xref: /openbmc/linux/drivers/scsi/sd.c (revision b34e08d5)
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 <asm/uaccess.h>
55 #include <asm/unaligned.h>
56 
57 #include <scsi/scsi.h>
58 #include <scsi/scsi_cmnd.h>
59 #include <scsi/scsi_dbg.h>
60 #include <scsi/scsi_device.h>
61 #include <scsi/scsi_driver.h>
62 #include <scsi/scsi_eh.h>
63 #include <scsi/scsi_host.h>
64 #include <scsi/scsi_ioctl.h>
65 #include <scsi/scsicam.h>
66 
67 #include "sd.h"
68 #include "scsi_priv.h"
69 #include "scsi_logging.h"
70 
71 MODULE_AUTHOR("Eric Youngdale");
72 MODULE_DESCRIPTION("SCSI disk (sd) driver");
73 MODULE_LICENSE("GPL");
74 
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
94 
95 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
96 #define SD_MINORS	16
97 #else
98 #define SD_MINORS	0
99 #endif
100 
101 static void sd_config_discard(struct scsi_disk *, unsigned int);
102 static void sd_config_write_same(struct scsi_disk *);
103 static int  sd_revalidate_disk(struct gendisk *);
104 static void sd_unlock_native_capacity(struct gendisk *disk);
105 static int  sd_probe(struct device *);
106 static int  sd_remove(struct device *);
107 static void sd_shutdown(struct device *);
108 static int sd_suspend_system(struct device *);
109 static int sd_suspend_runtime(struct device *);
110 static int sd_resume(struct device *);
111 static void sd_rescan(struct device *);
112 static int sd_done(struct scsi_cmnd *);
113 static int sd_eh_action(struct scsi_cmnd *, int);
114 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
115 static void scsi_disk_release(struct device *cdev);
116 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
117 static void sd_print_result(struct scsi_disk *, int);
118 
119 static DEFINE_SPINLOCK(sd_index_lock);
120 static DEFINE_IDA(sd_index_ida);
121 
122 /* This semaphore is used to mediate the 0->1 reference get in the
123  * face of object destruction (i.e. we can't allow a get on an
124  * object after last put) */
125 static DEFINE_MUTEX(sd_ref_mutex);
126 
127 static struct kmem_cache *sd_cdb_cache;
128 static mempool_t *sd_cdb_pool;
129 
130 static const char *sd_cache_types[] = {
131 	"write through", "none", "write back",
132 	"write back, no read (daft)"
133 };
134 
135 static ssize_t
136 cache_type_store(struct device *dev, struct device_attribute *attr,
137 		 const char *buf, size_t count)
138 {
139 	int i, ct = -1, rcd, wce, sp;
140 	struct scsi_disk *sdkp = to_scsi_disk(dev);
141 	struct scsi_device *sdp = sdkp->device;
142 	char buffer[64];
143 	char *buffer_data;
144 	struct scsi_mode_data data;
145 	struct scsi_sense_hdr sshdr;
146 	static const char temp[] = "temporary ";
147 	int len;
148 
149 	if (sdp->type != TYPE_DISK)
150 		/* no cache control on RBC devices; theoretically they
151 		 * can do it, but there's probably so many exceptions
152 		 * it's not worth the risk */
153 		return -EINVAL;
154 
155 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
156 		buf += sizeof(temp) - 1;
157 		sdkp->cache_override = 1;
158 	} else {
159 		sdkp->cache_override = 0;
160 	}
161 
162 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
163 		len = strlen(sd_cache_types[i]);
164 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
165 		    buf[len] == '\n') {
166 			ct = i;
167 			break;
168 		}
169 	}
170 	if (ct < 0)
171 		return -EINVAL;
172 	rcd = ct & 0x01 ? 1 : 0;
173 	wce = ct & 0x02 ? 1 : 0;
174 
175 	if (sdkp->cache_override) {
176 		sdkp->WCE = wce;
177 		sdkp->RCD = rcd;
178 		return count;
179 	}
180 
181 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
182 			    SD_MAX_RETRIES, &data, NULL))
183 		return -EINVAL;
184 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
185 		  data.block_descriptor_length);
186 	buffer_data = buffer + data.header_length +
187 		data.block_descriptor_length;
188 	buffer_data[2] &= ~0x05;
189 	buffer_data[2] |= wce << 2 | rcd;
190 	sp = buffer_data[0] & 0x80 ? 1 : 0;
191 
192 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
193 			     SD_MAX_RETRIES, &data, &sshdr)) {
194 		if (scsi_sense_valid(&sshdr))
195 			sd_print_sense_hdr(sdkp, &sshdr);
196 		return -EINVAL;
197 	}
198 	revalidate_disk(sdkp->disk);
199 	return count;
200 }
201 
202 static ssize_t
203 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
204 		       char *buf)
205 {
206 	struct scsi_disk *sdkp = to_scsi_disk(dev);
207 	struct scsi_device *sdp = sdkp->device;
208 
209 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
210 }
211 
212 static ssize_t
213 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
214 			const char *buf, size_t count)
215 {
216 	struct scsi_disk *sdkp = to_scsi_disk(dev);
217 	struct scsi_device *sdp = sdkp->device;
218 
219 	if (!capable(CAP_SYS_ADMIN))
220 		return -EACCES;
221 
222 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
223 
224 	return count;
225 }
226 static DEVICE_ATTR_RW(manage_start_stop);
227 
228 static ssize_t
229 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
230 {
231 	struct scsi_disk *sdkp = to_scsi_disk(dev);
232 
233 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
234 }
235 
236 static ssize_t
237 allow_restart_store(struct device *dev, struct device_attribute *attr,
238 		    const char *buf, size_t count)
239 {
240 	struct scsi_disk *sdkp = to_scsi_disk(dev);
241 	struct scsi_device *sdp = sdkp->device;
242 
243 	if (!capable(CAP_SYS_ADMIN))
244 		return -EACCES;
245 
246 	if (sdp->type != TYPE_DISK)
247 		return -EINVAL;
248 
249 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
250 
251 	return count;
252 }
253 static DEVICE_ATTR_RW(allow_restart);
254 
255 static ssize_t
256 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
257 {
258 	struct scsi_disk *sdkp = to_scsi_disk(dev);
259 	int ct = sdkp->RCD + 2*sdkp->WCE;
260 
261 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
262 }
263 static DEVICE_ATTR_RW(cache_type);
264 
265 static ssize_t
266 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
267 {
268 	struct scsi_disk *sdkp = to_scsi_disk(dev);
269 
270 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
271 }
272 static DEVICE_ATTR_RO(FUA);
273 
274 static ssize_t
275 protection_type_show(struct device *dev, struct device_attribute *attr,
276 		     char *buf)
277 {
278 	struct scsi_disk *sdkp = to_scsi_disk(dev);
279 
280 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
281 }
282 
283 static ssize_t
284 protection_type_store(struct device *dev, struct device_attribute *attr,
285 		      const char *buf, size_t count)
286 {
287 	struct scsi_disk *sdkp = to_scsi_disk(dev);
288 	unsigned int val;
289 	int err;
290 
291 	if (!capable(CAP_SYS_ADMIN))
292 		return -EACCES;
293 
294 	err = kstrtouint(buf, 10, &val);
295 
296 	if (err)
297 		return err;
298 
299 	if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
300 		sdkp->protection_type = val;
301 
302 	return count;
303 }
304 static DEVICE_ATTR_RW(protection_type);
305 
306 static ssize_t
307 protection_mode_show(struct device *dev, struct device_attribute *attr,
308 		     char *buf)
309 {
310 	struct scsi_disk *sdkp = to_scsi_disk(dev);
311 	struct scsi_device *sdp = sdkp->device;
312 	unsigned int dif, dix;
313 
314 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
315 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
316 
317 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
318 		dif = 0;
319 		dix = 1;
320 	}
321 
322 	if (!dif && !dix)
323 		return snprintf(buf, 20, "none\n");
324 
325 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
326 }
327 static DEVICE_ATTR_RO(protection_mode);
328 
329 static ssize_t
330 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
331 {
332 	struct scsi_disk *sdkp = to_scsi_disk(dev);
333 
334 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
335 }
336 static DEVICE_ATTR_RO(app_tag_own);
337 
338 static ssize_t
339 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
340 		       char *buf)
341 {
342 	struct scsi_disk *sdkp = to_scsi_disk(dev);
343 
344 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
345 }
346 static DEVICE_ATTR_RO(thin_provisioning);
347 
348 static const char *lbp_mode[] = {
349 	[SD_LBP_FULL]		= "full",
350 	[SD_LBP_UNMAP]		= "unmap",
351 	[SD_LBP_WS16]		= "writesame_16",
352 	[SD_LBP_WS10]		= "writesame_10",
353 	[SD_LBP_ZERO]		= "writesame_zero",
354 	[SD_LBP_DISABLE]	= "disabled",
355 };
356 
357 static ssize_t
358 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
359 		       char *buf)
360 {
361 	struct scsi_disk *sdkp = to_scsi_disk(dev);
362 
363 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
364 }
365 
366 static ssize_t
367 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
368 			const char *buf, size_t count)
369 {
370 	struct scsi_disk *sdkp = to_scsi_disk(dev);
371 	struct scsi_device *sdp = sdkp->device;
372 
373 	if (!capable(CAP_SYS_ADMIN))
374 		return -EACCES;
375 
376 	if (sdp->type != TYPE_DISK)
377 		return -EINVAL;
378 
379 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
380 		sd_config_discard(sdkp, SD_LBP_UNMAP);
381 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
382 		sd_config_discard(sdkp, SD_LBP_WS16);
383 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
384 		sd_config_discard(sdkp, SD_LBP_WS10);
385 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
386 		sd_config_discard(sdkp, SD_LBP_ZERO);
387 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
388 		sd_config_discard(sdkp, SD_LBP_DISABLE);
389 	else
390 		return -EINVAL;
391 
392 	return count;
393 }
394 static DEVICE_ATTR_RW(provisioning_mode);
395 
396 static ssize_t
397 max_medium_access_timeouts_show(struct device *dev,
398 				struct device_attribute *attr, char *buf)
399 {
400 	struct scsi_disk *sdkp = to_scsi_disk(dev);
401 
402 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
403 }
404 
405 static ssize_t
406 max_medium_access_timeouts_store(struct device *dev,
407 				 struct device_attribute *attr, const char *buf,
408 				 size_t count)
409 {
410 	struct scsi_disk *sdkp = to_scsi_disk(dev);
411 	int err;
412 
413 	if (!capable(CAP_SYS_ADMIN))
414 		return -EACCES;
415 
416 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
417 
418 	return err ? err : count;
419 }
420 static DEVICE_ATTR_RW(max_medium_access_timeouts);
421 
422 static ssize_t
423 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
424 			   char *buf)
425 {
426 	struct scsi_disk *sdkp = to_scsi_disk(dev);
427 
428 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
429 }
430 
431 static ssize_t
432 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
433 			    const char *buf, size_t count)
434 {
435 	struct scsi_disk *sdkp = to_scsi_disk(dev);
436 	struct scsi_device *sdp = sdkp->device;
437 	unsigned long max;
438 	int err;
439 
440 	if (!capable(CAP_SYS_ADMIN))
441 		return -EACCES;
442 
443 	if (sdp->type != TYPE_DISK)
444 		return -EINVAL;
445 
446 	err = kstrtoul(buf, 10, &max);
447 
448 	if (err)
449 		return err;
450 
451 	if (max == 0)
452 		sdp->no_write_same = 1;
453 	else if (max <= SD_MAX_WS16_BLOCKS) {
454 		sdp->no_write_same = 0;
455 		sdkp->max_ws_blocks = max;
456 	}
457 
458 	sd_config_write_same(sdkp);
459 
460 	return count;
461 }
462 static DEVICE_ATTR_RW(max_write_same_blocks);
463 
464 static struct attribute *sd_disk_attrs[] = {
465 	&dev_attr_cache_type.attr,
466 	&dev_attr_FUA.attr,
467 	&dev_attr_allow_restart.attr,
468 	&dev_attr_manage_start_stop.attr,
469 	&dev_attr_protection_type.attr,
470 	&dev_attr_protection_mode.attr,
471 	&dev_attr_app_tag_own.attr,
472 	&dev_attr_thin_provisioning.attr,
473 	&dev_attr_provisioning_mode.attr,
474 	&dev_attr_max_write_same_blocks.attr,
475 	&dev_attr_max_medium_access_timeouts.attr,
476 	NULL,
477 };
478 ATTRIBUTE_GROUPS(sd_disk);
479 
480 static struct class sd_disk_class = {
481 	.name		= "scsi_disk",
482 	.owner		= THIS_MODULE,
483 	.dev_release	= scsi_disk_release,
484 	.dev_groups	= sd_disk_groups,
485 };
486 
487 static const struct dev_pm_ops sd_pm_ops = {
488 	.suspend		= sd_suspend_system,
489 	.resume			= sd_resume,
490 	.poweroff		= sd_suspend_system,
491 	.restore		= sd_resume,
492 	.runtime_suspend	= sd_suspend_runtime,
493 	.runtime_resume		= sd_resume,
494 };
495 
496 static struct scsi_driver sd_template = {
497 	.owner			= THIS_MODULE,
498 	.gendrv = {
499 		.name		= "sd",
500 		.probe		= sd_probe,
501 		.remove		= sd_remove,
502 		.shutdown	= sd_shutdown,
503 		.pm		= &sd_pm_ops,
504 	},
505 	.rescan			= sd_rescan,
506 	.done			= sd_done,
507 	.eh_action		= sd_eh_action,
508 };
509 
510 /*
511  * Dummy kobj_map->probe function.
512  * The default ->probe function will call modprobe, which is
513  * pointless as this module is already loaded.
514  */
515 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
516 {
517 	return NULL;
518 }
519 
520 /*
521  * Device no to disk mapping:
522  *
523  *       major         disc2     disc  p1
524  *   |............|.............|....|....| <- dev_t
525  *    31        20 19          8 7  4 3  0
526  *
527  * Inside a major, we have 16k disks, however mapped non-
528  * contiguously. The first 16 disks are for major0, the next
529  * ones with major1, ... Disk 256 is for major0 again, disk 272
530  * for major1, ...
531  * As we stay compatible with our numbering scheme, we can reuse
532  * the well-know SCSI majors 8, 65--71, 136--143.
533  */
534 static int sd_major(int major_idx)
535 {
536 	switch (major_idx) {
537 	case 0:
538 		return SCSI_DISK0_MAJOR;
539 	case 1 ... 7:
540 		return SCSI_DISK1_MAJOR + major_idx - 1;
541 	case 8 ... 15:
542 		return SCSI_DISK8_MAJOR + major_idx - 8;
543 	default:
544 		BUG();
545 		return 0;	/* shut up gcc */
546 	}
547 }
548 
549 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
550 {
551 	struct scsi_disk *sdkp = NULL;
552 
553 	if (disk->private_data) {
554 		sdkp = scsi_disk(disk);
555 		if (scsi_device_get(sdkp->device) == 0)
556 			get_device(&sdkp->dev);
557 		else
558 			sdkp = NULL;
559 	}
560 	return sdkp;
561 }
562 
563 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
564 {
565 	struct scsi_disk *sdkp;
566 
567 	mutex_lock(&sd_ref_mutex);
568 	sdkp = __scsi_disk_get(disk);
569 	mutex_unlock(&sd_ref_mutex);
570 	return sdkp;
571 }
572 
573 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
574 {
575 	struct scsi_disk *sdkp;
576 
577 	mutex_lock(&sd_ref_mutex);
578 	sdkp = dev_get_drvdata(dev);
579 	if (sdkp)
580 		sdkp = __scsi_disk_get(sdkp->disk);
581 	mutex_unlock(&sd_ref_mutex);
582 	return sdkp;
583 }
584 
585 static void scsi_disk_put(struct scsi_disk *sdkp)
586 {
587 	struct scsi_device *sdev = sdkp->device;
588 
589 	mutex_lock(&sd_ref_mutex);
590 	put_device(&sdkp->dev);
591 	scsi_device_put(sdev);
592 	mutex_unlock(&sd_ref_mutex);
593 }
594 
595 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
596 {
597 	unsigned int prot_op = SCSI_PROT_NORMAL;
598 	unsigned int dix = scsi_prot_sg_count(scmd);
599 
600 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
601 		if (dif && dix)
602 			prot_op = SCSI_PROT_READ_PASS;
603 		else if (dif && !dix)
604 			prot_op = SCSI_PROT_READ_STRIP;
605 		else if (!dif && dix)
606 			prot_op = SCSI_PROT_READ_INSERT;
607 	} else {
608 		if (dif && dix)
609 			prot_op = SCSI_PROT_WRITE_PASS;
610 		else if (dif && !dix)
611 			prot_op = SCSI_PROT_WRITE_INSERT;
612 		else if (!dif && dix)
613 			prot_op = SCSI_PROT_WRITE_STRIP;
614 	}
615 
616 	scsi_set_prot_op(scmd, prot_op);
617 	scsi_set_prot_type(scmd, dif);
618 }
619 
620 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
621 {
622 	struct request_queue *q = sdkp->disk->queue;
623 	unsigned int logical_block_size = sdkp->device->sector_size;
624 	unsigned int max_blocks = 0;
625 
626 	q->limits.discard_zeroes_data = sdkp->lbprz;
627 	q->limits.discard_alignment = sdkp->unmap_alignment *
628 		logical_block_size;
629 	q->limits.discard_granularity =
630 		max(sdkp->physical_block_size,
631 		    sdkp->unmap_granularity * logical_block_size);
632 
633 	sdkp->provisioning_mode = mode;
634 
635 	switch (mode) {
636 
637 	case SD_LBP_DISABLE:
638 		q->limits.max_discard_sectors = 0;
639 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
640 		return;
641 
642 	case SD_LBP_UNMAP:
643 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
644 					  (u32)SD_MAX_WS16_BLOCKS);
645 		break;
646 
647 	case SD_LBP_WS16:
648 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
649 					  (u32)SD_MAX_WS16_BLOCKS);
650 		break;
651 
652 	case SD_LBP_WS10:
653 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
654 					  (u32)SD_MAX_WS10_BLOCKS);
655 		break;
656 
657 	case SD_LBP_ZERO:
658 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
659 					  (u32)SD_MAX_WS10_BLOCKS);
660 		q->limits.discard_zeroes_data = 1;
661 		break;
662 	}
663 
664 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
665 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
666 }
667 
668 /**
669  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
670  * @sdp: scsi device to operate one
671  * @rq: Request to prepare
672  *
673  * Will issue either UNMAP or WRITE SAME(16) depending on preference
674  * indicated by target device.
675  **/
676 static int sd_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
677 {
678 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
679 	sector_t sector = blk_rq_pos(rq);
680 	unsigned int nr_sectors = blk_rq_sectors(rq);
681 	unsigned int nr_bytes = blk_rq_bytes(rq);
682 	unsigned int len;
683 	int ret;
684 	char *buf;
685 	struct page *page;
686 
687 	sector >>= ilog2(sdp->sector_size) - 9;
688 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
689 	rq->timeout = SD_TIMEOUT;
690 
691 	memset(rq->cmd, 0, rq->cmd_len);
692 
693 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
694 	if (!page)
695 		return BLKPREP_DEFER;
696 
697 	switch (sdkp->provisioning_mode) {
698 	case SD_LBP_UNMAP:
699 		buf = page_address(page);
700 
701 		rq->cmd_len = 10;
702 		rq->cmd[0] = UNMAP;
703 		rq->cmd[8] = 24;
704 
705 		put_unaligned_be16(6 + 16, &buf[0]);
706 		put_unaligned_be16(16, &buf[2]);
707 		put_unaligned_be64(sector, &buf[8]);
708 		put_unaligned_be32(nr_sectors, &buf[16]);
709 
710 		len = 24;
711 		break;
712 
713 	case SD_LBP_WS16:
714 		rq->cmd_len = 16;
715 		rq->cmd[0] = WRITE_SAME_16;
716 		rq->cmd[1] = 0x8; /* UNMAP */
717 		put_unaligned_be64(sector, &rq->cmd[2]);
718 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
719 
720 		len = sdkp->device->sector_size;
721 		break;
722 
723 	case SD_LBP_WS10:
724 	case SD_LBP_ZERO:
725 		rq->cmd_len = 10;
726 		rq->cmd[0] = WRITE_SAME;
727 		if (sdkp->provisioning_mode == SD_LBP_WS10)
728 			rq->cmd[1] = 0x8; /* UNMAP */
729 		put_unaligned_be32(sector, &rq->cmd[2]);
730 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
731 
732 		len = sdkp->device->sector_size;
733 		break;
734 
735 	default:
736 		ret = BLKPREP_KILL;
737 		goto out;
738 	}
739 
740 	blk_add_request_payload(rq, page, len);
741 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
742 	rq->buffer = page_address(page);
743 	rq->__data_len = nr_bytes;
744 
745 out:
746 	if (ret != BLKPREP_OK) {
747 		__free_page(page);
748 		rq->buffer = NULL;
749 	}
750 	return ret;
751 }
752 
753 static void sd_config_write_same(struct scsi_disk *sdkp)
754 {
755 	struct request_queue *q = sdkp->disk->queue;
756 	unsigned int logical_block_size = sdkp->device->sector_size;
757 
758 	if (sdkp->device->no_write_same) {
759 		sdkp->max_ws_blocks = 0;
760 		goto out;
761 	}
762 
763 	/* Some devices can not handle block counts above 0xffff despite
764 	 * supporting WRITE SAME(16). Consequently we default to 64k
765 	 * blocks per I/O unless the device explicitly advertises a
766 	 * bigger limit.
767 	 */
768 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
769 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
770 						   (u32)SD_MAX_WS16_BLOCKS);
771 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
772 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
773 						   (u32)SD_MAX_WS10_BLOCKS);
774 	else {
775 		sdkp->device->no_write_same = 1;
776 		sdkp->max_ws_blocks = 0;
777 	}
778 
779 out:
780 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
781 					 (logical_block_size >> 9));
782 }
783 
784 /**
785  * sd_setup_write_same_cmnd - write the same data to multiple blocks
786  * @sdp: scsi device to operate one
787  * @rq: Request to prepare
788  *
789  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
790  * preference indicated by target device.
791  **/
792 static int sd_setup_write_same_cmnd(struct scsi_device *sdp, struct request *rq)
793 {
794 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
795 	struct bio *bio = rq->bio;
796 	sector_t sector = blk_rq_pos(rq);
797 	unsigned int nr_sectors = blk_rq_sectors(rq);
798 	unsigned int nr_bytes = blk_rq_bytes(rq);
799 	int ret;
800 
801 	if (sdkp->device->no_write_same)
802 		return BLKPREP_KILL;
803 
804 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
805 
806 	sector >>= ilog2(sdp->sector_size) - 9;
807 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
808 
809 	rq->__data_len = sdp->sector_size;
810 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
811 	memset(rq->cmd, 0, rq->cmd_len);
812 
813 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
814 		rq->cmd_len = 16;
815 		rq->cmd[0] = WRITE_SAME_16;
816 		put_unaligned_be64(sector, &rq->cmd[2]);
817 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
818 	} else {
819 		rq->cmd_len = 10;
820 		rq->cmd[0] = WRITE_SAME;
821 		put_unaligned_be32(sector, &rq->cmd[2]);
822 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
823 	}
824 
825 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
826 	rq->__data_len = nr_bytes;
827 
828 	return ret;
829 }
830 
831 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
832 {
833 	rq->timeout *= SD_FLUSH_TIMEOUT_MULTIPLIER;
834 	rq->retries = SD_MAX_RETRIES;
835 	rq->cmd[0] = SYNCHRONIZE_CACHE;
836 	rq->cmd_len = 10;
837 
838 	return scsi_setup_blk_pc_cmnd(sdp, rq);
839 }
840 
841 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
842 {
843 	struct scsi_cmnd *SCpnt = rq->special;
844 
845 	if (rq->cmd_flags & REQ_DISCARD) {
846 		free_page((unsigned long)rq->buffer);
847 		rq->buffer = NULL;
848 	}
849 	if (SCpnt->cmnd != rq->cmd) {
850 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
851 		SCpnt->cmnd = NULL;
852 		SCpnt->cmd_len = 0;
853 	}
854 }
855 
856 /**
857  *	sd_prep_fn - build a scsi (read or write) command from
858  *	information in the request structure.
859  *	@SCpnt: pointer to mid-level's per scsi command structure that
860  *	contains request and into which the scsi command is written
861  *
862  *	Returns 1 if successful and 0 if error (or cannot be done now).
863  **/
864 static int sd_prep_fn(struct request_queue *q, struct request *rq)
865 {
866 	struct scsi_cmnd *SCpnt;
867 	struct scsi_device *sdp = q->queuedata;
868 	struct gendisk *disk = rq->rq_disk;
869 	struct scsi_disk *sdkp;
870 	sector_t block = blk_rq_pos(rq);
871 	sector_t threshold;
872 	unsigned int this_count = blk_rq_sectors(rq);
873 	int ret, host_dif;
874 	unsigned char protect;
875 
876 	/*
877 	 * Discard request come in as REQ_TYPE_FS but we turn them into
878 	 * block PC requests to make life easier.
879 	 */
880 	if (rq->cmd_flags & REQ_DISCARD) {
881 		ret = sd_setup_discard_cmnd(sdp, rq);
882 		goto out;
883 	} else if (rq->cmd_flags & REQ_WRITE_SAME) {
884 		ret = sd_setup_write_same_cmnd(sdp, rq);
885 		goto out;
886 	} else if (rq->cmd_flags & REQ_FLUSH) {
887 		ret = scsi_setup_flush_cmnd(sdp, rq);
888 		goto out;
889 	} else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
890 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
891 		goto out;
892 	} else if (rq->cmd_type != REQ_TYPE_FS) {
893 		ret = BLKPREP_KILL;
894 		goto out;
895 	}
896 	ret = scsi_setup_fs_cmnd(sdp, rq);
897 	if (ret != BLKPREP_OK)
898 		goto out;
899 	SCpnt = rq->special;
900 	sdkp = scsi_disk(disk);
901 
902 	/* from here on until we're complete, any goto out
903 	 * is used for a killable error condition */
904 	ret = BLKPREP_KILL;
905 
906 	SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
907 					"sd_prep_fn: block=%llu, "
908 					"count=%d\n",
909 					(unsigned long long)block,
910 					this_count));
911 
912 	if (!sdp || !scsi_device_online(sdp) ||
913 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
914 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
915 						"Finishing %u sectors\n",
916 						blk_rq_sectors(rq)));
917 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
918 						"Retry with 0x%p\n", SCpnt));
919 		goto out;
920 	}
921 
922 	if (sdp->changed) {
923 		/*
924 		 * quietly refuse to do anything to a changed disc until
925 		 * the changed bit has been reset
926 		 */
927 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
928 		goto out;
929 	}
930 
931 	/*
932 	 * Some SD card readers can't handle multi-sector accesses which touch
933 	 * the last one or two hardware sectors.  Split accesses as needed.
934 	 */
935 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
936 		(sdp->sector_size / 512);
937 
938 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
939 		if (block < threshold) {
940 			/* Access up to the threshold but not beyond */
941 			this_count = threshold - block;
942 		} else {
943 			/* Access only a single hardware sector */
944 			this_count = sdp->sector_size / 512;
945 		}
946 	}
947 
948 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
949 					(unsigned long long)block));
950 
951 	/*
952 	 * If we have a 1K hardware sectorsize, prevent access to single
953 	 * 512 byte sectors.  In theory we could handle this - in fact
954 	 * the scsi cdrom driver must be able to handle this because
955 	 * we typically use 1K blocksizes, and cdroms typically have
956 	 * 2K hardware sectorsizes.  Of course, things are simpler
957 	 * with the cdrom, since it is read-only.  For performance
958 	 * reasons, the filesystems should be able to handle this
959 	 * and not force the scsi disk driver to use bounce buffers
960 	 * for this.
961 	 */
962 	if (sdp->sector_size == 1024) {
963 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
964 			scmd_printk(KERN_ERR, SCpnt,
965 				    "Bad block number requested\n");
966 			goto out;
967 		} else {
968 			block = block >> 1;
969 			this_count = this_count >> 1;
970 		}
971 	}
972 	if (sdp->sector_size == 2048) {
973 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
974 			scmd_printk(KERN_ERR, SCpnt,
975 				    "Bad block number requested\n");
976 			goto out;
977 		} else {
978 			block = block >> 2;
979 			this_count = this_count >> 2;
980 		}
981 	}
982 	if (sdp->sector_size == 4096) {
983 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
984 			scmd_printk(KERN_ERR, SCpnt,
985 				    "Bad block number requested\n");
986 			goto out;
987 		} else {
988 			block = block >> 3;
989 			this_count = this_count >> 3;
990 		}
991 	}
992 	if (rq_data_dir(rq) == WRITE) {
993 		if (!sdp->writeable) {
994 			goto out;
995 		}
996 		SCpnt->cmnd[0] = WRITE_6;
997 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
998 
999 		if (blk_integrity_rq(rq))
1000 			sd_dif_prepare(rq, block, sdp->sector_size);
1001 
1002 	} else if (rq_data_dir(rq) == READ) {
1003 		SCpnt->cmnd[0] = READ_6;
1004 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
1005 	} else {
1006 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %llx\n", (unsigned long long) rq->cmd_flags);
1007 		goto out;
1008 	}
1009 
1010 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1011 					"%s %d/%u 512 byte blocks.\n",
1012 					(rq_data_dir(rq) == WRITE) ?
1013 					"writing" : "reading", this_count,
1014 					blk_rq_sectors(rq)));
1015 
1016 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
1017 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
1018 	if (host_dif)
1019 		protect = 1 << 5;
1020 	else
1021 		protect = 0;
1022 
1023 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
1024 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1025 
1026 		if (unlikely(SCpnt->cmnd == NULL)) {
1027 			ret = BLKPREP_DEFER;
1028 			goto out;
1029 		}
1030 
1031 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1032 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1033 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1034 		SCpnt->cmnd[7] = 0x18;
1035 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1036 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1037 
1038 		/* LBA */
1039 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1040 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1041 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1042 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1043 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1044 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1045 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1046 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1047 
1048 		/* Expected Indirect LBA */
1049 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1050 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1051 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1052 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1053 
1054 		/* Transfer length */
1055 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1056 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1057 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1058 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1059 	} else if (sdp->use_16_for_rw) {
1060 		SCpnt->cmnd[0] += READ_16 - READ_6;
1061 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1062 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1063 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1064 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1065 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1066 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1067 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1068 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1069 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1070 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1071 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1072 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1073 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1074 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1075 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1076 		   scsi_device_protection(SCpnt->device) ||
1077 		   SCpnt->device->use_10_for_rw) {
1078 		if (this_count > 0xffff)
1079 			this_count = 0xffff;
1080 
1081 		SCpnt->cmnd[0] += READ_10 - READ_6;
1082 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1083 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1084 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1085 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1086 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1087 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1088 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1089 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1090 	} else {
1091 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1092 			/*
1093 			 * This happens only if this drive failed
1094 			 * 10byte rw command with ILLEGAL_REQUEST
1095 			 * during operation and thus turned off
1096 			 * use_10_for_rw.
1097 			 */
1098 			scmd_printk(KERN_ERR, SCpnt,
1099 				    "FUA write on READ/WRITE(6) drive\n");
1100 			goto out;
1101 		}
1102 
1103 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1104 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1105 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1106 		SCpnt->cmnd[4] = (unsigned char) this_count;
1107 		SCpnt->cmnd[5] = 0;
1108 	}
1109 	SCpnt->sdb.length = this_count * sdp->sector_size;
1110 
1111 	/* If DIF or DIX is enabled, tell HBA how to handle request */
1112 	if (host_dif || scsi_prot_sg_count(SCpnt))
1113 		sd_prot_op(SCpnt, host_dif);
1114 
1115 	/*
1116 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1117 	 * host adapter, it's safe to assume that we can at least transfer
1118 	 * this many bytes between each connect / disconnect.
1119 	 */
1120 	SCpnt->transfersize = sdp->sector_size;
1121 	SCpnt->underflow = this_count << 9;
1122 	SCpnt->allowed = SD_MAX_RETRIES;
1123 
1124 	/*
1125 	 * This indicates that the command is ready from our end to be
1126 	 * queued.
1127 	 */
1128 	ret = BLKPREP_OK;
1129  out:
1130 	return scsi_prep_return(q, rq, ret);
1131 }
1132 
1133 /**
1134  *	sd_open - open a scsi disk device
1135  *	@inode: only i_rdev member may be used
1136  *	@filp: only f_mode and f_flags may be used
1137  *
1138  *	Returns 0 if successful. Returns a negated errno value in case
1139  *	of error.
1140  *
1141  *	Note: This can be called from a user context (e.g. fsck(1) )
1142  *	or from within the kernel (e.g. as a result of a mount(1) ).
1143  *	In the latter case @inode and @filp carry an abridged amount
1144  *	of information as noted above.
1145  *
1146  *	Locking: called with bdev->bd_mutex held.
1147  **/
1148 static int sd_open(struct block_device *bdev, fmode_t mode)
1149 {
1150 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1151 	struct scsi_device *sdev;
1152 	int retval;
1153 
1154 	if (!sdkp)
1155 		return -ENXIO;
1156 
1157 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1158 
1159 	sdev = sdkp->device;
1160 
1161 	/*
1162 	 * If the device is in error recovery, wait until it is done.
1163 	 * If the device is offline, then disallow any access to it.
1164 	 */
1165 	retval = -ENXIO;
1166 	if (!scsi_block_when_processing_errors(sdev))
1167 		goto error_out;
1168 
1169 	if (sdev->removable || sdkp->write_prot)
1170 		check_disk_change(bdev);
1171 
1172 	/*
1173 	 * If the drive is empty, just let the open fail.
1174 	 */
1175 	retval = -ENOMEDIUM;
1176 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1177 		goto error_out;
1178 
1179 	/*
1180 	 * If the device has the write protect tab set, have the open fail
1181 	 * if the user expects to be able to write to the thing.
1182 	 */
1183 	retval = -EROFS;
1184 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1185 		goto error_out;
1186 
1187 	/*
1188 	 * It is possible that the disk changing stuff resulted in
1189 	 * the device being taken offline.  If this is the case,
1190 	 * report this to the user, and don't pretend that the
1191 	 * open actually succeeded.
1192 	 */
1193 	retval = -ENXIO;
1194 	if (!scsi_device_online(sdev))
1195 		goto error_out;
1196 
1197 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1198 		if (scsi_block_when_processing_errors(sdev))
1199 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1200 	}
1201 
1202 	return 0;
1203 
1204 error_out:
1205 	scsi_disk_put(sdkp);
1206 	return retval;
1207 }
1208 
1209 /**
1210  *	sd_release - invoked when the (last) close(2) is called on this
1211  *	scsi disk.
1212  *	@inode: only i_rdev member may be used
1213  *	@filp: only f_mode and f_flags may be used
1214  *
1215  *	Returns 0.
1216  *
1217  *	Note: may block (uninterruptible) if error recovery is underway
1218  *	on this disk.
1219  *
1220  *	Locking: called with bdev->bd_mutex held.
1221  **/
1222 static void sd_release(struct gendisk *disk, fmode_t mode)
1223 {
1224 	struct scsi_disk *sdkp = scsi_disk(disk);
1225 	struct scsi_device *sdev = sdkp->device;
1226 
1227 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1228 
1229 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1230 		if (scsi_block_when_processing_errors(sdev))
1231 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1232 	}
1233 
1234 	/*
1235 	 * XXX and what if there are packets in flight and this close()
1236 	 * XXX is followed by a "rmmod sd_mod"?
1237 	 */
1238 
1239 	scsi_disk_put(sdkp);
1240 }
1241 
1242 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1243 {
1244 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1245 	struct scsi_device *sdp = sdkp->device;
1246 	struct Scsi_Host *host = sdp->host;
1247 	int diskinfo[4];
1248 
1249 	/* default to most commonly used values */
1250         diskinfo[0] = 0x40;	/* 1 << 6 */
1251        	diskinfo[1] = 0x20;	/* 1 << 5 */
1252        	diskinfo[2] = sdkp->capacity >> 11;
1253 
1254 	/* override with calculated, extended default, or driver values */
1255 	if (host->hostt->bios_param)
1256 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1257 	else
1258 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1259 
1260 	geo->heads = diskinfo[0];
1261 	geo->sectors = diskinfo[1];
1262 	geo->cylinders = diskinfo[2];
1263 	return 0;
1264 }
1265 
1266 /**
1267  *	sd_ioctl - process an ioctl
1268  *	@inode: only i_rdev/i_bdev members may be used
1269  *	@filp: only f_mode and f_flags may be used
1270  *	@cmd: ioctl command number
1271  *	@arg: this is third argument given to ioctl(2) system call.
1272  *	Often contains a pointer.
1273  *
1274  *	Returns 0 if successful (some ioctls return positive numbers on
1275  *	success as well). Returns a negated errno value in case of error.
1276  *
1277  *	Note: most ioctls are forward onto the block subsystem or further
1278  *	down in the scsi subsystem.
1279  **/
1280 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1281 		    unsigned int cmd, unsigned long arg)
1282 {
1283 	struct gendisk *disk = bdev->bd_disk;
1284 	struct scsi_disk *sdkp = scsi_disk(disk);
1285 	struct scsi_device *sdp = sdkp->device;
1286 	void __user *p = (void __user *)arg;
1287 	int error;
1288 
1289 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1290 				    "cmd=0x%x\n", disk->disk_name, cmd));
1291 
1292 	error = scsi_verify_blk_ioctl(bdev, cmd);
1293 	if (error < 0)
1294 		return error;
1295 
1296 	/*
1297 	 * If we are in the middle of error recovery, don't let anyone
1298 	 * else try and use this device.  Also, if error recovery fails, it
1299 	 * may try and take the device offline, in which case all further
1300 	 * access to the device is prohibited.
1301 	 */
1302 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1303 					(mode & FMODE_NDELAY) != 0);
1304 	if (!scsi_block_when_processing_errors(sdp) || !error)
1305 		goto out;
1306 
1307 	/*
1308 	 * Send SCSI addressing ioctls directly to mid level, send other
1309 	 * ioctls to block level and then onto mid level if they can't be
1310 	 * resolved.
1311 	 */
1312 	switch (cmd) {
1313 		case SCSI_IOCTL_GET_IDLUN:
1314 		case SCSI_IOCTL_GET_BUS_NUMBER:
1315 			error = scsi_ioctl(sdp, cmd, p);
1316 			break;
1317 		default:
1318 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1319 			if (error != -ENOTTY)
1320 				break;
1321 			error = scsi_ioctl(sdp, cmd, p);
1322 			break;
1323 	}
1324 out:
1325 	return error;
1326 }
1327 
1328 static void set_media_not_present(struct scsi_disk *sdkp)
1329 {
1330 	if (sdkp->media_present)
1331 		sdkp->device->changed = 1;
1332 
1333 	if (sdkp->device->removable) {
1334 		sdkp->media_present = 0;
1335 		sdkp->capacity = 0;
1336 	}
1337 }
1338 
1339 static int media_not_present(struct scsi_disk *sdkp,
1340 			     struct scsi_sense_hdr *sshdr)
1341 {
1342 	if (!scsi_sense_valid(sshdr))
1343 		return 0;
1344 
1345 	/* not invoked for commands that could return deferred errors */
1346 	switch (sshdr->sense_key) {
1347 	case UNIT_ATTENTION:
1348 	case NOT_READY:
1349 		/* medium not present */
1350 		if (sshdr->asc == 0x3A) {
1351 			set_media_not_present(sdkp);
1352 			return 1;
1353 		}
1354 	}
1355 	return 0;
1356 }
1357 
1358 /**
1359  *	sd_check_events - check media events
1360  *	@disk: kernel device descriptor
1361  *	@clearing: disk events currently being cleared
1362  *
1363  *	Returns mask of DISK_EVENT_*.
1364  *
1365  *	Note: this function is invoked from the block subsystem.
1366  **/
1367 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1368 {
1369 	struct scsi_disk *sdkp = scsi_disk(disk);
1370 	struct scsi_device *sdp = sdkp->device;
1371 	struct scsi_sense_hdr *sshdr = NULL;
1372 	int retval;
1373 
1374 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1375 
1376 	/*
1377 	 * If the device is offline, don't send any commands - just pretend as
1378 	 * if the command failed.  If the device ever comes back online, we
1379 	 * can deal with it then.  It is only because of unrecoverable errors
1380 	 * that we would ever take a device offline in the first place.
1381 	 */
1382 	if (!scsi_device_online(sdp)) {
1383 		set_media_not_present(sdkp);
1384 		goto out;
1385 	}
1386 
1387 	/*
1388 	 * Using TEST_UNIT_READY enables differentiation between drive with
1389 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1390 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1391 	 *
1392 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1393 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1394 	 * sd_revalidate() is called.
1395 	 */
1396 	retval = -ENODEV;
1397 
1398 	if (scsi_block_when_processing_errors(sdp)) {
1399 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1400 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1401 					      sshdr);
1402 	}
1403 
1404 	/* failed to execute TUR, assume media not present */
1405 	if (host_byte(retval)) {
1406 		set_media_not_present(sdkp);
1407 		goto out;
1408 	}
1409 
1410 	if (media_not_present(sdkp, sshdr))
1411 		goto out;
1412 
1413 	/*
1414 	 * For removable scsi disk we have to recognise the presence
1415 	 * of a disk in the drive.
1416 	 */
1417 	if (!sdkp->media_present)
1418 		sdp->changed = 1;
1419 	sdkp->media_present = 1;
1420 out:
1421 	/*
1422 	 * sdp->changed is set under the following conditions:
1423 	 *
1424 	 *	Medium present state has changed in either direction.
1425 	 *	Device has indicated UNIT_ATTENTION.
1426 	 */
1427 	kfree(sshdr);
1428 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1429 	sdp->changed = 0;
1430 	return retval;
1431 }
1432 
1433 static int sd_sync_cache(struct scsi_disk *sdkp)
1434 {
1435 	int retries, res;
1436 	struct scsi_device *sdp = sdkp->device;
1437 	const int timeout = sdp->request_queue->rq_timeout
1438 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1439 	struct scsi_sense_hdr sshdr;
1440 
1441 	if (!scsi_device_online(sdp))
1442 		return -ENODEV;
1443 
1444 	for (retries = 3; retries > 0; --retries) {
1445 		unsigned char cmd[10] = { 0 };
1446 
1447 		cmd[0] = SYNCHRONIZE_CACHE;
1448 		/*
1449 		 * Leave the rest of the command zero to indicate
1450 		 * flush everything.
1451 		 */
1452 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1453 					     &sshdr, timeout, SD_MAX_RETRIES,
1454 					     NULL, REQ_PM);
1455 		if (res == 0)
1456 			break;
1457 	}
1458 
1459 	if (res) {
1460 		sd_print_result(sdkp, res);
1461 
1462 		if (driver_byte(res) & DRIVER_SENSE)
1463 			sd_print_sense_hdr(sdkp, &sshdr);
1464 		/* we need to evaluate the error return  */
1465 		if (scsi_sense_valid(&sshdr) &&
1466 			(sshdr.asc == 0x3a ||	/* medium not present */
1467 			 sshdr.asc == 0x20))	/* invalid command */
1468 				/* this is no error here */
1469 				return 0;
1470 
1471 		switch (host_byte(res)) {
1472 		/* ignore errors due to racing a disconnection */
1473 		case DID_BAD_TARGET:
1474 		case DID_NO_CONNECT:
1475 			return 0;
1476 		/* signal the upper layer it might try again */
1477 		case DID_BUS_BUSY:
1478 		case DID_IMM_RETRY:
1479 		case DID_REQUEUE:
1480 		case DID_SOFT_ERROR:
1481 			return -EBUSY;
1482 		default:
1483 			return -EIO;
1484 		}
1485 	}
1486 	return 0;
1487 }
1488 
1489 static void sd_rescan(struct device *dev)
1490 {
1491 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1492 
1493 	if (sdkp) {
1494 		revalidate_disk(sdkp->disk);
1495 		scsi_disk_put(sdkp);
1496 	}
1497 }
1498 
1499 
1500 #ifdef CONFIG_COMPAT
1501 /*
1502  * This gets directly called from VFS. When the ioctl
1503  * is not recognized we go back to the other translation paths.
1504  */
1505 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1506 			   unsigned int cmd, unsigned long arg)
1507 {
1508 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1509 	int ret;
1510 
1511 	ret = scsi_verify_blk_ioctl(bdev, cmd);
1512 	if (ret < 0)
1513 		return ret;
1514 
1515 	/*
1516 	 * If we are in the middle of error recovery, don't let anyone
1517 	 * else try and use this device.  Also, if error recovery fails, it
1518 	 * may try and take the device offline, in which case all further
1519 	 * access to the device is prohibited.
1520 	 */
1521 	if (!scsi_block_when_processing_errors(sdev))
1522 		return -ENODEV;
1523 
1524 	if (sdev->host->hostt->compat_ioctl) {
1525 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1526 
1527 		return ret;
1528 	}
1529 
1530 	/*
1531 	 * Let the static ioctl translation table take care of it.
1532 	 */
1533 	return -ENOIOCTLCMD;
1534 }
1535 #endif
1536 
1537 static const struct block_device_operations sd_fops = {
1538 	.owner			= THIS_MODULE,
1539 	.open			= sd_open,
1540 	.release		= sd_release,
1541 	.ioctl			= sd_ioctl,
1542 	.getgeo			= sd_getgeo,
1543 #ifdef CONFIG_COMPAT
1544 	.compat_ioctl		= sd_compat_ioctl,
1545 #endif
1546 	.check_events		= sd_check_events,
1547 	.revalidate_disk	= sd_revalidate_disk,
1548 	.unlock_native_capacity	= sd_unlock_native_capacity,
1549 };
1550 
1551 /**
1552  *	sd_eh_action - error handling callback
1553  *	@scmd:		sd-issued command that has failed
1554  *	@eh_disp:	The recovery disposition suggested by the midlayer
1555  *
1556  *	This function is called by the SCSI midlayer upon completion of an
1557  *	error test command (currently TEST UNIT READY). The result of sending
1558  *	the eh command is passed in eh_disp.  We're looking for devices that
1559  *	fail medium access commands but are OK with non access commands like
1560  *	test unit ready (so wrongly see the device as having a successful
1561  *	recovery)
1562  **/
1563 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1564 {
1565 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1566 
1567 	if (!scsi_device_online(scmd->device) ||
1568 	    !scsi_medium_access_command(scmd) ||
1569 	    host_byte(scmd->result) != DID_TIME_OUT ||
1570 	    eh_disp != SUCCESS)
1571 		return eh_disp;
1572 
1573 	/*
1574 	 * The device has timed out executing a medium access command.
1575 	 * However, the TEST UNIT READY command sent during error
1576 	 * handling completed successfully. Either the device is in the
1577 	 * process of recovering or has it suffered an internal failure
1578 	 * that prevents access to the storage medium.
1579 	 */
1580 	sdkp->medium_access_timed_out++;
1581 
1582 	/*
1583 	 * If the device keeps failing read/write commands but TEST UNIT
1584 	 * READY always completes successfully we assume that medium
1585 	 * access is no longer possible and take the device offline.
1586 	 */
1587 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1588 		scmd_printk(KERN_ERR, scmd,
1589 			    "Medium access timeout failure. Offlining disk!\n");
1590 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1591 
1592 		return FAILED;
1593 	}
1594 
1595 	return eh_disp;
1596 }
1597 
1598 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1599 {
1600 	u64 start_lba = blk_rq_pos(scmd->request);
1601 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1602 	u64 bad_lba;
1603 	int info_valid;
1604 	/*
1605 	 * resid is optional but mostly filled in.  When it's unused,
1606 	 * its value is zero, so we assume the whole buffer transferred
1607 	 */
1608 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1609 	unsigned int good_bytes;
1610 
1611 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1612 		return 0;
1613 
1614 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1615 					     SCSI_SENSE_BUFFERSIZE,
1616 					     &bad_lba);
1617 	if (!info_valid)
1618 		return 0;
1619 
1620 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1621 		return 0;
1622 
1623 	if (scmd->device->sector_size < 512) {
1624 		/* only legitimate sector_size here is 256 */
1625 		start_lba <<= 1;
1626 		end_lba <<= 1;
1627 	} else {
1628 		/* be careful ... don't want any overflows */
1629 		unsigned int factor = scmd->device->sector_size / 512;
1630 		do_div(start_lba, factor);
1631 		do_div(end_lba, factor);
1632 	}
1633 
1634 	/* The bad lba was reported incorrectly, we have no idea where
1635 	 * the error is.
1636 	 */
1637 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1638 		return 0;
1639 
1640 	/* This computation should always be done in terms of
1641 	 * the resolution of the device's medium.
1642 	 */
1643 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1644 	return min(good_bytes, transferred);
1645 }
1646 
1647 /**
1648  *	sd_done - bottom half handler: called when the lower level
1649  *	driver has completed (successfully or otherwise) a scsi command.
1650  *	@SCpnt: mid-level's per command structure.
1651  *
1652  *	Note: potentially run from within an ISR. Must not block.
1653  **/
1654 static int sd_done(struct scsi_cmnd *SCpnt)
1655 {
1656 	int result = SCpnt->result;
1657 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1658 	struct scsi_sense_hdr sshdr;
1659 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1660 	struct request *req = SCpnt->request;
1661 	int sense_valid = 0;
1662 	int sense_deferred = 0;
1663 	unsigned char op = SCpnt->cmnd[0];
1664 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1665 
1666 	if (req->cmd_flags & REQ_DISCARD || req->cmd_flags & REQ_WRITE_SAME) {
1667 		if (!result) {
1668 			good_bytes = blk_rq_bytes(req);
1669 			scsi_set_resid(SCpnt, 0);
1670 		} else {
1671 			good_bytes = 0;
1672 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1673 		}
1674 	}
1675 
1676 	if (result) {
1677 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1678 		if (sense_valid)
1679 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1680 	}
1681 #ifdef CONFIG_SCSI_LOGGING
1682 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1683 	if (sense_valid) {
1684 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1685 						   "sd_done: sb[respc,sk,asc,"
1686 						   "ascq]=%x,%x,%x,%x\n",
1687 						   sshdr.response_code,
1688 						   sshdr.sense_key, sshdr.asc,
1689 						   sshdr.ascq));
1690 	}
1691 #endif
1692 	if (driver_byte(result) != DRIVER_SENSE &&
1693 	    (!sense_valid || sense_deferred))
1694 		goto out;
1695 
1696 	sdkp->medium_access_timed_out = 0;
1697 
1698 	switch (sshdr.sense_key) {
1699 	case HARDWARE_ERROR:
1700 	case MEDIUM_ERROR:
1701 		good_bytes = sd_completed_bytes(SCpnt);
1702 		break;
1703 	case RECOVERED_ERROR:
1704 		good_bytes = scsi_bufflen(SCpnt);
1705 		break;
1706 	case NO_SENSE:
1707 		/* This indicates a false check condition, so ignore it.  An
1708 		 * unknown amount of data was transferred so treat it as an
1709 		 * error.
1710 		 */
1711 		scsi_print_sense("sd", SCpnt);
1712 		SCpnt->result = 0;
1713 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1714 		break;
1715 	case ABORTED_COMMAND:
1716 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1717 			good_bytes = sd_completed_bytes(SCpnt);
1718 		break;
1719 	case ILLEGAL_REQUEST:
1720 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1721 			good_bytes = sd_completed_bytes(SCpnt);
1722 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1723 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1724 			switch (op) {
1725 			case UNMAP:
1726 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1727 				break;
1728 			case WRITE_SAME_16:
1729 			case WRITE_SAME:
1730 				if (unmap)
1731 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1732 				else {
1733 					sdkp->device->no_write_same = 1;
1734 					sd_config_write_same(sdkp);
1735 
1736 					good_bytes = 0;
1737 					req->__data_len = blk_rq_bytes(req);
1738 					req->cmd_flags |= REQ_QUIET;
1739 				}
1740 			}
1741 		}
1742 		break;
1743 	default:
1744 		break;
1745 	}
1746  out:
1747 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1748 		sd_dif_complete(SCpnt, good_bytes);
1749 
1750 	return good_bytes;
1751 }
1752 
1753 /*
1754  * spinup disk - called only in sd_revalidate_disk()
1755  */
1756 static void
1757 sd_spinup_disk(struct scsi_disk *sdkp)
1758 {
1759 	unsigned char cmd[10];
1760 	unsigned long spintime_expire = 0;
1761 	int retries, spintime;
1762 	unsigned int the_result;
1763 	struct scsi_sense_hdr sshdr;
1764 	int sense_valid = 0;
1765 
1766 	spintime = 0;
1767 
1768 	/* Spin up drives, as required.  Only do this at boot time */
1769 	/* Spinup needs to be done for module loads too. */
1770 	do {
1771 		retries = 0;
1772 
1773 		do {
1774 			cmd[0] = TEST_UNIT_READY;
1775 			memset((void *) &cmd[1], 0, 9);
1776 
1777 			the_result = scsi_execute_req(sdkp->device, cmd,
1778 						      DMA_NONE, NULL, 0,
1779 						      &sshdr, SD_TIMEOUT,
1780 						      SD_MAX_RETRIES, NULL);
1781 
1782 			/*
1783 			 * If the drive has indicated to us that it
1784 			 * doesn't have any media in it, don't bother
1785 			 * with any more polling.
1786 			 */
1787 			if (media_not_present(sdkp, &sshdr))
1788 				return;
1789 
1790 			if (the_result)
1791 				sense_valid = scsi_sense_valid(&sshdr);
1792 			retries++;
1793 		} while (retries < 3 &&
1794 			 (!scsi_status_is_good(the_result) ||
1795 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1796 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1797 
1798 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1799 			/* no sense, TUR either succeeded or failed
1800 			 * with a status error */
1801 			if(!spintime && !scsi_status_is_good(the_result)) {
1802 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1803 				sd_print_result(sdkp, the_result);
1804 			}
1805 			break;
1806 		}
1807 
1808 		/*
1809 		 * The device does not want the automatic start to be issued.
1810 		 */
1811 		if (sdkp->device->no_start_on_add)
1812 			break;
1813 
1814 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1815 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1816 				break;	/* manual intervention required */
1817 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1818 				break;	/* standby */
1819 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1820 				break;	/* unavailable */
1821 			/*
1822 			 * Issue command to spin up drive when not ready
1823 			 */
1824 			if (!spintime) {
1825 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1826 				cmd[0] = START_STOP;
1827 				cmd[1] = 1;	/* Return immediately */
1828 				memset((void *) &cmd[2], 0, 8);
1829 				cmd[4] = 1;	/* Start spin cycle */
1830 				if (sdkp->device->start_stop_pwr_cond)
1831 					cmd[4] |= 1 << 4;
1832 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1833 						 NULL, 0, &sshdr,
1834 						 SD_TIMEOUT, SD_MAX_RETRIES,
1835 						 NULL);
1836 				spintime_expire = jiffies + 100 * HZ;
1837 				spintime = 1;
1838 			}
1839 			/* Wait 1 second for next try */
1840 			msleep(1000);
1841 			printk(".");
1842 
1843 		/*
1844 		 * Wait for USB flash devices with slow firmware.
1845 		 * Yes, this sense key/ASC combination shouldn't
1846 		 * occur here.  It's characteristic of these devices.
1847 		 */
1848 		} else if (sense_valid &&
1849 				sshdr.sense_key == UNIT_ATTENTION &&
1850 				sshdr.asc == 0x28) {
1851 			if (!spintime) {
1852 				spintime_expire = jiffies + 5 * HZ;
1853 				spintime = 1;
1854 			}
1855 			/* Wait 1 second for next try */
1856 			msleep(1000);
1857 		} else {
1858 			/* we don't understand the sense code, so it's
1859 			 * probably pointless to loop */
1860 			if(!spintime) {
1861 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1862 				sd_print_sense_hdr(sdkp, &sshdr);
1863 			}
1864 			break;
1865 		}
1866 
1867 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1868 
1869 	if (spintime) {
1870 		if (scsi_status_is_good(the_result))
1871 			printk("ready\n");
1872 		else
1873 			printk("not responding...\n");
1874 	}
1875 }
1876 
1877 
1878 /*
1879  * Determine whether disk supports Data Integrity Field.
1880  */
1881 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1882 {
1883 	struct scsi_device *sdp = sdkp->device;
1884 	u8 type;
1885 	int ret = 0;
1886 
1887 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1888 		return ret;
1889 
1890 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1891 
1892 	if (type > SD_DIF_TYPE3_PROTECTION)
1893 		ret = -ENODEV;
1894 	else if (scsi_host_dif_capable(sdp->host, type))
1895 		ret = 1;
1896 
1897 	if (sdkp->first_scan || type != sdkp->protection_type)
1898 		switch (ret) {
1899 		case -ENODEV:
1900 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
1901 				  " protection type %u. Disabling disk!\n",
1902 				  type);
1903 			break;
1904 		case 1:
1905 			sd_printk(KERN_NOTICE, sdkp,
1906 				  "Enabling DIF Type %u protection\n", type);
1907 			break;
1908 		case 0:
1909 			sd_printk(KERN_NOTICE, sdkp,
1910 				  "Disabling DIF Type %u protection\n", type);
1911 			break;
1912 		}
1913 
1914 	sdkp->protection_type = type;
1915 
1916 	return ret;
1917 }
1918 
1919 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1920 			struct scsi_sense_hdr *sshdr, int sense_valid,
1921 			int the_result)
1922 {
1923 	sd_print_result(sdkp, the_result);
1924 	if (driver_byte(the_result) & DRIVER_SENSE)
1925 		sd_print_sense_hdr(sdkp, sshdr);
1926 	else
1927 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1928 
1929 	/*
1930 	 * Set dirty bit for removable devices if not ready -
1931 	 * sometimes drives will not report this properly.
1932 	 */
1933 	if (sdp->removable &&
1934 	    sense_valid && sshdr->sense_key == NOT_READY)
1935 		set_media_not_present(sdkp);
1936 
1937 	/*
1938 	 * We used to set media_present to 0 here to indicate no media
1939 	 * in the drive, but some drives fail read capacity even with
1940 	 * media present, so we can't do that.
1941 	 */
1942 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1943 }
1944 
1945 #define RC16_LEN 32
1946 #if RC16_LEN > SD_BUF_SIZE
1947 #error RC16_LEN must not be more than SD_BUF_SIZE
1948 #endif
1949 
1950 #define READ_CAPACITY_RETRIES_ON_RESET	10
1951 
1952 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1953 						unsigned char *buffer)
1954 {
1955 	unsigned char cmd[16];
1956 	struct scsi_sense_hdr sshdr;
1957 	int sense_valid = 0;
1958 	int the_result;
1959 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1960 	unsigned int alignment;
1961 	unsigned long long lba;
1962 	unsigned sector_size;
1963 
1964 	if (sdp->no_read_capacity_16)
1965 		return -EINVAL;
1966 
1967 	do {
1968 		memset(cmd, 0, 16);
1969 		cmd[0] = SERVICE_ACTION_IN;
1970 		cmd[1] = SAI_READ_CAPACITY_16;
1971 		cmd[13] = RC16_LEN;
1972 		memset(buffer, 0, RC16_LEN);
1973 
1974 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1975 					buffer, RC16_LEN, &sshdr,
1976 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1977 
1978 		if (media_not_present(sdkp, &sshdr))
1979 			return -ENODEV;
1980 
1981 		if (the_result) {
1982 			sense_valid = scsi_sense_valid(&sshdr);
1983 			if (sense_valid &&
1984 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1985 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1986 			    sshdr.ascq == 0x00)
1987 				/* Invalid Command Operation Code or
1988 				 * Invalid Field in CDB, just retry
1989 				 * silently with RC10 */
1990 				return -EINVAL;
1991 			if (sense_valid &&
1992 			    sshdr.sense_key == UNIT_ATTENTION &&
1993 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1994 				/* Device reset might occur several times,
1995 				 * give it one more chance */
1996 				if (--reset_retries > 0)
1997 					continue;
1998 		}
1999 		retries--;
2000 
2001 	} while (the_result && retries);
2002 
2003 	if (the_result) {
2004 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
2005 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2006 		return -EINVAL;
2007 	}
2008 
2009 	sector_size = get_unaligned_be32(&buffer[8]);
2010 	lba = get_unaligned_be64(&buffer[0]);
2011 
2012 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2013 		sdkp->capacity = 0;
2014 		return -ENODEV;
2015 	}
2016 
2017 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2018 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2019 			"kernel compiled with support for large block "
2020 			"devices.\n");
2021 		sdkp->capacity = 0;
2022 		return -EOVERFLOW;
2023 	}
2024 
2025 	/* Logical blocks per physical block exponent */
2026 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2027 
2028 	/* Lowest aligned logical block */
2029 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2030 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2031 	if (alignment && sdkp->first_scan)
2032 		sd_printk(KERN_NOTICE, sdkp,
2033 			  "physical block alignment offset: %u\n", alignment);
2034 
2035 	if (buffer[14] & 0x80) { /* LBPME */
2036 		sdkp->lbpme = 1;
2037 
2038 		if (buffer[14] & 0x40) /* LBPRZ */
2039 			sdkp->lbprz = 1;
2040 
2041 		sd_config_discard(sdkp, SD_LBP_WS16);
2042 	}
2043 
2044 	sdkp->capacity = lba + 1;
2045 	return sector_size;
2046 }
2047 
2048 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2049 						unsigned char *buffer)
2050 {
2051 	unsigned char cmd[16];
2052 	struct scsi_sense_hdr sshdr;
2053 	int sense_valid = 0;
2054 	int the_result;
2055 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2056 	sector_t lba;
2057 	unsigned sector_size;
2058 
2059 	do {
2060 		cmd[0] = READ_CAPACITY;
2061 		memset(&cmd[1], 0, 9);
2062 		memset(buffer, 0, 8);
2063 
2064 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2065 					buffer, 8, &sshdr,
2066 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2067 
2068 		if (media_not_present(sdkp, &sshdr))
2069 			return -ENODEV;
2070 
2071 		if (the_result) {
2072 			sense_valid = scsi_sense_valid(&sshdr);
2073 			if (sense_valid &&
2074 			    sshdr.sense_key == UNIT_ATTENTION &&
2075 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2076 				/* Device reset might occur several times,
2077 				 * give it one more chance */
2078 				if (--reset_retries > 0)
2079 					continue;
2080 		}
2081 		retries--;
2082 
2083 	} while (the_result && retries);
2084 
2085 	if (the_result) {
2086 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
2087 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2088 		return -EINVAL;
2089 	}
2090 
2091 	sector_size = get_unaligned_be32(&buffer[4]);
2092 	lba = get_unaligned_be32(&buffer[0]);
2093 
2094 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2095 		/* Some buggy (usb cardreader) devices return an lba of
2096 		   0xffffffff when the want to report a size of 0 (with
2097 		   which they really mean no media is present) */
2098 		sdkp->capacity = 0;
2099 		sdkp->physical_block_size = sector_size;
2100 		return sector_size;
2101 	}
2102 
2103 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2104 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2105 			"kernel compiled with support for large block "
2106 			"devices.\n");
2107 		sdkp->capacity = 0;
2108 		return -EOVERFLOW;
2109 	}
2110 
2111 	sdkp->capacity = lba + 1;
2112 	sdkp->physical_block_size = sector_size;
2113 	return sector_size;
2114 }
2115 
2116 static int sd_try_rc16_first(struct scsi_device *sdp)
2117 {
2118 	if (sdp->host->max_cmd_len < 16)
2119 		return 0;
2120 	if (sdp->try_rc_10_first)
2121 		return 0;
2122 	if (sdp->scsi_level > SCSI_SPC_2)
2123 		return 1;
2124 	if (scsi_device_protection(sdp))
2125 		return 1;
2126 	return 0;
2127 }
2128 
2129 /*
2130  * read disk capacity
2131  */
2132 static void
2133 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2134 {
2135 	int sector_size;
2136 	struct scsi_device *sdp = sdkp->device;
2137 	sector_t old_capacity = sdkp->capacity;
2138 
2139 	if (sd_try_rc16_first(sdp)) {
2140 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2141 		if (sector_size == -EOVERFLOW)
2142 			goto got_data;
2143 		if (sector_size == -ENODEV)
2144 			return;
2145 		if (sector_size < 0)
2146 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2147 		if (sector_size < 0)
2148 			return;
2149 	} else {
2150 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2151 		if (sector_size == -EOVERFLOW)
2152 			goto got_data;
2153 		if (sector_size < 0)
2154 			return;
2155 		if ((sizeof(sdkp->capacity) > 4) &&
2156 		    (sdkp->capacity > 0xffffffffULL)) {
2157 			int old_sector_size = sector_size;
2158 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2159 					"Trying to use READ CAPACITY(16).\n");
2160 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2161 			if (sector_size < 0) {
2162 				sd_printk(KERN_NOTICE, sdkp,
2163 					"Using 0xffffffff as device size\n");
2164 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2165 				sector_size = old_sector_size;
2166 				goto got_data;
2167 			}
2168 		}
2169 	}
2170 
2171 	/* Some devices are known to return the total number of blocks,
2172 	 * not the highest block number.  Some devices have versions
2173 	 * which do this and others which do not.  Some devices we might
2174 	 * suspect of doing this but we don't know for certain.
2175 	 *
2176 	 * If we know the reported capacity is wrong, decrement it.  If
2177 	 * we can only guess, then assume the number of blocks is even
2178 	 * (usually true but not always) and err on the side of lowering
2179 	 * the capacity.
2180 	 */
2181 	if (sdp->fix_capacity ||
2182 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2183 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2184 				"from its reported value: %llu\n",
2185 				(unsigned long long) sdkp->capacity);
2186 		--sdkp->capacity;
2187 	}
2188 
2189 got_data:
2190 	if (sector_size == 0) {
2191 		sector_size = 512;
2192 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2193 			  "assuming 512.\n");
2194 	}
2195 
2196 	if (sector_size != 512 &&
2197 	    sector_size != 1024 &&
2198 	    sector_size != 2048 &&
2199 	    sector_size != 4096 &&
2200 	    sector_size != 256) {
2201 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2202 			  sector_size);
2203 		/*
2204 		 * The user might want to re-format the drive with
2205 		 * a supported sectorsize.  Once this happens, it
2206 		 * would be relatively trivial to set the thing up.
2207 		 * For this reason, we leave the thing in the table.
2208 		 */
2209 		sdkp->capacity = 0;
2210 		/*
2211 		 * set a bogus sector size so the normal read/write
2212 		 * logic in the block layer will eventually refuse any
2213 		 * request on this device without tripping over power
2214 		 * of two sector size assumptions
2215 		 */
2216 		sector_size = 512;
2217 	}
2218 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2219 
2220 	{
2221 		char cap_str_2[10], cap_str_10[10];
2222 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
2223 
2224 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
2225 				sizeof(cap_str_2));
2226 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
2227 				sizeof(cap_str_10));
2228 
2229 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2230 			sd_printk(KERN_NOTICE, sdkp,
2231 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2232 				  (unsigned long long)sdkp->capacity,
2233 				  sector_size, cap_str_10, cap_str_2);
2234 
2235 			if (sdkp->physical_block_size != sector_size)
2236 				sd_printk(KERN_NOTICE, sdkp,
2237 					  "%u-byte physical blocks\n",
2238 					  sdkp->physical_block_size);
2239 		}
2240 	}
2241 
2242 	sdp->use_16_for_rw = (sdkp->capacity > 0xffffffff);
2243 
2244 	/* Rescale capacity to 512-byte units */
2245 	if (sector_size == 4096)
2246 		sdkp->capacity <<= 3;
2247 	else if (sector_size == 2048)
2248 		sdkp->capacity <<= 2;
2249 	else if (sector_size == 1024)
2250 		sdkp->capacity <<= 1;
2251 	else if (sector_size == 256)
2252 		sdkp->capacity >>= 1;
2253 
2254 	blk_queue_physical_block_size(sdp->request_queue,
2255 				      sdkp->physical_block_size);
2256 	sdkp->device->sector_size = sector_size;
2257 }
2258 
2259 /* called with buffer of length 512 */
2260 static inline int
2261 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2262 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2263 		 struct scsi_sense_hdr *sshdr)
2264 {
2265 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2266 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2267 			       sshdr);
2268 }
2269 
2270 /*
2271  * read write protect setting, if possible - called only in sd_revalidate_disk()
2272  * called with buffer of length SD_BUF_SIZE
2273  */
2274 static void
2275 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2276 {
2277 	int res;
2278 	struct scsi_device *sdp = sdkp->device;
2279 	struct scsi_mode_data data;
2280 	int old_wp = sdkp->write_prot;
2281 
2282 	set_disk_ro(sdkp->disk, 0);
2283 	if (sdp->skip_ms_page_3f) {
2284 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2285 		return;
2286 	}
2287 
2288 	if (sdp->use_192_bytes_for_3f) {
2289 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2290 	} else {
2291 		/*
2292 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2293 		 * We have to start carefully: some devices hang if we ask
2294 		 * for more than is available.
2295 		 */
2296 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2297 
2298 		/*
2299 		 * Second attempt: ask for page 0 When only page 0 is
2300 		 * implemented, a request for page 3F may return Sense Key
2301 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2302 		 * CDB.
2303 		 */
2304 		if (!scsi_status_is_good(res))
2305 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2306 
2307 		/*
2308 		 * Third attempt: ask 255 bytes, as we did earlier.
2309 		 */
2310 		if (!scsi_status_is_good(res))
2311 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2312 					       &data, NULL);
2313 	}
2314 
2315 	if (!scsi_status_is_good(res)) {
2316 		sd_first_printk(KERN_WARNING, sdkp,
2317 			  "Test WP failed, assume Write Enabled\n");
2318 	} else {
2319 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2320 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2321 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2322 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2323 				  sdkp->write_prot ? "on" : "off");
2324 			sd_printk(KERN_DEBUG, sdkp,
2325 				  "Mode Sense: %02x %02x %02x %02x\n",
2326 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2327 		}
2328 	}
2329 }
2330 
2331 /*
2332  * sd_read_cache_type - called only from sd_revalidate_disk()
2333  * called with buffer of length SD_BUF_SIZE
2334  */
2335 static void
2336 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2337 {
2338 	int len = 0, res;
2339 	struct scsi_device *sdp = sdkp->device;
2340 
2341 	int dbd;
2342 	int modepage;
2343 	int first_len;
2344 	struct scsi_mode_data data;
2345 	struct scsi_sense_hdr sshdr;
2346 	int old_wce = sdkp->WCE;
2347 	int old_rcd = sdkp->RCD;
2348 	int old_dpofua = sdkp->DPOFUA;
2349 
2350 
2351 	if (sdkp->cache_override)
2352 		return;
2353 
2354 	first_len = 4;
2355 	if (sdp->skip_ms_page_8) {
2356 		if (sdp->type == TYPE_RBC)
2357 			goto defaults;
2358 		else {
2359 			if (sdp->skip_ms_page_3f)
2360 				goto defaults;
2361 			modepage = 0x3F;
2362 			if (sdp->use_192_bytes_for_3f)
2363 				first_len = 192;
2364 			dbd = 0;
2365 		}
2366 	} else if (sdp->type == TYPE_RBC) {
2367 		modepage = 6;
2368 		dbd = 8;
2369 	} else {
2370 		modepage = 8;
2371 		dbd = 0;
2372 	}
2373 
2374 	/* cautiously ask */
2375 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2376 			&data, &sshdr);
2377 
2378 	if (!scsi_status_is_good(res))
2379 		goto bad_sense;
2380 
2381 	if (!data.header_length) {
2382 		modepage = 6;
2383 		first_len = 0;
2384 		sd_first_printk(KERN_ERR, sdkp,
2385 				"Missing header in MODE_SENSE response\n");
2386 	}
2387 
2388 	/* that went OK, now ask for the proper length */
2389 	len = data.length;
2390 
2391 	/*
2392 	 * We're only interested in the first three bytes, actually.
2393 	 * But the data cache page is defined for the first 20.
2394 	 */
2395 	if (len < 3)
2396 		goto bad_sense;
2397 	else if (len > SD_BUF_SIZE) {
2398 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2399 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2400 		len = SD_BUF_SIZE;
2401 	}
2402 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2403 		len = 192;
2404 
2405 	/* Get the data */
2406 	if (len > first_len)
2407 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2408 				&data, &sshdr);
2409 
2410 	if (scsi_status_is_good(res)) {
2411 		int offset = data.header_length + data.block_descriptor_length;
2412 
2413 		while (offset < len) {
2414 			u8 page_code = buffer[offset] & 0x3F;
2415 			u8 spf       = buffer[offset] & 0x40;
2416 
2417 			if (page_code == 8 || page_code == 6) {
2418 				/* We're interested only in the first 3 bytes.
2419 				 */
2420 				if (len - offset <= 2) {
2421 					sd_first_printk(KERN_ERR, sdkp,
2422 						"Incomplete mode parameter "
2423 							"data\n");
2424 					goto defaults;
2425 				} else {
2426 					modepage = page_code;
2427 					goto Page_found;
2428 				}
2429 			} else {
2430 				/* Go to the next page */
2431 				if (spf && len - offset > 3)
2432 					offset += 4 + (buffer[offset+2] << 8) +
2433 						buffer[offset+3];
2434 				else if (!spf && len - offset > 1)
2435 					offset += 2 + buffer[offset+1];
2436 				else {
2437 					sd_first_printk(KERN_ERR, sdkp,
2438 							"Incomplete mode "
2439 							"parameter data\n");
2440 					goto defaults;
2441 				}
2442 			}
2443 		}
2444 
2445 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2446 		goto defaults;
2447 
2448 	Page_found:
2449 		if (modepage == 8) {
2450 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2451 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2452 		} else {
2453 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2454 			sdkp->RCD = 0;
2455 		}
2456 
2457 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2458 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2459 			sd_first_printk(KERN_NOTICE, sdkp,
2460 				  "Uses READ/WRITE(6), disabling FUA\n");
2461 			sdkp->DPOFUA = 0;
2462 		}
2463 
2464 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2465 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2466 			sd_printk(KERN_NOTICE, sdkp,
2467 				  "Write cache: %s, read cache: %s, %s\n",
2468 				  sdkp->WCE ? "enabled" : "disabled",
2469 				  sdkp->RCD ? "disabled" : "enabled",
2470 				  sdkp->DPOFUA ? "supports DPO and FUA"
2471 				  : "doesn't support DPO or FUA");
2472 
2473 		return;
2474 	}
2475 
2476 bad_sense:
2477 	if (scsi_sense_valid(&sshdr) &&
2478 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2479 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2480 		/* Invalid field in CDB */
2481 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2482 	else
2483 		sd_first_printk(KERN_ERR, sdkp,
2484 				"Asking for cache data failed\n");
2485 
2486 defaults:
2487 	if (sdp->wce_default_on) {
2488 		sd_first_printk(KERN_NOTICE, sdkp,
2489 				"Assuming drive cache: write back\n");
2490 		sdkp->WCE = 1;
2491 	} else {
2492 		sd_first_printk(KERN_ERR, sdkp,
2493 				"Assuming drive cache: write through\n");
2494 		sdkp->WCE = 0;
2495 	}
2496 	sdkp->RCD = 0;
2497 	sdkp->DPOFUA = 0;
2498 }
2499 
2500 /*
2501  * The ATO bit indicates whether the DIF application tag is available
2502  * for use by the operating system.
2503  */
2504 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2505 {
2506 	int res, offset;
2507 	struct scsi_device *sdp = sdkp->device;
2508 	struct scsi_mode_data data;
2509 	struct scsi_sense_hdr sshdr;
2510 
2511 	if (sdp->type != TYPE_DISK)
2512 		return;
2513 
2514 	if (sdkp->protection_type == 0)
2515 		return;
2516 
2517 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2518 			      SD_MAX_RETRIES, &data, &sshdr);
2519 
2520 	if (!scsi_status_is_good(res) || !data.header_length ||
2521 	    data.length < 6) {
2522 		sd_first_printk(KERN_WARNING, sdkp,
2523 			  "getting Control mode page failed, assume no ATO\n");
2524 
2525 		if (scsi_sense_valid(&sshdr))
2526 			sd_print_sense_hdr(sdkp, &sshdr);
2527 
2528 		return;
2529 	}
2530 
2531 	offset = data.header_length + data.block_descriptor_length;
2532 
2533 	if ((buffer[offset] & 0x3f) != 0x0a) {
2534 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2535 		return;
2536 	}
2537 
2538 	if ((buffer[offset + 5] & 0x80) == 0)
2539 		return;
2540 
2541 	sdkp->ATO = 1;
2542 
2543 	return;
2544 }
2545 
2546 /**
2547  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2548  * @disk: disk to query
2549  */
2550 static void sd_read_block_limits(struct scsi_disk *sdkp)
2551 {
2552 	unsigned int sector_sz = sdkp->device->sector_size;
2553 	const int vpd_len = 64;
2554 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2555 
2556 	if (!buffer ||
2557 	    /* Block Limits VPD */
2558 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2559 		goto out;
2560 
2561 	blk_queue_io_min(sdkp->disk->queue,
2562 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2563 	blk_queue_io_opt(sdkp->disk->queue,
2564 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2565 
2566 	if (buffer[3] == 0x3c) {
2567 		unsigned int lba_count, desc_count;
2568 
2569 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2570 
2571 		if (!sdkp->lbpme)
2572 			goto out;
2573 
2574 		lba_count = get_unaligned_be32(&buffer[20]);
2575 		desc_count = get_unaligned_be32(&buffer[24]);
2576 
2577 		if (lba_count && desc_count)
2578 			sdkp->max_unmap_blocks = lba_count;
2579 
2580 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2581 
2582 		if (buffer[32] & 0x80)
2583 			sdkp->unmap_alignment =
2584 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2585 
2586 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2587 
2588 			if (sdkp->max_unmap_blocks)
2589 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2590 			else
2591 				sd_config_discard(sdkp, SD_LBP_WS16);
2592 
2593 		} else {	/* LBP VPD page tells us what to use */
2594 
2595 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2596 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2597 			else if (sdkp->lbpws)
2598 				sd_config_discard(sdkp, SD_LBP_WS16);
2599 			else if (sdkp->lbpws10)
2600 				sd_config_discard(sdkp, SD_LBP_WS10);
2601 			else
2602 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2603 		}
2604 	}
2605 
2606  out:
2607 	kfree(buffer);
2608 }
2609 
2610 /**
2611  * sd_read_block_characteristics - Query block dev. characteristics
2612  * @disk: disk to query
2613  */
2614 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2615 {
2616 	unsigned char *buffer;
2617 	u16 rot;
2618 	const int vpd_len = 64;
2619 
2620 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2621 
2622 	if (!buffer ||
2623 	    /* Block Device Characteristics VPD */
2624 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2625 		goto out;
2626 
2627 	rot = get_unaligned_be16(&buffer[4]);
2628 
2629 	if (rot == 1)
2630 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2631 
2632  out:
2633 	kfree(buffer);
2634 }
2635 
2636 /**
2637  * sd_read_block_provisioning - Query provisioning VPD page
2638  * @disk: disk to query
2639  */
2640 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2641 {
2642 	unsigned char *buffer;
2643 	const int vpd_len = 8;
2644 
2645 	if (sdkp->lbpme == 0)
2646 		return;
2647 
2648 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2649 
2650 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2651 		goto out;
2652 
2653 	sdkp->lbpvpd	= 1;
2654 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2655 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2656 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2657 
2658  out:
2659 	kfree(buffer);
2660 }
2661 
2662 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2663 {
2664 	struct scsi_device *sdev = sdkp->device;
2665 
2666 	if (sdev->host->no_write_same) {
2667 		sdev->no_write_same = 1;
2668 
2669 		return;
2670 	}
2671 
2672 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2673 		/* too large values might cause issues with arcmsr */
2674 		int vpd_buf_len = 64;
2675 
2676 		sdev->no_report_opcodes = 1;
2677 
2678 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2679 		 * CODES is unsupported and the device has an ATA
2680 		 * Information VPD page (SAT).
2681 		 */
2682 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2683 			sdev->no_write_same = 1;
2684 	}
2685 
2686 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2687 		sdkp->ws16 = 1;
2688 
2689 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2690 		sdkp->ws10 = 1;
2691 }
2692 
2693 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2694 {
2695 	/*
2696 	 * Although VPD inquiries can go to SCSI-2 type devices,
2697 	 * some USB ones crash on receiving them, and the pages
2698 	 * we currently ask for are for SPC-3 and beyond
2699 	 */
2700 	if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2701 		return 1;
2702 	return 0;
2703 }
2704 
2705 /**
2706  *	sd_revalidate_disk - called the first time a new disk is seen,
2707  *	performs disk spin up, read_capacity, etc.
2708  *	@disk: struct gendisk we care about
2709  **/
2710 static int sd_revalidate_disk(struct gendisk *disk)
2711 {
2712 	struct scsi_disk *sdkp = scsi_disk(disk);
2713 	struct scsi_device *sdp = sdkp->device;
2714 	unsigned char *buffer;
2715 	unsigned flush = 0;
2716 
2717 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2718 				      "sd_revalidate_disk\n"));
2719 
2720 	/*
2721 	 * If the device is offline, don't try and read capacity or any
2722 	 * of the other niceties.
2723 	 */
2724 	if (!scsi_device_online(sdp))
2725 		goto out;
2726 
2727 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2728 	if (!buffer) {
2729 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2730 			  "allocation failure.\n");
2731 		goto out;
2732 	}
2733 
2734 	sd_spinup_disk(sdkp);
2735 
2736 	/*
2737 	 * Without media there is no reason to ask; moreover, some devices
2738 	 * react badly if we do.
2739 	 */
2740 	if (sdkp->media_present) {
2741 		sd_read_capacity(sdkp, buffer);
2742 
2743 		if (sd_try_extended_inquiry(sdp)) {
2744 			sd_read_block_provisioning(sdkp);
2745 			sd_read_block_limits(sdkp);
2746 			sd_read_block_characteristics(sdkp);
2747 		}
2748 
2749 		sd_read_write_protect_flag(sdkp, buffer);
2750 		sd_read_cache_type(sdkp, buffer);
2751 		sd_read_app_tag_own(sdkp, buffer);
2752 		sd_read_write_same(sdkp, buffer);
2753 	}
2754 
2755 	sdkp->first_scan = 0;
2756 
2757 	/*
2758 	 * We now have all cache related info, determine how we deal
2759 	 * with flush requests.
2760 	 */
2761 	if (sdkp->WCE) {
2762 		flush |= REQ_FLUSH;
2763 		if (sdkp->DPOFUA)
2764 			flush |= REQ_FUA;
2765 	}
2766 
2767 	blk_queue_flush(sdkp->disk->queue, flush);
2768 
2769 	set_capacity(disk, sdkp->capacity);
2770 	sd_config_write_same(sdkp);
2771 	kfree(buffer);
2772 
2773  out:
2774 	return 0;
2775 }
2776 
2777 /**
2778  *	sd_unlock_native_capacity - unlock native capacity
2779  *	@disk: struct gendisk to set capacity for
2780  *
2781  *	Block layer calls this function if it detects that partitions
2782  *	on @disk reach beyond the end of the device.  If the SCSI host
2783  *	implements ->unlock_native_capacity() method, it's invoked to
2784  *	give it a chance to adjust the device capacity.
2785  *
2786  *	CONTEXT:
2787  *	Defined by block layer.  Might sleep.
2788  */
2789 static void sd_unlock_native_capacity(struct gendisk *disk)
2790 {
2791 	struct scsi_device *sdev = scsi_disk(disk)->device;
2792 
2793 	if (sdev->host->hostt->unlock_native_capacity)
2794 		sdev->host->hostt->unlock_native_capacity(sdev);
2795 }
2796 
2797 /**
2798  *	sd_format_disk_name - format disk name
2799  *	@prefix: name prefix - ie. "sd" for SCSI disks
2800  *	@index: index of the disk to format name for
2801  *	@buf: output buffer
2802  *	@buflen: length of the output buffer
2803  *
2804  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2805  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2806  *	which is followed by sdaaa.
2807  *
2808  *	This is basically 26 base counting with one extra 'nil' entry
2809  *	at the beginning from the second digit on and can be
2810  *	determined using similar method as 26 base conversion with the
2811  *	index shifted -1 after each digit is computed.
2812  *
2813  *	CONTEXT:
2814  *	Don't care.
2815  *
2816  *	RETURNS:
2817  *	0 on success, -errno on failure.
2818  */
2819 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2820 {
2821 	const int base = 'z' - 'a' + 1;
2822 	char *begin = buf + strlen(prefix);
2823 	char *end = buf + buflen;
2824 	char *p;
2825 	int unit;
2826 
2827 	p = end - 1;
2828 	*p = '\0';
2829 	unit = base;
2830 	do {
2831 		if (p == begin)
2832 			return -EINVAL;
2833 		*--p = 'a' + (index % unit);
2834 		index = (index / unit) - 1;
2835 	} while (index >= 0);
2836 
2837 	memmove(begin, p, end - p);
2838 	memcpy(buf, prefix, strlen(prefix));
2839 
2840 	return 0;
2841 }
2842 
2843 /*
2844  * The asynchronous part of sd_probe
2845  */
2846 static void sd_probe_async(void *data, async_cookie_t cookie)
2847 {
2848 	struct scsi_disk *sdkp = data;
2849 	struct scsi_device *sdp;
2850 	struct gendisk *gd;
2851 	u32 index;
2852 	struct device *dev;
2853 
2854 	sdp = sdkp->device;
2855 	gd = sdkp->disk;
2856 	index = sdkp->index;
2857 	dev = &sdp->sdev_gendev;
2858 
2859 	gd->major = sd_major((index & 0xf0) >> 4);
2860 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2861 	gd->minors = SD_MINORS;
2862 
2863 	gd->fops = &sd_fops;
2864 	gd->private_data = &sdkp->driver;
2865 	gd->queue = sdkp->device->request_queue;
2866 
2867 	/* defaults, until the device tells us otherwise */
2868 	sdp->sector_size = 512;
2869 	sdkp->capacity = 0;
2870 	sdkp->media_present = 1;
2871 	sdkp->write_prot = 0;
2872 	sdkp->cache_override = 0;
2873 	sdkp->WCE = 0;
2874 	sdkp->RCD = 0;
2875 	sdkp->ATO = 0;
2876 	sdkp->first_scan = 1;
2877 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2878 
2879 	sd_revalidate_disk(gd);
2880 
2881 	blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2882 	blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2883 
2884 	gd->driverfs_dev = &sdp->sdev_gendev;
2885 	gd->flags = GENHD_FL_EXT_DEVT;
2886 	if (sdp->removable) {
2887 		gd->flags |= GENHD_FL_REMOVABLE;
2888 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2889 	}
2890 
2891 	blk_pm_runtime_init(sdp->request_queue, dev);
2892 	add_disk(gd);
2893 	if (sdkp->capacity)
2894 		sd_dif_config_host(sdkp);
2895 
2896 	sd_revalidate_disk(gd);
2897 
2898 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2899 		  sdp->removable ? "removable " : "");
2900 	scsi_autopm_put_device(sdp);
2901 	put_device(&sdkp->dev);
2902 }
2903 
2904 /**
2905  *	sd_probe - called during driver initialization and whenever a
2906  *	new scsi device is attached to the system. It is called once
2907  *	for each scsi device (not just disks) present.
2908  *	@dev: pointer to device object
2909  *
2910  *	Returns 0 if successful (or not interested in this scsi device
2911  *	(e.g. scanner)); 1 when there is an error.
2912  *
2913  *	Note: this function is invoked from the scsi mid-level.
2914  *	This function sets up the mapping between a given
2915  *	<host,channel,id,lun> (found in sdp) and new device name
2916  *	(e.g. /dev/sda). More precisely it is the block device major
2917  *	and minor number that is chosen here.
2918  *
2919  *	Assume sd_probe is not re-entrant (for time being)
2920  *	Also think about sd_probe() and sd_remove() running coincidentally.
2921  **/
2922 static int sd_probe(struct device *dev)
2923 {
2924 	struct scsi_device *sdp = to_scsi_device(dev);
2925 	struct scsi_disk *sdkp;
2926 	struct gendisk *gd;
2927 	int index;
2928 	int error;
2929 
2930 	error = -ENODEV;
2931 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2932 		goto out;
2933 
2934 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2935 					"sd_probe\n"));
2936 
2937 	error = -ENOMEM;
2938 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2939 	if (!sdkp)
2940 		goto out;
2941 
2942 	gd = alloc_disk(SD_MINORS);
2943 	if (!gd)
2944 		goto out_free;
2945 
2946 	do {
2947 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2948 			goto out_put;
2949 
2950 		spin_lock(&sd_index_lock);
2951 		error = ida_get_new(&sd_index_ida, &index);
2952 		spin_unlock(&sd_index_lock);
2953 	} while (error == -EAGAIN);
2954 
2955 	if (error) {
2956 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2957 		goto out_put;
2958 	}
2959 
2960 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2961 	if (error) {
2962 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2963 		goto out_free_index;
2964 	}
2965 
2966 	sdkp->device = sdp;
2967 	sdkp->driver = &sd_template;
2968 	sdkp->disk = gd;
2969 	sdkp->index = index;
2970 	atomic_set(&sdkp->openers, 0);
2971 	atomic_set(&sdkp->device->ioerr_cnt, 0);
2972 
2973 	if (!sdp->request_queue->rq_timeout) {
2974 		if (sdp->type != TYPE_MOD)
2975 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2976 		else
2977 			blk_queue_rq_timeout(sdp->request_queue,
2978 					     SD_MOD_TIMEOUT);
2979 	}
2980 
2981 	device_initialize(&sdkp->dev);
2982 	sdkp->dev.parent = dev;
2983 	sdkp->dev.class = &sd_disk_class;
2984 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
2985 
2986 	if (device_add(&sdkp->dev))
2987 		goto out_free_index;
2988 
2989 	get_device(dev);
2990 	dev_set_drvdata(dev, sdkp);
2991 
2992 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2993 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
2994 
2995 	return 0;
2996 
2997  out_free_index:
2998 	spin_lock(&sd_index_lock);
2999 	ida_remove(&sd_index_ida, index);
3000 	spin_unlock(&sd_index_lock);
3001  out_put:
3002 	put_disk(gd);
3003  out_free:
3004 	kfree(sdkp);
3005  out:
3006 	return error;
3007 }
3008 
3009 /**
3010  *	sd_remove - called whenever a scsi disk (previously recognized by
3011  *	sd_probe) is detached from the system. It is called (potentially
3012  *	multiple times) during sd module unload.
3013  *	@sdp: pointer to mid level scsi device object
3014  *
3015  *	Note: this function is invoked from the scsi mid-level.
3016  *	This function potentially frees up a device name (e.g. /dev/sdc)
3017  *	that could be re-used by a subsequent sd_probe().
3018  *	This function is not called when the built-in sd driver is "exit-ed".
3019  **/
3020 static int sd_remove(struct device *dev)
3021 {
3022 	struct scsi_disk *sdkp;
3023 	dev_t devt;
3024 
3025 	sdkp = dev_get_drvdata(dev);
3026 	devt = disk_devt(sdkp->disk);
3027 	scsi_autopm_get_device(sdkp->device);
3028 
3029 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3030 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3031 	blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
3032 	blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
3033 	device_del(&sdkp->dev);
3034 	del_gendisk(sdkp->disk);
3035 	sd_shutdown(dev);
3036 
3037 	blk_register_region(devt, SD_MINORS, NULL,
3038 			    sd_default_probe, NULL, NULL);
3039 
3040 	mutex_lock(&sd_ref_mutex);
3041 	dev_set_drvdata(dev, NULL);
3042 	put_device(&sdkp->dev);
3043 	mutex_unlock(&sd_ref_mutex);
3044 
3045 	return 0;
3046 }
3047 
3048 /**
3049  *	scsi_disk_release - Called to free the scsi_disk structure
3050  *	@dev: pointer to embedded class device
3051  *
3052  *	sd_ref_mutex must be held entering this routine.  Because it is
3053  *	called on last put, you should always use the scsi_disk_get()
3054  *	scsi_disk_put() helpers which manipulate the semaphore directly
3055  *	and never do a direct put_device.
3056  **/
3057 static void scsi_disk_release(struct device *dev)
3058 {
3059 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3060 	struct gendisk *disk = sdkp->disk;
3061 
3062 	spin_lock(&sd_index_lock);
3063 	ida_remove(&sd_index_ida, sdkp->index);
3064 	spin_unlock(&sd_index_lock);
3065 
3066 	disk->private_data = NULL;
3067 	put_disk(disk);
3068 	put_device(&sdkp->device->sdev_gendev);
3069 
3070 	kfree(sdkp);
3071 }
3072 
3073 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3074 {
3075 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3076 	struct scsi_sense_hdr sshdr;
3077 	struct scsi_device *sdp = sdkp->device;
3078 	int res;
3079 
3080 	if (start)
3081 		cmd[4] |= 1;	/* START */
3082 
3083 	if (sdp->start_stop_pwr_cond)
3084 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3085 
3086 	if (!scsi_device_online(sdp))
3087 		return -ENODEV;
3088 
3089 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3090 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3091 	if (res) {
3092 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
3093 		sd_print_result(sdkp, res);
3094 		if (driver_byte(res) & DRIVER_SENSE)
3095 			sd_print_sense_hdr(sdkp, &sshdr);
3096 		if (scsi_sense_valid(&sshdr) &&
3097 			/* 0x3a is medium not present */
3098 			sshdr.asc == 0x3a)
3099 			res = 0;
3100 	}
3101 
3102 	/* SCSI error codes must not go to the generic layer */
3103 	if (res)
3104 		return -EIO;
3105 
3106 	return 0;
3107 }
3108 
3109 /*
3110  * Send a SYNCHRONIZE CACHE instruction down to the device through
3111  * the normal SCSI command structure.  Wait for the command to
3112  * complete.
3113  */
3114 static void sd_shutdown(struct device *dev)
3115 {
3116 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3117 
3118 	if (!sdkp)
3119 		return;         /* this can happen */
3120 
3121 	if (pm_runtime_suspended(dev))
3122 		goto exit;
3123 
3124 	if (sdkp->WCE && sdkp->media_present) {
3125 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3126 		sd_sync_cache(sdkp);
3127 	}
3128 
3129 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3130 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3131 		sd_start_stop_device(sdkp, 0);
3132 	}
3133 
3134 exit:
3135 	scsi_disk_put(sdkp);
3136 }
3137 
3138 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3139 {
3140 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3141 	int ret = 0;
3142 
3143 	if (!sdkp)
3144 		return 0;	/* this can happen */
3145 
3146 	if (sdkp->WCE && sdkp->media_present) {
3147 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3148 		ret = sd_sync_cache(sdkp);
3149 		if (ret) {
3150 			/* ignore OFFLINE device */
3151 			if (ret == -ENODEV)
3152 				ret = 0;
3153 			goto done;
3154 		}
3155 	}
3156 
3157 	if (sdkp->device->manage_start_stop) {
3158 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3159 		/* an error is not worth aborting a system sleep */
3160 		ret = sd_start_stop_device(sdkp, 0);
3161 		if (ignore_stop_errors)
3162 			ret = 0;
3163 	}
3164 
3165 done:
3166 	scsi_disk_put(sdkp);
3167 	return ret;
3168 }
3169 
3170 static int sd_suspend_system(struct device *dev)
3171 {
3172 	return sd_suspend_common(dev, true);
3173 }
3174 
3175 static int sd_suspend_runtime(struct device *dev)
3176 {
3177 	return sd_suspend_common(dev, false);
3178 }
3179 
3180 static int sd_resume(struct device *dev)
3181 {
3182 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3183 	int ret = 0;
3184 
3185 	if (!sdkp->device->manage_start_stop)
3186 		goto done;
3187 
3188 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3189 	ret = sd_start_stop_device(sdkp, 1);
3190 
3191 done:
3192 	scsi_disk_put(sdkp);
3193 	return ret;
3194 }
3195 
3196 /**
3197  *	init_sd - entry point for this driver (both when built in or when
3198  *	a module).
3199  *
3200  *	Note: this function registers this driver with the scsi mid-level.
3201  **/
3202 static int __init init_sd(void)
3203 {
3204 	int majors = 0, i, err;
3205 
3206 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3207 
3208 	for (i = 0; i < SD_MAJORS; i++) {
3209 		if (register_blkdev(sd_major(i), "sd") != 0)
3210 			continue;
3211 		majors++;
3212 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3213 				    sd_default_probe, NULL, NULL);
3214 	}
3215 
3216 	if (!majors)
3217 		return -ENODEV;
3218 
3219 	err = class_register(&sd_disk_class);
3220 	if (err)
3221 		goto err_out;
3222 
3223 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3224 					 0, 0, NULL);
3225 	if (!sd_cdb_cache) {
3226 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3227 		goto err_out_class;
3228 	}
3229 
3230 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3231 	if (!sd_cdb_pool) {
3232 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3233 		goto err_out_cache;
3234 	}
3235 
3236 	err = scsi_register_driver(&sd_template.gendrv);
3237 	if (err)
3238 		goto err_out_driver;
3239 
3240 	return 0;
3241 
3242 err_out_driver:
3243 	mempool_destroy(sd_cdb_pool);
3244 
3245 err_out_cache:
3246 	kmem_cache_destroy(sd_cdb_cache);
3247 
3248 err_out_class:
3249 	class_unregister(&sd_disk_class);
3250 err_out:
3251 	for (i = 0; i < SD_MAJORS; i++)
3252 		unregister_blkdev(sd_major(i), "sd");
3253 	return err;
3254 }
3255 
3256 /**
3257  *	exit_sd - exit point for this driver (when it is a module).
3258  *
3259  *	Note: this function unregisters this driver from the scsi mid-level.
3260  **/
3261 static void __exit exit_sd(void)
3262 {
3263 	int i;
3264 
3265 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3266 
3267 	scsi_unregister_driver(&sd_template.gendrv);
3268 	mempool_destroy(sd_cdb_pool);
3269 	kmem_cache_destroy(sd_cdb_cache);
3270 
3271 	class_unregister(&sd_disk_class);
3272 
3273 	for (i = 0; i < SD_MAJORS; i++) {
3274 		blk_unregister_region(sd_major(i), SD_MINORS);
3275 		unregister_blkdev(sd_major(i), "sd");
3276 	}
3277 }
3278 
3279 module_init(init_sd);
3280 module_exit(exit_sd);
3281 
3282 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3283 			       struct scsi_sense_hdr *sshdr)
3284 {
3285 	sd_printk(KERN_INFO, sdkp, " ");
3286 	scsi_show_sense_hdr(sshdr);
3287 	sd_printk(KERN_INFO, sdkp, " ");
3288 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
3289 }
3290 
3291 static void sd_print_result(struct scsi_disk *sdkp, int result)
3292 {
3293 	sd_printk(KERN_INFO, sdkp, " ");
3294 	scsi_show_result(result);
3295 }
3296 
3297