xref: /linux/drivers/scsi/sd.c (revision 492c826b9facefa84995f4dea917e301b5ee0884)
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 <asm/uaccess.h>
54 #include <asm/unaligned.h>
55 
56 #include <scsi/scsi.h>
57 #include <scsi/scsi_cmnd.h>
58 #include <scsi/scsi_dbg.h>
59 #include <scsi/scsi_device.h>
60 #include <scsi/scsi_driver.h>
61 #include <scsi/scsi_eh.h>
62 #include <scsi/scsi_host.h>
63 #include <scsi/scsi_ioctl.h>
64 #include <scsi/scsicam.h>
65 
66 #include "sd.h"
67 #include "scsi_logging.h"
68 
69 MODULE_AUTHOR("Eric Youngdale");
70 MODULE_DESCRIPTION("SCSI disk (sd) driver");
71 MODULE_LICENSE("GPL");
72 
73 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
74 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
89 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
90 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
92 
93 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
94 #define SD_MINORS	16
95 #else
96 #define SD_MINORS	0
97 #endif
98 
99 static void sd_config_discard(struct scsi_disk *, unsigned int);
100 static int  sd_revalidate_disk(struct gendisk *);
101 static void sd_unlock_native_capacity(struct gendisk *disk);
102 static int  sd_probe(struct device *);
103 static int  sd_remove(struct device *);
104 static void sd_shutdown(struct device *);
105 static int sd_suspend(struct device *, pm_message_t state);
106 static int sd_resume(struct device *);
107 static void sd_rescan(struct device *);
108 static int sd_done(struct scsi_cmnd *);
109 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
110 static void scsi_disk_release(struct device *cdev);
111 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
112 static void sd_print_result(struct scsi_disk *, int);
113 
114 static DEFINE_SPINLOCK(sd_index_lock);
115 static DEFINE_IDA(sd_index_ida);
116 
117 /* This semaphore is used to mediate the 0->1 reference get in the
118  * face of object destruction (i.e. we can't allow a get on an
119  * object after last put) */
120 static DEFINE_MUTEX(sd_ref_mutex);
121 
122 static struct kmem_cache *sd_cdb_cache;
123 static mempool_t *sd_cdb_pool;
124 
125 static const char *sd_cache_types[] = {
126 	"write through", "none", "write back",
127 	"write back, no read (daft)"
128 };
129 
130 static ssize_t
131 sd_store_cache_type(struct device *dev, struct device_attribute *attr,
132 		    const char *buf, size_t count)
133 {
134 	int i, ct = -1, rcd, wce, sp;
135 	struct scsi_disk *sdkp = to_scsi_disk(dev);
136 	struct scsi_device *sdp = sdkp->device;
137 	char buffer[64];
138 	char *buffer_data;
139 	struct scsi_mode_data data;
140 	struct scsi_sense_hdr sshdr;
141 	int len;
142 
143 	if (sdp->type != TYPE_DISK)
144 		/* no cache control on RBC devices; theoretically they
145 		 * can do it, but there's probably so many exceptions
146 		 * it's not worth the risk */
147 		return -EINVAL;
148 
149 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
150 		len = strlen(sd_cache_types[i]);
151 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
152 		    buf[len] == '\n') {
153 			ct = i;
154 			break;
155 		}
156 	}
157 	if (ct < 0)
158 		return -EINVAL;
159 	rcd = ct & 0x01 ? 1 : 0;
160 	wce = ct & 0x02 ? 1 : 0;
161 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
162 			    SD_MAX_RETRIES, &data, NULL))
163 		return -EINVAL;
164 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
165 		  data.block_descriptor_length);
166 	buffer_data = buffer + data.header_length +
167 		data.block_descriptor_length;
168 	buffer_data[2] &= ~0x05;
169 	buffer_data[2] |= wce << 2 | rcd;
170 	sp = buffer_data[0] & 0x80 ? 1 : 0;
171 
172 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
173 			     SD_MAX_RETRIES, &data, &sshdr)) {
174 		if (scsi_sense_valid(&sshdr))
175 			sd_print_sense_hdr(sdkp, &sshdr);
176 		return -EINVAL;
177 	}
178 	revalidate_disk(sdkp->disk);
179 	return count;
180 }
181 
182 static ssize_t
183 sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
184 			   const char *buf, size_t count)
185 {
186 	struct scsi_disk *sdkp = to_scsi_disk(dev);
187 	struct scsi_device *sdp = sdkp->device;
188 
189 	if (!capable(CAP_SYS_ADMIN))
190 		return -EACCES;
191 
192 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
193 
194 	return count;
195 }
196 
197 static ssize_t
198 sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
199 		       const char *buf, size_t count)
200 {
201 	struct scsi_disk *sdkp = to_scsi_disk(dev);
202 	struct scsi_device *sdp = sdkp->device;
203 
204 	if (!capable(CAP_SYS_ADMIN))
205 		return -EACCES;
206 
207 	if (sdp->type != TYPE_DISK)
208 		return -EINVAL;
209 
210 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
211 
212 	return count;
213 }
214 
215 static ssize_t
216 sd_show_cache_type(struct device *dev, struct device_attribute *attr,
217 		   char *buf)
218 {
219 	struct scsi_disk *sdkp = to_scsi_disk(dev);
220 	int ct = sdkp->RCD + 2*sdkp->WCE;
221 
222 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
223 }
224 
225 static ssize_t
226 sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
227 {
228 	struct scsi_disk *sdkp = to_scsi_disk(dev);
229 
230 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
231 }
232 
233 static ssize_t
234 sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
235 			  char *buf)
236 {
237 	struct scsi_disk *sdkp = to_scsi_disk(dev);
238 	struct scsi_device *sdp = sdkp->device;
239 
240 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
241 }
242 
243 static ssize_t
244 sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
245 		      char *buf)
246 {
247 	struct scsi_disk *sdkp = to_scsi_disk(dev);
248 
249 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
250 }
251 
252 static ssize_t
253 sd_show_protection_type(struct device *dev, struct device_attribute *attr,
254 			char *buf)
255 {
256 	struct scsi_disk *sdkp = to_scsi_disk(dev);
257 
258 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
259 }
260 
261 static ssize_t
262 sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
263 			char *buf)
264 {
265 	struct scsi_disk *sdkp = to_scsi_disk(dev);
266 	struct scsi_device *sdp = sdkp->device;
267 	unsigned int dif, dix;
268 
269 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
270 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
271 
272 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
273 		dif = 0;
274 		dix = 1;
275 	}
276 
277 	if (!dif && !dix)
278 		return snprintf(buf, 20, "none\n");
279 
280 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
281 }
282 
283 static ssize_t
284 sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
285 		    char *buf)
286 {
287 	struct scsi_disk *sdkp = to_scsi_disk(dev);
288 
289 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
290 }
291 
292 static ssize_t
293 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
294 			  char *buf)
295 {
296 	struct scsi_disk *sdkp = to_scsi_disk(dev);
297 
298 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
299 }
300 
301 static const char *lbp_mode[] = {
302 	[SD_LBP_FULL]		= "full",
303 	[SD_LBP_UNMAP]		= "unmap",
304 	[SD_LBP_WS16]		= "writesame_16",
305 	[SD_LBP_WS10]		= "writesame_10",
306 	[SD_LBP_ZERO]		= "writesame_zero",
307 	[SD_LBP_DISABLE]	= "disabled",
308 };
309 
310 static ssize_t
311 sd_show_provisioning_mode(struct device *dev, struct device_attribute *attr,
312 			  char *buf)
313 {
314 	struct scsi_disk *sdkp = to_scsi_disk(dev);
315 
316 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
317 }
318 
319 static ssize_t
320 sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
321 			   const char *buf, size_t count)
322 {
323 	struct scsi_disk *sdkp = to_scsi_disk(dev);
324 	struct scsi_device *sdp = sdkp->device;
325 
326 	if (!capable(CAP_SYS_ADMIN))
327 		return -EACCES;
328 
329 	if (sdp->type != TYPE_DISK)
330 		return -EINVAL;
331 
332 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
333 		sd_config_discard(sdkp, SD_LBP_UNMAP);
334 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
335 		sd_config_discard(sdkp, SD_LBP_WS16);
336 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
337 		sd_config_discard(sdkp, SD_LBP_WS10);
338 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
339 		sd_config_discard(sdkp, SD_LBP_ZERO);
340 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
341 		sd_config_discard(sdkp, SD_LBP_DISABLE);
342 	else
343 		return -EINVAL;
344 
345 	return count;
346 }
347 
348 static struct device_attribute sd_disk_attrs[] = {
349 	__ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
350 	       sd_store_cache_type),
351 	__ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
352 	__ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
353 	       sd_store_allow_restart),
354 	__ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
355 	       sd_store_manage_start_stop),
356 	__ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
357 	__ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
358 	__ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
359 	__ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
360 	__ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
361 	       sd_store_provisioning_mode),
362 	__ATTR_NULL,
363 };
364 
365 static struct class sd_disk_class = {
366 	.name		= "scsi_disk",
367 	.owner		= THIS_MODULE,
368 	.dev_release	= scsi_disk_release,
369 	.dev_attrs	= sd_disk_attrs,
370 };
371 
372 static struct scsi_driver sd_template = {
373 	.owner			= THIS_MODULE,
374 	.gendrv = {
375 		.name		= "sd",
376 		.probe		= sd_probe,
377 		.remove		= sd_remove,
378 		.suspend	= sd_suspend,
379 		.resume		= sd_resume,
380 		.shutdown	= sd_shutdown,
381 	},
382 	.rescan			= sd_rescan,
383 	.done			= sd_done,
384 };
385 
386 /*
387  * Device no to disk mapping:
388  *
389  *       major         disc2     disc  p1
390  *   |............|.............|....|....| <- dev_t
391  *    31        20 19          8 7  4 3  0
392  *
393  * Inside a major, we have 16k disks, however mapped non-
394  * contiguously. The first 16 disks are for major0, the next
395  * ones with major1, ... Disk 256 is for major0 again, disk 272
396  * for major1, ...
397  * As we stay compatible with our numbering scheme, we can reuse
398  * the well-know SCSI majors 8, 65--71, 136--143.
399  */
400 static int sd_major(int major_idx)
401 {
402 	switch (major_idx) {
403 	case 0:
404 		return SCSI_DISK0_MAJOR;
405 	case 1 ... 7:
406 		return SCSI_DISK1_MAJOR + major_idx - 1;
407 	case 8 ... 15:
408 		return SCSI_DISK8_MAJOR + major_idx - 8;
409 	default:
410 		BUG();
411 		return 0;	/* shut up gcc */
412 	}
413 }
414 
415 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
416 {
417 	struct scsi_disk *sdkp = NULL;
418 
419 	if (disk->private_data) {
420 		sdkp = scsi_disk(disk);
421 		if (scsi_device_get(sdkp->device) == 0)
422 			get_device(&sdkp->dev);
423 		else
424 			sdkp = NULL;
425 	}
426 	return sdkp;
427 }
428 
429 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
430 {
431 	struct scsi_disk *sdkp;
432 
433 	mutex_lock(&sd_ref_mutex);
434 	sdkp = __scsi_disk_get(disk);
435 	mutex_unlock(&sd_ref_mutex);
436 	return sdkp;
437 }
438 
439 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
440 {
441 	struct scsi_disk *sdkp;
442 
443 	mutex_lock(&sd_ref_mutex);
444 	sdkp = dev_get_drvdata(dev);
445 	if (sdkp)
446 		sdkp = __scsi_disk_get(sdkp->disk);
447 	mutex_unlock(&sd_ref_mutex);
448 	return sdkp;
449 }
450 
451 static void scsi_disk_put(struct scsi_disk *sdkp)
452 {
453 	struct scsi_device *sdev = sdkp->device;
454 
455 	mutex_lock(&sd_ref_mutex);
456 	put_device(&sdkp->dev);
457 	scsi_device_put(sdev);
458 	mutex_unlock(&sd_ref_mutex);
459 }
460 
461 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
462 {
463 	unsigned int prot_op = SCSI_PROT_NORMAL;
464 	unsigned int dix = scsi_prot_sg_count(scmd);
465 
466 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
467 		if (dif && dix)
468 			prot_op = SCSI_PROT_READ_PASS;
469 		else if (dif && !dix)
470 			prot_op = SCSI_PROT_READ_STRIP;
471 		else if (!dif && dix)
472 			prot_op = SCSI_PROT_READ_INSERT;
473 	} else {
474 		if (dif && dix)
475 			prot_op = SCSI_PROT_WRITE_PASS;
476 		else if (dif && !dix)
477 			prot_op = SCSI_PROT_WRITE_INSERT;
478 		else if (!dif && dix)
479 			prot_op = SCSI_PROT_WRITE_STRIP;
480 	}
481 
482 	scsi_set_prot_op(scmd, prot_op);
483 	scsi_set_prot_type(scmd, dif);
484 }
485 
486 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
487 {
488 	struct request_queue *q = sdkp->disk->queue;
489 	unsigned int logical_block_size = sdkp->device->sector_size;
490 	unsigned int max_blocks = 0;
491 
492 	q->limits.discard_zeroes_data = sdkp->lbprz;
493 	q->limits.discard_alignment = sdkp->unmap_alignment;
494 	q->limits.discard_granularity =
495 		max(sdkp->physical_block_size,
496 		    sdkp->unmap_granularity * logical_block_size);
497 
498 	switch (mode) {
499 
500 	case SD_LBP_DISABLE:
501 		q->limits.max_discard_sectors = 0;
502 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
503 		return;
504 
505 	case SD_LBP_UNMAP:
506 		max_blocks = min_not_zero(sdkp->max_unmap_blocks, 0xffffffff);
507 		break;
508 
509 	case SD_LBP_WS16:
510 		max_blocks = min_not_zero(sdkp->max_ws_blocks, 0xffffffff);
511 		break;
512 
513 	case SD_LBP_WS10:
514 		max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
515 		break;
516 
517 	case SD_LBP_ZERO:
518 		max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
519 		q->limits.discard_zeroes_data = 1;
520 		break;
521 	}
522 
523 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
524 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
525 
526 	sdkp->provisioning_mode = mode;
527 }
528 
529 /**
530  * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
531  * @sdp: scsi device to operate one
532  * @rq: Request to prepare
533  *
534  * Will issue either UNMAP or WRITE SAME(16) depending on preference
535  * indicated by target device.
536  **/
537 static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
538 {
539 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
540 	struct bio *bio = rq->bio;
541 	sector_t sector = bio->bi_sector;
542 	unsigned int nr_sectors = bio_sectors(bio);
543 	unsigned int len;
544 	int ret;
545 	char *buf;
546 	struct page *page;
547 
548 	if (sdkp->device->sector_size == 4096) {
549 		sector >>= 3;
550 		nr_sectors >>= 3;
551 	}
552 
553 	rq->timeout = SD_TIMEOUT;
554 
555 	memset(rq->cmd, 0, rq->cmd_len);
556 
557 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
558 	if (!page)
559 		return BLKPREP_DEFER;
560 
561 	switch (sdkp->provisioning_mode) {
562 	case SD_LBP_UNMAP:
563 		buf = page_address(page);
564 
565 		rq->cmd_len = 10;
566 		rq->cmd[0] = UNMAP;
567 		rq->cmd[8] = 24;
568 
569 		put_unaligned_be16(6 + 16, &buf[0]);
570 		put_unaligned_be16(16, &buf[2]);
571 		put_unaligned_be64(sector, &buf[8]);
572 		put_unaligned_be32(nr_sectors, &buf[16]);
573 
574 		len = 24;
575 		break;
576 
577 	case SD_LBP_WS16:
578 		rq->cmd_len = 16;
579 		rq->cmd[0] = WRITE_SAME_16;
580 		rq->cmd[1] = 0x8; /* UNMAP */
581 		put_unaligned_be64(sector, &rq->cmd[2]);
582 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
583 
584 		len = sdkp->device->sector_size;
585 		break;
586 
587 	case SD_LBP_WS10:
588 	case SD_LBP_ZERO:
589 		rq->cmd_len = 10;
590 		rq->cmd[0] = WRITE_SAME;
591 		if (sdkp->provisioning_mode == SD_LBP_WS10)
592 			rq->cmd[1] = 0x8; /* UNMAP */
593 		put_unaligned_be32(sector, &rq->cmd[2]);
594 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
595 
596 		len = sdkp->device->sector_size;
597 		break;
598 
599 	default:
600 		ret = BLKPREP_KILL;
601 		goto out;
602 	}
603 
604 	blk_add_request_payload(rq, page, len);
605 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
606 	rq->buffer = page_address(page);
607 
608 out:
609 	if (ret != BLKPREP_OK) {
610 		__free_page(page);
611 		rq->buffer = NULL;
612 	}
613 	return ret;
614 }
615 
616 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
617 {
618 	rq->timeout = SD_FLUSH_TIMEOUT;
619 	rq->retries = SD_MAX_RETRIES;
620 	rq->cmd[0] = SYNCHRONIZE_CACHE;
621 	rq->cmd_len = 10;
622 
623 	return scsi_setup_blk_pc_cmnd(sdp, rq);
624 }
625 
626 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
627 {
628 	if (rq->cmd_flags & REQ_DISCARD) {
629 		free_page((unsigned long)rq->buffer);
630 		rq->buffer = NULL;
631 	}
632 }
633 
634 /**
635  *	sd_init_command - build a scsi (read or write) command from
636  *	information in the request structure.
637  *	@SCpnt: pointer to mid-level's per scsi command structure that
638  *	contains request and into which the scsi command is written
639  *
640  *	Returns 1 if successful and 0 if error (or cannot be done now).
641  **/
642 static int sd_prep_fn(struct request_queue *q, struct request *rq)
643 {
644 	struct scsi_cmnd *SCpnt;
645 	struct scsi_device *sdp = q->queuedata;
646 	struct gendisk *disk = rq->rq_disk;
647 	struct scsi_disk *sdkp;
648 	sector_t block = blk_rq_pos(rq);
649 	sector_t threshold;
650 	unsigned int this_count = blk_rq_sectors(rq);
651 	int ret, host_dif;
652 	unsigned char protect;
653 
654 	/*
655 	 * Discard request come in as REQ_TYPE_FS but we turn them into
656 	 * block PC requests to make life easier.
657 	 */
658 	if (rq->cmd_flags & REQ_DISCARD) {
659 		ret = scsi_setup_discard_cmnd(sdp, rq);
660 		goto out;
661 	} else if (rq->cmd_flags & REQ_FLUSH) {
662 		ret = scsi_setup_flush_cmnd(sdp, rq);
663 		goto out;
664 	} else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
665 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
666 		goto out;
667 	} else if (rq->cmd_type != REQ_TYPE_FS) {
668 		ret = BLKPREP_KILL;
669 		goto out;
670 	}
671 	ret = scsi_setup_fs_cmnd(sdp, rq);
672 	if (ret != BLKPREP_OK)
673 		goto out;
674 	SCpnt = rq->special;
675 	sdkp = scsi_disk(disk);
676 
677 	/* from here on until we're complete, any goto out
678 	 * is used for a killable error condition */
679 	ret = BLKPREP_KILL;
680 
681 	SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
682 					"sd_init_command: block=%llu, "
683 					"count=%d\n",
684 					(unsigned long long)block,
685 					this_count));
686 
687 	if (!sdp || !scsi_device_online(sdp) ||
688 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
689 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
690 						"Finishing %u sectors\n",
691 						blk_rq_sectors(rq)));
692 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
693 						"Retry with 0x%p\n", SCpnt));
694 		goto out;
695 	}
696 
697 	if (sdp->changed) {
698 		/*
699 		 * quietly refuse to do anything to a changed disc until
700 		 * the changed bit has been reset
701 		 */
702 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
703 		goto out;
704 	}
705 
706 	/*
707 	 * Some SD card readers can't handle multi-sector accesses which touch
708 	 * the last one or two hardware sectors.  Split accesses as needed.
709 	 */
710 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
711 		(sdp->sector_size / 512);
712 
713 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
714 		if (block < threshold) {
715 			/* Access up to the threshold but not beyond */
716 			this_count = threshold - block;
717 		} else {
718 			/* Access only a single hardware sector */
719 			this_count = sdp->sector_size / 512;
720 		}
721 	}
722 
723 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
724 					(unsigned long long)block));
725 
726 	/*
727 	 * If we have a 1K hardware sectorsize, prevent access to single
728 	 * 512 byte sectors.  In theory we could handle this - in fact
729 	 * the scsi cdrom driver must be able to handle this because
730 	 * we typically use 1K blocksizes, and cdroms typically have
731 	 * 2K hardware sectorsizes.  Of course, things are simpler
732 	 * with the cdrom, since it is read-only.  For performance
733 	 * reasons, the filesystems should be able to handle this
734 	 * and not force the scsi disk driver to use bounce buffers
735 	 * for this.
736 	 */
737 	if (sdp->sector_size == 1024) {
738 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
739 			scmd_printk(KERN_ERR, SCpnt,
740 				    "Bad block number requested\n");
741 			goto out;
742 		} else {
743 			block = block >> 1;
744 			this_count = this_count >> 1;
745 		}
746 	}
747 	if (sdp->sector_size == 2048) {
748 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
749 			scmd_printk(KERN_ERR, SCpnt,
750 				    "Bad block number requested\n");
751 			goto out;
752 		} else {
753 			block = block >> 2;
754 			this_count = this_count >> 2;
755 		}
756 	}
757 	if (sdp->sector_size == 4096) {
758 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
759 			scmd_printk(KERN_ERR, SCpnt,
760 				    "Bad block number requested\n");
761 			goto out;
762 		} else {
763 			block = block >> 3;
764 			this_count = this_count >> 3;
765 		}
766 	}
767 	if (rq_data_dir(rq) == WRITE) {
768 		if (!sdp->writeable) {
769 			goto out;
770 		}
771 		SCpnt->cmnd[0] = WRITE_6;
772 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
773 
774 		if (blk_integrity_rq(rq) &&
775 		    sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
776 			goto out;
777 
778 	} else if (rq_data_dir(rq) == READ) {
779 		SCpnt->cmnd[0] = READ_6;
780 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
781 	} else {
782 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
783 		goto out;
784 	}
785 
786 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
787 					"%s %d/%u 512 byte blocks.\n",
788 					(rq_data_dir(rq) == WRITE) ?
789 					"writing" : "reading", this_count,
790 					blk_rq_sectors(rq)));
791 
792 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
793 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
794 	if (host_dif)
795 		protect = 1 << 5;
796 	else
797 		protect = 0;
798 
799 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
800 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
801 
802 		if (unlikely(SCpnt->cmnd == NULL)) {
803 			ret = BLKPREP_DEFER;
804 			goto out;
805 		}
806 
807 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
808 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
809 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
810 		SCpnt->cmnd[7] = 0x18;
811 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
812 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
813 
814 		/* LBA */
815 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
816 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
817 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
818 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
819 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
820 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
821 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
822 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
823 
824 		/* Expected Indirect LBA */
825 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
826 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
827 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
828 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
829 
830 		/* Transfer length */
831 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
832 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
833 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
834 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
835 	} else if (block > 0xffffffff) {
836 		SCpnt->cmnd[0] += READ_16 - READ_6;
837 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
838 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
839 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
840 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
841 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
842 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
843 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
844 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
845 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
846 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
847 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
848 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
849 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
850 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
851 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
852 		   scsi_device_protection(SCpnt->device) ||
853 		   SCpnt->device->use_10_for_rw) {
854 		if (this_count > 0xffff)
855 			this_count = 0xffff;
856 
857 		SCpnt->cmnd[0] += READ_10 - READ_6;
858 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
859 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
860 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
861 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
862 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
863 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
864 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
865 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
866 	} else {
867 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
868 			/*
869 			 * This happens only if this drive failed
870 			 * 10byte rw command with ILLEGAL_REQUEST
871 			 * during operation and thus turned off
872 			 * use_10_for_rw.
873 			 */
874 			scmd_printk(KERN_ERR, SCpnt,
875 				    "FUA write on READ/WRITE(6) drive\n");
876 			goto out;
877 		}
878 
879 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
880 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
881 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
882 		SCpnt->cmnd[4] = (unsigned char) this_count;
883 		SCpnt->cmnd[5] = 0;
884 	}
885 	SCpnt->sdb.length = this_count * sdp->sector_size;
886 
887 	/* If DIF or DIX is enabled, tell HBA how to handle request */
888 	if (host_dif || scsi_prot_sg_count(SCpnt))
889 		sd_prot_op(SCpnt, host_dif);
890 
891 	/*
892 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
893 	 * host adapter, it's safe to assume that we can at least transfer
894 	 * this many bytes between each connect / disconnect.
895 	 */
896 	SCpnt->transfersize = sdp->sector_size;
897 	SCpnt->underflow = this_count << 9;
898 	SCpnt->allowed = SD_MAX_RETRIES;
899 
900 	/*
901 	 * This indicates that the command is ready from our end to be
902 	 * queued.
903 	 */
904 	ret = BLKPREP_OK;
905  out:
906 	return scsi_prep_return(q, rq, ret);
907 }
908 
909 /**
910  *	sd_open - open a scsi disk device
911  *	@inode: only i_rdev member may be used
912  *	@filp: only f_mode and f_flags may be used
913  *
914  *	Returns 0 if successful. Returns a negated errno value in case
915  *	of error.
916  *
917  *	Note: This can be called from a user context (e.g. fsck(1) )
918  *	or from within the kernel (e.g. as a result of a mount(1) ).
919  *	In the latter case @inode and @filp carry an abridged amount
920  *	of information as noted above.
921  *
922  *	Locking: called with bdev->bd_mutex held.
923  **/
924 static int sd_open(struct block_device *bdev, fmode_t mode)
925 {
926 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
927 	struct scsi_device *sdev;
928 	int retval;
929 
930 	if (!sdkp)
931 		return -ENXIO;
932 
933 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
934 
935 	sdev = sdkp->device;
936 
937 	retval = scsi_autopm_get_device(sdev);
938 	if (retval)
939 		goto error_autopm;
940 
941 	/*
942 	 * If the device is in error recovery, wait until it is done.
943 	 * If the device is offline, then disallow any access to it.
944 	 */
945 	retval = -ENXIO;
946 	if (!scsi_block_when_processing_errors(sdev))
947 		goto error_out;
948 
949 	if (sdev->removable || sdkp->write_prot)
950 		check_disk_change(bdev);
951 
952 	/*
953 	 * If the drive is empty, just let the open fail.
954 	 */
955 	retval = -ENOMEDIUM;
956 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
957 		goto error_out;
958 
959 	/*
960 	 * If the device has the write protect tab set, have the open fail
961 	 * if the user expects to be able to write to the thing.
962 	 */
963 	retval = -EROFS;
964 	if (sdkp->write_prot && (mode & FMODE_WRITE))
965 		goto error_out;
966 
967 	/*
968 	 * It is possible that the disk changing stuff resulted in
969 	 * the device being taken offline.  If this is the case,
970 	 * report this to the user, and don't pretend that the
971 	 * open actually succeeded.
972 	 */
973 	retval = -ENXIO;
974 	if (!scsi_device_online(sdev))
975 		goto error_out;
976 
977 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
978 		if (scsi_block_when_processing_errors(sdev))
979 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
980 	}
981 
982 	return 0;
983 
984 error_out:
985 	scsi_autopm_put_device(sdev);
986 error_autopm:
987 	scsi_disk_put(sdkp);
988 	return retval;
989 }
990 
991 /**
992  *	sd_release - invoked when the (last) close(2) is called on this
993  *	scsi disk.
994  *	@inode: only i_rdev member may be used
995  *	@filp: only f_mode and f_flags may be used
996  *
997  *	Returns 0.
998  *
999  *	Note: may block (uninterruptible) if error recovery is underway
1000  *	on this disk.
1001  *
1002  *	Locking: called with bdev->bd_mutex held.
1003  **/
1004 static int sd_release(struct gendisk *disk, fmode_t mode)
1005 {
1006 	struct scsi_disk *sdkp = scsi_disk(disk);
1007 	struct scsi_device *sdev = sdkp->device;
1008 
1009 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1010 
1011 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1012 		if (scsi_block_when_processing_errors(sdev))
1013 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1014 	}
1015 
1016 	/*
1017 	 * XXX and what if there are packets in flight and this close()
1018 	 * XXX is followed by a "rmmod sd_mod"?
1019 	 */
1020 
1021 	scsi_autopm_put_device(sdev);
1022 	scsi_disk_put(sdkp);
1023 	return 0;
1024 }
1025 
1026 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1027 {
1028 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1029 	struct scsi_device *sdp = sdkp->device;
1030 	struct Scsi_Host *host = sdp->host;
1031 	int diskinfo[4];
1032 
1033 	/* default to most commonly used values */
1034         diskinfo[0] = 0x40;	/* 1 << 6 */
1035        	diskinfo[1] = 0x20;	/* 1 << 5 */
1036        	diskinfo[2] = sdkp->capacity >> 11;
1037 
1038 	/* override with calculated, extended default, or driver values */
1039 	if (host->hostt->bios_param)
1040 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1041 	else
1042 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1043 
1044 	geo->heads = diskinfo[0];
1045 	geo->sectors = diskinfo[1];
1046 	geo->cylinders = diskinfo[2];
1047 	return 0;
1048 }
1049 
1050 /**
1051  *	sd_ioctl - process an ioctl
1052  *	@inode: only i_rdev/i_bdev members may be used
1053  *	@filp: only f_mode and f_flags may be used
1054  *	@cmd: ioctl command number
1055  *	@arg: this is third argument given to ioctl(2) system call.
1056  *	Often contains a pointer.
1057  *
1058  *	Returns 0 if successful (some ioctls return positive numbers on
1059  *	success as well). Returns a negated errno value in case of error.
1060  *
1061  *	Note: most ioctls are forward onto the block subsystem or further
1062  *	down in the scsi subsystem.
1063  **/
1064 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1065 		    unsigned int cmd, unsigned long arg)
1066 {
1067 	struct gendisk *disk = bdev->bd_disk;
1068 	struct scsi_device *sdp = scsi_disk(disk)->device;
1069 	void __user *p = (void __user *)arg;
1070 	int error;
1071 
1072 	SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
1073 						disk->disk_name, cmd));
1074 
1075 	/*
1076 	 * If we are in the middle of error recovery, don't let anyone
1077 	 * else try and use this device.  Also, if error recovery fails, it
1078 	 * may try and take the device offline, in which case all further
1079 	 * access to the device is prohibited.
1080 	 */
1081 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1082 					(mode & FMODE_NDELAY) != 0);
1083 	if (!scsi_block_when_processing_errors(sdp) || !error)
1084 		goto out;
1085 
1086 	/*
1087 	 * Send SCSI addressing ioctls directly to mid level, send other
1088 	 * ioctls to block level and then onto mid level if they can't be
1089 	 * resolved.
1090 	 */
1091 	switch (cmd) {
1092 		case SCSI_IOCTL_GET_IDLUN:
1093 		case SCSI_IOCTL_GET_BUS_NUMBER:
1094 			error = scsi_ioctl(sdp, cmd, p);
1095 			break;
1096 		default:
1097 			error = scsi_cmd_ioctl(disk->queue, disk, mode, cmd, p);
1098 			if (error != -ENOTTY)
1099 				break;
1100 			error = scsi_ioctl(sdp, cmd, p);
1101 			break;
1102 	}
1103 out:
1104 	return error;
1105 }
1106 
1107 static void set_media_not_present(struct scsi_disk *sdkp)
1108 {
1109 	if (sdkp->media_present)
1110 		sdkp->device->changed = 1;
1111 
1112 	if (sdkp->device->removable) {
1113 		sdkp->media_present = 0;
1114 		sdkp->capacity = 0;
1115 	}
1116 }
1117 
1118 static int media_not_present(struct scsi_disk *sdkp,
1119 			     struct scsi_sense_hdr *sshdr)
1120 {
1121 	if (!scsi_sense_valid(sshdr))
1122 		return 0;
1123 
1124 	/* not invoked for commands that could return deferred errors */
1125 	switch (sshdr->sense_key) {
1126 	case UNIT_ATTENTION:
1127 	case NOT_READY:
1128 		/* medium not present */
1129 		if (sshdr->asc == 0x3A) {
1130 			set_media_not_present(sdkp);
1131 			return 1;
1132 		}
1133 	}
1134 	return 0;
1135 }
1136 
1137 /**
1138  *	sd_check_events - check media events
1139  *	@disk: kernel device descriptor
1140  *	@clearing: disk events currently being cleared
1141  *
1142  *	Returns mask of DISK_EVENT_*.
1143  *
1144  *	Note: this function is invoked from the block subsystem.
1145  **/
1146 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1147 {
1148 	struct scsi_disk *sdkp = scsi_disk(disk);
1149 	struct scsi_device *sdp = sdkp->device;
1150 	struct scsi_sense_hdr *sshdr = NULL;
1151 	int retval;
1152 
1153 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1154 
1155 	/*
1156 	 * If the device is offline, don't send any commands - just pretend as
1157 	 * if the command failed.  If the device ever comes back online, we
1158 	 * can deal with it then.  It is only because of unrecoverable errors
1159 	 * that we would ever take a device offline in the first place.
1160 	 */
1161 	if (!scsi_device_online(sdp)) {
1162 		set_media_not_present(sdkp);
1163 		goto out;
1164 	}
1165 
1166 	/*
1167 	 * Using TEST_UNIT_READY enables differentiation between drive with
1168 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1169 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1170 	 *
1171 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1172 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1173 	 * sd_revalidate() is called.
1174 	 */
1175 	retval = -ENODEV;
1176 
1177 	if (scsi_block_when_processing_errors(sdp)) {
1178 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1179 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1180 					      sshdr);
1181 	}
1182 
1183 	/* failed to execute TUR, assume media not present */
1184 	if (host_byte(retval)) {
1185 		set_media_not_present(sdkp);
1186 		goto out;
1187 	}
1188 
1189 	if (media_not_present(sdkp, sshdr))
1190 		goto out;
1191 
1192 	/*
1193 	 * For removable scsi disk we have to recognise the presence
1194 	 * of a disk in the drive.
1195 	 */
1196 	if (!sdkp->media_present)
1197 		sdp->changed = 1;
1198 	sdkp->media_present = 1;
1199 out:
1200 	/*
1201 	 * sdp->changed is set under the following conditions:
1202 	 *
1203 	 *	Medium present state has changed in either direction.
1204 	 *	Device has indicated UNIT_ATTENTION.
1205 	 */
1206 	kfree(sshdr);
1207 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1208 	sdp->changed = 0;
1209 	return retval;
1210 }
1211 
1212 static int sd_sync_cache(struct scsi_disk *sdkp)
1213 {
1214 	int retries, res;
1215 	struct scsi_device *sdp = sdkp->device;
1216 	struct scsi_sense_hdr sshdr;
1217 
1218 	if (!scsi_device_online(sdp))
1219 		return -ENODEV;
1220 
1221 
1222 	for (retries = 3; retries > 0; --retries) {
1223 		unsigned char cmd[10] = { 0 };
1224 
1225 		cmd[0] = SYNCHRONIZE_CACHE;
1226 		/*
1227 		 * Leave the rest of the command zero to indicate
1228 		 * flush everything.
1229 		 */
1230 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1231 				       SD_FLUSH_TIMEOUT, SD_MAX_RETRIES, NULL);
1232 		if (res == 0)
1233 			break;
1234 	}
1235 
1236 	if (res) {
1237 		sd_print_result(sdkp, res);
1238 		if (driver_byte(res) & DRIVER_SENSE)
1239 			sd_print_sense_hdr(sdkp, &sshdr);
1240 	}
1241 
1242 	if (res)
1243 		return -EIO;
1244 	return 0;
1245 }
1246 
1247 static void sd_rescan(struct device *dev)
1248 {
1249 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1250 
1251 	if (sdkp) {
1252 		revalidate_disk(sdkp->disk);
1253 		scsi_disk_put(sdkp);
1254 	}
1255 }
1256 
1257 
1258 #ifdef CONFIG_COMPAT
1259 /*
1260  * This gets directly called from VFS. When the ioctl
1261  * is not recognized we go back to the other translation paths.
1262  */
1263 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1264 			   unsigned int cmd, unsigned long arg)
1265 {
1266 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1267 
1268 	/*
1269 	 * If we are in the middle of error recovery, don't let anyone
1270 	 * else try and use this device.  Also, if error recovery fails, it
1271 	 * may try and take the device offline, in which case all further
1272 	 * access to the device is prohibited.
1273 	 */
1274 	if (!scsi_block_when_processing_errors(sdev))
1275 		return -ENODEV;
1276 
1277 	if (sdev->host->hostt->compat_ioctl) {
1278 		int ret;
1279 
1280 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1281 
1282 		return ret;
1283 	}
1284 
1285 	/*
1286 	 * Let the static ioctl translation table take care of it.
1287 	 */
1288 	return -ENOIOCTLCMD;
1289 }
1290 #endif
1291 
1292 static const struct block_device_operations sd_fops = {
1293 	.owner			= THIS_MODULE,
1294 	.open			= sd_open,
1295 	.release		= sd_release,
1296 	.ioctl			= sd_ioctl,
1297 	.getgeo			= sd_getgeo,
1298 #ifdef CONFIG_COMPAT
1299 	.compat_ioctl		= sd_compat_ioctl,
1300 #endif
1301 	.check_events		= sd_check_events,
1302 	.revalidate_disk	= sd_revalidate_disk,
1303 	.unlock_native_capacity	= sd_unlock_native_capacity,
1304 };
1305 
1306 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1307 {
1308 	u64 start_lba = blk_rq_pos(scmd->request);
1309 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1310 	u64 bad_lba;
1311 	int info_valid;
1312 	/*
1313 	 * resid is optional but mostly filled in.  When it's unused,
1314 	 * its value is zero, so we assume the whole buffer transferred
1315 	 */
1316 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1317 	unsigned int good_bytes;
1318 
1319 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1320 		return 0;
1321 
1322 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1323 					     SCSI_SENSE_BUFFERSIZE,
1324 					     &bad_lba);
1325 	if (!info_valid)
1326 		return 0;
1327 
1328 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1329 		return 0;
1330 
1331 	if (scmd->device->sector_size < 512) {
1332 		/* only legitimate sector_size here is 256 */
1333 		start_lba <<= 1;
1334 		end_lba <<= 1;
1335 	} else {
1336 		/* be careful ... don't want any overflows */
1337 		u64 factor = scmd->device->sector_size / 512;
1338 		do_div(start_lba, factor);
1339 		do_div(end_lba, factor);
1340 	}
1341 
1342 	/* The bad lba was reported incorrectly, we have no idea where
1343 	 * the error is.
1344 	 */
1345 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1346 		return 0;
1347 
1348 	/* This computation should always be done in terms of
1349 	 * the resolution of the device's medium.
1350 	 */
1351 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1352 	return min(good_bytes, transferred);
1353 }
1354 
1355 /**
1356  *	sd_done - bottom half handler: called when the lower level
1357  *	driver has completed (successfully or otherwise) a scsi command.
1358  *	@SCpnt: mid-level's per command structure.
1359  *
1360  *	Note: potentially run from within an ISR. Must not block.
1361  **/
1362 static int sd_done(struct scsi_cmnd *SCpnt)
1363 {
1364 	int result = SCpnt->result;
1365 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1366 	struct scsi_sense_hdr sshdr;
1367 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1368 	int sense_valid = 0;
1369 	int sense_deferred = 0;
1370 	unsigned char op = SCpnt->cmnd[0];
1371 
1372 	if ((SCpnt->request->cmd_flags & REQ_DISCARD) && !result)
1373 		scsi_set_resid(SCpnt, 0);
1374 
1375 	if (result) {
1376 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1377 		if (sense_valid)
1378 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1379 	}
1380 #ifdef CONFIG_SCSI_LOGGING
1381 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1382 	if (sense_valid) {
1383 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1384 						   "sd_done: sb[respc,sk,asc,"
1385 						   "ascq]=%x,%x,%x,%x\n",
1386 						   sshdr.response_code,
1387 						   sshdr.sense_key, sshdr.asc,
1388 						   sshdr.ascq));
1389 	}
1390 #endif
1391 	if (driver_byte(result) != DRIVER_SENSE &&
1392 	    (!sense_valid || sense_deferred))
1393 		goto out;
1394 
1395 	switch (sshdr.sense_key) {
1396 	case HARDWARE_ERROR:
1397 	case MEDIUM_ERROR:
1398 		good_bytes = sd_completed_bytes(SCpnt);
1399 		break;
1400 	case RECOVERED_ERROR:
1401 		good_bytes = scsi_bufflen(SCpnt);
1402 		break;
1403 	case NO_SENSE:
1404 		/* This indicates a false check condition, so ignore it.  An
1405 		 * unknown amount of data was transferred so treat it as an
1406 		 * error.
1407 		 */
1408 		scsi_print_sense("sd", SCpnt);
1409 		SCpnt->result = 0;
1410 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1411 		break;
1412 	case ABORTED_COMMAND:
1413 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1414 			good_bytes = sd_completed_bytes(SCpnt);
1415 		break;
1416 	case ILLEGAL_REQUEST:
1417 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1418 			good_bytes = sd_completed_bytes(SCpnt);
1419 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1420 		if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1421 		    (op == UNMAP || op == WRITE_SAME_16 || op == WRITE_SAME))
1422 			sd_config_discard(sdkp, SD_LBP_DISABLE);
1423 		break;
1424 	default:
1425 		break;
1426 	}
1427  out:
1428 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1429 		sd_dif_complete(SCpnt, good_bytes);
1430 
1431 	if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1432 	    == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1433 
1434 		/* We have to print a failed command here as the
1435 		 * extended CDB gets freed before scsi_io_completion()
1436 		 * is called.
1437 		 */
1438 		if (result)
1439 			scsi_print_command(SCpnt);
1440 
1441 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1442 		SCpnt->cmnd = NULL;
1443 		SCpnt->cmd_len = 0;
1444 	}
1445 
1446 	return good_bytes;
1447 }
1448 
1449 /*
1450  * spinup disk - called only in sd_revalidate_disk()
1451  */
1452 static void
1453 sd_spinup_disk(struct scsi_disk *sdkp)
1454 {
1455 	unsigned char cmd[10];
1456 	unsigned long spintime_expire = 0;
1457 	int retries, spintime;
1458 	unsigned int the_result;
1459 	struct scsi_sense_hdr sshdr;
1460 	int sense_valid = 0;
1461 
1462 	spintime = 0;
1463 
1464 	/* Spin up drives, as required.  Only do this at boot time */
1465 	/* Spinup needs to be done for module loads too. */
1466 	do {
1467 		retries = 0;
1468 
1469 		do {
1470 			cmd[0] = TEST_UNIT_READY;
1471 			memset((void *) &cmd[1], 0, 9);
1472 
1473 			the_result = scsi_execute_req(sdkp->device, cmd,
1474 						      DMA_NONE, NULL, 0,
1475 						      &sshdr, SD_TIMEOUT,
1476 						      SD_MAX_RETRIES, NULL);
1477 
1478 			/*
1479 			 * If the drive has indicated to us that it
1480 			 * doesn't have any media in it, don't bother
1481 			 * with any more polling.
1482 			 */
1483 			if (media_not_present(sdkp, &sshdr))
1484 				return;
1485 
1486 			if (the_result)
1487 				sense_valid = scsi_sense_valid(&sshdr);
1488 			retries++;
1489 		} while (retries < 3 &&
1490 			 (!scsi_status_is_good(the_result) ||
1491 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1492 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1493 
1494 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1495 			/* no sense, TUR either succeeded or failed
1496 			 * with a status error */
1497 			if(!spintime && !scsi_status_is_good(the_result)) {
1498 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1499 				sd_print_result(sdkp, the_result);
1500 			}
1501 			break;
1502 		}
1503 
1504 		/*
1505 		 * The device does not want the automatic start to be issued.
1506 		 */
1507 		if (sdkp->device->no_start_on_add)
1508 			break;
1509 
1510 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1511 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1512 				break;	/* manual intervention required */
1513 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1514 				break;	/* standby */
1515 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1516 				break;	/* unavailable */
1517 			/*
1518 			 * Issue command to spin up drive when not ready
1519 			 */
1520 			if (!spintime) {
1521 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1522 				cmd[0] = START_STOP;
1523 				cmd[1] = 1;	/* Return immediately */
1524 				memset((void *) &cmd[2], 0, 8);
1525 				cmd[4] = 1;	/* Start spin cycle */
1526 				if (sdkp->device->start_stop_pwr_cond)
1527 					cmd[4] |= 1 << 4;
1528 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1529 						 NULL, 0, &sshdr,
1530 						 SD_TIMEOUT, SD_MAX_RETRIES,
1531 						 NULL);
1532 				spintime_expire = jiffies + 100 * HZ;
1533 				spintime = 1;
1534 			}
1535 			/* Wait 1 second for next try */
1536 			msleep(1000);
1537 			printk(".");
1538 
1539 		/*
1540 		 * Wait for USB flash devices with slow firmware.
1541 		 * Yes, this sense key/ASC combination shouldn't
1542 		 * occur here.  It's characteristic of these devices.
1543 		 */
1544 		} else if (sense_valid &&
1545 				sshdr.sense_key == UNIT_ATTENTION &&
1546 				sshdr.asc == 0x28) {
1547 			if (!spintime) {
1548 				spintime_expire = jiffies + 5 * HZ;
1549 				spintime = 1;
1550 			}
1551 			/* Wait 1 second for next try */
1552 			msleep(1000);
1553 		} else {
1554 			/* we don't understand the sense code, so it's
1555 			 * probably pointless to loop */
1556 			if(!spintime) {
1557 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1558 				sd_print_sense_hdr(sdkp, &sshdr);
1559 			}
1560 			break;
1561 		}
1562 
1563 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1564 
1565 	if (spintime) {
1566 		if (scsi_status_is_good(the_result))
1567 			printk("ready\n");
1568 		else
1569 			printk("not responding...\n");
1570 	}
1571 }
1572 
1573 
1574 /*
1575  * Determine whether disk supports Data Integrity Field.
1576  */
1577 static void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1578 {
1579 	struct scsi_device *sdp = sdkp->device;
1580 	u8 type;
1581 
1582 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1583 		return;
1584 
1585 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1586 
1587 	if (type == sdkp->protection_type || !sdkp->first_scan)
1588 		return;
1589 
1590 	sdkp->protection_type = type;
1591 
1592 	if (type > SD_DIF_TYPE3_PROTECTION) {
1593 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported "	\
1594 			  "protection type %u. Disabling disk!\n", type);
1595 		sdkp->capacity = 0;
1596 		return;
1597 	}
1598 
1599 	if (scsi_host_dif_capable(sdp->host, type))
1600 		sd_printk(KERN_NOTICE, sdkp,
1601 			  "Enabling DIF Type %u protection\n", type);
1602 	else
1603 		sd_printk(KERN_NOTICE, sdkp,
1604 			  "Disabling DIF Type %u protection\n", type);
1605 }
1606 
1607 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1608 			struct scsi_sense_hdr *sshdr, int sense_valid,
1609 			int the_result)
1610 {
1611 	sd_print_result(sdkp, the_result);
1612 	if (driver_byte(the_result) & DRIVER_SENSE)
1613 		sd_print_sense_hdr(sdkp, sshdr);
1614 	else
1615 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1616 
1617 	/*
1618 	 * Set dirty bit for removable devices if not ready -
1619 	 * sometimes drives will not report this properly.
1620 	 */
1621 	if (sdp->removable &&
1622 	    sense_valid && sshdr->sense_key == NOT_READY)
1623 		set_media_not_present(sdkp);
1624 
1625 	/*
1626 	 * We used to set media_present to 0 here to indicate no media
1627 	 * in the drive, but some drives fail read capacity even with
1628 	 * media present, so we can't do that.
1629 	 */
1630 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1631 }
1632 
1633 #define RC16_LEN 32
1634 #if RC16_LEN > SD_BUF_SIZE
1635 #error RC16_LEN must not be more than SD_BUF_SIZE
1636 #endif
1637 
1638 #define READ_CAPACITY_RETRIES_ON_RESET	10
1639 
1640 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1641 						unsigned char *buffer)
1642 {
1643 	unsigned char cmd[16];
1644 	struct scsi_sense_hdr sshdr;
1645 	int sense_valid = 0;
1646 	int the_result;
1647 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1648 	unsigned int alignment;
1649 	unsigned long long lba;
1650 	unsigned sector_size;
1651 
1652 	if (sdp->no_read_capacity_16)
1653 		return -EINVAL;
1654 
1655 	do {
1656 		memset(cmd, 0, 16);
1657 		cmd[0] = SERVICE_ACTION_IN;
1658 		cmd[1] = SAI_READ_CAPACITY_16;
1659 		cmd[13] = RC16_LEN;
1660 		memset(buffer, 0, RC16_LEN);
1661 
1662 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1663 					buffer, RC16_LEN, &sshdr,
1664 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1665 
1666 		if (media_not_present(sdkp, &sshdr))
1667 			return -ENODEV;
1668 
1669 		if (the_result) {
1670 			sense_valid = scsi_sense_valid(&sshdr);
1671 			if (sense_valid &&
1672 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1673 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1674 			    sshdr.ascq == 0x00)
1675 				/* Invalid Command Operation Code or
1676 				 * Invalid Field in CDB, just retry
1677 				 * silently with RC10 */
1678 				return -EINVAL;
1679 			if (sense_valid &&
1680 			    sshdr.sense_key == UNIT_ATTENTION &&
1681 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1682 				/* Device reset might occur several times,
1683 				 * give it one more chance */
1684 				if (--reset_retries > 0)
1685 					continue;
1686 		}
1687 		retries--;
1688 
1689 	} while (the_result && retries);
1690 
1691 	if (the_result) {
1692 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1693 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1694 		return -EINVAL;
1695 	}
1696 
1697 	sector_size = get_unaligned_be32(&buffer[8]);
1698 	lba = get_unaligned_be64(&buffer[0]);
1699 
1700 	sd_read_protection_type(sdkp, buffer);
1701 
1702 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1703 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1704 			"kernel compiled with support for large block "
1705 			"devices.\n");
1706 		sdkp->capacity = 0;
1707 		return -EOVERFLOW;
1708 	}
1709 
1710 	/* Logical blocks per physical block exponent */
1711 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
1712 
1713 	/* Lowest aligned logical block */
1714 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1715 	blk_queue_alignment_offset(sdp->request_queue, alignment);
1716 	if (alignment && sdkp->first_scan)
1717 		sd_printk(KERN_NOTICE, sdkp,
1718 			  "physical block alignment offset: %u\n", alignment);
1719 
1720 	if (buffer[14] & 0x80) { /* LBPME */
1721 		sdkp->lbpme = 1;
1722 
1723 		if (buffer[14] & 0x40) /* LBPRZ */
1724 			sdkp->lbprz = 1;
1725 
1726 		sd_config_discard(sdkp, SD_LBP_WS16);
1727 	}
1728 
1729 	sdkp->capacity = lba + 1;
1730 	return sector_size;
1731 }
1732 
1733 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1734 						unsigned char *buffer)
1735 {
1736 	unsigned char cmd[16];
1737 	struct scsi_sense_hdr sshdr;
1738 	int sense_valid = 0;
1739 	int the_result;
1740 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1741 	sector_t lba;
1742 	unsigned sector_size;
1743 
1744 	do {
1745 		cmd[0] = READ_CAPACITY;
1746 		memset(&cmd[1], 0, 9);
1747 		memset(buffer, 0, 8);
1748 
1749 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1750 					buffer, 8, &sshdr,
1751 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1752 
1753 		if (media_not_present(sdkp, &sshdr))
1754 			return -ENODEV;
1755 
1756 		if (the_result) {
1757 			sense_valid = scsi_sense_valid(&sshdr);
1758 			if (sense_valid &&
1759 			    sshdr.sense_key == UNIT_ATTENTION &&
1760 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1761 				/* Device reset might occur several times,
1762 				 * give it one more chance */
1763 				if (--reset_retries > 0)
1764 					continue;
1765 		}
1766 		retries--;
1767 
1768 	} while (the_result && retries);
1769 
1770 	if (the_result) {
1771 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1772 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1773 		return -EINVAL;
1774 	}
1775 
1776 	sector_size = get_unaligned_be32(&buffer[4]);
1777 	lba = get_unaligned_be32(&buffer[0]);
1778 
1779 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
1780 		/* Some buggy (usb cardreader) devices return an lba of
1781 		   0xffffffff when the want to report a size of 0 (with
1782 		   which they really mean no media is present) */
1783 		sdkp->capacity = 0;
1784 		sdkp->physical_block_size = sector_size;
1785 		return sector_size;
1786 	}
1787 
1788 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1789 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1790 			"kernel compiled with support for large block "
1791 			"devices.\n");
1792 		sdkp->capacity = 0;
1793 		return -EOVERFLOW;
1794 	}
1795 
1796 	sdkp->capacity = lba + 1;
1797 	sdkp->physical_block_size = sector_size;
1798 	return sector_size;
1799 }
1800 
1801 static int sd_try_rc16_first(struct scsi_device *sdp)
1802 {
1803 	if (sdp->host->max_cmd_len < 16)
1804 		return 0;
1805 	if (sdp->scsi_level > SCSI_SPC_2)
1806 		return 1;
1807 	if (scsi_device_protection(sdp))
1808 		return 1;
1809 	return 0;
1810 }
1811 
1812 /*
1813  * read disk capacity
1814  */
1815 static void
1816 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1817 {
1818 	int sector_size;
1819 	struct scsi_device *sdp = sdkp->device;
1820 	sector_t old_capacity = sdkp->capacity;
1821 
1822 	if (sd_try_rc16_first(sdp)) {
1823 		sector_size = read_capacity_16(sdkp, sdp, buffer);
1824 		if (sector_size == -EOVERFLOW)
1825 			goto got_data;
1826 		if (sector_size == -ENODEV)
1827 			return;
1828 		if (sector_size < 0)
1829 			sector_size = read_capacity_10(sdkp, sdp, buffer);
1830 		if (sector_size < 0)
1831 			return;
1832 	} else {
1833 		sector_size = read_capacity_10(sdkp, sdp, buffer);
1834 		if (sector_size == -EOVERFLOW)
1835 			goto got_data;
1836 		if (sector_size < 0)
1837 			return;
1838 		if ((sizeof(sdkp->capacity) > 4) &&
1839 		    (sdkp->capacity > 0xffffffffULL)) {
1840 			int old_sector_size = sector_size;
1841 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1842 					"Trying to use READ CAPACITY(16).\n");
1843 			sector_size = read_capacity_16(sdkp, sdp, buffer);
1844 			if (sector_size < 0) {
1845 				sd_printk(KERN_NOTICE, sdkp,
1846 					"Using 0xffffffff as device size\n");
1847 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
1848 				sector_size = old_sector_size;
1849 				goto got_data;
1850 			}
1851 		}
1852 	}
1853 
1854 	/* Some devices are known to return the total number of blocks,
1855 	 * not the highest block number.  Some devices have versions
1856 	 * which do this and others which do not.  Some devices we might
1857 	 * suspect of doing this but we don't know for certain.
1858 	 *
1859 	 * If we know the reported capacity is wrong, decrement it.  If
1860 	 * we can only guess, then assume the number of blocks is even
1861 	 * (usually true but not always) and err on the side of lowering
1862 	 * the capacity.
1863 	 */
1864 	if (sdp->fix_capacity ||
1865 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1866 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1867 				"from its reported value: %llu\n",
1868 				(unsigned long long) sdkp->capacity);
1869 		--sdkp->capacity;
1870 	}
1871 
1872 got_data:
1873 	if (sector_size == 0) {
1874 		sector_size = 512;
1875 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
1876 			  "assuming 512.\n");
1877 	}
1878 
1879 	if (sector_size != 512 &&
1880 	    sector_size != 1024 &&
1881 	    sector_size != 2048 &&
1882 	    sector_size != 4096 &&
1883 	    sector_size != 256) {
1884 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
1885 			  sector_size);
1886 		/*
1887 		 * The user might want to re-format the drive with
1888 		 * a supported sectorsize.  Once this happens, it
1889 		 * would be relatively trivial to set the thing up.
1890 		 * For this reason, we leave the thing in the table.
1891 		 */
1892 		sdkp->capacity = 0;
1893 		/*
1894 		 * set a bogus sector size so the normal read/write
1895 		 * logic in the block layer will eventually refuse any
1896 		 * request on this device without tripping over power
1897 		 * of two sector size assumptions
1898 		 */
1899 		sector_size = 512;
1900 	}
1901 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
1902 
1903 	{
1904 		char cap_str_2[10], cap_str_10[10];
1905 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
1906 
1907 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
1908 				sizeof(cap_str_2));
1909 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
1910 				sizeof(cap_str_10));
1911 
1912 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
1913 			sd_printk(KERN_NOTICE, sdkp,
1914 				  "%llu %d-byte logical blocks: (%s/%s)\n",
1915 				  (unsigned long long)sdkp->capacity,
1916 				  sector_size, cap_str_10, cap_str_2);
1917 
1918 			if (sdkp->physical_block_size != sector_size)
1919 				sd_printk(KERN_NOTICE, sdkp,
1920 					  "%u-byte physical blocks\n",
1921 					  sdkp->physical_block_size);
1922 		}
1923 	}
1924 
1925 	/* Rescale capacity to 512-byte units */
1926 	if (sector_size == 4096)
1927 		sdkp->capacity <<= 3;
1928 	else if (sector_size == 2048)
1929 		sdkp->capacity <<= 2;
1930 	else if (sector_size == 1024)
1931 		sdkp->capacity <<= 1;
1932 	else if (sector_size == 256)
1933 		sdkp->capacity >>= 1;
1934 
1935 	blk_queue_physical_block_size(sdp->request_queue,
1936 				      sdkp->physical_block_size);
1937 	sdkp->device->sector_size = sector_size;
1938 }
1939 
1940 /* called with buffer of length 512 */
1941 static inline int
1942 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1943 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
1944 		 struct scsi_sense_hdr *sshdr)
1945 {
1946 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1947 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
1948 			       sshdr);
1949 }
1950 
1951 /*
1952  * read write protect setting, if possible - called only in sd_revalidate_disk()
1953  * called with buffer of length SD_BUF_SIZE
1954  */
1955 static void
1956 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
1957 {
1958 	int res;
1959 	struct scsi_device *sdp = sdkp->device;
1960 	struct scsi_mode_data data;
1961 	int old_wp = sdkp->write_prot;
1962 
1963 	set_disk_ro(sdkp->disk, 0);
1964 	if (sdp->skip_ms_page_3f) {
1965 		sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
1966 		return;
1967 	}
1968 
1969 	if (sdp->use_192_bytes_for_3f) {
1970 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1971 	} else {
1972 		/*
1973 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1974 		 * We have to start carefully: some devices hang if we ask
1975 		 * for more than is available.
1976 		 */
1977 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1978 
1979 		/*
1980 		 * Second attempt: ask for page 0 When only page 0 is
1981 		 * implemented, a request for page 3F may return Sense Key
1982 		 * 5: Illegal Request, Sense Code 24: Invalid field in
1983 		 * CDB.
1984 		 */
1985 		if (!scsi_status_is_good(res))
1986 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1987 
1988 		/*
1989 		 * Third attempt: ask 255 bytes, as we did earlier.
1990 		 */
1991 		if (!scsi_status_is_good(res))
1992 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1993 					       &data, NULL);
1994 	}
1995 
1996 	if (!scsi_status_is_good(res)) {
1997 		sd_printk(KERN_WARNING, sdkp,
1998 			  "Test WP failed, assume Write Enabled\n");
1999 	} else {
2000 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2001 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2002 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2003 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2004 				  sdkp->write_prot ? "on" : "off");
2005 			sd_printk(KERN_DEBUG, sdkp,
2006 				  "Mode Sense: %02x %02x %02x %02x\n",
2007 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2008 		}
2009 	}
2010 }
2011 
2012 /*
2013  * sd_read_cache_type - called only from sd_revalidate_disk()
2014  * called with buffer of length SD_BUF_SIZE
2015  */
2016 static void
2017 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2018 {
2019 	int len = 0, res;
2020 	struct scsi_device *sdp = sdkp->device;
2021 
2022 	int dbd;
2023 	int modepage;
2024 	struct scsi_mode_data data;
2025 	struct scsi_sense_hdr sshdr;
2026 	int old_wce = sdkp->WCE;
2027 	int old_rcd = sdkp->RCD;
2028 	int old_dpofua = sdkp->DPOFUA;
2029 
2030 	if (sdp->skip_ms_page_8)
2031 		goto defaults;
2032 
2033 	if (sdp->type == TYPE_RBC) {
2034 		modepage = 6;
2035 		dbd = 8;
2036 	} else {
2037 		modepage = 8;
2038 		dbd = 0;
2039 	}
2040 
2041 	/* cautiously ask */
2042 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, 4, &data, &sshdr);
2043 
2044 	if (!scsi_status_is_good(res))
2045 		goto bad_sense;
2046 
2047 	if (!data.header_length) {
2048 		modepage = 6;
2049 		sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
2050 	}
2051 
2052 	/* that went OK, now ask for the proper length */
2053 	len = data.length;
2054 
2055 	/*
2056 	 * We're only interested in the first three bytes, actually.
2057 	 * But the data cache page is defined for the first 20.
2058 	 */
2059 	if (len < 3)
2060 		goto bad_sense;
2061 	if (len > 20)
2062 		len = 20;
2063 
2064 	/* Take headers and block descriptors into account */
2065 	len += data.header_length + data.block_descriptor_length;
2066 	if (len > SD_BUF_SIZE)
2067 		goto bad_sense;
2068 
2069 	/* Get the data */
2070 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
2071 
2072 	if (scsi_status_is_good(res)) {
2073 		int offset = data.header_length + data.block_descriptor_length;
2074 
2075 		if (offset >= SD_BUF_SIZE - 2) {
2076 			sd_printk(KERN_ERR, sdkp, "Malformed MODE SENSE response\n");
2077 			goto defaults;
2078 		}
2079 
2080 		if ((buffer[offset] & 0x3f) != modepage) {
2081 			sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
2082 			goto defaults;
2083 		}
2084 
2085 		if (modepage == 8) {
2086 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2087 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2088 		} else {
2089 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2090 			sdkp->RCD = 0;
2091 		}
2092 
2093 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2094 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2095 			sd_printk(KERN_NOTICE, sdkp,
2096 				  "Uses READ/WRITE(6), disabling FUA\n");
2097 			sdkp->DPOFUA = 0;
2098 		}
2099 
2100 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2101 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2102 			sd_printk(KERN_NOTICE, sdkp,
2103 				  "Write cache: %s, read cache: %s, %s\n",
2104 				  sdkp->WCE ? "enabled" : "disabled",
2105 				  sdkp->RCD ? "disabled" : "enabled",
2106 				  sdkp->DPOFUA ? "supports DPO and FUA"
2107 				  : "doesn't support DPO or FUA");
2108 
2109 		return;
2110 	}
2111 
2112 bad_sense:
2113 	if (scsi_sense_valid(&sshdr) &&
2114 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2115 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2116 		/* Invalid field in CDB */
2117 		sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2118 	else
2119 		sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
2120 
2121 defaults:
2122 	sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
2123 	sdkp->WCE = 0;
2124 	sdkp->RCD = 0;
2125 	sdkp->DPOFUA = 0;
2126 }
2127 
2128 /*
2129  * The ATO bit indicates whether the DIF application tag is available
2130  * for use by the operating system.
2131  */
2132 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2133 {
2134 	int res, offset;
2135 	struct scsi_device *sdp = sdkp->device;
2136 	struct scsi_mode_data data;
2137 	struct scsi_sense_hdr sshdr;
2138 
2139 	if (sdp->type != TYPE_DISK)
2140 		return;
2141 
2142 	if (sdkp->protection_type == 0)
2143 		return;
2144 
2145 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2146 			      SD_MAX_RETRIES, &data, &sshdr);
2147 
2148 	if (!scsi_status_is_good(res) || !data.header_length ||
2149 	    data.length < 6) {
2150 		sd_printk(KERN_WARNING, sdkp,
2151 			  "getting Control mode page failed, assume no ATO\n");
2152 
2153 		if (scsi_sense_valid(&sshdr))
2154 			sd_print_sense_hdr(sdkp, &sshdr);
2155 
2156 		return;
2157 	}
2158 
2159 	offset = data.header_length + data.block_descriptor_length;
2160 
2161 	if ((buffer[offset] & 0x3f) != 0x0a) {
2162 		sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2163 		return;
2164 	}
2165 
2166 	if ((buffer[offset + 5] & 0x80) == 0)
2167 		return;
2168 
2169 	sdkp->ATO = 1;
2170 
2171 	return;
2172 }
2173 
2174 /**
2175  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2176  * @disk: disk to query
2177  */
2178 static void sd_read_block_limits(struct scsi_disk *sdkp)
2179 {
2180 	unsigned int sector_sz = sdkp->device->sector_size;
2181 	const int vpd_len = 64;
2182 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2183 
2184 	if (!buffer ||
2185 	    /* Block Limits VPD */
2186 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2187 		goto out;
2188 
2189 	blk_queue_io_min(sdkp->disk->queue,
2190 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2191 	blk_queue_io_opt(sdkp->disk->queue,
2192 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2193 
2194 	if (buffer[3] == 0x3c) {
2195 		unsigned int lba_count, desc_count;
2196 
2197 		sdkp->max_ws_blocks =
2198 			(u32) min_not_zero(get_unaligned_be64(&buffer[36]),
2199 					   (u64)0xffffffff);
2200 
2201 		if (!sdkp->lbpme)
2202 			goto out;
2203 
2204 		lba_count = get_unaligned_be32(&buffer[20]);
2205 		desc_count = get_unaligned_be32(&buffer[24]);
2206 
2207 		if (lba_count && desc_count)
2208 			sdkp->max_unmap_blocks = lba_count;
2209 
2210 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2211 
2212 		if (buffer[32] & 0x80)
2213 			sdkp->unmap_alignment =
2214 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2215 
2216 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2217 
2218 			if (sdkp->max_unmap_blocks)
2219 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2220 			else
2221 				sd_config_discard(sdkp, SD_LBP_WS16);
2222 
2223 		} else {	/* LBP VPD page tells us what to use */
2224 
2225 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2226 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2227 			else if (sdkp->lbpws)
2228 				sd_config_discard(sdkp, SD_LBP_WS16);
2229 			else if (sdkp->lbpws10)
2230 				sd_config_discard(sdkp, SD_LBP_WS10);
2231 			else
2232 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2233 		}
2234 	}
2235 
2236  out:
2237 	kfree(buffer);
2238 }
2239 
2240 /**
2241  * sd_read_block_characteristics - Query block dev. characteristics
2242  * @disk: disk to query
2243  */
2244 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2245 {
2246 	unsigned char *buffer;
2247 	u16 rot;
2248 	const int vpd_len = 64;
2249 
2250 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2251 
2252 	if (!buffer ||
2253 	    /* Block Device Characteristics VPD */
2254 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2255 		goto out;
2256 
2257 	rot = get_unaligned_be16(&buffer[4]);
2258 
2259 	if (rot == 1)
2260 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2261 
2262  out:
2263 	kfree(buffer);
2264 }
2265 
2266 /**
2267  * sd_read_block_provisioning - Query provisioning VPD page
2268  * @disk: disk to query
2269  */
2270 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2271 {
2272 	unsigned char *buffer;
2273 	const int vpd_len = 8;
2274 
2275 	if (sdkp->lbpme == 0)
2276 		return;
2277 
2278 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2279 
2280 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2281 		goto out;
2282 
2283 	sdkp->lbpvpd	= 1;
2284 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2285 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2286 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2287 
2288  out:
2289 	kfree(buffer);
2290 }
2291 
2292 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2293 {
2294 	/*
2295 	 * Although VPD inquiries can go to SCSI-2 type devices,
2296 	 * some USB ones crash on receiving them, and the pages
2297 	 * we currently ask for are for SPC-3 and beyond
2298 	 */
2299 	if (sdp->scsi_level > SCSI_SPC_2)
2300 		return 1;
2301 	return 0;
2302 }
2303 
2304 /**
2305  *	sd_revalidate_disk - called the first time a new disk is seen,
2306  *	performs disk spin up, read_capacity, etc.
2307  *	@disk: struct gendisk we care about
2308  **/
2309 static int sd_revalidate_disk(struct gendisk *disk)
2310 {
2311 	struct scsi_disk *sdkp = scsi_disk(disk);
2312 	struct scsi_device *sdp = sdkp->device;
2313 	unsigned char *buffer;
2314 	unsigned flush = 0;
2315 
2316 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2317 				      "sd_revalidate_disk\n"));
2318 
2319 	/*
2320 	 * If the device is offline, don't try and read capacity or any
2321 	 * of the other niceties.
2322 	 */
2323 	if (!scsi_device_online(sdp))
2324 		goto out;
2325 
2326 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2327 	if (!buffer) {
2328 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2329 			  "allocation failure.\n");
2330 		goto out;
2331 	}
2332 
2333 	sd_spinup_disk(sdkp);
2334 
2335 	/*
2336 	 * Without media there is no reason to ask; moreover, some devices
2337 	 * react badly if we do.
2338 	 */
2339 	if (sdkp->media_present) {
2340 		sd_read_capacity(sdkp, buffer);
2341 
2342 		if (sd_try_extended_inquiry(sdp)) {
2343 			sd_read_block_provisioning(sdkp);
2344 			sd_read_block_limits(sdkp);
2345 			sd_read_block_characteristics(sdkp);
2346 		}
2347 
2348 		sd_read_write_protect_flag(sdkp, buffer);
2349 		sd_read_cache_type(sdkp, buffer);
2350 		sd_read_app_tag_own(sdkp, buffer);
2351 	}
2352 
2353 	sdkp->first_scan = 0;
2354 
2355 	/*
2356 	 * We now have all cache related info, determine how we deal
2357 	 * with flush requests.
2358 	 */
2359 	if (sdkp->WCE) {
2360 		flush |= REQ_FLUSH;
2361 		if (sdkp->DPOFUA)
2362 			flush |= REQ_FUA;
2363 	}
2364 
2365 	blk_queue_flush(sdkp->disk->queue, flush);
2366 
2367 	set_capacity(disk, sdkp->capacity);
2368 	kfree(buffer);
2369 
2370  out:
2371 	return 0;
2372 }
2373 
2374 /**
2375  *	sd_unlock_native_capacity - unlock native capacity
2376  *	@disk: struct gendisk to set capacity for
2377  *
2378  *	Block layer calls this function if it detects that partitions
2379  *	on @disk reach beyond the end of the device.  If the SCSI host
2380  *	implements ->unlock_native_capacity() method, it's invoked to
2381  *	give it a chance to adjust the device capacity.
2382  *
2383  *	CONTEXT:
2384  *	Defined by block layer.  Might sleep.
2385  */
2386 static void sd_unlock_native_capacity(struct gendisk *disk)
2387 {
2388 	struct scsi_device *sdev = scsi_disk(disk)->device;
2389 
2390 	if (sdev->host->hostt->unlock_native_capacity)
2391 		sdev->host->hostt->unlock_native_capacity(sdev);
2392 }
2393 
2394 /**
2395  *	sd_format_disk_name - format disk name
2396  *	@prefix: name prefix - ie. "sd" for SCSI disks
2397  *	@index: index of the disk to format name for
2398  *	@buf: output buffer
2399  *	@buflen: length of the output buffer
2400  *
2401  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2402  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2403  *	which is followed by sdaaa.
2404  *
2405  *	This is basically 26 base counting with one extra 'nil' entry
2406  *	at the beginning from the second digit on and can be
2407  *	determined using similar method as 26 base conversion with the
2408  *	index shifted -1 after each digit is computed.
2409  *
2410  *	CONTEXT:
2411  *	Don't care.
2412  *
2413  *	RETURNS:
2414  *	0 on success, -errno on failure.
2415  */
2416 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2417 {
2418 	const int base = 'z' - 'a' + 1;
2419 	char *begin = buf + strlen(prefix);
2420 	char *end = buf + buflen;
2421 	char *p;
2422 	int unit;
2423 
2424 	p = end - 1;
2425 	*p = '\0';
2426 	unit = base;
2427 	do {
2428 		if (p == begin)
2429 			return -EINVAL;
2430 		*--p = 'a' + (index % unit);
2431 		index = (index / unit) - 1;
2432 	} while (index >= 0);
2433 
2434 	memmove(begin, p, end - p);
2435 	memcpy(buf, prefix, strlen(prefix));
2436 
2437 	return 0;
2438 }
2439 
2440 /*
2441  * The asynchronous part of sd_probe
2442  */
2443 static void sd_probe_async(void *data, async_cookie_t cookie)
2444 {
2445 	struct scsi_disk *sdkp = data;
2446 	struct scsi_device *sdp;
2447 	struct gendisk *gd;
2448 	u32 index;
2449 	struct device *dev;
2450 
2451 	sdp = sdkp->device;
2452 	gd = sdkp->disk;
2453 	index = sdkp->index;
2454 	dev = &sdp->sdev_gendev;
2455 
2456 	gd->major = sd_major((index & 0xf0) >> 4);
2457 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2458 	gd->minors = SD_MINORS;
2459 
2460 	gd->fops = &sd_fops;
2461 	gd->private_data = &sdkp->driver;
2462 	gd->queue = sdkp->device->request_queue;
2463 
2464 	/* defaults, until the device tells us otherwise */
2465 	sdp->sector_size = 512;
2466 	sdkp->capacity = 0;
2467 	sdkp->media_present = 1;
2468 	sdkp->write_prot = 0;
2469 	sdkp->WCE = 0;
2470 	sdkp->RCD = 0;
2471 	sdkp->ATO = 0;
2472 	sdkp->first_scan = 1;
2473 
2474 	sd_revalidate_disk(gd);
2475 
2476 	blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2477 	blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2478 
2479 	gd->driverfs_dev = &sdp->sdev_gendev;
2480 	gd->flags = GENHD_FL_EXT_DEVT;
2481 	if (sdp->removable) {
2482 		gd->flags |= GENHD_FL_REMOVABLE;
2483 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2484 	}
2485 
2486 	add_disk(gd);
2487 	sd_dif_config_host(sdkp);
2488 
2489 	sd_revalidate_disk(gd);
2490 
2491 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2492 		  sdp->removable ? "removable " : "");
2493 	scsi_autopm_put_device(sdp);
2494 	put_device(&sdkp->dev);
2495 }
2496 
2497 /**
2498  *	sd_probe - called during driver initialization and whenever a
2499  *	new scsi device is attached to the system. It is called once
2500  *	for each scsi device (not just disks) present.
2501  *	@dev: pointer to device object
2502  *
2503  *	Returns 0 if successful (or not interested in this scsi device
2504  *	(e.g. scanner)); 1 when there is an error.
2505  *
2506  *	Note: this function is invoked from the scsi mid-level.
2507  *	This function sets up the mapping between a given
2508  *	<host,channel,id,lun> (found in sdp) and new device name
2509  *	(e.g. /dev/sda). More precisely it is the block device major
2510  *	and minor number that is chosen here.
2511  *
2512  *	Assume sd_attach is not re-entrant (for time being)
2513  *	Also think about sd_attach() and sd_remove() running coincidentally.
2514  **/
2515 static int sd_probe(struct device *dev)
2516 {
2517 	struct scsi_device *sdp = to_scsi_device(dev);
2518 	struct scsi_disk *sdkp;
2519 	struct gendisk *gd;
2520 	int index;
2521 	int error;
2522 
2523 	error = -ENODEV;
2524 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2525 		goto out;
2526 
2527 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2528 					"sd_attach\n"));
2529 
2530 	error = -ENOMEM;
2531 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2532 	if (!sdkp)
2533 		goto out;
2534 
2535 	gd = alloc_disk(SD_MINORS);
2536 	if (!gd)
2537 		goto out_free;
2538 
2539 	do {
2540 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2541 			goto out_put;
2542 
2543 		spin_lock(&sd_index_lock);
2544 		error = ida_get_new(&sd_index_ida, &index);
2545 		spin_unlock(&sd_index_lock);
2546 	} while (error == -EAGAIN);
2547 
2548 	if (error)
2549 		goto out_put;
2550 
2551 	if (index >= SD_MAX_DISKS) {
2552 		error = -ENODEV;
2553 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name space exhausted.\n");
2554 		goto out_free_index;
2555 	}
2556 
2557 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2558 	if (error)
2559 		goto out_free_index;
2560 
2561 	sdkp->device = sdp;
2562 	sdkp->driver = &sd_template;
2563 	sdkp->disk = gd;
2564 	sdkp->index = index;
2565 	atomic_set(&sdkp->openers, 0);
2566 
2567 	if (!sdp->request_queue->rq_timeout) {
2568 		if (sdp->type != TYPE_MOD)
2569 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2570 		else
2571 			blk_queue_rq_timeout(sdp->request_queue,
2572 					     SD_MOD_TIMEOUT);
2573 	}
2574 
2575 	device_initialize(&sdkp->dev);
2576 	sdkp->dev.parent = dev;
2577 	sdkp->dev.class = &sd_disk_class;
2578 	dev_set_name(&sdkp->dev, dev_name(dev));
2579 
2580 	if (device_add(&sdkp->dev))
2581 		goto out_free_index;
2582 
2583 	get_device(dev);
2584 	dev_set_drvdata(dev, sdkp);
2585 
2586 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2587 	async_schedule(sd_probe_async, sdkp);
2588 
2589 	return 0;
2590 
2591  out_free_index:
2592 	spin_lock(&sd_index_lock);
2593 	ida_remove(&sd_index_ida, index);
2594 	spin_unlock(&sd_index_lock);
2595  out_put:
2596 	put_disk(gd);
2597  out_free:
2598 	kfree(sdkp);
2599  out:
2600 	return error;
2601 }
2602 
2603 /**
2604  *	sd_remove - called whenever a scsi disk (previously recognized by
2605  *	sd_probe) is detached from the system. It is called (potentially
2606  *	multiple times) during sd module unload.
2607  *	@sdp: pointer to mid level scsi device object
2608  *
2609  *	Note: this function is invoked from the scsi mid-level.
2610  *	This function potentially frees up a device name (e.g. /dev/sdc)
2611  *	that could be re-used by a subsequent sd_probe().
2612  *	This function is not called when the built-in sd driver is "exit-ed".
2613  **/
2614 static int sd_remove(struct device *dev)
2615 {
2616 	struct scsi_disk *sdkp;
2617 
2618 	sdkp = dev_get_drvdata(dev);
2619 	scsi_autopm_get_device(sdkp->device);
2620 
2621 	async_synchronize_full();
2622 	blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2623 	blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2624 	device_del(&sdkp->dev);
2625 	del_gendisk(sdkp->disk);
2626 	sd_shutdown(dev);
2627 
2628 	mutex_lock(&sd_ref_mutex);
2629 	dev_set_drvdata(dev, NULL);
2630 	put_device(&sdkp->dev);
2631 	mutex_unlock(&sd_ref_mutex);
2632 
2633 	return 0;
2634 }
2635 
2636 /**
2637  *	scsi_disk_release - Called to free the scsi_disk structure
2638  *	@dev: pointer to embedded class device
2639  *
2640  *	sd_ref_mutex must be held entering this routine.  Because it is
2641  *	called on last put, you should always use the scsi_disk_get()
2642  *	scsi_disk_put() helpers which manipulate the semaphore directly
2643  *	and never do a direct put_device.
2644  **/
2645 static void scsi_disk_release(struct device *dev)
2646 {
2647 	struct scsi_disk *sdkp = to_scsi_disk(dev);
2648 	struct gendisk *disk = sdkp->disk;
2649 
2650 	spin_lock(&sd_index_lock);
2651 	ida_remove(&sd_index_ida, sdkp->index);
2652 	spin_unlock(&sd_index_lock);
2653 
2654 	disk->private_data = NULL;
2655 	put_disk(disk);
2656 	put_device(&sdkp->device->sdev_gendev);
2657 
2658 	kfree(sdkp);
2659 }
2660 
2661 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2662 {
2663 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
2664 	struct scsi_sense_hdr sshdr;
2665 	struct scsi_device *sdp = sdkp->device;
2666 	int res;
2667 
2668 	if (start)
2669 		cmd[4] |= 1;	/* START */
2670 
2671 	if (sdp->start_stop_pwr_cond)
2672 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
2673 
2674 	if (!scsi_device_online(sdp))
2675 		return -ENODEV;
2676 
2677 	res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2678 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2679 	if (res) {
2680 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2681 		sd_print_result(sdkp, res);
2682 		if (driver_byte(res) & DRIVER_SENSE)
2683 			sd_print_sense_hdr(sdkp, &sshdr);
2684 	}
2685 
2686 	return res;
2687 }
2688 
2689 /*
2690  * Send a SYNCHRONIZE CACHE instruction down to the device through
2691  * the normal SCSI command structure.  Wait for the command to
2692  * complete.
2693  */
2694 static void sd_shutdown(struct device *dev)
2695 {
2696 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2697 
2698 	if (!sdkp)
2699 		return;         /* this can happen */
2700 
2701 	if (sdkp->WCE) {
2702 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2703 		sd_sync_cache(sdkp);
2704 	}
2705 
2706 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2707 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2708 		sd_start_stop_device(sdkp, 0);
2709 	}
2710 
2711 	scsi_disk_put(sdkp);
2712 }
2713 
2714 static int sd_suspend(struct device *dev, pm_message_t mesg)
2715 {
2716 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2717 	int ret = 0;
2718 
2719 	if (!sdkp)
2720 		return 0;	/* this can happen */
2721 
2722 	if (sdkp->WCE) {
2723 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2724 		ret = sd_sync_cache(sdkp);
2725 		if (ret)
2726 			goto done;
2727 	}
2728 
2729 	if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2730 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2731 		ret = sd_start_stop_device(sdkp, 0);
2732 	}
2733 
2734 done:
2735 	scsi_disk_put(sdkp);
2736 	return ret;
2737 }
2738 
2739 static int sd_resume(struct device *dev)
2740 {
2741 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2742 	int ret = 0;
2743 
2744 	if (!sdkp->device->manage_start_stop)
2745 		goto done;
2746 
2747 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2748 	ret = sd_start_stop_device(sdkp, 1);
2749 
2750 done:
2751 	scsi_disk_put(sdkp);
2752 	return ret;
2753 }
2754 
2755 /**
2756  *	init_sd - entry point for this driver (both when built in or when
2757  *	a module).
2758  *
2759  *	Note: this function registers this driver with the scsi mid-level.
2760  **/
2761 static int __init init_sd(void)
2762 {
2763 	int majors = 0, i, err;
2764 
2765 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2766 
2767 	for (i = 0; i < SD_MAJORS; i++)
2768 		if (register_blkdev(sd_major(i), "sd") == 0)
2769 			majors++;
2770 
2771 	if (!majors)
2772 		return -ENODEV;
2773 
2774 	err = class_register(&sd_disk_class);
2775 	if (err)
2776 		goto err_out;
2777 
2778 	err = scsi_register_driver(&sd_template.gendrv);
2779 	if (err)
2780 		goto err_out_class;
2781 
2782 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2783 					 0, 0, NULL);
2784 	if (!sd_cdb_cache) {
2785 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
2786 		goto err_out_class;
2787 	}
2788 
2789 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2790 	if (!sd_cdb_pool) {
2791 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
2792 		goto err_out_cache;
2793 	}
2794 
2795 	return 0;
2796 
2797 err_out_cache:
2798 	kmem_cache_destroy(sd_cdb_cache);
2799 
2800 err_out_class:
2801 	class_unregister(&sd_disk_class);
2802 err_out:
2803 	for (i = 0; i < SD_MAJORS; i++)
2804 		unregister_blkdev(sd_major(i), "sd");
2805 	return err;
2806 }
2807 
2808 /**
2809  *	exit_sd - exit point for this driver (when it is a module).
2810  *
2811  *	Note: this function unregisters this driver from the scsi mid-level.
2812  **/
2813 static void __exit exit_sd(void)
2814 {
2815 	int i;
2816 
2817 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
2818 
2819 	mempool_destroy(sd_cdb_pool);
2820 	kmem_cache_destroy(sd_cdb_cache);
2821 
2822 	scsi_unregister_driver(&sd_template.gendrv);
2823 	class_unregister(&sd_disk_class);
2824 
2825 	for (i = 0; i < SD_MAJORS; i++)
2826 		unregister_blkdev(sd_major(i), "sd");
2827 }
2828 
2829 module_init(init_sd);
2830 module_exit(exit_sd);
2831 
2832 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
2833 			       struct scsi_sense_hdr *sshdr)
2834 {
2835 	sd_printk(KERN_INFO, sdkp, " ");
2836 	scsi_show_sense_hdr(sshdr);
2837 	sd_printk(KERN_INFO, sdkp, " ");
2838 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
2839 }
2840 
2841 static void sd_print_result(struct scsi_disk *sdkp, int result)
2842 {
2843 	sd_printk(KERN_INFO, sdkp, " ");
2844 	scsi_show_result(result);
2845 }
2846 
2847