xref: /linux/drivers/scsi/sd.c (revision 90ab5ee94171b3e28de6bb42ee30b527014e0be7)
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_logging.h"
69 
70 MODULE_AUTHOR("Eric Youngdale");
71 MODULE_DESCRIPTION("SCSI disk (sd) driver");
72 MODULE_LICENSE("GPL");
73 
74 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
90 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
93 
94 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
95 #define SD_MINORS	16
96 #else
97 #define SD_MINORS	0
98 #endif
99 
100 static void sd_config_discard(struct scsi_disk *, unsigned int);
101 static int  sd_revalidate_disk(struct gendisk *);
102 static void sd_unlock_native_capacity(struct gendisk *disk);
103 static int  sd_probe(struct device *);
104 static int  sd_remove(struct device *);
105 static void sd_shutdown(struct device *);
106 static int sd_suspend(struct device *, pm_message_t state);
107 static int sd_resume(struct device *);
108 static void sd_rescan(struct device *);
109 static int sd_done(struct scsi_cmnd *);
110 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
111 static void scsi_disk_release(struct device *cdev);
112 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
113 static void sd_print_result(struct scsi_disk *, int);
114 
115 static DEFINE_SPINLOCK(sd_index_lock);
116 static DEFINE_IDA(sd_index_ida);
117 
118 /* This semaphore is used to mediate the 0->1 reference get in the
119  * face of object destruction (i.e. we can't allow a get on an
120  * object after last put) */
121 static DEFINE_MUTEX(sd_ref_mutex);
122 
123 static struct kmem_cache *sd_cdb_cache;
124 static mempool_t *sd_cdb_pool;
125 
126 static const char *sd_cache_types[] = {
127 	"write through", "none", "write back",
128 	"write back, no read (daft)"
129 };
130 
131 static ssize_t
132 sd_store_cache_type(struct device *dev, struct device_attribute *attr,
133 		    const char *buf, size_t count)
134 {
135 	int i, ct = -1, rcd, wce, sp;
136 	struct scsi_disk *sdkp = to_scsi_disk(dev);
137 	struct scsi_device *sdp = sdkp->device;
138 	char buffer[64];
139 	char *buffer_data;
140 	struct scsi_mode_data data;
141 	struct scsi_sense_hdr sshdr;
142 	int len;
143 
144 	if (sdp->type != TYPE_DISK)
145 		/* no cache control on RBC devices; theoretically they
146 		 * can do it, but there's probably so many exceptions
147 		 * it's not worth the risk */
148 		return -EINVAL;
149 
150 	for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
151 		len = strlen(sd_cache_types[i]);
152 		if (strncmp(sd_cache_types[i], buf, len) == 0 &&
153 		    buf[len] == '\n') {
154 			ct = i;
155 			break;
156 		}
157 	}
158 	if (ct < 0)
159 		return -EINVAL;
160 	rcd = ct & 0x01 ? 1 : 0;
161 	wce = ct & 0x02 ? 1 : 0;
162 	if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
163 			    SD_MAX_RETRIES, &data, NULL))
164 		return -EINVAL;
165 	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
166 		  data.block_descriptor_length);
167 	buffer_data = buffer + data.header_length +
168 		data.block_descriptor_length;
169 	buffer_data[2] &= ~0x05;
170 	buffer_data[2] |= wce << 2 | rcd;
171 	sp = buffer_data[0] & 0x80 ? 1 : 0;
172 
173 	if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
174 			     SD_MAX_RETRIES, &data, &sshdr)) {
175 		if (scsi_sense_valid(&sshdr))
176 			sd_print_sense_hdr(sdkp, &sshdr);
177 		return -EINVAL;
178 	}
179 	revalidate_disk(sdkp->disk);
180 	return count;
181 }
182 
183 static ssize_t
184 sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
185 			   const char *buf, size_t count)
186 {
187 	struct scsi_disk *sdkp = to_scsi_disk(dev);
188 	struct scsi_device *sdp = sdkp->device;
189 
190 	if (!capable(CAP_SYS_ADMIN))
191 		return -EACCES;
192 
193 	sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
194 
195 	return count;
196 }
197 
198 static ssize_t
199 sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
200 		       const char *buf, size_t count)
201 {
202 	struct scsi_disk *sdkp = to_scsi_disk(dev);
203 	struct scsi_device *sdp = sdkp->device;
204 
205 	if (!capable(CAP_SYS_ADMIN))
206 		return -EACCES;
207 
208 	if (sdp->type != TYPE_DISK)
209 		return -EINVAL;
210 
211 	sdp->allow_restart = simple_strtoul(buf, NULL, 10);
212 
213 	return count;
214 }
215 
216 static ssize_t
217 sd_show_cache_type(struct device *dev, struct device_attribute *attr,
218 		   char *buf)
219 {
220 	struct scsi_disk *sdkp = to_scsi_disk(dev);
221 	int ct = sdkp->RCD + 2*sdkp->WCE;
222 
223 	return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
224 }
225 
226 static ssize_t
227 sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
228 {
229 	struct scsi_disk *sdkp = to_scsi_disk(dev);
230 
231 	return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
232 }
233 
234 static ssize_t
235 sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
236 			  char *buf)
237 {
238 	struct scsi_disk *sdkp = to_scsi_disk(dev);
239 	struct scsi_device *sdp = sdkp->device;
240 
241 	return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
242 }
243 
244 static ssize_t
245 sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
246 		      char *buf)
247 {
248 	struct scsi_disk *sdkp = to_scsi_disk(dev);
249 
250 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
251 }
252 
253 static ssize_t
254 sd_show_protection_type(struct device *dev, struct device_attribute *attr,
255 			char *buf)
256 {
257 	struct scsi_disk *sdkp = to_scsi_disk(dev);
258 
259 	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
260 }
261 
262 static ssize_t
263 sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
264 			char *buf)
265 {
266 	struct scsi_disk *sdkp = to_scsi_disk(dev);
267 	struct scsi_device *sdp = sdkp->device;
268 	unsigned int dif, dix;
269 
270 	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
271 	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
272 
273 	if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
274 		dif = 0;
275 		dix = 1;
276 	}
277 
278 	if (!dif && !dix)
279 		return snprintf(buf, 20, "none\n");
280 
281 	return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
282 }
283 
284 static ssize_t
285 sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
286 		    char *buf)
287 {
288 	struct scsi_disk *sdkp = to_scsi_disk(dev);
289 
290 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
291 }
292 
293 static ssize_t
294 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
295 			  char *buf)
296 {
297 	struct scsi_disk *sdkp = to_scsi_disk(dev);
298 
299 	return snprintf(buf, 20, "%u\n", sdkp->lbpme);
300 }
301 
302 static const char *lbp_mode[] = {
303 	[SD_LBP_FULL]		= "full",
304 	[SD_LBP_UNMAP]		= "unmap",
305 	[SD_LBP_WS16]		= "writesame_16",
306 	[SD_LBP_WS10]		= "writesame_10",
307 	[SD_LBP_ZERO]		= "writesame_zero",
308 	[SD_LBP_DISABLE]	= "disabled",
309 };
310 
311 static ssize_t
312 sd_show_provisioning_mode(struct device *dev, struct device_attribute *attr,
313 			  char *buf)
314 {
315 	struct scsi_disk *sdkp = to_scsi_disk(dev);
316 
317 	return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
318 }
319 
320 static ssize_t
321 sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
322 			   const char *buf, size_t count)
323 {
324 	struct scsi_disk *sdkp = to_scsi_disk(dev);
325 	struct scsi_device *sdp = sdkp->device;
326 
327 	if (!capable(CAP_SYS_ADMIN))
328 		return -EACCES;
329 
330 	if (sdp->type != TYPE_DISK)
331 		return -EINVAL;
332 
333 	if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
334 		sd_config_discard(sdkp, SD_LBP_UNMAP);
335 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
336 		sd_config_discard(sdkp, SD_LBP_WS16);
337 	else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
338 		sd_config_discard(sdkp, SD_LBP_WS10);
339 	else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
340 		sd_config_discard(sdkp, SD_LBP_ZERO);
341 	else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
342 		sd_config_discard(sdkp, SD_LBP_DISABLE);
343 	else
344 		return -EINVAL;
345 
346 	return count;
347 }
348 
349 static struct device_attribute sd_disk_attrs[] = {
350 	__ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
351 	       sd_store_cache_type),
352 	__ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
353 	__ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
354 	       sd_store_allow_restart),
355 	__ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
356 	       sd_store_manage_start_stop),
357 	__ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
358 	__ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
359 	__ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
360 	__ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
361 	__ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
362 	       sd_store_provisioning_mode),
363 	__ATTR_NULL,
364 };
365 
366 static struct class sd_disk_class = {
367 	.name		= "scsi_disk",
368 	.owner		= THIS_MODULE,
369 	.dev_release	= scsi_disk_release,
370 	.dev_attrs	= sd_disk_attrs,
371 };
372 
373 static struct scsi_driver sd_template = {
374 	.owner			= THIS_MODULE,
375 	.gendrv = {
376 		.name		= "sd",
377 		.probe		= sd_probe,
378 		.remove		= sd_remove,
379 		.suspend	= sd_suspend,
380 		.resume		= sd_resume,
381 		.shutdown	= sd_shutdown,
382 	},
383 	.rescan			= sd_rescan,
384 	.done			= sd_done,
385 };
386 
387 /*
388  * Device no to disk mapping:
389  *
390  *       major         disc2     disc  p1
391  *   |............|.............|....|....| <- dev_t
392  *    31        20 19          8 7  4 3  0
393  *
394  * Inside a major, we have 16k disks, however mapped non-
395  * contiguously. The first 16 disks are for major0, the next
396  * ones with major1, ... Disk 256 is for major0 again, disk 272
397  * for major1, ...
398  * As we stay compatible with our numbering scheme, we can reuse
399  * the well-know SCSI majors 8, 65--71, 136--143.
400  */
401 static int sd_major(int major_idx)
402 {
403 	switch (major_idx) {
404 	case 0:
405 		return SCSI_DISK0_MAJOR;
406 	case 1 ... 7:
407 		return SCSI_DISK1_MAJOR + major_idx - 1;
408 	case 8 ... 15:
409 		return SCSI_DISK8_MAJOR + major_idx - 8;
410 	default:
411 		BUG();
412 		return 0;	/* shut up gcc */
413 	}
414 }
415 
416 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
417 {
418 	struct scsi_disk *sdkp = NULL;
419 
420 	if (disk->private_data) {
421 		sdkp = scsi_disk(disk);
422 		if (scsi_device_get(sdkp->device) == 0)
423 			get_device(&sdkp->dev);
424 		else
425 			sdkp = NULL;
426 	}
427 	return sdkp;
428 }
429 
430 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
431 {
432 	struct scsi_disk *sdkp;
433 
434 	mutex_lock(&sd_ref_mutex);
435 	sdkp = __scsi_disk_get(disk);
436 	mutex_unlock(&sd_ref_mutex);
437 	return sdkp;
438 }
439 
440 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
441 {
442 	struct scsi_disk *sdkp;
443 
444 	mutex_lock(&sd_ref_mutex);
445 	sdkp = dev_get_drvdata(dev);
446 	if (sdkp)
447 		sdkp = __scsi_disk_get(sdkp->disk);
448 	mutex_unlock(&sd_ref_mutex);
449 	return sdkp;
450 }
451 
452 static void scsi_disk_put(struct scsi_disk *sdkp)
453 {
454 	struct scsi_device *sdev = sdkp->device;
455 
456 	mutex_lock(&sd_ref_mutex);
457 	put_device(&sdkp->dev);
458 	scsi_device_put(sdev);
459 	mutex_unlock(&sd_ref_mutex);
460 }
461 
462 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
463 {
464 	unsigned int prot_op = SCSI_PROT_NORMAL;
465 	unsigned int dix = scsi_prot_sg_count(scmd);
466 
467 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
468 		if (dif && dix)
469 			prot_op = SCSI_PROT_READ_PASS;
470 		else if (dif && !dix)
471 			prot_op = SCSI_PROT_READ_STRIP;
472 		else if (!dif && dix)
473 			prot_op = SCSI_PROT_READ_INSERT;
474 	} else {
475 		if (dif && dix)
476 			prot_op = SCSI_PROT_WRITE_PASS;
477 		else if (dif && !dix)
478 			prot_op = SCSI_PROT_WRITE_INSERT;
479 		else if (!dif && dix)
480 			prot_op = SCSI_PROT_WRITE_STRIP;
481 	}
482 
483 	scsi_set_prot_op(scmd, prot_op);
484 	scsi_set_prot_type(scmd, dif);
485 }
486 
487 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
488 {
489 	struct request_queue *q = sdkp->disk->queue;
490 	unsigned int logical_block_size = sdkp->device->sector_size;
491 	unsigned int max_blocks = 0;
492 
493 	q->limits.discard_zeroes_data = sdkp->lbprz;
494 	q->limits.discard_alignment = sdkp->unmap_alignment *
495 		logical_block_size;
496 	q->limits.discard_granularity =
497 		max(sdkp->physical_block_size,
498 		    sdkp->unmap_granularity * logical_block_size);
499 
500 	switch (mode) {
501 
502 	case SD_LBP_DISABLE:
503 		q->limits.max_discard_sectors = 0;
504 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
505 		return;
506 
507 	case SD_LBP_UNMAP:
508 		max_blocks = min_not_zero(sdkp->max_unmap_blocks, 0xffffffff);
509 		break;
510 
511 	case SD_LBP_WS16:
512 		max_blocks = min_not_zero(sdkp->max_ws_blocks, 0xffffffff);
513 		break;
514 
515 	case SD_LBP_WS10:
516 		max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
517 		break;
518 
519 	case SD_LBP_ZERO:
520 		max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
521 		q->limits.discard_zeroes_data = 1;
522 		break;
523 	}
524 
525 	q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
526 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
527 
528 	sdkp->provisioning_mode = mode;
529 }
530 
531 /**
532  * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
533  * @sdp: scsi device to operate one
534  * @rq: Request to prepare
535  *
536  * Will issue either UNMAP or WRITE SAME(16) depending on preference
537  * indicated by target device.
538  **/
539 static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
540 {
541 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
542 	struct bio *bio = rq->bio;
543 	sector_t sector = bio->bi_sector;
544 	unsigned int nr_sectors = bio_sectors(bio);
545 	unsigned int len;
546 	int ret;
547 	char *buf;
548 	struct page *page;
549 
550 	if (sdkp->device->sector_size == 4096) {
551 		sector >>= 3;
552 		nr_sectors >>= 3;
553 	}
554 
555 	rq->timeout = SD_TIMEOUT;
556 
557 	memset(rq->cmd, 0, rq->cmd_len);
558 
559 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
560 	if (!page)
561 		return BLKPREP_DEFER;
562 
563 	switch (sdkp->provisioning_mode) {
564 	case SD_LBP_UNMAP:
565 		buf = page_address(page);
566 
567 		rq->cmd_len = 10;
568 		rq->cmd[0] = UNMAP;
569 		rq->cmd[8] = 24;
570 
571 		put_unaligned_be16(6 + 16, &buf[0]);
572 		put_unaligned_be16(16, &buf[2]);
573 		put_unaligned_be64(sector, &buf[8]);
574 		put_unaligned_be32(nr_sectors, &buf[16]);
575 
576 		len = 24;
577 		break;
578 
579 	case SD_LBP_WS16:
580 		rq->cmd_len = 16;
581 		rq->cmd[0] = WRITE_SAME_16;
582 		rq->cmd[1] = 0x8; /* UNMAP */
583 		put_unaligned_be64(sector, &rq->cmd[2]);
584 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
585 
586 		len = sdkp->device->sector_size;
587 		break;
588 
589 	case SD_LBP_WS10:
590 	case SD_LBP_ZERO:
591 		rq->cmd_len = 10;
592 		rq->cmd[0] = WRITE_SAME;
593 		if (sdkp->provisioning_mode == SD_LBP_WS10)
594 			rq->cmd[1] = 0x8; /* UNMAP */
595 		put_unaligned_be32(sector, &rq->cmd[2]);
596 		put_unaligned_be16(nr_sectors, &rq->cmd[7]);
597 
598 		len = sdkp->device->sector_size;
599 		break;
600 
601 	default:
602 		ret = BLKPREP_KILL;
603 		goto out;
604 	}
605 
606 	blk_add_request_payload(rq, page, len);
607 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
608 	rq->buffer = page_address(page);
609 
610 out:
611 	if (ret != BLKPREP_OK) {
612 		__free_page(page);
613 		rq->buffer = NULL;
614 	}
615 	return ret;
616 }
617 
618 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
619 {
620 	rq->timeout = SD_FLUSH_TIMEOUT;
621 	rq->retries = SD_MAX_RETRIES;
622 	rq->cmd[0] = SYNCHRONIZE_CACHE;
623 	rq->cmd_len = 10;
624 
625 	return scsi_setup_blk_pc_cmnd(sdp, rq);
626 }
627 
628 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
629 {
630 	if (rq->cmd_flags & REQ_DISCARD) {
631 		free_page((unsigned long)rq->buffer);
632 		rq->buffer = NULL;
633 	}
634 }
635 
636 /**
637  *	sd_init_command - build a scsi (read or write) command from
638  *	information in the request structure.
639  *	@SCpnt: pointer to mid-level's per scsi command structure that
640  *	contains request and into which the scsi command is written
641  *
642  *	Returns 1 if successful and 0 if error (or cannot be done now).
643  **/
644 static int sd_prep_fn(struct request_queue *q, struct request *rq)
645 {
646 	struct scsi_cmnd *SCpnt;
647 	struct scsi_device *sdp = q->queuedata;
648 	struct gendisk *disk = rq->rq_disk;
649 	struct scsi_disk *sdkp;
650 	sector_t block = blk_rq_pos(rq);
651 	sector_t threshold;
652 	unsigned int this_count = blk_rq_sectors(rq);
653 	int ret, host_dif;
654 	unsigned char protect;
655 
656 	/*
657 	 * Discard request come in as REQ_TYPE_FS but we turn them into
658 	 * block PC requests to make life easier.
659 	 */
660 	if (rq->cmd_flags & REQ_DISCARD) {
661 		ret = scsi_setup_discard_cmnd(sdp, rq);
662 		goto out;
663 	} else if (rq->cmd_flags & REQ_FLUSH) {
664 		ret = scsi_setup_flush_cmnd(sdp, rq);
665 		goto out;
666 	} else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
667 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
668 		goto out;
669 	} else if (rq->cmd_type != REQ_TYPE_FS) {
670 		ret = BLKPREP_KILL;
671 		goto out;
672 	}
673 	ret = scsi_setup_fs_cmnd(sdp, rq);
674 	if (ret != BLKPREP_OK)
675 		goto out;
676 	SCpnt = rq->special;
677 	sdkp = scsi_disk(disk);
678 
679 	/* from here on until we're complete, any goto out
680 	 * is used for a killable error condition */
681 	ret = BLKPREP_KILL;
682 
683 	SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
684 					"sd_init_command: block=%llu, "
685 					"count=%d\n",
686 					(unsigned long long)block,
687 					this_count));
688 
689 	if (!sdp || !scsi_device_online(sdp) ||
690 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
691 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
692 						"Finishing %u sectors\n",
693 						blk_rq_sectors(rq)));
694 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
695 						"Retry with 0x%p\n", SCpnt));
696 		goto out;
697 	}
698 
699 	if (sdp->changed) {
700 		/*
701 		 * quietly refuse to do anything to a changed disc until
702 		 * the changed bit has been reset
703 		 */
704 		/* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
705 		goto out;
706 	}
707 
708 	/*
709 	 * Some SD card readers can't handle multi-sector accesses which touch
710 	 * the last one or two hardware sectors.  Split accesses as needed.
711 	 */
712 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
713 		(sdp->sector_size / 512);
714 
715 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
716 		if (block < threshold) {
717 			/* Access up to the threshold but not beyond */
718 			this_count = threshold - block;
719 		} else {
720 			/* Access only a single hardware sector */
721 			this_count = sdp->sector_size / 512;
722 		}
723 	}
724 
725 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
726 					(unsigned long long)block));
727 
728 	/*
729 	 * If we have a 1K hardware sectorsize, prevent access to single
730 	 * 512 byte sectors.  In theory we could handle this - in fact
731 	 * the scsi cdrom driver must be able to handle this because
732 	 * we typically use 1K blocksizes, and cdroms typically have
733 	 * 2K hardware sectorsizes.  Of course, things are simpler
734 	 * with the cdrom, since it is read-only.  For performance
735 	 * reasons, the filesystems should be able to handle this
736 	 * and not force the scsi disk driver to use bounce buffers
737 	 * for this.
738 	 */
739 	if (sdp->sector_size == 1024) {
740 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
741 			scmd_printk(KERN_ERR, SCpnt,
742 				    "Bad block number requested\n");
743 			goto out;
744 		} else {
745 			block = block >> 1;
746 			this_count = this_count >> 1;
747 		}
748 	}
749 	if (sdp->sector_size == 2048) {
750 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
751 			scmd_printk(KERN_ERR, SCpnt,
752 				    "Bad block number requested\n");
753 			goto out;
754 		} else {
755 			block = block >> 2;
756 			this_count = this_count >> 2;
757 		}
758 	}
759 	if (sdp->sector_size == 4096) {
760 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
761 			scmd_printk(KERN_ERR, SCpnt,
762 				    "Bad block number requested\n");
763 			goto out;
764 		} else {
765 			block = block >> 3;
766 			this_count = this_count >> 3;
767 		}
768 	}
769 	if (rq_data_dir(rq) == WRITE) {
770 		if (!sdp->writeable) {
771 			goto out;
772 		}
773 		SCpnt->cmnd[0] = WRITE_6;
774 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
775 
776 		if (blk_integrity_rq(rq) &&
777 		    sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
778 			goto out;
779 
780 	} else if (rq_data_dir(rq) == READ) {
781 		SCpnt->cmnd[0] = READ_6;
782 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
783 	} else {
784 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
785 		goto out;
786 	}
787 
788 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
789 					"%s %d/%u 512 byte blocks.\n",
790 					(rq_data_dir(rq) == WRITE) ?
791 					"writing" : "reading", this_count,
792 					blk_rq_sectors(rq)));
793 
794 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
795 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
796 	if (host_dif)
797 		protect = 1 << 5;
798 	else
799 		protect = 0;
800 
801 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
802 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
803 
804 		if (unlikely(SCpnt->cmnd == NULL)) {
805 			ret = BLKPREP_DEFER;
806 			goto out;
807 		}
808 
809 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
810 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
811 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
812 		SCpnt->cmnd[7] = 0x18;
813 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
814 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
815 
816 		/* LBA */
817 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
818 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
819 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
820 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
821 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
822 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
823 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
824 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
825 
826 		/* Expected Indirect LBA */
827 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
828 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
829 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
830 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
831 
832 		/* Transfer length */
833 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
834 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
835 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
836 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
837 	} else if (block > 0xffffffff) {
838 		SCpnt->cmnd[0] += READ_16 - READ_6;
839 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
840 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
841 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
842 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
843 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
844 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
845 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
846 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
847 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
848 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
849 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
850 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
851 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
852 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
853 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
854 		   scsi_device_protection(SCpnt->device) ||
855 		   SCpnt->device->use_10_for_rw) {
856 		if (this_count > 0xffff)
857 			this_count = 0xffff;
858 
859 		SCpnt->cmnd[0] += READ_10 - READ_6;
860 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
861 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
862 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
863 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
864 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
865 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
866 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
867 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
868 	} else {
869 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
870 			/*
871 			 * This happens only if this drive failed
872 			 * 10byte rw command with ILLEGAL_REQUEST
873 			 * during operation and thus turned off
874 			 * use_10_for_rw.
875 			 */
876 			scmd_printk(KERN_ERR, SCpnt,
877 				    "FUA write on READ/WRITE(6) drive\n");
878 			goto out;
879 		}
880 
881 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
882 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
883 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
884 		SCpnt->cmnd[4] = (unsigned char) this_count;
885 		SCpnt->cmnd[5] = 0;
886 	}
887 	SCpnt->sdb.length = this_count * sdp->sector_size;
888 
889 	/* If DIF or DIX is enabled, tell HBA how to handle request */
890 	if (host_dif || scsi_prot_sg_count(SCpnt))
891 		sd_prot_op(SCpnt, host_dif);
892 
893 	/*
894 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
895 	 * host adapter, it's safe to assume that we can at least transfer
896 	 * this many bytes between each connect / disconnect.
897 	 */
898 	SCpnt->transfersize = sdp->sector_size;
899 	SCpnt->underflow = this_count << 9;
900 	SCpnt->allowed = SD_MAX_RETRIES;
901 
902 	/*
903 	 * This indicates that the command is ready from our end to be
904 	 * queued.
905 	 */
906 	ret = BLKPREP_OK;
907  out:
908 	return scsi_prep_return(q, rq, ret);
909 }
910 
911 /**
912  *	sd_open - open a scsi disk device
913  *	@inode: only i_rdev member may be used
914  *	@filp: only f_mode and f_flags may be used
915  *
916  *	Returns 0 if successful. Returns a negated errno value in case
917  *	of error.
918  *
919  *	Note: This can be called from a user context (e.g. fsck(1) )
920  *	or from within the kernel (e.g. as a result of a mount(1) ).
921  *	In the latter case @inode and @filp carry an abridged amount
922  *	of information as noted above.
923  *
924  *	Locking: called with bdev->bd_mutex held.
925  **/
926 static int sd_open(struct block_device *bdev, fmode_t mode)
927 {
928 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
929 	struct scsi_device *sdev;
930 	int retval;
931 
932 	if (!sdkp)
933 		return -ENXIO;
934 
935 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
936 
937 	sdev = sdkp->device;
938 
939 	retval = scsi_autopm_get_device(sdev);
940 	if (retval)
941 		goto error_autopm;
942 
943 	/*
944 	 * If the device is in error recovery, wait until it is done.
945 	 * If the device is offline, then disallow any access to it.
946 	 */
947 	retval = -ENXIO;
948 	if (!scsi_block_when_processing_errors(sdev))
949 		goto error_out;
950 
951 	if (sdev->removable || sdkp->write_prot)
952 		check_disk_change(bdev);
953 
954 	/*
955 	 * If the drive is empty, just let the open fail.
956 	 */
957 	retval = -ENOMEDIUM;
958 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
959 		goto error_out;
960 
961 	/*
962 	 * If the device has the write protect tab set, have the open fail
963 	 * if the user expects to be able to write to the thing.
964 	 */
965 	retval = -EROFS;
966 	if (sdkp->write_prot && (mode & FMODE_WRITE))
967 		goto error_out;
968 
969 	/*
970 	 * It is possible that the disk changing stuff resulted in
971 	 * the device being taken offline.  If this is the case,
972 	 * report this to the user, and don't pretend that the
973 	 * open actually succeeded.
974 	 */
975 	retval = -ENXIO;
976 	if (!scsi_device_online(sdev))
977 		goto error_out;
978 
979 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
980 		if (scsi_block_when_processing_errors(sdev))
981 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
982 	}
983 
984 	return 0;
985 
986 error_out:
987 	scsi_autopm_put_device(sdev);
988 error_autopm:
989 	scsi_disk_put(sdkp);
990 	return retval;
991 }
992 
993 /**
994  *	sd_release - invoked when the (last) close(2) is called on this
995  *	scsi disk.
996  *	@inode: only i_rdev member may be used
997  *	@filp: only f_mode and f_flags may be used
998  *
999  *	Returns 0.
1000  *
1001  *	Note: may block (uninterruptible) if error recovery is underway
1002  *	on this disk.
1003  *
1004  *	Locking: called with bdev->bd_mutex held.
1005  **/
1006 static int sd_release(struct gendisk *disk, fmode_t mode)
1007 {
1008 	struct scsi_disk *sdkp = scsi_disk(disk);
1009 	struct scsi_device *sdev = sdkp->device;
1010 
1011 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1012 
1013 	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1014 		if (scsi_block_when_processing_errors(sdev))
1015 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1016 	}
1017 
1018 	/*
1019 	 * XXX and what if there are packets in flight and this close()
1020 	 * XXX is followed by a "rmmod sd_mod"?
1021 	 */
1022 
1023 	scsi_autopm_put_device(sdev);
1024 	scsi_disk_put(sdkp);
1025 	return 0;
1026 }
1027 
1028 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1029 {
1030 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1031 	struct scsi_device *sdp = sdkp->device;
1032 	struct Scsi_Host *host = sdp->host;
1033 	int diskinfo[4];
1034 
1035 	/* default to most commonly used values */
1036         diskinfo[0] = 0x40;	/* 1 << 6 */
1037        	diskinfo[1] = 0x20;	/* 1 << 5 */
1038        	diskinfo[2] = sdkp->capacity >> 11;
1039 
1040 	/* override with calculated, extended default, or driver values */
1041 	if (host->hostt->bios_param)
1042 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1043 	else
1044 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1045 
1046 	geo->heads = diskinfo[0];
1047 	geo->sectors = diskinfo[1];
1048 	geo->cylinders = diskinfo[2];
1049 	return 0;
1050 }
1051 
1052 /**
1053  *	sd_ioctl - process an ioctl
1054  *	@inode: only i_rdev/i_bdev members may be used
1055  *	@filp: only f_mode and f_flags may be used
1056  *	@cmd: ioctl command number
1057  *	@arg: this is third argument given to ioctl(2) system call.
1058  *	Often contains a pointer.
1059  *
1060  *	Returns 0 if successful (some ioctls return positive numbers on
1061  *	success as well). Returns a negated errno value in case of error.
1062  *
1063  *	Note: most ioctls are forward onto the block subsystem or further
1064  *	down in the scsi subsystem.
1065  **/
1066 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1067 		    unsigned int cmd, unsigned long arg)
1068 {
1069 	struct gendisk *disk = bdev->bd_disk;
1070 	struct scsi_disk *sdkp = scsi_disk(disk);
1071 	struct scsi_device *sdp = sdkp->device;
1072 	void __user *p = (void __user *)arg;
1073 	int error;
1074 
1075 	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1076 				    "cmd=0x%x\n", disk->disk_name, cmd));
1077 
1078 	/*
1079 	 * If we are in the middle of error recovery, don't let anyone
1080 	 * else try and use this device.  Also, if error recovery fails, it
1081 	 * may try and take the device offline, in which case all further
1082 	 * access to the device is prohibited.
1083 	 */
1084 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
1085 					(mode & FMODE_NDELAY) != 0);
1086 	if (!scsi_block_when_processing_errors(sdp) || !error)
1087 		goto out;
1088 
1089 	/*
1090 	 * Send SCSI addressing ioctls directly to mid level, send other
1091 	 * ioctls to block level and then onto mid level if they can't be
1092 	 * resolved.
1093 	 */
1094 	switch (cmd) {
1095 		case SCSI_IOCTL_GET_IDLUN:
1096 		case SCSI_IOCTL_GET_BUS_NUMBER:
1097 			error = scsi_ioctl(sdp, cmd, p);
1098 			break;
1099 		default:
1100 			error = scsi_cmd_ioctl(disk->queue, disk, mode, cmd, p);
1101 			if (error != -ENOTTY)
1102 				break;
1103 			error = scsi_ioctl(sdp, cmd, p);
1104 			break;
1105 	}
1106 out:
1107 	return error;
1108 }
1109 
1110 static void set_media_not_present(struct scsi_disk *sdkp)
1111 {
1112 	if (sdkp->media_present)
1113 		sdkp->device->changed = 1;
1114 
1115 	if (sdkp->device->removable) {
1116 		sdkp->media_present = 0;
1117 		sdkp->capacity = 0;
1118 	}
1119 }
1120 
1121 static int media_not_present(struct scsi_disk *sdkp,
1122 			     struct scsi_sense_hdr *sshdr)
1123 {
1124 	if (!scsi_sense_valid(sshdr))
1125 		return 0;
1126 
1127 	/* not invoked for commands that could return deferred errors */
1128 	switch (sshdr->sense_key) {
1129 	case UNIT_ATTENTION:
1130 	case NOT_READY:
1131 		/* medium not present */
1132 		if (sshdr->asc == 0x3A) {
1133 			set_media_not_present(sdkp);
1134 			return 1;
1135 		}
1136 	}
1137 	return 0;
1138 }
1139 
1140 /**
1141  *	sd_check_events - check media events
1142  *	@disk: kernel device descriptor
1143  *	@clearing: disk events currently being cleared
1144  *
1145  *	Returns mask of DISK_EVENT_*.
1146  *
1147  *	Note: this function is invoked from the block subsystem.
1148  **/
1149 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1150 {
1151 	struct scsi_disk *sdkp = scsi_disk(disk);
1152 	struct scsi_device *sdp = sdkp->device;
1153 	struct scsi_sense_hdr *sshdr = NULL;
1154 	int retval;
1155 
1156 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1157 
1158 	/*
1159 	 * If the device is offline, don't send any commands - just pretend as
1160 	 * if the command failed.  If the device ever comes back online, we
1161 	 * can deal with it then.  It is only because of unrecoverable errors
1162 	 * that we would ever take a device offline in the first place.
1163 	 */
1164 	if (!scsi_device_online(sdp)) {
1165 		set_media_not_present(sdkp);
1166 		goto out;
1167 	}
1168 
1169 	/*
1170 	 * Using TEST_UNIT_READY enables differentiation between drive with
1171 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1172 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1173 	 *
1174 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1175 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1176 	 * sd_revalidate() is called.
1177 	 */
1178 	retval = -ENODEV;
1179 
1180 	if (scsi_block_when_processing_errors(sdp)) {
1181 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1182 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1183 					      sshdr);
1184 	}
1185 
1186 	/* failed to execute TUR, assume media not present */
1187 	if (host_byte(retval)) {
1188 		set_media_not_present(sdkp);
1189 		goto out;
1190 	}
1191 
1192 	if (media_not_present(sdkp, sshdr))
1193 		goto out;
1194 
1195 	/*
1196 	 * For removable scsi disk we have to recognise the presence
1197 	 * of a disk in the drive.
1198 	 */
1199 	if (!sdkp->media_present)
1200 		sdp->changed = 1;
1201 	sdkp->media_present = 1;
1202 out:
1203 	/*
1204 	 * sdp->changed is set under the following conditions:
1205 	 *
1206 	 *	Medium present state has changed in either direction.
1207 	 *	Device has indicated UNIT_ATTENTION.
1208 	 */
1209 	kfree(sshdr);
1210 	retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1211 	sdp->changed = 0;
1212 	return retval;
1213 }
1214 
1215 static int sd_sync_cache(struct scsi_disk *sdkp)
1216 {
1217 	int retries, res;
1218 	struct scsi_device *sdp = sdkp->device;
1219 	struct scsi_sense_hdr sshdr;
1220 
1221 	if (!scsi_device_online(sdp))
1222 		return -ENODEV;
1223 
1224 
1225 	for (retries = 3; retries > 0; --retries) {
1226 		unsigned char cmd[10] = { 0 };
1227 
1228 		cmd[0] = SYNCHRONIZE_CACHE;
1229 		/*
1230 		 * Leave the rest of the command zero to indicate
1231 		 * flush everything.
1232 		 */
1233 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1234 				       SD_FLUSH_TIMEOUT, SD_MAX_RETRIES, NULL);
1235 		if (res == 0)
1236 			break;
1237 	}
1238 
1239 	if (res) {
1240 		sd_print_result(sdkp, res);
1241 		if (driver_byte(res) & DRIVER_SENSE)
1242 			sd_print_sense_hdr(sdkp, &sshdr);
1243 	}
1244 
1245 	if (res)
1246 		return -EIO;
1247 	return 0;
1248 }
1249 
1250 static void sd_rescan(struct device *dev)
1251 {
1252 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1253 
1254 	if (sdkp) {
1255 		revalidate_disk(sdkp->disk);
1256 		scsi_disk_put(sdkp);
1257 	}
1258 }
1259 
1260 
1261 #ifdef CONFIG_COMPAT
1262 /*
1263  * This gets directly called from VFS. When the ioctl
1264  * is not recognized we go back to the other translation paths.
1265  */
1266 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1267 			   unsigned int cmd, unsigned long arg)
1268 {
1269 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1270 
1271 	/*
1272 	 * If we are in the middle of error recovery, don't let anyone
1273 	 * else try and use this device.  Also, if error recovery fails, it
1274 	 * may try and take the device offline, in which case all further
1275 	 * access to the device is prohibited.
1276 	 */
1277 	if (!scsi_block_when_processing_errors(sdev))
1278 		return -ENODEV;
1279 
1280 	if (sdev->host->hostt->compat_ioctl) {
1281 		int ret;
1282 
1283 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1284 
1285 		return ret;
1286 	}
1287 
1288 	/*
1289 	 * Let the static ioctl translation table take care of it.
1290 	 */
1291 	return -ENOIOCTLCMD;
1292 }
1293 #endif
1294 
1295 static const struct block_device_operations sd_fops = {
1296 	.owner			= THIS_MODULE,
1297 	.open			= sd_open,
1298 	.release		= sd_release,
1299 	.ioctl			= sd_ioctl,
1300 	.getgeo			= sd_getgeo,
1301 #ifdef CONFIG_COMPAT
1302 	.compat_ioctl		= sd_compat_ioctl,
1303 #endif
1304 	.check_events		= sd_check_events,
1305 	.revalidate_disk	= sd_revalidate_disk,
1306 	.unlock_native_capacity	= sd_unlock_native_capacity,
1307 };
1308 
1309 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1310 {
1311 	u64 start_lba = blk_rq_pos(scmd->request);
1312 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1313 	u64 bad_lba;
1314 	int info_valid;
1315 	/*
1316 	 * resid is optional but mostly filled in.  When it's unused,
1317 	 * its value is zero, so we assume the whole buffer transferred
1318 	 */
1319 	unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1320 	unsigned int good_bytes;
1321 
1322 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1323 		return 0;
1324 
1325 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1326 					     SCSI_SENSE_BUFFERSIZE,
1327 					     &bad_lba);
1328 	if (!info_valid)
1329 		return 0;
1330 
1331 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1332 		return 0;
1333 
1334 	if (scmd->device->sector_size < 512) {
1335 		/* only legitimate sector_size here is 256 */
1336 		start_lba <<= 1;
1337 		end_lba <<= 1;
1338 	} else {
1339 		/* be careful ... don't want any overflows */
1340 		u64 factor = scmd->device->sector_size / 512;
1341 		do_div(start_lba, factor);
1342 		do_div(end_lba, factor);
1343 	}
1344 
1345 	/* The bad lba was reported incorrectly, we have no idea where
1346 	 * the error is.
1347 	 */
1348 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1349 		return 0;
1350 
1351 	/* This computation should always be done in terms of
1352 	 * the resolution of the device's medium.
1353 	 */
1354 	good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1355 	return min(good_bytes, transferred);
1356 }
1357 
1358 /**
1359  *	sd_done - bottom half handler: called when the lower level
1360  *	driver has completed (successfully or otherwise) a scsi command.
1361  *	@SCpnt: mid-level's per command structure.
1362  *
1363  *	Note: potentially run from within an ISR. Must not block.
1364  **/
1365 static int sd_done(struct scsi_cmnd *SCpnt)
1366 {
1367 	int result = SCpnt->result;
1368 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1369 	struct scsi_sense_hdr sshdr;
1370 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1371 	int sense_valid = 0;
1372 	int sense_deferred = 0;
1373 	unsigned char op = SCpnt->cmnd[0];
1374 
1375 	if ((SCpnt->request->cmd_flags & REQ_DISCARD) && !result)
1376 		scsi_set_resid(SCpnt, 0);
1377 
1378 	if (result) {
1379 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1380 		if (sense_valid)
1381 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1382 	}
1383 #ifdef CONFIG_SCSI_LOGGING
1384 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1385 	if (sense_valid) {
1386 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1387 						   "sd_done: sb[respc,sk,asc,"
1388 						   "ascq]=%x,%x,%x,%x\n",
1389 						   sshdr.response_code,
1390 						   sshdr.sense_key, sshdr.asc,
1391 						   sshdr.ascq));
1392 	}
1393 #endif
1394 	if (driver_byte(result) != DRIVER_SENSE &&
1395 	    (!sense_valid || sense_deferred))
1396 		goto out;
1397 
1398 	switch (sshdr.sense_key) {
1399 	case HARDWARE_ERROR:
1400 	case MEDIUM_ERROR:
1401 		good_bytes = sd_completed_bytes(SCpnt);
1402 		break;
1403 	case RECOVERED_ERROR:
1404 		good_bytes = scsi_bufflen(SCpnt);
1405 		break;
1406 	case NO_SENSE:
1407 		/* This indicates a false check condition, so ignore it.  An
1408 		 * unknown amount of data was transferred so treat it as an
1409 		 * error.
1410 		 */
1411 		scsi_print_sense("sd", SCpnt);
1412 		SCpnt->result = 0;
1413 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1414 		break;
1415 	case ABORTED_COMMAND:
1416 		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1417 			good_bytes = sd_completed_bytes(SCpnt);
1418 		break;
1419 	case ILLEGAL_REQUEST:
1420 		if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1421 			good_bytes = sd_completed_bytes(SCpnt);
1422 		/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1423 		if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1424 		    (op == UNMAP || op == WRITE_SAME_16 || op == WRITE_SAME))
1425 			sd_config_discard(sdkp, SD_LBP_DISABLE);
1426 		break;
1427 	default:
1428 		break;
1429 	}
1430  out:
1431 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1432 		sd_dif_complete(SCpnt, good_bytes);
1433 
1434 	if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1435 	    == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1436 
1437 		/* We have to print a failed command here as the
1438 		 * extended CDB gets freed before scsi_io_completion()
1439 		 * is called.
1440 		 */
1441 		if (result)
1442 			scsi_print_command(SCpnt);
1443 
1444 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1445 		SCpnt->cmnd = NULL;
1446 		SCpnt->cmd_len = 0;
1447 	}
1448 
1449 	return good_bytes;
1450 }
1451 
1452 /*
1453  * spinup disk - called only in sd_revalidate_disk()
1454  */
1455 static void
1456 sd_spinup_disk(struct scsi_disk *sdkp)
1457 {
1458 	unsigned char cmd[10];
1459 	unsigned long spintime_expire = 0;
1460 	int retries, spintime;
1461 	unsigned int the_result;
1462 	struct scsi_sense_hdr sshdr;
1463 	int sense_valid = 0;
1464 
1465 	spintime = 0;
1466 
1467 	/* Spin up drives, as required.  Only do this at boot time */
1468 	/* Spinup needs to be done for module loads too. */
1469 	do {
1470 		retries = 0;
1471 
1472 		do {
1473 			cmd[0] = TEST_UNIT_READY;
1474 			memset((void *) &cmd[1], 0, 9);
1475 
1476 			the_result = scsi_execute_req(sdkp->device, cmd,
1477 						      DMA_NONE, NULL, 0,
1478 						      &sshdr, SD_TIMEOUT,
1479 						      SD_MAX_RETRIES, NULL);
1480 
1481 			/*
1482 			 * If the drive has indicated to us that it
1483 			 * doesn't have any media in it, don't bother
1484 			 * with any more polling.
1485 			 */
1486 			if (media_not_present(sdkp, &sshdr))
1487 				return;
1488 
1489 			if (the_result)
1490 				sense_valid = scsi_sense_valid(&sshdr);
1491 			retries++;
1492 		} while (retries < 3 &&
1493 			 (!scsi_status_is_good(the_result) ||
1494 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1495 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1496 
1497 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1498 			/* no sense, TUR either succeeded or failed
1499 			 * with a status error */
1500 			if(!spintime && !scsi_status_is_good(the_result)) {
1501 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1502 				sd_print_result(sdkp, the_result);
1503 			}
1504 			break;
1505 		}
1506 
1507 		/*
1508 		 * The device does not want the automatic start to be issued.
1509 		 */
1510 		if (sdkp->device->no_start_on_add)
1511 			break;
1512 
1513 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1514 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1515 				break;	/* manual intervention required */
1516 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1517 				break;	/* standby */
1518 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1519 				break;	/* unavailable */
1520 			/*
1521 			 * Issue command to spin up drive when not ready
1522 			 */
1523 			if (!spintime) {
1524 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1525 				cmd[0] = START_STOP;
1526 				cmd[1] = 1;	/* Return immediately */
1527 				memset((void *) &cmd[2], 0, 8);
1528 				cmd[4] = 1;	/* Start spin cycle */
1529 				if (sdkp->device->start_stop_pwr_cond)
1530 					cmd[4] |= 1 << 4;
1531 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1532 						 NULL, 0, &sshdr,
1533 						 SD_TIMEOUT, SD_MAX_RETRIES,
1534 						 NULL);
1535 				spintime_expire = jiffies + 100 * HZ;
1536 				spintime = 1;
1537 			}
1538 			/* Wait 1 second for next try */
1539 			msleep(1000);
1540 			printk(".");
1541 
1542 		/*
1543 		 * Wait for USB flash devices with slow firmware.
1544 		 * Yes, this sense key/ASC combination shouldn't
1545 		 * occur here.  It's characteristic of these devices.
1546 		 */
1547 		} else if (sense_valid &&
1548 				sshdr.sense_key == UNIT_ATTENTION &&
1549 				sshdr.asc == 0x28) {
1550 			if (!spintime) {
1551 				spintime_expire = jiffies + 5 * HZ;
1552 				spintime = 1;
1553 			}
1554 			/* Wait 1 second for next try */
1555 			msleep(1000);
1556 		} else {
1557 			/* we don't understand the sense code, so it's
1558 			 * probably pointless to loop */
1559 			if(!spintime) {
1560 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1561 				sd_print_sense_hdr(sdkp, &sshdr);
1562 			}
1563 			break;
1564 		}
1565 
1566 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1567 
1568 	if (spintime) {
1569 		if (scsi_status_is_good(the_result))
1570 			printk("ready\n");
1571 		else
1572 			printk("not responding...\n");
1573 	}
1574 }
1575 
1576 
1577 /*
1578  * Determine whether disk supports Data Integrity Field.
1579  */
1580 static void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1581 {
1582 	struct scsi_device *sdp = sdkp->device;
1583 	u8 type;
1584 
1585 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1586 		return;
1587 
1588 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1589 
1590 	if (type == sdkp->protection_type || !sdkp->first_scan)
1591 		return;
1592 
1593 	sdkp->protection_type = type;
1594 
1595 	if (type > SD_DIF_TYPE3_PROTECTION) {
1596 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported "	\
1597 			  "protection type %u. Disabling disk!\n", type);
1598 		sdkp->capacity = 0;
1599 		return;
1600 	}
1601 
1602 	if (scsi_host_dif_capable(sdp->host, type))
1603 		sd_printk(KERN_NOTICE, sdkp,
1604 			  "Enabling DIF Type %u protection\n", type);
1605 	else
1606 		sd_printk(KERN_NOTICE, sdkp,
1607 			  "Disabling DIF Type %u protection\n", type);
1608 }
1609 
1610 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1611 			struct scsi_sense_hdr *sshdr, int sense_valid,
1612 			int the_result)
1613 {
1614 	sd_print_result(sdkp, the_result);
1615 	if (driver_byte(the_result) & DRIVER_SENSE)
1616 		sd_print_sense_hdr(sdkp, sshdr);
1617 	else
1618 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1619 
1620 	/*
1621 	 * Set dirty bit for removable devices if not ready -
1622 	 * sometimes drives will not report this properly.
1623 	 */
1624 	if (sdp->removable &&
1625 	    sense_valid && sshdr->sense_key == NOT_READY)
1626 		set_media_not_present(sdkp);
1627 
1628 	/*
1629 	 * We used to set media_present to 0 here to indicate no media
1630 	 * in the drive, but some drives fail read capacity even with
1631 	 * media present, so we can't do that.
1632 	 */
1633 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1634 }
1635 
1636 #define RC16_LEN 32
1637 #if RC16_LEN > SD_BUF_SIZE
1638 #error RC16_LEN must not be more than SD_BUF_SIZE
1639 #endif
1640 
1641 #define READ_CAPACITY_RETRIES_ON_RESET	10
1642 
1643 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1644 						unsigned char *buffer)
1645 {
1646 	unsigned char cmd[16];
1647 	struct scsi_sense_hdr sshdr;
1648 	int sense_valid = 0;
1649 	int the_result;
1650 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1651 	unsigned int alignment;
1652 	unsigned long long lba;
1653 	unsigned sector_size;
1654 
1655 	if (sdp->no_read_capacity_16)
1656 		return -EINVAL;
1657 
1658 	do {
1659 		memset(cmd, 0, 16);
1660 		cmd[0] = SERVICE_ACTION_IN;
1661 		cmd[1] = SAI_READ_CAPACITY_16;
1662 		cmd[13] = RC16_LEN;
1663 		memset(buffer, 0, RC16_LEN);
1664 
1665 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1666 					buffer, RC16_LEN, &sshdr,
1667 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1668 
1669 		if (media_not_present(sdkp, &sshdr))
1670 			return -ENODEV;
1671 
1672 		if (the_result) {
1673 			sense_valid = scsi_sense_valid(&sshdr);
1674 			if (sense_valid &&
1675 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1676 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1677 			    sshdr.ascq == 0x00)
1678 				/* Invalid Command Operation Code or
1679 				 * Invalid Field in CDB, just retry
1680 				 * silently with RC10 */
1681 				return -EINVAL;
1682 			if (sense_valid &&
1683 			    sshdr.sense_key == UNIT_ATTENTION &&
1684 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1685 				/* Device reset might occur several times,
1686 				 * give it one more chance */
1687 				if (--reset_retries > 0)
1688 					continue;
1689 		}
1690 		retries--;
1691 
1692 	} while (the_result && retries);
1693 
1694 	if (the_result) {
1695 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1696 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1697 		return -EINVAL;
1698 	}
1699 
1700 	sector_size = get_unaligned_be32(&buffer[8]);
1701 	lba = get_unaligned_be64(&buffer[0]);
1702 
1703 	sd_read_protection_type(sdkp, buffer);
1704 
1705 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1706 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1707 			"kernel compiled with support for large block "
1708 			"devices.\n");
1709 		sdkp->capacity = 0;
1710 		return -EOVERFLOW;
1711 	}
1712 
1713 	/* Logical blocks per physical block exponent */
1714 	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
1715 
1716 	/* Lowest aligned logical block */
1717 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1718 	blk_queue_alignment_offset(sdp->request_queue, alignment);
1719 	if (alignment && sdkp->first_scan)
1720 		sd_printk(KERN_NOTICE, sdkp,
1721 			  "physical block alignment offset: %u\n", alignment);
1722 
1723 	if (buffer[14] & 0x80) { /* LBPME */
1724 		sdkp->lbpme = 1;
1725 
1726 		if (buffer[14] & 0x40) /* LBPRZ */
1727 			sdkp->lbprz = 1;
1728 
1729 		sd_config_discard(sdkp, SD_LBP_WS16);
1730 	}
1731 
1732 	sdkp->capacity = lba + 1;
1733 	return sector_size;
1734 }
1735 
1736 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1737 						unsigned char *buffer)
1738 {
1739 	unsigned char cmd[16];
1740 	struct scsi_sense_hdr sshdr;
1741 	int sense_valid = 0;
1742 	int the_result;
1743 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1744 	sector_t lba;
1745 	unsigned sector_size;
1746 
1747 	do {
1748 		cmd[0] = READ_CAPACITY;
1749 		memset(&cmd[1], 0, 9);
1750 		memset(buffer, 0, 8);
1751 
1752 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1753 					buffer, 8, &sshdr,
1754 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1755 
1756 		if (media_not_present(sdkp, &sshdr))
1757 			return -ENODEV;
1758 
1759 		if (the_result) {
1760 			sense_valid = scsi_sense_valid(&sshdr);
1761 			if (sense_valid &&
1762 			    sshdr.sense_key == UNIT_ATTENTION &&
1763 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1764 				/* Device reset might occur several times,
1765 				 * give it one more chance */
1766 				if (--reset_retries > 0)
1767 					continue;
1768 		}
1769 		retries--;
1770 
1771 	} while (the_result && retries);
1772 
1773 	if (the_result) {
1774 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1775 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1776 		return -EINVAL;
1777 	}
1778 
1779 	sector_size = get_unaligned_be32(&buffer[4]);
1780 	lba = get_unaligned_be32(&buffer[0]);
1781 
1782 	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
1783 		/* Some buggy (usb cardreader) devices return an lba of
1784 		   0xffffffff when the want to report a size of 0 (with
1785 		   which they really mean no media is present) */
1786 		sdkp->capacity = 0;
1787 		sdkp->physical_block_size = sector_size;
1788 		return sector_size;
1789 	}
1790 
1791 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1792 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1793 			"kernel compiled with support for large block "
1794 			"devices.\n");
1795 		sdkp->capacity = 0;
1796 		return -EOVERFLOW;
1797 	}
1798 
1799 	sdkp->capacity = lba + 1;
1800 	sdkp->physical_block_size = sector_size;
1801 	return sector_size;
1802 }
1803 
1804 static int sd_try_rc16_first(struct scsi_device *sdp)
1805 {
1806 	if (sdp->host->max_cmd_len < 16)
1807 		return 0;
1808 	if (sdp->scsi_level > SCSI_SPC_2)
1809 		return 1;
1810 	if (scsi_device_protection(sdp))
1811 		return 1;
1812 	return 0;
1813 }
1814 
1815 /*
1816  * read disk capacity
1817  */
1818 static void
1819 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1820 {
1821 	int sector_size;
1822 	struct scsi_device *sdp = sdkp->device;
1823 	sector_t old_capacity = sdkp->capacity;
1824 
1825 	if (sd_try_rc16_first(sdp)) {
1826 		sector_size = read_capacity_16(sdkp, sdp, buffer);
1827 		if (sector_size == -EOVERFLOW)
1828 			goto got_data;
1829 		if (sector_size == -ENODEV)
1830 			return;
1831 		if (sector_size < 0)
1832 			sector_size = read_capacity_10(sdkp, sdp, buffer);
1833 		if (sector_size < 0)
1834 			return;
1835 	} else {
1836 		sector_size = read_capacity_10(sdkp, sdp, buffer);
1837 		if (sector_size == -EOVERFLOW)
1838 			goto got_data;
1839 		if (sector_size < 0)
1840 			return;
1841 		if ((sizeof(sdkp->capacity) > 4) &&
1842 		    (sdkp->capacity > 0xffffffffULL)) {
1843 			int old_sector_size = sector_size;
1844 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1845 					"Trying to use READ CAPACITY(16).\n");
1846 			sector_size = read_capacity_16(sdkp, sdp, buffer);
1847 			if (sector_size < 0) {
1848 				sd_printk(KERN_NOTICE, sdkp,
1849 					"Using 0xffffffff as device size\n");
1850 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
1851 				sector_size = old_sector_size;
1852 				goto got_data;
1853 			}
1854 		}
1855 	}
1856 
1857 	/* Some devices are known to return the total number of blocks,
1858 	 * not the highest block number.  Some devices have versions
1859 	 * which do this and others which do not.  Some devices we might
1860 	 * suspect of doing this but we don't know for certain.
1861 	 *
1862 	 * If we know the reported capacity is wrong, decrement it.  If
1863 	 * we can only guess, then assume the number of blocks is even
1864 	 * (usually true but not always) and err on the side of lowering
1865 	 * the capacity.
1866 	 */
1867 	if (sdp->fix_capacity ||
1868 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1869 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1870 				"from its reported value: %llu\n",
1871 				(unsigned long long) sdkp->capacity);
1872 		--sdkp->capacity;
1873 	}
1874 
1875 got_data:
1876 	if (sector_size == 0) {
1877 		sector_size = 512;
1878 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
1879 			  "assuming 512.\n");
1880 	}
1881 
1882 	if (sector_size != 512 &&
1883 	    sector_size != 1024 &&
1884 	    sector_size != 2048 &&
1885 	    sector_size != 4096 &&
1886 	    sector_size != 256) {
1887 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
1888 			  sector_size);
1889 		/*
1890 		 * The user might want to re-format the drive with
1891 		 * a supported sectorsize.  Once this happens, it
1892 		 * would be relatively trivial to set the thing up.
1893 		 * For this reason, we leave the thing in the table.
1894 		 */
1895 		sdkp->capacity = 0;
1896 		/*
1897 		 * set a bogus sector size so the normal read/write
1898 		 * logic in the block layer will eventually refuse any
1899 		 * request on this device without tripping over power
1900 		 * of two sector size assumptions
1901 		 */
1902 		sector_size = 512;
1903 	}
1904 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
1905 
1906 	{
1907 		char cap_str_2[10], cap_str_10[10];
1908 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
1909 
1910 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
1911 				sizeof(cap_str_2));
1912 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
1913 				sizeof(cap_str_10));
1914 
1915 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
1916 			sd_printk(KERN_NOTICE, sdkp,
1917 				  "%llu %d-byte logical blocks: (%s/%s)\n",
1918 				  (unsigned long long)sdkp->capacity,
1919 				  sector_size, cap_str_10, cap_str_2);
1920 
1921 			if (sdkp->physical_block_size != sector_size)
1922 				sd_printk(KERN_NOTICE, sdkp,
1923 					  "%u-byte physical blocks\n",
1924 					  sdkp->physical_block_size);
1925 		}
1926 	}
1927 
1928 	/* Rescale capacity to 512-byte units */
1929 	if (sector_size == 4096)
1930 		sdkp->capacity <<= 3;
1931 	else if (sector_size == 2048)
1932 		sdkp->capacity <<= 2;
1933 	else if (sector_size == 1024)
1934 		sdkp->capacity <<= 1;
1935 	else if (sector_size == 256)
1936 		sdkp->capacity >>= 1;
1937 
1938 	blk_queue_physical_block_size(sdp->request_queue,
1939 				      sdkp->physical_block_size);
1940 	sdkp->device->sector_size = sector_size;
1941 }
1942 
1943 /* called with buffer of length 512 */
1944 static inline int
1945 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1946 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
1947 		 struct scsi_sense_hdr *sshdr)
1948 {
1949 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1950 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
1951 			       sshdr);
1952 }
1953 
1954 /*
1955  * read write protect setting, if possible - called only in sd_revalidate_disk()
1956  * called with buffer of length SD_BUF_SIZE
1957  */
1958 static void
1959 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
1960 {
1961 	int res;
1962 	struct scsi_device *sdp = sdkp->device;
1963 	struct scsi_mode_data data;
1964 	int old_wp = sdkp->write_prot;
1965 
1966 	set_disk_ro(sdkp->disk, 0);
1967 	if (sdp->skip_ms_page_3f) {
1968 		sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
1969 		return;
1970 	}
1971 
1972 	if (sdp->use_192_bytes_for_3f) {
1973 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1974 	} else {
1975 		/*
1976 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1977 		 * We have to start carefully: some devices hang if we ask
1978 		 * for more than is available.
1979 		 */
1980 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1981 
1982 		/*
1983 		 * Second attempt: ask for page 0 When only page 0 is
1984 		 * implemented, a request for page 3F may return Sense Key
1985 		 * 5: Illegal Request, Sense Code 24: Invalid field in
1986 		 * CDB.
1987 		 */
1988 		if (!scsi_status_is_good(res))
1989 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1990 
1991 		/*
1992 		 * Third attempt: ask 255 bytes, as we did earlier.
1993 		 */
1994 		if (!scsi_status_is_good(res))
1995 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1996 					       &data, NULL);
1997 	}
1998 
1999 	if (!scsi_status_is_good(res)) {
2000 		sd_printk(KERN_WARNING, sdkp,
2001 			  "Test WP failed, assume Write Enabled\n");
2002 	} else {
2003 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2004 		set_disk_ro(sdkp->disk, sdkp->write_prot);
2005 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2006 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2007 				  sdkp->write_prot ? "on" : "off");
2008 			sd_printk(KERN_DEBUG, sdkp,
2009 				  "Mode Sense: %02x %02x %02x %02x\n",
2010 				  buffer[0], buffer[1], buffer[2], buffer[3]);
2011 		}
2012 	}
2013 }
2014 
2015 /*
2016  * sd_read_cache_type - called only from sd_revalidate_disk()
2017  * called with buffer of length SD_BUF_SIZE
2018  */
2019 static void
2020 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2021 {
2022 	int len = 0, res;
2023 	struct scsi_device *sdp = sdkp->device;
2024 
2025 	int dbd;
2026 	int modepage;
2027 	int first_len;
2028 	struct scsi_mode_data data;
2029 	struct scsi_sense_hdr sshdr;
2030 	int old_wce = sdkp->WCE;
2031 	int old_rcd = sdkp->RCD;
2032 	int old_dpofua = sdkp->DPOFUA;
2033 
2034 	first_len = 4;
2035 	if (sdp->skip_ms_page_8) {
2036 		if (sdp->type == TYPE_RBC)
2037 			goto defaults;
2038 		else {
2039 			if (sdp->skip_ms_page_3f)
2040 				goto defaults;
2041 			modepage = 0x3F;
2042 			if (sdp->use_192_bytes_for_3f)
2043 				first_len = 192;
2044 			dbd = 0;
2045 		}
2046 	} else if (sdp->type == TYPE_RBC) {
2047 		modepage = 6;
2048 		dbd = 8;
2049 	} else {
2050 		modepage = 8;
2051 		dbd = 0;
2052 	}
2053 
2054 	/* cautiously ask */
2055 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2056 			&data, &sshdr);
2057 
2058 	if (!scsi_status_is_good(res))
2059 		goto bad_sense;
2060 
2061 	if (!data.header_length) {
2062 		modepage = 6;
2063 		first_len = 0;
2064 		sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
2065 	}
2066 
2067 	/* that went OK, now ask for the proper length */
2068 	len = data.length;
2069 
2070 	/*
2071 	 * We're only interested in the first three bytes, actually.
2072 	 * But the data cache page is defined for the first 20.
2073 	 */
2074 	if (len < 3)
2075 		goto bad_sense;
2076 	else if (len > SD_BUF_SIZE) {
2077 		sd_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2078 			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2079 		len = SD_BUF_SIZE;
2080 	}
2081 	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2082 		len = 192;
2083 
2084 	/* Get the data */
2085 	if (len > first_len)
2086 		res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2087 				&data, &sshdr);
2088 
2089 	if (scsi_status_is_good(res)) {
2090 		int offset = data.header_length + data.block_descriptor_length;
2091 
2092 		while (offset < len) {
2093 			u8 page_code = buffer[offset] & 0x3F;
2094 			u8 spf       = buffer[offset] & 0x40;
2095 
2096 			if (page_code == 8 || page_code == 6) {
2097 				/* We're interested only in the first 3 bytes.
2098 				 */
2099 				if (len - offset <= 2) {
2100 					sd_printk(KERN_ERR, sdkp, "Incomplete "
2101 						  "mode parameter data\n");
2102 					goto defaults;
2103 				} else {
2104 					modepage = page_code;
2105 					goto Page_found;
2106 				}
2107 			} else {
2108 				/* Go to the next page */
2109 				if (spf && len - offset > 3)
2110 					offset += 4 + (buffer[offset+2] << 8) +
2111 						buffer[offset+3];
2112 				else if (!spf && len - offset > 1)
2113 					offset += 2 + buffer[offset+1];
2114 				else {
2115 					sd_printk(KERN_ERR, sdkp, "Incomplete "
2116 						  "mode parameter data\n");
2117 					goto defaults;
2118 				}
2119 			}
2120 		}
2121 
2122 		if (modepage == 0x3F) {
2123 			sd_printk(KERN_ERR, sdkp, "No Caching mode page "
2124 				  "present\n");
2125 			goto defaults;
2126 		} else if ((buffer[offset] & 0x3f) != modepage) {
2127 			sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
2128 			goto defaults;
2129 		}
2130 	Page_found:
2131 		if (modepage == 8) {
2132 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2133 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2134 		} else {
2135 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2136 			sdkp->RCD = 0;
2137 		}
2138 
2139 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2140 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2141 			sd_printk(KERN_NOTICE, sdkp,
2142 				  "Uses READ/WRITE(6), disabling FUA\n");
2143 			sdkp->DPOFUA = 0;
2144 		}
2145 
2146 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
2147 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2148 			sd_printk(KERN_NOTICE, sdkp,
2149 				  "Write cache: %s, read cache: %s, %s\n",
2150 				  sdkp->WCE ? "enabled" : "disabled",
2151 				  sdkp->RCD ? "disabled" : "enabled",
2152 				  sdkp->DPOFUA ? "supports DPO and FUA"
2153 				  : "doesn't support DPO or FUA");
2154 
2155 		return;
2156 	}
2157 
2158 bad_sense:
2159 	if (scsi_sense_valid(&sshdr) &&
2160 	    sshdr.sense_key == ILLEGAL_REQUEST &&
2161 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2162 		/* Invalid field in CDB */
2163 		sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2164 	else
2165 		sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
2166 
2167 defaults:
2168 	sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
2169 	sdkp->WCE = 0;
2170 	sdkp->RCD = 0;
2171 	sdkp->DPOFUA = 0;
2172 }
2173 
2174 /*
2175  * The ATO bit indicates whether the DIF application tag is available
2176  * for use by the operating system.
2177  */
2178 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2179 {
2180 	int res, offset;
2181 	struct scsi_device *sdp = sdkp->device;
2182 	struct scsi_mode_data data;
2183 	struct scsi_sense_hdr sshdr;
2184 
2185 	if (sdp->type != TYPE_DISK)
2186 		return;
2187 
2188 	if (sdkp->protection_type == 0)
2189 		return;
2190 
2191 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2192 			      SD_MAX_RETRIES, &data, &sshdr);
2193 
2194 	if (!scsi_status_is_good(res) || !data.header_length ||
2195 	    data.length < 6) {
2196 		sd_printk(KERN_WARNING, sdkp,
2197 			  "getting Control mode page failed, assume no ATO\n");
2198 
2199 		if (scsi_sense_valid(&sshdr))
2200 			sd_print_sense_hdr(sdkp, &sshdr);
2201 
2202 		return;
2203 	}
2204 
2205 	offset = data.header_length + data.block_descriptor_length;
2206 
2207 	if ((buffer[offset] & 0x3f) != 0x0a) {
2208 		sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2209 		return;
2210 	}
2211 
2212 	if ((buffer[offset + 5] & 0x80) == 0)
2213 		return;
2214 
2215 	sdkp->ATO = 1;
2216 
2217 	return;
2218 }
2219 
2220 /**
2221  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2222  * @disk: disk to query
2223  */
2224 static void sd_read_block_limits(struct scsi_disk *sdkp)
2225 {
2226 	unsigned int sector_sz = sdkp->device->sector_size;
2227 	const int vpd_len = 64;
2228 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2229 
2230 	if (!buffer ||
2231 	    /* Block Limits VPD */
2232 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2233 		goto out;
2234 
2235 	blk_queue_io_min(sdkp->disk->queue,
2236 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2237 	blk_queue_io_opt(sdkp->disk->queue,
2238 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2239 
2240 	if (buffer[3] == 0x3c) {
2241 		unsigned int lba_count, desc_count;
2242 
2243 		sdkp->max_ws_blocks =
2244 			(u32) min_not_zero(get_unaligned_be64(&buffer[36]),
2245 					   (u64)0xffffffff);
2246 
2247 		if (!sdkp->lbpme)
2248 			goto out;
2249 
2250 		lba_count = get_unaligned_be32(&buffer[20]);
2251 		desc_count = get_unaligned_be32(&buffer[24]);
2252 
2253 		if (lba_count && desc_count)
2254 			sdkp->max_unmap_blocks = lba_count;
2255 
2256 		sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2257 
2258 		if (buffer[32] & 0x80)
2259 			sdkp->unmap_alignment =
2260 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2261 
2262 		if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2263 
2264 			if (sdkp->max_unmap_blocks)
2265 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2266 			else
2267 				sd_config_discard(sdkp, SD_LBP_WS16);
2268 
2269 		} else {	/* LBP VPD page tells us what to use */
2270 
2271 			if (sdkp->lbpu && sdkp->max_unmap_blocks)
2272 				sd_config_discard(sdkp, SD_LBP_UNMAP);
2273 			else if (sdkp->lbpws)
2274 				sd_config_discard(sdkp, SD_LBP_WS16);
2275 			else if (sdkp->lbpws10)
2276 				sd_config_discard(sdkp, SD_LBP_WS10);
2277 			else
2278 				sd_config_discard(sdkp, SD_LBP_DISABLE);
2279 		}
2280 	}
2281 
2282  out:
2283 	kfree(buffer);
2284 }
2285 
2286 /**
2287  * sd_read_block_characteristics - Query block dev. characteristics
2288  * @disk: disk to query
2289  */
2290 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2291 {
2292 	unsigned char *buffer;
2293 	u16 rot;
2294 	const int vpd_len = 64;
2295 
2296 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2297 
2298 	if (!buffer ||
2299 	    /* Block Device Characteristics VPD */
2300 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2301 		goto out;
2302 
2303 	rot = get_unaligned_be16(&buffer[4]);
2304 
2305 	if (rot == 1)
2306 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2307 
2308  out:
2309 	kfree(buffer);
2310 }
2311 
2312 /**
2313  * sd_read_block_provisioning - Query provisioning VPD page
2314  * @disk: disk to query
2315  */
2316 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2317 {
2318 	unsigned char *buffer;
2319 	const int vpd_len = 8;
2320 
2321 	if (sdkp->lbpme == 0)
2322 		return;
2323 
2324 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2325 
2326 	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2327 		goto out;
2328 
2329 	sdkp->lbpvpd	= 1;
2330 	sdkp->lbpu	= (buffer[5] >> 7) & 1;	/* UNMAP */
2331 	sdkp->lbpws	= (buffer[5] >> 6) & 1;	/* WRITE SAME(16) with UNMAP */
2332 	sdkp->lbpws10	= (buffer[5] >> 5) & 1;	/* WRITE SAME(10) with UNMAP */
2333 
2334  out:
2335 	kfree(buffer);
2336 }
2337 
2338 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2339 {
2340 	/*
2341 	 * Although VPD inquiries can go to SCSI-2 type devices,
2342 	 * some USB ones crash on receiving them, and the pages
2343 	 * we currently ask for are for SPC-3 and beyond
2344 	 */
2345 	if (sdp->scsi_level > SCSI_SPC_2)
2346 		return 1;
2347 	return 0;
2348 }
2349 
2350 /**
2351  *	sd_revalidate_disk - called the first time a new disk is seen,
2352  *	performs disk spin up, read_capacity, etc.
2353  *	@disk: struct gendisk we care about
2354  **/
2355 static int sd_revalidate_disk(struct gendisk *disk)
2356 {
2357 	struct scsi_disk *sdkp = scsi_disk(disk);
2358 	struct scsi_device *sdp = sdkp->device;
2359 	unsigned char *buffer;
2360 	unsigned flush = 0;
2361 
2362 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2363 				      "sd_revalidate_disk\n"));
2364 
2365 	/*
2366 	 * If the device is offline, don't try and read capacity or any
2367 	 * of the other niceties.
2368 	 */
2369 	if (!scsi_device_online(sdp))
2370 		goto out;
2371 
2372 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2373 	if (!buffer) {
2374 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2375 			  "allocation failure.\n");
2376 		goto out;
2377 	}
2378 
2379 	sd_spinup_disk(sdkp);
2380 
2381 	/*
2382 	 * Without media there is no reason to ask; moreover, some devices
2383 	 * react badly if we do.
2384 	 */
2385 	if (sdkp->media_present) {
2386 		sd_read_capacity(sdkp, buffer);
2387 
2388 		if (sd_try_extended_inquiry(sdp)) {
2389 			sd_read_block_provisioning(sdkp);
2390 			sd_read_block_limits(sdkp);
2391 			sd_read_block_characteristics(sdkp);
2392 		}
2393 
2394 		sd_read_write_protect_flag(sdkp, buffer);
2395 		sd_read_cache_type(sdkp, buffer);
2396 		sd_read_app_tag_own(sdkp, buffer);
2397 	}
2398 
2399 	sdkp->first_scan = 0;
2400 
2401 	/*
2402 	 * We now have all cache related info, determine how we deal
2403 	 * with flush requests.
2404 	 */
2405 	if (sdkp->WCE) {
2406 		flush |= REQ_FLUSH;
2407 		if (sdkp->DPOFUA)
2408 			flush |= REQ_FUA;
2409 	}
2410 
2411 	blk_queue_flush(sdkp->disk->queue, flush);
2412 
2413 	set_capacity(disk, sdkp->capacity);
2414 	kfree(buffer);
2415 
2416  out:
2417 	return 0;
2418 }
2419 
2420 /**
2421  *	sd_unlock_native_capacity - unlock native capacity
2422  *	@disk: struct gendisk to set capacity for
2423  *
2424  *	Block layer calls this function if it detects that partitions
2425  *	on @disk reach beyond the end of the device.  If the SCSI host
2426  *	implements ->unlock_native_capacity() method, it's invoked to
2427  *	give it a chance to adjust the device capacity.
2428  *
2429  *	CONTEXT:
2430  *	Defined by block layer.  Might sleep.
2431  */
2432 static void sd_unlock_native_capacity(struct gendisk *disk)
2433 {
2434 	struct scsi_device *sdev = scsi_disk(disk)->device;
2435 
2436 	if (sdev->host->hostt->unlock_native_capacity)
2437 		sdev->host->hostt->unlock_native_capacity(sdev);
2438 }
2439 
2440 /**
2441  *	sd_format_disk_name - format disk name
2442  *	@prefix: name prefix - ie. "sd" for SCSI disks
2443  *	@index: index of the disk to format name for
2444  *	@buf: output buffer
2445  *	@buflen: length of the output buffer
2446  *
2447  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2448  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2449  *	which is followed by sdaaa.
2450  *
2451  *	This is basically 26 base counting with one extra 'nil' entry
2452  *	at the beginning from the second digit on and can be
2453  *	determined using similar method as 26 base conversion with the
2454  *	index shifted -1 after each digit is computed.
2455  *
2456  *	CONTEXT:
2457  *	Don't care.
2458  *
2459  *	RETURNS:
2460  *	0 on success, -errno on failure.
2461  */
2462 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2463 {
2464 	const int base = 'z' - 'a' + 1;
2465 	char *begin = buf + strlen(prefix);
2466 	char *end = buf + buflen;
2467 	char *p;
2468 	int unit;
2469 
2470 	p = end - 1;
2471 	*p = '\0';
2472 	unit = base;
2473 	do {
2474 		if (p == begin)
2475 			return -EINVAL;
2476 		*--p = 'a' + (index % unit);
2477 		index = (index / unit) - 1;
2478 	} while (index >= 0);
2479 
2480 	memmove(begin, p, end - p);
2481 	memcpy(buf, prefix, strlen(prefix));
2482 
2483 	return 0;
2484 }
2485 
2486 /*
2487  * The asynchronous part of sd_probe
2488  */
2489 static void sd_probe_async(void *data, async_cookie_t cookie)
2490 {
2491 	struct scsi_disk *sdkp = data;
2492 	struct scsi_device *sdp;
2493 	struct gendisk *gd;
2494 	u32 index;
2495 	struct device *dev;
2496 
2497 	sdp = sdkp->device;
2498 	gd = sdkp->disk;
2499 	index = sdkp->index;
2500 	dev = &sdp->sdev_gendev;
2501 
2502 	gd->major = sd_major((index & 0xf0) >> 4);
2503 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2504 	gd->minors = SD_MINORS;
2505 
2506 	gd->fops = &sd_fops;
2507 	gd->private_data = &sdkp->driver;
2508 	gd->queue = sdkp->device->request_queue;
2509 
2510 	/* defaults, until the device tells us otherwise */
2511 	sdp->sector_size = 512;
2512 	sdkp->capacity = 0;
2513 	sdkp->media_present = 1;
2514 	sdkp->write_prot = 0;
2515 	sdkp->WCE = 0;
2516 	sdkp->RCD = 0;
2517 	sdkp->ATO = 0;
2518 	sdkp->first_scan = 1;
2519 
2520 	sd_revalidate_disk(gd);
2521 
2522 	blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2523 	blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2524 
2525 	gd->driverfs_dev = &sdp->sdev_gendev;
2526 	gd->flags = GENHD_FL_EXT_DEVT;
2527 	if (sdp->removable) {
2528 		gd->flags |= GENHD_FL_REMOVABLE;
2529 		gd->events |= DISK_EVENT_MEDIA_CHANGE;
2530 	}
2531 
2532 	add_disk(gd);
2533 	sd_dif_config_host(sdkp);
2534 
2535 	sd_revalidate_disk(gd);
2536 
2537 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2538 		  sdp->removable ? "removable " : "");
2539 	scsi_autopm_put_device(sdp);
2540 	put_device(&sdkp->dev);
2541 }
2542 
2543 /**
2544  *	sd_probe - called during driver initialization and whenever a
2545  *	new scsi device is attached to the system. It is called once
2546  *	for each scsi device (not just disks) present.
2547  *	@dev: pointer to device object
2548  *
2549  *	Returns 0 if successful (or not interested in this scsi device
2550  *	(e.g. scanner)); 1 when there is an error.
2551  *
2552  *	Note: this function is invoked from the scsi mid-level.
2553  *	This function sets up the mapping between a given
2554  *	<host,channel,id,lun> (found in sdp) and new device name
2555  *	(e.g. /dev/sda). More precisely it is the block device major
2556  *	and minor number that is chosen here.
2557  *
2558  *	Assume sd_attach is not re-entrant (for time being)
2559  *	Also think about sd_attach() and sd_remove() running coincidentally.
2560  **/
2561 static int sd_probe(struct device *dev)
2562 {
2563 	struct scsi_device *sdp = to_scsi_device(dev);
2564 	struct scsi_disk *sdkp;
2565 	struct gendisk *gd;
2566 	int index;
2567 	int error;
2568 
2569 	error = -ENODEV;
2570 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2571 		goto out;
2572 
2573 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2574 					"sd_attach\n"));
2575 
2576 	error = -ENOMEM;
2577 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2578 	if (!sdkp)
2579 		goto out;
2580 
2581 	gd = alloc_disk(SD_MINORS);
2582 	if (!gd)
2583 		goto out_free;
2584 
2585 	do {
2586 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2587 			goto out_put;
2588 
2589 		spin_lock(&sd_index_lock);
2590 		error = ida_get_new(&sd_index_ida, &index);
2591 		spin_unlock(&sd_index_lock);
2592 	} while (error == -EAGAIN);
2593 
2594 	if (error) {
2595 		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2596 		goto out_put;
2597 	}
2598 
2599 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2600 	if (error) {
2601 		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2602 		goto out_free_index;
2603 	}
2604 
2605 	sdkp->device = sdp;
2606 	sdkp->driver = &sd_template;
2607 	sdkp->disk = gd;
2608 	sdkp->index = index;
2609 	atomic_set(&sdkp->openers, 0);
2610 
2611 	if (!sdp->request_queue->rq_timeout) {
2612 		if (sdp->type != TYPE_MOD)
2613 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2614 		else
2615 			blk_queue_rq_timeout(sdp->request_queue,
2616 					     SD_MOD_TIMEOUT);
2617 	}
2618 
2619 	device_initialize(&sdkp->dev);
2620 	sdkp->dev.parent = dev;
2621 	sdkp->dev.class = &sd_disk_class;
2622 	dev_set_name(&sdkp->dev, dev_name(dev));
2623 
2624 	if (device_add(&sdkp->dev))
2625 		goto out_free_index;
2626 
2627 	get_device(dev);
2628 	dev_set_drvdata(dev, sdkp);
2629 
2630 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2631 	async_schedule(sd_probe_async, sdkp);
2632 
2633 	return 0;
2634 
2635  out_free_index:
2636 	spin_lock(&sd_index_lock);
2637 	ida_remove(&sd_index_ida, index);
2638 	spin_unlock(&sd_index_lock);
2639  out_put:
2640 	put_disk(gd);
2641  out_free:
2642 	kfree(sdkp);
2643  out:
2644 	return error;
2645 }
2646 
2647 /**
2648  *	sd_remove - called whenever a scsi disk (previously recognized by
2649  *	sd_probe) is detached from the system. It is called (potentially
2650  *	multiple times) during sd module unload.
2651  *	@sdp: pointer to mid level scsi device object
2652  *
2653  *	Note: this function is invoked from the scsi mid-level.
2654  *	This function potentially frees up a device name (e.g. /dev/sdc)
2655  *	that could be re-used by a subsequent sd_probe().
2656  *	This function is not called when the built-in sd driver is "exit-ed".
2657  **/
2658 static int sd_remove(struct device *dev)
2659 {
2660 	struct scsi_disk *sdkp;
2661 
2662 	sdkp = dev_get_drvdata(dev);
2663 	scsi_autopm_get_device(sdkp->device);
2664 
2665 	async_synchronize_full();
2666 	blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2667 	blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2668 	device_del(&sdkp->dev);
2669 	del_gendisk(sdkp->disk);
2670 	sd_shutdown(dev);
2671 
2672 	mutex_lock(&sd_ref_mutex);
2673 	dev_set_drvdata(dev, NULL);
2674 	put_device(&sdkp->dev);
2675 	mutex_unlock(&sd_ref_mutex);
2676 
2677 	return 0;
2678 }
2679 
2680 /**
2681  *	scsi_disk_release - Called to free the scsi_disk structure
2682  *	@dev: pointer to embedded class device
2683  *
2684  *	sd_ref_mutex must be held entering this routine.  Because it is
2685  *	called on last put, you should always use the scsi_disk_get()
2686  *	scsi_disk_put() helpers which manipulate the semaphore directly
2687  *	and never do a direct put_device.
2688  **/
2689 static void scsi_disk_release(struct device *dev)
2690 {
2691 	struct scsi_disk *sdkp = to_scsi_disk(dev);
2692 	struct gendisk *disk = sdkp->disk;
2693 
2694 	spin_lock(&sd_index_lock);
2695 	ida_remove(&sd_index_ida, sdkp->index);
2696 	spin_unlock(&sd_index_lock);
2697 
2698 	disk->private_data = NULL;
2699 	put_disk(disk);
2700 	put_device(&sdkp->device->sdev_gendev);
2701 
2702 	kfree(sdkp);
2703 }
2704 
2705 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2706 {
2707 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
2708 	struct scsi_sense_hdr sshdr;
2709 	struct scsi_device *sdp = sdkp->device;
2710 	int res;
2711 
2712 	if (start)
2713 		cmd[4] |= 1;	/* START */
2714 
2715 	if (sdp->start_stop_pwr_cond)
2716 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
2717 
2718 	if (!scsi_device_online(sdp))
2719 		return -ENODEV;
2720 
2721 	res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2722 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2723 	if (res) {
2724 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2725 		sd_print_result(sdkp, res);
2726 		if (driver_byte(res) & DRIVER_SENSE)
2727 			sd_print_sense_hdr(sdkp, &sshdr);
2728 	}
2729 
2730 	return res;
2731 }
2732 
2733 /*
2734  * Send a SYNCHRONIZE CACHE instruction down to the device through
2735  * the normal SCSI command structure.  Wait for the command to
2736  * complete.
2737  */
2738 static void sd_shutdown(struct device *dev)
2739 {
2740 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2741 
2742 	if (!sdkp)
2743 		return;         /* this can happen */
2744 
2745 	if (pm_runtime_suspended(dev))
2746 		goto exit;
2747 
2748 	if (sdkp->WCE) {
2749 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2750 		sd_sync_cache(sdkp);
2751 	}
2752 
2753 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2754 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2755 		sd_start_stop_device(sdkp, 0);
2756 	}
2757 
2758 exit:
2759 	scsi_disk_put(sdkp);
2760 }
2761 
2762 static int sd_suspend(struct device *dev, pm_message_t mesg)
2763 {
2764 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2765 	int ret = 0;
2766 
2767 	if (!sdkp)
2768 		return 0;	/* this can happen */
2769 
2770 	if (sdkp->WCE) {
2771 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2772 		ret = sd_sync_cache(sdkp);
2773 		if (ret)
2774 			goto done;
2775 	}
2776 
2777 	if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2778 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2779 		ret = sd_start_stop_device(sdkp, 0);
2780 	}
2781 
2782 done:
2783 	scsi_disk_put(sdkp);
2784 	return ret;
2785 }
2786 
2787 static int sd_resume(struct device *dev)
2788 {
2789 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2790 	int ret = 0;
2791 
2792 	if (!sdkp->device->manage_start_stop)
2793 		goto done;
2794 
2795 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2796 	ret = sd_start_stop_device(sdkp, 1);
2797 
2798 done:
2799 	scsi_disk_put(sdkp);
2800 	return ret;
2801 }
2802 
2803 /**
2804  *	init_sd - entry point for this driver (both when built in or when
2805  *	a module).
2806  *
2807  *	Note: this function registers this driver with the scsi mid-level.
2808  **/
2809 static int __init init_sd(void)
2810 {
2811 	int majors = 0, i, err;
2812 
2813 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2814 
2815 	for (i = 0; i < SD_MAJORS; i++)
2816 		if (register_blkdev(sd_major(i), "sd") == 0)
2817 			majors++;
2818 
2819 	if (!majors)
2820 		return -ENODEV;
2821 
2822 	err = class_register(&sd_disk_class);
2823 	if (err)
2824 		goto err_out;
2825 
2826 	err = scsi_register_driver(&sd_template.gendrv);
2827 	if (err)
2828 		goto err_out_class;
2829 
2830 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2831 					 0, 0, NULL);
2832 	if (!sd_cdb_cache) {
2833 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
2834 		goto err_out_class;
2835 	}
2836 
2837 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2838 	if (!sd_cdb_pool) {
2839 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
2840 		goto err_out_cache;
2841 	}
2842 
2843 	return 0;
2844 
2845 err_out_cache:
2846 	kmem_cache_destroy(sd_cdb_cache);
2847 
2848 err_out_class:
2849 	class_unregister(&sd_disk_class);
2850 err_out:
2851 	for (i = 0; i < SD_MAJORS; i++)
2852 		unregister_blkdev(sd_major(i), "sd");
2853 	return err;
2854 }
2855 
2856 /**
2857  *	exit_sd - exit point for this driver (when it is a module).
2858  *
2859  *	Note: this function unregisters this driver from the scsi mid-level.
2860  **/
2861 static void __exit exit_sd(void)
2862 {
2863 	int i;
2864 
2865 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
2866 
2867 	mempool_destroy(sd_cdb_pool);
2868 	kmem_cache_destroy(sd_cdb_cache);
2869 
2870 	scsi_unregister_driver(&sd_template.gendrv);
2871 	class_unregister(&sd_disk_class);
2872 
2873 	for (i = 0; i < SD_MAJORS; i++)
2874 		unregister_blkdev(sd_major(i), "sd");
2875 }
2876 
2877 module_init(init_sd);
2878 module_exit(exit_sd);
2879 
2880 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
2881 			       struct scsi_sense_hdr *sshdr)
2882 {
2883 	sd_printk(KERN_INFO, sdkp, " ");
2884 	scsi_show_sense_hdr(sshdr);
2885 	sd_printk(KERN_INFO, sdkp, " ");
2886 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
2887 }
2888 
2889 static void sd_print_result(struct scsi_disk *sdkp, int result)
2890 {
2891 	sd_printk(KERN_INFO, sdkp, " ");
2892 	scsi_show_result(result);
2893 }
2894 
2895