xref: /openbmc/linux/drivers/scsi/sd.c (revision f3a8b664)
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *	Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17  *	   sd_init and cleanups.
18  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *	   not being read in sd_open. Fix problem where removable media
20  *	   could be ejected after sd_open.
21  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24  *	   Support 32k/1M disks.
25  *
26  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *	Note: when the logging level is set by the user, it must be greater
32  *	than the level indicated above to trigger output.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <linux/pr.h>
55 #include <linux/t10-pi.h>
56 #include <asm/uaccess.h>
57 #include <asm/unaligned.h>
58 
59 #include <scsi/scsi.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_dbg.h>
62 #include <scsi/scsi_device.h>
63 #include <scsi/scsi_driver.h>
64 #include <scsi/scsi_eh.h>
65 #include <scsi/scsi_host.h>
66 #include <scsi/scsi_ioctl.h>
67 #include <scsi/scsicam.h>
68 
69 #include "sd.h"
70 #include "scsi_priv.h"
71 #include "scsi_logging.h"
72 
73 MODULE_AUTHOR("Eric Youngdale");
74 MODULE_DESCRIPTION("SCSI disk (sd) driver");
75 MODULE_LICENSE("GPL");
76 
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
96 
97 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
98 #define SD_MINORS	16
99 #else
100 #define SD_MINORS	0
101 #endif
102 
103 static void sd_config_discard(struct scsi_disk *, unsigned int);
104 static void sd_config_write_same(struct scsi_disk *);
105 static int  sd_revalidate_disk(struct gendisk *);
106 static void sd_unlock_native_capacity(struct gendisk *disk);
107 static int  sd_probe(struct device *);
108 static int  sd_remove(struct device *);
109 static void sd_shutdown(struct device *);
110 static int sd_suspend_system(struct device *);
111 static int sd_suspend_runtime(struct device *);
112 static int sd_resume(struct device *);
113 static void sd_rescan(struct device *);
114 static int sd_init_command(struct scsi_cmnd *SCpnt);
115 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
116 static int sd_done(struct scsi_cmnd *);
117 static int sd_eh_action(struct scsi_cmnd *, int);
118 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
119 static void scsi_disk_release(struct device *cdev);
120 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
121 static void sd_print_result(const struct scsi_disk *, const char *, int);
122 
123 static DEFINE_SPINLOCK(sd_index_lock);
124 static DEFINE_IDA(sd_index_ida);
125 
126 /* This semaphore is used to mediate the 0->1 reference get in the
127  * face of object destruction (i.e. we can't allow a get on an
128  * object after last put) */
129 static DEFINE_MUTEX(sd_ref_mutex);
130 
131 static struct kmem_cache *sd_cdb_cache;
132 static mempool_t *sd_cdb_pool;
133 
134 static const char *sd_cache_types[] = {
135 	"write through", "none", "write back",
136 	"write back, no read (daft)"
137 };
138 
139 static void sd_set_flush_flag(struct scsi_disk *sdkp)
140 {
141 	bool wc = false, fua = false;
142 
143 	if (sdkp->WCE) {
144 		wc = true;
145 		if (sdkp->DPOFUA)
146 			fua = true;
147 	}
148 
149 	blk_queue_write_cache(sdkp->disk->queue, wc, fua);
150 }
151 
152 static ssize_t
153 cache_type_store(struct device *dev, struct device_attribute *attr,
154 		 const char *buf, size_t count)
155 {
156 	int i, ct = -1, rcd, wce, sp;
157 	struct scsi_disk *sdkp = to_scsi_disk(dev);
158 	struct scsi_device *sdp = sdkp->device;
159 	char buffer[64];
160 	char *buffer_data;
161 	struct scsi_mode_data data;
162 	struct scsi_sense_hdr sshdr;
163 	static const char temp[] = "temporary ";
164 	int len;
165 
166 	if (sdp->type != TYPE_DISK)
167 		/* no cache control on RBC devices; theoretically they
168 		 * can do it, but there's probably so many exceptions
169 		 * it's not worth the risk */
170 		return -EINVAL;
171 
172 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
173 		buf += sizeof(temp) - 1;
174 		sdkp->cache_override = 1;
175 	} else {
176 		sdkp->cache_override = 0;
177 	}
178 
179 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
180 		len = strlen(sd_cache_types[i]);
181 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
182 		    buf[len] == '\n') {
183 			ct = i;
184 			break;
185 		}
186 	}
187 	if (ct < 0)
188 		return -EINVAL;
189 	rcd = ct & 0x01 ? 1 : 0;
190 	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
191 
192 	if (sdkp->cache_override) {
193 		sdkp->WCE = wce;
194 		sdkp->RCD = rcd;
195 		sd_set_flush_flag(sdkp);
196 		return count;
197 	}
198 
199 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
200 			    SD_MAX_RETRIES, &data, NULL))
201 		return -EINVAL;
202 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
203 		  data.block_descriptor_length);
204 	buffer_data = buffer + data.header_length +
205 		data.block_descriptor_length;
206 	buffer_data[2] &= ~0x05;
207 	buffer_data[2] |= wce << 2 | rcd;
208 	sp = buffer_data[0] & 0x80 ? 1 : 0;
209 	buffer_data[0] &= ~0x80;
210 
211 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
212 			     SD_MAX_RETRIES, &data, &sshdr)) {
213 		if (scsi_sense_valid(&sshdr))
214 			sd_print_sense_hdr(sdkp, &sshdr);
215 		return -EINVAL;
216 	}
217 	revalidate_disk(sdkp->disk);
218 	return count;
219 }
220 
221 static ssize_t
222 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
223 		       char *buf)
224 {
225 	struct scsi_disk *sdkp = to_scsi_disk(dev);
226 	struct scsi_device *sdp = sdkp->device;
227 
228 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
229 }
230 
231 static ssize_t
232 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
233 			const char *buf, size_t count)
234 {
235 	struct scsi_disk *sdkp = to_scsi_disk(dev);
236 	struct scsi_device *sdp = sdkp->device;
237 
238 	if (!capable(CAP_SYS_ADMIN))
239 		return -EACCES;
240 
241 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
242 
243 	return count;
244 }
245 static DEVICE_ATTR_RW(manage_start_stop);
246 
247 static ssize_t
248 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
249 {
250 	struct scsi_disk *sdkp = to_scsi_disk(dev);
251 
252 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
253 }
254 
255 static ssize_t
256 allow_restart_store(struct device *dev, struct device_attribute *attr,
257 		    const char *buf, size_t count)
258 {
259 	struct scsi_disk *sdkp = to_scsi_disk(dev);
260 	struct scsi_device *sdp = sdkp->device;
261 
262 	if (!capable(CAP_SYS_ADMIN))
263 		return -EACCES;
264 
265 	if (sdp->type != TYPE_DISK)
266 		return -EINVAL;
267 
268 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
269 
270 	return count;
271 }
272 static DEVICE_ATTR_RW(allow_restart);
273 
274 static ssize_t
275 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
276 {
277 	struct scsi_disk *sdkp = to_scsi_disk(dev);
278 	int ct = sdkp->RCD + 2*sdkp->WCE;
279 
280 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
281 }
282 static DEVICE_ATTR_RW(cache_type);
283 
284 static ssize_t
285 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
286 {
287 	struct scsi_disk *sdkp = to_scsi_disk(dev);
288 
289 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
290 }
291 static DEVICE_ATTR_RO(FUA);
292 
293 static ssize_t
294 protection_type_show(struct device *dev, struct device_attribute *attr,
295 		     char *buf)
296 {
297 	struct scsi_disk *sdkp = to_scsi_disk(dev);
298 
299 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
300 }
301 
302 static ssize_t
303 protection_type_store(struct device *dev, struct device_attribute *attr,
304 		      const char *buf, size_t count)
305 {
306 	struct scsi_disk *sdkp = to_scsi_disk(dev);
307 	unsigned int val;
308 	int err;
309 
310 	if (!capable(CAP_SYS_ADMIN))
311 		return -EACCES;
312 
313 	err = kstrtouint(buf, 10, &val);
314 
315 	if (err)
316 		return err;
317 
318 	if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
319 		sdkp->protection_type = val;
320 
321 	return count;
322 }
323 static DEVICE_ATTR_RW(protection_type);
324 
325 static ssize_t
326 protection_mode_show(struct device *dev, struct device_attribute *attr,
327 		     char *buf)
328 {
329 	struct scsi_disk *sdkp = to_scsi_disk(dev);
330 	struct scsi_device *sdp = sdkp->device;
331 	unsigned int dif, dix;
332 
333 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
334 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
335 
336 	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
337 		dif = 0;
338 		dix = 1;
339 	}
340 
341 	if (!dif && !dix)
342 		return snprintf(buf, 20, "none\n");
343 
344 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
345 }
346 static DEVICE_ATTR_RO(protection_mode);
347 
348 static ssize_t
349 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
350 {
351 	struct scsi_disk *sdkp = to_scsi_disk(dev);
352 
353 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
354 }
355 static DEVICE_ATTR_RO(app_tag_own);
356 
357 static ssize_t
358 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
359 		       char *buf)
360 {
361 	struct scsi_disk *sdkp = to_scsi_disk(dev);
362 
363 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
364 }
365 static DEVICE_ATTR_RO(thin_provisioning);
366 
367 static const char *lbp_mode[] = {
368 	[SD_LBP_FULL]		= "full",
369 	[SD_LBP_UNMAP]		= "unmap",
370 	[SD_LBP_WS16]		= "writesame_16",
371 	[SD_LBP_WS10]		= "writesame_10",
372 	[SD_LBP_ZERO]		= "writesame_zero",
373 	[SD_LBP_DISABLE]	= "disabled",
374 };
375 
376 static ssize_t
377 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
378 		       char *buf)
379 {
380 	struct scsi_disk *sdkp = to_scsi_disk(dev);
381 
382 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
383 }
384 
385 static ssize_t
386 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
387 			const char *buf, size_t count)
388 {
389 	struct scsi_disk *sdkp = to_scsi_disk(dev);
390 	struct scsi_device *sdp = sdkp->device;
391 
392 	if (!capable(CAP_SYS_ADMIN))
393 		return -EACCES;
394 
395 	if (sdp->type != TYPE_DISK)
396 		return -EINVAL;
397 
398 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
399 		sd_config_discard(sdkp, SD_LBP_UNMAP);
400 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
401 		sd_config_discard(sdkp, SD_LBP_WS16);
402 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
403 		sd_config_discard(sdkp, SD_LBP_WS10);
404 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
405 		sd_config_discard(sdkp, SD_LBP_ZERO);
406 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
407 		sd_config_discard(sdkp, SD_LBP_DISABLE);
408 	else
409 		return -EINVAL;
410 
411 	return count;
412 }
413 static DEVICE_ATTR_RW(provisioning_mode);
414 
415 static ssize_t
416 max_medium_access_timeouts_show(struct device *dev,
417 				struct device_attribute *attr, char *buf)
418 {
419 	struct scsi_disk *sdkp = to_scsi_disk(dev);
420 
421 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
422 }
423 
424 static ssize_t
425 max_medium_access_timeouts_store(struct device *dev,
426 				 struct device_attribute *attr, const char *buf,
427 				 size_t count)
428 {
429 	struct scsi_disk *sdkp = to_scsi_disk(dev);
430 	int err;
431 
432 	if (!capable(CAP_SYS_ADMIN))
433 		return -EACCES;
434 
435 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
436 
437 	return err ? err : count;
438 }
439 static DEVICE_ATTR_RW(max_medium_access_timeouts);
440 
441 static ssize_t
442 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
443 			   char *buf)
444 {
445 	struct scsi_disk *sdkp = to_scsi_disk(dev);
446 
447 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
448 }
449 
450 static ssize_t
451 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
452 			    const char *buf, size_t count)
453 {
454 	struct scsi_disk *sdkp = to_scsi_disk(dev);
455 	struct scsi_device *sdp = sdkp->device;
456 	unsigned long max;
457 	int err;
458 
459 	if (!capable(CAP_SYS_ADMIN))
460 		return -EACCES;
461 
462 	if (sdp->type != TYPE_DISK)
463 		return -EINVAL;
464 
465 	err = kstrtoul(buf, 10, &max);
466 
467 	if (err)
468 		return err;
469 
470 	if (max == 0)
471 		sdp->no_write_same = 1;
472 	else if (max <= SD_MAX_WS16_BLOCKS) {
473 		sdp->no_write_same = 0;
474 		sdkp->max_ws_blocks = max;
475 	}
476 
477 	sd_config_write_same(sdkp);
478 
479 	return count;
480 }
481 static DEVICE_ATTR_RW(max_write_same_blocks);
482 
483 static struct attribute *sd_disk_attrs[] = {
484 	&dev_attr_cache_type.attr,
485 	&dev_attr_FUA.attr,
486 	&dev_attr_allow_restart.attr,
487 	&dev_attr_manage_start_stop.attr,
488 	&dev_attr_protection_type.attr,
489 	&dev_attr_protection_mode.attr,
490 	&dev_attr_app_tag_own.attr,
491 	&dev_attr_thin_provisioning.attr,
492 	&dev_attr_provisioning_mode.attr,
493 	&dev_attr_max_write_same_blocks.attr,
494 	&dev_attr_max_medium_access_timeouts.attr,
495 	NULL,
496 };
497 ATTRIBUTE_GROUPS(sd_disk);
498 
499 static struct class sd_disk_class = {
500 	.name		= "scsi_disk",
501 	.owner		= THIS_MODULE,
502 	.dev_release	= scsi_disk_release,
503 	.dev_groups	= sd_disk_groups,
504 };
505 
506 static const struct dev_pm_ops sd_pm_ops = {
507 	.suspend		= sd_suspend_system,
508 	.resume			= sd_resume,
509 	.poweroff		= sd_suspend_system,
510 	.restore		= sd_resume,
511 	.runtime_suspend	= sd_suspend_runtime,
512 	.runtime_resume		= sd_resume,
513 };
514 
515 static struct scsi_driver sd_template = {
516 	.gendrv = {
517 		.name		= "sd",
518 		.owner		= THIS_MODULE,
519 		.probe		= sd_probe,
520 		.remove		= sd_remove,
521 		.shutdown	= sd_shutdown,
522 		.pm		= &sd_pm_ops,
523 	},
524 	.rescan			= sd_rescan,
525 	.init_command		= sd_init_command,
526 	.uninit_command		= sd_uninit_command,
527 	.done			= sd_done,
528 	.eh_action		= sd_eh_action,
529 };
530 
531 /*
532  * Dummy kobj_map->probe function.
533  * The default ->probe function will call modprobe, which is
534  * pointless as this module is already loaded.
535  */
536 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
537 {
538 	return NULL;
539 }
540 
541 /*
542  * Device no to disk mapping:
543  *
544  *       major         disc2     disc  p1
545  *   |............|.............|....|....| <- dev_t
546  *    31        20 19          8 7  4 3  0
547  *
548  * Inside a major, we have 16k disks, however mapped non-
549  * contiguously. The first 16 disks are for major0, the next
550  * ones with major1, ... Disk 256 is for major0 again, disk 272
551  * for major1, ...
552  * As we stay compatible with our numbering scheme, we can reuse
553  * the well-know SCSI majors 8, 65--71, 136--143.
554  */
555 static int sd_major(int major_idx)
556 {
557 	switch (major_idx) {
558 	case 0:
559 		return SCSI_DISK0_MAJOR;
560 	case 1 ... 7:
561 		return SCSI_DISK1_MAJOR + major_idx - 1;
562 	case 8 ... 15:
563 		return SCSI_DISK8_MAJOR + major_idx - 8;
564 	default:
565 		BUG();
566 		return 0;	/* shut up gcc */
567 	}
568 }
569 
570 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
571 {
572 	struct scsi_disk *sdkp = NULL;
573 
574 	mutex_lock(&sd_ref_mutex);
575 
576 	if (disk->private_data) {
577 		sdkp = scsi_disk(disk);
578 		if (scsi_device_get(sdkp->device) == 0)
579 			get_device(&sdkp->dev);
580 		else
581 			sdkp = NULL;
582 	}
583 	mutex_unlock(&sd_ref_mutex);
584 	return sdkp;
585 }
586 
587 static void scsi_disk_put(struct scsi_disk *sdkp)
588 {
589 	struct scsi_device *sdev = sdkp->device;
590 
591 	mutex_lock(&sd_ref_mutex);
592 	put_device(&sdkp->dev);
593 	scsi_device_put(sdev);
594 	mutex_unlock(&sd_ref_mutex);
595 }
596 
597 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
598 					   unsigned int dix, unsigned int dif)
599 {
600 	struct bio *bio = scmd->request->bio;
601 	unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
602 	unsigned int protect = 0;
603 
604 	if (dix) {				/* DIX Type 0, 1, 2, 3 */
605 		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
606 			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
607 
608 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
609 			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
610 	}
611 
612 	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
613 		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
614 
615 		if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
616 			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
617 	}
618 
619 	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
620 		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
621 
622 		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
623 			protect = 3 << 5;	/* Disable target PI checking */
624 		else
625 			protect = 1 << 5;	/* Enable target PI checking */
626 	}
627 
628 	scsi_set_prot_op(scmd, prot_op);
629 	scsi_set_prot_type(scmd, dif);
630 	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
631 
632 	return protect;
633 }
634 
635 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
636 {
637 	struct request_queue *q = sdkp->disk->queue;
638 	unsigned int logical_block_size = sdkp->device->sector_size;
639 	unsigned int max_blocks = 0;
640 
641 	q->limits.discard_zeroes_data = 0;
642 
643 	/*
644 	 * When LBPRZ is reported, discard alignment and granularity
645 	 * must be fixed to the logical block size. Otherwise the block
646 	 * layer will drop misaligned portions of the request which can
647 	 * lead to data corruption. If LBPRZ is not set, we honor the
648 	 * device preference.
649 	 */
650 	if (sdkp->lbprz) {
651 		q->limits.discard_alignment = 0;
652 		q->limits.discard_granularity = logical_block_size;
653 	} else {
654 		q->limits.discard_alignment = sdkp->unmap_alignment *
655 			logical_block_size;
656 		q->limits.discard_granularity =
657 			max(sdkp->physical_block_size,
658 			    sdkp->unmap_granularity * logical_block_size);
659 	}
660 
661 	sdkp->provisioning_mode = mode;
662 
663 	switch (mode) {
664 
665 	case SD_LBP_DISABLE:
666 		blk_queue_max_discard_sectors(q, 0);
667 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
668 		return;
669 
670 	case SD_LBP_UNMAP:
671 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
672 					  (u32)SD_MAX_WS16_BLOCKS);
673 		break;
674 
675 	case SD_LBP_WS16:
676 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
677 					  (u32)SD_MAX_WS16_BLOCKS);
678 		q->limits.discard_zeroes_data = sdkp->lbprz;
679 		break;
680 
681 	case SD_LBP_WS10:
682 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
683 					  (u32)SD_MAX_WS10_BLOCKS);
684 		q->limits.discard_zeroes_data = sdkp->lbprz;
685 		break;
686 
687 	case SD_LBP_ZERO:
688 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
689 					  (u32)SD_MAX_WS10_BLOCKS);
690 		q->limits.discard_zeroes_data = 1;
691 		break;
692 	}
693 
694 	blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
695 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
696 }
697 
698 /**
699  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
700  * @sdp: scsi device to operate one
701  * @rq: Request to prepare
702  *
703  * Will issue either UNMAP or WRITE SAME(16) depending on preference
704  * indicated by target device.
705  **/
706 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
707 {
708 	struct request *rq = cmd->request;
709 	struct scsi_device *sdp = cmd->device;
710 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
711 	sector_t sector = blk_rq_pos(rq);
712 	unsigned int nr_sectors = blk_rq_sectors(rq);
713 	unsigned int nr_bytes = blk_rq_bytes(rq);
714 	unsigned int len;
715 	int ret;
716 	char *buf;
717 	struct page *page;
718 
719 	sector >>= ilog2(sdp->sector_size) - 9;
720 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
721 
722 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
723 	if (!page)
724 		return BLKPREP_DEFER;
725 
726 	switch (sdkp->provisioning_mode) {
727 	case SD_LBP_UNMAP:
728 		buf = page_address(page);
729 
730 		cmd->cmd_len = 10;
731 		cmd->cmnd[0] = UNMAP;
732 		cmd->cmnd[8] = 24;
733 
734 		put_unaligned_be16(6 + 16, &buf[0]);
735 		put_unaligned_be16(16, &buf[2]);
736 		put_unaligned_be64(sector, &buf[8]);
737 		put_unaligned_be32(nr_sectors, &buf[16]);
738 
739 		len = 24;
740 		break;
741 
742 	case SD_LBP_WS16:
743 		cmd->cmd_len = 16;
744 		cmd->cmnd[0] = WRITE_SAME_16;
745 		cmd->cmnd[1] = 0x8; /* UNMAP */
746 		put_unaligned_be64(sector, &cmd->cmnd[2]);
747 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
748 
749 		len = sdkp->device->sector_size;
750 		break;
751 
752 	case SD_LBP_WS10:
753 	case SD_LBP_ZERO:
754 		cmd->cmd_len = 10;
755 		cmd->cmnd[0] = WRITE_SAME;
756 		if (sdkp->provisioning_mode == SD_LBP_WS10)
757 			cmd->cmnd[1] = 0x8; /* UNMAP */
758 		put_unaligned_be32(sector, &cmd->cmnd[2]);
759 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
760 
761 		len = sdkp->device->sector_size;
762 		break;
763 
764 	default:
765 		ret = BLKPREP_INVALID;
766 		goto out;
767 	}
768 
769 	rq->completion_data = page;
770 	rq->timeout = SD_TIMEOUT;
771 
772 	cmd->transfersize = len;
773 	cmd->allowed = SD_MAX_RETRIES;
774 
775 	/*
776 	 * Initially __data_len is set to the amount of data that needs to be
777 	 * transferred to the target. This amount depends on whether WRITE SAME
778 	 * or UNMAP is being used. After the scatterlist has been mapped by
779 	 * scsi_init_io() we set __data_len to the size of the area to be
780 	 * discarded on disk. This allows us to report completion on the full
781 	 * amount of blocks described by the request.
782 	 */
783 	blk_add_request_payload(rq, page, 0, len);
784 	ret = scsi_init_io(cmd);
785 	rq->__data_len = nr_bytes;
786 
787 out:
788 	if (ret != BLKPREP_OK)
789 		__free_page(page);
790 	return ret;
791 }
792 
793 static void sd_config_write_same(struct scsi_disk *sdkp)
794 {
795 	struct request_queue *q = sdkp->disk->queue;
796 	unsigned int logical_block_size = sdkp->device->sector_size;
797 
798 	if (sdkp->device->no_write_same) {
799 		sdkp->max_ws_blocks = 0;
800 		goto out;
801 	}
802 
803 	/* Some devices can not handle block counts above 0xffff despite
804 	 * supporting WRITE SAME(16). Consequently we default to 64k
805 	 * blocks per I/O unless the device explicitly advertises a
806 	 * bigger limit.
807 	 */
808 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
809 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
810 						   (u32)SD_MAX_WS16_BLOCKS);
811 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
812 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
813 						   (u32)SD_MAX_WS10_BLOCKS);
814 	else {
815 		sdkp->device->no_write_same = 1;
816 		sdkp->max_ws_blocks = 0;
817 	}
818 
819 out:
820 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
821 					 (logical_block_size >> 9));
822 }
823 
824 /**
825  * sd_setup_write_same_cmnd - write the same data to multiple blocks
826  * @cmd: command to prepare
827  *
828  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
829  * preference indicated by target device.
830  **/
831 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
832 {
833 	struct request *rq = cmd->request;
834 	struct scsi_device *sdp = cmd->device;
835 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
836 	struct bio *bio = rq->bio;
837 	sector_t sector = blk_rq_pos(rq);
838 	unsigned int nr_sectors = blk_rq_sectors(rq);
839 	unsigned int nr_bytes = blk_rq_bytes(rq);
840 	int ret;
841 
842 	if (sdkp->device->no_write_same)
843 		return BLKPREP_INVALID;
844 
845 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
846 
847 	sector >>= ilog2(sdp->sector_size) - 9;
848 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
849 
850 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
851 
852 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
853 		cmd->cmd_len = 16;
854 		cmd->cmnd[0] = WRITE_SAME_16;
855 		put_unaligned_be64(sector, &cmd->cmnd[2]);
856 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
857 	} else {
858 		cmd->cmd_len = 10;
859 		cmd->cmnd[0] = WRITE_SAME;
860 		put_unaligned_be32(sector, &cmd->cmnd[2]);
861 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
862 	}
863 
864 	cmd->transfersize = sdp->sector_size;
865 	cmd->allowed = SD_MAX_RETRIES;
866 
867 	/*
868 	 * For WRITE_SAME the data transferred in the DATA IN buffer is
869 	 * different from the amount of data actually written to the target.
870 	 *
871 	 * We set up __data_len to the amount of data transferred from the
872 	 * DATA IN buffer so that blk_rq_map_sg set up the proper S/G list
873 	 * to transfer a single sector of data first, but then reset it to
874 	 * the amount of data to be written right after so that the I/O path
875 	 * knows how much to actually write.
876 	 */
877 	rq->__data_len = sdp->sector_size;
878 	ret = scsi_init_io(cmd);
879 	rq->__data_len = nr_bytes;
880 	return ret;
881 }
882 
883 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
884 {
885 	struct request *rq = cmd->request;
886 
887 	/* flush requests don't perform I/O, zero the S/G table */
888 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
889 
890 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
891 	cmd->cmd_len = 10;
892 	cmd->transfersize = 0;
893 	cmd->allowed = SD_MAX_RETRIES;
894 
895 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
896 	return BLKPREP_OK;
897 }
898 
899 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
900 {
901 	struct request *rq = SCpnt->request;
902 	struct scsi_device *sdp = SCpnt->device;
903 	struct gendisk *disk = rq->rq_disk;
904 	struct scsi_disk *sdkp;
905 	sector_t block = blk_rq_pos(rq);
906 	sector_t threshold;
907 	unsigned int this_count = blk_rq_sectors(rq);
908 	unsigned int dif, dix;
909 	int ret;
910 	unsigned char protect;
911 
912 	ret = scsi_init_io(SCpnt);
913 	if (ret != BLKPREP_OK)
914 		goto out;
915 	SCpnt = rq->special;
916 	sdkp = scsi_disk(disk);
917 
918 	/* from here on until we're complete, any goto out
919 	 * is used for a killable error condition */
920 	ret = BLKPREP_KILL;
921 
922 	SCSI_LOG_HLQUEUE(1,
923 		scmd_printk(KERN_INFO, SCpnt,
924 			"%s: block=%llu, count=%d\n",
925 			__func__, (unsigned long long)block, this_count));
926 
927 	if (!sdp || !scsi_device_online(sdp) ||
928 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
929 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
930 						"Finishing %u sectors\n",
931 						blk_rq_sectors(rq)));
932 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
933 						"Retry with 0x%p\n", SCpnt));
934 		goto out;
935 	}
936 
937 	if (sdp->changed) {
938 		/*
939 		 * quietly refuse to do anything to a changed disc until
940 		 * the changed bit has been reset
941 		 */
942 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
943 		goto out;
944 	}
945 
946 	/*
947 	 * Some SD card readers can't handle multi-sector accesses which touch
948 	 * the last one or two hardware sectors.  Split accesses as needed.
949 	 */
950 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
951 		(sdp->sector_size / 512);
952 
953 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
954 		if (block < threshold) {
955 			/* Access up to the threshold but not beyond */
956 			this_count = threshold - block;
957 		} else {
958 			/* Access only a single hardware sector */
959 			this_count = sdp->sector_size / 512;
960 		}
961 	}
962 
963 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
964 					(unsigned long long)block));
965 
966 	/*
967 	 * If we have a 1K hardware sectorsize, prevent access to single
968 	 * 512 byte sectors.  In theory we could handle this - in fact
969 	 * the scsi cdrom driver must be able to handle this because
970 	 * we typically use 1K blocksizes, and cdroms typically have
971 	 * 2K hardware sectorsizes.  Of course, things are simpler
972 	 * with the cdrom, since it is read-only.  For performance
973 	 * reasons, the filesystems should be able to handle this
974 	 * and not force the scsi disk driver to use bounce buffers
975 	 * for this.
976 	 */
977 	if (sdp->sector_size == 1024) {
978 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
979 			scmd_printk(KERN_ERR, SCpnt,
980 				    "Bad block number requested\n");
981 			goto out;
982 		} else {
983 			block = block >> 1;
984 			this_count = this_count >> 1;
985 		}
986 	}
987 	if (sdp->sector_size == 2048) {
988 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
989 			scmd_printk(KERN_ERR, SCpnt,
990 				    "Bad block number requested\n");
991 			goto out;
992 		} else {
993 			block = block >> 2;
994 			this_count = this_count >> 2;
995 		}
996 	}
997 	if (sdp->sector_size == 4096) {
998 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
999 			scmd_printk(KERN_ERR, SCpnt,
1000 				    "Bad block number requested\n");
1001 			goto out;
1002 		} else {
1003 			block = block >> 3;
1004 			this_count = this_count >> 3;
1005 		}
1006 	}
1007 	if (rq_data_dir(rq) == WRITE) {
1008 		SCpnt->cmnd[0] = WRITE_6;
1009 
1010 		if (blk_integrity_rq(rq))
1011 			sd_dif_prepare(SCpnt);
1012 
1013 	} else if (rq_data_dir(rq) == READ) {
1014 		SCpnt->cmnd[0] = READ_6;
1015 	} else {
1016 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %llu,%llx\n",
1017 			    req_op(rq), (unsigned long long) rq->cmd_flags);
1018 		goto out;
1019 	}
1020 
1021 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1022 					"%s %d/%u 512 byte blocks.\n",
1023 					(rq_data_dir(rq) == WRITE) ?
1024 					"writing" : "reading", this_count,
1025 					blk_rq_sectors(rq)));
1026 
1027 	dix = scsi_prot_sg_count(SCpnt);
1028 	dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1029 
1030 	if (dif || dix)
1031 		protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1032 	else
1033 		protect = 0;
1034 
1035 	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1036 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1037 
1038 		if (unlikely(SCpnt->cmnd == NULL)) {
1039 			ret = BLKPREP_DEFER;
1040 			goto out;
1041 		}
1042 
1043 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1044 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1045 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1046 		SCpnt->cmnd[7] = 0x18;
1047 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1048 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1049 
1050 		/* LBA */
1051 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1052 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1053 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1054 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1055 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1056 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1057 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1058 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1059 
1060 		/* Expected Indirect LBA */
1061 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1062 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1063 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1064 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1065 
1066 		/* Transfer length */
1067 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1068 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1069 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1070 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1071 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1072 		SCpnt->cmnd[0] += READ_16 - READ_6;
1073 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1074 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1075 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1076 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1077 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1078 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1079 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1080 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1081 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1082 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1083 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1084 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1085 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1086 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1087 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1088 		   scsi_device_protection(SCpnt->device) ||
1089 		   SCpnt->device->use_10_for_rw) {
1090 		SCpnt->cmnd[0] += READ_10 - READ_6;
1091 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1092 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1093 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1094 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1095 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1096 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1097 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1098 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1099 	} else {
1100 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1101 			/*
1102 			 * This happens only if this drive failed
1103 			 * 10byte rw command with ILLEGAL_REQUEST
1104 			 * during operation and thus turned off
1105 			 * use_10_for_rw.
1106 			 */
1107 			scmd_printk(KERN_ERR, SCpnt,
1108 				    "FUA write on READ/WRITE(6) drive\n");
1109 			goto out;
1110 		}
1111 
1112 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1113 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1114 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1115 		SCpnt->cmnd[4] = (unsigned char) this_count;
1116 		SCpnt->cmnd[5] = 0;
1117 	}
1118 	SCpnt->sdb.length = this_count * sdp->sector_size;
1119 
1120 	/*
1121 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1122 	 * host adapter, it's safe to assume that we can at least transfer
1123 	 * this many bytes between each connect / disconnect.
1124 	 */
1125 	SCpnt->transfersize = sdp->sector_size;
1126 	SCpnt->underflow = this_count << 9;
1127 	SCpnt->allowed = SD_MAX_RETRIES;
1128 
1129 	/*
1130 	 * This indicates that the command is ready from our end to be
1131 	 * queued.
1132 	 */
1133 	ret = BLKPREP_OK;
1134  out:
1135 	return ret;
1136 }
1137 
1138 static int sd_init_command(struct scsi_cmnd *cmd)
1139 {
1140 	struct request *rq = cmd->request;
1141 
1142 	switch (req_op(rq)) {
1143 	case REQ_OP_DISCARD:
1144 		return sd_setup_discard_cmnd(cmd);
1145 	case REQ_OP_WRITE_SAME:
1146 		return sd_setup_write_same_cmnd(cmd);
1147 	case REQ_OP_FLUSH:
1148 		return sd_setup_flush_cmnd(cmd);
1149 	case REQ_OP_READ:
1150 	case REQ_OP_WRITE:
1151 		return sd_setup_read_write_cmnd(cmd);
1152 	default:
1153 		BUG();
1154 	}
1155 }
1156 
1157 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1158 {
1159 	struct request *rq = SCpnt->request;
1160 
1161 	if (req_op(rq) == REQ_OP_DISCARD)
1162 		__free_page(rq->completion_data);
1163 
1164 	if (SCpnt->cmnd != rq->cmd) {
1165 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1166 		SCpnt->cmnd = NULL;
1167 		SCpnt->cmd_len = 0;
1168 	}
1169 }
1170 
1171 /**
1172  *	sd_open - open a scsi disk device
1173  *	@inode: only i_rdev member may be used
1174  *	@filp: only f_mode and f_flags may be used
1175  *
1176  *	Returns 0 if successful. Returns a negated errno value in case
1177  *	of error.
1178  *
1179  *	Note: This can be called from a user context (e.g. fsck(1) )
1180  *	or from within the kernel (e.g. as a result of a mount(1) ).
1181  *	In the latter case @inode and @filp carry an abridged amount
1182  *	of information as noted above.
1183  *
1184  *	Locking: called with bdev->bd_mutex held.
1185  **/
1186 static int sd_open(struct block_device *bdev, fmode_t mode)
1187 {
1188 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1189 	struct scsi_device *sdev;
1190 	int retval;
1191 
1192 	if (!sdkp)
1193 		return -ENXIO;
1194 
1195 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1196 
1197 	sdev = sdkp->device;
1198 
1199 	/*
1200 	 * If the device is in error recovery, wait until it is done.
1201 	 * If the device is offline, then disallow any access to it.
1202 	 */
1203 	retval = -ENXIO;
1204 	if (!scsi_block_when_processing_errors(sdev))
1205 		goto error_out;
1206 
1207 	if (sdev->removable || sdkp->write_prot)
1208 		check_disk_change(bdev);
1209 
1210 	/*
1211 	 * If the drive is empty, just let the open fail.
1212 	 */
1213 	retval = -ENOMEDIUM;
1214 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1215 		goto error_out;
1216 
1217 	/*
1218 	 * If the device has the write protect tab set, have the open fail
1219 	 * if the user expects to be able to write to the thing.
1220 	 */
1221 	retval = -EROFS;
1222 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1223 		goto error_out;
1224 
1225 	/*
1226 	 * It is possible that the disk changing stuff resulted in
1227 	 * the device being taken offline.  If this is the case,
1228 	 * report this to the user, and don't pretend that the
1229 	 * open actually succeeded.
1230 	 */
1231 	retval = -ENXIO;
1232 	if (!scsi_device_online(sdev))
1233 		goto error_out;
1234 
1235 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1236 		if (scsi_block_when_processing_errors(sdev))
1237 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1238 	}
1239 
1240 	return 0;
1241 
1242 error_out:
1243 	scsi_disk_put(sdkp);
1244 	return retval;
1245 }
1246 
1247 /**
1248  *	sd_release - invoked when the (last) close(2) is called on this
1249  *	scsi disk.
1250  *	@inode: only i_rdev member may be used
1251  *	@filp: only f_mode and f_flags may be used
1252  *
1253  *	Returns 0.
1254  *
1255  *	Note: may block (uninterruptible) if error recovery is underway
1256  *	on this disk.
1257  *
1258  *	Locking: called with bdev->bd_mutex held.
1259  **/
1260 static void sd_release(struct gendisk *disk, fmode_t mode)
1261 {
1262 	struct scsi_disk *sdkp = scsi_disk(disk);
1263 	struct scsi_device *sdev = sdkp->device;
1264 
1265 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1266 
1267 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1268 		if (scsi_block_when_processing_errors(sdev))
1269 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1270 	}
1271 
1272 	/*
1273 	 * XXX and what if there are packets in flight and this close()
1274 	 * XXX is followed by a "rmmod sd_mod"?
1275 	 */
1276 
1277 	scsi_disk_put(sdkp);
1278 }
1279 
1280 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1281 {
1282 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1283 	struct scsi_device *sdp = sdkp->device;
1284 	struct Scsi_Host *host = sdp->host;
1285 	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1286 	int diskinfo[4];
1287 
1288 	/* default to most commonly used values */
1289 	diskinfo[0] = 0x40;	/* 1 << 6 */
1290 	diskinfo[1] = 0x20;	/* 1 << 5 */
1291 	diskinfo[2] = capacity >> 11;
1292 
1293 	/* override with calculated, extended default, or driver values */
1294 	if (host->hostt->bios_param)
1295 		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1296 	else
1297 		scsicam_bios_param(bdev, capacity, diskinfo);
1298 
1299 	geo->heads = diskinfo[0];
1300 	geo->sectors = diskinfo[1];
1301 	geo->cylinders = diskinfo[2];
1302 	return 0;
1303 }
1304 
1305 /**
1306  *	sd_ioctl - process an ioctl
1307  *	@inode: only i_rdev/i_bdev members may be used
1308  *	@filp: only f_mode and f_flags may be used
1309  *	@cmd: ioctl command number
1310  *	@arg: this is third argument given to ioctl(2) system call.
1311  *	Often contains a pointer.
1312  *
1313  *	Returns 0 if successful (some ioctls return positive numbers on
1314  *	success as well). Returns a negated errno value in case of error.
1315  *
1316  *	Note: most ioctls are forward onto the block subsystem or further
1317  *	down in the scsi subsystem.
1318  **/
1319 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1320 		    unsigned int cmd, unsigned long arg)
1321 {
1322 	struct gendisk *disk = bdev->bd_disk;
1323 	struct scsi_disk *sdkp = scsi_disk(disk);
1324 	struct scsi_device *sdp = sdkp->device;
1325 	void __user *p = (void __user *)arg;
1326 	int error;
1327 
1328 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1329 				    "cmd=0x%x\n", disk->disk_name, cmd));
1330 
1331 	error = scsi_verify_blk_ioctl(bdev, cmd);
1332 	if (error < 0)
1333 		return error;
1334 
1335 	/*
1336 	 * If we are in the middle of error recovery, don't let anyone
1337 	 * else try and use this device.  Also, if error recovery fails, it
1338 	 * may try and take the device offline, in which case all further
1339 	 * access to the device is prohibited.
1340 	 */
1341 	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1342 			(mode & FMODE_NDELAY) != 0);
1343 	if (error)
1344 		goto out;
1345 
1346 	/*
1347 	 * Send SCSI addressing ioctls directly to mid level, send other
1348 	 * ioctls to block level and then onto mid level if they can't be
1349 	 * resolved.
1350 	 */
1351 	switch (cmd) {
1352 		case SCSI_IOCTL_GET_IDLUN:
1353 		case SCSI_IOCTL_GET_BUS_NUMBER:
1354 			error = scsi_ioctl(sdp, cmd, p);
1355 			break;
1356 		default:
1357 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1358 			if (error != -ENOTTY)
1359 				break;
1360 			error = scsi_ioctl(sdp, cmd, p);
1361 			break;
1362 	}
1363 out:
1364 	return error;
1365 }
1366 
1367 static void set_media_not_present(struct scsi_disk *sdkp)
1368 {
1369 	if (sdkp->media_present)
1370 		sdkp->device->changed = 1;
1371 
1372 	if (sdkp->device->removable) {
1373 		sdkp->media_present = 0;
1374 		sdkp->capacity = 0;
1375 	}
1376 }
1377 
1378 static int media_not_present(struct scsi_disk *sdkp,
1379 			     struct scsi_sense_hdr *sshdr)
1380 {
1381 	if (!scsi_sense_valid(sshdr))
1382 		return 0;
1383 
1384 	/* not invoked for commands that could return deferred errors */
1385 	switch (sshdr->sense_key) {
1386 	case UNIT_ATTENTION:
1387 	case NOT_READY:
1388 		/* medium not present */
1389 		if (sshdr->asc == 0x3A) {
1390 			set_media_not_present(sdkp);
1391 			return 1;
1392 		}
1393 	}
1394 	return 0;
1395 }
1396 
1397 /**
1398  *	sd_check_events - check media events
1399  *	@disk: kernel device descriptor
1400  *	@clearing: disk events currently being cleared
1401  *
1402  *	Returns mask of DISK_EVENT_*.
1403  *
1404  *	Note: this function is invoked from the block subsystem.
1405  **/
1406 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1407 {
1408 	struct scsi_disk *sdkp = scsi_disk_get(disk);
1409 	struct scsi_device *sdp;
1410 	struct scsi_sense_hdr *sshdr = NULL;
1411 	int retval;
1412 
1413 	if (!sdkp)
1414 		return 0;
1415 
1416 	sdp = sdkp->device;
1417 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1418 
1419 	/*
1420 	 * If the device is offline, don't send any commands - just pretend as
1421 	 * if the command failed.  If the device ever comes back online, we
1422 	 * can deal with it then.  It is only because of unrecoverable errors
1423 	 * that we would ever take a device offline in the first place.
1424 	 */
1425 	if (!scsi_device_online(sdp)) {
1426 		set_media_not_present(sdkp);
1427 		goto out;
1428 	}
1429 
1430 	/*
1431 	 * Using TEST_UNIT_READY enables differentiation between drive with
1432 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1433 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1434 	 *
1435 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1436 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1437 	 * sd_revalidate() is called.
1438 	 */
1439 	retval = -ENODEV;
1440 
1441 	if (scsi_block_when_processing_errors(sdp)) {
1442 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1443 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1444 					      sshdr);
1445 	}
1446 
1447 	/* failed to execute TUR, assume media not present */
1448 	if (host_byte(retval)) {
1449 		set_media_not_present(sdkp);
1450 		goto out;
1451 	}
1452 
1453 	if (media_not_present(sdkp, sshdr))
1454 		goto out;
1455 
1456 	/*
1457 	 * For removable scsi disk we have to recognise the presence
1458 	 * of a disk in the drive.
1459 	 */
1460 	if (!sdkp->media_present)
1461 		sdp->changed = 1;
1462 	sdkp->media_present = 1;
1463 out:
1464 	/*
1465 	 * sdp->changed is set under the following conditions:
1466 	 *
1467 	 *	Medium present state has changed in either direction.
1468 	 *	Device has indicated UNIT_ATTENTION.
1469 	 */
1470 	kfree(sshdr);
1471 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1472 	sdp->changed = 0;
1473 	scsi_disk_put(sdkp);
1474 	return retval;
1475 }
1476 
1477 static int sd_sync_cache(struct scsi_disk *sdkp)
1478 {
1479 	int retries, res;
1480 	struct scsi_device *sdp = sdkp->device;
1481 	const int timeout = sdp->request_queue->rq_timeout
1482 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1483 	struct scsi_sense_hdr sshdr;
1484 
1485 	if (!scsi_device_online(sdp))
1486 		return -ENODEV;
1487 
1488 	for (retries = 3; retries > 0; --retries) {
1489 		unsigned char cmd[10] = { 0 };
1490 
1491 		cmd[0] = SYNCHRONIZE_CACHE;
1492 		/*
1493 		 * Leave the rest of the command zero to indicate
1494 		 * flush everything.
1495 		 */
1496 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1497 					     &sshdr, timeout, SD_MAX_RETRIES,
1498 					     NULL, REQ_PM);
1499 		if (res == 0)
1500 			break;
1501 	}
1502 
1503 	if (res) {
1504 		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1505 
1506 		if (driver_byte(res) & DRIVER_SENSE)
1507 			sd_print_sense_hdr(sdkp, &sshdr);
1508 		/* we need to evaluate the error return  */
1509 		if (scsi_sense_valid(&sshdr) &&
1510 			(sshdr.asc == 0x3a ||	/* medium not present */
1511 			 sshdr.asc == 0x20))	/* invalid command */
1512 				/* this is no error here */
1513 				return 0;
1514 
1515 		switch (host_byte(res)) {
1516 		/* ignore errors due to racing a disconnection */
1517 		case DID_BAD_TARGET:
1518 		case DID_NO_CONNECT:
1519 			return 0;
1520 		/* signal the upper layer it might try again */
1521 		case DID_BUS_BUSY:
1522 		case DID_IMM_RETRY:
1523 		case DID_REQUEUE:
1524 		case DID_SOFT_ERROR:
1525 			return -EBUSY;
1526 		default:
1527 			return -EIO;
1528 		}
1529 	}
1530 	return 0;
1531 }
1532 
1533 static void sd_rescan(struct device *dev)
1534 {
1535 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1536 
1537 	revalidate_disk(sdkp->disk);
1538 }
1539 
1540 
1541 #ifdef CONFIG_COMPAT
1542 /*
1543  * This gets directly called from VFS. When the ioctl
1544  * is not recognized we go back to the other translation paths.
1545  */
1546 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1547 			   unsigned int cmd, unsigned long arg)
1548 {
1549 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1550 	int error;
1551 
1552 	error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1553 			(mode & FMODE_NDELAY) != 0);
1554 	if (error)
1555 		return error;
1556 
1557 	/*
1558 	 * Let the static ioctl translation table take care of it.
1559 	 */
1560 	if (!sdev->host->hostt->compat_ioctl)
1561 		return -ENOIOCTLCMD;
1562 	return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1563 }
1564 #endif
1565 
1566 static char sd_pr_type(enum pr_type type)
1567 {
1568 	switch (type) {
1569 	case PR_WRITE_EXCLUSIVE:
1570 		return 0x01;
1571 	case PR_EXCLUSIVE_ACCESS:
1572 		return 0x03;
1573 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1574 		return 0x05;
1575 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1576 		return 0x06;
1577 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1578 		return 0x07;
1579 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1580 		return 0x08;
1581 	default:
1582 		return 0;
1583 	}
1584 };
1585 
1586 static int sd_pr_command(struct block_device *bdev, u8 sa,
1587 		u64 key, u64 sa_key, u8 type, u8 flags)
1588 {
1589 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1590 	struct scsi_sense_hdr sshdr;
1591 	int result;
1592 	u8 cmd[16] = { 0, };
1593 	u8 data[24] = { 0, };
1594 
1595 	cmd[0] = PERSISTENT_RESERVE_OUT;
1596 	cmd[1] = sa;
1597 	cmd[2] = type;
1598 	put_unaligned_be32(sizeof(data), &cmd[5]);
1599 
1600 	put_unaligned_be64(key, &data[0]);
1601 	put_unaligned_be64(sa_key, &data[8]);
1602 	data[20] = flags;
1603 
1604 	result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1605 			&sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1606 
1607 	if ((driver_byte(result) & DRIVER_SENSE) &&
1608 	    (scsi_sense_valid(&sshdr))) {
1609 		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1610 		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1611 	}
1612 
1613 	return result;
1614 }
1615 
1616 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1617 		u32 flags)
1618 {
1619 	if (flags & ~PR_FL_IGNORE_KEY)
1620 		return -EOPNOTSUPP;
1621 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1622 			old_key, new_key, 0,
1623 			(1 << 0) /* APTPL */);
1624 }
1625 
1626 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1627 		u32 flags)
1628 {
1629 	if (flags)
1630 		return -EOPNOTSUPP;
1631 	return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1632 }
1633 
1634 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1635 {
1636 	return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1637 }
1638 
1639 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1640 		enum pr_type type, bool abort)
1641 {
1642 	return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1643 			     sd_pr_type(type), 0);
1644 }
1645 
1646 static int sd_pr_clear(struct block_device *bdev, u64 key)
1647 {
1648 	return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1649 }
1650 
1651 static const struct pr_ops sd_pr_ops = {
1652 	.pr_register	= sd_pr_register,
1653 	.pr_reserve	= sd_pr_reserve,
1654 	.pr_release	= sd_pr_release,
1655 	.pr_preempt	= sd_pr_preempt,
1656 	.pr_clear	= sd_pr_clear,
1657 };
1658 
1659 static const struct block_device_operations sd_fops = {
1660 	.owner			= THIS_MODULE,
1661 	.open			= sd_open,
1662 	.release		= sd_release,
1663 	.ioctl			= sd_ioctl,
1664 	.getgeo			= sd_getgeo,
1665 #ifdef CONFIG_COMPAT
1666 	.compat_ioctl		= sd_compat_ioctl,
1667 #endif
1668 	.check_events		= sd_check_events,
1669 	.revalidate_disk	= sd_revalidate_disk,
1670 	.unlock_native_capacity	= sd_unlock_native_capacity,
1671 	.pr_ops			= &sd_pr_ops,
1672 };
1673 
1674 /**
1675  *	sd_eh_action - error handling callback
1676  *	@scmd:		sd-issued command that has failed
1677  *	@eh_disp:	The recovery disposition suggested by the midlayer
1678  *
1679  *	This function is called by the SCSI midlayer upon completion of an
1680  *	error test command (currently TEST UNIT READY). The result of sending
1681  *	the eh command is passed in eh_disp.  We're looking for devices that
1682  *	fail medium access commands but are OK with non access commands like
1683  *	test unit ready (so wrongly see the device as having a successful
1684  *	recovery)
1685  **/
1686 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1687 {
1688 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1689 
1690 	if (!scsi_device_online(scmd->device) ||
1691 	    !scsi_medium_access_command(scmd) ||
1692 	    host_byte(scmd->result) != DID_TIME_OUT ||
1693 	    eh_disp != SUCCESS)
1694 		return eh_disp;
1695 
1696 	/*
1697 	 * The device has timed out executing a medium access command.
1698 	 * However, the TEST UNIT READY command sent during error
1699 	 * handling completed successfully. Either the device is in the
1700 	 * process of recovering or has it suffered an internal failure
1701 	 * that prevents access to the storage medium.
1702 	 */
1703 	sdkp->medium_access_timed_out++;
1704 
1705 	/*
1706 	 * If the device keeps failing read/write commands but TEST UNIT
1707 	 * READY always completes successfully we assume that medium
1708 	 * access is no longer possible and take the device offline.
1709 	 */
1710 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1711 		scmd_printk(KERN_ERR, scmd,
1712 			    "Medium access timeout failure. Offlining disk!\n");
1713 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1714 
1715 		return FAILED;
1716 	}
1717 
1718 	return eh_disp;
1719 }
1720 
1721 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1722 {
1723 	u64 start_lba = blk_rq_pos(scmd->request);
1724 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1725 	u64 factor = scmd->device->sector_size / 512;
1726 	u64 bad_lba;
1727 	int info_valid;
1728 	/*
1729 	 * resid is optional but mostly filled in.  When it's unused,
1730 	 * its value is zero, so we assume the whole buffer transferred
1731 	 */
1732 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1733 	unsigned int good_bytes;
1734 
1735 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1736 		return 0;
1737 
1738 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1739 					     SCSI_SENSE_BUFFERSIZE,
1740 					     &bad_lba);
1741 	if (!info_valid)
1742 		return 0;
1743 
1744 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1745 		return 0;
1746 
1747 	/* be careful ... don't want any overflows */
1748 	do_div(start_lba, factor);
1749 	do_div(end_lba, factor);
1750 
1751 	/* The bad lba was reported incorrectly, we have no idea where
1752 	 * the error is.
1753 	 */
1754 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1755 		return 0;
1756 
1757 	/* This computation should always be done in terms of
1758 	 * the resolution of the device's medium.
1759 	 */
1760 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1761 	return min(good_bytes, transferred);
1762 }
1763 
1764 /**
1765  *	sd_done - bottom half handler: called when the lower level
1766  *	driver has completed (successfully or otherwise) a scsi command.
1767  *	@SCpnt: mid-level's per command structure.
1768  *
1769  *	Note: potentially run from within an ISR. Must not block.
1770  **/
1771 static int sd_done(struct scsi_cmnd *SCpnt)
1772 {
1773 	int result = SCpnt->result;
1774 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1775 	struct scsi_sense_hdr sshdr;
1776 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1777 	struct request *req = SCpnt->request;
1778 	int sense_valid = 0;
1779 	int sense_deferred = 0;
1780 	unsigned char op = SCpnt->cmnd[0];
1781 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1782 
1783 	if (req_op(req) == REQ_OP_DISCARD || req_op(req) == REQ_OP_WRITE_SAME) {
1784 		if (!result) {
1785 			good_bytes = blk_rq_bytes(req);
1786 			scsi_set_resid(SCpnt, 0);
1787 		} else {
1788 			good_bytes = 0;
1789 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1790 		}
1791 	}
1792 
1793 	if (result) {
1794 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1795 		if (sense_valid)
1796 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1797 	}
1798 	sdkp->medium_access_timed_out = 0;
1799 
1800 	if (driver_byte(result) != DRIVER_SENSE &&
1801 	    (!sense_valid || sense_deferred))
1802 		goto out;
1803 
1804 	switch (sshdr.sense_key) {
1805 	case HARDWARE_ERROR:
1806 	case MEDIUM_ERROR:
1807 		good_bytes = sd_completed_bytes(SCpnt);
1808 		break;
1809 	case RECOVERED_ERROR:
1810 		good_bytes = scsi_bufflen(SCpnt);
1811 		break;
1812 	case NO_SENSE:
1813 		/* This indicates a false check condition, so ignore it.  An
1814 		 * unknown amount of data was transferred so treat it as an
1815 		 * error.
1816 		 */
1817 		SCpnt->result = 0;
1818 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1819 		break;
1820 	case ABORTED_COMMAND:
1821 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1822 			good_bytes = sd_completed_bytes(SCpnt);
1823 		break;
1824 	case ILLEGAL_REQUEST:
1825 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1826 			good_bytes = sd_completed_bytes(SCpnt);
1827 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1828 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1829 			switch (op) {
1830 			case UNMAP:
1831 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1832 				break;
1833 			case WRITE_SAME_16:
1834 			case WRITE_SAME:
1835 				if (unmap)
1836 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1837 				else {
1838 					sdkp->device->no_write_same = 1;
1839 					sd_config_write_same(sdkp);
1840 
1841 					good_bytes = 0;
1842 					req->__data_len = blk_rq_bytes(req);
1843 					req->cmd_flags |= REQ_QUIET;
1844 				}
1845 			}
1846 		}
1847 		break;
1848 	default:
1849 		break;
1850 	}
1851  out:
1852 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1853 					   "sd_done: completed %d of %d bytes\n",
1854 					   good_bytes, scsi_bufflen(SCpnt)));
1855 
1856 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1857 		sd_dif_complete(SCpnt, good_bytes);
1858 
1859 	return good_bytes;
1860 }
1861 
1862 /*
1863  * spinup disk - called only in sd_revalidate_disk()
1864  */
1865 static void
1866 sd_spinup_disk(struct scsi_disk *sdkp)
1867 {
1868 	unsigned char cmd[10];
1869 	unsigned long spintime_expire = 0;
1870 	int retries, spintime;
1871 	unsigned int the_result;
1872 	struct scsi_sense_hdr sshdr;
1873 	int sense_valid = 0;
1874 
1875 	spintime = 0;
1876 
1877 	/* Spin up drives, as required.  Only do this at boot time */
1878 	/* Spinup needs to be done for module loads too. */
1879 	do {
1880 		retries = 0;
1881 
1882 		do {
1883 			cmd[0] = TEST_UNIT_READY;
1884 			memset((void *) &cmd[1], 0, 9);
1885 
1886 			the_result = scsi_execute_req(sdkp->device, cmd,
1887 						      DMA_NONE, NULL, 0,
1888 						      &sshdr, SD_TIMEOUT,
1889 						      SD_MAX_RETRIES, NULL);
1890 
1891 			/*
1892 			 * If the drive has indicated to us that it
1893 			 * doesn't have any media in it, don't bother
1894 			 * with any more polling.
1895 			 */
1896 			if (media_not_present(sdkp, &sshdr))
1897 				return;
1898 
1899 			if (the_result)
1900 				sense_valid = scsi_sense_valid(&sshdr);
1901 			retries++;
1902 		} while (retries < 3 &&
1903 			 (!scsi_status_is_good(the_result) ||
1904 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1905 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1906 
1907 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1908 			/* no sense, TUR either succeeded or failed
1909 			 * with a status error */
1910 			if(!spintime && !scsi_status_is_good(the_result)) {
1911 				sd_print_result(sdkp, "Test Unit Ready failed",
1912 						the_result);
1913 			}
1914 			break;
1915 		}
1916 
1917 		/*
1918 		 * The device does not want the automatic start to be issued.
1919 		 */
1920 		if (sdkp->device->no_start_on_add)
1921 			break;
1922 
1923 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1924 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1925 				break;	/* manual intervention required */
1926 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1927 				break;	/* standby */
1928 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1929 				break;	/* unavailable */
1930 			/*
1931 			 * Issue command to spin up drive when not ready
1932 			 */
1933 			if (!spintime) {
1934 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1935 				cmd[0] = START_STOP;
1936 				cmd[1] = 1;	/* Return immediately */
1937 				memset((void *) &cmd[2], 0, 8);
1938 				cmd[4] = 1;	/* Start spin cycle */
1939 				if (sdkp->device->start_stop_pwr_cond)
1940 					cmd[4] |= 1 << 4;
1941 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1942 						 NULL, 0, &sshdr,
1943 						 SD_TIMEOUT, SD_MAX_RETRIES,
1944 						 NULL);
1945 				spintime_expire = jiffies + 100 * HZ;
1946 				spintime = 1;
1947 			}
1948 			/* Wait 1 second for next try */
1949 			msleep(1000);
1950 			printk(".");
1951 
1952 		/*
1953 		 * Wait for USB flash devices with slow firmware.
1954 		 * Yes, this sense key/ASC combination shouldn't
1955 		 * occur here.  It's characteristic of these devices.
1956 		 */
1957 		} else if (sense_valid &&
1958 				sshdr.sense_key == UNIT_ATTENTION &&
1959 				sshdr.asc == 0x28) {
1960 			if (!spintime) {
1961 				spintime_expire = jiffies + 5 * HZ;
1962 				spintime = 1;
1963 			}
1964 			/* Wait 1 second for next try */
1965 			msleep(1000);
1966 		} else {
1967 			/* we don't understand the sense code, so it's
1968 			 * probably pointless to loop */
1969 			if(!spintime) {
1970 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1971 				sd_print_sense_hdr(sdkp, &sshdr);
1972 			}
1973 			break;
1974 		}
1975 
1976 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1977 
1978 	if (spintime) {
1979 		if (scsi_status_is_good(the_result))
1980 			printk("ready\n");
1981 		else
1982 			printk("not responding...\n");
1983 	}
1984 }
1985 
1986 
1987 /*
1988  * Determine whether disk supports Data Integrity Field.
1989  */
1990 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1991 {
1992 	struct scsi_device *sdp = sdkp->device;
1993 	u8 type;
1994 	int ret = 0;
1995 
1996 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1997 		return ret;
1998 
1999 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2000 
2001 	if (type > T10_PI_TYPE3_PROTECTION)
2002 		ret = -ENODEV;
2003 	else if (scsi_host_dif_capable(sdp->host, type))
2004 		ret = 1;
2005 
2006 	if (sdkp->first_scan || type != sdkp->protection_type)
2007 		switch (ret) {
2008 		case -ENODEV:
2009 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2010 				  " protection type %u. Disabling disk!\n",
2011 				  type);
2012 			break;
2013 		case 1:
2014 			sd_printk(KERN_NOTICE, sdkp,
2015 				  "Enabling DIF Type %u protection\n", type);
2016 			break;
2017 		case 0:
2018 			sd_printk(KERN_NOTICE, sdkp,
2019 				  "Disabling DIF Type %u protection\n", type);
2020 			break;
2021 		}
2022 
2023 	sdkp->protection_type = type;
2024 
2025 	return ret;
2026 }
2027 
2028 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2029 			struct scsi_sense_hdr *sshdr, int sense_valid,
2030 			int the_result)
2031 {
2032 	if (driver_byte(the_result) & DRIVER_SENSE)
2033 		sd_print_sense_hdr(sdkp, sshdr);
2034 	else
2035 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2036 
2037 	/*
2038 	 * Set dirty bit for removable devices if not ready -
2039 	 * sometimes drives will not report this properly.
2040 	 */
2041 	if (sdp->removable &&
2042 	    sense_valid && sshdr->sense_key == NOT_READY)
2043 		set_media_not_present(sdkp);
2044 
2045 	/*
2046 	 * We used to set media_present to 0 here to indicate no media
2047 	 * in the drive, but some drives fail read capacity even with
2048 	 * media present, so we can't do that.
2049 	 */
2050 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2051 }
2052 
2053 #define RC16_LEN 32
2054 #if RC16_LEN > SD_BUF_SIZE
2055 #error RC16_LEN must not be more than SD_BUF_SIZE
2056 #endif
2057 
2058 #define READ_CAPACITY_RETRIES_ON_RESET	10
2059 
2060 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2061 						unsigned char *buffer)
2062 {
2063 	unsigned char cmd[16];
2064 	struct scsi_sense_hdr sshdr;
2065 	int sense_valid = 0;
2066 	int the_result;
2067 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2068 	unsigned int alignment;
2069 	unsigned long long lba;
2070 	unsigned sector_size;
2071 
2072 	if (sdp->no_read_capacity_16)
2073 		return -EINVAL;
2074 
2075 	do {
2076 		memset(cmd, 0, 16);
2077 		cmd[0] = SERVICE_ACTION_IN_16;
2078 		cmd[1] = SAI_READ_CAPACITY_16;
2079 		cmd[13] = RC16_LEN;
2080 		memset(buffer, 0, RC16_LEN);
2081 
2082 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2083 					buffer, RC16_LEN, &sshdr,
2084 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2085 
2086 		if (media_not_present(sdkp, &sshdr))
2087 			return -ENODEV;
2088 
2089 		if (the_result) {
2090 			sense_valid = scsi_sense_valid(&sshdr);
2091 			if (sense_valid &&
2092 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2093 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2094 			    sshdr.ascq == 0x00)
2095 				/* Invalid Command Operation Code or
2096 				 * Invalid Field in CDB, just retry
2097 				 * silently with RC10 */
2098 				return -EINVAL;
2099 			if (sense_valid &&
2100 			    sshdr.sense_key == UNIT_ATTENTION &&
2101 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2102 				/* Device reset might occur several times,
2103 				 * give it one more chance */
2104 				if (--reset_retries > 0)
2105 					continue;
2106 		}
2107 		retries--;
2108 
2109 	} while (the_result && retries);
2110 
2111 	if (the_result) {
2112 		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2113 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2114 		return -EINVAL;
2115 	}
2116 
2117 	sector_size = get_unaligned_be32(&buffer[8]);
2118 	lba = get_unaligned_be64(&buffer[0]);
2119 
2120 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2121 		sdkp->capacity = 0;
2122 		return -ENODEV;
2123 	}
2124 
2125 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2126 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2127 			"kernel compiled with support for large block "
2128 			"devices.\n");
2129 		sdkp->capacity = 0;
2130 		return -EOVERFLOW;
2131 	}
2132 
2133 	/* Logical blocks per physical block exponent */
2134 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2135 
2136 	/* Lowest aligned logical block */
2137 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2138 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2139 	if (alignment && sdkp->first_scan)
2140 		sd_printk(KERN_NOTICE, sdkp,
2141 			  "physical block alignment offset: %u\n", alignment);
2142 
2143 	if (buffer[14] & 0x80) { /* LBPME */
2144 		sdkp->lbpme = 1;
2145 
2146 		if (buffer[14] & 0x40) /* LBPRZ */
2147 			sdkp->lbprz = 1;
2148 
2149 		sd_config_discard(sdkp, SD_LBP_WS16);
2150 	}
2151 
2152 	sdkp->capacity = lba + 1;
2153 	return sector_size;
2154 }
2155 
2156 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2157 						unsigned char *buffer)
2158 {
2159 	unsigned char cmd[16];
2160 	struct scsi_sense_hdr sshdr;
2161 	int sense_valid = 0;
2162 	int the_result;
2163 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2164 	sector_t lba;
2165 	unsigned sector_size;
2166 
2167 	do {
2168 		cmd[0] = READ_CAPACITY;
2169 		memset(&cmd[1], 0, 9);
2170 		memset(buffer, 0, 8);
2171 
2172 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2173 					buffer, 8, &sshdr,
2174 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2175 
2176 		if (media_not_present(sdkp, &sshdr))
2177 			return -ENODEV;
2178 
2179 		if (the_result) {
2180 			sense_valid = scsi_sense_valid(&sshdr);
2181 			if (sense_valid &&
2182 			    sshdr.sense_key == UNIT_ATTENTION &&
2183 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2184 				/* Device reset might occur several times,
2185 				 * give it one more chance */
2186 				if (--reset_retries > 0)
2187 					continue;
2188 		}
2189 		retries--;
2190 
2191 	} while (the_result && retries);
2192 
2193 	if (the_result) {
2194 		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2195 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2196 		return -EINVAL;
2197 	}
2198 
2199 	sector_size = get_unaligned_be32(&buffer[4]);
2200 	lba = get_unaligned_be32(&buffer[0]);
2201 
2202 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2203 		/* Some buggy (usb cardreader) devices return an lba of
2204 		   0xffffffff when the want to report a size of 0 (with
2205 		   which they really mean no media is present) */
2206 		sdkp->capacity = 0;
2207 		sdkp->physical_block_size = sector_size;
2208 		return sector_size;
2209 	}
2210 
2211 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2212 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2213 			"kernel compiled with support for large block "
2214 			"devices.\n");
2215 		sdkp->capacity = 0;
2216 		return -EOVERFLOW;
2217 	}
2218 
2219 	sdkp->capacity = lba + 1;
2220 	sdkp->physical_block_size = sector_size;
2221 	return sector_size;
2222 }
2223 
2224 static int sd_try_rc16_first(struct scsi_device *sdp)
2225 {
2226 	if (sdp->host->max_cmd_len < 16)
2227 		return 0;
2228 	if (sdp->try_rc_10_first)
2229 		return 0;
2230 	if (sdp->scsi_level > SCSI_SPC_2)
2231 		return 1;
2232 	if (scsi_device_protection(sdp))
2233 		return 1;
2234 	return 0;
2235 }
2236 
2237 /*
2238  * read disk capacity
2239  */
2240 static void
2241 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2242 {
2243 	int sector_size;
2244 	struct scsi_device *sdp = sdkp->device;
2245 	sector_t old_capacity = sdkp->capacity;
2246 
2247 	if (sd_try_rc16_first(sdp)) {
2248 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2249 		if (sector_size == -EOVERFLOW)
2250 			goto got_data;
2251 		if (sector_size == -ENODEV)
2252 			return;
2253 		if (sector_size < 0)
2254 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2255 		if (sector_size < 0)
2256 			return;
2257 	} else {
2258 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2259 		if (sector_size == -EOVERFLOW)
2260 			goto got_data;
2261 		if (sector_size < 0)
2262 			return;
2263 		if ((sizeof(sdkp->capacity) > 4) &&
2264 		    (sdkp->capacity > 0xffffffffULL)) {
2265 			int old_sector_size = sector_size;
2266 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2267 					"Trying to use READ CAPACITY(16).\n");
2268 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2269 			if (sector_size < 0) {
2270 				sd_printk(KERN_NOTICE, sdkp,
2271 					"Using 0xffffffff as device size\n");
2272 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2273 				sector_size = old_sector_size;
2274 				goto got_data;
2275 			}
2276 		}
2277 	}
2278 
2279 	/* Some devices are known to return the total number of blocks,
2280 	 * not the highest block number.  Some devices have versions
2281 	 * which do this and others which do not.  Some devices we might
2282 	 * suspect of doing this but we don't know for certain.
2283 	 *
2284 	 * If we know the reported capacity is wrong, decrement it.  If
2285 	 * we can only guess, then assume the number of blocks is even
2286 	 * (usually true but not always) and err on the side of lowering
2287 	 * the capacity.
2288 	 */
2289 	if (sdp->fix_capacity ||
2290 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2291 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2292 				"from its reported value: %llu\n",
2293 				(unsigned long long) sdkp->capacity);
2294 		--sdkp->capacity;
2295 	}
2296 
2297 got_data:
2298 	if (sector_size == 0) {
2299 		sector_size = 512;
2300 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2301 			  "assuming 512.\n");
2302 	}
2303 
2304 	if (sector_size != 512 &&
2305 	    sector_size != 1024 &&
2306 	    sector_size != 2048 &&
2307 	    sector_size != 4096) {
2308 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2309 			  sector_size);
2310 		/*
2311 		 * The user might want to re-format the drive with
2312 		 * a supported sectorsize.  Once this happens, it
2313 		 * would be relatively trivial to set the thing up.
2314 		 * For this reason, we leave the thing in the table.
2315 		 */
2316 		sdkp->capacity = 0;
2317 		/*
2318 		 * set a bogus sector size so the normal read/write
2319 		 * logic in the block layer will eventually refuse any
2320 		 * request on this device without tripping over power
2321 		 * of two sector size assumptions
2322 		 */
2323 		sector_size = 512;
2324 	}
2325 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2326 
2327 	{
2328 		char cap_str_2[10], cap_str_10[10];
2329 
2330 		string_get_size(sdkp->capacity, sector_size,
2331 				STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2332 		string_get_size(sdkp->capacity, sector_size,
2333 				STRING_UNITS_10, cap_str_10,
2334 				sizeof(cap_str_10));
2335 
2336 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2337 			sd_printk(KERN_NOTICE, sdkp,
2338 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2339 				  (unsigned long long)sdkp->capacity,
2340 				  sector_size, cap_str_10, cap_str_2);
2341 
2342 			if (sdkp->physical_block_size != sector_size)
2343 				sd_printk(KERN_NOTICE, sdkp,
2344 					  "%u-byte physical blocks\n",
2345 					  sdkp->physical_block_size);
2346 		}
2347 	}
2348 
2349 	if (sdkp->capacity > 0xffffffff)
2350 		sdp->use_16_for_rw = 1;
2351 
2352 	blk_queue_physical_block_size(sdp->request_queue,
2353 				      sdkp->physical_block_size);
2354 	sdkp->device->sector_size = sector_size;
2355 }
2356 
2357 /* called with buffer of length 512 */
2358 static inline int
2359 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2360 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2361 		 struct scsi_sense_hdr *sshdr)
2362 {
2363 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2364 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2365 			       sshdr);
2366 }
2367 
2368 /*
2369  * read write protect setting, if possible - called only in sd_revalidate_disk()
2370  * called with buffer of length SD_BUF_SIZE
2371  */
2372 static void
2373 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2374 {
2375 	int res;
2376 	struct scsi_device *sdp = sdkp->device;
2377 	struct scsi_mode_data data;
2378 	int old_wp = sdkp->write_prot;
2379 
2380 	set_disk_ro(sdkp->disk, 0);
2381 	if (sdp->skip_ms_page_3f) {
2382 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2383 		return;
2384 	}
2385 
2386 	if (sdp->use_192_bytes_for_3f) {
2387 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2388 	} else {
2389 		/*
2390 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2391 		 * We have to start carefully: some devices hang if we ask
2392 		 * for more than is available.
2393 		 */
2394 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2395 
2396 		/*
2397 		 * Second attempt: ask for page 0 When only page 0 is
2398 		 * implemented, a request for page 3F may return Sense Key
2399 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2400 		 * CDB.
2401 		 */
2402 		if (!scsi_status_is_good(res))
2403 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2404 
2405 		/*
2406 		 * Third attempt: ask 255 bytes, as we did earlier.
2407 		 */
2408 		if (!scsi_status_is_good(res))
2409 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2410 					       &data, NULL);
2411 	}
2412 
2413 	if (!scsi_status_is_good(res)) {
2414 		sd_first_printk(KERN_WARNING, sdkp,
2415 			  "Test WP failed, assume Write Enabled\n");
2416 	} else {
2417 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2418 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2419 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2420 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2421 				  sdkp->write_prot ? "on" : "off");
2422 			sd_printk(KERN_DEBUG, sdkp,
2423 				  "Mode Sense: %02x %02x %02x %02x\n",
2424 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2425 		}
2426 	}
2427 }
2428 
2429 /*
2430  * sd_read_cache_type - called only from sd_revalidate_disk()
2431  * called with buffer of length SD_BUF_SIZE
2432  */
2433 static void
2434 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2435 {
2436 	int len = 0, res;
2437 	struct scsi_device *sdp = sdkp->device;
2438 
2439 	int dbd;
2440 	int modepage;
2441 	int first_len;
2442 	struct scsi_mode_data data;
2443 	struct scsi_sense_hdr sshdr;
2444 	int old_wce = sdkp->WCE;
2445 	int old_rcd = sdkp->RCD;
2446 	int old_dpofua = sdkp->DPOFUA;
2447 
2448 
2449 	if (sdkp->cache_override)
2450 		return;
2451 
2452 	first_len = 4;
2453 	if (sdp->skip_ms_page_8) {
2454 		if (sdp->type == TYPE_RBC)
2455 			goto defaults;
2456 		else {
2457 			if (sdp->skip_ms_page_3f)
2458 				goto defaults;
2459 			modepage = 0x3F;
2460 			if (sdp->use_192_bytes_for_3f)
2461 				first_len = 192;
2462 			dbd = 0;
2463 		}
2464 	} else if (sdp->type == TYPE_RBC) {
2465 		modepage = 6;
2466 		dbd = 8;
2467 	} else {
2468 		modepage = 8;
2469 		dbd = 0;
2470 	}
2471 
2472 	/* cautiously ask */
2473 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2474 			&data, &sshdr);
2475 
2476 	if (!scsi_status_is_good(res))
2477 		goto bad_sense;
2478 
2479 	if (!data.header_length) {
2480 		modepage = 6;
2481 		first_len = 0;
2482 		sd_first_printk(KERN_ERR, sdkp,
2483 				"Missing header in MODE_SENSE response\n");
2484 	}
2485 
2486 	/* that went OK, now ask for the proper length */
2487 	len = data.length;
2488 
2489 	/*
2490 	 * We're only interested in the first three bytes, actually.
2491 	 * But the data cache page is defined for the first 20.
2492 	 */
2493 	if (len < 3)
2494 		goto bad_sense;
2495 	else if (len > SD_BUF_SIZE) {
2496 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2497 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2498 		len = SD_BUF_SIZE;
2499 	}
2500 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2501 		len = 192;
2502 
2503 	/* Get the data */
2504 	if (len > first_len)
2505 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2506 				&data, &sshdr);
2507 
2508 	if (scsi_status_is_good(res)) {
2509 		int offset = data.header_length + data.block_descriptor_length;
2510 
2511 		while (offset < len) {
2512 			u8 page_code = buffer[offset] & 0x3F;
2513 			u8 spf       = buffer[offset] & 0x40;
2514 
2515 			if (page_code == 8 || page_code == 6) {
2516 				/* We're interested only in the first 3 bytes.
2517 				 */
2518 				if (len - offset <= 2) {
2519 					sd_first_printk(KERN_ERR, sdkp,
2520 						"Incomplete mode parameter "
2521 							"data\n");
2522 					goto defaults;
2523 				} else {
2524 					modepage = page_code;
2525 					goto Page_found;
2526 				}
2527 			} else {
2528 				/* Go to the next page */
2529 				if (spf && len - offset > 3)
2530 					offset += 4 + (buffer[offset+2] << 8) +
2531 						buffer[offset+3];
2532 				else if (!spf && len - offset > 1)
2533 					offset += 2 + buffer[offset+1];
2534 				else {
2535 					sd_first_printk(KERN_ERR, sdkp,
2536 							"Incomplete mode "
2537 							"parameter data\n");
2538 					goto defaults;
2539 				}
2540 			}
2541 		}
2542 
2543 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2544 		goto defaults;
2545 
2546 	Page_found:
2547 		if (modepage == 8) {
2548 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2549 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2550 		} else {
2551 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2552 			sdkp->RCD = 0;
2553 		}
2554 
2555 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2556 		if (sdp->broken_fua) {
2557 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2558 			sdkp->DPOFUA = 0;
2559 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2560 			sd_first_printk(KERN_NOTICE, sdkp,
2561 				  "Uses READ/WRITE(6), disabling FUA\n");
2562 			sdkp->DPOFUA = 0;
2563 		}
2564 
2565 		/* No cache flush allowed for write protected devices */
2566 		if (sdkp->WCE && sdkp->write_prot)
2567 			sdkp->WCE = 0;
2568 
2569 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2570 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2571 			sd_printk(KERN_NOTICE, sdkp,
2572 				  "Write cache: %s, read cache: %s, %s\n",
2573 				  sdkp->WCE ? "enabled" : "disabled",
2574 				  sdkp->RCD ? "disabled" : "enabled",
2575 				  sdkp->DPOFUA ? "supports DPO and FUA"
2576 				  : "doesn't support DPO or FUA");
2577 
2578 		return;
2579 	}
2580 
2581 bad_sense:
2582 	if (scsi_sense_valid(&sshdr) &&
2583 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2584 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2585 		/* Invalid field in CDB */
2586 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2587 	else
2588 		sd_first_printk(KERN_ERR, sdkp,
2589 				"Asking for cache data failed\n");
2590 
2591 defaults:
2592 	if (sdp->wce_default_on) {
2593 		sd_first_printk(KERN_NOTICE, sdkp,
2594 				"Assuming drive cache: write back\n");
2595 		sdkp->WCE = 1;
2596 	} else {
2597 		sd_first_printk(KERN_ERR, sdkp,
2598 				"Assuming drive cache: write through\n");
2599 		sdkp->WCE = 0;
2600 	}
2601 	sdkp->RCD = 0;
2602 	sdkp->DPOFUA = 0;
2603 }
2604 
2605 /*
2606  * The ATO bit indicates whether the DIF application tag is available
2607  * for use by the operating system.
2608  */
2609 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2610 {
2611 	int res, offset;
2612 	struct scsi_device *sdp = sdkp->device;
2613 	struct scsi_mode_data data;
2614 	struct scsi_sense_hdr sshdr;
2615 
2616 	if (sdp->type != TYPE_DISK)
2617 		return;
2618 
2619 	if (sdkp->protection_type == 0)
2620 		return;
2621 
2622 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2623 			      SD_MAX_RETRIES, &data, &sshdr);
2624 
2625 	if (!scsi_status_is_good(res) || !data.header_length ||
2626 	    data.length < 6) {
2627 		sd_first_printk(KERN_WARNING, sdkp,
2628 			  "getting Control mode page failed, assume no ATO\n");
2629 
2630 		if (scsi_sense_valid(&sshdr))
2631 			sd_print_sense_hdr(sdkp, &sshdr);
2632 
2633 		return;
2634 	}
2635 
2636 	offset = data.header_length + data.block_descriptor_length;
2637 
2638 	if ((buffer[offset] & 0x3f) != 0x0a) {
2639 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2640 		return;
2641 	}
2642 
2643 	if ((buffer[offset + 5] & 0x80) == 0)
2644 		return;
2645 
2646 	sdkp->ATO = 1;
2647 
2648 	return;
2649 }
2650 
2651 /**
2652  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2653  * @disk: disk to query
2654  */
2655 static void sd_read_block_limits(struct scsi_disk *sdkp)
2656 {
2657 	unsigned int sector_sz = sdkp->device->sector_size;
2658 	const int vpd_len = 64;
2659 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2660 
2661 	if (!buffer ||
2662 	    /* Block Limits VPD */
2663 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2664 		goto out;
2665 
2666 	blk_queue_io_min(sdkp->disk->queue,
2667 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2668 
2669 	sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2670 	sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2671 
2672 	if (buffer[3] == 0x3c) {
2673 		unsigned int lba_count, desc_count;
2674 
2675 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2676 
2677 		if (!sdkp->lbpme)
2678 			goto out;
2679 
2680 		lba_count = get_unaligned_be32(&buffer[20]);
2681 		desc_count = get_unaligned_be32(&buffer[24]);
2682 
2683 		if (lba_count && desc_count)
2684 			sdkp->max_unmap_blocks = lba_count;
2685 
2686 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2687 
2688 		if (buffer[32] & 0x80)
2689 			sdkp->unmap_alignment =
2690 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2691 
2692 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2693 
2694 			if (sdkp->max_unmap_blocks)
2695 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2696 			else
2697 				sd_config_discard(sdkp, SD_LBP_WS16);
2698 
2699 		} else {	/* LBP VPD page tells us what to use */
2700 			if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2701 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2702 			else if (sdkp->lbpws)
2703 				sd_config_discard(sdkp, SD_LBP_WS16);
2704 			else if (sdkp->lbpws10)
2705 				sd_config_discard(sdkp, SD_LBP_WS10);
2706 			else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2707 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2708 			else
2709 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2710 		}
2711 	}
2712 
2713  out:
2714 	kfree(buffer);
2715 }
2716 
2717 /**
2718  * sd_read_block_characteristics - Query block dev. characteristics
2719  * @disk: disk to query
2720  */
2721 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2722 {
2723 	unsigned char *buffer;
2724 	u16 rot;
2725 	const int vpd_len = 64;
2726 
2727 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2728 
2729 	if (!buffer ||
2730 	    /* Block Device Characteristics VPD */
2731 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2732 		goto out;
2733 
2734 	rot = get_unaligned_be16(&buffer[4]);
2735 
2736 	if (rot == 1) {
2737 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2738 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, sdkp->disk->queue);
2739 	}
2740 
2741  out:
2742 	kfree(buffer);
2743 }
2744 
2745 /**
2746  * sd_read_block_provisioning - Query provisioning VPD page
2747  * @disk: disk to query
2748  */
2749 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2750 {
2751 	unsigned char *buffer;
2752 	const int vpd_len = 8;
2753 
2754 	if (sdkp->lbpme == 0)
2755 		return;
2756 
2757 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2758 
2759 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2760 		goto out;
2761 
2762 	sdkp->lbpvpd	= 1;
2763 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2764 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2765 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2766 
2767  out:
2768 	kfree(buffer);
2769 }
2770 
2771 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2772 {
2773 	struct scsi_device *sdev = sdkp->device;
2774 
2775 	if (sdev->host->no_write_same) {
2776 		sdev->no_write_same = 1;
2777 
2778 		return;
2779 	}
2780 
2781 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2782 		/* too large values might cause issues with arcmsr */
2783 		int vpd_buf_len = 64;
2784 
2785 		sdev->no_report_opcodes = 1;
2786 
2787 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2788 		 * CODES is unsupported and the device has an ATA
2789 		 * Information VPD page (SAT).
2790 		 */
2791 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2792 			sdev->no_write_same = 1;
2793 	}
2794 
2795 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2796 		sdkp->ws16 = 1;
2797 
2798 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2799 		sdkp->ws10 = 1;
2800 }
2801 
2802 /**
2803  *	sd_revalidate_disk - called the first time a new disk is seen,
2804  *	performs disk spin up, read_capacity, etc.
2805  *	@disk: struct gendisk we care about
2806  **/
2807 static int sd_revalidate_disk(struct gendisk *disk)
2808 {
2809 	struct scsi_disk *sdkp = scsi_disk(disk);
2810 	struct scsi_device *sdp = sdkp->device;
2811 	struct request_queue *q = sdkp->disk->queue;
2812 	unsigned char *buffer;
2813 	unsigned int dev_max, rw_max;
2814 
2815 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2816 				      "sd_revalidate_disk\n"));
2817 
2818 	/*
2819 	 * If the device is offline, don't try and read capacity or any
2820 	 * of the other niceties.
2821 	 */
2822 	if (!scsi_device_online(sdp))
2823 		goto out;
2824 
2825 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2826 	if (!buffer) {
2827 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2828 			  "allocation failure.\n");
2829 		goto out;
2830 	}
2831 
2832 	sd_spinup_disk(sdkp);
2833 
2834 	/*
2835 	 * Without media there is no reason to ask; moreover, some devices
2836 	 * react badly if we do.
2837 	 */
2838 	if (sdkp->media_present) {
2839 		sd_read_capacity(sdkp, buffer);
2840 
2841 		if (scsi_device_supports_vpd(sdp)) {
2842 			sd_read_block_provisioning(sdkp);
2843 			sd_read_block_limits(sdkp);
2844 			sd_read_block_characteristics(sdkp);
2845 		}
2846 
2847 		sd_read_write_protect_flag(sdkp, buffer);
2848 		sd_read_cache_type(sdkp, buffer);
2849 		sd_read_app_tag_own(sdkp, buffer);
2850 		sd_read_write_same(sdkp, buffer);
2851 	}
2852 
2853 	sdkp->first_scan = 0;
2854 
2855 	/*
2856 	 * We now have all cache related info, determine how we deal
2857 	 * with flush requests.
2858 	 */
2859 	sd_set_flush_flag(sdkp);
2860 
2861 	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2862 	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2863 
2864 	/* Some devices report a maximum block count for READ/WRITE requests. */
2865 	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2866 	q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2867 
2868 	/*
2869 	 * Use the device's preferred I/O size for reads and writes
2870 	 * unless the reported value is unreasonably small, large, or
2871 	 * garbage.
2872 	 */
2873 	if (sdkp->opt_xfer_blocks &&
2874 	    sdkp->opt_xfer_blocks <= dev_max &&
2875 	    sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2876 	    logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2877 		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2878 		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2879 	} else
2880 		rw_max = BLK_DEF_MAX_SECTORS;
2881 
2882 	/* Combine with controller limits */
2883 	q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2884 
2885 	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2886 	sd_config_write_same(sdkp);
2887 	kfree(buffer);
2888 
2889  out:
2890 	return 0;
2891 }
2892 
2893 /**
2894  *	sd_unlock_native_capacity - unlock native capacity
2895  *	@disk: struct gendisk to set capacity for
2896  *
2897  *	Block layer calls this function if it detects that partitions
2898  *	on @disk reach beyond the end of the device.  If the SCSI host
2899  *	implements ->unlock_native_capacity() method, it's invoked to
2900  *	give it a chance to adjust the device capacity.
2901  *
2902  *	CONTEXT:
2903  *	Defined by block layer.  Might sleep.
2904  */
2905 static void sd_unlock_native_capacity(struct gendisk *disk)
2906 {
2907 	struct scsi_device *sdev = scsi_disk(disk)->device;
2908 
2909 	if (sdev->host->hostt->unlock_native_capacity)
2910 		sdev->host->hostt->unlock_native_capacity(sdev);
2911 }
2912 
2913 /**
2914  *	sd_format_disk_name - format disk name
2915  *	@prefix: name prefix - ie. "sd" for SCSI disks
2916  *	@index: index of the disk to format name for
2917  *	@buf: output buffer
2918  *	@buflen: length of the output buffer
2919  *
2920  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2921  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2922  *	which is followed by sdaaa.
2923  *
2924  *	This is basically 26 base counting with one extra 'nil' entry
2925  *	at the beginning from the second digit on and can be
2926  *	determined using similar method as 26 base conversion with the
2927  *	index shifted -1 after each digit is computed.
2928  *
2929  *	CONTEXT:
2930  *	Don't care.
2931  *
2932  *	RETURNS:
2933  *	0 on success, -errno on failure.
2934  */
2935 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2936 {
2937 	const int base = 'z' - 'a' + 1;
2938 	char *begin = buf + strlen(prefix);
2939 	char *end = buf + buflen;
2940 	char *p;
2941 	int unit;
2942 
2943 	p = end - 1;
2944 	*p = '\0';
2945 	unit = base;
2946 	do {
2947 		if (p == begin)
2948 			return -EINVAL;
2949 		*--p = 'a' + (index % unit);
2950 		index = (index / unit) - 1;
2951 	} while (index >= 0);
2952 
2953 	memmove(begin, p, end - p);
2954 	memcpy(buf, prefix, strlen(prefix));
2955 
2956 	return 0;
2957 }
2958 
2959 /*
2960  * The asynchronous part of sd_probe
2961  */
2962 static void sd_probe_async(void *data, async_cookie_t cookie)
2963 {
2964 	struct scsi_disk *sdkp = data;
2965 	struct scsi_device *sdp;
2966 	struct gendisk *gd;
2967 	u32 index;
2968 	struct device *dev;
2969 
2970 	sdp = sdkp->device;
2971 	gd = sdkp->disk;
2972 	index = sdkp->index;
2973 	dev = &sdp->sdev_gendev;
2974 
2975 	gd->major = sd_major((index & 0xf0) >> 4);
2976 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2977 	gd->minors = SD_MINORS;
2978 
2979 	gd->fops = &sd_fops;
2980 	gd->private_data = &sdkp->driver;
2981 	gd->queue = sdkp->device->request_queue;
2982 
2983 	/* defaults, until the device tells us otherwise */
2984 	sdp->sector_size = 512;
2985 	sdkp->capacity = 0;
2986 	sdkp->media_present = 1;
2987 	sdkp->write_prot = 0;
2988 	sdkp->cache_override = 0;
2989 	sdkp->WCE = 0;
2990 	sdkp->RCD = 0;
2991 	sdkp->ATO = 0;
2992 	sdkp->first_scan = 1;
2993 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2994 
2995 	sd_revalidate_disk(gd);
2996 
2997 	gd->flags = GENHD_FL_EXT_DEVT;
2998 	if (sdp->removable) {
2999 		gd->flags |= GENHD_FL_REMOVABLE;
3000 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
3001 	}
3002 
3003 	blk_pm_runtime_init(sdp->request_queue, dev);
3004 	device_add_disk(dev, gd);
3005 	if (sdkp->capacity)
3006 		sd_dif_config_host(sdkp);
3007 
3008 	sd_revalidate_disk(gd);
3009 
3010 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3011 		  sdp->removable ? "removable " : "");
3012 	scsi_autopm_put_device(sdp);
3013 	put_device(&sdkp->dev);
3014 }
3015 
3016 /**
3017  *	sd_probe - called during driver initialization and whenever a
3018  *	new scsi device is attached to the system. It is called once
3019  *	for each scsi device (not just disks) present.
3020  *	@dev: pointer to device object
3021  *
3022  *	Returns 0 if successful (or not interested in this scsi device
3023  *	(e.g. scanner)); 1 when there is an error.
3024  *
3025  *	Note: this function is invoked from the scsi mid-level.
3026  *	This function sets up the mapping between a given
3027  *	<host,channel,id,lun> (found in sdp) and new device name
3028  *	(e.g. /dev/sda). More precisely it is the block device major
3029  *	and minor number that is chosen here.
3030  *
3031  *	Assume sd_probe is not re-entrant (for time being)
3032  *	Also think about sd_probe() and sd_remove() running coincidentally.
3033  **/
3034 static int sd_probe(struct device *dev)
3035 {
3036 	struct scsi_device *sdp = to_scsi_device(dev);
3037 	struct scsi_disk *sdkp;
3038 	struct gendisk *gd;
3039 	int index;
3040 	int error;
3041 
3042 	scsi_autopm_get_device(sdp);
3043 	error = -ENODEV;
3044 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
3045 		goto out;
3046 
3047 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3048 					"sd_probe\n"));
3049 
3050 	error = -ENOMEM;
3051 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3052 	if (!sdkp)
3053 		goto out;
3054 
3055 	gd = alloc_disk(SD_MINORS);
3056 	if (!gd)
3057 		goto out_free;
3058 
3059 	do {
3060 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3061 			goto out_put;
3062 
3063 		spin_lock(&sd_index_lock);
3064 		error = ida_get_new(&sd_index_ida, &index);
3065 		spin_unlock(&sd_index_lock);
3066 	} while (error == -EAGAIN);
3067 
3068 	if (error) {
3069 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3070 		goto out_put;
3071 	}
3072 
3073 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3074 	if (error) {
3075 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3076 		goto out_free_index;
3077 	}
3078 
3079 	sdkp->device = sdp;
3080 	sdkp->driver = &sd_template;
3081 	sdkp->disk = gd;
3082 	sdkp->index = index;
3083 	atomic_set(&sdkp->openers, 0);
3084 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3085 
3086 	if (!sdp->request_queue->rq_timeout) {
3087 		if (sdp->type != TYPE_MOD)
3088 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3089 		else
3090 			blk_queue_rq_timeout(sdp->request_queue,
3091 					     SD_MOD_TIMEOUT);
3092 	}
3093 
3094 	device_initialize(&sdkp->dev);
3095 	sdkp->dev.parent = dev;
3096 	sdkp->dev.class = &sd_disk_class;
3097 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3098 
3099 	error = device_add(&sdkp->dev);
3100 	if (error)
3101 		goto out_free_index;
3102 
3103 	get_device(dev);
3104 	dev_set_drvdata(dev, sdkp);
3105 
3106 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3107 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3108 
3109 	return 0;
3110 
3111  out_free_index:
3112 	spin_lock(&sd_index_lock);
3113 	ida_remove(&sd_index_ida, index);
3114 	spin_unlock(&sd_index_lock);
3115  out_put:
3116 	put_disk(gd);
3117  out_free:
3118 	kfree(sdkp);
3119  out:
3120 	scsi_autopm_put_device(sdp);
3121 	return error;
3122 }
3123 
3124 /**
3125  *	sd_remove - called whenever a scsi disk (previously recognized by
3126  *	sd_probe) is detached from the system. It is called (potentially
3127  *	multiple times) during sd module unload.
3128  *	@sdp: pointer to mid level scsi device object
3129  *
3130  *	Note: this function is invoked from the scsi mid-level.
3131  *	This function potentially frees up a device name (e.g. /dev/sdc)
3132  *	that could be re-used by a subsequent sd_probe().
3133  *	This function is not called when the built-in sd driver is "exit-ed".
3134  **/
3135 static int sd_remove(struct device *dev)
3136 {
3137 	struct scsi_disk *sdkp;
3138 	dev_t devt;
3139 
3140 	sdkp = dev_get_drvdata(dev);
3141 	devt = disk_devt(sdkp->disk);
3142 	scsi_autopm_get_device(sdkp->device);
3143 
3144 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3145 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3146 	device_del(&sdkp->dev);
3147 	del_gendisk(sdkp->disk);
3148 	sd_shutdown(dev);
3149 
3150 	blk_register_region(devt, SD_MINORS, NULL,
3151 			    sd_default_probe, NULL, NULL);
3152 
3153 	mutex_lock(&sd_ref_mutex);
3154 	dev_set_drvdata(dev, NULL);
3155 	put_device(&sdkp->dev);
3156 	mutex_unlock(&sd_ref_mutex);
3157 
3158 	return 0;
3159 }
3160 
3161 /**
3162  *	scsi_disk_release - Called to free the scsi_disk structure
3163  *	@dev: pointer to embedded class device
3164  *
3165  *	sd_ref_mutex must be held entering this routine.  Because it is
3166  *	called on last put, you should always use the scsi_disk_get()
3167  *	scsi_disk_put() helpers which manipulate the semaphore directly
3168  *	and never do a direct put_device.
3169  **/
3170 static void scsi_disk_release(struct device *dev)
3171 {
3172 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3173 	struct gendisk *disk = sdkp->disk;
3174 
3175 	spin_lock(&sd_index_lock);
3176 	ida_remove(&sd_index_ida, sdkp->index);
3177 	spin_unlock(&sd_index_lock);
3178 
3179 	disk->private_data = NULL;
3180 	put_disk(disk);
3181 	put_device(&sdkp->device->sdev_gendev);
3182 
3183 	kfree(sdkp);
3184 }
3185 
3186 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3187 {
3188 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3189 	struct scsi_sense_hdr sshdr;
3190 	struct scsi_device *sdp = sdkp->device;
3191 	int res;
3192 
3193 	if (start)
3194 		cmd[4] |= 1;	/* START */
3195 
3196 	if (sdp->start_stop_pwr_cond)
3197 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3198 
3199 	if (!scsi_device_online(sdp))
3200 		return -ENODEV;
3201 
3202 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3203 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3204 	if (res) {
3205 		sd_print_result(sdkp, "Start/Stop Unit failed", res);
3206 		if (driver_byte(res) & DRIVER_SENSE)
3207 			sd_print_sense_hdr(sdkp, &sshdr);
3208 		if (scsi_sense_valid(&sshdr) &&
3209 			/* 0x3a is medium not present */
3210 			sshdr.asc == 0x3a)
3211 			res = 0;
3212 	}
3213 
3214 	/* SCSI error codes must not go to the generic layer */
3215 	if (res)
3216 		return -EIO;
3217 
3218 	return 0;
3219 }
3220 
3221 /*
3222  * Send a SYNCHRONIZE CACHE instruction down to the device through
3223  * the normal SCSI command structure.  Wait for the command to
3224  * complete.
3225  */
3226 static void sd_shutdown(struct device *dev)
3227 {
3228 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3229 
3230 	if (!sdkp)
3231 		return;         /* this can happen */
3232 
3233 	if (pm_runtime_suspended(dev))
3234 		return;
3235 
3236 	if (sdkp->WCE && sdkp->media_present) {
3237 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3238 		sd_sync_cache(sdkp);
3239 	}
3240 
3241 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3242 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3243 		sd_start_stop_device(sdkp, 0);
3244 	}
3245 }
3246 
3247 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3248 {
3249 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3250 	int ret = 0;
3251 
3252 	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
3253 		return 0;
3254 
3255 	if (sdkp->WCE && sdkp->media_present) {
3256 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3257 		ret = sd_sync_cache(sdkp);
3258 		if (ret) {
3259 			/* ignore OFFLINE device */
3260 			if (ret == -ENODEV)
3261 				ret = 0;
3262 			goto done;
3263 		}
3264 	}
3265 
3266 	if (sdkp->device->manage_start_stop) {
3267 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3268 		/* an error is not worth aborting a system sleep */
3269 		ret = sd_start_stop_device(sdkp, 0);
3270 		if (ignore_stop_errors)
3271 			ret = 0;
3272 	}
3273 
3274 done:
3275 	return ret;
3276 }
3277 
3278 static int sd_suspend_system(struct device *dev)
3279 {
3280 	return sd_suspend_common(dev, true);
3281 }
3282 
3283 static int sd_suspend_runtime(struct device *dev)
3284 {
3285 	return sd_suspend_common(dev, false);
3286 }
3287 
3288 static int sd_resume(struct device *dev)
3289 {
3290 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
3291 
3292 	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
3293 		return 0;
3294 
3295 	if (!sdkp->device->manage_start_stop)
3296 		return 0;
3297 
3298 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3299 	return sd_start_stop_device(sdkp, 1);
3300 }
3301 
3302 /**
3303  *	init_sd - entry point for this driver (both when built in or when
3304  *	a module).
3305  *
3306  *	Note: this function registers this driver with the scsi mid-level.
3307  **/
3308 static int __init init_sd(void)
3309 {
3310 	int majors = 0, i, err;
3311 
3312 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3313 
3314 	for (i = 0; i < SD_MAJORS; i++) {
3315 		if (register_blkdev(sd_major(i), "sd") != 0)
3316 			continue;
3317 		majors++;
3318 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3319 				    sd_default_probe, NULL, NULL);
3320 	}
3321 
3322 	if (!majors)
3323 		return -ENODEV;
3324 
3325 	err = class_register(&sd_disk_class);
3326 	if (err)
3327 		goto err_out;
3328 
3329 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3330 					 0, 0, NULL);
3331 	if (!sd_cdb_cache) {
3332 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3333 		err = -ENOMEM;
3334 		goto err_out_class;
3335 	}
3336 
3337 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3338 	if (!sd_cdb_pool) {
3339 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3340 		err = -ENOMEM;
3341 		goto err_out_cache;
3342 	}
3343 
3344 	err = scsi_register_driver(&sd_template.gendrv);
3345 	if (err)
3346 		goto err_out_driver;
3347 
3348 	return 0;
3349 
3350 err_out_driver:
3351 	mempool_destroy(sd_cdb_pool);
3352 
3353 err_out_cache:
3354 	kmem_cache_destroy(sd_cdb_cache);
3355 
3356 err_out_class:
3357 	class_unregister(&sd_disk_class);
3358 err_out:
3359 	for (i = 0; i < SD_MAJORS; i++)
3360 		unregister_blkdev(sd_major(i), "sd");
3361 	return err;
3362 }
3363 
3364 /**
3365  *	exit_sd - exit point for this driver (when it is a module).
3366  *
3367  *	Note: this function unregisters this driver from the scsi mid-level.
3368  **/
3369 static void __exit exit_sd(void)
3370 {
3371 	int i;
3372 
3373 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3374 
3375 	scsi_unregister_driver(&sd_template.gendrv);
3376 	mempool_destroy(sd_cdb_pool);
3377 	kmem_cache_destroy(sd_cdb_cache);
3378 
3379 	class_unregister(&sd_disk_class);
3380 
3381 	for (i = 0; i < SD_MAJORS; i++) {
3382 		blk_unregister_region(sd_major(i), SD_MINORS);
3383 		unregister_blkdev(sd_major(i), "sd");
3384 	}
3385 }
3386 
3387 module_init(init_sd);
3388 module_exit(exit_sd);
3389 
3390 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3391 			       struct scsi_sense_hdr *sshdr)
3392 {
3393 	scsi_print_sense_hdr(sdkp->device,
3394 			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3395 }
3396 
3397 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3398 			    int result)
3399 {
3400 	const char *hb_string = scsi_hostbyte_string(result);
3401 	const char *db_string = scsi_driverbyte_string(result);
3402 
3403 	if (hb_string || db_string)
3404 		sd_printk(KERN_INFO, sdkp,
3405 			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3406 			  hb_string ? hb_string : "invalid",
3407 			  db_string ? db_string : "invalid");
3408 	else
3409 		sd_printk(KERN_INFO, sdkp,
3410 			  "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3411 			  msg, host_byte(result), driver_byte(result));
3412 }
3413 
3414