xref: /openbmc/linux/drivers/scsi/sd.c (revision 4f205687)
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 <asm/uaccess.h>
56 #include <asm/unaligned.h>
57 
58 #include <scsi/scsi.h>
59 #include <scsi/scsi_cmnd.h>
60 #include <scsi/scsi_dbg.h>
61 #include <scsi/scsi_device.h>
62 #include <scsi/scsi_driver.h>
63 #include <scsi/scsi_eh.h>
64 #include <scsi/scsi_host.h>
65 #include <scsi/scsi_ioctl.h>
66 #include <scsi/scsicam.h>
67 
68 #include "sd.h"
69 #include "scsi_priv.h"
70 #include "scsi_logging.h"
71 
72 MODULE_AUTHOR("Eric Youngdale");
73 MODULE_DESCRIPTION("SCSI disk (sd) driver");
74 MODULE_LICENSE("GPL");
75 
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
95 
96 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
97 #define SD_MINORS	16
98 #else
99 #define SD_MINORS	0
100 #endif
101 
102 static void sd_config_discard(struct scsi_disk *, unsigned int);
103 static void sd_config_write_same(struct scsi_disk *);
104 static int  sd_revalidate_disk(struct gendisk *);
105 static void sd_unlock_native_capacity(struct gendisk *disk);
106 static int  sd_probe(struct device *);
107 static int  sd_remove(struct device *);
108 static void sd_shutdown(struct device *);
109 static int sd_suspend_system(struct device *);
110 static int sd_suspend_runtime(struct device *);
111 static int sd_resume(struct device *);
112 static void sd_rescan(struct device *);
113 static int sd_init_command(struct scsi_cmnd *SCpnt);
114 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
115 static int sd_done(struct scsi_cmnd *);
116 static int sd_eh_action(struct scsi_cmnd *, int);
117 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
118 static void scsi_disk_release(struct device *cdev);
119 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
120 static void sd_print_result(const struct scsi_disk *, const char *, int);
121 
122 static DEFINE_SPINLOCK(sd_index_lock);
123 static DEFINE_IDA(sd_index_ida);
124 
125 /* This semaphore is used to mediate the 0->1 reference get in the
126  * face of object destruction (i.e. we can't allow a get on an
127  * object after last put) */
128 static DEFINE_MUTEX(sd_ref_mutex);
129 
130 static struct kmem_cache *sd_cdb_cache;
131 static mempool_t *sd_cdb_pool;
132 
133 static const char *sd_cache_types[] = {
134 	"write through", "none", "write back",
135 	"write back, no read (daft)"
136 };
137 
138 static void sd_set_flush_flag(struct scsi_disk *sdkp)
139 {
140 	bool wc = false, fua = false;
141 
142 	if (sdkp->WCE) {
143 		wc = true;
144 		if (sdkp->DPOFUA)
145 			fua = true;
146 	}
147 
148 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
149 }
150 
151 static ssize_t
152 cache_type_store(struct device *dev, struct device_attribute *attr,
153 		 const char *buf, size_t count)
154 {
155 	int i, ct = -1, rcd, wce, sp;
156 	struct scsi_disk *sdkp = to_scsi_disk(dev);
157 	struct scsi_device *sdp = sdkp->device;
158 	char buffer[64];
159 	char *buffer_data;
160 	struct scsi_mode_data data;
161 	struct scsi_sense_hdr sshdr;
162 	static const char temp[] = "temporary ";
163 	int len;
164 
165 	if (sdp->type != TYPE_DISK)
166 		/* no cache control on RBC devices; theoretically they
167 		 * can do it, but there's probably so many exceptions
168 		 * it's not worth the risk */
169 		return -EINVAL;
170 
171 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
172 		buf += sizeof(temp) - 1;
173 		sdkp->cache_override = 1;
174 	} else {
175 		sdkp->cache_override = 0;
176 	}
177 
178 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
179 		len = strlen(sd_cache_types[i]);
180 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
181 		    buf[len] == '\n') {
182 			ct = i;
183 			break;
184 		}
185 	}
186 	if (ct < 0)
187 		return -EINVAL;
188 	rcd = ct & 0x01 ? 1 : 0;
189 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 
191 	if (sdkp->cache_override) {
192 		sdkp->WCE = wce;
193 		sdkp->RCD = rcd;
194 		sd_set_flush_flag(sdkp);
195 		return count;
196 	}
197 
198 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 			    SD_MAX_RETRIES, &data, NULL))
200 		return -EINVAL;
201 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 		  data.block_descriptor_length);
203 	buffer_data = buffer + data.header_length +
204 		data.block_descriptor_length;
205 	buffer_data[2] &= ~0x05;
206 	buffer_data[2] |= wce << 2 | rcd;
207 	sp = buffer_data[0] & 0x80 ? 1 : 0;
208 	buffer_data[0] &= ~0x80;
209 
210 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
211 			     SD_MAX_RETRIES, &data, &sshdr)) {
212 		if (scsi_sense_valid(&sshdr))
213 			sd_print_sense_hdr(sdkp, &sshdr);
214 		return -EINVAL;
215 	}
216 	revalidate_disk(sdkp->disk);
217 	return count;
218 }
219 
220 static ssize_t
221 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
222 		       char *buf)
223 {
224 	struct scsi_disk *sdkp = to_scsi_disk(dev);
225 	struct scsi_device *sdp = sdkp->device;
226 
227 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
228 }
229 
230 static ssize_t
231 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
232 			const char *buf, size_t count)
233 {
234 	struct scsi_disk *sdkp = to_scsi_disk(dev);
235 	struct scsi_device *sdp = sdkp->device;
236 
237 	if (!capable(CAP_SYS_ADMIN))
238 		return -EACCES;
239 
240 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
241 
242 	return count;
243 }
244 static DEVICE_ATTR_RW(manage_start_stop);
245 
246 static ssize_t
247 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
248 {
249 	struct scsi_disk *sdkp = to_scsi_disk(dev);
250 
251 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
252 }
253 
254 static ssize_t
255 allow_restart_store(struct device *dev, struct device_attribute *attr,
256 		    const char *buf, size_t count)
257 {
258 	struct scsi_disk *sdkp = to_scsi_disk(dev);
259 	struct scsi_device *sdp = sdkp->device;
260 
261 	if (!capable(CAP_SYS_ADMIN))
262 		return -EACCES;
263 
264 	if (sdp->type != TYPE_DISK)
265 		return -EINVAL;
266 
267 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
268 
269 	return count;
270 }
271 static DEVICE_ATTR_RW(allow_restart);
272 
273 static ssize_t
274 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
275 {
276 	struct scsi_disk *sdkp = to_scsi_disk(dev);
277 	int ct = sdkp->RCD + 2*sdkp->WCE;
278 
279 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
280 }
281 static DEVICE_ATTR_RW(cache_type);
282 
283 static ssize_t
284 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
285 {
286 	struct scsi_disk *sdkp = to_scsi_disk(dev);
287 
288 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
289 }
290 static DEVICE_ATTR_RO(FUA);
291 
292 static ssize_t
293 protection_type_show(struct device *dev, struct device_attribute *attr,
294 		     char *buf)
295 {
296 	struct scsi_disk *sdkp = to_scsi_disk(dev);
297 
298 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
299 }
300 
301 static ssize_t
302 protection_type_store(struct device *dev, struct device_attribute *attr,
303 		      const char *buf, size_t count)
304 {
305 	struct scsi_disk *sdkp = to_scsi_disk(dev);
306 	unsigned int val;
307 	int err;
308 
309 	if (!capable(CAP_SYS_ADMIN))
310 		return -EACCES;
311 
312 	err = kstrtouint(buf, 10, &val);
313 
314 	if (err)
315 		return err;
316 
317 	if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
318 		sdkp->protection_type = val;
319 
320 	return count;
321 }
322 static DEVICE_ATTR_RW(protection_type);
323 
324 static ssize_t
325 protection_mode_show(struct device *dev, struct device_attribute *attr,
326 		     char *buf)
327 {
328 	struct scsi_disk *sdkp = to_scsi_disk(dev);
329 	struct scsi_device *sdp = sdkp->device;
330 	unsigned int dif, dix;
331 
332 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
333 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
334 
335 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
336 		dif = 0;
337 		dix = 1;
338 	}
339 
340 	if (!dif && !dix)
341 		return snprintf(buf, 20, "none\n");
342 
343 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
344 }
345 static DEVICE_ATTR_RO(protection_mode);
346 
347 static ssize_t
348 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
349 {
350 	struct scsi_disk *sdkp = to_scsi_disk(dev);
351 
352 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
353 }
354 static DEVICE_ATTR_RO(app_tag_own);
355 
356 static ssize_t
357 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
358 		       char *buf)
359 {
360 	struct scsi_disk *sdkp = to_scsi_disk(dev);
361 
362 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
363 }
364 static DEVICE_ATTR_RO(thin_provisioning);
365 
366 static const char *lbp_mode[] = {
367 	[SD_LBP_FULL]		= "full",
368 	[SD_LBP_UNMAP]		= "unmap",
369 	[SD_LBP_WS16]		= "writesame_16",
370 	[SD_LBP_WS10]		= "writesame_10",
371 	[SD_LBP_ZERO]		= "writesame_zero",
372 	[SD_LBP_DISABLE]	= "disabled",
373 };
374 
375 static ssize_t
376 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
377 		       char *buf)
378 {
379 	struct scsi_disk *sdkp = to_scsi_disk(dev);
380 
381 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
382 }
383 
384 static ssize_t
385 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
386 			const char *buf, size_t count)
387 {
388 	struct scsi_disk *sdkp = to_scsi_disk(dev);
389 	struct scsi_device *sdp = sdkp->device;
390 
391 	if (!capable(CAP_SYS_ADMIN))
392 		return -EACCES;
393 
394 	if (sdp->type != TYPE_DISK)
395 		return -EINVAL;
396 
397 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
398 		sd_config_discard(sdkp, SD_LBP_UNMAP);
399 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
400 		sd_config_discard(sdkp, SD_LBP_WS16);
401 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
402 		sd_config_discard(sdkp, SD_LBP_WS10);
403 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
404 		sd_config_discard(sdkp, SD_LBP_ZERO);
405 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
406 		sd_config_discard(sdkp, SD_LBP_DISABLE);
407 	else
408 		return -EINVAL;
409 
410 	return count;
411 }
412 static DEVICE_ATTR_RW(provisioning_mode);
413 
414 static ssize_t
415 max_medium_access_timeouts_show(struct device *dev,
416 				struct device_attribute *attr, char *buf)
417 {
418 	struct scsi_disk *sdkp = to_scsi_disk(dev);
419 
420 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
421 }
422 
423 static ssize_t
424 max_medium_access_timeouts_store(struct device *dev,
425 				 struct device_attribute *attr, const char *buf,
426 				 size_t count)
427 {
428 	struct scsi_disk *sdkp = to_scsi_disk(dev);
429 	int err;
430 
431 	if (!capable(CAP_SYS_ADMIN))
432 		return -EACCES;
433 
434 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
435 
436 	return err ? err : count;
437 }
438 static DEVICE_ATTR_RW(max_medium_access_timeouts);
439 
440 static ssize_t
441 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
442 			   char *buf)
443 {
444 	struct scsi_disk *sdkp = to_scsi_disk(dev);
445 
446 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
447 }
448 
449 static ssize_t
450 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
451 			    const char *buf, size_t count)
452 {
453 	struct scsi_disk *sdkp = to_scsi_disk(dev);
454 	struct scsi_device *sdp = sdkp->device;
455 	unsigned long max;
456 	int err;
457 
458 	if (!capable(CAP_SYS_ADMIN))
459 		return -EACCES;
460 
461 	if (sdp->type != TYPE_DISK)
462 		return -EINVAL;
463 
464 	err = kstrtoul(buf, 10, &max);
465 
466 	if (err)
467 		return err;
468 
469 	if (max == 0)
470 		sdp->no_write_same = 1;
471 	else if (max <= SD_MAX_WS16_BLOCKS) {
472 		sdp->no_write_same = 0;
473 		sdkp->max_ws_blocks = max;
474 	}
475 
476 	sd_config_write_same(sdkp);
477 
478 	return count;
479 }
480 static DEVICE_ATTR_RW(max_write_same_blocks);
481 
482 static struct attribute *sd_disk_attrs[] = {
483 	&dev_attr_cache_type.attr,
484 	&dev_attr_FUA.attr,
485 	&dev_attr_allow_restart.attr,
486 	&dev_attr_manage_start_stop.attr,
487 	&dev_attr_protection_type.attr,
488 	&dev_attr_protection_mode.attr,
489 	&dev_attr_app_tag_own.attr,
490 	&dev_attr_thin_provisioning.attr,
491 	&dev_attr_provisioning_mode.attr,
492 	&dev_attr_max_write_same_blocks.attr,
493 	&dev_attr_max_medium_access_timeouts.attr,
494 	NULL,
495 };
496 ATTRIBUTE_GROUPS(sd_disk);
497 
498 static struct class sd_disk_class = {
499 	.name		= "scsi_disk",
500 	.owner		= THIS_MODULE,
501 	.dev_release	= scsi_disk_release,
502 	.dev_groups	= sd_disk_groups,
503 };
504 
505 static const struct dev_pm_ops sd_pm_ops = {
506 	.suspend		= sd_suspend_system,
507 	.resume			= sd_resume,
508 	.poweroff		= sd_suspend_system,
509 	.restore		= sd_resume,
510 	.runtime_suspend	= sd_suspend_runtime,
511 	.runtime_resume		= sd_resume,
512 };
513 
514 static struct scsi_driver sd_template = {
515 	.gendrv = {
516 		.name		= "sd",
517 		.owner		= THIS_MODULE,
518 		.probe		= sd_probe,
519 		.remove		= sd_remove,
520 		.shutdown	= sd_shutdown,
521 		.pm		= &sd_pm_ops,
522 	},
523 	.rescan			= sd_rescan,
524 	.init_command		= sd_init_command,
525 	.uninit_command		= sd_uninit_command,
526 	.done			= sd_done,
527 	.eh_action		= sd_eh_action,
528 };
529 
530 /*
531  * Dummy kobj_map->probe function.
532  * The default ->probe function will call modprobe, which is
533  * pointless as this module is already loaded.
534  */
535 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
536 {
537 	return NULL;
538 }
539 
540 /*
541  * Device no to disk mapping:
542  *
543  *       major         disc2     disc  p1
544  *   |............|.............|....|....| <- dev_t
545  *    31        20 19          8 7  4 3  0
546  *
547  * Inside a major, we have 16k disks, however mapped non-
548  * contiguously. The first 16 disks are for major0, the next
549  * ones with major1, ... Disk 256 is for major0 again, disk 272
550  * for major1, ...
551  * As we stay compatible with our numbering scheme, we can reuse
552  * the well-know SCSI majors 8, 65--71, 136--143.
553  */
554 static int sd_major(int major_idx)
555 {
556 	switch (major_idx) {
557 	case 0:
558 		return SCSI_DISK0_MAJOR;
559 	case 1 ... 7:
560 		return SCSI_DISK1_MAJOR + major_idx - 1;
561 	case 8 ... 15:
562 		return SCSI_DISK8_MAJOR + major_idx - 8;
563 	default:
564 		BUG();
565 		return 0;	/* shut up gcc */
566 	}
567 }
568 
569 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
570 {
571 	struct scsi_disk *sdkp = NULL;
572 
573 	mutex_lock(&sd_ref_mutex);
574 
575 	if (disk->private_data) {
576 		sdkp = scsi_disk(disk);
577 		if (scsi_device_get(sdkp->device) == 0)
578 			get_device(&sdkp->dev);
579 		else
580 			sdkp = NULL;
581 	}
582 	mutex_unlock(&sd_ref_mutex);
583 	return sdkp;
584 }
585 
586 static void scsi_disk_put(struct scsi_disk *sdkp)
587 {
588 	struct scsi_device *sdev = sdkp->device;
589 
590 	mutex_lock(&sd_ref_mutex);
591 	put_device(&sdkp->dev);
592 	scsi_device_put(sdev);
593 	mutex_unlock(&sd_ref_mutex);
594 }
595 
596 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
597 					   unsigned int dix, unsigned int dif)
598 {
599 	struct bio *bio = scmd->request->bio;
600 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
601 	unsigned int protect = 0;
602 
603 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
604 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
605 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
606 
607 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
608 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
609 	}
610 
611 	if (dif != SD_DIF_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
612 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
613 
614 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
615 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
616 	}
617 
618 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
619 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
620 
621 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
622 			protect = 3 << 5;	/* Disable target PI checking */
623 		else
624 			protect = 1 << 5;	/* Enable target PI checking */
625 	}
626 
627 	scsi_set_prot_op(scmd, prot_op);
628 	scsi_set_prot_type(scmd, dif);
629 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
630 
631 	return protect;
632 }
633 
634 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
635 {
636 	struct request_queue *q = sdkp->disk->queue;
637 	unsigned int logical_block_size = sdkp->device->sector_size;
638 	unsigned int max_blocks = 0;
639 
640 	q->limits.discard_zeroes_data = 0;
641 
642 	/*
643 	 * When LBPRZ is reported, discard alignment and granularity
644 	 * must be fixed to the logical block size. Otherwise the block
645 	 * layer will drop misaligned portions of the request which can
646 	 * lead to data corruption. If LBPRZ is not set, we honor the
647 	 * device preference.
648 	 */
649 	if (sdkp->lbprz) {
650 		q->limits.discard_alignment = 0;
651 		q->limits.discard_granularity = logical_block_size;
652 	} else {
653 		q->limits.discard_alignment = sdkp->unmap_alignment *
654 			logical_block_size;
655 		q->limits.discard_granularity =
656 			max(sdkp->physical_block_size,
657 			    sdkp->unmap_granularity * logical_block_size);
658 	}
659 
660 	sdkp->provisioning_mode = mode;
661 
662 	switch (mode) {
663 
664 	case SD_LBP_DISABLE:
665 		blk_queue_max_discard_sectors(q, 0);
666 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
667 		return;
668 
669 	case SD_LBP_UNMAP:
670 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
671 					  (u32)SD_MAX_WS16_BLOCKS);
672 		break;
673 
674 	case SD_LBP_WS16:
675 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
676 					  (u32)SD_MAX_WS16_BLOCKS);
677 		q->limits.discard_zeroes_data = sdkp->lbprz;
678 		break;
679 
680 	case SD_LBP_WS10:
681 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
682 					  (u32)SD_MAX_WS10_BLOCKS);
683 		q->limits.discard_zeroes_data = sdkp->lbprz;
684 		break;
685 
686 	case SD_LBP_ZERO:
687 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
688 					  (u32)SD_MAX_WS10_BLOCKS);
689 		q->limits.discard_zeroes_data = 1;
690 		break;
691 	}
692 
693 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
694 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
695 }
696 
697 /**
698  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
699  * @sdp: scsi device to operate one
700  * @rq: Request to prepare
701  *
702  * Will issue either UNMAP or WRITE SAME(16) depending on preference
703  * indicated by target device.
704  **/
705 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
706 {
707 	struct request *rq = cmd->request;
708 	struct scsi_device *sdp = cmd->device;
709 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
710 	sector_t sector = blk_rq_pos(rq);
711 	unsigned int nr_sectors = blk_rq_sectors(rq);
712 	unsigned int nr_bytes = blk_rq_bytes(rq);
713 	unsigned int len;
714 	int ret;
715 	char *buf;
716 	struct page *page;
717 
718 	sector >>= ilog2(sdp->sector_size) - 9;
719 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
720 
721 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
722 	if (!page)
723 		return BLKPREP_DEFER;
724 
725 	switch (sdkp->provisioning_mode) {
726 	case SD_LBP_UNMAP:
727 		buf = page_address(page);
728 
729 		cmd->cmd_len = 10;
730 		cmd->cmnd[0] = UNMAP;
731 		cmd->cmnd[8] = 24;
732 
733 		put_unaligned_be16(6 + 16, &buf[0]);
734 		put_unaligned_be16(16, &buf[2]);
735 		put_unaligned_be64(sector, &buf[8]);
736 		put_unaligned_be32(nr_sectors, &buf[16]);
737 
738 		len = 24;
739 		break;
740 
741 	case SD_LBP_WS16:
742 		cmd->cmd_len = 16;
743 		cmd->cmnd[0] = WRITE_SAME_16;
744 		cmd->cmnd[1] = 0x8; /* UNMAP */
745 		put_unaligned_be64(sector, &cmd->cmnd[2]);
746 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
747 
748 		len = sdkp->device->sector_size;
749 		break;
750 
751 	case SD_LBP_WS10:
752 	case SD_LBP_ZERO:
753 		cmd->cmd_len = 10;
754 		cmd->cmnd[0] = WRITE_SAME;
755 		if (sdkp->provisioning_mode == SD_LBP_WS10)
756 			cmd->cmnd[1] = 0x8; /* UNMAP */
757 		put_unaligned_be32(sector, &cmd->cmnd[2]);
758 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
759 
760 		len = sdkp->device->sector_size;
761 		break;
762 
763 	default:
764 		ret = BLKPREP_INVALID;
765 		goto out;
766 	}
767 
768 	rq->completion_data = page;
769 	rq->timeout = SD_TIMEOUT;
770 
771 	cmd->transfersize = len;
772 	cmd->allowed = SD_MAX_RETRIES;
773 
774 	/*
775 	 * Initially __data_len is set to the amount of data that needs to be
776 	 * transferred to the target. This amount depends on whether WRITE SAME
777 	 * or UNMAP is being used. After the scatterlist has been mapped by
778 	 * scsi_init_io() we set __data_len to the size of the area to be
779 	 * discarded on disk. This allows us to report completion on the full
780 	 * amount of blocks described by the request.
781 	 */
782 	blk_add_request_payload(rq, page, 0, len);
783 	ret = scsi_init_io(cmd);
784 	rq->__data_len = nr_bytes;
785 
786 out:
787 	if (ret != BLKPREP_OK)
788 		__free_page(page);
789 	return ret;
790 }
791 
792 static void sd_config_write_same(struct scsi_disk *sdkp)
793 {
794 	struct request_queue *q = sdkp->disk->queue;
795 	unsigned int logical_block_size = sdkp->device->sector_size;
796 
797 	if (sdkp->device->no_write_same) {
798 		sdkp->max_ws_blocks = 0;
799 		goto out;
800 	}
801 
802 	/* Some devices can not handle block counts above 0xffff despite
803 	 * supporting WRITE SAME(16). Consequently we default to 64k
804 	 * blocks per I/O unless the device explicitly advertises a
805 	 * bigger limit.
806 	 */
807 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
808 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
809 						   (u32)SD_MAX_WS16_BLOCKS);
810 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
811 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
812 						   (u32)SD_MAX_WS10_BLOCKS);
813 	else {
814 		sdkp->device->no_write_same = 1;
815 		sdkp->max_ws_blocks = 0;
816 	}
817 
818 out:
819 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
820 					 (logical_block_size >> 9));
821 }
822 
823 /**
824  * sd_setup_write_same_cmnd - write the same data to multiple blocks
825  * @cmd: command to prepare
826  *
827  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
828  * preference indicated by target device.
829  **/
830 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
831 {
832 	struct request *rq = cmd->request;
833 	struct scsi_device *sdp = cmd->device;
834 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
835 	struct bio *bio = rq->bio;
836 	sector_t sector = blk_rq_pos(rq);
837 	unsigned int nr_sectors = blk_rq_sectors(rq);
838 	unsigned int nr_bytes = blk_rq_bytes(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 	sector >>= ilog2(sdp->sector_size) - 9;
847 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
848 
849 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
850 
851 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
852 		cmd->cmd_len = 16;
853 		cmd->cmnd[0] = WRITE_SAME_16;
854 		put_unaligned_be64(sector, &cmd->cmnd[2]);
855 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
856 	} else {
857 		cmd->cmd_len = 10;
858 		cmd->cmnd[0] = WRITE_SAME;
859 		put_unaligned_be32(sector, &cmd->cmnd[2]);
860 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
861 	}
862 
863 	cmd->transfersize = sdp->sector_size;
864 	cmd->allowed = SD_MAX_RETRIES;
865 
866 	/*
867 	 * For WRITE_SAME the data transferred in the DATA IN buffer is
868 	 * different from the amount of data actually written to the target.
869 	 *
870 	 * We set up __data_len to the amount of data transferred from the
871 	 * DATA IN buffer so that blk_rq_map_sg set up the proper S/G list
872 	 * to transfer a single sector of data first, but then reset it to
873 	 * the amount of data to be written right after so that the I/O path
874 	 * knows how much to actually write.
875 	 */
876 	rq->__data_len = sdp->sector_size;
877 	ret = scsi_init_io(cmd);
878 	rq->__data_len = nr_bytes;
879 	return ret;
880 }
881 
882 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
883 {
884 	struct request *rq = cmd->request;
885 
886 	/* flush requests don't perform I/O, zero the S/G table */
887 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
888 
889 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
890 	cmd->cmd_len = 10;
891 	cmd->transfersize = 0;
892 	cmd->allowed = SD_MAX_RETRIES;
893 
894 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
895 	return BLKPREP_OK;
896 }
897 
898 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
899 {
900 	struct request *rq = SCpnt->request;
901 	struct scsi_device *sdp = SCpnt->device;
902 	struct gendisk *disk = rq->rq_disk;
903 	struct scsi_disk *sdkp;
904 	sector_t block = blk_rq_pos(rq);
905 	sector_t threshold;
906 	unsigned int this_count = blk_rq_sectors(rq);
907 	unsigned int dif, dix;
908 	int ret;
909 	unsigned char protect;
910 
911 	ret = scsi_init_io(SCpnt);
912 	if (ret != BLKPREP_OK)
913 		goto out;
914 	SCpnt = rq->special;
915 	sdkp = scsi_disk(disk);
916 
917 	/* from here on until we're complete, any goto out
918 	 * is used for a killable error condition */
919 	ret = BLKPREP_KILL;
920 
921 	SCSI_LOG_HLQUEUE(1,
922 		scmd_printk(KERN_INFO, SCpnt,
923 			"%s: block=%llu, count=%d\n",
924 			__func__, (unsigned long long)block, this_count));
925 
926 	if (!sdp || !scsi_device_online(sdp) ||
927 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
928 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
929 						"Finishing %u sectors\n",
930 						blk_rq_sectors(rq)));
931 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
932 						"Retry with 0x%p\n", SCpnt));
933 		goto out;
934 	}
935 
936 	if (sdp->changed) {
937 		/*
938 		 * quietly refuse to do anything to a changed disc until
939 		 * the changed bit has been reset
940 		 */
941 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
942 		goto out;
943 	}
944 
945 	/*
946 	 * Some SD card readers can't handle multi-sector accesses which touch
947 	 * the last one or two hardware sectors.  Split accesses as needed.
948 	 */
949 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
950 		(sdp->sector_size / 512);
951 
952 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
953 		if (block < threshold) {
954 			/* Access up to the threshold but not beyond */
955 			this_count = threshold - block;
956 		} else {
957 			/* Access only a single hardware sector */
958 			this_count = sdp->sector_size / 512;
959 		}
960 	}
961 
962 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
963 					(unsigned long long)block));
964 
965 	/*
966 	 * If we have a 1K hardware sectorsize, prevent access to single
967 	 * 512 byte sectors.  In theory we could handle this - in fact
968 	 * the scsi cdrom driver must be able to handle this because
969 	 * we typically use 1K blocksizes, and cdroms typically have
970 	 * 2K hardware sectorsizes.  Of course, things are simpler
971 	 * with the cdrom, since it is read-only.  For performance
972 	 * reasons, the filesystems should be able to handle this
973 	 * and not force the scsi disk driver to use bounce buffers
974 	 * for this.
975 	 */
976 	if (sdp->sector_size == 1024) {
977 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
978 			scmd_printk(KERN_ERR, SCpnt,
979 				    "Bad block number requested\n");
980 			goto out;
981 		} else {
982 			block = block >> 1;
983 			this_count = this_count >> 1;
984 		}
985 	}
986 	if (sdp->sector_size == 2048) {
987 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
988 			scmd_printk(KERN_ERR, SCpnt,
989 				    "Bad block number requested\n");
990 			goto out;
991 		} else {
992 			block = block >> 2;
993 			this_count = this_count >> 2;
994 		}
995 	}
996 	if (sdp->sector_size == 4096) {
997 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
998 			scmd_printk(KERN_ERR, SCpnt,
999 				    "Bad block number requested\n");
1000 			goto out;
1001 		} else {
1002 			block = block >> 3;
1003 			this_count = this_count >> 3;
1004 		}
1005 	}
1006 	if (rq_data_dir(rq) == WRITE) {
1007 		SCpnt->cmnd[0] = WRITE_6;
1008 
1009 		if (blk_integrity_rq(rq))
1010 			sd_dif_prepare(SCpnt);
1011 
1012 	} else if (rq_data_dir(rq) == READ) {
1013 		SCpnt->cmnd[0] = READ_6;
1014 	} else {
1015 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %llx\n", (unsigned long long) rq->cmd_flags);
1016 		goto out;
1017 	}
1018 
1019 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1020 					"%s %d/%u 512 byte blocks.\n",
1021 					(rq_data_dir(rq) == WRITE) ?
1022 					"writing" : "reading", this_count,
1023 					blk_rq_sectors(rq)));
1024 
1025 	dix = scsi_prot_sg_count(SCpnt);
1026 	dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1027 
1028 	if (dif || dix)
1029 		protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1030 	else
1031 		protect = 0;
1032 
1033 	if (protect && sdkp->protection_type == SD_DIF_TYPE2_PROTECTION) {
1034 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1035 
1036 		if (unlikely(SCpnt->cmnd == NULL)) {
1037 			ret = BLKPREP_DEFER;
1038 			goto out;
1039 		}
1040 
1041 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1042 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1043 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1044 		SCpnt->cmnd[7] = 0x18;
1045 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1046 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1047 
1048 		/* LBA */
1049 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1050 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1051 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1052 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1053 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1054 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1055 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1056 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1057 
1058 		/* Expected Indirect LBA */
1059 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1060 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1061 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1062 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1063 
1064 		/* Transfer length */
1065 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1066 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1067 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1068 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1069 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1070 		SCpnt->cmnd[0] += READ_16 - READ_6;
1071 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1072 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1073 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1074 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1075 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1076 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1077 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1078 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1079 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1080 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1081 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1082 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1083 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1084 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1085 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1086 		   scsi_device_protection(SCpnt->device) ||
1087 		   SCpnt->device->use_10_for_rw) {
1088 		SCpnt->cmnd[0] += READ_10 - READ_6;
1089 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1090 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1091 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1092 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1093 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1094 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1095 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1096 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1097 	} else {
1098 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1099 			/*
1100 			 * This happens only if this drive failed
1101 			 * 10byte rw command with ILLEGAL_REQUEST
1102 			 * during operation and thus turned off
1103 			 * use_10_for_rw.
1104 			 */
1105 			scmd_printk(KERN_ERR, SCpnt,
1106 				    "FUA write on READ/WRITE(6) drive\n");
1107 			goto out;
1108 		}
1109 
1110 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1111 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1112 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1113 		SCpnt->cmnd[4] = (unsigned char) this_count;
1114 		SCpnt->cmnd[5] = 0;
1115 	}
1116 	SCpnt->sdb.length = this_count * sdp->sector_size;
1117 
1118 	/*
1119 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1120 	 * host adapter, it's safe to assume that we can at least transfer
1121 	 * this many bytes between each connect / disconnect.
1122 	 */
1123 	SCpnt->transfersize = sdp->sector_size;
1124 	SCpnt->underflow = this_count << 9;
1125 	SCpnt->allowed = SD_MAX_RETRIES;
1126 
1127 	/*
1128 	 * This indicates that the command is ready from our end to be
1129 	 * queued.
1130 	 */
1131 	ret = BLKPREP_OK;
1132  out:
1133 	return ret;
1134 }
1135 
1136 static int sd_init_command(struct scsi_cmnd *cmd)
1137 {
1138 	struct request *rq = cmd->request;
1139 
1140 	if (rq->cmd_flags & REQ_DISCARD)
1141 		return sd_setup_discard_cmnd(cmd);
1142 	else if (rq->cmd_flags & REQ_WRITE_SAME)
1143 		return sd_setup_write_same_cmnd(cmd);
1144 	else if (rq->cmd_flags & REQ_FLUSH)
1145 		return sd_setup_flush_cmnd(cmd);
1146 	else
1147 		return sd_setup_read_write_cmnd(cmd);
1148 }
1149 
1150 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1151 {
1152 	struct request *rq = SCpnt->request;
1153 
1154 	if (rq->cmd_flags & REQ_DISCARD)
1155 		__free_page(rq->completion_data);
1156 
1157 	if (SCpnt->cmnd != rq->cmd) {
1158 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1159 		SCpnt->cmnd = NULL;
1160 		SCpnt->cmd_len = 0;
1161 	}
1162 }
1163 
1164 /**
1165  *	sd_open - open a scsi disk device
1166  *	@inode: only i_rdev member may be used
1167  *	@filp: only f_mode and f_flags may be used
1168  *
1169  *	Returns 0 if successful. Returns a negated errno value in case
1170  *	of error.
1171  *
1172  *	Note: This can be called from a user context (e.g. fsck(1) )
1173  *	or from within the kernel (e.g. as a result of a mount(1) ).
1174  *	In the latter case @inode and @filp carry an abridged amount
1175  *	of information as noted above.
1176  *
1177  *	Locking: called with bdev->bd_mutex held.
1178  **/
1179 static int sd_open(struct block_device *bdev, fmode_t mode)
1180 {
1181 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1182 	struct scsi_device *sdev;
1183 	int retval;
1184 
1185 	if (!sdkp)
1186 		return -ENXIO;
1187 
1188 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1189 
1190 	sdev = sdkp->device;
1191 
1192 	/*
1193 	 * If the device is in error recovery, wait until it is done.
1194 	 * If the device is offline, then disallow any access to it.
1195 	 */
1196 	retval = -ENXIO;
1197 	if (!scsi_block_when_processing_errors(sdev))
1198 		goto error_out;
1199 
1200 	if (sdev->removable || sdkp->write_prot)
1201 		check_disk_change(bdev);
1202 
1203 	/*
1204 	 * If the drive is empty, just let the open fail.
1205 	 */
1206 	retval = -ENOMEDIUM;
1207 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1208 		goto error_out;
1209 
1210 	/*
1211 	 * If the device has the write protect tab set, have the open fail
1212 	 * if the user expects to be able to write to the thing.
1213 	 */
1214 	retval = -EROFS;
1215 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1216 		goto error_out;
1217 
1218 	/*
1219 	 * It is possible that the disk changing stuff resulted in
1220 	 * the device being taken offline.  If this is the case,
1221 	 * report this to the user, and don't pretend that the
1222 	 * open actually succeeded.
1223 	 */
1224 	retval = -ENXIO;
1225 	if (!scsi_device_online(sdev))
1226 		goto error_out;
1227 
1228 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1229 		if (scsi_block_when_processing_errors(sdev))
1230 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1231 	}
1232 
1233 	return 0;
1234 
1235 error_out:
1236 	scsi_disk_put(sdkp);
1237 	return retval;
1238 }
1239 
1240 /**
1241  *	sd_release - invoked when the (last) close(2) is called on this
1242  *	scsi disk.
1243  *	@inode: only i_rdev member may be used
1244  *	@filp: only f_mode and f_flags may be used
1245  *
1246  *	Returns 0.
1247  *
1248  *	Note: may block (uninterruptible) if error recovery is underway
1249  *	on this disk.
1250  *
1251  *	Locking: called with bdev->bd_mutex held.
1252  **/
1253 static void sd_release(struct gendisk *disk, fmode_t mode)
1254 {
1255 	struct scsi_disk *sdkp = scsi_disk(disk);
1256 	struct scsi_device *sdev = sdkp->device;
1257 
1258 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1259 
1260 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1261 		if (scsi_block_when_processing_errors(sdev))
1262 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1263 	}
1264 
1265 	/*
1266 	 * XXX and what if there are packets in flight and this close()
1267 	 * XXX is followed by a "rmmod sd_mod"?
1268 	 */
1269 
1270 	scsi_disk_put(sdkp);
1271 }
1272 
1273 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1274 {
1275 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1276 	struct scsi_device *sdp = sdkp->device;
1277 	struct Scsi_Host *host = sdp->host;
1278 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1279 	int diskinfo[4];
1280 
1281 	/* default to most commonly used values */
1282 	diskinfo[0] = 0x40;	/* 1 << 6 */
1283 	diskinfo[1] = 0x20;	/* 1 << 5 */
1284 	diskinfo[2] = capacity >> 11;
1285 
1286 	/* override with calculated, extended default, or driver values */
1287 	if (host->hostt->bios_param)
1288 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1289 	else
1290 		scsicam_bios_param(bdev, capacity, diskinfo);
1291 
1292 	geo->heads = diskinfo[0];
1293 	geo->sectors = diskinfo[1];
1294 	geo->cylinders = diskinfo[2];
1295 	return 0;
1296 }
1297 
1298 /**
1299  *	sd_ioctl - process an ioctl
1300  *	@inode: only i_rdev/i_bdev members may be used
1301  *	@filp: only f_mode and f_flags may be used
1302  *	@cmd: ioctl command number
1303  *	@arg: this is third argument given to ioctl(2) system call.
1304  *	Often contains a pointer.
1305  *
1306  *	Returns 0 if successful (some ioctls return positive numbers on
1307  *	success as well). Returns a negated errno value in case of error.
1308  *
1309  *	Note: most ioctls are forward onto the block subsystem or further
1310  *	down in the scsi subsystem.
1311  **/
1312 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1313 		    unsigned int cmd, unsigned long arg)
1314 {
1315 	struct gendisk *disk = bdev->bd_disk;
1316 	struct scsi_disk *sdkp = scsi_disk(disk);
1317 	struct scsi_device *sdp = sdkp->device;
1318 	void __user *p = (void __user *)arg;
1319 	int error;
1320 
1321 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1322 				    "cmd=0x%x\n", disk->disk_name, cmd));
1323 
1324 	error = scsi_verify_blk_ioctl(bdev, cmd);
1325 	if (error < 0)
1326 		return error;
1327 
1328 	/*
1329 	 * If we are in the middle of error recovery, don't let anyone
1330 	 * else try and use this device.  Also, if error recovery fails, it
1331 	 * may try and take the device offline, in which case all further
1332 	 * access to the device is prohibited.
1333 	 */
1334 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1335 			(mode & FMODE_NDELAY) != 0);
1336 	if (error)
1337 		goto out;
1338 
1339 	/*
1340 	 * Send SCSI addressing ioctls directly to mid level, send other
1341 	 * ioctls to block level and then onto mid level if they can't be
1342 	 * resolved.
1343 	 */
1344 	switch (cmd) {
1345 		case SCSI_IOCTL_GET_IDLUN:
1346 		case SCSI_IOCTL_GET_BUS_NUMBER:
1347 			error = scsi_ioctl(sdp, cmd, p);
1348 			break;
1349 		default:
1350 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1351 			if (error != -ENOTTY)
1352 				break;
1353 			error = scsi_ioctl(sdp, cmd, p);
1354 			break;
1355 	}
1356 out:
1357 	return error;
1358 }
1359 
1360 static void set_media_not_present(struct scsi_disk *sdkp)
1361 {
1362 	if (sdkp->media_present)
1363 		sdkp->device->changed = 1;
1364 
1365 	if (sdkp->device->removable) {
1366 		sdkp->media_present = 0;
1367 		sdkp->capacity = 0;
1368 	}
1369 }
1370 
1371 static int media_not_present(struct scsi_disk *sdkp,
1372 			     struct scsi_sense_hdr *sshdr)
1373 {
1374 	if (!scsi_sense_valid(sshdr))
1375 		return 0;
1376 
1377 	/* not invoked for commands that could return deferred errors */
1378 	switch (sshdr->sense_key) {
1379 	case UNIT_ATTENTION:
1380 	case NOT_READY:
1381 		/* medium not present */
1382 		if (sshdr->asc == 0x3A) {
1383 			set_media_not_present(sdkp);
1384 			return 1;
1385 		}
1386 	}
1387 	return 0;
1388 }
1389 
1390 /**
1391  *	sd_check_events - check media events
1392  *	@disk: kernel device descriptor
1393  *	@clearing: disk events currently being cleared
1394  *
1395  *	Returns mask of DISK_EVENT_*.
1396  *
1397  *	Note: this function is invoked from the block subsystem.
1398  **/
1399 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1400 {
1401 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1402 	struct scsi_device *sdp;
1403 	struct scsi_sense_hdr *sshdr = NULL;
1404 	int retval;
1405 
1406 	if (!sdkp)
1407 		return 0;
1408 
1409 	sdp = sdkp->device;
1410 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1411 
1412 	/*
1413 	 * If the device is offline, don't send any commands - just pretend as
1414 	 * if the command failed.  If the device ever comes back online, we
1415 	 * can deal with it then.  It is only because of unrecoverable errors
1416 	 * that we would ever take a device offline in the first place.
1417 	 */
1418 	if (!scsi_device_online(sdp)) {
1419 		set_media_not_present(sdkp);
1420 		goto out;
1421 	}
1422 
1423 	/*
1424 	 * Using TEST_UNIT_READY enables differentiation between drive with
1425 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1426 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1427 	 *
1428 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1429 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1430 	 * sd_revalidate() is called.
1431 	 */
1432 	retval = -ENODEV;
1433 
1434 	if (scsi_block_when_processing_errors(sdp)) {
1435 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1436 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1437 					      sshdr);
1438 	}
1439 
1440 	/* failed to execute TUR, assume media not present */
1441 	if (host_byte(retval)) {
1442 		set_media_not_present(sdkp);
1443 		goto out;
1444 	}
1445 
1446 	if (media_not_present(sdkp, sshdr))
1447 		goto out;
1448 
1449 	/*
1450 	 * For removable scsi disk we have to recognise the presence
1451 	 * of a disk in the drive.
1452 	 */
1453 	if (!sdkp->media_present)
1454 		sdp->changed = 1;
1455 	sdkp->media_present = 1;
1456 out:
1457 	/*
1458 	 * sdp->changed is set under the following conditions:
1459 	 *
1460 	 *	Medium present state has changed in either direction.
1461 	 *	Device has indicated UNIT_ATTENTION.
1462 	 */
1463 	kfree(sshdr);
1464 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1465 	sdp->changed = 0;
1466 	scsi_disk_put(sdkp);
1467 	return retval;
1468 }
1469 
1470 static int sd_sync_cache(struct scsi_disk *sdkp)
1471 {
1472 	int retries, res;
1473 	struct scsi_device *sdp = sdkp->device;
1474 	const int timeout = sdp->request_queue->rq_timeout
1475 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1476 	struct scsi_sense_hdr sshdr;
1477 
1478 	if (!scsi_device_online(sdp))
1479 		return -ENODEV;
1480 
1481 	for (retries = 3; retries > 0; --retries) {
1482 		unsigned char cmd[10] = { 0 };
1483 
1484 		cmd[0] = SYNCHRONIZE_CACHE;
1485 		/*
1486 		 * Leave the rest of the command zero to indicate
1487 		 * flush everything.
1488 		 */
1489 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1490 					     &sshdr, timeout, SD_MAX_RETRIES,
1491 					     NULL, REQ_PM);
1492 		if (res == 0)
1493 			break;
1494 	}
1495 
1496 	if (res) {
1497 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1498 
1499 		if (driver_byte(res) & DRIVER_SENSE)
1500 			sd_print_sense_hdr(sdkp, &sshdr);
1501 		/* we need to evaluate the error return  */
1502 		if (scsi_sense_valid(&sshdr) &&
1503 			(sshdr.asc == 0x3a ||	/* medium not present */
1504 			 sshdr.asc == 0x20))	/* invalid command */
1505 				/* this is no error here */
1506 				return 0;
1507 
1508 		switch (host_byte(res)) {
1509 		/* ignore errors due to racing a disconnection */
1510 		case DID_BAD_TARGET:
1511 		case DID_NO_CONNECT:
1512 			return 0;
1513 		/* signal the upper layer it might try again */
1514 		case DID_BUS_BUSY:
1515 		case DID_IMM_RETRY:
1516 		case DID_REQUEUE:
1517 		case DID_SOFT_ERROR:
1518 			return -EBUSY;
1519 		default:
1520 			return -EIO;
1521 		}
1522 	}
1523 	return 0;
1524 }
1525 
1526 static void sd_rescan(struct device *dev)
1527 {
1528 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1529 
1530 	revalidate_disk(sdkp->disk);
1531 }
1532 
1533 
1534 #ifdef CONFIG_COMPAT
1535 /*
1536  * This gets directly called from VFS. When the ioctl
1537  * is not recognized we go back to the other translation paths.
1538  */
1539 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1540 			   unsigned int cmd, unsigned long arg)
1541 {
1542 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1543 	int error;
1544 
1545 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1546 			(mode & FMODE_NDELAY) != 0);
1547 	if (error)
1548 		return error;
1549 
1550 	/*
1551 	 * Let the static ioctl translation table take care of it.
1552 	 */
1553 	if (!sdev->host->hostt->compat_ioctl)
1554 		return -ENOIOCTLCMD;
1555 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1556 }
1557 #endif
1558 
1559 static char sd_pr_type(enum pr_type type)
1560 {
1561 	switch (type) {
1562 	case PR_WRITE_EXCLUSIVE:
1563 		return 0x01;
1564 	case PR_EXCLUSIVE_ACCESS:
1565 		return 0x03;
1566 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1567 		return 0x05;
1568 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1569 		return 0x06;
1570 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1571 		return 0x07;
1572 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1573 		return 0x08;
1574 	default:
1575 		return 0;
1576 	}
1577 };
1578 
1579 static int sd_pr_command(struct block_device *bdev, u8 sa,
1580 		u64 key, u64 sa_key, u8 type, u8 flags)
1581 {
1582 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1583 	struct scsi_sense_hdr sshdr;
1584 	int result;
1585 	u8 cmd[16] = { 0, };
1586 	u8 data[24] = { 0, };
1587 
1588 	cmd[0] = PERSISTENT_RESERVE_OUT;
1589 	cmd[1] = sa;
1590 	cmd[2] = type;
1591 	put_unaligned_be32(sizeof(data), &cmd[5]);
1592 
1593 	put_unaligned_be64(key, &data[0]);
1594 	put_unaligned_be64(sa_key, &data[8]);
1595 	data[20] = flags;
1596 
1597 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1598 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1599 
1600 	if ((driver_byte(result) & DRIVER_SENSE) &&
1601 	    (scsi_sense_valid(&sshdr))) {
1602 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1603 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1604 	}
1605 
1606 	return result;
1607 }
1608 
1609 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1610 		u32 flags)
1611 {
1612 	if (flags & ~PR_FL_IGNORE_KEY)
1613 		return -EOPNOTSUPP;
1614 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1615 			old_key, new_key, 0,
1616 			(1 << 0) /* APTPL */ |
1617 			(1 << 2) /* ALL_TG_PT */);
1618 }
1619 
1620 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1621 		u32 flags)
1622 {
1623 	if (flags)
1624 		return -EOPNOTSUPP;
1625 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1626 }
1627 
1628 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1629 {
1630 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1631 }
1632 
1633 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1634 		enum pr_type type, bool abort)
1635 {
1636 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1637 			     sd_pr_type(type), 0);
1638 }
1639 
1640 static int sd_pr_clear(struct block_device *bdev, u64 key)
1641 {
1642 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1643 }
1644 
1645 static const struct pr_ops sd_pr_ops = {
1646 	.pr_register	= sd_pr_register,
1647 	.pr_reserve	= sd_pr_reserve,
1648 	.pr_release	= sd_pr_release,
1649 	.pr_preempt	= sd_pr_preempt,
1650 	.pr_clear	= sd_pr_clear,
1651 };
1652 
1653 static const struct block_device_operations sd_fops = {
1654 	.owner			= THIS_MODULE,
1655 	.open			= sd_open,
1656 	.release		= sd_release,
1657 	.ioctl			= sd_ioctl,
1658 	.getgeo			= sd_getgeo,
1659 #ifdef CONFIG_COMPAT
1660 	.compat_ioctl		= sd_compat_ioctl,
1661 #endif
1662 	.check_events		= sd_check_events,
1663 	.revalidate_disk	= sd_revalidate_disk,
1664 	.unlock_native_capacity	= sd_unlock_native_capacity,
1665 	.pr_ops			= &sd_pr_ops,
1666 };
1667 
1668 /**
1669  *	sd_eh_action - error handling callback
1670  *	@scmd:		sd-issued command that has failed
1671  *	@eh_disp:	The recovery disposition suggested by the midlayer
1672  *
1673  *	This function is called by the SCSI midlayer upon completion of an
1674  *	error test command (currently TEST UNIT READY). The result of sending
1675  *	the eh command is passed in eh_disp.  We're looking for devices that
1676  *	fail medium access commands but are OK with non access commands like
1677  *	test unit ready (so wrongly see the device as having a successful
1678  *	recovery)
1679  **/
1680 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1681 {
1682 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1683 
1684 	if (!scsi_device_online(scmd->device) ||
1685 	    !scsi_medium_access_command(scmd) ||
1686 	    host_byte(scmd->result) != DID_TIME_OUT ||
1687 	    eh_disp != SUCCESS)
1688 		return eh_disp;
1689 
1690 	/*
1691 	 * The device has timed out executing a medium access command.
1692 	 * However, the TEST UNIT READY command sent during error
1693 	 * handling completed successfully. Either the device is in the
1694 	 * process of recovering or has it suffered an internal failure
1695 	 * that prevents access to the storage medium.
1696 	 */
1697 	sdkp->medium_access_timed_out++;
1698 
1699 	/*
1700 	 * If the device keeps failing read/write commands but TEST UNIT
1701 	 * READY always completes successfully we assume that medium
1702 	 * access is no longer possible and take the device offline.
1703 	 */
1704 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1705 		scmd_printk(KERN_ERR, scmd,
1706 			    "Medium access timeout failure. Offlining disk!\n");
1707 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1708 
1709 		return FAILED;
1710 	}
1711 
1712 	return eh_disp;
1713 }
1714 
1715 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1716 {
1717 	u64 start_lba = blk_rq_pos(scmd->request);
1718 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1719 	u64 factor = scmd->device->sector_size / 512;
1720 	u64 bad_lba;
1721 	int info_valid;
1722 	/*
1723 	 * resid is optional but mostly filled in.  When it's unused,
1724 	 * its value is zero, so we assume the whole buffer transferred
1725 	 */
1726 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1727 	unsigned int good_bytes;
1728 
1729 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1730 		return 0;
1731 
1732 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1733 					     SCSI_SENSE_BUFFERSIZE,
1734 					     &bad_lba);
1735 	if (!info_valid)
1736 		return 0;
1737 
1738 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1739 		return 0;
1740 
1741 	/* be careful ... don't want any overflows */
1742 	do_div(start_lba, factor);
1743 	do_div(end_lba, factor);
1744 
1745 	/* The bad lba was reported incorrectly, we have no idea where
1746 	 * the error is.
1747 	 */
1748 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1749 		return 0;
1750 
1751 	/* This computation should always be done in terms of
1752 	 * the resolution of the device's medium.
1753 	 */
1754 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1755 	return min(good_bytes, transferred);
1756 }
1757 
1758 /**
1759  *	sd_done - bottom half handler: called when the lower level
1760  *	driver has completed (successfully or otherwise) a scsi command.
1761  *	@SCpnt: mid-level's per command structure.
1762  *
1763  *	Note: potentially run from within an ISR. Must not block.
1764  **/
1765 static int sd_done(struct scsi_cmnd *SCpnt)
1766 {
1767 	int result = SCpnt->result;
1768 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1769 	struct scsi_sense_hdr sshdr;
1770 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1771 	struct request *req = SCpnt->request;
1772 	int sense_valid = 0;
1773 	int sense_deferred = 0;
1774 	unsigned char op = SCpnt->cmnd[0];
1775 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1776 
1777 	if (req->cmd_flags & REQ_DISCARD || req->cmd_flags & REQ_WRITE_SAME) {
1778 		if (!result) {
1779 			good_bytes = blk_rq_bytes(req);
1780 			scsi_set_resid(SCpnt, 0);
1781 		} else {
1782 			good_bytes = 0;
1783 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1784 		}
1785 	}
1786 
1787 	if (result) {
1788 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1789 		if (sense_valid)
1790 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1791 	}
1792 	sdkp->medium_access_timed_out = 0;
1793 
1794 	if (driver_byte(result) != DRIVER_SENSE &&
1795 	    (!sense_valid || sense_deferred))
1796 		goto out;
1797 
1798 	switch (sshdr.sense_key) {
1799 	case HARDWARE_ERROR:
1800 	case MEDIUM_ERROR:
1801 		good_bytes = sd_completed_bytes(SCpnt);
1802 		break;
1803 	case RECOVERED_ERROR:
1804 		good_bytes = scsi_bufflen(SCpnt);
1805 		break;
1806 	case NO_SENSE:
1807 		/* This indicates a false check condition, so ignore it.  An
1808 		 * unknown amount of data was transferred so treat it as an
1809 		 * error.
1810 		 */
1811 		SCpnt->result = 0;
1812 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1813 		break;
1814 	case ABORTED_COMMAND:
1815 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1816 			good_bytes = sd_completed_bytes(SCpnt);
1817 		break;
1818 	case ILLEGAL_REQUEST:
1819 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1820 			good_bytes = sd_completed_bytes(SCpnt);
1821 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1822 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1823 			switch (op) {
1824 			case UNMAP:
1825 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1826 				break;
1827 			case WRITE_SAME_16:
1828 			case WRITE_SAME:
1829 				if (unmap)
1830 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1831 				else {
1832 					sdkp->device->no_write_same = 1;
1833 					sd_config_write_same(sdkp);
1834 
1835 					good_bytes = 0;
1836 					req->__data_len = blk_rq_bytes(req);
1837 					req->cmd_flags |= REQ_QUIET;
1838 				}
1839 			}
1840 		}
1841 		break;
1842 	default:
1843 		break;
1844 	}
1845  out:
1846 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1847 					   "sd_done: completed %d of %d bytes\n",
1848 					   good_bytes, scsi_bufflen(SCpnt)));
1849 
1850 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1851 		sd_dif_complete(SCpnt, good_bytes);
1852 
1853 	return good_bytes;
1854 }
1855 
1856 /*
1857  * spinup disk - called only in sd_revalidate_disk()
1858  */
1859 static void
1860 sd_spinup_disk(struct scsi_disk *sdkp)
1861 {
1862 	unsigned char cmd[10];
1863 	unsigned long spintime_expire = 0;
1864 	int retries, spintime;
1865 	unsigned int the_result;
1866 	struct scsi_sense_hdr sshdr;
1867 	int sense_valid = 0;
1868 
1869 	spintime = 0;
1870 
1871 	/* Spin up drives, as required.  Only do this at boot time */
1872 	/* Spinup needs to be done for module loads too. */
1873 	do {
1874 		retries = 0;
1875 
1876 		do {
1877 			cmd[0] = TEST_UNIT_READY;
1878 			memset((void *) &cmd[1], 0, 9);
1879 
1880 			the_result = scsi_execute_req(sdkp->device, cmd,
1881 						      DMA_NONE, NULL, 0,
1882 						      &sshdr, SD_TIMEOUT,
1883 						      SD_MAX_RETRIES, NULL);
1884 
1885 			/*
1886 			 * If the drive has indicated to us that it
1887 			 * doesn't have any media in it, don't bother
1888 			 * with any more polling.
1889 			 */
1890 			if (media_not_present(sdkp, &sshdr))
1891 				return;
1892 
1893 			if (the_result)
1894 				sense_valid = scsi_sense_valid(&sshdr);
1895 			retries++;
1896 		} while (retries < 3 &&
1897 			 (!scsi_status_is_good(the_result) ||
1898 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1899 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1900 
1901 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1902 			/* no sense, TUR either succeeded or failed
1903 			 * with a status error */
1904 			if(!spintime && !scsi_status_is_good(the_result)) {
1905 				sd_print_result(sdkp, "Test Unit Ready failed",
1906 						the_result);
1907 			}
1908 			break;
1909 		}
1910 
1911 		/*
1912 		 * The device does not want the automatic start to be issued.
1913 		 */
1914 		if (sdkp->device->no_start_on_add)
1915 			break;
1916 
1917 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1918 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1919 				break;	/* manual intervention required */
1920 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1921 				break;	/* standby */
1922 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1923 				break;	/* unavailable */
1924 			/*
1925 			 * Issue command to spin up drive when not ready
1926 			 */
1927 			if (!spintime) {
1928 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1929 				cmd[0] = START_STOP;
1930 				cmd[1] = 1;	/* Return immediately */
1931 				memset((void *) &cmd[2], 0, 8);
1932 				cmd[4] = 1;	/* Start spin cycle */
1933 				if (sdkp->device->start_stop_pwr_cond)
1934 					cmd[4] |= 1 << 4;
1935 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1936 						 NULL, 0, &sshdr,
1937 						 SD_TIMEOUT, SD_MAX_RETRIES,
1938 						 NULL);
1939 				spintime_expire = jiffies + 100 * HZ;
1940 				spintime = 1;
1941 			}
1942 			/* Wait 1 second for next try */
1943 			msleep(1000);
1944 			printk(".");
1945 
1946 		/*
1947 		 * Wait for USB flash devices with slow firmware.
1948 		 * Yes, this sense key/ASC combination shouldn't
1949 		 * occur here.  It's characteristic of these devices.
1950 		 */
1951 		} else if (sense_valid &&
1952 				sshdr.sense_key == UNIT_ATTENTION &&
1953 				sshdr.asc == 0x28) {
1954 			if (!spintime) {
1955 				spintime_expire = jiffies + 5 * HZ;
1956 				spintime = 1;
1957 			}
1958 			/* Wait 1 second for next try */
1959 			msleep(1000);
1960 		} else {
1961 			/* we don't understand the sense code, so it's
1962 			 * probably pointless to loop */
1963 			if(!spintime) {
1964 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1965 				sd_print_sense_hdr(sdkp, &sshdr);
1966 			}
1967 			break;
1968 		}
1969 
1970 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1971 
1972 	if (spintime) {
1973 		if (scsi_status_is_good(the_result))
1974 			printk("ready\n");
1975 		else
1976 			printk("not responding...\n");
1977 	}
1978 }
1979 
1980 
1981 /*
1982  * Determine whether disk supports Data Integrity Field.
1983  */
1984 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1985 {
1986 	struct scsi_device *sdp = sdkp->device;
1987 	u8 type;
1988 	int ret = 0;
1989 
1990 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1991 		return ret;
1992 
1993 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1994 
1995 	if (type > SD_DIF_TYPE3_PROTECTION)
1996 		ret = -ENODEV;
1997 	else if (scsi_host_dif_capable(sdp->host, type))
1998 		ret = 1;
1999 
2000 	if (sdkp->first_scan || type != sdkp->protection_type)
2001 		switch (ret) {
2002 		case -ENODEV:
2003 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2004 				  " protection type %u. Disabling disk!\n",
2005 				  type);
2006 			break;
2007 		case 1:
2008 			sd_printk(KERN_NOTICE, sdkp,
2009 				  "Enabling DIF Type %u protection\n", type);
2010 			break;
2011 		case 0:
2012 			sd_printk(KERN_NOTICE, sdkp,
2013 				  "Disabling DIF Type %u protection\n", type);
2014 			break;
2015 		}
2016 
2017 	sdkp->protection_type = type;
2018 
2019 	return ret;
2020 }
2021 
2022 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2023 			struct scsi_sense_hdr *sshdr, int sense_valid,
2024 			int the_result)
2025 {
2026 	if (driver_byte(the_result) & DRIVER_SENSE)
2027 		sd_print_sense_hdr(sdkp, sshdr);
2028 	else
2029 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2030 
2031 	/*
2032 	 * Set dirty bit for removable devices if not ready -
2033 	 * sometimes drives will not report this properly.
2034 	 */
2035 	if (sdp->removable &&
2036 	    sense_valid && sshdr->sense_key == NOT_READY)
2037 		set_media_not_present(sdkp);
2038 
2039 	/*
2040 	 * We used to set media_present to 0 here to indicate no media
2041 	 * in the drive, but some drives fail read capacity even with
2042 	 * media present, so we can't do that.
2043 	 */
2044 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2045 }
2046 
2047 #define RC16_LEN 32
2048 #if RC16_LEN > SD_BUF_SIZE
2049 #error RC16_LEN must not be more than SD_BUF_SIZE
2050 #endif
2051 
2052 #define READ_CAPACITY_RETRIES_ON_RESET	10
2053 
2054 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2055 						unsigned char *buffer)
2056 {
2057 	unsigned char cmd[16];
2058 	struct scsi_sense_hdr sshdr;
2059 	int sense_valid = 0;
2060 	int the_result;
2061 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2062 	unsigned int alignment;
2063 	unsigned long long lba;
2064 	unsigned sector_size;
2065 
2066 	if (sdp->no_read_capacity_16)
2067 		return -EINVAL;
2068 
2069 	do {
2070 		memset(cmd, 0, 16);
2071 		cmd[0] = SERVICE_ACTION_IN_16;
2072 		cmd[1] = SAI_READ_CAPACITY_16;
2073 		cmd[13] = RC16_LEN;
2074 		memset(buffer, 0, RC16_LEN);
2075 
2076 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2077 					buffer, RC16_LEN, &sshdr,
2078 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2079 
2080 		if (media_not_present(sdkp, &sshdr))
2081 			return -ENODEV;
2082 
2083 		if (the_result) {
2084 			sense_valid = scsi_sense_valid(&sshdr);
2085 			if (sense_valid &&
2086 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2087 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2088 			    sshdr.ascq == 0x00)
2089 				/* Invalid Command Operation Code or
2090 				 * Invalid Field in CDB, just retry
2091 				 * silently with RC10 */
2092 				return -EINVAL;
2093 			if (sense_valid &&
2094 			    sshdr.sense_key == UNIT_ATTENTION &&
2095 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2096 				/* Device reset might occur several times,
2097 				 * give it one more chance */
2098 				if (--reset_retries > 0)
2099 					continue;
2100 		}
2101 		retries--;
2102 
2103 	} while (the_result && retries);
2104 
2105 	if (the_result) {
2106 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2107 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2108 		return -EINVAL;
2109 	}
2110 
2111 	sector_size = get_unaligned_be32(&buffer[8]);
2112 	lba = get_unaligned_be64(&buffer[0]);
2113 
2114 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2115 		sdkp->capacity = 0;
2116 		return -ENODEV;
2117 	}
2118 
2119 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2120 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2121 			"kernel compiled with support for large block "
2122 			"devices.\n");
2123 		sdkp->capacity = 0;
2124 		return -EOVERFLOW;
2125 	}
2126 
2127 	/* Logical blocks per physical block exponent */
2128 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2129 
2130 	/* Lowest aligned logical block */
2131 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2132 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2133 	if (alignment && sdkp->first_scan)
2134 		sd_printk(KERN_NOTICE, sdkp,
2135 			  "physical block alignment offset: %u\n", alignment);
2136 
2137 	if (buffer[14] & 0x80) { /* LBPME */
2138 		sdkp->lbpme = 1;
2139 
2140 		if (buffer[14] & 0x40) /* LBPRZ */
2141 			sdkp->lbprz = 1;
2142 
2143 		sd_config_discard(sdkp, SD_LBP_WS16);
2144 	}
2145 
2146 	sdkp->capacity = lba + 1;
2147 	return sector_size;
2148 }
2149 
2150 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2151 						unsigned char *buffer)
2152 {
2153 	unsigned char cmd[16];
2154 	struct scsi_sense_hdr sshdr;
2155 	int sense_valid = 0;
2156 	int the_result;
2157 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2158 	sector_t lba;
2159 	unsigned sector_size;
2160 
2161 	do {
2162 		cmd[0] = READ_CAPACITY;
2163 		memset(&cmd[1], 0, 9);
2164 		memset(buffer, 0, 8);
2165 
2166 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2167 					buffer, 8, &sshdr,
2168 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2169 
2170 		if (media_not_present(sdkp, &sshdr))
2171 			return -ENODEV;
2172 
2173 		if (the_result) {
2174 			sense_valid = scsi_sense_valid(&sshdr);
2175 			if (sense_valid &&
2176 			    sshdr.sense_key == UNIT_ATTENTION &&
2177 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2178 				/* Device reset might occur several times,
2179 				 * give it one more chance */
2180 				if (--reset_retries > 0)
2181 					continue;
2182 		}
2183 		retries--;
2184 
2185 	} while (the_result && retries);
2186 
2187 	if (the_result) {
2188 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2189 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2190 		return -EINVAL;
2191 	}
2192 
2193 	sector_size = get_unaligned_be32(&buffer[4]);
2194 	lba = get_unaligned_be32(&buffer[0]);
2195 
2196 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2197 		/* Some buggy (usb cardreader) devices return an lba of
2198 		   0xffffffff when the want to report a size of 0 (with
2199 		   which they really mean no media is present) */
2200 		sdkp->capacity = 0;
2201 		sdkp->physical_block_size = sector_size;
2202 		return sector_size;
2203 	}
2204 
2205 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2206 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2207 			"kernel compiled with support for large block "
2208 			"devices.\n");
2209 		sdkp->capacity = 0;
2210 		return -EOVERFLOW;
2211 	}
2212 
2213 	sdkp->capacity = lba + 1;
2214 	sdkp->physical_block_size = sector_size;
2215 	return sector_size;
2216 }
2217 
2218 static int sd_try_rc16_first(struct scsi_device *sdp)
2219 {
2220 	if (sdp->host->max_cmd_len < 16)
2221 		return 0;
2222 	if (sdp->try_rc_10_first)
2223 		return 0;
2224 	if (sdp->scsi_level > SCSI_SPC_2)
2225 		return 1;
2226 	if (scsi_device_protection(sdp))
2227 		return 1;
2228 	return 0;
2229 }
2230 
2231 /*
2232  * read disk capacity
2233  */
2234 static void
2235 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2236 {
2237 	int sector_size;
2238 	struct scsi_device *sdp = sdkp->device;
2239 	sector_t old_capacity = sdkp->capacity;
2240 
2241 	if (sd_try_rc16_first(sdp)) {
2242 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2243 		if (sector_size == -EOVERFLOW)
2244 			goto got_data;
2245 		if (sector_size == -ENODEV)
2246 			return;
2247 		if (sector_size < 0)
2248 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2249 		if (sector_size < 0)
2250 			return;
2251 	} else {
2252 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2253 		if (sector_size == -EOVERFLOW)
2254 			goto got_data;
2255 		if (sector_size < 0)
2256 			return;
2257 		if ((sizeof(sdkp->capacity) > 4) &&
2258 		    (sdkp->capacity > 0xffffffffULL)) {
2259 			int old_sector_size = sector_size;
2260 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2261 					"Trying to use READ CAPACITY(16).\n");
2262 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2263 			if (sector_size < 0) {
2264 				sd_printk(KERN_NOTICE, sdkp,
2265 					"Using 0xffffffff as device size\n");
2266 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2267 				sector_size = old_sector_size;
2268 				goto got_data;
2269 			}
2270 		}
2271 	}
2272 
2273 	/* Some devices are known to return the total number of blocks,
2274 	 * not the highest block number.  Some devices have versions
2275 	 * which do this and others which do not.  Some devices we might
2276 	 * suspect of doing this but we don't know for certain.
2277 	 *
2278 	 * If we know the reported capacity is wrong, decrement it.  If
2279 	 * we can only guess, then assume the number of blocks is even
2280 	 * (usually true but not always) and err on the side of lowering
2281 	 * the capacity.
2282 	 */
2283 	if (sdp->fix_capacity ||
2284 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2285 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2286 				"from its reported value: %llu\n",
2287 				(unsigned long long) sdkp->capacity);
2288 		--sdkp->capacity;
2289 	}
2290 
2291 got_data:
2292 	if (sector_size == 0) {
2293 		sector_size = 512;
2294 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2295 			  "assuming 512.\n");
2296 	}
2297 
2298 	if (sector_size != 512 &&
2299 	    sector_size != 1024 &&
2300 	    sector_size != 2048 &&
2301 	    sector_size != 4096) {
2302 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2303 			  sector_size);
2304 		/*
2305 		 * The user might want to re-format the drive with
2306 		 * a supported sectorsize.  Once this happens, it
2307 		 * would be relatively trivial to set the thing up.
2308 		 * For this reason, we leave the thing in the table.
2309 		 */
2310 		sdkp->capacity = 0;
2311 		/*
2312 		 * set a bogus sector size so the normal read/write
2313 		 * logic in the block layer will eventually refuse any
2314 		 * request on this device without tripping over power
2315 		 * of two sector size assumptions
2316 		 */
2317 		sector_size = 512;
2318 	}
2319 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2320 
2321 	{
2322 		char cap_str_2[10], cap_str_10[10];
2323 
2324 		string_get_size(sdkp->capacity, sector_size,
2325 				STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2326 		string_get_size(sdkp->capacity, sector_size,
2327 				STRING_UNITS_10, cap_str_10,
2328 				sizeof(cap_str_10));
2329 
2330 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2331 			sd_printk(KERN_NOTICE, sdkp,
2332 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2333 				  (unsigned long long)sdkp->capacity,
2334 				  sector_size, cap_str_10, cap_str_2);
2335 
2336 			if (sdkp->physical_block_size != sector_size)
2337 				sd_printk(KERN_NOTICE, sdkp,
2338 					  "%u-byte physical blocks\n",
2339 					  sdkp->physical_block_size);
2340 		}
2341 	}
2342 
2343 	if (sdkp->capacity > 0xffffffff)
2344 		sdp->use_16_for_rw = 1;
2345 
2346 	blk_queue_physical_block_size(sdp->request_queue,
2347 				      sdkp->physical_block_size);
2348 	sdkp->device->sector_size = sector_size;
2349 }
2350 
2351 /* called with buffer of length 512 */
2352 static inline int
2353 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2354 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2355 		 struct scsi_sense_hdr *sshdr)
2356 {
2357 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2358 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2359 			       sshdr);
2360 }
2361 
2362 /*
2363  * read write protect setting, if possible - called only in sd_revalidate_disk()
2364  * called with buffer of length SD_BUF_SIZE
2365  */
2366 static void
2367 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2368 {
2369 	int res;
2370 	struct scsi_device *sdp = sdkp->device;
2371 	struct scsi_mode_data data;
2372 	int old_wp = sdkp->write_prot;
2373 
2374 	set_disk_ro(sdkp->disk, 0);
2375 	if (sdp->skip_ms_page_3f) {
2376 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2377 		return;
2378 	}
2379 
2380 	if (sdp->use_192_bytes_for_3f) {
2381 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2382 	} else {
2383 		/*
2384 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2385 		 * We have to start carefully: some devices hang if we ask
2386 		 * for more than is available.
2387 		 */
2388 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2389 
2390 		/*
2391 		 * Second attempt: ask for page 0 When only page 0 is
2392 		 * implemented, a request for page 3F may return Sense Key
2393 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2394 		 * CDB.
2395 		 */
2396 		if (!scsi_status_is_good(res))
2397 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2398 
2399 		/*
2400 		 * Third attempt: ask 255 bytes, as we did earlier.
2401 		 */
2402 		if (!scsi_status_is_good(res))
2403 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2404 					       &data, NULL);
2405 	}
2406 
2407 	if (!scsi_status_is_good(res)) {
2408 		sd_first_printk(KERN_WARNING, sdkp,
2409 			  "Test WP failed, assume Write Enabled\n");
2410 	} else {
2411 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2412 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2413 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2414 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2415 				  sdkp->write_prot ? "on" : "off");
2416 			sd_printk(KERN_DEBUG, sdkp,
2417 				  "Mode Sense: %02x %02x %02x %02x\n",
2418 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2419 		}
2420 	}
2421 }
2422 
2423 /*
2424  * sd_read_cache_type - called only from sd_revalidate_disk()
2425  * called with buffer of length SD_BUF_SIZE
2426  */
2427 static void
2428 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2429 {
2430 	int len = 0, res;
2431 	struct scsi_device *sdp = sdkp->device;
2432 
2433 	int dbd;
2434 	int modepage;
2435 	int first_len;
2436 	struct scsi_mode_data data;
2437 	struct scsi_sense_hdr sshdr;
2438 	int old_wce = sdkp->WCE;
2439 	int old_rcd = sdkp->RCD;
2440 	int old_dpofua = sdkp->DPOFUA;
2441 
2442 
2443 	if (sdkp->cache_override)
2444 		return;
2445 
2446 	first_len = 4;
2447 	if (sdp->skip_ms_page_8) {
2448 		if (sdp->type == TYPE_RBC)
2449 			goto defaults;
2450 		else {
2451 			if (sdp->skip_ms_page_3f)
2452 				goto defaults;
2453 			modepage = 0x3F;
2454 			if (sdp->use_192_bytes_for_3f)
2455 				first_len = 192;
2456 			dbd = 0;
2457 		}
2458 	} else if (sdp->type == TYPE_RBC) {
2459 		modepage = 6;
2460 		dbd = 8;
2461 	} else {
2462 		modepage = 8;
2463 		dbd = 0;
2464 	}
2465 
2466 	/* cautiously ask */
2467 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2468 			&data, &sshdr);
2469 
2470 	if (!scsi_status_is_good(res))
2471 		goto bad_sense;
2472 
2473 	if (!data.header_length) {
2474 		modepage = 6;
2475 		first_len = 0;
2476 		sd_first_printk(KERN_ERR, sdkp,
2477 				"Missing header in MODE_SENSE response\n");
2478 	}
2479 
2480 	/* that went OK, now ask for the proper length */
2481 	len = data.length;
2482 
2483 	/*
2484 	 * We're only interested in the first three bytes, actually.
2485 	 * But the data cache page is defined for the first 20.
2486 	 */
2487 	if (len < 3)
2488 		goto bad_sense;
2489 	else if (len > SD_BUF_SIZE) {
2490 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2491 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2492 		len = SD_BUF_SIZE;
2493 	}
2494 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2495 		len = 192;
2496 
2497 	/* Get the data */
2498 	if (len > first_len)
2499 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2500 				&data, &sshdr);
2501 
2502 	if (scsi_status_is_good(res)) {
2503 		int offset = data.header_length + data.block_descriptor_length;
2504 
2505 		while (offset < len) {
2506 			u8 page_code = buffer[offset] & 0x3F;
2507 			u8 spf       = buffer[offset] & 0x40;
2508 
2509 			if (page_code == 8 || page_code == 6) {
2510 				/* We're interested only in the first 3 bytes.
2511 				 */
2512 				if (len - offset <= 2) {
2513 					sd_first_printk(KERN_ERR, sdkp,
2514 						"Incomplete mode parameter "
2515 							"data\n");
2516 					goto defaults;
2517 				} else {
2518 					modepage = page_code;
2519 					goto Page_found;
2520 				}
2521 			} else {
2522 				/* Go to the next page */
2523 				if (spf && len - offset > 3)
2524 					offset += 4 + (buffer[offset+2] << 8) +
2525 						buffer[offset+3];
2526 				else if (!spf && len - offset > 1)
2527 					offset += 2 + buffer[offset+1];
2528 				else {
2529 					sd_first_printk(KERN_ERR, sdkp,
2530 							"Incomplete mode "
2531 							"parameter data\n");
2532 					goto defaults;
2533 				}
2534 			}
2535 		}
2536 
2537 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2538 		goto defaults;
2539 
2540 	Page_found:
2541 		if (modepage == 8) {
2542 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2543 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2544 		} else {
2545 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2546 			sdkp->RCD = 0;
2547 		}
2548 
2549 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2550 		if (sdp->broken_fua) {
2551 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2552 			sdkp->DPOFUA = 0;
2553 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2554 			sd_first_printk(KERN_NOTICE, sdkp,
2555 				  "Uses READ/WRITE(6), disabling FUA\n");
2556 			sdkp->DPOFUA = 0;
2557 		}
2558 
2559 		/* No cache flush allowed for write protected devices */
2560 		if (sdkp->WCE && sdkp->write_prot)
2561 			sdkp->WCE = 0;
2562 
2563 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2564 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2565 			sd_printk(KERN_NOTICE, sdkp,
2566 				  "Write cache: %s, read cache: %s, %s\n",
2567 				  sdkp->WCE ? "enabled" : "disabled",
2568 				  sdkp->RCD ? "disabled" : "enabled",
2569 				  sdkp->DPOFUA ? "supports DPO and FUA"
2570 				  : "doesn't support DPO or FUA");
2571 
2572 		return;
2573 	}
2574 
2575 bad_sense:
2576 	if (scsi_sense_valid(&sshdr) &&
2577 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2578 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2579 		/* Invalid field in CDB */
2580 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2581 	else
2582 		sd_first_printk(KERN_ERR, sdkp,
2583 				"Asking for cache data failed\n");
2584 
2585 defaults:
2586 	if (sdp->wce_default_on) {
2587 		sd_first_printk(KERN_NOTICE, sdkp,
2588 				"Assuming drive cache: write back\n");
2589 		sdkp->WCE = 1;
2590 	} else {
2591 		sd_first_printk(KERN_ERR, sdkp,
2592 				"Assuming drive cache: write through\n");
2593 		sdkp->WCE = 0;
2594 	}
2595 	sdkp->RCD = 0;
2596 	sdkp->DPOFUA = 0;
2597 }
2598 
2599 /*
2600  * The ATO bit indicates whether the DIF application tag is available
2601  * for use by the operating system.
2602  */
2603 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2604 {
2605 	int res, offset;
2606 	struct scsi_device *sdp = sdkp->device;
2607 	struct scsi_mode_data data;
2608 	struct scsi_sense_hdr sshdr;
2609 
2610 	if (sdp->type != TYPE_DISK)
2611 		return;
2612 
2613 	if (sdkp->protection_type == 0)
2614 		return;
2615 
2616 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2617 			      SD_MAX_RETRIES, &data, &sshdr);
2618 
2619 	if (!scsi_status_is_good(res) || !data.header_length ||
2620 	    data.length < 6) {
2621 		sd_first_printk(KERN_WARNING, sdkp,
2622 			  "getting Control mode page failed, assume no ATO\n");
2623 
2624 		if (scsi_sense_valid(&sshdr))
2625 			sd_print_sense_hdr(sdkp, &sshdr);
2626 
2627 		return;
2628 	}
2629 
2630 	offset = data.header_length + data.block_descriptor_length;
2631 
2632 	if ((buffer[offset] & 0x3f) != 0x0a) {
2633 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2634 		return;
2635 	}
2636 
2637 	if ((buffer[offset + 5] & 0x80) == 0)
2638 		return;
2639 
2640 	sdkp->ATO = 1;
2641 
2642 	return;
2643 }
2644 
2645 /**
2646  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2647  * @disk: disk to query
2648  */
2649 static void sd_read_block_limits(struct scsi_disk *sdkp)
2650 {
2651 	unsigned int sector_sz = sdkp->device->sector_size;
2652 	const int vpd_len = 64;
2653 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2654 
2655 	if (!buffer ||
2656 	    /* Block Limits VPD */
2657 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2658 		goto out;
2659 
2660 	blk_queue_io_min(sdkp->disk->queue,
2661 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2662 
2663 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2664 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2665 
2666 	if (buffer[3] == 0x3c) {
2667 		unsigned int lba_count, desc_count;
2668 
2669 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2670 
2671 		if (!sdkp->lbpme)
2672 			goto out;
2673 
2674 		lba_count = get_unaligned_be32(&buffer[20]);
2675 		desc_count = get_unaligned_be32(&buffer[24]);
2676 
2677 		if (lba_count && desc_count)
2678 			sdkp->max_unmap_blocks = lba_count;
2679 
2680 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2681 
2682 		if (buffer[32] & 0x80)
2683 			sdkp->unmap_alignment =
2684 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2685 
2686 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2687 
2688 			if (sdkp->max_unmap_blocks)
2689 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2690 			else
2691 				sd_config_discard(sdkp, SD_LBP_WS16);
2692 
2693 		} else {	/* LBP VPD page tells us what to use */
2694 			if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2695 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2696 			else if (sdkp->lbpws)
2697 				sd_config_discard(sdkp, SD_LBP_WS16);
2698 			else if (sdkp->lbpws10)
2699 				sd_config_discard(sdkp, SD_LBP_WS10);
2700 			else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2701 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2702 			else
2703 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2704 		}
2705 	}
2706 
2707  out:
2708 	kfree(buffer);
2709 }
2710 
2711 /**
2712  * sd_read_block_characteristics - Query block dev. characteristics
2713  * @disk: disk to query
2714  */
2715 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2716 {
2717 	unsigned char *buffer;
2718 	u16 rot;
2719 	const int vpd_len = 64;
2720 
2721 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2722 
2723 	if (!buffer ||
2724 	    /* Block Device Characteristics VPD */
2725 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2726 		goto out;
2727 
2728 	rot = get_unaligned_be16(&buffer[4]);
2729 
2730 	if (rot == 1) {
2731 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2732 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, sdkp->disk->queue);
2733 	}
2734 
2735  out:
2736 	kfree(buffer);
2737 }
2738 
2739 /**
2740  * sd_read_block_provisioning - Query provisioning VPD page
2741  * @disk: disk to query
2742  */
2743 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2744 {
2745 	unsigned char *buffer;
2746 	const int vpd_len = 8;
2747 
2748 	if (sdkp->lbpme == 0)
2749 		return;
2750 
2751 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2752 
2753 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2754 		goto out;
2755 
2756 	sdkp->lbpvpd	= 1;
2757 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2758 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2759 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2760 
2761  out:
2762 	kfree(buffer);
2763 }
2764 
2765 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2766 {
2767 	struct scsi_device *sdev = sdkp->device;
2768 
2769 	if (sdev->host->no_write_same) {
2770 		sdev->no_write_same = 1;
2771 
2772 		return;
2773 	}
2774 
2775 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2776 		/* too large values might cause issues with arcmsr */
2777 		int vpd_buf_len = 64;
2778 
2779 		sdev->no_report_opcodes = 1;
2780 
2781 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2782 		 * CODES is unsupported and the device has an ATA
2783 		 * Information VPD page (SAT).
2784 		 */
2785 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2786 			sdev->no_write_same = 1;
2787 	}
2788 
2789 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2790 		sdkp->ws16 = 1;
2791 
2792 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2793 		sdkp->ws10 = 1;
2794 }
2795 
2796 /**
2797  *	sd_revalidate_disk - called the first time a new disk is seen,
2798  *	performs disk spin up, read_capacity, etc.
2799  *	@disk: struct gendisk we care about
2800  **/
2801 static int sd_revalidate_disk(struct gendisk *disk)
2802 {
2803 	struct scsi_disk *sdkp = scsi_disk(disk);
2804 	struct scsi_device *sdp = sdkp->device;
2805 	struct request_queue *q = sdkp->disk->queue;
2806 	unsigned char *buffer;
2807 	unsigned int dev_max, rw_max;
2808 
2809 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2810 				      "sd_revalidate_disk\n"));
2811 
2812 	/*
2813 	 * If the device is offline, don't try and read capacity or any
2814 	 * of the other niceties.
2815 	 */
2816 	if (!scsi_device_online(sdp))
2817 		goto out;
2818 
2819 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2820 	if (!buffer) {
2821 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2822 			  "allocation failure.\n");
2823 		goto out;
2824 	}
2825 
2826 	sd_spinup_disk(sdkp);
2827 
2828 	/*
2829 	 * Without media there is no reason to ask; moreover, some devices
2830 	 * react badly if we do.
2831 	 */
2832 	if (sdkp->media_present) {
2833 		sd_read_capacity(sdkp, buffer);
2834 
2835 		if (scsi_device_supports_vpd(sdp)) {
2836 			sd_read_block_provisioning(sdkp);
2837 			sd_read_block_limits(sdkp);
2838 			sd_read_block_characteristics(sdkp);
2839 		}
2840 
2841 		sd_read_write_protect_flag(sdkp, buffer);
2842 		sd_read_cache_type(sdkp, buffer);
2843 		sd_read_app_tag_own(sdkp, buffer);
2844 		sd_read_write_same(sdkp, buffer);
2845 	}
2846 
2847 	sdkp->first_scan = 0;
2848 
2849 	/*
2850 	 * We now have all cache related info, determine how we deal
2851 	 * with flush requests.
2852 	 */
2853 	sd_set_flush_flag(sdkp);
2854 
2855 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2856 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2857 
2858 	/* Some devices report a maximum block count for READ/WRITE requests. */
2859 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2860 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2861 
2862 	/*
2863 	 * Use the device's preferred I/O size for reads and writes
2864 	 * unless the reported value is unreasonably small, large, or
2865 	 * garbage.
2866 	 */
2867 	if (sdkp->opt_xfer_blocks &&
2868 	    sdkp->opt_xfer_blocks <= dev_max &&
2869 	    sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2870 	    logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2871 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2872 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2873 	} else
2874 		rw_max = BLK_DEF_MAX_SECTORS;
2875 
2876 	/* Combine with controller limits */
2877 	q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2878 
2879 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2880 	sd_config_write_same(sdkp);
2881 	kfree(buffer);
2882 
2883  out:
2884 	return 0;
2885 }
2886 
2887 /**
2888  *	sd_unlock_native_capacity - unlock native capacity
2889  *	@disk: struct gendisk to set capacity for
2890  *
2891  *	Block layer calls this function if it detects that partitions
2892  *	on @disk reach beyond the end of the device.  If the SCSI host
2893  *	implements ->unlock_native_capacity() method, it's invoked to
2894  *	give it a chance to adjust the device capacity.
2895  *
2896  *	CONTEXT:
2897  *	Defined by block layer.  Might sleep.
2898  */
2899 static void sd_unlock_native_capacity(struct gendisk *disk)
2900 {
2901 	struct scsi_device *sdev = scsi_disk(disk)->device;
2902 
2903 	if (sdev->host->hostt->unlock_native_capacity)
2904 		sdev->host->hostt->unlock_native_capacity(sdev);
2905 }
2906 
2907 /**
2908  *	sd_format_disk_name - format disk name
2909  *	@prefix: name prefix - ie. "sd" for SCSI disks
2910  *	@index: index of the disk to format name for
2911  *	@buf: output buffer
2912  *	@buflen: length of the output buffer
2913  *
2914  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2915  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2916  *	which is followed by sdaaa.
2917  *
2918  *	This is basically 26 base counting with one extra 'nil' entry
2919  *	at the beginning from the second digit on and can be
2920  *	determined using similar method as 26 base conversion with the
2921  *	index shifted -1 after each digit is computed.
2922  *
2923  *	CONTEXT:
2924  *	Don't care.
2925  *
2926  *	RETURNS:
2927  *	0 on success, -errno on failure.
2928  */
2929 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2930 {
2931 	const int base = 'z' - 'a' + 1;
2932 	char *begin = buf + strlen(prefix);
2933 	char *end = buf + buflen;
2934 	char *p;
2935 	int unit;
2936 
2937 	p = end - 1;
2938 	*p = '\0';
2939 	unit = base;
2940 	do {
2941 		if (p == begin)
2942 			return -EINVAL;
2943 		*--p = 'a' + (index % unit);
2944 		index = (index / unit) - 1;
2945 	} while (index >= 0);
2946 
2947 	memmove(begin, p, end - p);
2948 	memcpy(buf, prefix, strlen(prefix));
2949 
2950 	return 0;
2951 }
2952 
2953 /*
2954  * The asynchronous part of sd_probe
2955  */
2956 static void sd_probe_async(void *data, async_cookie_t cookie)
2957 {
2958 	struct scsi_disk *sdkp = data;
2959 	struct scsi_device *sdp;
2960 	struct gendisk *gd;
2961 	u32 index;
2962 	struct device *dev;
2963 
2964 	sdp = sdkp->device;
2965 	gd = sdkp->disk;
2966 	index = sdkp->index;
2967 	dev = &sdp->sdev_gendev;
2968 
2969 	gd->major = sd_major((index & 0xf0) >> 4);
2970 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2971 	gd->minors = SD_MINORS;
2972 
2973 	gd->fops = &sd_fops;
2974 	gd->private_data = &sdkp->driver;
2975 	gd->queue = sdkp->device->request_queue;
2976 
2977 	/* defaults, until the device tells us otherwise */
2978 	sdp->sector_size = 512;
2979 	sdkp->capacity = 0;
2980 	sdkp->media_present = 1;
2981 	sdkp->write_prot = 0;
2982 	sdkp->cache_override = 0;
2983 	sdkp->WCE = 0;
2984 	sdkp->RCD = 0;
2985 	sdkp->ATO = 0;
2986 	sdkp->first_scan = 1;
2987 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2988 
2989 	sd_revalidate_disk(gd);
2990 
2991 	gd->driverfs_dev = &sdp->sdev_gendev;
2992 	gd->flags = GENHD_FL_EXT_DEVT;
2993 	if (sdp->removable) {
2994 		gd->flags |= GENHD_FL_REMOVABLE;
2995 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2996 	}
2997 
2998 	blk_pm_runtime_init(sdp->request_queue, dev);
2999 	add_disk(gd);
3000 	if (sdkp->capacity)
3001 		sd_dif_config_host(sdkp);
3002 
3003 	sd_revalidate_disk(gd);
3004 
3005 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3006 		  sdp->removable ? "removable " : "");
3007 	scsi_autopm_put_device(sdp);
3008 	put_device(&sdkp->dev);
3009 }
3010 
3011 /**
3012  *	sd_probe - called during driver initialization and whenever a
3013  *	new scsi device is attached to the system. It is called once
3014  *	for each scsi device (not just disks) present.
3015  *	@dev: pointer to device object
3016  *
3017  *	Returns 0 if successful (or not interested in this scsi device
3018  *	(e.g. scanner)); 1 when there is an error.
3019  *
3020  *	Note: this function is invoked from the scsi mid-level.
3021  *	This function sets up the mapping between a given
3022  *	<host,channel,id,lun> (found in sdp) and new device name
3023  *	(e.g. /dev/sda). More precisely it is the block device major
3024  *	and minor number that is chosen here.
3025  *
3026  *	Assume sd_probe is not re-entrant (for time being)
3027  *	Also think about sd_probe() and sd_remove() running coincidentally.
3028  **/
3029 static int sd_probe(struct device *dev)
3030 {
3031 	struct scsi_device *sdp = to_scsi_device(dev);
3032 	struct scsi_disk *sdkp;
3033 	struct gendisk *gd;
3034 	int index;
3035 	int error;
3036 
3037 	scsi_autopm_get_device(sdp);
3038 	error = -ENODEV;
3039 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
3040 		goto out;
3041 
3042 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3043 					"sd_probe\n"));
3044 
3045 	error = -ENOMEM;
3046 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3047 	if (!sdkp)
3048 		goto out;
3049 
3050 	gd = alloc_disk(SD_MINORS);
3051 	if (!gd)
3052 		goto out_free;
3053 
3054 	do {
3055 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3056 			goto out_put;
3057 
3058 		spin_lock(&sd_index_lock);
3059 		error = ida_get_new(&sd_index_ida, &index);
3060 		spin_unlock(&sd_index_lock);
3061 	} while (error == -EAGAIN);
3062 
3063 	if (error) {
3064 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3065 		goto out_put;
3066 	}
3067 
3068 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3069 	if (error) {
3070 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3071 		goto out_free_index;
3072 	}
3073 
3074 	sdkp->device = sdp;
3075 	sdkp->driver = &sd_template;
3076 	sdkp->disk = gd;
3077 	sdkp->index = index;
3078 	atomic_set(&sdkp->openers, 0);
3079 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3080 
3081 	if (!sdp->request_queue->rq_timeout) {
3082 		if (sdp->type != TYPE_MOD)
3083 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3084 		else
3085 			blk_queue_rq_timeout(sdp->request_queue,
3086 					     SD_MOD_TIMEOUT);
3087 	}
3088 
3089 	device_initialize(&sdkp->dev);
3090 	sdkp->dev.parent = dev;
3091 	sdkp->dev.class = &sd_disk_class;
3092 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3093 
3094 	error = device_add(&sdkp->dev);
3095 	if (error)
3096 		goto out_free_index;
3097 
3098 	get_device(dev);
3099 	dev_set_drvdata(dev, sdkp);
3100 
3101 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3102 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3103 
3104 	return 0;
3105 
3106  out_free_index:
3107 	spin_lock(&sd_index_lock);
3108 	ida_remove(&sd_index_ida, index);
3109 	spin_unlock(&sd_index_lock);
3110  out_put:
3111 	put_disk(gd);
3112  out_free:
3113 	kfree(sdkp);
3114  out:
3115 	scsi_autopm_put_device(sdp);
3116 	return error;
3117 }
3118 
3119 /**
3120  *	sd_remove - called whenever a scsi disk (previously recognized by
3121  *	sd_probe) is detached from the system. It is called (potentially
3122  *	multiple times) during sd module unload.
3123  *	@sdp: pointer to mid level scsi device object
3124  *
3125  *	Note: this function is invoked from the scsi mid-level.
3126  *	This function potentially frees up a device name (e.g. /dev/sdc)
3127  *	that could be re-used by a subsequent sd_probe().
3128  *	This function is not called when the built-in sd driver is "exit-ed".
3129  **/
3130 static int sd_remove(struct device *dev)
3131 {
3132 	struct scsi_disk *sdkp;
3133 	dev_t devt;
3134 
3135 	sdkp = dev_get_drvdata(dev);
3136 	devt = disk_devt(sdkp->disk);
3137 	scsi_autopm_get_device(sdkp->device);
3138 
3139 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3140 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3141 	device_del(&sdkp->dev);
3142 	del_gendisk(sdkp->disk);
3143 	sd_shutdown(dev);
3144 
3145 	blk_register_region(devt, SD_MINORS, NULL,
3146 			    sd_default_probe, NULL, NULL);
3147 
3148 	mutex_lock(&sd_ref_mutex);
3149 	dev_set_drvdata(dev, NULL);
3150 	put_device(&sdkp->dev);
3151 	mutex_unlock(&sd_ref_mutex);
3152 
3153 	return 0;
3154 }
3155 
3156 /**
3157  *	scsi_disk_release - Called to free the scsi_disk structure
3158  *	@dev: pointer to embedded class device
3159  *
3160  *	sd_ref_mutex must be held entering this routine.  Because it is
3161  *	called on last put, you should always use the scsi_disk_get()
3162  *	scsi_disk_put() helpers which manipulate the semaphore directly
3163  *	and never do a direct put_device.
3164  **/
3165 static void scsi_disk_release(struct device *dev)
3166 {
3167 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3168 	struct gendisk *disk = sdkp->disk;
3169 
3170 	spin_lock(&sd_index_lock);
3171 	ida_remove(&sd_index_ida, sdkp->index);
3172 	spin_unlock(&sd_index_lock);
3173 
3174 	disk->private_data = NULL;
3175 	put_disk(disk);
3176 	put_device(&sdkp->device->sdev_gendev);
3177 
3178 	kfree(sdkp);
3179 }
3180 
3181 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3182 {
3183 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3184 	struct scsi_sense_hdr sshdr;
3185 	struct scsi_device *sdp = sdkp->device;
3186 	int res;
3187 
3188 	if (start)
3189 		cmd[4] |= 1;	/* START */
3190 
3191 	if (sdp->start_stop_pwr_cond)
3192 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3193 
3194 	if (!scsi_device_online(sdp))
3195 		return -ENODEV;
3196 
3197 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3198 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3199 	if (res) {
3200 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3201 		if (driver_byte(res) & DRIVER_SENSE)
3202 			sd_print_sense_hdr(sdkp, &sshdr);
3203 		if (scsi_sense_valid(&sshdr) &&
3204 			/* 0x3a is medium not present */
3205 			sshdr.asc == 0x3a)
3206 			res = 0;
3207 	}
3208 
3209 	/* SCSI error codes must not go to the generic layer */
3210 	if (res)
3211 		return -EIO;
3212 
3213 	return 0;
3214 }
3215 
3216 /*
3217  * Send a SYNCHRONIZE CACHE instruction down to the device through
3218  * the normal SCSI command structure.  Wait for the command to
3219  * complete.
3220  */
3221 static void sd_shutdown(struct device *dev)
3222 {
3223 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3224 
3225 	if (!sdkp)
3226 		return;         /* this can happen */
3227 
3228 	if (pm_runtime_suspended(dev))
3229 		return;
3230 
3231 	if (sdkp->WCE && sdkp->media_present) {
3232 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3233 		sd_sync_cache(sdkp);
3234 	}
3235 
3236 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3237 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3238 		sd_start_stop_device(sdkp, 0);
3239 	}
3240 }
3241 
3242 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3243 {
3244 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3245 	int ret = 0;
3246 
3247 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3248 		return 0;
3249 
3250 	if (sdkp->WCE && sdkp->media_present) {
3251 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3252 		ret = sd_sync_cache(sdkp);
3253 		if (ret) {
3254 			/* ignore OFFLINE device */
3255 			if (ret == -ENODEV)
3256 				ret = 0;
3257 			goto done;
3258 		}
3259 	}
3260 
3261 	if (sdkp->device->manage_start_stop) {
3262 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3263 		/* an error is not worth aborting a system sleep */
3264 		ret = sd_start_stop_device(sdkp, 0);
3265 		if (ignore_stop_errors)
3266 			ret = 0;
3267 	}
3268 
3269 done:
3270 	return ret;
3271 }
3272 
3273 static int sd_suspend_system(struct device *dev)
3274 {
3275 	return sd_suspend_common(dev, true);
3276 }
3277 
3278 static int sd_suspend_runtime(struct device *dev)
3279 {
3280 	return sd_suspend_common(dev, false);
3281 }
3282 
3283 static int sd_resume(struct device *dev)
3284 {
3285 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3286 
3287 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3288 		return 0;
3289 
3290 	if (!sdkp->device->manage_start_stop)
3291 		return 0;
3292 
3293 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3294 	return sd_start_stop_device(sdkp, 1);
3295 }
3296 
3297 /**
3298  *	init_sd - entry point for this driver (both when built in or when
3299  *	a module).
3300  *
3301  *	Note: this function registers this driver with the scsi mid-level.
3302  **/
3303 static int __init init_sd(void)
3304 {
3305 	int majors = 0, i, err;
3306 
3307 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3308 
3309 	for (i = 0; i < SD_MAJORS; i++) {
3310 		if (register_blkdev(sd_major(i), "sd") != 0)
3311 			continue;
3312 		majors++;
3313 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3314 				    sd_default_probe, NULL, NULL);
3315 	}
3316 
3317 	if (!majors)
3318 		return -ENODEV;
3319 
3320 	err = class_register(&sd_disk_class);
3321 	if (err)
3322 		goto err_out;
3323 
3324 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3325 					 0, 0, NULL);
3326 	if (!sd_cdb_cache) {
3327 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3328 		err = -ENOMEM;
3329 		goto err_out_class;
3330 	}
3331 
3332 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3333 	if (!sd_cdb_pool) {
3334 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3335 		err = -ENOMEM;
3336 		goto err_out_cache;
3337 	}
3338 
3339 	err = scsi_register_driver(&sd_template.gendrv);
3340 	if (err)
3341 		goto err_out_driver;
3342 
3343 	return 0;
3344 
3345 err_out_driver:
3346 	mempool_destroy(sd_cdb_pool);
3347 
3348 err_out_cache:
3349 	kmem_cache_destroy(sd_cdb_cache);
3350 
3351 err_out_class:
3352 	class_unregister(&sd_disk_class);
3353 err_out:
3354 	for (i = 0; i < SD_MAJORS; i++)
3355 		unregister_blkdev(sd_major(i), "sd");
3356 	return err;
3357 }
3358 
3359 /**
3360  *	exit_sd - exit point for this driver (when it is a module).
3361  *
3362  *	Note: this function unregisters this driver from the scsi mid-level.
3363  **/
3364 static void __exit exit_sd(void)
3365 {
3366 	int i;
3367 
3368 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3369 
3370 	scsi_unregister_driver(&sd_template.gendrv);
3371 	mempool_destroy(sd_cdb_pool);
3372 	kmem_cache_destroy(sd_cdb_cache);
3373 
3374 	class_unregister(&sd_disk_class);
3375 
3376 	for (i = 0; i < SD_MAJORS; i++) {
3377 		blk_unregister_region(sd_major(i), SD_MINORS);
3378 		unregister_blkdev(sd_major(i), "sd");
3379 	}
3380 }
3381 
3382 module_init(init_sd);
3383 module_exit(exit_sd);
3384 
3385 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3386 			       struct scsi_sense_hdr *sshdr)
3387 {
3388 	scsi_print_sense_hdr(sdkp->device,
3389 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3390 }
3391 
3392 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3393 			    int result)
3394 {
3395 	const char *hb_string = scsi_hostbyte_string(result);
3396 	const char *db_string = scsi_driverbyte_string(result);
3397 
3398 	if (hb_string || db_string)
3399 		sd_printk(KERN_INFO, sdkp,
3400 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3401 			  hb_string ? hb_string : "invalid",
3402 			  db_string ? db_string : "invalid");
3403 	else
3404 		sd_printk(KERN_INFO, sdkp,
3405 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3406 			  msg, host_byte(result), driver_byte(result));
3407 }
3408 
3409