xref: /openbmc/linux/drivers/scsi/sd.c (revision fc28ab18)
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 one
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 	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 	int ret;
840 
841 	if (sdkp->device->no_write_same)
842 		return BLKPREP_INVALID;
843 
844 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
845 
846 	if (sd_is_zoned(sdkp)) {
847 		ret = sd_zbc_setup_write_cmnd(cmd);
848 		if (ret != BLKPREP_OK)
849 			return ret;
850 	}
851 
852 	sector >>= ilog2(sdp->sector_size) - 9;
853 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
854 
855 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
856 
857 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
858 		cmd->cmd_len = 16;
859 		cmd->cmnd[0] = WRITE_SAME_16;
860 		put_unaligned_be64(sector, &cmd->cmnd[2]);
861 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
862 	} else {
863 		cmd->cmd_len = 10;
864 		cmd->cmnd[0] = WRITE_SAME;
865 		put_unaligned_be32(sector, &cmd->cmnd[2]);
866 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
867 	}
868 
869 	cmd->transfersize = sdp->sector_size;
870 	cmd->allowed = SD_MAX_RETRIES;
871 	return scsi_init_io(cmd);
872 }
873 
874 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
875 {
876 	struct request *rq = cmd->request;
877 
878 	/* flush requests don't perform I/O, zero the S/G table */
879 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
880 
881 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
882 	cmd->cmd_len = 10;
883 	cmd->transfersize = 0;
884 	cmd->allowed = SD_MAX_RETRIES;
885 
886 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
887 	return BLKPREP_OK;
888 }
889 
890 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
891 {
892 	struct request *rq = SCpnt->request;
893 	struct scsi_device *sdp = SCpnt->device;
894 	struct gendisk *disk = rq->rq_disk;
895 	struct scsi_disk *sdkp = scsi_disk(disk);
896 	sector_t block = blk_rq_pos(rq);
897 	sector_t threshold;
898 	unsigned int this_count = blk_rq_sectors(rq);
899 	unsigned int dif, dix;
900 	bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
901 	int ret;
902 	unsigned char protect;
903 
904 	if (zoned_write) {
905 		ret = sd_zbc_setup_write_cmnd(SCpnt);
906 		if (ret != BLKPREP_OK)
907 			return ret;
908 	}
909 
910 	ret = scsi_init_io(SCpnt);
911 	if (ret != BLKPREP_OK)
912 		goto out;
913 	SCpnt = rq->special;
914 
915 	/* from here on until we're complete, any goto out
916 	 * is used for a killable error condition */
917 	ret = BLKPREP_KILL;
918 
919 	SCSI_LOG_HLQUEUE(1,
920 		scmd_printk(KERN_INFO, SCpnt,
921 			"%s: block=%llu, count=%d\n",
922 			__func__, (unsigned long long)block, this_count));
923 
924 	if (!sdp || !scsi_device_online(sdp) ||
925 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
926 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
927 						"Finishing %u sectors\n",
928 						blk_rq_sectors(rq)));
929 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
930 						"Retry with 0x%p\n", SCpnt));
931 		goto out;
932 	}
933 
934 	if (sdp->changed) {
935 		/*
936 		 * quietly refuse to do anything to a changed disc until
937 		 * the changed bit has been reset
938 		 */
939 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
940 		goto out;
941 	}
942 
943 	/*
944 	 * Some SD card readers can't handle multi-sector accesses which touch
945 	 * the last one or two hardware sectors.  Split accesses as needed.
946 	 */
947 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
948 		(sdp->sector_size / 512);
949 
950 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
951 		if (block < threshold) {
952 			/* Access up to the threshold but not beyond */
953 			this_count = threshold - block;
954 		} else {
955 			/* Access only a single hardware sector */
956 			this_count = sdp->sector_size / 512;
957 		}
958 	}
959 
960 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
961 					(unsigned long long)block));
962 
963 	/*
964 	 * If we have a 1K hardware sectorsize, prevent access to single
965 	 * 512 byte sectors.  In theory we could handle this - in fact
966 	 * the scsi cdrom driver must be able to handle this because
967 	 * we typically use 1K blocksizes, and cdroms typically have
968 	 * 2K hardware sectorsizes.  Of course, things are simpler
969 	 * with the cdrom, since it is read-only.  For performance
970 	 * reasons, the filesystems should be able to handle this
971 	 * and not force the scsi disk driver to use bounce buffers
972 	 * for this.
973 	 */
974 	if (sdp->sector_size == 1024) {
975 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
976 			scmd_printk(KERN_ERR, SCpnt,
977 				    "Bad block number requested\n");
978 			goto out;
979 		} else {
980 			block = block >> 1;
981 			this_count = this_count >> 1;
982 		}
983 	}
984 	if (sdp->sector_size == 2048) {
985 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
986 			scmd_printk(KERN_ERR, SCpnt,
987 				    "Bad block number requested\n");
988 			goto out;
989 		} else {
990 			block = block >> 2;
991 			this_count = this_count >> 2;
992 		}
993 	}
994 	if (sdp->sector_size == 4096) {
995 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
996 			scmd_printk(KERN_ERR, SCpnt,
997 				    "Bad block number requested\n");
998 			goto out;
999 		} else {
1000 			block = block >> 3;
1001 			this_count = this_count >> 3;
1002 		}
1003 	}
1004 	if (rq_data_dir(rq) == WRITE) {
1005 		SCpnt->cmnd[0] = WRITE_6;
1006 
1007 		if (blk_integrity_rq(rq))
1008 			sd_dif_prepare(SCpnt);
1009 
1010 	} else if (rq_data_dir(rq) == READ) {
1011 		SCpnt->cmnd[0] = READ_6;
1012 	} else {
1013 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1014 		goto out;
1015 	}
1016 
1017 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1018 					"%s %d/%u 512 byte blocks.\n",
1019 					(rq_data_dir(rq) == WRITE) ?
1020 					"writing" : "reading", this_count,
1021 					blk_rq_sectors(rq)));
1022 
1023 	dix = scsi_prot_sg_count(SCpnt);
1024 	dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1025 
1026 	if (dif || dix)
1027 		protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1028 	else
1029 		protect = 0;
1030 
1031 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1032 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1033 
1034 		if (unlikely(SCpnt->cmnd == NULL)) {
1035 			ret = BLKPREP_DEFER;
1036 			goto out;
1037 		}
1038 
1039 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1040 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1041 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1042 		SCpnt->cmnd[7] = 0x18;
1043 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1044 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1045 
1046 		/* LBA */
1047 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1048 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1049 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1050 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1051 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1052 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1053 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1054 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1055 
1056 		/* Expected Indirect LBA */
1057 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1058 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1059 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1060 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1061 
1062 		/* Transfer length */
1063 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1064 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1065 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1066 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1067 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1068 		SCpnt->cmnd[0] += READ_16 - READ_6;
1069 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1070 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1071 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1072 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1073 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1074 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1075 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1076 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1077 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1078 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1079 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1080 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1081 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1082 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1083 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1084 		   scsi_device_protection(SCpnt->device) ||
1085 		   SCpnt->device->use_10_for_rw) {
1086 		SCpnt->cmnd[0] += READ_10 - READ_6;
1087 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1088 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1089 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1090 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1091 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1092 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1093 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1094 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1095 	} else {
1096 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1097 			/*
1098 			 * This happens only if this drive failed
1099 			 * 10byte rw command with ILLEGAL_REQUEST
1100 			 * during operation and thus turned off
1101 			 * use_10_for_rw.
1102 			 */
1103 			scmd_printk(KERN_ERR, SCpnt,
1104 				    "FUA write on READ/WRITE(6) drive\n");
1105 			goto out;
1106 		}
1107 
1108 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1109 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1110 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1111 		SCpnt->cmnd[4] = (unsigned char) this_count;
1112 		SCpnt->cmnd[5] = 0;
1113 	}
1114 	SCpnt->sdb.length = this_count * sdp->sector_size;
1115 
1116 	/*
1117 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1118 	 * host adapter, it's safe to assume that we can at least transfer
1119 	 * this many bytes between each connect / disconnect.
1120 	 */
1121 	SCpnt->transfersize = sdp->sector_size;
1122 	SCpnt->underflow = this_count << 9;
1123 	SCpnt->allowed = SD_MAX_RETRIES;
1124 
1125 	/*
1126 	 * This indicates that the command is ready from our end to be
1127 	 * queued.
1128 	 */
1129 	ret = BLKPREP_OK;
1130  out:
1131 	if (zoned_write && ret != BLKPREP_OK)
1132 		sd_zbc_cancel_write_cmnd(SCpnt);
1133 
1134 	return ret;
1135 }
1136 
1137 static int sd_init_command(struct scsi_cmnd *cmd)
1138 {
1139 	struct request *rq = cmd->request;
1140 
1141 	switch (req_op(rq)) {
1142 	case REQ_OP_DISCARD:
1143 		return sd_setup_discard_cmnd(cmd);
1144 	case REQ_OP_WRITE_SAME:
1145 		return sd_setup_write_same_cmnd(cmd);
1146 	case REQ_OP_FLUSH:
1147 		return sd_setup_flush_cmnd(cmd);
1148 	case REQ_OP_READ:
1149 	case REQ_OP_WRITE:
1150 		return sd_setup_read_write_cmnd(cmd);
1151 	case REQ_OP_ZONE_REPORT:
1152 		return sd_zbc_setup_report_cmnd(cmd);
1153 	case REQ_OP_ZONE_RESET:
1154 		return sd_zbc_setup_reset_cmnd(cmd);
1155 	default:
1156 		BUG();
1157 	}
1158 }
1159 
1160 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1161 {
1162 	struct request *rq = SCpnt->request;
1163 
1164 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1165 		__free_page(rq->special_vec.bv_page);
1166 
1167 	if (SCpnt->cmnd != rq->cmd) {
1168 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1169 		SCpnt->cmnd = NULL;
1170 		SCpnt->cmd_len = 0;
1171 	}
1172 }
1173 
1174 /**
1175  *	sd_open - open a scsi disk device
1176  *	@inode: only i_rdev member may be used
1177  *	@filp: only f_mode and f_flags may be used
1178  *
1179  *	Returns 0 if successful. Returns a negated errno value in case
1180  *	of error.
1181  *
1182  *	Note: This can be called from a user context (e.g. fsck(1) )
1183  *	or from within the kernel (e.g. as a result of a mount(1) ).
1184  *	In the latter case @inode and @filp carry an abridged amount
1185  *	of information as noted above.
1186  *
1187  *	Locking: called with bdev->bd_mutex held.
1188  **/
1189 static int sd_open(struct block_device *bdev, fmode_t mode)
1190 {
1191 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1192 	struct scsi_device *sdev;
1193 	int retval;
1194 
1195 	if (!sdkp)
1196 		return -ENXIO;
1197 
1198 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1199 
1200 	sdev = sdkp->device;
1201 
1202 	/*
1203 	 * If the device is in error recovery, wait until it is done.
1204 	 * If the device is offline, then disallow any access to it.
1205 	 */
1206 	retval = -ENXIO;
1207 	if (!scsi_block_when_processing_errors(sdev))
1208 		goto error_out;
1209 
1210 	if (sdev->removable || sdkp->write_prot)
1211 		check_disk_change(bdev);
1212 
1213 	/*
1214 	 * If the drive is empty, just let the open fail.
1215 	 */
1216 	retval = -ENOMEDIUM;
1217 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1218 		goto error_out;
1219 
1220 	/*
1221 	 * If the device has the write protect tab set, have the open fail
1222 	 * if the user expects to be able to write to the thing.
1223 	 */
1224 	retval = -EROFS;
1225 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1226 		goto error_out;
1227 
1228 	/*
1229 	 * It is possible that the disk changing stuff resulted in
1230 	 * the device being taken offline.  If this is the case,
1231 	 * report this to the user, and don't pretend that the
1232 	 * open actually succeeded.
1233 	 */
1234 	retval = -ENXIO;
1235 	if (!scsi_device_online(sdev))
1236 		goto error_out;
1237 
1238 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1239 		if (scsi_block_when_processing_errors(sdev))
1240 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1241 	}
1242 
1243 	return 0;
1244 
1245 error_out:
1246 	scsi_disk_put(sdkp);
1247 	return retval;
1248 }
1249 
1250 /**
1251  *	sd_release - invoked when the (last) close(2) is called on this
1252  *	scsi disk.
1253  *	@inode: only i_rdev member may be used
1254  *	@filp: only f_mode and f_flags may be used
1255  *
1256  *	Returns 0.
1257  *
1258  *	Note: may block (uninterruptible) if error recovery is underway
1259  *	on this disk.
1260  *
1261  *	Locking: called with bdev->bd_mutex held.
1262  **/
1263 static void sd_release(struct gendisk *disk, fmode_t mode)
1264 {
1265 	struct scsi_disk *sdkp = scsi_disk(disk);
1266 	struct scsi_device *sdev = sdkp->device;
1267 
1268 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1269 
1270 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1271 		if (scsi_block_when_processing_errors(sdev))
1272 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1273 	}
1274 
1275 	/*
1276 	 * XXX and what if there are packets in flight and this close()
1277 	 * XXX is followed by a "rmmod sd_mod"?
1278 	 */
1279 
1280 	scsi_disk_put(sdkp);
1281 }
1282 
1283 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1284 {
1285 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1286 	struct scsi_device *sdp = sdkp->device;
1287 	struct Scsi_Host *host = sdp->host;
1288 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1289 	int diskinfo[4];
1290 
1291 	/* default to most commonly used values */
1292 	diskinfo[0] = 0x40;	/* 1 << 6 */
1293 	diskinfo[1] = 0x20;	/* 1 << 5 */
1294 	diskinfo[2] = capacity >> 11;
1295 
1296 	/* override with calculated, extended default, or driver values */
1297 	if (host->hostt->bios_param)
1298 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1299 	else
1300 		scsicam_bios_param(bdev, capacity, diskinfo);
1301 
1302 	geo->heads = diskinfo[0];
1303 	geo->sectors = diskinfo[1];
1304 	geo->cylinders = diskinfo[2];
1305 	return 0;
1306 }
1307 
1308 /**
1309  *	sd_ioctl - process an ioctl
1310  *	@inode: only i_rdev/i_bdev members may be used
1311  *	@filp: only f_mode and f_flags may be used
1312  *	@cmd: ioctl command number
1313  *	@arg: this is third argument given to ioctl(2) system call.
1314  *	Often contains a pointer.
1315  *
1316  *	Returns 0 if successful (some ioctls return positive numbers on
1317  *	success as well). Returns a negated errno value in case of error.
1318  *
1319  *	Note: most ioctls are forward onto the block subsystem or further
1320  *	down in the scsi subsystem.
1321  **/
1322 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1323 		    unsigned int cmd, unsigned long arg)
1324 {
1325 	struct gendisk *disk = bdev->bd_disk;
1326 	struct scsi_disk *sdkp = scsi_disk(disk);
1327 	struct scsi_device *sdp = sdkp->device;
1328 	void __user *p = (void __user *)arg;
1329 	int error;
1330 
1331 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1332 				    "cmd=0x%x\n", disk->disk_name, cmd));
1333 
1334 	error = scsi_verify_blk_ioctl(bdev, cmd);
1335 	if (error < 0)
1336 		return error;
1337 
1338 	/*
1339 	 * If we are in the middle of error recovery, don't let anyone
1340 	 * else try and use this device.  Also, if error recovery fails, it
1341 	 * may try and take the device offline, in which case all further
1342 	 * access to the device is prohibited.
1343 	 */
1344 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1345 			(mode & FMODE_NDELAY) != 0);
1346 	if (error)
1347 		goto out;
1348 
1349 	/*
1350 	 * Send SCSI addressing ioctls directly to mid level, send other
1351 	 * ioctls to block level and then onto mid level if they can't be
1352 	 * resolved.
1353 	 */
1354 	switch (cmd) {
1355 		case SCSI_IOCTL_GET_IDLUN:
1356 		case SCSI_IOCTL_GET_BUS_NUMBER:
1357 			error = scsi_ioctl(sdp, cmd, p);
1358 			break;
1359 		default:
1360 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1361 			if (error != -ENOTTY)
1362 				break;
1363 			error = scsi_ioctl(sdp, cmd, p);
1364 			break;
1365 	}
1366 out:
1367 	return error;
1368 }
1369 
1370 static void set_media_not_present(struct scsi_disk *sdkp)
1371 {
1372 	if (sdkp->media_present)
1373 		sdkp->device->changed = 1;
1374 
1375 	if (sdkp->device->removable) {
1376 		sdkp->media_present = 0;
1377 		sdkp->capacity = 0;
1378 	}
1379 }
1380 
1381 static int media_not_present(struct scsi_disk *sdkp,
1382 			     struct scsi_sense_hdr *sshdr)
1383 {
1384 	if (!scsi_sense_valid(sshdr))
1385 		return 0;
1386 
1387 	/* not invoked for commands that could return deferred errors */
1388 	switch (sshdr->sense_key) {
1389 	case UNIT_ATTENTION:
1390 	case NOT_READY:
1391 		/* medium not present */
1392 		if (sshdr->asc == 0x3A) {
1393 			set_media_not_present(sdkp);
1394 			return 1;
1395 		}
1396 	}
1397 	return 0;
1398 }
1399 
1400 /**
1401  *	sd_check_events - check media events
1402  *	@disk: kernel device descriptor
1403  *	@clearing: disk events currently being cleared
1404  *
1405  *	Returns mask of DISK_EVENT_*.
1406  *
1407  *	Note: this function is invoked from the block subsystem.
1408  **/
1409 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1410 {
1411 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1412 	struct scsi_device *sdp;
1413 	struct scsi_sense_hdr *sshdr = NULL;
1414 	int retval;
1415 
1416 	if (!sdkp)
1417 		return 0;
1418 
1419 	sdp = sdkp->device;
1420 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1421 
1422 	/*
1423 	 * If the device is offline, don't send any commands - just pretend as
1424 	 * if the command failed.  If the device ever comes back online, we
1425 	 * can deal with it then.  It is only because of unrecoverable errors
1426 	 * that we would ever take a device offline in the first place.
1427 	 */
1428 	if (!scsi_device_online(sdp)) {
1429 		set_media_not_present(sdkp);
1430 		goto out;
1431 	}
1432 
1433 	/*
1434 	 * Using TEST_UNIT_READY enables differentiation between drive with
1435 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1436 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1437 	 *
1438 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1439 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1440 	 * sd_revalidate() is called.
1441 	 */
1442 	retval = -ENODEV;
1443 
1444 	if (scsi_block_when_processing_errors(sdp)) {
1445 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1446 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1447 					      sshdr);
1448 	}
1449 
1450 	/* failed to execute TUR, assume media not present */
1451 	if (host_byte(retval)) {
1452 		set_media_not_present(sdkp);
1453 		goto out;
1454 	}
1455 
1456 	if (media_not_present(sdkp, sshdr))
1457 		goto out;
1458 
1459 	/*
1460 	 * For removable scsi disk we have to recognise the presence
1461 	 * of a disk in the drive.
1462 	 */
1463 	if (!sdkp->media_present)
1464 		sdp->changed = 1;
1465 	sdkp->media_present = 1;
1466 out:
1467 	/*
1468 	 * sdp->changed is set under the following conditions:
1469 	 *
1470 	 *	Medium present state has changed in either direction.
1471 	 *	Device has indicated UNIT_ATTENTION.
1472 	 */
1473 	kfree(sshdr);
1474 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1475 	sdp->changed = 0;
1476 	scsi_disk_put(sdkp);
1477 	return retval;
1478 }
1479 
1480 static int sd_sync_cache(struct scsi_disk *sdkp)
1481 {
1482 	int retries, res;
1483 	struct scsi_device *sdp = sdkp->device;
1484 	const int timeout = sdp->request_queue->rq_timeout
1485 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1486 	struct scsi_sense_hdr sshdr;
1487 
1488 	if (!scsi_device_online(sdp))
1489 		return -ENODEV;
1490 
1491 	for (retries = 3; retries > 0; --retries) {
1492 		unsigned char cmd[10] = { 0 };
1493 
1494 		cmd[0] = SYNCHRONIZE_CACHE;
1495 		/*
1496 		 * Leave the rest of the command zero to indicate
1497 		 * flush everything.
1498 		 */
1499 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1500 					     &sshdr, timeout, SD_MAX_RETRIES,
1501 					     NULL, 0, RQF_PM);
1502 		if (res == 0)
1503 			break;
1504 	}
1505 
1506 	if (res) {
1507 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1508 
1509 		if (driver_byte(res) & DRIVER_SENSE)
1510 			sd_print_sense_hdr(sdkp, &sshdr);
1511 		/* we need to evaluate the error return  */
1512 		if (scsi_sense_valid(&sshdr) &&
1513 			(sshdr.asc == 0x3a ||	/* medium not present */
1514 			 sshdr.asc == 0x20))	/* invalid command */
1515 				/* this is no error here */
1516 				return 0;
1517 
1518 		switch (host_byte(res)) {
1519 		/* ignore errors due to racing a disconnection */
1520 		case DID_BAD_TARGET:
1521 		case DID_NO_CONNECT:
1522 			return 0;
1523 		/* signal the upper layer it might try again */
1524 		case DID_BUS_BUSY:
1525 		case DID_IMM_RETRY:
1526 		case DID_REQUEUE:
1527 		case DID_SOFT_ERROR:
1528 			return -EBUSY;
1529 		default:
1530 			return -EIO;
1531 		}
1532 	}
1533 	return 0;
1534 }
1535 
1536 static void sd_rescan(struct device *dev)
1537 {
1538 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1539 
1540 	revalidate_disk(sdkp->disk);
1541 }
1542 
1543 
1544 #ifdef CONFIG_COMPAT
1545 /*
1546  * This gets directly called from VFS. When the ioctl
1547  * is not recognized we go back to the other translation paths.
1548  */
1549 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1550 			   unsigned int cmd, unsigned long arg)
1551 {
1552 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1553 	int error;
1554 
1555 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1556 			(mode & FMODE_NDELAY) != 0);
1557 	if (error)
1558 		return error;
1559 
1560 	/*
1561 	 * Let the static ioctl translation table take care of it.
1562 	 */
1563 	if (!sdev->host->hostt->compat_ioctl)
1564 		return -ENOIOCTLCMD;
1565 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1566 }
1567 #endif
1568 
1569 static char sd_pr_type(enum pr_type type)
1570 {
1571 	switch (type) {
1572 	case PR_WRITE_EXCLUSIVE:
1573 		return 0x01;
1574 	case PR_EXCLUSIVE_ACCESS:
1575 		return 0x03;
1576 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1577 		return 0x05;
1578 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1579 		return 0x06;
1580 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1581 		return 0x07;
1582 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1583 		return 0x08;
1584 	default:
1585 		return 0;
1586 	}
1587 };
1588 
1589 static int sd_pr_command(struct block_device *bdev, u8 sa,
1590 		u64 key, u64 sa_key, u8 type, u8 flags)
1591 {
1592 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1593 	struct scsi_sense_hdr sshdr;
1594 	int result;
1595 	u8 cmd[16] = { 0, };
1596 	u8 data[24] = { 0, };
1597 
1598 	cmd[0] = PERSISTENT_RESERVE_OUT;
1599 	cmd[1] = sa;
1600 	cmd[2] = type;
1601 	put_unaligned_be32(sizeof(data), &cmd[5]);
1602 
1603 	put_unaligned_be64(key, &data[0]);
1604 	put_unaligned_be64(sa_key, &data[8]);
1605 	data[20] = flags;
1606 
1607 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1608 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1609 
1610 	if ((driver_byte(result) & DRIVER_SENSE) &&
1611 	    (scsi_sense_valid(&sshdr))) {
1612 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1613 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1614 	}
1615 
1616 	return result;
1617 }
1618 
1619 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1620 		u32 flags)
1621 {
1622 	if (flags & ~PR_FL_IGNORE_KEY)
1623 		return -EOPNOTSUPP;
1624 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1625 			old_key, new_key, 0,
1626 			(1 << 0) /* APTPL */);
1627 }
1628 
1629 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1630 		u32 flags)
1631 {
1632 	if (flags)
1633 		return -EOPNOTSUPP;
1634 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1635 }
1636 
1637 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1638 {
1639 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1640 }
1641 
1642 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1643 		enum pr_type type, bool abort)
1644 {
1645 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1646 			     sd_pr_type(type), 0);
1647 }
1648 
1649 static int sd_pr_clear(struct block_device *bdev, u64 key)
1650 {
1651 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1652 }
1653 
1654 static const struct pr_ops sd_pr_ops = {
1655 	.pr_register	= sd_pr_register,
1656 	.pr_reserve	= sd_pr_reserve,
1657 	.pr_release	= sd_pr_release,
1658 	.pr_preempt	= sd_pr_preempt,
1659 	.pr_clear	= sd_pr_clear,
1660 };
1661 
1662 static const struct block_device_operations sd_fops = {
1663 	.owner			= THIS_MODULE,
1664 	.open			= sd_open,
1665 	.release		= sd_release,
1666 	.ioctl			= sd_ioctl,
1667 	.getgeo			= sd_getgeo,
1668 #ifdef CONFIG_COMPAT
1669 	.compat_ioctl		= sd_compat_ioctl,
1670 #endif
1671 	.check_events		= sd_check_events,
1672 	.revalidate_disk	= sd_revalidate_disk,
1673 	.unlock_native_capacity	= sd_unlock_native_capacity,
1674 	.pr_ops			= &sd_pr_ops,
1675 };
1676 
1677 /**
1678  *	sd_eh_action - error handling callback
1679  *	@scmd:		sd-issued command that has failed
1680  *	@eh_disp:	The recovery disposition suggested by the midlayer
1681  *
1682  *	This function is called by the SCSI midlayer upon completion of an
1683  *	error test command (currently TEST UNIT READY). The result of sending
1684  *	the eh command is passed in eh_disp.  We're looking for devices that
1685  *	fail medium access commands but are OK with non access commands like
1686  *	test unit ready (so wrongly see the device as having a successful
1687  *	recovery)
1688  **/
1689 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1690 {
1691 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1692 
1693 	if (!scsi_device_online(scmd->device) ||
1694 	    !scsi_medium_access_command(scmd) ||
1695 	    host_byte(scmd->result) != DID_TIME_OUT ||
1696 	    eh_disp != SUCCESS)
1697 		return eh_disp;
1698 
1699 	/*
1700 	 * The device has timed out executing a medium access command.
1701 	 * However, the TEST UNIT READY command sent during error
1702 	 * handling completed successfully. Either the device is in the
1703 	 * process of recovering or has it suffered an internal failure
1704 	 * that prevents access to the storage medium.
1705 	 */
1706 	sdkp->medium_access_timed_out++;
1707 
1708 	/*
1709 	 * If the device keeps failing read/write commands but TEST UNIT
1710 	 * READY always completes successfully we assume that medium
1711 	 * access is no longer possible and take the device offline.
1712 	 */
1713 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1714 		scmd_printk(KERN_ERR, scmd,
1715 			    "Medium access timeout failure. Offlining disk!\n");
1716 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1717 
1718 		return FAILED;
1719 	}
1720 
1721 	return eh_disp;
1722 }
1723 
1724 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1725 {
1726 	u64 start_lba = blk_rq_pos(scmd->request);
1727 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1728 	u64 factor = scmd->device->sector_size / 512;
1729 	u64 bad_lba;
1730 	int info_valid;
1731 	/*
1732 	 * resid is optional but mostly filled in.  When it's unused,
1733 	 * its value is zero, so we assume the whole buffer transferred
1734 	 */
1735 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1736 	unsigned int good_bytes;
1737 
1738 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1739 		return 0;
1740 
1741 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1742 					     SCSI_SENSE_BUFFERSIZE,
1743 					     &bad_lba);
1744 	if (!info_valid)
1745 		return 0;
1746 
1747 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1748 		return 0;
1749 
1750 	/* be careful ... don't want any overflows */
1751 	do_div(start_lba, factor);
1752 	do_div(end_lba, factor);
1753 
1754 	/* The bad lba was reported incorrectly, we have no idea where
1755 	 * the error is.
1756 	 */
1757 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1758 		return 0;
1759 
1760 	/* This computation should always be done in terms of
1761 	 * the resolution of the device's medium.
1762 	 */
1763 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1764 	return min(good_bytes, transferred);
1765 }
1766 
1767 /**
1768  *	sd_done - bottom half handler: called when the lower level
1769  *	driver has completed (successfully or otherwise) a scsi command.
1770  *	@SCpnt: mid-level's per command structure.
1771  *
1772  *	Note: potentially run from within an ISR. Must not block.
1773  **/
1774 static int sd_done(struct scsi_cmnd *SCpnt)
1775 {
1776 	int result = SCpnt->result;
1777 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1778 	struct scsi_sense_hdr sshdr;
1779 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1780 	struct request *req = SCpnt->request;
1781 	int sense_valid = 0;
1782 	int sense_deferred = 0;
1783 	unsigned char op = SCpnt->cmnd[0];
1784 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1785 
1786 	switch (req_op(req)) {
1787 	case REQ_OP_DISCARD:
1788 	case REQ_OP_WRITE_SAME:
1789 	case REQ_OP_ZONE_RESET:
1790 		if (!result) {
1791 			good_bytes = blk_rq_bytes(req);
1792 			scsi_set_resid(SCpnt, 0);
1793 		} else {
1794 			good_bytes = 0;
1795 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1796 		}
1797 		break;
1798 	case REQ_OP_ZONE_REPORT:
1799 		if (!result) {
1800 			good_bytes = scsi_bufflen(SCpnt)
1801 				- scsi_get_resid(SCpnt);
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 	}
1809 
1810 	if (result) {
1811 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1812 		if (sense_valid)
1813 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1814 	}
1815 	sdkp->medium_access_timed_out = 0;
1816 
1817 	if (driver_byte(result) != DRIVER_SENSE &&
1818 	    (!sense_valid || sense_deferred))
1819 		goto out;
1820 
1821 	switch (sshdr.sense_key) {
1822 	case HARDWARE_ERROR:
1823 	case MEDIUM_ERROR:
1824 		good_bytes = sd_completed_bytes(SCpnt);
1825 		break;
1826 	case RECOVERED_ERROR:
1827 		good_bytes = scsi_bufflen(SCpnt);
1828 		break;
1829 	case NO_SENSE:
1830 		/* This indicates a false check condition, so ignore it.  An
1831 		 * unknown amount of data was transferred so treat it as an
1832 		 * error.
1833 		 */
1834 		SCpnt->result = 0;
1835 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1836 		break;
1837 	case ABORTED_COMMAND:
1838 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1839 			good_bytes = sd_completed_bytes(SCpnt);
1840 		break;
1841 	case ILLEGAL_REQUEST:
1842 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1843 			good_bytes = sd_completed_bytes(SCpnt);
1844 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1845 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1846 			switch (op) {
1847 			case UNMAP:
1848 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1849 				break;
1850 			case WRITE_SAME_16:
1851 			case WRITE_SAME:
1852 				if (unmap)
1853 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1854 				else {
1855 					sdkp->device->no_write_same = 1;
1856 					sd_config_write_same(sdkp);
1857 
1858 					good_bytes = 0;
1859 					req->__data_len = blk_rq_bytes(req);
1860 					req->rq_flags |= RQF_QUIET;
1861 				}
1862 			}
1863 		}
1864 		break;
1865 	default:
1866 		break;
1867 	}
1868 
1869  out:
1870 	if (sd_is_zoned(sdkp))
1871 		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
1872 
1873 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1874 					   "sd_done: completed %d of %d bytes\n",
1875 					   good_bytes, scsi_bufflen(SCpnt)));
1876 
1877 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1878 		sd_dif_complete(SCpnt, good_bytes);
1879 
1880 	return good_bytes;
1881 }
1882 
1883 /*
1884  * spinup disk - called only in sd_revalidate_disk()
1885  */
1886 static void
1887 sd_spinup_disk(struct scsi_disk *sdkp)
1888 {
1889 	unsigned char cmd[10];
1890 	unsigned long spintime_expire = 0;
1891 	int retries, spintime;
1892 	unsigned int the_result;
1893 	struct scsi_sense_hdr sshdr;
1894 	int sense_valid = 0;
1895 
1896 	spintime = 0;
1897 
1898 	/* Spin up drives, as required.  Only do this at boot time */
1899 	/* Spinup needs to be done for module loads too. */
1900 	do {
1901 		retries = 0;
1902 
1903 		do {
1904 			cmd[0] = TEST_UNIT_READY;
1905 			memset((void *) &cmd[1], 0, 9);
1906 
1907 			the_result = scsi_execute_req(sdkp->device, cmd,
1908 						      DMA_NONE, NULL, 0,
1909 						      &sshdr, SD_TIMEOUT,
1910 						      SD_MAX_RETRIES, NULL);
1911 
1912 			/*
1913 			 * If the drive has indicated to us that it
1914 			 * doesn't have any media in it, don't bother
1915 			 * with any more polling.
1916 			 */
1917 			if (media_not_present(sdkp, &sshdr))
1918 				return;
1919 
1920 			if (the_result)
1921 				sense_valid = scsi_sense_valid(&sshdr);
1922 			retries++;
1923 		} while (retries < 3 &&
1924 			 (!scsi_status_is_good(the_result) ||
1925 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1926 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1927 
1928 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1929 			/* no sense, TUR either succeeded or failed
1930 			 * with a status error */
1931 			if(!spintime && !scsi_status_is_good(the_result)) {
1932 				sd_print_result(sdkp, "Test Unit Ready failed",
1933 						the_result);
1934 			}
1935 			break;
1936 		}
1937 
1938 		/*
1939 		 * The device does not want the automatic start to be issued.
1940 		 */
1941 		if (sdkp->device->no_start_on_add)
1942 			break;
1943 
1944 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1945 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1946 				break;	/* manual intervention required */
1947 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1948 				break;	/* standby */
1949 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1950 				break;	/* unavailable */
1951 			/*
1952 			 * Issue command to spin up drive when not ready
1953 			 */
1954 			if (!spintime) {
1955 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1956 				cmd[0] = START_STOP;
1957 				cmd[1] = 1;	/* Return immediately */
1958 				memset((void *) &cmd[2], 0, 8);
1959 				cmd[4] = 1;	/* Start spin cycle */
1960 				if (sdkp->device->start_stop_pwr_cond)
1961 					cmd[4] |= 1 << 4;
1962 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1963 						 NULL, 0, &sshdr,
1964 						 SD_TIMEOUT, SD_MAX_RETRIES,
1965 						 NULL);
1966 				spintime_expire = jiffies + 100 * HZ;
1967 				spintime = 1;
1968 			}
1969 			/* Wait 1 second for next try */
1970 			msleep(1000);
1971 			printk(".");
1972 
1973 		/*
1974 		 * Wait for USB flash devices with slow firmware.
1975 		 * Yes, this sense key/ASC combination shouldn't
1976 		 * occur here.  It's characteristic of these devices.
1977 		 */
1978 		} else if (sense_valid &&
1979 				sshdr.sense_key == UNIT_ATTENTION &&
1980 				sshdr.asc == 0x28) {
1981 			if (!spintime) {
1982 				spintime_expire = jiffies + 5 * HZ;
1983 				spintime = 1;
1984 			}
1985 			/* Wait 1 second for next try */
1986 			msleep(1000);
1987 		} else {
1988 			/* we don't understand the sense code, so it's
1989 			 * probably pointless to loop */
1990 			if(!spintime) {
1991 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1992 				sd_print_sense_hdr(sdkp, &sshdr);
1993 			}
1994 			break;
1995 		}
1996 
1997 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1998 
1999 	if (spintime) {
2000 		if (scsi_status_is_good(the_result))
2001 			printk("ready\n");
2002 		else
2003 			printk("not responding...\n");
2004 	}
2005 }
2006 
2007 /*
2008  * Determine whether disk supports Data Integrity Field.
2009  */
2010 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2011 {
2012 	struct scsi_device *sdp = sdkp->device;
2013 	u8 type;
2014 	int ret = 0;
2015 
2016 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2017 		return ret;
2018 
2019 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2020 
2021 	if (type > T10_PI_TYPE3_PROTECTION)
2022 		ret = -ENODEV;
2023 	else if (scsi_host_dif_capable(sdp->host, type))
2024 		ret = 1;
2025 
2026 	if (sdkp->first_scan || type != sdkp->protection_type)
2027 		switch (ret) {
2028 		case -ENODEV:
2029 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2030 				  " protection type %u. Disabling disk!\n",
2031 				  type);
2032 			break;
2033 		case 1:
2034 			sd_printk(KERN_NOTICE, sdkp,
2035 				  "Enabling DIF Type %u protection\n", type);
2036 			break;
2037 		case 0:
2038 			sd_printk(KERN_NOTICE, sdkp,
2039 				  "Disabling DIF Type %u protection\n", type);
2040 			break;
2041 		}
2042 
2043 	sdkp->protection_type = type;
2044 
2045 	return ret;
2046 }
2047 
2048 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2049 			struct scsi_sense_hdr *sshdr, int sense_valid,
2050 			int the_result)
2051 {
2052 	if (driver_byte(the_result) & DRIVER_SENSE)
2053 		sd_print_sense_hdr(sdkp, sshdr);
2054 	else
2055 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2056 
2057 	/*
2058 	 * Set dirty bit for removable devices if not ready -
2059 	 * sometimes drives will not report this properly.
2060 	 */
2061 	if (sdp->removable &&
2062 	    sense_valid && sshdr->sense_key == NOT_READY)
2063 		set_media_not_present(sdkp);
2064 
2065 	/*
2066 	 * We used to set media_present to 0 here to indicate no media
2067 	 * in the drive, but some drives fail read capacity even with
2068 	 * media present, so we can't do that.
2069 	 */
2070 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2071 }
2072 
2073 #define RC16_LEN 32
2074 #if RC16_LEN > SD_BUF_SIZE
2075 #error RC16_LEN must not be more than SD_BUF_SIZE
2076 #endif
2077 
2078 #define READ_CAPACITY_RETRIES_ON_RESET	10
2079 
2080 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2081 						unsigned char *buffer)
2082 {
2083 	unsigned char cmd[16];
2084 	struct scsi_sense_hdr sshdr;
2085 	int sense_valid = 0;
2086 	int the_result;
2087 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2088 	unsigned int alignment;
2089 	unsigned long long lba;
2090 	unsigned sector_size;
2091 
2092 	if (sdp->no_read_capacity_16)
2093 		return -EINVAL;
2094 
2095 	do {
2096 		memset(cmd, 0, 16);
2097 		cmd[0] = SERVICE_ACTION_IN_16;
2098 		cmd[1] = SAI_READ_CAPACITY_16;
2099 		cmd[13] = RC16_LEN;
2100 		memset(buffer, 0, RC16_LEN);
2101 
2102 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2103 					buffer, RC16_LEN, &sshdr,
2104 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2105 
2106 		if (media_not_present(sdkp, &sshdr))
2107 			return -ENODEV;
2108 
2109 		if (the_result) {
2110 			sense_valid = scsi_sense_valid(&sshdr);
2111 			if (sense_valid &&
2112 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2113 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2114 			    sshdr.ascq == 0x00)
2115 				/* Invalid Command Operation Code or
2116 				 * Invalid Field in CDB, just retry
2117 				 * silently with RC10 */
2118 				return -EINVAL;
2119 			if (sense_valid &&
2120 			    sshdr.sense_key == UNIT_ATTENTION &&
2121 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2122 				/* Device reset might occur several times,
2123 				 * give it one more chance */
2124 				if (--reset_retries > 0)
2125 					continue;
2126 		}
2127 		retries--;
2128 
2129 	} while (the_result && retries);
2130 
2131 	if (the_result) {
2132 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2133 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2134 		return -EINVAL;
2135 	}
2136 
2137 	sector_size = get_unaligned_be32(&buffer[8]);
2138 	lba = get_unaligned_be64(&buffer[0]);
2139 
2140 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2141 		sdkp->capacity = 0;
2142 		return -ENODEV;
2143 	}
2144 
2145 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2146 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2147 			"kernel compiled with support for large block "
2148 			"devices.\n");
2149 		sdkp->capacity = 0;
2150 		return -EOVERFLOW;
2151 	}
2152 
2153 	/* Logical blocks per physical block exponent */
2154 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2155 
2156 	/* RC basis */
2157 	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2158 
2159 	/* Lowest aligned logical block */
2160 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2161 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2162 	if (alignment && sdkp->first_scan)
2163 		sd_printk(KERN_NOTICE, sdkp,
2164 			  "physical block alignment offset: %u\n", alignment);
2165 
2166 	if (buffer[14] & 0x80) { /* LBPME */
2167 		sdkp->lbpme = 1;
2168 
2169 		if (buffer[14] & 0x40) /* LBPRZ */
2170 			sdkp->lbprz = 1;
2171 
2172 		sd_config_discard(sdkp, SD_LBP_WS16);
2173 	}
2174 
2175 	sdkp->capacity = lba + 1;
2176 	return sector_size;
2177 }
2178 
2179 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2180 						unsigned char *buffer)
2181 {
2182 	unsigned char cmd[16];
2183 	struct scsi_sense_hdr sshdr;
2184 	int sense_valid = 0;
2185 	int the_result;
2186 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2187 	sector_t lba;
2188 	unsigned sector_size;
2189 
2190 	do {
2191 		cmd[0] = READ_CAPACITY;
2192 		memset(&cmd[1], 0, 9);
2193 		memset(buffer, 0, 8);
2194 
2195 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2196 					buffer, 8, &sshdr,
2197 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2198 
2199 		if (media_not_present(sdkp, &sshdr))
2200 			return -ENODEV;
2201 
2202 		if (the_result) {
2203 			sense_valid = scsi_sense_valid(&sshdr);
2204 			if (sense_valid &&
2205 			    sshdr.sense_key == UNIT_ATTENTION &&
2206 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2207 				/* Device reset might occur several times,
2208 				 * give it one more chance */
2209 				if (--reset_retries > 0)
2210 					continue;
2211 		}
2212 		retries--;
2213 
2214 	} while (the_result && retries);
2215 
2216 	if (the_result) {
2217 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2218 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2219 		return -EINVAL;
2220 	}
2221 
2222 	sector_size = get_unaligned_be32(&buffer[4]);
2223 	lba = get_unaligned_be32(&buffer[0]);
2224 
2225 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2226 		/* Some buggy (usb cardreader) devices return an lba of
2227 		   0xffffffff when the want to report a size of 0 (with
2228 		   which they really mean no media is present) */
2229 		sdkp->capacity = 0;
2230 		sdkp->physical_block_size = sector_size;
2231 		return sector_size;
2232 	}
2233 
2234 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2235 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2236 			"kernel compiled with support for large block "
2237 			"devices.\n");
2238 		sdkp->capacity = 0;
2239 		return -EOVERFLOW;
2240 	}
2241 
2242 	sdkp->capacity = lba + 1;
2243 	sdkp->physical_block_size = sector_size;
2244 	return sector_size;
2245 }
2246 
2247 static int sd_try_rc16_first(struct scsi_device *sdp)
2248 {
2249 	if (sdp->host->max_cmd_len < 16)
2250 		return 0;
2251 	if (sdp->try_rc_10_first)
2252 		return 0;
2253 	if (sdp->scsi_level > SCSI_SPC_2)
2254 		return 1;
2255 	if (scsi_device_protection(sdp))
2256 		return 1;
2257 	return 0;
2258 }
2259 
2260 /*
2261  * read disk capacity
2262  */
2263 static void
2264 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2265 {
2266 	int sector_size;
2267 	struct scsi_device *sdp = sdkp->device;
2268 
2269 	if (sd_try_rc16_first(sdp)) {
2270 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2271 		if (sector_size == -EOVERFLOW)
2272 			goto got_data;
2273 		if (sector_size == -ENODEV)
2274 			return;
2275 		if (sector_size < 0)
2276 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2277 		if (sector_size < 0)
2278 			return;
2279 	} else {
2280 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2281 		if (sector_size == -EOVERFLOW)
2282 			goto got_data;
2283 		if (sector_size < 0)
2284 			return;
2285 		if ((sizeof(sdkp->capacity) > 4) &&
2286 		    (sdkp->capacity > 0xffffffffULL)) {
2287 			int old_sector_size = sector_size;
2288 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2289 					"Trying to use READ CAPACITY(16).\n");
2290 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2291 			if (sector_size < 0) {
2292 				sd_printk(KERN_NOTICE, sdkp,
2293 					"Using 0xffffffff as device size\n");
2294 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2295 				sector_size = old_sector_size;
2296 				goto got_data;
2297 			}
2298 		}
2299 	}
2300 
2301 	/* Some devices are known to return the total number of blocks,
2302 	 * not the highest block number.  Some devices have versions
2303 	 * which do this and others which do not.  Some devices we might
2304 	 * suspect of doing this but we don't know for certain.
2305 	 *
2306 	 * If we know the reported capacity is wrong, decrement it.  If
2307 	 * we can only guess, then assume the number of blocks is even
2308 	 * (usually true but not always) and err on the side of lowering
2309 	 * the capacity.
2310 	 */
2311 	if (sdp->fix_capacity ||
2312 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2313 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2314 				"from its reported value: %llu\n",
2315 				(unsigned long long) sdkp->capacity);
2316 		--sdkp->capacity;
2317 	}
2318 
2319 got_data:
2320 	if (sector_size == 0) {
2321 		sector_size = 512;
2322 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2323 			  "assuming 512.\n");
2324 	}
2325 
2326 	if (sector_size != 512 &&
2327 	    sector_size != 1024 &&
2328 	    sector_size != 2048 &&
2329 	    sector_size != 4096) {
2330 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2331 			  sector_size);
2332 		/*
2333 		 * The user might want to re-format the drive with
2334 		 * a supported sectorsize.  Once this happens, it
2335 		 * would be relatively trivial to set the thing up.
2336 		 * For this reason, we leave the thing in the table.
2337 		 */
2338 		sdkp->capacity = 0;
2339 		/*
2340 		 * set a bogus sector size so the normal read/write
2341 		 * logic in the block layer will eventually refuse any
2342 		 * request on this device without tripping over power
2343 		 * of two sector size assumptions
2344 		 */
2345 		sector_size = 512;
2346 	}
2347 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2348 	blk_queue_physical_block_size(sdp->request_queue,
2349 				      sdkp->physical_block_size);
2350 	sdkp->device->sector_size = sector_size;
2351 
2352 	if (sdkp->capacity > 0xffffffff)
2353 		sdp->use_16_for_rw = 1;
2354 
2355 }
2356 
2357 /*
2358  * Print disk capacity
2359  */
2360 static void
2361 sd_print_capacity(struct scsi_disk *sdkp,
2362 		  sector_t old_capacity)
2363 {
2364 	int sector_size = sdkp->device->sector_size;
2365 	char cap_str_2[10], cap_str_10[10];
2366 
2367 	string_get_size(sdkp->capacity, sector_size,
2368 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2369 	string_get_size(sdkp->capacity, sector_size,
2370 			STRING_UNITS_10, cap_str_10,
2371 			sizeof(cap_str_10));
2372 
2373 	if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2374 		sd_printk(KERN_NOTICE, sdkp,
2375 			  "%llu %d-byte logical blocks: (%s/%s)\n",
2376 			  (unsigned long long)sdkp->capacity,
2377 			  sector_size, cap_str_10, cap_str_2);
2378 
2379 		if (sdkp->physical_block_size != sector_size)
2380 			sd_printk(KERN_NOTICE, sdkp,
2381 				  "%u-byte physical blocks\n",
2382 				  sdkp->physical_block_size);
2383 
2384 		sd_zbc_print_zones(sdkp);
2385 	}
2386 }
2387 
2388 /* called with buffer of length 512 */
2389 static inline int
2390 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2391 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2392 		 struct scsi_sense_hdr *sshdr)
2393 {
2394 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2395 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2396 			       sshdr);
2397 }
2398 
2399 /*
2400  * read write protect setting, if possible - called only in sd_revalidate_disk()
2401  * called with buffer of length SD_BUF_SIZE
2402  */
2403 static void
2404 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2405 {
2406 	int res;
2407 	struct scsi_device *sdp = sdkp->device;
2408 	struct scsi_mode_data data;
2409 	int old_wp = sdkp->write_prot;
2410 
2411 	set_disk_ro(sdkp->disk, 0);
2412 	if (sdp->skip_ms_page_3f) {
2413 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2414 		return;
2415 	}
2416 
2417 	if (sdp->use_192_bytes_for_3f) {
2418 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2419 	} else {
2420 		/*
2421 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2422 		 * We have to start carefully: some devices hang if we ask
2423 		 * for more than is available.
2424 		 */
2425 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2426 
2427 		/*
2428 		 * Second attempt: ask for page 0 When only page 0 is
2429 		 * implemented, a request for page 3F may return Sense Key
2430 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2431 		 * CDB.
2432 		 */
2433 		if (!scsi_status_is_good(res))
2434 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2435 
2436 		/*
2437 		 * Third attempt: ask 255 bytes, as we did earlier.
2438 		 */
2439 		if (!scsi_status_is_good(res))
2440 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2441 					       &data, NULL);
2442 	}
2443 
2444 	if (!scsi_status_is_good(res)) {
2445 		sd_first_printk(KERN_WARNING, sdkp,
2446 			  "Test WP failed, assume Write Enabled\n");
2447 	} else {
2448 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2449 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2450 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2451 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2452 				  sdkp->write_prot ? "on" : "off");
2453 			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2454 		}
2455 	}
2456 }
2457 
2458 /*
2459  * sd_read_cache_type - called only from sd_revalidate_disk()
2460  * called with buffer of length SD_BUF_SIZE
2461  */
2462 static void
2463 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2464 {
2465 	int len = 0, res;
2466 	struct scsi_device *sdp = sdkp->device;
2467 
2468 	int dbd;
2469 	int modepage;
2470 	int first_len;
2471 	struct scsi_mode_data data;
2472 	struct scsi_sense_hdr sshdr;
2473 	int old_wce = sdkp->WCE;
2474 	int old_rcd = sdkp->RCD;
2475 	int old_dpofua = sdkp->DPOFUA;
2476 
2477 
2478 	if (sdkp->cache_override)
2479 		return;
2480 
2481 	first_len = 4;
2482 	if (sdp->skip_ms_page_8) {
2483 		if (sdp->type == TYPE_RBC)
2484 			goto defaults;
2485 		else {
2486 			if (sdp->skip_ms_page_3f)
2487 				goto defaults;
2488 			modepage = 0x3F;
2489 			if (sdp->use_192_bytes_for_3f)
2490 				first_len = 192;
2491 			dbd = 0;
2492 		}
2493 	} else if (sdp->type == TYPE_RBC) {
2494 		modepage = 6;
2495 		dbd = 8;
2496 	} else {
2497 		modepage = 8;
2498 		dbd = 0;
2499 	}
2500 
2501 	/* cautiously ask */
2502 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2503 			&data, &sshdr);
2504 
2505 	if (!scsi_status_is_good(res))
2506 		goto bad_sense;
2507 
2508 	if (!data.header_length) {
2509 		modepage = 6;
2510 		first_len = 0;
2511 		sd_first_printk(KERN_ERR, sdkp,
2512 				"Missing header in MODE_SENSE response\n");
2513 	}
2514 
2515 	/* that went OK, now ask for the proper length */
2516 	len = data.length;
2517 
2518 	/*
2519 	 * We're only interested in the first three bytes, actually.
2520 	 * But the data cache page is defined for the first 20.
2521 	 */
2522 	if (len < 3)
2523 		goto bad_sense;
2524 	else if (len > SD_BUF_SIZE) {
2525 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2526 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2527 		len = SD_BUF_SIZE;
2528 	}
2529 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2530 		len = 192;
2531 
2532 	/* Get the data */
2533 	if (len > first_len)
2534 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2535 				&data, &sshdr);
2536 
2537 	if (scsi_status_is_good(res)) {
2538 		int offset = data.header_length + data.block_descriptor_length;
2539 
2540 		while (offset < len) {
2541 			u8 page_code = buffer[offset] & 0x3F;
2542 			u8 spf       = buffer[offset] & 0x40;
2543 
2544 			if (page_code == 8 || page_code == 6) {
2545 				/* We're interested only in the first 3 bytes.
2546 				 */
2547 				if (len - offset <= 2) {
2548 					sd_first_printk(KERN_ERR, sdkp,
2549 						"Incomplete mode parameter "
2550 							"data\n");
2551 					goto defaults;
2552 				} else {
2553 					modepage = page_code;
2554 					goto Page_found;
2555 				}
2556 			} else {
2557 				/* Go to the next page */
2558 				if (spf && len - offset > 3)
2559 					offset += 4 + (buffer[offset+2] << 8) +
2560 						buffer[offset+3];
2561 				else if (!spf && len - offset > 1)
2562 					offset += 2 + buffer[offset+1];
2563 				else {
2564 					sd_first_printk(KERN_ERR, sdkp,
2565 							"Incomplete mode "
2566 							"parameter data\n");
2567 					goto defaults;
2568 				}
2569 			}
2570 		}
2571 
2572 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2573 		goto defaults;
2574 
2575 	Page_found:
2576 		if (modepage == 8) {
2577 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2578 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2579 		} else {
2580 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2581 			sdkp->RCD = 0;
2582 		}
2583 
2584 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2585 		if (sdp->broken_fua) {
2586 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2587 			sdkp->DPOFUA = 0;
2588 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2589 			   !sdkp->device->use_16_for_rw) {
2590 			sd_first_printk(KERN_NOTICE, sdkp,
2591 				  "Uses READ/WRITE(6), disabling FUA\n");
2592 			sdkp->DPOFUA = 0;
2593 		}
2594 
2595 		/* No cache flush allowed for write protected devices */
2596 		if (sdkp->WCE && sdkp->write_prot)
2597 			sdkp->WCE = 0;
2598 
2599 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2600 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2601 			sd_printk(KERN_NOTICE, sdkp,
2602 				  "Write cache: %s, read cache: %s, %s\n",
2603 				  sdkp->WCE ? "enabled" : "disabled",
2604 				  sdkp->RCD ? "disabled" : "enabled",
2605 				  sdkp->DPOFUA ? "supports DPO and FUA"
2606 				  : "doesn't support DPO or FUA");
2607 
2608 		return;
2609 	}
2610 
2611 bad_sense:
2612 	if (scsi_sense_valid(&sshdr) &&
2613 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2614 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2615 		/* Invalid field in CDB */
2616 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2617 	else
2618 		sd_first_printk(KERN_ERR, sdkp,
2619 				"Asking for cache data failed\n");
2620 
2621 defaults:
2622 	if (sdp->wce_default_on) {
2623 		sd_first_printk(KERN_NOTICE, sdkp,
2624 				"Assuming drive cache: write back\n");
2625 		sdkp->WCE = 1;
2626 	} else {
2627 		sd_first_printk(KERN_ERR, sdkp,
2628 				"Assuming drive cache: write through\n");
2629 		sdkp->WCE = 0;
2630 	}
2631 	sdkp->RCD = 0;
2632 	sdkp->DPOFUA = 0;
2633 }
2634 
2635 /*
2636  * The ATO bit indicates whether the DIF application tag is available
2637  * for use by the operating system.
2638  */
2639 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2640 {
2641 	int res, offset;
2642 	struct scsi_device *sdp = sdkp->device;
2643 	struct scsi_mode_data data;
2644 	struct scsi_sense_hdr sshdr;
2645 
2646 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2647 		return;
2648 
2649 	if (sdkp->protection_type == 0)
2650 		return;
2651 
2652 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2653 			      SD_MAX_RETRIES, &data, &sshdr);
2654 
2655 	if (!scsi_status_is_good(res) || !data.header_length ||
2656 	    data.length < 6) {
2657 		sd_first_printk(KERN_WARNING, sdkp,
2658 			  "getting Control mode page failed, assume no ATO\n");
2659 
2660 		if (scsi_sense_valid(&sshdr))
2661 			sd_print_sense_hdr(sdkp, &sshdr);
2662 
2663 		return;
2664 	}
2665 
2666 	offset = data.header_length + data.block_descriptor_length;
2667 
2668 	if ((buffer[offset] & 0x3f) != 0x0a) {
2669 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2670 		return;
2671 	}
2672 
2673 	if ((buffer[offset + 5] & 0x80) == 0)
2674 		return;
2675 
2676 	sdkp->ATO = 1;
2677 
2678 	return;
2679 }
2680 
2681 /**
2682  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2683  * @disk: disk to query
2684  */
2685 static void sd_read_block_limits(struct scsi_disk *sdkp)
2686 {
2687 	unsigned int sector_sz = sdkp->device->sector_size;
2688 	const int vpd_len = 64;
2689 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2690 
2691 	if (!buffer ||
2692 	    /* Block Limits VPD */
2693 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2694 		goto out;
2695 
2696 	blk_queue_io_min(sdkp->disk->queue,
2697 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2698 
2699 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2700 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2701 
2702 	if (buffer[3] == 0x3c) {
2703 		unsigned int lba_count, desc_count;
2704 
2705 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2706 
2707 		if (!sdkp->lbpme)
2708 			goto out;
2709 
2710 		lba_count = get_unaligned_be32(&buffer[20]);
2711 		desc_count = get_unaligned_be32(&buffer[24]);
2712 
2713 		if (lba_count && desc_count)
2714 			sdkp->max_unmap_blocks = lba_count;
2715 
2716 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2717 
2718 		if (buffer[32] & 0x80)
2719 			sdkp->unmap_alignment =
2720 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2721 
2722 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2723 
2724 			if (sdkp->max_unmap_blocks)
2725 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2726 			else
2727 				sd_config_discard(sdkp, SD_LBP_WS16);
2728 
2729 		} else {	/* LBP VPD page tells us what to use */
2730 			if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2731 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2732 			else if (sdkp->lbpws)
2733 				sd_config_discard(sdkp, SD_LBP_WS16);
2734 			else if (sdkp->lbpws10)
2735 				sd_config_discard(sdkp, SD_LBP_WS10);
2736 			else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2737 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2738 			else
2739 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2740 		}
2741 	}
2742 
2743  out:
2744 	kfree(buffer);
2745 }
2746 
2747 /**
2748  * sd_read_block_characteristics - Query block dev. characteristics
2749  * @disk: disk to query
2750  */
2751 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2752 {
2753 	struct request_queue *q = sdkp->disk->queue;
2754 	unsigned char *buffer;
2755 	u16 rot;
2756 	const int vpd_len = 64;
2757 
2758 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2759 
2760 	if (!buffer ||
2761 	    /* Block Device Characteristics VPD */
2762 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2763 		goto out;
2764 
2765 	rot = get_unaligned_be16(&buffer[4]);
2766 
2767 	if (rot == 1) {
2768 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2769 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2770 	}
2771 
2772 	if (sdkp->device->type == TYPE_ZBC) {
2773 		/* Host-managed */
2774 		q->limits.zoned = BLK_ZONED_HM;
2775 	} else {
2776 		sdkp->zoned = (buffer[8] >> 4) & 3;
2777 		if (sdkp->zoned == 1)
2778 			/* Host-aware */
2779 			q->limits.zoned = BLK_ZONED_HA;
2780 		else
2781 			/*
2782 			 * Treat drive-managed devices as
2783 			 * regular block devices.
2784 			 */
2785 			q->limits.zoned = BLK_ZONED_NONE;
2786 	}
2787 	if (blk_queue_is_zoned(q) && sdkp->first_scan)
2788 		sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2789 		      q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2790 
2791  out:
2792 	kfree(buffer);
2793 }
2794 
2795 /**
2796  * sd_read_block_provisioning - Query provisioning VPD page
2797  * @disk: disk to query
2798  */
2799 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2800 {
2801 	unsigned char *buffer;
2802 	const int vpd_len = 8;
2803 
2804 	if (sdkp->lbpme == 0)
2805 		return;
2806 
2807 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2808 
2809 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2810 		goto out;
2811 
2812 	sdkp->lbpvpd	= 1;
2813 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2814 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2815 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2816 
2817  out:
2818 	kfree(buffer);
2819 }
2820 
2821 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2822 {
2823 	struct scsi_device *sdev = sdkp->device;
2824 
2825 	if (sdev->host->no_write_same) {
2826 		sdev->no_write_same = 1;
2827 
2828 		return;
2829 	}
2830 
2831 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2832 		/* too large values might cause issues with arcmsr */
2833 		int vpd_buf_len = 64;
2834 
2835 		sdev->no_report_opcodes = 1;
2836 
2837 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2838 		 * CODES is unsupported and the device has an ATA
2839 		 * Information VPD page (SAT).
2840 		 */
2841 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2842 			sdev->no_write_same = 1;
2843 	}
2844 
2845 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2846 		sdkp->ws16 = 1;
2847 
2848 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2849 		sdkp->ws10 = 1;
2850 }
2851 
2852 /**
2853  *	sd_revalidate_disk - called the first time a new disk is seen,
2854  *	performs disk spin up, read_capacity, etc.
2855  *	@disk: struct gendisk we care about
2856  **/
2857 static int sd_revalidate_disk(struct gendisk *disk)
2858 {
2859 	struct scsi_disk *sdkp = scsi_disk(disk);
2860 	struct scsi_device *sdp = sdkp->device;
2861 	struct request_queue *q = sdkp->disk->queue;
2862 	sector_t old_capacity = sdkp->capacity;
2863 	unsigned char *buffer;
2864 	unsigned int dev_max, rw_max;
2865 
2866 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2867 				      "sd_revalidate_disk\n"));
2868 
2869 	/*
2870 	 * If the device is offline, don't try and read capacity or any
2871 	 * of the other niceties.
2872 	 */
2873 	if (!scsi_device_online(sdp))
2874 		goto out;
2875 
2876 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2877 	if (!buffer) {
2878 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2879 			  "allocation failure.\n");
2880 		goto out;
2881 	}
2882 
2883 	sd_spinup_disk(sdkp);
2884 
2885 	/*
2886 	 * Without media there is no reason to ask; moreover, some devices
2887 	 * react badly if we do.
2888 	 */
2889 	if (sdkp->media_present) {
2890 		sd_read_capacity(sdkp, buffer);
2891 
2892 		if (scsi_device_supports_vpd(sdp)) {
2893 			sd_read_block_provisioning(sdkp);
2894 			sd_read_block_limits(sdkp);
2895 			sd_read_block_characteristics(sdkp);
2896 			sd_zbc_read_zones(sdkp, buffer);
2897 		}
2898 
2899 		sd_print_capacity(sdkp, old_capacity);
2900 
2901 		sd_read_write_protect_flag(sdkp, buffer);
2902 		sd_read_cache_type(sdkp, buffer);
2903 		sd_read_app_tag_own(sdkp, buffer);
2904 		sd_read_write_same(sdkp, buffer);
2905 	}
2906 
2907 	sdkp->first_scan = 0;
2908 
2909 	/*
2910 	 * We now have all cache related info, determine how we deal
2911 	 * with flush requests.
2912 	 */
2913 	sd_set_flush_flag(sdkp);
2914 
2915 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2916 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2917 
2918 	/* Some devices report a maximum block count for READ/WRITE requests. */
2919 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2920 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2921 
2922 	/*
2923 	 * Use the device's preferred I/O size for reads and writes
2924 	 * unless the reported value is unreasonably small, large, or
2925 	 * garbage.
2926 	 */
2927 	if (sdkp->opt_xfer_blocks &&
2928 	    sdkp->opt_xfer_blocks <= dev_max &&
2929 	    sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2930 	    logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2931 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2932 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2933 	} else
2934 		rw_max = BLK_DEF_MAX_SECTORS;
2935 
2936 	/* Combine with controller limits */
2937 	q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2938 
2939 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2940 	sd_config_write_same(sdkp);
2941 	kfree(buffer);
2942 
2943  out:
2944 	return 0;
2945 }
2946 
2947 /**
2948  *	sd_unlock_native_capacity - unlock native capacity
2949  *	@disk: struct gendisk to set capacity for
2950  *
2951  *	Block layer calls this function if it detects that partitions
2952  *	on @disk reach beyond the end of the device.  If the SCSI host
2953  *	implements ->unlock_native_capacity() method, it's invoked to
2954  *	give it a chance to adjust the device capacity.
2955  *
2956  *	CONTEXT:
2957  *	Defined by block layer.  Might sleep.
2958  */
2959 static void sd_unlock_native_capacity(struct gendisk *disk)
2960 {
2961 	struct scsi_device *sdev = scsi_disk(disk)->device;
2962 
2963 	if (sdev->host->hostt->unlock_native_capacity)
2964 		sdev->host->hostt->unlock_native_capacity(sdev);
2965 }
2966 
2967 /**
2968  *	sd_format_disk_name - format disk name
2969  *	@prefix: name prefix - ie. "sd" for SCSI disks
2970  *	@index: index of the disk to format name for
2971  *	@buf: output buffer
2972  *	@buflen: length of the output buffer
2973  *
2974  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2975  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2976  *	which is followed by sdaaa.
2977  *
2978  *	This is basically 26 base counting with one extra 'nil' entry
2979  *	at the beginning from the second digit on and can be
2980  *	determined using similar method as 26 base conversion with the
2981  *	index shifted -1 after each digit is computed.
2982  *
2983  *	CONTEXT:
2984  *	Don't care.
2985  *
2986  *	RETURNS:
2987  *	0 on success, -errno on failure.
2988  */
2989 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2990 {
2991 	const int base = 'z' - 'a' + 1;
2992 	char *begin = buf + strlen(prefix);
2993 	char *end = buf + buflen;
2994 	char *p;
2995 	int unit;
2996 
2997 	p = end - 1;
2998 	*p = '\0';
2999 	unit = base;
3000 	do {
3001 		if (p == begin)
3002 			return -EINVAL;
3003 		*--p = 'a' + (index % unit);
3004 		index = (index / unit) - 1;
3005 	} while (index >= 0);
3006 
3007 	memmove(begin, p, end - p);
3008 	memcpy(buf, prefix, strlen(prefix));
3009 
3010 	return 0;
3011 }
3012 
3013 /*
3014  * The asynchronous part of sd_probe
3015  */
3016 static void sd_probe_async(void *data, async_cookie_t cookie)
3017 {
3018 	struct scsi_disk *sdkp = data;
3019 	struct scsi_device *sdp;
3020 	struct gendisk *gd;
3021 	u32 index;
3022 	struct device *dev;
3023 
3024 	sdp = sdkp->device;
3025 	gd = sdkp->disk;
3026 	index = sdkp->index;
3027 	dev = &sdp->sdev_gendev;
3028 
3029 	gd->major = sd_major((index & 0xf0) >> 4);
3030 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3031 	gd->minors = SD_MINORS;
3032 
3033 	gd->fops = &sd_fops;
3034 	gd->private_data = &sdkp->driver;
3035 	gd->queue = sdkp->device->request_queue;
3036 
3037 	/* defaults, until the device tells us otherwise */
3038 	sdp->sector_size = 512;
3039 	sdkp->capacity = 0;
3040 	sdkp->media_present = 1;
3041 	sdkp->write_prot = 0;
3042 	sdkp->cache_override = 0;
3043 	sdkp->WCE = 0;
3044 	sdkp->RCD = 0;
3045 	sdkp->ATO = 0;
3046 	sdkp->first_scan = 1;
3047 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3048 
3049 	sd_revalidate_disk(gd);
3050 
3051 	gd->flags = GENHD_FL_EXT_DEVT;
3052 	if (sdp->removable) {
3053 		gd->flags |= GENHD_FL_REMOVABLE;
3054 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3055 	}
3056 
3057 	blk_pm_runtime_init(sdp->request_queue, dev);
3058 	device_add_disk(dev, gd);
3059 	if (sdkp->capacity)
3060 		sd_dif_config_host(sdkp);
3061 
3062 	sd_revalidate_disk(gd);
3063 
3064 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3065 		  sdp->removable ? "removable " : "");
3066 	scsi_autopm_put_device(sdp);
3067 	put_device(&sdkp->dev);
3068 }
3069 
3070 /**
3071  *	sd_probe - called during driver initialization and whenever a
3072  *	new scsi device is attached to the system. It is called once
3073  *	for each scsi device (not just disks) present.
3074  *	@dev: pointer to device object
3075  *
3076  *	Returns 0 if successful (or not interested in this scsi device
3077  *	(e.g. scanner)); 1 when there is an error.
3078  *
3079  *	Note: this function is invoked from the scsi mid-level.
3080  *	This function sets up the mapping between a given
3081  *	<host,channel,id,lun> (found in sdp) and new device name
3082  *	(e.g. /dev/sda). More precisely it is the block device major
3083  *	and minor number that is chosen here.
3084  *
3085  *	Assume sd_probe is not re-entrant (for time being)
3086  *	Also think about sd_probe() and sd_remove() running coincidentally.
3087  **/
3088 static int sd_probe(struct device *dev)
3089 {
3090 	struct scsi_device *sdp = to_scsi_device(dev);
3091 	struct scsi_disk *sdkp;
3092 	struct gendisk *gd;
3093 	int index;
3094 	int error;
3095 
3096 	scsi_autopm_get_device(sdp);
3097 	error = -ENODEV;
3098 	if (sdp->type != TYPE_DISK &&
3099 	    sdp->type != TYPE_ZBC &&
3100 	    sdp->type != TYPE_MOD &&
3101 	    sdp->type != TYPE_RBC)
3102 		goto out;
3103 
3104 #ifndef CONFIG_BLK_DEV_ZONED
3105 	if (sdp->type == TYPE_ZBC)
3106 		goto out;
3107 #endif
3108 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3109 					"sd_probe\n"));
3110 
3111 	error = -ENOMEM;
3112 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3113 	if (!sdkp)
3114 		goto out;
3115 
3116 	gd = alloc_disk(SD_MINORS);
3117 	if (!gd)
3118 		goto out_free;
3119 
3120 	do {
3121 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3122 			goto out_put;
3123 
3124 		spin_lock(&sd_index_lock);
3125 		error = ida_get_new(&sd_index_ida, &index);
3126 		spin_unlock(&sd_index_lock);
3127 	} while (error == -EAGAIN);
3128 
3129 	if (error) {
3130 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3131 		goto out_put;
3132 	}
3133 
3134 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3135 	if (error) {
3136 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3137 		goto out_free_index;
3138 	}
3139 
3140 	sdkp->device = sdp;
3141 	sdkp->driver = &sd_template;
3142 	sdkp->disk = gd;
3143 	sdkp->index = index;
3144 	atomic_set(&sdkp->openers, 0);
3145 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3146 
3147 	if (!sdp->request_queue->rq_timeout) {
3148 		if (sdp->type != TYPE_MOD)
3149 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3150 		else
3151 			blk_queue_rq_timeout(sdp->request_queue,
3152 					     SD_MOD_TIMEOUT);
3153 	}
3154 
3155 	device_initialize(&sdkp->dev);
3156 	sdkp->dev.parent = dev;
3157 	sdkp->dev.class = &sd_disk_class;
3158 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3159 
3160 	error = device_add(&sdkp->dev);
3161 	if (error)
3162 		goto out_free_index;
3163 
3164 	get_device(dev);
3165 	dev_set_drvdata(dev, sdkp);
3166 
3167 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3168 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3169 
3170 	return 0;
3171 
3172  out_free_index:
3173 	spin_lock(&sd_index_lock);
3174 	ida_remove(&sd_index_ida, index);
3175 	spin_unlock(&sd_index_lock);
3176  out_put:
3177 	put_disk(gd);
3178  out_free:
3179 	kfree(sdkp);
3180  out:
3181 	scsi_autopm_put_device(sdp);
3182 	return error;
3183 }
3184 
3185 /**
3186  *	sd_remove - called whenever a scsi disk (previously recognized by
3187  *	sd_probe) is detached from the system. It is called (potentially
3188  *	multiple times) during sd module unload.
3189  *	@sdp: pointer to mid level scsi device object
3190  *
3191  *	Note: this function is invoked from the scsi mid-level.
3192  *	This function potentially frees up a device name (e.g. /dev/sdc)
3193  *	that could be re-used by a subsequent sd_probe().
3194  *	This function is not called when the built-in sd driver is "exit-ed".
3195  **/
3196 static int sd_remove(struct device *dev)
3197 {
3198 	struct scsi_disk *sdkp;
3199 	dev_t devt;
3200 
3201 	sdkp = dev_get_drvdata(dev);
3202 	devt = disk_devt(sdkp->disk);
3203 	scsi_autopm_get_device(sdkp->device);
3204 
3205 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3206 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3207 	device_del(&sdkp->dev);
3208 	del_gendisk(sdkp->disk);
3209 	sd_shutdown(dev);
3210 
3211 	sd_zbc_remove(sdkp);
3212 
3213 	blk_register_region(devt, SD_MINORS, NULL,
3214 			    sd_default_probe, NULL, NULL);
3215 
3216 	mutex_lock(&sd_ref_mutex);
3217 	dev_set_drvdata(dev, NULL);
3218 	put_device(&sdkp->dev);
3219 	mutex_unlock(&sd_ref_mutex);
3220 
3221 	return 0;
3222 }
3223 
3224 /**
3225  *	scsi_disk_release - Called to free the scsi_disk structure
3226  *	@dev: pointer to embedded class device
3227  *
3228  *	sd_ref_mutex must be held entering this routine.  Because it is
3229  *	called on last put, you should always use the scsi_disk_get()
3230  *	scsi_disk_put() helpers which manipulate the semaphore directly
3231  *	and never do a direct put_device.
3232  **/
3233 static void scsi_disk_release(struct device *dev)
3234 {
3235 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3236 	struct gendisk *disk = sdkp->disk;
3237 
3238 	spin_lock(&sd_index_lock);
3239 	ida_remove(&sd_index_ida, sdkp->index);
3240 	spin_unlock(&sd_index_lock);
3241 
3242 	disk->private_data = NULL;
3243 	put_disk(disk);
3244 	put_device(&sdkp->device->sdev_gendev);
3245 
3246 	kfree(sdkp);
3247 }
3248 
3249 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3250 {
3251 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3252 	struct scsi_sense_hdr sshdr;
3253 	struct scsi_device *sdp = sdkp->device;
3254 	int res;
3255 
3256 	if (start)
3257 		cmd[4] |= 1;	/* START */
3258 
3259 	if (sdp->start_stop_pwr_cond)
3260 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3261 
3262 	if (!scsi_device_online(sdp))
3263 		return -ENODEV;
3264 
3265 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3266 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, 0, RQF_PM);
3267 	if (res) {
3268 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3269 		if (driver_byte(res) & DRIVER_SENSE)
3270 			sd_print_sense_hdr(sdkp, &sshdr);
3271 		if (scsi_sense_valid(&sshdr) &&
3272 			/* 0x3a is medium not present */
3273 			sshdr.asc == 0x3a)
3274 			res = 0;
3275 	}
3276 
3277 	/* SCSI error codes must not go to the generic layer */
3278 	if (res)
3279 		return -EIO;
3280 
3281 	return 0;
3282 }
3283 
3284 /*
3285  * Send a SYNCHRONIZE CACHE instruction down to the device through
3286  * the normal SCSI command structure.  Wait for the command to
3287  * complete.
3288  */
3289 static void sd_shutdown(struct device *dev)
3290 {
3291 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3292 
3293 	if (!sdkp)
3294 		return;         /* this can happen */
3295 
3296 	if (pm_runtime_suspended(dev))
3297 		return;
3298 
3299 	if (sdkp->WCE && sdkp->media_present) {
3300 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3301 		sd_sync_cache(sdkp);
3302 	}
3303 
3304 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3305 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3306 		sd_start_stop_device(sdkp, 0);
3307 	}
3308 }
3309 
3310 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3311 {
3312 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3313 	int ret = 0;
3314 
3315 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3316 		return 0;
3317 
3318 	if (sdkp->WCE && sdkp->media_present) {
3319 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3320 		ret = sd_sync_cache(sdkp);
3321 		if (ret) {
3322 			/* ignore OFFLINE device */
3323 			if (ret == -ENODEV)
3324 				ret = 0;
3325 			goto done;
3326 		}
3327 	}
3328 
3329 	if (sdkp->device->manage_start_stop) {
3330 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3331 		/* an error is not worth aborting a system sleep */
3332 		ret = sd_start_stop_device(sdkp, 0);
3333 		if (ignore_stop_errors)
3334 			ret = 0;
3335 	}
3336 
3337 done:
3338 	return ret;
3339 }
3340 
3341 static int sd_suspend_system(struct device *dev)
3342 {
3343 	return sd_suspend_common(dev, true);
3344 }
3345 
3346 static int sd_suspend_runtime(struct device *dev)
3347 {
3348 	return sd_suspend_common(dev, false);
3349 }
3350 
3351 static int sd_resume(struct device *dev)
3352 {
3353 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3354 
3355 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3356 		return 0;
3357 
3358 	if (!sdkp->device->manage_start_stop)
3359 		return 0;
3360 
3361 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3362 	return sd_start_stop_device(sdkp, 1);
3363 }
3364 
3365 /**
3366  *	init_sd - entry point for this driver (both when built in or when
3367  *	a module).
3368  *
3369  *	Note: this function registers this driver with the scsi mid-level.
3370  **/
3371 static int __init init_sd(void)
3372 {
3373 	int majors = 0, i, err;
3374 
3375 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3376 
3377 	for (i = 0; i < SD_MAJORS; i++) {
3378 		if (register_blkdev(sd_major(i), "sd") != 0)
3379 			continue;
3380 		majors++;
3381 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3382 				    sd_default_probe, NULL, NULL);
3383 	}
3384 
3385 	if (!majors)
3386 		return -ENODEV;
3387 
3388 	err = class_register(&sd_disk_class);
3389 	if (err)
3390 		goto err_out;
3391 
3392 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3393 					 0, 0, NULL);
3394 	if (!sd_cdb_cache) {
3395 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3396 		err = -ENOMEM;
3397 		goto err_out_class;
3398 	}
3399 
3400 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3401 	if (!sd_cdb_pool) {
3402 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3403 		err = -ENOMEM;
3404 		goto err_out_cache;
3405 	}
3406 
3407 	err = scsi_register_driver(&sd_template.gendrv);
3408 	if (err)
3409 		goto err_out_driver;
3410 
3411 	return 0;
3412 
3413 err_out_driver:
3414 	mempool_destroy(sd_cdb_pool);
3415 
3416 err_out_cache:
3417 	kmem_cache_destroy(sd_cdb_cache);
3418 
3419 err_out_class:
3420 	class_unregister(&sd_disk_class);
3421 err_out:
3422 	for (i = 0; i < SD_MAJORS; i++)
3423 		unregister_blkdev(sd_major(i), "sd");
3424 	return err;
3425 }
3426 
3427 /**
3428  *	exit_sd - exit point for this driver (when it is a module).
3429  *
3430  *	Note: this function unregisters this driver from the scsi mid-level.
3431  **/
3432 static void __exit exit_sd(void)
3433 {
3434 	int i;
3435 
3436 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3437 
3438 	scsi_unregister_driver(&sd_template.gendrv);
3439 	mempool_destroy(sd_cdb_pool);
3440 	kmem_cache_destroy(sd_cdb_cache);
3441 
3442 	class_unregister(&sd_disk_class);
3443 
3444 	for (i = 0; i < SD_MAJORS; i++) {
3445 		blk_unregister_region(sd_major(i), SD_MINORS);
3446 		unregister_blkdev(sd_major(i), "sd");
3447 	}
3448 }
3449 
3450 module_init(init_sd);
3451 module_exit(exit_sd);
3452 
3453 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3454 			       struct scsi_sense_hdr *sshdr)
3455 {
3456 	scsi_print_sense_hdr(sdkp->device,
3457 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3458 }
3459 
3460 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3461 			    int result)
3462 {
3463 	const char *hb_string = scsi_hostbyte_string(result);
3464 	const char *db_string = scsi_driverbyte_string(result);
3465 
3466 	if (hb_string || db_string)
3467 		sd_printk(KERN_INFO, sdkp,
3468 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3469 			  hb_string ? hb_string : "invalid",
3470 			  db_string ? db_string : "invalid");
3471 	else
3472 		sd_printk(KERN_INFO, sdkp,
3473 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3474 			  msg, host_byte(result), driver_byte(result));
3475 }
3476 
3477