xref: /linux/drivers/scsi/sd.c (revision a5359ddd052860bacf957e65fe819c63e974b3a6)
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 
71 #include "sd.h"
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
74 
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
78 
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 
100 #define SD_MINORS	16
101 
102 static void sd_config_discard(struct scsi_disk *, unsigned int);
103 static void sd_config_write_same(struct scsi_disk *);
104 static int  sd_revalidate_disk(struct gendisk *);
105 static void sd_unlock_native_capacity(struct gendisk *disk);
106 static int  sd_probe(struct device *);
107 static int  sd_remove(struct device *);
108 static void sd_shutdown(struct device *);
109 static int sd_suspend_system(struct device *);
110 static int sd_suspend_runtime(struct device *);
111 static int sd_resume_system(struct device *);
112 static int sd_resume_runtime(struct device *);
113 static void sd_rescan(struct device *);
114 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
115 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
116 static int sd_done(struct scsi_cmnd *);
117 static void sd_eh_reset(struct scsi_cmnd *);
118 static int sd_eh_action(struct scsi_cmnd *, int);
119 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
120 static void scsi_disk_release(struct device *cdev);
121 
122 static DEFINE_IDA(sd_index_ida);
123 
124 /* This semaphore is used to mediate the 0->1 reference get in the
125  * face of object destruction (i.e. we can't allow a get on an
126  * object after last put) */
127 static DEFINE_MUTEX(sd_ref_mutex);
128 
129 static struct kmem_cache *sd_cdb_cache;
130 static mempool_t *sd_cdb_pool;
131 static mempool_t *sd_page_pool;
132 static struct lock_class_key sd_bio_compl_lkclass;
133 
134 static const char *sd_cache_types[] = {
135 	"write through", "none", "write back",
136 	"write back, no read (daft)"
137 };
138 
139 static void sd_set_flush_flag(struct scsi_disk *sdkp)
140 {
141 	bool wc = false, fua = false;
142 
143 	if (sdkp->WCE) {
144 		wc = true;
145 		if (sdkp->DPOFUA)
146 			fua = true;
147 	}
148 
149 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
150 }
151 
152 static ssize_t
153 cache_type_store(struct device *dev, struct device_attribute *attr,
154 		 const char *buf, size_t count)
155 {
156 	int ct, rcd, wce, sp;
157 	struct scsi_disk *sdkp = to_scsi_disk(dev);
158 	struct scsi_device *sdp = sdkp->device;
159 	char buffer[64];
160 	char *buffer_data;
161 	struct scsi_mode_data data;
162 	struct scsi_sense_hdr sshdr;
163 	static const char temp[] = "temporary ";
164 	int len;
165 
166 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
167 		/* no cache control on RBC devices; theoretically they
168 		 * can do it, but there's probably so many exceptions
169 		 * it's not worth the risk */
170 		return -EINVAL;
171 
172 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
173 		buf += sizeof(temp) - 1;
174 		sdkp->cache_override = 1;
175 	} else {
176 		sdkp->cache_override = 0;
177 	}
178 
179 	ct = sysfs_match_string(sd_cache_types, buf);
180 	if (ct < 0)
181 		return -EINVAL;
182 
183 	rcd = ct & 0x01 ? 1 : 0;
184 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
185 
186 	if (sdkp->cache_override) {
187 		sdkp->WCE = wce;
188 		sdkp->RCD = rcd;
189 		sd_set_flush_flag(sdkp);
190 		return count;
191 	}
192 
193 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
194 			    sdkp->max_retries, &data, NULL))
195 		return -EINVAL;
196 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
197 		  data.block_descriptor_length);
198 	buffer_data = buffer + data.header_length +
199 		data.block_descriptor_length;
200 	buffer_data[2] &= ~0x05;
201 	buffer_data[2] |= wce << 2 | rcd;
202 	sp = buffer_data[0] & 0x80 ? 1 : 0;
203 	buffer_data[0] &= ~0x80;
204 
205 	/*
206 	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
207 	 * received mode parameter buffer before doing MODE SELECT.
208 	 */
209 	data.device_specific = 0;
210 
211 	if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
212 			     sdkp->max_retries, &data, &sshdr)) {
213 		if (scsi_sense_valid(&sshdr))
214 			sd_print_sense_hdr(sdkp, &sshdr);
215 		return -EINVAL;
216 	}
217 	sd_revalidate_disk(sdkp->disk);
218 	return count;
219 }
220 
221 static ssize_t
222 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
223 		       char *buf)
224 {
225 	struct scsi_disk *sdkp = to_scsi_disk(dev);
226 	struct scsi_device *sdp = sdkp->device;
227 
228 	return sprintf(buf, "%u\n", sdp->manage_start_stop);
229 }
230 
231 static ssize_t
232 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
233 			const char *buf, size_t count)
234 {
235 	struct scsi_disk *sdkp = to_scsi_disk(dev);
236 	struct scsi_device *sdp = sdkp->device;
237 	bool v;
238 
239 	if (!capable(CAP_SYS_ADMIN))
240 		return -EACCES;
241 
242 	if (kstrtobool(buf, &v))
243 		return -EINVAL;
244 
245 	sdp->manage_start_stop = v;
246 
247 	return count;
248 }
249 static DEVICE_ATTR_RW(manage_start_stop);
250 
251 static ssize_t
252 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
253 {
254 	struct scsi_disk *sdkp = to_scsi_disk(dev);
255 
256 	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
257 }
258 
259 static ssize_t
260 allow_restart_store(struct device *dev, struct device_attribute *attr,
261 		    const char *buf, size_t count)
262 {
263 	bool v;
264 	struct scsi_disk *sdkp = to_scsi_disk(dev);
265 	struct scsi_device *sdp = sdkp->device;
266 
267 	if (!capable(CAP_SYS_ADMIN))
268 		return -EACCES;
269 
270 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
271 		return -EINVAL;
272 
273 	if (kstrtobool(buf, &v))
274 		return -EINVAL;
275 
276 	sdp->allow_restart = v;
277 
278 	return count;
279 }
280 static DEVICE_ATTR_RW(allow_restart);
281 
282 static ssize_t
283 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
284 {
285 	struct scsi_disk *sdkp = to_scsi_disk(dev);
286 	int ct = sdkp->RCD + 2*sdkp->WCE;
287 
288 	return sprintf(buf, "%s\n", sd_cache_types[ct]);
289 }
290 static DEVICE_ATTR_RW(cache_type);
291 
292 static ssize_t
293 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
294 {
295 	struct scsi_disk *sdkp = to_scsi_disk(dev);
296 
297 	return sprintf(buf, "%u\n", sdkp->DPOFUA);
298 }
299 static DEVICE_ATTR_RO(FUA);
300 
301 static ssize_t
302 protection_type_show(struct device *dev, struct device_attribute *attr,
303 		     char *buf)
304 {
305 	struct scsi_disk *sdkp = to_scsi_disk(dev);
306 
307 	return sprintf(buf, "%u\n", sdkp->protection_type);
308 }
309 
310 static ssize_t
311 protection_type_store(struct device *dev, struct device_attribute *attr,
312 		      const char *buf, size_t count)
313 {
314 	struct scsi_disk *sdkp = to_scsi_disk(dev);
315 	unsigned int val;
316 	int err;
317 
318 	if (!capable(CAP_SYS_ADMIN))
319 		return -EACCES;
320 
321 	err = kstrtouint(buf, 10, &val);
322 
323 	if (err)
324 		return err;
325 
326 	if (val <= T10_PI_TYPE3_PROTECTION)
327 		sdkp->protection_type = val;
328 
329 	return count;
330 }
331 static DEVICE_ATTR_RW(protection_type);
332 
333 static ssize_t
334 protection_mode_show(struct device *dev, struct device_attribute *attr,
335 		     char *buf)
336 {
337 	struct scsi_disk *sdkp = to_scsi_disk(dev);
338 	struct scsi_device *sdp = sdkp->device;
339 	unsigned int dif, dix;
340 
341 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
342 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
343 
344 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
345 		dif = 0;
346 		dix = 1;
347 	}
348 
349 	if (!dif && !dix)
350 		return sprintf(buf, "none\n");
351 
352 	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
353 }
354 static DEVICE_ATTR_RO(protection_mode);
355 
356 static ssize_t
357 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
358 {
359 	struct scsi_disk *sdkp = to_scsi_disk(dev);
360 
361 	return sprintf(buf, "%u\n", sdkp->ATO);
362 }
363 static DEVICE_ATTR_RO(app_tag_own);
364 
365 static ssize_t
366 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
367 		       char *buf)
368 {
369 	struct scsi_disk *sdkp = to_scsi_disk(dev);
370 
371 	return sprintf(buf, "%u\n", sdkp->lbpme);
372 }
373 static DEVICE_ATTR_RO(thin_provisioning);
374 
375 /* sysfs_match_string() requires dense arrays */
376 static const char *lbp_mode[] = {
377 	[SD_LBP_FULL]		= "full",
378 	[SD_LBP_UNMAP]		= "unmap",
379 	[SD_LBP_WS16]		= "writesame_16",
380 	[SD_LBP_WS10]		= "writesame_10",
381 	[SD_LBP_ZERO]		= "writesame_zero",
382 	[SD_LBP_DISABLE]	= "disabled",
383 };
384 
385 static ssize_t
386 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
387 		       char *buf)
388 {
389 	struct scsi_disk *sdkp = to_scsi_disk(dev);
390 
391 	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
392 }
393 
394 static ssize_t
395 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
396 			const char *buf, size_t count)
397 {
398 	struct scsi_disk *sdkp = to_scsi_disk(dev);
399 	struct scsi_device *sdp = sdkp->device;
400 	int mode;
401 
402 	if (!capable(CAP_SYS_ADMIN))
403 		return -EACCES;
404 
405 	if (sd_is_zoned(sdkp)) {
406 		sd_config_discard(sdkp, SD_LBP_DISABLE);
407 		return count;
408 	}
409 
410 	if (sdp->type != TYPE_DISK)
411 		return -EINVAL;
412 
413 	mode = sysfs_match_string(lbp_mode, buf);
414 	if (mode < 0)
415 		return -EINVAL;
416 
417 	sd_config_discard(sdkp, mode);
418 
419 	return count;
420 }
421 static DEVICE_ATTR_RW(provisioning_mode);
422 
423 /* sysfs_match_string() requires dense arrays */
424 static const char *zeroing_mode[] = {
425 	[SD_ZERO_WRITE]		= "write",
426 	[SD_ZERO_WS]		= "writesame",
427 	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
428 	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
429 };
430 
431 static ssize_t
432 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
433 		  char *buf)
434 {
435 	struct scsi_disk *sdkp = to_scsi_disk(dev);
436 
437 	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
438 }
439 
440 static ssize_t
441 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
442 		   const char *buf, size_t count)
443 {
444 	struct scsi_disk *sdkp = to_scsi_disk(dev);
445 	int mode;
446 
447 	if (!capable(CAP_SYS_ADMIN))
448 		return -EACCES;
449 
450 	mode = sysfs_match_string(zeroing_mode, buf);
451 	if (mode < 0)
452 		return -EINVAL;
453 
454 	sdkp->zeroing_mode = mode;
455 
456 	return count;
457 }
458 static DEVICE_ATTR_RW(zeroing_mode);
459 
460 static ssize_t
461 max_medium_access_timeouts_show(struct device *dev,
462 				struct device_attribute *attr, char *buf)
463 {
464 	struct scsi_disk *sdkp = to_scsi_disk(dev);
465 
466 	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
467 }
468 
469 static ssize_t
470 max_medium_access_timeouts_store(struct device *dev,
471 				 struct device_attribute *attr, const char *buf,
472 				 size_t count)
473 {
474 	struct scsi_disk *sdkp = to_scsi_disk(dev);
475 	int err;
476 
477 	if (!capable(CAP_SYS_ADMIN))
478 		return -EACCES;
479 
480 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
481 
482 	return err ? err : count;
483 }
484 static DEVICE_ATTR_RW(max_medium_access_timeouts);
485 
486 static ssize_t
487 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
488 			   char *buf)
489 {
490 	struct scsi_disk *sdkp = to_scsi_disk(dev);
491 
492 	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
493 }
494 
495 static ssize_t
496 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
497 			    const char *buf, size_t count)
498 {
499 	struct scsi_disk *sdkp = to_scsi_disk(dev);
500 	struct scsi_device *sdp = sdkp->device;
501 	unsigned long max;
502 	int err;
503 
504 	if (!capable(CAP_SYS_ADMIN))
505 		return -EACCES;
506 
507 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
508 		return -EINVAL;
509 
510 	err = kstrtoul(buf, 10, &max);
511 
512 	if (err)
513 		return err;
514 
515 	if (max == 0)
516 		sdp->no_write_same = 1;
517 	else if (max <= SD_MAX_WS16_BLOCKS) {
518 		sdp->no_write_same = 0;
519 		sdkp->max_ws_blocks = max;
520 	}
521 
522 	sd_config_write_same(sdkp);
523 
524 	return count;
525 }
526 static DEVICE_ATTR_RW(max_write_same_blocks);
527 
528 static ssize_t
529 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
530 {
531 	struct scsi_disk *sdkp = to_scsi_disk(dev);
532 
533 	if (sdkp->device->type == TYPE_ZBC)
534 		return sprintf(buf, "host-managed\n");
535 	if (sdkp->zoned == 1)
536 		return sprintf(buf, "host-aware\n");
537 	if (sdkp->zoned == 2)
538 		return sprintf(buf, "drive-managed\n");
539 	return sprintf(buf, "none\n");
540 }
541 static DEVICE_ATTR_RO(zoned_cap);
542 
543 static ssize_t
544 max_retries_store(struct device *dev, struct device_attribute *attr,
545 		  const char *buf, size_t count)
546 {
547 	struct scsi_disk *sdkp = to_scsi_disk(dev);
548 	struct scsi_device *sdev = sdkp->device;
549 	int retries, err;
550 
551 	err = kstrtoint(buf, 10, &retries);
552 	if (err)
553 		return err;
554 
555 	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
556 		sdkp->max_retries = retries;
557 		return count;
558 	}
559 
560 	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
561 		    SD_MAX_RETRIES);
562 	return -EINVAL;
563 }
564 
565 static ssize_t
566 max_retries_show(struct device *dev, struct device_attribute *attr,
567 		 char *buf)
568 {
569 	struct scsi_disk *sdkp = to_scsi_disk(dev);
570 
571 	return sprintf(buf, "%d\n", sdkp->max_retries);
572 }
573 
574 static DEVICE_ATTR_RW(max_retries);
575 
576 static struct attribute *sd_disk_attrs[] = {
577 	&dev_attr_cache_type.attr,
578 	&dev_attr_FUA.attr,
579 	&dev_attr_allow_restart.attr,
580 	&dev_attr_manage_start_stop.attr,
581 	&dev_attr_protection_type.attr,
582 	&dev_attr_protection_mode.attr,
583 	&dev_attr_app_tag_own.attr,
584 	&dev_attr_thin_provisioning.attr,
585 	&dev_attr_provisioning_mode.attr,
586 	&dev_attr_zeroing_mode.attr,
587 	&dev_attr_max_write_same_blocks.attr,
588 	&dev_attr_max_medium_access_timeouts.attr,
589 	&dev_attr_zoned_cap.attr,
590 	&dev_attr_max_retries.attr,
591 	NULL,
592 };
593 ATTRIBUTE_GROUPS(sd_disk);
594 
595 static struct class sd_disk_class = {
596 	.name		= "scsi_disk",
597 	.owner		= THIS_MODULE,
598 	.dev_release	= scsi_disk_release,
599 	.dev_groups	= sd_disk_groups,
600 };
601 
602 static const struct dev_pm_ops sd_pm_ops = {
603 	.suspend		= sd_suspend_system,
604 	.resume			= sd_resume_system,
605 	.poweroff		= sd_suspend_system,
606 	.restore		= sd_resume_system,
607 	.runtime_suspend	= sd_suspend_runtime,
608 	.runtime_resume		= sd_resume_runtime,
609 };
610 
611 static struct scsi_driver sd_template = {
612 	.gendrv = {
613 		.name		= "sd",
614 		.owner		= THIS_MODULE,
615 		.probe		= sd_probe,
616 		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
617 		.remove		= sd_remove,
618 		.shutdown	= sd_shutdown,
619 		.pm		= &sd_pm_ops,
620 	},
621 	.rescan			= sd_rescan,
622 	.init_command		= sd_init_command,
623 	.uninit_command		= sd_uninit_command,
624 	.done			= sd_done,
625 	.eh_action		= sd_eh_action,
626 	.eh_reset		= sd_eh_reset,
627 };
628 
629 /*
630  * Don't request a new module, as that could deadlock in multipath
631  * environment.
632  */
633 static void sd_default_probe(dev_t devt)
634 {
635 }
636 
637 /*
638  * Device no to disk mapping:
639  *
640  *       major         disc2     disc  p1
641  *   |............|.............|....|....| <- dev_t
642  *    31        20 19          8 7  4 3  0
643  *
644  * Inside a major, we have 16k disks, however mapped non-
645  * contiguously. The first 16 disks are for major0, the next
646  * ones with major1, ... Disk 256 is for major0 again, disk 272
647  * for major1, ...
648  * As we stay compatible with our numbering scheme, we can reuse
649  * the well-know SCSI majors 8, 65--71, 136--143.
650  */
651 static int sd_major(int major_idx)
652 {
653 	switch (major_idx) {
654 	case 0:
655 		return SCSI_DISK0_MAJOR;
656 	case 1 ... 7:
657 		return SCSI_DISK1_MAJOR + major_idx - 1;
658 	case 8 ... 15:
659 		return SCSI_DISK8_MAJOR + major_idx - 8;
660 	default:
661 		BUG();
662 		return 0;	/* shut up gcc */
663 	}
664 }
665 
666 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
667 {
668 	struct scsi_disk *sdkp = NULL;
669 
670 	mutex_lock(&sd_ref_mutex);
671 
672 	if (disk->private_data) {
673 		sdkp = scsi_disk(disk);
674 		if (scsi_device_get(sdkp->device) == 0)
675 			get_device(&sdkp->dev);
676 		else
677 			sdkp = NULL;
678 	}
679 	mutex_unlock(&sd_ref_mutex);
680 	return sdkp;
681 }
682 
683 static void scsi_disk_put(struct scsi_disk *sdkp)
684 {
685 	struct scsi_device *sdev = sdkp->device;
686 
687 	mutex_lock(&sd_ref_mutex);
688 	put_device(&sdkp->dev);
689 	scsi_device_put(sdev);
690 	mutex_unlock(&sd_ref_mutex);
691 }
692 
693 #ifdef CONFIG_BLK_SED_OPAL
694 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
695 		size_t len, bool send)
696 {
697 	struct scsi_disk *sdkp = data;
698 	struct scsi_device *sdev = sdkp->device;
699 	u8 cdb[12] = { 0, };
700 	int ret;
701 
702 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
703 	cdb[1] = secp;
704 	put_unaligned_be16(spsp, &cdb[2]);
705 	put_unaligned_be32(len, &cdb[6]);
706 
707 	ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
708 		buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
709 		RQF_PM, NULL);
710 	return ret <= 0 ? ret : -EIO;
711 }
712 #endif /* CONFIG_BLK_SED_OPAL */
713 
714 /*
715  * Look up the DIX operation based on whether the command is read or
716  * write and whether dix and dif are enabled.
717  */
718 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
719 {
720 	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
721 	static const unsigned int ops[] = {	/* wrt dix dif */
722 		SCSI_PROT_NORMAL,		/*  0	0   0  */
723 		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
724 		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
725 		SCSI_PROT_READ_PASS,		/*  0	1   1  */
726 		SCSI_PROT_NORMAL,		/*  1	0   0  */
727 		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
728 		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
729 		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
730 	};
731 
732 	return ops[write << 2 | dix << 1 | dif];
733 }
734 
735 /*
736  * Returns a mask of the protection flags that are valid for a given DIX
737  * operation.
738  */
739 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
740 {
741 	static const unsigned int flag_mask[] = {
742 		[SCSI_PROT_NORMAL]		= 0,
743 
744 		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
745 						  SCSI_PROT_GUARD_CHECK |
746 						  SCSI_PROT_REF_CHECK |
747 						  SCSI_PROT_REF_INCREMENT,
748 
749 		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
750 						  SCSI_PROT_IP_CHECKSUM,
751 
752 		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
753 						  SCSI_PROT_GUARD_CHECK |
754 						  SCSI_PROT_REF_CHECK |
755 						  SCSI_PROT_REF_INCREMENT |
756 						  SCSI_PROT_IP_CHECKSUM,
757 
758 		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
759 						  SCSI_PROT_REF_INCREMENT,
760 
761 		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
762 						  SCSI_PROT_REF_CHECK |
763 						  SCSI_PROT_REF_INCREMENT |
764 						  SCSI_PROT_IP_CHECKSUM,
765 
766 		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
767 						  SCSI_PROT_GUARD_CHECK |
768 						  SCSI_PROT_REF_CHECK |
769 						  SCSI_PROT_REF_INCREMENT |
770 						  SCSI_PROT_IP_CHECKSUM,
771 	};
772 
773 	return flag_mask[prot_op];
774 }
775 
776 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
777 					   unsigned int dix, unsigned int dif)
778 {
779 	struct request *rq = scsi_cmd_to_rq(scmd);
780 	struct bio *bio = rq->bio;
781 	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
782 	unsigned int protect = 0;
783 
784 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
785 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
786 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
787 
788 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
789 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
790 	}
791 
792 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
793 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
794 
795 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
796 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
797 	}
798 
799 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
800 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
801 
802 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
803 			protect = 3 << 5;	/* Disable target PI checking */
804 		else
805 			protect = 1 << 5;	/* Enable target PI checking */
806 	}
807 
808 	scsi_set_prot_op(scmd, prot_op);
809 	scsi_set_prot_type(scmd, dif);
810 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
811 
812 	return protect;
813 }
814 
815 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
816 {
817 	struct request_queue *q = sdkp->disk->queue;
818 	unsigned int logical_block_size = sdkp->device->sector_size;
819 	unsigned int max_blocks = 0;
820 
821 	q->limits.discard_alignment =
822 		sdkp->unmap_alignment * logical_block_size;
823 	q->limits.discard_granularity =
824 		max(sdkp->physical_block_size,
825 		    sdkp->unmap_granularity * logical_block_size);
826 	sdkp->provisioning_mode = mode;
827 
828 	switch (mode) {
829 
830 	case SD_LBP_FULL:
831 	case SD_LBP_DISABLE:
832 		blk_queue_max_discard_sectors(q, 0);
833 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
834 		return;
835 
836 	case SD_LBP_UNMAP:
837 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
838 					  (u32)SD_MAX_WS16_BLOCKS);
839 		break;
840 
841 	case SD_LBP_WS16:
842 		if (sdkp->device->unmap_limit_for_ws)
843 			max_blocks = sdkp->max_unmap_blocks;
844 		else
845 			max_blocks = sdkp->max_ws_blocks;
846 
847 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
848 		break;
849 
850 	case SD_LBP_WS10:
851 		if (sdkp->device->unmap_limit_for_ws)
852 			max_blocks = sdkp->max_unmap_blocks;
853 		else
854 			max_blocks = sdkp->max_ws_blocks;
855 
856 		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
857 		break;
858 
859 	case SD_LBP_ZERO:
860 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
861 					  (u32)SD_MAX_WS10_BLOCKS);
862 		break;
863 	}
864 
865 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
866 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
867 }
868 
869 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
870 {
871 	struct scsi_device *sdp = cmd->device;
872 	struct request *rq = scsi_cmd_to_rq(cmd);
873 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
874 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
875 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
876 	unsigned int data_len = 24;
877 	char *buf;
878 
879 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
880 	if (!rq->special_vec.bv_page)
881 		return BLK_STS_RESOURCE;
882 	clear_highpage(rq->special_vec.bv_page);
883 	rq->special_vec.bv_offset = 0;
884 	rq->special_vec.bv_len = data_len;
885 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
886 
887 	cmd->cmd_len = 10;
888 	cmd->cmnd[0] = UNMAP;
889 	cmd->cmnd[8] = 24;
890 
891 	buf = bvec_virt(&rq->special_vec);
892 	put_unaligned_be16(6 + 16, &buf[0]);
893 	put_unaligned_be16(16, &buf[2]);
894 	put_unaligned_be64(lba, &buf[8]);
895 	put_unaligned_be32(nr_blocks, &buf[16]);
896 
897 	cmd->allowed = sdkp->max_retries;
898 	cmd->transfersize = data_len;
899 	rq->timeout = SD_TIMEOUT;
900 
901 	return scsi_alloc_sgtables(cmd);
902 }
903 
904 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
905 		bool unmap)
906 {
907 	struct scsi_device *sdp = cmd->device;
908 	struct request *rq = scsi_cmd_to_rq(cmd);
909 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
910 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
911 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
912 	u32 data_len = sdp->sector_size;
913 
914 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
915 	if (!rq->special_vec.bv_page)
916 		return BLK_STS_RESOURCE;
917 	clear_highpage(rq->special_vec.bv_page);
918 	rq->special_vec.bv_offset = 0;
919 	rq->special_vec.bv_len = data_len;
920 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
921 
922 	cmd->cmd_len = 16;
923 	cmd->cmnd[0] = WRITE_SAME_16;
924 	if (unmap)
925 		cmd->cmnd[1] = 0x8; /* UNMAP */
926 	put_unaligned_be64(lba, &cmd->cmnd[2]);
927 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
928 
929 	cmd->allowed = sdkp->max_retries;
930 	cmd->transfersize = data_len;
931 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
932 
933 	return scsi_alloc_sgtables(cmd);
934 }
935 
936 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
937 		bool unmap)
938 {
939 	struct scsi_device *sdp = cmd->device;
940 	struct request *rq = scsi_cmd_to_rq(cmd);
941 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
942 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
943 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
944 	u32 data_len = sdp->sector_size;
945 
946 	rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
947 	if (!rq->special_vec.bv_page)
948 		return BLK_STS_RESOURCE;
949 	clear_highpage(rq->special_vec.bv_page);
950 	rq->special_vec.bv_offset = 0;
951 	rq->special_vec.bv_len = data_len;
952 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
953 
954 	cmd->cmd_len = 10;
955 	cmd->cmnd[0] = WRITE_SAME;
956 	if (unmap)
957 		cmd->cmnd[1] = 0x8; /* UNMAP */
958 	put_unaligned_be32(lba, &cmd->cmnd[2]);
959 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
960 
961 	cmd->allowed = sdkp->max_retries;
962 	cmd->transfersize = data_len;
963 	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
964 
965 	return scsi_alloc_sgtables(cmd);
966 }
967 
968 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
969 {
970 	struct request *rq = scsi_cmd_to_rq(cmd);
971 	struct scsi_device *sdp = cmd->device;
972 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
973 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
974 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
975 
976 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
977 		switch (sdkp->zeroing_mode) {
978 		case SD_ZERO_WS16_UNMAP:
979 			return sd_setup_write_same16_cmnd(cmd, true);
980 		case SD_ZERO_WS10_UNMAP:
981 			return sd_setup_write_same10_cmnd(cmd, true);
982 		}
983 	}
984 
985 	if (sdp->no_write_same) {
986 		rq->rq_flags |= RQF_QUIET;
987 		return BLK_STS_TARGET;
988 	}
989 
990 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
991 		return sd_setup_write_same16_cmnd(cmd, false);
992 
993 	return sd_setup_write_same10_cmnd(cmd, false);
994 }
995 
996 static void sd_config_write_same(struct scsi_disk *sdkp)
997 {
998 	struct request_queue *q = sdkp->disk->queue;
999 	unsigned int logical_block_size = sdkp->device->sector_size;
1000 
1001 	if (sdkp->device->no_write_same) {
1002 		sdkp->max_ws_blocks = 0;
1003 		goto out;
1004 	}
1005 
1006 	/* Some devices can not handle block counts above 0xffff despite
1007 	 * supporting WRITE SAME(16). Consequently we default to 64k
1008 	 * blocks per I/O unless the device explicitly advertises a
1009 	 * bigger limit.
1010 	 */
1011 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1012 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1013 						   (u32)SD_MAX_WS16_BLOCKS);
1014 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1015 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1016 						   (u32)SD_MAX_WS10_BLOCKS);
1017 	else {
1018 		sdkp->device->no_write_same = 1;
1019 		sdkp->max_ws_blocks = 0;
1020 	}
1021 
1022 	if (sdkp->lbprz && sdkp->lbpws)
1023 		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1024 	else if (sdkp->lbprz && sdkp->lbpws10)
1025 		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1026 	else if (sdkp->max_ws_blocks)
1027 		sdkp->zeroing_mode = SD_ZERO_WS;
1028 	else
1029 		sdkp->zeroing_mode = SD_ZERO_WRITE;
1030 
1031 	if (sdkp->max_ws_blocks &&
1032 	    sdkp->physical_block_size > logical_block_size) {
1033 		/*
1034 		 * Reporting a maximum number of blocks that is not aligned
1035 		 * on the device physical size would cause a large write same
1036 		 * request to be split into physically unaligned chunks by
1037 		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1038 		 * even if the caller of these functions took care to align the
1039 		 * large request. So make sure the maximum reported is aligned
1040 		 * to the device physical block size. This is only an optional
1041 		 * optimization for regular disks, but this is mandatory to
1042 		 * avoid failure of large write same requests directed at
1043 		 * sequential write required zones of host-managed ZBC disks.
1044 		 */
1045 		sdkp->max_ws_blocks =
1046 			round_down(sdkp->max_ws_blocks,
1047 				   bytes_to_logical(sdkp->device,
1048 						    sdkp->physical_block_size));
1049 	}
1050 
1051 out:
1052 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1053 					 (logical_block_size >> 9));
1054 	blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1055 					 (logical_block_size >> 9));
1056 }
1057 
1058 /**
1059  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1060  * @cmd: command to prepare
1061  *
1062  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1063  * the preference indicated by the target device.
1064  **/
1065 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1066 {
1067 	struct request *rq = scsi_cmd_to_rq(cmd);
1068 	struct scsi_device *sdp = cmd->device;
1069 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1070 	struct bio *bio = rq->bio;
1071 	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1072 	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1073 	blk_status_t ret;
1074 
1075 	if (sdkp->device->no_write_same)
1076 		return BLK_STS_TARGET;
1077 
1078 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1079 
1080 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
1081 
1082 	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1083 		cmd->cmd_len = 16;
1084 		cmd->cmnd[0] = WRITE_SAME_16;
1085 		put_unaligned_be64(lba, &cmd->cmnd[2]);
1086 		put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1087 	} else {
1088 		cmd->cmd_len = 10;
1089 		cmd->cmnd[0] = WRITE_SAME;
1090 		put_unaligned_be32(lba, &cmd->cmnd[2]);
1091 		put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1092 	}
1093 
1094 	cmd->transfersize = sdp->sector_size;
1095 	cmd->allowed = sdkp->max_retries;
1096 
1097 	/*
1098 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
1099 	 * different from the amount of data actually written to the target.
1100 	 *
1101 	 * We set up __data_len to the amount of data transferred via the
1102 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1103 	 * to transfer a single sector of data first, but then reset it to
1104 	 * the amount of data to be written right after so that the I/O path
1105 	 * knows how much to actually write.
1106 	 */
1107 	rq->__data_len = sdp->sector_size;
1108 	ret = scsi_alloc_sgtables(cmd);
1109 	rq->__data_len = blk_rq_bytes(rq);
1110 
1111 	return ret;
1112 }
1113 
1114 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1115 {
1116 	struct request *rq = scsi_cmd_to_rq(cmd);
1117 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1118 
1119 	/* flush requests don't perform I/O, zero the S/G table */
1120 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1121 
1122 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1123 	cmd->cmd_len = 10;
1124 	cmd->transfersize = 0;
1125 	cmd->allowed = sdkp->max_retries;
1126 
1127 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1128 	return BLK_STS_OK;
1129 }
1130 
1131 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1132 				       sector_t lba, unsigned int nr_blocks,
1133 				       unsigned char flags)
1134 {
1135 	cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1136 	if (unlikely(cmd->cmnd == NULL))
1137 		return BLK_STS_RESOURCE;
1138 
1139 	cmd->cmd_len = SD_EXT_CDB_SIZE;
1140 	memset(cmd->cmnd, 0, cmd->cmd_len);
1141 
1142 	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1143 	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1144 	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1145 	cmd->cmnd[10] = flags;
1146 	put_unaligned_be64(lba, &cmd->cmnd[12]);
1147 	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1148 	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1149 
1150 	return BLK_STS_OK;
1151 }
1152 
1153 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1154 				       sector_t lba, unsigned int nr_blocks,
1155 				       unsigned char flags)
1156 {
1157 	cmd->cmd_len  = 16;
1158 	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1159 	cmd->cmnd[1]  = flags;
1160 	cmd->cmnd[14] = 0;
1161 	cmd->cmnd[15] = 0;
1162 	put_unaligned_be64(lba, &cmd->cmnd[2]);
1163 	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1164 
1165 	return BLK_STS_OK;
1166 }
1167 
1168 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1169 				       sector_t lba, unsigned int nr_blocks,
1170 				       unsigned char flags)
1171 {
1172 	cmd->cmd_len = 10;
1173 	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1174 	cmd->cmnd[1] = flags;
1175 	cmd->cmnd[6] = 0;
1176 	cmd->cmnd[9] = 0;
1177 	put_unaligned_be32(lba, &cmd->cmnd[2]);
1178 	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1179 
1180 	return BLK_STS_OK;
1181 }
1182 
1183 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1184 				      sector_t lba, unsigned int nr_blocks,
1185 				      unsigned char flags)
1186 {
1187 	/* Avoid that 0 blocks gets translated into 256 blocks. */
1188 	if (WARN_ON_ONCE(nr_blocks == 0))
1189 		return BLK_STS_IOERR;
1190 
1191 	if (unlikely(flags & 0x8)) {
1192 		/*
1193 		 * This happens only if this drive failed 10byte rw
1194 		 * command with ILLEGAL_REQUEST during operation and
1195 		 * thus turned off use_10_for_rw.
1196 		 */
1197 		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1198 		return BLK_STS_IOERR;
1199 	}
1200 
1201 	cmd->cmd_len = 6;
1202 	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1203 	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1204 	cmd->cmnd[2] = (lba >> 8) & 0xff;
1205 	cmd->cmnd[3] = lba & 0xff;
1206 	cmd->cmnd[4] = nr_blocks;
1207 	cmd->cmnd[5] = 0;
1208 
1209 	return BLK_STS_OK;
1210 }
1211 
1212 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1213 {
1214 	struct request *rq = scsi_cmd_to_rq(cmd);
1215 	struct scsi_device *sdp = cmd->device;
1216 	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1217 	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1218 	sector_t threshold;
1219 	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1220 	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1221 	bool write = rq_data_dir(rq) == WRITE;
1222 	unsigned char protect, fua;
1223 	blk_status_t ret;
1224 	unsigned int dif;
1225 	bool dix;
1226 
1227 	ret = scsi_alloc_sgtables(cmd);
1228 	if (ret != BLK_STS_OK)
1229 		return ret;
1230 
1231 	ret = BLK_STS_IOERR;
1232 	if (!scsi_device_online(sdp) || sdp->changed) {
1233 		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1234 		goto fail;
1235 	}
1236 
1237 	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1238 		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1239 		goto fail;
1240 	}
1241 
1242 	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1243 		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1244 		goto fail;
1245 	}
1246 
1247 	/*
1248 	 * Some SD card readers can't handle accesses which touch the
1249 	 * last one or two logical blocks. Split accesses as needed.
1250 	 */
1251 	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1252 
1253 	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1254 		if (lba < threshold) {
1255 			/* Access up to the threshold but not beyond */
1256 			nr_blocks = threshold - lba;
1257 		} else {
1258 			/* Access only a single logical block */
1259 			nr_blocks = 1;
1260 		}
1261 	}
1262 
1263 	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1264 		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1265 		if (ret)
1266 			goto fail;
1267 	}
1268 
1269 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1270 	dix = scsi_prot_sg_count(cmd);
1271 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1272 
1273 	if (dif || dix)
1274 		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1275 	else
1276 		protect = 0;
1277 
1278 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1279 		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1280 					 protect | fua);
1281 	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1282 		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1283 					 protect | fua);
1284 	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1285 		   sdp->use_10_for_rw || protect) {
1286 		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1287 					 protect | fua);
1288 	} else {
1289 		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1290 					protect | fua);
1291 	}
1292 
1293 	if (unlikely(ret != BLK_STS_OK))
1294 		goto fail;
1295 
1296 	/*
1297 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1298 	 * host adapter, it's safe to assume that we can at least transfer
1299 	 * this many bytes between each connect / disconnect.
1300 	 */
1301 	cmd->transfersize = sdp->sector_size;
1302 	cmd->underflow = nr_blocks << 9;
1303 	cmd->allowed = sdkp->max_retries;
1304 	cmd->sdb.length = nr_blocks * sdp->sector_size;
1305 
1306 	SCSI_LOG_HLQUEUE(1,
1307 			 scmd_printk(KERN_INFO, cmd,
1308 				     "%s: block=%llu, count=%d\n", __func__,
1309 				     (unsigned long long)blk_rq_pos(rq),
1310 				     blk_rq_sectors(rq)));
1311 	SCSI_LOG_HLQUEUE(2,
1312 			 scmd_printk(KERN_INFO, cmd,
1313 				     "%s %d/%u 512 byte blocks.\n",
1314 				     write ? "writing" : "reading", nr_blocks,
1315 				     blk_rq_sectors(rq)));
1316 
1317 	/*
1318 	 * This indicates that the command is ready from our end to be queued.
1319 	 */
1320 	return BLK_STS_OK;
1321 fail:
1322 	scsi_free_sgtables(cmd);
1323 	return ret;
1324 }
1325 
1326 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1327 {
1328 	struct request *rq = scsi_cmd_to_rq(cmd);
1329 
1330 	switch (req_op(rq)) {
1331 	case REQ_OP_DISCARD:
1332 		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1333 		case SD_LBP_UNMAP:
1334 			return sd_setup_unmap_cmnd(cmd);
1335 		case SD_LBP_WS16:
1336 			return sd_setup_write_same16_cmnd(cmd, true);
1337 		case SD_LBP_WS10:
1338 			return sd_setup_write_same10_cmnd(cmd, true);
1339 		case SD_LBP_ZERO:
1340 			return sd_setup_write_same10_cmnd(cmd, false);
1341 		default:
1342 			return BLK_STS_TARGET;
1343 		}
1344 	case REQ_OP_WRITE_ZEROES:
1345 		return sd_setup_write_zeroes_cmnd(cmd);
1346 	case REQ_OP_WRITE_SAME:
1347 		return sd_setup_write_same_cmnd(cmd);
1348 	case REQ_OP_FLUSH:
1349 		return sd_setup_flush_cmnd(cmd);
1350 	case REQ_OP_READ:
1351 	case REQ_OP_WRITE:
1352 	case REQ_OP_ZONE_APPEND:
1353 		return sd_setup_read_write_cmnd(cmd);
1354 	case REQ_OP_ZONE_RESET:
1355 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1356 						   false);
1357 	case REQ_OP_ZONE_RESET_ALL:
1358 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1359 						   true);
1360 	case REQ_OP_ZONE_OPEN:
1361 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1362 	case REQ_OP_ZONE_CLOSE:
1363 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1364 	case REQ_OP_ZONE_FINISH:
1365 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1366 	default:
1367 		WARN_ON_ONCE(1);
1368 		return BLK_STS_NOTSUPP;
1369 	}
1370 }
1371 
1372 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1373 {
1374 	struct request *rq = scsi_cmd_to_rq(SCpnt);
1375 	u8 *cmnd;
1376 
1377 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1378 		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1379 
1380 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1381 		cmnd = SCpnt->cmnd;
1382 		SCpnt->cmnd = NULL;
1383 		SCpnt->cmd_len = 0;
1384 		mempool_free(cmnd, sd_cdb_pool);
1385 	}
1386 }
1387 
1388 static bool sd_need_revalidate(struct block_device *bdev,
1389 		struct scsi_disk *sdkp)
1390 {
1391 	if (sdkp->device->removable || sdkp->write_prot) {
1392 		if (bdev_check_media_change(bdev))
1393 			return true;
1394 	}
1395 
1396 	/*
1397 	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1398 	 * nothing to do with partitions, BLKRRPART is used to force a full
1399 	 * revalidate after things like a format for historical reasons.
1400 	 */
1401 	return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1402 }
1403 
1404 /**
1405  *	sd_open - open a scsi disk device
1406  *	@bdev: Block device of the scsi disk to open
1407  *	@mode: FMODE_* mask
1408  *
1409  *	Returns 0 if successful. Returns a negated errno value in case
1410  *	of error.
1411  *
1412  *	Note: This can be called from a user context (e.g. fsck(1) )
1413  *	or from within the kernel (e.g. as a result of a mount(1) ).
1414  *	In the latter case @inode and @filp carry an abridged amount
1415  *	of information as noted above.
1416  *
1417  *	Locking: called with bdev->bd_disk->open_mutex held.
1418  **/
1419 static int sd_open(struct block_device *bdev, fmode_t mode)
1420 {
1421 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1422 	struct scsi_device *sdev;
1423 	int retval;
1424 
1425 	if (!sdkp)
1426 		return -ENXIO;
1427 
1428 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1429 
1430 	sdev = sdkp->device;
1431 
1432 	/*
1433 	 * If the device is in error recovery, wait until it is done.
1434 	 * If the device is offline, then disallow any access to it.
1435 	 */
1436 	retval = -ENXIO;
1437 	if (!scsi_block_when_processing_errors(sdev))
1438 		goto error_out;
1439 
1440 	if (sd_need_revalidate(bdev, sdkp))
1441 		sd_revalidate_disk(bdev->bd_disk);
1442 
1443 	/*
1444 	 * If the drive is empty, just let the open fail.
1445 	 */
1446 	retval = -ENOMEDIUM;
1447 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1448 		goto error_out;
1449 
1450 	/*
1451 	 * If the device has the write protect tab set, have the open fail
1452 	 * if the user expects to be able to write to the thing.
1453 	 */
1454 	retval = -EROFS;
1455 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1456 		goto error_out;
1457 
1458 	/*
1459 	 * It is possible that the disk changing stuff resulted in
1460 	 * the device being taken offline.  If this is the case,
1461 	 * report this to the user, and don't pretend that the
1462 	 * open actually succeeded.
1463 	 */
1464 	retval = -ENXIO;
1465 	if (!scsi_device_online(sdev))
1466 		goto error_out;
1467 
1468 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1469 		if (scsi_block_when_processing_errors(sdev))
1470 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1471 	}
1472 
1473 	return 0;
1474 
1475 error_out:
1476 	scsi_disk_put(sdkp);
1477 	return retval;
1478 }
1479 
1480 /**
1481  *	sd_release - invoked when the (last) close(2) is called on this
1482  *	scsi disk.
1483  *	@disk: disk to release
1484  *	@mode: FMODE_* mask
1485  *
1486  *	Returns 0.
1487  *
1488  *	Note: may block (uninterruptible) if error recovery is underway
1489  *	on this disk.
1490  *
1491  *	Locking: called with bdev->bd_disk->open_mutex held.
1492  **/
1493 static void sd_release(struct gendisk *disk, fmode_t mode)
1494 {
1495 	struct scsi_disk *sdkp = scsi_disk(disk);
1496 	struct scsi_device *sdev = sdkp->device;
1497 
1498 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1499 
1500 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1501 		if (scsi_block_when_processing_errors(sdev))
1502 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1503 	}
1504 
1505 	scsi_disk_put(sdkp);
1506 }
1507 
1508 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1509 {
1510 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1511 	struct scsi_device *sdp = sdkp->device;
1512 	struct Scsi_Host *host = sdp->host;
1513 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1514 	int diskinfo[4];
1515 
1516 	/* default to most commonly used values */
1517 	diskinfo[0] = 0x40;	/* 1 << 6 */
1518 	diskinfo[1] = 0x20;	/* 1 << 5 */
1519 	diskinfo[2] = capacity >> 11;
1520 
1521 	/* override with calculated, extended default, or driver values */
1522 	if (host->hostt->bios_param)
1523 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1524 	else
1525 		scsicam_bios_param(bdev, capacity, diskinfo);
1526 
1527 	geo->heads = diskinfo[0];
1528 	geo->sectors = diskinfo[1];
1529 	geo->cylinders = diskinfo[2];
1530 	return 0;
1531 }
1532 
1533 /**
1534  *	sd_ioctl - process an ioctl
1535  *	@bdev: target block device
1536  *	@mode: FMODE_* mask
1537  *	@cmd: ioctl command number
1538  *	@arg: this is third argument given to ioctl(2) system call.
1539  *	Often contains a pointer.
1540  *
1541  *	Returns 0 if successful (some ioctls return positive numbers on
1542  *	success as well). Returns a negated errno value in case of error.
1543  *
1544  *	Note: most ioctls are forward onto the block subsystem or further
1545  *	down in the scsi subsystem.
1546  **/
1547 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1548 		    unsigned int cmd, unsigned long arg)
1549 {
1550 	struct gendisk *disk = bdev->bd_disk;
1551 	struct scsi_disk *sdkp = scsi_disk(disk);
1552 	struct scsi_device *sdp = sdkp->device;
1553 	void __user *p = (void __user *)arg;
1554 	int error;
1555 
1556 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1557 				    "cmd=0x%x\n", disk->disk_name, cmd));
1558 
1559 	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1560 		return -ENOIOCTLCMD;
1561 
1562 	/*
1563 	 * If we are in the middle of error recovery, don't let anyone
1564 	 * else try and use this device.  Also, if error recovery fails, it
1565 	 * may try and take the device offline, in which case all further
1566 	 * access to the device is prohibited.
1567 	 */
1568 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1569 			(mode & FMODE_NDELAY) != 0);
1570 	if (error)
1571 		return error;
1572 
1573 	if (is_sed_ioctl(cmd))
1574 		return sed_ioctl(sdkp->opal_dev, cmd, p);
1575 	return scsi_ioctl(sdp, mode, cmd, p);
1576 }
1577 
1578 static void set_media_not_present(struct scsi_disk *sdkp)
1579 {
1580 	if (sdkp->media_present)
1581 		sdkp->device->changed = 1;
1582 
1583 	if (sdkp->device->removable) {
1584 		sdkp->media_present = 0;
1585 		sdkp->capacity = 0;
1586 	}
1587 }
1588 
1589 static int media_not_present(struct scsi_disk *sdkp,
1590 			     struct scsi_sense_hdr *sshdr)
1591 {
1592 	if (!scsi_sense_valid(sshdr))
1593 		return 0;
1594 
1595 	/* not invoked for commands that could return deferred errors */
1596 	switch (sshdr->sense_key) {
1597 	case UNIT_ATTENTION:
1598 	case NOT_READY:
1599 		/* medium not present */
1600 		if (sshdr->asc == 0x3A) {
1601 			set_media_not_present(sdkp);
1602 			return 1;
1603 		}
1604 	}
1605 	return 0;
1606 }
1607 
1608 /**
1609  *	sd_check_events - check media events
1610  *	@disk: kernel device descriptor
1611  *	@clearing: disk events currently being cleared
1612  *
1613  *	Returns mask of DISK_EVENT_*.
1614  *
1615  *	Note: this function is invoked from the block subsystem.
1616  **/
1617 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1618 {
1619 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1620 	struct scsi_device *sdp;
1621 	int retval;
1622 	bool disk_changed;
1623 
1624 	if (!sdkp)
1625 		return 0;
1626 
1627 	sdp = sdkp->device;
1628 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1629 
1630 	/*
1631 	 * If the device is offline, don't send any commands - just pretend as
1632 	 * if the command failed.  If the device ever comes back online, we
1633 	 * can deal with it then.  It is only because of unrecoverable errors
1634 	 * that we would ever take a device offline in the first place.
1635 	 */
1636 	if (!scsi_device_online(sdp)) {
1637 		set_media_not_present(sdkp);
1638 		goto out;
1639 	}
1640 
1641 	/*
1642 	 * Using TEST_UNIT_READY enables differentiation between drive with
1643 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1644 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1645 	 *
1646 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1647 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1648 	 * sd_revalidate() is called.
1649 	 */
1650 	if (scsi_block_when_processing_errors(sdp)) {
1651 		struct scsi_sense_hdr sshdr = { 0, };
1652 
1653 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1654 					      &sshdr);
1655 
1656 		/* failed to execute TUR, assume media not present */
1657 		if (retval < 0 || host_byte(retval)) {
1658 			set_media_not_present(sdkp);
1659 			goto out;
1660 		}
1661 
1662 		if (media_not_present(sdkp, &sshdr))
1663 			goto out;
1664 	}
1665 
1666 	/*
1667 	 * For removable scsi disk we have to recognise the presence
1668 	 * of a disk in the drive.
1669 	 */
1670 	if (!sdkp->media_present)
1671 		sdp->changed = 1;
1672 	sdkp->media_present = 1;
1673 out:
1674 	/*
1675 	 * sdp->changed is set under the following conditions:
1676 	 *
1677 	 *	Medium present state has changed in either direction.
1678 	 *	Device has indicated UNIT_ATTENTION.
1679 	 */
1680 	disk_changed = sdp->changed;
1681 	sdp->changed = 0;
1682 	scsi_disk_put(sdkp);
1683 	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1684 }
1685 
1686 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1687 {
1688 	int retries, res;
1689 	struct scsi_device *sdp = sdkp->device;
1690 	const int timeout = sdp->request_queue->rq_timeout
1691 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1692 	struct scsi_sense_hdr my_sshdr;
1693 
1694 	if (!scsi_device_online(sdp))
1695 		return -ENODEV;
1696 
1697 	/* caller might not be interested in sense, but we need it */
1698 	if (!sshdr)
1699 		sshdr = &my_sshdr;
1700 
1701 	for (retries = 3; retries > 0; --retries) {
1702 		unsigned char cmd[10] = { 0 };
1703 
1704 		cmd[0] = SYNCHRONIZE_CACHE;
1705 		/*
1706 		 * Leave the rest of the command zero to indicate
1707 		 * flush everything.
1708 		 */
1709 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1710 				timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1711 		if (res == 0)
1712 			break;
1713 	}
1714 
1715 	if (res) {
1716 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1717 
1718 		if (res < 0)
1719 			return res;
1720 
1721 		if (scsi_status_is_check_condition(res) &&
1722 		    scsi_sense_valid(sshdr)) {
1723 			sd_print_sense_hdr(sdkp, sshdr);
1724 
1725 			/* we need to evaluate the error return  */
1726 			if (sshdr->asc == 0x3a ||	/* medium not present */
1727 			    sshdr->asc == 0x20 ||	/* invalid command */
1728 			    (sshdr->asc == 0x74 && sshdr->ascq == 0x71))	/* drive is password locked */
1729 				/* this is no error here */
1730 				return 0;
1731 		}
1732 
1733 		switch (host_byte(res)) {
1734 		/* ignore errors due to racing a disconnection */
1735 		case DID_BAD_TARGET:
1736 		case DID_NO_CONNECT:
1737 			return 0;
1738 		/* signal the upper layer it might try again */
1739 		case DID_BUS_BUSY:
1740 		case DID_IMM_RETRY:
1741 		case DID_REQUEUE:
1742 		case DID_SOFT_ERROR:
1743 			return -EBUSY;
1744 		default:
1745 			return -EIO;
1746 		}
1747 	}
1748 	return 0;
1749 }
1750 
1751 static void sd_rescan(struct device *dev)
1752 {
1753 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1754 
1755 	sd_revalidate_disk(sdkp->disk);
1756 }
1757 
1758 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1759 		enum blk_unique_id type)
1760 {
1761 	struct scsi_device *sdev = scsi_disk(disk)->device;
1762 	const struct scsi_vpd *vpd;
1763 	const unsigned char *d;
1764 	int ret = -ENXIO, len;
1765 
1766 	rcu_read_lock();
1767 	vpd = rcu_dereference(sdev->vpd_pg83);
1768 	if (!vpd)
1769 		goto out_unlock;
1770 
1771 	ret = -EINVAL;
1772 	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1773 		/* we only care about designators with LU association */
1774 		if (((d[1] >> 4) & 0x3) != 0x00)
1775 			continue;
1776 		if ((d[1] & 0xf) != type)
1777 			continue;
1778 
1779 		/*
1780 		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1781 		 * keep looking as one with more entropy might still show up.
1782 		 */
1783 		len = d[3];
1784 		if (len != 8 && len != 12 && len != 16)
1785 			continue;
1786 		ret = len;
1787 		memcpy(id, d + 4, len);
1788 		if (len == 16)
1789 			break;
1790 	}
1791 out_unlock:
1792 	rcu_read_unlock();
1793 	return ret;
1794 }
1795 
1796 static char sd_pr_type(enum pr_type type)
1797 {
1798 	switch (type) {
1799 	case PR_WRITE_EXCLUSIVE:
1800 		return 0x01;
1801 	case PR_EXCLUSIVE_ACCESS:
1802 		return 0x03;
1803 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1804 		return 0x05;
1805 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1806 		return 0x06;
1807 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1808 		return 0x07;
1809 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1810 		return 0x08;
1811 	default:
1812 		return 0;
1813 	}
1814 };
1815 
1816 static int sd_pr_command(struct block_device *bdev, u8 sa,
1817 		u64 key, u64 sa_key, u8 type, u8 flags)
1818 {
1819 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1820 	struct scsi_device *sdev = sdkp->device;
1821 	struct scsi_sense_hdr sshdr;
1822 	int result;
1823 	u8 cmd[16] = { 0, };
1824 	u8 data[24] = { 0, };
1825 
1826 	cmd[0] = PERSISTENT_RESERVE_OUT;
1827 	cmd[1] = sa;
1828 	cmd[2] = type;
1829 	put_unaligned_be32(sizeof(data), &cmd[5]);
1830 
1831 	put_unaligned_be64(key, &data[0]);
1832 	put_unaligned_be64(sa_key, &data[8]);
1833 	data[20] = flags;
1834 
1835 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1836 			&sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1837 
1838 	if (scsi_status_is_check_condition(result) &&
1839 	    scsi_sense_valid(&sshdr)) {
1840 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1841 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1842 	}
1843 
1844 	return result;
1845 }
1846 
1847 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1848 		u32 flags)
1849 {
1850 	if (flags & ~PR_FL_IGNORE_KEY)
1851 		return -EOPNOTSUPP;
1852 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1853 			old_key, new_key, 0,
1854 			(1 << 0) /* APTPL */);
1855 }
1856 
1857 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1858 		u32 flags)
1859 {
1860 	if (flags)
1861 		return -EOPNOTSUPP;
1862 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1863 }
1864 
1865 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1866 {
1867 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1868 }
1869 
1870 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1871 		enum pr_type type, bool abort)
1872 {
1873 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1874 			     sd_pr_type(type), 0);
1875 }
1876 
1877 static int sd_pr_clear(struct block_device *bdev, u64 key)
1878 {
1879 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1880 }
1881 
1882 static const struct pr_ops sd_pr_ops = {
1883 	.pr_register	= sd_pr_register,
1884 	.pr_reserve	= sd_pr_reserve,
1885 	.pr_release	= sd_pr_release,
1886 	.pr_preempt	= sd_pr_preempt,
1887 	.pr_clear	= sd_pr_clear,
1888 };
1889 
1890 static const struct block_device_operations sd_fops = {
1891 	.owner			= THIS_MODULE,
1892 	.open			= sd_open,
1893 	.release		= sd_release,
1894 	.ioctl			= sd_ioctl,
1895 	.getgeo			= sd_getgeo,
1896 	.compat_ioctl		= blkdev_compat_ptr_ioctl,
1897 	.check_events		= sd_check_events,
1898 	.unlock_native_capacity	= sd_unlock_native_capacity,
1899 	.report_zones		= sd_zbc_report_zones,
1900 	.get_unique_id		= sd_get_unique_id,
1901 	.pr_ops			= &sd_pr_ops,
1902 };
1903 
1904 /**
1905  *	sd_eh_reset - reset error handling callback
1906  *	@scmd:		sd-issued command that has failed
1907  *
1908  *	This function is called by the SCSI midlayer before starting
1909  *	SCSI EH. When counting medium access failures we have to be
1910  *	careful to register it only only once per device and SCSI EH run;
1911  *	there might be several timed out commands which will cause the
1912  *	'max_medium_access_timeouts' counter to trigger after the first
1913  *	SCSI EH run already and set the device to offline.
1914  *	So this function resets the internal counter before starting SCSI EH.
1915  **/
1916 static void sd_eh_reset(struct scsi_cmnd *scmd)
1917 {
1918 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1919 
1920 	/* New SCSI EH run, reset gate variable */
1921 	sdkp->ignore_medium_access_errors = false;
1922 }
1923 
1924 /**
1925  *	sd_eh_action - error handling callback
1926  *	@scmd:		sd-issued command that has failed
1927  *	@eh_disp:	The recovery disposition suggested by the midlayer
1928  *
1929  *	This function is called by the SCSI midlayer upon completion of an
1930  *	error test command (currently TEST UNIT READY). The result of sending
1931  *	the eh command is passed in eh_disp.  We're looking for devices that
1932  *	fail medium access commands but are OK with non access commands like
1933  *	test unit ready (so wrongly see the device as having a successful
1934  *	recovery)
1935  **/
1936 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1937 {
1938 	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1939 	struct scsi_device *sdev = scmd->device;
1940 
1941 	if (!scsi_device_online(sdev) ||
1942 	    !scsi_medium_access_command(scmd) ||
1943 	    host_byte(scmd->result) != DID_TIME_OUT ||
1944 	    eh_disp != SUCCESS)
1945 		return eh_disp;
1946 
1947 	/*
1948 	 * The device has timed out executing a medium access command.
1949 	 * However, the TEST UNIT READY command sent during error
1950 	 * handling completed successfully. Either the device is in the
1951 	 * process of recovering or has it suffered an internal failure
1952 	 * that prevents access to the storage medium.
1953 	 */
1954 	if (!sdkp->ignore_medium_access_errors) {
1955 		sdkp->medium_access_timed_out++;
1956 		sdkp->ignore_medium_access_errors = true;
1957 	}
1958 
1959 	/*
1960 	 * If the device keeps failing read/write commands but TEST UNIT
1961 	 * READY always completes successfully we assume that medium
1962 	 * access is no longer possible and take the device offline.
1963 	 */
1964 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1965 		scmd_printk(KERN_ERR, scmd,
1966 			    "Medium access timeout failure. Offlining disk!\n");
1967 		mutex_lock(&sdev->state_mutex);
1968 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1969 		mutex_unlock(&sdev->state_mutex);
1970 
1971 		return SUCCESS;
1972 	}
1973 
1974 	return eh_disp;
1975 }
1976 
1977 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1978 {
1979 	struct request *req = scsi_cmd_to_rq(scmd);
1980 	struct scsi_device *sdev = scmd->device;
1981 	unsigned int transferred, good_bytes;
1982 	u64 start_lba, end_lba, bad_lba;
1983 
1984 	/*
1985 	 * Some commands have a payload smaller than the device logical
1986 	 * block size (e.g. INQUIRY on a 4K disk).
1987 	 */
1988 	if (scsi_bufflen(scmd) <= sdev->sector_size)
1989 		return 0;
1990 
1991 	/* Check if we have a 'bad_lba' information */
1992 	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1993 				     SCSI_SENSE_BUFFERSIZE,
1994 				     &bad_lba))
1995 		return 0;
1996 
1997 	/*
1998 	 * If the bad lba was reported incorrectly, we have no idea where
1999 	 * the error is.
2000 	 */
2001 	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2002 	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2003 	if (bad_lba < start_lba || bad_lba >= end_lba)
2004 		return 0;
2005 
2006 	/*
2007 	 * resid is optional but mostly filled in.  When it's unused,
2008 	 * its value is zero, so we assume the whole buffer transferred
2009 	 */
2010 	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2011 
2012 	/* This computation should always be done in terms of the
2013 	 * resolution of the device's medium.
2014 	 */
2015 	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2016 
2017 	return min(good_bytes, transferred);
2018 }
2019 
2020 /**
2021  *	sd_done - bottom half handler: called when the lower level
2022  *	driver has completed (successfully or otherwise) a scsi command.
2023  *	@SCpnt: mid-level's per command structure.
2024  *
2025  *	Note: potentially run from within an ISR. Must not block.
2026  **/
2027 static int sd_done(struct scsi_cmnd *SCpnt)
2028 {
2029 	int result = SCpnt->result;
2030 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2031 	unsigned int sector_size = SCpnt->device->sector_size;
2032 	unsigned int resid;
2033 	struct scsi_sense_hdr sshdr;
2034 	struct request *req = scsi_cmd_to_rq(SCpnt);
2035 	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2036 	int sense_valid = 0;
2037 	int sense_deferred = 0;
2038 
2039 	switch (req_op(req)) {
2040 	case REQ_OP_DISCARD:
2041 	case REQ_OP_WRITE_ZEROES:
2042 	case REQ_OP_WRITE_SAME:
2043 	case REQ_OP_ZONE_RESET:
2044 	case REQ_OP_ZONE_RESET_ALL:
2045 	case REQ_OP_ZONE_OPEN:
2046 	case REQ_OP_ZONE_CLOSE:
2047 	case REQ_OP_ZONE_FINISH:
2048 		if (!result) {
2049 			good_bytes = blk_rq_bytes(req);
2050 			scsi_set_resid(SCpnt, 0);
2051 		} else {
2052 			good_bytes = 0;
2053 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2054 		}
2055 		break;
2056 	default:
2057 		/*
2058 		 * In case of bogus fw or device, we could end up having
2059 		 * an unaligned partial completion. Check this here and force
2060 		 * alignment.
2061 		 */
2062 		resid = scsi_get_resid(SCpnt);
2063 		if (resid & (sector_size - 1)) {
2064 			sd_printk(KERN_INFO, sdkp,
2065 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2066 				resid, sector_size);
2067 			scsi_print_command(SCpnt);
2068 			resid = min(scsi_bufflen(SCpnt),
2069 				    round_up(resid, sector_size));
2070 			scsi_set_resid(SCpnt, resid);
2071 		}
2072 	}
2073 
2074 	if (result) {
2075 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2076 		if (sense_valid)
2077 			sense_deferred = scsi_sense_is_deferred(&sshdr);
2078 	}
2079 	sdkp->medium_access_timed_out = 0;
2080 
2081 	if (!scsi_status_is_check_condition(result) &&
2082 	    (!sense_valid || sense_deferred))
2083 		goto out;
2084 
2085 	switch (sshdr.sense_key) {
2086 	case HARDWARE_ERROR:
2087 	case MEDIUM_ERROR:
2088 		good_bytes = sd_completed_bytes(SCpnt);
2089 		break;
2090 	case RECOVERED_ERROR:
2091 		good_bytes = scsi_bufflen(SCpnt);
2092 		break;
2093 	case NO_SENSE:
2094 		/* This indicates a false check condition, so ignore it.  An
2095 		 * unknown amount of data was transferred so treat it as an
2096 		 * error.
2097 		 */
2098 		SCpnt->result = 0;
2099 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2100 		break;
2101 	case ABORTED_COMMAND:
2102 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2103 			good_bytes = sd_completed_bytes(SCpnt);
2104 		break;
2105 	case ILLEGAL_REQUEST:
2106 		switch (sshdr.asc) {
2107 		case 0x10:	/* DIX: Host detected corruption */
2108 			good_bytes = sd_completed_bytes(SCpnt);
2109 			break;
2110 		case 0x20:	/* INVALID COMMAND OPCODE */
2111 		case 0x24:	/* INVALID FIELD IN CDB */
2112 			switch (SCpnt->cmnd[0]) {
2113 			case UNMAP:
2114 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2115 				break;
2116 			case WRITE_SAME_16:
2117 			case WRITE_SAME:
2118 				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2119 					sd_config_discard(sdkp, SD_LBP_DISABLE);
2120 				} else {
2121 					sdkp->device->no_write_same = 1;
2122 					sd_config_write_same(sdkp);
2123 					req->rq_flags |= RQF_QUIET;
2124 				}
2125 				break;
2126 			}
2127 		}
2128 		break;
2129 	default:
2130 		break;
2131 	}
2132 
2133  out:
2134 	if (sd_is_zoned(sdkp))
2135 		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2136 
2137 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2138 					   "sd_done: completed %d of %d bytes\n",
2139 					   good_bytes, scsi_bufflen(SCpnt)));
2140 
2141 	return good_bytes;
2142 }
2143 
2144 /*
2145  * spinup disk - called only in sd_revalidate_disk()
2146  */
2147 static void
2148 sd_spinup_disk(struct scsi_disk *sdkp)
2149 {
2150 	unsigned char cmd[10];
2151 	unsigned long spintime_expire = 0;
2152 	int retries, spintime;
2153 	unsigned int the_result;
2154 	struct scsi_sense_hdr sshdr;
2155 	int sense_valid = 0;
2156 
2157 	spintime = 0;
2158 
2159 	/* Spin up drives, as required.  Only do this at boot time */
2160 	/* Spinup needs to be done for module loads too. */
2161 	do {
2162 		retries = 0;
2163 
2164 		do {
2165 			bool media_was_present = sdkp->media_present;
2166 
2167 			cmd[0] = TEST_UNIT_READY;
2168 			memset((void *) &cmd[1], 0, 9);
2169 
2170 			the_result = scsi_execute_req(sdkp->device, cmd,
2171 						      DMA_NONE, NULL, 0,
2172 						      &sshdr, SD_TIMEOUT,
2173 						      sdkp->max_retries, NULL);
2174 
2175 			/*
2176 			 * If the drive has indicated to us that it
2177 			 * doesn't have any media in it, don't bother
2178 			 * with any more polling.
2179 			 */
2180 			if (media_not_present(sdkp, &sshdr)) {
2181 				if (media_was_present)
2182 					sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2183 				return;
2184 			}
2185 
2186 			if (the_result)
2187 				sense_valid = scsi_sense_valid(&sshdr);
2188 			retries++;
2189 		} while (retries < 3 &&
2190 			 (!scsi_status_is_good(the_result) ||
2191 			  (scsi_status_is_check_condition(the_result) &&
2192 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2193 
2194 		if (!scsi_status_is_check_condition(the_result)) {
2195 			/* no sense, TUR either succeeded or failed
2196 			 * with a status error */
2197 			if(!spintime && !scsi_status_is_good(the_result)) {
2198 				sd_print_result(sdkp, "Test Unit Ready failed",
2199 						the_result);
2200 			}
2201 			break;
2202 		}
2203 
2204 		/*
2205 		 * The device does not want the automatic start to be issued.
2206 		 */
2207 		if (sdkp->device->no_start_on_add)
2208 			break;
2209 
2210 		if (sense_valid && sshdr.sense_key == NOT_READY) {
2211 			if (sshdr.asc == 4 && sshdr.ascq == 3)
2212 				break;	/* manual intervention required */
2213 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2214 				break;	/* standby */
2215 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2216 				break;	/* unavailable */
2217 			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2218 				break;	/* sanitize in progress */
2219 			/*
2220 			 * Issue command to spin up drive when not ready
2221 			 */
2222 			if (!spintime) {
2223 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2224 				cmd[0] = START_STOP;
2225 				cmd[1] = 1;	/* Return immediately */
2226 				memset((void *) &cmd[2], 0, 8);
2227 				cmd[4] = 1;	/* Start spin cycle */
2228 				if (sdkp->device->start_stop_pwr_cond)
2229 					cmd[4] |= 1 << 4;
2230 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2231 						 NULL, 0, &sshdr,
2232 						 SD_TIMEOUT, sdkp->max_retries,
2233 						 NULL);
2234 				spintime_expire = jiffies + 100 * HZ;
2235 				spintime = 1;
2236 			}
2237 			/* Wait 1 second for next try */
2238 			msleep(1000);
2239 			printk(KERN_CONT ".");
2240 
2241 		/*
2242 		 * Wait for USB flash devices with slow firmware.
2243 		 * Yes, this sense key/ASC combination shouldn't
2244 		 * occur here.  It's characteristic of these devices.
2245 		 */
2246 		} else if (sense_valid &&
2247 				sshdr.sense_key == UNIT_ATTENTION &&
2248 				sshdr.asc == 0x28) {
2249 			if (!spintime) {
2250 				spintime_expire = jiffies + 5 * HZ;
2251 				spintime = 1;
2252 			}
2253 			/* Wait 1 second for next try */
2254 			msleep(1000);
2255 		} else {
2256 			/* we don't understand the sense code, so it's
2257 			 * probably pointless to loop */
2258 			if(!spintime) {
2259 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2260 				sd_print_sense_hdr(sdkp, &sshdr);
2261 			}
2262 			break;
2263 		}
2264 
2265 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2266 
2267 	if (spintime) {
2268 		if (scsi_status_is_good(the_result))
2269 			printk(KERN_CONT "ready\n");
2270 		else
2271 			printk(KERN_CONT "not responding...\n");
2272 	}
2273 }
2274 
2275 /*
2276  * Determine whether disk supports Data Integrity Field.
2277  */
2278 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2279 {
2280 	struct scsi_device *sdp = sdkp->device;
2281 	u8 type;
2282 	int ret = 0;
2283 
2284 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2285 		sdkp->protection_type = 0;
2286 		return ret;
2287 	}
2288 
2289 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2290 
2291 	if (type > T10_PI_TYPE3_PROTECTION)
2292 		ret = -ENODEV;
2293 	else if (scsi_host_dif_capable(sdp->host, type))
2294 		ret = 1;
2295 
2296 	if (sdkp->first_scan || type != sdkp->protection_type)
2297 		switch (ret) {
2298 		case -ENODEV:
2299 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2300 				  " protection type %u. Disabling disk!\n",
2301 				  type);
2302 			break;
2303 		case 1:
2304 			sd_printk(KERN_NOTICE, sdkp,
2305 				  "Enabling DIF Type %u protection\n", type);
2306 			break;
2307 		case 0:
2308 			sd_printk(KERN_NOTICE, sdkp,
2309 				  "Disabling DIF Type %u protection\n", type);
2310 			break;
2311 		}
2312 
2313 	sdkp->protection_type = type;
2314 
2315 	return ret;
2316 }
2317 
2318 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2319 			struct scsi_sense_hdr *sshdr, int sense_valid,
2320 			int the_result)
2321 {
2322 	if (sense_valid)
2323 		sd_print_sense_hdr(sdkp, sshdr);
2324 	else
2325 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2326 
2327 	/*
2328 	 * Set dirty bit for removable devices if not ready -
2329 	 * sometimes drives will not report this properly.
2330 	 */
2331 	if (sdp->removable &&
2332 	    sense_valid && sshdr->sense_key == NOT_READY)
2333 		set_media_not_present(sdkp);
2334 
2335 	/*
2336 	 * We used to set media_present to 0 here to indicate no media
2337 	 * in the drive, but some drives fail read capacity even with
2338 	 * media present, so we can't do that.
2339 	 */
2340 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2341 }
2342 
2343 #define RC16_LEN 32
2344 #if RC16_LEN > SD_BUF_SIZE
2345 #error RC16_LEN must not be more than SD_BUF_SIZE
2346 #endif
2347 
2348 #define READ_CAPACITY_RETRIES_ON_RESET	10
2349 
2350 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2351 						unsigned char *buffer)
2352 {
2353 	unsigned char cmd[16];
2354 	struct scsi_sense_hdr sshdr;
2355 	int sense_valid = 0;
2356 	int the_result;
2357 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2358 	unsigned int alignment;
2359 	unsigned long long lba;
2360 	unsigned sector_size;
2361 
2362 	if (sdp->no_read_capacity_16)
2363 		return -EINVAL;
2364 
2365 	do {
2366 		memset(cmd, 0, 16);
2367 		cmd[0] = SERVICE_ACTION_IN_16;
2368 		cmd[1] = SAI_READ_CAPACITY_16;
2369 		cmd[13] = RC16_LEN;
2370 		memset(buffer, 0, RC16_LEN);
2371 
2372 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2373 					buffer, RC16_LEN, &sshdr,
2374 					SD_TIMEOUT, sdkp->max_retries, NULL);
2375 
2376 		if (media_not_present(sdkp, &sshdr))
2377 			return -ENODEV;
2378 
2379 		if (the_result > 0) {
2380 			sense_valid = scsi_sense_valid(&sshdr);
2381 			if (sense_valid &&
2382 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2383 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2384 			    sshdr.ascq == 0x00)
2385 				/* Invalid Command Operation Code or
2386 				 * Invalid Field in CDB, just retry
2387 				 * silently with RC10 */
2388 				return -EINVAL;
2389 			if (sense_valid &&
2390 			    sshdr.sense_key == UNIT_ATTENTION &&
2391 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2392 				/* Device reset might occur several times,
2393 				 * give it one more chance */
2394 				if (--reset_retries > 0)
2395 					continue;
2396 		}
2397 		retries--;
2398 
2399 	} while (the_result && retries);
2400 
2401 	if (the_result) {
2402 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2403 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2404 		return -EINVAL;
2405 	}
2406 
2407 	sector_size = get_unaligned_be32(&buffer[8]);
2408 	lba = get_unaligned_be64(&buffer[0]);
2409 
2410 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2411 		sdkp->capacity = 0;
2412 		return -ENODEV;
2413 	}
2414 
2415 	/* Logical blocks per physical block exponent */
2416 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2417 
2418 	/* RC basis */
2419 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2420 
2421 	/* Lowest aligned logical block */
2422 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2423 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2424 	if (alignment && sdkp->first_scan)
2425 		sd_printk(KERN_NOTICE, sdkp,
2426 			  "physical block alignment offset: %u\n", alignment);
2427 
2428 	if (buffer[14] & 0x80) { /* LBPME */
2429 		sdkp->lbpme = 1;
2430 
2431 		if (buffer[14] & 0x40) /* LBPRZ */
2432 			sdkp->lbprz = 1;
2433 
2434 		sd_config_discard(sdkp, SD_LBP_WS16);
2435 	}
2436 
2437 	sdkp->capacity = lba + 1;
2438 	return sector_size;
2439 }
2440 
2441 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2442 						unsigned char *buffer)
2443 {
2444 	unsigned char cmd[16];
2445 	struct scsi_sense_hdr sshdr;
2446 	int sense_valid = 0;
2447 	int the_result;
2448 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2449 	sector_t lba;
2450 	unsigned sector_size;
2451 
2452 	do {
2453 		cmd[0] = READ_CAPACITY;
2454 		memset(&cmd[1], 0, 9);
2455 		memset(buffer, 0, 8);
2456 
2457 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2458 					buffer, 8, &sshdr,
2459 					SD_TIMEOUT, sdkp->max_retries, NULL);
2460 
2461 		if (media_not_present(sdkp, &sshdr))
2462 			return -ENODEV;
2463 
2464 		if (the_result > 0) {
2465 			sense_valid = scsi_sense_valid(&sshdr);
2466 			if (sense_valid &&
2467 			    sshdr.sense_key == UNIT_ATTENTION &&
2468 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2469 				/* Device reset might occur several times,
2470 				 * give it one more chance */
2471 				if (--reset_retries > 0)
2472 					continue;
2473 		}
2474 		retries--;
2475 
2476 	} while (the_result && retries);
2477 
2478 	if (the_result) {
2479 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2480 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2481 		return -EINVAL;
2482 	}
2483 
2484 	sector_size = get_unaligned_be32(&buffer[4]);
2485 	lba = get_unaligned_be32(&buffer[0]);
2486 
2487 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2488 		/* Some buggy (usb cardreader) devices return an lba of
2489 		   0xffffffff when the want to report a size of 0 (with
2490 		   which they really mean no media is present) */
2491 		sdkp->capacity = 0;
2492 		sdkp->physical_block_size = sector_size;
2493 		return sector_size;
2494 	}
2495 
2496 	sdkp->capacity = lba + 1;
2497 	sdkp->physical_block_size = sector_size;
2498 	return sector_size;
2499 }
2500 
2501 static int sd_try_rc16_first(struct scsi_device *sdp)
2502 {
2503 	if (sdp->host->max_cmd_len < 16)
2504 		return 0;
2505 	if (sdp->try_rc_10_first)
2506 		return 0;
2507 	if (sdp->scsi_level > SCSI_SPC_2)
2508 		return 1;
2509 	if (scsi_device_protection(sdp))
2510 		return 1;
2511 	return 0;
2512 }
2513 
2514 /*
2515  * read disk capacity
2516  */
2517 static void
2518 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2519 {
2520 	int sector_size;
2521 	struct scsi_device *sdp = sdkp->device;
2522 
2523 	if (sd_try_rc16_first(sdp)) {
2524 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2525 		if (sector_size == -EOVERFLOW)
2526 			goto got_data;
2527 		if (sector_size == -ENODEV)
2528 			return;
2529 		if (sector_size < 0)
2530 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2531 		if (sector_size < 0)
2532 			return;
2533 	} else {
2534 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2535 		if (sector_size == -EOVERFLOW)
2536 			goto got_data;
2537 		if (sector_size < 0)
2538 			return;
2539 		if ((sizeof(sdkp->capacity) > 4) &&
2540 		    (sdkp->capacity > 0xffffffffULL)) {
2541 			int old_sector_size = sector_size;
2542 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2543 					"Trying to use READ CAPACITY(16).\n");
2544 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2545 			if (sector_size < 0) {
2546 				sd_printk(KERN_NOTICE, sdkp,
2547 					"Using 0xffffffff as device size\n");
2548 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2549 				sector_size = old_sector_size;
2550 				goto got_data;
2551 			}
2552 			/* Remember that READ CAPACITY(16) succeeded */
2553 			sdp->try_rc_10_first = 0;
2554 		}
2555 	}
2556 
2557 	/* Some devices are known to return the total number of blocks,
2558 	 * not the highest block number.  Some devices have versions
2559 	 * which do this and others which do not.  Some devices we might
2560 	 * suspect of doing this but we don't know for certain.
2561 	 *
2562 	 * If we know the reported capacity is wrong, decrement it.  If
2563 	 * we can only guess, then assume the number of blocks is even
2564 	 * (usually true but not always) and err on the side of lowering
2565 	 * the capacity.
2566 	 */
2567 	if (sdp->fix_capacity ||
2568 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2569 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2570 				"from its reported value: %llu\n",
2571 				(unsigned long long) sdkp->capacity);
2572 		--sdkp->capacity;
2573 	}
2574 
2575 got_data:
2576 	if (sector_size == 0) {
2577 		sector_size = 512;
2578 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2579 			  "assuming 512.\n");
2580 	}
2581 
2582 	if (sector_size != 512 &&
2583 	    sector_size != 1024 &&
2584 	    sector_size != 2048 &&
2585 	    sector_size != 4096) {
2586 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2587 			  sector_size);
2588 		/*
2589 		 * The user might want to re-format the drive with
2590 		 * a supported sectorsize.  Once this happens, it
2591 		 * would be relatively trivial to set the thing up.
2592 		 * For this reason, we leave the thing in the table.
2593 		 */
2594 		sdkp->capacity = 0;
2595 		/*
2596 		 * set a bogus sector size so the normal read/write
2597 		 * logic in the block layer will eventually refuse any
2598 		 * request on this device without tripping over power
2599 		 * of two sector size assumptions
2600 		 */
2601 		sector_size = 512;
2602 	}
2603 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2604 	blk_queue_physical_block_size(sdp->request_queue,
2605 				      sdkp->physical_block_size);
2606 	sdkp->device->sector_size = sector_size;
2607 
2608 	if (sdkp->capacity > 0xffffffff)
2609 		sdp->use_16_for_rw = 1;
2610 
2611 }
2612 
2613 /*
2614  * Print disk capacity
2615  */
2616 static void
2617 sd_print_capacity(struct scsi_disk *sdkp,
2618 		  sector_t old_capacity)
2619 {
2620 	int sector_size = sdkp->device->sector_size;
2621 	char cap_str_2[10], cap_str_10[10];
2622 
2623 	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2624 		return;
2625 
2626 	string_get_size(sdkp->capacity, sector_size,
2627 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2628 	string_get_size(sdkp->capacity, sector_size,
2629 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2630 
2631 	sd_printk(KERN_NOTICE, sdkp,
2632 		  "%llu %d-byte logical blocks: (%s/%s)\n",
2633 		  (unsigned long long)sdkp->capacity,
2634 		  sector_size, cap_str_10, cap_str_2);
2635 
2636 	if (sdkp->physical_block_size != sector_size)
2637 		sd_printk(KERN_NOTICE, sdkp,
2638 			  "%u-byte physical blocks\n",
2639 			  sdkp->physical_block_size);
2640 }
2641 
2642 /* called with buffer of length 512 */
2643 static inline int
2644 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2645 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2646 		 struct scsi_sense_hdr *sshdr)
2647 {
2648 	/*
2649 	 * If we must use MODE SENSE(10), make sure that the buffer length
2650 	 * is at least 8 bytes so that the mode sense header fits.
2651 	 */
2652 	if (sdkp->device->use_10_for_ms && len < 8)
2653 		len = 8;
2654 
2655 	return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2656 			       SD_TIMEOUT, sdkp->max_retries, data,
2657 			       sshdr);
2658 }
2659 
2660 /*
2661  * read write protect setting, if possible - called only in sd_revalidate_disk()
2662  * called with buffer of length SD_BUF_SIZE
2663  */
2664 static void
2665 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2666 {
2667 	int res;
2668 	struct scsi_device *sdp = sdkp->device;
2669 	struct scsi_mode_data data;
2670 	int old_wp = sdkp->write_prot;
2671 
2672 	set_disk_ro(sdkp->disk, 0);
2673 	if (sdp->skip_ms_page_3f) {
2674 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2675 		return;
2676 	}
2677 
2678 	if (sdp->use_192_bytes_for_3f) {
2679 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2680 	} else {
2681 		/*
2682 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2683 		 * We have to start carefully: some devices hang if we ask
2684 		 * for more than is available.
2685 		 */
2686 		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2687 
2688 		/*
2689 		 * Second attempt: ask for page 0 When only page 0 is
2690 		 * implemented, a request for page 3F may return Sense Key
2691 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2692 		 * CDB.
2693 		 */
2694 		if (res < 0)
2695 			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2696 
2697 		/*
2698 		 * Third attempt: ask 255 bytes, as we did earlier.
2699 		 */
2700 		if (res < 0)
2701 			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2702 					       &data, NULL);
2703 	}
2704 
2705 	if (res < 0) {
2706 		sd_first_printk(KERN_WARNING, sdkp,
2707 			  "Test WP failed, assume Write Enabled\n");
2708 	} else {
2709 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2710 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2711 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2712 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2713 				  sdkp->write_prot ? "on" : "off");
2714 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2715 		}
2716 	}
2717 }
2718 
2719 /*
2720  * sd_read_cache_type - called only from sd_revalidate_disk()
2721  * called with buffer of length SD_BUF_SIZE
2722  */
2723 static void
2724 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2725 {
2726 	int len = 0, res;
2727 	struct scsi_device *sdp = sdkp->device;
2728 
2729 	int dbd;
2730 	int modepage;
2731 	int first_len;
2732 	struct scsi_mode_data data;
2733 	struct scsi_sense_hdr sshdr;
2734 	int old_wce = sdkp->WCE;
2735 	int old_rcd = sdkp->RCD;
2736 	int old_dpofua = sdkp->DPOFUA;
2737 
2738 
2739 	if (sdkp->cache_override)
2740 		return;
2741 
2742 	first_len = 4;
2743 	if (sdp->skip_ms_page_8) {
2744 		if (sdp->type == TYPE_RBC)
2745 			goto defaults;
2746 		else {
2747 			if (sdp->skip_ms_page_3f)
2748 				goto defaults;
2749 			modepage = 0x3F;
2750 			if (sdp->use_192_bytes_for_3f)
2751 				first_len = 192;
2752 			dbd = 0;
2753 		}
2754 	} else if (sdp->type == TYPE_RBC) {
2755 		modepage = 6;
2756 		dbd = 8;
2757 	} else {
2758 		modepage = 8;
2759 		dbd = 0;
2760 	}
2761 
2762 	/* cautiously ask */
2763 	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2764 			&data, &sshdr);
2765 
2766 	if (res < 0)
2767 		goto bad_sense;
2768 
2769 	if (!data.header_length) {
2770 		modepage = 6;
2771 		first_len = 0;
2772 		sd_first_printk(KERN_ERR, sdkp,
2773 				"Missing header in MODE_SENSE response\n");
2774 	}
2775 
2776 	/* that went OK, now ask for the proper length */
2777 	len = data.length;
2778 
2779 	/*
2780 	 * We're only interested in the first three bytes, actually.
2781 	 * But the data cache page is defined for the first 20.
2782 	 */
2783 	if (len < 3)
2784 		goto bad_sense;
2785 	else if (len > SD_BUF_SIZE) {
2786 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2787 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2788 		len = SD_BUF_SIZE;
2789 	}
2790 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2791 		len = 192;
2792 
2793 	/* Get the data */
2794 	if (len > first_len)
2795 		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2796 				&data, &sshdr);
2797 
2798 	if (!res) {
2799 		int offset = data.header_length + data.block_descriptor_length;
2800 
2801 		while (offset < len) {
2802 			u8 page_code = buffer[offset] & 0x3F;
2803 			u8 spf       = buffer[offset] & 0x40;
2804 
2805 			if (page_code == 8 || page_code == 6) {
2806 				/* We're interested only in the first 3 bytes.
2807 				 */
2808 				if (len - offset <= 2) {
2809 					sd_first_printk(KERN_ERR, sdkp,
2810 						"Incomplete mode parameter "
2811 							"data\n");
2812 					goto defaults;
2813 				} else {
2814 					modepage = page_code;
2815 					goto Page_found;
2816 				}
2817 			} else {
2818 				/* Go to the next page */
2819 				if (spf && len - offset > 3)
2820 					offset += 4 + (buffer[offset+2] << 8) +
2821 						buffer[offset+3];
2822 				else if (!spf && len - offset > 1)
2823 					offset += 2 + buffer[offset+1];
2824 				else {
2825 					sd_first_printk(KERN_ERR, sdkp,
2826 							"Incomplete mode "
2827 							"parameter data\n");
2828 					goto defaults;
2829 				}
2830 			}
2831 		}
2832 
2833 		sd_first_printk(KERN_WARNING, sdkp,
2834 				"No Caching mode page found\n");
2835 		goto defaults;
2836 
2837 	Page_found:
2838 		if (modepage == 8) {
2839 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2840 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2841 		} else {
2842 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2843 			sdkp->RCD = 0;
2844 		}
2845 
2846 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2847 		if (sdp->broken_fua) {
2848 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2849 			sdkp->DPOFUA = 0;
2850 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2851 			   !sdkp->device->use_16_for_rw) {
2852 			sd_first_printk(KERN_NOTICE, sdkp,
2853 				  "Uses READ/WRITE(6), disabling FUA\n");
2854 			sdkp->DPOFUA = 0;
2855 		}
2856 
2857 		/* No cache flush allowed for write protected devices */
2858 		if (sdkp->WCE && sdkp->write_prot)
2859 			sdkp->WCE = 0;
2860 
2861 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2862 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2863 			sd_printk(KERN_NOTICE, sdkp,
2864 				  "Write cache: %s, read cache: %s, %s\n",
2865 				  sdkp->WCE ? "enabled" : "disabled",
2866 				  sdkp->RCD ? "disabled" : "enabled",
2867 				  sdkp->DPOFUA ? "supports DPO and FUA"
2868 				  : "doesn't support DPO or FUA");
2869 
2870 		return;
2871 	}
2872 
2873 bad_sense:
2874 	if (scsi_sense_valid(&sshdr) &&
2875 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2876 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2877 		/* Invalid field in CDB */
2878 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2879 	else
2880 		sd_first_printk(KERN_ERR, sdkp,
2881 				"Asking for cache data failed\n");
2882 
2883 defaults:
2884 	if (sdp->wce_default_on) {
2885 		sd_first_printk(KERN_NOTICE, sdkp,
2886 				"Assuming drive cache: write back\n");
2887 		sdkp->WCE = 1;
2888 	} else {
2889 		sd_first_printk(KERN_WARNING, sdkp,
2890 				"Assuming drive cache: write through\n");
2891 		sdkp->WCE = 0;
2892 	}
2893 	sdkp->RCD = 0;
2894 	sdkp->DPOFUA = 0;
2895 }
2896 
2897 /*
2898  * The ATO bit indicates whether the DIF application tag is available
2899  * for use by the operating system.
2900  */
2901 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2902 {
2903 	int res, offset;
2904 	struct scsi_device *sdp = sdkp->device;
2905 	struct scsi_mode_data data;
2906 	struct scsi_sense_hdr sshdr;
2907 
2908 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2909 		return;
2910 
2911 	if (sdkp->protection_type == 0)
2912 		return;
2913 
2914 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2915 			      sdkp->max_retries, &data, &sshdr);
2916 
2917 	if (res < 0 || !data.header_length ||
2918 	    data.length < 6) {
2919 		sd_first_printk(KERN_WARNING, sdkp,
2920 			  "getting Control mode page failed, assume no ATO\n");
2921 
2922 		if (scsi_sense_valid(&sshdr))
2923 			sd_print_sense_hdr(sdkp, &sshdr);
2924 
2925 		return;
2926 	}
2927 
2928 	offset = data.header_length + data.block_descriptor_length;
2929 
2930 	if ((buffer[offset] & 0x3f) != 0x0a) {
2931 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2932 		return;
2933 	}
2934 
2935 	if ((buffer[offset + 5] & 0x80) == 0)
2936 		return;
2937 
2938 	sdkp->ATO = 1;
2939 
2940 	return;
2941 }
2942 
2943 /**
2944  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2945  * @sdkp: disk to query
2946  */
2947 static void sd_read_block_limits(struct scsi_disk *sdkp)
2948 {
2949 	unsigned int sector_sz = sdkp->device->sector_size;
2950 	const int vpd_len = 64;
2951 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2952 
2953 	if (!buffer ||
2954 	    /* Block Limits VPD */
2955 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2956 		goto out;
2957 
2958 	blk_queue_io_min(sdkp->disk->queue,
2959 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2960 
2961 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2962 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2963 
2964 	if (buffer[3] == 0x3c) {
2965 		unsigned int lba_count, desc_count;
2966 
2967 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2968 
2969 		if (!sdkp->lbpme)
2970 			goto out;
2971 
2972 		lba_count = get_unaligned_be32(&buffer[20]);
2973 		desc_count = get_unaligned_be32(&buffer[24]);
2974 
2975 		if (lba_count && desc_count)
2976 			sdkp->max_unmap_blocks = lba_count;
2977 
2978 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2979 
2980 		if (buffer[32] & 0x80)
2981 			sdkp->unmap_alignment =
2982 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2983 
2984 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2985 
2986 			if (sdkp->max_unmap_blocks)
2987 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2988 			else
2989 				sd_config_discard(sdkp, SD_LBP_WS16);
2990 
2991 		} else {	/* LBP VPD page tells us what to use */
2992 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2993 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2994 			else if (sdkp->lbpws)
2995 				sd_config_discard(sdkp, SD_LBP_WS16);
2996 			else if (sdkp->lbpws10)
2997 				sd_config_discard(sdkp, SD_LBP_WS10);
2998 			else
2999 				sd_config_discard(sdkp, SD_LBP_DISABLE);
3000 		}
3001 	}
3002 
3003  out:
3004 	kfree(buffer);
3005 }
3006 
3007 /**
3008  * sd_read_block_characteristics - Query block dev. characteristics
3009  * @sdkp: disk to query
3010  */
3011 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3012 {
3013 	struct request_queue *q = sdkp->disk->queue;
3014 	unsigned char *buffer;
3015 	u16 rot;
3016 	const int vpd_len = 64;
3017 
3018 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3019 
3020 	if (!buffer ||
3021 	    /* Block Device Characteristics VPD */
3022 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
3023 		goto out;
3024 
3025 	rot = get_unaligned_be16(&buffer[4]);
3026 
3027 	if (rot == 1) {
3028 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3029 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3030 	}
3031 
3032 	if (sdkp->device->type == TYPE_ZBC) {
3033 		/* Host-managed */
3034 		blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3035 	} else {
3036 		sdkp->zoned = (buffer[8] >> 4) & 3;
3037 		if (sdkp->zoned == 1) {
3038 			/* Host-aware */
3039 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3040 		} else {
3041 			/* Regular disk or drive managed disk */
3042 			blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3043 		}
3044 	}
3045 
3046 	if (!sdkp->first_scan)
3047 		goto out;
3048 
3049 	if (blk_queue_is_zoned(q)) {
3050 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3051 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3052 	} else {
3053 		if (sdkp->zoned == 1)
3054 			sd_printk(KERN_NOTICE, sdkp,
3055 				  "Host-aware SMR disk used as regular disk\n");
3056 		else if (sdkp->zoned == 2)
3057 			sd_printk(KERN_NOTICE, sdkp,
3058 				  "Drive-managed SMR disk\n");
3059 	}
3060 
3061  out:
3062 	kfree(buffer);
3063 }
3064 
3065 /**
3066  * sd_read_block_provisioning - Query provisioning VPD page
3067  * @sdkp: disk to query
3068  */
3069 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3070 {
3071 	unsigned char *buffer;
3072 	const int vpd_len = 8;
3073 
3074 	if (sdkp->lbpme == 0)
3075 		return;
3076 
3077 	buffer = kmalloc(vpd_len, GFP_KERNEL);
3078 
3079 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3080 		goto out;
3081 
3082 	sdkp->lbpvpd	= 1;
3083 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
3084 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
3085 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
3086 
3087  out:
3088 	kfree(buffer);
3089 }
3090 
3091 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3092 {
3093 	struct scsi_device *sdev = sdkp->device;
3094 
3095 	if (sdev->host->no_write_same) {
3096 		sdev->no_write_same = 1;
3097 
3098 		return;
3099 	}
3100 
3101 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3102 		/* too large values might cause issues with arcmsr */
3103 		int vpd_buf_len = 64;
3104 
3105 		sdev->no_report_opcodes = 1;
3106 
3107 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3108 		 * CODES is unsupported and the device has an ATA
3109 		 * Information VPD page (SAT).
3110 		 */
3111 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3112 			sdev->no_write_same = 1;
3113 	}
3114 
3115 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3116 		sdkp->ws16 = 1;
3117 
3118 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3119 		sdkp->ws10 = 1;
3120 }
3121 
3122 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3123 {
3124 	struct scsi_device *sdev = sdkp->device;
3125 
3126 	if (!sdev->security_supported)
3127 		return;
3128 
3129 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3130 			SECURITY_PROTOCOL_IN) == 1 &&
3131 	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3132 			SECURITY_PROTOCOL_OUT) == 1)
3133 		sdkp->security = 1;
3134 }
3135 
3136 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3137 {
3138 	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3139 }
3140 
3141 /**
3142  * sd_read_cpr - Query concurrent positioning ranges
3143  * @sdkp:	disk to query
3144  */
3145 static void sd_read_cpr(struct scsi_disk *sdkp)
3146 {
3147 	struct blk_independent_access_ranges *iars = NULL;
3148 	unsigned char *buffer = NULL;
3149 	unsigned int nr_cpr = 0;
3150 	int i, vpd_len, buf_len = SD_BUF_SIZE;
3151 	u8 *desc;
3152 
3153 	/*
3154 	 * We need to have the capacity set first for the block layer to be
3155 	 * able to check the ranges.
3156 	 */
3157 	if (sdkp->first_scan)
3158 		return;
3159 
3160 	if (!sdkp->capacity)
3161 		goto out;
3162 
3163 	/*
3164 	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3165 	 * leading to a maximum page size of 64 + 256*32 bytes.
3166 	 */
3167 	buf_len = 64 + 256*32;
3168 	buffer = kmalloc(buf_len, GFP_KERNEL);
3169 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3170 		goto out;
3171 
3172 	/* We must have at least a 64B header and one 32B range descriptor */
3173 	vpd_len = get_unaligned_be16(&buffer[2]) + 3;
3174 	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3175 		sd_printk(KERN_ERR, sdkp,
3176 			  "Invalid Concurrent Positioning Ranges VPD page\n");
3177 		goto out;
3178 	}
3179 
3180 	nr_cpr = (vpd_len - 64) / 32;
3181 	if (nr_cpr == 1) {
3182 		nr_cpr = 0;
3183 		goto out;
3184 	}
3185 
3186 	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3187 	if (!iars) {
3188 		nr_cpr = 0;
3189 		goto out;
3190 	}
3191 
3192 	desc = &buffer[64];
3193 	for (i = 0; i < nr_cpr; i++, desc += 32) {
3194 		if (desc[0] != i) {
3195 			sd_printk(KERN_ERR, sdkp,
3196 				"Invalid Concurrent Positioning Range number\n");
3197 			nr_cpr = 0;
3198 			break;
3199 		}
3200 
3201 		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3202 		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3203 	}
3204 
3205 out:
3206 	disk_set_independent_access_ranges(sdkp->disk, iars);
3207 	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3208 		sd_printk(KERN_NOTICE, sdkp,
3209 			  "%u concurrent positioning ranges\n", nr_cpr);
3210 		sdkp->nr_actuators = nr_cpr;
3211 	}
3212 
3213 	kfree(buffer);
3214 }
3215 
3216 /*
3217  * Determine the device's preferred I/O size for reads and writes
3218  * unless the reported value is unreasonably small, large, not a
3219  * multiple of the physical block size, or simply garbage.
3220  */
3221 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3222 				      unsigned int dev_max)
3223 {
3224 	struct scsi_device *sdp = sdkp->device;
3225 	unsigned int opt_xfer_bytes =
3226 		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3227 
3228 	if (sdkp->opt_xfer_blocks == 0)
3229 		return false;
3230 
3231 	if (sdkp->opt_xfer_blocks > dev_max) {
3232 		sd_first_printk(KERN_WARNING, sdkp,
3233 				"Optimal transfer size %u logical blocks " \
3234 				"> dev_max (%u logical blocks)\n",
3235 				sdkp->opt_xfer_blocks, dev_max);
3236 		return false;
3237 	}
3238 
3239 	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3240 		sd_first_printk(KERN_WARNING, sdkp,
3241 				"Optimal transfer size %u logical blocks " \
3242 				"> sd driver limit (%u logical blocks)\n",
3243 				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3244 		return false;
3245 	}
3246 
3247 	if (opt_xfer_bytes < PAGE_SIZE) {
3248 		sd_first_printk(KERN_WARNING, sdkp,
3249 				"Optimal transfer size %u bytes < " \
3250 				"PAGE_SIZE (%u bytes)\n",
3251 				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3252 		return false;
3253 	}
3254 
3255 	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3256 		sd_first_printk(KERN_WARNING, sdkp,
3257 				"Optimal transfer size %u bytes not a " \
3258 				"multiple of physical block size (%u bytes)\n",
3259 				opt_xfer_bytes, sdkp->physical_block_size);
3260 		return false;
3261 	}
3262 
3263 	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3264 			opt_xfer_bytes);
3265 	return true;
3266 }
3267 
3268 /**
3269  *	sd_revalidate_disk - called the first time a new disk is seen,
3270  *	performs disk spin up, read_capacity, etc.
3271  *	@disk: struct gendisk we care about
3272  **/
3273 static int sd_revalidate_disk(struct gendisk *disk)
3274 {
3275 	struct scsi_disk *sdkp = scsi_disk(disk);
3276 	struct scsi_device *sdp = sdkp->device;
3277 	struct request_queue *q = sdkp->disk->queue;
3278 	sector_t old_capacity = sdkp->capacity;
3279 	unsigned char *buffer;
3280 	unsigned int dev_max, rw_max;
3281 
3282 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3283 				      "sd_revalidate_disk\n"));
3284 
3285 	/*
3286 	 * If the device is offline, don't try and read capacity or any
3287 	 * of the other niceties.
3288 	 */
3289 	if (!scsi_device_online(sdp))
3290 		goto out;
3291 
3292 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3293 	if (!buffer) {
3294 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3295 			  "allocation failure.\n");
3296 		goto out;
3297 	}
3298 
3299 	sd_spinup_disk(sdkp);
3300 
3301 	/*
3302 	 * Without media there is no reason to ask; moreover, some devices
3303 	 * react badly if we do.
3304 	 */
3305 	if (sdkp->media_present) {
3306 		sd_read_capacity(sdkp, buffer);
3307 
3308 		/*
3309 		 * set the default to rotational.  All non-rotational devices
3310 		 * support the block characteristics VPD page, which will
3311 		 * cause this to be updated correctly and any device which
3312 		 * doesn't support it should be treated as rotational.
3313 		 */
3314 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3315 		blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3316 
3317 		if (scsi_device_supports_vpd(sdp)) {
3318 			sd_read_block_provisioning(sdkp);
3319 			sd_read_block_limits(sdkp);
3320 			sd_read_block_characteristics(sdkp);
3321 			sd_zbc_read_zones(sdkp, buffer);
3322 		}
3323 
3324 		sd_print_capacity(sdkp, old_capacity);
3325 
3326 		sd_read_write_protect_flag(sdkp, buffer);
3327 		sd_read_cache_type(sdkp, buffer);
3328 		sd_read_app_tag_own(sdkp, buffer);
3329 		sd_read_write_same(sdkp, buffer);
3330 		sd_read_security(sdkp, buffer);
3331 		sd_read_cpr(sdkp);
3332 	}
3333 
3334 	/*
3335 	 * We now have all cache related info, determine how we deal
3336 	 * with flush requests.
3337 	 */
3338 	sd_set_flush_flag(sdkp);
3339 
3340 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3341 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3342 
3343 	/* Some devices report a maximum block count for READ/WRITE requests. */
3344 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3345 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3346 
3347 	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3348 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3349 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3350 	} else {
3351 		q->limits.io_opt = 0;
3352 		rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3353 				      (sector_t)BLK_DEF_MAX_SECTORS);
3354 	}
3355 
3356 	/* Do not exceed controller limit */
3357 	rw_max = min(rw_max, queue_max_hw_sectors(q));
3358 
3359 	/*
3360 	 * Only update max_sectors if previously unset or if the current value
3361 	 * exceeds the capabilities of the hardware.
3362 	 */
3363 	if (sdkp->first_scan ||
3364 	    q->limits.max_sectors > q->limits.max_dev_sectors ||
3365 	    q->limits.max_sectors > q->limits.max_hw_sectors)
3366 		q->limits.max_sectors = rw_max;
3367 
3368 	sdkp->first_scan = 0;
3369 
3370 	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3371 	sd_config_write_same(sdkp);
3372 	kfree(buffer);
3373 
3374 	/*
3375 	 * For a zoned drive, revalidating the zones can be done only once
3376 	 * the gendisk capacity is set. So if this fails, set back the gendisk
3377 	 * capacity to 0.
3378 	 */
3379 	if (sd_zbc_revalidate_zones(sdkp))
3380 		set_capacity_and_notify(disk, 0);
3381 
3382  out:
3383 	return 0;
3384 }
3385 
3386 /**
3387  *	sd_unlock_native_capacity - unlock native capacity
3388  *	@disk: struct gendisk to set capacity for
3389  *
3390  *	Block layer calls this function if it detects that partitions
3391  *	on @disk reach beyond the end of the device.  If the SCSI host
3392  *	implements ->unlock_native_capacity() method, it's invoked to
3393  *	give it a chance to adjust the device capacity.
3394  *
3395  *	CONTEXT:
3396  *	Defined by block layer.  Might sleep.
3397  */
3398 static void sd_unlock_native_capacity(struct gendisk *disk)
3399 {
3400 	struct scsi_device *sdev = scsi_disk(disk)->device;
3401 
3402 	if (sdev->host->hostt->unlock_native_capacity)
3403 		sdev->host->hostt->unlock_native_capacity(sdev);
3404 }
3405 
3406 /**
3407  *	sd_format_disk_name - format disk name
3408  *	@prefix: name prefix - ie. "sd" for SCSI disks
3409  *	@index: index of the disk to format name for
3410  *	@buf: output buffer
3411  *	@buflen: length of the output buffer
3412  *
3413  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3414  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3415  *	which is followed by sdaaa.
3416  *
3417  *	This is basically 26 base counting with one extra 'nil' entry
3418  *	at the beginning from the second digit on and can be
3419  *	determined using similar method as 26 base conversion with the
3420  *	index shifted -1 after each digit is computed.
3421  *
3422  *	CONTEXT:
3423  *	Don't care.
3424  *
3425  *	RETURNS:
3426  *	0 on success, -errno on failure.
3427  */
3428 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3429 {
3430 	const int base = 'z' - 'a' + 1;
3431 	char *begin = buf + strlen(prefix);
3432 	char *end = buf + buflen;
3433 	char *p;
3434 	int unit;
3435 
3436 	p = end - 1;
3437 	*p = '\0';
3438 	unit = base;
3439 	do {
3440 		if (p == begin)
3441 			return -EINVAL;
3442 		*--p = 'a' + (index % unit);
3443 		index = (index / unit) - 1;
3444 	} while (index >= 0);
3445 
3446 	memmove(begin, p, end - p);
3447 	memcpy(buf, prefix, strlen(prefix));
3448 
3449 	return 0;
3450 }
3451 
3452 /**
3453  *	sd_probe - called during driver initialization and whenever a
3454  *	new scsi device is attached to the system. It is called once
3455  *	for each scsi device (not just disks) present.
3456  *	@dev: pointer to device object
3457  *
3458  *	Returns 0 if successful (or not interested in this scsi device
3459  *	(e.g. scanner)); 1 when there is an error.
3460  *
3461  *	Note: this function is invoked from the scsi mid-level.
3462  *	This function sets up the mapping between a given
3463  *	<host,channel,id,lun> (found in sdp) and new device name
3464  *	(e.g. /dev/sda). More precisely it is the block device major
3465  *	and minor number that is chosen here.
3466  *
3467  *	Assume sd_probe is not re-entrant (for time being)
3468  *	Also think about sd_probe() and sd_remove() running coincidentally.
3469  **/
3470 static int sd_probe(struct device *dev)
3471 {
3472 	struct scsi_device *sdp = to_scsi_device(dev);
3473 	struct scsi_disk *sdkp;
3474 	struct gendisk *gd;
3475 	int index;
3476 	int error;
3477 
3478 	scsi_autopm_get_device(sdp);
3479 	error = -ENODEV;
3480 	if (sdp->type != TYPE_DISK &&
3481 	    sdp->type != TYPE_ZBC &&
3482 	    sdp->type != TYPE_MOD &&
3483 	    sdp->type != TYPE_RBC)
3484 		goto out;
3485 
3486 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3487 		sdev_printk(KERN_WARNING, sdp,
3488 			    "Unsupported ZBC host-managed device.\n");
3489 		goto out;
3490 	}
3491 
3492 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3493 					"sd_probe\n"));
3494 
3495 	error = -ENOMEM;
3496 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3497 	if (!sdkp)
3498 		goto out;
3499 
3500 	gd = __alloc_disk_node(sdp->request_queue, NUMA_NO_NODE,
3501 			       &sd_bio_compl_lkclass);
3502 	if (!gd)
3503 		goto out_free;
3504 
3505 	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3506 	if (index < 0) {
3507 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3508 		goto out_put;
3509 	}
3510 
3511 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3512 	if (error) {
3513 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3514 		goto out_free_index;
3515 	}
3516 
3517 	sdkp->device = sdp;
3518 	sdkp->driver = &sd_template;
3519 	sdkp->disk = gd;
3520 	sdkp->index = index;
3521 	sdkp->max_retries = SD_MAX_RETRIES;
3522 	atomic_set(&sdkp->openers, 0);
3523 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3524 
3525 	if (!sdp->request_queue->rq_timeout) {
3526 		if (sdp->type != TYPE_MOD)
3527 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3528 		else
3529 			blk_queue_rq_timeout(sdp->request_queue,
3530 					     SD_MOD_TIMEOUT);
3531 	}
3532 
3533 	device_initialize(&sdkp->dev);
3534 	sdkp->dev.parent = get_device(dev);
3535 	sdkp->dev.class = &sd_disk_class;
3536 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3537 
3538 	error = device_add(&sdkp->dev);
3539 	if (error) {
3540 		put_device(&sdkp->dev);
3541 		goto out;
3542 	}
3543 
3544 	dev_set_drvdata(dev, sdkp);
3545 
3546 	gd->major = sd_major((index & 0xf0) >> 4);
3547 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3548 	gd->minors = SD_MINORS;
3549 
3550 	gd->fops = &sd_fops;
3551 	gd->private_data = &sdkp->driver;
3552 
3553 	/* defaults, until the device tells us otherwise */
3554 	sdp->sector_size = 512;
3555 	sdkp->capacity = 0;
3556 	sdkp->media_present = 1;
3557 	sdkp->write_prot = 0;
3558 	sdkp->cache_override = 0;
3559 	sdkp->WCE = 0;
3560 	sdkp->RCD = 0;
3561 	sdkp->ATO = 0;
3562 	sdkp->first_scan = 1;
3563 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3564 
3565 	sd_revalidate_disk(gd);
3566 
3567 	if (sdp->removable) {
3568 		gd->flags |= GENHD_FL_REMOVABLE;
3569 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3570 		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3571 	}
3572 
3573 	blk_pm_runtime_init(sdp->request_queue, dev);
3574 	if (sdp->rpm_autosuspend) {
3575 		pm_runtime_set_autosuspend_delay(dev,
3576 			sdp->host->hostt->rpm_autosuspend_delay);
3577 	}
3578 
3579 	error = device_add_disk(dev, gd, NULL);
3580 	if (error) {
3581 		put_device(&sdkp->dev);
3582 		goto out;
3583 	}
3584 
3585 	if (sdkp->capacity)
3586 		sd_dif_config_host(sdkp);
3587 
3588 	sd_revalidate_disk(gd);
3589 
3590 	if (sdkp->security) {
3591 		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3592 		if (sdkp->opal_dev)
3593 			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3594 	}
3595 
3596 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3597 		  sdp->removable ? "removable " : "");
3598 	scsi_autopm_put_device(sdp);
3599 
3600 	return 0;
3601 
3602  out_free_index:
3603 	ida_free(&sd_index_ida, index);
3604  out_put:
3605 	put_disk(gd);
3606  out_free:
3607 	sd_zbc_release_disk(sdkp);
3608 	kfree(sdkp);
3609  out:
3610 	scsi_autopm_put_device(sdp);
3611 	return error;
3612 }
3613 
3614 /**
3615  *	sd_remove - called whenever a scsi disk (previously recognized by
3616  *	sd_probe) is detached from the system. It is called (potentially
3617  *	multiple times) during sd module unload.
3618  *	@dev: pointer to device object
3619  *
3620  *	Note: this function is invoked from the scsi mid-level.
3621  *	This function potentially frees up a device name (e.g. /dev/sdc)
3622  *	that could be re-used by a subsequent sd_probe().
3623  *	This function is not called when the built-in sd driver is "exit-ed".
3624  **/
3625 static int sd_remove(struct device *dev)
3626 {
3627 	struct scsi_disk *sdkp;
3628 
3629 	sdkp = dev_get_drvdata(dev);
3630 	scsi_autopm_get_device(sdkp->device);
3631 
3632 	device_del(&sdkp->dev);
3633 	del_gendisk(sdkp->disk);
3634 	sd_shutdown(dev);
3635 
3636 	free_opal_dev(sdkp->opal_dev);
3637 
3638 	mutex_lock(&sd_ref_mutex);
3639 	dev_set_drvdata(dev, NULL);
3640 	put_device(&sdkp->dev);
3641 	mutex_unlock(&sd_ref_mutex);
3642 
3643 	return 0;
3644 }
3645 
3646 /**
3647  *	scsi_disk_release - Called to free the scsi_disk structure
3648  *	@dev: pointer to embedded class device
3649  *
3650  *	sd_ref_mutex must be held entering this routine.  Because it is
3651  *	called on last put, you should always use the scsi_disk_get()
3652  *	scsi_disk_put() helpers which manipulate the semaphore directly
3653  *	and never do a direct put_device.
3654  **/
3655 static void scsi_disk_release(struct device *dev)
3656 {
3657 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3658 	struct gendisk *disk = sdkp->disk;
3659 	struct request_queue *q = disk->queue;
3660 
3661 	ida_free(&sd_index_ida, sdkp->index);
3662 
3663 	/*
3664 	 * Wait until all requests that are in progress have completed.
3665 	 * This is necessary to avoid that e.g. scsi_end_request() crashes
3666 	 * due to clearing the disk->private_data pointer. Wait from inside
3667 	 * scsi_disk_release() instead of from sd_release() to avoid that
3668 	 * freezing and unfreezing the request queue affects user space I/O
3669 	 * in case multiple processes open a /dev/sd... node concurrently.
3670 	 */
3671 	blk_mq_freeze_queue(q);
3672 	blk_mq_unfreeze_queue(q);
3673 
3674 	disk->private_data = NULL;
3675 	put_disk(disk);
3676 	put_device(&sdkp->device->sdev_gendev);
3677 
3678 	sd_zbc_release_disk(sdkp);
3679 
3680 	kfree(sdkp);
3681 }
3682 
3683 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3684 {
3685 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3686 	struct scsi_sense_hdr sshdr;
3687 	struct scsi_device *sdp = sdkp->device;
3688 	int res;
3689 
3690 	if (start)
3691 		cmd[4] |= 1;	/* START */
3692 
3693 	if (sdp->start_stop_pwr_cond)
3694 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3695 
3696 	if (!scsi_device_online(sdp))
3697 		return -ENODEV;
3698 
3699 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3700 			SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3701 	if (res) {
3702 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3703 		if (res > 0 && scsi_sense_valid(&sshdr)) {
3704 			sd_print_sense_hdr(sdkp, &sshdr);
3705 			/* 0x3a is medium not present */
3706 			if (sshdr.asc == 0x3a)
3707 				res = 0;
3708 		}
3709 	}
3710 
3711 	/* SCSI error codes must not go to the generic layer */
3712 	if (res)
3713 		return -EIO;
3714 
3715 	return 0;
3716 }
3717 
3718 /*
3719  * Send a SYNCHRONIZE CACHE instruction down to the device through
3720  * the normal SCSI command structure.  Wait for the command to
3721  * complete.
3722  */
3723 static void sd_shutdown(struct device *dev)
3724 {
3725 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3726 
3727 	if (!sdkp)
3728 		return;         /* this can happen */
3729 
3730 	if (pm_runtime_suspended(dev))
3731 		return;
3732 
3733 	if (sdkp->WCE && sdkp->media_present) {
3734 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3735 		sd_sync_cache(sdkp, NULL);
3736 	}
3737 
3738 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3739 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3740 		sd_start_stop_device(sdkp, 0);
3741 	}
3742 }
3743 
3744 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3745 {
3746 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3747 	struct scsi_sense_hdr sshdr;
3748 	int ret = 0;
3749 
3750 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3751 		return 0;
3752 
3753 	if (sdkp->WCE && sdkp->media_present) {
3754 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3755 		ret = sd_sync_cache(sdkp, &sshdr);
3756 
3757 		if (ret) {
3758 			/* ignore OFFLINE device */
3759 			if (ret == -ENODEV)
3760 				return 0;
3761 
3762 			if (!scsi_sense_valid(&sshdr) ||
3763 			    sshdr.sense_key != ILLEGAL_REQUEST)
3764 				return ret;
3765 
3766 			/*
3767 			 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3768 			 * doesn't support sync. There's not much to do and
3769 			 * suspend shouldn't fail.
3770 			 */
3771 			ret = 0;
3772 		}
3773 	}
3774 
3775 	if (sdkp->device->manage_start_stop) {
3776 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3777 		/* an error is not worth aborting a system sleep */
3778 		ret = sd_start_stop_device(sdkp, 0);
3779 		if (ignore_stop_errors)
3780 			ret = 0;
3781 	}
3782 
3783 	return ret;
3784 }
3785 
3786 static int sd_suspend_system(struct device *dev)
3787 {
3788 	if (pm_runtime_suspended(dev))
3789 		return 0;
3790 
3791 	return sd_suspend_common(dev, true);
3792 }
3793 
3794 static int sd_suspend_runtime(struct device *dev)
3795 {
3796 	return sd_suspend_common(dev, false);
3797 }
3798 
3799 static int sd_resume(struct device *dev)
3800 {
3801 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3802 	int ret;
3803 
3804 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3805 		return 0;
3806 
3807 	if (!sdkp->device->manage_start_stop)
3808 		return 0;
3809 
3810 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3811 	ret = sd_start_stop_device(sdkp, 1);
3812 	if (!ret)
3813 		opal_unlock_from_suspend(sdkp->opal_dev);
3814 	return ret;
3815 }
3816 
3817 static int sd_resume_system(struct device *dev)
3818 {
3819 	if (pm_runtime_suspended(dev))
3820 		return 0;
3821 
3822 	return sd_resume(dev);
3823 }
3824 
3825 static int sd_resume_runtime(struct device *dev)
3826 {
3827 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3828 	struct scsi_device *sdp;
3829 
3830 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3831 		return 0;
3832 
3833 	sdp = sdkp->device;
3834 
3835 	if (sdp->ignore_media_change) {
3836 		/* clear the device's sense data */
3837 		static const u8 cmd[10] = { REQUEST_SENSE };
3838 
3839 		if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3840 				 NULL, sdp->request_queue->rq_timeout, 1, 0,
3841 				 RQF_PM, NULL))
3842 			sd_printk(KERN_NOTICE, sdkp,
3843 				  "Failed to clear sense data\n");
3844 	}
3845 
3846 	return sd_resume(dev);
3847 }
3848 
3849 /**
3850  *	init_sd - entry point for this driver (both when built in or when
3851  *	a module).
3852  *
3853  *	Note: this function registers this driver with the scsi mid-level.
3854  **/
3855 static int __init init_sd(void)
3856 {
3857 	int majors = 0, i, err;
3858 
3859 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3860 
3861 	for (i = 0; i < SD_MAJORS; i++) {
3862 		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3863 			continue;
3864 		majors++;
3865 	}
3866 
3867 	if (!majors)
3868 		return -ENODEV;
3869 
3870 	err = class_register(&sd_disk_class);
3871 	if (err)
3872 		goto err_out;
3873 
3874 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3875 					 0, 0, NULL);
3876 	if (!sd_cdb_cache) {
3877 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3878 		err = -ENOMEM;
3879 		goto err_out_class;
3880 	}
3881 
3882 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3883 	if (!sd_cdb_pool) {
3884 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3885 		err = -ENOMEM;
3886 		goto err_out_cache;
3887 	}
3888 
3889 	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3890 	if (!sd_page_pool) {
3891 		printk(KERN_ERR "sd: can't init discard page pool\n");
3892 		err = -ENOMEM;
3893 		goto err_out_ppool;
3894 	}
3895 
3896 	err = scsi_register_driver(&sd_template.gendrv);
3897 	if (err)
3898 		goto err_out_driver;
3899 
3900 	return 0;
3901 
3902 err_out_driver:
3903 	mempool_destroy(sd_page_pool);
3904 
3905 err_out_ppool:
3906 	mempool_destroy(sd_cdb_pool);
3907 
3908 err_out_cache:
3909 	kmem_cache_destroy(sd_cdb_cache);
3910 
3911 err_out_class:
3912 	class_unregister(&sd_disk_class);
3913 err_out:
3914 	for (i = 0; i < SD_MAJORS; i++)
3915 		unregister_blkdev(sd_major(i), "sd");
3916 	return err;
3917 }
3918 
3919 /**
3920  *	exit_sd - exit point for this driver (when it is a module).
3921  *
3922  *	Note: this function unregisters this driver from the scsi mid-level.
3923  **/
3924 static void __exit exit_sd(void)
3925 {
3926 	int i;
3927 
3928 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3929 
3930 	scsi_unregister_driver(&sd_template.gendrv);
3931 	mempool_destroy(sd_cdb_pool);
3932 	mempool_destroy(sd_page_pool);
3933 	kmem_cache_destroy(sd_cdb_cache);
3934 
3935 	class_unregister(&sd_disk_class);
3936 
3937 	for (i = 0; i < SD_MAJORS; i++)
3938 		unregister_blkdev(sd_major(i), "sd");
3939 }
3940 
3941 module_init(init_sd);
3942 module_exit(exit_sd);
3943 
3944 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3945 {
3946 	scsi_print_sense_hdr(sdkp->device,
3947 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3948 }
3949 
3950 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3951 {
3952 	const char *hb_string = scsi_hostbyte_string(result);
3953 
3954 	if (hb_string)
3955 		sd_printk(KERN_INFO, sdkp,
3956 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3957 			  hb_string ? hb_string : "invalid",
3958 			  "DRIVER_OK");
3959 	else
3960 		sd_printk(KERN_INFO, sdkp,
3961 			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3962 			  msg, host_byte(result), "DRIVER_OK");
3963 }
3964