xref: /openbmc/linux/drivers/scsi/sd.c (revision c4c3c32d)
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 int  sd_probe(struct device *);
108 static int  sd_remove(struct device *);
109 static void sd_shutdown(struct device *);
110 static int sd_suspend_system(struct device *);
111 static int sd_suspend_runtime(struct device *);
112 static int sd_resume_system(struct device *);
113 static int sd_resume_runtime(struct device *);
114 static void sd_rescan(struct device *);
115 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
116 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117 static int sd_done(struct scsi_cmnd *);
118 static void sd_eh_reset(struct scsi_cmnd *);
119 static int sd_eh_action(struct scsi_cmnd *, int);
120 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
121 static void scsi_disk_release(struct device *cdev);
122 
123 static DEFINE_IDA(sd_index_ida);
124 
125 static mempool_t *sd_page_pool;
126 static struct lock_class_key sd_bio_compl_lkclass;
127 
128 static const char *sd_cache_types[] = {
129 	"write through", "none", "write back",
130 	"write back, no read (daft)"
131 };
132 
133 static void sd_set_flush_flag(struct scsi_disk *sdkp)
134 {
135 	bool wc = false, fua = false;
136 
137 	if (sdkp->WCE) {
138 		wc = true;
139 		if (sdkp->DPOFUA)
140 			fua = true;
141 	}
142 
143 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
144 }
145 
146 static ssize_t
147 cache_type_store(struct device *dev, struct device_attribute *attr,
148 		 const char *buf, size_t count)
149 {
150 	int ct, rcd, wce, sp;
151 	struct scsi_disk *sdkp = to_scsi_disk(dev);
152 	struct scsi_device *sdp = sdkp->device;
153 	char buffer[64];
154 	char *buffer_data;
155 	struct scsi_mode_data data;
156 	struct scsi_sense_hdr sshdr;
157 	static const char temp[] = "temporary ";
158 	int len;
159 
160 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
161 		/* no cache control on RBC devices; theoretically they
162 		 * can do it, but there's probably so many exceptions
163 		 * it's not worth the risk */
164 		return -EINVAL;
165 
166 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
167 		buf += sizeof(temp) - 1;
168 		sdkp->cache_override = 1;
169 	} else {
170 		sdkp->cache_override = 0;
171 	}
172 
173 	ct = sysfs_match_string(sd_cache_types, buf);
174 	if (ct < 0)
175 		return -EINVAL;
176 
177 	rcd = ct & 0x01 ? 1 : 0;
178 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
179 
180 	if (sdkp->cache_override) {
181 		sdkp->WCE = wce;
182 		sdkp->RCD = rcd;
183 		sd_set_flush_flag(sdkp);
184 		return count;
185 	}
186 
187 	if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
188 			    sdkp->max_retries, &data, NULL))
189 		return -EINVAL;
190 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
191 		  data.block_descriptor_length);
192 	buffer_data = buffer + data.header_length +
193 		data.block_descriptor_length;
194 	buffer_data[2] &= ~0x05;
195 	buffer_data[2] |= wce << 2 | rcd;
196 	sp = buffer_data[0] & 0x80 ? 1 : 0;
197 	buffer_data[0] &= ~0x80;
198 
199 	/*
200 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
201 	 * received mode parameter buffer before doing MODE SELECT.
202 	 */
203 	data.device_specific = 0;
204 
205 	if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
206 			     sdkp->max_retries, &data, &sshdr)) {
207 		if (scsi_sense_valid(&sshdr))
208 			sd_print_sense_hdr(sdkp, &sshdr);
209 		return -EINVAL;
210 	}
211 	sd_revalidate_disk(sdkp->disk);
212 	return count;
213 }
214 
215 static ssize_t
216 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
217 		       char *buf)
218 {
219 	struct scsi_disk *sdkp = to_scsi_disk(dev);
220 	struct scsi_device *sdp = sdkp->device;
221 
222 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
223 }
224 
225 static ssize_t
226 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
227 			const char *buf, size_t count)
228 {
229 	struct scsi_disk *sdkp = to_scsi_disk(dev);
230 	struct scsi_device *sdp = sdkp->device;
231 	bool v;
232 
233 	if (!capable(CAP_SYS_ADMIN))
234 		return -EACCES;
235 
236 	if (kstrtobool(buf, &v))
237 		return -EINVAL;
238 
239 	sdp->manage_start_stop = v;
240 
241 	return count;
242 }
243 static DEVICE_ATTR_RW(manage_start_stop);
244 
245 static ssize_t
246 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
247 {
248 	struct scsi_disk *sdkp = to_scsi_disk(dev);
249 
250 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
251 }
252 
253 static ssize_t
254 allow_restart_store(struct device *dev, struct device_attribute *attr,
255 		    const char *buf, size_t count)
256 {
257 	bool v;
258 	struct scsi_disk *sdkp = to_scsi_disk(dev);
259 	struct scsi_device *sdp = sdkp->device;
260 
261 	if (!capable(CAP_SYS_ADMIN))
262 		return -EACCES;
263 
264 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
265 		return -EINVAL;
266 
267 	if (kstrtobool(buf, &v))
268 		return -EINVAL;
269 
270 	sdp->allow_restart = v;
271 
272 	return count;
273 }
274 static DEVICE_ATTR_RW(allow_restart);
275 
276 static ssize_t
277 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
278 {
279 	struct scsi_disk *sdkp = to_scsi_disk(dev);
280 	int ct = sdkp->RCD + 2*sdkp->WCE;
281 
282 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
283 }
284 static DEVICE_ATTR_RW(cache_type);
285 
286 static ssize_t
287 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
288 {
289 	struct scsi_disk *sdkp = to_scsi_disk(dev);
290 
291 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
292 }
293 static DEVICE_ATTR_RO(FUA);
294 
295 static ssize_t
296 protection_type_show(struct device *dev, struct device_attribute *attr,
297 		     char *buf)
298 {
299 	struct scsi_disk *sdkp = to_scsi_disk(dev);
300 
301 	return sprintf(buf, "%u\n", sdkp->protection_type);
302 }
303 
304 static ssize_t
305 protection_type_store(struct device *dev, struct device_attribute *attr,
306 		      const char *buf, size_t count)
307 {
308 	struct scsi_disk *sdkp = to_scsi_disk(dev);
309 	unsigned int val;
310 	int err;
311 
312 	if (!capable(CAP_SYS_ADMIN))
313 		return -EACCES;
314 
315 	err = kstrtouint(buf, 10, &val);
316 
317 	if (err)
318 		return err;
319 
320 	if (val <= T10_PI_TYPE3_PROTECTION)
321 		sdkp->protection_type = val;
322 
323 	return count;
324 }
325 static DEVICE_ATTR_RW(protection_type);
326 
327 static ssize_t
328 protection_mode_show(struct device *dev, struct device_attribute *attr,
329 		     char *buf)
330 {
331 	struct scsi_disk *sdkp = to_scsi_disk(dev);
332 	struct scsi_device *sdp = sdkp->device;
333 	unsigned int dif, dix;
334 
335 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
336 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
337 
338 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
339 		dif = 0;
340 		dix = 1;
341 	}
342 
343 	if (!dif && !dix)
344 		return sprintf(buf, "none\n");
345 
346 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
347 }
348 static DEVICE_ATTR_RO(protection_mode);
349 
350 static ssize_t
351 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
352 {
353 	struct scsi_disk *sdkp = to_scsi_disk(dev);
354 
355 	return sprintf(buf, "%u\n", sdkp->ATO);
356 }
357 static DEVICE_ATTR_RO(app_tag_own);
358 
359 static ssize_t
360 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
361 		       char *buf)
362 {
363 	struct scsi_disk *sdkp = to_scsi_disk(dev);
364 
365 	return sprintf(buf, "%u\n", sdkp->lbpme);
366 }
367 static DEVICE_ATTR_RO(thin_provisioning);
368 
369 /* sysfs_match_string() requires dense arrays */
370 static const char *lbp_mode[] = {
371 	[SD_LBP_FULL]		= "full",
372 	[SD_LBP_UNMAP]		= "unmap",
373 	[SD_LBP_WS16]		= "writesame_16",
374 	[SD_LBP_WS10]		= "writesame_10",
375 	[SD_LBP_ZERO]		= "writesame_zero",
376 	[SD_LBP_DISABLE]	= "disabled",
377 };
378 
379 static ssize_t
380 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
381 		       char *buf)
382 {
383 	struct scsi_disk *sdkp = to_scsi_disk(dev);
384 
385 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
386 }
387 
388 static ssize_t
389 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
390 			const char *buf, size_t count)
391 {
392 	struct scsi_disk *sdkp = to_scsi_disk(dev);
393 	struct scsi_device *sdp = sdkp->device;
394 	int mode;
395 
396 	if (!capable(CAP_SYS_ADMIN))
397 		return -EACCES;
398 
399 	if (sd_is_zoned(sdkp)) {
400 		sd_config_discard(sdkp, SD_LBP_DISABLE);
401 		return count;
402 	}
403 
404 	if (sdp->type != TYPE_DISK)
405 		return -EINVAL;
406 
407 	mode = sysfs_match_string(lbp_mode, buf);
408 	if (mode < 0)
409 		return -EINVAL;
410 
411 	sd_config_discard(sdkp, mode);
412 
413 	return count;
414 }
415 static DEVICE_ATTR_RW(provisioning_mode);
416 
417 /* sysfs_match_string() requires dense arrays */
418 static const char *zeroing_mode[] = {
419 	[SD_ZERO_WRITE]		= "write",
420 	[SD_ZERO_WS]		= "writesame",
421 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
422 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
423 };
424 
425 static ssize_t
426 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
427 		  char *buf)
428 {
429 	struct scsi_disk *sdkp = to_scsi_disk(dev);
430 
431 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
432 }
433 
434 static ssize_t
435 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
436 		   const char *buf, size_t count)
437 {
438 	struct scsi_disk *sdkp = to_scsi_disk(dev);
439 	int mode;
440 
441 	if (!capable(CAP_SYS_ADMIN))
442 		return -EACCES;
443 
444 	mode = sysfs_match_string(zeroing_mode, buf);
445 	if (mode < 0)
446 		return -EINVAL;
447 
448 	sdkp->zeroing_mode = mode;
449 
450 	return count;
451 }
452 static DEVICE_ATTR_RW(zeroing_mode);
453 
454 static ssize_t
455 max_medium_access_timeouts_show(struct device *dev,
456 				struct device_attribute *attr, char *buf)
457 {
458 	struct scsi_disk *sdkp = to_scsi_disk(dev);
459 
460 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
461 }
462 
463 static ssize_t
464 max_medium_access_timeouts_store(struct device *dev,
465 				 struct device_attribute *attr, const char *buf,
466 				 size_t count)
467 {
468 	struct scsi_disk *sdkp = to_scsi_disk(dev);
469 	int err;
470 
471 	if (!capable(CAP_SYS_ADMIN))
472 		return -EACCES;
473 
474 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
475 
476 	return err ? err : count;
477 }
478 static DEVICE_ATTR_RW(max_medium_access_timeouts);
479 
480 static ssize_t
481 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
482 			   char *buf)
483 {
484 	struct scsi_disk *sdkp = to_scsi_disk(dev);
485 
486 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
487 }
488 
489 static ssize_t
490 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
491 			    const char *buf, size_t count)
492 {
493 	struct scsi_disk *sdkp = to_scsi_disk(dev);
494 	struct scsi_device *sdp = sdkp->device;
495 	unsigned long max;
496 	int err;
497 
498 	if (!capable(CAP_SYS_ADMIN))
499 		return -EACCES;
500 
501 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
502 		return -EINVAL;
503 
504 	err = kstrtoul(buf, 10, &max);
505 
506 	if (err)
507 		return err;
508 
509 	if (max == 0)
510 		sdp->no_write_same = 1;
511 	else if (max <= SD_MAX_WS16_BLOCKS) {
512 		sdp->no_write_same = 0;
513 		sdkp->max_ws_blocks = max;
514 	}
515 
516 	sd_config_write_same(sdkp);
517 
518 	return count;
519 }
520 static DEVICE_ATTR_RW(max_write_same_blocks);
521 
522 static ssize_t
523 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
524 {
525 	struct scsi_disk *sdkp = to_scsi_disk(dev);
526 
527 	if (sdkp->device->type == TYPE_ZBC)
528 		return sprintf(buf, "host-managed\n");
529 	if (sdkp->zoned == 1)
530 		return sprintf(buf, "host-aware\n");
531 	if (sdkp->zoned == 2)
532 		return sprintf(buf, "drive-managed\n");
533 	return sprintf(buf, "none\n");
534 }
535 static DEVICE_ATTR_RO(zoned_cap);
536 
537 static ssize_t
538 max_retries_store(struct device *dev, struct device_attribute *attr,
539 		  const char *buf, size_t count)
540 {
541 	struct scsi_disk *sdkp = to_scsi_disk(dev);
542 	struct scsi_device *sdev = sdkp->device;
543 	int retries, err;
544 
545 	err = kstrtoint(buf, 10, &retries);
546 	if (err)
547 		return err;
548 
549 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
550 		sdkp->max_retries = retries;
551 		return count;
552 	}
553 
554 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
555 		    SD_MAX_RETRIES);
556 	return -EINVAL;
557 }
558 
559 static ssize_t
560 max_retries_show(struct device *dev, struct device_attribute *attr,
561 		 char *buf)
562 {
563 	struct scsi_disk *sdkp = to_scsi_disk(dev);
564 
565 	return sprintf(buf, "%d\n", sdkp->max_retries);
566 }
567 
568 static DEVICE_ATTR_RW(max_retries);
569 
570 static struct attribute *sd_disk_attrs[] = {
571 	&dev_attr_cache_type.attr,
572 	&dev_attr_FUA.attr,
573 	&dev_attr_allow_restart.attr,
574 	&dev_attr_manage_start_stop.attr,
575 	&dev_attr_protection_type.attr,
576 	&dev_attr_protection_mode.attr,
577 	&dev_attr_app_tag_own.attr,
578 	&dev_attr_thin_provisioning.attr,
579 	&dev_attr_provisioning_mode.attr,
580 	&dev_attr_zeroing_mode.attr,
581 	&dev_attr_max_write_same_blocks.attr,
582 	&dev_attr_max_medium_access_timeouts.attr,
583 	&dev_attr_zoned_cap.attr,
584 	&dev_attr_max_retries.attr,
585 	NULL,
586 };
587 ATTRIBUTE_GROUPS(sd_disk);
588 
589 static struct class sd_disk_class = {
590 	.name		= "scsi_disk",
591 	.dev_release	= scsi_disk_release,
592 	.dev_groups	= sd_disk_groups,
593 };
594 
595 static const struct dev_pm_ops sd_pm_ops = {
596 	.suspend		= sd_suspend_system,
597 	.resume			= sd_resume_system,
598 	.poweroff		= sd_suspend_system,
599 	.restore		= sd_resume_system,
600 	.runtime_suspend	= sd_suspend_runtime,
601 	.runtime_resume		= sd_resume_runtime,
602 };
603 
604 static struct scsi_driver sd_template = {
605 	.gendrv = {
606 		.name		= "sd",
607 		.owner		= THIS_MODULE,
608 		.probe		= sd_probe,
609 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
610 		.remove		= sd_remove,
611 		.shutdown	= sd_shutdown,
612 		.pm		= &sd_pm_ops,
613 	},
614 	.rescan			= sd_rescan,
615 	.init_command		= sd_init_command,
616 	.uninit_command		= sd_uninit_command,
617 	.done			= sd_done,
618 	.eh_action		= sd_eh_action,
619 	.eh_reset		= sd_eh_reset,
620 };
621 
622 /*
623  * Don't request a new module, as that could deadlock in multipath
624  * environment.
625  */
626 static void sd_default_probe(dev_t devt)
627 {
628 }
629 
630 /*
631  * Device no to disk mapping:
632  *
633  *       major         disc2     disc  p1
634  *   |............|.............|....|....| <- dev_t
635  *    31        20 19          8 7  4 3  0
636  *
637  * Inside a major, we have 16k disks, however mapped non-
638  * contiguously. The first 16 disks are for major0, the next
639  * ones with major1, ... Disk 256 is for major0 again, disk 272
640  * for major1, ...
641  * As we stay compatible with our numbering scheme, we can reuse
642  * the well-know SCSI majors 8, 65--71, 136--143.
643  */
644 static int sd_major(int major_idx)
645 {
646 	switch (major_idx) {
647 	case 0:
648 		return SCSI_DISK0_MAJOR;
649 	case 1 ... 7:
650 		return SCSI_DISK1_MAJOR + major_idx - 1;
651 	case 8 ... 15:
652 		return SCSI_DISK8_MAJOR + major_idx - 8;
653 	default:
654 		BUG();
655 		return 0;	/* shut up gcc */
656 	}
657 }
658 
659 #ifdef CONFIG_BLK_SED_OPAL
660 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
661 		size_t len, bool send)
662 {
663 	struct scsi_disk *sdkp = data;
664 	struct scsi_device *sdev = sdkp->device;
665 	u8 cdb[12] = { 0, };
666 	const struct scsi_exec_args exec_args = {
667 		.req_flags = BLK_MQ_REQ_PM,
668 	};
669 	int ret;
670 
671 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
672 	cdb[1] = secp;
673 	put_unaligned_be16(spsp, &cdb[2]);
674 	put_unaligned_be32(len, &cdb[6]);
675 
676 	ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
677 			       buffer, len, SD_TIMEOUT, sdkp->max_retries,
678 			       &exec_args);
679 	return ret <= 0 ? ret : -EIO;
680 }
681 #endif /* CONFIG_BLK_SED_OPAL */
682 
683 /*
684  * Look up the DIX operation based on whether the command is read or
685  * write and whether dix and dif are enabled.
686  */
687 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
688 {
689 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
690 	static const unsigned int ops[] = {	/* wrt dix dif */
691 		SCSI_PROT_NORMAL,		/*  0	0   0  */
692 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
693 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
694 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
695 		SCSI_PROT_NORMAL,		/*  1	0   0  */
696 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
697 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
698 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
699 	};
700 
701 	return ops[write << 2 | dix << 1 | dif];
702 }
703 
704 /*
705  * Returns a mask of the protection flags that are valid for a given DIX
706  * operation.
707  */
708 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
709 {
710 	static const unsigned int flag_mask[] = {
711 		[SCSI_PROT_NORMAL]		= 0,
712 
713 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
714 						  SCSI_PROT_GUARD_CHECK |
715 						  SCSI_PROT_REF_CHECK |
716 						  SCSI_PROT_REF_INCREMENT,
717 
718 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
719 						  SCSI_PROT_IP_CHECKSUM,
720 
721 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
722 						  SCSI_PROT_GUARD_CHECK |
723 						  SCSI_PROT_REF_CHECK |
724 						  SCSI_PROT_REF_INCREMENT |
725 						  SCSI_PROT_IP_CHECKSUM,
726 
727 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
728 						  SCSI_PROT_REF_INCREMENT,
729 
730 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
731 						  SCSI_PROT_REF_CHECK |
732 						  SCSI_PROT_REF_INCREMENT |
733 						  SCSI_PROT_IP_CHECKSUM,
734 
735 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
736 						  SCSI_PROT_GUARD_CHECK |
737 						  SCSI_PROT_REF_CHECK |
738 						  SCSI_PROT_REF_INCREMENT |
739 						  SCSI_PROT_IP_CHECKSUM,
740 	};
741 
742 	return flag_mask[prot_op];
743 }
744 
745 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
746 					   unsigned int dix, unsigned int dif)
747 {
748 	struct request *rq = scsi_cmd_to_rq(scmd);
749 	struct bio *bio = rq->bio;
750 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
751 	unsigned int protect = 0;
752 
753 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
754 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
755 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
756 
757 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
758 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
759 	}
760 
761 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
762 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
763 
764 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
765 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
766 	}
767 
768 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
769 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
770 
771 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
772 			protect = 3 << 5;	/* Disable target PI checking */
773 		else
774 			protect = 1 << 5;	/* Enable target PI checking */
775 	}
776 
777 	scsi_set_prot_op(scmd, prot_op);
778 	scsi_set_prot_type(scmd, dif);
779 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
780 
781 	return protect;
782 }
783 
784 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
785 {
786 	struct request_queue *q = sdkp->disk->queue;
787 	unsigned int logical_block_size = sdkp->device->sector_size;
788 	unsigned int max_blocks = 0;
789 
790 	q->limits.discard_alignment =
791 		sdkp->unmap_alignment * logical_block_size;
792 	q->limits.discard_granularity =
793 		max(sdkp->physical_block_size,
794 		    sdkp->unmap_granularity * logical_block_size);
795 	sdkp->provisioning_mode = mode;
796 
797 	switch (mode) {
798 
799 	case SD_LBP_FULL:
800 	case SD_LBP_DISABLE:
801 		blk_queue_max_discard_sectors(q, 0);
802 		return;
803 
804 	case SD_LBP_UNMAP:
805 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
806 					  (u32)SD_MAX_WS16_BLOCKS);
807 		break;
808 
809 	case SD_LBP_WS16:
810 		if (sdkp->device->unmap_limit_for_ws)
811 			max_blocks = sdkp->max_unmap_blocks;
812 		else
813 			max_blocks = sdkp->max_ws_blocks;
814 
815 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
816 		break;
817 
818 	case SD_LBP_WS10:
819 		if (sdkp->device->unmap_limit_for_ws)
820 			max_blocks = sdkp->max_unmap_blocks;
821 		else
822 			max_blocks = sdkp->max_ws_blocks;
823 
824 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
825 		break;
826 
827 	case SD_LBP_ZERO:
828 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
829 					  (u32)SD_MAX_WS10_BLOCKS);
830 		break;
831 	}
832 
833 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
834 }
835 
836 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
837 {
838 	struct page *page;
839 
840 	page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
841 	if (!page)
842 		return NULL;
843 	clear_highpage(page);
844 	bvec_set_page(&rq->special_vec, page, data_len, 0);
845 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
846 	return bvec_virt(&rq->special_vec);
847 }
848 
849 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
850 {
851 	struct scsi_device *sdp = cmd->device;
852 	struct request *rq = scsi_cmd_to_rq(cmd);
853 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
854 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
855 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
856 	unsigned int data_len = 24;
857 	char *buf;
858 
859 	buf = sd_set_special_bvec(rq, data_len);
860 	if (!buf)
861 		return BLK_STS_RESOURCE;
862 
863 	cmd->cmd_len = 10;
864 	cmd->cmnd[0] = UNMAP;
865 	cmd->cmnd[8] = 24;
866 
867 	put_unaligned_be16(6 + 16, &buf[0]);
868 	put_unaligned_be16(16, &buf[2]);
869 	put_unaligned_be64(lba, &buf[8]);
870 	put_unaligned_be32(nr_blocks, &buf[16]);
871 
872 	cmd->allowed = sdkp->max_retries;
873 	cmd->transfersize = data_len;
874 	rq->timeout = SD_TIMEOUT;
875 
876 	return scsi_alloc_sgtables(cmd);
877 }
878 
879 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
880 		bool unmap)
881 {
882 	struct scsi_device *sdp = cmd->device;
883 	struct request *rq = scsi_cmd_to_rq(cmd);
884 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
885 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
886 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
887 	u32 data_len = sdp->sector_size;
888 
889 	if (!sd_set_special_bvec(rq, data_len))
890 		return BLK_STS_RESOURCE;
891 
892 	cmd->cmd_len = 16;
893 	cmd->cmnd[0] = WRITE_SAME_16;
894 	if (unmap)
895 		cmd->cmnd[1] = 0x8; /* UNMAP */
896 	put_unaligned_be64(lba, &cmd->cmnd[2]);
897 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
898 
899 	cmd->allowed = sdkp->max_retries;
900 	cmd->transfersize = data_len;
901 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
902 
903 	return scsi_alloc_sgtables(cmd);
904 }
905 
906 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
907 		bool unmap)
908 {
909 	struct scsi_device *sdp = cmd->device;
910 	struct request *rq = scsi_cmd_to_rq(cmd);
911 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
912 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
913 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
914 	u32 data_len = sdp->sector_size;
915 
916 	if (!sd_set_special_bvec(rq, data_len))
917 		return BLK_STS_RESOURCE;
918 
919 	cmd->cmd_len = 10;
920 	cmd->cmnd[0] = WRITE_SAME;
921 	if (unmap)
922 		cmd->cmnd[1] = 0x8; /* UNMAP */
923 	put_unaligned_be32(lba, &cmd->cmnd[2]);
924 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
925 
926 	cmd->allowed = sdkp->max_retries;
927 	cmd->transfersize = data_len;
928 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
929 
930 	return scsi_alloc_sgtables(cmd);
931 }
932 
933 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
934 {
935 	struct request *rq = scsi_cmd_to_rq(cmd);
936 	struct scsi_device *sdp = cmd->device;
937 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
938 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
939 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
940 
941 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
942 		switch (sdkp->zeroing_mode) {
943 		case SD_ZERO_WS16_UNMAP:
944 			return sd_setup_write_same16_cmnd(cmd, true);
945 		case SD_ZERO_WS10_UNMAP:
946 			return sd_setup_write_same10_cmnd(cmd, true);
947 		}
948 	}
949 
950 	if (sdp->no_write_same) {
951 		rq->rq_flags |= RQF_QUIET;
952 		return BLK_STS_TARGET;
953 	}
954 
955 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
956 		return sd_setup_write_same16_cmnd(cmd, false);
957 
958 	return sd_setup_write_same10_cmnd(cmd, false);
959 }
960 
961 static void sd_config_write_same(struct scsi_disk *sdkp)
962 {
963 	struct request_queue *q = sdkp->disk->queue;
964 	unsigned int logical_block_size = sdkp->device->sector_size;
965 
966 	if (sdkp->device->no_write_same) {
967 		sdkp->max_ws_blocks = 0;
968 		goto out;
969 	}
970 
971 	/* Some devices can not handle block counts above 0xffff despite
972 	 * supporting WRITE SAME(16). Consequently we default to 64k
973 	 * blocks per I/O unless the device explicitly advertises a
974 	 * bigger limit.
975 	 */
976 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
977 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
978 						   (u32)SD_MAX_WS16_BLOCKS);
979 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
980 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
981 						   (u32)SD_MAX_WS10_BLOCKS);
982 	else {
983 		sdkp->device->no_write_same = 1;
984 		sdkp->max_ws_blocks = 0;
985 	}
986 
987 	if (sdkp->lbprz && sdkp->lbpws)
988 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
989 	else if (sdkp->lbprz && sdkp->lbpws10)
990 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
991 	else if (sdkp->max_ws_blocks)
992 		sdkp->zeroing_mode = SD_ZERO_WS;
993 	else
994 		sdkp->zeroing_mode = SD_ZERO_WRITE;
995 
996 	if (sdkp->max_ws_blocks &&
997 	    sdkp->physical_block_size > logical_block_size) {
998 		/*
999 		 * Reporting a maximum number of blocks that is not aligned
1000 		 * on the device physical size would cause a large write same
1001 		 * request to be split into physically unaligned chunks by
1002 		 * __blkdev_issue_write_zeroes() even if the caller of this
1003 		 * functions took care to align the large request. So make sure
1004 		 * the maximum reported is aligned to the device physical block
1005 		 * size. This is only an optional optimization for regular
1006 		 * disks, but this is mandatory to avoid failure of large write
1007 		 * same requests directed at sequential write required zones of
1008 		 * host-managed ZBC disks.
1009 		 */
1010 		sdkp->max_ws_blocks =
1011 			round_down(sdkp->max_ws_blocks,
1012 				   bytes_to_logical(sdkp->device,
1013 						    sdkp->physical_block_size));
1014 	}
1015 
1016 out:
1017 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1018 					 (logical_block_size >> 9));
1019 }
1020 
1021 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1022 {
1023 	struct request *rq = scsi_cmd_to_rq(cmd);
1024 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1025 
1026 	/* flush requests don't perform I/O, zero the S/G table */
1027 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1028 
1029 	if (cmd->device->use_16_for_sync) {
1030 		cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1031 		cmd->cmd_len = 16;
1032 	} else {
1033 		cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1034 		cmd->cmd_len = 10;
1035 	}
1036 	cmd->transfersize = 0;
1037 	cmd->allowed = sdkp->max_retries;
1038 
1039 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1040 	return BLK_STS_OK;
1041 }
1042 
1043 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1044 				       sector_t lba, unsigned int nr_blocks,
1045 				       unsigned char flags, unsigned int dld)
1046 {
1047 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1048 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1049 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1050 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1051 	cmd->cmnd[10] = flags;
1052 	cmd->cmnd[11] = dld & 0x07;
1053 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1054 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1055 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1056 
1057 	return BLK_STS_OK;
1058 }
1059 
1060 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1061 				       sector_t lba, unsigned int nr_blocks,
1062 				       unsigned char flags, unsigned int dld)
1063 {
1064 	cmd->cmd_len  = 16;
1065 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1066 	cmd->cmnd[1]  = flags | ((dld >> 2) & 0x01);
1067 	cmd->cmnd[14] = (dld & 0x03) << 6;
1068 	cmd->cmnd[15] = 0;
1069 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1070 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1071 
1072 	return BLK_STS_OK;
1073 }
1074 
1075 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1076 				       sector_t lba, unsigned int nr_blocks,
1077 				       unsigned char flags)
1078 {
1079 	cmd->cmd_len = 10;
1080 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1081 	cmd->cmnd[1] = flags;
1082 	cmd->cmnd[6] = 0;
1083 	cmd->cmnd[9] = 0;
1084 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1085 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1086 
1087 	return BLK_STS_OK;
1088 }
1089 
1090 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1091 				      sector_t lba, unsigned int nr_blocks,
1092 				      unsigned char flags)
1093 {
1094 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1095 	if (WARN_ON_ONCE(nr_blocks == 0))
1096 		return BLK_STS_IOERR;
1097 
1098 	if (unlikely(flags & 0x8)) {
1099 		/*
1100 		 * This happens only if this drive failed 10byte rw
1101 		 * command with ILLEGAL_REQUEST during operation and
1102 		 * thus turned off use_10_for_rw.
1103 		 */
1104 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1105 		return BLK_STS_IOERR;
1106 	}
1107 
1108 	cmd->cmd_len = 6;
1109 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1110 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1111 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1112 	cmd->cmnd[3] = lba & 0xff;
1113 	cmd->cmnd[4] = nr_blocks;
1114 	cmd->cmnd[5] = 0;
1115 
1116 	return BLK_STS_OK;
1117 }
1118 
1119 /*
1120  * Check if a command has a duration limit set. If it does, and the target
1121  * device supports CDL and the feature is enabled, return the limit
1122  * descriptor index to use. Return 0 (no limit) otherwise.
1123  */
1124 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1125 {
1126 	struct scsi_device *sdp = sdkp->device;
1127 	int hint;
1128 
1129 	if (!sdp->cdl_supported || !sdp->cdl_enable)
1130 		return 0;
1131 
1132 	/*
1133 	 * Use "no limit" if the request ioprio does not specify a duration
1134 	 * limit hint.
1135 	 */
1136 	hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1137 	if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1138 	    hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1139 		return 0;
1140 
1141 	return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1142 }
1143 
1144 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1145 {
1146 	struct request *rq = scsi_cmd_to_rq(cmd);
1147 	struct scsi_device *sdp = cmd->device;
1148 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1149 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1150 	sector_t threshold;
1151 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1152 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1153 	bool write = rq_data_dir(rq) == WRITE;
1154 	unsigned char protect, fua;
1155 	unsigned int dld;
1156 	blk_status_t ret;
1157 	unsigned int dif;
1158 	bool dix;
1159 
1160 	ret = scsi_alloc_sgtables(cmd);
1161 	if (ret != BLK_STS_OK)
1162 		return ret;
1163 
1164 	ret = BLK_STS_IOERR;
1165 	if (!scsi_device_online(sdp) || sdp->changed) {
1166 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1167 		goto fail;
1168 	}
1169 
1170 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1171 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1172 		goto fail;
1173 	}
1174 
1175 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1176 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1177 		goto fail;
1178 	}
1179 
1180 	/*
1181 	 * Some SD card readers can't handle accesses which touch the
1182 	 * last one or two logical blocks. Split accesses as needed.
1183 	 */
1184 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1185 
1186 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1187 		if (lba < threshold) {
1188 			/* Access up to the threshold but not beyond */
1189 			nr_blocks = threshold - lba;
1190 		} else {
1191 			/* Access only a single logical block */
1192 			nr_blocks = 1;
1193 		}
1194 	}
1195 
1196 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1197 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1198 		if (ret)
1199 			goto fail;
1200 	}
1201 
1202 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1203 	dix = scsi_prot_sg_count(cmd);
1204 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1205 	dld = sd_cdl_dld(sdkp, cmd);
1206 
1207 	if (dif || dix)
1208 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1209 	else
1210 		protect = 0;
1211 
1212 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1213 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1214 					 protect | fua, dld);
1215 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1216 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1217 					 protect | fua, dld);
1218 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1219 		   sdp->use_10_for_rw || protect) {
1220 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1221 					 protect | fua);
1222 	} else {
1223 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1224 					protect | fua);
1225 	}
1226 
1227 	if (unlikely(ret != BLK_STS_OK))
1228 		goto fail;
1229 
1230 	/*
1231 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1232 	 * host adapter, it's safe to assume that we can at least transfer
1233 	 * this many bytes between each connect / disconnect.
1234 	 */
1235 	cmd->transfersize = sdp->sector_size;
1236 	cmd->underflow = nr_blocks << 9;
1237 	cmd->allowed = sdkp->max_retries;
1238 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1239 
1240 	SCSI_LOG_HLQUEUE(1,
1241 			 scmd_printk(KERN_INFO, cmd,
1242 				     "%s: block=%llu, count=%d\n", __func__,
1243 				     (unsigned long long)blk_rq_pos(rq),
1244 				     blk_rq_sectors(rq)));
1245 	SCSI_LOG_HLQUEUE(2,
1246 			 scmd_printk(KERN_INFO, cmd,
1247 				     "%s %d/%u 512 byte blocks.\n",
1248 				     write ? "writing" : "reading", nr_blocks,
1249 				     blk_rq_sectors(rq)));
1250 
1251 	/*
1252 	 * This indicates that the command is ready from our end to be queued.
1253 	 */
1254 	return BLK_STS_OK;
1255 fail:
1256 	scsi_free_sgtables(cmd);
1257 	return ret;
1258 }
1259 
1260 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1261 {
1262 	struct request *rq = scsi_cmd_to_rq(cmd);
1263 
1264 	switch (req_op(rq)) {
1265 	case REQ_OP_DISCARD:
1266 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1267 		case SD_LBP_UNMAP:
1268 			return sd_setup_unmap_cmnd(cmd);
1269 		case SD_LBP_WS16:
1270 			return sd_setup_write_same16_cmnd(cmd, true);
1271 		case SD_LBP_WS10:
1272 			return sd_setup_write_same10_cmnd(cmd, true);
1273 		case SD_LBP_ZERO:
1274 			return sd_setup_write_same10_cmnd(cmd, false);
1275 		default:
1276 			return BLK_STS_TARGET;
1277 		}
1278 	case REQ_OP_WRITE_ZEROES:
1279 		return sd_setup_write_zeroes_cmnd(cmd);
1280 	case REQ_OP_FLUSH:
1281 		return sd_setup_flush_cmnd(cmd);
1282 	case REQ_OP_READ:
1283 	case REQ_OP_WRITE:
1284 	case REQ_OP_ZONE_APPEND:
1285 		return sd_setup_read_write_cmnd(cmd);
1286 	case REQ_OP_ZONE_RESET:
1287 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1288 						   false);
1289 	case REQ_OP_ZONE_RESET_ALL:
1290 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1291 						   true);
1292 	case REQ_OP_ZONE_OPEN:
1293 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1294 	case REQ_OP_ZONE_CLOSE:
1295 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1296 	case REQ_OP_ZONE_FINISH:
1297 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1298 	default:
1299 		WARN_ON_ONCE(1);
1300 		return BLK_STS_NOTSUPP;
1301 	}
1302 }
1303 
1304 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1305 {
1306 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1307 
1308 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1309 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1310 }
1311 
1312 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1313 {
1314 	if (sdkp->device->removable || sdkp->write_prot) {
1315 		if (disk_check_media_change(disk))
1316 			return true;
1317 	}
1318 
1319 	/*
1320 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1321 	 * nothing to do with partitions, BLKRRPART is used to force a full
1322 	 * revalidate after things like a format for historical reasons.
1323 	 */
1324 	return test_bit(GD_NEED_PART_SCAN, &disk->state);
1325 }
1326 
1327 /**
1328  *	sd_open - open a scsi disk device
1329  *	@disk: disk to open
1330  *	@mode: open mode
1331  *
1332  *	Returns 0 if successful. Returns a negated errno value in case
1333  *	of error.
1334  *
1335  *	Note: This can be called from a user context (e.g. fsck(1) )
1336  *	or from within the kernel (e.g. as a result of a mount(1) ).
1337  *	In the latter case @inode and @filp carry an abridged amount
1338  *	of information as noted above.
1339  *
1340  *	Locking: called with disk->open_mutex held.
1341  **/
1342 static int sd_open(struct gendisk *disk, blk_mode_t mode)
1343 {
1344 	struct scsi_disk *sdkp = scsi_disk(disk);
1345 	struct scsi_device *sdev = sdkp->device;
1346 	int retval;
1347 
1348 	if (scsi_device_get(sdev))
1349 		return -ENXIO;
1350 
1351 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1352 
1353 	/*
1354 	 * If the device is in error recovery, wait until it is done.
1355 	 * If the device is offline, then disallow any access to it.
1356 	 */
1357 	retval = -ENXIO;
1358 	if (!scsi_block_when_processing_errors(sdev))
1359 		goto error_out;
1360 
1361 	if (sd_need_revalidate(disk, sdkp))
1362 		sd_revalidate_disk(disk);
1363 
1364 	/*
1365 	 * If the drive is empty, just let the open fail.
1366 	 */
1367 	retval = -ENOMEDIUM;
1368 	if (sdev->removable && !sdkp->media_present &&
1369 	    !(mode & BLK_OPEN_NDELAY))
1370 		goto error_out;
1371 
1372 	/*
1373 	 * If the device has the write protect tab set, have the open fail
1374 	 * if the user expects to be able to write to the thing.
1375 	 */
1376 	retval = -EROFS;
1377 	if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1378 		goto error_out;
1379 
1380 	/*
1381 	 * It is possible that the disk changing stuff resulted in
1382 	 * the device being taken offline.  If this is the case,
1383 	 * report this to the user, and don't pretend that the
1384 	 * open actually succeeded.
1385 	 */
1386 	retval = -ENXIO;
1387 	if (!scsi_device_online(sdev))
1388 		goto error_out;
1389 
1390 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1391 		if (scsi_block_when_processing_errors(sdev))
1392 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1393 	}
1394 
1395 	return 0;
1396 
1397 error_out:
1398 	scsi_device_put(sdev);
1399 	return retval;
1400 }
1401 
1402 /**
1403  *	sd_release - invoked when the (last) close(2) is called on this
1404  *	scsi disk.
1405  *	@disk: disk to release
1406  *
1407  *	Returns 0.
1408  *
1409  *	Note: may block (uninterruptible) if error recovery is underway
1410  *	on this disk.
1411  *
1412  *	Locking: called with disk->open_mutex held.
1413  **/
1414 static void sd_release(struct gendisk *disk)
1415 {
1416 	struct scsi_disk *sdkp = scsi_disk(disk);
1417 	struct scsi_device *sdev = sdkp->device;
1418 
1419 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1420 
1421 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1422 		if (scsi_block_when_processing_errors(sdev))
1423 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1424 	}
1425 
1426 	scsi_device_put(sdev);
1427 }
1428 
1429 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1430 {
1431 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1432 	struct scsi_device *sdp = sdkp->device;
1433 	struct Scsi_Host *host = sdp->host;
1434 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1435 	int diskinfo[4];
1436 
1437 	/* default to most commonly used values */
1438 	diskinfo[0] = 0x40;	/* 1 << 6 */
1439 	diskinfo[1] = 0x20;	/* 1 << 5 */
1440 	diskinfo[2] = capacity >> 11;
1441 
1442 	/* override with calculated, extended default, or driver values */
1443 	if (host->hostt->bios_param)
1444 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1445 	else
1446 		scsicam_bios_param(bdev, capacity, diskinfo);
1447 
1448 	geo->heads = diskinfo[0];
1449 	geo->sectors = diskinfo[1];
1450 	geo->cylinders = diskinfo[2];
1451 	return 0;
1452 }
1453 
1454 /**
1455  *	sd_ioctl - process an ioctl
1456  *	@bdev: target block device
1457  *	@mode: open mode
1458  *	@cmd: ioctl command number
1459  *	@arg: this is third argument given to ioctl(2) system call.
1460  *	Often contains a pointer.
1461  *
1462  *	Returns 0 if successful (some ioctls return positive numbers on
1463  *	success as well). Returns a negated errno value in case of error.
1464  *
1465  *	Note: most ioctls are forward onto the block subsystem or further
1466  *	down in the scsi subsystem.
1467  **/
1468 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1469 		    unsigned int cmd, unsigned long arg)
1470 {
1471 	struct gendisk *disk = bdev->bd_disk;
1472 	struct scsi_disk *sdkp = scsi_disk(disk);
1473 	struct scsi_device *sdp = sdkp->device;
1474 	void __user *p = (void __user *)arg;
1475 	int error;
1476 
1477 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1478 				    "cmd=0x%x\n", disk->disk_name, cmd));
1479 
1480 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1481 		return -ENOIOCTLCMD;
1482 
1483 	/*
1484 	 * If we are in the middle of error recovery, don't let anyone
1485 	 * else try and use this device.  Also, if error recovery fails, it
1486 	 * may try and take the device offline, in which case all further
1487 	 * access to the device is prohibited.
1488 	 */
1489 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1490 			(mode & BLK_OPEN_NDELAY));
1491 	if (error)
1492 		return error;
1493 
1494 	if (is_sed_ioctl(cmd))
1495 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1496 	return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1497 }
1498 
1499 static void set_media_not_present(struct scsi_disk *sdkp)
1500 {
1501 	if (sdkp->media_present)
1502 		sdkp->device->changed = 1;
1503 
1504 	if (sdkp->device->removable) {
1505 		sdkp->media_present = 0;
1506 		sdkp->capacity = 0;
1507 	}
1508 }
1509 
1510 static int media_not_present(struct scsi_disk *sdkp,
1511 			     struct scsi_sense_hdr *sshdr)
1512 {
1513 	if (!scsi_sense_valid(sshdr))
1514 		return 0;
1515 
1516 	/* not invoked for commands that could return deferred errors */
1517 	switch (sshdr->sense_key) {
1518 	case UNIT_ATTENTION:
1519 	case NOT_READY:
1520 		/* medium not present */
1521 		if (sshdr->asc == 0x3A) {
1522 			set_media_not_present(sdkp);
1523 			return 1;
1524 		}
1525 	}
1526 	return 0;
1527 }
1528 
1529 /**
1530  *	sd_check_events - check media events
1531  *	@disk: kernel device descriptor
1532  *	@clearing: disk events currently being cleared
1533  *
1534  *	Returns mask of DISK_EVENT_*.
1535  *
1536  *	Note: this function is invoked from the block subsystem.
1537  **/
1538 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1539 {
1540 	struct scsi_disk *sdkp = disk->private_data;
1541 	struct scsi_device *sdp;
1542 	int retval;
1543 	bool disk_changed;
1544 
1545 	if (!sdkp)
1546 		return 0;
1547 
1548 	sdp = sdkp->device;
1549 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1550 
1551 	/*
1552 	 * If the device is offline, don't send any commands - just pretend as
1553 	 * if the command failed.  If the device ever comes back online, we
1554 	 * can deal with it then.  It is only because of unrecoverable errors
1555 	 * that we would ever take a device offline in the first place.
1556 	 */
1557 	if (!scsi_device_online(sdp)) {
1558 		set_media_not_present(sdkp);
1559 		goto out;
1560 	}
1561 
1562 	/*
1563 	 * Using TEST_UNIT_READY enables differentiation between drive with
1564 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1565 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1566 	 *
1567 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1568 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1569 	 * sd_revalidate() is called.
1570 	 */
1571 	if (scsi_block_when_processing_errors(sdp)) {
1572 		struct scsi_sense_hdr sshdr = { 0, };
1573 
1574 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1575 					      &sshdr);
1576 
1577 		/* failed to execute TUR, assume media not present */
1578 		if (retval < 0 || host_byte(retval)) {
1579 			set_media_not_present(sdkp);
1580 			goto out;
1581 		}
1582 
1583 		if (media_not_present(sdkp, &sshdr))
1584 			goto out;
1585 	}
1586 
1587 	/*
1588 	 * For removable scsi disk we have to recognise the presence
1589 	 * of a disk in the drive.
1590 	 */
1591 	if (!sdkp->media_present)
1592 		sdp->changed = 1;
1593 	sdkp->media_present = 1;
1594 out:
1595 	/*
1596 	 * sdp->changed is set under the following conditions:
1597 	 *
1598 	 *	Medium present state has changed in either direction.
1599 	 *	Device has indicated UNIT_ATTENTION.
1600 	 */
1601 	disk_changed = sdp->changed;
1602 	sdp->changed = 0;
1603 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1604 }
1605 
1606 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1607 {
1608 	int retries, res;
1609 	struct scsi_device *sdp = sdkp->device;
1610 	const int timeout = sdp->request_queue->rq_timeout
1611 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1612 	struct scsi_sense_hdr my_sshdr;
1613 	const struct scsi_exec_args exec_args = {
1614 		.req_flags = BLK_MQ_REQ_PM,
1615 		/* caller might not be interested in sense, but we need it */
1616 		.sshdr = sshdr ? : &my_sshdr,
1617 	};
1618 
1619 	if (!scsi_device_online(sdp))
1620 		return -ENODEV;
1621 
1622 	sshdr = exec_args.sshdr;
1623 
1624 	for (retries = 3; retries > 0; --retries) {
1625 		unsigned char cmd[16] = { 0 };
1626 
1627 		if (sdp->use_16_for_sync)
1628 			cmd[0] = SYNCHRONIZE_CACHE_16;
1629 		else
1630 			cmd[0] = SYNCHRONIZE_CACHE;
1631 		/*
1632 		 * Leave the rest of the command zero to indicate
1633 		 * flush everything.
1634 		 */
1635 		res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1636 				       timeout, sdkp->max_retries, &exec_args);
1637 		if (res == 0)
1638 			break;
1639 	}
1640 
1641 	if (res) {
1642 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1643 
1644 		if (res < 0)
1645 			return res;
1646 
1647 		if (scsi_status_is_check_condition(res) &&
1648 		    scsi_sense_valid(sshdr)) {
1649 			sd_print_sense_hdr(sdkp, sshdr);
1650 
1651 			/* we need to evaluate the error return  */
1652 			if (sshdr->asc == 0x3a ||	/* medium not present */
1653 			    sshdr->asc == 0x20 ||	/* invalid command */
1654 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1655 				/* this is no error here */
1656 				return 0;
1657 		}
1658 
1659 		switch (host_byte(res)) {
1660 		/* ignore errors due to racing a disconnection */
1661 		case DID_BAD_TARGET:
1662 		case DID_NO_CONNECT:
1663 			return 0;
1664 		/* signal the upper layer it might try again */
1665 		case DID_BUS_BUSY:
1666 		case DID_IMM_RETRY:
1667 		case DID_REQUEUE:
1668 		case DID_SOFT_ERROR:
1669 			return -EBUSY;
1670 		default:
1671 			return -EIO;
1672 		}
1673 	}
1674 	return 0;
1675 }
1676 
1677 static void sd_rescan(struct device *dev)
1678 {
1679 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1680 
1681 	sd_revalidate_disk(sdkp->disk);
1682 }
1683 
1684 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1685 		enum blk_unique_id type)
1686 {
1687 	struct scsi_device *sdev = scsi_disk(disk)->device;
1688 	const struct scsi_vpd *vpd;
1689 	const unsigned char *d;
1690 	int ret = -ENXIO, len;
1691 
1692 	rcu_read_lock();
1693 	vpd = rcu_dereference(sdev->vpd_pg83);
1694 	if (!vpd)
1695 		goto out_unlock;
1696 
1697 	ret = -EINVAL;
1698 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1699 		/* we only care about designators with LU association */
1700 		if (((d[1] >> 4) & 0x3) != 0x00)
1701 			continue;
1702 		if ((d[1] & 0xf) != type)
1703 			continue;
1704 
1705 		/*
1706 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1707 		 * keep looking as one with more entropy might still show up.
1708 		 */
1709 		len = d[3];
1710 		if (len != 8 && len != 12 && len != 16)
1711 			continue;
1712 		ret = len;
1713 		memcpy(id, d + 4, len);
1714 		if (len == 16)
1715 			break;
1716 	}
1717 out_unlock:
1718 	rcu_read_unlock();
1719 	return ret;
1720 }
1721 
1722 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1723 {
1724 	switch (host_byte(result)) {
1725 	case DID_TRANSPORT_MARGINAL:
1726 	case DID_TRANSPORT_DISRUPTED:
1727 	case DID_BUS_BUSY:
1728 		return PR_STS_RETRY_PATH_FAILURE;
1729 	case DID_NO_CONNECT:
1730 		return PR_STS_PATH_FAILED;
1731 	case DID_TRANSPORT_FAILFAST:
1732 		return PR_STS_PATH_FAST_FAILED;
1733 	}
1734 
1735 	switch (status_byte(result)) {
1736 	case SAM_STAT_RESERVATION_CONFLICT:
1737 		return PR_STS_RESERVATION_CONFLICT;
1738 	case SAM_STAT_CHECK_CONDITION:
1739 		if (!scsi_sense_valid(sshdr))
1740 			return PR_STS_IOERR;
1741 
1742 		if (sshdr->sense_key == ILLEGAL_REQUEST &&
1743 		    (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1744 			return -EINVAL;
1745 
1746 		fallthrough;
1747 	default:
1748 		return PR_STS_IOERR;
1749 	}
1750 }
1751 
1752 static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1753 			    unsigned char *data, int data_len)
1754 {
1755 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1756 	struct scsi_device *sdev = sdkp->device;
1757 	struct scsi_sense_hdr sshdr;
1758 	u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1759 	const struct scsi_exec_args exec_args = {
1760 		.sshdr = &sshdr,
1761 	};
1762 	int result;
1763 
1764 	put_unaligned_be16(data_len, &cmd[7]);
1765 
1766 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1767 				  SD_TIMEOUT, sdkp->max_retries, &exec_args);
1768 	if (scsi_status_is_check_condition(result) &&
1769 	    scsi_sense_valid(&sshdr)) {
1770 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1771 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1772 	}
1773 
1774 	if (result <= 0)
1775 		return result;
1776 
1777 	return sd_scsi_to_pr_err(&sshdr, result);
1778 }
1779 
1780 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1781 {
1782 	int result, i, data_offset, num_copy_keys;
1783 	u32 num_keys = keys_info->num_keys;
1784 	int data_len = num_keys * 8 + 8;
1785 	u8 *data;
1786 
1787 	data = kzalloc(data_len, GFP_KERNEL);
1788 	if (!data)
1789 		return -ENOMEM;
1790 
1791 	result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1792 	if (result)
1793 		goto free_data;
1794 
1795 	keys_info->generation = get_unaligned_be32(&data[0]);
1796 	keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1797 
1798 	data_offset = 8;
1799 	num_copy_keys = min(num_keys, keys_info->num_keys);
1800 
1801 	for (i = 0; i < num_copy_keys; i++) {
1802 		keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1803 		data_offset += 8;
1804 	}
1805 
1806 free_data:
1807 	kfree(data);
1808 	return result;
1809 }
1810 
1811 static int sd_pr_read_reservation(struct block_device *bdev,
1812 				  struct pr_held_reservation *rsv)
1813 {
1814 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1815 	struct scsi_device *sdev = sdkp->device;
1816 	u8 data[24] = { };
1817 	int result, len;
1818 
1819 	result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1820 	if (result)
1821 		return result;
1822 
1823 	len = get_unaligned_be32(&data[4]);
1824 	if (!len)
1825 		return 0;
1826 
1827 	/* Make sure we have at least the key and type */
1828 	if (len < 14) {
1829 		sdev_printk(KERN_INFO, sdev,
1830 			    "READ RESERVATION failed due to short return buffer of %d bytes\n",
1831 			    len);
1832 		return -EINVAL;
1833 	}
1834 
1835 	rsv->generation = get_unaligned_be32(&data[0]);
1836 	rsv->key = get_unaligned_be64(&data[8]);
1837 	rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1838 	return 0;
1839 }
1840 
1841 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1842 			     u64 sa_key, enum scsi_pr_type type, u8 flags)
1843 {
1844 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1845 	struct scsi_device *sdev = sdkp->device;
1846 	struct scsi_sense_hdr sshdr;
1847 	const struct scsi_exec_args exec_args = {
1848 		.sshdr = &sshdr,
1849 	};
1850 	int result;
1851 	u8 cmd[16] = { 0, };
1852 	u8 data[24] = { 0, };
1853 
1854 	cmd[0] = PERSISTENT_RESERVE_OUT;
1855 	cmd[1] = sa;
1856 	cmd[2] = type;
1857 	put_unaligned_be32(sizeof(data), &cmd[5]);
1858 
1859 	put_unaligned_be64(key, &data[0]);
1860 	put_unaligned_be64(sa_key, &data[8]);
1861 	data[20] = flags;
1862 
1863 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1864 				  sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1865 				  &exec_args);
1866 
1867 	if (scsi_status_is_check_condition(result) &&
1868 	    scsi_sense_valid(&sshdr)) {
1869 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1870 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1871 	}
1872 
1873 	if (result <= 0)
1874 		return result;
1875 
1876 	return sd_scsi_to_pr_err(&sshdr, result);
1877 }
1878 
1879 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1880 		u32 flags)
1881 {
1882 	if (flags & ~PR_FL_IGNORE_KEY)
1883 		return -EOPNOTSUPP;
1884 	return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1885 			old_key, new_key, 0,
1886 			(1 << 0) /* APTPL */);
1887 }
1888 
1889 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1890 		u32 flags)
1891 {
1892 	if (flags)
1893 		return -EOPNOTSUPP;
1894 	return sd_pr_out_command(bdev, 0x01, key, 0,
1895 				 block_pr_type_to_scsi(type), 0);
1896 }
1897 
1898 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1899 {
1900 	return sd_pr_out_command(bdev, 0x02, key, 0,
1901 				 block_pr_type_to_scsi(type), 0);
1902 }
1903 
1904 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1905 		enum pr_type type, bool abort)
1906 {
1907 	return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1908 				 block_pr_type_to_scsi(type), 0);
1909 }
1910 
1911 static int sd_pr_clear(struct block_device *bdev, u64 key)
1912 {
1913 	return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
1914 }
1915 
1916 static const struct pr_ops sd_pr_ops = {
1917 	.pr_register	= sd_pr_register,
1918 	.pr_reserve	= sd_pr_reserve,
1919 	.pr_release	= sd_pr_release,
1920 	.pr_preempt	= sd_pr_preempt,
1921 	.pr_clear	= sd_pr_clear,
1922 	.pr_read_keys	= sd_pr_read_keys,
1923 	.pr_read_reservation = sd_pr_read_reservation,
1924 };
1925 
1926 static void scsi_disk_free_disk(struct gendisk *disk)
1927 {
1928 	struct scsi_disk *sdkp = scsi_disk(disk);
1929 
1930 	put_device(&sdkp->disk_dev);
1931 }
1932 
1933 static const struct block_device_operations sd_fops = {
1934 	.owner			= THIS_MODULE,
1935 	.open			= sd_open,
1936 	.release		= sd_release,
1937 	.ioctl			= sd_ioctl,
1938 	.getgeo			= sd_getgeo,
1939 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1940 	.check_events		= sd_check_events,
1941 	.unlock_native_capacity	= sd_unlock_native_capacity,
1942 	.report_zones		= sd_zbc_report_zones,
1943 	.get_unique_id		= sd_get_unique_id,
1944 	.free_disk		= scsi_disk_free_disk,
1945 	.pr_ops			= &sd_pr_ops,
1946 };
1947 
1948 /**
1949  *	sd_eh_reset - reset error handling callback
1950  *	@scmd:		sd-issued command that has failed
1951  *
1952  *	This function is called by the SCSI midlayer before starting
1953  *	SCSI EH. When counting medium access failures we have to be
1954  *	careful to register it only only once per device and SCSI EH run;
1955  *	there might be several timed out commands which will cause the
1956  *	'max_medium_access_timeouts' counter to trigger after the first
1957  *	SCSI EH run already and set the device to offline.
1958  *	So this function resets the internal counter before starting SCSI EH.
1959  **/
1960 static void sd_eh_reset(struct scsi_cmnd *scmd)
1961 {
1962 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1963 
1964 	/* New SCSI EH run, reset gate variable */
1965 	sdkp->ignore_medium_access_errors = false;
1966 }
1967 
1968 /**
1969  *	sd_eh_action - error handling callback
1970  *	@scmd:		sd-issued command that has failed
1971  *	@eh_disp:	The recovery disposition suggested by the midlayer
1972  *
1973  *	This function is called by the SCSI midlayer upon completion of an
1974  *	error test command (currently TEST UNIT READY). The result of sending
1975  *	the eh command is passed in eh_disp.  We're looking for devices that
1976  *	fail medium access commands but are OK with non access commands like
1977  *	test unit ready (so wrongly see the device as having a successful
1978  *	recovery)
1979  **/
1980 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1981 {
1982 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1983 	struct scsi_device *sdev = scmd->device;
1984 
1985 	if (!scsi_device_online(sdev) ||
1986 	    !scsi_medium_access_command(scmd) ||
1987 	    host_byte(scmd->result) != DID_TIME_OUT ||
1988 	    eh_disp != SUCCESS)
1989 		return eh_disp;
1990 
1991 	/*
1992 	 * The device has timed out executing a medium access command.
1993 	 * However, the TEST UNIT READY command sent during error
1994 	 * handling completed successfully. Either the device is in the
1995 	 * process of recovering or has it suffered an internal failure
1996 	 * that prevents access to the storage medium.
1997 	 */
1998 	if (!sdkp->ignore_medium_access_errors) {
1999 		sdkp->medium_access_timed_out++;
2000 		sdkp->ignore_medium_access_errors = true;
2001 	}
2002 
2003 	/*
2004 	 * If the device keeps failing read/write commands but TEST UNIT
2005 	 * READY always completes successfully we assume that medium
2006 	 * access is no longer possible and take the device offline.
2007 	 */
2008 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2009 		scmd_printk(KERN_ERR, scmd,
2010 			    "Medium access timeout failure. Offlining disk!\n");
2011 		mutex_lock(&sdev->state_mutex);
2012 		scsi_device_set_state(sdev, SDEV_OFFLINE);
2013 		mutex_unlock(&sdev->state_mutex);
2014 
2015 		return SUCCESS;
2016 	}
2017 
2018 	return eh_disp;
2019 }
2020 
2021 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2022 {
2023 	struct request *req = scsi_cmd_to_rq(scmd);
2024 	struct scsi_device *sdev = scmd->device;
2025 	unsigned int transferred, good_bytes;
2026 	u64 start_lba, end_lba, bad_lba;
2027 
2028 	/*
2029 	 * Some commands have a payload smaller than the device logical
2030 	 * block size (e.g. INQUIRY on a 4K disk).
2031 	 */
2032 	if (scsi_bufflen(scmd) <= sdev->sector_size)
2033 		return 0;
2034 
2035 	/* Check if we have a 'bad_lba' information */
2036 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2037 				     SCSI_SENSE_BUFFERSIZE,
2038 				     &bad_lba))
2039 		return 0;
2040 
2041 	/*
2042 	 * If the bad lba was reported incorrectly, we have no idea where
2043 	 * the error is.
2044 	 */
2045 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2046 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2047 	if (bad_lba < start_lba || bad_lba >= end_lba)
2048 		return 0;
2049 
2050 	/*
2051 	 * resid is optional but mostly filled in.  When it's unused,
2052 	 * its value is zero, so we assume the whole buffer transferred
2053 	 */
2054 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2055 
2056 	/* This computation should always be done in terms of the
2057 	 * resolution of the device's medium.
2058 	 */
2059 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2060 
2061 	return min(good_bytes, transferred);
2062 }
2063 
2064 /**
2065  *	sd_done - bottom half handler: called when the lower level
2066  *	driver has completed (successfully or otherwise) a scsi command.
2067  *	@SCpnt: mid-level's per command structure.
2068  *
2069  *	Note: potentially run from within an ISR. Must not block.
2070  **/
2071 static int sd_done(struct scsi_cmnd *SCpnt)
2072 {
2073 	int result = SCpnt->result;
2074 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2075 	unsigned int sector_size = SCpnt->device->sector_size;
2076 	unsigned int resid;
2077 	struct scsi_sense_hdr sshdr;
2078 	struct request *req = scsi_cmd_to_rq(SCpnt);
2079 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2080 	int sense_valid = 0;
2081 	int sense_deferred = 0;
2082 
2083 	switch (req_op(req)) {
2084 	case REQ_OP_DISCARD:
2085 	case REQ_OP_WRITE_ZEROES:
2086 	case REQ_OP_ZONE_RESET:
2087 	case REQ_OP_ZONE_RESET_ALL:
2088 	case REQ_OP_ZONE_OPEN:
2089 	case REQ_OP_ZONE_CLOSE:
2090 	case REQ_OP_ZONE_FINISH:
2091 		if (!result) {
2092 			good_bytes = blk_rq_bytes(req);
2093 			scsi_set_resid(SCpnt, 0);
2094 		} else {
2095 			good_bytes = 0;
2096 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2097 		}
2098 		break;
2099 	default:
2100 		/*
2101 		 * In case of bogus fw or device, we could end up having
2102 		 * an unaligned partial completion. Check this here and force
2103 		 * alignment.
2104 		 */
2105 		resid = scsi_get_resid(SCpnt);
2106 		if (resid & (sector_size - 1)) {
2107 			sd_printk(KERN_INFO, sdkp,
2108 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2109 				resid, sector_size);
2110 			scsi_print_command(SCpnt);
2111 			resid = min(scsi_bufflen(SCpnt),
2112 				    round_up(resid, sector_size));
2113 			scsi_set_resid(SCpnt, resid);
2114 		}
2115 	}
2116 
2117 	if (result) {
2118 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2119 		if (sense_valid)
2120 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2121 	}
2122 	sdkp->medium_access_timed_out = 0;
2123 
2124 	if (!scsi_status_is_check_condition(result) &&
2125 	    (!sense_valid || sense_deferred))
2126 		goto out;
2127 
2128 	switch (sshdr.sense_key) {
2129 	case HARDWARE_ERROR:
2130 	case MEDIUM_ERROR:
2131 		good_bytes = sd_completed_bytes(SCpnt);
2132 		break;
2133 	case RECOVERED_ERROR:
2134 		good_bytes = scsi_bufflen(SCpnt);
2135 		break;
2136 	case NO_SENSE:
2137 		/* This indicates a false check condition, so ignore it.  An
2138 		 * unknown amount of data was transferred so treat it as an
2139 		 * error.
2140 		 */
2141 		SCpnt->result = 0;
2142 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2143 		break;
2144 	case ABORTED_COMMAND:
2145 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2146 			good_bytes = sd_completed_bytes(SCpnt);
2147 		break;
2148 	case ILLEGAL_REQUEST:
2149 		switch (sshdr.asc) {
2150 		case 0x10:	/* DIX: Host detected corruption */
2151 			good_bytes = sd_completed_bytes(SCpnt);
2152 			break;
2153 		case 0x20:	/* INVALID COMMAND OPCODE */
2154 		case 0x24:	/* INVALID FIELD IN CDB */
2155 			switch (SCpnt->cmnd[0]) {
2156 			case UNMAP:
2157 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2158 				break;
2159 			case WRITE_SAME_16:
2160 			case WRITE_SAME:
2161 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2162 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2163 				} else {
2164 					sdkp->device->no_write_same = 1;
2165 					sd_config_write_same(sdkp);
2166 					req->rq_flags |= RQF_QUIET;
2167 				}
2168 				break;
2169 			}
2170 		}
2171 		break;
2172 	default:
2173 		break;
2174 	}
2175 
2176  out:
2177 	if (sd_is_zoned(sdkp))
2178 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2179 
2180 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2181 					   "sd_done: completed %d of %d bytes\n",
2182 					   good_bytes, scsi_bufflen(SCpnt)));
2183 
2184 	return good_bytes;
2185 }
2186 
2187 /*
2188  * spinup disk - called only in sd_revalidate_disk()
2189  */
2190 static void
2191 sd_spinup_disk(struct scsi_disk *sdkp)
2192 {
2193 	unsigned char cmd[10];
2194 	unsigned long spintime_expire = 0;
2195 	int retries, spintime;
2196 	unsigned int the_result;
2197 	struct scsi_sense_hdr sshdr;
2198 	const struct scsi_exec_args exec_args = {
2199 		.sshdr = &sshdr,
2200 	};
2201 	int sense_valid = 0;
2202 
2203 	spintime = 0;
2204 
2205 	/* Spin up drives, as required.  Only do this at boot time */
2206 	/* Spinup needs to be done for module loads too. */
2207 	do {
2208 		retries = 0;
2209 
2210 		do {
2211 			bool media_was_present = sdkp->media_present;
2212 
2213 			cmd[0] = TEST_UNIT_READY;
2214 			memset((void *) &cmd[1], 0, 9);
2215 
2216 			the_result = scsi_execute_cmd(sdkp->device, cmd,
2217 						      REQ_OP_DRV_IN, NULL, 0,
2218 						      SD_TIMEOUT,
2219 						      sdkp->max_retries,
2220 						      &exec_args);
2221 
2222 			/*
2223 			 * If the drive has indicated to us that it
2224 			 * doesn't have any media in it, don't bother
2225 			 * with any more polling.
2226 			 */
2227 			if (media_not_present(sdkp, &sshdr)) {
2228 				if (media_was_present)
2229 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2230 				return;
2231 			}
2232 
2233 			if (the_result)
2234 				sense_valid = scsi_sense_valid(&sshdr);
2235 			retries++;
2236 		} while (retries < 3 &&
2237 			 (!scsi_status_is_good(the_result) ||
2238 			  (scsi_status_is_check_condition(the_result) &&
2239 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2240 
2241 		if (!scsi_status_is_check_condition(the_result)) {
2242 			/* no sense, TUR either succeeded or failed
2243 			 * with a status error */
2244 			if(!spintime && !scsi_status_is_good(the_result)) {
2245 				sd_print_result(sdkp, "Test Unit Ready failed",
2246 						the_result);
2247 			}
2248 			break;
2249 		}
2250 
2251 		/*
2252 		 * The device does not want the automatic start to be issued.
2253 		 */
2254 		if (sdkp->device->no_start_on_add)
2255 			break;
2256 
2257 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2258 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2259 				break;	/* manual intervention required */
2260 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2261 				break;	/* standby */
2262 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2263 				break;	/* unavailable */
2264 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2265 				break;	/* sanitize in progress */
2266 			/*
2267 			 * Issue command to spin up drive when not ready
2268 			 */
2269 			if (!spintime) {
2270 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2271 				cmd[0] = START_STOP;
2272 				cmd[1] = 1;	/* Return immediately */
2273 				memset((void *) &cmd[2], 0, 8);
2274 				cmd[4] = 1;	/* Start spin cycle */
2275 				if (sdkp->device->start_stop_pwr_cond)
2276 					cmd[4] |= 1 << 4;
2277 				scsi_execute_cmd(sdkp->device, cmd,
2278 						 REQ_OP_DRV_IN, NULL, 0,
2279 						 SD_TIMEOUT, sdkp->max_retries,
2280 						 &exec_args);
2281 				spintime_expire = jiffies + 100 * HZ;
2282 				spintime = 1;
2283 			}
2284 			/* Wait 1 second for next try */
2285 			msleep(1000);
2286 			printk(KERN_CONT ".");
2287 
2288 		/*
2289 		 * Wait for USB flash devices with slow firmware.
2290 		 * Yes, this sense key/ASC combination shouldn't
2291 		 * occur here.  It's characteristic of these devices.
2292 		 */
2293 		} else if (sense_valid &&
2294 				sshdr.sense_key == UNIT_ATTENTION &&
2295 				sshdr.asc == 0x28) {
2296 			if (!spintime) {
2297 				spintime_expire = jiffies + 5 * HZ;
2298 				spintime = 1;
2299 			}
2300 			/* Wait 1 second for next try */
2301 			msleep(1000);
2302 		} else {
2303 			/* we don't understand the sense code, so it's
2304 			 * probably pointless to loop */
2305 			if(!spintime) {
2306 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2307 				sd_print_sense_hdr(sdkp, &sshdr);
2308 			}
2309 			break;
2310 		}
2311 
2312 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2313 
2314 	if (spintime) {
2315 		if (scsi_status_is_good(the_result))
2316 			printk(KERN_CONT "ready\n");
2317 		else
2318 			printk(KERN_CONT "not responding...\n");
2319 	}
2320 }
2321 
2322 /*
2323  * Determine whether disk supports Data Integrity Field.
2324  */
2325 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2326 {
2327 	struct scsi_device *sdp = sdkp->device;
2328 	u8 type;
2329 
2330 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2331 		sdkp->protection_type = 0;
2332 		return 0;
2333 	}
2334 
2335 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2336 
2337 	if (type > T10_PI_TYPE3_PROTECTION) {
2338 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported"	\
2339 			  " protection type %u. Disabling disk!\n",
2340 			  type);
2341 		sdkp->protection_type = 0;
2342 		return -ENODEV;
2343 	}
2344 
2345 	sdkp->protection_type = type;
2346 
2347 	return 0;
2348 }
2349 
2350 static void sd_config_protection(struct scsi_disk *sdkp)
2351 {
2352 	struct scsi_device *sdp = sdkp->device;
2353 
2354 	sd_dif_config_host(sdkp);
2355 
2356 	if (!sdkp->protection_type)
2357 		return;
2358 
2359 	if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2360 		sd_first_printk(KERN_NOTICE, sdkp,
2361 				"Disabling DIF Type %u protection\n",
2362 				sdkp->protection_type);
2363 		sdkp->protection_type = 0;
2364 	}
2365 
2366 	sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2367 			sdkp->protection_type);
2368 }
2369 
2370 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2371 			struct scsi_sense_hdr *sshdr, int sense_valid,
2372 			int the_result)
2373 {
2374 	if (sense_valid)
2375 		sd_print_sense_hdr(sdkp, sshdr);
2376 	else
2377 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2378 
2379 	/*
2380 	 * Set dirty bit for removable devices if not ready -
2381 	 * sometimes drives will not report this properly.
2382 	 */
2383 	if (sdp->removable &&
2384 	    sense_valid && sshdr->sense_key == NOT_READY)
2385 		set_media_not_present(sdkp);
2386 
2387 	/*
2388 	 * We used to set media_present to 0 here to indicate no media
2389 	 * in the drive, but some drives fail read capacity even with
2390 	 * media present, so we can't do that.
2391 	 */
2392 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2393 }
2394 
2395 #define RC16_LEN 32
2396 #if RC16_LEN > SD_BUF_SIZE
2397 #error RC16_LEN must not be more than SD_BUF_SIZE
2398 #endif
2399 
2400 #define READ_CAPACITY_RETRIES_ON_RESET	10
2401 
2402 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2403 						unsigned char *buffer)
2404 {
2405 	unsigned char cmd[16];
2406 	struct scsi_sense_hdr sshdr;
2407 	const struct scsi_exec_args exec_args = {
2408 		.sshdr = &sshdr,
2409 	};
2410 	int sense_valid = 0;
2411 	int the_result;
2412 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2413 	unsigned int alignment;
2414 	unsigned long long lba;
2415 	unsigned sector_size;
2416 
2417 	if (sdp->no_read_capacity_16)
2418 		return -EINVAL;
2419 
2420 	do {
2421 		memset(cmd, 0, 16);
2422 		cmd[0] = SERVICE_ACTION_IN_16;
2423 		cmd[1] = SAI_READ_CAPACITY_16;
2424 		cmd[13] = RC16_LEN;
2425 		memset(buffer, 0, RC16_LEN);
2426 
2427 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2428 					      buffer, RC16_LEN, SD_TIMEOUT,
2429 					      sdkp->max_retries, &exec_args);
2430 
2431 		if (media_not_present(sdkp, &sshdr))
2432 			return -ENODEV;
2433 
2434 		if (the_result > 0) {
2435 			sense_valid = scsi_sense_valid(&sshdr);
2436 			if (sense_valid &&
2437 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2438 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2439 			    sshdr.ascq == 0x00)
2440 				/* Invalid Command Operation Code or
2441 				 * Invalid Field in CDB, just retry
2442 				 * silently with RC10 */
2443 				return -EINVAL;
2444 			if (sense_valid &&
2445 			    sshdr.sense_key == UNIT_ATTENTION &&
2446 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2447 				/* Device reset might occur several times,
2448 				 * give it one more chance */
2449 				if (--reset_retries > 0)
2450 					continue;
2451 		}
2452 		retries--;
2453 
2454 	} while (the_result && retries);
2455 
2456 	if (the_result) {
2457 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2458 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2459 		return -EINVAL;
2460 	}
2461 
2462 	sector_size = get_unaligned_be32(&buffer[8]);
2463 	lba = get_unaligned_be64(&buffer[0]);
2464 
2465 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2466 		sdkp->capacity = 0;
2467 		return -ENODEV;
2468 	}
2469 
2470 	/* Logical blocks per physical block exponent */
2471 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2472 
2473 	/* RC basis */
2474 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2475 
2476 	/* Lowest aligned logical block */
2477 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2478 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2479 	if (alignment && sdkp->first_scan)
2480 		sd_printk(KERN_NOTICE, sdkp,
2481 			  "physical block alignment offset: %u\n", alignment);
2482 
2483 	if (buffer[14] & 0x80) { /* LBPME */
2484 		sdkp->lbpme = 1;
2485 
2486 		if (buffer[14] & 0x40) /* LBPRZ */
2487 			sdkp->lbprz = 1;
2488 
2489 		sd_config_discard(sdkp, SD_LBP_WS16);
2490 	}
2491 
2492 	sdkp->capacity = lba + 1;
2493 	return sector_size;
2494 }
2495 
2496 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2497 						unsigned char *buffer)
2498 {
2499 	unsigned char cmd[16];
2500 	struct scsi_sense_hdr sshdr;
2501 	const struct scsi_exec_args exec_args = {
2502 		.sshdr = &sshdr,
2503 	};
2504 	int sense_valid = 0;
2505 	int the_result;
2506 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2507 	sector_t lba;
2508 	unsigned sector_size;
2509 
2510 	do {
2511 		cmd[0] = READ_CAPACITY;
2512 		memset(&cmd[1], 0, 9);
2513 		memset(buffer, 0, 8);
2514 
2515 		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2516 					      8, SD_TIMEOUT, sdkp->max_retries,
2517 					      &exec_args);
2518 
2519 		if (media_not_present(sdkp, &sshdr))
2520 			return -ENODEV;
2521 
2522 		if (the_result > 0) {
2523 			sense_valid = scsi_sense_valid(&sshdr);
2524 			if (sense_valid &&
2525 			    sshdr.sense_key == UNIT_ATTENTION &&
2526 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2527 				/* Device reset might occur several times,
2528 				 * give it one more chance */
2529 				if (--reset_retries > 0)
2530 					continue;
2531 		}
2532 		retries--;
2533 
2534 	} while (the_result && retries);
2535 
2536 	if (the_result) {
2537 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2538 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2539 		return -EINVAL;
2540 	}
2541 
2542 	sector_size = get_unaligned_be32(&buffer[4]);
2543 	lba = get_unaligned_be32(&buffer[0]);
2544 
2545 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2546 		/* Some buggy (usb cardreader) devices return an lba of
2547 		   0xffffffff when the want to report a size of 0 (with
2548 		   which they really mean no media is present) */
2549 		sdkp->capacity = 0;
2550 		sdkp->physical_block_size = sector_size;
2551 		return sector_size;
2552 	}
2553 
2554 	sdkp->capacity = lba + 1;
2555 	sdkp->physical_block_size = sector_size;
2556 	return sector_size;
2557 }
2558 
2559 static int sd_try_rc16_first(struct scsi_device *sdp)
2560 {
2561 	if (sdp->host->max_cmd_len < 16)
2562 		return 0;
2563 	if (sdp->try_rc_10_first)
2564 		return 0;
2565 	if (sdp->scsi_level > SCSI_SPC_2)
2566 		return 1;
2567 	if (scsi_device_protection(sdp))
2568 		return 1;
2569 	return 0;
2570 }
2571 
2572 /*
2573  * read disk capacity
2574  */
2575 static void
2576 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2577 {
2578 	int sector_size;
2579 	struct scsi_device *sdp = sdkp->device;
2580 
2581 	if (sd_try_rc16_first(sdp)) {
2582 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2583 		if (sector_size == -EOVERFLOW)
2584 			goto got_data;
2585 		if (sector_size == -ENODEV)
2586 			return;
2587 		if (sector_size < 0)
2588 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2589 		if (sector_size < 0)
2590 			return;
2591 	} else {
2592 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2593 		if (sector_size == -EOVERFLOW)
2594 			goto got_data;
2595 		if (sector_size < 0)
2596 			return;
2597 		if ((sizeof(sdkp->capacity) > 4) &&
2598 		    (sdkp->capacity > 0xffffffffULL)) {
2599 			int old_sector_size = sector_size;
2600 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2601 					"Trying to use READ CAPACITY(16).\n");
2602 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2603 			if (sector_size < 0) {
2604 				sd_printk(KERN_NOTICE, sdkp,
2605 					"Using 0xffffffff as device size\n");
2606 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2607 				sector_size = old_sector_size;
2608 				goto got_data;
2609 			}
2610 			/* Remember that READ CAPACITY(16) succeeded */
2611 			sdp->try_rc_10_first = 0;
2612 		}
2613 	}
2614 
2615 	/* Some devices are known to return the total number of blocks,
2616 	 * not the highest block number.  Some devices have versions
2617 	 * which do this and others which do not.  Some devices we might
2618 	 * suspect of doing this but we don't know for certain.
2619 	 *
2620 	 * If we know the reported capacity is wrong, decrement it.  If
2621 	 * we can only guess, then assume the number of blocks is even
2622 	 * (usually true but not always) and err on the side of lowering
2623 	 * the capacity.
2624 	 */
2625 	if (sdp->fix_capacity ||
2626 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2627 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2628 				"from its reported value: %llu\n",
2629 				(unsigned long long) sdkp->capacity);
2630 		--sdkp->capacity;
2631 	}
2632 
2633 got_data:
2634 	if (sector_size == 0) {
2635 		sector_size = 512;
2636 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2637 			  "assuming 512.\n");
2638 	}
2639 
2640 	if (sector_size != 512 &&
2641 	    sector_size != 1024 &&
2642 	    sector_size != 2048 &&
2643 	    sector_size != 4096) {
2644 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2645 			  sector_size);
2646 		/*
2647 		 * The user might want to re-format the drive with
2648 		 * a supported sectorsize.  Once this happens, it
2649 		 * would be relatively trivial to set the thing up.
2650 		 * For this reason, we leave the thing in the table.
2651 		 */
2652 		sdkp->capacity = 0;
2653 		/*
2654 		 * set a bogus sector size so the normal read/write
2655 		 * logic in the block layer will eventually refuse any
2656 		 * request on this device without tripping over power
2657 		 * of two sector size assumptions
2658 		 */
2659 		sector_size = 512;
2660 	}
2661 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2662 	blk_queue_physical_block_size(sdp->request_queue,
2663 				      sdkp->physical_block_size);
2664 	sdkp->device->sector_size = sector_size;
2665 
2666 	if (sdkp->capacity > 0xffffffff)
2667 		sdp->use_16_for_rw = 1;
2668 
2669 }
2670 
2671 /*
2672  * Print disk capacity
2673  */
2674 static void
2675 sd_print_capacity(struct scsi_disk *sdkp,
2676 		  sector_t old_capacity)
2677 {
2678 	int sector_size = sdkp->device->sector_size;
2679 	char cap_str_2[10], cap_str_10[10];
2680 
2681 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2682 		return;
2683 
2684 	string_get_size(sdkp->capacity, sector_size,
2685 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2686 	string_get_size(sdkp->capacity, sector_size,
2687 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2688 
2689 	sd_printk(KERN_NOTICE, sdkp,
2690 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2691 		  (unsigned long long)sdkp->capacity,
2692 		  sector_size, cap_str_10, cap_str_2);
2693 
2694 	if (sdkp->physical_block_size != sector_size)
2695 		sd_printk(KERN_NOTICE, sdkp,
2696 			  "%u-byte physical blocks\n",
2697 			  sdkp->physical_block_size);
2698 }
2699 
2700 /* called with buffer of length 512 */
2701 static inline int
2702 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2703 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2704 		 struct scsi_sense_hdr *sshdr)
2705 {
2706 	/*
2707 	 * If we must use MODE SENSE(10), make sure that the buffer length
2708 	 * is at least 8 bytes so that the mode sense header fits.
2709 	 */
2710 	if (sdkp->device->use_10_for_ms && len < 8)
2711 		len = 8;
2712 
2713 	return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2714 			       SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2715 }
2716 
2717 /*
2718  * read write protect setting, if possible - called only in sd_revalidate_disk()
2719  * called with buffer of length SD_BUF_SIZE
2720  */
2721 static void
2722 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2723 {
2724 	int res;
2725 	struct scsi_device *sdp = sdkp->device;
2726 	struct scsi_mode_data data;
2727 	int old_wp = sdkp->write_prot;
2728 
2729 	set_disk_ro(sdkp->disk, 0);
2730 	if (sdp->skip_ms_page_3f) {
2731 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2732 		return;
2733 	}
2734 
2735 	if (sdp->use_192_bytes_for_3f) {
2736 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2737 	} else {
2738 		/*
2739 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2740 		 * We have to start carefully: some devices hang if we ask
2741 		 * for more than is available.
2742 		 */
2743 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2744 
2745 		/*
2746 		 * Second attempt: ask for page 0 When only page 0 is
2747 		 * implemented, a request for page 3F may return Sense Key
2748 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2749 		 * CDB.
2750 		 */
2751 		if (res < 0)
2752 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2753 
2754 		/*
2755 		 * Third attempt: ask 255 bytes, as we did earlier.
2756 		 */
2757 		if (res < 0)
2758 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2759 					       &data, NULL);
2760 	}
2761 
2762 	if (res < 0) {
2763 		sd_first_printk(KERN_WARNING, sdkp,
2764 			  "Test WP failed, assume Write Enabled\n");
2765 	} else {
2766 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2767 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2768 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2769 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2770 				  sdkp->write_prot ? "on" : "off");
2771 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2772 		}
2773 	}
2774 }
2775 
2776 /*
2777  * sd_read_cache_type - called only from sd_revalidate_disk()
2778  * called with buffer of length SD_BUF_SIZE
2779  */
2780 static void
2781 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2782 {
2783 	int len = 0, res;
2784 	struct scsi_device *sdp = sdkp->device;
2785 
2786 	int dbd;
2787 	int modepage;
2788 	int first_len;
2789 	struct scsi_mode_data data;
2790 	struct scsi_sense_hdr sshdr;
2791 	int old_wce = sdkp->WCE;
2792 	int old_rcd = sdkp->RCD;
2793 	int old_dpofua = sdkp->DPOFUA;
2794 
2795 
2796 	if (sdkp->cache_override)
2797 		return;
2798 
2799 	first_len = 4;
2800 	if (sdp->skip_ms_page_8) {
2801 		if (sdp->type == TYPE_RBC)
2802 			goto defaults;
2803 		else {
2804 			if (sdp->skip_ms_page_3f)
2805 				goto defaults;
2806 			modepage = 0x3F;
2807 			if (sdp->use_192_bytes_for_3f)
2808 				first_len = 192;
2809 			dbd = 0;
2810 		}
2811 	} else if (sdp->type == TYPE_RBC) {
2812 		modepage = 6;
2813 		dbd = 8;
2814 	} else {
2815 		modepage = 8;
2816 		dbd = 0;
2817 	}
2818 
2819 	/* cautiously ask */
2820 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2821 			&data, &sshdr);
2822 
2823 	if (res < 0)
2824 		goto bad_sense;
2825 
2826 	if (!data.header_length) {
2827 		modepage = 6;
2828 		first_len = 0;
2829 		sd_first_printk(KERN_ERR, sdkp,
2830 				"Missing header in MODE_SENSE response\n");
2831 	}
2832 
2833 	/* that went OK, now ask for the proper length */
2834 	len = data.length;
2835 
2836 	/*
2837 	 * We're only interested in the first three bytes, actually.
2838 	 * But the data cache page is defined for the first 20.
2839 	 */
2840 	if (len < 3)
2841 		goto bad_sense;
2842 	else if (len > SD_BUF_SIZE) {
2843 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2844 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2845 		len = SD_BUF_SIZE;
2846 	}
2847 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2848 		len = 192;
2849 
2850 	/* Get the data */
2851 	if (len > first_len)
2852 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2853 				&data, &sshdr);
2854 
2855 	if (!res) {
2856 		int offset = data.header_length + data.block_descriptor_length;
2857 
2858 		while (offset < len) {
2859 			u8 page_code = buffer[offset] & 0x3F;
2860 			u8 spf       = buffer[offset] & 0x40;
2861 
2862 			if (page_code == 8 || page_code == 6) {
2863 				/* We're interested only in the first 3 bytes.
2864 				 */
2865 				if (len - offset <= 2) {
2866 					sd_first_printk(KERN_ERR, sdkp,
2867 						"Incomplete mode parameter "
2868 							"data\n");
2869 					goto defaults;
2870 				} else {
2871 					modepage = page_code;
2872 					goto Page_found;
2873 				}
2874 			} else {
2875 				/* Go to the next page */
2876 				if (spf && len - offset > 3)
2877 					offset += 4 + (buffer[offset+2] << 8) +
2878 						buffer[offset+3];
2879 				else if (!spf && len - offset > 1)
2880 					offset += 2 + buffer[offset+1];
2881 				else {
2882 					sd_first_printk(KERN_ERR, sdkp,
2883 							"Incomplete mode "
2884 							"parameter data\n");
2885 					goto defaults;
2886 				}
2887 			}
2888 		}
2889 
2890 		sd_first_printk(KERN_WARNING, sdkp,
2891 				"No Caching mode page found\n");
2892 		goto defaults;
2893 
2894 	Page_found:
2895 		if (modepage == 8) {
2896 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2897 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2898 		} else {
2899 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2900 			sdkp->RCD = 0;
2901 		}
2902 
2903 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2904 		if (sdp->broken_fua) {
2905 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2906 			sdkp->DPOFUA = 0;
2907 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2908 			   !sdkp->device->use_16_for_rw) {
2909 			sd_first_printk(KERN_NOTICE, sdkp,
2910 				  "Uses READ/WRITE(6), disabling FUA\n");
2911 			sdkp->DPOFUA = 0;
2912 		}
2913 
2914 		/* No cache flush allowed for write protected devices */
2915 		if (sdkp->WCE && sdkp->write_prot)
2916 			sdkp->WCE = 0;
2917 
2918 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2919 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2920 			sd_printk(KERN_NOTICE, sdkp,
2921 				  "Write cache: %s, read cache: %s, %s\n",
2922 				  sdkp->WCE ? "enabled" : "disabled",
2923 				  sdkp->RCD ? "disabled" : "enabled",
2924 				  sdkp->DPOFUA ? "supports DPO and FUA"
2925 				  : "doesn't support DPO or FUA");
2926 
2927 		return;
2928 	}
2929 
2930 bad_sense:
2931 	if (scsi_sense_valid(&sshdr) &&
2932 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2933 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2934 		/* Invalid field in CDB */
2935 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2936 	else
2937 		sd_first_printk(KERN_ERR, sdkp,
2938 				"Asking for cache data failed\n");
2939 
2940 defaults:
2941 	if (sdp->wce_default_on) {
2942 		sd_first_printk(KERN_NOTICE, sdkp,
2943 				"Assuming drive cache: write back\n");
2944 		sdkp->WCE = 1;
2945 	} else {
2946 		sd_first_printk(KERN_WARNING, sdkp,
2947 				"Assuming drive cache: write through\n");
2948 		sdkp->WCE = 0;
2949 	}
2950 	sdkp->RCD = 0;
2951 	sdkp->DPOFUA = 0;
2952 }
2953 
2954 /*
2955  * The ATO bit indicates whether the DIF application tag is available
2956  * for use by the operating system.
2957  */
2958 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2959 {
2960 	int res, offset;
2961 	struct scsi_device *sdp = sdkp->device;
2962 	struct scsi_mode_data data;
2963 	struct scsi_sense_hdr sshdr;
2964 
2965 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2966 		return;
2967 
2968 	if (sdkp->protection_type == 0)
2969 		return;
2970 
2971 	res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
2972 			      sdkp->max_retries, &data, &sshdr);
2973 
2974 	if (res < 0 || !data.header_length ||
2975 	    data.length < 6) {
2976 		sd_first_printk(KERN_WARNING, sdkp,
2977 			  "getting Control mode page failed, assume no ATO\n");
2978 
2979 		if (scsi_sense_valid(&sshdr))
2980 			sd_print_sense_hdr(sdkp, &sshdr);
2981 
2982 		return;
2983 	}
2984 
2985 	offset = data.header_length + data.block_descriptor_length;
2986 
2987 	if ((buffer[offset] & 0x3f) != 0x0a) {
2988 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2989 		return;
2990 	}
2991 
2992 	if ((buffer[offset + 5] & 0x80) == 0)
2993 		return;
2994 
2995 	sdkp->ATO = 1;
2996 
2997 	return;
2998 }
2999 
3000 /**
3001  * sd_read_block_limits - Query disk device for preferred I/O sizes.
3002  * @sdkp: disk to query
3003  */
3004 static void sd_read_block_limits(struct scsi_disk *sdkp)
3005 {
3006 	struct scsi_vpd *vpd;
3007 
3008 	rcu_read_lock();
3009 
3010 	vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3011 	if (!vpd || vpd->len < 16)
3012 		goto out;
3013 
3014 	sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3015 	sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3016 	sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3017 
3018 	if (vpd->len >= 64) {
3019 		unsigned int lba_count, desc_count;
3020 
3021 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3022 
3023 		if (!sdkp->lbpme)
3024 			goto out;
3025 
3026 		lba_count = get_unaligned_be32(&vpd->data[20]);
3027 		desc_count = get_unaligned_be32(&vpd->data[24]);
3028 
3029 		if (lba_count && desc_count)
3030 			sdkp->max_unmap_blocks = lba_count;
3031 
3032 		sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3033 
3034 		if (vpd->data[32] & 0x80)
3035 			sdkp->unmap_alignment =
3036 				get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3037 
3038 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3039 
3040 			if (sdkp->max_unmap_blocks)
3041 				sd_config_discard(sdkp, SD_LBP_UNMAP);
3042 			else
3043 				sd_config_discard(sdkp, SD_LBP_WS16);
3044 
3045 		} else {	/* LBP VPD page tells us what to use */
3046 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
3047 				sd_config_discard(sdkp, SD_LBP_UNMAP);
3048 			else if (sdkp->lbpws)
3049 				sd_config_discard(sdkp, SD_LBP_WS16);
3050 			else if (sdkp->lbpws10)
3051 				sd_config_discard(sdkp, SD_LBP_WS10);
3052 			else
3053 				sd_config_discard(sdkp, SD_LBP_DISABLE);
3054 		}
3055 	}
3056 
3057  out:
3058 	rcu_read_unlock();
3059 }
3060 
3061 /**
3062  * sd_read_block_characteristics - Query block dev. characteristics
3063  * @sdkp: disk to query
3064  */
3065 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3066 {
3067 	struct request_queue *q = sdkp->disk->queue;
3068 	struct scsi_vpd *vpd;
3069 	u16 rot;
3070 	u8 zoned;
3071 
3072 	rcu_read_lock();
3073 	vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3074 
3075 	if (!vpd || vpd->len < 8) {
3076 		rcu_read_unlock();
3077 	        return;
3078 	}
3079 
3080 	rot = get_unaligned_be16(&vpd->data[4]);
3081 	zoned = (vpd->data[8] >> 4) & 3;
3082 	rcu_read_unlock();
3083 
3084 	if (rot == 1) {
3085 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3086 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3087 	}
3088 
3089 	if (sdkp->device->type == TYPE_ZBC) {
3090 		/*
3091 		 * Host-managed: Per ZBC and ZAC specifications, writes in
3092 		 * sequential write required zones of host-managed devices must
3093 		 * be aligned to the device physical block size.
3094 		 */
3095 		disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
3096 		blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3097 	} else {
3098 		sdkp->zoned = zoned;
3099 		if (sdkp->zoned == 1) {
3100 			/* Host-aware */
3101 			disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
3102 		} else {
3103 			/* Regular disk or drive managed disk */
3104 			disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3105 		}
3106 	}
3107 
3108 	if (!sdkp->first_scan)
3109 		return;
3110 
3111 	if (blk_queue_is_zoned(q)) {
3112 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3113 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3114 	} else {
3115 		if (sdkp->zoned == 1)
3116 			sd_printk(KERN_NOTICE, sdkp,
3117 				  "Host-aware SMR disk used as regular disk\n");
3118 		else if (sdkp->zoned == 2)
3119 			sd_printk(KERN_NOTICE, sdkp,
3120 				  "Drive-managed SMR disk\n");
3121 	}
3122 }
3123 
3124 /**
3125  * sd_read_block_provisioning - Query provisioning VPD page
3126  * @sdkp: disk to query
3127  */
3128 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3129 {
3130 	struct scsi_vpd *vpd;
3131 
3132 	if (sdkp->lbpme == 0)
3133 		return;
3134 
3135 	rcu_read_lock();
3136 	vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3137 
3138 	if (!vpd || vpd->len < 8) {
3139 		rcu_read_unlock();
3140 		return;
3141 	}
3142 
3143 	sdkp->lbpvpd	= 1;
3144 	sdkp->lbpu	= (vpd->data[5] >> 7) & 1; /* UNMAP */
3145 	sdkp->lbpws	= (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3146 	sdkp->lbpws10	= (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3147 	rcu_read_unlock();
3148 }
3149 
3150 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3151 {
3152 	struct scsi_device *sdev = sdkp->device;
3153 
3154 	if (sdev->host->no_write_same) {
3155 		sdev->no_write_same = 1;
3156 
3157 		return;
3158 	}
3159 
3160 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3161 		struct scsi_vpd *vpd;
3162 
3163 		sdev->no_report_opcodes = 1;
3164 
3165 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3166 		 * CODES is unsupported and the device has an ATA
3167 		 * Information VPD page (SAT).
3168 		 */
3169 		rcu_read_lock();
3170 		vpd = rcu_dereference(sdev->vpd_pg89);
3171 		if (vpd)
3172 			sdev->no_write_same = 1;
3173 		rcu_read_unlock();
3174 	}
3175 
3176 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3177 		sdkp->ws16 = 1;
3178 
3179 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3180 		sdkp->ws10 = 1;
3181 }
3182 
3183 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3184 {
3185 	struct scsi_device *sdev = sdkp->device;
3186 
3187 	if (!sdev->security_supported)
3188 		return;
3189 
3190 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3191 			SECURITY_PROTOCOL_IN, 0) == 1 &&
3192 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3193 			SECURITY_PROTOCOL_OUT, 0) == 1)
3194 		sdkp->security = 1;
3195 }
3196 
3197 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3198 {
3199 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3200 }
3201 
3202 /**
3203  * sd_read_cpr - Query concurrent positioning ranges
3204  * @sdkp:	disk to query
3205  */
3206 static void sd_read_cpr(struct scsi_disk *sdkp)
3207 {
3208 	struct blk_independent_access_ranges *iars = NULL;
3209 	unsigned char *buffer = NULL;
3210 	unsigned int nr_cpr = 0;
3211 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3212 	u8 *desc;
3213 
3214 	/*
3215 	 * We need to have the capacity set first for the block layer to be
3216 	 * able to check the ranges.
3217 	 */
3218 	if (sdkp->first_scan)
3219 		return;
3220 
3221 	if (!sdkp->capacity)
3222 		goto out;
3223 
3224 	/*
3225 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3226 	 * leading to a maximum page size of 64 + 256*32 bytes.
3227 	 */
3228 	buf_len = 64 + 256*32;
3229 	buffer = kmalloc(buf_len, GFP_KERNEL);
3230 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3231 		goto out;
3232 
3233 	/* We must have at least a 64B header and one 32B range descriptor */
3234 	vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3235 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3236 		sd_printk(KERN_ERR, sdkp,
3237 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3238 		goto out;
3239 	}
3240 
3241 	nr_cpr = (vpd_len - 64) / 32;
3242 	if (nr_cpr == 1) {
3243 		nr_cpr = 0;
3244 		goto out;
3245 	}
3246 
3247 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3248 	if (!iars) {
3249 		nr_cpr = 0;
3250 		goto out;
3251 	}
3252 
3253 	desc = &buffer[64];
3254 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3255 		if (desc[0] != i) {
3256 			sd_printk(KERN_ERR, sdkp,
3257 				"Invalid Concurrent Positioning Range number\n");
3258 			nr_cpr = 0;
3259 			break;
3260 		}
3261 
3262 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3263 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3264 	}
3265 
3266 out:
3267 	disk_set_independent_access_ranges(sdkp->disk, iars);
3268 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3269 		sd_printk(KERN_NOTICE, sdkp,
3270 			  "%u concurrent positioning ranges\n", nr_cpr);
3271 		sdkp->nr_actuators = nr_cpr;
3272 	}
3273 
3274 	kfree(buffer);
3275 }
3276 
3277 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3278 {
3279 	struct scsi_device *sdp = sdkp->device;
3280 	unsigned int min_xfer_bytes =
3281 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3282 
3283 	if (sdkp->min_xfer_blocks == 0)
3284 		return false;
3285 
3286 	if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3287 		sd_first_printk(KERN_WARNING, sdkp,
3288 				"Preferred minimum I/O size %u bytes not a " \
3289 				"multiple of physical block size (%u bytes)\n",
3290 				min_xfer_bytes, sdkp->physical_block_size);
3291 		sdkp->min_xfer_blocks = 0;
3292 		return false;
3293 	}
3294 
3295 	sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3296 			min_xfer_bytes);
3297 	return true;
3298 }
3299 
3300 /*
3301  * Determine the device's preferred I/O size for reads and writes
3302  * unless the reported value is unreasonably small, large, not a
3303  * multiple of the physical block size, or simply garbage.
3304  */
3305 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3306 				      unsigned int dev_max)
3307 {
3308 	struct scsi_device *sdp = sdkp->device;
3309 	unsigned int opt_xfer_bytes =
3310 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3311 	unsigned int min_xfer_bytes =
3312 		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3313 
3314 	if (sdkp->opt_xfer_blocks == 0)
3315 		return false;
3316 
3317 	if (sdkp->opt_xfer_blocks > dev_max) {
3318 		sd_first_printk(KERN_WARNING, sdkp,
3319 				"Optimal transfer size %u logical blocks " \
3320 				"> dev_max (%u logical blocks)\n",
3321 				sdkp->opt_xfer_blocks, dev_max);
3322 		return false;
3323 	}
3324 
3325 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3326 		sd_first_printk(KERN_WARNING, sdkp,
3327 				"Optimal transfer size %u logical blocks " \
3328 				"> sd driver limit (%u logical blocks)\n",
3329 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3330 		return false;
3331 	}
3332 
3333 	if (opt_xfer_bytes < PAGE_SIZE) {
3334 		sd_first_printk(KERN_WARNING, sdkp,
3335 				"Optimal transfer size %u bytes < " \
3336 				"PAGE_SIZE (%u bytes)\n",
3337 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3338 		return false;
3339 	}
3340 
3341 	if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3342 		sd_first_printk(KERN_WARNING, sdkp,
3343 				"Optimal transfer size %u bytes not a " \
3344 				"multiple of preferred minimum block " \
3345 				"size (%u bytes)\n",
3346 				opt_xfer_bytes, min_xfer_bytes);
3347 		return false;
3348 	}
3349 
3350 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3351 		sd_first_printk(KERN_WARNING, sdkp,
3352 				"Optimal transfer size %u bytes not a " \
3353 				"multiple of physical block size (%u bytes)\n",
3354 				opt_xfer_bytes, sdkp->physical_block_size);
3355 		return false;
3356 	}
3357 
3358 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3359 			opt_xfer_bytes);
3360 	return true;
3361 }
3362 
3363 /**
3364  *	sd_revalidate_disk - called the first time a new disk is seen,
3365  *	performs disk spin up, read_capacity, etc.
3366  *	@disk: struct gendisk we care about
3367  **/
3368 static int sd_revalidate_disk(struct gendisk *disk)
3369 {
3370 	struct scsi_disk *sdkp = scsi_disk(disk);
3371 	struct scsi_device *sdp = sdkp->device;
3372 	struct request_queue *q = sdkp->disk->queue;
3373 	sector_t old_capacity = sdkp->capacity;
3374 	unsigned char *buffer;
3375 	unsigned int dev_max, rw_max;
3376 
3377 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3378 				      "sd_revalidate_disk\n"));
3379 
3380 	/*
3381 	 * If the device is offline, don't try and read capacity or any
3382 	 * of the other niceties.
3383 	 */
3384 	if (!scsi_device_online(sdp))
3385 		goto out;
3386 
3387 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3388 	if (!buffer) {
3389 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3390 			  "allocation failure.\n");
3391 		goto out;
3392 	}
3393 
3394 	sd_spinup_disk(sdkp);
3395 
3396 	/*
3397 	 * Without media there is no reason to ask; moreover, some devices
3398 	 * react badly if we do.
3399 	 */
3400 	if (sdkp->media_present) {
3401 		sd_read_capacity(sdkp, buffer);
3402 
3403 		/*
3404 		 * set the default to rotational.  All non-rotational devices
3405 		 * support the block characteristics VPD page, which will
3406 		 * cause this to be updated correctly and any device which
3407 		 * doesn't support it should be treated as rotational.
3408 		 */
3409 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3410 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3411 
3412 		if (scsi_device_supports_vpd(sdp)) {
3413 			sd_read_block_provisioning(sdkp);
3414 			sd_read_block_limits(sdkp);
3415 			sd_read_block_characteristics(sdkp);
3416 			sd_zbc_read_zones(sdkp, buffer);
3417 			sd_read_cpr(sdkp);
3418 		}
3419 
3420 		sd_print_capacity(sdkp, old_capacity);
3421 
3422 		sd_read_write_protect_flag(sdkp, buffer);
3423 		sd_read_cache_type(sdkp, buffer);
3424 		sd_read_app_tag_own(sdkp, buffer);
3425 		sd_read_write_same(sdkp, buffer);
3426 		sd_read_security(sdkp, buffer);
3427 		sd_config_protection(sdkp);
3428 	}
3429 
3430 	/*
3431 	 * We now have all cache related info, determine how we deal
3432 	 * with flush requests.
3433 	 */
3434 	sd_set_flush_flag(sdkp);
3435 
3436 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3437 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3438 
3439 	/* Some devices report a maximum block count for READ/WRITE requests. */
3440 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3441 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3442 
3443 	if (sd_validate_min_xfer_size(sdkp))
3444 		blk_queue_io_min(sdkp->disk->queue,
3445 				 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3446 	else
3447 		blk_queue_io_min(sdkp->disk->queue, 0);
3448 
3449 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3450 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3451 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3452 	} else {
3453 		q->limits.io_opt = 0;
3454 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3455 				      (sector_t)BLK_DEF_MAX_SECTORS);
3456 	}
3457 
3458 	/*
3459 	 * Limit default to SCSI host optimal sector limit if set. There may be
3460 	 * an impact on performance for when the size of a request exceeds this
3461 	 * host limit.
3462 	 */
3463 	rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3464 
3465 	/* Do not exceed controller limit */
3466 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3467 
3468 	/*
3469 	 * Only update max_sectors if previously unset or if the current value
3470 	 * exceeds the capabilities of the hardware.
3471 	 */
3472 	if (sdkp->first_scan ||
3473 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3474 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3475 		q->limits.max_sectors = rw_max;
3476 
3477 	sdkp->first_scan = 0;
3478 
3479 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3480 	sd_config_write_same(sdkp);
3481 	kfree(buffer);
3482 
3483 	/*
3484 	 * For a zoned drive, revalidating the zones can be done only once
3485 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3486 	 * capacity to 0.
3487 	 */
3488 	if (sd_zbc_revalidate_zones(sdkp))
3489 		set_capacity_and_notify(disk, 0);
3490 
3491  out:
3492 	return 0;
3493 }
3494 
3495 /**
3496  *	sd_unlock_native_capacity - unlock native capacity
3497  *	@disk: struct gendisk to set capacity for
3498  *
3499  *	Block layer calls this function if it detects that partitions
3500  *	on @disk reach beyond the end of the device.  If the SCSI host
3501  *	implements ->unlock_native_capacity() method, it's invoked to
3502  *	give it a chance to adjust the device capacity.
3503  *
3504  *	CONTEXT:
3505  *	Defined by block layer.  Might sleep.
3506  */
3507 static void sd_unlock_native_capacity(struct gendisk *disk)
3508 {
3509 	struct scsi_device *sdev = scsi_disk(disk)->device;
3510 
3511 	if (sdev->host->hostt->unlock_native_capacity)
3512 		sdev->host->hostt->unlock_native_capacity(sdev);
3513 }
3514 
3515 /**
3516  *	sd_format_disk_name - format disk name
3517  *	@prefix: name prefix - ie. "sd" for SCSI disks
3518  *	@index: index of the disk to format name for
3519  *	@buf: output buffer
3520  *	@buflen: length of the output buffer
3521  *
3522  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3523  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3524  *	which is followed by sdaaa.
3525  *
3526  *	This is basically 26 base counting with one extra 'nil' entry
3527  *	at the beginning from the second digit on and can be
3528  *	determined using similar method as 26 base conversion with the
3529  *	index shifted -1 after each digit is computed.
3530  *
3531  *	CONTEXT:
3532  *	Don't care.
3533  *
3534  *	RETURNS:
3535  *	0 on success, -errno on failure.
3536  */
3537 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3538 {
3539 	const int base = 'z' - 'a' + 1;
3540 	char *begin = buf + strlen(prefix);
3541 	char *end = buf + buflen;
3542 	char *p;
3543 	int unit;
3544 
3545 	p = end - 1;
3546 	*p = '\0';
3547 	unit = base;
3548 	do {
3549 		if (p == begin)
3550 			return -EINVAL;
3551 		*--p = 'a' + (index % unit);
3552 		index = (index / unit) - 1;
3553 	} while (index >= 0);
3554 
3555 	memmove(begin, p, end - p);
3556 	memcpy(buf, prefix, strlen(prefix));
3557 
3558 	return 0;
3559 }
3560 
3561 /**
3562  *	sd_probe - called during driver initialization and whenever a
3563  *	new scsi device is attached to the system. It is called once
3564  *	for each scsi device (not just disks) present.
3565  *	@dev: pointer to device object
3566  *
3567  *	Returns 0 if successful (or not interested in this scsi device
3568  *	(e.g. scanner)); 1 when there is an error.
3569  *
3570  *	Note: this function is invoked from the scsi mid-level.
3571  *	This function sets up the mapping between a given
3572  *	<host,channel,id,lun> (found in sdp) and new device name
3573  *	(e.g. /dev/sda). More precisely it is the block device major
3574  *	and minor number that is chosen here.
3575  *
3576  *	Assume sd_probe is not re-entrant (for time being)
3577  *	Also think about sd_probe() and sd_remove() running coincidentally.
3578  **/
3579 static int sd_probe(struct device *dev)
3580 {
3581 	struct scsi_device *sdp = to_scsi_device(dev);
3582 	struct scsi_disk *sdkp;
3583 	struct gendisk *gd;
3584 	int index;
3585 	int error;
3586 
3587 	scsi_autopm_get_device(sdp);
3588 	error = -ENODEV;
3589 	if (sdp->type != TYPE_DISK &&
3590 	    sdp->type != TYPE_ZBC &&
3591 	    sdp->type != TYPE_MOD &&
3592 	    sdp->type != TYPE_RBC)
3593 		goto out;
3594 
3595 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3596 		sdev_printk(KERN_WARNING, sdp,
3597 			    "Unsupported ZBC host-managed device.\n");
3598 		goto out;
3599 	}
3600 
3601 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3602 					"sd_probe\n"));
3603 
3604 	error = -ENOMEM;
3605 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3606 	if (!sdkp)
3607 		goto out;
3608 
3609 	gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3610 					 &sd_bio_compl_lkclass);
3611 	if (!gd)
3612 		goto out_free;
3613 
3614 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3615 	if (index < 0) {
3616 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3617 		goto out_put;
3618 	}
3619 
3620 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3621 	if (error) {
3622 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3623 		goto out_free_index;
3624 	}
3625 
3626 	sdkp->device = sdp;
3627 	sdkp->disk = gd;
3628 	sdkp->index = index;
3629 	sdkp->max_retries = SD_MAX_RETRIES;
3630 	atomic_set(&sdkp->openers, 0);
3631 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3632 
3633 	if (!sdp->request_queue->rq_timeout) {
3634 		if (sdp->type != TYPE_MOD)
3635 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3636 		else
3637 			blk_queue_rq_timeout(sdp->request_queue,
3638 					     SD_MOD_TIMEOUT);
3639 	}
3640 
3641 	device_initialize(&sdkp->disk_dev);
3642 	sdkp->disk_dev.parent = get_device(dev);
3643 	sdkp->disk_dev.class = &sd_disk_class;
3644 	dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3645 
3646 	error = device_add(&sdkp->disk_dev);
3647 	if (error) {
3648 		put_device(&sdkp->disk_dev);
3649 		goto out;
3650 	}
3651 
3652 	dev_set_drvdata(dev, sdkp);
3653 
3654 	gd->major = sd_major((index & 0xf0) >> 4);
3655 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3656 	gd->minors = SD_MINORS;
3657 
3658 	gd->fops = &sd_fops;
3659 	gd->private_data = sdkp;
3660 
3661 	/* defaults, until the device tells us otherwise */
3662 	sdp->sector_size = 512;
3663 	sdkp->capacity = 0;
3664 	sdkp->media_present = 1;
3665 	sdkp->write_prot = 0;
3666 	sdkp->cache_override = 0;
3667 	sdkp->WCE = 0;
3668 	sdkp->RCD = 0;
3669 	sdkp->ATO = 0;
3670 	sdkp->first_scan = 1;
3671 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3672 
3673 	sd_revalidate_disk(gd);
3674 
3675 	if (sdp->removable) {
3676 		gd->flags |= GENHD_FL_REMOVABLE;
3677 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3678 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3679 	}
3680 
3681 	blk_pm_runtime_init(sdp->request_queue, dev);
3682 	if (sdp->rpm_autosuspend) {
3683 		pm_runtime_set_autosuspend_delay(dev,
3684 			sdp->host->hostt->rpm_autosuspend_delay);
3685 	}
3686 
3687 	error = device_add_disk(dev, gd, NULL);
3688 	if (error) {
3689 		put_device(&sdkp->disk_dev);
3690 		put_disk(gd);
3691 		goto out;
3692 	}
3693 
3694 	if (sdkp->security) {
3695 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3696 		if (sdkp->opal_dev)
3697 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3698 	}
3699 
3700 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3701 		  sdp->removable ? "removable " : "");
3702 	scsi_autopm_put_device(sdp);
3703 
3704 	return 0;
3705 
3706  out_free_index:
3707 	ida_free(&sd_index_ida, index);
3708  out_put:
3709 	put_disk(gd);
3710  out_free:
3711 	kfree(sdkp);
3712  out:
3713 	scsi_autopm_put_device(sdp);
3714 	return error;
3715 }
3716 
3717 /**
3718  *	sd_remove - called whenever a scsi disk (previously recognized by
3719  *	sd_probe) is detached from the system. It is called (potentially
3720  *	multiple times) during sd module unload.
3721  *	@dev: pointer to device object
3722  *
3723  *	Note: this function is invoked from the scsi mid-level.
3724  *	This function potentially frees up a device name (e.g. /dev/sdc)
3725  *	that could be re-used by a subsequent sd_probe().
3726  *	This function is not called when the built-in sd driver is "exit-ed".
3727  **/
3728 static int sd_remove(struct device *dev)
3729 {
3730 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3731 
3732 	scsi_autopm_get_device(sdkp->device);
3733 
3734 	device_del(&sdkp->disk_dev);
3735 	del_gendisk(sdkp->disk);
3736 	sd_shutdown(dev);
3737 
3738 	put_disk(sdkp->disk);
3739 	return 0;
3740 }
3741 
3742 static void scsi_disk_release(struct device *dev)
3743 {
3744 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3745 
3746 	ida_free(&sd_index_ida, sdkp->index);
3747 	sd_zbc_free_zone_info(sdkp);
3748 	put_device(&sdkp->device->sdev_gendev);
3749 	free_opal_dev(sdkp->opal_dev);
3750 
3751 	kfree(sdkp);
3752 }
3753 
3754 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3755 {
3756 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3757 	struct scsi_sense_hdr sshdr;
3758 	const struct scsi_exec_args exec_args = {
3759 		.sshdr = &sshdr,
3760 		.req_flags = BLK_MQ_REQ_PM,
3761 	};
3762 	struct scsi_device *sdp = sdkp->device;
3763 	int res;
3764 
3765 	if (start)
3766 		cmd[4] |= 1;	/* START */
3767 
3768 	if (sdp->start_stop_pwr_cond)
3769 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3770 
3771 	if (!scsi_device_online(sdp))
3772 		return -ENODEV;
3773 
3774 	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
3775 			       sdkp->max_retries, &exec_args);
3776 	if (res) {
3777 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3778 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3779 			sd_print_sense_hdr(sdkp, &sshdr);
3780 			/* 0x3a is medium not present */
3781 			if (sshdr.asc == 0x3a)
3782 				res = 0;
3783 		}
3784 	}
3785 
3786 	/* SCSI error codes must not go to the generic layer */
3787 	if (res)
3788 		return -EIO;
3789 
3790 	return 0;
3791 }
3792 
3793 /*
3794  * Send a SYNCHRONIZE CACHE instruction down to the device through
3795  * the normal SCSI command structure.  Wait for the command to
3796  * complete.
3797  */
3798 static void sd_shutdown(struct device *dev)
3799 {
3800 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3801 
3802 	if (!sdkp)
3803 		return;         /* this can happen */
3804 
3805 	if (pm_runtime_suspended(dev))
3806 		return;
3807 
3808 	if (sdkp->WCE && sdkp->media_present) {
3809 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3810 		sd_sync_cache(sdkp, NULL);
3811 	}
3812 
3813 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3814 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3815 		sd_start_stop_device(sdkp, 0);
3816 	}
3817 }
3818 
3819 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3820 {
3821 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3822 	struct scsi_sense_hdr sshdr;
3823 	int ret = 0;
3824 
3825 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3826 		return 0;
3827 
3828 	if (sdkp->WCE && sdkp->media_present) {
3829 		if (!sdkp->device->silence_suspend)
3830 			sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3831 		ret = sd_sync_cache(sdkp, &sshdr);
3832 
3833 		if (ret) {
3834 			/* ignore OFFLINE device */
3835 			if (ret == -ENODEV)
3836 				return 0;
3837 
3838 			if (!scsi_sense_valid(&sshdr) ||
3839 			    sshdr.sense_key != ILLEGAL_REQUEST)
3840 				return ret;
3841 
3842 			/*
3843 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3844 			 * doesn't support sync. There's not much to do and
3845 			 * suspend shouldn't fail.
3846 			 */
3847 			ret = 0;
3848 		}
3849 	}
3850 
3851 	if (sdkp->device->manage_start_stop) {
3852 		if (!sdkp->device->silence_suspend)
3853 			sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3854 		/* an error is not worth aborting a system sleep */
3855 		ret = sd_start_stop_device(sdkp, 0);
3856 		if (ignore_stop_errors)
3857 			ret = 0;
3858 	}
3859 
3860 	return ret;
3861 }
3862 
3863 static int sd_suspend_system(struct device *dev)
3864 {
3865 	if (pm_runtime_suspended(dev))
3866 		return 0;
3867 
3868 	return sd_suspend_common(dev, true);
3869 }
3870 
3871 static int sd_suspend_runtime(struct device *dev)
3872 {
3873 	return sd_suspend_common(dev, false);
3874 }
3875 
3876 static int sd_resume(struct device *dev)
3877 {
3878 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3879 	int ret;
3880 
3881 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3882 		return 0;
3883 
3884 	if (!sdkp->device->manage_start_stop)
3885 		return 0;
3886 
3887 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3888 	ret = sd_start_stop_device(sdkp, 1);
3889 	if (!ret)
3890 		opal_unlock_from_suspend(sdkp->opal_dev);
3891 	return ret;
3892 }
3893 
3894 static int sd_resume_system(struct device *dev)
3895 {
3896 	if (pm_runtime_suspended(dev))
3897 		return 0;
3898 
3899 	return sd_resume(dev);
3900 }
3901 
3902 static int sd_resume_runtime(struct device *dev)
3903 {
3904 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3905 	struct scsi_device *sdp;
3906 
3907 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3908 		return 0;
3909 
3910 	sdp = sdkp->device;
3911 
3912 	if (sdp->ignore_media_change) {
3913 		/* clear the device's sense data */
3914 		static const u8 cmd[10] = { REQUEST_SENSE };
3915 		const struct scsi_exec_args exec_args = {
3916 			.req_flags = BLK_MQ_REQ_PM,
3917 		};
3918 
3919 		if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
3920 				     sdp->request_queue->rq_timeout, 1,
3921 				     &exec_args))
3922 			sd_printk(KERN_NOTICE, sdkp,
3923 				  "Failed to clear sense data\n");
3924 	}
3925 
3926 	return sd_resume(dev);
3927 }
3928 
3929 /**
3930  *	init_sd - entry point for this driver (both when built in or when
3931  *	a module).
3932  *
3933  *	Note: this function registers this driver with the scsi mid-level.
3934  **/
3935 static int __init init_sd(void)
3936 {
3937 	int majors = 0, i, err;
3938 
3939 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3940 
3941 	for (i = 0; i < SD_MAJORS; i++) {
3942 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3943 			continue;
3944 		majors++;
3945 	}
3946 
3947 	if (!majors)
3948 		return -ENODEV;
3949 
3950 	err = class_register(&sd_disk_class);
3951 	if (err)
3952 		goto err_out;
3953 
3954 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3955 	if (!sd_page_pool) {
3956 		printk(KERN_ERR "sd: can't init discard page pool\n");
3957 		err = -ENOMEM;
3958 		goto err_out_class;
3959 	}
3960 
3961 	err = scsi_register_driver(&sd_template.gendrv);
3962 	if (err)
3963 		goto err_out_driver;
3964 
3965 	return 0;
3966 
3967 err_out_driver:
3968 	mempool_destroy(sd_page_pool);
3969 err_out_class:
3970 	class_unregister(&sd_disk_class);
3971 err_out:
3972 	for (i = 0; i < SD_MAJORS; i++)
3973 		unregister_blkdev(sd_major(i), "sd");
3974 	return err;
3975 }
3976 
3977 /**
3978  *	exit_sd - exit point for this driver (when it is a module).
3979  *
3980  *	Note: this function unregisters this driver from the scsi mid-level.
3981  **/
3982 static void __exit exit_sd(void)
3983 {
3984 	int i;
3985 
3986 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3987 
3988 	scsi_unregister_driver(&sd_template.gendrv);
3989 	mempool_destroy(sd_page_pool);
3990 
3991 	class_unregister(&sd_disk_class);
3992 
3993 	for (i = 0; i < SD_MAJORS; i++)
3994 		unregister_blkdev(sd_major(i), "sd");
3995 }
3996 
3997 module_init(init_sd);
3998 module_exit(exit_sd);
3999 
4000 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4001 {
4002 	scsi_print_sense_hdr(sdkp->device,
4003 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4004 }
4005 
4006 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4007 {
4008 	const char *hb_string = scsi_hostbyte_string(result);
4009 
4010 	if (hb_string)
4011 		sd_printk(KERN_INFO, sdkp,
4012 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4013 			  hb_string ? hb_string : "invalid",
4014 			  "DRIVER_OK");
4015 	else
4016 		sd_printk(KERN_INFO, sdkp,
4017 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4018 			  msg, host_byte(result), "DRIVER_OK");
4019 }
4020