xref: /openbmc/linux/drivers/scsi/sd.c (revision 77a87824)
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *	Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17  *	   sd_init and cleanups.
18  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *	   not being read in sd_open. Fix problem where removable media
20  *	   could be ejected after sd_open.
21  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24  *	   Support 32k/1M disks.
25  *
26  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *	Note: when the logging level is set by the user, it must be greater
32  *	than the level indicated above to trigger output.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <linux/pr.h>
55 #include <asm/uaccess.h>
56 #include <asm/unaligned.h>
57 
58 #include <scsi/scsi.h>
59 #include <scsi/scsi_cmnd.h>
60 #include <scsi/scsi_dbg.h>
61 #include <scsi/scsi_device.h>
62 #include <scsi/scsi_driver.h>
63 #include <scsi/scsi_eh.h>
64 #include <scsi/scsi_host.h>
65 #include <scsi/scsi_ioctl.h>
66 #include <scsi/scsicam.h>
67 
68 #include "sd.h"
69 #include "scsi_priv.h"
70 #include "scsi_logging.h"
71 
72 MODULE_AUTHOR("Eric Youngdale");
73 MODULE_DESCRIPTION("SCSI disk (sd) driver");
74 MODULE_LICENSE("GPL");
75 
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
95 
96 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
97 #define SD_MINORS	16
98 #else
99 #define SD_MINORS	0
100 #endif
101 
102 static void sd_config_discard(struct scsi_disk *, unsigned int);
103 static void sd_config_write_same(struct scsi_disk *);
104 static int  sd_revalidate_disk(struct gendisk *);
105 static void sd_unlock_native_capacity(struct gendisk *disk);
106 static int  sd_probe(struct device *);
107 static int  sd_remove(struct device *);
108 static void sd_shutdown(struct device *);
109 static int sd_suspend_system(struct device *);
110 static int sd_suspend_runtime(struct device *);
111 static int sd_resume(struct device *);
112 static void sd_rescan(struct device *);
113 static int sd_init_command(struct scsi_cmnd *SCpnt);
114 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
115 static int sd_done(struct scsi_cmnd *);
116 static int sd_eh_action(struct scsi_cmnd *, int);
117 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
118 static void scsi_disk_release(struct device *cdev);
119 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
120 static void sd_print_result(const struct scsi_disk *, const char *, int);
121 
122 static DEFINE_SPINLOCK(sd_index_lock);
123 static DEFINE_IDA(sd_index_ida);
124 
125 /* This semaphore is used to mediate the 0->1 reference get in the
126  * face of object destruction (i.e. we can't allow a get on an
127  * object after last put) */
128 static DEFINE_MUTEX(sd_ref_mutex);
129 
130 static struct kmem_cache *sd_cdb_cache;
131 static mempool_t *sd_cdb_pool;
132 
133 static const char *sd_cache_types[] = {
134 	"write through", "none", "write back",
135 	"write back, no read (daft)"
136 };
137 
138 static void sd_set_flush_flag(struct scsi_disk *sdkp)
139 {
140 	bool wc = false, fua = false;
141 
142 	if (sdkp->WCE) {
143 		wc = true;
144 		if (sdkp->DPOFUA)
145 			fua = true;
146 	}
147 
148 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
149 }
150 
151 static ssize_t
152 cache_type_store(struct device *dev, struct device_attribute *attr,
153 		 const char *buf, size_t count)
154 {
155 	int i, ct = -1, rcd, wce, sp;
156 	struct scsi_disk *sdkp = to_scsi_disk(dev);
157 	struct scsi_device *sdp = sdkp->device;
158 	char buffer[64];
159 	char *buffer_data;
160 	struct scsi_mode_data data;
161 	struct scsi_sense_hdr sshdr;
162 	static const char temp[] = "temporary ";
163 	int len;
164 
165 	if (sdp->type != TYPE_DISK)
166 		/* no cache control on RBC devices; theoretically they
167 		 * can do it, but there's probably so many exceptions
168 		 * it's not worth the risk */
169 		return -EINVAL;
170 
171 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
172 		buf += sizeof(temp) - 1;
173 		sdkp->cache_override = 1;
174 	} else {
175 		sdkp->cache_override = 0;
176 	}
177 
178 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
179 		len = strlen(sd_cache_types[i]);
180 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
181 		    buf[len] == '\n') {
182 			ct = i;
183 			break;
184 		}
185 	}
186 	if (ct < 0)
187 		return -EINVAL;
188 	rcd = ct & 0x01 ? 1 : 0;
189 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 
191 	if (sdkp->cache_override) {
192 		sdkp->WCE = wce;
193 		sdkp->RCD = rcd;
194 		sd_set_flush_flag(sdkp);
195 		return count;
196 	}
197 
198 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 			    SD_MAX_RETRIES, &data, NULL))
200 		return -EINVAL;
201 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 		  data.block_descriptor_length);
203 	buffer_data = buffer + data.header_length +
204 		data.block_descriptor_length;
205 	buffer_data[2] &= ~0x05;
206 	buffer_data[2] |= wce << 2 | rcd;
207 	sp = buffer_data[0] & 0x80 ? 1 : 0;
208 	buffer_data[0] &= ~0x80;
209 
210 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
211 			     SD_MAX_RETRIES, &data, &sshdr)) {
212 		if (scsi_sense_valid(&sshdr))
213 			sd_print_sense_hdr(sdkp, &sshdr);
214 		return -EINVAL;
215 	}
216 	revalidate_disk(sdkp->disk);
217 	return count;
218 }
219 
220 static ssize_t
221 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
222 		       char *buf)
223 {
224 	struct scsi_disk *sdkp = to_scsi_disk(dev);
225 	struct scsi_device *sdp = sdkp->device;
226 
227 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
228 }
229 
230 static ssize_t
231 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
232 			const char *buf, size_t count)
233 {
234 	struct scsi_disk *sdkp = to_scsi_disk(dev);
235 	struct scsi_device *sdp = sdkp->device;
236 
237 	if (!capable(CAP_SYS_ADMIN))
238 		return -EACCES;
239 
240 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
241 
242 	return count;
243 }
244 static DEVICE_ATTR_RW(manage_start_stop);
245 
246 static ssize_t
247 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
248 {
249 	struct scsi_disk *sdkp = to_scsi_disk(dev);
250 
251 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
252 }
253 
254 static ssize_t
255 allow_restart_store(struct device *dev, struct device_attribute *attr,
256 		    const char *buf, size_t count)
257 {
258 	struct scsi_disk *sdkp = to_scsi_disk(dev);
259 	struct scsi_device *sdp = sdkp->device;
260 
261 	if (!capable(CAP_SYS_ADMIN))
262 		return -EACCES;
263 
264 	if (sdp->type != TYPE_DISK)
265 		return -EINVAL;
266 
267 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
268 
269 	return count;
270 }
271 static DEVICE_ATTR_RW(allow_restart);
272 
273 static ssize_t
274 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
275 {
276 	struct scsi_disk *sdkp = to_scsi_disk(dev);
277 	int ct = sdkp->RCD + 2*sdkp->WCE;
278 
279 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
280 }
281 static DEVICE_ATTR_RW(cache_type);
282 
283 static ssize_t
284 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
285 {
286 	struct scsi_disk *sdkp = to_scsi_disk(dev);
287 
288 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
289 }
290 static DEVICE_ATTR_RO(FUA);
291 
292 static ssize_t
293 protection_type_show(struct device *dev, struct device_attribute *attr,
294 		     char *buf)
295 {
296 	struct scsi_disk *sdkp = to_scsi_disk(dev);
297 
298 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
299 }
300 
301 static ssize_t
302 protection_type_store(struct device *dev, struct device_attribute *attr,
303 		      const char *buf, size_t count)
304 {
305 	struct scsi_disk *sdkp = to_scsi_disk(dev);
306 	unsigned int val;
307 	int err;
308 
309 	if (!capable(CAP_SYS_ADMIN))
310 		return -EACCES;
311 
312 	err = kstrtouint(buf, 10, &val);
313 
314 	if (err)
315 		return err;
316 
317 	if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
318 		sdkp->protection_type = val;
319 
320 	return count;
321 }
322 static DEVICE_ATTR_RW(protection_type);
323 
324 static ssize_t
325 protection_mode_show(struct device *dev, struct device_attribute *attr,
326 		     char *buf)
327 {
328 	struct scsi_disk *sdkp = to_scsi_disk(dev);
329 	struct scsi_device *sdp = sdkp->device;
330 	unsigned int dif, dix;
331 
332 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
333 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
334 
335 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
336 		dif = 0;
337 		dix = 1;
338 	}
339 
340 	if (!dif && !dix)
341 		return snprintf(buf, 20, "none\n");
342 
343 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
344 }
345 static DEVICE_ATTR_RO(protection_mode);
346 
347 static ssize_t
348 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
349 {
350 	struct scsi_disk *sdkp = to_scsi_disk(dev);
351 
352 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
353 }
354 static DEVICE_ATTR_RO(app_tag_own);
355 
356 static ssize_t
357 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
358 		       char *buf)
359 {
360 	struct scsi_disk *sdkp = to_scsi_disk(dev);
361 
362 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
363 }
364 static DEVICE_ATTR_RO(thin_provisioning);
365 
366 static const char *lbp_mode[] = {
367 	[SD_LBP_FULL]		= "full",
368 	[SD_LBP_UNMAP]		= "unmap",
369 	[SD_LBP_WS16]		= "writesame_16",
370 	[SD_LBP_WS10]		= "writesame_10",
371 	[SD_LBP_ZERO]		= "writesame_zero",
372 	[SD_LBP_DISABLE]	= "disabled",
373 };
374 
375 static ssize_t
376 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
377 		       char *buf)
378 {
379 	struct scsi_disk *sdkp = to_scsi_disk(dev);
380 
381 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
382 }
383 
384 static ssize_t
385 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
386 			const char *buf, size_t count)
387 {
388 	struct scsi_disk *sdkp = to_scsi_disk(dev);
389 	struct scsi_device *sdp = sdkp->device;
390 
391 	if (!capable(CAP_SYS_ADMIN))
392 		return -EACCES;
393 
394 	if (sdp->type != TYPE_DISK)
395 		return -EINVAL;
396 
397 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
398 		sd_config_discard(sdkp, SD_LBP_UNMAP);
399 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
400 		sd_config_discard(sdkp, SD_LBP_WS16);
401 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
402 		sd_config_discard(sdkp, SD_LBP_WS10);
403 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
404 		sd_config_discard(sdkp, SD_LBP_ZERO);
405 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
406 		sd_config_discard(sdkp, SD_LBP_DISABLE);
407 	else
408 		return -EINVAL;
409 
410 	return count;
411 }
412 static DEVICE_ATTR_RW(provisioning_mode);
413 
414 static ssize_t
415 max_medium_access_timeouts_show(struct device *dev,
416 				struct device_attribute *attr, char *buf)
417 {
418 	struct scsi_disk *sdkp = to_scsi_disk(dev);
419 
420 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
421 }
422 
423 static ssize_t
424 max_medium_access_timeouts_store(struct device *dev,
425 				 struct device_attribute *attr, const char *buf,
426 				 size_t count)
427 {
428 	struct scsi_disk *sdkp = to_scsi_disk(dev);
429 	int err;
430 
431 	if (!capable(CAP_SYS_ADMIN))
432 		return -EACCES;
433 
434 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
435 
436 	return err ? err : count;
437 }
438 static DEVICE_ATTR_RW(max_medium_access_timeouts);
439 
440 static ssize_t
441 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
442 			   char *buf)
443 {
444 	struct scsi_disk *sdkp = to_scsi_disk(dev);
445 
446 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
447 }
448 
449 static ssize_t
450 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
451 			    const char *buf, size_t count)
452 {
453 	struct scsi_disk *sdkp = to_scsi_disk(dev);
454 	struct scsi_device *sdp = sdkp->device;
455 	unsigned long max;
456 	int err;
457 
458 	if (!capable(CAP_SYS_ADMIN))
459 		return -EACCES;
460 
461 	if (sdp->type != TYPE_DISK)
462 		return -EINVAL;
463 
464 	err = kstrtoul(buf, 10, &max);
465 
466 	if (err)
467 		return err;
468 
469 	if (max == 0)
470 		sdp->no_write_same = 1;
471 	else if (max <= SD_MAX_WS16_BLOCKS) {
472 		sdp->no_write_same = 0;
473 		sdkp->max_ws_blocks = max;
474 	}
475 
476 	sd_config_write_same(sdkp);
477 
478 	return count;
479 }
480 static DEVICE_ATTR_RW(max_write_same_blocks);
481 
482 static struct attribute *sd_disk_attrs[] = {
483 	&dev_attr_cache_type.attr,
484 	&dev_attr_FUA.attr,
485 	&dev_attr_allow_restart.attr,
486 	&dev_attr_manage_start_stop.attr,
487 	&dev_attr_protection_type.attr,
488 	&dev_attr_protection_mode.attr,
489 	&dev_attr_app_tag_own.attr,
490 	&dev_attr_thin_provisioning.attr,
491 	&dev_attr_provisioning_mode.attr,
492 	&dev_attr_max_write_same_blocks.attr,
493 	&dev_attr_max_medium_access_timeouts.attr,
494 	NULL,
495 };
496 ATTRIBUTE_GROUPS(sd_disk);
497 
498 static struct class sd_disk_class = {
499 	.name		= "scsi_disk",
500 	.owner		= THIS_MODULE,
501 	.dev_release	= scsi_disk_release,
502 	.dev_groups	= sd_disk_groups,
503 };
504 
505 static const struct dev_pm_ops sd_pm_ops = {
506 	.suspend		= sd_suspend_system,
507 	.resume			= sd_resume,
508 	.poweroff		= sd_suspend_system,
509 	.restore		= sd_resume,
510 	.runtime_suspend	= sd_suspend_runtime,
511 	.runtime_resume		= sd_resume,
512 };
513 
514 static struct scsi_driver sd_template = {
515 	.gendrv = {
516 		.name		= "sd",
517 		.owner		= THIS_MODULE,
518 		.probe		= sd_probe,
519 		.remove		= sd_remove,
520 		.shutdown	= sd_shutdown,
521 		.pm		= &sd_pm_ops,
522 	},
523 	.rescan			= sd_rescan,
524 	.init_command		= sd_init_command,
525 	.uninit_command		= sd_uninit_command,
526 	.done			= sd_done,
527 	.eh_action		= sd_eh_action,
528 };
529 
530 /*
531  * Dummy kobj_map->probe function.
532  * The default ->probe function will call modprobe, which is
533  * pointless as this module is already loaded.
534  */
535 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
536 {
537 	return NULL;
538 }
539 
540 /*
541  * Device no to disk mapping:
542  *
543  *       major         disc2     disc  p1
544  *   |............|.............|....|....| <- dev_t
545  *    31        20 19          8 7  4 3  0
546  *
547  * Inside a major, we have 16k disks, however mapped non-
548  * contiguously. The first 16 disks are for major0, the next
549  * ones with major1, ... Disk 256 is for major0 again, disk 272
550  * for major1, ...
551  * As we stay compatible with our numbering scheme, we can reuse
552  * the well-know SCSI majors 8, 65--71, 136--143.
553  */
554 static int sd_major(int major_idx)
555 {
556 	switch (major_idx) {
557 	case 0:
558 		return SCSI_DISK0_MAJOR;
559 	case 1 ... 7:
560 		return SCSI_DISK1_MAJOR + major_idx - 1;
561 	case 8 ... 15:
562 		return SCSI_DISK8_MAJOR + major_idx - 8;
563 	default:
564 		BUG();
565 		return 0;	/* shut up gcc */
566 	}
567 }
568 
569 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
570 {
571 	struct scsi_disk *sdkp = NULL;
572 
573 	mutex_lock(&sd_ref_mutex);
574 
575 	if (disk->private_data) {
576 		sdkp = scsi_disk(disk);
577 		if (scsi_device_get(sdkp->device) == 0)
578 			get_device(&sdkp->dev);
579 		else
580 			sdkp = NULL;
581 	}
582 	mutex_unlock(&sd_ref_mutex);
583 	return sdkp;
584 }
585 
586 static void scsi_disk_put(struct scsi_disk *sdkp)
587 {
588 	struct scsi_device *sdev = sdkp->device;
589 
590 	mutex_lock(&sd_ref_mutex);
591 	put_device(&sdkp->dev);
592 	scsi_device_put(sdev);
593 	mutex_unlock(&sd_ref_mutex);
594 }
595 
596 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
597 					   unsigned int dix, unsigned int dif)
598 {
599 	struct bio *bio = scmd->request->bio;
600 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
601 	unsigned int protect = 0;
602 
603 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
604 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
605 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
606 
607 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
608 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
609 	}
610 
611 	if (dif != SD_DIF_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
612 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
613 
614 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
615 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
616 	}
617 
618 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
619 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
620 
621 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
622 			protect = 3 << 5;	/* Disable target PI checking */
623 		else
624 			protect = 1 << 5;	/* Enable target PI checking */
625 	}
626 
627 	scsi_set_prot_op(scmd, prot_op);
628 	scsi_set_prot_type(scmd, dif);
629 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
630 
631 	return protect;
632 }
633 
634 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
635 {
636 	struct request_queue *q = sdkp->disk->queue;
637 	unsigned int logical_block_size = sdkp->device->sector_size;
638 	unsigned int max_blocks = 0;
639 
640 	q->limits.discard_zeroes_data = 0;
641 
642 	/*
643 	 * When LBPRZ is reported, discard alignment and granularity
644 	 * must be fixed to the logical block size. Otherwise the block
645 	 * layer will drop misaligned portions of the request which can
646 	 * lead to data corruption. If LBPRZ is not set, we honor the
647 	 * device preference.
648 	 */
649 	if (sdkp->lbprz) {
650 		q->limits.discard_alignment = 0;
651 		q->limits.discard_granularity = logical_block_size;
652 	} else {
653 		q->limits.discard_alignment = sdkp->unmap_alignment *
654 			logical_block_size;
655 		q->limits.discard_granularity =
656 			max(sdkp->physical_block_size,
657 			    sdkp->unmap_granularity * logical_block_size);
658 	}
659 
660 	sdkp->provisioning_mode = mode;
661 
662 	switch (mode) {
663 
664 	case SD_LBP_DISABLE:
665 		blk_queue_max_discard_sectors(q, 0);
666 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
667 		return;
668 
669 	case SD_LBP_UNMAP:
670 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
671 					  (u32)SD_MAX_WS16_BLOCKS);
672 		break;
673 
674 	case SD_LBP_WS16:
675 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
676 					  (u32)SD_MAX_WS16_BLOCKS);
677 		q->limits.discard_zeroes_data = sdkp->lbprz;
678 		break;
679 
680 	case SD_LBP_WS10:
681 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
682 					  (u32)SD_MAX_WS10_BLOCKS);
683 		q->limits.discard_zeroes_data = sdkp->lbprz;
684 		break;
685 
686 	case SD_LBP_ZERO:
687 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
688 					  (u32)SD_MAX_WS10_BLOCKS);
689 		q->limits.discard_zeroes_data = 1;
690 		break;
691 	}
692 
693 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
694 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
695 }
696 
697 /**
698  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
699  * @sdp: scsi device to operate one
700  * @rq: Request to prepare
701  *
702  * Will issue either UNMAP or WRITE SAME(16) depending on preference
703  * indicated by target device.
704  **/
705 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
706 {
707 	struct request *rq = cmd->request;
708 	struct scsi_device *sdp = cmd->device;
709 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
710 	sector_t sector = blk_rq_pos(rq);
711 	unsigned int nr_sectors = blk_rq_sectors(rq);
712 	unsigned int nr_bytes = blk_rq_bytes(rq);
713 	unsigned int len;
714 	int ret;
715 	char *buf;
716 	struct page *page;
717 
718 	sector >>= ilog2(sdp->sector_size) - 9;
719 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
720 
721 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
722 	if (!page)
723 		return BLKPREP_DEFER;
724 
725 	switch (sdkp->provisioning_mode) {
726 	case SD_LBP_UNMAP:
727 		buf = page_address(page);
728 
729 		cmd->cmd_len = 10;
730 		cmd->cmnd[0] = UNMAP;
731 		cmd->cmnd[8] = 24;
732 
733 		put_unaligned_be16(6 + 16, &buf[0]);
734 		put_unaligned_be16(16, &buf[2]);
735 		put_unaligned_be64(sector, &buf[8]);
736 		put_unaligned_be32(nr_sectors, &buf[16]);
737 
738 		len = 24;
739 		break;
740 
741 	case SD_LBP_WS16:
742 		cmd->cmd_len = 16;
743 		cmd->cmnd[0] = WRITE_SAME_16;
744 		cmd->cmnd[1] = 0x8; /* UNMAP */
745 		put_unaligned_be64(sector, &cmd->cmnd[2]);
746 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
747 
748 		len = sdkp->device->sector_size;
749 		break;
750 
751 	case SD_LBP_WS10:
752 	case SD_LBP_ZERO:
753 		cmd->cmd_len = 10;
754 		cmd->cmnd[0] = WRITE_SAME;
755 		if (sdkp->provisioning_mode == SD_LBP_WS10)
756 			cmd->cmnd[1] = 0x8; /* UNMAP */
757 		put_unaligned_be32(sector, &cmd->cmnd[2]);
758 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
759 
760 		len = sdkp->device->sector_size;
761 		break;
762 
763 	default:
764 		ret = BLKPREP_INVALID;
765 		goto out;
766 	}
767 
768 	rq->completion_data = page;
769 	rq->timeout = SD_TIMEOUT;
770 
771 	cmd->transfersize = len;
772 	cmd->allowed = SD_MAX_RETRIES;
773 
774 	/*
775 	 * Initially __data_len is set to the amount of data that needs to be
776 	 * transferred to the target. This amount depends on whether WRITE SAME
777 	 * or UNMAP is being used. After the scatterlist has been mapped by
778 	 * scsi_init_io() we set __data_len to the size of the area to be
779 	 * discarded on disk. This allows us to report completion on the full
780 	 * amount of blocks described by the request.
781 	 */
782 	blk_add_request_payload(rq, page, 0, len);
783 	ret = scsi_init_io(cmd);
784 	rq->__data_len = nr_bytes;
785 
786 out:
787 	if (ret != BLKPREP_OK)
788 		__free_page(page);
789 	return ret;
790 }
791 
792 static void sd_config_write_same(struct scsi_disk *sdkp)
793 {
794 	struct request_queue *q = sdkp->disk->queue;
795 	unsigned int logical_block_size = sdkp->device->sector_size;
796 
797 	if (sdkp->device->no_write_same) {
798 		sdkp->max_ws_blocks = 0;
799 		goto out;
800 	}
801 
802 	/* Some devices can not handle block counts above 0xffff despite
803 	 * supporting WRITE SAME(16). Consequently we default to 64k
804 	 * blocks per I/O unless the device explicitly advertises a
805 	 * bigger limit.
806 	 */
807 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
808 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
809 						   (u32)SD_MAX_WS16_BLOCKS);
810 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
811 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
812 						   (u32)SD_MAX_WS10_BLOCKS);
813 	else {
814 		sdkp->device->no_write_same = 1;
815 		sdkp->max_ws_blocks = 0;
816 	}
817 
818 out:
819 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
820 					 (logical_block_size >> 9));
821 }
822 
823 /**
824  * sd_setup_write_same_cmnd - write the same data to multiple blocks
825  * @cmd: command to prepare
826  *
827  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
828  * preference indicated by target device.
829  **/
830 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
831 {
832 	struct request *rq = cmd->request;
833 	struct scsi_device *sdp = cmd->device;
834 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
835 	struct bio *bio = rq->bio;
836 	sector_t sector = blk_rq_pos(rq);
837 	unsigned int nr_sectors = blk_rq_sectors(rq);
838 	unsigned int nr_bytes = blk_rq_bytes(rq);
839 	int ret;
840 
841 	if (sdkp->device->no_write_same)
842 		return BLKPREP_INVALID;
843 
844 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
845 
846 	sector >>= ilog2(sdp->sector_size) - 9;
847 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
848 
849 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
850 
851 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
852 		cmd->cmd_len = 16;
853 		cmd->cmnd[0] = WRITE_SAME_16;
854 		put_unaligned_be64(sector, &cmd->cmnd[2]);
855 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
856 	} else {
857 		cmd->cmd_len = 10;
858 		cmd->cmnd[0] = WRITE_SAME;
859 		put_unaligned_be32(sector, &cmd->cmnd[2]);
860 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
861 	}
862 
863 	cmd->transfersize = sdp->sector_size;
864 	cmd->allowed = SD_MAX_RETRIES;
865 
866 	/*
867 	 * For WRITE_SAME the data transferred in the DATA IN buffer is
868 	 * different from the amount of data actually written to the target.
869 	 *
870 	 * We set up __data_len to the amount of data transferred from the
871 	 * DATA IN buffer so that blk_rq_map_sg set up the proper S/G list
872 	 * to transfer a single sector of data first, but then reset it to
873 	 * the amount of data to be written right after so that the I/O path
874 	 * knows how much to actually write.
875 	 */
876 	rq->__data_len = sdp->sector_size;
877 	ret = scsi_init_io(cmd);
878 	rq->__data_len = nr_bytes;
879 	return ret;
880 }
881 
882 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
883 {
884 	struct request *rq = cmd->request;
885 
886 	/* flush requests don't perform I/O, zero the S/G table */
887 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
888 
889 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
890 	cmd->cmd_len = 10;
891 	cmd->transfersize = 0;
892 	cmd->allowed = SD_MAX_RETRIES;
893 
894 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
895 	return BLKPREP_OK;
896 }
897 
898 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
899 {
900 	struct request *rq = SCpnt->request;
901 	struct scsi_device *sdp = SCpnt->device;
902 	struct gendisk *disk = rq->rq_disk;
903 	struct scsi_disk *sdkp;
904 	sector_t block = blk_rq_pos(rq);
905 	sector_t threshold;
906 	unsigned int this_count = blk_rq_sectors(rq);
907 	unsigned int dif, dix;
908 	int ret;
909 	unsigned char protect;
910 
911 	ret = scsi_init_io(SCpnt);
912 	if (ret != BLKPREP_OK)
913 		goto out;
914 	SCpnt = rq->special;
915 	sdkp = scsi_disk(disk);
916 
917 	/* from here on until we're complete, any goto out
918 	 * is used for a killable error condition */
919 	ret = BLKPREP_KILL;
920 
921 	SCSI_LOG_HLQUEUE(1,
922 		scmd_printk(KERN_INFO, SCpnt,
923 			"%s: block=%llu, count=%d\n",
924 			__func__, (unsigned long long)block, this_count));
925 
926 	if (!sdp || !scsi_device_online(sdp) ||
927 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
928 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
929 						"Finishing %u sectors\n",
930 						blk_rq_sectors(rq)));
931 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
932 						"Retry with 0x%p\n", SCpnt));
933 		goto out;
934 	}
935 
936 	if (sdp->changed) {
937 		/*
938 		 * quietly refuse to do anything to a changed disc until
939 		 * the changed bit has been reset
940 		 */
941 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
942 		goto out;
943 	}
944 
945 	/*
946 	 * Some SD card readers can't handle multi-sector accesses which touch
947 	 * the last one or two hardware sectors.  Split accesses as needed.
948 	 */
949 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
950 		(sdp->sector_size / 512);
951 
952 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
953 		if (block < threshold) {
954 			/* Access up to the threshold but not beyond */
955 			this_count = threshold - block;
956 		} else {
957 			/* Access only a single hardware sector */
958 			this_count = sdp->sector_size / 512;
959 		}
960 	}
961 
962 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
963 					(unsigned long long)block));
964 
965 	/*
966 	 * If we have a 1K hardware sectorsize, prevent access to single
967 	 * 512 byte sectors.  In theory we could handle this - in fact
968 	 * the scsi cdrom driver must be able to handle this because
969 	 * we typically use 1K blocksizes, and cdroms typically have
970 	 * 2K hardware sectorsizes.  Of course, things are simpler
971 	 * with the cdrom, since it is read-only.  For performance
972 	 * reasons, the filesystems should be able to handle this
973 	 * and not force the scsi disk driver to use bounce buffers
974 	 * for this.
975 	 */
976 	if (sdp->sector_size == 1024) {
977 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
978 			scmd_printk(KERN_ERR, SCpnt,
979 				    "Bad block number requested\n");
980 			goto out;
981 		} else {
982 			block = block >> 1;
983 			this_count = this_count >> 1;
984 		}
985 	}
986 	if (sdp->sector_size == 2048) {
987 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
988 			scmd_printk(KERN_ERR, SCpnt,
989 				    "Bad block number requested\n");
990 			goto out;
991 		} else {
992 			block = block >> 2;
993 			this_count = this_count >> 2;
994 		}
995 	}
996 	if (sdp->sector_size == 4096) {
997 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
998 			scmd_printk(KERN_ERR, SCpnt,
999 				    "Bad block number requested\n");
1000 			goto out;
1001 		} else {
1002 			block = block >> 3;
1003 			this_count = this_count >> 3;
1004 		}
1005 	}
1006 	if (rq_data_dir(rq) == WRITE) {
1007 		SCpnt->cmnd[0] = WRITE_6;
1008 
1009 		if (blk_integrity_rq(rq))
1010 			sd_dif_prepare(SCpnt);
1011 
1012 	} else if (rq_data_dir(rq) == READ) {
1013 		SCpnt->cmnd[0] = READ_6;
1014 	} else {
1015 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %llu,%llx\n",
1016 			    req_op(rq), (unsigned long long) rq->cmd_flags);
1017 		goto out;
1018 	}
1019 
1020 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1021 					"%s %d/%u 512 byte blocks.\n",
1022 					(rq_data_dir(rq) == WRITE) ?
1023 					"writing" : "reading", this_count,
1024 					blk_rq_sectors(rq)));
1025 
1026 	dix = scsi_prot_sg_count(SCpnt);
1027 	dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1028 
1029 	if (dif || dix)
1030 		protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1031 	else
1032 		protect = 0;
1033 
1034 	if (protect && sdkp->protection_type == SD_DIF_TYPE2_PROTECTION) {
1035 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1036 
1037 		if (unlikely(SCpnt->cmnd == NULL)) {
1038 			ret = BLKPREP_DEFER;
1039 			goto out;
1040 		}
1041 
1042 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1043 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1044 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1045 		SCpnt->cmnd[7] = 0x18;
1046 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1047 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1048 
1049 		/* LBA */
1050 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1051 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1052 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1053 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1054 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1055 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1056 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1057 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1058 
1059 		/* Expected Indirect LBA */
1060 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1061 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1062 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1063 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1064 
1065 		/* Transfer length */
1066 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1067 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1068 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1069 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1070 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1071 		SCpnt->cmnd[0] += READ_16 - READ_6;
1072 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1073 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1074 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1075 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1076 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1077 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1078 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1079 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1080 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1081 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1082 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1083 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1084 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1085 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1086 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1087 		   scsi_device_protection(SCpnt->device) ||
1088 		   SCpnt->device->use_10_for_rw) {
1089 		SCpnt->cmnd[0] += READ_10 - READ_6;
1090 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1091 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1092 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1093 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1094 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1095 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1096 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1097 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1098 	} else {
1099 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1100 			/*
1101 			 * This happens only if this drive failed
1102 			 * 10byte rw command with ILLEGAL_REQUEST
1103 			 * during operation and thus turned off
1104 			 * use_10_for_rw.
1105 			 */
1106 			scmd_printk(KERN_ERR, SCpnt,
1107 				    "FUA write on READ/WRITE(6) drive\n");
1108 			goto out;
1109 		}
1110 
1111 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1112 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1113 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1114 		SCpnt->cmnd[4] = (unsigned char) this_count;
1115 		SCpnt->cmnd[5] = 0;
1116 	}
1117 	SCpnt->sdb.length = this_count * sdp->sector_size;
1118 
1119 	/*
1120 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1121 	 * host adapter, it's safe to assume that we can at least transfer
1122 	 * this many bytes between each connect / disconnect.
1123 	 */
1124 	SCpnt->transfersize = sdp->sector_size;
1125 	SCpnt->underflow = this_count << 9;
1126 	SCpnt->allowed = SD_MAX_RETRIES;
1127 
1128 	/*
1129 	 * This indicates that the command is ready from our end to be
1130 	 * queued.
1131 	 */
1132 	ret = BLKPREP_OK;
1133  out:
1134 	return ret;
1135 }
1136 
1137 static int sd_init_command(struct scsi_cmnd *cmd)
1138 {
1139 	struct request *rq = cmd->request;
1140 
1141 	switch (req_op(rq)) {
1142 	case REQ_OP_DISCARD:
1143 		return sd_setup_discard_cmnd(cmd);
1144 	case REQ_OP_WRITE_SAME:
1145 		return sd_setup_write_same_cmnd(cmd);
1146 	case REQ_OP_FLUSH:
1147 		return sd_setup_flush_cmnd(cmd);
1148 	case REQ_OP_READ:
1149 	case REQ_OP_WRITE:
1150 		return sd_setup_read_write_cmnd(cmd);
1151 	default:
1152 		BUG();
1153 	}
1154 }
1155 
1156 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1157 {
1158 	struct request *rq = SCpnt->request;
1159 
1160 	if (req_op(rq) == REQ_OP_DISCARD)
1161 		__free_page(rq->completion_data);
1162 
1163 	if (SCpnt->cmnd != rq->cmd) {
1164 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1165 		SCpnt->cmnd = NULL;
1166 		SCpnt->cmd_len = 0;
1167 	}
1168 }
1169 
1170 /**
1171  *	sd_open - open a scsi disk device
1172  *	@inode: only i_rdev member may be used
1173  *	@filp: only f_mode and f_flags may be used
1174  *
1175  *	Returns 0 if successful. Returns a negated errno value in case
1176  *	of error.
1177  *
1178  *	Note: This can be called from a user context (e.g. fsck(1) )
1179  *	or from within the kernel (e.g. as a result of a mount(1) ).
1180  *	In the latter case @inode and @filp carry an abridged amount
1181  *	of information as noted above.
1182  *
1183  *	Locking: called with bdev->bd_mutex held.
1184  **/
1185 static int sd_open(struct block_device *bdev, fmode_t mode)
1186 {
1187 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1188 	struct scsi_device *sdev;
1189 	int retval;
1190 
1191 	if (!sdkp)
1192 		return -ENXIO;
1193 
1194 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1195 
1196 	sdev = sdkp->device;
1197 
1198 	/*
1199 	 * If the device is in error recovery, wait until it is done.
1200 	 * If the device is offline, then disallow any access to it.
1201 	 */
1202 	retval = -ENXIO;
1203 	if (!scsi_block_when_processing_errors(sdev))
1204 		goto error_out;
1205 
1206 	if (sdev->removable || sdkp->write_prot)
1207 		check_disk_change(bdev);
1208 
1209 	/*
1210 	 * If the drive is empty, just let the open fail.
1211 	 */
1212 	retval = -ENOMEDIUM;
1213 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1214 		goto error_out;
1215 
1216 	/*
1217 	 * If the device has the write protect tab set, have the open fail
1218 	 * if the user expects to be able to write to the thing.
1219 	 */
1220 	retval = -EROFS;
1221 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1222 		goto error_out;
1223 
1224 	/*
1225 	 * It is possible that the disk changing stuff resulted in
1226 	 * the device being taken offline.  If this is the case,
1227 	 * report this to the user, and don't pretend that the
1228 	 * open actually succeeded.
1229 	 */
1230 	retval = -ENXIO;
1231 	if (!scsi_device_online(sdev))
1232 		goto error_out;
1233 
1234 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1235 		if (scsi_block_when_processing_errors(sdev))
1236 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1237 	}
1238 
1239 	return 0;
1240 
1241 error_out:
1242 	scsi_disk_put(sdkp);
1243 	return retval;
1244 }
1245 
1246 /**
1247  *	sd_release - invoked when the (last) close(2) is called on this
1248  *	scsi disk.
1249  *	@inode: only i_rdev member may be used
1250  *	@filp: only f_mode and f_flags may be used
1251  *
1252  *	Returns 0.
1253  *
1254  *	Note: may block (uninterruptible) if error recovery is underway
1255  *	on this disk.
1256  *
1257  *	Locking: called with bdev->bd_mutex held.
1258  **/
1259 static void sd_release(struct gendisk *disk, fmode_t mode)
1260 {
1261 	struct scsi_disk *sdkp = scsi_disk(disk);
1262 	struct scsi_device *sdev = sdkp->device;
1263 
1264 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1265 
1266 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1267 		if (scsi_block_when_processing_errors(sdev))
1268 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1269 	}
1270 
1271 	/*
1272 	 * XXX and what if there are packets in flight and this close()
1273 	 * XXX is followed by a "rmmod sd_mod"?
1274 	 */
1275 
1276 	scsi_disk_put(sdkp);
1277 }
1278 
1279 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1280 {
1281 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1282 	struct scsi_device *sdp = sdkp->device;
1283 	struct Scsi_Host *host = sdp->host;
1284 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1285 	int diskinfo[4];
1286 
1287 	/* default to most commonly used values */
1288 	diskinfo[0] = 0x40;	/* 1 << 6 */
1289 	diskinfo[1] = 0x20;	/* 1 << 5 */
1290 	diskinfo[2] = capacity >> 11;
1291 
1292 	/* override with calculated, extended default, or driver values */
1293 	if (host->hostt->bios_param)
1294 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1295 	else
1296 		scsicam_bios_param(bdev, capacity, diskinfo);
1297 
1298 	geo->heads = diskinfo[0];
1299 	geo->sectors = diskinfo[1];
1300 	geo->cylinders = diskinfo[2];
1301 	return 0;
1302 }
1303 
1304 /**
1305  *	sd_ioctl - process an ioctl
1306  *	@inode: only i_rdev/i_bdev members may be used
1307  *	@filp: only f_mode and f_flags may be used
1308  *	@cmd: ioctl command number
1309  *	@arg: this is third argument given to ioctl(2) system call.
1310  *	Often contains a pointer.
1311  *
1312  *	Returns 0 if successful (some ioctls return positive numbers on
1313  *	success as well). Returns a negated errno value in case of error.
1314  *
1315  *	Note: most ioctls are forward onto the block subsystem or further
1316  *	down in the scsi subsystem.
1317  **/
1318 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1319 		    unsigned int cmd, unsigned long arg)
1320 {
1321 	struct gendisk *disk = bdev->bd_disk;
1322 	struct scsi_disk *sdkp = scsi_disk(disk);
1323 	struct scsi_device *sdp = sdkp->device;
1324 	void __user *p = (void __user *)arg;
1325 	int error;
1326 
1327 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1328 				    "cmd=0x%x\n", disk->disk_name, cmd));
1329 
1330 	error = scsi_verify_blk_ioctl(bdev, cmd);
1331 	if (error < 0)
1332 		return error;
1333 
1334 	/*
1335 	 * If we are in the middle of error recovery, don't let anyone
1336 	 * else try and use this device.  Also, if error recovery fails, it
1337 	 * may try and take the device offline, in which case all further
1338 	 * access to the device is prohibited.
1339 	 */
1340 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1341 			(mode & FMODE_NDELAY) != 0);
1342 	if (error)
1343 		goto out;
1344 
1345 	/*
1346 	 * Send SCSI addressing ioctls directly to mid level, send other
1347 	 * ioctls to block level and then onto mid level if they can't be
1348 	 * resolved.
1349 	 */
1350 	switch (cmd) {
1351 		case SCSI_IOCTL_GET_IDLUN:
1352 		case SCSI_IOCTL_GET_BUS_NUMBER:
1353 			error = scsi_ioctl(sdp, cmd, p);
1354 			break;
1355 		default:
1356 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1357 			if (error != -ENOTTY)
1358 				break;
1359 			error = scsi_ioctl(sdp, cmd, p);
1360 			break;
1361 	}
1362 out:
1363 	return error;
1364 }
1365 
1366 static void set_media_not_present(struct scsi_disk *sdkp)
1367 {
1368 	if (sdkp->media_present)
1369 		sdkp->device->changed = 1;
1370 
1371 	if (sdkp->device->removable) {
1372 		sdkp->media_present = 0;
1373 		sdkp->capacity = 0;
1374 	}
1375 }
1376 
1377 static int media_not_present(struct scsi_disk *sdkp,
1378 			     struct scsi_sense_hdr *sshdr)
1379 {
1380 	if (!scsi_sense_valid(sshdr))
1381 		return 0;
1382 
1383 	/* not invoked for commands that could return deferred errors */
1384 	switch (sshdr->sense_key) {
1385 	case UNIT_ATTENTION:
1386 	case NOT_READY:
1387 		/* medium not present */
1388 		if (sshdr->asc == 0x3A) {
1389 			set_media_not_present(sdkp);
1390 			return 1;
1391 		}
1392 	}
1393 	return 0;
1394 }
1395 
1396 /**
1397  *	sd_check_events - check media events
1398  *	@disk: kernel device descriptor
1399  *	@clearing: disk events currently being cleared
1400  *
1401  *	Returns mask of DISK_EVENT_*.
1402  *
1403  *	Note: this function is invoked from the block subsystem.
1404  **/
1405 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1406 {
1407 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1408 	struct scsi_device *sdp;
1409 	struct scsi_sense_hdr *sshdr = NULL;
1410 	int retval;
1411 
1412 	if (!sdkp)
1413 		return 0;
1414 
1415 	sdp = sdkp->device;
1416 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1417 
1418 	/*
1419 	 * If the device is offline, don't send any commands - just pretend as
1420 	 * if the command failed.  If the device ever comes back online, we
1421 	 * can deal with it then.  It is only because of unrecoverable errors
1422 	 * that we would ever take a device offline in the first place.
1423 	 */
1424 	if (!scsi_device_online(sdp)) {
1425 		set_media_not_present(sdkp);
1426 		goto out;
1427 	}
1428 
1429 	/*
1430 	 * Using TEST_UNIT_READY enables differentiation between drive with
1431 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1432 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1433 	 *
1434 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1435 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1436 	 * sd_revalidate() is called.
1437 	 */
1438 	retval = -ENODEV;
1439 
1440 	if (scsi_block_when_processing_errors(sdp)) {
1441 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1442 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1443 					      sshdr);
1444 	}
1445 
1446 	/* failed to execute TUR, assume media not present */
1447 	if (host_byte(retval)) {
1448 		set_media_not_present(sdkp);
1449 		goto out;
1450 	}
1451 
1452 	if (media_not_present(sdkp, sshdr))
1453 		goto out;
1454 
1455 	/*
1456 	 * For removable scsi disk we have to recognise the presence
1457 	 * of a disk in the drive.
1458 	 */
1459 	if (!sdkp->media_present)
1460 		sdp->changed = 1;
1461 	sdkp->media_present = 1;
1462 out:
1463 	/*
1464 	 * sdp->changed is set under the following conditions:
1465 	 *
1466 	 *	Medium present state has changed in either direction.
1467 	 *	Device has indicated UNIT_ATTENTION.
1468 	 */
1469 	kfree(sshdr);
1470 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1471 	sdp->changed = 0;
1472 	scsi_disk_put(sdkp);
1473 	return retval;
1474 }
1475 
1476 static int sd_sync_cache(struct scsi_disk *sdkp)
1477 {
1478 	int retries, res;
1479 	struct scsi_device *sdp = sdkp->device;
1480 	const int timeout = sdp->request_queue->rq_timeout
1481 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1482 	struct scsi_sense_hdr sshdr;
1483 
1484 	if (!scsi_device_online(sdp))
1485 		return -ENODEV;
1486 
1487 	for (retries = 3; retries > 0; --retries) {
1488 		unsigned char cmd[10] = { 0 };
1489 
1490 		cmd[0] = SYNCHRONIZE_CACHE;
1491 		/*
1492 		 * Leave the rest of the command zero to indicate
1493 		 * flush everything.
1494 		 */
1495 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1496 					     &sshdr, timeout, SD_MAX_RETRIES,
1497 					     NULL, REQ_PM);
1498 		if (res == 0)
1499 			break;
1500 	}
1501 
1502 	if (res) {
1503 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1504 
1505 		if (driver_byte(res) & DRIVER_SENSE)
1506 			sd_print_sense_hdr(sdkp, &sshdr);
1507 		/* we need to evaluate the error return  */
1508 		if (scsi_sense_valid(&sshdr) &&
1509 			(sshdr.asc == 0x3a ||	/* medium not present */
1510 			 sshdr.asc == 0x20))	/* invalid command */
1511 				/* this is no error here */
1512 				return 0;
1513 
1514 		switch (host_byte(res)) {
1515 		/* ignore errors due to racing a disconnection */
1516 		case DID_BAD_TARGET:
1517 		case DID_NO_CONNECT:
1518 			return 0;
1519 		/* signal the upper layer it might try again */
1520 		case DID_BUS_BUSY:
1521 		case DID_IMM_RETRY:
1522 		case DID_REQUEUE:
1523 		case DID_SOFT_ERROR:
1524 			return -EBUSY;
1525 		default:
1526 			return -EIO;
1527 		}
1528 	}
1529 	return 0;
1530 }
1531 
1532 static void sd_rescan(struct device *dev)
1533 {
1534 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1535 
1536 	revalidate_disk(sdkp->disk);
1537 }
1538 
1539 
1540 #ifdef CONFIG_COMPAT
1541 /*
1542  * This gets directly called from VFS. When the ioctl
1543  * is not recognized we go back to the other translation paths.
1544  */
1545 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1546 			   unsigned int cmd, unsigned long arg)
1547 {
1548 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1549 	int error;
1550 
1551 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1552 			(mode & FMODE_NDELAY) != 0);
1553 	if (error)
1554 		return error;
1555 
1556 	/*
1557 	 * Let the static ioctl translation table take care of it.
1558 	 */
1559 	if (!sdev->host->hostt->compat_ioctl)
1560 		return -ENOIOCTLCMD;
1561 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1562 }
1563 #endif
1564 
1565 static char sd_pr_type(enum pr_type type)
1566 {
1567 	switch (type) {
1568 	case PR_WRITE_EXCLUSIVE:
1569 		return 0x01;
1570 	case PR_EXCLUSIVE_ACCESS:
1571 		return 0x03;
1572 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1573 		return 0x05;
1574 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1575 		return 0x06;
1576 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1577 		return 0x07;
1578 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1579 		return 0x08;
1580 	default:
1581 		return 0;
1582 	}
1583 };
1584 
1585 static int sd_pr_command(struct block_device *bdev, u8 sa,
1586 		u64 key, u64 sa_key, u8 type, u8 flags)
1587 {
1588 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1589 	struct scsi_sense_hdr sshdr;
1590 	int result;
1591 	u8 cmd[16] = { 0, };
1592 	u8 data[24] = { 0, };
1593 
1594 	cmd[0] = PERSISTENT_RESERVE_OUT;
1595 	cmd[1] = sa;
1596 	cmd[2] = type;
1597 	put_unaligned_be32(sizeof(data), &cmd[5]);
1598 
1599 	put_unaligned_be64(key, &data[0]);
1600 	put_unaligned_be64(sa_key, &data[8]);
1601 	data[20] = flags;
1602 
1603 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1604 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1605 
1606 	if ((driver_byte(result) & DRIVER_SENSE) &&
1607 	    (scsi_sense_valid(&sshdr))) {
1608 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1609 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1610 	}
1611 
1612 	return result;
1613 }
1614 
1615 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1616 		u32 flags)
1617 {
1618 	if (flags & ~PR_FL_IGNORE_KEY)
1619 		return -EOPNOTSUPP;
1620 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1621 			old_key, new_key, 0,
1622 			(1 << 0) /* APTPL */);
1623 }
1624 
1625 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1626 		u32 flags)
1627 {
1628 	if (flags)
1629 		return -EOPNOTSUPP;
1630 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1631 }
1632 
1633 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1634 {
1635 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1636 }
1637 
1638 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1639 		enum pr_type type, bool abort)
1640 {
1641 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1642 			     sd_pr_type(type), 0);
1643 }
1644 
1645 static int sd_pr_clear(struct block_device *bdev, u64 key)
1646 {
1647 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1648 }
1649 
1650 static const struct pr_ops sd_pr_ops = {
1651 	.pr_register	= sd_pr_register,
1652 	.pr_reserve	= sd_pr_reserve,
1653 	.pr_release	= sd_pr_release,
1654 	.pr_preempt	= sd_pr_preempt,
1655 	.pr_clear	= sd_pr_clear,
1656 };
1657 
1658 static const struct block_device_operations sd_fops = {
1659 	.owner			= THIS_MODULE,
1660 	.open			= sd_open,
1661 	.release		= sd_release,
1662 	.ioctl			= sd_ioctl,
1663 	.getgeo			= sd_getgeo,
1664 #ifdef CONFIG_COMPAT
1665 	.compat_ioctl		= sd_compat_ioctl,
1666 #endif
1667 	.check_events		= sd_check_events,
1668 	.revalidate_disk	= sd_revalidate_disk,
1669 	.unlock_native_capacity	= sd_unlock_native_capacity,
1670 	.pr_ops			= &sd_pr_ops,
1671 };
1672 
1673 /**
1674  *	sd_eh_action - error handling callback
1675  *	@scmd:		sd-issued command that has failed
1676  *	@eh_disp:	The recovery disposition suggested by the midlayer
1677  *
1678  *	This function is called by the SCSI midlayer upon completion of an
1679  *	error test command (currently TEST UNIT READY). The result of sending
1680  *	the eh command is passed in eh_disp.  We're looking for devices that
1681  *	fail medium access commands but are OK with non access commands like
1682  *	test unit ready (so wrongly see the device as having a successful
1683  *	recovery)
1684  **/
1685 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1686 {
1687 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1688 
1689 	if (!scsi_device_online(scmd->device) ||
1690 	    !scsi_medium_access_command(scmd) ||
1691 	    host_byte(scmd->result) != DID_TIME_OUT ||
1692 	    eh_disp != SUCCESS)
1693 		return eh_disp;
1694 
1695 	/*
1696 	 * The device has timed out executing a medium access command.
1697 	 * However, the TEST UNIT READY command sent during error
1698 	 * handling completed successfully. Either the device is in the
1699 	 * process of recovering or has it suffered an internal failure
1700 	 * that prevents access to the storage medium.
1701 	 */
1702 	sdkp->medium_access_timed_out++;
1703 
1704 	/*
1705 	 * If the device keeps failing read/write commands but TEST UNIT
1706 	 * READY always completes successfully we assume that medium
1707 	 * access is no longer possible and take the device offline.
1708 	 */
1709 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1710 		scmd_printk(KERN_ERR, scmd,
1711 			    "Medium access timeout failure. Offlining disk!\n");
1712 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1713 
1714 		return FAILED;
1715 	}
1716 
1717 	return eh_disp;
1718 }
1719 
1720 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1721 {
1722 	u64 start_lba = blk_rq_pos(scmd->request);
1723 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1724 	u64 factor = scmd->device->sector_size / 512;
1725 	u64 bad_lba;
1726 	int info_valid;
1727 	/*
1728 	 * resid is optional but mostly filled in.  When it's unused,
1729 	 * its value is zero, so we assume the whole buffer transferred
1730 	 */
1731 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1732 	unsigned int good_bytes;
1733 
1734 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1735 		return 0;
1736 
1737 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1738 					     SCSI_SENSE_BUFFERSIZE,
1739 					     &bad_lba);
1740 	if (!info_valid)
1741 		return 0;
1742 
1743 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1744 		return 0;
1745 
1746 	/* be careful ... don't want any overflows */
1747 	do_div(start_lba, factor);
1748 	do_div(end_lba, factor);
1749 
1750 	/* The bad lba was reported incorrectly, we have no idea where
1751 	 * the error is.
1752 	 */
1753 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1754 		return 0;
1755 
1756 	/* This computation should always be done in terms of
1757 	 * the resolution of the device's medium.
1758 	 */
1759 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1760 	return min(good_bytes, transferred);
1761 }
1762 
1763 /**
1764  *	sd_done - bottom half handler: called when the lower level
1765  *	driver has completed (successfully or otherwise) a scsi command.
1766  *	@SCpnt: mid-level's per command structure.
1767  *
1768  *	Note: potentially run from within an ISR. Must not block.
1769  **/
1770 static int sd_done(struct scsi_cmnd *SCpnt)
1771 {
1772 	int result = SCpnt->result;
1773 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1774 	struct scsi_sense_hdr sshdr;
1775 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1776 	struct request *req = SCpnt->request;
1777 	int sense_valid = 0;
1778 	int sense_deferred = 0;
1779 	unsigned char op = SCpnt->cmnd[0];
1780 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1781 
1782 	if (req_op(req) == REQ_OP_DISCARD || req_op(req) == REQ_OP_WRITE_SAME) {
1783 		if (!result) {
1784 			good_bytes = blk_rq_bytes(req);
1785 			scsi_set_resid(SCpnt, 0);
1786 		} else {
1787 			good_bytes = 0;
1788 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1789 		}
1790 	}
1791 
1792 	if (result) {
1793 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1794 		if (sense_valid)
1795 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1796 	}
1797 	sdkp->medium_access_timed_out = 0;
1798 
1799 	if (driver_byte(result) != DRIVER_SENSE &&
1800 	    (!sense_valid || sense_deferred))
1801 		goto out;
1802 
1803 	switch (sshdr.sense_key) {
1804 	case HARDWARE_ERROR:
1805 	case MEDIUM_ERROR:
1806 		good_bytes = sd_completed_bytes(SCpnt);
1807 		break;
1808 	case RECOVERED_ERROR:
1809 		good_bytes = scsi_bufflen(SCpnt);
1810 		break;
1811 	case NO_SENSE:
1812 		/* This indicates a false check condition, so ignore it.  An
1813 		 * unknown amount of data was transferred so treat it as an
1814 		 * error.
1815 		 */
1816 		SCpnt->result = 0;
1817 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1818 		break;
1819 	case ABORTED_COMMAND:
1820 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1821 			good_bytes = sd_completed_bytes(SCpnt);
1822 		break;
1823 	case ILLEGAL_REQUEST:
1824 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1825 			good_bytes = sd_completed_bytes(SCpnt);
1826 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1827 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1828 			switch (op) {
1829 			case UNMAP:
1830 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1831 				break;
1832 			case WRITE_SAME_16:
1833 			case WRITE_SAME:
1834 				if (unmap)
1835 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1836 				else {
1837 					sdkp->device->no_write_same = 1;
1838 					sd_config_write_same(sdkp);
1839 
1840 					good_bytes = 0;
1841 					req->__data_len = blk_rq_bytes(req);
1842 					req->cmd_flags |= REQ_QUIET;
1843 				}
1844 			}
1845 		}
1846 		break;
1847 	default:
1848 		break;
1849 	}
1850  out:
1851 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1852 					   "sd_done: completed %d of %d bytes\n",
1853 					   good_bytes, scsi_bufflen(SCpnt)));
1854 
1855 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1856 		sd_dif_complete(SCpnt, good_bytes);
1857 
1858 	return good_bytes;
1859 }
1860 
1861 /*
1862  * spinup disk - called only in sd_revalidate_disk()
1863  */
1864 static void
1865 sd_spinup_disk(struct scsi_disk *sdkp)
1866 {
1867 	unsigned char cmd[10];
1868 	unsigned long spintime_expire = 0;
1869 	int retries, spintime;
1870 	unsigned int the_result;
1871 	struct scsi_sense_hdr sshdr;
1872 	int sense_valid = 0;
1873 
1874 	spintime = 0;
1875 
1876 	/* Spin up drives, as required.  Only do this at boot time */
1877 	/* Spinup needs to be done for module loads too. */
1878 	do {
1879 		retries = 0;
1880 
1881 		do {
1882 			cmd[0] = TEST_UNIT_READY;
1883 			memset((void *) &cmd[1], 0, 9);
1884 
1885 			the_result = scsi_execute_req(sdkp->device, cmd,
1886 						      DMA_NONE, NULL, 0,
1887 						      &sshdr, SD_TIMEOUT,
1888 						      SD_MAX_RETRIES, NULL);
1889 
1890 			/*
1891 			 * If the drive has indicated to us that it
1892 			 * doesn't have any media in it, don't bother
1893 			 * with any more polling.
1894 			 */
1895 			if (media_not_present(sdkp, &sshdr))
1896 				return;
1897 
1898 			if (the_result)
1899 				sense_valid = scsi_sense_valid(&sshdr);
1900 			retries++;
1901 		} while (retries < 3 &&
1902 			 (!scsi_status_is_good(the_result) ||
1903 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1904 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1905 
1906 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1907 			/* no sense, TUR either succeeded or failed
1908 			 * with a status error */
1909 			if(!spintime && !scsi_status_is_good(the_result)) {
1910 				sd_print_result(sdkp, "Test Unit Ready failed",
1911 						the_result);
1912 			}
1913 			break;
1914 		}
1915 
1916 		/*
1917 		 * The device does not want the automatic start to be issued.
1918 		 */
1919 		if (sdkp->device->no_start_on_add)
1920 			break;
1921 
1922 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1923 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1924 				break;	/* manual intervention required */
1925 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1926 				break;	/* standby */
1927 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1928 				break;	/* unavailable */
1929 			/*
1930 			 * Issue command to spin up drive when not ready
1931 			 */
1932 			if (!spintime) {
1933 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1934 				cmd[0] = START_STOP;
1935 				cmd[1] = 1;	/* Return immediately */
1936 				memset((void *) &cmd[2], 0, 8);
1937 				cmd[4] = 1;	/* Start spin cycle */
1938 				if (sdkp->device->start_stop_pwr_cond)
1939 					cmd[4] |= 1 << 4;
1940 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1941 						 NULL, 0, &sshdr,
1942 						 SD_TIMEOUT, SD_MAX_RETRIES,
1943 						 NULL);
1944 				spintime_expire = jiffies + 100 * HZ;
1945 				spintime = 1;
1946 			}
1947 			/* Wait 1 second for next try */
1948 			msleep(1000);
1949 			printk(".");
1950 
1951 		/*
1952 		 * Wait for USB flash devices with slow firmware.
1953 		 * Yes, this sense key/ASC combination shouldn't
1954 		 * occur here.  It's characteristic of these devices.
1955 		 */
1956 		} else if (sense_valid &&
1957 				sshdr.sense_key == UNIT_ATTENTION &&
1958 				sshdr.asc == 0x28) {
1959 			if (!spintime) {
1960 				spintime_expire = jiffies + 5 * HZ;
1961 				spintime = 1;
1962 			}
1963 			/* Wait 1 second for next try */
1964 			msleep(1000);
1965 		} else {
1966 			/* we don't understand the sense code, so it's
1967 			 * probably pointless to loop */
1968 			if(!spintime) {
1969 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1970 				sd_print_sense_hdr(sdkp, &sshdr);
1971 			}
1972 			break;
1973 		}
1974 
1975 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1976 
1977 	if (spintime) {
1978 		if (scsi_status_is_good(the_result))
1979 			printk("ready\n");
1980 		else
1981 			printk("not responding...\n");
1982 	}
1983 }
1984 
1985 
1986 /*
1987  * Determine whether disk supports Data Integrity Field.
1988  */
1989 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1990 {
1991 	struct scsi_device *sdp = sdkp->device;
1992 	u8 type;
1993 	int ret = 0;
1994 
1995 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1996 		return ret;
1997 
1998 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1999 
2000 	if (type > SD_DIF_TYPE3_PROTECTION)
2001 		ret = -ENODEV;
2002 	else if (scsi_host_dif_capable(sdp->host, type))
2003 		ret = 1;
2004 
2005 	if (sdkp->first_scan || type != sdkp->protection_type)
2006 		switch (ret) {
2007 		case -ENODEV:
2008 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2009 				  " protection type %u. Disabling disk!\n",
2010 				  type);
2011 			break;
2012 		case 1:
2013 			sd_printk(KERN_NOTICE, sdkp,
2014 				  "Enabling DIF Type %u protection\n", type);
2015 			break;
2016 		case 0:
2017 			sd_printk(KERN_NOTICE, sdkp,
2018 				  "Disabling DIF Type %u protection\n", type);
2019 			break;
2020 		}
2021 
2022 	sdkp->protection_type = type;
2023 
2024 	return ret;
2025 }
2026 
2027 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2028 			struct scsi_sense_hdr *sshdr, int sense_valid,
2029 			int the_result)
2030 {
2031 	if (driver_byte(the_result) & DRIVER_SENSE)
2032 		sd_print_sense_hdr(sdkp, sshdr);
2033 	else
2034 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2035 
2036 	/*
2037 	 * Set dirty bit for removable devices if not ready -
2038 	 * sometimes drives will not report this properly.
2039 	 */
2040 	if (sdp->removable &&
2041 	    sense_valid && sshdr->sense_key == NOT_READY)
2042 		set_media_not_present(sdkp);
2043 
2044 	/*
2045 	 * We used to set media_present to 0 here to indicate no media
2046 	 * in the drive, but some drives fail read capacity even with
2047 	 * media present, so we can't do that.
2048 	 */
2049 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2050 }
2051 
2052 #define RC16_LEN 32
2053 #if RC16_LEN > SD_BUF_SIZE
2054 #error RC16_LEN must not be more than SD_BUF_SIZE
2055 #endif
2056 
2057 #define READ_CAPACITY_RETRIES_ON_RESET	10
2058 
2059 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2060 						unsigned char *buffer)
2061 {
2062 	unsigned char cmd[16];
2063 	struct scsi_sense_hdr sshdr;
2064 	int sense_valid = 0;
2065 	int the_result;
2066 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2067 	unsigned int alignment;
2068 	unsigned long long lba;
2069 	unsigned sector_size;
2070 
2071 	if (sdp->no_read_capacity_16)
2072 		return -EINVAL;
2073 
2074 	do {
2075 		memset(cmd, 0, 16);
2076 		cmd[0] = SERVICE_ACTION_IN_16;
2077 		cmd[1] = SAI_READ_CAPACITY_16;
2078 		cmd[13] = RC16_LEN;
2079 		memset(buffer, 0, RC16_LEN);
2080 
2081 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2082 					buffer, RC16_LEN, &sshdr,
2083 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2084 
2085 		if (media_not_present(sdkp, &sshdr))
2086 			return -ENODEV;
2087 
2088 		if (the_result) {
2089 			sense_valid = scsi_sense_valid(&sshdr);
2090 			if (sense_valid &&
2091 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2092 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2093 			    sshdr.ascq == 0x00)
2094 				/* Invalid Command Operation Code or
2095 				 * Invalid Field in CDB, just retry
2096 				 * silently with RC10 */
2097 				return -EINVAL;
2098 			if (sense_valid &&
2099 			    sshdr.sense_key == UNIT_ATTENTION &&
2100 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2101 				/* Device reset might occur several times,
2102 				 * give it one more chance */
2103 				if (--reset_retries > 0)
2104 					continue;
2105 		}
2106 		retries--;
2107 
2108 	} while (the_result && retries);
2109 
2110 	if (the_result) {
2111 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2112 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2113 		return -EINVAL;
2114 	}
2115 
2116 	sector_size = get_unaligned_be32(&buffer[8]);
2117 	lba = get_unaligned_be64(&buffer[0]);
2118 
2119 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2120 		sdkp->capacity = 0;
2121 		return -ENODEV;
2122 	}
2123 
2124 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2125 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2126 			"kernel compiled with support for large block "
2127 			"devices.\n");
2128 		sdkp->capacity = 0;
2129 		return -EOVERFLOW;
2130 	}
2131 
2132 	/* Logical blocks per physical block exponent */
2133 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2134 
2135 	/* Lowest aligned logical block */
2136 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2137 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2138 	if (alignment && sdkp->first_scan)
2139 		sd_printk(KERN_NOTICE, sdkp,
2140 			  "physical block alignment offset: %u\n", alignment);
2141 
2142 	if (buffer[14] & 0x80) { /* LBPME */
2143 		sdkp->lbpme = 1;
2144 
2145 		if (buffer[14] & 0x40) /* LBPRZ */
2146 			sdkp->lbprz = 1;
2147 
2148 		sd_config_discard(sdkp, SD_LBP_WS16);
2149 	}
2150 
2151 	sdkp->capacity = lba + 1;
2152 	return sector_size;
2153 }
2154 
2155 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2156 						unsigned char *buffer)
2157 {
2158 	unsigned char cmd[16];
2159 	struct scsi_sense_hdr sshdr;
2160 	int sense_valid = 0;
2161 	int the_result;
2162 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2163 	sector_t lba;
2164 	unsigned sector_size;
2165 
2166 	do {
2167 		cmd[0] = READ_CAPACITY;
2168 		memset(&cmd[1], 0, 9);
2169 		memset(buffer, 0, 8);
2170 
2171 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2172 					buffer, 8, &sshdr,
2173 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2174 
2175 		if (media_not_present(sdkp, &sshdr))
2176 			return -ENODEV;
2177 
2178 		if (the_result) {
2179 			sense_valid = scsi_sense_valid(&sshdr);
2180 			if (sense_valid &&
2181 			    sshdr.sense_key == UNIT_ATTENTION &&
2182 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2183 				/* Device reset might occur several times,
2184 				 * give it one more chance */
2185 				if (--reset_retries > 0)
2186 					continue;
2187 		}
2188 		retries--;
2189 
2190 	} while (the_result && retries);
2191 
2192 	if (the_result) {
2193 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2194 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2195 		return -EINVAL;
2196 	}
2197 
2198 	sector_size = get_unaligned_be32(&buffer[4]);
2199 	lba = get_unaligned_be32(&buffer[0]);
2200 
2201 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2202 		/* Some buggy (usb cardreader) devices return an lba of
2203 		   0xffffffff when the want to report a size of 0 (with
2204 		   which they really mean no media is present) */
2205 		sdkp->capacity = 0;
2206 		sdkp->physical_block_size = sector_size;
2207 		return sector_size;
2208 	}
2209 
2210 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2211 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2212 			"kernel compiled with support for large block "
2213 			"devices.\n");
2214 		sdkp->capacity = 0;
2215 		return -EOVERFLOW;
2216 	}
2217 
2218 	sdkp->capacity = lba + 1;
2219 	sdkp->physical_block_size = sector_size;
2220 	return sector_size;
2221 }
2222 
2223 static int sd_try_rc16_first(struct scsi_device *sdp)
2224 {
2225 	if (sdp->host->max_cmd_len < 16)
2226 		return 0;
2227 	if (sdp->try_rc_10_first)
2228 		return 0;
2229 	if (sdp->scsi_level > SCSI_SPC_2)
2230 		return 1;
2231 	if (scsi_device_protection(sdp))
2232 		return 1;
2233 	return 0;
2234 }
2235 
2236 /*
2237  * read disk capacity
2238  */
2239 static void
2240 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2241 {
2242 	int sector_size;
2243 	struct scsi_device *sdp = sdkp->device;
2244 	sector_t old_capacity = sdkp->capacity;
2245 
2246 	if (sd_try_rc16_first(sdp)) {
2247 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2248 		if (sector_size == -EOVERFLOW)
2249 			goto got_data;
2250 		if (sector_size == -ENODEV)
2251 			return;
2252 		if (sector_size < 0)
2253 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2254 		if (sector_size < 0)
2255 			return;
2256 	} else {
2257 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2258 		if (sector_size == -EOVERFLOW)
2259 			goto got_data;
2260 		if (sector_size < 0)
2261 			return;
2262 		if ((sizeof(sdkp->capacity) > 4) &&
2263 		    (sdkp->capacity > 0xffffffffULL)) {
2264 			int old_sector_size = sector_size;
2265 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2266 					"Trying to use READ CAPACITY(16).\n");
2267 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2268 			if (sector_size < 0) {
2269 				sd_printk(KERN_NOTICE, sdkp,
2270 					"Using 0xffffffff as device size\n");
2271 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2272 				sector_size = old_sector_size;
2273 				goto got_data;
2274 			}
2275 		}
2276 	}
2277 
2278 	/* Some devices are known to return the total number of blocks,
2279 	 * not the highest block number.  Some devices have versions
2280 	 * which do this and others which do not.  Some devices we might
2281 	 * suspect of doing this but we don't know for certain.
2282 	 *
2283 	 * If we know the reported capacity is wrong, decrement it.  If
2284 	 * we can only guess, then assume the number of blocks is even
2285 	 * (usually true but not always) and err on the side of lowering
2286 	 * the capacity.
2287 	 */
2288 	if (sdp->fix_capacity ||
2289 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2290 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2291 				"from its reported value: %llu\n",
2292 				(unsigned long long) sdkp->capacity);
2293 		--sdkp->capacity;
2294 	}
2295 
2296 got_data:
2297 	if (sector_size == 0) {
2298 		sector_size = 512;
2299 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2300 			  "assuming 512.\n");
2301 	}
2302 
2303 	if (sector_size != 512 &&
2304 	    sector_size != 1024 &&
2305 	    sector_size != 2048 &&
2306 	    sector_size != 4096) {
2307 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2308 			  sector_size);
2309 		/*
2310 		 * The user might want to re-format the drive with
2311 		 * a supported sectorsize.  Once this happens, it
2312 		 * would be relatively trivial to set the thing up.
2313 		 * For this reason, we leave the thing in the table.
2314 		 */
2315 		sdkp->capacity = 0;
2316 		/*
2317 		 * set a bogus sector size so the normal read/write
2318 		 * logic in the block layer will eventually refuse any
2319 		 * request on this device without tripping over power
2320 		 * of two sector size assumptions
2321 		 */
2322 		sector_size = 512;
2323 	}
2324 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2325 
2326 	{
2327 		char cap_str_2[10], cap_str_10[10];
2328 
2329 		string_get_size(sdkp->capacity, sector_size,
2330 				STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2331 		string_get_size(sdkp->capacity, sector_size,
2332 				STRING_UNITS_10, cap_str_10,
2333 				sizeof(cap_str_10));
2334 
2335 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2336 			sd_printk(KERN_NOTICE, sdkp,
2337 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2338 				  (unsigned long long)sdkp->capacity,
2339 				  sector_size, cap_str_10, cap_str_2);
2340 
2341 			if (sdkp->physical_block_size != sector_size)
2342 				sd_printk(KERN_NOTICE, sdkp,
2343 					  "%u-byte physical blocks\n",
2344 					  sdkp->physical_block_size);
2345 		}
2346 	}
2347 
2348 	if (sdkp->capacity > 0xffffffff)
2349 		sdp->use_16_for_rw = 1;
2350 
2351 	blk_queue_physical_block_size(sdp->request_queue,
2352 				      sdkp->physical_block_size);
2353 	sdkp->device->sector_size = sector_size;
2354 }
2355 
2356 /* called with buffer of length 512 */
2357 static inline int
2358 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2359 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2360 		 struct scsi_sense_hdr *sshdr)
2361 {
2362 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2363 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2364 			       sshdr);
2365 }
2366 
2367 /*
2368  * read write protect setting, if possible - called only in sd_revalidate_disk()
2369  * called with buffer of length SD_BUF_SIZE
2370  */
2371 static void
2372 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2373 {
2374 	int res;
2375 	struct scsi_device *sdp = sdkp->device;
2376 	struct scsi_mode_data data;
2377 	int old_wp = sdkp->write_prot;
2378 
2379 	set_disk_ro(sdkp->disk, 0);
2380 	if (sdp->skip_ms_page_3f) {
2381 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2382 		return;
2383 	}
2384 
2385 	if (sdp->use_192_bytes_for_3f) {
2386 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2387 	} else {
2388 		/*
2389 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2390 		 * We have to start carefully: some devices hang if we ask
2391 		 * for more than is available.
2392 		 */
2393 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2394 
2395 		/*
2396 		 * Second attempt: ask for page 0 When only page 0 is
2397 		 * implemented, a request for page 3F may return Sense Key
2398 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2399 		 * CDB.
2400 		 */
2401 		if (!scsi_status_is_good(res))
2402 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2403 
2404 		/*
2405 		 * Third attempt: ask 255 bytes, as we did earlier.
2406 		 */
2407 		if (!scsi_status_is_good(res))
2408 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2409 					       &data, NULL);
2410 	}
2411 
2412 	if (!scsi_status_is_good(res)) {
2413 		sd_first_printk(KERN_WARNING, sdkp,
2414 			  "Test WP failed, assume Write Enabled\n");
2415 	} else {
2416 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2417 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2418 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2419 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2420 				  sdkp->write_prot ? "on" : "off");
2421 			sd_printk(KERN_DEBUG, sdkp,
2422 				  "Mode Sense: %02x %02x %02x %02x\n",
2423 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2424 		}
2425 	}
2426 }
2427 
2428 /*
2429  * sd_read_cache_type - called only from sd_revalidate_disk()
2430  * called with buffer of length SD_BUF_SIZE
2431  */
2432 static void
2433 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2434 {
2435 	int len = 0, res;
2436 	struct scsi_device *sdp = sdkp->device;
2437 
2438 	int dbd;
2439 	int modepage;
2440 	int first_len;
2441 	struct scsi_mode_data data;
2442 	struct scsi_sense_hdr sshdr;
2443 	int old_wce = sdkp->WCE;
2444 	int old_rcd = sdkp->RCD;
2445 	int old_dpofua = sdkp->DPOFUA;
2446 
2447 
2448 	if (sdkp->cache_override)
2449 		return;
2450 
2451 	first_len = 4;
2452 	if (sdp->skip_ms_page_8) {
2453 		if (sdp->type == TYPE_RBC)
2454 			goto defaults;
2455 		else {
2456 			if (sdp->skip_ms_page_3f)
2457 				goto defaults;
2458 			modepage = 0x3F;
2459 			if (sdp->use_192_bytes_for_3f)
2460 				first_len = 192;
2461 			dbd = 0;
2462 		}
2463 	} else if (sdp->type == TYPE_RBC) {
2464 		modepage = 6;
2465 		dbd = 8;
2466 	} else {
2467 		modepage = 8;
2468 		dbd = 0;
2469 	}
2470 
2471 	/* cautiously ask */
2472 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2473 			&data, &sshdr);
2474 
2475 	if (!scsi_status_is_good(res))
2476 		goto bad_sense;
2477 
2478 	if (!data.header_length) {
2479 		modepage = 6;
2480 		first_len = 0;
2481 		sd_first_printk(KERN_ERR, sdkp,
2482 				"Missing header in MODE_SENSE response\n");
2483 	}
2484 
2485 	/* that went OK, now ask for the proper length */
2486 	len = data.length;
2487 
2488 	/*
2489 	 * We're only interested in the first three bytes, actually.
2490 	 * But the data cache page is defined for the first 20.
2491 	 */
2492 	if (len < 3)
2493 		goto bad_sense;
2494 	else if (len > SD_BUF_SIZE) {
2495 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2496 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2497 		len = SD_BUF_SIZE;
2498 	}
2499 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2500 		len = 192;
2501 
2502 	/* Get the data */
2503 	if (len > first_len)
2504 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2505 				&data, &sshdr);
2506 
2507 	if (scsi_status_is_good(res)) {
2508 		int offset = data.header_length + data.block_descriptor_length;
2509 
2510 		while (offset < len) {
2511 			u8 page_code = buffer[offset] & 0x3F;
2512 			u8 spf       = buffer[offset] & 0x40;
2513 
2514 			if (page_code == 8 || page_code == 6) {
2515 				/* We're interested only in the first 3 bytes.
2516 				 */
2517 				if (len - offset <= 2) {
2518 					sd_first_printk(KERN_ERR, sdkp,
2519 						"Incomplete mode parameter "
2520 							"data\n");
2521 					goto defaults;
2522 				} else {
2523 					modepage = page_code;
2524 					goto Page_found;
2525 				}
2526 			} else {
2527 				/* Go to the next page */
2528 				if (spf && len - offset > 3)
2529 					offset += 4 + (buffer[offset+2] << 8) +
2530 						buffer[offset+3];
2531 				else if (!spf && len - offset > 1)
2532 					offset += 2 + buffer[offset+1];
2533 				else {
2534 					sd_first_printk(KERN_ERR, sdkp,
2535 							"Incomplete mode "
2536 							"parameter data\n");
2537 					goto defaults;
2538 				}
2539 			}
2540 		}
2541 
2542 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2543 		goto defaults;
2544 
2545 	Page_found:
2546 		if (modepage == 8) {
2547 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2548 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2549 		} else {
2550 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2551 			sdkp->RCD = 0;
2552 		}
2553 
2554 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2555 		if (sdp->broken_fua) {
2556 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2557 			sdkp->DPOFUA = 0;
2558 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2559 			sd_first_printk(KERN_NOTICE, sdkp,
2560 				  "Uses READ/WRITE(6), disabling FUA\n");
2561 			sdkp->DPOFUA = 0;
2562 		}
2563 
2564 		/* No cache flush allowed for write protected devices */
2565 		if (sdkp->WCE && sdkp->write_prot)
2566 			sdkp->WCE = 0;
2567 
2568 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2569 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2570 			sd_printk(KERN_NOTICE, sdkp,
2571 				  "Write cache: %s, read cache: %s, %s\n",
2572 				  sdkp->WCE ? "enabled" : "disabled",
2573 				  sdkp->RCD ? "disabled" : "enabled",
2574 				  sdkp->DPOFUA ? "supports DPO and FUA"
2575 				  : "doesn't support DPO or FUA");
2576 
2577 		return;
2578 	}
2579 
2580 bad_sense:
2581 	if (scsi_sense_valid(&sshdr) &&
2582 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2583 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2584 		/* Invalid field in CDB */
2585 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2586 	else
2587 		sd_first_printk(KERN_ERR, sdkp,
2588 				"Asking for cache data failed\n");
2589 
2590 defaults:
2591 	if (sdp->wce_default_on) {
2592 		sd_first_printk(KERN_NOTICE, sdkp,
2593 				"Assuming drive cache: write back\n");
2594 		sdkp->WCE = 1;
2595 	} else {
2596 		sd_first_printk(KERN_ERR, sdkp,
2597 				"Assuming drive cache: write through\n");
2598 		sdkp->WCE = 0;
2599 	}
2600 	sdkp->RCD = 0;
2601 	sdkp->DPOFUA = 0;
2602 }
2603 
2604 /*
2605  * The ATO bit indicates whether the DIF application tag is available
2606  * for use by the operating system.
2607  */
2608 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2609 {
2610 	int res, offset;
2611 	struct scsi_device *sdp = sdkp->device;
2612 	struct scsi_mode_data data;
2613 	struct scsi_sense_hdr sshdr;
2614 
2615 	if (sdp->type != TYPE_DISK)
2616 		return;
2617 
2618 	if (sdkp->protection_type == 0)
2619 		return;
2620 
2621 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2622 			      SD_MAX_RETRIES, &data, &sshdr);
2623 
2624 	if (!scsi_status_is_good(res) || !data.header_length ||
2625 	    data.length < 6) {
2626 		sd_first_printk(KERN_WARNING, sdkp,
2627 			  "getting Control mode page failed, assume no ATO\n");
2628 
2629 		if (scsi_sense_valid(&sshdr))
2630 			sd_print_sense_hdr(sdkp, &sshdr);
2631 
2632 		return;
2633 	}
2634 
2635 	offset = data.header_length + data.block_descriptor_length;
2636 
2637 	if ((buffer[offset] & 0x3f) != 0x0a) {
2638 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2639 		return;
2640 	}
2641 
2642 	if ((buffer[offset + 5] & 0x80) == 0)
2643 		return;
2644 
2645 	sdkp->ATO = 1;
2646 
2647 	return;
2648 }
2649 
2650 /**
2651  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2652  * @disk: disk to query
2653  */
2654 static void sd_read_block_limits(struct scsi_disk *sdkp)
2655 {
2656 	unsigned int sector_sz = sdkp->device->sector_size;
2657 	const int vpd_len = 64;
2658 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2659 
2660 	if (!buffer ||
2661 	    /* Block Limits VPD */
2662 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2663 		goto out;
2664 
2665 	blk_queue_io_min(sdkp->disk->queue,
2666 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2667 
2668 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2669 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2670 
2671 	if (buffer[3] == 0x3c) {
2672 		unsigned int lba_count, desc_count;
2673 
2674 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2675 
2676 		if (!sdkp->lbpme)
2677 			goto out;
2678 
2679 		lba_count = get_unaligned_be32(&buffer[20]);
2680 		desc_count = get_unaligned_be32(&buffer[24]);
2681 
2682 		if (lba_count && desc_count)
2683 			sdkp->max_unmap_blocks = lba_count;
2684 
2685 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2686 
2687 		if (buffer[32] & 0x80)
2688 			sdkp->unmap_alignment =
2689 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2690 
2691 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2692 
2693 			if (sdkp->max_unmap_blocks)
2694 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2695 			else
2696 				sd_config_discard(sdkp, SD_LBP_WS16);
2697 
2698 		} else {	/* LBP VPD page tells us what to use */
2699 			if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2700 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2701 			else if (sdkp->lbpws)
2702 				sd_config_discard(sdkp, SD_LBP_WS16);
2703 			else if (sdkp->lbpws10)
2704 				sd_config_discard(sdkp, SD_LBP_WS10);
2705 			else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2706 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2707 			else
2708 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2709 		}
2710 	}
2711 
2712  out:
2713 	kfree(buffer);
2714 }
2715 
2716 /**
2717  * sd_read_block_characteristics - Query block dev. characteristics
2718  * @disk: disk to query
2719  */
2720 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2721 {
2722 	unsigned char *buffer;
2723 	u16 rot;
2724 	const int vpd_len = 64;
2725 
2726 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2727 
2728 	if (!buffer ||
2729 	    /* Block Device Characteristics VPD */
2730 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2731 		goto out;
2732 
2733 	rot = get_unaligned_be16(&buffer[4]);
2734 
2735 	if (rot == 1) {
2736 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2737 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, sdkp->disk->queue);
2738 	}
2739 
2740  out:
2741 	kfree(buffer);
2742 }
2743 
2744 /**
2745  * sd_read_block_provisioning - Query provisioning VPD page
2746  * @disk: disk to query
2747  */
2748 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2749 {
2750 	unsigned char *buffer;
2751 	const int vpd_len = 8;
2752 
2753 	if (sdkp->lbpme == 0)
2754 		return;
2755 
2756 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2757 
2758 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2759 		goto out;
2760 
2761 	sdkp->lbpvpd	= 1;
2762 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2763 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2764 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2765 
2766  out:
2767 	kfree(buffer);
2768 }
2769 
2770 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2771 {
2772 	struct scsi_device *sdev = sdkp->device;
2773 
2774 	if (sdev->host->no_write_same) {
2775 		sdev->no_write_same = 1;
2776 
2777 		return;
2778 	}
2779 
2780 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2781 		/* too large values might cause issues with arcmsr */
2782 		int vpd_buf_len = 64;
2783 
2784 		sdev->no_report_opcodes = 1;
2785 
2786 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2787 		 * CODES is unsupported and the device has an ATA
2788 		 * Information VPD page (SAT).
2789 		 */
2790 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2791 			sdev->no_write_same = 1;
2792 	}
2793 
2794 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2795 		sdkp->ws16 = 1;
2796 
2797 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2798 		sdkp->ws10 = 1;
2799 }
2800 
2801 /**
2802  *	sd_revalidate_disk - called the first time a new disk is seen,
2803  *	performs disk spin up, read_capacity, etc.
2804  *	@disk: struct gendisk we care about
2805  **/
2806 static int sd_revalidate_disk(struct gendisk *disk)
2807 {
2808 	struct scsi_disk *sdkp = scsi_disk(disk);
2809 	struct scsi_device *sdp = sdkp->device;
2810 	struct request_queue *q = sdkp->disk->queue;
2811 	unsigned char *buffer;
2812 	unsigned int dev_max, rw_max;
2813 
2814 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2815 				      "sd_revalidate_disk\n"));
2816 
2817 	/*
2818 	 * If the device is offline, don't try and read capacity or any
2819 	 * of the other niceties.
2820 	 */
2821 	if (!scsi_device_online(sdp))
2822 		goto out;
2823 
2824 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2825 	if (!buffer) {
2826 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2827 			  "allocation failure.\n");
2828 		goto out;
2829 	}
2830 
2831 	sd_spinup_disk(sdkp);
2832 
2833 	/*
2834 	 * Without media there is no reason to ask; moreover, some devices
2835 	 * react badly if we do.
2836 	 */
2837 	if (sdkp->media_present) {
2838 		sd_read_capacity(sdkp, buffer);
2839 
2840 		if (scsi_device_supports_vpd(sdp)) {
2841 			sd_read_block_provisioning(sdkp);
2842 			sd_read_block_limits(sdkp);
2843 			sd_read_block_characteristics(sdkp);
2844 		}
2845 
2846 		sd_read_write_protect_flag(sdkp, buffer);
2847 		sd_read_cache_type(sdkp, buffer);
2848 		sd_read_app_tag_own(sdkp, buffer);
2849 		sd_read_write_same(sdkp, buffer);
2850 	}
2851 
2852 	sdkp->first_scan = 0;
2853 
2854 	/*
2855 	 * We now have all cache related info, determine how we deal
2856 	 * with flush requests.
2857 	 */
2858 	sd_set_flush_flag(sdkp);
2859 
2860 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2861 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2862 
2863 	/* Some devices report a maximum block count for READ/WRITE requests. */
2864 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2865 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2866 
2867 	/*
2868 	 * Use the device's preferred I/O size for reads and writes
2869 	 * unless the reported value is unreasonably small, large, or
2870 	 * garbage.
2871 	 */
2872 	if (sdkp->opt_xfer_blocks &&
2873 	    sdkp->opt_xfer_blocks <= dev_max &&
2874 	    sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2875 	    logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2876 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2877 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2878 	} else
2879 		rw_max = BLK_DEF_MAX_SECTORS;
2880 
2881 	/* Combine with controller limits */
2882 	q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2883 
2884 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2885 	sd_config_write_same(sdkp);
2886 	kfree(buffer);
2887 
2888  out:
2889 	return 0;
2890 }
2891 
2892 /**
2893  *	sd_unlock_native_capacity - unlock native capacity
2894  *	@disk: struct gendisk to set capacity for
2895  *
2896  *	Block layer calls this function if it detects that partitions
2897  *	on @disk reach beyond the end of the device.  If the SCSI host
2898  *	implements ->unlock_native_capacity() method, it's invoked to
2899  *	give it a chance to adjust the device capacity.
2900  *
2901  *	CONTEXT:
2902  *	Defined by block layer.  Might sleep.
2903  */
2904 static void sd_unlock_native_capacity(struct gendisk *disk)
2905 {
2906 	struct scsi_device *sdev = scsi_disk(disk)->device;
2907 
2908 	if (sdev->host->hostt->unlock_native_capacity)
2909 		sdev->host->hostt->unlock_native_capacity(sdev);
2910 }
2911 
2912 /**
2913  *	sd_format_disk_name - format disk name
2914  *	@prefix: name prefix - ie. "sd" for SCSI disks
2915  *	@index: index of the disk to format name for
2916  *	@buf: output buffer
2917  *	@buflen: length of the output buffer
2918  *
2919  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2920  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2921  *	which is followed by sdaaa.
2922  *
2923  *	This is basically 26 base counting with one extra 'nil' entry
2924  *	at the beginning from the second digit on and can be
2925  *	determined using similar method as 26 base conversion with the
2926  *	index shifted -1 after each digit is computed.
2927  *
2928  *	CONTEXT:
2929  *	Don't care.
2930  *
2931  *	RETURNS:
2932  *	0 on success, -errno on failure.
2933  */
2934 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2935 {
2936 	const int base = 'z' - 'a' + 1;
2937 	char *begin = buf + strlen(prefix);
2938 	char *end = buf + buflen;
2939 	char *p;
2940 	int unit;
2941 
2942 	p = end - 1;
2943 	*p = '\0';
2944 	unit = base;
2945 	do {
2946 		if (p == begin)
2947 			return -EINVAL;
2948 		*--p = 'a' + (index % unit);
2949 		index = (index / unit) - 1;
2950 	} while (index >= 0);
2951 
2952 	memmove(begin, p, end - p);
2953 	memcpy(buf, prefix, strlen(prefix));
2954 
2955 	return 0;
2956 }
2957 
2958 /*
2959  * The asynchronous part of sd_probe
2960  */
2961 static void sd_probe_async(void *data, async_cookie_t cookie)
2962 {
2963 	struct scsi_disk *sdkp = data;
2964 	struct scsi_device *sdp;
2965 	struct gendisk *gd;
2966 	u32 index;
2967 	struct device *dev;
2968 
2969 	sdp = sdkp->device;
2970 	gd = sdkp->disk;
2971 	index = sdkp->index;
2972 	dev = &sdp->sdev_gendev;
2973 
2974 	gd->major = sd_major((index & 0xf0) >> 4);
2975 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2976 	gd->minors = SD_MINORS;
2977 
2978 	gd->fops = &sd_fops;
2979 	gd->private_data = &sdkp->driver;
2980 	gd->queue = sdkp->device->request_queue;
2981 
2982 	/* defaults, until the device tells us otherwise */
2983 	sdp->sector_size = 512;
2984 	sdkp->capacity = 0;
2985 	sdkp->media_present = 1;
2986 	sdkp->write_prot = 0;
2987 	sdkp->cache_override = 0;
2988 	sdkp->WCE = 0;
2989 	sdkp->RCD = 0;
2990 	sdkp->ATO = 0;
2991 	sdkp->first_scan = 1;
2992 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2993 
2994 	sd_revalidate_disk(gd);
2995 
2996 	gd->flags = GENHD_FL_EXT_DEVT;
2997 	if (sdp->removable) {
2998 		gd->flags |= GENHD_FL_REMOVABLE;
2999 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3000 	}
3001 
3002 	blk_pm_runtime_init(sdp->request_queue, dev);
3003 	device_add_disk(dev, gd);
3004 	if (sdkp->capacity)
3005 		sd_dif_config_host(sdkp);
3006 
3007 	sd_revalidate_disk(gd);
3008 
3009 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3010 		  sdp->removable ? "removable " : "");
3011 	scsi_autopm_put_device(sdp);
3012 	put_device(&sdkp->dev);
3013 }
3014 
3015 /**
3016  *	sd_probe - called during driver initialization and whenever a
3017  *	new scsi device is attached to the system. It is called once
3018  *	for each scsi device (not just disks) present.
3019  *	@dev: pointer to device object
3020  *
3021  *	Returns 0 if successful (or not interested in this scsi device
3022  *	(e.g. scanner)); 1 when there is an error.
3023  *
3024  *	Note: this function is invoked from the scsi mid-level.
3025  *	This function sets up the mapping between a given
3026  *	<host,channel,id,lun> (found in sdp) and new device name
3027  *	(e.g. /dev/sda). More precisely it is the block device major
3028  *	and minor number that is chosen here.
3029  *
3030  *	Assume sd_probe is not re-entrant (for time being)
3031  *	Also think about sd_probe() and sd_remove() running coincidentally.
3032  **/
3033 static int sd_probe(struct device *dev)
3034 {
3035 	struct scsi_device *sdp = to_scsi_device(dev);
3036 	struct scsi_disk *sdkp;
3037 	struct gendisk *gd;
3038 	int index;
3039 	int error;
3040 
3041 	scsi_autopm_get_device(sdp);
3042 	error = -ENODEV;
3043 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
3044 		goto out;
3045 
3046 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3047 					"sd_probe\n"));
3048 
3049 	error = -ENOMEM;
3050 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3051 	if (!sdkp)
3052 		goto out;
3053 
3054 	gd = alloc_disk(SD_MINORS);
3055 	if (!gd)
3056 		goto out_free;
3057 
3058 	do {
3059 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3060 			goto out_put;
3061 
3062 		spin_lock(&sd_index_lock);
3063 		error = ida_get_new(&sd_index_ida, &index);
3064 		spin_unlock(&sd_index_lock);
3065 	} while (error == -EAGAIN);
3066 
3067 	if (error) {
3068 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3069 		goto out_put;
3070 	}
3071 
3072 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3073 	if (error) {
3074 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3075 		goto out_free_index;
3076 	}
3077 
3078 	sdkp->device = sdp;
3079 	sdkp->driver = &sd_template;
3080 	sdkp->disk = gd;
3081 	sdkp->index = index;
3082 	atomic_set(&sdkp->openers, 0);
3083 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3084 
3085 	if (!sdp->request_queue->rq_timeout) {
3086 		if (sdp->type != TYPE_MOD)
3087 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3088 		else
3089 			blk_queue_rq_timeout(sdp->request_queue,
3090 					     SD_MOD_TIMEOUT);
3091 	}
3092 
3093 	device_initialize(&sdkp->dev);
3094 	sdkp->dev.parent = dev;
3095 	sdkp->dev.class = &sd_disk_class;
3096 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3097 
3098 	error = device_add(&sdkp->dev);
3099 	if (error)
3100 		goto out_free_index;
3101 
3102 	get_device(dev);
3103 	dev_set_drvdata(dev, sdkp);
3104 
3105 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3106 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3107 
3108 	return 0;
3109 
3110  out_free_index:
3111 	spin_lock(&sd_index_lock);
3112 	ida_remove(&sd_index_ida, index);
3113 	spin_unlock(&sd_index_lock);
3114  out_put:
3115 	put_disk(gd);
3116  out_free:
3117 	kfree(sdkp);
3118  out:
3119 	scsi_autopm_put_device(sdp);
3120 	return error;
3121 }
3122 
3123 /**
3124  *	sd_remove - called whenever a scsi disk (previously recognized by
3125  *	sd_probe) is detached from the system. It is called (potentially
3126  *	multiple times) during sd module unload.
3127  *	@sdp: pointer to mid level scsi device object
3128  *
3129  *	Note: this function is invoked from the scsi mid-level.
3130  *	This function potentially frees up a device name (e.g. /dev/sdc)
3131  *	that could be re-used by a subsequent sd_probe().
3132  *	This function is not called when the built-in sd driver is "exit-ed".
3133  **/
3134 static int sd_remove(struct device *dev)
3135 {
3136 	struct scsi_disk *sdkp;
3137 	dev_t devt;
3138 
3139 	sdkp = dev_get_drvdata(dev);
3140 	devt = disk_devt(sdkp->disk);
3141 	scsi_autopm_get_device(sdkp->device);
3142 
3143 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3144 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3145 	device_del(&sdkp->dev);
3146 	del_gendisk(sdkp->disk);
3147 	sd_shutdown(dev);
3148 
3149 	blk_register_region(devt, SD_MINORS, NULL,
3150 			    sd_default_probe, NULL, NULL);
3151 
3152 	mutex_lock(&sd_ref_mutex);
3153 	dev_set_drvdata(dev, NULL);
3154 	put_device(&sdkp->dev);
3155 	mutex_unlock(&sd_ref_mutex);
3156 
3157 	return 0;
3158 }
3159 
3160 /**
3161  *	scsi_disk_release - Called to free the scsi_disk structure
3162  *	@dev: pointer to embedded class device
3163  *
3164  *	sd_ref_mutex must be held entering this routine.  Because it is
3165  *	called on last put, you should always use the scsi_disk_get()
3166  *	scsi_disk_put() helpers which manipulate the semaphore directly
3167  *	and never do a direct put_device.
3168  **/
3169 static void scsi_disk_release(struct device *dev)
3170 {
3171 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3172 	struct gendisk *disk = sdkp->disk;
3173 
3174 	spin_lock(&sd_index_lock);
3175 	ida_remove(&sd_index_ida, sdkp->index);
3176 	spin_unlock(&sd_index_lock);
3177 
3178 	disk->private_data = NULL;
3179 	put_disk(disk);
3180 	put_device(&sdkp->device->sdev_gendev);
3181 
3182 	kfree(sdkp);
3183 }
3184 
3185 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3186 {
3187 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3188 	struct scsi_sense_hdr sshdr;
3189 	struct scsi_device *sdp = sdkp->device;
3190 	int res;
3191 
3192 	if (start)
3193 		cmd[4] |= 1;	/* START */
3194 
3195 	if (sdp->start_stop_pwr_cond)
3196 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3197 
3198 	if (!scsi_device_online(sdp))
3199 		return -ENODEV;
3200 
3201 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3202 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3203 	if (res) {
3204 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3205 		if (driver_byte(res) & DRIVER_SENSE)
3206 			sd_print_sense_hdr(sdkp, &sshdr);
3207 		if (scsi_sense_valid(&sshdr) &&
3208 			/* 0x3a is medium not present */
3209 			sshdr.asc == 0x3a)
3210 			res = 0;
3211 	}
3212 
3213 	/* SCSI error codes must not go to the generic layer */
3214 	if (res)
3215 		return -EIO;
3216 
3217 	return 0;
3218 }
3219 
3220 /*
3221  * Send a SYNCHRONIZE CACHE instruction down to the device through
3222  * the normal SCSI command structure.  Wait for the command to
3223  * complete.
3224  */
3225 static void sd_shutdown(struct device *dev)
3226 {
3227 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3228 
3229 	if (!sdkp)
3230 		return;         /* this can happen */
3231 
3232 	if (pm_runtime_suspended(dev))
3233 		return;
3234 
3235 	if (sdkp->WCE && sdkp->media_present) {
3236 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3237 		sd_sync_cache(sdkp);
3238 	}
3239 
3240 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3241 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3242 		sd_start_stop_device(sdkp, 0);
3243 	}
3244 }
3245 
3246 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3247 {
3248 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3249 	int ret = 0;
3250 
3251 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3252 		return 0;
3253 
3254 	if (sdkp->WCE && sdkp->media_present) {
3255 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3256 		ret = sd_sync_cache(sdkp);
3257 		if (ret) {
3258 			/* ignore OFFLINE device */
3259 			if (ret == -ENODEV)
3260 				ret = 0;
3261 			goto done;
3262 		}
3263 	}
3264 
3265 	if (sdkp->device->manage_start_stop) {
3266 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3267 		/* an error is not worth aborting a system sleep */
3268 		ret = sd_start_stop_device(sdkp, 0);
3269 		if (ignore_stop_errors)
3270 			ret = 0;
3271 	}
3272 
3273 done:
3274 	return ret;
3275 }
3276 
3277 static int sd_suspend_system(struct device *dev)
3278 {
3279 	return sd_suspend_common(dev, true);
3280 }
3281 
3282 static int sd_suspend_runtime(struct device *dev)
3283 {
3284 	return sd_suspend_common(dev, false);
3285 }
3286 
3287 static int sd_resume(struct device *dev)
3288 {
3289 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3290 
3291 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3292 		return 0;
3293 
3294 	if (!sdkp->device->manage_start_stop)
3295 		return 0;
3296 
3297 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3298 	return sd_start_stop_device(sdkp, 1);
3299 }
3300 
3301 /**
3302  *	init_sd - entry point for this driver (both when built in or when
3303  *	a module).
3304  *
3305  *	Note: this function registers this driver with the scsi mid-level.
3306  **/
3307 static int __init init_sd(void)
3308 {
3309 	int majors = 0, i, err;
3310 
3311 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3312 
3313 	for (i = 0; i < SD_MAJORS; i++) {
3314 		if (register_blkdev(sd_major(i), "sd") != 0)
3315 			continue;
3316 		majors++;
3317 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3318 				    sd_default_probe, NULL, NULL);
3319 	}
3320 
3321 	if (!majors)
3322 		return -ENODEV;
3323 
3324 	err = class_register(&sd_disk_class);
3325 	if (err)
3326 		goto err_out;
3327 
3328 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3329 					 0, 0, NULL);
3330 	if (!sd_cdb_cache) {
3331 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3332 		err = -ENOMEM;
3333 		goto err_out_class;
3334 	}
3335 
3336 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3337 	if (!sd_cdb_pool) {
3338 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3339 		err = -ENOMEM;
3340 		goto err_out_cache;
3341 	}
3342 
3343 	err = scsi_register_driver(&sd_template.gendrv);
3344 	if (err)
3345 		goto err_out_driver;
3346 
3347 	return 0;
3348 
3349 err_out_driver:
3350 	mempool_destroy(sd_cdb_pool);
3351 
3352 err_out_cache:
3353 	kmem_cache_destroy(sd_cdb_cache);
3354 
3355 err_out_class:
3356 	class_unregister(&sd_disk_class);
3357 err_out:
3358 	for (i = 0; i < SD_MAJORS; i++)
3359 		unregister_blkdev(sd_major(i), "sd");
3360 	return err;
3361 }
3362 
3363 /**
3364  *	exit_sd - exit point for this driver (when it is a module).
3365  *
3366  *	Note: this function unregisters this driver from the scsi mid-level.
3367  **/
3368 static void __exit exit_sd(void)
3369 {
3370 	int i;
3371 
3372 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3373 
3374 	scsi_unregister_driver(&sd_template.gendrv);
3375 	mempool_destroy(sd_cdb_pool);
3376 	kmem_cache_destroy(sd_cdb_cache);
3377 
3378 	class_unregister(&sd_disk_class);
3379 
3380 	for (i = 0; i < SD_MAJORS; i++) {
3381 		blk_unregister_region(sd_major(i), SD_MINORS);
3382 		unregister_blkdev(sd_major(i), "sd");
3383 	}
3384 }
3385 
3386 module_init(init_sd);
3387 module_exit(exit_sd);
3388 
3389 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3390 			       struct scsi_sense_hdr *sshdr)
3391 {
3392 	scsi_print_sense_hdr(sdkp->device,
3393 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3394 }
3395 
3396 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3397 			    int result)
3398 {
3399 	const char *hb_string = scsi_hostbyte_string(result);
3400 	const char *db_string = scsi_driverbyte_string(result);
3401 
3402 	if (hb_string || db_string)
3403 		sd_printk(KERN_INFO, sdkp,
3404 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3405 			  hb_string ? hb_string : "invalid",
3406 			  db_string ? db_string : "invalid");
3407 	else
3408 		sd_printk(KERN_INFO, sdkp,
3409 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3410 			  msg, host_byte(result), driver_byte(result));
3411 }
3412 
3413