xref: /openbmc/linux/drivers/scsi/sd.c (revision af958a38)
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *	Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17  *	   sd_init and cleanups.
18  *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *	   not being read in sd_open. Fix problem where removable media
20  *	   could be ejected after sd_open.
21  *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23  *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24  *	   Support 32k/1M disks.
25  *
26  *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *	Note: when the logging level is set by the user, it must be greater
32  *	than the level indicated above to trigger output.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <asm/uaccess.h>
55 #include <asm/unaligned.h>
56 
57 #include <scsi/scsi.h>
58 #include <scsi/scsi_cmnd.h>
59 #include <scsi/scsi_dbg.h>
60 #include <scsi/scsi_device.h>
61 #include <scsi/scsi_driver.h>
62 #include <scsi/scsi_eh.h>
63 #include <scsi/scsi_host.h>
64 #include <scsi/scsi_ioctl.h>
65 #include <scsi/scsicam.h>
66 
67 #include "sd.h"
68 #include "scsi_priv.h"
69 #include "scsi_logging.h"
70 
71 MODULE_AUTHOR("Eric Youngdale");
72 MODULE_DESCRIPTION("SCSI disk (sd) driver");
73 MODULE_LICENSE("GPL");
74 
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
94 
95 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
96 #define SD_MINORS	16
97 #else
98 #define SD_MINORS	0
99 #endif
100 
101 static void sd_config_discard(struct scsi_disk *, unsigned int);
102 static void sd_config_write_same(struct scsi_disk *);
103 static int  sd_revalidate_disk(struct gendisk *);
104 static void sd_unlock_native_capacity(struct gendisk *disk);
105 static int  sd_probe(struct device *);
106 static int  sd_remove(struct device *);
107 static void sd_shutdown(struct device *);
108 static int sd_suspend_system(struct device *);
109 static int sd_suspend_runtime(struct device *);
110 static int sd_resume(struct device *);
111 static void sd_rescan(struct device *);
112 static int sd_init_command(struct scsi_cmnd *SCpnt);
113 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
114 static int sd_done(struct scsi_cmnd *);
115 static int sd_eh_action(struct scsi_cmnd *, int);
116 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
117 static void scsi_disk_release(struct device *cdev);
118 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
119 static void sd_print_result(struct scsi_disk *, int);
120 
121 static DEFINE_SPINLOCK(sd_index_lock);
122 static DEFINE_IDA(sd_index_ida);
123 
124 /* This semaphore is used to mediate the 0->1 reference get in the
125  * face of object destruction (i.e. we can't allow a get on an
126  * object after last put) */
127 static DEFINE_MUTEX(sd_ref_mutex);
128 
129 static struct kmem_cache *sd_cdb_cache;
130 static mempool_t *sd_cdb_pool;
131 
132 static const char *sd_cache_types[] = {
133 	"write through", "none", "write back",
134 	"write back, no read (daft)"
135 };
136 
137 static void sd_set_flush_flag(struct scsi_disk *sdkp)
138 {
139 	unsigned flush = 0;
140 
141 	if (sdkp->WCE) {
142 		flush |= REQ_FLUSH;
143 		if (sdkp->DPOFUA)
144 			flush |= REQ_FUA;
145 	}
146 
147 	blk_queue_flush(sdkp->disk->queue, flush);
148 }
149 
150 static ssize_t
151 cache_type_store(struct device *dev, struct device_attribute *attr,
152 		 const char *buf, size_t count)
153 {
154 	int i, ct = -1, rcd, wce, sp;
155 	struct scsi_disk *sdkp = to_scsi_disk(dev);
156 	struct scsi_device *sdp = sdkp->device;
157 	char buffer[64];
158 	char *buffer_data;
159 	struct scsi_mode_data data;
160 	struct scsi_sense_hdr sshdr;
161 	static const char temp[] = "temporary ";
162 	int len;
163 
164 	if (sdp->type != TYPE_DISK)
165 		/* no cache control on RBC devices; theoretically they
166 		 * can do it, but there's probably so many exceptions
167 		 * it's not worth the risk */
168 		return -EINVAL;
169 
170 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
171 		buf += sizeof(temp) - 1;
172 		sdkp->cache_override = 1;
173 	} else {
174 		sdkp->cache_override = 0;
175 	}
176 
177 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
178 		len = strlen(sd_cache_types[i]);
179 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
180 		    buf[len] == '\n') {
181 			ct = i;
182 			break;
183 		}
184 	}
185 	if (ct < 0)
186 		return -EINVAL;
187 	rcd = ct & 0x01 ? 1 : 0;
188 	wce = ct & 0x02 ? 1 : 0;
189 
190 	if (sdkp->cache_override) {
191 		sdkp->WCE = wce;
192 		sdkp->RCD = rcd;
193 		sd_set_flush_flag(sdkp);
194 		return count;
195 	}
196 
197 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198 			    SD_MAX_RETRIES, &data, NULL))
199 		return -EINVAL;
200 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201 		  data.block_descriptor_length);
202 	buffer_data = buffer + data.header_length +
203 		data.block_descriptor_length;
204 	buffer_data[2] &= ~0x05;
205 	buffer_data[2] |= wce << 2 | rcd;
206 	sp = buffer_data[0] & 0x80 ? 1 : 0;
207 
208 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209 			     SD_MAX_RETRIES, &data, &sshdr)) {
210 		if (scsi_sense_valid(&sshdr))
211 			sd_print_sense_hdr(sdkp, &sshdr);
212 		return -EINVAL;
213 	}
214 	revalidate_disk(sdkp->disk);
215 	return count;
216 }
217 
218 static ssize_t
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
220 		       char *buf)
221 {
222 	struct scsi_disk *sdkp = to_scsi_disk(dev);
223 	struct scsi_device *sdp = sdkp->device;
224 
225 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
226 }
227 
228 static ssize_t
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230 			const char *buf, size_t count)
231 {
232 	struct scsi_disk *sdkp = to_scsi_disk(dev);
233 	struct scsi_device *sdp = sdkp->device;
234 
235 	if (!capable(CAP_SYS_ADMIN))
236 		return -EACCES;
237 
238 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
239 
240 	return count;
241 }
242 static DEVICE_ATTR_RW(manage_start_stop);
243 
244 static ssize_t
245 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
246 {
247 	struct scsi_disk *sdkp = to_scsi_disk(dev);
248 
249 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
250 }
251 
252 static ssize_t
253 allow_restart_store(struct device *dev, struct device_attribute *attr,
254 		    const char *buf, size_t count)
255 {
256 	struct scsi_disk *sdkp = to_scsi_disk(dev);
257 	struct scsi_device *sdp = sdkp->device;
258 
259 	if (!capable(CAP_SYS_ADMIN))
260 		return -EACCES;
261 
262 	if (sdp->type != TYPE_DISK)
263 		return -EINVAL;
264 
265 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
266 
267 	return count;
268 }
269 static DEVICE_ATTR_RW(allow_restart);
270 
271 static ssize_t
272 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
273 {
274 	struct scsi_disk *sdkp = to_scsi_disk(dev);
275 	int ct = sdkp->RCD + 2*sdkp->WCE;
276 
277 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
278 }
279 static DEVICE_ATTR_RW(cache_type);
280 
281 static ssize_t
282 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
283 {
284 	struct scsi_disk *sdkp = to_scsi_disk(dev);
285 
286 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
287 }
288 static DEVICE_ATTR_RO(FUA);
289 
290 static ssize_t
291 protection_type_show(struct device *dev, struct device_attribute *attr,
292 		     char *buf)
293 {
294 	struct scsi_disk *sdkp = to_scsi_disk(dev);
295 
296 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
297 }
298 
299 static ssize_t
300 protection_type_store(struct device *dev, struct device_attribute *attr,
301 		      const char *buf, size_t count)
302 {
303 	struct scsi_disk *sdkp = to_scsi_disk(dev);
304 	unsigned int val;
305 	int err;
306 
307 	if (!capable(CAP_SYS_ADMIN))
308 		return -EACCES;
309 
310 	err = kstrtouint(buf, 10, &val);
311 
312 	if (err)
313 		return err;
314 
315 	if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
316 		sdkp->protection_type = val;
317 
318 	return count;
319 }
320 static DEVICE_ATTR_RW(protection_type);
321 
322 static ssize_t
323 protection_mode_show(struct device *dev, struct device_attribute *attr,
324 		     char *buf)
325 {
326 	struct scsi_disk *sdkp = to_scsi_disk(dev);
327 	struct scsi_device *sdp = sdkp->device;
328 	unsigned int dif, dix;
329 
330 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
331 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
332 
333 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
334 		dif = 0;
335 		dix = 1;
336 	}
337 
338 	if (!dif && !dix)
339 		return snprintf(buf, 20, "none\n");
340 
341 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
342 }
343 static DEVICE_ATTR_RO(protection_mode);
344 
345 static ssize_t
346 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
347 {
348 	struct scsi_disk *sdkp = to_scsi_disk(dev);
349 
350 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
351 }
352 static DEVICE_ATTR_RO(app_tag_own);
353 
354 static ssize_t
355 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
356 		       char *buf)
357 {
358 	struct scsi_disk *sdkp = to_scsi_disk(dev);
359 
360 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
361 }
362 static DEVICE_ATTR_RO(thin_provisioning);
363 
364 static const char *lbp_mode[] = {
365 	[SD_LBP_FULL]		= "full",
366 	[SD_LBP_UNMAP]		= "unmap",
367 	[SD_LBP_WS16]		= "writesame_16",
368 	[SD_LBP_WS10]		= "writesame_10",
369 	[SD_LBP_ZERO]		= "writesame_zero",
370 	[SD_LBP_DISABLE]	= "disabled",
371 };
372 
373 static ssize_t
374 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
375 		       char *buf)
376 {
377 	struct scsi_disk *sdkp = to_scsi_disk(dev);
378 
379 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
380 }
381 
382 static ssize_t
383 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
384 			const char *buf, size_t count)
385 {
386 	struct scsi_disk *sdkp = to_scsi_disk(dev);
387 	struct scsi_device *sdp = sdkp->device;
388 
389 	if (!capable(CAP_SYS_ADMIN))
390 		return -EACCES;
391 
392 	if (sdp->type != TYPE_DISK)
393 		return -EINVAL;
394 
395 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
396 		sd_config_discard(sdkp, SD_LBP_UNMAP);
397 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
398 		sd_config_discard(sdkp, SD_LBP_WS16);
399 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
400 		sd_config_discard(sdkp, SD_LBP_WS10);
401 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
402 		sd_config_discard(sdkp, SD_LBP_ZERO);
403 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
404 		sd_config_discard(sdkp, SD_LBP_DISABLE);
405 	else
406 		return -EINVAL;
407 
408 	return count;
409 }
410 static DEVICE_ATTR_RW(provisioning_mode);
411 
412 static ssize_t
413 max_medium_access_timeouts_show(struct device *dev,
414 				struct device_attribute *attr, char *buf)
415 {
416 	struct scsi_disk *sdkp = to_scsi_disk(dev);
417 
418 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
419 }
420 
421 static ssize_t
422 max_medium_access_timeouts_store(struct device *dev,
423 				 struct device_attribute *attr, const char *buf,
424 				 size_t count)
425 {
426 	struct scsi_disk *sdkp = to_scsi_disk(dev);
427 	int err;
428 
429 	if (!capable(CAP_SYS_ADMIN))
430 		return -EACCES;
431 
432 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
433 
434 	return err ? err : count;
435 }
436 static DEVICE_ATTR_RW(max_medium_access_timeouts);
437 
438 static ssize_t
439 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
440 			   char *buf)
441 {
442 	struct scsi_disk *sdkp = to_scsi_disk(dev);
443 
444 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
445 }
446 
447 static ssize_t
448 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
449 			    const char *buf, size_t count)
450 {
451 	struct scsi_disk *sdkp = to_scsi_disk(dev);
452 	struct scsi_device *sdp = sdkp->device;
453 	unsigned long max;
454 	int err;
455 
456 	if (!capable(CAP_SYS_ADMIN))
457 		return -EACCES;
458 
459 	if (sdp->type != TYPE_DISK)
460 		return -EINVAL;
461 
462 	err = kstrtoul(buf, 10, &max);
463 
464 	if (err)
465 		return err;
466 
467 	if (max == 0)
468 		sdp->no_write_same = 1;
469 	else if (max <= SD_MAX_WS16_BLOCKS) {
470 		sdp->no_write_same = 0;
471 		sdkp->max_ws_blocks = max;
472 	}
473 
474 	sd_config_write_same(sdkp);
475 
476 	return count;
477 }
478 static DEVICE_ATTR_RW(max_write_same_blocks);
479 
480 static struct attribute *sd_disk_attrs[] = {
481 	&dev_attr_cache_type.attr,
482 	&dev_attr_FUA.attr,
483 	&dev_attr_allow_restart.attr,
484 	&dev_attr_manage_start_stop.attr,
485 	&dev_attr_protection_type.attr,
486 	&dev_attr_protection_mode.attr,
487 	&dev_attr_app_tag_own.attr,
488 	&dev_attr_thin_provisioning.attr,
489 	&dev_attr_provisioning_mode.attr,
490 	&dev_attr_max_write_same_blocks.attr,
491 	&dev_attr_max_medium_access_timeouts.attr,
492 	NULL,
493 };
494 ATTRIBUTE_GROUPS(sd_disk);
495 
496 static struct class sd_disk_class = {
497 	.name		= "scsi_disk",
498 	.owner		= THIS_MODULE,
499 	.dev_release	= scsi_disk_release,
500 	.dev_groups	= sd_disk_groups,
501 };
502 
503 static const struct dev_pm_ops sd_pm_ops = {
504 	.suspend		= sd_suspend_system,
505 	.resume			= sd_resume,
506 	.poweroff		= sd_suspend_system,
507 	.restore		= sd_resume,
508 	.runtime_suspend	= sd_suspend_runtime,
509 	.runtime_resume		= sd_resume,
510 };
511 
512 static struct scsi_driver sd_template = {
513 	.owner			= THIS_MODULE,
514 	.gendrv = {
515 		.name		= "sd",
516 		.probe		= sd_probe,
517 		.remove		= sd_remove,
518 		.shutdown	= sd_shutdown,
519 		.pm		= &sd_pm_ops,
520 	},
521 	.rescan			= sd_rescan,
522 	.init_command		= sd_init_command,
523 	.uninit_command		= sd_uninit_command,
524 	.done			= sd_done,
525 	.eh_action		= sd_eh_action,
526 };
527 
528 /*
529  * Dummy kobj_map->probe function.
530  * The default ->probe function will call modprobe, which is
531  * pointless as this module is already loaded.
532  */
533 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
534 {
535 	return NULL;
536 }
537 
538 /*
539  * Device no to disk mapping:
540  *
541  *       major         disc2     disc  p1
542  *   |............|.............|....|....| <- dev_t
543  *    31        20 19          8 7  4 3  0
544  *
545  * Inside a major, we have 16k disks, however mapped non-
546  * contiguously. The first 16 disks are for major0, the next
547  * ones with major1, ... Disk 256 is for major0 again, disk 272
548  * for major1, ...
549  * As we stay compatible with our numbering scheme, we can reuse
550  * the well-know SCSI majors 8, 65--71, 136--143.
551  */
552 static int sd_major(int major_idx)
553 {
554 	switch (major_idx) {
555 	case 0:
556 		return SCSI_DISK0_MAJOR;
557 	case 1 ... 7:
558 		return SCSI_DISK1_MAJOR + major_idx - 1;
559 	case 8 ... 15:
560 		return SCSI_DISK8_MAJOR + major_idx - 8;
561 	default:
562 		BUG();
563 		return 0;	/* shut up gcc */
564 	}
565 }
566 
567 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
568 {
569 	struct scsi_disk *sdkp = NULL;
570 
571 	if (disk->private_data) {
572 		sdkp = scsi_disk(disk);
573 		if (scsi_device_get(sdkp->device) == 0)
574 			get_device(&sdkp->dev);
575 		else
576 			sdkp = NULL;
577 	}
578 	return sdkp;
579 }
580 
581 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
582 {
583 	struct scsi_disk *sdkp;
584 
585 	mutex_lock(&sd_ref_mutex);
586 	sdkp = __scsi_disk_get(disk);
587 	mutex_unlock(&sd_ref_mutex);
588 	return sdkp;
589 }
590 
591 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
592 {
593 	struct scsi_disk *sdkp;
594 
595 	mutex_lock(&sd_ref_mutex);
596 	sdkp = dev_get_drvdata(dev);
597 	if (sdkp)
598 		sdkp = __scsi_disk_get(sdkp->disk);
599 	mutex_unlock(&sd_ref_mutex);
600 	return sdkp;
601 }
602 
603 static void scsi_disk_put(struct scsi_disk *sdkp)
604 {
605 	struct scsi_device *sdev = sdkp->device;
606 
607 	mutex_lock(&sd_ref_mutex);
608 	put_device(&sdkp->dev);
609 	scsi_device_put(sdev);
610 	mutex_unlock(&sd_ref_mutex);
611 }
612 
613 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
614 {
615 	unsigned int prot_op = SCSI_PROT_NORMAL;
616 	unsigned int dix = scsi_prot_sg_count(scmd);
617 
618 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
619 		if (dif && dix)
620 			prot_op = SCSI_PROT_READ_PASS;
621 		else if (dif && !dix)
622 			prot_op = SCSI_PROT_READ_STRIP;
623 		else if (!dif && dix)
624 			prot_op = SCSI_PROT_READ_INSERT;
625 	} else {
626 		if (dif && dix)
627 			prot_op = SCSI_PROT_WRITE_PASS;
628 		else if (dif && !dix)
629 			prot_op = SCSI_PROT_WRITE_INSERT;
630 		else if (!dif && dix)
631 			prot_op = SCSI_PROT_WRITE_STRIP;
632 	}
633 
634 	scsi_set_prot_op(scmd, prot_op);
635 	scsi_set_prot_type(scmd, dif);
636 }
637 
638 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
639 {
640 	struct request_queue *q = sdkp->disk->queue;
641 	unsigned int logical_block_size = sdkp->device->sector_size;
642 	unsigned int max_blocks = 0;
643 
644 	q->limits.discard_zeroes_data = sdkp->lbprz;
645 	q->limits.discard_alignment = sdkp->unmap_alignment *
646 		logical_block_size;
647 	q->limits.discard_granularity =
648 		max(sdkp->physical_block_size,
649 		    sdkp->unmap_granularity * logical_block_size);
650 
651 	sdkp->provisioning_mode = mode;
652 
653 	switch (mode) {
654 
655 	case SD_LBP_DISABLE:
656 		q->limits.max_discard_sectors = 0;
657 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
658 		return;
659 
660 	case SD_LBP_UNMAP:
661 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
662 					  (u32)SD_MAX_WS16_BLOCKS);
663 		break;
664 
665 	case SD_LBP_WS16:
666 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
667 					  (u32)SD_MAX_WS16_BLOCKS);
668 		break;
669 
670 	case SD_LBP_WS10:
671 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
672 					  (u32)SD_MAX_WS10_BLOCKS);
673 		break;
674 
675 	case SD_LBP_ZERO:
676 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
677 					  (u32)SD_MAX_WS10_BLOCKS);
678 		q->limits.discard_zeroes_data = 1;
679 		break;
680 	}
681 
682 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
683 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
684 }
685 
686 /**
687  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
688  * @sdp: scsi device to operate one
689  * @rq: Request to prepare
690  *
691  * Will issue either UNMAP or WRITE SAME(16) depending on preference
692  * indicated by target device.
693  **/
694 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
695 {
696 	struct request *rq = cmd->request;
697 	struct scsi_device *sdp = cmd->device;
698 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
699 	sector_t sector = blk_rq_pos(rq);
700 	unsigned int nr_sectors = blk_rq_sectors(rq);
701 	unsigned int nr_bytes = blk_rq_bytes(rq);
702 	unsigned int len;
703 	int ret;
704 	char *buf;
705 	struct page *page;
706 
707 	sector >>= ilog2(sdp->sector_size) - 9;
708 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
709 
710 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
711 	if (!page)
712 		return BLKPREP_DEFER;
713 
714 	switch (sdkp->provisioning_mode) {
715 	case SD_LBP_UNMAP:
716 		buf = page_address(page);
717 
718 		cmd->cmd_len = 10;
719 		cmd->cmnd[0] = UNMAP;
720 		cmd->cmnd[8] = 24;
721 
722 		put_unaligned_be16(6 + 16, &buf[0]);
723 		put_unaligned_be16(16, &buf[2]);
724 		put_unaligned_be64(sector, &buf[8]);
725 		put_unaligned_be32(nr_sectors, &buf[16]);
726 
727 		len = 24;
728 		break;
729 
730 	case SD_LBP_WS16:
731 		cmd->cmd_len = 16;
732 		cmd->cmnd[0] = WRITE_SAME_16;
733 		cmd->cmnd[1] = 0x8; /* UNMAP */
734 		put_unaligned_be64(sector, &cmd->cmnd[2]);
735 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
736 
737 		len = sdkp->device->sector_size;
738 		break;
739 
740 	case SD_LBP_WS10:
741 	case SD_LBP_ZERO:
742 		cmd->cmd_len = 10;
743 		cmd->cmnd[0] = WRITE_SAME;
744 		if (sdkp->provisioning_mode == SD_LBP_WS10)
745 			cmd->cmnd[1] = 0x8; /* UNMAP */
746 		put_unaligned_be32(sector, &cmd->cmnd[2]);
747 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
748 
749 		len = sdkp->device->sector_size;
750 		break;
751 
752 	default:
753 		ret = BLKPREP_KILL;
754 		goto out;
755 	}
756 
757 	rq->completion_data = page;
758 	rq->timeout = SD_TIMEOUT;
759 
760 	cmd->transfersize = len;
761 	cmd->allowed = SD_MAX_RETRIES;
762 
763 	/*
764 	 * Initially __data_len is set to the amount of data that needs to be
765 	 * transferred to the target. This amount depends on whether WRITE SAME
766 	 * or UNMAP is being used. After the scatterlist has been mapped by
767 	 * scsi_init_io() we set __data_len to the size of the area to be
768 	 * discarded on disk. This allows us to report completion on the full
769 	 * amount of blocks described by the request.
770 	 */
771 	blk_add_request_payload(rq, page, len);
772 	ret = scsi_init_io(cmd, GFP_ATOMIC);
773 	rq->__data_len = nr_bytes;
774 
775 out:
776 	if (ret != BLKPREP_OK)
777 		__free_page(page);
778 	return ret;
779 }
780 
781 static void sd_config_write_same(struct scsi_disk *sdkp)
782 {
783 	struct request_queue *q = sdkp->disk->queue;
784 	unsigned int logical_block_size = sdkp->device->sector_size;
785 
786 	if (sdkp->device->no_write_same) {
787 		sdkp->max_ws_blocks = 0;
788 		goto out;
789 	}
790 
791 	/* Some devices can not handle block counts above 0xffff despite
792 	 * supporting WRITE SAME(16). Consequently we default to 64k
793 	 * blocks per I/O unless the device explicitly advertises a
794 	 * bigger limit.
795 	 */
796 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
797 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
798 						   (u32)SD_MAX_WS16_BLOCKS);
799 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
800 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
801 						   (u32)SD_MAX_WS10_BLOCKS);
802 	else {
803 		sdkp->device->no_write_same = 1;
804 		sdkp->max_ws_blocks = 0;
805 	}
806 
807 out:
808 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
809 					 (logical_block_size >> 9));
810 }
811 
812 /**
813  * sd_setup_write_same_cmnd - write the same data to multiple blocks
814  * @cmd: command to prepare
815  *
816  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
817  * preference indicated by target device.
818  **/
819 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
820 {
821 	struct request *rq = cmd->request;
822 	struct scsi_device *sdp = cmd->device;
823 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
824 	struct bio *bio = rq->bio;
825 	sector_t sector = blk_rq_pos(rq);
826 	unsigned int nr_sectors = blk_rq_sectors(rq);
827 	unsigned int nr_bytes = blk_rq_bytes(rq);
828 	int ret;
829 
830 	if (sdkp->device->no_write_same)
831 		return BLKPREP_KILL;
832 
833 	BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
834 
835 	sector >>= ilog2(sdp->sector_size) - 9;
836 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
837 
838 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
839 
840 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
841 		cmd->cmd_len = 16;
842 		cmd->cmnd[0] = WRITE_SAME_16;
843 		put_unaligned_be64(sector, &cmd->cmnd[2]);
844 		put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
845 	} else {
846 		cmd->cmd_len = 10;
847 		cmd->cmnd[0] = WRITE_SAME;
848 		put_unaligned_be32(sector, &cmd->cmnd[2]);
849 		put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
850 	}
851 
852 	cmd->transfersize = sdp->sector_size;
853 	cmd->allowed = SD_MAX_RETRIES;
854 
855 	/*
856 	 * For WRITE_SAME the data transferred in the DATA IN buffer is
857 	 * different from the amount of data actually written to the target.
858 	 *
859 	 * We set up __data_len to the amount of data transferred from the
860 	 * DATA IN buffer so that blk_rq_map_sg set up the proper S/G list
861 	 * to transfer a single sector of data first, but then reset it to
862 	 * the amount of data to be written right after so that the I/O path
863 	 * knows how much to actually write.
864 	 */
865 	rq->__data_len = sdp->sector_size;
866 	ret = scsi_init_io(cmd, GFP_ATOMIC);
867 	rq->__data_len = nr_bytes;
868 	return ret;
869 }
870 
871 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
872 {
873 	struct request *rq = cmd->request;
874 
875 	/* flush requests don't perform I/O, zero the S/G table */
876 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
877 
878 	cmd->cmnd[0] = SYNCHRONIZE_CACHE;
879 	cmd->cmd_len = 10;
880 	cmd->transfersize = 0;
881 	cmd->allowed = SD_MAX_RETRIES;
882 
883 	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
884 	return BLKPREP_OK;
885 }
886 
887 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
888 {
889 	struct request *rq = SCpnt->request;
890 	struct scsi_device *sdp = SCpnt->device;
891 	struct gendisk *disk = rq->rq_disk;
892 	struct scsi_disk *sdkp;
893 	sector_t block = blk_rq_pos(rq);
894 	sector_t threshold;
895 	unsigned int this_count = blk_rq_sectors(rq);
896 	int ret, host_dif;
897 	unsigned char protect;
898 
899 	ret = scsi_init_io(SCpnt, GFP_ATOMIC);
900 	if (ret != BLKPREP_OK)
901 		goto out;
902 	SCpnt = rq->special;
903 	sdkp = scsi_disk(disk);
904 
905 	/* from here on until we're complete, any goto out
906 	 * is used for a killable error condition */
907 	ret = BLKPREP_KILL;
908 
909 	SCSI_LOG_HLQUEUE(1,
910 		scmd_printk(KERN_INFO, SCpnt,
911 			"%s: block=%llu, count=%d\n",
912 			__func__, (unsigned long long)block, this_count));
913 
914 	if (!sdp || !scsi_device_online(sdp) ||
915 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
916 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
917 						"Finishing %u sectors\n",
918 						blk_rq_sectors(rq)));
919 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
920 						"Retry with 0x%p\n", SCpnt));
921 		goto out;
922 	}
923 
924 	if (sdp->changed) {
925 		/*
926 		 * quietly refuse to do anything to a changed disc until
927 		 * the changed bit has been reset
928 		 */
929 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
930 		goto out;
931 	}
932 
933 	/*
934 	 * Some SD card readers can't handle multi-sector accesses which touch
935 	 * the last one or two hardware sectors.  Split accesses as needed.
936 	 */
937 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
938 		(sdp->sector_size / 512);
939 
940 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
941 		if (block < threshold) {
942 			/* Access up to the threshold but not beyond */
943 			this_count = threshold - block;
944 		} else {
945 			/* Access only a single hardware sector */
946 			this_count = sdp->sector_size / 512;
947 		}
948 	}
949 
950 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
951 					(unsigned long long)block));
952 
953 	/*
954 	 * If we have a 1K hardware sectorsize, prevent access to single
955 	 * 512 byte sectors.  In theory we could handle this - in fact
956 	 * the scsi cdrom driver must be able to handle this because
957 	 * we typically use 1K blocksizes, and cdroms typically have
958 	 * 2K hardware sectorsizes.  Of course, things are simpler
959 	 * with the cdrom, since it is read-only.  For performance
960 	 * reasons, the filesystems should be able to handle this
961 	 * and not force the scsi disk driver to use bounce buffers
962 	 * for this.
963 	 */
964 	if (sdp->sector_size == 1024) {
965 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
966 			scmd_printk(KERN_ERR, SCpnt,
967 				    "Bad block number requested\n");
968 			goto out;
969 		} else {
970 			block = block >> 1;
971 			this_count = this_count >> 1;
972 		}
973 	}
974 	if (sdp->sector_size == 2048) {
975 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
976 			scmd_printk(KERN_ERR, SCpnt,
977 				    "Bad block number requested\n");
978 			goto out;
979 		} else {
980 			block = block >> 2;
981 			this_count = this_count >> 2;
982 		}
983 	}
984 	if (sdp->sector_size == 4096) {
985 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
986 			scmd_printk(KERN_ERR, SCpnt,
987 				    "Bad block number requested\n");
988 			goto out;
989 		} else {
990 			block = block >> 3;
991 			this_count = this_count >> 3;
992 		}
993 	}
994 	if (rq_data_dir(rq) == WRITE) {
995 		SCpnt->cmnd[0] = WRITE_6;
996 
997 		if (blk_integrity_rq(rq))
998 			sd_dif_prepare(rq, block, sdp->sector_size);
999 
1000 	} else if (rq_data_dir(rq) == READ) {
1001 		SCpnt->cmnd[0] = READ_6;
1002 	} else {
1003 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %llx\n", (unsigned long long) rq->cmd_flags);
1004 		goto out;
1005 	}
1006 
1007 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1008 					"%s %d/%u 512 byte blocks.\n",
1009 					(rq_data_dir(rq) == WRITE) ?
1010 					"writing" : "reading", this_count,
1011 					blk_rq_sectors(rq)));
1012 
1013 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
1014 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
1015 	if (host_dif)
1016 		protect = 1 << 5;
1017 	else
1018 		protect = 0;
1019 
1020 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
1021 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1022 
1023 		if (unlikely(SCpnt->cmnd == NULL)) {
1024 			ret = BLKPREP_DEFER;
1025 			goto out;
1026 		}
1027 
1028 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1029 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1030 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1031 		SCpnt->cmnd[7] = 0x18;
1032 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1033 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1034 
1035 		/* LBA */
1036 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1037 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1038 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1039 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1040 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1041 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1042 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1043 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1044 
1045 		/* Expected Indirect LBA */
1046 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1047 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1048 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1049 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1050 
1051 		/* Transfer length */
1052 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1053 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1054 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1055 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1056 	} else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1057 		SCpnt->cmnd[0] += READ_16 - READ_6;
1058 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1059 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1060 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1061 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1062 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1063 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1064 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1065 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1066 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1067 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1068 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1069 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1070 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1071 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1072 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1073 		   scsi_device_protection(SCpnt->device) ||
1074 		   SCpnt->device->use_10_for_rw) {
1075 		SCpnt->cmnd[0] += READ_10 - READ_6;
1076 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1077 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1078 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1079 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1080 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1081 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1082 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1083 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1084 	} else {
1085 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1086 			/*
1087 			 * This happens only if this drive failed
1088 			 * 10byte rw command with ILLEGAL_REQUEST
1089 			 * during operation and thus turned off
1090 			 * use_10_for_rw.
1091 			 */
1092 			scmd_printk(KERN_ERR, SCpnt,
1093 				    "FUA write on READ/WRITE(6) drive\n");
1094 			goto out;
1095 		}
1096 
1097 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1098 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1099 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1100 		SCpnt->cmnd[4] = (unsigned char) this_count;
1101 		SCpnt->cmnd[5] = 0;
1102 	}
1103 	SCpnt->sdb.length = this_count * sdp->sector_size;
1104 
1105 	/* If DIF or DIX is enabled, tell HBA how to handle request */
1106 	if (host_dif || scsi_prot_sg_count(SCpnt))
1107 		sd_prot_op(SCpnt, host_dif);
1108 
1109 	/*
1110 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1111 	 * host adapter, it's safe to assume that we can at least transfer
1112 	 * this many bytes between each connect / disconnect.
1113 	 */
1114 	SCpnt->transfersize = sdp->sector_size;
1115 	SCpnt->underflow = this_count << 9;
1116 	SCpnt->allowed = SD_MAX_RETRIES;
1117 
1118 	/*
1119 	 * This indicates that the command is ready from our end to be
1120 	 * queued.
1121 	 */
1122 	ret = BLKPREP_OK;
1123  out:
1124 	return ret;
1125 }
1126 
1127 static int sd_init_command(struct scsi_cmnd *cmd)
1128 {
1129 	struct request *rq = cmd->request;
1130 
1131 	if (rq->cmd_flags & REQ_DISCARD)
1132 		return sd_setup_discard_cmnd(cmd);
1133 	else if (rq->cmd_flags & REQ_WRITE_SAME)
1134 		return sd_setup_write_same_cmnd(cmd);
1135 	else if (rq->cmd_flags & REQ_FLUSH)
1136 		return sd_setup_flush_cmnd(cmd);
1137 	else
1138 		return sd_setup_read_write_cmnd(cmd);
1139 }
1140 
1141 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1142 {
1143 	struct request *rq = SCpnt->request;
1144 
1145 	if (rq->cmd_flags & REQ_DISCARD)
1146 		__free_page(rq->completion_data);
1147 
1148 	if (SCpnt->cmnd != rq->cmd) {
1149 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1150 		SCpnt->cmnd = NULL;
1151 		SCpnt->cmd_len = 0;
1152 	}
1153 }
1154 
1155 /**
1156  *	sd_open - open a scsi disk device
1157  *	@inode: only i_rdev member may be used
1158  *	@filp: only f_mode and f_flags may be used
1159  *
1160  *	Returns 0 if successful. Returns a negated errno value in case
1161  *	of error.
1162  *
1163  *	Note: This can be called from a user context (e.g. fsck(1) )
1164  *	or from within the kernel (e.g. as a result of a mount(1) ).
1165  *	In the latter case @inode and @filp carry an abridged amount
1166  *	of information as noted above.
1167  *
1168  *	Locking: called with bdev->bd_mutex held.
1169  **/
1170 static int sd_open(struct block_device *bdev, fmode_t mode)
1171 {
1172 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1173 	struct scsi_device *sdev;
1174 	int retval;
1175 
1176 	if (!sdkp)
1177 		return -ENXIO;
1178 
1179 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1180 
1181 	sdev = sdkp->device;
1182 
1183 	/*
1184 	 * If the device is in error recovery, wait until it is done.
1185 	 * If the device is offline, then disallow any access to it.
1186 	 */
1187 	retval = -ENXIO;
1188 	if (!scsi_block_when_processing_errors(sdev))
1189 		goto error_out;
1190 
1191 	if (sdev->removable || sdkp->write_prot)
1192 		check_disk_change(bdev);
1193 
1194 	/*
1195 	 * If the drive is empty, just let the open fail.
1196 	 */
1197 	retval = -ENOMEDIUM;
1198 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1199 		goto error_out;
1200 
1201 	/*
1202 	 * If the device has the write protect tab set, have the open fail
1203 	 * if the user expects to be able to write to the thing.
1204 	 */
1205 	retval = -EROFS;
1206 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1207 		goto error_out;
1208 
1209 	/*
1210 	 * It is possible that the disk changing stuff resulted in
1211 	 * the device being taken offline.  If this is the case,
1212 	 * report this to the user, and don't pretend that the
1213 	 * open actually succeeded.
1214 	 */
1215 	retval = -ENXIO;
1216 	if (!scsi_device_online(sdev))
1217 		goto error_out;
1218 
1219 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1220 		if (scsi_block_when_processing_errors(sdev))
1221 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1222 	}
1223 
1224 	return 0;
1225 
1226 error_out:
1227 	scsi_disk_put(sdkp);
1228 	return retval;
1229 }
1230 
1231 /**
1232  *	sd_release - invoked when the (last) close(2) is called on this
1233  *	scsi disk.
1234  *	@inode: only i_rdev member may be used
1235  *	@filp: only f_mode and f_flags may be used
1236  *
1237  *	Returns 0.
1238  *
1239  *	Note: may block (uninterruptible) if error recovery is underway
1240  *	on this disk.
1241  *
1242  *	Locking: called with bdev->bd_mutex held.
1243  **/
1244 static void sd_release(struct gendisk *disk, fmode_t mode)
1245 {
1246 	struct scsi_disk *sdkp = scsi_disk(disk);
1247 	struct scsi_device *sdev = sdkp->device;
1248 
1249 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1250 
1251 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1252 		if (scsi_block_when_processing_errors(sdev))
1253 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1254 	}
1255 
1256 	/*
1257 	 * XXX and what if there are packets in flight and this close()
1258 	 * XXX is followed by a "rmmod sd_mod"?
1259 	 */
1260 
1261 	scsi_disk_put(sdkp);
1262 }
1263 
1264 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1265 {
1266 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1267 	struct scsi_device *sdp = sdkp->device;
1268 	struct Scsi_Host *host = sdp->host;
1269 	int diskinfo[4];
1270 
1271 	/* default to most commonly used values */
1272         diskinfo[0] = 0x40;	/* 1 << 6 */
1273        	diskinfo[1] = 0x20;	/* 1 << 5 */
1274        	diskinfo[2] = sdkp->capacity >> 11;
1275 
1276 	/* override with calculated, extended default, or driver values */
1277 	if (host->hostt->bios_param)
1278 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1279 	else
1280 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1281 
1282 	geo->heads = diskinfo[0];
1283 	geo->sectors = diskinfo[1];
1284 	geo->cylinders = diskinfo[2];
1285 	return 0;
1286 }
1287 
1288 /**
1289  *	sd_ioctl - process an ioctl
1290  *	@inode: only i_rdev/i_bdev members may be used
1291  *	@filp: only f_mode and f_flags may be used
1292  *	@cmd: ioctl command number
1293  *	@arg: this is third argument given to ioctl(2) system call.
1294  *	Often contains a pointer.
1295  *
1296  *	Returns 0 if successful (some ioctls return positive numbers on
1297  *	success as well). Returns a negated errno value in case of error.
1298  *
1299  *	Note: most ioctls are forward onto the block subsystem or further
1300  *	down in the scsi subsystem.
1301  **/
1302 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1303 		    unsigned int cmd, unsigned long arg)
1304 {
1305 	struct gendisk *disk = bdev->bd_disk;
1306 	struct scsi_disk *sdkp = scsi_disk(disk);
1307 	struct scsi_device *sdp = sdkp->device;
1308 	void __user *p = (void __user *)arg;
1309 	int error;
1310 
1311 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1312 				    "cmd=0x%x\n", disk->disk_name, cmd));
1313 
1314 	error = scsi_verify_blk_ioctl(bdev, cmd);
1315 	if (error < 0)
1316 		return error;
1317 
1318 	/*
1319 	 * If we are in the middle of error recovery, don't let anyone
1320 	 * else try and use this device.  Also, if error recovery fails, it
1321 	 * may try and take the device offline, in which case all further
1322 	 * access to the device is prohibited.
1323 	 */
1324 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1325 					(mode & FMODE_NDELAY) != 0);
1326 	if (!scsi_block_when_processing_errors(sdp) || !error)
1327 		goto out;
1328 
1329 	/*
1330 	 * Send SCSI addressing ioctls directly to mid level, send other
1331 	 * ioctls to block level and then onto mid level if they can't be
1332 	 * resolved.
1333 	 */
1334 	switch (cmd) {
1335 		case SCSI_IOCTL_GET_IDLUN:
1336 		case SCSI_IOCTL_GET_BUS_NUMBER:
1337 			error = scsi_ioctl(sdp, cmd, p);
1338 			break;
1339 		default:
1340 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1341 			if (error != -ENOTTY)
1342 				break;
1343 			error = scsi_ioctl(sdp, cmd, p);
1344 			break;
1345 	}
1346 out:
1347 	return error;
1348 }
1349 
1350 static void set_media_not_present(struct scsi_disk *sdkp)
1351 {
1352 	if (sdkp->media_present)
1353 		sdkp->device->changed = 1;
1354 
1355 	if (sdkp->device->removable) {
1356 		sdkp->media_present = 0;
1357 		sdkp->capacity = 0;
1358 	}
1359 }
1360 
1361 static int media_not_present(struct scsi_disk *sdkp,
1362 			     struct scsi_sense_hdr *sshdr)
1363 {
1364 	if (!scsi_sense_valid(sshdr))
1365 		return 0;
1366 
1367 	/* not invoked for commands that could return deferred errors */
1368 	switch (sshdr->sense_key) {
1369 	case UNIT_ATTENTION:
1370 	case NOT_READY:
1371 		/* medium not present */
1372 		if (sshdr->asc == 0x3A) {
1373 			set_media_not_present(sdkp);
1374 			return 1;
1375 		}
1376 	}
1377 	return 0;
1378 }
1379 
1380 /**
1381  *	sd_check_events - check media events
1382  *	@disk: kernel device descriptor
1383  *	@clearing: disk events currently being cleared
1384  *
1385  *	Returns mask of DISK_EVENT_*.
1386  *
1387  *	Note: this function is invoked from the block subsystem.
1388  **/
1389 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1390 {
1391 	struct scsi_disk *sdkp = scsi_disk(disk);
1392 	struct scsi_device *sdp = sdkp->device;
1393 	struct scsi_sense_hdr *sshdr = NULL;
1394 	int retval;
1395 
1396 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1397 
1398 	/*
1399 	 * If the device is offline, don't send any commands - just pretend as
1400 	 * if the command failed.  If the device ever comes back online, we
1401 	 * can deal with it then.  It is only because of unrecoverable errors
1402 	 * that we would ever take a device offline in the first place.
1403 	 */
1404 	if (!scsi_device_online(sdp)) {
1405 		set_media_not_present(sdkp);
1406 		goto out;
1407 	}
1408 
1409 	/*
1410 	 * Using TEST_UNIT_READY enables differentiation between drive with
1411 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1412 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1413 	 *
1414 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1415 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1416 	 * sd_revalidate() is called.
1417 	 */
1418 	retval = -ENODEV;
1419 
1420 	if (scsi_block_when_processing_errors(sdp)) {
1421 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1422 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1423 					      sshdr);
1424 	}
1425 
1426 	/* failed to execute TUR, assume media not present */
1427 	if (host_byte(retval)) {
1428 		set_media_not_present(sdkp);
1429 		goto out;
1430 	}
1431 
1432 	if (media_not_present(sdkp, sshdr))
1433 		goto out;
1434 
1435 	/*
1436 	 * For removable scsi disk we have to recognise the presence
1437 	 * of a disk in the drive.
1438 	 */
1439 	if (!sdkp->media_present)
1440 		sdp->changed = 1;
1441 	sdkp->media_present = 1;
1442 out:
1443 	/*
1444 	 * sdp->changed is set under the following conditions:
1445 	 *
1446 	 *	Medium present state has changed in either direction.
1447 	 *	Device has indicated UNIT_ATTENTION.
1448 	 */
1449 	kfree(sshdr);
1450 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1451 	sdp->changed = 0;
1452 	return retval;
1453 }
1454 
1455 static int sd_sync_cache(struct scsi_disk *sdkp)
1456 {
1457 	int retries, res;
1458 	struct scsi_device *sdp = sdkp->device;
1459 	const int timeout = sdp->request_queue->rq_timeout
1460 		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1461 	struct scsi_sense_hdr sshdr;
1462 
1463 	if (!scsi_device_online(sdp))
1464 		return -ENODEV;
1465 
1466 	for (retries = 3; retries > 0; --retries) {
1467 		unsigned char cmd[10] = { 0 };
1468 
1469 		cmd[0] = SYNCHRONIZE_CACHE;
1470 		/*
1471 		 * Leave the rest of the command zero to indicate
1472 		 * flush everything.
1473 		 */
1474 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1475 					     &sshdr, timeout, SD_MAX_RETRIES,
1476 					     NULL, REQ_PM);
1477 		if (res == 0)
1478 			break;
1479 	}
1480 
1481 	if (res) {
1482 		sd_print_result(sdkp, res);
1483 
1484 		if (driver_byte(res) & DRIVER_SENSE)
1485 			sd_print_sense_hdr(sdkp, &sshdr);
1486 		/* we need to evaluate the error return  */
1487 		if (scsi_sense_valid(&sshdr) &&
1488 			(sshdr.asc == 0x3a ||	/* medium not present */
1489 			 sshdr.asc == 0x20))	/* invalid command */
1490 				/* this is no error here */
1491 				return 0;
1492 
1493 		switch (host_byte(res)) {
1494 		/* ignore errors due to racing a disconnection */
1495 		case DID_BAD_TARGET:
1496 		case DID_NO_CONNECT:
1497 			return 0;
1498 		/* signal the upper layer it might try again */
1499 		case DID_BUS_BUSY:
1500 		case DID_IMM_RETRY:
1501 		case DID_REQUEUE:
1502 		case DID_SOFT_ERROR:
1503 			return -EBUSY;
1504 		default:
1505 			return -EIO;
1506 		}
1507 	}
1508 	return 0;
1509 }
1510 
1511 static void sd_rescan(struct device *dev)
1512 {
1513 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1514 
1515 	if (sdkp) {
1516 		revalidate_disk(sdkp->disk);
1517 		scsi_disk_put(sdkp);
1518 	}
1519 }
1520 
1521 
1522 #ifdef CONFIG_COMPAT
1523 /*
1524  * This gets directly called from VFS. When the ioctl
1525  * is not recognized we go back to the other translation paths.
1526  */
1527 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1528 			   unsigned int cmd, unsigned long arg)
1529 {
1530 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1531 	int ret;
1532 
1533 	ret = scsi_verify_blk_ioctl(bdev, cmd);
1534 	if (ret < 0)
1535 		return ret;
1536 
1537 	/*
1538 	 * If we are in the middle of error recovery, don't let anyone
1539 	 * else try and use this device.  Also, if error recovery fails, it
1540 	 * may try and take the device offline, in which case all further
1541 	 * access to the device is prohibited.
1542 	 */
1543 	if (!scsi_block_when_processing_errors(sdev))
1544 		return -ENODEV;
1545 
1546 	if (sdev->host->hostt->compat_ioctl) {
1547 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1548 
1549 		return ret;
1550 	}
1551 
1552 	/*
1553 	 * Let the static ioctl translation table take care of it.
1554 	 */
1555 	return -ENOIOCTLCMD;
1556 }
1557 #endif
1558 
1559 static const struct block_device_operations sd_fops = {
1560 	.owner			= THIS_MODULE,
1561 	.open			= sd_open,
1562 	.release		= sd_release,
1563 	.ioctl			= sd_ioctl,
1564 	.getgeo			= sd_getgeo,
1565 #ifdef CONFIG_COMPAT
1566 	.compat_ioctl		= sd_compat_ioctl,
1567 #endif
1568 	.check_events		= sd_check_events,
1569 	.revalidate_disk	= sd_revalidate_disk,
1570 	.unlock_native_capacity	= sd_unlock_native_capacity,
1571 };
1572 
1573 /**
1574  *	sd_eh_action - error handling callback
1575  *	@scmd:		sd-issued command that has failed
1576  *	@eh_disp:	The recovery disposition suggested by the midlayer
1577  *
1578  *	This function is called by the SCSI midlayer upon completion of an
1579  *	error test command (currently TEST UNIT READY). The result of sending
1580  *	the eh command is passed in eh_disp.  We're looking for devices that
1581  *	fail medium access commands but are OK with non access commands like
1582  *	test unit ready (so wrongly see the device as having a successful
1583  *	recovery)
1584  **/
1585 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1586 {
1587 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1588 
1589 	if (!scsi_device_online(scmd->device) ||
1590 	    !scsi_medium_access_command(scmd) ||
1591 	    host_byte(scmd->result) != DID_TIME_OUT ||
1592 	    eh_disp != SUCCESS)
1593 		return eh_disp;
1594 
1595 	/*
1596 	 * The device has timed out executing a medium access command.
1597 	 * However, the TEST UNIT READY command sent during error
1598 	 * handling completed successfully. Either the device is in the
1599 	 * process of recovering or has it suffered an internal failure
1600 	 * that prevents access to the storage medium.
1601 	 */
1602 	sdkp->medium_access_timed_out++;
1603 
1604 	/*
1605 	 * If the device keeps failing read/write commands but TEST UNIT
1606 	 * READY always completes successfully we assume that medium
1607 	 * access is no longer possible and take the device offline.
1608 	 */
1609 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1610 		scmd_printk(KERN_ERR, scmd,
1611 			    "Medium access timeout failure. Offlining disk!\n");
1612 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1613 
1614 		return FAILED;
1615 	}
1616 
1617 	return eh_disp;
1618 }
1619 
1620 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1621 {
1622 	u64 start_lba = blk_rq_pos(scmd->request);
1623 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1624 	u64 bad_lba;
1625 	int info_valid;
1626 	/*
1627 	 * resid is optional but mostly filled in.  When it's unused,
1628 	 * its value is zero, so we assume the whole buffer transferred
1629 	 */
1630 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1631 	unsigned int good_bytes;
1632 
1633 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1634 		return 0;
1635 
1636 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1637 					     SCSI_SENSE_BUFFERSIZE,
1638 					     &bad_lba);
1639 	if (!info_valid)
1640 		return 0;
1641 
1642 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1643 		return 0;
1644 
1645 	if (scmd->device->sector_size < 512) {
1646 		/* only legitimate sector_size here is 256 */
1647 		start_lba <<= 1;
1648 		end_lba <<= 1;
1649 	} else {
1650 		/* be careful ... don't want any overflows */
1651 		unsigned int factor = scmd->device->sector_size / 512;
1652 		do_div(start_lba, factor);
1653 		do_div(end_lba, factor);
1654 	}
1655 
1656 	/* The bad lba was reported incorrectly, we have no idea where
1657 	 * the error is.
1658 	 */
1659 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1660 		return 0;
1661 
1662 	/* This computation should always be done in terms of
1663 	 * the resolution of the device's medium.
1664 	 */
1665 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1666 	return min(good_bytes, transferred);
1667 }
1668 
1669 /**
1670  *	sd_done - bottom half handler: called when the lower level
1671  *	driver has completed (successfully or otherwise) a scsi command.
1672  *	@SCpnt: mid-level's per command structure.
1673  *
1674  *	Note: potentially run from within an ISR. Must not block.
1675  **/
1676 static int sd_done(struct scsi_cmnd *SCpnt)
1677 {
1678 	int result = SCpnt->result;
1679 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1680 	struct scsi_sense_hdr sshdr;
1681 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1682 	struct request *req = SCpnt->request;
1683 	int sense_valid = 0;
1684 	int sense_deferred = 0;
1685 	unsigned char op = SCpnt->cmnd[0];
1686 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1687 
1688 	if (req->cmd_flags & REQ_DISCARD || req->cmd_flags & REQ_WRITE_SAME) {
1689 		if (!result) {
1690 			good_bytes = blk_rq_bytes(req);
1691 			scsi_set_resid(SCpnt, 0);
1692 		} else {
1693 			good_bytes = 0;
1694 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1695 		}
1696 	}
1697 
1698 	if (result) {
1699 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1700 		if (sense_valid)
1701 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1702 	}
1703 #ifdef CONFIG_SCSI_LOGGING
1704 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1705 	if (sense_valid) {
1706 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1707 						   "sd_done: sb[respc,sk,asc,"
1708 						   "ascq]=%x,%x,%x,%x\n",
1709 						   sshdr.response_code,
1710 						   sshdr.sense_key, sshdr.asc,
1711 						   sshdr.ascq));
1712 	}
1713 #endif
1714 	sdkp->medium_access_timed_out = 0;
1715 
1716 	if (driver_byte(result) != DRIVER_SENSE &&
1717 	    (!sense_valid || sense_deferred))
1718 		goto out;
1719 
1720 	switch (sshdr.sense_key) {
1721 	case HARDWARE_ERROR:
1722 	case MEDIUM_ERROR:
1723 		good_bytes = sd_completed_bytes(SCpnt);
1724 		break;
1725 	case RECOVERED_ERROR:
1726 		good_bytes = scsi_bufflen(SCpnt);
1727 		break;
1728 	case NO_SENSE:
1729 		/* This indicates a false check condition, so ignore it.  An
1730 		 * unknown amount of data was transferred so treat it as an
1731 		 * error.
1732 		 */
1733 		scsi_print_sense("sd", SCpnt);
1734 		SCpnt->result = 0;
1735 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1736 		break;
1737 	case ABORTED_COMMAND:
1738 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1739 			good_bytes = sd_completed_bytes(SCpnt);
1740 		break;
1741 	case ILLEGAL_REQUEST:
1742 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1743 			good_bytes = sd_completed_bytes(SCpnt);
1744 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1745 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1746 			switch (op) {
1747 			case UNMAP:
1748 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1749 				break;
1750 			case WRITE_SAME_16:
1751 			case WRITE_SAME:
1752 				if (unmap)
1753 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1754 				else {
1755 					sdkp->device->no_write_same = 1;
1756 					sd_config_write_same(sdkp);
1757 
1758 					good_bytes = 0;
1759 					req->__data_len = blk_rq_bytes(req);
1760 					req->cmd_flags |= REQ_QUIET;
1761 				}
1762 			}
1763 		}
1764 		break;
1765 	default:
1766 		break;
1767 	}
1768  out:
1769 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1770 		sd_dif_complete(SCpnt, good_bytes);
1771 
1772 	return good_bytes;
1773 }
1774 
1775 /*
1776  * spinup disk - called only in sd_revalidate_disk()
1777  */
1778 static void
1779 sd_spinup_disk(struct scsi_disk *sdkp)
1780 {
1781 	unsigned char cmd[10];
1782 	unsigned long spintime_expire = 0;
1783 	int retries, spintime;
1784 	unsigned int the_result;
1785 	struct scsi_sense_hdr sshdr;
1786 	int sense_valid = 0;
1787 
1788 	spintime = 0;
1789 
1790 	/* Spin up drives, as required.  Only do this at boot time */
1791 	/* Spinup needs to be done for module loads too. */
1792 	do {
1793 		retries = 0;
1794 
1795 		do {
1796 			cmd[0] = TEST_UNIT_READY;
1797 			memset((void *) &cmd[1], 0, 9);
1798 
1799 			the_result = scsi_execute_req(sdkp->device, cmd,
1800 						      DMA_NONE, NULL, 0,
1801 						      &sshdr, SD_TIMEOUT,
1802 						      SD_MAX_RETRIES, NULL);
1803 
1804 			/*
1805 			 * If the drive has indicated to us that it
1806 			 * doesn't have any media in it, don't bother
1807 			 * with any more polling.
1808 			 */
1809 			if (media_not_present(sdkp, &sshdr))
1810 				return;
1811 
1812 			if (the_result)
1813 				sense_valid = scsi_sense_valid(&sshdr);
1814 			retries++;
1815 		} while (retries < 3 &&
1816 			 (!scsi_status_is_good(the_result) ||
1817 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1818 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1819 
1820 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1821 			/* no sense, TUR either succeeded or failed
1822 			 * with a status error */
1823 			if(!spintime && !scsi_status_is_good(the_result)) {
1824 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1825 				sd_print_result(sdkp, the_result);
1826 			}
1827 			break;
1828 		}
1829 
1830 		/*
1831 		 * The device does not want the automatic start to be issued.
1832 		 */
1833 		if (sdkp->device->no_start_on_add)
1834 			break;
1835 
1836 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1837 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1838 				break;	/* manual intervention required */
1839 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1840 				break;	/* standby */
1841 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1842 				break;	/* unavailable */
1843 			/*
1844 			 * Issue command to spin up drive when not ready
1845 			 */
1846 			if (!spintime) {
1847 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1848 				cmd[0] = START_STOP;
1849 				cmd[1] = 1;	/* Return immediately */
1850 				memset((void *) &cmd[2], 0, 8);
1851 				cmd[4] = 1;	/* Start spin cycle */
1852 				if (sdkp->device->start_stop_pwr_cond)
1853 					cmd[4] |= 1 << 4;
1854 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1855 						 NULL, 0, &sshdr,
1856 						 SD_TIMEOUT, SD_MAX_RETRIES,
1857 						 NULL);
1858 				spintime_expire = jiffies + 100 * HZ;
1859 				spintime = 1;
1860 			}
1861 			/* Wait 1 second for next try */
1862 			msleep(1000);
1863 			printk(".");
1864 
1865 		/*
1866 		 * Wait for USB flash devices with slow firmware.
1867 		 * Yes, this sense key/ASC combination shouldn't
1868 		 * occur here.  It's characteristic of these devices.
1869 		 */
1870 		} else if (sense_valid &&
1871 				sshdr.sense_key == UNIT_ATTENTION &&
1872 				sshdr.asc == 0x28) {
1873 			if (!spintime) {
1874 				spintime_expire = jiffies + 5 * HZ;
1875 				spintime = 1;
1876 			}
1877 			/* Wait 1 second for next try */
1878 			msleep(1000);
1879 		} else {
1880 			/* we don't understand the sense code, so it's
1881 			 * probably pointless to loop */
1882 			if(!spintime) {
1883 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1884 				sd_print_sense_hdr(sdkp, &sshdr);
1885 			}
1886 			break;
1887 		}
1888 
1889 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1890 
1891 	if (spintime) {
1892 		if (scsi_status_is_good(the_result))
1893 			printk("ready\n");
1894 		else
1895 			printk("not responding...\n");
1896 	}
1897 }
1898 
1899 
1900 /*
1901  * Determine whether disk supports Data Integrity Field.
1902  */
1903 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1904 {
1905 	struct scsi_device *sdp = sdkp->device;
1906 	u8 type;
1907 	int ret = 0;
1908 
1909 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1910 		return ret;
1911 
1912 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1913 
1914 	if (type > SD_DIF_TYPE3_PROTECTION)
1915 		ret = -ENODEV;
1916 	else if (scsi_host_dif_capable(sdp->host, type))
1917 		ret = 1;
1918 
1919 	if (sdkp->first_scan || type != sdkp->protection_type)
1920 		switch (ret) {
1921 		case -ENODEV:
1922 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
1923 				  " protection type %u. Disabling disk!\n",
1924 				  type);
1925 			break;
1926 		case 1:
1927 			sd_printk(KERN_NOTICE, sdkp,
1928 				  "Enabling DIF Type %u protection\n", type);
1929 			break;
1930 		case 0:
1931 			sd_printk(KERN_NOTICE, sdkp,
1932 				  "Disabling DIF Type %u protection\n", type);
1933 			break;
1934 		}
1935 
1936 	sdkp->protection_type = type;
1937 
1938 	return ret;
1939 }
1940 
1941 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1942 			struct scsi_sense_hdr *sshdr, int sense_valid,
1943 			int the_result)
1944 {
1945 	sd_print_result(sdkp, the_result);
1946 	if (driver_byte(the_result) & DRIVER_SENSE)
1947 		sd_print_sense_hdr(sdkp, sshdr);
1948 	else
1949 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1950 
1951 	/*
1952 	 * Set dirty bit for removable devices if not ready -
1953 	 * sometimes drives will not report this properly.
1954 	 */
1955 	if (sdp->removable &&
1956 	    sense_valid && sshdr->sense_key == NOT_READY)
1957 		set_media_not_present(sdkp);
1958 
1959 	/*
1960 	 * We used to set media_present to 0 here to indicate no media
1961 	 * in the drive, but some drives fail read capacity even with
1962 	 * media present, so we can't do that.
1963 	 */
1964 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1965 }
1966 
1967 #define RC16_LEN 32
1968 #if RC16_LEN > SD_BUF_SIZE
1969 #error RC16_LEN must not be more than SD_BUF_SIZE
1970 #endif
1971 
1972 #define READ_CAPACITY_RETRIES_ON_RESET	10
1973 
1974 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1975 						unsigned char *buffer)
1976 {
1977 	unsigned char cmd[16];
1978 	struct scsi_sense_hdr sshdr;
1979 	int sense_valid = 0;
1980 	int the_result;
1981 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1982 	unsigned int alignment;
1983 	unsigned long long lba;
1984 	unsigned sector_size;
1985 
1986 	if (sdp->no_read_capacity_16)
1987 		return -EINVAL;
1988 
1989 	do {
1990 		memset(cmd, 0, 16);
1991 		cmd[0] = SERVICE_ACTION_IN;
1992 		cmd[1] = SAI_READ_CAPACITY_16;
1993 		cmd[13] = RC16_LEN;
1994 		memset(buffer, 0, RC16_LEN);
1995 
1996 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1997 					buffer, RC16_LEN, &sshdr,
1998 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1999 
2000 		if (media_not_present(sdkp, &sshdr))
2001 			return -ENODEV;
2002 
2003 		if (the_result) {
2004 			sense_valid = scsi_sense_valid(&sshdr);
2005 			if (sense_valid &&
2006 			    sshdr.sense_key == ILLEGAL_REQUEST &&
2007 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2008 			    sshdr.ascq == 0x00)
2009 				/* Invalid Command Operation Code or
2010 				 * Invalid Field in CDB, just retry
2011 				 * silently with RC10 */
2012 				return -EINVAL;
2013 			if (sense_valid &&
2014 			    sshdr.sense_key == UNIT_ATTENTION &&
2015 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2016 				/* Device reset might occur several times,
2017 				 * give it one more chance */
2018 				if (--reset_retries > 0)
2019 					continue;
2020 		}
2021 		retries--;
2022 
2023 	} while (the_result && retries);
2024 
2025 	if (the_result) {
2026 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
2027 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2028 		return -EINVAL;
2029 	}
2030 
2031 	sector_size = get_unaligned_be32(&buffer[8]);
2032 	lba = get_unaligned_be64(&buffer[0]);
2033 
2034 	if (sd_read_protection_type(sdkp, buffer) < 0) {
2035 		sdkp->capacity = 0;
2036 		return -ENODEV;
2037 	}
2038 
2039 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2040 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2041 			"kernel compiled with support for large block "
2042 			"devices.\n");
2043 		sdkp->capacity = 0;
2044 		return -EOVERFLOW;
2045 	}
2046 
2047 	/* Logical blocks per physical block exponent */
2048 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2049 
2050 	/* Lowest aligned logical block */
2051 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2052 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2053 	if (alignment && sdkp->first_scan)
2054 		sd_printk(KERN_NOTICE, sdkp,
2055 			  "physical block alignment offset: %u\n", alignment);
2056 
2057 	if (buffer[14] & 0x80) { /* LBPME */
2058 		sdkp->lbpme = 1;
2059 
2060 		if (buffer[14] & 0x40) /* LBPRZ */
2061 			sdkp->lbprz = 1;
2062 
2063 		sd_config_discard(sdkp, SD_LBP_WS16);
2064 	}
2065 
2066 	sdkp->capacity = lba + 1;
2067 	return sector_size;
2068 }
2069 
2070 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2071 						unsigned char *buffer)
2072 {
2073 	unsigned char cmd[16];
2074 	struct scsi_sense_hdr sshdr;
2075 	int sense_valid = 0;
2076 	int the_result;
2077 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2078 	sector_t lba;
2079 	unsigned sector_size;
2080 
2081 	do {
2082 		cmd[0] = READ_CAPACITY;
2083 		memset(&cmd[1], 0, 9);
2084 		memset(buffer, 0, 8);
2085 
2086 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2087 					buffer, 8, &sshdr,
2088 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2089 
2090 		if (media_not_present(sdkp, &sshdr))
2091 			return -ENODEV;
2092 
2093 		if (the_result) {
2094 			sense_valid = scsi_sense_valid(&sshdr);
2095 			if (sense_valid &&
2096 			    sshdr.sense_key == UNIT_ATTENTION &&
2097 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2098 				/* Device reset might occur several times,
2099 				 * give it one more chance */
2100 				if (--reset_retries > 0)
2101 					continue;
2102 		}
2103 		retries--;
2104 
2105 	} while (the_result && retries);
2106 
2107 	if (the_result) {
2108 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
2109 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2110 		return -EINVAL;
2111 	}
2112 
2113 	sector_size = get_unaligned_be32(&buffer[4]);
2114 	lba = get_unaligned_be32(&buffer[0]);
2115 
2116 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2117 		/* Some buggy (usb cardreader) devices return an lba of
2118 		   0xffffffff when the want to report a size of 0 (with
2119 		   which they really mean no media is present) */
2120 		sdkp->capacity = 0;
2121 		sdkp->physical_block_size = sector_size;
2122 		return sector_size;
2123 	}
2124 
2125 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
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 	sdkp->capacity = lba + 1;
2134 	sdkp->physical_block_size = sector_size;
2135 	return sector_size;
2136 }
2137 
2138 static int sd_try_rc16_first(struct scsi_device *sdp)
2139 {
2140 	if (sdp->host->max_cmd_len < 16)
2141 		return 0;
2142 	if (sdp->try_rc_10_first)
2143 		return 0;
2144 	if (sdp->scsi_level > SCSI_SPC_2)
2145 		return 1;
2146 	if (scsi_device_protection(sdp))
2147 		return 1;
2148 	return 0;
2149 }
2150 
2151 /*
2152  * read disk capacity
2153  */
2154 static void
2155 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2156 {
2157 	int sector_size;
2158 	struct scsi_device *sdp = sdkp->device;
2159 	sector_t old_capacity = sdkp->capacity;
2160 
2161 	if (sd_try_rc16_first(sdp)) {
2162 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2163 		if (sector_size == -EOVERFLOW)
2164 			goto got_data;
2165 		if (sector_size == -ENODEV)
2166 			return;
2167 		if (sector_size < 0)
2168 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2169 		if (sector_size < 0)
2170 			return;
2171 	} else {
2172 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2173 		if (sector_size == -EOVERFLOW)
2174 			goto got_data;
2175 		if (sector_size < 0)
2176 			return;
2177 		if ((sizeof(sdkp->capacity) > 4) &&
2178 		    (sdkp->capacity > 0xffffffffULL)) {
2179 			int old_sector_size = sector_size;
2180 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2181 					"Trying to use READ CAPACITY(16).\n");
2182 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2183 			if (sector_size < 0) {
2184 				sd_printk(KERN_NOTICE, sdkp,
2185 					"Using 0xffffffff as device size\n");
2186 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2187 				sector_size = old_sector_size;
2188 				goto got_data;
2189 			}
2190 		}
2191 	}
2192 
2193 	/* Some devices are known to return the total number of blocks,
2194 	 * not the highest block number.  Some devices have versions
2195 	 * which do this and others which do not.  Some devices we might
2196 	 * suspect of doing this but we don't know for certain.
2197 	 *
2198 	 * If we know the reported capacity is wrong, decrement it.  If
2199 	 * we can only guess, then assume the number of blocks is even
2200 	 * (usually true but not always) and err on the side of lowering
2201 	 * the capacity.
2202 	 */
2203 	if (sdp->fix_capacity ||
2204 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2205 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2206 				"from its reported value: %llu\n",
2207 				(unsigned long long) sdkp->capacity);
2208 		--sdkp->capacity;
2209 	}
2210 
2211 got_data:
2212 	if (sector_size == 0) {
2213 		sector_size = 512;
2214 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2215 			  "assuming 512.\n");
2216 	}
2217 
2218 	if (sector_size != 512 &&
2219 	    sector_size != 1024 &&
2220 	    sector_size != 2048 &&
2221 	    sector_size != 4096 &&
2222 	    sector_size != 256) {
2223 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2224 			  sector_size);
2225 		/*
2226 		 * The user might want to re-format the drive with
2227 		 * a supported sectorsize.  Once this happens, it
2228 		 * would be relatively trivial to set the thing up.
2229 		 * For this reason, we leave the thing in the table.
2230 		 */
2231 		sdkp->capacity = 0;
2232 		/*
2233 		 * set a bogus sector size so the normal read/write
2234 		 * logic in the block layer will eventually refuse any
2235 		 * request on this device without tripping over power
2236 		 * of two sector size assumptions
2237 		 */
2238 		sector_size = 512;
2239 	}
2240 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2241 
2242 	{
2243 		char cap_str_2[10], cap_str_10[10];
2244 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
2245 
2246 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
2247 				sizeof(cap_str_2));
2248 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
2249 				sizeof(cap_str_10));
2250 
2251 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2252 			sd_printk(KERN_NOTICE, sdkp,
2253 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2254 				  (unsigned long long)sdkp->capacity,
2255 				  sector_size, cap_str_10, cap_str_2);
2256 
2257 			if (sdkp->physical_block_size != sector_size)
2258 				sd_printk(KERN_NOTICE, sdkp,
2259 					  "%u-byte physical blocks\n",
2260 					  sdkp->physical_block_size);
2261 		}
2262 	}
2263 
2264 	if (sdkp->capacity > 0xffffffff) {
2265 		sdp->use_16_for_rw = 1;
2266 		sdkp->max_xfer_blocks = SD_MAX_XFER_BLOCKS;
2267 	} else
2268 		sdkp->max_xfer_blocks = SD_DEF_XFER_BLOCKS;
2269 
2270 	/* Rescale capacity to 512-byte units */
2271 	if (sector_size == 4096)
2272 		sdkp->capacity <<= 3;
2273 	else if (sector_size == 2048)
2274 		sdkp->capacity <<= 2;
2275 	else if (sector_size == 1024)
2276 		sdkp->capacity <<= 1;
2277 	else if (sector_size == 256)
2278 		sdkp->capacity >>= 1;
2279 
2280 	blk_queue_physical_block_size(sdp->request_queue,
2281 				      sdkp->physical_block_size);
2282 	sdkp->device->sector_size = sector_size;
2283 }
2284 
2285 /* called with buffer of length 512 */
2286 static inline int
2287 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2288 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2289 		 struct scsi_sense_hdr *sshdr)
2290 {
2291 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2292 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2293 			       sshdr);
2294 }
2295 
2296 /*
2297  * read write protect setting, if possible - called only in sd_revalidate_disk()
2298  * called with buffer of length SD_BUF_SIZE
2299  */
2300 static void
2301 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2302 {
2303 	int res;
2304 	struct scsi_device *sdp = sdkp->device;
2305 	struct scsi_mode_data data;
2306 	int old_wp = sdkp->write_prot;
2307 
2308 	set_disk_ro(sdkp->disk, 0);
2309 	if (sdp->skip_ms_page_3f) {
2310 		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2311 		return;
2312 	}
2313 
2314 	if (sdp->use_192_bytes_for_3f) {
2315 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2316 	} else {
2317 		/*
2318 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2319 		 * We have to start carefully: some devices hang if we ask
2320 		 * for more than is available.
2321 		 */
2322 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2323 
2324 		/*
2325 		 * Second attempt: ask for page 0 When only page 0 is
2326 		 * implemented, a request for page 3F may return Sense Key
2327 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2328 		 * CDB.
2329 		 */
2330 		if (!scsi_status_is_good(res))
2331 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2332 
2333 		/*
2334 		 * Third attempt: ask 255 bytes, as we did earlier.
2335 		 */
2336 		if (!scsi_status_is_good(res))
2337 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2338 					       &data, NULL);
2339 	}
2340 
2341 	if (!scsi_status_is_good(res)) {
2342 		sd_first_printk(KERN_WARNING, sdkp,
2343 			  "Test WP failed, assume Write Enabled\n");
2344 	} else {
2345 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2346 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2347 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2348 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2349 				  sdkp->write_prot ? "on" : "off");
2350 			sd_printk(KERN_DEBUG, sdkp,
2351 				  "Mode Sense: %02x %02x %02x %02x\n",
2352 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2353 		}
2354 	}
2355 }
2356 
2357 /*
2358  * sd_read_cache_type - called only from sd_revalidate_disk()
2359  * called with buffer of length SD_BUF_SIZE
2360  */
2361 static void
2362 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2363 {
2364 	int len = 0, res;
2365 	struct scsi_device *sdp = sdkp->device;
2366 
2367 	int dbd;
2368 	int modepage;
2369 	int first_len;
2370 	struct scsi_mode_data data;
2371 	struct scsi_sense_hdr sshdr;
2372 	int old_wce = sdkp->WCE;
2373 	int old_rcd = sdkp->RCD;
2374 	int old_dpofua = sdkp->DPOFUA;
2375 
2376 
2377 	if (sdkp->cache_override)
2378 		return;
2379 
2380 	first_len = 4;
2381 	if (sdp->skip_ms_page_8) {
2382 		if (sdp->type == TYPE_RBC)
2383 			goto defaults;
2384 		else {
2385 			if (sdp->skip_ms_page_3f)
2386 				goto defaults;
2387 			modepage = 0x3F;
2388 			if (sdp->use_192_bytes_for_3f)
2389 				first_len = 192;
2390 			dbd = 0;
2391 		}
2392 	} else if (sdp->type == TYPE_RBC) {
2393 		modepage = 6;
2394 		dbd = 8;
2395 	} else {
2396 		modepage = 8;
2397 		dbd = 0;
2398 	}
2399 
2400 	/* cautiously ask */
2401 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2402 			&data, &sshdr);
2403 
2404 	if (!scsi_status_is_good(res))
2405 		goto bad_sense;
2406 
2407 	if (!data.header_length) {
2408 		modepage = 6;
2409 		first_len = 0;
2410 		sd_first_printk(KERN_ERR, sdkp,
2411 				"Missing header in MODE_SENSE response\n");
2412 	}
2413 
2414 	/* that went OK, now ask for the proper length */
2415 	len = data.length;
2416 
2417 	/*
2418 	 * We're only interested in the first three bytes, actually.
2419 	 * But the data cache page is defined for the first 20.
2420 	 */
2421 	if (len < 3)
2422 		goto bad_sense;
2423 	else if (len > SD_BUF_SIZE) {
2424 		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2425 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2426 		len = SD_BUF_SIZE;
2427 	}
2428 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2429 		len = 192;
2430 
2431 	/* Get the data */
2432 	if (len > first_len)
2433 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2434 				&data, &sshdr);
2435 
2436 	if (scsi_status_is_good(res)) {
2437 		int offset = data.header_length + data.block_descriptor_length;
2438 
2439 		while (offset < len) {
2440 			u8 page_code = buffer[offset] & 0x3F;
2441 			u8 spf       = buffer[offset] & 0x40;
2442 
2443 			if (page_code == 8 || page_code == 6) {
2444 				/* We're interested only in the first 3 bytes.
2445 				 */
2446 				if (len - offset <= 2) {
2447 					sd_first_printk(KERN_ERR, sdkp,
2448 						"Incomplete mode parameter "
2449 							"data\n");
2450 					goto defaults;
2451 				} else {
2452 					modepage = page_code;
2453 					goto Page_found;
2454 				}
2455 			} else {
2456 				/* Go to the next page */
2457 				if (spf && len - offset > 3)
2458 					offset += 4 + (buffer[offset+2] << 8) +
2459 						buffer[offset+3];
2460 				else if (!spf && len - offset > 1)
2461 					offset += 2 + buffer[offset+1];
2462 				else {
2463 					sd_first_printk(KERN_ERR, sdkp,
2464 							"Incomplete mode "
2465 							"parameter data\n");
2466 					goto defaults;
2467 				}
2468 			}
2469 		}
2470 
2471 		sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2472 		goto defaults;
2473 
2474 	Page_found:
2475 		if (modepage == 8) {
2476 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2477 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2478 		} else {
2479 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2480 			sdkp->RCD = 0;
2481 		}
2482 
2483 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2484 		if (sdp->broken_fua) {
2485 			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2486 			sdkp->DPOFUA = 0;
2487 		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2488 			sd_first_printk(KERN_NOTICE, sdkp,
2489 				  "Uses READ/WRITE(6), disabling FUA\n");
2490 			sdkp->DPOFUA = 0;
2491 		}
2492 
2493 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2494 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2495 			sd_printk(KERN_NOTICE, sdkp,
2496 				  "Write cache: %s, read cache: %s, %s\n",
2497 				  sdkp->WCE ? "enabled" : "disabled",
2498 				  sdkp->RCD ? "disabled" : "enabled",
2499 				  sdkp->DPOFUA ? "supports DPO and FUA"
2500 				  : "doesn't support DPO or FUA");
2501 
2502 		return;
2503 	}
2504 
2505 bad_sense:
2506 	if (scsi_sense_valid(&sshdr) &&
2507 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2508 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2509 		/* Invalid field in CDB */
2510 		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2511 	else
2512 		sd_first_printk(KERN_ERR, sdkp,
2513 				"Asking for cache data failed\n");
2514 
2515 defaults:
2516 	if (sdp->wce_default_on) {
2517 		sd_first_printk(KERN_NOTICE, sdkp,
2518 				"Assuming drive cache: write back\n");
2519 		sdkp->WCE = 1;
2520 	} else {
2521 		sd_first_printk(KERN_ERR, sdkp,
2522 				"Assuming drive cache: write through\n");
2523 		sdkp->WCE = 0;
2524 	}
2525 	sdkp->RCD = 0;
2526 	sdkp->DPOFUA = 0;
2527 }
2528 
2529 /*
2530  * The ATO bit indicates whether the DIF application tag is available
2531  * for use by the operating system.
2532  */
2533 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2534 {
2535 	int res, offset;
2536 	struct scsi_device *sdp = sdkp->device;
2537 	struct scsi_mode_data data;
2538 	struct scsi_sense_hdr sshdr;
2539 
2540 	if (sdp->type != TYPE_DISK)
2541 		return;
2542 
2543 	if (sdkp->protection_type == 0)
2544 		return;
2545 
2546 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2547 			      SD_MAX_RETRIES, &data, &sshdr);
2548 
2549 	if (!scsi_status_is_good(res) || !data.header_length ||
2550 	    data.length < 6) {
2551 		sd_first_printk(KERN_WARNING, sdkp,
2552 			  "getting Control mode page failed, assume no ATO\n");
2553 
2554 		if (scsi_sense_valid(&sshdr))
2555 			sd_print_sense_hdr(sdkp, &sshdr);
2556 
2557 		return;
2558 	}
2559 
2560 	offset = data.header_length + data.block_descriptor_length;
2561 
2562 	if ((buffer[offset] & 0x3f) != 0x0a) {
2563 		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2564 		return;
2565 	}
2566 
2567 	if ((buffer[offset + 5] & 0x80) == 0)
2568 		return;
2569 
2570 	sdkp->ATO = 1;
2571 
2572 	return;
2573 }
2574 
2575 /**
2576  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2577  * @disk: disk to query
2578  */
2579 static void sd_read_block_limits(struct scsi_disk *sdkp)
2580 {
2581 	unsigned int sector_sz = sdkp->device->sector_size;
2582 	const int vpd_len = 64;
2583 	u32 max_xfer_length;
2584 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2585 
2586 	if (!buffer ||
2587 	    /* Block Limits VPD */
2588 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2589 		goto out;
2590 
2591 	max_xfer_length = get_unaligned_be32(&buffer[8]);
2592 	if (max_xfer_length)
2593 		sdkp->max_xfer_blocks = max_xfer_length;
2594 
2595 	blk_queue_io_min(sdkp->disk->queue,
2596 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2597 	blk_queue_io_opt(sdkp->disk->queue,
2598 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2599 
2600 	if (buffer[3] == 0x3c) {
2601 		unsigned int lba_count, desc_count;
2602 
2603 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2604 
2605 		if (!sdkp->lbpme)
2606 			goto out;
2607 
2608 		lba_count = get_unaligned_be32(&buffer[20]);
2609 		desc_count = get_unaligned_be32(&buffer[24]);
2610 
2611 		if (lba_count && desc_count)
2612 			sdkp->max_unmap_blocks = lba_count;
2613 
2614 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2615 
2616 		if (buffer[32] & 0x80)
2617 			sdkp->unmap_alignment =
2618 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2619 
2620 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2621 
2622 			if (sdkp->max_unmap_blocks)
2623 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2624 			else
2625 				sd_config_discard(sdkp, SD_LBP_WS16);
2626 
2627 		} else {	/* LBP VPD page tells us what to use */
2628 
2629 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2630 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2631 			else if (sdkp->lbpws)
2632 				sd_config_discard(sdkp, SD_LBP_WS16);
2633 			else if (sdkp->lbpws10)
2634 				sd_config_discard(sdkp, SD_LBP_WS10);
2635 			else
2636 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2637 		}
2638 	}
2639 
2640  out:
2641 	kfree(buffer);
2642 }
2643 
2644 /**
2645  * sd_read_block_characteristics - Query block dev. characteristics
2646  * @disk: disk to query
2647  */
2648 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2649 {
2650 	unsigned char *buffer;
2651 	u16 rot;
2652 	const int vpd_len = 64;
2653 
2654 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2655 
2656 	if (!buffer ||
2657 	    /* Block Device Characteristics VPD */
2658 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2659 		goto out;
2660 
2661 	rot = get_unaligned_be16(&buffer[4]);
2662 
2663 	if (rot == 1)
2664 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2665 
2666  out:
2667 	kfree(buffer);
2668 }
2669 
2670 /**
2671  * sd_read_block_provisioning - Query provisioning VPD page
2672  * @disk: disk to query
2673  */
2674 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2675 {
2676 	unsigned char *buffer;
2677 	const int vpd_len = 8;
2678 
2679 	if (sdkp->lbpme == 0)
2680 		return;
2681 
2682 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2683 
2684 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2685 		goto out;
2686 
2687 	sdkp->lbpvpd	= 1;
2688 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2689 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2690 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2691 
2692  out:
2693 	kfree(buffer);
2694 }
2695 
2696 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2697 {
2698 	struct scsi_device *sdev = sdkp->device;
2699 
2700 	if (sdev->host->no_write_same) {
2701 		sdev->no_write_same = 1;
2702 
2703 		return;
2704 	}
2705 
2706 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2707 		/* too large values might cause issues with arcmsr */
2708 		int vpd_buf_len = 64;
2709 
2710 		sdev->no_report_opcodes = 1;
2711 
2712 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2713 		 * CODES is unsupported and the device has an ATA
2714 		 * Information VPD page (SAT).
2715 		 */
2716 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2717 			sdev->no_write_same = 1;
2718 	}
2719 
2720 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2721 		sdkp->ws16 = 1;
2722 
2723 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2724 		sdkp->ws10 = 1;
2725 }
2726 
2727 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2728 {
2729 	/* Attempt VPD inquiry if the device blacklist explicitly calls
2730 	 * for it.
2731 	 */
2732 	if (sdp->try_vpd_pages)
2733 		return 1;
2734 	/*
2735 	 * Although VPD inquiries can go to SCSI-2 type devices,
2736 	 * some USB ones crash on receiving them, and the pages
2737 	 * we currently ask for are for SPC-3 and beyond
2738 	 */
2739 	if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2740 		return 1;
2741 	return 0;
2742 }
2743 
2744 /**
2745  *	sd_revalidate_disk - called the first time a new disk is seen,
2746  *	performs disk spin up, read_capacity, etc.
2747  *	@disk: struct gendisk we care about
2748  **/
2749 static int sd_revalidate_disk(struct gendisk *disk)
2750 {
2751 	struct scsi_disk *sdkp = scsi_disk(disk);
2752 	struct scsi_device *sdp = sdkp->device;
2753 	unsigned char *buffer;
2754 	unsigned int max_xfer;
2755 
2756 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2757 				      "sd_revalidate_disk\n"));
2758 
2759 	/*
2760 	 * If the device is offline, don't try and read capacity or any
2761 	 * of the other niceties.
2762 	 */
2763 	if (!scsi_device_online(sdp))
2764 		goto out;
2765 
2766 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2767 	if (!buffer) {
2768 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2769 			  "allocation failure.\n");
2770 		goto out;
2771 	}
2772 
2773 	sd_spinup_disk(sdkp);
2774 
2775 	/*
2776 	 * Without media there is no reason to ask; moreover, some devices
2777 	 * react badly if we do.
2778 	 */
2779 	if (sdkp->media_present) {
2780 		sd_read_capacity(sdkp, buffer);
2781 
2782 		if (sd_try_extended_inquiry(sdp)) {
2783 			sd_read_block_provisioning(sdkp);
2784 			sd_read_block_limits(sdkp);
2785 			sd_read_block_characteristics(sdkp);
2786 		}
2787 
2788 		sd_read_write_protect_flag(sdkp, buffer);
2789 		sd_read_cache_type(sdkp, buffer);
2790 		sd_read_app_tag_own(sdkp, buffer);
2791 		sd_read_write_same(sdkp, buffer);
2792 	}
2793 
2794 	sdkp->first_scan = 0;
2795 
2796 	/*
2797 	 * We now have all cache related info, determine how we deal
2798 	 * with flush requests.
2799 	 */
2800 	sd_set_flush_flag(sdkp);
2801 
2802 	max_xfer = min_not_zero(queue_max_hw_sectors(sdkp->disk->queue),
2803 				sdkp->max_xfer_blocks);
2804 	max_xfer <<= ilog2(sdp->sector_size) - 9;
2805 	blk_queue_max_hw_sectors(sdkp->disk->queue, max_xfer);
2806 	set_capacity(disk, sdkp->capacity);
2807 	sd_config_write_same(sdkp);
2808 	kfree(buffer);
2809 
2810  out:
2811 	return 0;
2812 }
2813 
2814 /**
2815  *	sd_unlock_native_capacity - unlock native capacity
2816  *	@disk: struct gendisk to set capacity for
2817  *
2818  *	Block layer calls this function if it detects that partitions
2819  *	on @disk reach beyond the end of the device.  If the SCSI host
2820  *	implements ->unlock_native_capacity() method, it's invoked to
2821  *	give it a chance to adjust the device capacity.
2822  *
2823  *	CONTEXT:
2824  *	Defined by block layer.  Might sleep.
2825  */
2826 static void sd_unlock_native_capacity(struct gendisk *disk)
2827 {
2828 	struct scsi_device *sdev = scsi_disk(disk)->device;
2829 
2830 	if (sdev->host->hostt->unlock_native_capacity)
2831 		sdev->host->hostt->unlock_native_capacity(sdev);
2832 }
2833 
2834 /**
2835  *	sd_format_disk_name - format disk name
2836  *	@prefix: name prefix - ie. "sd" for SCSI disks
2837  *	@index: index of the disk to format name for
2838  *	@buf: output buffer
2839  *	@buflen: length of the output buffer
2840  *
2841  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2842  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2843  *	which is followed by sdaaa.
2844  *
2845  *	This is basically 26 base counting with one extra 'nil' entry
2846  *	at the beginning from the second digit on and can be
2847  *	determined using similar method as 26 base conversion with the
2848  *	index shifted -1 after each digit is computed.
2849  *
2850  *	CONTEXT:
2851  *	Don't care.
2852  *
2853  *	RETURNS:
2854  *	0 on success, -errno on failure.
2855  */
2856 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2857 {
2858 	const int base = 'z' - 'a' + 1;
2859 	char *begin = buf + strlen(prefix);
2860 	char *end = buf + buflen;
2861 	char *p;
2862 	int unit;
2863 
2864 	p = end - 1;
2865 	*p = '\0';
2866 	unit = base;
2867 	do {
2868 		if (p == begin)
2869 			return -EINVAL;
2870 		*--p = 'a' + (index % unit);
2871 		index = (index / unit) - 1;
2872 	} while (index >= 0);
2873 
2874 	memmove(begin, p, end - p);
2875 	memcpy(buf, prefix, strlen(prefix));
2876 
2877 	return 0;
2878 }
2879 
2880 /*
2881  * The asynchronous part of sd_probe
2882  */
2883 static void sd_probe_async(void *data, async_cookie_t cookie)
2884 {
2885 	struct scsi_disk *sdkp = data;
2886 	struct scsi_device *sdp;
2887 	struct gendisk *gd;
2888 	u32 index;
2889 	struct device *dev;
2890 
2891 	sdp = sdkp->device;
2892 	gd = sdkp->disk;
2893 	index = sdkp->index;
2894 	dev = &sdp->sdev_gendev;
2895 
2896 	gd->major = sd_major((index & 0xf0) >> 4);
2897 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2898 	gd->minors = SD_MINORS;
2899 
2900 	gd->fops = &sd_fops;
2901 	gd->private_data = &sdkp->driver;
2902 	gd->queue = sdkp->device->request_queue;
2903 
2904 	/* defaults, until the device tells us otherwise */
2905 	sdp->sector_size = 512;
2906 	sdkp->capacity = 0;
2907 	sdkp->media_present = 1;
2908 	sdkp->write_prot = 0;
2909 	sdkp->cache_override = 0;
2910 	sdkp->WCE = 0;
2911 	sdkp->RCD = 0;
2912 	sdkp->ATO = 0;
2913 	sdkp->first_scan = 1;
2914 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2915 
2916 	sd_revalidate_disk(gd);
2917 
2918 	gd->driverfs_dev = &sdp->sdev_gendev;
2919 	gd->flags = GENHD_FL_EXT_DEVT;
2920 	if (sdp->removable) {
2921 		gd->flags |= GENHD_FL_REMOVABLE;
2922 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2923 	}
2924 
2925 	blk_pm_runtime_init(sdp->request_queue, dev);
2926 	add_disk(gd);
2927 	if (sdkp->capacity)
2928 		sd_dif_config_host(sdkp);
2929 
2930 	sd_revalidate_disk(gd);
2931 
2932 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2933 		  sdp->removable ? "removable " : "");
2934 	scsi_autopm_put_device(sdp);
2935 	put_device(&sdkp->dev);
2936 }
2937 
2938 /**
2939  *	sd_probe - called during driver initialization and whenever a
2940  *	new scsi device is attached to the system. It is called once
2941  *	for each scsi device (not just disks) present.
2942  *	@dev: pointer to device object
2943  *
2944  *	Returns 0 if successful (or not interested in this scsi device
2945  *	(e.g. scanner)); 1 when there is an error.
2946  *
2947  *	Note: this function is invoked from the scsi mid-level.
2948  *	This function sets up the mapping between a given
2949  *	<host,channel,id,lun> (found in sdp) and new device name
2950  *	(e.g. /dev/sda). More precisely it is the block device major
2951  *	and minor number that is chosen here.
2952  *
2953  *	Assume sd_probe is not re-entrant (for time being)
2954  *	Also think about sd_probe() and sd_remove() running coincidentally.
2955  **/
2956 static int sd_probe(struct device *dev)
2957 {
2958 	struct scsi_device *sdp = to_scsi_device(dev);
2959 	struct scsi_disk *sdkp;
2960 	struct gendisk *gd;
2961 	int index;
2962 	int error;
2963 
2964 	error = -ENODEV;
2965 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2966 		goto out;
2967 
2968 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2969 					"sd_probe\n"));
2970 
2971 	error = -ENOMEM;
2972 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2973 	if (!sdkp)
2974 		goto out;
2975 
2976 	gd = alloc_disk(SD_MINORS);
2977 	if (!gd)
2978 		goto out_free;
2979 
2980 	do {
2981 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2982 			goto out_put;
2983 
2984 		spin_lock(&sd_index_lock);
2985 		error = ida_get_new(&sd_index_ida, &index);
2986 		spin_unlock(&sd_index_lock);
2987 	} while (error == -EAGAIN);
2988 
2989 	if (error) {
2990 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2991 		goto out_put;
2992 	}
2993 
2994 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2995 	if (error) {
2996 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2997 		goto out_free_index;
2998 	}
2999 
3000 	sdkp->device = sdp;
3001 	sdkp->driver = &sd_template;
3002 	sdkp->disk = gd;
3003 	sdkp->index = index;
3004 	atomic_set(&sdkp->openers, 0);
3005 	atomic_set(&sdkp->device->ioerr_cnt, 0);
3006 
3007 	if (!sdp->request_queue->rq_timeout) {
3008 		if (sdp->type != TYPE_MOD)
3009 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3010 		else
3011 			blk_queue_rq_timeout(sdp->request_queue,
3012 					     SD_MOD_TIMEOUT);
3013 	}
3014 
3015 	device_initialize(&sdkp->dev);
3016 	sdkp->dev.parent = dev;
3017 	sdkp->dev.class = &sd_disk_class;
3018 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3019 
3020 	if (device_add(&sdkp->dev))
3021 		goto out_free_index;
3022 
3023 	get_device(dev);
3024 	dev_set_drvdata(dev, sdkp);
3025 
3026 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
3027 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3028 
3029 	return 0;
3030 
3031  out_free_index:
3032 	spin_lock(&sd_index_lock);
3033 	ida_remove(&sd_index_ida, index);
3034 	spin_unlock(&sd_index_lock);
3035  out_put:
3036 	put_disk(gd);
3037  out_free:
3038 	kfree(sdkp);
3039  out:
3040 	return error;
3041 }
3042 
3043 /**
3044  *	sd_remove - called whenever a scsi disk (previously recognized by
3045  *	sd_probe) is detached from the system. It is called (potentially
3046  *	multiple times) during sd module unload.
3047  *	@sdp: pointer to mid level scsi device object
3048  *
3049  *	Note: this function is invoked from the scsi mid-level.
3050  *	This function potentially frees up a device name (e.g. /dev/sdc)
3051  *	that could be re-used by a subsequent sd_probe().
3052  *	This function is not called when the built-in sd driver is "exit-ed".
3053  **/
3054 static int sd_remove(struct device *dev)
3055 {
3056 	struct scsi_disk *sdkp;
3057 	dev_t devt;
3058 
3059 	sdkp = dev_get_drvdata(dev);
3060 	devt = disk_devt(sdkp->disk);
3061 	scsi_autopm_get_device(sdkp->device);
3062 
3063 	async_synchronize_full_domain(&scsi_sd_pm_domain);
3064 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3065 	device_del(&sdkp->dev);
3066 	del_gendisk(sdkp->disk);
3067 	sd_shutdown(dev);
3068 
3069 	blk_register_region(devt, SD_MINORS, NULL,
3070 			    sd_default_probe, NULL, NULL);
3071 
3072 	mutex_lock(&sd_ref_mutex);
3073 	dev_set_drvdata(dev, NULL);
3074 	put_device(&sdkp->dev);
3075 	mutex_unlock(&sd_ref_mutex);
3076 
3077 	return 0;
3078 }
3079 
3080 /**
3081  *	scsi_disk_release - Called to free the scsi_disk structure
3082  *	@dev: pointer to embedded class device
3083  *
3084  *	sd_ref_mutex must be held entering this routine.  Because it is
3085  *	called on last put, you should always use the scsi_disk_get()
3086  *	scsi_disk_put() helpers which manipulate the semaphore directly
3087  *	and never do a direct put_device.
3088  **/
3089 static void scsi_disk_release(struct device *dev)
3090 {
3091 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3092 	struct gendisk *disk = sdkp->disk;
3093 
3094 	spin_lock(&sd_index_lock);
3095 	ida_remove(&sd_index_ida, sdkp->index);
3096 	spin_unlock(&sd_index_lock);
3097 
3098 	disk->private_data = NULL;
3099 	put_disk(disk);
3100 	put_device(&sdkp->device->sdev_gendev);
3101 
3102 	kfree(sdkp);
3103 }
3104 
3105 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3106 {
3107 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3108 	struct scsi_sense_hdr sshdr;
3109 	struct scsi_device *sdp = sdkp->device;
3110 	int res;
3111 
3112 	if (start)
3113 		cmd[4] |= 1;	/* START */
3114 
3115 	if (sdp->start_stop_pwr_cond)
3116 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3117 
3118 	if (!scsi_device_online(sdp))
3119 		return -ENODEV;
3120 
3121 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3122 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3123 	if (res) {
3124 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
3125 		sd_print_result(sdkp, res);
3126 		if (driver_byte(res) & DRIVER_SENSE)
3127 			sd_print_sense_hdr(sdkp, &sshdr);
3128 		if (scsi_sense_valid(&sshdr) &&
3129 			/* 0x3a is medium not present */
3130 			sshdr.asc == 0x3a)
3131 			res = 0;
3132 	}
3133 
3134 	/* SCSI error codes must not go to the generic layer */
3135 	if (res)
3136 		return -EIO;
3137 
3138 	return 0;
3139 }
3140 
3141 /*
3142  * Send a SYNCHRONIZE CACHE instruction down to the device through
3143  * the normal SCSI command structure.  Wait for the command to
3144  * complete.
3145  */
3146 static void sd_shutdown(struct device *dev)
3147 {
3148 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3149 
3150 	if (!sdkp)
3151 		return;         /* this can happen */
3152 
3153 	if (pm_runtime_suspended(dev))
3154 		goto exit;
3155 
3156 	if (sdkp->WCE && sdkp->media_present) {
3157 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3158 		sd_sync_cache(sdkp);
3159 	}
3160 
3161 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3162 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3163 		sd_start_stop_device(sdkp, 0);
3164 	}
3165 
3166 exit:
3167 	scsi_disk_put(sdkp);
3168 }
3169 
3170 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3171 {
3172 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3173 	int ret = 0;
3174 
3175 	if (!sdkp)
3176 		return 0;	/* this can happen */
3177 
3178 	if (sdkp->WCE && sdkp->media_present) {
3179 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3180 		ret = sd_sync_cache(sdkp);
3181 		if (ret) {
3182 			/* ignore OFFLINE device */
3183 			if (ret == -ENODEV)
3184 				ret = 0;
3185 			goto done;
3186 		}
3187 	}
3188 
3189 	if (sdkp->device->manage_start_stop) {
3190 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3191 		/* an error is not worth aborting a system sleep */
3192 		ret = sd_start_stop_device(sdkp, 0);
3193 		if (ignore_stop_errors)
3194 			ret = 0;
3195 	}
3196 
3197 done:
3198 	scsi_disk_put(sdkp);
3199 	return ret;
3200 }
3201 
3202 static int sd_suspend_system(struct device *dev)
3203 {
3204 	return sd_suspend_common(dev, true);
3205 }
3206 
3207 static int sd_suspend_runtime(struct device *dev)
3208 {
3209 	return sd_suspend_common(dev, false);
3210 }
3211 
3212 static int sd_resume(struct device *dev)
3213 {
3214 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3215 	int ret = 0;
3216 
3217 	if (!sdkp->device->manage_start_stop)
3218 		goto done;
3219 
3220 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3221 	ret = sd_start_stop_device(sdkp, 1);
3222 
3223 done:
3224 	scsi_disk_put(sdkp);
3225 	return ret;
3226 }
3227 
3228 /**
3229  *	init_sd - entry point for this driver (both when built in or when
3230  *	a module).
3231  *
3232  *	Note: this function registers this driver with the scsi mid-level.
3233  **/
3234 static int __init init_sd(void)
3235 {
3236 	int majors = 0, i, err;
3237 
3238 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3239 
3240 	for (i = 0; i < SD_MAJORS; i++) {
3241 		if (register_blkdev(sd_major(i), "sd") != 0)
3242 			continue;
3243 		majors++;
3244 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3245 				    sd_default_probe, NULL, NULL);
3246 	}
3247 
3248 	if (!majors)
3249 		return -ENODEV;
3250 
3251 	err = class_register(&sd_disk_class);
3252 	if (err)
3253 		goto err_out;
3254 
3255 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3256 					 0, 0, NULL);
3257 	if (!sd_cdb_cache) {
3258 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3259 		err = -ENOMEM;
3260 		goto err_out_class;
3261 	}
3262 
3263 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3264 	if (!sd_cdb_pool) {
3265 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3266 		err = -ENOMEM;
3267 		goto err_out_cache;
3268 	}
3269 
3270 	err = scsi_register_driver(&sd_template.gendrv);
3271 	if (err)
3272 		goto err_out_driver;
3273 
3274 	return 0;
3275 
3276 err_out_driver:
3277 	mempool_destroy(sd_cdb_pool);
3278 
3279 err_out_cache:
3280 	kmem_cache_destroy(sd_cdb_cache);
3281 
3282 err_out_class:
3283 	class_unregister(&sd_disk_class);
3284 err_out:
3285 	for (i = 0; i < SD_MAJORS; i++)
3286 		unregister_blkdev(sd_major(i), "sd");
3287 	return err;
3288 }
3289 
3290 /**
3291  *	exit_sd - exit point for this driver (when it is a module).
3292  *
3293  *	Note: this function unregisters this driver from the scsi mid-level.
3294  **/
3295 static void __exit exit_sd(void)
3296 {
3297 	int i;
3298 
3299 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3300 
3301 	scsi_unregister_driver(&sd_template.gendrv);
3302 	mempool_destroy(sd_cdb_pool);
3303 	kmem_cache_destroy(sd_cdb_cache);
3304 
3305 	class_unregister(&sd_disk_class);
3306 
3307 	for (i = 0; i < SD_MAJORS; i++) {
3308 		blk_unregister_region(sd_major(i), SD_MINORS);
3309 		unregister_blkdev(sd_major(i), "sd");
3310 	}
3311 }
3312 
3313 module_init(init_sd);
3314 module_exit(exit_sd);
3315 
3316 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3317 			       struct scsi_sense_hdr *sshdr)
3318 {
3319 	sd_printk(KERN_INFO, sdkp, " ");
3320 	scsi_show_sense_hdr(sshdr);
3321 	sd_printk(KERN_INFO, sdkp, " ");
3322 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
3323 }
3324 
3325 static void sd_print_result(struct scsi_disk *sdkp, int result)
3326 {
3327 	sd_printk(KERN_INFO, sdkp, " ");
3328 	scsi_show_result(result);
3329 }
3330 
3331