xref: /openbmc/linux/drivers/scsi/sd.c (revision 4f6cce39)
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *	Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17  *	   sd_init and cleanups.
18  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *	   not being read in sd_open. Fix problem where removable media
20  *	   could be ejected after sd_open.
21  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24  *	   Support 32k/1M disks.
25  *
26  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *	Note: when the logging level is set by the user, it must be greater
32  *	than the level indicated above to trigger output.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.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/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <linux/pr.h>
55 #include <linux/t10-pi.h>
56 #include <linux/uaccess.h>
57 #include <asm/unaligned.h>
58 
59 #include <scsi/scsi.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_dbg.h>
62 #include <scsi/scsi_device.h>
63 #include <scsi/scsi_driver.h>
64 #include <scsi/scsi_eh.h>
65 #include <scsi/scsi_host.h>
66 #include <scsi/scsi_ioctl.h>
67 #include <scsi/scsicam.h>
68 
69 #include "sd.h"
70 #include "scsi_priv.h"
71 #include "scsi_logging.h"
72 
73 MODULE_AUTHOR("Eric Youngdale");
74 MODULE_DESCRIPTION("SCSI disk (sd) driver");
75 MODULE_LICENSE("GPL");
76 
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
97 
98 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
99 #define SD_MINORS	16
100 #else
101 #define SD_MINORS	0
102 #endif
103 
104 static void sd_config_discard(struct scsi_disk *, unsigned int);
105 static void sd_config_write_same(struct scsi_disk *);
106 static int  sd_revalidate_disk(struct gendisk *);
107 static void sd_unlock_native_capacity(struct gendisk *disk);
108 static int  sd_probe(struct device *);
109 static int  sd_remove(struct device *);
110 static void sd_shutdown(struct device *);
111 static int sd_suspend_system(struct device *);
112 static int sd_suspend_runtime(struct device *);
113 static int sd_resume(struct device *);
114 static void sd_rescan(struct device *);
115 static int sd_init_command(struct scsi_cmnd *SCpnt);
116 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117 static int sd_done(struct scsi_cmnd *);
118 static 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 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
122 static void sd_print_result(const struct scsi_disk *, const char *, int);
123 
124 static DEFINE_SPINLOCK(sd_index_lock);
125 static DEFINE_IDA(sd_index_ida);
126 
127 /* This semaphore is used to mediate the 0->1 reference get in the
128  * face of object destruction (i.e. we can't allow a get on an
129  * object after last put) */
130 static DEFINE_MUTEX(sd_ref_mutex);
131 
132 static struct kmem_cache *sd_cdb_cache;
133 static mempool_t *sd_cdb_pool;
134 
135 static const char *sd_cache_types[] = {
136 	"write through", "none", "write back",
137 	"write back, no read (daft)"
138 };
139 
140 static void sd_set_flush_flag(struct scsi_disk *sdkp)
141 {
142 	bool wc = false, fua = false;
143 
144 	if (sdkp->WCE) {
145 		wc = true;
146 		if (sdkp->DPOFUA)
147 			fua = true;
148 	}
149 
150 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
151 }
152 
153 static ssize_t
154 cache_type_store(struct device *dev, struct device_attribute *attr,
155 		 const char *buf, size_t count)
156 {
157 	int i, ct = -1, rcd, wce, sp;
158 	struct scsi_disk *sdkp = to_scsi_disk(dev);
159 	struct scsi_device *sdp = sdkp->device;
160 	char buffer[64];
161 	char *buffer_data;
162 	struct scsi_mode_data data;
163 	struct scsi_sense_hdr sshdr;
164 	static const char temp[] = "temporary ";
165 	int len;
166 
167 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
168 		/* no cache control on RBC devices; theoretically they
169 		 * can do it, but there's probably so many exceptions
170 		 * it's not worth the risk */
171 		return -EINVAL;
172 
173 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
174 		buf += sizeof(temp) - 1;
175 		sdkp->cache_override = 1;
176 	} else {
177 		sdkp->cache_override = 0;
178 	}
179 
180 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
181 		len = strlen(sd_cache_types[i]);
182 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
183 		    buf[len] == '\n') {
184 			ct = i;
185 			break;
186 		}
187 	}
188 	if (ct < 0)
189 		return -EINVAL;
190 	rcd = ct & 0x01 ? 1 : 0;
191 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
192 
193 	if (sdkp->cache_override) {
194 		sdkp->WCE = wce;
195 		sdkp->RCD = rcd;
196 		sd_set_flush_flag(sdkp);
197 		return count;
198 	}
199 
200 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
201 			    SD_MAX_RETRIES, &data, NULL))
202 		return -EINVAL;
203 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
204 		  data.block_descriptor_length);
205 	buffer_data = buffer + data.header_length +
206 		data.block_descriptor_length;
207 	buffer_data[2] &= ~0x05;
208 	buffer_data[2] |= wce << 2 | rcd;
209 	sp = buffer_data[0] & 0x80 ? 1 : 0;
210 	buffer_data[0] &= ~0x80;
211 
212 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
213 			     SD_MAX_RETRIES, &data, &sshdr)) {
214 		if (scsi_sense_valid(&sshdr))
215 			sd_print_sense_hdr(sdkp, &sshdr);
216 		return -EINVAL;
217 	}
218 	revalidate_disk(sdkp->disk);
219 	return count;
220 }
221 
222 static ssize_t
223 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
224 		       char *buf)
225 {
226 	struct scsi_disk *sdkp = to_scsi_disk(dev);
227 	struct scsi_device *sdp = sdkp->device;
228 
229 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
230 }
231 
232 static ssize_t
233 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
234 			const char *buf, size_t count)
235 {
236 	struct scsi_disk *sdkp = to_scsi_disk(dev);
237 	struct scsi_device *sdp = sdkp->device;
238 
239 	if (!capable(CAP_SYS_ADMIN))
240 		return -EACCES;
241 
242 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
243 
244 	return count;
245 }
246 static DEVICE_ATTR_RW(manage_start_stop);
247 
248 static ssize_t
249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
250 {
251 	struct scsi_disk *sdkp = to_scsi_disk(dev);
252 
253 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
254 }
255 
256 static ssize_t
257 allow_restart_store(struct device *dev, struct device_attribute *attr,
258 		    const char *buf, size_t count)
259 {
260 	struct scsi_disk *sdkp = to_scsi_disk(dev);
261 	struct scsi_device *sdp = sdkp->device;
262 
263 	if (!capable(CAP_SYS_ADMIN))
264 		return -EACCES;
265 
266 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
267 		return -EINVAL;
268 
269 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
270 
271 	return count;
272 }
273 static DEVICE_ATTR_RW(allow_restart);
274 
275 static ssize_t
276 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
277 {
278 	struct scsi_disk *sdkp = to_scsi_disk(dev);
279 	int ct = sdkp->RCD + 2*sdkp->WCE;
280 
281 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
282 }
283 static DEVICE_ATTR_RW(cache_type);
284 
285 static ssize_t
286 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
287 {
288 	struct scsi_disk *sdkp = to_scsi_disk(dev);
289 
290 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
291 }
292 static DEVICE_ATTR_RO(FUA);
293 
294 static ssize_t
295 protection_type_show(struct device *dev, struct device_attribute *attr,
296 		     char *buf)
297 {
298 	struct scsi_disk *sdkp = to_scsi_disk(dev);
299 
300 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
301 }
302 
303 static ssize_t
304 protection_type_store(struct device *dev, struct device_attribute *attr,
305 		      const char *buf, size_t count)
306 {
307 	struct scsi_disk *sdkp = to_scsi_disk(dev);
308 	unsigned int val;
309 	int err;
310 
311 	if (!capable(CAP_SYS_ADMIN))
312 		return -EACCES;
313 
314 	err = kstrtouint(buf, 10, &val);
315 
316 	if (err)
317 		return err;
318 
319 	if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
320 		sdkp->protection_type = val;
321 
322 	return count;
323 }
324 static DEVICE_ATTR_RW(protection_type);
325 
326 static ssize_t
327 protection_mode_show(struct device *dev, struct device_attribute *attr,
328 		     char *buf)
329 {
330 	struct scsi_disk *sdkp = to_scsi_disk(dev);
331 	struct scsi_device *sdp = sdkp->device;
332 	unsigned int dif, dix;
333 
334 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
335 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
336 
337 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
338 		dif = 0;
339 		dix = 1;
340 	}
341 
342 	if (!dif && !dix)
343 		return snprintf(buf, 20, "none\n");
344 
345 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
346 }
347 static DEVICE_ATTR_RO(protection_mode);
348 
349 static ssize_t
350 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
351 {
352 	struct scsi_disk *sdkp = to_scsi_disk(dev);
353 
354 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
355 }
356 static DEVICE_ATTR_RO(app_tag_own);
357 
358 static ssize_t
359 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
360 		       char *buf)
361 {
362 	struct scsi_disk *sdkp = to_scsi_disk(dev);
363 
364 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
365 }
366 static DEVICE_ATTR_RO(thin_provisioning);
367 
368 static const char *lbp_mode[] = {
369 	[SD_LBP_FULL]		= "full",
370 	[SD_LBP_UNMAP]		= "unmap",
371 	[SD_LBP_WS16]		= "writesame_16",
372 	[SD_LBP_WS10]		= "writesame_10",
373 	[SD_LBP_ZERO]		= "writesame_zero",
374 	[SD_LBP_DISABLE]	= "disabled",
375 };
376 
377 static ssize_t
378 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
379 		       char *buf)
380 {
381 	struct scsi_disk *sdkp = to_scsi_disk(dev);
382 
383 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
384 }
385 
386 static ssize_t
387 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
388 			const char *buf, size_t count)
389 {
390 	struct scsi_disk *sdkp = to_scsi_disk(dev);
391 	struct scsi_device *sdp = sdkp->device;
392 
393 	if (!capable(CAP_SYS_ADMIN))
394 		return -EACCES;
395 
396 	if (sd_is_zoned(sdkp)) {
397 		sd_config_discard(sdkp, SD_LBP_DISABLE);
398 		return count;
399 	}
400 
401 	if (sdp->type != TYPE_DISK)
402 		return -EINVAL;
403 
404 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
405 		sd_config_discard(sdkp, SD_LBP_UNMAP);
406 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
407 		sd_config_discard(sdkp, SD_LBP_WS16);
408 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
409 		sd_config_discard(sdkp, SD_LBP_WS10);
410 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
411 		sd_config_discard(sdkp, SD_LBP_ZERO);
412 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
413 		sd_config_discard(sdkp, SD_LBP_DISABLE);
414 	else
415 		return -EINVAL;
416 
417 	return count;
418 }
419 static DEVICE_ATTR_RW(provisioning_mode);
420 
421 static ssize_t
422 max_medium_access_timeouts_show(struct device *dev,
423 				struct device_attribute *attr, char *buf)
424 {
425 	struct scsi_disk *sdkp = to_scsi_disk(dev);
426 
427 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
428 }
429 
430 static ssize_t
431 max_medium_access_timeouts_store(struct device *dev,
432 				 struct device_attribute *attr, const char *buf,
433 				 size_t count)
434 {
435 	struct scsi_disk *sdkp = to_scsi_disk(dev);
436 	int err;
437 
438 	if (!capable(CAP_SYS_ADMIN))
439 		return -EACCES;
440 
441 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
442 
443 	return err ? err : count;
444 }
445 static DEVICE_ATTR_RW(max_medium_access_timeouts);
446 
447 static ssize_t
448 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
449 			   char *buf)
450 {
451 	struct scsi_disk *sdkp = to_scsi_disk(dev);
452 
453 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
454 }
455 
456 static ssize_t
457 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
458 			    const char *buf, size_t count)
459 {
460 	struct scsi_disk *sdkp = to_scsi_disk(dev);
461 	struct scsi_device *sdp = sdkp->device;
462 	unsigned long max;
463 	int err;
464 
465 	if (!capable(CAP_SYS_ADMIN))
466 		return -EACCES;
467 
468 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
469 		return -EINVAL;
470 
471 	err = kstrtoul(buf, 10, &max);
472 
473 	if (err)
474 		return err;
475 
476 	if (max == 0)
477 		sdp->no_write_same = 1;
478 	else if (max <= SD_MAX_WS16_BLOCKS) {
479 		sdp->no_write_same = 0;
480 		sdkp->max_ws_blocks = max;
481 	}
482 
483 	sd_config_write_same(sdkp);
484 
485 	return count;
486 }
487 static DEVICE_ATTR_RW(max_write_same_blocks);
488 
489 static struct attribute *sd_disk_attrs[] = {
490 	&dev_attr_cache_type.attr,
491 	&dev_attr_FUA.attr,
492 	&dev_attr_allow_restart.attr,
493 	&dev_attr_manage_start_stop.attr,
494 	&dev_attr_protection_type.attr,
495 	&dev_attr_protection_mode.attr,
496 	&dev_attr_app_tag_own.attr,
497 	&dev_attr_thin_provisioning.attr,
498 	&dev_attr_provisioning_mode.attr,
499 	&dev_attr_max_write_same_blocks.attr,
500 	&dev_attr_max_medium_access_timeouts.attr,
501 	NULL,
502 };
503 ATTRIBUTE_GROUPS(sd_disk);
504 
505 static struct class sd_disk_class = {
506 	.name		= "scsi_disk",
507 	.owner		= THIS_MODULE,
508 	.dev_release	= scsi_disk_release,
509 	.dev_groups	= sd_disk_groups,
510 };
511 
512 static const struct dev_pm_ops sd_pm_ops = {
513 	.suspend		= sd_suspend_system,
514 	.resume			= sd_resume,
515 	.poweroff		= sd_suspend_system,
516 	.restore		= sd_resume,
517 	.runtime_suspend	= sd_suspend_runtime,
518 	.runtime_resume		= sd_resume,
519 };
520 
521 static struct scsi_driver sd_template = {
522 	.gendrv = {
523 		.name		= "sd",
524 		.owner		= THIS_MODULE,
525 		.probe		= sd_probe,
526 		.remove		= sd_remove,
527 		.shutdown	= sd_shutdown,
528 		.pm		= &sd_pm_ops,
529 	},
530 	.rescan			= sd_rescan,
531 	.init_command		= sd_init_command,
532 	.uninit_command		= sd_uninit_command,
533 	.done			= sd_done,
534 	.eh_action		= sd_eh_action,
535 };
536 
537 /*
538  * Dummy kobj_map->probe function.
539  * The default ->probe function will call modprobe, which is
540  * pointless as this module is already loaded.
541  */
542 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
543 {
544 	return NULL;
545 }
546 
547 /*
548  * Device no to disk mapping:
549  *
550  *       major         disc2     disc  p1
551  *   |............|.............|....|....| <- dev_t
552  *    31        20 19          8 7  4 3  0
553  *
554  * Inside a major, we have 16k disks, however mapped non-
555  * contiguously. The first 16 disks are for major0, the next
556  * ones with major1, ... Disk 256 is for major0 again, disk 272
557  * for major1, ...
558  * As we stay compatible with our numbering scheme, we can reuse
559  * the well-know SCSI majors 8, 65--71, 136--143.
560  */
561 static int sd_major(int major_idx)
562 {
563 	switch (major_idx) {
564 	case 0:
565 		return SCSI_DISK0_MAJOR;
566 	case 1 ... 7:
567 		return SCSI_DISK1_MAJOR + major_idx - 1;
568 	case 8 ... 15:
569 		return SCSI_DISK8_MAJOR + major_idx - 8;
570 	default:
571 		BUG();
572 		return 0;	/* shut up gcc */
573 	}
574 }
575 
576 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
577 {
578 	struct scsi_disk *sdkp = NULL;
579 
580 	mutex_lock(&sd_ref_mutex);
581 
582 	if (disk->private_data) {
583 		sdkp = scsi_disk(disk);
584 		if (scsi_device_get(sdkp->device) == 0)
585 			get_device(&sdkp->dev);
586 		else
587 			sdkp = NULL;
588 	}
589 	mutex_unlock(&sd_ref_mutex);
590 	return sdkp;
591 }
592 
593 static void scsi_disk_put(struct scsi_disk *sdkp)
594 {
595 	struct scsi_device *sdev = sdkp->device;
596 
597 	mutex_lock(&sd_ref_mutex);
598 	put_device(&sdkp->dev);
599 	scsi_device_put(sdev);
600 	mutex_unlock(&sd_ref_mutex);
601 }
602 
603 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
604 					   unsigned int dix, unsigned int dif)
605 {
606 	struct bio *bio = scmd->request->bio;
607 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
608 	unsigned int protect = 0;
609 
610 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
611 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
612 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
613 
614 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
615 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
616 	}
617 
618 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
619 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
620 
621 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
622 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
623 	}
624 
625 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
626 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
627 
628 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
629 			protect = 3 << 5;	/* Disable target PI checking */
630 		else
631 			protect = 1 << 5;	/* Enable target PI checking */
632 	}
633 
634 	scsi_set_prot_op(scmd, prot_op);
635 	scsi_set_prot_type(scmd, dif);
636 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
637 
638 	return protect;
639 }
640 
641 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
642 {
643 	struct request_queue *q = sdkp->disk->queue;
644 	unsigned int logical_block_size = sdkp->device->sector_size;
645 	unsigned int max_blocks = 0;
646 
647 	q->limits.discard_zeroes_data = 0;
648 
649 	/*
650 	 * When LBPRZ is reported, discard alignment and granularity
651 	 * must be fixed to the logical block size. Otherwise the block
652 	 * layer will drop misaligned portions of the request which can
653 	 * lead to data corruption. If LBPRZ is not set, we honor the
654 	 * device preference.
655 	 */
656 	if (sdkp->lbprz) {
657 		q->limits.discard_alignment = 0;
658 		q->limits.discard_granularity = logical_block_size;
659 	} else {
660 		q->limits.discard_alignment = sdkp->unmap_alignment *
661 			logical_block_size;
662 		q->limits.discard_granularity =
663 			max(sdkp->physical_block_size,
664 			    sdkp->unmap_granularity * logical_block_size);
665 	}
666 
667 	sdkp->provisioning_mode = mode;
668 
669 	switch (mode) {
670 
671 	case SD_LBP_DISABLE:
672 		blk_queue_max_discard_sectors(q, 0);
673 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
674 		return;
675 
676 	case SD_LBP_UNMAP:
677 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
678 					  (u32)SD_MAX_WS16_BLOCKS);
679 		break;
680 
681 	case SD_LBP_WS16:
682 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
683 					  (u32)SD_MAX_WS16_BLOCKS);
684 		q->limits.discard_zeroes_data = sdkp->lbprz;
685 		break;
686 
687 	case SD_LBP_WS10:
688 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
689 					  (u32)SD_MAX_WS10_BLOCKS);
690 		q->limits.discard_zeroes_data = sdkp->lbprz;
691 		break;
692 
693 	case SD_LBP_ZERO:
694 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
695 					  (u32)SD_MAX_WS10_BLOCKS);
696 		q->limits.discard_zeroes_data = 1;
697 		break;
698 	}
699 
700 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
701 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
702 }
703 
704 /**
705  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
706  * @sdp: scsi device to operate on
707  * @rq: Request to prepare
708  *
709  * Will issue either UNMAP or WRITE SAME(16) depending on preference
710  * indicated by target device.
711  **/
712 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
713 {
714 	struct request *rq = cmd->request;
715 	struct scsi_device *sdp = cmd->device;
716 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
717 	sector_t sector = blk_rq_pos(rq);
718 	unsigned int nr_sectors = blk_rq_sectors(rq);
719 	unsigned int len;
720 	int ret;
721 	char *buf;
722 	struct page *page;
723 
724 	sector >>= ilog2(sdp->sector_size) - 9;
725 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
726 
727 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
728 	if (!page)
729 		return BLKPREP_DEFER;
730 
731 	switch (sdkp->provisioning_mode) {
732 	case SD_LBP_UNMAP:
733 		buf = page_address(page);
734 
735 		cmd->cmd_len = 10;
736 		cmd->cmnd[0] = UNMAP;
737 		cmd->cmnd[8] = 24;
738 
739 		put_unaligned_be16(6 + 16, &buf[0]);
740 		put_unaligned_be16(16, &buf[2]);
741 		put_unaligned_be64(sector, &buf[8]);
742 		put_unaligned_be32(nr_sectors, &buf[16]);
743 
744 		len = 24;
745 		break;
746 
747 	case SD_LBP_WS16:
748 		cmd->cmd_len = 16;
749 		cmd->cmnd[0] = WRITE_SAME_16;
750 		cmd->cmnd[1] = 0x8; /* UNMAP */
751 		put_unaligned_be64(sector, &cmd->cmnd[2]);
752 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
753 
754 		len = sdkp->device->sector_size;
755 		break;
756 
757 	case SD_LBP_WS10:
758 	case SD_LBP_ZERO:
759 		cmd->cmd_len = 10;
760 		cmd->cmnd[0] = WRITE_SAME;
761 		if (sdkp->provisioning_mode == SD_LBP_WS10)
762 			cmd->cmnd[1] = 0x8; /* UNMAP */
763 		put_unaligned_be32(sector, &cmd->cmnd[2]);
764 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
765 
766 		len = sdkp->device->sector_size;
767 		break;
768 
769 	default:
770 		ret = BLKPREP_INVALID;
771 		goto out;
772 	}
773 
774 	rq->timeout = SD_TIMEOUT;
775 
776 	cmd->transfersize = len;
777 	cmd->allowed = SD_MAX_RETRIES;
778 
779 	rq->special_vec.bv_page = page;
780 	rq->special_vec.bv_offset = 0;
781 	rq->special_vec.bv_len = len;
782 
783 	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
784 	scsi_req(rq)->resid_len = len;
785 
786 	ret = scsi_init_io(cmd);
787 out:
788 	if (ret != BLKPREP_OK)
789 		__free_page(page);
790 	return ret;
791 }
792 
793 static void sd_config_write_same(struct scsi_disk *sdkp)
794 {
795 	struct request_queue *q = sdkp->disk->queue;
796 	unsigned int logical_block_size = sdkp->device->sector_size;
797 
798 	if (sdkp->device->no_write_same) {
799 		sdkp->max_ws_blocks = 0;
800 		goto out;
801 	}
802 
803 	/* Some devices can not handle block counts above 0xffff despite
804 	 * supporting WRITE SAME(16). Consequently we default to 64k
805 	 * blocks per I/O unless the device explicitly advertises a
806 	 * bigger limit.
807 	 */
808 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
809 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
810 						   (u32)SD_MAX_WS16_BLOCKS);
811 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
812 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
813 						   (u32)SD_MAX_WS10_BLOCKS);
814 	else {
815 		sdkp->device->no_write_same = 1;
816 		sdkp->max_ws_blocks = 0;
817 	}
818 
819 out:
820 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
821 					 (logical_block_size >> 9));
822 }
823 
824 /**
825  * sd_setup_write_same_cmnd - write the same data to multiple blocks
826  * @cmd: command to prepare
827  *
828  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
829  * preference indicated by target device.
830  **/
831 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
832 {
833 	struct request *rq = cmd->request;
834 	struct scsi_device *sdp = cmd->device;
835 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
836 	struct bio *bio = rq->bio;
837 	sector_t sector = blk_rq_pos(rq);
838 	unsigned int nr_sectors = blk_rq_sectors(rq);
839 	unsigned int nr_bytes = blk_rq_bytes(rq);
840 	int ret;
841 
842 	if (sdkp->device->no_write_same)
843 		return BLKPREP_INVALID;
844 
845 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
846 
847 	if (sd_is_zoned(sdkp)) {
848 		ret = sd_zbc_setup_write_cmnd(cmd);
849 		if (ret != BLKPREP_OK)
850 			return ret;
851 	}
852 
853 	sector >>= ilog2(sdp->sector_size) - 9;
854 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
855 
856 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
857 
858 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
859 		cmd->cmd_len = 16;
860 		cmd->cmnd[0] = WRITE_SAME_16;
861 		put_unaligned_be64(sector, &cmd->cmnd[2]);
862 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
863 	} else {
864 		cmd->cmd_len = 10;
865 		cmd->cmnd[0] = WRITE_SAME;
866 		put_unaligned_be32(sector, &cmd->cmnd[2]);
867 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
868 	}
869 
870 	cmd->transfersize = sdp->sector_size;
871 	cmd->allowed = SD_MAX_RETRIES;
872 
873 	/*
874 	 * For WRITE SAME the data transferred via the DATA OUT buffer is
875 	 * different from the amount of data actually written to the target.
876 	 *
877 	 * We set up __data_len to the amount of data transferred via the
878 	 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
879 	 * to transfer a single sector of data first, but then reset it to
880 	 * the amount of data to be written right after so that the I/O path
881 	 * knows how much to actually write.
882 	 */
883 	rq->__data_len = sdp->sector_size;
884 	ret = scsi_init_io(cmd);
885 	rq->__data_len = nr_bytes;
886 	return ret;
887 }
888 
889 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
890 {
891 	struct request *rq = cmd->request;
892 
893 	/* flush requests don't perform I/O, zero the S/G table */
894 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
895 
896 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
897 	cmd->cmd_len = 10;
898 	cmd->transfersize = 0;
899 	cmd->allowed = SD_MAX_RETRIES;
900 
901 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
902 	return BLKPREP_OK;
903 }
904 
905 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
906 {
907 	struct request *rq = SCpnt->request;
908 	struct scsi_device *sdp = SCpnt->device;
909 	struct gendisk *disk = rq->rq_disk;
910 	struct scsi_disk *sdkp = scsi_disk(disk);
911 	sector_t block = blk_rq_pos(rq);
912 	sector_t threshold;
913 	unsigned int this_count = blk_rq_sectors(rq);
914 	unsigned int dif, dix;
915 	bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
916 	int ret;
917 	unsigned char protect;
918 
919 	if (zoned_write) {
920 		ret = sd_zbc_setup_write_cmnd(SCpnt);
921 		if (ret != BLKPREP_OK)
922 			return ret;
923 	}
924 
925 	ret = scsi_init_io(SCpnt);
926 	if (ret != BLKPREP_OK)
927 		goto out;
928 	SCpnt = rq->special;
929 
930 	/* from here on until we're complete, any goto out
931 	 * is used for a killable error condition */
932 	ret = BLKPREP_KILL;
933 
934 	SCSI_LOG_HLQUEUE(1,
935 		scmd_printk(KERN_INFO, SCpnt,
936 			"%s: block=%llu, count=%d\n",
937 			__func__, (unsigned long long)block, this_count));
938 
939 	if (!sdp || !scsi_device_online(sdp) ||
940 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
941 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
942 						"Finishing %u sectors\n",
943 						blk_rq_sectors(rq)));
944 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
945 						"Retry with 0x%p\n", SCpnt));
946 		goto out;
947 	}
948 
949 	if (sdp->changed) {
950 		/*
951 		 * quietly refuse to do anything to a changed disc until
952 		 * the changed bit has been reset
953 		 */
954 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
955 		goto out;
956 	}
957 
958 	/*
959 	 * Some SD card readers can't handle multi-sector accesses which touch
960 	 * the last one or two hardware sectors.  Split accesses as needed.
961 	 */
962 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
963 		(sdp->sector_size / 512);
964 
965 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
966 		if (block < threshold) {
967 			/* Access up to the threshold but not beyond */
968 			this_count = threshold - block;
969 		} else {
970 			/* Access only a single hardware sector */
971 			this_count = sdp->sector_size / 512;
972 		}
973 	}
974 
975 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
976 					(unsigned long long)block));
977 
978 	/*
979 	 * If we have a 1K hardware sectorsize, prevent access to single
980 	 * 512 byte sectors.  In theory we could handle this - in fact
981 	 * the scsi cdrom driver must be able to handle this because
982 	 * we typically use 1K blocksizes, and cdroms typically have
983 	 * 2K hardware sectorsizes.  Of course, things are simpler
984 	 * with the cdrom, since it is read-only.  For performance
985 	 * reasons, the filesystems should be able to handle this
986 	 * and not force the scsi disk driver to use bounce buffers
987 	 * for this.
988 	 */
989 	if (sdp->sector_size == 1024) {
990 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
991 			scmd_printk(KERN_ERR, SCpnt,
992 				    "Bad block number requested\n");
993 			goto out;
994 		} else {
995 			block = block >> 1;
996 			this_count = this_count >> 1;
997 		}
998 	}
999 	if (sdp->sector_size == 2048) {
1000 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1001 			scmd_printk(KERN_ERR, SCpnt,
1002 				    "Bad block number requested\n");
1003 			goto out;
1004 		} else {
1005 			block = block >> 2;
1006 			this_count = this_count >> 2;
1007 		}
1008 	}
1009 	if (sdp->sector_size == 4096) {
1010 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1011 			scmd_printk(KERN_ERR, SCpnt,
1012 				    "Bad block number requested\n");
1013 			goto out;
1014 		} else {
1015 			block = block >> 3;
1016 			this_count = this_count >> 3;
1017 		}
1018 	}
1019 	if (rq_data_dir(rq) == WRITE) {
1020 		SCpnt->cmnd[0] = WRITE_6;
1021 
1022 		if (blk_integrity_rq(rq))
1023 			sd_dif_prepare(SCpnt);
1024 
1025 	} else if (rq_data_dir(rq) == READ) {
1026 		SCpnt->cmnd[0] = READ_6;
1027 	} else {
1028 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1029 		goto out;
1030 	}
1031 
1032 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1033 					"%s %d/%u 512 byte blocks.\n",
1034 					(rq_data_dir(rq) == WRITE) ?
1035 					"writing" : "reading", this_count,
1036 					blk_rq_sectors(rq)));
1037 
1038 	dix = scsi_prot_sg_count(SCpnt);
1039 	dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1040 
1041 	if (dif || dix)
1042 		protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1043 	else
1044 		protect = 0;
1045 
1046 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1047 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1048 
1049 		if (unlikely(SCpnt->cmnd == NULL)) {
1050 			ret = BLKPREP_DEFER;
1051 			goto out;
1052 		}
1053 
1054 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1055 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1056 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1057 		SCpnt->cmnd[7] = 0x18;
1058 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1059 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1060 
1061 		/* LBA */
1062 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1063 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1064 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1065 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1066 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1067 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1068 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1069 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1070 
1071 		/* Expected Indirect LBA */
1072 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1073 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1074 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1075 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1076 
1077 		/* Transfer length */
1078 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1079 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1080 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1081 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1082 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1083 		SCpnt->cmnd[0] += READ_16 - READ_6;
1084 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1085 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1086 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1087 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1088 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1089 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1090 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1091 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1092 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1093 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1094 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1095 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1096 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1097 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1098 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1099 		   scsi_device_protection(SCpnt->device) ||
1100 		   SCpnt->device->use_10_for_rw) {
1101 		SCpnt->cmnd[0] += READ_10 - READ_6;
1102 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1103 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1104 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1105 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1106 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1107 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1108 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1109 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1110 	} else {
1111 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1112 			/*
1113 			 * This happens only if this drive failed
1114 			 * 10byte rw command with ILLEGAL_REQUEST
1115 			 * during operation and thus turned off
1116 			 * use_10_for_rw.
1117 			 */
1118 			scmd_printk(KERN_ERR, SCpnt,
1119 				    "FUA write on READ/WRITE(6) drive\n");
1120 			goto out;
1121 		}
1122 
1123 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1124 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1125 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1126 		SCpnt->cmnd[4] = (unsigned char) this_count;
1127 		SCpnt->cmnd[5] = 0;
1128 	}
1129 	SCpnt->sdb.length = this_count * sdp->sector_size;
1130 
1131 	/*
1132 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1133 	 * host adapter, it's safe to assume that we can at least transfer
1134 	 * this many bytes between each connect / disconnect.
1135 	 */
1136 	SCpnt->transfersize = sdp->sector_size;
1137 	SCpnt->underflow = this_count << 9;
1138 	SCpnt->allowed = SD_MAX_RETRIES;
1139 
1140 	/*
1141 	 * This indicates that the command is ready from our end to be
1142 	 * queued.
1143 	 */
1144 	ret = BLKPREP_OK;
1145  out:
1146 	if (zoned_write && ret != BLKPREP_OK)
1147 		sd_zbc_cancel_write_cmnd(SCpnt);
1148 
1149 	return ret;
1150 }
1151 
1152 static int sd_init_command(struct scsi_cmnd *cmd)
1153 {
1154 	struct request *rq = cmd->request;
1155 
1156 	switch (req_op(rq)) {
1157 	case REQ_OP_DISCARD:
1158 		return sd_setup_discard_cmnd(cmd);
1159 	case REQ_OP_WRITE_SAME:
1160 		return sd_setup_write_same_cmnd(cmd);
1161 	case REQ_OP_FLUSH:
1162 		return sd_setup_flush_cmnd(cmd);
1163 	case REQ_OP_READ:
1164 	case REQ_OP_WRITE:
1165 		return sd_setup_read_write_cmnd(cmd);
1166 	case REQ_OP_ZONE_REPORT:
1167 		return sd_zbc_setup_report_cmnd(cmd);
1168 	case REQ_OP_ZONE_RESET:
1169 		return sd_zbc_setup_reset_cmnd(cmd);
1170 	default:
1171 		BUG();
1172 	}
1173 }
1174 
1175 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1176 {
1177 	struct request *rq = SCpnt->request;
1178 
1179 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1180 		__free_page(rq->special_vec.bv_page);
1181 
1182 	if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1183 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1184 		SCpnt->cmnd = NULL;
1185 		SCpnt->cmd_len = 0;
1186 	}
1187 }
1188 
1189 /**
1190  *	sd_open - open a scsi disk device
1191  *	@inode: only i_rdev member may be used
1192  *	@filp: only f_mode and f_flags may be used
1193  *
1194  *	Returns 0 if successful. Returns a negated errno value in case
1195  *	of error.
1196  *
1197  *	Note: This can be called from a user context (e.g. fsck(1) )
1198  *	or from within the kernel (e.g. as a result of a mount(1) ).
1199  *	In the latter case @inode and @filp carry an abridged amount
1200  *	of information as noted above.
1201  *
1202  *	Locking: called with bdev->bd_mutex held.
1203  **/
1204 static int sd_open(struct block_device *bdev, fmode_t mode)
1205 {
1206 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1207 	struct scsi_device *sdev;
1208 	int retval;
1209 
1210 	if (!sdkp)
1211 		return -ENXIO;
1212 
1213 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1214 
1215 	sdev = sdkp->device;
1216 
1217 	/*
1218 	 * If the device is in error recovery, wait until it is done.
1219 	 * If the device is offline, then disallow any access to it.
1220 	 */
1221 	retval = -ENXIO;
1222 	if (!scsi_block_when_processing_errors(sdev))
1223 		goto error_out;
1224 
1225 	if (sdev->removable || sdkp->write_prot)
1226 		check_disk_change(bdev);
1227 
1228 	/*
1229 	 * If the drive is empty, just let the open fail.
1230 	 */
1231 	retval = -ENOMEDIUM;
1232 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1233 		goto error_out;
1234 
1235 	/*
1236 	 * If the device has the write protect tab set, have the open fail
1237 	 * if the user expects to be able to write to the thing.
1238 	 */
1239 	retval = -EROFS;
1240 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1241 		goto error_out;
1242 
1243 	/*
1244 	 * It is possible that the disk changing stuff resulted in
1245 	 * the device being taken offline.  If this is the case,
1246 	 * report this to the user, and don't pretend that the
1247 	 * open actually succeeded.
1248 	 */
1249 	retval = -ENXIO;
1250 	if (!scsi_device_online(sdev))
1251 		goto error_out;
1252 
1253 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1254 		if (scsi_block_when_processing_errors(sdev))
1255 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1256 	}
1257 
1258 	return 0;
1259 
1260 error_out:
1261 	scsi_disk_put(sdkp);
1262 	return retval;
1263 }
1264 
1265 /**
1266  *	sd_release - invoked when the (last) close(2) is called on this
1267  *	scsi disk.
1268  *	@inode: only i_rdev member may be used
1269  *	@filp: only f_mode and f_flags may be used
1270  *
1271  *	Returns 0.
1272  *
1273  *	Note: may block (uninterruptible) if error recovery is underway
1274  *	on this disk.
1275  *
1276  *	Locking: called with bdev->bd_mutex held.
1277  **/
1278 static void sd_release(struct gendisk *disk, fmode_t mode)
1279 {
1280 	struct scsi_disk *sdkp = scsi_disk(disk);
1281 	struct scsi_device *sdev = sdkp->device;
1282 
1283 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1284 
1285 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1286 		if (scsi_block_when_processing_errors(sdev))
1287 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1288 	}
1289 
1290 	/*
1291 	 * XXX and what if there are packets in flight and this close()
1292 	 * XXX is followed by a "rmmod sd_mod"?
1293 	 */
1294 
1295 	scsi_disk_put(sdkp);
1296 }
1297 
1298 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1299 {
1300 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1301 	struct scsi_device *sdp = sdkp->device;
1302 	struct Scsi_Host *host = sdp->host;
1303 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1304 	int diskinfo[4];
1305 
1306 	/* default to most commonly used values */
1307 	diskinfo[0] = 0x40;	/* 1 << 6 */
1308 	diskinfo[1] = 0x20;	/* 1 << 5 */
1309 	diskinfo[2] = capacity >> 11;
1310 
1311 	/* override with calculated, extended default, or driver values */
1312 	if (host->hostt->bios_param)
1313 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1314 	else
1315 		scsicam_bios_param(bdev, capacity, diskinfo);
1316 
1317 	geo->heads = diskinfo[0];
1318 	geo->sectors = diskinfo[1];
1319 	geo->cylinders = diskinfo[2];
1320 	return 0;
1321 }
1322 
1323 /**
1324  *	sd_ioctl - process an ioctl
1325  *	@inode: only i_rdev/i_bdev members may be used
1326  *	@filp: only f_mode and f_flags may be used
1327  *	@cmd: ioctl command number
1328  *	@arg: this is third argument given to ioctl(2) system call.
1329  *	Often contains a pointer.
1330  *
1331  *	Returns 0 if successful (some ioctls return positive numbers on
1332  *	success as well). Returns a negated errno value in case of error.
1333  *
1334  *	Note: most ioctls are forward onto the block subsystem or further
1335  *	down in the scsi subsystem.
1336  **/
1337 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1338 		    unsigned int cmd, unsigned long arg)
1339 {
1340 	struct gendisk *disk = bdev->bd_disk;
1341 	struct scsi_disk *sdkp = scsi_disk(disk);
1342 	struct scsi_device *sdp = sdkp->device;
1343 	void __user *p = (void __user *)arg;
1344 	int error;
1345 
1346 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1347 				    "cmd=0x%x\n", disk->disk_name, cmd));
1348 
1349 	error = scsi_verify_blk_ioctl(bdev, cmd);
1350 	if (error < 0)
1351 		return error;
1352 
1353 	/*
1354 	 * If we are in the middle of error recovery, don't let anyone
1355 	 * else try and use this device.  Also, if error recovery fails, it
1356 	 * may try and take the device offline, in which case all further
1357 	 * access to the device is prohibited.
1358 	 */
1359 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1360 			(mode & FMODE_NDELAY) != 0);
1361 	if (error)
1362 		goto out;
1363 
1364 	/*
1365 	 * Send SCSI addressing ioctls directly to mid level, send other
1366 	 * ioctls to block level and then onto mid level if they can't be
1367 	 * resolved.
1368 	 */
1369 	switch (cmd) {
1370 		case SCSI_IOCTL_GET_IDLUN:
1371 		case SCSI_IOCTL_GET_BUS_NUMBER:
1372 			error = scsi_ioctl(sdp, cmd, p);
1373 			break;
1374 		default:
1375 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1376 			if (error != -ENOTTY)
1377 				break;
1378 			error = scsi_ioctl(sdp, cmd, p);
1379 			break;
1380 	}
1381 out:
1382 	return error;
1383 }
1384 
1385 static void set_media_not_present(struct scsi_disk *sdkp)
1386 {
1387 	if (sdkp->media_present)
1388 		sdkp->device->changed = 1;
1389 
1390 	if (sdkp->device->removable) {
1391 		sdkp->media_present = 0;
1392 		sdkp->capacity = 0;
1393 	}
1394 }
1395 
1396 static int media_not_present(struct scsi_disk *sdkp,
1397 			     struct scsi_sense_hdr *sshdr)
1398 {
1399 	if (!scsi_sense_valid(sshdr))
1400 		return 0;
1401 
1402 	/* not invoked for commands that could return deferred errors */
1403 	switch (sshdr->sense_key) {
1404 	case UNIT_ATTENTION:
1405 	case NOT_READY:
1406 		/* medium not present */
1407 		if (sshdr->asc == 0x3A) {
1408 			set_media_not_present(sdkp);
1409 			return 1;
1410 		}
1411 	}
1412 	return 0;
1413 }
1414 
1415 /**
1416  *	sd_check_events - check media events
1417  *	@disk: kernel device descriptor
1418  *	@clearing: disk events currently being cleared
1419  *
1420  *	Returns mask of DISK_EVENT_*.
1421  *
1422  *	Note: this function is invoked from the block subsystem.
1423  **/
1424 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1425 {
1426 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1427 	struct scsi_device *sdp;
1428 	int retval;
1429 
1430 	if (!sdkp)
1431 		return 0;
1432 
1433 	sdp = sdkp->device;
1434 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1435 
1436 	/*
1437 	 * If the device is offline, don't send any commands - just pretend as
1438 	 * if the command failed.  If the device ever comes back online, we
1439 	 * can deal with it then.  It is only because of unrecoverable errors
1440 	 * that we would ever take a device offline in the first place.
1441 	 */
1442 	if (!scsi_device_online(sdp)) {
1443 		set_media_not_present(sdkp);
1444 		goto out;
1445 	}
1446 
1447 	/*
1448 	 * Using TEST_UNIT_READY enables differentiation between drive with
1449 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1450 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1451 	 *
1452 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1453 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1454 	 * sd_revalidate() is called.
1455 	 */
1456 	if (scsi_block_when_processing_errors(sdp)) {
1457 		struct scsi_sense_hdr sshdr = { 0, };
1458 
1459 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1460 					      &sshdr);
1461 
1462 		/* failed to execute TUR, assume media not present */
1463 		if (host_byte(retval)) {
1464 			set_media_not_present(sdkp);
1465 			goto out;
1466 		}
1467 
1468 		if (media_not_present(sdkp, &sshdr))
1469 			goto out;
1470 	}
1471 
1472 	/*
1473 	 * For removable scsi disk we have to recognise the presence
1474 	 * of a disk in the drive.
1475 	 */
1476 	if (!sdkp->media_present)
1477 		sdp->changed = 1;
1478 	sdkp->media_present = 1;
1479 out:
1480 	/*
1481 	 * sdp->changed is set under the following conditions:
1482 	 *
1483 	 *	Medium present state has changed in either direction.
1484 	 *	Device has indicated UNIT_ATTENTION.
1485 	 */
1486 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1487 	sdp->changed = 0;
1488 	scsi_disk_put(sdkp);
1489 	return retval;
1490 }
1491 
1492 static int sd_sync_cache(struct scsi_disk *sdkp)
1493 {
1494 	int retries, res;
1495 	struct scsi_device *sdp = sdkp->device;
1496 	const int timeout = sdp->request_queue->rq_timeout
1497 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1498 	struct scsi_sense_hdr sshdr;
1499 
1500 	if (!scsi_device_online(sdp))
1501 		return -ENODEV;
1502 
1503 	for (retries = 3; retries > 0; --retries) {
1504 		unsigned char cmd[10] = { 0 };
1505 
1506 		cmd[0] = SYNCHRONIZE_CACHE;
1507 		/*
1508 		 * Leave the rest of the command zero to indicate
1509 		 * flush everything.
1510 		 */
1511 		res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
1512 				timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1513 		if (res == 0)
1514 			break;
1515 	}
1516 
1517 	if (res) {
1518 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1519 
1520 		if (driver_byte(res) & DRIVER_SENSE)
1521 			sd_print_sense_hdr(sdkp, &sshdr);
1522 		/* we need to evaluate the error return  */
1523 		if (scsi_sense_valid(&sshdr) &&
1524 			(sshdr.asc == 0x3a ||	/* medium not present */
1525 			 sshdr.asc == 0x20))	/* invalid command */
1526 				/* this is no error here */
1527 				return 0;
1528 
1529 		switch (host_byte(res)) {
1530 		/* ignore errors due to racing a disconnection */
1531 		case DID_BAD_TARGET:
1532 		case DID_NO_CONNECT:
1533 			return 0;
1534 		/* signal the upper layer it might try again */
1535 		case DID_BUS_BUSY:
1536 		case DID_IMM_RETRY:
1537 		case DID_REQUEUE:
1538 		case DID_SOFT_ERROR:
1539 			return -EBUSY;
1540 		default:
1541 			return -EIO;
1542 		}
1543 	}
1544 	return 0;
1545 }
1546 
1547 static void sd_rescan(struct device *dev)
1548 {
1549 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1550 
1551 	revalidate_disk(sdkp->disk);
1552 }
1553 
1554 
1555 #ifdef CONFIG_COMPAT
1556 /*
1557  * This gets directly called from VFS. When the ioctl
1558  * is not recognized we go back to the other translation paths.
1559  */
1560 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1561 			   unsigned int cmd, unsigned long arg)
1562 {
1563 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1564 	int error;
1565 
1566 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1567 			(mode & FMODE_NDELAY) != 0);
1568 	if (error)
1569 		return error;
1570 
1571 	/*
1572 	 * Let the static ioctl translation table take care of it.
1573 	 */
1574 	if (!sdev->host->hostt->compat_ioctl)
1575 		return -ENOIOCTLCMD;
1576 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1577 }
1578 #endif
1579 
1580 static char sd_pr_type(enum pr_type type)
1581 {
1582 	switch (type) {
1583 	case PR_WRITE_EXCLUSIVE:
1584 		return 0x01;
1585 	case PR_EXCLUSIVE_ACCESS:
1586 		return 0x03;
1587 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1588 		return 0x05;
1589 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1590 		return 0x06;
1591 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1592 		return 0x07;
1593 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1594 		return 0x08;
1595 	default:
1596 		return 0;
1597 	}
1598 };
1599 
1600 static int sd_pr_command(struct block_device *bdev, u8 sa,
1601 		u64 key, u64 sa_key, u8 type, u8 flags)
1602 {
1603 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1604 	struct scsi_sense_hdr sshdr;
1605 	int result;
1606 	u8 cmd[16] = { 0, };
1607 	u8 data[24] = { 0, };
1608 
1609 	cmd[0] = PERSISTENT_RESERVE_OUT;
1610 	cmd[1] = sa;
1611 	cmd[2] = type;
1612 	put_unaligned_be32(sizeof(data), &cmd[5]);
1613 
1614 	put_unaligned_be64(key, &data[0]);
1615 	put_unaligned_be64(sa_key, &data[8]);
1616 	data[20] = flags;
1617 
1618 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1619 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1620 
1621 	if ((driver_byte(result) & DRIVER_SENSE) &&
1622 	    (scsi_sense_valid(&sshdr))) {
1623 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1624 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1625 	}
1626 
1627 	return result;
1628 }
1629 
1630 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1631 		u32 flags)
1632 {
1633 	if (flags & ~PR_FL_IGNORE_KEY)
1634 		return -EOPNOTSUPP;
1635 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1636 			old_key, new_key, 0,
1637 			(1 << 0) /* APTPL */);
1638 }
1639 
1640 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1641 		u32 flags)
1642 {
1643 	if (flags)
1644 		return -EOPNOTSUPP;
1645 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1646 }
1647 
1648 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1649 {
1650 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1651 }
1652 
1653 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1654 		enum pr_type type, bool abort)
1655 {
1656 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1657 			     sd_pr_type(type), 0);
1658 }
1659 
1660 static int sd_pr_clear(struct block_device *bdev, u64 key)
1661 {
1662 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1663 }
1664 
1665 static const struct pr_ops sd_pr_ops = {
1666 	.pr_register	= sd_pr_register,
1667 	.pr_reserve	= sd_pr_reserve,
1668 	.pr_release	= sd_pr_release,
1669 	.pr_preempt	= sd_pr_preempt,
1670 	.pr_clear	= sd_pr_clear,
1671 };
1672 
1673 static const struct block_device_operations sd_fops = {
1674 	.owner			= THIS_MODULE,
1675 	.open			= sd_open,
1676 	.release		= sd_release,
1677 	.ioctl			= sd_ioctl,
1678 	.getgeo			= sd_getgeo,
1679 #ifdef CONFIG_COMPAT
1680 	.compat_ioctl		= sd_compat_ioctl,
1681 #endif
1682 	.check_events		= sd_check_events,
1683 	.revalidate_disk	= sd_revalidate_disk,
1684 	.unlock_native_capacity	= sd_unlock_native_capacity,
1685 	.pr_ops			= &sd_pr_ops,
1686 };
1687 
1688 /**
1689  *	sd_eh_action - error handling callback
1690  *	@scmd:		sd-issued command that has failed
1691  *	@eh_disp:	The recovery disposition suggested by the midlayer
1692  *
1693  *	This function is called by the SCSI midlayer upon completion of an
1694  *	error test command (currently TEST UNIT READY). The result of sending
1695  *	the eh command is passed in eh_disp.  We're looking for devices that
1696  *	fail medium access commands but are OK with non access commands like
1697  *	test unit ready (so wrongly see the device as having a successful
1698  *	recovery)
1699  **/
1700 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1701 {
1702 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1703 
1704 	if (!scsi_device_online(scmd->device) ||
1705 	    !scsi_medium_access_command(scmd) ||
1706 	    host_byte(scmd->result) != DID_TIME_OUT ||
1707 	    eh_disp != SUCCESS)
1708 		return eh_disp;
1709 
1710 	/*
1711 	 * The device has timed out executing a medium access command.
1712 	 * However, the TEST UNIT READY command sent during error
1713 	 * handling completed successfully. Either the device is in the
1714 	 * process of recovering or has it suffered an internal failure
1715 	 * that prevents access to the storage medium.
1716 	 */
1717 	sdkp->medium_access_timed_out++;
1718 
1719 	/*
1720 	 * If the device keeps failing read/write commands but TEST UNIT
1721 	 * READY always completes successfully we assume that medium
1722 	 * access is no longer possible and take the device offline.
1723 	 */
1724 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1725 		scmd_printk(KERN_ERR, scmd,
1726 			    "Medium access timeout failure. Offlining disk!\n");
1727 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1728 
1729 		return FAILED;
1730 	}
1731 
1732 	return eh_disp;
1733 }
1734 
1735 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1736 {
1737 	u64 start_lba = blk_rq_pos(scmd->request);
1738 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1739 	u64 factor = scmd->device->sector_size / 512;
1740 	u64 bad_lba;
1741 	int info_valid;
1742 	/*
1743 	 * resid is optional but mostly filled in.  When it's unused,
1744 	 * its value is zero, so we assume the whole buffer transferred
1745 	 */
1746 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1747 	unsigned int good_bytes;
1748 
1749 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1750 					     SCSI_SENSE_BUFFERSIZE,
1751 					     &bad_lba);
1752 	if (!info_valid)
1753 		return 0;
1754 
1755 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1756 		return 0;
1757 
1758 	/* be careful ... don't want any overflows */
1759 	do_div(start_lba, factor);
1760 	do_div(end_lba, factor);
1761 
1762 	/* The bad lba was reported incorrectly, we have no idea where
1763 	 * the error is.
1764 	 */
1765 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1766 		return 0;
1767 
1768 	/* This computation should always be done in terms of
1769 	 * the resolution of the device's medium.
1770 	 */
1771 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1772 	return min(good_bytes, transferred);
1773 }
1774 
1775 /**
1776  *	sd_done - bottom half handler: called when the lower level
1777  *	driver has completed (successfully or otherwise) a scsi command.
1778  *	@SCpnt: mid-level's per command structure.
1779  *
1780  *	Note: potentially run from within an ISR. Must not block.
1781  **/
1782 static int sd_done(struct scsi_cmnd *SCpnt)
1783 {
1784 	int result = SCpnt->result;
1785 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1786 	unsigned int sector_size = SCpnt->device->sector_size;
1787 	unsigned int resid;
1788 	struct scsi_sense_hdr sshdr;
1789 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1790 	struct request *req = SCpnt->request;
1791 	int sense_valid = 0;
1792 	int sense_deferred = 0;
1793 	unsigned char op = SCpnt->cmnd[0];
1794 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1795 
1796 	switch (req_op(req)) {
1797 	case REQ_OP_DISCARD:
1798 	case REQ_OP_WRITE_SAME:
1799 	case REQ_OP_ZONE_RESET:
1800 		if (!result) {
1801 			good_bytes = blk_rq_bytes(req);
1802 			scsi_set_resid(SCpnt, 0);
1803 		} else {
1804 			good_bytes = 0;
1805 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1806 		}
1807 		break;
1808 	case REQ_OP_ZONE_REPORT:
1809 		if (!result) {
1810 			good_bytes = scsi_bufflen(SCpnt)
1811 				- scsi_get_resid(SCpnt);
1812 			scsi_set_resid(SCpnt, 0);
1813 		} else {
1814 			good_bytes = 0;
1815 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1816 		}
1817 		break;
1818 	default:
1819 		/*
1820 		 * In case of bogus fw or device, we could end up having
1821 		 * an unaligned partial completion. Check this here and force
1822 		 * alignment.
1823 		 */
1824 		resid = scsi_get_resid(SCpnt);
1825 		if (resid & (sector_size - 1)) {
1826 			sd_printk(KERN_INFO, sdkp,
1827 				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1828 				resid, sector_size);
1829 			resid = min(scsi_bufflen(SCpnt),
1830 				    round_up(resid, sector_size));
1831 			scsi_set_resid(SCpnt, resid);
1832 		}
1833 	}
1834 
1835 	if (result) {
1836 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1837 		if (sense_valid)
1838 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1839 	}
1840 	sdkp->medium_access_timed_out = 0;
1841 
1842 	if (driver_byte(result) != DRIVER_SENSE &&
1843 	    (!sense_valid || sense_deferred))
1844 		goto out;
1845 
1846 	switch (sshdr.sense_key) {
1847 	case HARDWARE_ERROR:
1848 	case MEDIUM_ERROR:
1849 		good_bytes = sd_completed_bytes(SCpnt);
1850 		break;
1851 	case RECOVERED_ERROR:
1852 		good_bytes = scsi_bufflen(SCpnt);
1853 		break;
1854 	case NO_SENSE:
1855 		/* This indicates a false check condition, so ignore it.  An
1856 		 * unknown amount of data was transferred so treat it as an
1857 		 * error.
1858 		 */
1859 		SCpnt->result = 0;
1860 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1861 		break;
1862 	case ABORTED_COMMAND:
1863 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1864 			good_bytes = sd_completed_bytes(SCpnt);
1865 		break;
1866 	case ILLEGAL_REQUEST:
1867 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1868 			good_bytes = sd_completed_bytes(SCpnt);
1869 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1870 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1871 			switch (op) {
1872 			case UNMAP:
1873 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1874 				break;
1875 			case WRITE_SAME_16:
1876 			case WRITE_SAME:
1877 				if (unmap)
1878 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1879 				else {
1880 					sdkp->device->no_write_same = 1;
1881 					sd_config_write_same(sdkp);
1882 
1883 					good_bytes = 0;
1884 					req->__data_len = blk_rq_bytes(req);
1885 					req->rq_flags |= RQF_QUIET;
1886 				}
1887 			}
1888 		}
1889 		break;
1890 	default:
1891 		break;
1892 	}
1893 
1894  out:
1895 	if (sd_is_zoned(sdkp))
1896 		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
1897 
1898 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1899 					   "sd_done: completed %d of %d bytes\n",
1900 					   good_bytes, scsi_bufflen(SCpnt)));
1901 
1902 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1903 		sd_dif_complete(SCpnt, good_bytes);
1904 
1905 	return good_bytes;
1906 }
1907 
1908 /*
1909  * spinup disk - called only in sd_revalidate_disk()
1910  */
1911 static void
1912 sd_spinup_disk(struct scsi_disk *sdkp)
1913 {
1914 	unsigned char cmd[10];
1915 	unsigned long spintime_expire = 0;
1916 	int retries, spintime;
1917 	unsigned int the_result;
1918 	struct scsi_sense_hdr sshdr;
1919 	int sense_valid = 0;
1920 
1921 	spintime = 0;
1922 
1923 	/* Spin up drives, as required.  Only do this at boot time */
1924 	/* Spinup needs to be done for module loads too. */
1925 	do {
1926 		retries = 0;
1927 
1928 		do {
1929 			cmd[0] = TEST_UNIT_READY;
1930 			memset((void *) &cmd[1], 0, 9);
1931 
1932 			the_result = scsi_execute_req(sdkp->device, cmd,
1933 						      DMA_NONE, NULL, 0,
1934 						      &sshdr, SD_TIMEOUT,
1935 						      SD_MAX_RETRIES, NULL);
1936 
1937 			/*
1938 			 * If the drive has indicated to us that it
1939 			 * doesn't have any media in it, don't bother
1940 			 * with any more polling.
1941 			 */
1942 			if (media_not_present(sdkp, &sshdr))
1943 				return;
1944 
1945 			if (the_result)
1946 				sense_valid = scsi_sense_valid(&sshdr);
1947 			retries++;
1948 		} while (retries < 3 &&
1949 			 (!scsi_status_is_good(the_result) ||
1950 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1951 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1952 
1953 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1954 			/* no sense, TUR either succeeded or failed
1955 			 * with a status error */
1956 			if(!spintime && !scsi_status_is_good(the_result)) {
1957 				sd_print_result(sdkp, "Test Unit Ready failed",
1958 						the_result);
1959 			}
1960 			break;
1961 		}
1962 
1963 		/*
1964 		 * The device does not want the automatic start to be issued.
1965 		 */
1966 		if (sdkp->device->no_start_on_add)
1967 			break;
1968 
1969 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1970 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1971 				break;	/* manual intervention required */
1972 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1973 				break;	/* standby */
1974 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1975 				break;	/* unavailable */
1976 			/*
1977 			 * Issue command to spin up drive when not ready
1978 			 */
1979 			if (!spintime) {
1980 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1981 				cmd[0] = START_STOP;
1982 				cmd[1] = 1;	/* Return immediately */
1983 				memset((void *) &cmd[2], 0, 8);
1984 				cmd[4] = 1;	/* Start spin cycle */
1985 				if (sdkp->device->start_stop_pwr_cond)
1986 					cmd[4] |= 1 << 4;
1987 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1988 						 NULL, 0, &sshdr,
1989 						 SD_TIMEOUT, SD_MAX_RETRIES,
1990 						 NULL);
1991 				spintime_expire = jiffies + 100 * HZ;
1992 				spintime = 1;
1993 			}
1994 			/* Wait 1 second for next try */
1995 			msleep(1000);
1996 			printk(".");
1997 
1998 		/*
1999 		 * Wait for USB flash devices with slow firmware.
2000 		 * Yes, this sense key/ASC combination shouldn't
2001 		 * occur here.  It's characteristic of these devices.
2002 		 */
2003 		} else if (sense_valid &&
2004 				sshdr.sense_key == UNIT_ATTENTION &&
2005 				sshdr.asc == 0x28) {
2006 			if (!spintime) {
2007 				spintime_expire = jiffies + 5 * HZ;
2008 				spintime = 1;
2009 			}
2010 			/* Wait 1 second for next try */
2011 			msleep(1000);
2012 		} else {
2013 			/* we don't understand the sense code, so it's
2014 			 * probably pointless to loop */
2015 			if(!spintime) {
2016 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2017 				sd_print_sense_hdr(sdkp, &sshdr);
2018 			}
2019 			break;
2020 		}
2021 
2022 	} while (spintime && time_before_eq(jiffies, spintime_expire));
2023 
2024 	if (spintime) {
2025 		if (scsi_status_is_good(the_result))
2026 			printk("ready\n");
2027 		else
2028 			printk("not responding...\n");
2029 	}
2030 }
2031 
2032 /*
2033  * Determine whether disk supports Data Integrity Field.
2034  */
2035 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2036 {
2037 	struct scsi_device *sdp = sdkp->device;
2038 	u8 type;
2039 	int ret = 0;
2040 
2041 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2042 		return ret;
2043 
2044 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2045 
2046 	if (type > T10_PI_TYPE3_PROTECTION)
2047 		ret = -ENODEV;
2048 	else if (scsi_host_dif_capable(sdp->host, type))
2049 		ret = 1;
2050 
2051 	if (sdkp->first_scan || type != sdkp->protection_type)
2052 		switch (ret) {
2053 		case -ENODEV:
2054 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2055 				  " protection type %u. Disabling disk!\n",
2056 				  type);
2057 			break;
2058 		case 1:
2059 			sd_printk(KERN_NOTICE, sdkp,
2060 				  "Enabling DIF Type %u protection\n", type);
2061 			break;
2062 		case 0:
2063 			sd_printk(KERN_NOTICE, sdkp,
2064 				  "Disabling DIF Type %u protection\n", type);
2065 			break;
2066 		}
2067 
2068 	sdkp->protection_type = type;
2069 
2070 	return ret;
2071 }
2072 
2073 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2074 			struct scsi_sense_hdr *sshdr, int sense_valid,
2075 			int the_result)
2076 {
2077 	if (driver_byte(the_result) & DRIVER_SENSE)
2078 		sd_print_sense_hdr(sdkp, sshdr);
2079 	else
2080 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2081 
2082 	/*
2083 	 * Set dirty bit for removable devices if not ready -
2084 	 * sometimes drives will not report this properly.
2085 	 */
2086 	if (sdp->removable &&
2087 	    sense_valid && sshdr->sense_key == NOT_READY)
2088 		set_media_not_present(sdkp);
2089 
2090 	/*
2091 	 * We used to set media_present to 0 here to indicate no media
2092 	 * in the drive, but some drives fail read capacity even with
2093 	 * media present, so we can't do that.
2094 	 */
2095 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2096 }
2097 
2098 #define RC16_LEN 32
2099 #if RC16_LEN > SD_BUF_SIZE
2100 #error RC16_LEN must not be more than SD_BUF_SIZE
2101 #endif
2102 
2103 #define READ_CAPACITY_RETRIES_ON_RESET	10
2104 
2105 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2106 						unsigned char *buffer)
2107 {
2108 	unsigned char cmd[16];
2109 	struct scsi_sense_hdr sshdr;
2110 	int sense_valid = 0;
2111 	int the_result;
2112 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2113 	unsigned int alignment;
2114 	unsigned long long lba;
2115 	unsigned sector_size;
2116 
2117 	if (sdp->no_read_capacity_16)
2118 		return -EINVAL;
2119 
2120 	do {
2121 		memset(cmd, 0, 16);
2122 		cmd[0] = SERVICE_ACTION_IN_16;
2123 		cmd[1] = SAI_READ_CAPACITY_16;
2124 		cmd[13] = RC16_LEN;
2125 		memset(buffer, 0, RC16_LEN);
2126 
2127 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2128 					buffer, RC16_LEN, &sshdr,
2129 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2130 
2131 		if (media_not_present(sdkp, &sshdr))
2132 			return -ENODEV;
2133 
2134 		if (the_result) {
2135 			sense_valid = scsi_sense_valid(&sshdr);
2136 			if (sense_valid &&
2137 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2138 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2139 			    sshdr.ascq == 0x00)
2140 				/* Invalid Command Operation Code or
2141 				 * Invalid Field in CDB, just retry
2142 				 * silently with RC10 */
2143 				return -EINVAL;
2144 			if (sense_valid &&
2145 			    sshdr.sense_key == UNIT_ATTENTION &&
2146 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2147 				/* Device reset might occur several times,
2148 				 * give it one more chance */
2149 				if (--reset_retries > 0)
2150 					continue;
2151 		}
2152 		retries--;
2153 
2154 	} while (the_result && retries);
2155 
2156 	if (the_result) {
2157 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2158 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2159 		return -EINVAL;
2160 	}
2161 
2162 	sector_size = get_unaligned_be32(&buffer[8]);
2163 	lba = get_unaligned_be64(&buffer[0]);
2164 
2165 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2166 		sdkp->capacity = 0;
2167 		return -ENODEV;
2168 	}
2169 
2170 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2171 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2172 			"kernel compiled with support for large block "
2173 			"devices.\n");
2174 		sdkp->capacity = 0;
2175 		return -EOVERFLOW;
2176 	}
2177 
2178 	/* Logical blocks per physical block exponent */
2179 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2180 
2181 	/* RC basis */
2182 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2183 
2184 	/* Lowest aligned logical block */
2185 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2186 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2187 	if (alignment && sdkp->first_scan)
2188 		sd_printk(KERN_NOTICE, sdkp,
2189 			  "physical block alignment offset: %u\n", alignment);
2190 
2191 	if (buffer[14] & 0x80) { /* LBPME */
2192 		sdkp->lbpme = 1;
2193 
2194 		if (buffer[14] & 0x40) /* LBPRZ */
2195 			sdkp->lbprz = 1;
2196 
2197 		sd_config_discard(sdkp, SD_LBP_WS16);
2198 	}
2199 
2200 	sdkp->capacity = lba + 1;
2201 	return sector_size;
2202 }
2203 
2204 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2205 						unsigned char *buffer)
2206 {
2207 	unsigned char cmd[16];
2208 	struct scsi_sense_hdr sshdr;
2209 	int sense_valid = 0;
2210 	int the_result;
2211 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2212 	sector_t lba;
2213 	unsigned sector_size;
2214 
2215 	do {
2216 		cmd[0] = READ_CAPACITY;
2217 		memset(&cmd[1], 0, 9);
2218 		memset(buffer, 0, 8);
2219 
2220 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2221 					buffer, 8, &sshdr,
2222 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2223 
2224 		if (media_not_present(sdkp, &sshdr))
2225 			return -ENODEV;
2226 
2227 		if (the_result) {
2228 			sense_valid = scsi_sense_valid(&sshdr);
2229 			if (sense_valid &&
2230 			    sshdr.sense_key == UNIT_ATTENTION &&
2231 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2232 				/* Device reset might occur several times,
2233 				 * give it one more chance */
2234 				if (--reset_retries > 0)
2235 					continue;
2236 		}
2237 		retries--;
2238 
2239 	} while (the_result && retries);
2240 
2241 	if (the_result) {
2242 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2243 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2244 		return -EINVAL;
2245 	}
2246 
2247 	sector_size = get_unaligned_be32(&buffer[4]);
2248 	lba = get_unaligned_be32(&buffer[0]);
2249 
2250 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2251 		/* Some buggy (usb cardreader) devices return an lba of
2252 		   0xffffffff when the want to report a size of 0 (with
2253 		   which they really mean no media is present) */
2254 		sdkp->capacity = 0;
2255 		sdkp->physical_block_size = sector_size;
2256 		return sector_size;
2257 	}
2258 
2259 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2260 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2261 			"kernel compiled with support for large block "
2262 			"devices.\n");
2263 		sdkp->capacity = 0;
2264 		return -EOVERFLOW;
2265 	}
2266 
2267 	sdkp->capacity = lba + 1;
2268 	sdkp->physical_block_size = sector_size;
2269 	return sector_size;
2270 }
2271 
2272 static int sd_try_rc16_first(struct scsi_device *sdp)
2273 {
2274 	if (sdp->host->max_cmd_len < 16)
2275 		return 0;
2276 	if (sdp->try_rc_10_first)
2277 		return 0;
2278 	if (sdp->scsi_level > SCSI_SPC_2)
2279 		return 1;
2280 	if (scsi_device_protection(sdp))
2281 		return 1;
2282 	return 0;
2283 }
2284 
2285 /*
2286  * read disk capacity
2287  */
2288 static void
2289 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2290 {
2291 	int sector_size;
2292 	struct scsi_device *sdp = sdkp->device;
2293 
2294 	if (sd_try_rc16_first(sdp)) {
2295 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2296 		if (sector_size == -EOVERFLOW)
2297 			goto got_data;
2298 		if (sector_size == -ENODEV)
2299 			return;
2300 		if (sector_size < 0)
2301 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2302 		if (sector_size < 0)
2303 			return;
2304 	} else {
2305 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2306 		if (sector_size == -EOVERFLOW)
2307 			goto got_data;
2308 		if (sector_size < 0)
2309 			return;
2310 		if ((sizeof(sdkp->capacity) > 4) &&
2311 		    (sdkp->capacity > 0xffffffffULL)) {
2312 			int old_sector_size = sector_size;
2313 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2314 					"Trying to use READ CAPACITY(16).\n");
2315 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2316 			if (sector_size < 0) {
2317 				sd_printk(KERN_NOTICE, sdkp,
2318 					"Using 0xffffffff as device size\n");
2319 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2320 				sector_size = old_sector_size;
2321 				goto got_data;
2322 			}
2323 		}
2324 	}
2325 
2326 	/* Some devices are known to return the total number of blocks,
2327 	 * not the highest block number.  Some devices have versions
2328 	 * which do this and others which do not.  Some devices we might
2329 	 * suspect of doing this but we don't know for certain.
2330 	 *
2331 	 * If we know the reported capacity is wrong, decrement it.  If
2332 	 * we can only guess, then assume the number of blocks is even
2333 	 * (usually true but not always) and err on the side of lowering
2334 	 * the capacity.
2335 	 */
2336 	if (sdp->fix_capacity ||
2337 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2338 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2339 				"from its reported value: %llu\n",
2340 				(unsigned long long) sdkp->capacity);
2341 		--sdkp->capacity;
2342 	}
2343 
2344 got_data:
2345 	if (sector_size == 0) {
2346 		sector_size = 512;
2347 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2348 			  "assuming 512.\n");
2349 	}
2350 
2351 	if (sector_size != 512 &&
2352 	    sector_size != 1024 &&
2353 	    sector_size != 2048 &&
2354 	    sector_size != 4096) {
2355 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2356 			  sector_size);
2357 		/*
2358 		 * The user might want to re-format the drive with
2359 		 * a supported sectorsize.  Once this happens, it
2360 		 * would be relatively trivial to set the thing up.
2361 		 * For this reason, we leave the thing in the table.
2362 		 */
2363 		sdkp->capacity = 0;
2364 		/*
2365 		 * set a bogus sector size so the normal read/write
2366 		 * logic in the block layer will eventually refuse any
2367 		 * request on this device without tripping over power
2368 		 * of two sector size assumptions
2369 		 */
2370 		sector_size = 512;
2371 	}
2372 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2373 	blk_queue_physical_block_size(sdp->request_queue,
2374 				      sdkp->physical_block_size);
2375 	sdkp->device->sector_size = sector_size;
2376 
2377 	if (sdkp->capacity > 0xffffffff)
2378 		sdp->use_16_for_rw = 1;
2379 
2380 }
2381 
2382 /*
2383  * Print disk capacity
2384  */
2385 static void
2386 sd_print_capacity(struct scsi_disk *sdkp,
2387 		  sector_t old_capacity)
2388 {
2389 	int sector_size = sdkp->device->sector_size;
2390 	char cap_str_2[10], cap_str_10[10];
2391 
2392 	string_get_size(sdkp->capacity, sector_size,
2393 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2394 	string_get_size(sdkp->capacity, sector_size,
2395 			STRING_UNITS_10, cap_str_10,
2396 			sizeof(cap_str_10));
2397 
2398 	if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2399 		sd_printk(KERN_NOTICE, sdkp,
2400 			  "%llu %d-byte logical blocks: (%s/%s)\n",
2401 			  (unsigned long long)sdkp->capacity,
2402 			  sector_size, cap_str_10, cap_str_2);
2403 
2404 		if (sdkp->physical_block_size != sector_size)
2405 			sd_printk(KERN_NOTICE, sdkp,
2406 				  "%u-byte physical blocks\n",
2407 				  sdkp->physical_block_size);
2408 
2409 		sd_zbc_print_zones(sdkp);
2410 	}
2411 }
2412 
2413 /* called with buffer of length 512 */
2414 static inline int
2415 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2416 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2417 		 struct scsi_sense_hdr *sshdr)
2418 {
2419 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2420 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2421 			       sshdr);
2422 }
2423 
2424 /*
2425  * read write protect setting, if possible - called only in sd_revalidate_disk()
2426  * called with buffer of length SD_BUF_SIZE
2427  */
2428 static void
2429 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2430 {
2431 	int res;
2432 	struct scsi_device *sdp = sdkp->device;
2433 	struct scsi_mode_data data;
2434 	int old_wp = sdkp->write_prot;
2435 
2436 	set_disk_ro(sdkp->disk, 0);
2437 	if (sdp->skip_ms_page_3f) {
2438 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2439 		return;
2440 	}
2441 
2442 	if (sdp->use_192_bytes_for_3f) {
2443 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2444 	} else {
2445 		/*
2446 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2447 		 * We have to start carefully: some devices hang if we ask
2448 		 * for more than is available.
2449 		 */
2450 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2451 
2452 		/*
2453 		 * Second attempt: ask for page 0 When only page 0 is
2454 		 * implemented, a request for page 3F may return Sense Key
2455 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2456 		 * CDB.
2457 		 */
2458 		if (!scsi_status_is_good(res))
2459 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2460 
2461 		/*
2462 		 * Third attempt: ask 255 bytes, as we did earlier.
2463 		 */
2464 		if (!scsi_status_is_good(res))
2465 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2466 					       &data, NULL);
2467 	}
2468 
2469 	if (!scsi_status_is_good(res)) {
2470 		sd_first_printk(KERN_WARNING, sdkp,
2471 			  "Test WP failed, assume Write Enabled\n");
2472 	} else {
2473 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2474 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2475 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2476 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2477 				  sdkp->write_prot ? "on" : "off");
2478 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2479 		}
2480 	}
2481 }
2482 
2483 /*
2484  * sd_read_cache_type - called only from sd_revalidate_disk()
2485  * called with buffer of length SD_BUF_SIZE
2486  */
2487 static void
2488 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2489 {
2490 	int len = 0, res;
2491 	struct scsi_device *sdp = sdkp->device;
2492 
2493 	int dbd;
2494 	int modepage;
2495 	int first_len;
2496 	struct scsi_mode_data data;
2497 	struct scsi_sense_hdr sshdr;
2498 	int old_wce = sdkp->WCE;
2499 	int old_rcd = sdkp->RCD;
2500 	int old_dpofua = sdkp->DPOFUA;
2501 
2502 
2503 	if (sdkp->cache_override)
2504 		return;
2505 
2506 	first_len = 4;
2507 	if (sdp->skip_ms_page_8) {
2508 		if (sdp->type == TYPE_RBC)
2509 			goto defaults;
2510 		else {
2511 			if (sdp->skip_ms_page_3f)
2512 				goto defaults;
2513 			modepage = 0x3F;
2514 			if (sdp->use_192_bytes_for_3f)
2515 				first_len = 192;
2516 			dbd = 0;
2517 		}
2518 	} else if (sdp->type == TYPE_RBC) {
2519 		modepage = 6;
2520 		dbd = 8;
2521 	} else {
2522 		modepage = 8;
2523 		dbd = 0;
2524 	}
2525 
2526 	/* cautiously ask */
2527 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2528 			&data, &sshdr);
2529 
2530 	if (!scsi_status_is_good(res))
2531 		goto bad_sense;
2532 
2533 	if (!data.header_length) {
2534 		modepage = 6;
2535 		first_len = 0;
2536 		sd_first_printk(KERN_ERR, sdkp,
2537 				"Missing header in MODE_SENSE response\n");
2538 	}
2539 
2540 	/* that went OK, now ask for the proper length */
2541 	len = data.length;
2542 
2543 	/*
2544 	 * We're only interested in the first three bytes, actually.
2545 	 * But the data cache page is defined for the first 20.
2546 	 */
2547 	if (len < 3)
2548 		goto bad_sense;
2549 	else if (len > SD_BUF_SIZE) {
2550 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2551 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2552 		len = SD_BUF_SIZE;
2553 	}
2554 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2555 		len = 192;
2556 
2557 	/* Get the data */
2558 	if (len > first_len)
2559 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2560 				&data, &sshdr);
2561 
2562 	if (scsi_status_is_good(res)) {
2563 		int offset = data.header_length + data.block_descriptor_length;
2564 
2565 		while (offset < len) {
2566 			u8 page_code = buffer[offset] & 0x3F;
2567 			u8 spf       = buffer[offset] & 0x40;
2568 
2569 			if (page_code == 8 || page_code == 6) {
2570 				/* We're interested only in the first 3 bytes.
2571 				 */
2572 				if (len - offset <= 2) {
2573 					sd_first_printk(KERN_ERR, sdkp,
2574 						"Incomplete mode parameter "
2575 							"data\n");
2576 					goto defaults;
2577 				} else {
2578 					modepage = page_code;
2579 					goto Page_found;
2580 				}
2581 			} else {
2582 				/* Go to the next page */
2583 				if (spf && len - offset > 3)
2584 					offset += 4 + (buffer[offset+2] << 8) +
2585 						buffer[offset+3];
2586 				else if (!spf && len - offset > 1)
2587 					offset += 2 + buffer[offset+1];
2588 				else {
2589 					sd_first_printk(KERN_ERR, sdkp,
2590 							"Incomplete mode "
2591 							"parameter data\n");
2592 					goto defaults;
2593 				}
2594 			}
2595 		}
2596 
2597 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2598 		goto defaults;
2599 
2600 	Page_found:
2601 		if (modepage == 8) {
2602 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2603 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2604 		} else {
2605 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2606 			sdkp->RCD = 0;
2607 		}
2608 
2609 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2610 		if (sdp->broken_fua) {
2611 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2612 			sdkp->DPOFUA = 0;
2613 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2614 			   !sdkp->device->use_16_for_rw) {
2615 			sd_first_printk(KERN_NOTICE, sdkp,
2616 				  "Uses READ/WRITE(6), disabling FUA\n");
2617 			sdkp->DPOFUA = 0;
2618 		}
2619 
2620 		/* No cache flush allowed for write protected devices */
2621 		if (sdkp->WCE && sdkp->write_prot)
2622 			sdkp->WCE = 0;
2623 
2624 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2625 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2626 			sd_printk(KERN_NOTICE, sdkp,
2627 				  "Write cache: %s, read cache: %s, %s\n",
2628 				  sdkp->WCE ? "enabled" : "disabled",
2629 				  sdkp->RCD ? "disabled" : "enabled",
2630 				  sdkp->DPOFUA ? "supports DPO and FUA"
2631 				  : "doesn't support DPO or FUA");
2632 
2633 		return;
2634 	}
2635 
2636 bad_sense:
2637 	if (scsi_sense_valid(&sshdr) &&
2638 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2639 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2640 		/* Invalid field in CDB */
2641 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2642 	else
2643 		sd_first_printk(KERN_ERR, sdkp,
2644 				"Asking for cache data failed\n");
2645 
2646 defaults:
2647 	if (sdp->wce_default_on) {
2648 		sd_first_printk(KERN_NOTICE, sdkp,
2649 				"Assuming drive cache: write back\n");
2650 		sdkp->WCE = 1;
2651 	} else {
2652 		sd_first_printk(KERN_ERR, sdkp,
2653 				"Assuming drive cache: write through\n");
2654 		sdkp->WCE = 0;
2655 	}
2656 	sdkp->RCD = 0;
2657 	sdkp->DPOFUA = 0;
2658 }
2659 
2660 /*
2661  * The ATO bit indicates whether the DIF application tag is available
2662  * for use by the operating system.
2663  */
2664 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2665 {
2666 	int res, offset;
2667 	struct scsi_device *sdp = sdkp->device;
2668 	struct scsi_mode_data data;
2669 	struct scsi_sense_hdr sshdr;
2670 
2671 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2672 		return;
2673 
2674 	if (sdkp->protection_type == 0)
2675 		return;
2676 
2677 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2678 			      SD_MAX_RETRIES, &data, &sshdr);
2679 
2680 	if (!scsi_status_is_good(res) || !data.header_length ||
2681 	    data.length < 6) {
2682 		sd_first_printk(KERN_WARNING, sdkp,
2683 			  "getting Control mode page failed, assume no ATO\n");
2684 
2685 		if (scsi_sense_valid(&sshdr))
2686 			sd_print_sense_hdr(sdkp, &sshdr);
2687 
2688 		return;
2689 	}
2690 
2691 	offset = data.header_length + data.block_descriptor_length;
2692 
2693 	if ((buffer[offset] & 0x3f) != 0x0a) {
2694 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2695 		return;
2696 	}
2697 
2698 	if ((buffer[offset + 5] & 0x80) == 0)
2699 		return;
2700 
2701 	sdkp->ATO = 1;
2702 
2703 	return;
2704 }
2705 
2706 /**
2707  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2708  * @disk: disk to query
2709  */
2710 static void sd_read_block_limits(struct scsi_disk *sdkp)
2711 {
2712 	unsigned int sector_sz = sdkp->device->sector_size;
2713 	const int vpd_len = 64;
2714 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2715 
2716 	if (!buffer ||
2717 	    /* Block Limits VPD */
2718 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2719 		goto out;
2720 
2721 	blk_queue_io_min(sdkp->disk->queue,
2722 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2723 
2724 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2725 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2726 
2727 	if (buffer[3] == 0x3c) {
2728 		unsigned int lba_count, desc_count;
2729 
2730 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2731 
2732 		if (!sdkp->lbpme)
2733 			goto out;
2734 
2735 		lba_count = get_unaligned_be32(&buffer[20]);
2736 		desc_count = get_unaligned_be32(&buffer[24]);
2737 
2738 		if (lba_count && desc_count)
2739 			sdkp->max_unmap_blocks = lba_count;
2740 
2741 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2742 
2743 		if (buffer[32] & 0x80)
2744 			sdkp->unmap_alignment =
2745 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2746 
2747 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2748 
2749 			if (sdkp->max_unmap_blocks)
2750 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2751 			else
2752 				sd_config_discard(sdkp, SD_LBP_WS16);
2753 
2754 		} else {	/* LBP VPD page tells us what to use */
2755 			if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2756 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2757 			else if (sdkp->lbpws)
2758 				sd_config_discard(sdkp, SD_LBP_WS16);
2759 			else if (sdkp->lbpws10)
2760 				sd_config_discard(sdkp, SD_LBP_WS10);
2761 			else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2762 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2763 			else
2764 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2765 		}
2766 	}
2767 
2768  out:
2769 	kfree(buffer);
2770 }
2771 
2772 /**
2773  * sd_read_block_characteristics - Query block dev. characteristics
2774  * @disk: disk to query
2775  */
2776 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2777 {
2778 	struct request_queue *q = sdkp->disk->queue;
2779 	unsigned char *buffer;
2780 	u16 rot;
2781 	const int vpd_len = 64;
2782 
2783 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2784 
2785 	if (!buffer ||
2786 	    /* Block Device Characteristics VPD */
2787 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2788 		goto out;
2789 
2790 	rot = get_unaligned_be16(&buffer[4]);
2791 
2792 	if (rot == 1) {
2793 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2794 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2795 	}
2796 
2797 	if (sdkp->device->type == TYPE_ZBC) {
2798 		/* Host-managed */
2799 		q->limits.zoned = BLK_ZONED_HM;
2800 	} else {
2801 		sdkp->zoned = (buffer[8] >> 4) & 3;
2802 		if (sdkp->zoned == 1)
2803 			/* Host-aware */
2804 			q->limits.zoned = BLK_ZONED_HA;
2805 		else
2806 			/*
2807 			 * Treat drive-managed devices as
2808 			 * regular block devices.
2809 			 */
2810 			q->limits.zoned = BLK_ZONED_NONE;
2811 	}
2812 	if (blk_queue_is_zoned(q) && sdkp->first_scan)
2813 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2814 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2815 
2816  out:
2817 	kfree(buffer);
2818 }
2819 
2820 /**
2821  * sd_read_block_provisioning - Query provisioning VPD page
2822  * @disk: disk to query
2823  */
2824 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2825 {
2826 	unsigned char *buffer;
2827 	const int vpd_len = 8;
2828 
2829 	if (sdkp->lbpme == 0)
2830 		return;
2831 
2832 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2833 
2834 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2835 		goto out;
2836 
2837 	sdkp->lbpvpd	= 1;
2838 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2839 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2840 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2841 
2842  out:
2843 	kfree(buffer);
2844 }
2845 
2846 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2847 {
2848 	struct scsi_device *sdev = sdkp->device;
2849 
2850 	if (sdev->host->no_write_same) {
2851 		sdev->no_write_same = 1;
2852 
2853 		return;
2854 	}
2855 
2856 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2857 		/* too large values might cause issues with arcmsr */
2858 		int vpd_buf_len = 64;
2859 
2860 		sdev->no_report_opcodes = 1;
2861 
2862 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2863 		 * CODES is unsupported and the device has an ATA
2864 		 * Information VPD page (SAT).
2865 		 */
2866 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2867 			sdev->no_write_same = 1;
2868 	}
2869 
2870 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2871 		sdkp->ws16 = 1;
2872 
2873 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2874 		sdkp->ws10 = 1;
2875 }
2876 
2877 /**
2878  *	sd_revalidate_disk - called the first time a new disk is seen,
2879  *	performs disk spin up, read_capacity, etc.
2880  *	@disk: struct gendisk we care about
2881  **/
2882 static int sd_revalidate_disk(struct gendisk *disk)
2883 {
2884 	struct scsi_disk *sdkp = scsi_disk(disk);
2885 	struct scsi_device *sdp = sdkp->device;
2886 	struct request_queue *q = sdkp->disk->queue;
2887 	sector_t old_capacity = sdkp->capacity;
2888 	unsigned char *buffer;
2889 	unsigned int dev_max, rw_max;
2890 
2891 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2892 				      "sd_revalidate_disk\n"));
2893 
2894 	/*
2895 	 * If the device is offline, don't try and read capacity or any
2896 	 * of the other niceties.
2897 	 */
2898 	if (!scsi_device_online(sdp))
2899 		goto out;
2900 
2901 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2902 	if (!buffer) {
2903 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2904 			  "allocation failure.\n");
2905 		goto out;
2906 	}
2907 
2908 	sd_spinup_disk(sdkp);
2909 
2910 	/*
2911 	 * Without media there is no reason to ask; moreover, some devices
2912 	 * react badly if we do.
2913 	 */
2914 	if (sdkp->media_present) {
2915 		sd_read_capacity(sdkp, buffer);
2916 
2917 		if (scsi_device_supports_vpd(sdp)) {
2918 			sd_read_block_provisioning(sdkp);
2919 			sd_read_block_limits(sdkp);
2920 			sd_read_block_characteristics(sdkp);
2921 			sd_zbc_read_zones(sdkp, buffer);
2922 		}
2923 
2924 		sd_print_capacity(sdkp, old_capacity);
2925 
2926 		sd_read_write_protect_flag(sdkp, buffer);
2927 		sd_read_cache_type(sdkp, buffer);
2928 		sd_read_app_tag_own(sdkp, buffer);
2929 		sd_read_write_same(sdkp, buffer);
2930 	}
2931 
2932 	sdkp->first_scan = 0;
2933 
2934 	/*
2935 	 * We now have all cache related info, determine how we deal
2936 	 * with flush requests.
2937 	 */
2938 	sd_set_flush_flag(sdkp);
2939 
2940 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2941 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2942 
2943 	/* Some devices report a maximum block count for READ/WRITE requests. */
2944 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2945 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2946 
2947 	/*
2948 	 * Use the device's preferred I/O size for reads and writes
2949 	 * unless the reported value is unreasonably small, large, or
2950 	 * garbage.
2951 	 */
2952 	if (sdkp->opt_xfer_blocks &&
2953 	    sdkp->opt_xfer_blocks <= dev_max &&
2954 	    sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2955 	    logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2956 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2957 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2958 	} else
2959 		rw_max = BLK_DEF_MAX_SECTORS;
2960 
2961 	/* Combine with controller limits */
2962 	q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2963 
2964 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2965 	sd_config_write_same(sdkp);
2966 	kfree(buffer);
2967 
2968  out:
2969 	return 0;
2970 }
2971 
2972 /**
2973  *	sd_unlock_native_capacity - unlock native capacity
2974  *	@disk: struct gendisk to set capacity for
2975  *
2976  *	Block layer calls this function if it detects that partitions
2977  *	on @disk reach beyond the end of the device.  If the SCSI host
2978  *	implements ->unlock_native_capacity() method, it's invoked to
2979  *	give it a chance to adjust the device capacity.
2980  *
2981  *	CONTEXT:
2982  *	Defined by block layer.  Might sleep.
2983  */
2984 static void sd_unlock_native_capacity(struct gendisk *disk)
2985 {
2986 	struct scsi_device *sdev = scsi_disk(disk)->device;
2987 
2988 	if (sdev->host->hostt->unlock_native_capacity)
2989 		sdev->host->hostt->unlock_native_capacity(sdev);
2990 }
2991 
2992 /**
2993  *	sd_format_disk_name - format disk name
2994  *	@prefix: name prefix - ie. "sd" for SCSI disks
2995  *	@index: index of the disk to format name for
2996  *	@buf: output buffer
2997  *	@buflen: length of the output buffer
2998  *
2999  *	SCSI disk names starts at sda.  The 26th device is sdz and the
3000  *	27th is sdaa.  The last one for two lettered suffix is sdzz
3001  *	which is followed by sdaaa.
3002  *
3003  *	This is basically 26 base counting with one extra 'nil' entry
3004  *	at the beginning from the second digit on and can be
3005  *	determined using similar method as 26 base conversion with the
3006  *	index shifted -1 after each digit is computed.
3007  *
3008  *	CONTEXT:
3009  *	Don't care.
3010  *
3011  *	RETURNS:
3012  *	0 on success, -errno on failure.
3013  */
3014 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3015 {
3016 	const int base = 'z' - 'a' + 1;
3017 	char *begin = buf + strlen(prefix);
3018 	char *end = buf + buflen;
3019 	char *p;
3020 	int unit;
3021 
3022 	p = end - 1;
3023 	*p = '\0';
3024 	unit = base;
3025 	do {
3026 		if (p == begin)
3027 			return -EINVAL;
3028 		*--p = 'a' + (index % unit);
3029 		index = (index / unit) - 1;
3030 	} while (index >= 0);
3031 
3032 	memmove(begin, p, end - p);
3033 	memcpy(buf, prefix, strlen(prefix));
3034 
3035 	return 0;
3036 }
3037 
3038 /*
3039  * The asynchronous part of sd_probe
3040  */
3041 static void sd_probe_async(void *data, async_cookie_t cookie)
3042 {
3043 	struct scsi_disk *sdkp = data;
3044 	struct scsi_device *sdp;
3045 	struct gendisk *gd;
3046 	u32 index;
3047 	struct device *dev;
3048 
3049 	sdp = sdkp->device;
3050 	gd = sdkp->disk;
3051 	index = sdkp->index;
3052 	dev = &sdp->sdev_gendev;
3053 
3054 	gd->major = sd_major((index & 0xf0) >> 4);
3055 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3056 	gd->minors = SD_MINORS;
3057 
3058 	gd->fops = &sd_fops;
3059 	gd->private_data = &sdkp->driver;
3060 	gd->queue = sdkp->device->request_queue;
3061 
3062 	/* defaults, until the device tells us otherwise */
3063 	sdp->sector_size = 512;
3064 	sdkp->capacity = 0;
3065 	sdkp->media_present = 1;
3066 	sdkp->write_prot = 0;
3067 	sdkp->cache_override = 0;
3068 	sdkp->WCE = 0;
3069 	sdkp->RCD = 0;
3070 	sdkp->ATO = 0;
3071 	sdkp->first_scan = 1;
3072 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3073 
3074 	sd_revalidate_disk(gd);
3075 
3076 	gd->flags = GENHD_FL_EXT_DEVT;
3077 	if (sdp->removable) {
3078 		gd->flags |= GENHD_FL_REMOVABLE;
3079 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3080 	}
3081 
3082 	blk_pm_runtime_init(sdp->request_queue, dev);
3083 	device_add_disk(dev, gd);
3084 	if (sdkp->capacity)
3085 		sd_dif_config_host(sdkp);
3086 
3087 	sd_revalidate_disk(gd);
3088 
3089 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3090 		  sdp->removable ? "removable " : "");
3091 	scsi_autopm_put_device(sdp);
3092 	put_device(&sdkp->dev);
3093 }
3094 
3095 /**
3096  *	sd_probe - called during driver initialization and whenever a
3097  *	new scsi device is attached to the system. It is called once
3098  *	for each scsi device (not just disks) present.
3099  *	@dev: pointer to device object
3100  *
3101  *	Returns 0 if successful (or not interested in this scsi device
3102  *	(e.g. scanner)); 1 when there is an error.
3103  *
3104  *	Note: this function is invoked from the scsi mid-level.
3105  *	This function sets up the mapping between a given
3106  *	<host,channel,id,lun> (found in sdp) and new device name
3107  *	(e.g. /dev/sda). More precisely it is the block device major
3108  *	and minor number that is chosen here.
3109  *
3110  *	Assume sd_probe is not re-entrant (for time being)
3111  *	Also think about sd_probe() and sd_remove() running coincidentally.
3112  **/
3113 static int sd_probe(struct device *dev)
3114 {
3115 	struct scsi_device *sdp = to_scsi_device(dev);
3116 	struct scsi_disk *sdkp;
3117 	struct gendisk *gd;
3118 	int index;
3119 	int error;
3120 
3121 	scsi_autopm_get_device(sdp);
3122 	error = -ENODEV;
3123 	if (sdp->type != TYPE_DISK &&
3124 	    sdp->type != TYPE_ZBC &&
3125 	    sdp->type != TYPE_MOD &&
3126 	    sdp->type != TYPE_RBC)
3127 		goto out;
3128 
3129 #ifndef CONFIG_BLK_DEV_ZONED
3130 	if (sdp->type == TYPE_ZBC)
3131 		goto out;
3132 #endif
3133 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3134 					"sd_probe\n"));
3135 
3136 	error = -ENOMEM;
3137 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3138 	if (!sdkp)
3139 		goto out;
3140 
3141 	gd = alloc_disk(SD_MINORS);
3142 	if (!gd)
3143 		goto out_free;
3144 
3145 	do {
3146 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3147 			goto out_put;
3148 
3149 		spin_lock(&sd_index_lock);
3150 		error = ida_get_new(&sd_index_ida, &index);
3151 		spin_unlock(&sd_index_lock);
3152 	} while (error == -EAGAIN);
3153 
3154 	if (error) {
3155 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3156 		goto out_put;
3157 	}
3158 
3159 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3160 	if (error) {
3161 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3162 		goto out_free_index;
3163 	}
3164 
3165 	sdkp->device = sdp;
3166 	sdkp->driver = &sd_template;
3167 	sdkp->disk = gd;
3168 	sdkp->index = index;
3169 	atomic_set(&sdkp->openers, 0);
3170 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3171 
3172 	if (!sdp->request_queue->rq_timeout) {
3173 		if (sdp->type != TYPE_MOD)
3174 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3175 		else
3176 			blk_queue_rq_timeout(sdp->request_queue,
3177 					     SD_MOD_TIMEOUT);
3178 	}
3179 
3180 	device_initialize(&sdkp->dev);
3181 	sdkp->dev.parent = dev;
3182 	sdkp->dev.class = &sd_disk_class;
3183 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3184 
3185 	error = device_add(&sdkp->dev);
3186 	if (error)
3187 		goto out_free_index;
3188 
3189 	get_device(dev);
3190 	dev_set_drvdata(dev, sdkp);
3191 
3192 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3193 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3194 
3195 	return 0;
3196 
3197  out_free_index:
3198 	spin_lock(&sd_index_lock);
3199 	ida_remove(&sd_index_ida, index);
3200 	spin_unlock(&sd_index_lock);
3201  out_put:
3202 	put_disk(gd);
3203  out_free:
3204 	kfree(sdkp);
3205  out:
3206 	scsi_autopm_put_device(sdp);
3207 	return error;
3208 }
3209 
3210 /**
3211  *	sd_remove - called whenever a scsi disk (previously recognized by
3212  *	sd_probe) is detached from the system. It is called (potentially
3213  *	multiple times) during sd module unload.
3214  *	@dev: pointer to device object
3215  *
3216  *	Note: this function is invoked from the scsi mid-level.
3217  *	This function potentially frees up a device name (e.g. /dev/sdc)
3218  *	that could be re-used by a subsequent sd_probe().
3219  *	This function is not called when the built-in sd driver is "exit-ed".
3220  **/
3221 static int sd_remove(struct device *dev)
3222 {
3223 	struct scsi_disk *sdkp;
3224 	dev_t devt;
3225 
3226 	sdkp = dev_get_drvdata(dev);
3227 	devt = disk_devt(sdkp->disk);
3228 	scsi_autopm_get_device(sdkp->device);
3229 
3230 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3231 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3232 	device_del(&sdkp->dev);
3233 	del_gendisk(sdkp->disk);
3234 	sd_shutdown(dev);
3235 
3236 	sd_zbc_remove(sdkp);
3237 
3238 	blk_register_region(devt, SD_MINORS, NULL,
3239 			    sd_default_probe, NULL, NULL);
3240 
3241 	mutex_lock(&sd_ref_mutex);
3242 	dev_set_drvdata(dev, NULL);
3243 	put_device(&sdkp->dev);
3244 	mutex_unlock(&sd_ref_mutex);
3245 
3246 	return 0;
3247 }
3248 
3249 /**
3250  *	scsi_disk_release - Called to free the scsi_disk structure
3251  *	@dev: pointer to embedded class device
3252  *
3253  *	sd_ref_mutex must be held entering this routine.  Because it is
3254  *	called on last put, you should always use the scsi_disk_get()
3255  *	scsi_disk_put() helpers which manipulate the semaphore directly
3256  *	and never do a direct put_device.
3257  **/
3258 static void scsi_disk_release(struct device *dev)
3259 {
3260 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3261 	struct gendisk *disk = sdkp->disk;
3262 
3263 	spin_lock(&sd_index_lock);
3264 	ida_remove(&sd_index_ida, sdkp->index);
3265 	spin_unlock(&sd_index_lock);
3266 
3267 	disk->private_data = NULL;
3268 	put_disk(disk);
3269 	put_device(&sdkp->device->sdev_gendev);
3270 
3271 	kfree(sdkp);
3272 }
3273 
3274 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3275 {
3276 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3277 	struct scsi_sense_hdr sshdr;
3278 	struct scsi_device *sdp = sdkp->device;
3279 	int res;
3280 
3281 	if (start)
3282 		cmd[4] |= 1;	/* START */
3283 
3284 	if (sdp->start_stop_pwr_cond)
3285 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3286 
3287 	if (!scsi_device_online(sdp))
3288 		return -ENODEV;
3289 
3290 	res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3291 			SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3292 	if (res) {
3293 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3294 		if (driver_byte(res) & DRIVER_SENSE)
3295 			sd_print_sense_hdr(sdkp, &sshdr);
3296 		if (scsi_sense_valid(&sshdr) &&
3297 			/* 0x3a is medium not present */
3298 			sshdr.asc == 0x3a)
3299 			res = 0;
3300 	}
3301 
3302 	/* SCSI error codes must not go to the generic layer */
3303 	if (res)
3304 		return -EIO;
3305 
3306 	return 0;
3307 }
3308 
3309 /*
3310  * Send a SYNCHRONIZE CACHE instruction down to the device through
3311  * the normal SCSI command structure.  Wait for the command to
3312  * complete.
3313  */
3314 static void sd_shutdown(struct device *dev)
3315 {
3316 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3317 
3318 	if (!sdkp)
3319 		return;         /* this can happen */
3320 
3321 	if (pm_runtime_suspended(dev))
3322 		return;
3323 
3324 	if (sdkp->WCE && sdkp->media_present) {
3325 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3326 		sd_sync_cache(sdkp);
3327 	}
3328 
3329 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3330 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3331 		sd_start_stop_device(sdkp, 0);
3332 	}
3333 }
3334 
3335 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3336 {
3337 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3338 	int ret = 0;
3339 
3340 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3341 		return 0;
3342 
3343 	if (sdkp->WCE && sdkp->media_present) {
3344 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3345 		ret = sd_sync_cache(sdkp);
3346 		if (ret) {
3347 			/* ignore OFFLINE device */
3348 			if (ret == -ENODEV)
3349 				ret = 0;
3350 			goto done;
3351 		}
3352 	}
3353 
3354 	if (sdkp->device->manage_start_stop) {
3355 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3356 		/* an error is not worth aborting a system sleep */
3357 		ret = sd_start_stop_device(sdkp, 0);
3358 		if (ignore_stop_errors)
3359 			ret = 0;
3360 	}
3361 
3362 done:
3363 	return ret;
3364 }
3365 
3366 static int sd_suspend_system(struct device *dev)
3367 {
3368 	return sd_suspend_common(dev, true);
3369 }
3370 
3371 static int sd_suspend_runtime(struct device *dev)
3372 {
3373 	return sd_suspend_common(dev, false);
3374 }
3375 
3376 static int sd_resume(struct device *dev)
3377 {
3378 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3379 
3380 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3381 		return 0;
3382 
3383 	if (!sdkp->device->manage_start_stop)
3384 		return 0;
3385 
3386 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3387 	return sd_start_stop_device(sdkp, 1);
3388 }
3389 
3390 /**
3391  *	init_sd - entry point for this driver (both when built in or when
3392  *	a module).
3393  *
3394  *	Note: this function registers this driver with the scsi mid-level.
3395  **/
3396 static int __init init_sd(void)
3397 {
3398 	int majors = 0, i, err;
3399 
3400 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3401 
3402 	for (i = 0; i < SD_MAJORS; i++) {
3403 		if (register_blkdev(sd_major(i), "sd") != 0)
3404 			continue;
3405 		majors++;
3406 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3407 				    sd_default_probe, NULL, NULL);
3408 	}
3409 
3410 	if (!majors)
3411 		return -ENODEV;
3412 
3413 	err = class_register(&sd_disk_class);
3414 	if (err)
3415 		goto err_out;
3416 
3417 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3418 					 0, 0, NULL);
3419 	if (!sd_cdb_cache) {
3420 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3421 		err = -ENOMEM;
3422 		goto err_out_class;
3423 	}
3424 
3425 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3426 	if (!sd_cdb_pool) {
3427 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3428 		err = -ENOMEM;
3429 		goto err_out_cache;
3430 	}
3431 
3432 	err = scsi_register_driver(&sd_template.gendrv);
3433 	if (err)
3434 		goto err_out_driver;
3435 
3436 	return 0;
3437 
3438 err_out_driver:
3439 	mempool_destroy(sd_cdb_pool);
3440 
3441 err_out_cache:
3442 	kmem_cache_destroy(sd_cdb_cache);
3443 
3444 err_out_class:
3445 	class_unregister(&sd_disk_class);
3446 err_out:
3447 	for (i = 0; i < SD_MAJORS; i++)
3448 		unregister_blkdev(sd_major(i), "sd");
3449 	return err;
3450 }
3451 
3452 /**
3453  *	exit_sd - exit point for this driver (when it is a module).
3454  *
3455  *	Note: this function unregisters this driver from the scsi mid-level.
3456  **/
3457 static void __exit exit_sd(void)
3458 {
3459 	int i;
3460 
3461 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3462 
3463 	scsi_unregister_driver(&sd_template.gendrv);
3464 	mempool_destroy(sd_cdb_pool);
3465 	kmem_cache_destroy(sd_cdb_cache);
3466 
3467 	class_unregister(&sd_disk_class);
3468 
3469 	for (i = 0; i < SD_MAJORS; i++) {
3470 		blk_unregister_region(sd_major(i), SD_MINORS);
3471 		unregister_blkdev(sd_major(i), "sd");
3472 	}
3473 }
3474 
3475 module_init(init_sd);
3476 module_exit(exit_sd);
3477 
3478 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3479 			       struct scsi_sense_hdr *sshdr)
3480 {
3481 	scsi_print_sense_hdr(sdkp->device,
3482 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3483 }
3484 
3485 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3486 			    int result)
3487 {
3488 	const char *hb_string = scsi_hostbyte_string(result);
3489 	const char *db_string = scsi_driverbyte_string(result);
3490 
3491 	if (hb_string || db_string)
3492 		sd_printk(KERN_INFO, sdkp,
3493 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3494 			  hb_string ? hb_string : "invalid",
3495 			  db_string ? db_string : "invalid");
3496 	else
3497 		sd_printk(KERN_INFO, sdkp,
3498 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3499 			  msg, host_byte(result), driver_byte(result));
3500 }
3501 
3502