xref: /openbmc/linux/drivers/scsi/sd.c (revision 21278aea)
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_init_command(struct scsi_cmnd *SCpnt);
113 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
114 static int sd_done(struct scsi_cmnd *);
115 static int sd_eh_action(struct scsi_cmnd *, int);
116 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
117 static void scsi_disk_release(struct device *cdev);
118 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
119 static void sd_print_result(struct scsi_disk *, int);
120 
121 static DEFINE_SPINLOCK(sd_index_lock);
122 static DEFINE_IDA(sd_index_ida);
123 
124 /* This semaphore is used to mediate the 0->1 reference get in the
125  * face of object destruction (i.e. we can't allow a get on an
126  * object after last put) */
127 static DEFINE_MUTEX(sd_ref_mutex);
128 
129 static struct kmem_cache *sd_cdb_cache;
130 static mempool_t *sd_cdb_pool;
131 
132 static const char *sd_cache_types[] = {
133 	"write through", "none", "write back",
134 	"write back, no read (daft)"
135 };
136 
137 static ssize_t
138 cache_type_store(struct device *dev, struct device_attribute *attr,
139 		 const char *buf, size_t count)
140 {
141 	int i, ct = -1, rcd, wce, sp;
142 	struct scsi_disk *sdkp = to_scsi_disk(dev);
143 	struct scsi_device *sdp = sdkp->device;
144 	char buffer[64];
145 	char *buffer_data;
146 	struct scsi_mode_data data;
147 	struct scsi_sense_hdr sshdr;
148 	static const char temp[] = "temporary ";
149 	int len;
150 
151 	if (sdp->type != TYPE_DISK)
152 		/* no cache control on RBC devices; theoretically they
153 		 * can do it, but there's probably so many exceptions
154 		 * it's not worth the risk */
155 		return -EINVAL;
156 
157 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
158 		buf += sizeof(temp) - 1;
159 		sdkp->cache_override = 1;
160 	} else {
161 		sdkp->cache_override = 0;
162 	}
163 
164 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
165 		len = strlen(sd_cache_types[i]);
166 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
167 		    buf[len] == '\n') {
168 			ct = i;
169 			break;
170 		}
171 	}
172 	if (ct < 0)
173 		return -EINVAL;
174 	rcd = ct & 0x01 ? 1 : 0;
175 	wce = ct & 0x02 ? 1 : 0;
176 
177 	if (sdkp->cache_override) {
178 		sdkp->WCE = wce;
179 		sdkp->RCD = rcd;
180 		return count;
181 	}
182 
183 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
184 			    SD_MAX_RETRIES, &data, NULL))
185 		return -EINVAL;
186 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
187 		  data.block_descriptor_length);
188 	buffer_data = buffer + data.header_length +
189 		data.block_descriptor_length;
190 	buffer_data[2] &= ~0x05;
191 	buffer_data[2] |= wce << 2 | rcd;
192 	sp = buffer_data[0] & 0x80 ? 1 : 0;
193 
194 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
195 			     SD_MAX_RETRIES, &data, &sshdr)) {
196 		if (scsi_sense_valid(&sshdr))
197 			sd_print_sense_hdr(sdkp, &sshdr);
198 		return -EINVAL;
199 	}
200 	revalidate_disk(sdkp->disk);
201 	return count;
202 }
203 
204 static ssize_t
205 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
206 		       char *buf)
207 {
208 	struct scsi_disk *sdkp = to_scsi_disk(dev);
209 	struct scsi_device *sdp = sdkp->device;
210 
211 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
212 }
213 
214 static ssize_t
215 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
216 			const char *buf, size_t count)
217 {
218 	struct scsi_disk *sdkp = to_scsi_disk(dev);
219 	struct scsi_device *sdp = sdkp->device;
220 
221 	if (!capable(CAP_SYS_ADMIN))
222 		return -EACCES;
223 
224 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
225 
226 	return count;
227 }
228 static DEVICE_ATTR_RW(manage_start_stop);
229 
230 static ssize_t
231 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
232 {
233 	struct scsi_disk *sdkp = to_scsi_disk(dev);
234 
235 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
236 }
237 
238 static ssize_t
239 allow_restart_store(struct device *dev, struct device_attribute *attr,
240 		    const char *buf, size_t count)
241 {
242 	struct scsi_disk *sdkp = to_scsi_disk(dev);
243 	struct scsi_device *sdp = sdkp->device;
244 
245 	if (!capable(CAP_SYS_ADMIN))
246 		return -EACCES;
247 
248 	if (sdp->type != TYPE_DISK)
249 		return -EINVAL;
250 
251 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
252 
253 	return count;
254 }
255 static DEVICE_ATTR_RW(allow_restart);
256 
257 static ssize_t
258 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
259 {
260 	struct scsi_disk *sdkp = to_scsi_disk(dev);
261 	int ct = sdkp->RCD + 2*sdkp->WCE;
262 
263 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
264 }
265 static DEVICE_ATTR_RW(cache_type);
266 
267 static ssize_t
268 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
269 {
270 	struct scsi_disk *sdkp = to_scsi_disk(dev);
271 
272 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
273 }
274 static DEVICE_ATTR_RO(FUA);
275 
276 static ssize_t
277 protection_type_show(struct device *dev, struct device_attribute *attr,
278 		     char *buf)
279 {
280 	struct scsi_disk *sdkp = to_scsi_disk(dev);
281 
282 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
283 }
284 
285 static ssize_t
286 protection_type_store(struct device *dev, struct device_attribute *attr,
287 		      const char *buf, size_t count)
288 {
289 	struct scsi_disk *sdkp = to_scsi_disk(dev);
290 	unsigned int val;
291 	int err;
292 
293 	if (!capable(CAP_SYS_ADMIN))
294 		return -EACCES;
295 
296 	err = kstrtouint(buf, 10, &val);
297 
298 	if (err)
299 		return err;
300 
301 	if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
302 		sdkp->protection_type = val;
303 
304 	return count;
305 }
306 static DEVICE_ATTR_RW(protection_type);
307 
308 static ssize_t
309 protection_mode_show(struct device *dev, struct device_attribute *attr,
310 		     char *buf)
311 {
312 	struct scsi_disk *sdkp = to_scsi_disk(dev);
313 	struct scsi_device *sdp = sdkp->device;
314 	unsigned int dif, dix;
315 
316 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
317 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
318 
319 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
320 		dif = 0;
321 		dix = 1;
322 	}
323 
324 	if (!dif && !dix)
325 		return snprintf(buf, 20, "none\n");
326 
327 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
328 }
329 static DEVICE_ATTR_RO(protection_mode);
330 
331 static ssize_t
332 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
333 {
334 	struct scsi_disk *sdkp = to_scsi_disk(dev);
335 
336 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
337 }
338 static DEVICE_ATTR_RO(app_tag_own);
339 
340 static ssize_t
341 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
342 		       char *buf)
343 {
344 	struct scsi_disk *sdkp = to_scsi_disk(dev);
345 
346 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
347 }
348 static DEVICE_ATTR_RO(thin_provisioning);
349 
350 static const char *lbp_mode[] = {
351 	[SD_LBP_FULL]		= "full",
352 	[SD_LBP_UNMAP]		= "unmap",
353 	[SD_LBP_WS16]		= "writesame_16",
354 	[SD_LBP_WS10]		= "writesame_10",
355 	[SD_LBP_ZERO]		= "writesame_zero",
356 	[SD_LBP_DISABLE]	= "disabled",
357 };
358 
359 static ssize_t
360 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
361 		       char *buf)
362 {
363 	struct scsi_disk *sdkp = to_scsi_disk(dev);
364 
365 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
366 }
367 
368 static ssize_t
369 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
370 			const char *buf, size_t count)
371 {
372 	struct scsi_disk *sdkp = to_scsi_disk(dev);
373 	struct scsi_device *sdp = sdkp->device;
374 
375 	if (!capable(CAP_SYS_ADMIN))
376 		return -EACCES;
377 
378 	if (sdp->type != TYPE_DISK)
379 		return -EINVAL;
380 
381 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
382 		sd_config_discard(sdkp, SD_LBP_UNMAP);
383 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
384 		sd_config_discard(sdkp, SD_LBP_WS16);
385 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
386 		sd_config_discard(sdkp, SD_LBP_WS10);
387 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
388 		sd_config_discard(sdkp, SD_LBP_ZERO);
389 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
390 		sd_config_discard(sdkp, SD_LBP_DISABLE);
391 	else
392 		return -EINVAL;
393 
394 	return count;
395 }
396 static DEVICE_ATTR_RW(provisioning_mode);
397 
398 static ssize_t
399 max_medium_access_timeouts_show(struct device *dev,
400 				struct device_attribute *attr, char *buf)
401 {
402 	struct scsi_disk *sdkp = to_scsi_disk(dev);
403 
404 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
405 }
406 
407 static ssize_t
408 max_medium_access_timeouts_store(struct device *dev,
409 				 struct device_attribute *attr, const char *buf,
410 				 size_t count)
411 {
412 	struct scsi_disk *sdkp = to_scsi_disk(dev);
413 	int err;
414 
415 	if (!capable(CAP_SYS_ADMIN))
416 		return -EACCES;
417 
418 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
419 
420 	return err ? err : count;
421 }
422 static DEVICE_ATTR_RW(max_medium_access_timeouts);
423 
424 static ssize_t
425 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
426 			   char *buf)
427 {
428 	struct scsi_disk *sdkp = to_scsi_disk(dev);
429 
430 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
431 }
432 
433 static ssize_t
434 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
435 			    const char *buf, size_t count)
436 {
437 	struct scsi_disk *sdkp = to_scsi_disk(dev);
438 	struct scsi_device *sdp = sdkp->device;
439 	unsigned long max;
440 	int err;
441 
442 	if (!capable(CAP_SYS_ADMIN))
443 		return -EACCES;
444 
445 	if (sdp->type != TYPE_DISK)
446 		return -EINVAL;
447 
448 	err = kstrtoul(buf, 10, &max);
449 
450 	if (err)
451 		return err;
452 
453 	if (max == 0)
454 		sdp->no_write_same = 1;
455 	else if (max <= SD_MAX_WS16_BLOCKS) {
456 		sdp->no_write_same = 0;
457 		sdkp->max_ws_blocks = max;
458 	}
459 
460 	sd_config_write_same(sdkp);
461 
462 	return count;
463 }
464 static DEVICE_ATTR_RW(max_write_same_blocks);
465 
466 static struct attribute *sd_disk_attrs[] = {
467 	&dev_attr_cache_type.attr,
468 	&dev_attr_FUA.attr,
469 	&dev_attr_allow_restart.attr,
470 	&dev_attr_manage_start_stop.attr,
471 	&dev_attr_protection_type.attr,
472 	&dev_attr_protection_mode.attr,
473 	&dev_attr_app_tag_own.attr,
474 	&dev_attr_thin_provisioning.attr,
475 	&dev_attr_provisioning_mode.attr,
476 	&dev_attr_max_write_same_blocks.attr,
477 	&dev_attr_max_medium_access_timeouts.attr,
478 	NULL,
479 };
480 ATTRIBUTE_GROUPS(sd_disk);
481 
482 static struct class sd_disk_class = {
483 	.name		= "scsi_disk",
484 	.owner		= THIS_MODULE,
485 	.dev_release	= scsi_disk_release,
486 	.dev_groups	= sd_disk_groups,
487 };
488 
489 static const struct dev_pm_ops sd_pm_ops = {
490 	.suspend		= sd_suspend_system,
491 	.resume			= sd_resume,
492 	.poweroff		= sd_suspend_system,
493 	.restore		= sd_resume,
494 	.runtime_suspend	= sd_suspend_runtime,
495 	.runtime_resume		= sd_resume,
496 };
497 
498 static struct scsi_driver sd_template = {
499 	.owner			= THIS_MODULE,
500 	.gendrv = {
501 		.name		= "sd",
502 		.probe		= sd_probe,
503 		.remove		= sd_remove,
504 		.shutdown	= sd_shutdown,
505 		.pm		= &sd_pm_ops,
506 	},
507 	.rescan			= sd_rescan,
508 	.init_command		= sd_init_command,
509 	.uninit_command		= sd_uninit_command,
510 	.done			= sd_done,
511 	.eh_action		= sd_eh_action,
512 };
513 
514 /*
515  * Dummy kobj_map->probe function.
516  * The default ->probe function will call modprobe, which is
517  * pointless as this module is already loaded.
518  */
519 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
520 {
521 	return NULL;
522 }
523 
524 /*
525  * Device no to disk mapping:
526  *
527  *       major         disc2     disc  p1
528  *   |............|.............|....|....| <- dev_t
529  *    31        20 19          8 7  4 3  0
530  *
531  * Inside a major, we have 16k disks, however mapped non-
532  * contiguously. The first 16 disks are for major0, the next
533  * ones with major1, ... Disk 256 is for major0 again, disk 272
534  * for major1, ...
535  * As we stay compatible with our numbering scheme, we can reuse
536  * the well-know SCSI majors 8, 65--71, 136--143.
537  */
538 static int sd_major(int major_idx)
539 {
540 	switch (major_idx) {
541 	case 0:
542 		return SCSI_DISK0_MAJOR;
543 	case 1 ... 7:
544 		return SCSI_DISK1_MAJOR + major_idx - 1;
545 	case 8 ... 15:
546 		return SCSI_DISK8_MAJOR + major_idx - 8;
547 	default:
548 		BUG();
549 		return 0;	/* shut up gcc */
550 	}
551 }
552 
553 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
554 {
555 	struct scsi_disk *sdkp = NULL;
556 
557 	if (disk->private_data) {
558 		sdkp = scsi_disk(disk);
559 		if (scsi_device_get(sdkp->device) == 0)
560 			get_device(&sdkp->dev);
561 		else
562 			sdkp = NULL;
563 	}
564 	return sdkp;
565 }
566 
567 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
568 {
569 	struct scsi_disk *sdkp;
570 
571 	mutex_lock(&sd_ref_mutex);
572 	sdkp = __scsi_disk_get(disk);
573 	mutex_unlock(&sd_ref_mutex);
574 	return sdkp;
575 }
576 
577 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
578 {
579 	struct scsi_disk *sdkp;
580 
581 	mutex_lock(&sd_ref_mutex);
582 	sdkp = dev_get_drvdata(dev);
583 	if (sdkp)
584 		sdkp = __scsi_disk_get(sdkp->disk);
585 	mutex_unlock(&sd_ref_mutex);
586 	return sdkp;
587 }
588 
589 static void scsi_disk_put(struct scsi_disk *sdkp)
590 {
591 	struct scsi_device *sdev = sdkp->device;
592 
593 	mutex_lock(&sd_ref_mutex);
594 	put_device(&sdkp->dev);
595 	scsi_device_put(sdev);
596 	mutex_unlock(&sd_ref_mutex);
597 }
598 
599 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
600 {
601 	unsigned int prot_op = SCSI_PROT_NORMAL;
602 	unsigned int dix = scsi_prot_sg_count(scmd);
603 
604 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
605 		if (dif && dix)
606 			prot_op = SCSI_PROT_READ_PASS;
607 		else if (dif && !dix)
608 			prot_op = SCSI_PROT_READ_STRIP;
609 		else if (!dif && dix)
610 			prot_op = SCSI_PROT_READ_INSERT;
611 	} else {
612 		if (dif && dix)
613 			prot_op = SCSI_PROT_WRITE_PASS;
614 		else if (dif && !dix)
615 			prot_op = SCSI_PROT_WRITE_INSERT;
616 		else if (!dif && dix)
617 			prot_op = SCSI_PROT_WRITE_STRIP;
618 	}
619 
620 	scsi_set_prot_op(scmd, prot_op);
621 	scsi_set_prot_type(scmd, dif);
622 }
623 
624 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
625 {
626 	struct request_queue *q = sdkp->disk->queue;
627 	unsigned int logical_block_size = sdkp->device->sector_size;
628 	unsigned int max_blocks = 0;
629 
630 	q->limits.discard_zeroes_data = sdkp->lbprz;
631 	q->limits.discard_alignment = sdkp->unmap_alignment *
632 		logical_block_size;
633 	q->limits.discard_granularity =
634 		max(sdkp->physical_block_size,
635 		    sdkp->unmap_granularity * logical_block_size);
636 
637 	sdkp->provisioning_mode = mode;
638 
639 	switch (mode) {
640 
641 	case SD_LBP_DISABLE:
642 		q->limits.max_discard_sectors = 0;
643 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
644 		return;
645 
646 	case SD_LBP_UNMAP:
647 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
648 					  (u32)SD_MAX_WS16_BLOCKS);
649 		break;
650 
651 	case SD_LBP_WS16:
652 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
653 					  (u32)SD_MAX_WS16_BLOCKS);
654 		break;
655 
656 	case SD_LBP_WS10:
657 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
658 					  (u32)SD_MAX_WS10_BLOCKS);
659 		break;
660 
661 	case SD_LBP_ZERO:
662 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
663 					  (u32)SD_MAX_WS10_BLOCKS);
664 		q->limits.discard_zeroes_data = 1;
665 		break;
666 	}
667 
668 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
669 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
670 }
671 
672 /**
673  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
674  * @sdp: scsi device to operate one
675  * @rq: Request to prepare
676  *
677  * Will issue either UNMAP or WRITE SAME(16) depending on preference
678  * indicated by target device.
679  **/
680 static int sd_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
681 {
682 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
683 	sector_t sector = blk_rq_pos(rq);
684 	unsigned int nr_sectors = blk_rq_sectors(rq);
685 	unsigned int nr_bytes = blk_rq_bytes(rq);
686 	unsigned int len;
687 	int ret;
688 	char *buf;
689 	struct page *page;
690 
691 	sector >>= ilog2(sdp->sector_size) - 9;
692 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
693 	rq->timeout = SD_TIMEOUT;
694 
695 	memset(rq->cmd, 0, rq->cmd_len);
696 
697 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
698 	if (!page)
699 		return BLKPREP_DEFER;
700 
701 	switch (sdkp->provisioning_mode) {
702 	case SD_LBP_UNMAP:
703 		buf = page_address(page);
704 
705 		rq->cmd_len = 10;
706 		rq->cmd[0] = UNMAP;
707 		rq->cmd[8] = 24;
708 
709 		put_unaligned_be16(6 + 16, &buf[0]);
710 		put_unaligned_be16(16, &buf[2]);
711 		put_unaligned_be64(sector, &buf[8]);
712 		put_unaligned_be32(nr_sectors, &buf[16]);
713 
714 		len = 24;
715 		break;
716 
717 	case SD_LBP_WS16:
718 		rq->cmd_len = 16;
719 		rq->cmd[0] = WRITE_SAME_16;
720 		rq->cmd[1] = 0x8; /* UNMAP */
721 		put_unaligned_be64(sector, &rq->cmd[2]);
722 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
723 
724 		len = sdkp->device->sector_size;
725 		break;
726 
727 	case SD_LBP_WS10:
728 	case SD_LBP_ZERO:
729 		rq->cmd_len = 10;
730 		rq->cmd[0] = WRITE_SAME;
731 		if (sdkp->provisioning_mode == SD_LBP_WS10)
732 			rq->cmd[1] = 0x8; /* UNMAP */
733 		put_unaligned_be32(sector, &rq->cmd[2]);
734 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
735 
736 		len = sdkp->device->sector_size;
737 		break;
738 
739 	default:
740 		ret = BLKPREP_KILL;
741 		goto out;
742 	}
743 
744 	rq->completion_data = page;
745 	blk_add_request_payload(rq, page, len);
746 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
747 	rq->__data_len = nr_bytes;
748 
749 out:
750 	if (ret != BLKPREP_OK)
751 		__free_page(page);
752 	return ret;
753 }
754 
755 static void sd_config_write_same(struct scsi_disk *sdkp)
756 {
757 	struct request_queue *q = sdkp->disk->queue;
758 	unsigned int logical_block_size = sdkp->device->sector_size;
759 
760 	if (sdkp->device->no_write_same) {
761 		sdkp->max_ws_blocks = 0;
762 		goto out;
763 	}
764 
765 	/* Some devices can not handle block counts above 0xffff despite
766 	 * supporting WRITE SAME(16). Consequently we default to 64k
767 	 * blocks per I/O unless the device explicitly advertises a
768 	 * bigger limit.
769 	 */
770 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
771 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
772 						   (u32)SD_MAX_WS16_BLOCKS);
773 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
774 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
775 						   (u32)SD_MAX_WS10_BLOCKS);
776 	else {
777 		sdkp->device->no_write_same = 1;
778 		sdkp->max_ws_blocks = 0;
779 	}
780 
781 out:
782 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
783 					 (logical_block_size >> 9));
784 }
785 
786 /**
787  * sd_setup_write_same_cmnd - write the same data to multiple blocks
788  * @sdp: scsi device to operate one
789  * @rq: Request to prepare
790  *
791  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
792  * preference indicated by target device.
793  **/
794 static int sd_setup_write_same_cmnd(struct scsi_device *sdp, struct request *rq)
795 {
796 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
797 	struct bio *bio = rq->bio;
798 	sector_t sector = blk_rq_pos(rq);
799 	unsigned int nr_sectors = blk_rq_sectors(rq);
800 	unsigned int nr_bytes = blk_rq_bytes(rq);
801 	int ret;
802 
803 	if (sdkp->device->no_write_same)
804 		return BLKPREP_KILL;
805 
806 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
807 
808 	sector >>= ilog2(sdp->sector_size) - 9;
809 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
810 
811 	rq->__data_len = sdp->sector_size;
812 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
813 	memset(rq->cmd, 0, rq->cmd_len);
814 
815 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
816 		rq->cmd_len = 16;
817 		rq->cmd[0] = WRITE_SAME_16;
818 		put_unaligned_be64(sector, &rq->cmd[2]);
819 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
820 	} else {
821 		rq->cmd_len = 10;
822 		rq->cmd[0] = WRITE_SAME;
823 		put_unaligned_be32(sector, &rq->cmd[2]);
824 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
825 	}
826 
827 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
828 	rq->__data_len = nr_bytes;
829 
830 	return ret;
831 }
832 
833 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
834 {
835 	rq->timeout *= SD_FLUSH_TIMEOUT_MULTIPLIER;
836 	rq->retries = SD_MAX_RETRIES;
837 	rq->cmd[0] = SYNCHRONIZE_CACHE;
838 	rq->cmd_len = 10;
839 
840 	return scsi_setup_blk_pc_cmnd(sdp, rq);
841 }
842 
843 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
844 {
845 	struct request *rq = SCpnt->request;
846 
847 	if (rq->cmd_flags & REQ_DISCARD)
848 		__free_page(rq->completion_data);
849 
850 	if (SCpnt->cmnd != rq->cmd) {
851 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
852 		SCpnt->cmnd = NULL;
853 		SCpnt->cmd_len = 0;
854 	}
855 }
856 
857 static int sd_init_command(struct scsi_cmnd *SCpnt)
858 {
859 	struct request *rq = SCpnt->request;
860 	struct scsi_device *sdp = SCpnt->device;
861 	struct gendisk *disk = rq->rq_disk;
862 	struct scsi_disk *sdkp;
863 	sector_t block = blk_rq_pos(rq);
864 	sector_t threshold;
865 	unsigned int this_count = blk_rq_sectors(rq);
866 	int ret, host_dif;
867 	unsigned char protect;
868 
869 	/*
870 	 * Discard request come in as REQ_TYPE_FS but we turn them into
871 	 * block PC requests to make life easier.
872 	 */
873 	if (rq->cmd_flags & REQ_DISCARD) {
874 		ret = sd_setup_discard_cmnd(sdp, rq);
875 		goto out;
876 	} else if (rq->cmd_flags & REQ_WRITE_SAME) {
877 		ret = sd_setup_write_same_cmnd(sdp, rq);
878 		goto out;
879 	} else if (rq->cmd_flags & REQ_FLUSH) {
880 		ret = scsi_setup_flush_cmnd(sdp, rq);
881 		goto out;
882 	}
883 	ret = scsi_setup_fs_cmnd(sdp, rq);
884 	if (ret != BLKPREP_OK)
885 		goto out;
886 	SCpnt = rq->special;
887 	sdkp = scsi_disk(disk);
888 
889 	/* from here on until we're complete, any goto out
890 	 * is used for a killable error condition */
891 	ret = BLKPREP_KILL;
892 
893 	SCSI_LOG_HLQUEUE(1,
894 		scmd_printk(KERN_INFO, SCpnt,
895 			"%s: block=%llu, count=%d\n",
896 			__func__, (unsigned long long)block, this_count));
897 
898 	if (!sdp || !scsi_device_online(sdp) ||
899 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
900 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
901 						"Finishing %u sectors\n",
902 						blk_rq_sectors(rq)));
903 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
904 						"Retry with 0x%p\n", SCpnt));
905 		goto out;
906 	}
907 
908 	if (sdp->changed) {
909 		/*
910 		 * quietly refuse to do anything to a changed disc until
911 		 * the changed bit has been reset
912 		 */
913 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
914 		goto out;
915 	}
916 
917 	/*
918 	 * Some SD card readers can't handle multi-sector accesses which touch
919 	 * the last one or two hardware sectors.  Split accesses as needed.
920 	 */
921 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
922 		(sdp->sector_size / 512);
923 
924 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
925 		if (block < threshold) {
926 			/* Access up to the threshold but not beyond */
927 			this_count = threshold - block;
928 		} else {
929 			/* Access only a single hardware sector */
930 			this_count = sdp->sector_size / 512;
931 		}
932 	}
933 
934 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
935 					(unsigned long long)block));
936 
937 	/*
938 	 * If we have a 1K hardware sectorsize, prevent access to single
939 	 * 512 byte sectors.  In theory we could handle this - in fact
940 	 * the scsi cdrom driver must be able to handle this because
941 	 * we typically use 1K blocksizes, and cdroms typically have
942 	 * 2K hardware sectorsizes.  Of course, things are simpler
943 	 * with the cdrom, since it is read-only.  For performance
944 	 * reasons, the filesystems should be able to handle this
945 	 * and not force the scsi disk driver to use bounce buffers
946 	 * for this.
947 	 */
948 	if (sdp->sector_size == 1024) {
949 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
950 			scmd_printk(KERN_ERR, SCpnt,
951 				    "Bad block number requested\n");
952 			goto out;
953 		} else {
954 			block = block >> 1;
955 			this_count = this_count >> 1;
956 		}
957 	}
958 	if (sdp->sector_size == 2048) {
959 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
960 			scmd_printk(KERN_ERR, SCpnt,
961 				    "Bad block number requested\n");
962 			goto out;
963 		} else {
964 			block = block >> 2;
965 			this_count = this_count >> 2;
966 		}
967 	}
968 	if (sdp->sector_size == 4096) {
969 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
970 			scmd_printk(KERN_ERR, SCpnt,
971 				    "Bad block number requested\n");
972 			goto out;
973 		} else {
974 			block = block >> 3;
975 			this_count = this_count >> 3;
976 		}
977 	}
978 	if (rq_data_dir(rq) == WRITE) {
979 		if (!sdp->writeable) {
980 			goto out;
981 		}
982 		SCpnt->cmnd[0] = WRITE_6;
983 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
984 
985 		if (blk_integrity_rq(rq))
986 			sd_dif_prepare(rq, block, sdp->sector_size);
987 
988 	} else if (rq_data_dir(rq) == READ) {
989 		SCpnt->cmnd[0] = READ_6;
990 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
991 	} else {
992 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %llx\n", (unsigned long long) rq->cmd_flags);
993 		goto out;
994 	}
995 
996 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
997 					"%s %d/%u 512 byte blocks.\n",
998 					(rq_data_dir(rq) == WRITE) ?
999 					"writing" : "reading", this_count,
1000 					blk_rq_sectors(rq)));
1001 
1002 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
1003 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
1004 	if (host_dif)
1005 		protect = 1 << 5;
1006 	else
1007 		protect = 0;
1008 
1009 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
1010 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1011 
1012 		if (unlikely(SCpnt->cmnd == NULL)) {
1013 			ret = BLKPREP_DEFER;
1014 			goto out;
1015 		}
1016 
1017 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1018 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1019 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1020 		SCpnt->cmnd[7] = 0x18;
1021 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1022 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1023 
1024 		/* LBA */
1025 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1026 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1027 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1028 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1029 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1030 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1031 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1032 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1033 
1034 		/* Expected Indirect LBA */
1035 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1036 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1037 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1038 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1039 
1040 		/* Transfer length */
1041 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1042 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1043 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1044 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1045 	} else if (sdp->use_16_for_rw) {
1046 		SCpnt->cmnd[0] += READ_16 - READ_6;
1047 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1048 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1049 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1050 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1051 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1052 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1053 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1054 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1055 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1056 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1057 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1058 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1059 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1060 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1061 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1062 		   scsi_device_protection(SCpnt->device) ||
1063 		   SCpnt->device->use_10_for_rw) {
1064 		if (this_count > 0xffff)
1065 			this_count = 0xffff;
1066 
1067 		SCpnt->cmnd[0] += READ_10 - READ_6;
1068 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1069 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1070 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1071 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1072 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1073 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1074 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1075 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1076 	} else {
1077 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1078 			/*
1079 			 * This happens only if this drive failed
1080 			 * 10byte rw command with ILLEGAL_REQUEST
1081 			 * during operation and thus turned off
1082 			 * use_10_for_rw.
1083 			 */
1084 			scmd_printk(KERN_ERR, SCpnt,
1085 				    "FUA write on READ/WRITE(6) drive\n");
1086 			goto out;
1087 		}
1088 
1089 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1090 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1091 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1092 		SCpnt->cmnd[4] = (unsigned char) this_count;
1093 		SCpnt->cmnd[5] = 0;
1094 	}
1095 	SCpnt->sdb.length = this_count * sdp->sector_size;
1096 
1097 	/* If DIF or DIX is enabled, tell HBA how to handle request */
1098 	if (host_dif || scsi_prot_sg_count(SCpnt))
1099 		sd_prot_op(SCpnt, host_dif);
1100 
1101 	/*
1102 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1103 	 * host adapter, it's safe to assume that we can at least transfer
1104 	 * this many bytes between each connect / disconnect.
1105 	 */
1106 	SCpnt->transfersize = sdp->sector_size;
1107 	SCpnt->underflow = this_count << 9;
1108 	SCpnt->allowed = SD_MAX_RETRIES;
1109 
1110 	/*
1111 	 * This indicates that the command is ready from our end to be
1112 	 * queued.
1113 	 */
1114 	ret = BLKPREP_OK;
1115  out:
1116 	return ret;
1117 }
1118 
1119 /**
1120  *	sd_open - open a scsi disk device
1121  *	@inode: only i_rdev member may be used
1122  *	@filp: only f_mode and f_flags may be used
1123  *
1124  *	Returns 0 if successful. Returns a negated errno value in case
1125  *	of error.
1126  *
1127  *	Note: This can be called from a user context (e.g. fsck(1) )
1128  *	or from within the kernel (e.g. as a result of a mount(1) ).
1129  *	In the latter case @inode and @filp carry an abridged amount
1130  *	of information as noted above.
1131  *
1132  *	Locking: called with bdev->bd_mutex held.
1133  **/
1134 static int sd_open(struct block_device *bdev, fmode_t mode)
1135 {
1136 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1137 	struct scsi_device *sdev;
1138 	int retval;
1139 
1140 	if (!sdkp)
1141 		return -ENXIO;
1142 
1143 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1144 
1145 	sdev = sdkp->device;
1146 
1147 	/*
1148 	 * If the device is in error recovery, wait until it is done.
1149 	 * If the device is offline, then disallow any access to it.
1150 	 */
1151 	retval = -ENXIO;
1152 	if (!scsi_block_when_processing_errors(sdev))
1153 		goto error_out;
1154 
1155 	if (sdev->removable || sdkp->write_prot)
1156 		check_disk_change(bdev);
1157 
1158 	/*
1159 	 * If the drive is empty, just let the open fail.
1160 	 */
1161 	retval = -ENOMEDIUM;
1162 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1163 		goto error_out;
1164 
1165 	/*
1166 	 * If the device has the write protect tab set, have the open fail
1167 	 * if the user expects to be able to write to the thing.
1168 	 */
1169 	retval = -EROFS;
1170 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1171 		goto error_out;
1172 
1173 	/*
1174 	 * It is possible that the disk changing stuff resulted in
1175 	 * the device being taken offline.  If this is the case,
1176 	 * report this to the user, and don't pretend that the
1177 	 * open actually succeeded.
1178 	 */
1179 	retval = -ENXIO;
1180 	if (!scsi_device_online(sdev))
1181 		goto error_out;
1182 
1183 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1184 		if (scsi_block_when_processing_errors(sdev))
1185 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1186 	}
1187 
1188 	return 0;
1189 
1190 error_out:
1191 	scsi_disk_put(sdkp);
1192 	return retval;
1193 }
1194 
1195 /**
1196  *	sd_release - invoked when the (last) close(2) is called on this
1197  *	scsi disk.
1198  *	@inode: only i_rdev member may be used
1199  *	@filp: only f_mode and f_flags may be used
1200  *
1201  *	Returns 0.
1202  *
1203  *	Note: may block (uninterruptible) if error recovery is underway
1204  *	on this disk.
1205  *
1206  *	Locking: called with bdev->bd_mutex held.
1207  **/
1208 static void sd_release(struct gendisk *disk, fmode_t mode)
1209 {
1210 	struct scsi_disk *sdkp = scsi_disk(disk);
1211 	struct scsi_device *sdev = sdkp->device;
1212 
1213 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1214 
1215 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1216 		if (scsi_block_when_processing_errors(sdev))
1217 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1218 	}
1219 
1220 	/*
1221 	 * XXX and what if there are packets in flight and this close()
1222 	 * XXX is followed by a "rmmod sd_mod"?
1223 	 */
1224 
1225 	scsi_disk_put(sdkp);
1226 }
1227 
1228 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1229 {
1230 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1231 	struct scsi_device *sdp = sdkp->device;
1232 	struct Scsi_Host *host = sdp->host;
1233 	int diskinfo[4];
1234 
1235 	/* default to most commonly used values */
1236         diskinfo[0] = 0x40;	/* 1 << 6 */
1237        	diskinfo[1] = 0x20;	/* 1 << 5 */
1238        	diskinfo[2] = sdkp->capacity >> 11;
1239 
1240 	/* override with calculated, extended default, or driver values */
1241 	if (host->hostt->bios_param)
1242 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1243 	else
1244 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1245 
1246 	geo->heads = diskinfo[0];
1247 	geo->sectors = diskinfo[1];
1248 	geo->cylinders = diskinfo[2];
1249 	return 0;
1250 }
1251 
1252 /**
1253  *	sd_ioctl - process an ioctl
1254  *	@inode: only i_rdev/i_bdev members may be used
1255  *	@filp: only f_mode and f_flags may be used
1256  *	@cmd: ioctl command number
1257  *	@arg: this is third argument given to ioctl(2) system call.
1258  *	Often contains a pointer.
1259  *
1260  *	Returns 0 if successful (some ioctls return positive numbers on
1261  *	success as well). Returns a negated errno value in case of error.
1262  *
1263  *	Note: most ioctls are forward onto the block subsystem or further
1264  *	down in the scsi subsystem.
1265  **/
1266 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1267 		    unsigned int cmd, unsigned long arg)
1268 {
1269 	struct gendisk *disk = bdev->bd_disk;
1270 	struct scsi_disk *sdkp = scsi_disk(disk);
1271 	struct scsi_device *sdp = sdkp->device;
1272 	void __user *p = (void __user *)arg;
1273 	int error;
1274 
1275 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1276 				    "cmd=0x%x\n", disk->disk_name, cmd));
1277 
1278 	error = scsi_verify_blk_ioctl(bdev, cmd);
1279 	if (error < 0)
1280 		return error;
1281 
1282 	/*
1283 	 * If we are in the middle of error recovery, don't let anyone
1284 	 * else try and use this device.  Also, if error recovery fails, it
1285 	 * may try and take the device offline, in which case all further
1286 	 * access to the device is prohibited.
1287 	 */
1288 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1289 					(mode & FMODE_NDELAY) != 0);
1290 	if (!scsi_block_when_processing_errors(sdp) || !error)
1291 		goto out;
1292 
1293 	/*
1294 	 * Send SCSI addressing ioctls directly to mid level, send other
1295 	 * ioctls to block level and then onto mid level if they can't be
1296 	 * resolved.
1297 	 */
1298 	switch (cmd) {
1299 		case SCSI_IOCTL_GET_IDLUN:
1300 		case SCSI_IOCTL_GET_BUS_NUMBER:
1301 			error = scsi_ioctl(sdp, cmd, p);
1302 			break;
1303 		default:
1304 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1305 			if (error != -ENOTTY)
1306 				break;
1307 			error = scsi_ioctl(sdp, cmd, p);
1308 			break;
1309 	}
1310 out:
1311 	return error;
1312 }
1313 
1314 static void set_media_not_present(struct scsi_disk *sdkp)
1315 {
1316 	if (sdkp->media_present)
1317 		sdkp->device->changed = 1;
1318 
1319 	if (sdkp->device->removable) {
1320 		sdkp->media_present = 0;
1321 		sdkp->capacity = 0;
1322 	}
1323 }
1324 
1325 static int media_not_present(struct scsi_disk *sdkp,
1326 			     struct scsi_sense_hdr *sshdr)
1327 {
1328 	if (!scsi_sense_valid(sshdr))
1329 		return 0;
1330 
1331 	/* not invoked for commands that could return deferred errors */
1332 	switch (sshdr->sense_key) {
1333 	case UNIT_ATTENTION:
1334 	case NOT_READY:
1335 		/* medium not present */
1336 		if (sshdr->asc == 0x3A) {
1337 			set_media_not_present(sdkp);
1338 			return 1;
1339 		}
1340 	}
1341 	return 0;
1342 }
1343 
1344 /**
1345  *	sd_check_events - check media events
1346  *	@disk: kernel device descriptor
1347  *	@clearing: disk events currently being cleared
1348  *
1349  *	Returns mask of DISK_EVENT_*.
1350  *
1351  *	Note: this function is invoked from the block subsystem.
1352  **/
1353 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1354 {
1355 	struct scsi_disk *sdkp = scsi_disk(disk);
1356 	struct scsi_device *sdp = sdkp->device;
1357 	struct scsi_sense_hdr *sshdr = NULL;
1358 	int retval;
1359 
1360 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1361 
1362 	/*
1363 	 * If the device is offline, don't send any commands - just pretend as
1364 	 * if the command failed.  If the device ever comes back online, we
1365 	 * can deal with it then.  It is only because of unrecoverable errors
1366 	 * that we would ever take a device offline in the first place.
1367 	 */
1368 	if (!scsi_device_online(sdp)) {
1369 		set_media_not_present(sdkp);
1370 		goto out;
1371 	}
1372 
1373 	/*
1374 	 * Using TEST_UNIT_READY enables differentiation between drive with
1375 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1376 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1377 	 *
1378 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1379 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1380 	 * sd_revalidate() is called.
1381 	 */
1382 	retval = -ENODEV;
1383 
1384 	if (scsi_block_when_processing_errors(sdp)) {
1385 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1386 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1387 					      sshdr);
1388 	}
1389 
1390 	/* failed to execute TUR, assume media not present */
1391 	if (host_byte(retval)) {
1392 		set_media_not_present(sdkp);
1393 		goto out;
1394 	}
1395 
1396 	if (media_not_present(sdkp, sshdr))
1397 		goto out;
1398 
1399 	/*
1400 	 * For removable scsi disk we have to recognise the presence
1401 	 * of a disk in the drive.
1402 	 */
1403 	if (!sdkp->media_present)
1404 		sdp->changed = 1;
1405 	sdkp->media_present = 1;
1406 out:
1407 	/*
1408 	 * sdp->changed is set under the following conditions:
1409 	 *
1410 	 *	Medium present state has changed in either direction.
1411 	 *	Device has indicated UNIT_ATTENTION.
1412 	 */
1413 	kfree(sshdr);
1414 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1415 	sdp->changed = 0;
1416 	return retval;
1417 }
1418 
1419 static int sd_sync_cache(struct scsi_disk *sdkp)
1420 {
1421 	int retries, res;
1422 	struct scsi_device *sdp = sdkp->device;
1423 	const int timeout = sdp->request_queue->rq_timeout
1424 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1425 	struct scsi_sense_hdr sshdr;
1426 
1427 	if (!scsi_device_online(sdp))
1428 		return -ENODEV;
1429 
1430 	for (retries = 3; retries > 0; --retries) {
1431 		unsigned char cmd[10] = { 0 };
1432 
1433 		cmd[0] = SYNCHRONIZE_CACHE;
1434 		/*
1435 		 * Leave the rest of the command zero to indicate
1436 		 * flush everything.
1437 		 */
1438 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1439 					     &sshdr, timeout, SD_MAX_RETRIES,
1440 					     NULL, REQ_PM);
1441 		if (res == 0)
1442 			break;
1443 	}
1444 
1445 	if (res) {
1446 		sd_print_result(sdkp, res);
1447 
1448 		if (driver_byte(res) & DRIVER_SENSE)
1449 			sd_print_sense_hdr(sdkp, &sshdr);
1450 		/* we need to evaluate the error return  */
1451 		if (scsi_sense_valid(&sshdr) &&
1452 			(sshdr.asc == 0x3a ||	/* medium not present */
1453 			 sshdr.asc == 0x20))	/* invalid command */
1454 				/* this is no error here */
1455 				return 0;
1456 
1457 		switch (host_byte(res)) {
1458 		/* ignore errors due to racing a disconnection */
1459 		case DID_BAD_TARGET:
1460 		case DID_NO_CONNECT:
1461 			return 0;
1462 		/* signal the upper layer it might try again */
1463 		case DID_BUS_BUSY:
1464 		case DID_IMM_RETRY:
1465 		case DID_REQUEUE:
1466 		case DID_SOFT_ERROR:
1467 			return -EBUSY;
1468 		default:
1469 			return -EIO;
1470 		}
1471 	}
1472 	return 0;
1473 }
1474 
1475 static void sd_rescan(struct device *dev)
1476 {
1477 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1478 
1479 	if (sdkp) {
1480 		revalidate_disk(sdkp->disk);
1481 		scsi_disk_put(sdkp);
1482 	}
1483 }
1484 
1485 
1486 #ifdef CONFIG_COMPAT
1487 /*
1488  * This gets directly called from VFS. When the ioctl
1489  * is not recognized we go back to the other translation paths.
1490  */
1491 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1492 			   unsigned int cmd, unsigned long arg)
1493 {
1494 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1495 	int ret;
1496 
1497 	ret = scsi_verify_blk_ioctl(bdev, cmd);
1498 	if (ret < 0)
1499 		return ret;
1500 
1501 	/*
1502 	 * If we are in the middle of error recovery, don't let anyone
1503 	 * else try and use this device.  Also, if error recovery fails, it
1504 	 * may try and take the device offline, in which case all further
1505 	 * access to the device is prohibited.
1506 	 */
1507 	if (!scsi_block_when_processing_errors(sdev))
1508 		return -ENODEV;
1509 
1510 	if (sdev->host->hostt->compat_ioctl) {
1511 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1512 
1513 		return ret;
1514 	}
1515 
1516 	/*
1517 	 * Let the static ioctl translation table take care of it.
1518 	 */
1519 	return -ENOIOCTLCMD;
1520 }
1521 #endif
1522 
1523 static const struct block_device_operations sd_fops = {
1524 	.owner			= THIS_MODULE,
1525 	.open			= sd_open,
1526 	.release		= sd_release,
1527 	.ioctl			= sd_ioctl,
1528 	.getgeo			= sd_getgeo,
1529 #ifdef CONFIG_COMPAT
1530 	.compat_ioctl		= sd_compat_ioctl,
1531 #endif
1532 	.check_events		= sd_check_events,
1533 	.revalidate_disk	= sd_revalidate_disk,
1534 	.unlock_native_capacity	= sd_unlock_native_capacity,
1535 };
1536 
1537 /**
1538  *	sd_eh_action - error handling callback
1539  *	@scmd:		sd-issued command that has failed
1540  *	@eh_disp:	The recovery disposition suggested by the midlayer
1541  *
1542  *	This function is called by the SCSI midlayer upon completion of an
1543  *	error test command (currently TEST UNIT READY). The result of sending
1544  *	the eh command is passed in eh_disp.  We're looking for devices that
1545  *	fail medium access commands but are OK with non access commands like
1546  *	test unit ready (so wrongly see the device as having a successful
1547  *	recovery)
1548  **/
1549 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1550 {
1551 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1552 
1553 	if (!scsi_device_online(scmd->device) ||
1554 	    !scsi_medium_access_command(scmd) ||
1555 	    host_byte(scmd->result) != DID_TIME_OUT ||
1556 	    eh_disp != SUCCESS)
1557 		return eh_disp;
1558 
1559 	/*
1560 	 * The device has timed out executing a medium access command.
1561 	 * However, the TEST UNIT READY command sent during error
1562 	 * handling completed successfully. Either the device is in the
1563 	 * process of recovering or has it suffered an internal failure
1564 	 * that prevents access to the storage medium.
1565 	 */
1566 	sdkp->medium_access_timed_out++;
1567 
1568 	/*
1569 	 * If the device keeps failing read/write commands but TEST UNIT
1570 	 * READY always completes successfully we assume that medium
1571 	 * access is no longer possible and take the device offline.
1572 	 */
1573 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1574 		scmd_printk(KERN_ERR, scmd,
1575 			    "Medium access timeout failure. Offlining disk!\n");
1576 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1577 
1578 		return FAILED;
1579 	}
1580 
1581 	return eh_disp;
1582 }
1583 
1584 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1585 {
1586 	u64 start_lba = blk_rq_pos(scmd->request);
1587 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1588 	u64 bad_lba;
1589 	int info_valid;
1590 	/*
1591 	 * resid is optional but mostly filled in.  When it's unused,
1592 	 * its value is zero, so we assume the whole buffer transferred
1593 	 */
1594 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1595 	unsigned int good_bytes;
1596 
1597 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1598 		return 0;
1599 
1600 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1601 					     SCSI_SENSE_BUFFERSIZE,
1602 					     &bad_lba);
1603 	if (!info_valid)
1604 		return 0;
1605 
1606 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1607 		return 0;
1608 
1609 	if (scmd->device->sector_size < 512) {
1610 		/* only legitimate sector_size here is 256 */
1611 		start_lba <<= 1;
1612 		end_lba <<= 1;
1613 	} else {
1614 		/* be careful ... don't want any overflows */
1615 		unsigned int factor = scmd->device->sector_size / 512;
1616 		do_div(start_lba, factor);
1617 		do_div(end_lba, factor);
1618 	}
1619 
1620 	/* The bad lba was reported incorrectly, we have no idea where
1621 	 * the error is.
1622 	 */
1623 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1624 		return 0;
1625 
1626 	/* This computation should always be done in terms of
1627 	 * the resolution of the device's medium.
1628 	 */
1629 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1630 	return min(good_bytes, transferred);
1631 }
1632 
1633 /**
1634  *	sd_done - bottom half handler: called when the lower level
1635  *	driver has completed (successfully or otherwise) a scsi command.
1636  *	@SCpnt: mid-level's per command structure.
1637  *
1638  *	Note: potentially run from within an ISR. Must not block.
1639  **/
1640 static int sd_done(struct scsi_cmnd *SCpnt)
1641 {
1642 	int result = SCpnt->result;
1643 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1644 	struct scsi_sense_hdr sshdr;
1645 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1646 	struct request *req = SCpnt->request;
1647 	int sense_valid = 0;
1648 	int sense_deferred = 0;
1649 	unsigned char op = SCpnt->cmnd[0];
1650 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1651 
1652 	if (req->cmd_flags & REQ_DISCARD || req->cmd_flags & REQ_WRITE_SAME) {
1653 		if (!result) {
1654 			good_bytes = blk_rq_bytes(req);
1655 			scsi_set_resid(SCpnt, 0);
1656 		} else {
1657 			good_bytes = 0;
1658 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1659 		}
1660 	}
1661 
1662 	if (result) {
1663 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1664 		if (sense_valid)
1665 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1666 	}
1667 #ifdef CONFIG_SCSI_LOGGING
1668 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1669 	if (sense_valid) {
1670 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1671 						   "sd_done: sb[respc,sk,asc,"
1672 						   "ascq]=%x,%x,%x,%x\n",
1673 						   sshdr.response_code,
1674 						   sshdr.sense_key, sshdr.asc,
1675 						   sshdr.ascq));
1676 	}
1677 #endif
1678 	sdkp->medium_access_timed_out = 0;
1679 
1680 	if (driver_byte(result) != DRIVER_SENSE &&
1681 	    (!sense_valid || sense_deferred))
1682 		goto out;
1683 
1684 	switch (sshdr.sense_key) {
1685 	case HARDWARE_ERROR:
1686 	case MEDIUM_ERROR:
1687 		good_bytes = sd_completed_bytes(SCpnt);
1688 		break;
1689 	case RECOVERED_ERROR:
1690 		good_bytes = scsi_bufflen(SCpnt);
1691 		break;
1692 	case NO_SENSE:
1693 		/* This indicates a false check condition, so ignore it.  An
1694 		 * unknown amount of data was transferred so treat it as an
1695 		 * error.
1696 		 */
1697 		scsi_print_sense("sd", SCpnt);
1698 		SCpnt->result = 0;
1699 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1700 		break;
1701 	case ABORTED_COMMAND:
1702 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1703 			good_bytes = sd_completed_bytes(SCpnt);
1704 		break;
1705 	case ILLEGAL_REQUEST:
1706 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1707 			good_bytes = sd_completed_bytes(SCpnt);
1708 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1709 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1710 			switch (op) {
1711 			case UNMAP:
1712 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1713 				break;
1714 			case WRITE_SAME_16:
1715 			case WRITE_SAME:
1716 				if (unmap)
1717 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1718 				else {
1719 					sdkp->device->no_write_same = 1;
1720 					sd_config_write_same(sdkp);
1721 
1722 					good_bytes = 0;
1723 					req->__data_len = blk_rq_bytes(req);
1724 					req->cmd_flags |= REQ_QUIET;
1725 				}
1726 			}
1727 		}
1728 		break;
1729 	default:
1730 		break;
1731 	}
1732  out:
1733 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1734 		sd_dif_complete(SCpnt, good_bytes);
1735 
1736 	return good_bytes;
1737 }
1738 
1739 /*
1740  * spinup disk - called only in sd_revalidate_disk()
1741  */
1742 static void
1743 sd_spinup_disk(struct scsi_disk *sdkp)
1744 {
1745 	unsigned char cmd[10];
1746 	unsigned long spintime_expire = 0;
1747 	int retries, spintime;
1748 	unsigned int the_result;
1749 	struct scsi_sense_hdr sshdr;
1750 	int sense_valid = 0;
1751 
1752 	spintime = 0;
1753 
1754 	/* Spin up drives, as required.  Only do this at boot time */
1755 	/* Spinup needs to be done for module loads too. */
1756 	do {
1757 		retries = 0;
1758 
1759 		do {
1760 			cmd[0] = TEST_UNIT_READY;
1761 			memset((void *) &cmd[1], 0, 9);
1762 
1763 			the_result = scsi_execute_req(sdkp->device, cmd,
1764 						      DMA_NONE, NULL, 0,
1765 						      &sshdr, SD_TIMEOUT,
1766 						      SD_MAX_RETRIES, NULL);
1767 
1768 			/*
1769 			 * If the drive has indicated to us that it
1770 			 * doesn't have any media in it, don't bother
1771 			 * with any more polling.
1772 			 */
1773 			if (media_not_present(sdkp, &sshdr))
1774 				return;
1775 
1776 			if (the_result)
1777 				sense_valid = scsi_sense_valid(&sshdr);
1778 			retries++;
1779 		} while (retries < 3 &&
1780 			 (!scsi_status_is_good(the_result) ||
1781 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1782 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1783 
1784 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1785 			/* no sense, TUR either succeeded or failed
1786 			 * with a status error */
1787 			if(!spintime && !scsi_status_is_good(the_result)) {
1788 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1789 				sd_print_result(sdkp, the_result);
1790 			}
1791 			break;
1792 		}
1793 
1794 		/*
1795 		 * The device does not want the automatic start to be issued.
1796 		 */
1797 		if (sdkp->device->no_start_on_add)
1798 			break;
1799 
1800 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1801 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1802 				break;	/* manual intervention required */
1803 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1804 				break;	/* standby */
1805 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1806 				break;	/* unavailable */
1807 			/*
1808 			 * Issue command to spin up drive when not ready
1809 			 */
1810 			if (!spintime) {
1811 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1812 				cmd[0] = START_STOP;
1813 				cmd[1] = 1;	/* Return immediately */
1814 				memset((void *) &cmd[2], 0, 8);
1815 				cmd[4] = 1;	/* Start spin cycle */
1816 				if (sdkp->device->start_stop_pwr_cond)
1817 					cmd[4] |= 1 << 4;
1818 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1819 						 NULL, 0, &sshdr,
1820 						 SD_TIMEOUT, SD_MAX_RETRIES,
1821 						 NULL);
1822 				spintime_expire = jiffies + 100 * HZ;
1823 				spintime = 1;
1824 			}
1825 			/* Wait 1 second for next try */
1826 			msleep(1000);
1827 			printk(".");
1828 
1829 		/*
1830 		 * Wait for USB flash devices with slow firmware.
1831 		 * Yes, this sense key/ASC combination shouldn't
1832 		 * occur here.  It's characteristic of these devices.
1833 		 */
1834 		} else if (sense_valid &&
1835 				sshdr.sense_key == UNIT_ATTENTION &&
1836 				sshdr.asc == 0x28) {
1837 			if (!spintime) {
1838 				spintime_expire = jiffies + 5 * HZ;
1839 				spintime = 1;
1840 			}
1841 			/* Wait 1 second for next try */
1842 			msleep(1000);
1843 		} else {
1844 			/* we don't understand the sense code, so it's
1845 			 * probably pointless to loop */
1846 			if(!spintime) {
1847 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1848 				sd_print_sense_hdr(sdkp, &sshdr);
1849 			}
1850 			break;
1851 		}
1852 
1853 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1854 
1855 	if (spintime) {
1856 		if (scsi_status_is_good(the_result))
1857 			printk("ready\n");
1858 		else
1859 			printk("not responding...\n");
1860 	}
1861 }
1862 
1863 
1864 /*
1865  * Determine whether disk supports Data Integrity Field.
1866  */
1867 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1868 {
1869 	struct scsi_device *sdp = sdkp->device;
1870 	u8 type;
1871 	int ret = 0;
1872 
1873 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1874 		return ret;
1875 
1876 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1877 
1878 	if (type > SD_DIF_TYPE3_PROTECTION)
1879 		ret = -ENODEV;
1880 	else if (scsi_host_dif_capable(sdp->host, type))
1881 		ret = 1;
1882 
1883 	if (sdkp->first_scan || type != sdkp->protection_type)
1884 		switch (ret) {
1885 		case -ENODEV:
1886 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
1887 				  " protection type %u. Disabling disk!\n",
1888 				  type);
1889 			break;
1890 		case 1:
1891 			sd_printk(KERN_NOTICE, sdkp,
1892 				  "Enabling DIF Type %u protection\n", type);
1893 			break;
1894 		case 0:
1895 			sd_printk(KERN_NOTICE, sdkp,
1896 				  "Disabling DIF Type %u protection\n", type);
1897 			break;
1898 		}
1899 
1900 	sdkp->protection_type = type;
1901 
1902 	return ret;
1903 }
1904 
1905 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1906 			struct scsi_sense_hdr *sshdr, int sense_valid,
1907 			int the_result)
1908 {
1909 	sd_print_result(sdkp, the_result);
1910 	if (driver_byte(the_result) & DRIVER_SENSE)
1911 		sd_print_sense_hdr(sdkp, sshdr);
1912 	else
1913 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1914 
1915 	/*
1916 	 * Set dirty bit for removable devices if not ready -
1917 	 * sometimes drives will not report this properly.
1918 	 */
1919 	if (sdp->removable &&
1920 	    sense_valid && sshdr->sense_key == NOT_READY)
1921 		set_media_not_present(sdkp);
1922 
1923 	/*
1924 	 * We used to set media_present to 0 here to indicate no media
1925 	 * in the drive, but some drives fail read capacity even with
1926 	 * media present, so we can't do that.
1927 	 */
1928 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1929 }
1930 
1931 #define RC16_LEN 32
1932 #if RC16_LEN > SD_BUF_SIZE
1933 #error RC16_LEN must not be more than SD_BUF_SIZE
1934 #endif
1935 
1936 #define READ_CAPACITY_RETRIES_ON_RESET	10
1937 
1938 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1939 						unsigned char *buffer)
1940 {
1941 	unsigned char cmd[16];
1942 	struct scsi_sense_hdr sshdr;
1943 	int sense_valid = 0;
1944 	int the_result;
1945 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1946 	unsigned int alignment;
1947 	unsigned long long lba;
1948 	unsigned sector_size;
1949 
1950 	if (sdp->no_read_capacity_16)
1951 		return -EINVAL;
1952 
1953 	do {
1954 		memset(cmd, 0, 16);
1955 		cmd[0] = SERVICE_ACTION_IN;
1956 		cmd[1] = SAI_READ_CAPACITY_16;
1957 		cmd[13] = RC16_LEN;
1958 		memset(buffer, 0, RC16_LEN);
1959 
1960 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1961 					buffer, RC16_LEN, &sshdr,
1962 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1963 
1964 		if (media_not_present(sdkp, &sshdr))
1965 			return -ENODEV;
1966 
1967 		if (the_result) {
1968 			sense_valid = scsi_sense_valid(&sshdr);
1969 			if (sense_valid &&
1970 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1971 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1972 			    sshdr.ascq == 0x00)
1973 				/* Invalid Command Operation Code or
1974 				 * Invalid Field in CDB, just retry
1975 				 * silently with RC10 */
1976 				return -EINVAL;
1977 			if (sense_valid &&
1978 			    sshdr.sense_key == UNIT_ATTENTION &&
1979 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1980 				/* Device reset might occur several times,
1981 				 * give it one more chance */
1982 				if (--reset_retries > 0)
1983 					continue;
1984 		}
1985 		retries--;
1986 
1987 	} while (the_result && retries);
1988 
1989 	if (the_result) {
1990 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1991 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1992 		return -EINVAL;
1993 	}
1994 
1995 	sector_size = get_unaligned_be32(&buffer[8]);
1996 	lba = get_unaligned_be64(&buffer[0]);
1997 
1998 	if (sd_read_protection_type(sdkp, buffer) < 0) {
1999 		sdkp->capacity = 0;
2000 		return -ENODEV;
2001 	}
2002 
2003 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2004 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2005 			"kernel compiled with support for large block "
2006 			"devices.\n");
2007 		sdkp->capacity = 0;
2008 		return -EOVERFLOW;
2009 	}
2010 
2011 	/* Logical blocks per physical block exponent */
2012 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2013 
2014 	/* Lowest aligned logical block */
2015 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2016 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2017 	if (alignment && sdkp->first_scan)
2018 		sd_printk(KERN_NOTICE, sdkp,
2019 			  "physical block alignment offset: %u\n", alignment);
2020 
2021 	if (buffer[14] & 0x80) { /* LBPME */
2022 		sdkp->lbpme = 1;
2023 
2024 		if (buffer[14] & 0x40) /* LBPRZ */
2025 			sdkp->lbprz = 1;
2026 
2027 		sd_config_discard(sdkp, SD_LBP_WS16);
2028 	}
2029 
2030 	sdkp->capacity = lba + 1;
2031 	return sector_size;
2032 }
2033 
2034 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2035 						unsigned char *buffer)
2036 {
2037 	unsigned char cmd[16];
2038 	struct scsi_sense_hdr sshdr;
2039 	int sense_valid = 0;
2040 	int the_result;
2041 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2042 	sector_t lba;
2043 	unsigned sector_size;
2044 
2045 	do {
2046 		cmd[0] = READ_CAPACITY;
2047 		memset(&cmd[1], 0, 9);
2048 		memset(buffer, 0, 8);
2049 
2050 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2051 					buffer, 8, &sshdr,
2052 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2053 
2054 		if (media_not_present(sdkp, &sshdr))
2055 			return -ENODEV;
2056 
2057 		if (the_result) {
2058 			sense_valid = scsi_sense_valid(&sshdr);
2059 			if (sense_valid &&
2060 			    sshdr.sense_key == UNIT_ATTENTION &&
2061 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2062 				/* Device reset might occur several times,
2063 				 * give it one more chance */
2064 				if (--reset_retries > 0)
2065 					continue;
2066 		}
2067 		retries--;
2068 
2069 	} while (the_result && retries);
2070 
2071 	if (the_result) {
2072 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
2073 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2074 		return -EINVAL;
2075 	}
2076 
2077 	sector_size = get_unaligned_be32(&buffer[4]);
2078 	lba = get_unaligned_be32(&buffer[0]);
2079 
2080 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2081 		/* Some buggy (usb cardreader) devices return an lba of
2082 		   0xffffffff when the want to report a size of 0 (with
2083 		   which they really mean no media is present) */
2084 		sdkp->capacity = 0;
2085 		sdkp->physical_block_size = sector_size;
2086 		return sector_size;
2087 	}
2088 
2089 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2090 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2091 			"kernel compiled with support for large block "
2092 			"devices.\n");
2093 		sdkp->capacity = 0;
2094 		return -EOVERFLOW;
2095 	}
2096 
2097 	sdkp->capacity = lba + 1;
2098 	sdkp->physical_block_size = sector_size;
2099 	return sector_size;
2100 }
2101 
2102 static int sd_try_rc16_first(struct scsi_device *sdp)
2103 {
2104 	if (sdp->host->max_cmd_len < 16)
2105 		return 0;
2106 	if (sdp->try_rc_10_first)
2107 		return 0;
2108 	if (sdp->scsi_level > SCSI_SPC_2)
2109 		return 1;
2110 	if (scsi_device_protection(sdp))
2111 		return 1;
2112 	return 0;
2113 }
2114 
2115 /*
2116  * read disk capacity
2117  */
2118 static void
2119 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2120 {
2121 	int sector_size;
2122 	struct scsi_device *sdp = sdkp->device;
2123 	sector_t old_capacity = sdkp->capacity;
2124 
2125 	if (sd_try_rc16_first(sdp)) {
2126 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2127 		if (sector_size == -EOVERFLOW)
2128 			goto got_data;
2129 		if (sector_size == -ENODEV)
2130 			return;
2131 		if (sector_size < 0)
2132 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2133 		if (sector_size < 0)
2134 			return;
2135 	} else {
2136 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2137 		if (sector_size == -EOVERFLOW)
2138 			goto got_data;
2139 		if (sector_size < 0)
2140 			return;
2141 		if ((sizeof(sdkp->capacity) > 4) &&
2142 		    (sdkp->capacity > 0xffffffffULL)) {
2143 			int old_sector_size = sector_size;
2144 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2145 					"Trying to use READ CAPACITY(16).\n");
2146 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2147 			if (sector_size < 0) {
2148 				sd_printk(KERN_NOTICE, sdkp,
2149 					"Using 0xffffffff as device size\n");
2150 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2151 				sector_size = old_sector_size;
2152 				goto got_data;
2153 			}
2154 		}
2155 	}
2156 
2157 	/* Some devices are known to return the total number of blocks,
2158 	 * not the highest block number.  Some devices have versions
2159 	 * which do this and others which do not.  Some devices we might
2160 	 * suspect of doing this but we don't know for certain.
2161 	 *
2162 	 * If we know the reported capacity is wrong, decrement it.  If
2163 	 * we can only guess, then assume the number of blocks is even
2164 	 * (usually true but not always) and err on the side of lowering
2165 	 * the capacity.
2166 	 */
2167 	if (sdp->fix_capacity ||
2168 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2169 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2170 				"from its reported value: %llu\n",
2171 				(unsigned long long) sdkp->capacity);
2172 		--sdkp->capacity;
2173 	}
2174 
2175 got_data:
2176 	if (sector_size == 0) {
2177 		sector_size = 512;
2178 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2179 			  "assuming 512.\n");
2180 	}
2181 
2182 	if (sector_size != 512 &&
2183 	    sector_size != 1024 &&
2184 	    sector_size != 2048 &&
2185 	    sector_size != 4096 &&
2186 	    sector_size != 256) {
2187 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2188 			  sector_size);
2189 		/*
2190 		 * The user might want to re-format the drive with
2191 		 * a supported sectorsize.  Once this happens, it
2192 		 * would be relatively trivial to set the thing up.
2193 		 * For this reason, we leave the thing in the table.
2194 		 */
2195 		sdkp->capacity = 0;
2196 		/*
2197 		 * set a bogus sector size so the normal read/write
2198 		 * logic in the block layer will eventually refuse any
2199 		 * request on this device without tripping over power
2200 		 * of two sector size assumptions
2201 		 */
2202 		sector_size = 512;
2203 	}
2204 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2205 
2206 	{
2207 		char cap_str_2[10], cap_str_10[10];
2208 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
2209 
2210 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
2211 				sizeof(cap_str_2));
2212 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
2213 				sizeof(cap_str_10));
2214 
2215 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2216 			sd_printk(KERN_NOTICE, sdkp,
2217 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2218 				  (unsigned long long)sdkp->capacity,
2219 				  sector_size, cap_str_10, cap_str_2);
2220 
2221 			if (sdkp->physical_block_size != sector_size)
2222 				sd_printk(KERN_NOTICE, sdkp,
2223 					  "%u-byte physical blocks\n",
2224 					  sdkp->physical_block_size);
2225 		}
2226 	}
2227 
2228 	sdp->use_16_for_rw = (sdkp->capacity > 0xffffffff);
2229 
2230 	/* Rescale capacity to 512-byte units */
2231 	if (sector_size == 4096)
2232 		sdkp->capacity <<= 3;
2233 	else if (sector_size == 2048)
2234 		sdkp->capacity <<= 2;
2235 	else if (sector_size == 1024)
2236 		sdkp->capacity <<= 1;
2237 	else if (sector_size == 256)
2238 		sdkp->capacity >>= 1;
2239 
2240 	blk_queue_physical_block_size(sdp->request_queue,
2241 				      sdkp->physical_block_size);
2242 	sdkp->device->sector_size = sector_size;
2243 }
2244 
2245 /* called with buffer of length 512 */
2246 static inline int
2247 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2248 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2249 		 struct scsi_sense_hdr *sshdr)
2250 {
2251 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2252 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2253 			       sshdr);
2254 }
2255 
2256 /*
2257  * read write protect setting, if possible - called only in sd_revalidate_disk()
2258  * called with buffer of length SD_BUF_SIZE
2259  */
2260 static void
2261 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2262 {
2263 	int res;
2264 	struct scsi_device *sdp = sdkp->device;
2265 	struct scsi_mode_data data;
2266 	int old_wp = sdkp->write_prot;
2267 
2268 	set_disk_ro(sdkp->disk, 0);
2269 	if (sdp->skip_ms_page_3f) {
2270 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2271 		return;
2272 	}
2273 
2274 	if (sdp->use_192_bytes_for_3f) {
2275 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2276 	} else {
2277 		/*
2278 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2279 		 * We have to start carefully: some devices hang if we ask
2280 		 * for more than is available.
2281 		 */
2282 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2283 
2284 		/*
2285 		 * Second attempt: ask for page 0 When only page 0 is
2286 		 * implemented, a request for page 3F may return Sense Key
2287 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2288 		 * CDB.
2289 		 */
2290 		if (!scsi_status_is_good(res))
2291 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2292 
2293 		/*
2294 		 * Third attempt: ask 255 bytes, as we did earlier.
2295 		 */
2296 		if (!scsi_status_is_good(res))
2297 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2298 					       &data, NULL);
2299 	}
2300 
2301 	if (!scsi_status_is_good(res)) {
2302 		sd_first_printk(KERN_WARNING, sdkp,
2303 			  "Test WP failed, assume Write Enabled\n");
2304 	} else {
2305 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2306 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2307 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2308 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2309 				  sdkp->write_prot ? "on" : "off");
2310 			sd_printk(KERN_DEBUG, sdkp,
2311 				  "Mode Sense: %02x %02x %02x %02x\n",
2312 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2313 		}
2314 	}
2315 }
2316 
2317 /*
2318  * sd_read_cache_type - called only from sd_revalidate_disk()
2319  * called with buffer of length SD_BUF_SIZE
2320  */
2321 static void
2322 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2323 {
2324 	int len = 0, res;
2325 	struct scsi_device *sdp = sdkp->device;
2326 
2327 	int dbd;
2328 	int modepage;
2329 	int first_len;
2330 	struct scsi_mode_data data;
2331 	struct scsi_sense_hdr sshdr;
2332 	int old_wce = sdkp->WCE;
2333 	int old_rcd = sdkp->RCD;
2334 	int old_dpofua = sdkp->DPOFUA;
2335 
2336 
2337 	if (sdkp->cache_override)
2338 		return;
2339 
2340 	first_len = 4;
2341 	if (sdp->skip_ms_page_8) {
2342 		if (sdp->type == TYPE_RBC)
2343 			goto defaults;
2344 		else {
2345 			if (sdp->skip_ms_page_3f)
2346 				goto defaults;
2347 			modepage = 0x3F;
2348 			if (sdp->use_192_bytes_for_3f)
2349 				first_len = 192;
2350 			dbd = 0;
2351 		}
2352 	} else if (sdp->type == TYPE_RBC) {
2353 		modepage = 6;
2354 		dbd = 8;
2355 	} else {
2356 		modepage = 8;
2357 		dbd = 0;
2358 	}
2359 
2360 	/* cautiously ask */
2361 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2362 			&data, &sshdr);
2363 
2364 	if (!scsi_status_is_good(res))
2365 		goto bad_sense;
2366 
2367 	if (!data.header_length) {
2368 		modepage = 6;
2369 		first_len = 0;
2370 		sd_first_printk(KERN_ERR, sdkp,
2371 				"Missing header in MODE_SENSE response\n");
2372 	}
2373 
2374 	/* that went OK, now ask for the proper length */
2375 	len = data.length;
2376 
2377 	/*
2378 	 * We're only interested in the first three bytes, actually.
2379 	 * But the data cache page is defined for the first 20.
2380 	 */
2381 	if (len < 3)
2382 		goto bad_sense;
2383 	else if (len > SD_BUF_SIZE) {
2384 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2385 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2386 		len = SD_BUF_SIZE;
2387 	}
2388 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2389 		len = 192;
2390 
2391 	/* Get the data */
2392 	if (len > first_len)
2393 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2394 				&data, &sshdr);
2395 
2396 	if (scsi_status_is_good(res)) {
2397 		int offset = data.header_length + data.block_descriptor_length;
2398 
2399 		while (offset < len) {
2400 			u8 page_code = buffer[offset] & 0x3F;
2401 			u8 spf       = buffer[offset] & 0x40;
2402 
2403 			if (page_code == 8 || page_code == 6) {
2404 				/* We're interested only in the first 3 bytes.
2405 				 */
2406 				if (len - offset <= 2) {
2407 					sd_first_printk(KERN_ERR, sdkp,
2408 						"Incomplete mode parameter "
2409 							"data\n");
2410 					goto defaults;
2411 				} else {
2412 					modepage = page_code;
2413 					goto Page_found;
2414 				}
2415 			} else {
2416 				/* Go to the next page */
2417 				if (spf && len - offset > 3)
2418 					offset += 4 + (buffer[offset+2] << 8) +
2419 						buffer[offset+3];
2420 				else if (!spf && len - offset > 1)
2421 					offset += 2 + buffer[offset+1];
2422 				else {
2423 					sd_first_printk(KERN_ERR, sdkp,
2424 							"Incomplete mode "
2425 							"parameter data\n");
2426 					goto defaults;
2427 				}
2428 			}
2429 		}
2430 
2431 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2432 		goto defaults;
2433 
2434 	Page_found:
2435 		if (modepage == 8) {
2436 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2437 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2438 		} else {
2439 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2440 			sdkp->RCD = 0;
2441 		}
2442 
2443 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2444 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2445 			sd_first_printk(KERN_NOTICE, sdkp,
2446 				  "Uses READ/WRITE(6), disabling FUA\n");
2447 			sdkp->DPOFUA = 0;
2448 		}
2449 
2450 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2451 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2452 			sd_printk(KERN_NOTICE, sdkp,
2453 				  "Write cache: %s, read cache: %s, %s\n",
2454 				  sdkp->WCE ? "enabled" : "disabled",
2455 				  sdkp->RCD ? "disabled" : "enabled",
2456 				  sdkp->DPOFUA ? "supports DPO and FUA"
2457 				  : "doesn't support DPO or FUA");
2458 
2459 		return;
2460 	}
2461 
2462 bad_sense:
2463 	if (scsi_sense_valid(&sshdr) &&
2464 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2465 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2466 		/* Invalid field in CDB */
2467 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2468 	else
2469 		sd_first_printk(KERN_ERR, sdkp,
2470 				"Asking for cache data failed\n");
2471 
2472 defaults:
2473 	if (sdp->wce_default_on) {
2474 		sd_first_printk(KERN_NOTICE, sdkp,
2475 				"Assuming drive cache: write back\n");
2476 		sdkp->WCE = 1;
2477 	} else {
2478 		sd_first_printk(KERN_ERR, sdkp,
2479 				"Assuming drive cache: write through\n");
2480 		sdkp->WCE = 0;
2481 	}
2482 	sdkp->RCD = 0;
2483 	sdkp->DPOFUA = 0;
2484 }
2485 
2486 /*
2487  * The ATO bit indicates whether the DIF application tag is available
2488  * for use by the operating system.
2489  */
2490 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2491 {
2492 	int res, offset;
2493 	struct scsi_device *sdp = sdkp->device;
2494 	struct scsi_mode_data data;
2495 	struct scsi_sense_hdr sshdr;
2496 
2497 	if (sdp->type != TYPE_DISK)
2498 		return;
2499 
2500 	if (sdkp->protection_type == 0)
2501 		return;
2502 
2503 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2504 			      SD_MAX_RETRIES, &data, &sshdr);
2505 
2506 	if (!scsi_status_is_good(res) || !data.header_length ||
2507 	    data.length < 6) {
2508 		sd_first_printk(KERN_WARNING, sdkp,
2509 			  "getting Control mode page failed, assume no ATO\n");
2510 
2511 		if (scsi_sense_valid(&sshdr))
2512 			sd_print_sense_hdr(sdkp, &sshdr);
2513 
2514 		return;
2515 	}
2516 
2517 	offset = data.header_length + data.block_descriptor_length;
2518 
2519 	if ((buffer[offset] & 0x3f) != 0x0a) {
2520 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2521 		return;
2522 	}
2523 
2524 	if ((buffer[offset + 5] & 0x80) == 0)
2525 		return;
2526 
2527 	sdkp->ATO = 1;
2528 
2529 	return;
2530 }
2531 
2532 /**
2533  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2534  * @disk: disk to query
2535  */
2536 static void sd_read_block_limits(struct scsi_disk *sdkp)
2537 {
2538 	unsigned int sector_sz = sdkp->device->sector_size;
2539 	const int vpd_len = 64;
2540 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2541 
2542 	if (!buffer ||
2543 	    /* Block Limits VPD */
2544 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2545 		goto out;
2546 
2547 	blk_queue_io_min(sdkp->disk->queue,
2548 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2549 	blk_queue_io_opt(sdkp->disk->queue,
2550 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2551 
2552 	if (buffer[3] == 0x3c) {
2553 		unsigned int lba_count, desc_count;
2554 
2555 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2556 
2557 		if (!sdkp->lbpme)
2558 			goto out;
2559 
2560 		lba_count = get_unaligned_be32(&buffer[20]);
2561 		desc_count = get_unaligned_be32(&buffer[24]);
2562 
2563 		if (lba_count && desc_count)
2564 			sdkp->max_unmap_blocks = lba_count;
2565 
2566 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2567 
2568 		if (buffer[32] & 0x80)
2569 			sdkp->unmap_alignment =
2570 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2571 
2572 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2573 
2574 			if (sdkp->max_unmap_blocks)
2575 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2576 			else
2577 				sd_config_discard(sdkp, SD_LBP_WS16);
2578 
2579 		} else {	/* LBP VPD page tells us what to use */
2580 
2581 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2582 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2583 			else if (sdkp->lbpws)
2584 				sd_config_discard(sdkp, SD_LBP_WS16);
2585 			else if (sdkp->lbpws10)
2586 				sd_config_discard(sdkp, SD_LBP_WS10);
2587 			else
2588 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2589 		}
2590 	}
2591 
2592  out:
2593 	kfree(buffer);
2594 }
2595 
2596 /**
2597  * sd_read_block_characteristics - Query block dev. characteristics
2598  * @disk: disk to query
2599  */
2600 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2601 {
2602 	unsigned char *buffer;
2603 	u16 rot;
2604 	const int vpd_len = 64;
2605 
2606 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2607 
2608 	if (!buffer ||
2609 	    /* Block Device Characteristics VPD */
2610 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2611 		goto out;
2612 
2613 	rot = get_unaligned_be16(&buffer[4]);
2614 
2615 	if (rot == 1)
2616 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2617 
2618  out:
2619 	kfree(buffer);
2620 }
2621 
2622 /**
2623  * sd_read_block_provisioning - Query provisioning VPD page
2624  * @disk: disk to query
2625  */
2626 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2627 {
2628 	unsigned char *buffer;
2629 	const int vpd_len = 8;
2630 
2631 	if (sdkp->lbpme == 0)
2632 		return;
2633 
2634 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2635 
2636 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2637 		goto out;
2638 
2639 	sdkp->lbpvpd	= 1;
2640 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2641 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2642 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2643 
2644  out:
2645 	kfree(buffer);
2646 }
2647 
2648 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2649 {
2650 	struct scsi_device *sdev = sdkp->device;
2651 
2652 	if (sdev->host->no_write_same) {
2653 		sdev->no_write_same = 1;
2654 
2655 		return;
2656 	}
2657 
2658 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2659 		/* too large values might cause issues with arcmsr */
2660 		int vpd_buf_len = 64;
2661 
2662 		sdev->no_report_opcodes = 1;
2663 
2664 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2665 		 * CODES is unsupported and the device has an ATA
2666 		 * Information VPD page (SAT).
2667 		 */
2668 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2669 			sdev->no_write_same = 1;
2670 	}
2671 
2672 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2673 		sdkp->ws16 = 1;
2674 
2675 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2676 		sdkp->ws10 = 1;
2677 }
2678 
2679 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2680 {
2681 	/*
2682 	 * Although VPD inquiries can go to SCSI-2 type devices,
2683 	 * some USB ones crash on receiving them, and the pages
2684 	 * we currently ask for are for SPC-3 and beyond
2685 	 */
2686 	if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2687 		return 1;
2688 	return 0;
2689 }
2690 
2691 /**
2692  *	sd_revalidate_disk - called the first time a new disk is seen,
2693  *	performs disk spin up, read_capacity, etc.
2694  *	@disk: struct gendisk we care about
2695  **/
2696 static int sd_revalidate_disk(struct gendisk *disk)
2697 {
2698 	struct scsi_disk *sdkp = scsi_disk(disk);
2699 	struct scsi_device *sdp = sdkp->device;
2700 	unsigned char *buffer;
2701 	unsigned flush = 0;
2702 
2703 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2704 				      "sd_revalidate_disk\n"));
2705 
2706 	/*
2707 	 * If the device is offline, don't try and read capacity or any
2708 	 * of the other niceties.
2709 	 */
2710 	if (!scsi_device_online(sdp))
2711 		goto out;
2712 
2713 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2714 	if (!buffer) {
2715 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2716 			  "allocation failure.\n");
2717 		goto out;
2718 	}
2719 
2720 	sd_spinup_disk(sdkp);
2721 
2722 	/*
2723 	 * Without media there is no reason to ask; moreover, some devices
2724 	 * react badly if we do.
2725 	 */
2726 	if (sdkp->media_present) {
2727 		sd_read_capacity(sdkp, buffer);
2728 
2729 		if (sd_try_extended_inquiry(sdp)) {
2730 			sd_read_block_provisioning(sdkp);
2731 			sd_read_block_limits(sdkp);
2732 			sd_read_block_characteristics(sdkp);
2733 		}
2734 
2735 		sd_read_write_protect_flag(sdkp, buffer);
2736 		sd_read_cache_type(sdkp, buffer);
2737 		sd_read_app_tag_own(sdkp, buffer);
2738 		sd_read_write_same(sdkp, buffer);
2739 	}
2740 
2741 	sdkp->first_scan = 0;
2742 
2743 	/*
2744 	 * We now have all cache related info, determine how we deal
2745 	 * with flush requests.
2746 	 */
2747 	if (sdkp->WCE) {
2748 		flush |= REQ_FLUSH;
2749 		if (sdkp->DPOFUA)
2750 			flush |= REQ_FUA;
2751 	}
2752 
2753 	blk_queue_flush(sdkp->disk->queue, flush);
2754 
2755 	set_capacity(disk, sdkp->capacity);
2756 	sd_config_write_same(sdkp);
2757 	kfree(buffer);
2758 
2759  out:
2760 	return 0;
2761 }
2762 
2763 /**
2764  *	sd_unlock_native_capacity - unlock native capacity
2765  *	@disk: struct gendisk to set capacity for
2766  *
2767  *	Block layer calls this function if it detects that partitions
2768  *	on @disk reach beyond the end of the device.  If the SCSI host
2769  *	implements ->unlock_native_capacity() method, it's invoked to
2770  *	give it a chance to adjust the device capacity.
2771  *
2772  *	CONTEXT:
2773  *	Defined by block layer.  Might sleep.
2774  */
2775 static void sd_unlock_native_capacity(struct gendisk *disk)
2776 {
2777 	struct scsi_device *sdev = scsi_disk(disk)->device;
2778 
2779 	if (sdev->host->hostt->unlock_native_capacity)
2780 		sdev->host->hostt->unlock_native_capacity(sdev);
2781 }
2782 
2783 /**
2784  *	sd_format_disk_name - format disk name
2785  *	@prefix: name prefix - ie. "sd" for SCSI disks
2786  *	@index: index of the disk to format name for
2787  *	@buf: output buffer
2788  *	@buflen: length of the output buffer
2789  *
2790  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2791  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2792  *	which is followed by sdaaa.
2793  *
2794  *	This is basically 26 base counting with one extra 'nil' entry
2795  *	at the beginning from the second digit on and can be
2796  *	determined using similar method as 26 base conversion with the
2797  *	index shifted -1 after each digit is computed.
2798  *
2799  *	CONTEXT:
2800  *	Don't care.
2801  *
2802  *	RETURNS:
2803  *	0 on success, -errno on failure.
2804  */
2805 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2806 {
2807 	const int base = 'z' - 'a' + 1;
2808 	char *begin = buf + strlen(prefix);
2809 	char *end = buf + buflen;
2810 	char *p;
2811 	int unit;
2812 
2813 	p = end - 1;
2814 	*p = '\0';
2815 	unit = base;
2816 	do {
2817 		if (p == begin)
2818 			return -EINVAL;
2819 		*--p = 'a' + (index % unit);
2820 		index = (index / unit) - 1;
2821 	} while (index >= 0);
2822 
2823 	memmove(begin, p, end - p);
2824 	memcpy(buf, prefix, strlen(prefix));
2825 
2826 	return 0;
2827 }
2828 
2829 /*
2830  * The asynchronous part of sd_probe
2831  */
2832 static void sd_probe_async(void *data, async_cookie_t cookie)
2833 {
2834 	struct scsi_disk *sdkp = data;
2835 	struct scsi_device *sdp;
2836 	struct gendisk *gd;
2837 	u32 index;
2838 	struct device *dev;
2839 
2840 	sdp = sdkp->device;
2841 	gd = sdkp->disk;
2842 	index = sdkp->index;
2843 	dev = &sdp->sdev_gendev;
2844 
2845 	gd->major = sd_major((index & 0xf0) >> 4);
2846 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2847 	gd->minors = SD_MINORS;
2848 
2849 	gd->fops = &sd_fops;
2850 	gd->private_data = &sdkp->driver;
2851 	gd->queue = sdkp->device->request_queue;
2852 
2853 	/* defaults, until the device tells us otherwise */
2854 	sdp->sector_size = 512;
2855 	sdkp->capacity = 0;
2856 	sdkp->media_present = 1;
2857 	sdkp->write_prot = 0;
2858 	sdkp->cache_override = 0;
2859 	sdkp->WCE = 0;
2860 	sdkp->RCD = 0;
2861 	sdkp->ATO = 0;
2862 	sdkp->first_scan = 1;
2863 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2864 
2865 	sd_revalidate_disk(gd);
2866 
2867 	gd->driverfs_dev = &sdp->sdev_gendev;
2868 	gd->flags = GENHD_FL_EXT_DEVT;
2869 	if (sdp->removable) {
2870 		gd->flags |= GENHD_FL_REMOVABLE;
2871 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2872 	}
2873 
2874 	blk_pm_runtime_init(sdp->request_queue, dev);
2875 	add_disk(gd);
2876 	if (sdkp->capacity)
2877 		sd_dif_config_host(sdkp);
2878 
2879 	sd_revalidate_disk(gd);
2880 
2881 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2882 		  sdp->removable ? "removable " : "");
2883 	scsi_autopm_put_device(sdp);
2884 	put_device(&sdkp->dev);
2885 }
2886 
2887 /**
2888  *	sd_probe - called during driver initialization and whenever a
2889  *	new scsi device is attached to the system. It is called once
2890  *	for each scsi device (not just disks) present.
2891  *	@dev: pointer to device object
2892  *
2893  *	Returns 0 if successful (or not interested in this scsi device
2894  *	(e.g. scanner)); 1 when there is an error.
2895  *
2896  *	Note: this function is invoked from the scsi mid-level.
2897  *	This function sets up the mapping between a given
2898  *	<host,channel,id,lun> (found in sdp) and new device name
2899  *	(e.g. /dev/sda). More precisely it is the block device major
2900  *	and minor number that is chosen here.
2901  *
2902  *	Assume sd_probe is not re-entrant (for time being)
2903  *	Also think about sd_probe() and sd_remove() running coincidentally.
2904  **/
2905 static int sd_probe(struct device *dev)
2906 {
2907 	struct scsi_device *sdp = to_scsi_device(dev);
2908 	struct scsi_disk *sdkp;
2909 	struct gendisk *gd;
2910 	int index;
2911 	int error;
2912 
2913 	error = -ENODEV;
2914 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2915 		goto out;
2916 
2917 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2918 					"sd_probe\n"));
2919 
2920 	error = -ENOMEM;
2921 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2922 	if (!sdkp)
2923 		goto out;
2924 
2925 	gd = alloc_disk(SD_MINORS);
2926 	if (!gd)
2927 		goto out_free;
2928 
2929 	do {
2930 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2931 			goto out_put;
2932 
2933 		spin_lock(&sd_index_lock);
2934 		error = ida_get_new(&sd_index_ida, &index);
2935 		spin_unlock(&sd_index_lock);
2936 	} while (error == -EAGAIN);
2937 
2938 	if (error) {
2939 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2940 		goto out_put;
2941 	}
2942 
2943 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2944 	if (error) {
2945 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2946 		goto out_free_index;
2947 	}
2948 
2949 	sdkp->device = sdp;
2950 	sdkp->driver = &sd_template;
2951 	sdkp->disk = gd;
2952 	sdkp->index = index;
2953 	atomic_set(&sdkp->openers, 0);
2954 	atomic_set(&sdkp->device->ioerr_cnt, 0);
2955 
2956 	if (!sdp->request_queue->rq_timeout) {
2957 		if (sdp->type != TYPE_MOD)
2958 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2959 		else
2960 			blk_queue_rq_timeout(sdp->request_queue,
2961 					     SD_MOD_TIMEOUT);
2962 	}
2963 
2964 	device_initialize(&sdkp->dev);
2965 	sdkp->dev.parent = dev;
2966 	sdkp->dev.class = &sd_disk_class;
2967 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
2968 
2969 	if (device_add(&sdkp->dev))
2970 		goto out_free_index;
2971 
2972 	get_device(dev);
2973 	dev_set_drvdata(dev, sdkp);
2974 
2975 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2976 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
2977 
2978 	return 0;
2979 
2980  out_free_index:
2981 	spin_lock(&sd_index_lock);
2982 	ida_remove(&sd_index_ida, index);
2983 	spin_unlock(&sd_index_lock);
2984  out_put:
2985 	put_disk(gd);
2986  out_free:
2987 	kfree(sdkp);
2988  out:
2989 	return error;
2990 }
2991 
2992 /**
2993  *	sd_remove - called whenever a scsi disk (previously recognized by
2994  *	sd_probe) is detached from the system. It is called (potentially
2995  *	multiple times) during sd module unload.
2996  *	@sdp: pointer to mid level scsi device object
2997  *
2998  *	Note: this function is invoked from the scsi mid-level.
2999  *	This function potentially frees up a device name (e.g. /dev/sdc)
3000  *	that could be re-used by a subsequent sd_probe().
3001  *	This function is not called when the built-in sd driver is "exit-ed".
3002  **/
3003 static int sd_remove(struct device *dev)
3004 {
3005 	struct scsi_disk *sdkp;
3006 	dev_t devt;
3007 
3008 	sdkp = dev_get_drvdata(dev);
3009 	devt = disk_devt(sdkp->disk);
3010 	scsi_autopm_get_device(sdkp->device);
3011 
3012 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3013 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3014 	device_del(&sdkp->dev);
3015 	del_gendisk(sdkp->disk);
3016 	sd_shutdown(dev);
3017 
3018 	blk_register_region(devt, SD_MINORS, NULL,
3019 			    sd_default_probe, NULL, NULL);
3020 
3021 	mutex_lock(&sd_ref_mutex);
3022 	dev_set_drvdata(dev, NULL);
3023 	put_device(&sdkp->dev);
3024 	mutex_unlock(&sd_ref_mutex);
3025 
3026 	return 0;
3027 }
3028 
3029 /**
3030  *	scsi_disk_release - Called to free the scsi_disk structure
3031  *	@dev: pointer to embedded class device
3032  *
3033  *	sd_ref_mutex must be held entering this routine.  Because it is
3034  *	called on last put, you should always use the scsi_disk_get()
3035  *	scsi_disk_put() helpers which manipulate the semaphore directly
3036  *	and never do a direct put_device.
3037  **/
3038 static void scsi_disk_release(struct device *dev)
3039 {
3040 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3041 	struct gendisk *disk = sdkp->disk;
3042 
3043 	spin_lock(&sd_index_lock);
3044 	ida_remove(&sd_index_ida, sdkp->index);
3045 	spin_unlock(&sd_index_lock);
3046 
3047 	disk->private_data = NULL;
3048 	put_disk(disk);
3049 	put_device(&sdkp->device->sdev_gendev);
3050 
3051 	kfree(sdkp);
3052 }
3053 
3054 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3055 {
3056 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3057 	struct scsi_sense_hdr sshdr;
3058 	struct scsi_device *sdp = sdkp->device;
3059 	int res;
3060 
3061 	if (start)
3062 		cmd[4] |= 1;	/* START */
3063 
3064 	if (sdp->start_stop_pwr_cond)
3065 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3066 
3067 	if (!scsi_device_online(sdp))
3068 		return -ENODEV;
3069 
3070 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3071 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3072 	if (res) {
3073 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
3074 		sd_print_result(sdkp, res);
3075 		if (driver_byte(res) & DRIVER_SENSE)
3076 			sd_print_sense_hdr(sdkp, &sshdr);
3077 		if (scsi_sense_valid(&sshdr) &&
3078 			/* 0x3a is medium not present */
3079 			sshdr.asc == 0x3a)
3080 			res = 0;
3081 	}
3082 
3083 	/* SCSI error codes must not go to the generic layer */
3084 	if (res)
3085 		return -EIO;
3086 
3087 	return 0;
3088 }
3089 
3090 /*
3091  * Send a SYNCHRONIZE CACHE instruction down to the device through
3092  * the normal SCSI command structure.  Wait for the command to
3093  * complete.
3094  */
3095 static void sd_shutdown(struct device *dev)
3096 {
3097 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3098 
3099 	if (!sdkp)
3100 		return;         /* this can happen */
3101 
3102 	if (pm_runtime_suspended(dev))
3103 		goto exit;
3104 
3105 	if (sdkp->WCE && sdkp->media_present) {
3106 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3107 		sd_sync_cache(sdkp);
3108 	}
3109 
3110 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3111 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3112 		sd_start_stop_device(sdkp, 0);
3113 	}
3114 
3115 exit:
3116 	scsi_disk_put(sdkp);
3117 }
3118 
3119 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3120 {
3121 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3122 	int ret = 0;
3123 
3124 	if (!sdkp)
3125 		return 0;	/* this can happen */
3126 
3127 	if (sdkp->WCE && sdkp->media_present) {
3128 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3129 		ret = sd_sync_cache(sdkp);
3130 		if (ret) {
3131 			/* ignore OFFLINE device */
3132 			if (ret == -ENODEV)
3133 				ret = 0;
3134 			goto done;
3135 		}
3136 	}
3137 
3138 	if (sdkp->device->manage_start_stop) {
3139 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3140 		/* an error is not worth aborting a system sleep */
3141 		ret = sd_start_stop_device(sdkp, 0);
3142 		if (ignore_stop_errors)
3143 			ret = 0;
3144 	}
3145 
3146 done:
3147 	scsi_disk_put(sdkp);
3148 	return ret;
3149 }
3150 
3151 static int sd_suspend_system(struct device *dev)
3152 {
3153 	return sd_suspend_common(dev, true);
3154 }
3155 
3156 static int sd_suspend_runtime(struct device *dev)
3157 {
3158 	return sd_suspend_common(dev, false);
3159 }
3160 
3161 static int sd_resume(struct device *dev)
3162 {
3163 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3164 	int ret = 0;
3165 
3166 	if (!sdkp->device->manage_start_stop)
3167 		goto done;
3168 
3169 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3170 	ret = sd_start_stop_device(sdkp, 1);
3171 
3172 done:
3173 	scsi_disk_put(sdkp);
3174 	return ret;
3175 }
3176 
3177 /**
3178  *	init_sd - entry point for this driver (both when built in or when
3179  *	a module).
3180  *
3181  *	Note: this function registers this driver with the scsi mid-level.
3182  **/
3183 static int __init init_sd(void)
3184 {
3185 	int majors = 0, i, err;
3186 
3187 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3188 
3189 	for (i = 0; i < SD_MAJORS; i++) {
3190 		if (register_blkdev(sd_major(i), "sd") != 0)
3191 			continue;
3192 		majors++;
3193 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3194 				    sd_default_probe, NULL, NULL);
3195 	}
3196 
3197 	if (!majors)
3198 		return -ENODEV;
3199 
3200 	err = class_register(&sd_disk_class);
3201 	if (err)
3202 		goto err_out;
3203 
3204 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3205 					 0, 0, NULL);
3206 	if (!sd_cdb_cache) {
3207 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3208 		goto err_out_class;
3209 	}
3210 
3211 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3212 	if (!sd_cdb_pool) {
3213 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3214 		goto err_out_cache;
3215 	}
3216 
3217 	err = scsi_register_driver(&sd_template.gendrv);
3218 	if (err)
3219 		goto err_out_driver;
3220 
3221 	return 0;
3222 
3223 err_out_driver:
3224 	mempool_destroy(sd_cdb_pool);
3225 
3226 err_out_cache:
3227 	kmem_cache_destroy(sd_cdb_cache);
3228 
3229 err_out_class:
3230 	class_unregister(&sd_disk_class);
3231 err_out:
3232 	for (i = 0; i < SD_MAJORS; i++)
3233 		unregister_blkdev(sd_major(i), "sd");
3234 	return err;
3235 }
3236 
3237 /**
3238  *	exit_sd - exit point for this driver (when it is a module).
3239  *
3240  *	Note: this function unregisters this driver from the scsi mid-level.
3241  **/
3242 static void __exit exit_sd(void)
3243 {
3244 	int i;
3245 
3246 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3247 
3248 	scsi_unregister_driver(&sd_template.gendrv);
3249 	mempool_destroy(sd_cdb_pool);
3250 	kmem_cache_destroy(sd_cdb_cache);
3251 
3252 	class_unregister(&sd_disk_class);
3253 
3254 	for (i = 0; i < SD_MAJORS; i++) {
3255 		blk_unregister_region(sd_major(i), SD_MINORS);
3256 		unregister_blkdev(sd_major(i), "sd");
3257 	}
3258 }
3259 
3260 module_init(init_sd);
3261 module_exit(exit_sd);
3262 
3263 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3264 			       struct scsi_sense_hdr *sshdr)
3265 {
3266 	sd_printk(KERN_INFO, sdkp, " ");
3267 	scsi_show_sense_hdr(sshdr);
3268 	sd_printk(KERN_INFO, sdkp, " ");
3269 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
3270 }
3271 
3272 static void sd_print_result(struct scsi_disk *sdkp, int result)
3273 {
3274 	sd_printk(KERN_INFO, sdkp, " ");
3275 	scsi_show_result(result);
3276 }
3277 
3278