xref: /linux/drivers/scsi/sd.c (revision 31a1b26f16e822577def5402ffc79cfe4aed2db9)
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(struct device *);
109 static int sd_resume(struct device *);
110 static void sd_rescan(struct device *);
111 static int sd_done(struct scsi_cmnd *);
112 static int sd_eh_action(struct scsi_cmnd *, unsigned char *, int, int);
113 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
114 static void scsi_disk_release(struct device *cdev);
115 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
116 static void sd_print_result(struct scsi_disk *, int);
117 
118 static DEFINE_SPINLOCK(sd_index_lock);
119 static DEFINE_IDA(sd_index_ida);
120 
121 /* This semaphore is used to mediate the 0->1 reference get in the
122  * face of object destruction (i.e. we can't allow a get on an
123  * object after last put) */
124 static DEFINE_MUTEX(sd_ref_mutex);
125 
126 static struct kmem_cache *sd_cdb_cache;
127 static mempool_t *sd_cdb_pool;
128 
129 static const char *sd_cache_types[] = {
130 	"write through", "none", "write back",
131 	"write back, no read (daft)"
132 };
133 
134 static ssize_t
135 sd_store_cache_type(struct device *dev, struct device_attribute *attr,
136 		    const char *buf, size_t count)
137 {
138 	int i, ct = -1, rcd, wce, sp;
139 	struct scsi_disk *sdkp = to_scsi_disk(dev);
140 	struct scsi_device *sdp = sdkp->device;
141 	char buffer[64];
142 	char *buffer_data;
143 	struct scsi_mode_data data;
144 	struct scsi_sense_hdr sshdr;
145 	static const char temp[] = "temporary ";
146 	int len;
147 
148 	if (sdp->type != TYPE_DISK)
149 		/* no cache control on RBC devices; theoretically they
150 		 * can do it, but there's probably so many exceptions
151 		 * it's not worth the risk */
152 		return -EINVAL;
153 
154 	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
155 		buf += sizeof(temp) - 1;
156 		sdkp->cache_override = 1;
157 	} else {
158 		sdkp->cache_override = 0;
159 	}
160 
161 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
162 		len = strlen(sd_cache_types[i]);
163 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
164 		    buf[len] == '\n') {
165 			ct = i;
166 			break;
167 		}
168 	}
169 	if (ct < 0)
170 		return -EINVAL;
171 	rcd = ct & 0x01 ? 1 : 0;
172 	wce = ct & 0x02 ? 1 : 0;
173 
174 	if (sdkp->cache_override) {
175 		sdkp->WCE = wce;
176 		sdkp->RCD = rcd;
177 		return count;
178 	}
179 
180 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
181 			    SD_MAX_RETRIES, &data, NULL))
182 		return -EINVAL;
183 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
184 		  data.block_descriptor_length);
185 	buffer_data = buffer + data.header_length +
186 		data.block_descriptor_length;
187 	buffer_data[2] &= ~0x05;
188 	buffer_data[2] |= wce << 2 | rcd;
189 	sp = buffer_data[0] & 0x80 ? 1 : 0;
190 
191 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
192 			     SD_MAX_RETRIES, &data, &sshdr)) {
193 		if (scsi_sense_valid(&sshdr))
194 			sd_print_sense_hdr(sdkp, &sshdr);
195 		return -EINVAL;
196 	}
197 	revalidate_disk(sdkp->disk);
198 	return count;
199 }
200 
201 static ssize_t
202 sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
203 			   const char *buf, size_t count)
204 {
205 	struct scsi_disk *sdkp = to_scsi_disk(dev);
206 	struct scsi_device *sdp = sdkp->device;
207 
208 	if (!capable(CAP_SYS_ADMIN))
209 		return -EACCES;
210 
211 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
212 
213 	return count;
214 }
215 
216 static ssize_t
217 sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
218 		       const char *buf, size_t count)
219 {
220 	struct scsi_disk *sdkp = to_scsi_disk(dev);
221 	struct scsi_device *sdp = sdkp->device;
222 
223 	if (!capable(CAP_SYS_ADMIN))
224 		return -EACCES;
225 
226 	if (sdp->type != TYPE_DISK)
227 		return -EINVAL;
228 
229 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
230 
231 	return count;
232 }
233 
234 static ssize_t
235 sd_show_cache_type(struct device *dev, struct device_attribute *attr,
236 		   char *buf)
237 {
238 	struct scsi_disk *sdkp = to_scsi_disk(dev);
239 	int ct = sdkp->RCD + 2*sdkp->WCE;
240 
241 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
242 }
243 
244 static ssize_t
245 sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
246 {
247 	struct scsi_disk *sdkp = to_scsi_disk(dev);
248 
249 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
250 }
251 
252 static ssize_t
253 sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
254 			  char *buf)
255 {
256 	struct scsi_disk *sdkp = to_scsi_disk(dev);
257 	struct scsi_device *sdp = sdkp->device;
258 
259 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
260 }
261 
262 static ssize_t
263 sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
264 		      char *buf)
265 {
266 	struct scsi_disk *sdkp = to_scsi_disk(dev);
267 
268 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
269 }
270 
271 static ssize_t
272 sd_show_protection_type(struct device *dev, struct device_attribute *attr,
273 			char *buf)
274 {
275 	struct scsi_disk *sdkp = to_scsi_disk(dev);
276 
277 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
278 }
279 
280 static ssize_t
281 sd_store_protection_type(struct device *dev, struct device_attribute *attr,
282 			 const char *buf, size_t count)
283 {
284 	struct scsi_disk *sdkp = to_scsi_disk(dev);
285 	unsigned int val;
286 	int err;
287 
288 	if (!capable(CAP_SYS_ADMIN))
289 		return -EACCES;
290 
291 	err = kstrtouint(buf, 10, &val);
292 
293 	if (err)
294 		return err;
295 
296 	if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
297 		sdkp->protection_type = val;
298 
299 	return count;
300 }
301 
302 static ssize_t
303 sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
304 			char *buf)
305 {
306 	struct scsi_disk *sdkp = to_scsi_disk(dev);
307 	struct scsi_device *sdp = sdkp->device;
308 	unsigned int dif, dix;
309 
310 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
311 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
312 
313 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
314 		dif = 0;
315 		dix = 1;
316 	}
317 
318 	if (!dif && !dix)
319 		return snprintf(buf, 20, "none\n");
320 
321 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
322 }
323 
324 static ssize_t
325 sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
326 		    char *buf)
327 {
328 	struct scsi_disk *sdkp = to_scsi_disk(dev);
329 
330 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
331 }
332 
333 static ssize_t
334 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
335 			  char *buf)
336 {
337 	struct scsi_disk *sdkp = to_scsi_disk(dev);
338 
339 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
340 }
341 
342 static const char *lbp_mode[] = {
343 	[SD_LBP_FULL]		= "full",
344 	[SD_LBP_UNMAP]		= "unmap",
345 	[SD_LBP_WS16]		= "writesame_16",
346 	[SD_LBP_WS10]		= "writesame_10",
347 	[SD_LBP_ZERO]		= "writesame_zero",
348 	[SD_LBP_DISABLE]	= "disabled",
349 };
350 
351 static ssize_t
352 sd_show_provisioning_mode(struct device *dev, struct device_attribute *attr,
353 			  char *buf)
354 {
355 	struct scsi_disk *sdkp = to_scsi_disk(dev);
356 
357 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
358 }
359 
360 static ssize_t
361 sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
362 			   const char *buf, size_t count)
363 {
364 	struct scsi_disk *sdkp = to_scsi_disk(dev);
365 	struct scsi_device *sdp = sdkp->device;
366 
367 	if (!capable(CAP_SYS_ADMIN))
368 		return -EACCES;
369 
370 	if (sdp->type != TYPE_DISK)
371 		return -EINVAL;
372 
373 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
374 		sd_config_discard(sdkp, SD_LBP_UNMAP);
375 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
376 		sd_config_discard(sdkp, SD_LBP_WS16);
377 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
378 		sd_config_discard(sdkp, SD_LBP_WS10);
379 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
380 		sd_config_discard(sdkp, SD_LBP_ZERO);
381 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
382 		sd_config_discard(sdkp, SD_LBP_DISABLE);
383 	else
384 		return -EINVAL;
385 
386 	return count;
387 }
388 
389 static ssize_t
390 sd_show_max_medium_access_timeouts(struct device *dev,
391 				   struct device_attribute *attr, char *buf)
392 {
393 	struct scsi_disk *sdkp = to_scsi_disk(dev);
394 
395 	return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
396 }
397 
398 static ssize_t
399 sd_store_max_medium_access_timeouts(struct device *dev,
400 				    struct device_attribute *attr,
401 				    const char *buf, size_t count)
402 {
403 	struct scsi_disk *sdkp = to_scsi_disk(dev);
404 	int err;
405 
406 	if (!capable(CAP_SYS_ADMIN))
407 		return -EACCES;
408 
409 	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
410 
411 	return err ? err : count;
412 }
413 
414 static ssize_t
415 sd_show_write_same_blocks(struct device *dev, struct device_attribute *attr,
416 			  char *buf)
417 {
418 	struct scsi_disk *sdkp = to_scsi_disk(dev);
419 
420 	return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
421 }
422 
423 static ssize_t
424 sd_store_write_same_blocks(struct device *dev, struct device_attribute *attr,
425 			   const char *buf, size_t count)
426 {
427 	struct scsi_disk *sdkp = to_scsi_disk(dev);
428 	struct scsi_device *sdp = sdkp->device;
429 	unsigned long max;
430 	int err;
431 
432 	if (!capable(CAP_SYS_ADMIN))
433 		return -EACCES;
434 
435 	if (sdp->type != TYPE_DISK)
436 		return -EINVAL;
437 
438 	err = kstrtoul(buf, 10, &max);
439 
440 	if (err)
441 		return err;
442 
443 	if (max == 0)
444 		sdp->no_write_same = 1;
445 	else if (max <= SD_MAX_WS16_BLOCKS) {
446 		sdp->no_write_same = 0;
447 		sdkp->max_ws_blocks = max;
448 	}
449 
450 	sd_config_write_same(sdkp);
451 
452 	return count;
453 }
454 
455 static struct device_attribute sd_disk_attrs[] = {
456 	__ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
457 	       sd_store_cache_type),
458 	__ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
459 	__ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
460 	       sd_store_allow_restart),
461 	__ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
462 	       sd_store_manage_start_stop),
463 	__ATTR(protection_type, S_IRUGO|S_IWUSR, sd_show_protection_type,
464 	       sd_store_protection_type),
465 	__ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
466 	__ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
467 	__ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
468 	__ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
469 	       sd_store_provisioning_mode),
470 	__ATTR(max_write_same_blocks, S_IRUGO|S_IWUSR,
471 	       sd_show_write_same_blocks, sd_store_write_same_blocks),
472 	__ATTR(max_medium_access_timeouts, S_IRUGO|S_IWUSR,
473 	       sd_show_max_medium_access_timeouts,
474 	       sd_store_max_medium_access_timeouts),
475 	__ATTR_NULL,
476 };
477 
478 static struct class sd_disk_class = {
479 	.name		= "scsi_disk",
480 	.owner		= THIS_MODULE,
481 	.dev_release	= scsi_disk_release,
482 	.dev_attrs	= sd_disk_attrs,
483 };
484 
485 static const struct dev_pm_ops sd_pm_ops = {
486 	.suspend		= sd_suspend,
487 	.resume			= sd_resume,
488 	.poweroff		= sd_suspend,
489 	.restore		= sd_resume,
490 	.runtime_suspend	= sd_suspend,
491 	.runtime_resume		= sd_resume,
492 };
493 
494 static struct scsi_driver sd_template = {
495 	.owner			= THIS_MODULE,
496 	.gendrv = {
497 		.name		= "sd",
498 		.probe		= sd_probe,
499 		.remove		= sd_remove,
500 		.shutdown	= sd_shutdown,
501 		.pm		= &sd_pm_ops,
502 	},
503 	.rescan			= sd_rescan,
504 	.done			= sd_done,
505 	.eh_action		= sd_eh_action,
506 };
507 
508 /*
509  * Dummy kobj_map->probe function.
510  * The default ->probe function will call modprobe, which is
511  * pointless as this module is already loaded.
512  */
513 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
514 {
515 	return NULL;
516 }
517 
518 /*
519  * Device no to disk mapping:
520  *
521  *       major         disc2     disc  p1
522  *   |............|.............|....|....| <- dev_t
523  *    31        20 19          8 7  4 3  0
524  *
525  * Inside a major, we have 16k disks, however mapped non-
526  * contiguously. The first 16 disks are for major0, the next
527  * ones with major1, ... Disk 256 is for major0 again, disk 272
528  * for major1, ...
529  * As we stay compatible with our numbering scheme, we can reuse
530  * the well-know SCSI majors 8, 65--71, 136--143.
531  */
532 static int sd_major(int major_idx)
533 {
534 	switch (major_idx) {
535 	case 0:
536 		return SCSI_DISK0_MAJOR;
537 	case 1 ... 7:
538 		return SCSI_DISK1_MAJOR + major_idx - 1;
539 	case 8 ... 15:
540 		return SCSI_DISK8_MAJOR + major_idx - 8;
541 	default:
542 		BUG();
543 		return 0;	/* shut up gcc */
544 	}
545 }
546 
547 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
548 {
549 	struct scsi_disk *sdkp = NULL;
550 
551 	if (disk->private_data) {
552 		sdkp = scsi_disk(disk);
553 		if (scsi_device_get(sdkp->device) == 0)
554 			get_device(&sdkp->dev);
555 		else
556 			sdkp = NULL;
557 	}
558 	return sdkp;
559 }
560 
561 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
562 {
563 	struct scsi_disk *sdkp;
564 
565 	mutex_lock(&sd_ref_mutex);
566 	sdkp = __scsi_disk_get(disk);
567 	mutex_unlock(&sd_ref_mutex);
568 	return sdkp;
569 }
570 
571 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
572 {
573 	struct scsi_disk *sdkp;
574 
575 	mutex_lock(&sd_ref_mutex);
576 	sdkp = dev_get_drvdata(dev);
577 	if (sdkp)
578 		sdkp = __scsi_disk_get(sdkp->disk);
579 	mutex_unlock(&sd_ref_mutex);
580 	return sdkp;
581 }
582 
583 static void scsi_disk_put(struct scsi_disk *sdkp)
584 {
585 	struct scsi_device *sdev = sdkp->device;
586 
587 	mutex_lock(&sd_ref_mutex);
588 	put_device(&sdkp->dev);
589 	scsi_device_put(sdev);
590 	mutex_unlock(&sd_ref_mutex);
591 }
592 
593 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
594 {
595 	unsigned int prot_op = SCSI_PROT_NORMAL;
596 	unsigned int dix = scsi_prot_sg_count(scmd);
597 
598 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
599 		if (dif && dix)
600 			prot_op = SCSI_PROT_READ_PASS;
601 		else if (dif && !dix)
602 			prot_op = SCSI_PROT_READ_STRIP;
603 		else if (!dif && dix)
604 			prot_op = SCSI_PROT_READ_INSERT;
605 	} else {
606 		if (dif && dix)
607 			prot_op = SCSI_PROT_WRITE_PASS;
608 		else if (dif && !dix)
609 			prot_op = SCSI_PROT_WRITE_INSERT;
610 		else if (!dif && dix)
611 			prot_op = SCSI_PROT_WRITE_STRIP;
612 	}
613 
614 	scsi_set_prot_op(scmd, prot_op);
615 	scsi_set_prot_type(scmd, dif);
616 }
617 
618 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
619 {
620 	struct request_queue *q = sdkp->disk->queue;
621 	unsigned int logical_block_size = sdkp->device->sector_size;
622 	unsigned int max_blocks = 0;
623 
624 	q->limits.discard_zeroes_data = sdkp->lbprz;
625 	q->limits.discard_alignment = sdkp->unmap_alignment *
626 		logical_block_size;
627 	q->limits.discard_granularity =
628 		max(sdkp->physical_block_size,
629 		    sdkp->unmap_granularity * logical_block_size);
630 
631 	sdkp->provisioning_mode = mode;
632 
633 	switch (mode) {
634 
635 	case SD_LBP_DISABLE:
636 		q->limits.max_discard_sectors = 0;
637 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
638 		return;
639 
640 	case SD_LBP_UNMAP:
641 		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
642 					  (u32)SD_MAX_WS16_BLOCKS);
643 		break;
644 
645 	case SD_LBP_WS16:
646 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
647 					  (u32)SD_MAX_WS16_BLOCKS);
648 		break;
649 
650 	case SD_LBP_WS10:
651 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
652 					  (u32)SD_MAX_WS10_BLOCKS);
653 		break;
654 
655 	case SD_LBP_ZERO:
656 		max_blocks = min_not_zero(sdkp->max_ws_blocks,
657 					  (u32)SD_MAX_WS10_BLOCKS);
658 		q->limits.discard_zeroes_data = 1;
659 		break;
660 	}
661 
662 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
663 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
664 }
665 
666 /**
667  * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
668  * @sdp: scsi device to operate one
669  * @rq: Request to prepare
670  *
671  * Will issue either UNMAP or WRITE SAME(16) depending on preference
672  * indicated by target device.
673  **/
674 static int sd_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
675 {
676 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
677 	sector_t sector = blk_rq_pos(rq);
678 	unsigned int nr_sectors = blk_rq_sectors(rq);
679 	unsigned int nr_bytes = blk_rq_bytes(rq);
680 	unsigned int len;
681 	int ret;
682 	char *buf;
683 	struct page *page;
684 
685 	sector >>= ilog2(sdp->sector_size) - 9;
686 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
687 	rq->timeout = SD_TIMEOUT;
688 
689 	memset(rq->cmd, 0, rq->cmd_len);
690 
691 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
692 	if (!page)
693 		return BLKPREP_DEFER;
694 
695 	switch (sdkp->provisioning_mode) {
696 	case SD_LBP_UNMAP:
697 		buf = page_address(page);
698 
699 		rq->cmd_len = 10;
700 		rq->cmd[0] = UNMAP;
701 		rq->cmd[8] = 24;
702 
703 		put_unaligned_be16(6 + 16, &buf[0]);
704 		put_unaligned_be16(16, &buf[2]);
705 		put_unaligned_be64(sector, &buf[8]);
706 		put_unaligned_be32(nr_sectors, &buf[16]);
707 
708 		len = 24;
709 		break;
710 
711 	case SD_LBP_WS16:
712 		rq->cmd_len = 16;
713 		rq->cmd[0] = WRITE_SAME_16;
714 		rq->cmd[1] = 0x8; /* UNMAP */
715 		put_unaligned_be64(sector, &rq->cmd[2]);
716 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
717 
718 		len = sdkp->device->sector_size;
719 		break;
720 
721 	case SD_LBP_WS10:
722 	case SD_LBP_ZERO:
723 		rq->cmd_len = 10;
724 		rq->cmd[0] = WRITE_SAME;
725 		if (sdkp->provisioning_mode == SD_LBP_WS10)
726 			rq->cmd[1] = 0x8; /* UNMAP */
727 		put_unaligned_be32(sector, &rq->cmd[2]);
728 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
729 
730 		len = sdkp->device->sector_size;
731 		break;
732 
733 	default:
734 		ret = BLKPREP_KILL;
735 		goto out;
736 	}
737 
738 	blk_add_request_payload(rq, page, len);
739 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
740 	rq->buffer = page_address(page);
741 	rq->__data_len = nr_bytes;
742 
743 out:
744 	if (ret != BLKPREP_OK) {
745 		__free_page(page);
746 		rq->buffer = NULL;
747 	}
748 	return ret;
749 }
750 
751 static void sd_config_write_same(struct scsi_disk *sdkp)
752 {
753 	struct request_queue *q = sdkp->disk->queue;
754 	unsigned int logical_block_size = sdkp->device->sector_size;
755 
756 	if (sdkp->device->no_write_same) {
757 		sdkp->max_ws_blocks = 0;
758 		goto out;
759 	}
760 
761 	/* Some devices can not handle block counts above 0xffff despite
762 	 * supporting WRITE SAME(16). Consequently we default to 64k
763 	 * blocks per I/O unless the device explicitly advertises a
764 	 * bigger limit.
765 	 */
766 	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
767 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
768 						   (u32)SD_MAX_WS16_BLOCKS);
769 	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
770 		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
771 						   (u32)SD_MAX_WS10_BLOCKS);
772 	else {
773 		sdkp->device->no_write_same = 1;
774 		sdkp->max_ws_blocks = 0;
775 	}
776 
777 out:
778 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
779 					 (logical_block_size >> 9));
780 }
781 
782 /**
783  * sd_setup_write_same_cmnd - write the same data to multiple blocks
784  * @sdp: scsi device to operate one
785  * @rq: Request to prepare
786  *
787  * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
788  * preference indicated by target device.
789  **/
790 static int sd_setup_write_same_cmnd(struct scsi_device *sdp, struct request *rq)
791 {
792 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
793 	struct bio *bio = rq->bio;
794 	sector_t sector = blk_rq_pos(rq);
795 	unsigned int nr_sectors = blk_rq_sectors(rq);
796 	unsigned int nr_bytes = blk_rq_bytes(rq);
797 	int ret;
798 
799 	if (sdkp->device->no_write_same)
800 		return BLKPREP_KILL;
801 
802 	BUG_ON(bio_offset(bio) || bio_iovec(bio)->bv_len != sdp->sector_size);
803 
804 	sector >>= ilog2(sdp->sector_size) - 9;
805 	nr_sectors >>= ilog2(sdp->sector_size) - 9;
806 
807 	rq->__data_len = sdp->sector_size;
808 	rq->timeout = SD_WRITE_SAME_TIMEOUT;
809 	memset(rq->cmd, 0, rq->cmd_len);
810 
811 	if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
812 		rq->cmd_len = 16;
813 		rq->cmd[0] = WRITE_SAME_16;
814 		put_unaligned_be64(sector, &rq->cmd[2]);
815 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
816 	} else {
817 		rq->cmd_len = 10;
818 		rq->cmd[0] = WRITE_SAME;
819 		put_unaligned_be32(sector, &rq->cmd[2]);
820 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
821 	}
822 
823 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
824 	rq->__data_len = nr_bytes;
825 
826 	return ret;
827 }
828 
829 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
830 {
831 	rq->timeout = SD_FLUSH_TIMEOUT;
832 	rq->retries = SD_MAX_RETRIES;
833 	rq->cmd[0] = SYNCHRONIZE_CACHE;
834 	rq->cmd_len = 10;
835 
836 	return scsi_setup_blk_pc_cmnd(sdp, rq);
837 }
838 
839 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
840 {
841 	struct scsi_cmnd *SCpnt = rq->special;
842 
843 	if (rq->cmd_flags & REQ_DISCARD) {
844 		free_page((unsigned long)rq->buffer);
845 		rq->buffer = NULL;
846 	}
847 	if (SCpnt->cmnd != rq->cmd) {
848 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
849 		SCpnt->cmnd = NULL;
850 		SCpnt->cmd_len = 0;
851 	}
852 }
853 
854 /**
855  *	sd_prep_fn - build a scsi (read or write) command from
856  *	information in the request structure.
857  *	@SCpnt: pointer to mid-level's per scsi command structure that
858  *	contains request and into which the scsi command is written
859  *
860  *	Returns 1 if successful and 0 if error (or cannot be done now).
861  **/
862 static int sd_prep_fn(struct request_queue *q, struct request *rq)
863 {
864 	struct scsi_cmnd *SCpnt;
865 	struct scsi_device *sdp = q->queuedata;
866 	struct gendisk *disk = rq->rq_disk;
867 	struct scsi_disk *sdkp;
868 	sector_t block = blk_rq_pos(rq);
869 	sector_t threshold;
870 	unsigned int this_count = blk_rq_sectors(rq);
871 	int ret, host_dif;
872 	unsigned char protect;
873 
874 	/*
875 	 * Discard request come in as REQ_TYPE_FS but we turn them into
876 	 * block PC requests to make life easier.
877 	 */
878 	if (rq->cmd_flags & REQ_DISCARD) {
879 		ret = sd_setup_discard_cmnd(sdp, rq);
880 		goto out;
881 	} else if (rq->cmd_flags & REQ_WRITE_SAME) {
882 		ret = sd_setup_write_same_cmnd(sdp, rq);
883 		goto out;
884 	} else if (rq->cmd_flags & REQ_FLUSH) {
885 		ret = scsi_setup_flush_cmnd(sdp, rq);
886 		goto out;
887 	} else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
888 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
889 		goto out;
890 	} else if (rq->cmd_type != REQ_TYPE_FS) {
891 		ret = BLKPREP_KILL;
892 		goto out;
893 	}
894 	ret = scsi_setup_fs_cmnd(sdp, rq);
895 	if (ret != BLKPREP_OK)
896 		goto out;
897 	SCpnt = rq->special;
898 	sdkp = scsi_disk(disk);
899 
900 	/* from here on until we're complete, any goto out
901 	 * is used for a killable error condition */
902 	ret = BLKPREP_KILL;
903 
904 	SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
905 					"sd_prep_fn: block=%llu, "
906 					"count=%d\n",
907 					(unsigned long long)block,
908 					this_count));
909 
910 	if (!sdp || !scsi_device_online(sdp) ||
911 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
912 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
913 						"Finishing %u sectors\n",
914 						blk_rq_sectors(rq)));
915 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
916 						"Retry with 0x%p\n", SCpnt));
917 		goto out;
918 	}
919 
920 	if (sdp->changed) {
921 		/*
922 		 * quietly refuse to do anything to a changed disc until
923 		 * the changed bit has been reset
924 		 */
925 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
926 		goto out;
927 	}
928 
929 	/*
930 	 * Some SD card readers can't handle multi-sector accesses which touch
931 	 * the last one or two hardware sectors.  Split accesses as needed.
932 	 */
933 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
934 		(sdp->sector_size / 512);
935 
936 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
937 		if (block < threshold) {
938 			/* Access up to the threshold but not beyond */
939 			this_count = threshold - block;
940 		} else {
941 			/* Access only a single hardware sector */
942 			this_count = sdp->sector_size / 512;
943 		}
944 	}
945 
946 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
947 					(unsigned long long)block));
948 
949 	/*
950 	 * If we have a 1K hardware sectorsize, prevent access to single
951 	 * 512 byte sectors.  In theory we could handle this - in fact
952 	 * the scsi cdrom driver must be able to handle this because
953 	 * we typically use 1K blocksizes, and cdroms typically have
954 	 * 2K hardware sectorsizes.  Of course, things are simpler
955 	 * with the cdrom, since it is read-only.  For performance
956 	 * reasons, the filesystems should be able to handle this
957 	 * and not force the scsi disk driver to use bounce buffers
958 	 * for this.
959 	 */
960 	if (sdp->sector_size == 1024) {
961 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
962 			scmd_printk(KERN_ERR, SCpnt,
963 				    "Bad block number requested\n");
964 			goto out;
965 		} else {
966 			block = block >> 1;
967 			this_count = this_count >> 1;
968 		}
969 	}
970 	if (sdp->sector_size == 2048) {
971 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
972 			scmd_printk(KERN_ERR, SCpnt,
973 				    "Bad block number requested\n");
974 			goto out;
975 		} else {
976 			block = block >> 2;
977 			this_count = this_count >> 2;
978 		}
979 	}
980 	if (sdp->sector_size == 4096) {
981 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
982 			scmd_printk(KERN_ERR, SCpnt,
983 				    "Bad block number requested\n");
984 			goto out;
985 		} else {
986 			block = block >> 3;
987 			this_count = this_count >> 3;
988 		}
989 	}
990 	if (rq_data_dir(rq) == WRITE) {
991 		if (!sdp->writeable) {
992 			goto out;
993 		}
994 		SCpnt->cmnd[0] = WRITE_6;
995 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
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 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
1003 	} else {
1004 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
1005 		goto out;
1006 	}
1007 
1008 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1009 					"%s %d/%u 512 byte blocks.\n",
1010 					(rq_data_dir(rq) == WRITE) ?
1011 					"writing" : "reading", this_count,
1012 					blk_rq_sectors(rq)));
1013 
1014 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
1015 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
1016 	if (host_dif)
1017 		protect = 1 << 5;
1018 	else
1019 		protect = 0;
1020 
1021 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
1022 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1023 
1024 		if (unlikely(SCpnt->cmnd == NULL)) {
1025 			ret = BLKPREP_DEFER;
1026 			goto out;
1027 		}
1028 
1029 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1030 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1031 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1032 		SCpnt->cmnd[7] = 0x18;
1033 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1034 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1035 
1036 		/* LBA */
1037 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1038 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1039 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1040 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1041 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1042 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1043 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1044 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1045 
1046 		/* Expected Indirect LBA */
1047 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1048 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1049 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1050 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1051 
1052 		/* Transfer length */
1053 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1054 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1055 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1056 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1057 	} else if (sdp->use_16_for_rw) {
1058 		SCpnt->cmnd[0] += READ_16 - READ_6;
1059 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1060 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1061 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1062 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1063 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1064 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1065 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1066 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1067 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1068 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1069 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1070 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1071 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1072 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1073 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
1074 		   scsi_device_protection(SCpnt->device) ||
1075 		   SCpnt->device->use_10_for_rw) {
1076 		if (this_count > 0xffff)
1077 			this_count = 0xffff;
1078 
1079 		SCpnt->cmnd[0] += READ_10 - READ_6;
1080 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1081 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1082 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1083 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1084 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1085 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1086 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1087 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1088 	} else {
1089 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
1090 			/*
1091 			 * This happens only if this drive failed
1092 			 * 10byte rw command with ILLEGAL_REQUEST
1093 			 * during operation and thus turned off
1094 			 * use_10_for_rw.
1095 			 */
1096 			scmd_printk(KERN_ERR, SCpnt,
1097 				    "FUA write on READ/WRITE(6) drive\n");
1098 			goto out;
1099 		}
1100 
1101 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1102 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1103 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1104 		SCpnt->cmnd[4] = (unsigned char) this_count;
1105 		SCpnt->cmnd[5] = 0;
1106 	}
1107 	SCpnt->sdb.length = this_count * sdp->sector_size;
1108 
1109 	/* If DIF or DIX is enabled, tell HBA how to handle request */
1110 	if (host_dif || scsi_prot_sg_count(SCpnt))
1111 		sd_prot_op(SCpnt, host_dif);
1112 
1113 	/*
1114 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1115 	 * host adapter, it's safe to assume that we can at least transfer
1116 	 * this many bytes between each connect / disconnect.
1117 	 */
1118 	SCpnt->transfersize = sdp->sector_size;
1119 	SCpnt->underflow = this_count << 9;
1120 	SCpnt->allowed = SD_MAX_RETRIES;
1121 
1122 	/*
1123 	 * This indicates that the command is ready from our end to be
1124 	 * queued.
1125 	 */
1126 	ret = BLKPREP_OK;
1127  out:
1128 	return scsi_prep_return(q, rq, ret);
1129 }
1130 
1131 /**
1132  *	sd_open - open a scsi disk device
1133  *	@inode: only i_rdev member may be used
1134  *	@filp: only f_mode and f_flags may be used
1135  *
1136  *	Returns 0 if successful. Returns a negated errno value in case
1137  *	of error.
1138  *
1139  *	Note: This can be called from a user context (e.g. fsck(1) )
1140  *	or from within the kernel (e.g. as a result of a mount(1) ).
1141  *	In the latter case @inode and @filp carry an abridged amount
1142  *	of information as noted above.
1143  *
1144  *	Locking: called with bdev->bd_mutex held.
1145  **/
1146 static int sd_open(struct block_device *bdev, fmode_t mode)
1147 {
1148 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1149 	struct scsi_device *sdev;
1150 	int retval;
1151 
1152 	if (!sdkp)
1153 		return -ENXIO;
1154 
1155 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1156 
1157 	sdev = sdkp->device;
1158 
1159 	/*
1160 	 * If the device is in error recovery, wait until it is done.
1161 	 * If the device is offline, then disallow any access to it.
1162 	 */
1163 	retval = -ENXIO;
1164 	if (!scsi_block_when_processing_errors(sdev))
1165 		goto error_out;
1166 
1167 	if (sdev->removable || sdkp->write_prot)
1168 		check_disk_change(bdev);
1169 
1170 	/*
1171 	 * If the drive is empty, just let the open fail.
1172 	 */
1173 	retval = -ENOMEDIUM;
1174 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1175 		goto error_out;
1176 
1177 	/*
1178 	 * If the device has the write protect tab set, have the open fail
1179 	 * if the user expects to be able to write to the thing.
1180 	 */
1181 	retval = -EROFS;
1182 	if (sdkp->write_prot && (mode & FMODE_WRITE))
1183 		goto error_out;
1184 
1185 	/*
1186 	 * It is possible that the disk changing stuff resulted in
1187 	 * the device being taken offline.  If this is the case,
1188 	 * report this to the user, and don't pretend that the
1189 	 * open actually succeeded.
1190 	 */
1191 	retval = -ENXIO;
1192 	if (!scsi_device_online(sdev))
1193 		goto error_out;
1194 
1195 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1196 		if (scsi_block_when_processing_errors(sdev))
1197 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1198 	}
1199 
1200 	return 0;
1201 
1202 error_out:
1203 	scsi_disk_put(sdkp);
1204 	return retval;
1205 }
1206 
1207 /**
1208  *	sd_release - invoked when the (last) close(2) is called on this
1209  *	scsi disk.
1210  *	@inode: only i_rdev member may be used
1211  *	@filp: only f_mode and f_flags may be used
1212  *
1213  *	Returns 0.
1214  *
1215  *	Note: may block (uninterruptible) if error recovery is underway
1216  *	on this disk.
1217  *
1218  *	Locking: called with bdev->bd_mutex held.
1219  **/
1220 static void sd_release(struct gendisk *disk, fmode_t mode)
1221 {
1222 	struct scsi_disk *sdkp = scsi_disk(disk);
1223 	struct scsi_device *sdev = sdkp->device;
1224 
1225 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1226 
1227 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1228 		if (scsi_block_when_processing_errors(sdev))
1229 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1230 	}
1231 
1232 	/*
1233 	 * XXX and what if there are packets in flight and this close()
1234 	 * XXX is followed by a "rmmod sd_mod"?
1235 	 */
1236 
1237 	scsi_disk_put(sdkp);
1238 }
1239 
1240 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1241 {
1242 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1243 	struct scsi_device *sdp = sdkp->device;
1244 	struct Scsi_Host *host = sdp->host;
1245 	int diskinfo[4];
1246 
1247 	/* default to most commonly used values */
1248         diskinfo[0] = 0x40;	/* 1 << 6 */
1249        	diskinfo[1] = 0x20;	/* 1 << 5 */
1250        	diskinfo[2] = sdkp->capacity >> 11;
1251 
1252 	/* override with calculated, extended default, or driver values */
1253 	if (host->hostt->bios_param)
1254 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1255 	else
1256 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1257 
1258 	geo->heads = diskinfo[0];
1259 	geo->sectors = diskinfo[1];
1260 	geo->cylinders = diskinfo[2];
1261 	return 0;
1262 }
1263 
1264 /**
1265  *	sd_ioctl - process an ioctl
1266  *	@inode: only i_rdev/i_bdev members may be used
1267  *	@filp: only f_mode and f_flags may be used
1268  *	@cmd: ioctl command number
1269  *	@arg: this is third argument given to ioctl(2) system call.
1270  *	Often contains a pointer.
1271  *
1272  *	Returns 0 if successful (some ioctls return positive numbers on
1273  *	success as well). Returns a negated errno value in case of error.
1274  *
1275  *	Note: most ioctls are forward onto the block subsystem or further
1276  *	down in the scsi subsystem.
1277  **/
1278 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1279 		    unsigned int cmd, unsigned long arg)
1280 {
1281 	struct gendisk *disk = bdev->bd_disk;
1282 	struct scsi_disk *sdkp = scsi_disk(disk);
1283 	struct scsi_device *sdp = sdkp->device;
1284 	void __user *p = (void __user *)arg;
1285 	int error;
1286 
1287 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1288 				    "cmd=0x%x\n", disk->disk_name, cmd));
1289 
1290 	error = scsi_verify_blk_ioctl(bdev, cmd);
1291 	if (error < 0)
1292 		return error;
1293 
1294 	/*
1295 	 * If we are in the middle of error recovery, don't let anyone
1296 	 * else try and use this device.  Also, if error recovery fails, it
1297 	 * may try and take the device offline, in which case all further
1298 	 * access to the device is prohibited.
1299 	 */
1300 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1301 					(mode & FMODE_NDELAY) != 0);
1302 	if (!scsi_block_when_processing_errors(sdp) || !error)
1303 		goto out;
1304 
1305 	/*
1306 	 * Send SCSI addressing ioctls directly to mid level, send other
1307 	 * ioctls to block level and then onto mid level if they can't be
1308 	 * resolved.
1309 	 */
1310 	switch (cmd) {
1311 		case SCSI_IOCTL_GET_IDLUN:
1312 		case SCSI_IOCTL_GET_BUS_NUMBER:
1313 			error = scsi_ioctl(sdp, cmd, p);
1314 			break;
1315 		default:
1316 			error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1317 			if (error != -ENOTTY)
1318 				break;
1319 			error = scsi_ioctl(sdp, cmd, p);
1320 			break;
1321 	}
1322 out:
1323 	return error;
1324 }
1325 
1326 static void set_media_not_present(struct scsi_disk *sdkp)
1327 {
1328 	if (sdkp->media_present)
1329 		sdkp->device->changed = 1;
1330 
1331 	if (sdkp->device->removable) {
1332 		sdkp->media_present = 0;
1333 		sdkp->capacity = 0;
1334 	}
1335 }
1336 
1337 static int media_not_present(struct scsi_disk *sdkp,
1338 			     struct scsi_sense_hdr *sshdr)
1339 {
1340 	if (!scsi_sense_valid(sshdr))
1341 		return 0;
1342 
1343 	/* not invoked for commands that could return deferred errors */
1344 	switch (sshdr->sense_key) {
1345 	case UNIT_ATTENTION:
1346 	case NOT_READY:
1347 		/* medium not present */
1348 		if (sshdr->asc == 0x3A) {
1349 			set_media_not_present(sdkp);
1350 			return 1;
1351 		}
1352 	}
1353 	return 0;
1354 }
1355 
1356 /**
1357  *	sd_check_events - check media events
1358  *	@disk: kernel device descriptor
1359  *	@clearing: disk events currently being cleared
1360  *
1361  *	Returns mask of DISK_EVENT_*.
1362  *
1363  *	Note: this function is invoked from the block subsystem.
1364  **/
1365 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1366 {
1367 	struct scsi_disk *sdkp = scsi_disk(disk);
1368 	struct scsi_device *sdp = sdkp->device;
1369 	struct scsi_sense_hdr *sshdr = NULL;
1370 	int retval;
1371 
1372 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1373 
1374 	/*
1375 	 * If the device is offline, don't send any commands - just pretend as
1376 	 * if the command failed.  If the device ever comes back online, we
1377 	 * can deal with it then.  It is only because of unrecoverable errors
1378 	 * that we would ever take a device offline in the first place.
1379 	 */
1380 	if (!scsi_device_online(sdp)) {
1381 		set_media_not_present(sdkp);
1382 		goto out;
1383 	}
1384 
1385 	/*
1386 	 * Using TEST_UNIT_READY enables differentiation between drive with
1387 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1388 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1389 	 *
1390 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1391 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1392 	 * sd_revalidate() is called.
1393 	 */
1394 	retval = -ENODEV;
1395 
1396 	if (scsi_block_when_processing_errors(sdp)) {
1397 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1398 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1399 					      sshdr);
1400 	}
1401 
1402 	/* failed to execute TUR, assume media not present */
1403 	if (host_byte(retval)) {
1404 		set_media_not_present(sdkp);
1405 		goto out;
1406 	}
1407 
1408 	if (media_not_present(sdkp, sshdr))
1409 		goto out;
1410 
1411 	/*
1412 	 * For removable scsi disk we have to recognise the presence
1413 	 * of a disk in the drive.
1414 	 */
1415 	if (!sdkp->media_present)
1416 		sdp->changed = 1;
1417 	sdkp->media_present = 1;
1418 out:
1419 	/*
1420 	 * sdp->changed is set under the following conditions:
1421 	 *
1422 	 *	Medium present state has changed in either direction.
1423 	 *	Device has indicated UNIT_ATTENTION.
1424 	 */
1425 	kfree(sshdr);
1426 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1427 	sdp->changed = 0;
1428 	return retval;
1429 }
1430 
1431 static int sd_sync_cache(struct scsi_disk *sdkp)
1432 {
1433 	int retries, res;
1434 	struct scsi_device *sdp = sdkp->device;
1435 	struct scsi_sense_hdr sshdr;
1436 
1437 	if (!scsi_device_online(sdp))
1438 		return -ENODEV;
1439 
1440 
1441 	for (retries = 3; retries > 0; --retries) {
1442 		unsigned char cmd[10] = { 0 };
1443 
1444 		cmd[0] = SYNCHRONIZE_CACHE;
1445 		/*
1446 		 * Leave the rest of the command zero to indicate
1447 		 * flush everything.
1448 		 */
1449 		res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1450 					     &sshdr, SD_FLUSH_TIMEOUT,
1451 					     SD_MAX_RETRIES, NULL, REQ_PM);
1452 		if (res == 0)
1453 			break;
1454 	}
1455 
1456 	if (res) {
1457 		sd_print_result(sdkp, res);
1458 		if (driver_byte(res) & DRIVER_SENSE)
1459 			sd_print_sense_hdr(sdkp, &sshdr);
1460 	}
1461 
1462 	if (res)
1463 		return -EIO;
1464 	return 0;
1465 }
1466 
1467 static void sd_rescan(struct device *dev)
1468 {
1469 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1470 
1471 	if (sdkp) {
1472 		revalidate_disk(sdkp->disk);
1473 		scsi_disk_put(sdkp);
1474 	}
1475 }
1476 
1477 
1478 #ifdef CONFIG_COMPAT
1479 /*
1480  * This gets directly called from VFS. When the ioctl
1481  * is not recognized we go back to the other translation paths.
1482  */
1483 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1484 			   unsigned int cmd, unsigned long arg)
1485 {
1486 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1487 	int ret;
1488 
1489 	ret = scsi_verify_blk_ioctl(bdev, cmd);
1490 	if (ret < 0)
1491 		return ret;
1492 
1493 	/*
1494 	 * If we are in the middle of error recovery, don't let anyone
1495 	 * else try and use this device.  Also, if error recovery fails, it
1496 	 * may try and take the device offline, in which case all further
1497 	 * access to the device is prohibited.
1498 	 */
1499 	if (!scsi_block_when_processing_errors(sdev))
1500 		return -ENODEV;
1501 
1502 	if (sdev->host->hostt->compat_ioctl) {
1503 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1504 
1505 		return ret;
1506 	}
1507 
1508 	/*
1509 	 * Let the static ioctl translation table take care of it.
1510 	 */
1511 	return -ENOIOCTLCMD;
1512 }
1513 #endif
1514 
1515 static const struct block_device_operations sd_fops = {
1516 	.owner			= THIS_MODULE,
1517 	.open			= sd_open,
1518 	.release		= sd_release,
1519 	.ioctl			= sd_ioctl,
1520 	.getgeo			= sd_getgeo,
1521 #ifdef CONFIG_COMPAT
1522 	.compat_ioctl		= sd_compat_ioctl,
1523 #endif
1524 	.check_events		= sd_check_events,
1525 	.revalidate_disk	= sd_revalidate_disk,
1526 	.unlock_native_capacity	= sd_unlock_native_capacity,
1527 };
1528 
1529 /**
1530  *	sd_eh_action - error handling callback
1531  *	@scmd:		sd-issued command that has failed
1532  *	@eh_cmnd:	The command that was sent during error handling
1533  *	@eh_cmnd_len:	Length of eh_cmnd in bytes
1534  *	@eh_disp:	The recovery disposition suggested by the midlayer
1535  *
1536  *	This function is called by the SCSI midlayer upon completion of
1537  *	an error handling command (TEST UNIT READY, START STOP UNIT,
1538  *	etc.) The command sent to the device by the error handler is
1539  *	stored in eh_cmnd. The result of sending the eh command is
1540  *	passed in eh_disp.
1541  **/
1542 static int sd_eh_action(struct scsi_cmnd *scmd, unsigned char *eh_cmnd,
1543 			int eh_cmnd_len, int eh_disp)
1544 {
1545 	struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1546 
1547 	if (!scsi_device_online(scmd->device) ||
1548 	    !scsi_medium_access_command(scmd))
1549 		return eh_disp;
1550 
1551 	/*
1552 	 * The device has timed out executing a medium access command.
1553 	 * However, the TEST UNIT READY command sent during error
1554 	 * handling completed successfully. Either the device is in the
1555 	 * process of recovering or has it suffered an internal failure
1556 	 * that prevents access to the storage medium.
1557 	 */
1558 	if (host_byte(scmd->result) == DID_TIME_OUT && eh_disp == SUCCESS &&
1559 	    eh_cmnd_len && eh_cmnd[0] == TEST_UNIT_READY)
1560 		sdkp->medium_access_timed_out++;
1561 
1562 	/*
1563 	 * If the device keeps failing read/write commands but TEST UNIT
1564 	 * READY always completes successfully we assume that medium
1565 	 * access is no longer possible and take the device offline.
1566 	 */
1567 	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1568 		scmd_printk(KERN_ERR, scmd,
1569 			    "Medium access timeout failure. Offlining disk!\n");
1570 		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1571 
1572 		return FAILED;
1573 	}
1574 
1575 	return eh_disp;
1576 }
1577 
1578 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1579 {
1580 	u64 start_lba = blk_rq_pos(scmd->request);
1581 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1582 	u64 bad_lba;
1583 	int info_valid;
1584 	/*
1585 	 * resid is optional but mostly filled in.  When it's unused,
1586 	 * its value is zero, so we assume the whole buffer transferred
1587 	 */
1588 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1589 	unsigned int good_bytes;
1590 
1591 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1592 		return 0;
1593 
1594 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1595 					     SCSI_SENSE_BUFFERSIZE,
1596 					     &bad_lba);
1597 	if (!info_valid)
1598 		return 0;
1599 
1600 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1601 		return 0;
1602 
1603 	if (scmd->device->sector_size < 512) {
1604 		/* only legitimate sector_size here is 256 */
1605 		start_lba <<= 1;
1606 		end_lba <<= 1;
1607 	} else {
1608 		/* be careful ... don't want any overflows */
1609 		u64 factor = scmd->device->sector_size / 512;
1610 		do_div(start_lba, factor);
1611 		do_div(end_lba, factor);
1612 	}
1613 
1614 	/* The bad lba was reported incorrectly, we have no idea where
1615 	 * the error is.
1616 	 */
1617 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1618 		return 0;
1619 
1620 	/* This computation should always be done in terms of
1621 	 * the resolution of the device's medium.
1622 	 */
1623 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1624 	return min(good_bytes, transferred);
1625 }
1626 
1627 /**
1628  *	sd_done - bottom half handler: called when the lower level
1629  *	driver has completed (successfully or otherwise) a scsi command.
1630  *	@SCpnt: mid-level's per command structure.
1631  *
1632  *	Note: potentially run from within an ISR. Must not block.
1633  **/
1634 static int sd_done(struct scsi_cmnd *SCpnt)
1635 {
1636 	int result = SCpnt->result;
1637 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1638 	struct scsi_sense_hdr sshdr;
1639 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1640 	struct request *req = SCpnt->request;
1641 	int sense_valid = 0;
1642 	int sense_deferred = 0;
1643 	unsigned char op = SCpnt->cmnd[0];
1644 	unsigned char unmap = SCpnt->cmnd[1] & 8;
1645 
1646 	if (req->cmd_flags & REQ_DISCARD || req->cmd_flags & REQ_WRITE_SAME) {
1647 		if (!result) {
1648 			good_bytes = blk_rq_bytes(req);
1649 			scsi_set_resid(SCpnt, 0);
1650 		} else {
1651 			good_bytes = 0;
1652 			scsi_set_resid(SCpnt, blk_rq_bytes(req));
1653 		}
1654 	}
1655 
1656 	if (result) {
1657 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1658 		if (sense_valid)
1659 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1660 	}
1661 #ifdef CONFIG_SCSI_LOGGING
1662 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1663 	if (sense_valid) {
1664 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1665 						   "sd_done: sb[respc,sk,asc,"
1666 						   "ascq]=%x,%x,%x,%x\n",
1667 						   sshdr.response_code,
1668 						   sshdr.sense_key, sshdr.asc,
1669 						   sshdr.ascq));
1670 	}
1671 #endif
1672 	if (driver_byte(result) != DRIVER_SENSE &&
1673 	    (!sense_valid || sense_deferred))
1674 		goto out;
1675 
1676 	sdkp->medium_access_timed_out = 0;
1677 
1678 	switch (sshdr.sense_key) {
1679 	case HARDWARE_ERROR:
1680 	case MEDIUM_ERROR:
1681 		good_bytes = sd_completed_bytes(SCpnt);
1682 		break;
1683 	case RECOVERED_ERROR:
1684 		good_bytes = scsi_bufflen(SCpnt);
1685 		break;
1686 	case NO_SENSE:
1687 		/* This indicates a false check condition, so ignore it.  An
1688 		 * unknown amount of data was transferred so treat it as an
1689 		 * error.
1690 		 */
1691 		scsi_print_sense("sd", SCpnt);
1692 		SCpnt->result = 0;
1693 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1694 		break;
1695 	case ABORTED_COMMAND:
1696 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1697 			good_bytes = sd_completed_bytes(SCpnt);
1698 		break;
1699 	case ILLEGAL_REQUEST:
1700 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1701 			good_bytes = sd_completed_bytes(SCpnt);
1702 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1703 		if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1704 			switch (op) {
1705 			case UNMAP:
1706 				sd_config_discard(sdkp, SD_LBP_DISABLE);
1707 				break;
1708 			case WRITE_SAME_16:
1709 			case WRITE_SAME:
1710 				if (unmap)
1711 					sd_config_discard(sdkp, SD_LBP_DISABLE);
1712 				else {
1713 					sdkp->device->no_write_same = 1;
1714 					sd_config_write_same(sdkp);
1715 
1716 					good_bytes = 0;
1717 					req->__data_len = blk_rq_bytes(req);
1718 					req->cmd_flags |= REQ_QUIET;
1719 				}
1720 			}
1721 		}
1722 		break;
1723 	default:
1724 		break;
1725 	}
1726  out:
1727 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1728 		sd_dif_complete(SCpnt, good_bytes);
1729 
1730 	return good_bytes;
1731 }
1732 
1733 /*
1734  * spinup disk - called only in sd_revalidate_disk()
1735  */
1736 static void
1737 sd_spinup_disk(struct scsi_disk *sdkp)
1738 {
1739 	unsigned char cmd[10];
1740 	unsigned long spintime_expire = 0;
1741 	int retries, spintime;
1742 	unsigned int the_result;
1743 	struct scsi_sense_hdr sshdr;
1744 	int sense_valid = 0;
1745 
1746 	spintime = 0;
1747 
1748 	/* Spin up drives, as required.  Only do this at boot time */
1749 	/* Spinup needs to be done for module loads too. */
1750 	do {
1751 		retries = 0;
1752 
1753 		do {
1754 			cmd[0] = TEST_UNIT_READY;
1755 			memset((void *) &cmd[1], 0, 9);
1756 
1757 			the_result = scsi_execute_req(sdkp->device, cmd,
1758 						      DMA_NONE, NULL, 0,
1759 						      &sshdr, SD_TIMEOUT,
1760 						      SD_MAX_RETRIES, NULL);
1761 
1762 			/*
1763 			 * If the drive has indicated to us that it
1764 			 * doesn't have any media in it, don't bother
1765 			 * with any more polling.
1766 			 */
1767 			if (media_not_present(sdkp, &sshdr))
1768 				return;
1769 
1770 			if (the_result)
1771 				sense_valid = scsi_sense_valid(&sshdr);
1772 			retries++;
1773 		} while (retries < 3 &&
1774 			 (!scsi_status_is_good(the_result) ||
1775 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1776 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1777 
1778 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1779 			/* no sense, TUR either succeeded or failed
1780 			 * with a status error */
1781 			if(!spintime && !scsi_status_is_good(the_result)) {
1782 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1783 				sd_print_result(sdkp, the_result);
1784 			}
1785 			break;
1786 		}
1787 
1788 		/*
1789 		 * The device does not want the automatic start to be issued.
1790 		 */
1791 		if (sdkp->device->no_start_on_add)
1792 			break;
1793 
1794 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1795 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1796 				break;	/* manual intervention required */
1797 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1798 				break;	/* standby */
1799 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1800 				break;	/* unavailable */
1801 			/*
1802 			 * Issue command to spin up drive when not ready
1803 			 */
1804 			if (!spintime) {
1805 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1806 				cmd[0] = START_STOP;
1807 				cmd[1] = 1;	/* Return immediately */
1808 				memset((void *) &cmd[2], 0, 8);
1809 				cmd[4] = 1;	/* Start spin cycle */
1810 				if (sdkp->device->start_stop_pwr_cond)
1811 					cmd[4] |= 1 << 4;
1812 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1813 						 NULL, 0, &sshdr,
1814 						 SD_TIMEOUT, SD_MAX_RETRIES,
1815 						 NULL);
1816 				spintime_expire = jiffies + 100 * HZ;
1817 				spintime = 1;
1818 			}
1819 			/* Wait 1 second for next try */
1820 			msleep(1000);
1821 			printk(".");
1822 
1823 		/*
1824 		 * Wait for USB flash devices with slow firmware.
1825 		 * Yes, this sense key/ASC combination shouldn't
1826 		 * occur here.  It's characteristic of these devices.
1827 		 */
1828 		} else if (sense_valid &&
1829 				sshdr.sense_key == UNIT_ATTENTION &&
1830 				sshdr.asc == 0x28) {
1831 			if (!spintime) {
1832 				spintime_expire = jiffies + 5 * HZ;
1833 				spintime = 1;
1834 			}
1835 			/* Wait 1 second for next try */
1836 			msleep(1000);
1837 		} else {
1838 			/* we don't understand the sense code, so it's
1839 			 * probably pointless to loop */
1840 			if(!spintime) {
1841 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1842 				sd_print_sense_hdr(sdkp, &sshdr);
1843 			}
1844 			break;
1845 		}
1846 
1847 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1848 
1849 	if (spintime) {
1850 		if (scsi_status_is_good(the_result))
1851 			printk("ready\n");
1852 		else
1853 			printk("not responding...\n");
1854 	}
1855 }
1856 
1857 
1858 /*
1859  * Determine whether disk supports Data Integrity Field.
1860  */
1861 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1862 {
1863 	struct scsi_device *sdp = sdkp->device;
1864 	u8 type;
1865 	int ret = 0;
1866 
1867 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1868 		return ret;
1869 
1870 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1871 
1872 	if (type > SD_DIF_TYPE3_PROTECTION)
1873 		ret = -ENODEV;
1874 	else if (scsi_host_dif_capable(sdp->host, type))
1875 		ret = 1;
1876 
1877 	if (sdkp->first_scan || type != sdkp->protection_type)
1878 		switch (ret) {
1879 		case -ENODEV:
1880 			sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
1881 				  " protection type %u. Disabling disk!\n",
1882 				  type);
1883 			break;
1884 		case 1:
1885 			sd_printk(KERN_NOTICE, sdkp,
1886 				  "Enabling DIF Type %u protection\n", type);
1887 			break;
1888 		case 0:
1889 			sd_printk(KERN_NOTICE, sdkp,
1890 				  "Disabling DIF Type %u protection\n", type);
1891 			break;
1892 		}
1893 
1894 	sdkp->protection_type = type;
1895 
1896 	return ret;
1897 }
1898 
1899 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1900 			struct scsi_sense_hdr *sshdr, int sense_valid,
1901 			int the_result)
1902 {
1903 	sd_print_result(sdkp, the_result);
1904 	if (driver_byte(the_result) & DRIVER_SENSE)
1905 		sd_print_sense_hdr(sdkp, sshdr);
1906 	else
1907 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1908 
1909 	/*
1910 	 * Set dirty bit for removable devices if not ready -
1911 	 * sometimes drives will not report this properly.
1912 	 */
1913 	if (sdp->removable &&
1914 	    sense_valid && sshdr->sense_key == NOT_READY)
1915 		set_media_not_present(sdkp);
1916 
1917 	/*
1918 	 * We used to set media_present to 0 here to indicate no media
1919 	 * in the drive, but some drives fail read capacity even with
1920 	 * media present, so we can't do that.
1921 	 */
1922 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1923 }
1924 
1925 #define RC16_LEN 32
1926 #if RC16_LEN > SD_BUF_SIZE
1927 #error RC16_LEN must not be more than SD_BUF_SIZE
1928 #endif
1929 
1930 #define READ_CAPACITY_RETRIES_ON_RESET	10
1931 
1932 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1933 						unsigned char *buffer)
1934 {
1935 	unsigned char cmd[16];
1936 	struct scsi_sense_hdr sshdr;
1937 	int sense_valid = 0;
1938 	int the_result;
1939 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1940 	unsigned int alignment;
1941 	unsigned long long lba;
1942 	unsigned sector_size;
1943 
1944 	if (sdp->no_read_capacity_16)
1945 		return -EINVAL;
1946 
1947 	do {
1948 		memset(cmd, 0, 16);
1949 		cmd[0] = SERVICE_ACTION_IN;
1950 		cmd[1] = SAI_READ_CAPACITY_16;
1951 		cmd[13] = RC16_LEN;
1952 		memset(buffer, 0, RC16_LEN);
1953 
1954 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1955 					buffer, RC16_LEN, &sshdr,
1956 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1957 
1958 		if (media_not_present(sdkp, &sshdr))
1959 			return -ENODEV;
1960 
1961 		if (the_result) {
1962 			sense_valid = scsi_sense_valid(&sshdr);
1963 			if (sense_valid &&
1964 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1965 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1966 			    sshdr.ascq == 0x00)
1967 				/* Invalid Command Operation Code or
1968 				 * Invalid Field in CDB, just retry
1969 				 * silently with RC10 */
1970 				return -EINVAL;
1971 			if (sense_valid &&
1972 			    sshdr.sense_key == UNIT_ATTENTION &&
1973 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1974 				/* Device reset might occur several times,
1975 				 * give it one more chance */
1976 				if (--reset_retries > 0)
1977 					continue;
1978 		}
1979 		retries--;
1980 
1981 	} while (the_result && retries);
1982 
1983 	if (the_result) {
1984 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1985 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1986 		return -EINVAL;
1987 	}
1988 
1989 	sector_size = get_unaligned_be32(&buffer[8]);
1990 	lba = get_unaligned_be64(&buffer[0]);
1991 
1992 	if (sd_read_protection_type(sdkp, buffer) < 0) {
1993 		sdkp->capacity = 0;
1994 		return -ENODEV;
1995 	}
1996 
1997 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1998 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1999 			"kernel compiled with support for large block "
2000 			"devices.\n");
2001 		sdkp->capacity = 0;
2002 		return -EOVERFLOW;
2003 	}
2004 
2005 	/* Logical blocks per physical block exponent */
2006 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2007 
2008 	/* Lowest aligned logical block */
2009 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2010 	blk_queue_alignment_offset(sdp->request_queue, alignment);
2011 	if (alignment && sdkp->first_scan)
2012 		sd_printk(KERN_NOTICE, sdkp,
2013 			  "physical block alignment offset: %u\n", alignment);
2014 
2015 	if (buffer[14] & 0x80) { /* LBPME */
2016 		sdkp->lbpme = 1;
2017 
2018 		if (buffer[14] & 0x40) /* LBPRZ */
2019 			sdkp->lbprz = 1;
2020 
2021 		sd_config_discard(sdkp, SD_LBP_WS16);
2022 	}
2023 
2024 	sdkp->capacity = lba + 1;
2025 	return sector_size;
2026 }
2027 
2028 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2029 						unsigned char *buffer)
2030 {
2031 	unsigned char cmd[16];
2032 	struct scsi_sense_hdr sshdr;
2033 	int sense_valid = 0;
2034 	int the_result;
2035 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2036 	sector_t lba;
2037 	unsigned sector_size;
2038 
2039 	do {
2040 		cmd[0] = READ_CAPACITY;
2041 		memset(&cmd[1], 0, 9);
2042 		memset(buffer, 0, 8);
2043 
2044 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2045 					buffer, 8, &sshdr,
2046 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2047 
2048 		if (media_not_present(sdkp, &sshdr))
2049 			return -ENODEV;
2050 
2051 		if (the_result) {
2052 			sense_valid = scsi_sense_valid(&sshdr);
2053 			if (sense_valid &&
2054 			    sshdr.sense_key == UNIT_ATTENTION &&
2055 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2056 				/* Device reset might occur several times,
2057 				 * give it one more chance */
2058 				if (--reset_retries > 0)
2059 					continue;
2060 		}
2061 		retries--;
2062 
2063 	} while (the_result && retries);
2064 
2065 	if (the_result) {
2066 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
2067 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2068 		return -EINVAL;
2069 	}
2070 
2071 	sector_size = get_unaligned_be32(&buffer[4]);
2072 	lba = get_unaligned_be32(&buffer[0]);
2073 
2074 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2075 		/* Some buggy (usb cardreader) devices return an lba of
2076 		   0xffffffff when the want to report a size of 0 (with
2077 		   which they really mean no media is present) */
2078 		sdkp->capacity = 0;
2079 		sdkp->physical_block_size = sector_size;
2080 		return sector_size;
2081 	}
2082 
2083 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2084 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2085 			"kernel compiled with support for large block "
2086 			"devices.\n");
2087 		sdkp->capacity = 0;
2088 		return -EOVERFLOW;
2089 	}
2090 
2091 	sdkp->capacity = lba + 1;
2092 	sdkp->physical_block_size = sector_size;
2093 	return sector_size;
2094 }
2095 
2096 static int sd_try_rc16_first(struct scsi_device *sdp)
2097 {
2098 	if (sdp->host->max_cmd_len < 16)
2099 		return 0;
2100 	if (sdp->try_rc_10_first)
2101 		return 0;
2102 	if (sdp->scsi_level > SCSI_SPC_2)
2103 		return 1;
2104 	if (scsi_device_protection(sdp))
2105 		return 1;
2106 	return 0;
2107 }
2108 
2109 /*
2110  * read disk capacity
2111  */
2112 static void
2113 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2114 {
2115 	int sector_size;
2116 	struct scsi_device *sdp = sdkp->device;
2117 	sector_t old_capacity = sdkp->capacity;
2118 
2119 	if (sd_try_rc16_first(sdp)) {
2120 		sector_size = read_capacity_16(sdkp, sdp, buffer);
2121 		if (sector_size == -EOVERFLOW)
2122 			goto got_data;
2123 		if (sector_size == -ENODEV)
2124 			return;
2125 		if (sector_size < 0)
2126 			sector_size = read_capacity_10(sdkp, sdp, buffer);
2127 		if (sector_size < 0)
2128 			return;
2129 	} else {
2130 		sector_size = read_capacity_10(sdkp, sdp, buffer);
2131 		if (sector_size == -EOVERFLOW)
2132 			goto got_data;
2133 		if (sector_size < 0)
2134 			return;
2135 		if ((sizeof(sdkp->capacity) > 4) &&
2136 		    (sdkp->capacity > 0xffffffffULL)) {
2137 			int old_sector_size = sector_size;
2138 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2139 					"Trying to use READ CAPACITY(16).\n");
2140 			sector_size = read_capacity_16(sdkp, sdp, buffer);
2141 			if (sector_size < 0) {
2142 				sd_printk(KERN_NOTICE, sdkp,
2143 					"Using 0xffffffff as device size\n");
2144 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2145 				sector_size = old_sector_size;
2146 				goto got_data;
2147 			}
2148 		}
2149 	}
2150 
2151 	/* Some devices are known to return the total number of blocks,
2152 	 * not the highest block number.  Some devices have versions
2153 	 * which do this and others which do not.  Some devices we might
2154 	 * suspect of doing this but we don't know for certain.
2155 	 *
2156 	 * If we know the reported capacity is wrong, decrement it.  If
2157 	 * we can only guess, then assume the number of blocks is even
2158 	 * (usually true but not always) and err on the side of lowering
2159 	 * the capacity.
2160 	 */
2161 	if (sdp->fix_capacity ||
2162 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2163 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2164 				"from its reported value: %llu\n",
2165 				(unsigned long long) sdkp->capacity);
2166 		--sdkp->capacity;
2167 	}
2168 
2169 got_data:
2170 	if (sector_size == 0) {
2171 		sector_size = 512;
2172 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2173 			  "assuming 512.\n");
2174 	}
2175 
2176 	if (sector_size != 512 &&
2177 	    sector_size != 1024 &&
2178 	    sector_size != 2048 &&
2179 	    sector_size != 4096 &&
2180 	    sector_size != 256) {
2181 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2182 			  sector_size);
2183 		/*
2184 		 * The user might want to re-format the drive with
2185 		 * a supported sectorsize.  Once this happens, it
2186 		 * would be relatively trivial to set the thing up.
2187 		 * For this reason, we leave the thing in the table.
2188 		 */
2189 		sdkp->capacity = 0;
2190 		/*
2191 		 * set a bogus sector size so the normal read/write
2192 		 * logic in the block layer will eventually refuse any
2193 		 * request on this device without tripping over power
2194 		 * of two sector size assumptions
2195 		 */
2196 		sector_size = 512;
2197 	}
2198 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
2199 
2200 	{
2201 		char cap_str_2[10], cap_str_10[10];
2202 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
2203 
2204 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
2205 				sizeof(cap_str_2));
2206 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
2207 				sizeof(cap_str_10));
2208 
2209 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2210 			sd_printk(KERN_NOTICE, sdkp,
2211 				  "%llu %d-byte logical blocks: (%s/%s)\n",
2212 				  (unsigned long long)sdkp->capacity,
2213 				  sector_size, cap_str_10, cap_str_2);
2214 
2215 			if (sdkp->physical_block_size != sector_size)
2216 				sd_printk(KERN_NOTICE, sdkp,
2217 					  "%u-byte physical blocks\n",
2218 					  sdkp->physical_block_size);
2219 		}
2220 	}
2221 
2222 	sdp->use_16_for_rw = (sdkp->capacity > 0xffffffff);
2223 
2224 	/* Rescale capacity to 512-byte units */
2225 	if (sector_size == 4096)
2226 		sdkp->capacity <<= 3;
2227 	else if (sector_size == 2048)
2228 		sdkp->capacity <<= 2;
2229 	else if (sector_size == 1024)
2230 		sdkp->capacity <<= 1;
2231 	else if (sector_size == 256)
2232 		sdkp->capacity >>= 1;
2233 
2234 	blk_queue_physical_block_size(sdp->request_queue,
2235 				      sdkp->physical_block_size);
2236 	sdkp->device->sector_size = sector_size;
2237 }
2238 
2239 /* called with buffer of length 512 */
2240 static inline int
2241 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2242 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2243 		 struct scsi_sense_hdr *sshdr)
2244 {
2245 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2246 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
2247 			       sshdr);
2248 }
2249 
2250 /*
2251  * read write protect setting, if possible - called only in sd_revalidate_disk()
2252  * called with buffer of length SD_BUF_SIZE
2253  */
2254 static void
2255 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2256 {
2257 	int res;
2258 	struct scsi_device *sdp = sdkp->device;
2259 	struct scsi_mode_data data;
2260 	int old_wp = sdkp->write_prot;
2261 
2262 	set_disk_ro(sdkp->disk, 0);
2263 	if (sdp->skip_ms_page_3f) {
2264 		sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2265 		return;
2266 	}
2267 
2268 	if (sdp->use_192_bytes_for_3f) {
2269 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2270 	} else {
2271 		/*
2272 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2273 		 * We have to start carefully: some devices hang if we ask
2274 		 * for more than is available.
2275 		 */
2276 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2277 
2278 		/*
2279 		 * Second attempt: ask for page 0 When only page 0 is
2280 		 * implemented, a request for page 3F may return Sense Key
2281 		 * 5: Illegal Request, Sense Code 24: Invalid field in
2282 		 * CDB.
2283 		 */
2284 		if (!scsi_status_is_good(res))
2285 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2286 
2287 		/*
2288 		 * Third attempt: ask 255 bytes, as we did earlier.
2289 		 */
2290 		if (!scsi_status_is_good(res))
2291 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2292 					       &data, NULL);
2293 	}
2294 
2295 	if (!scsi_status_is_good(res)) {
2296 		sd_printk(KERN_WARNING, sdkp,
2297 			  "Test WP failed, assume Write Enabled\n");
2298 	} else {
2299 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2300 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2301 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2302 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2303 				  sdkp->write_prot ? "on" : "off");
2304 			sd_printk(KERN_DEBUG, sdkp,
2305 				  "Mode Sense: %02x %02x %02x %02x\n",
2306 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2307 		}
2308 	}
2309 }
2310 
2311 /*
2312  * sd_read_cache_type - called only from sd_revalidate_disk()
2313  * called with buffer of length SD_BUF_SIZE
2314  */
2315 static void
2316 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2317 {
2318 	int len = 0, res;
2319 	struct scsi_device *sdp = sdkp->device;
2320 
2321 	int dbd;
2322 	int modepage;
2323 	int first_len;
2324 	struct scsi_mode_data data;
2325 	struct scsi_sense_hdr sshdr;
2326 	int old_wce = sdkp->WCE;
2327 	int old_rcd = sdkp->RCD;
2328 	int old_dpofua = sdkp->DPOFUA;
2329 
2330 
2331 	if (sdkp->cache_override)
2332 		return;
2333 
2334 	first_len = 4;
2335 	if (sdp->skip_ms_page_8) {
2336 		if (sdp->type == TYPE_RBC)
2337 			goto defaults;
2338 		else {
2339 			if (sdp->skip_ms_page_3f)
2340 				goto defaults;
2341 			modepage = 0x3F;
2342 			if (sdp->use_192_bytes_for_3f)
2343 				first_len = 192;
2344 			dbd = 0;
2345 		}
2346 	} else if (sdp->type == TYPE_RBC) {
2347 		modepage = 6;
2348 		dbd = 8;
2349 	} else {
2350 		modepage = 8;
2351 		dbd = 0;
2352 	}
2353 
2354 	/* cautiously ask */
2355 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2356 			&data, &sshdr);
2357 
2358 	if (!scsi_status_is_good(res))
2359 		goto bad_sense;
2360 
2361 	if (!data.header_length) {
2362 		modepage = 6;
2363 		first_len = 0;
2364 		sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
2365 	}
2366 
2367 	/* that went OK, now ask for the proper length */
2368 	len = data.length;
2369 
2370 	/*
2371 	 * We're only interested in the first three bytes, actually.
2372 	 * But the data cache page is defined for the first 20.
2373 	 */
2374 	if (len < 3)
2375 		goto bad_sense;
2376 	else if (len > SD_BUF_SIZE) {
2377 		sd_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2378 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2379 		len = SD_BUF_SIZE;
2380 	}
2381 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2382 		len = 192;
2383 
2384 	/* Get the data */
2385 	if (len > first_len)
2386 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2387 				&data, &sshdr);
2388 
2389 	if (scsi_status_is_good(res)) {
2390 		int offset = data.header_length + data.block_descriptor_length;
2391 
2392 		while (offset < len) {
2393 			u8 page_code = buffer[offset] & 0x3F;
2394 			u8 spf       = buffer[offset] & 0x40;
2395 
2396 			if (page_code == 8 || page_code == 6) {
2397 				/* We're interested only in the first 3 bytes.
2398 				 */
2399 				if (len - offset <= 2) {
2400 					sd_printk(KERN_ERR, sdkp, "Incomplete "
2401 						  "mode parameter data\n");
2402 					goto defaults;
2403 				} else {
2404 					modepage = page_code;
2405 					goto Page_found;
2406 				}
2407 			} else {
2408 				/* Go to the next page */
2409 				if (spf && len - offset > 3)
2410 					offset += 4 + (buffer[offset+2] << 8) +
2411 						buffer[offset+3];
2412 				else if (!spf && len - offset > 1)
2413 					offset += 2 + buffer[offset+1];
2414 				else {
2415 					sd_printk(KERN_ERR, sdkp, "Incomplete "
2416 						  "mode parameter data\n");
2417 					goto defaults;
2418 				}
2419 			}
2420 		}
2421 
2422 		if (modepage == 0x3F) {
2423 			sd_printk(KERN_ERR, sdkp, "No Caching mode page "
2424 				  "present\n");
2425 			goto defaults;
2426 		} else if ((buffer[offset] & 0x3f) != modepage) {
2427 			sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
2428 			goto defaults;
2429 		}
2430 	Page_found:
2431 		if (modepage == 8) {
2432 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2433 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2434 		} else {
2435 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2436 			sdkp->RCD = 0;
2437 		}
2438 
2439 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2440 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2441 			sd_printk(KERN_NOTICE, sdkp,
2442 				  "Uses READ/WRITE(6), disabling FUA\n");
2443 			sdkp->DPOFUA = 0;
2444 		}
2445 
2446 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2447 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2448 			sd_printk(KERN_NOTICE, sdkp,
2449 				  "Write cache: %s, read cache: %s, %s\n",
2450 				  sdkp->WCE ? "enabled" : "disabled",
2451 				  sdkp->RCD ? "disabled" : "enabled",
2452 				  sdkp->DPOFUA ? "supports DPO and FUA"
2453 				  : "doesn't support DPO or FUA");
2454 
2455 		return;
2456 	}
2457 
2458 bad_sense:
2459 	if (scsi_sense_valid(&sshdr) &&
2460 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2461 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2462 		/* Invalid field in CDB */
2463 		sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2464 	else
2465 		sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
2466 
2467 defaults:
2468 	if (sdp->wce_default_on) {
2469 		sd_printk(KERN_NOTICE, sdkp, "Assuming drive cache: write back\n");
2470 		sdkp->WCE = 1;
2471 	} else {
2472 		sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
2473 		sdkp->WCE = 0;
2474 	}
2475 	sdkp->RCD = 0;
2476 	sdkp->DPOFUA = 0;
2477 }
2478 
2479 /*
2480  * The ATO bit indicates whether the DIF application tag is available
2481  * for use by the operating system.
2482  */
2483 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2484 {
2485 	int res, offset;
2486 	struct scsi_device *sdp = sdkp->device;
2487 	struct scsi_mode_data data;
2488 	struct scsi_sense_hdr sshdr;
2489 
2490 	if (sdp->type != TYPE_DISK)
2491 		return;
2492 
2493 	if (sdkp->protection_type == 0)
2494 		return;
2495 
2496 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2497 			      SD_MAX_RETRIES, &data, &sshdr);
2498 
2499 	if (!scsi_status_is_good(res) || !data.header_length ||
2500 	    data.length < 6) {
2501 		sd_printk(KERN_WARNING, sdkp,
2502 			  "getting Control mode page failed, assume no ATO\n");
2503 
2504 		if (scsi_sense_valid(&sshdr))
2505 			sd_print_sense_hdr(sdkp, &sshdr);
2506 
2507 		return;
2508 	}
2509 
2510 	offset = data.header_length + data.block_descriptor_length;
2511 
2512 	if ((buffer[offset] & 0x3f) != 0x0a) {
2513 		sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2514 		return;
2515 	}
2516 
2517 	if ((buffer[offset + 5] & 0x80) == 0)
2518 		return;
2519 
2520 	sdkp->ATO = 1;
2521 
2522 	return;
2523 }
2524 
2525 /**
2526  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2527  * @disk: disk to query
2528  */
2529 static void sd_read_block_limits(struct scsi_disk *sdkp)
2530 {
2531 	unsigned int sector_sz = sdkp->device->sector_size;
2532 	const int vpd_len = 64;
2533 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2534 
2535 	if (!buffer ||
2536 	    /* Block Limits VPD */
2537 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2538 		goto out;
2539 
2540 	blk_queue_io_min(sdkp->disk->queue,
2541 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2542 	blk_queue_io_opt(sdkp->disk->queue,
2543 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2544 
2545 	if (buffer[3] == 0x3c) {
2546 		unsigned int lba_count, desc_count;
2547 
2548 		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2549 
2550 		if (!sdkp->lbpme)
2551 			goto out;
2552 
2553 		lba_count = get_unaligned_be32(&buffer[20]);
2554 		desc_count = get_unaligned_be32(&buffer[24]);
2555 
2556 		if (lba_count && desc_count)
2557 			sdkp->max_unmap_blocks = lba_count;
2558 
2559 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2560 
2561 		if (buffer[32] & 0x80)
2562 			sdkp->unmap_alignment =
2563 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2564 
2565 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2566 
2567 			if (sdkp->max_unmap_blocks)
2568 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2569 			else
2570 				sd_config_discard(sdkp, SD_LBP_WS16);
2571 
2572 		} else {	/* LBP VPD page tells us what to use */
2573 
2574 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2575 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2576 			else if (sdkp->lbpws)
2577 				sd_config_discard(sdkp, SD_LBP_WS16);
2578 			else if (sdkp->lbpws10)
2579 				sd_config_discard(sdkp, SD_LBP_WS10);
2580 			else
2581 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2582 		}
2583 	}
2584 
2585  out:
2586 	kfree(buffer);
2587 }
2588 
2589 /**
2590  * sd_read_block_characteristics - Query block dev. characteristics
2591  * @disk: disk to query
2592  */
2593 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2594 {
2595 	unsigned char *buffer;
2596 	u16 rot;
2597 	const int vpd_len = 64;
2598 
2599 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2600 
2601 	if (!buffer ||
2602 	    /* Block Device Characteristics VPD */
2603 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2604 		goto out;
2605 
2606 	rot = get_unaligned_be16(&buffer[4]);
2607 
2608 	if (rot == 1)
2609 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2610 
2611  out:
2612 	kfree(buffer);
2613 }
2614 
2615 /**
2616  * sd_read_block_provisioning - Query provisioning VPD page
2617  * @disk: disk to query
2618  */
2619 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2620 {
2621 	unsigned char *buffer;
2622 	const int vpd_len = 8;
2623 
2624 	if (sdkp->lbpme == 0)
2625 		return;
2626 
2627 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2628 
2629 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2630 		goto out;
2631 
2632 	sdkp->lbpvpd	= 1;
2633 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2634 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2635 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2636 
2637  out:
2638 	kfree(buffer);
2639 }
2640 
2641 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2642 {
2643 	struct scsi_device *sdev = sdkp->device;
2644 
2645 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2646 		sdev->no_report_opcodes = 1;
2647 
2648 		/* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2649 		 * CODES is unsupported and the device has an ATA
2650 		 * Information VPD page (SAT).
2651 		 */
2652 		if (!scsi_get_vpd_page(sdev, 0x89, buffer, SD_BUF_SIZE))
2653 			sdev->no_write_same = 1;
2654 	}
2655 
2656 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2657 		sdkp->ws16 = 1;
2658 
2659 	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2660 		sdkp->ws10 = 1;
2661 }
2662 
2663 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2664 {
2665 	/*
2666 	 * Although VPD inquiries can go to SCSI-2 type devices,
2667 	 * some USB ones crash on receiving them, and the pages
2668 	 * we currently ask for are for SPC-3 and beyond
2669 	 */
2670 	if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2671 		return 1;
2672 	return 0;
2673 }
2674 
2675 /**
2676  *	sd_revalidate_disk - called the first time a new disk is seen,
2677  *	performs disk spin up, read_capacity, etc.
2678  *	@disk: struct gendisk we care about
2679  **/
2680 static int sd_revalidate_disk(struct gendisk *disk)
2681 {
2682 	struct scsi_disk *sdkp = scsi_disk(disk);
2683 	struct scsi_device *sdp = sdkp->device;
2684 	unsigned char *buffer;
2685 	unsigned flush = 0;
2686 
2687 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2688 				      "sd_revalidate_disk\n"));
2689 
2690 	/*
2691 	 * If the device is offline, don't try and read capacity or any
2692 	 * of the other niceties.
2693 	 */
2694 	if (!scsi_device_online(sdp))
2695 		goto out;
2696 
2697 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2698 	if (!buffer) {
2699 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2700 			  "allocation failure.\n");
2701 		goto out;
2702 	}
2703 
2704 	sd_spinup_disk(sdkp);
2705 
2706 	/*
2707 	 * Without media there is no reason to ask; moreover, some devices
2708 	 * react badly if we do.
2709 	 */
2710 	if (sdkp->media_present) {
2711 		sd_read_capacity(sdkp, buffer);
2712 
2713 		if (sd_try_extended_inquiry(sdp)) {
2714 			sd_read_block_provisioning(sdkp);
2715 			sd_read_block_limits(sdkp);
2716 			sd_read_block_characteristics(sdkp);
2717 		}
2718 
2719 		sd_read_write_protect_flag(sdkp, buffer);
2720 		sd_read_cache_type(sdkp, buffer);
2721 		sd_read_app_tag_own(sdkp, buffer);
2722 		sd_read_write_same(sdkp, buffer);
2723 	}
2724 
2725 	sdkp->first_scan = 0;
2726 
2727 	/*
2728 	 * We now have all cache related info, determine how we deal
2729 	 * with flush requests.
2730 	 */
2731 	if (sdkp->WCE) {
2732 		flush |= REQ_FLUSH;
2733 		if (sdkp->DPOFUA)
2734 			flush |= REQ_FUA;
2735 	}
2736 
2737 	blk_queue_flush(sdkp->disk->queue, flush);
2738 
2739 	set_capacity(disk, sdkp->capacity);
2740 	sd_config_write_same(sdkp);
2741 	kfree(buffer);
2742 
2743  out:
2744 	return 0;
2745 }
2746 
2747 /**
2748  *	sd_unlock_native_capacity - unlock native capacity
2749  *	@disk: struct gendisk to set capacity for
2750  *
2751  *	Block layer calls this function if it detects that partitions
2752  *	on @disk reach beyond the end of the device.  If the SCSI host
2753  *	implements ->unlock_native_capacity() method, it's invoked to
2754  *	give it a chance to adjust the device capacity.
2755  *
2756  *	CONTEXT:
2757  *	Defined by block layer.  Might sleep.
2758  */
2759 static void sd_unlock_native_capacity(struct gendisk *disk)
2760 {
2761 	struct scsi_device *sdev = scsi_disk(disk)->device;
2762 
2763 	if (sdev->host->hostt->unlock_native_capacity)
2764 		sdev->host->hostt->unlock_native_capacity(sdev);
2765 }
2766 
2767 /**
2768  *	sd_format_disk_name - format disk name
2769  *	@prefix: name prefix - ie. "sd" for SCSI disks
2770  *	@index: index of the disk to format name for
2771  *	@buf: output buffer
2772  *	@buflen: length of the output buffer
2773  *
2774  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2775  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2776  *	which is followed by sdaaa.
2777  *
2778  *	This is basically 26 base counting with one extra 'nil' entry
2779  *	at the beginning from the second digit on and can be
2780  *	determined using similar method as 26 base conversion with the
2781  *	index shifted -1 after each digit is computed.
2782  *
2783  *	CONTEXT:
2784  *	Don't care.
2785  *
2786  *	RETURNS:
2787  *	0 on success, -errno on failure.
2788  */
2789 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2790 {
2791 	const int base = 'z' - 'a' + 1;
2792 	char *begin = buf + strlen(prefix);
2793 	char *end = buf + buflen;
2794 	char *p;
2795 	int unit;
2796 
2797 	p = end - 1;
2798 	*p = '\0';
2799 	unit = base;
2800 	do {
2801 		if (p == begin)
2802 			return -EINVAL;
2803 		*--p = 'a' + (index % unit);
2804 		index = (index / unit) - 1;
2805 	} while (index >= 0);
2806 
2807 	memmove(begin, p, end - p);
2808 	memcpy(buf, prefix, strlen(prefix));
2809 
2810 	return 0;
2811 }
2812 
2813 /*
2814  * The asynchronous part of sd_probe
2815  */
2816 static void sd_probe_async(void *data, async_cookie_t cookie)
2817 {
2818 	struct scsi_disk *sdkp = data;
2819 	struct scsi_device *sdp;
2820 	struct gendisk *gd;
2821 	u32 index;
2822 	struct device *dev;
2823 
2824 	sdp = sdkp->device;
2825 	gd = sdkp->disk;
2826 	index = sdkp->index;
2827 	dev = &sdp->sdev_gendev;
2828 
2829 	gd->major = sd_major((index & 0xf0) >> 4);
2830 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2831 	gd->minors = SD_MINORS;
2832 
2833 	gd->fops = &sd_fops;
2834 	gd->private_data = &sdkp->driver;
2835 	gd->queue = sdkp->device->request_queue;
2836 
2837 	/* defaults, until the device tells us otherwise */
2838 	sdp->sector_size = 512;
2839 	sdkp->capacity = 0;
2840 	sdkp->media_present = 1;
2841 	sdkp->write_prot = 0;
2842 	sdkp->cache_override = 0;
2843 	sdkp->WCE = 0;
2844 	sdkp->RCD = 0;
2845 	sdkp->ATO = 0;
2846 	sdkp->first_scan = 1;
2847 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2848 
2849 	sd_revalidate_disk(gd);
2850 
2851 	blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2852 	blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2853 
2854 	gd->driverfs_dev = &sdp->sdev_gendev;
2855 	gd->flags = GENHD_FL_EXT_DEVT;
2856 	if (sdp->removable) {
2857 		gd->flags |= GENHD_FL_REMOVABLE;
2858 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2859 	}
2860 
2861 	add_disk(gd);
2862 	if (sdkp->capacity)
2863 		sd_dif_config_host(sdkp);
2864 
2865 	sd_revalidate_disk(gd);
2866 
2867 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2868 		  sdp->removable ? "removable " : "");
2869 	blk_pm_runtime_init(sdp->request_queue, dev);
2870 	scsi_autopm_put_device(sdp);
2871 	put_device(&sdkp->dev);
2872 }
2873 
2874 /**
2875  *	sd_probe - called during driver initialization and whenever a
2876  *	new scsi device is attached to the system. It is called once
2877  *	for each scsi device (not just disks) present.
2878  *	@dev: pointer to device object
2879  *
2880  *	Returns 0 if successful (or not interested in this scsi device
2881  *	(e.g. scanner)); 1 when there is an error.
2882  *
2883  *	Note: this function is invoked from the scsi mid-level.
2884  *	This function sets up the mapping between a given
2885  *	<host,channel,id,lun> (found in sdp) and new device name
2886  *	(e.g. /dev/sda). More precisely it is the block device major
2887  *	and minor number that is chosen here.
2888  *
2889  *	Assume sd_probe is not re-entrant (for time being)
2890  *	Also think about sd_probe() and sd_remove() running coincidentally.
2891  **/
2892 static int sd_probe(struct device *dev)
2893 {
2894 	struct scsi_device *sdp = to_scsi_device(dev);
2895 	struct scsi_disk *sdkp;
2896 	struct gendisk *gd;
2897 	int index;
2898 	int error;
2899 
2900 	error = -ENODEV;
2901 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2902 		goto out;
2903 
2904 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2905 					"sd_probe\n"));
2906 
2907 	error = -ENOMEM;
2908 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2909 	if (!sdkp)
2910 		goto out;
2911 
2912 	gd = alloc_disk(SD_MINORS);
2913 	if (!gd)
2914 		goto out_free;
2915 
2916 	do {
2917 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2918 			goto out_put;
2919 
2920 		spin_lock(&sd_index_lock);
2921 		error = ida_get_new(&sd_index_ida, &index);
2922 		spin_unlock(&sd_index_lock);
2923 	} while (error == -EAGAIN);
2924 
2925 	if (error) {
2926 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2927 		goto out_put;
2928 	}
2929 
2930 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2931 	if (error) {
2932 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2933 		goto out_free_index;
2934 	}
2935 
2936 	sdkp->device = sdp;
2937 	sdkp->driver = &sd_template;
2938 	sdkp->disk = gd;
2939 	sdkp->index = index;
2940 	atomic_set(&sdkp->openers, 0);
2941 	atomic_set(&sdkp->device->ioerr_cnt, 0);
2942 
2943 	if (!sdp->request_queue->rq_timeout) {
2944 		if (sdp->type != TYPE_MOD)
2945 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2946 		else
2947 			blk_queue_rq_timeout(sdp->request_queue,
2948 					     SD_MOD_TIMEOUT);
2949 	}
2950 
2951 	device_initialize(&sdkp->dev);
2952 	sdkp->dev.parent = dev;
2953 	sdkp->dev.class = &sd_disk_class;
2954 	dev_set_name(&sdkp->dev, "%s", dev_name(dev));
2955 
2956 	if (device_add(&sdkp->dev))
2957 		goto out_free_index;
2958 
2959 	get_device(dev);
2960 	dev_set_drvdata(dev, sdkp);
2961 
2962 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2963 	async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
2964 
2965 	return 0;
2966 
2967  out_free_index:
2968 	spin_lock(&sd_index_lock);
2969 	ida_remove(&sd_index_ida, index);
2970 	spin_unlock(&sd_index_lock);
2971  out_put:
2972 	put_disk(gd);
2973  out_free:
2974 	kfree(sdkp);
2975  out:
2976 	return error;
2977 }
2978 
2979 /**
2980  *	sd_remove - called whenever a scsi disk (previously recognized by
2981  *	sd_probe) is detached from the system. It is called (potentially
2982  *	multiple times) during sd module unload.
2983  *	@sdp: pointer to mid level scsi device object
2984  *
2985  *	Note: this function is invoked from the scsi mid-level.
2986  *	This function potentially frees up a device name (e.g. /dev/sdc)
2987  *	that could be re-used by a subsequent sd_probe().
2988  *	This function is not called when the built-in sd driver is "exit-ed".
2989  **/
2990 static int sd_remove(struct device *dev)
2991 {
2992 	struct scsi_disk *sdkp;
2993 	dev_t devt;
2994 
2995 	sdkp = dev_get_drvdata(dev);
2996 	devt = disk_devt(sdkp->disk);
2997 	scsi_autopm_get_device(sdkp->device);
2998 
2999 	async_synchronize_full_domain(&scsi_sd_probe_domain);
3000 	blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
3001 	blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
3002 	device_del(&sdkp->dev);
3003 	del_gendisk(sdkp->disk);
3004 	sd_shutdown(dev);
3005 
3006 	blk_register_region(devt, SD_MINORS, NULL,
3007 			    sd_default_probe, NULL, NULL);
3008 
3009 	mutex_lock(&sd_ref_mutex);
3010 	dev_set_drvdata(dev, NULL);
3011 	put_device(&sdkp->dev);
3012 	mutex_unlock(&sd_ref_mutex);
3013 
3014 	return 0;
3015 }
3016 
3017 /**
3018  *	scsi_disk_release - Called to free the scsi_disk structure
3019  *	@dev: pointer to embedded class device
3020  *
3021  *	sd_ref_mutex must be held entering this routine.  Because it is
3022  *	called on last put, you should always use the scsi_disk_get()
3023  *	scsi_disk_put() helpers which manipulate the semaphore directly
3024  *	and never do a direct put_device.
3025  **/
3026 static void scsi_disk_release(struct device *dev)
3027 {
3028 	struct scsi_disk *sdkp = to_scsi_disk(dev);
3029 	struct gendisk *disk = sdkp->disk;
3030 
3031 	spin_lock(&sd_index_lock);
3032 	ida_remove(&sd_index_ida, sdkp->index);
3033 	spin_unlock(&sd_index_lock);
3034 
3035 	disk->private_data = NULL;
3036 	put_disk(disk);
3037 	put_device(&sdkp->device->sdev_gendev);
3038 
3039 	kfree(sdkp);
3040 }
3041 
3042 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3043 {
3044 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
3045 	struct scsi_sense_hdr sshdr;
3046 	struct scsi_device *sdp = sdkp->device;
3047 	int res;
3048 
3049 	if (start)
3050 		cmd[4] |= 1;	/* START */
3051 
3052 	if (sdp->start_stop_pwr_cond)
3053 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
3054 
3055 	if (!scsi_device_online(sdp))
3056 		return -ENODEV;
3057 
3058 	res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3059 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3060 	if (res) {
3061 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
3062 		sd_print_result(sdkp, res);
3063 		if (driver_byte(res) & DRIVER_SENSE)
3064 			sd_print_sense_hdr(sdkp, &sshdr);
3065 	}
3066 
3067 	return res;
3068 }
3069 
3070 /*
3071  * Send a SYNCHRONIZE CACHE instruction down to the device through
3072  * the normal SCSI command structure.  Wait for the command to
3073  * complete.
3074  */
3075 static void sd_shutdown(struct device *dev)
3076 {
3077 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3078 
3079 	if (!sdkp)
3080 		return;         /* this can happen */
3081 
3082 	if (pm_runtime_suspended(dev))
3083 		goto exit;
3084 
3085 	if (sdkp->WCE) {
3086 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3087 		sd_sync_cache(sdkp);
3088 	}
3089 
3090 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3091 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3092 		sd_start_stop_device(sdkp, 0);
3093 	}
3094 
3095 exit:
3096 	scsi_disk_put(sdkp);
3097 }
3098 
3099 static int sd_suspend(struct device *dev)
3100 {
3101 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3102 	int ret = 0;
3103 
3104 	if (!sdkp)
3105 		return 0;	/* this can happen */
3106 
3107 	if (sdkp->WCE) {
3108 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3109 		ret = sd_sync_cache(sdkp);
3110 		if (ret)
3111 			goto done;
3112 	}
3113 
3114 	if (sdkp->device->manage_start_stop) {
3115 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3116 		ret = sd_start_stop_device(sdkp, 0);
3117 	}
3118 
3119 done:
3120 	scsi_disk_put(sdkp);
3121 	return ret;
3122 }
3123 
3124 static int sd_resume(struct device *dev)
3125 {
3126 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
3127 	int ret = 0;
3128 
3129 	if (!sdkp->device->manage_start_stop)
3130 		goto done;
3131 
3132 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3133 	ret = sd_start_stop_device(sdkp, 1);
3134 
3135 done:
3136 	scsi_disk_put(sdkp);
3137 	return ret;
3138 }
3139 
3140 /**
3141  *	init_sd - entry point for this driver (both when built in or when
3142  *	a module).
3143  *
3144  *	Note: this function registers this driver with the scsi mid-level.
3145  **/
3146 static int __init init_sd(void)
3147 {
3148 	int majors = 0, i, err;
3149 
3150 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3151 
3152 	for (i = 0; i < SD_MAJORS; i++) {
3153 		if (register_blkdev(sd_major(i), "sd") != 0)
3154 			continue;
3155 		majors++;
3156 		blk_register_region(sd_major(i), SD_MINORS, NULL,
3157 				    sd_default_probe, NULL, NULL);
3158 	}
3159 
3160 	if (!majors)
3161 		return -ENODEV;
3162 
3163 	err = class_register(&sd_disk_class);
3164 	if (err)
3165 		goto err_out;
3166 
3167 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3168 					 0, 0, NULL);
3169 	if (!sd_cdb_cache) {
3170 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
3171 		goto err_out_class;
3172 	}
3173 
3174 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3175 	if (!sd_cdb_pool) {
3176 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
3177 		goto err_out_cache;
3178 	}
3179 
3180 	err = scsi_register_driver(&sd_template.gendrv);
3181 	if (err)
3182 		goto err_out_driver;
3183 
3184 	return 0;
3185 
3186 err_out_driver:
3187 	mempool_destroy(sd_cdb_pool);
3188 
3189 err_out_cache:
3190 	kmem_cache_destroy(sd_cdb_cache);
3191 
3192 err_out_class:
3193 	class_unregister(&sd_disk_class);
3194 err_out:
3195 	for (i = 0; i < SD_MAJORS; i++)
3196 		unregister_blkdev(sd_major(i), "sd");
3197 	return err;
3198 }
3199 
3200 /**
3201  *	exit_sd - exit point for this driver (when it is a module).
3202  *
3203  *	Note: this function unregisters this driver from the scsi mid-level.
3204  **/
3205 static void __exit exit_sd(void)
3206 {
3207 	int i;
3208 
3209 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3210 
3211 	scsi_unregister_driver(&sd_template.gendrv);
3212 	mempool_destroy(sd_cdb_pool);
3213 	kmem_cache_destroy(sd_cdb_cache);
3214 
3215 	class_unregister(&sd_disk_class);
3216 
3217 	for (i = 0; i < SD_MAJORS; i++) {
3218 		blk_unregister_region(sd_major(i), SD_MINORS);
3219 		unregister_blkdev(sd_major(i), "sd");
3220 	}
3221 }
3222 
3223 module_init(init_sd);
3224 module_exit(exit_sd);
3225 
3226 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3227 			       struct scsi_sense_hdr *sshdr)
3228 {
3229 	sd_printk(KERN_INFO, sdkp, " ");
3230 	scsi_show_sense_hdr(sshdr);
3231 	sd_printk(KERN_INFO, sdkp, " ");
3232 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
3233 }
3234 
3235 static void sd_print_result(struct scsi_disk *sdkp, int result)
3236 {
3237 	sd_printk(KERN_INFO, sdkp, " ");
3238 	scsi_show_result(result);
3239 }
3240 
3241