updated correctly and any device which
3484 		 * doesn't support it should be treated as rotational.
3485 		 */
3486 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3487 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3488 
3489 		if (scsi_device_supports_vpd(sdp)) {
3490 			sd_read_block_provisioning(sdkp);
3491 			sd_read_block_limits(sdkp);
3492 			sd_read_block_characteristics(sdkp);
3493 			sd_zbc_read_zones(sdkp, buffer);
3494 			sd_read_cpr(sdkp);
3495 		}
3496 
3497 		sd_print_capacity(sdkp, old_capacity);
3498 
3499 		sd_read_write_protect_flag(sdkp, buffer);
3500 		sd_read_cache_type(sdkp, buffer);
3501 		sd_read_app_tag_own(sdkp, buffer);
3502 		sd_read_write_same(sdkp, buffer);
3503 		sd_read_security(sdkp, buffer);
3504 		sd_config_protection(sdkp);
3505 	}
3506 
3507 	/*
3508 	 * We now have all cache related info, determine how we deal
3509 	 * with flush requests.
3510 	 */
3511 	sd_set_flush_flag(sdkp);
3512 
3513 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3514 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3515 
3516 	/* Some devices report a maximum block count for READ/WRITE requests. */
3517 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3518 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3519 
3520 	if (sd_validate_min_xfer_size(sdkp))
3521 		blk_queue_io_min(sdkp->disk->queue,
3522 				 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3523 	else
3524 		blk_queue_io_min(sdkp->disk->queue, 0);
3525 
3526 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3527 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3528 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3529 	} else {
3530 		q->limits.io_opt = 0;
3531 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3532 				      (sector_t)BLK_DEF_MAX_SECTORS);
3533 	}
3534 
3535 	/*
3536 	 * Limit default to SCSI host optimal sector limit if set. There may be
3537 	 * an impact on performance for when the size of a request exceeds this
3538 	 * host limit.
3539 	 */
3540 	rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3541 
3542 	/* Do not exceed controller limit */
3543 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3544 
3545 	/*
3546 	 * Only update max_sectors if previously unset or if the current value
3547 	 * exceeds the capabilities of the hardware.
3548 	 */
3549 	if (sdkp->first_scan ||
3550 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3551 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3552 		q->limits.max_sectors = rw_max;
3553 
3554 	sdkp->first_scan = 0;
3555 
3556 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3557 	sd_config_write_same(sdkp);
3558 	kfree(buffer);
3559 
3560 	/*
3561 	 * For a zoned drive, revalidating the zones can be done only once
3562 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3563 	 * capacity to 0.
3564 	 */
3565 	if (sd_zbc_revalidate_zones(sdkp))
3566 		set_capacity_and_notify(disk, 0);
3567 
3568  out:
3569 	return 0;
3570 }
3571 
3572 /**
3573  *	sd_unlock_native_capacity - unlock native capacity
3574  *	@disk: struct gendisk to set capacity for
3575  *
3576  *	Block layer calls this function if it detects that partitions
3577  *	on @disk reach beyond the end of the device.  If the SCSI host
3578  *	implements ->unlock_native_capacity() method, it's invoked to
3579  *	give it a chance to adjust the device capacity.
3580  *
3581  *	CONTEXT:
3582  *	Defined by block layer.  Might sleep.
3583  */
3584 static void sd_unlock_native_capacity(struct gendisk *disk)
3585 {
3586 	struct scsi_device *sdev = scsi_disk(disk)->device;
3587 
3588 	if (sdev->host->hostt->unlock_native_capacity)
3589 		sdev->host->hostt->unlock_native_capacity(sdev);
3590 }
3591 
3592 /**
3593  *	sd_format_disk_name - format disk name
3594  *	@prefix: name prefix - ie. "sd" for SCSI disks
3595  *	@index: index of the disk to format name for
3596  *	@buf: output buffer
3597  *	@buflen: length of the output buffer
3598  *
3599  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3600  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3601  *	which is followed by sdaaa.
3602  *
3603  *	This is basically 26 base counting with one extra 'nil' entry
3604  *	at the beginning from the second digit on and can be
3605  *	determined using similar method as 26 base conversion with the
3606  *	index shifted -1 after each digit is computed.
3607  *
3608  *	CONTEXT:
3609  *	Don't care.
3610  *
3611  *	RETURNS:
3612  *	0 on success, -errno on failure.
3613  */
3614 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3615 {
3616 	const int base = 'z' - 'a' + 1;
3617 	char *begin = buf + strlen(prefix);
3618 	char *end = buf + buflen;
3619 	char *p;
3620 	int unit;
3621 
3622 	p = end - 1;
3623 	*p = '\0';
3624 	unit = base;
3625 	do {
3626 		if (p == begin)
3627 			return -EINVAL;
3628 		*--p = 'a' + (index % unit);
3629 		index = (index / unit) - 1;
3630 	} while (index >= 0);
3631 
3632 	memmove(begin, p, end - p);
3633 	memcpy(buf, prefix, strlen(prefix));
3634 
3635 	return 0;
3636 }
3637 
3638 /**
3639  *	sd_probe - called during driver initialization and whenever a
3640  *	new scsi device is attached to the system. It is called once
3641  *	for each scsi device (not just disks) present.
3642  *	@dev: pointer to device object
3643  *
3644  *	Returns 0 if successful (or not interested in this scsi device
3645  *	(e.g. scanner)); 1 when there is an error.
3646  *
3647  *	Note: this function is invoked from the scsi mid-level.
3648  *	This function sets up the mapping between a given
3649  *	<host,channel,id,lun> (found in sdp) and new device name
3650  *	(e.g. /dev/sda). More precisely it is the block device major
3651  *	and minor number that is chosen here.
3652  *
3653  *	Assume sd_probe is not re-entrant (for time being)
3654  *	Also think about sd_probe() and sd_remove() running coincidentally.
3655  **/
3656 static int sd_probe(struct device *dev)
3657 {
3658 	struct scsi_device *sdp = to_scsi_device(dev);
3659 	struct scsi_disk *sdkp;
3660 	struct gendisk *gd;
3661 	int index;
3662 	int error;
3663 
3664 	scsi_autopm_get_device(sdp);
3665 	error = -ENODEV;
3666 	if (sdp->type != TYPE_DISK &&
3667 	    sdp->type != TYPE_ZBC &&
3668 	    sdp->type != TYPE_MOD &&
3669 	    sdp->type != TYPE_RBC)
3670 		goto out;
3671 
3672 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3673 		sdev_printk(KERN_WARNING, sdp,
3674 			    "Unsupported ZBC host-managed device.\n");
3675 		goto out;
3676 	}
3677 
3678 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3679 					"sd_probe\n"));
3680 
3681 	error = -ENOMEM;
3682 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3683 	if (!sdkp)
3684 		goto out;
3685 
3686 	gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3687 					 &sd_bio_compl_lkclass);
3688 	if (!gd)
3689 		goto out_free;
3690 
3691 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3692 	if (index < 0) {
3693 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3694 		goto out_put;
3695 	}
3696 
3697 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3698 	if (error) {
3699 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3700 		goto out_free_index;
3701 	}
3702 
3703 	sdkp->device = sdp;
3704 	sdkp->disk = gd;
3705 	sdkp->index = index;
3706 	sdkp->max_retries = SD_MAX_RETRIES;
3707 	atomic_set(&sdkp->openers, 0);
3708 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3709 
3710 	if (!sdp->request_queue->rq_timeout) {
3711 		if (sdp->type != TYPE_MOD)
3712 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3713 		else
3714 			blk_queue_rq_timeout(sdp->request_queue,
3715 					     SD_MOD_TIMEOUT);
3716 	}
3717 
3718 	device_initialize(&sdkp->disk_dev);
3719 	sdkp->disk_dev.parent = get_device(dev);
3720 	sdkp->disk_dev.class = &sd_disk_class;
3721 	dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3722 
3723 	error = device_add(&sdkp->disk_dev);
3724 	if (error) {
3725 		put_device(&sdkp->disk_dev);
3726 		goto out;
3727 	}
3728 
3729 	dev_set_drvdata(dev, sdkp);
3730 
3731 	gd->major = sd_major((index & 0xf0) >> 4);
3732 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3733 	gd->minors = SD_MINORS;
3734 
3735 	gd->fops = &sd_fops;
3736 	gd->private_data = sdkp;
3737 
3738 	/* defaults, until the device tells us otherwise */
3739 	sdp->sector_size = 512;
3740 	sdkp->capacity = 0;
3741 	sdkp->media_present = 1;
3742 	sdkp->write_prot = 0;
3743 	sdkp->cache_override = 0;
3744 	sdkp->WCE = 0;
3745 	sdkp->RCD = 0;
3746 	sdkp->ATO = 0;
3747 	sdkp->first_scan = 1;
3748 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3749 
3750 	sd_revalidate_disk(gd);
3751 
3752 	if (sdp->removable) {
3753 		gd->flags |= GENHD_FL_REMOVABLE;
3754 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3755 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3756 	}
3757 
3758 	blk_pm_runtime_init(sdp->request_queue, dev);
3759 	if (sdp->rpm_autosuspend) {
3760 		pm_runtime_set_autosuspend_delay(dev,
3761 			sdp->host->hostt->rpm_autosuspend_delay);
3762 	}
3763 
3764 	error = device_add_disk(dev, gd, NULL);
3765 	if (error) {
3766 		device_unregister(&sdkp->disk_dev);
3767 		put_disk(gd);
3768 		goto out;
3769 	}
3770 
3771 	if (sdkp->security) {
3772 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3773 		if (sdkp->opal_dev)
3774 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3775 	}
3776 
3777 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3778 		  sdp->removable ? "removable " : "");
3779 	scsi_autopm_put_device(sdp);
3780 
3781 	return 0;
3782 
3783  out_free_index:
3784 	ida_free(&sd_index_ida, index);
3785  out_put:
3786 	put_disk(gd);
3787  out_free:
3788 	kfree(sdkp);
3789  out:
3790 	scsi_autopm_put_device(sdp);
3791 	return error;
3792 }
3793 
3794 /**
3795  *	sd_remove - called whenever a scsi disk (previously recognized by
3796  *	sd_probe) is detached from the system. It is called (potentially
3797  *	multiple times) during sd module unload.
3798  *	@dev: pointer to device object
3799  *
3800  *	Note: this function is invoked from the scsi mid-level.
3801  *	This function potentially frees up a device name (e.g. /dev/sdc)
3802  *	that could be re-used by a subsequent sd_probe().
3803  *	This function is not called when the built-in sd driver is "exit-ed".
3804  **/
3805 static int sd_remove(struct device *dev)
3806 {
3807 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3808 
3809 	scsi_autopm_get_device(sdkp->device);
3810 
3811 	device_del(&sdkp->disk_dev);
3812 	del_gendisk(sdkp->disk);
3813 	if (!sdkp->suspended)
3814 		sd_shutdown(dev);
3815 
3816 	put_disk(sdkp->disk);
3817 	return 0;
3818 }
3819 
3820 static void scsi_disk_release(struct device *dev)
3821 {
3822 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3823 
3824 	ida_free(&sd_index_ida, sdkp->index);
3825 	sd_zbc_free_zone_info(sdkp);
3826 	put_device(&sdkp->device->sdev_gendev);
3827 	free_opal_dev(sdkp->opal_dev);
3828 
3829 	kfree(sdkp);
3830 }
3831 
3832 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3833 {
3834 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3835 	struct scsi_sense_hdr sshdr;
3836 	const struct scsi_exec_args exec_args = {
3837 		.sshdr = &sshdr,
3838 		.req_flags = BLK_MQ_REQ_PM,
3839 	};
3840 	struct scsi_device *sdp = sdkp->device;
3841 	int res;
3842 
3843 	if (start)
3844 		cmd[4] |= 1;	/* START */
3845 
3846 	if (sdp->start_stop_pwr_cond)
3847 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3848 
3849 	if (!scsi_device_online(sdp))
3850 		return -ENODEV;
3851 
3852 	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
3853 			       sdkp->max_retries, &exec_args);
3854 	if (res) {
3855 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3856 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3857 			sd_print_sense_hdr(sdkp, &sshdr);
3858 			/* 0x3a is medium not present */
3859 			if (sshdr.asc == 0x3a)
3860 				res = 0;
3861 		}
3862 	}
3863 
3864 	/* SCSI error codes must not go to the generic layer */
3865 	if (res)
3866 		return -EIO;
3867 
3868 	return 0;
3869 }
3870 
3871 /*
3872  * Send a SYNCHRONIZE CACHE instruction down to the device through
3873  * the normal SCSI command structure.  Wait for the command to
3874  * complete.
3875  */
3876 static void sd_shutdown(struct device *dev)
3877 {
3878 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3879 
3880 	if (!sdkp)
3881 		return;         /* this can happen */
3882 
3883 	if (pm_runtime_suspended(dev))
3884 		return;
3885 
3886 	if (sdkp->WCE && sdkp->media_present) {
3887 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3888 		sd_sync_cache(sdkp);
3889 	}
3890 
3891 	if ((system_state != SYSTEM_RESTART &&
3892 	     sdkp->device->manage_system_start_stop) ||
3893 	    (system_state == SYSTEM_POWER_OFF &&
3894 	     sdkp->device->manage_shutdown)) {
3895 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3896 		sd_start_stop_device(sdkp, 0);
3897 	}
3898 }
3899 
3900 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
3901 {
3902 	return (sdev->manage_system_start_stop && !runtime) ||
3903 		(sdev->manage_runtime_start_stop && runtime);
3904 }
3905 
3906 static int sd_suspend_common(struct device *dev, bool runtime)
3907 {
3908 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3909 	int ret = 0;
3910 
3911 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3912 		return 0;
3913 
3914 	if (sdkp->WCE && sdkp->media_present) {
3915 		if (!sdkp->device->silence_suspend)
3916 			sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3917 		ret = sd_sync_cache(sdkp);
3918 		/* ignore OFFLINE device */
3919 		if (ret == -ENODEV)
3920 			return 0;
3921 
3922 		if (ret)
3923 			return ret;
3924 	}
3925 
3926 	if (sd_do_start_stop(sdkp->device, runtime)) {
3927 		if (!sdkp->device->silence_suspend)
3928 			sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3929 		/* an error is not worth aborting a system sleep */
3930 		ret = sd_start_stop_device(sdkp, 0);
3931 		if (!runtime)
3932 			ret = 0;
3933 	}
3934 
3935 	if (!ret)
3936 		sdkp->suspended = true;
3937 
3938 	return ret;
3939 }
3940 
3941 static int sd_suspend_system(struct device *dev)
3942 {
3943 	if (pm_runtime_suspended(dev))
3944 		return 0;
3945 
3946 	return sd_suspend_common(dev, false);
3947 }
3948 
3949 static int sd_suspend_runtime(struct device *dev)
3950 {
3951 	return sd_suspend_common(dev, true);
3952 }
3953 
3954 static int sd_resume(struct device *dev)
3955 {
3956 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3957 
3958 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3959 
3960 	if (opal_unlock_from_suspend(sdkp->opal_dev)) {
3961 		sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
3962 		return -EIO;
3963 	}
3964 
3965 	return 0;
3966 }
3967 
3968 static int sd_resume_common(struct device *dev, bool runtime)
3969 {
3970 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3971 	int ret = 0;
3972 
3973 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3974 		return 0;
3975 
3976 	if (!sd_do_start_stop(sdkp->device, runtime)) {
3977 		sdkp->suspended = false;
3978 		return 0;
3979 	}
3980 
3981 	if (!sdkp->device->no_start_on_resume) {
3982 		sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3983 		ret = sd_start_stop_device(sdkp, 1);
3984 	}
3985 
3986 	if (!ret) {
3987 		sd_resume(dev);
3988 		sdkp->suspended = false;
3989 	}
3990 
3991 	return ret;
3992 }
3993 
3994 static int sd_resume_system(struct device *dev)
3995 {
3996 	if (pm_runtime_suspended(dev)) {
3997 		struct scsi_disk *sdkp = dev_get_drvdata(dev);
3998 		struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
3999 
4000 		if (sdp && sdp->force_runtime_start_on_system_start)
4001 			pm_request_resume(dev);
4002 
4003 		return 0;
4004 	}
4005 
4006 	return sd_resume_common(dev, false);
4007 }
4008 
4009 static int sd_resume_runtime(struct device *dev)
4010 {
4011 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4012 	struct scsi_device *sdp;
4013 
4014 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
4015 		return 0;
4016 
4017 	sdp = sdkp->device;
4018 
4019 	if (sdp->ignore_media_change) {
4020 		/* clear the device's sense data */
4021 		static const u8 cmd[10] = { REQUEST_SENSE };
4022 		const struct scsi_exec_args exec_args = {
4023 			.req_flags = BLK_MQ_REQ_PM,
4024 		};
4025 
4026 		if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4027 				     sdp->request_queue->rq_timeout, 1,
4028 				     &exec_args))
4029 			sd_printk(KERN_NOTICE, sdkp,
4030 				  "Failed to clear sense data\n");
4031 	}
4032 
4033 	return sd_resume_common(dev, true);
4034 }
4035 
4036 static const struct dev_pm_ops sd_pm_ops = {
4037 	.suspend		= sd_suspend_system,
4038 	.resume			= sd_resume_system,
4039 	.poweroff		= sd_suspend_system,
4040 	.restore		= sd_resume_system,
4041 	.runtime_suspend	= sd_suspend_runtime,
4042 	.runtime_resume		= sd_resume_runtime,
4043 };
4044 
4045 static struct scsi_driver sd_template = {
4046 	.gendrv = {
4047 		.name		= "sd",
4048 		.owner		= THIS_MODULE,
4049 		.probe		= sd_probe,
4050 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
4051 		.remove		= sd_remove,
4052 		.shutdown	= sd_shutdown,
4053 		.pm		= &sd_pm_ops,
4054 	},
4055 	.rescan			= sd_rescan,
4056 	.resume			= sd_resume,
4057 	.init_command		= sd_init_command,
4058 	.uninit_command		= sd_uninit_command,
4059 	.done			= sd_done,
4060 	.eh_action		= sd_eh_action,
4061 	.eh_reset		= sd_eh_reset,
4062 };
4063 
4064 /**
4065  *	init_sd - entry point for this driver (both when built in or when
4066  *	a module).
4067  *
4068  *	Note: this function registers this driver with the scsi mid-level.
4069  **/
4070 static int __init init_sd(void)
4071 {
4072 	int majors = 0, i, err;
4073 
4074 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4075 
4076 	for (i = 0; i < SD_MAJORS; i++) {
4077 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4078 			continue;
4079 		majors++;
4080 	}
4081 
4082 	if (!majors)
4083 		return -ENODEV;
4084 
4085 	err = class_register(&sd_disk_class);
4086 	if (err)
4087 		goto err_out;
4088 
4089 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4090 	if (!sd_page_pool) {
4091 		printk(KERN_ERR "sd: can't init discard page pool\n");
4092 		err = -ENOMEM;
4093 		goto err_out_class;
4094 	}
4095 
4096 	err = scsi_register_driver(&sd_template.gendrv);
4097 	if (err)
4098 		goto err_out_driver;
4099 
4100 	return 0;
4101 
4102 err_out_driver:
4103 	mempool_destroy(sd_page_pool);
4104 err_out_class:
4105 	class_unregister(&sd_disk_class);
4106 err_out:
4107 	for (i = 0; i < SD_MAJORS; i++)
4108 		unregister_blkdev(sd_major(i), "sd");
4109 	return err;
4110 }
4111 
4112 /**
4113  *	exit_sd - exit point for this driver (when it is a module).
4114  *
4115  *	Note: this function unregisters this driver from the scsi mid-level.
4116  **/
4117 static void __exit exit_sd(void)
4118 {
4119 	int i;
4120 
4121 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4122 
4123 	scsi_unregister_driver(&sd_template.gendrv);
4124 	mempool_destroy(sd_page_pool);
4125 
4126 	class_unregister(&sd_disk_class);
4127 
4128 	for (i = 0; i < SD_MAJORS; i++)
4129 		unregister_blkdev(sd_major(i), "sd");
4130 }
4131 
4132 module_init(init_sd);
4133 module_exit(exit_sd);
4134 
4135 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4136 {
4137 	scsi_print_sense_hdr(sdkp->device,
4138 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4139 }
4140 
4141 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4142 {
4143 	const char *hb_string = scsi_hostbyte_string(result);
4144 
4145 	if (hb_string)
4146 		sd_printk(KERN_INFO, sdkp,
4147 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4148 			  hb_string ? hb_string : "invalid",
4149 			  "DRIVER_OK");
4150 	else
4151 		sd_printk(KERN_INFO, sdkp,
4152 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4153 			  msg, host_byte(result), "DRIVER_OK");
4154 }
4155