xref: /linux/drivers/scsi/sd.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
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/smp_lock.h>
50 #include <linux/mutex.h>
51 #include <linux/string_helpers.h>
52 #include <linux/async.h>
53 #include <linux/slab.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 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_app_tag_own(struct device *dev, struct device_attribute *attr,
263 		    char *buf)
264 {
265 	struct scsi_disk *sdkp = to_scsi_disk(dev);
266 
267 	return snprintf(buf, 20, "%u\n", sdkp->ATO);
268 }
269 
270 static ssize_t
271 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
272 			  char *buf)
273 {
274 	struct scsi_disk *sdkp = to_scsi_disk(dev);
275 
276 	return snprintf(buf, 20, "%u\n", sdkp->thin_provisioning);
277 }
278 
279 static struct device_attribute sd_disk_attrs[] = {
280 	__ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
281 	       sd_store_cache_type),
282 	__ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
283 	__ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
284 	       sd_store_allow_restart),
285 	__ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
286 	       sd_store_manage_start_stop),
287 	__ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
288 	__ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
289 	__ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
290 	__ATTR_NULL,
291 };
292 
293 static struct class sd_disk_class = {
294 	.name		= "scsi_disk",
295 	.owner		= THIS_MODULE,
296 	.dev_release	= scsi_disk_release,
297 	.dev_attrs	= sd_disk_attrs,
298 };
299 
300 static struct scsi_driver sd_template = {
301 	.owner			= THIS_MODULE,
302 	.gendrv = {
303 		.name		= "sd",
304 		.probe		= sd_probe,
305 		.remove		= sd_remove,
306 		.suspend	= sd_suspend,
307 		.resume		= sd_resume,
308 		.shutdown	= sd_shutdown,
309 	},
310 	.rescan			= sd_rescan,
311 	.done			= sd_done,
312 };
313 
314 /*
315  * Device no to disk mapping:
316  *
317  *       major         disc2     disc  p1
318  *   |............|.............|....|....| <- dev_t
319  *    31        20 19          8 7  4 3  0
320  *
321  * Inside a major, we have 16k disks, however mapped non-
322  * contiguously. The first 16 disks are for major0, the next
323  * ones with major1, ... Disk 256 is for major0 again, disk 272
324  * for major1, ...
325  * As we stay compatible with our numbering scheme, we can reuse
326  * the well-know SCSI majors 8, 65--71, 136--143.
327  */
328 static int sd_major(int major_idx)
329 {
330 	switch (major_idx) {
331 	case 0:
332 		return SCSI_DISK0_MAJOR;
333 	case 1 ... 7:
334 		return SCSI_DISK1_MAJOR + major_idx - 1;
335 	case 8 ... 15:
336 		return SCSI_DISK8_MAJOR + major_idx - 8;
337 	default:
338 		BUG();
339 		return 0;	/* shut up gcc */
340 	}
341 }
342 
343 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
344 {
345 	struct scsi_disk *sdkp = NULL;
346 
347 	if (disk->private_data) {
348 		sdkp = scsi_disk(disk);
349 		if (scsi_device_get(sdkp->device) == 0)
350 			get_device(&sdkp->dev);
351 		else
352 			sdkp = NULL;
353 	}
354 	return sdkp;
355 }
356 
357 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
358 {
359 	struct scsi_disk *sdkp;
360 
361 	mutex_lock(&sd_ref_mutex);
362 	sdkp = __scsi_disk_get(disk);
363 	mutex_unlock(&sd_ref_mutex);
364 	return sdkp;
365 }
366 
367 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
368 {
369 	struct scsi_disk *sdkp;
370 
371 	mutex_lock(&sd_ref_mutex);
372 	sdkp = dev_get_drvdata(dev);
373 	if (sdkp)
374 		sdkp = __scsi_disk_get(sdkp->disk);
375 	mutex_unlock(&sd_ref_mutex);
376 	return sdkp;
377 }
378 
379 static void scsi_disk_put(struct scsi_disk *sdkp)
380 {
381 	struct scsi_device *sdev = sdkp->device;
382 
383 	mutex_lock(&sd_ref_mutex);
384 	put_device(&sdkp->dev);
385 	scsi_device_put(sdev);
386 	mutex_unlock(&sd_ref_mutex);
387 }
388 
389 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
390 {
391 	unsigned int prot_op = SCSI_PROT_NORMAL;
392 	unsigned int dix = scsi_prot_sg_count(scmd);
393 
394 	if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
395 		if (dif && dix)
396 			prot_op = SCSI_PROT_READ_PASS;
397 		else if (dif && !dix)
398 			prot_op = SCSI_PROT_READ_STRIP;
399 		else if (!dif && dix)
400 			prot_op = SCSI_PROT_READ_INSERT;
401 	} else {
402 		if (dif && dix)
403 			prot_op = SCSI_PROT_WRITE_PASS;
404 		else if (dif && !dix)
405 			prot_op = SCSI_PROT_WRITE_INSERT;
406 		else if (!dif && dix)
407 			prot_op = SCSI_PROT_WRITE_STRIP;
408 	}
409 
410 	scsi_set_prot_op(scmd, prot_op);
411 	scsi_set_prot_type(scmd, dif);
412 }
413 
414 /**
415  * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
416  * @sdp: scsi device to operate one
417  * @rq: Request to prepare
418  *
419  * Will issue either UNMAP or WRITE SAME(16) depending on preference
420  * indicated by target device.
421  **/
422 static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
423 {
424 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
425 	struct bio *bio = rq->bio;
426 	sector_t sector = bio->bi_sector;
427 	unsigned int nr_sectors = bio_sectors(bio);
428 	unsigned int len;
429 	int ret;
430 	struct page *page;
431 
432 	if (sdkp->device->sector_size == 4096) {
433 		sector >>= 3;
434 		nr_sectors >>= 3;
435 	}
436 
437 	rq->timeout = SD_TIMEOUT;
438 
439 	memset(rq->cmd, 0, rq->cmd_len);
440 
441 	page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
442 	if (!page)
443 		return BLKPREP_DEFER;
444 
445 	if (sdkp->unmap) {
446 		char *buf = page_address(page);
447 
448 		rq->cmd_len = 10;
449 		rq->cmd[0] = UNMAP;
450 		rq->cmd[8] = 24;
451 
452 		put_unaligned_be16(6 + 16, &buf[0]);
453 		put_unaligned_be16(16, &buf[2]);
454 		put_unaligned_be64(sector, &buf[8]);
455 		put_unaligned_be32(nr_sectors, &buf[16]);
456 
457 		len = 24;
458 	} else {
459 		rq->cmd_len = 16;
460 		rq->cmd[0] = WRITE_SAME_16;
461 		rq->cmd[1] = 0x8; /* UNMAP */
462 		put_unaligned_be64(sector, &rq->cmd[2]);
463 		put_unaligned_be32(nr_sectors, &rq->cmd[10]);
464 
465 		len = sdkp->device->sector_size;
466 	}
467 
468 	blk_add_request_payload(rq, page, len);
469 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
470 	rq->buffer = page_address(page);
471 	if (ret != BLKPREP_OK) {
472 		__free_page(page);
473 		rq->buffer = NULL;
474 	}
475 	return ret;
476 }
477 
478 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
479 {
480 	rq->timeout = SD_TIMEOUT;
481 	rq->retries = SD_MAX_RETRIES;
482 	rq->cmd[0] = SYNCHRONIZE_CACHE;
483 	rq->cmd_len = 10;
484 
485 	return scsi_setup_blk_pc_cmnd(sdp, rq);
486 }
487 
488 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
489 {
490 	if (rq->cmd_flags & REQ_DISCARD) {
491 		free_page((unsigned long)rq->buffer);
492 		rq->buffer = NULL;
493 	}
494 }
495 
496 /**
497  *	sd_init_command - build a scsi (read or write) command from
498  *	information in the request structure.
499  *	@SCpnt: pointer to mid-level's per scsi command structure that
500  *	contains request and into which the scsi command is written
501  *
502  *	Returns 1 if successful and 0 if error (or cannot be done now).
503  **/
504 static int sd_prep_fn(struct request_queue *q, struct request *rq)
505 {
506 	struct scsi_cmnd *SCpnt;
507 	struct scsi_device *sdp = q->queuedata;
508 	struct gendisk *disk = rq->rq_disk;
509 	struct scsi_disk *sdkp;
510 	sector_t block = blk_rq_pos(rq);
511 	sector_t threshold;
512 	unsigned int this_count = blk_rq_sectors(rq);
513 	int ret, host_dif;
514 	unsigned char protect;
515 
516 	/*
517 	 * Discard request come in as REQ_TYPE_FS but we turn them into
518 	 * block PC requests to make life easier.
519 	 */
520 	if (rq->cmd_flags & REQ_DISCARD) {
521 		ret = scsi_setup_discard_cmnd(sdp, rq);
522 		goto out;
523 	} else if (rq->cmd_flags & REQ_FLUSH) {
524 		ret = scsi_setup_flush_cmnd(sdp, rq);
525 		goto out;
526 	} else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
527 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
528 		goto out;
529 	} else if (rq->cmd_type != REQ_TYPE_FS) {
530 		ret = BLKPREP_KILL;
531 		goto out;
532 	}
533 	ret = scsi_setup_fs_cmnd(sdp, rq);
534 	if (ret != BLKPREP_OK)
535 		goto out;
536 	SCpnt = rq->special;
537 	sdkp = scsi_disk(disk);
538 
539 	/* from here on until we're complete, any goto out
540 	 * is used for a killable error condition */
541 	ret = BLKPREP_KILL;
542 
543 	SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
544 					"sd_init_command: block=%llu, "
545 					"count=%d\n",
546 					(unsigned long long)block,
547 					this_count));
548 
549 	if (!sdp || !scsi_device_online(sdp) ||
550 	    block + blk_rq_sectors(rq) > get_capacity(disk)) {
551 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
552 						"Finishing %u sectors\n",
553 						blk_rq_sectors(rq)));
554 		SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
555 						"Retry with 0x%p\n", SCpnt));
556 		goto out;
557 	}
558 
559 	if (sdp->changed) {
560 		/*
561 		 * quietly refuse to do anything to a changed disc until
562 		 * the changed bit has been reset
563 		 */
564 		/* printk("SCSI disk has been changed. Prohibiting further I/O.\n"); */
565 		goto out;
566 	}
567 
568 	/*
569 	 * Some SD card readers can't handle multi-sector accesses which touch
570 	 * the last one or two hardware sectors.  Split accesses as needed.
571 	 */
572 	threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
573 		(sdp->sector_size / 512);
574 
575 	if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
576 		if (block < threshold) {
577 			/* Access up to the threshold but not beyond */
578 			this_count = threshold - block;
579 		} else {
580 			/* Access only a single hardware sector */
581 			this_count = sdp->sector_size / 512;
582 		}
583 	}
584 
585 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
586 					(unsigned long long)block));
587 
588 	/*
589 	 * If we have a 1K hardware sectorsize, prevent access to single
590 	 * 512 byte sectors.  In theory we could handle this - in fact
591 	 * the scsi cdrom driver must be able to handle this because
592 	 * we typically use 1K blocksizes, and cdroms typically have
593 	 * 2K hardware sectorsizes.  Of course, things are simpler
594 	 * with the cdrom, since it is read-only.  For performance
595 	 * reasons, the filesystems should be able to handle this
596 	 * and not force the scsi disk driver to use bounce buffers
597 	 * for this.
598 	 */
599 	if (sdp->sector_size == 1024) {
600 		if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
601 			scmd_printk(KERN_ERR, SCpnt,
602 				    "Bad block number requested\n");
603 			goto out;
604 		} else {
605 			block = block >> 1;
606 			this_count = this_count >> 1;
607 		}
608 	}
609 	if (sdp->sector_size == 2048) {
610 		if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
611 			scmd_printk(KERN_ERR, SCpnt,
612 				    "Bad block number requested\n");
613 			goto out;
614 		} else {
615 			block = block >> 2;
616 			this_count = this_count >> 2;
617 		}
618 	}
619 	if (sdp->sector_size == 4096) {
620 		if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
621 			scmd_printk(KERN_ERR, SCpnt,
622 				    "Bad block number requested\n");
623 			goto out;
624 		} else {
625 			block = block >> 3;
626 			this_count = this_count >> 3;
627 		}
628 	}
629 	if (rq_data_dir(rq) == WRITE) {
630 		if (!sdp->writeable) {
631 			goto out;
632 		}
633 		SCpnt->cmnd[0] = WRITE_6;
634 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
635 
636 		if (blk_integrity_rq(rq) &&
637 		    sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
638 			goto out;
639 
640 	} else if (rq_data_dir(rq) == READ) {
641 		SCpnt->cmnd[0] = READ_6;
642 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
643 	} else {
644 		scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
645 		goto out;
646 	}
647 
648 	SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
649 					"%s %d/%u 512 byte blocks.\n",
650 					(rq_data_dir(rq) == WRITE) ?
651 					"writing" : "reading", this_count,
652 					blk_rq_sectors(rq)));
653 
654 	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
655 	host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
656 	if (host_dif)
657 		protect = 1 << 5;
658 	else
659 		protect = 0;
660 
661 	if (host_dif == SD_DIF_TYPE2_PROTECTION) {
662 		SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
663 
664 		if (unlikely(SCpnt->cmnd == NULL)) {
665 			ret = BLKPREP_DEFER;
666 			goto out;
667 		}
668 
669 		SCpnt->cmd_len = SD_EXT_CDB_SIZE;
670 		memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
671 		SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
672 		SCpnt->cmnd[7] = 0x18;
673 		SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
674 		SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
675 
676 		/* LBA */
677 		SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
678 		SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
679 		SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
680 		SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
681 		SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
682 		SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
683 		SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
684 		SCpnt->cmnd[19] = (unsigned char) block & 0xff;
685 
686 		/* Expected Indirect LBA */
687 		SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
688 		SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
689 		SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
690 		SCpnt->cmnd[23] = (unsigned char) block & 0xff;
691 
692 		/* Transfer length */
693 		SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
694 		SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
695 		SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
696 		SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
697 	} else if (block > 0xffffffff) {
698 		SCpnt->cmnd[0] += READ_16 - READ_6;
699 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
700 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
701 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
702 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
703 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
704 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
705 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
706 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
707 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
708 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
709 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
710 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
711 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
712 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
713 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
714 		   scsi_device_protection(SCpnt->device) ||
715 		   SCpnt->device->use_10_for_rw) {
716 		if (this_count > 0xffff)
717 			this_count = 0xffff;
718 
719 		SCpnt->cmnd[0] += READ_10 - READ_6;
720 		SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
721 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
722 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
723 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
724 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
725 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
726 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
727 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
728 	} else {
729 		if (unlikely(rq->cmd_flags & REQ_FUA)) {
730 			/*
731 			 * This happens only if this drive failed
732 			 * 10byte rw command with ILLEGAL_REQUEST
733 			 * during operation and thus turned off
734 			 * use_10_for_rw.
735 			 */
736 			scmd_printk(KERN_ERR, SCpnt,
737 				    "FUA write on READ/WRITE(6) drive\n");
738 			goto out;
739 		}
740 
741 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
742 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
743 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
744 		SCpnt->cmnd[4] = (unsigned char) this_count;
745 		SCpnt->cmnd[5] = 0;
746 	}
747 	SCpnt->sdb.length = this_count * sdp->sector_size;
748 
749 	/* If DIF or DIX is enabled, tell HBA how to handle request */
750 	if (host_dif || scsi_prot_sg_count(SCpnt))
751 		sd_prot_op(SCpnt, host_dif);
752 
753 	/*
754 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
755 	 * host adapter, it's safe to assume that we can at least transfer
756 	 * this many bytes between each connect / disconnect.
757 	 */
758 	SCpnt->transfersize = sdp->sector_size;
759 	SCpnt->underflow = this_count << 9;
760 	SCpnt->allowed = SD_MAX_RETRIES;
761 
762 	/*
763 	 * This indicates that the command is ready from our end to be
764 	 * queued.
765 	 */
766 	ret = BLKPREP_OK;
767  out:
768 	return scsi_prep_return(q, rq, ret);
769 }
770 
771 /**
772  *	sd_open - open a scsi disk device
773  *	@inode: only i_rdev member may be used
774  *	@filp: only f_mode and f_flags may be used
775  *
776  *	Returns 0 if successful. Returns a negated errno value in case
777  *	of error.
778  *
779  *	Note: This can be called from a user context (e.g. fsck(1) )
780  *	or from within the kernel (e.g. as a result of a mount(1) ).
781  *	In the latter case @inode and @filp carry an abridged amount
782  *	of information as noted above.
783  *
784  *	Locking: called with bdev->bd_mutex held.
785  **/
786 static int sd_open(struct block_device *bdev, fmode_t mode)
787 {
788 	struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
789 	struct scsi_device *sdev;
790 	int retval;
791 
792 	if (!sdkp)
793 		return -ENXIO;
794 
795 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
796 
797 	sdev = sdkp->device;
798 
799 	retval = scsi_autopm_get_device(sdev);
800 	if (retval)
801 		goto error_autopm;
802 
803 	/*
804 	 * If the device is in error recovery, wait until it is done.
805 	 * If the device is offline, then disallow any access to it.
806 	 */
807 	retval = -ENXIO;
808 	if (!scsi_block_when_processing_errors(sdev))
809 		goto error_out;
810 
811 	if (sdev->removable || sdkp->write_prot)
812 		check_disk_change(bdev);
813 
814 	/*
815 	 * If the drive is empty, just let the open fail.
816 	 */
817 	retval = -ENOMEDIUM;
818 	if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
819 		goto error_out;
820 
821 	/*
822 	 * If the device has the write protect tab set, have the open fail
823 	 * if the user expects to be able to write to the thing.
824 	 */
825 	retval = -EROFS;
826 	if (sdkp->write_prot && (mode & FMODE_WRITE))
827 		goto error_out;
828 
829 	/*
830 	 * It is possible that the disk changing stuff resulted in
831 	 * the device being taken offline.  If this is the case,
832 	 * report this to the user, and don't pretend that the
833 	 * open actually succeeded.
834 	 */
835 	retval = -ENXIO;
836 	if (!scsi_device_online(sdev))
837 		goto error_out;
838 
839 	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
840 		if (scsi_block_when_processing_errors(sdev))
841 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
842 	}
843 
844 	return 0;
845 
846 error_out:
847 	scsi_autopm_put_device(sdev);
848 error_autopm:
849 	scsi_disk_put(sdkp);
850 	return retval;
851 }
852 
853 /**
854  *	sd_release - invoked when the (last) close(2) is called on this
855  *	scsi disk.
856  *	@inode: only i_rdev member may be used
857  *	@filp: only f_mode and f_flags may be used
858  *
859  *	Returns 0.
860  *
861  *	Note: may block (uninterruptible) if error recovery is underway
862  *	on this disk.
863  *
864  *	Locking: called with bdev->bd_mutex held.
865  **/
866 static int sd_release(struct gendisk *disk, fmode_t mode)
867 {
868 	struct scsi_disk *sdkp = scsi_disk(disk);
869 	struct scsi_device *sdev = sdkp->device;
870 
871 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
872 
873 	if (atomic_dec_return(&sdkp->openers) && sdev->removable) {
874 		if (scsi_block_when_processing_errors(sdev))
875 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
876 	}
877 
878 	/*
879 	 * XXX and what if there are packets in flight and this close()
880 	 * XXX is followed by a "rmmod sd_mod"?
881 	 */
882 
883 	scsi_autopm_put_device(sdev);
884 	scsi_disk_put(sdkp);
885 	return 0;
886 }
887 
888 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
889 {
890 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
891 	struct scsi_device *sdp = sdkp->device;
892 	struct Scsi_Host *host = sdp->host;
893 	int diskinfo[4];
894 
895 	/* default to most commonly used values */
896         diskinfo[0] = 0x40;	/* 1 << 6 */
897        	diskinfo[1] = 0x20;	/* 1 << 5 */
898        	diskinfo[2] = sdkp->capacity >> 11;
899 
900 	/* override with calculated, extended default, or driver values */
901 	if (host->hostt->bios_param)
902 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
903 	else
904 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
905 
906 	geo->heads = diskinfo[0];
907 	geo->sectors = diskinfo[1];
908 	geo->cylinders = diskinfo[2];
909 	return 0;
910 }
911 
912 /**
913  *	sd_ioctl - process an ioctl
914  *	@inode: only i_rdev/i_bdev members may be used
915  *	@filp: only f_mode and f_flags may be used
916  *	@cmd: ioctl command number
917  *	@arg: this is third argument given to ioctl(2) system call.
918  *	Often contains a pointer.
919  *
920  *	Returns 0 if successful (some ioctls return postive numbers on
921  *	success as well). Returns a negated errno value in case of error.
922  *
923  *	Note: most ioctls are forward onto the block subsystem or further
924  *	down in the scsi subsystem.
925  **/
926 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
927 		    unsigned int cmd, unsigned long arg)
928 {
929 	struct gendisk *disk = bdev->bd_disk;
930 	struct scsi_device *sdp = scsi_disk(disk)->device;
931 	void __user *p = (void __user *)arg;
932 	int error;
933 
934 	SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
935 						disk->disk_name, cmd));
936 
937 	/*
938 	 * If we are in the middle of error recovery, don't let anyone
939 	 * else try and use this device.  Also, if error recovery fails, it
940 	 * may try and take the device offline, in which case all further
941 	 * access to the device is prohibited.
942 	 */
943 	error = scsi_nonblockable_ioctl(sdp, cmd, p,
944 					(mode & FMODE_NDELAY) != 0);
945 	if (!scsi_block_when_processing_errors(sdp) || !error)
946 		goto out;
947 
948 	/*
949 	 * Send SCSI addressing ioctls directly to mid level, send other
950 	 * ioctls to block level and then onto mid level if they can't be
951 	 * resolved.
952 	 */
953 	switch (cmd) {
954 		case SCSI_IOCTL_GET_IDLUN:
955 		case SCSI_IOCTL_GET_BUS_NUMBER:
956 			error = scsi_ioctl(sdp, cmd, p);
957 			break;
958 		default:
959 			error = scsi_cmd_ioctl(disk->queue, disk, mode, cmd, p);
960 			if (error != -ENOTTY)
961 				break;
962 			error = scsi_ioctl(sdp, cmd, p);
963 			break;
964 	}
965 out:
966 	return error;
967 }
968 
969 static void set_media_not_present(struct scsi_disk *sdkp)
970 {
971 	sdkp->media_present = 0;
972 	sdkp->capacity = 0;
973 	sdkp->device->changed = 1;
974 }
975 
976 /**
977  *	sd_media_changed - check if our medium changed
978  *	@disk: kernel device descriptor
979  *
980  *	Returns 0 if not applicable or no change; 1 if change
981  *
982  *	Note: this function is invoked from the block subsystem.
983  **/
984 static int sd_media_changed(struct gendisk *disk)
985 {
986 	struct scsi_disk *sdkp = scsi_disk(disk);
987 	struct scsi_device *sdp = sdkp->device;
988 	struct scsi_sense_hdr *sshdr = NULL;
989 	int retval;
990 
991 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_media_changed\n"));
992 
993 	if (!sdp->removable)
994 		return 0;
995 
996 	/*
997 	 * If the device is offline, don't send any commands - just pretend as
998 	 * if the command failed.  If the device ever comes back online, we
999 	 * can deal with it then.  It is only because of unrecoverable errors
1000 	 * that we would ever take a device offline in the first place.
1001 	 */
1002 	if (!scsi_device_online(sdp)) {
1003 		set_media_not_present(sdkp);
1004 		retval = 1;
1005 		goto out;
1006 	}
1007 
1008 	/*
1009 	 * Using TEST_UNIT_READY enables differentiation between drive with
1010 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1011 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1012 	 *
1013 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1014 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1015 	 * sd_revalidate() is called.
1016 	 */
1017 	retval = -ENODEV;
1018 
1019 	if (scsi_block_when_processing_errors(sdp)) {
1020 		sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1021 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1022 					      sshdr);
1023 	}
1024 
1025 	/*
1026 	 * Unable to test, unit probably not ready.   This usually
1027 	 * means there is no disc in the drive.  Mark as changed,
1028 	 * and we will figure it out later once the drive is
1029 	 * available again.
1030 	 */
1031 	if (retval || (scsi_sense_valid(sshdr) &&
1032 		       /* 0x3a is medium not present */
1033 		       sshdr->asc == 0x3a)) {
1034 		set_media_not_present(sdkp);
1035 		retval = 1;
1036 		goto out;
1037 	}
1038 
1039 	/*
1040 	 * For removable scsi disk we have to recognise the presence
1041 	 * of a disk in the drive. This is kept in the struct scsi_disk
1042 	 * struct and tested at open !  Daniel Roche (dan@lectra.fr)
1043 	 */
1044 	sdkp->media_present = 1;
1045 
1046 	retval = sdp->changed;
1047 	sdp->changed = 0;
1048 out:
1049 	if (retval != sdkp->previous_state)
1050 		sdev_evt_send_simple(sdp, SDEV_EVT_MEDIA_CHANGE, GFP_KERNEL);
1051 	sdkp->previous_state = retval;
1052 	kfree(sshdr);
1053 	return retval;
1054 }
1055 
1056 static int sd_sync_cache(struct scsi_disk *sdkp)
1057 {
1058 	int retries, res;
1059 	struct scsi_device *sdp = sdkp->device;
1060 	struct scsi_sense_hdr sshdr;
1061 
1062 	if (!scsi_device_online(sdp))
1063 		return -ENODEV;
1064 
1065 
1066 	for (retries = 3; retries > 0; --retries) {
1067 		unsigned char cmd[10] = { 0 };
1068 
1069 		cmd[0] = SYNCHRONIZE_CACHE;
1070 		/*
1071 		 * Leave the rest of the command zero to indicate
1072 		 * flush everything.
1073 		 */
1074 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1075 				       SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1076 		if (res == 0)
1077 			break;
1078 	}
1079 
1080 	if (res) {
1081 		sd_print_result(sdkp, res);
1082 		if (driver_byte(res) & DRIVER_SENSE)
1083 			sd_print_sense_hdr(sdkp, &sshdr);
1084 	}
1085 
1086 	if (res)
1087 		return -EIO;
1088 	return 0;
1089 }
1090 
1091 static void sd_rescan(struct device *dev)
1092 {
1093 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1094 
1095 	if (sdkp) {
1096 		revalidate_disk(sdkp->disk);
1097 		scsi_disk_put(sdkp);
1098 	}
1099 }
1100 
1101 
1102 #ifdef CONFIG_COMPAT
1103 /*
1104  * This gets directly called from VFS. When the ioctl
1105  * is not recognized we go back to the other translation paths.
1106  */
1107 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1108 			   unsigned int cmd, unsigned long arg)
1109 {
1110 	struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1111 
1112 	/*
1113 	 * If we are in the middle of error recovery, don't let anyone
1114 	 * else try and use this device.  Also, if error recovery fails, it
1115 	 * may try and take the device offline, in which case all further
1116 	 * access to the device is prohibited.
1117 	 */
1118 	if (!scsi_block_when_processing_errors(sdev))
1119 		return -ENODEV;
1120 
1121 	if (sdev->host->hostt->compat_ioctl) {
1122 		int ret;
1123 
1124 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1125 
1126 		return ret;
1127 	}
1128 
1129 	/*
1130 	 * Let the static ioctl translation table take care of it.
1131 	 */
1132 	return -ENOIOCTLCMD;
1133 }
1134 #endif
1135 
1136 static const struct block_device_operations sd_fops = {
1137 	.owner			= THIS_MODULE,
1138 	.open			= sd_open,
1139 	.release		= sd_release,
1140 	.ioctl			= sd_ioctl,
1141 	.getgeo			= sd_getgeo,
1142 #ifdef CONFIG_COMPAT
1143 	.compat_ioctl		= sd_compat_ioctl,
1144 #endif
1145 	.media_changed		= sd_media_changed,
1146 	.revalidate_disk	= sd_revalidate_disk,
1147 	.unlock_native_capacity	= sd_unlock_native_capacity,
1148 };
1149 
1150 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1151 {
1152 	u64 start_lba = blk_rq_pos(scmd->request);
1153 	u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1154 	u64 bad_lba;
1155 	int info_valid;
1156 
1157 	if (scmd->request->cmd_type != REQ_TYPE_FS)
1158 		return 0;
1159 
1160 	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1161 					     SCSI_SENSE_BUFFERSIZE,
1162 					     &bad_lba);
1163 	if (!info_valid)
1164 		return 0;
1165 
1166 	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1167 		return 0;
1168 
1169 	if (scmd->device->sector_size < 512) {
1170 		/* only legitimate sector_size here is 256 */
1171 		start_lba <<= 1;
1172 		end_lba <<= 1;
1173 	} else {
1174 		/* be careful ... don't want any overflows */
1175 		u64 factor = scmd->device->sector_size / 512;
1176 		do_div(start_lba, factor);
1177 		do_div(end_lba, factor);
1178 	}
1179 
1180 	/* The bad lba was reported incorrectly, we have no idea where
1181 	 * the error is.
1182 	 */
1183 	if (bad_lba < start_lba  || bad_lba >= end_lba)
1184 		return 0;
1185 
1186 	/* This computation should always be done in terms of
1187 	 * the resolution of the device's medium.
1188 	 */
1189 	return (bad_lba - start_lba) * scmd->device->sector_size;
1190 }
1191 
1192 /**
1193  *	sd_done - bottom half handler: called when the lower level
1194  *	driver has completed (successfully or otherwise) a scsi command.
1195  *	@SCpnt: mid-level's per command structure.
1196  *
1197  *	Note: potentially run from within an ISR. Must not block.
1198  **/
1199 static int sd_done(struct scsi_cmnd *SCpnt)
1200 {
1201 	int result = SCpnt->result;
1202 	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1203 	struct scsi_sense_hdr sshdr;
1204 	struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1205 	int sense_valid = 0;
1206 	int sense_deferred = 0;
1207 
1208 	if (SCpnt->request->cmd_flags & REQ_DISCARD) {
1209 		if (!result)
1210 			scsi_set_resid(SCpnt, 0);
1211 		return good_bytes;
1212 	}
1213 
1214 	if (result) {
1215 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1216 		if (sense_valid)
1217 			sense_deferred = scsi_sense_is_deferred(&sshdr);
1218 	}
1219 #ifdef CONFIG_SCSI_LOGGING
1220 	SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1221 	if (sense_valid) {
1222 		SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1223 						   "sd_done: sb[respc,sk,asc,"
1224 						   "ascq]=%x,%x,%x,%x\n",
1225 						   sshdr.response_code,
1226 						   sshdr.sense_key, sshdr.asc,
1227 						   sshdr.ascq));
1228 	}
1229 #endif
1230 	if (driver_byte(result) != DRIVER_SENSE &&
1231 	    (!sense_valid || sense_deferred))
1232 		goto out;
1233 
1234 	switch (sshdr.sense_key) {
1235 	case HARDWARE_ERROR:
1236 	case MEDIUM_ERROR:
1237 		good_bytes = sd_completed_bytes(SCpnt);
1238 		break;
1239 	case RECOVERED_ERROR:
1240 		good_bytes = scsi_bufflen(SCpnt);
1241 		break;
1242 	case NO_SENSE:
1243 		/* This indicates a false check condition, so ignore it.  An
1244 		 * unknown amount of data was transferred so treat it as an
1245 		 * error.
1246 		 */
1247 		scsi_print_sense("sd", SCpnt);
1248 		SCpnt->result = 0;
1249 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1250 		break;
1251 	case ABORTED_COMMAND: /* DIF: Target detected corruption */
1252 	case ILLEGAL_REQUEST: /* DIX: Host detected corruption */
1253 		if (sshdr.asc == 0x10)
1254 			good_bytes = sd_completed_bytes(SCpnt);
1255 		break;
1256 	default:
1257 		break;
1258 	}
1259  out:
1260 	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1261 		sd_dif_complete(SCpnt, good_bytes);
1262 
1263 	if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1264 	    == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1265 
1266 		/* We have to print a failed command here as the
1267 		 * extended CDB gets freed before scsi_io_completion()
1268 		 * is called.
1269 		 */
1270 		if (result)
1271 			scsi_print_command(SCpnt);
1272 
1273 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
1274 		SCpnt->cmnd = NULL;
1275 		SCpnt->cmd_len = 0;
1276 	}
1277 
1278 	return good_bytes;
1279 }
1280 
1281 static int media_not_present(struct scsi_disk *sdkp,
1282 			     struct scsi_sense_hdr *sshdr)
1283 {
1284 
1285 	if (!scsi_sense_valid(sshdr))
1286 		return 0;
1287 	/* not invoked for commands that could return deferred errors */
1288 	if (sshdr->sense_key != NOT_READY &&
1289 	    sshdr->sense_key != UNIT_ATTENTION)
1290 		return 0;
1291 	if (sshdr->asc != 0x3A) /* medium not present */
1292 		return 0;
1293 
1294 	set_media_not_present(sdkp);
1295 	return 1;
1296 }
1297 
1298 /*
1299  * spinup disk - called only in sd_revalidate_disk()
1300  */
1301 static void
1302 sd_spinup_disk(struct scsi_disk *sdkp)
1303 {
1304 	unsigned char cmd[10];
1305 	unsigned long spintime_expire = 0;
1306 	int retries, spintime;
1307 	unsigned int the_result;
1308 	struct scsi_sense_hdr sshdr;
1309 	int sense_valid = 0;
1310 
1311 	spintime = 0;
1312 
1313 	/* Spin up drives, as required.  Only do this at boot time */
1314 	/* Spinup needs to be done for module loads too. */
1315 	do {
1316 		retries = 0;
1317 
1318 		do {
1319 			cmd[0] = TEST_UNIT_READY;
1320 			memset((void *) &cmd[1], 0, 9);
1321 
1322 			the_result = scsi_execute_req(sdkp->device, cmd,
1323 						      DMA_NONE, NULL, 0,
1324 						      &sshdr, SD_TIMEOUT,
1325 						      SD_MAX_RETRIES, NULL);
1326 
1327 			/*
1328 			 * If the drive has indicated to us that it
1329 			 * doesn't have any media in it, don't bother
1330 			 * with any more polling.
1331 			 */
1332 			if (media_not_present(sdkp, &sshdr))
1333 				return;
1334 
1335 			if (the_result)
1336 				sense_valid = scsi_sense_valid(&sshdr);
1337 			retries++;
1338 		} while (retries < 3 &&
1339 			 (!scsi_status_is_good(the_result) ||
1340 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1341 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1342 
1343 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1344 			/* no sense, TUR either succeeded or failed
1345 			 * with a status error */
1346 			if(!spintime && !scsi_status_is_good(the_result)) {
1347 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1348 				sd_print_result(sdkp, the_result);
1349 			}
1350 			break;
1351 		}
1352 
1353 		/*
1354 		 * The device does not want the automatic start to be issued.
1355 		 */
1356 		if (sdkp->device->no_start_on_add)
1357 			break;
1358 
1359 		if (sense_valid && sshdr.sense_key == NOT_READY) {
1360 			if (sshdr.asc == 4 && sshdr.ascq == 3)
1361 				break;	/* manual intervention required */
1362 			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1363 				break;	/* standby */
1364 			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1365 				break;	/* unavailable */
1366 			/*
1367 			 * Issue command to spin up drive when not ready
1368 			 */
1369 			if (!spintime) {
1370 				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1371 				cmd[0] = START_STOP;
1372 				cmd[1] = 1;	/* Return immediately */
1373 				memset((void *) &cmd[2], 0, 8);
1374 				cmd[4] = 1;	/* Start spin cycle */
1375 				if (sdkp->device->start_stop_pwr_cond)
1376 					cmd[4] |= 1 << 4;
1377 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1378 						 NULL, 0, &sshdr,
1379 						 SD_TIMEOUT, SD_MAX_RETRIES,
1380 						 NULL);
1381 				spintime_expire = jiffies + 100 * HZ;
1382 				spintime = 1;
1383 			}
1384 			/* Wait 1 second for next try */
1385 			msleep(1000);
1386 			printk(".");
1387 
1388 		/*
1389 		 * Wait for USB flash devices with slow firmware.
1390 		 * Yes, this sense key/ASC combination shouldn't
1391 		 * occur here.  It's characteristic of these devices.
1392 		 */
1393 		} else if (sense_valid &&
1394 				sshdr.sense_key == UNIT_ATTENTION &&
1395 				sshdr.asc == 0x28) {
1396 			if (!spintime) {
1397 				spintime_expire = jiffies + 5 * HZ;
1398 				spintime = 1;
1399 			}
1400 			/* Wait 1 second for next try */
1401 			msleep(1000);
1402 		} else {
1403 			/* we don't understand the sense code, so it's
1404 			 * probably pointless to loop */
1405 			if(!spintime) {
1406 				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1407 				sd_print_sense_hdr(sdkp, &sshdr);
1408 			}
1409 			break;
1410 		}
1411 
1412 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1413 
1414 	if (spintime) {
1415 		if (scsi_status_is_good(the_result))
1416 			printk("ready\n");
1417 		else
1418 			printk("not responding...\n");
1419 	}
1420 }
1421 
1422 
1423 /*
1424  * Determine whether disk supports Data Integrity Field.
1425  */
1426 static void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1427 {
1428 	struct scsi_device *sdp = sdkp->device;
1429 	u8 type;
1430 
1431 	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1432 		return;
1433 
1434 	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1435 
1436 	if (type == sdkp->protection_type || !sdkp->first_scan)
1437 		return;
1438 
1439 	sdkp->protection_type = type;
1440 
1441 	if (type > SD_DIF_TYPE3_PROTECTION) {
1442 		sd_printk(KERN_ERR, sdkp, "formatted with unsupported "	\
1443 			  "protection type %u. Disabling disk!\n", type);
1444 		sdkp->capacity = 0;
1445 		return;
1446 	}
1447 
1448 	if (scsi_host_dif_capable(sdp->host, type))
1449 		sd_printk(KERN_NOTICE, sdkp,
1450 			  "Enabling DIF Type %u protection\n", type);
1451 	else
1452 		sd_printk(KERN_NOTICE, sdkp,
1453 			  "Disabling DIF Type %u protection\n", type);
1454 }
1455 
1456 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1457 			struct scsi_sense_hdr *sshdr, int sense_valid,
1458 			int the_result)
1459 {
1460 	sd_print_result(sdkp, the_result);
1461 	if (driver_byte(the_result) & DRIVER_SENSE)
1462 		sd_print_sense_hdr(sdkp, sshdr);
1463 	else
1464 		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1465 
1466 	/*
1467 	 * Set dirty bit for removable devices if not ready -
1468 	 * sometimes drives will not report this properly.
1469 	 */
1470 	if (sdp->removable &&
1471 	    sense_valid && sshdr->sense_key == NOT_READY)
1472 		sdp->changed = 1;
1473 
1474 	/*
1475 	 * We used to set media_present to 0 here to indicate no media
1476 	 * in the drive, but some drives fail read capacity even with
1477 	 * media present, so we can't do that.
1478 	 */
1479 	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1480 }
1481 
1482 #define RC16_LEN 32
1483 #if RC16_LEN > SD_BUF_SIZE
1484 #error RC16_LEN must not be more than SD_BUF_SIZE
1485 #endif
1486 
1487 #define READ_CAPACITY_RETRIES_ON_RESET	10
1488 
1489 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1490 						unsigned char *buffer)
1491 {
1492 	unsigned char cmd[16];
1493 	struct scsi_sense_hdr sshdr;
1494 	int sense_valid = 0;
1495 	int the_result;
1496 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1497 	unsigned int alignment;
1498 	unsigned long long lba;
1499 	unsigned sector_size;
1500 
1501 	do {
1502 		memset(cmd, 0, 16);
1503 		cmd[0] = SERVICE_ACTION_IN;
1504 		cmd[1] = SAI_READ_CAPACITY_16;
1505 		cmd[13] = RC16_LEN;
1506 		memset(buffer, 0, RC16_LEN);
1507 
1508 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1509 					buffer, RC16_LEN, &sshdr,
1510 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1511 
1512 		if (media_not_present(sdkp, &sshdr))
1513 			return -ENODEV;
1514 
1515 		if (the_result) {
1516 			sense_valid = scsi_sense_valid(&sshdr);
1517 			if (sense_valid &&
1518 			    sshdr.sense_key == ILLEGAL_REQUEST &&
1519 			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1520 			    sshdr.ascq == 0x00)
1521 				/* Invalid Command Operation Code or
1522 				 * Invalid Field in CDB, just retry
1523 				 * silently with RC10 */
1524 				return -EINVAL;
1525 			if (sense_valid &&
1526 			    sshdr.sense_key == UNIT_ATTENTION &&
1527 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1528 				/* Device reset might occur several times,
1529 				 * give it one more chance */
1530 				if (--reset_retries > 0)
1531 					continue;
1532 		}
1533 		retries--;
1534 
1535 	} while (the_result && retries);
1536 
1537 	if (the_result) {
1538 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1539 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1540 		return -EINVAL;
1541 	}
1542 
1543 	sector_size = get_unaligned_be32(&buffer[8]);
1544 	lba = get_unaligned_be64(&buffer[0]);
1545 
1546 	sd_read_protection_type(sdkp, buffer);
1547 
1548 	if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1549 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1550 			"kernel compiled with support for large block "
1551 			"devices.\n");
1552 		sdkp->capacity = 0;
1553 		return -EOVERFLOW;
1554 	}
1555 
1556 	/* Logical blocks per physical block exponent */
1557 	sdkp->hw_sector_size = (1 << (buffer[13] & 0xf)) * sector_size;
1558 
1559 	/* Lowest aligned logical block */
1560 	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1561 	blk_queue_alignment_offset(sdp->request_queue, alignment);
1562 	if (alignment && sdkp->first_scan)
1563 		sd_printk(KERN_NOTICE, sdkp,
1564 			  "physical block alignment offset: %u\n", alignment);
1565 
1566 	if (buffer[14] & 0x80) { /* TPE */
1567 		struct request_queue *q = sdp->request_queue;
1568 
1569 		sdkp->thin_provisioning = 1;
1570 		q->limits.discard_granularity = sdkp->hw_sector_size;
1571 		q->limits.max_discard_sectors = 0xffffffff;
1572 
1573 		if (buffer[14] & 0x40) /* TPRZ */
1574 			q->limits.discard_zeroes_data = 1;
1575 
1576 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1577 	}
1578 
1579 	sdkp->capacity = lba + 1;
1580 	return sector_size;
1581 }
1582 
1583 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1584 						unsigned char *buffer)
1585 {
1586 	unsigned char cmd[16];
1587 	struct scsi_sense_hdr sshdr;
1588 	int sense_valid = 0;
1589 	int the_result;
1590 	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1591 	sector_t lba;
1592 	unsigned sector_size;
1593 
1594 	do {
1595 		cmd[0] = READ_CAPACITY;
1596 		memset(&cmd[1], 0, 9);
1597 		memset(buffer, 0, 8);
1598 
1599 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1600 					buffer, 8, &sshdr,
1601 					SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1602 
1603 		if (media_not_present(sdkp, &sshdr))
1604 			return -ENODEV;
1605 
1606 		if (the_result) {
1607 			sense_valid = scsi_sense_valid(&sshdr);
1608 			if (sense_valid &&
1609 			    sshdr.sense_key == UNIT_ATTENTION &&
1610 			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1611 				/* Device reset might occur several times,
1612 				 * give it one more chance */
1613 				if (--reset_retries > 0)
1614 					continue;
1615 		}
1616 		retries--;
1617 
1618 	} while (the_result && retries);
1619 
1620 	if (the_result) {
1621 		sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1622 		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1623 		return -EINVAL;
1624 	}
1625 
1626 	sector_size = get_unaligned_be32(&buffer[4]);
1627 	lba = get_unaligned_be32(&buffer[0]);
1628 
1629 	if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1630 		sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1631 			"kernel compiled with support for large block "
1632 			"devices.\n");
1633 		sdkp->capacity = 0;
1634 		return -EOVERFLOW;
1635 	}
1636 
1637 	sdkp->capacity = lba + 1;
1638 	sdkp->hw_sector_size = sector_size;
1639 	return sector_size;
1640 }
1641 
1642 static int sd_try_rc16_first(struct scsi_device *sdp)
1643 {
1644 	if (sdp->host->max_cmd_len < 16)
1645 		return 0;
1646 	if (sdp->scsi_level > SCSI_SPC_2)
1647 		return 1;
1648 	if (scsi_device_protection(sdp))
1649 		return 1;
1650 	return 0;
1651 }
1652 
1653 /*
1654  * read disk capacity
1655  */
1656 static void
1657 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1658 {
1659 	int sector_size;
1660 	struct scsi_device *sdp = sdkp->device;
1661 	sector_t old_capacity = sdkp->capacity;
1662 
1663 	if (sd_try_rc16_first(sdp)) {
1664 		sector_size = read_capacity_16(sdkp, sdp, buffer);
1665 		if (sector_size == -EOVERFLOW)
1666 			goto got_data;
1667 		if (sector_size == -ENODEV)
1668 			return;
1669 		if (sector_size < 0)
1670 			sector_size = read_capacity_10(sdkp, sdp, buffer);
1671 		if (sector_size < 0)
1672 			return;
1673 	} else {
1674 		sector_size = read_capacity_10(sdkp, sdp, buffer);
1675 		if (sector_size == -EOVERFLOW)
1676 			goto got_data;
1677 		if (sector_size < 0)
1678 			return;
1679 		if ((sizeof(sdkp->capacity) > 4) &&
1680 		    (sdkp->capacity > 0xffffffffULL)) {
1681 			int old_sector_size = sector_size;
1682 			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1683 					"Trying to use READ CAPACITY(16).\n");
1684 			sector_size = read_capacity_16(sdkp, sdp, buffer);
1685 			if (sector_size < 0) {
1686 				sd_printk(KERN_NOTICE, sdkp,
1687 					"Using 0xffffffff as device size\n");
1688 				sdkp->capacity = 1 + (sector_t) 0xffffffff;
1689 				sector_size = old_sector_size;
1690 				goto got_data;
1691 			}
1692 		}
1693 	}
1694 
1695 	/* Some devices are known to return the total number of blocks,
1696 	 * not the highest block number.  Some devices have versions
1697 	 * which do this and others which do not.  Some devices we might
1698 	 * suspect of doing this but we don't know for certain.
1699 	 *
1700 	 * If we know the reported capacity is wrong, decrement it.  If
1701 	 * we can only guess, then assume the number of blocks is even
1702 	 * (usually true but not always) and err on the side of lowering
1703 	 * the capacity.
1704 	 */
1705 	if (sdp->fix_capacity ||
1706 	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1707 		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1708 				"from its reported value: %llu\n",
1709 				(unsigned long long) sdkp->capacity);
1710 		--sdkp->capacity;
1711 	}
1712 
1713 got_data:
1714 	if (sector_size == 0) {
1715 		sector_size = 512;
1716 		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
1717 			  "assuming 512.\n");
1718 	}
1719 
1720 	if (sector_size != 512 &&
1721 	    sector_size != 1024 &&
1722 	    sector_size != 2048 &&
1723 	    sector_size != 4096 &&
1724 	    sector_size != 256) {
1725 		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
1726 			  sector_size);
1727 		/*
1728 		 * The user might want to re-format the drive with
1729 		 * a supported sectorsize.  Once this happens, it
1730 		 * would be relatively trivial to set the thing up.
1731 		 * For this reason, we leave the thing in the table.
1732 		 */
1733 		sdkp->capacity = 0;
1734 		/*
1735 		 * set a bogus sector size so the normal read/write
1736 		 * logic in the block layer will eventually refuse any
1737 		 * request on this device without tripping over power
1738 		 * of two sector size assumptions
1739 		 */
1740 		sector_size = 512;
1741 	}
1742 	blk_queue_logical_block_size(sdp->request_queue, sector_size);
1743 
1744 	{
1745 		char cap_str_2[10], cap_str_10[10];
1746 		u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
1747 
1748 		string_get_size(sz, STRING_UNITS_2, cap_str_2,
1749 				sizeof(cap_str_2));
1750 		string_get_size(sz, STRING_UNITS_10, cap_str_10,
1751 				sizeof(cap_str_10));
1752 
1753 		if (sdkp->first_scan || old_capacity != sdkp->capacity) {
1754 			sd_printk(KERN_NOTICE, sdkp,
1755 				  "%llu %d-byte logical blocks: (%s/%s)\n",
1756 				  (unsigned long long)sdkp->capacity,
1757 				  sector_size, cap_str_10, cap_str_2);
1758 
1759 			if (sdkp->hw_sector_size != sector_size)
1760 				sd_printk(KERN_NOTICE, sdkp,
1761 					  "%u-byte physical blocks\n",
1762 					  sdkp->hw_sector_size);
1763 		}
1764 	}
1765 
1766 	/* Rescale capacity to 512-byte units */
1767 	if (sector_size == 4096)
1768 		sdkp->capacity <<= 3;
1769 	else if (sector_size == 2048)
1770 		sdkp->capacity <<= 2;
1771 	else if (sector_size == 1024)
1772 		sdkp->capacity <<= 1;
1773 	else if (sector_size == 256)
1774 		sdkp->capacity >>= 1;
1775 
1776 	blk_queue_physical_block_size(sdp->request_queue, sdkp->hw_sector_size);
1777 	sdkp->device->sector_size = sector_size;
1778 }
1779 
1780 /* called with buffer of length 512 */
1781 static inline int
1782 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1783 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
1784 		 struct scsi_sense_hdr *sshdr)
1785 {
1786 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1787 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
1788 			       sshdr);
1789 }
1790 
1791 /*
1792  * read write protect setting, if possible - called only in sd_revalidate_disk()
1793  * called with buffer of length SD_BUF_SIZE
1794  */
1795 static void
1796 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
1797 {
1798 	int res;
1799 	struct scsi_device *sdp = sdkp->device;
1800 	struct scsi_mode_data data;
1801 	int old_wp = sdkp->write_prot;
1802 
1803 	set_disk_ro(sdkp->disk, 0);
1804 	if (sdp->skip_ms_page_3f) {
1805 		sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
1806 		return;
1807 	}
1808 
1809 	if (sdp->use_192_bytes_for_3f) {
1810 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1811 	} else {
1812 		/*
1813 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1814 		 * We have to start carefully: some devices hang if we ask
1815 		 * for more than is available.
1816 		 */
1817 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1818 
1819 		/*
1820 		 * Second attempt: ask for page 0 When only page 0 is
1821 		 * implemented, a request for page 3F may return Sense Key
1822 		 * 5: Illegal Request, Sense Code 24: Invalid field in
1823 		 * CDB.
1824 		 */
1825 		if (!scsi_status_is_good(res))
1826 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1827 
1828 		/*
1829 		 * Third attempt: ask 255 bytes, as we did earlier.
1830 		 */
1831 		if (!scsi_status_is_good(res))
1832 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1833 					       &data, NULL);
1834 	}
1835 
1836 	if (!scsi_status_is_good(res)) {
1837 		sd_printk(KERN_WARNING, sdkp,
1838 			  "Test WP failed, assume Write Enabled\n");
1839 	} else {
1840 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
1841 		set_disk_ro(sdkp->disk, sdkp->write_prot);
1842 		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
1843 			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
1844 				  sdkp->write_prot ? "on" : "off");
1845 			sd_printk(KERN_DEBUG, sdkp,
1846 				  "Mode Sense: %02x %02x %02x %02x\n",
1847 				  buffer[0], buffer[1], buffer[2], buffer[3]);
1848 		}
1849 	}
1850 }
1851 
1852 /*
1853  * sd_read_cache_type - called only from sd_revalidate_disk()
1854  * called with buffer of length SD_BUF_SIZE
1855  */
1856 static void
1857 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
1858 {
1859 	int len = 0, res;
1860 	struct scsi_device *sdp = sdkp->device;
1861 
1862 	int dbd;
1863 	int modepage;
1864 	struct scsi_mode_data data;
1865 	struct scsi_sense_hdr sshdr;
1866 	int old_wce = sdkp->WCE;
1867 	int old_rcd = sdkp->RCD;
1868 	int old_dpofua = sdkp->DPOFUA;
1869 
1870 	if (sdp->skip_ms_page_8)
1871 		goto defaults;
1872 
1873 	if (sdp->type == TYPE_RBC) {
1874 		modepage = 6;
1875 		dbd = 8;
1876 	} else {
1877 		modepage = 8;
1878 		dbd = 0;
1879 	}
1880 
1881 	/* cautiously ask */
1882 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, 4, &data, &sshdr);
1883 
1884 	if (!scsi_status_is_good(res))
1885 		goto bad_sense;
1886 
1887 	if (!data.header_length) {
1888 		modepage = 6;
1889 		sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
1890 	}
1891 
1892 	/* that went OK, now ask for the proper length */
1893 	len = data.length;
1894 
1895 	/*
1896 	 * We're only interested in the first three bytes, actually.
1897 	 * But the data cache page is defined for the first 20.
1898 	 */
1899 	if (len < 3)
1900 		goto bad_sense;
1901 	if (len > 20)
1902 		len = 20;
1903 
1904 	/* Take headers and block descriptors into account */
1905 	len += data.header_length + data.block_descriptor_length;
1906 	if (len > SD_BUF_SIZE)
1907 		goto bad_sense;
1908 
1909 	/* Get the data */
1910 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
1911 
1912 	if (scsi_status_is_good(res)) {
1913 		int offset = data.header_length + data.block_descriptor_length;
1914 
1915 		if (offset >= SD_BUF_SIZE - 2) {
1916 			sd_printk(KERN_ERR, sdkp, "Malformed MODE SENSE response\n");
1917 			goto defaults;
1918 		}
1919 
1920 		if ((buffer[offset] & 0x3f) != modepage) {
1921 			sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
1922 			goto defaults;
1923 		}
1924 
1925 		if (modepage == 8) {
1926 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
1927 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
1928 		} else {
1929 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
1930 			sdkp->RCD = 0;
1931 		}
1932 
1933 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
1934 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
1935 			sd_printk(KERN_NOTICE, sdkp,
1936 				  "Uses READ/WRITE(6), disabling FUA\n");
1937 			sdkp->DPOFUA = 0;
1938 		}
1939 
1940 		if (sdkp->first_scan || old_wce != sdkp->WCE ||
1941 		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
1942 			sd_printk(KERN_NOTICE, sdkp,
1943 				  "Write cache: %s, read cache: %s, %s\n",
1944 				  sdkp->WCE ? "enabled" : "disabled",
1945 				  sdkp->RCD ? "disabled" : "enabled",
1946 				  sdkp->DPOFUA ? "supports DPO and FUA"
1947 				  : "doesn't support DPO or FUA");
1948 
1949 		return;
1950 	}
1951 
1952 bad_sense:
1953 	if (scsi_sense_valid(&sshdr) &&
1954 	    sshdr.sense_key == ILLEGAL_REQUEST &&
1955 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
1956 		/* Invalid field in CDB */
1957 		sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
1958 	else
1959 		sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
1960 
1961 defaults:
1962 	sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
1963 	sdkp->WCE = 0;
1964 	sdkp->RCD = 0;
1965 	sdkp->DPOFUA = 0;
1966 }
1967 
1968 /*
1969  * The ATO bit indicates whether the DIF application tag is available
1970  * for use by the operating system.
1971  */
1972 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
1973 {
1974 	int res, offset;
1975 	struct scsi_device *sdp = sdkp->device;
1976 	struct scsi_mode_data data;
1977 	struct scsi_sense_hdr sshdr;
1978 
1979 	if (sdp->type != TYPE_DISK)
1980 		return;
1981 
1982 	if (sdkp->protection_type == 0)
1983 		return;
1984 
1985 	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
1986 			      SD_MAX_RETRIES, &data, &sshdr);
1987 
1988 	if (!scsi_status_is_good(res) || !data.header_length ||
1989 	    data.length < 6) {
1990 		sd_printk(KERN_WARNING, sdkp,
1991 			  "getting Control mode page failed, assume no ATO\n");
1992 
1993 		if (scsi_sense_valid(&sshdr))
1994 			sd_print_sense_hdr(sdkp, &sshdr);
1995 
1996 		return;
1997 	}
1998 
1999 	offset = data.header_length + data.block_descriptor_length;
2000 
2001 	if ((buffer[offset] & 0x3f) != 0x0a) {
2002 		sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2003 		return;
2004 	}
2005 
2006 	if ((buffer[offset + 5] & 0x80) == 0)
2007 		return;
2008 
2009 	sdkp->ATO = 1;
2010 
2011 	return;
2012 }
2013 
2014 /**
2015  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2016  * @disk: disk to query
2017  */
2018 static void sd_read_block_limits(struct scsi_disk *sdkp)
2019 {
2020 	struct request_queue *q = sdkp->disk->queue;
2021 	unsigned int sector_sz = sdkp->device->sector_size;
2022 	const int vpd_len = 64;
2023 	unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2024 
2025 	if (!buffer ||
2026 	    /* Block Limits VPD */
2027 	    scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2028 		goto out;
2029 
2030 	blk_queue_io_min(sdkp->disk->queue,
2031 			 get_unaligned_be16(&buffer[6]) * sector_sz);
2032 	blk_queue_io_opt(sdkp->disk->queue,
2033 			 get_unaligned_be32(&buffer[12]) * sector_sz);
2034 
2035 	/* Thin provisioning enabled and page length indicates TP support */
2036 	if (sdkp->thin_provisioning && buffer[3] == 0x3c) {
2037 		unsigned int lba_count, desc_count, granularity;
2038 
2039 		lba_count = get_unaligned_be32(&buffer[20]);
2040 		desc_count = get_unaligned_be32(&buffer[24]);
2041 
2042 		if (lba_count) {
2043 			q->limits.max_discard_sectors =
2044 				lba_count * sector_sz >> 9;
2045 
2046 			if (desc_count)
2047 				sdkp->unmap = 1;
2048 		}
2049 
2050 		granularity = get_unaligned_be32(&buffer[28]);
2051 
2052 		if (granularity)
2053 			q->limits.discard_granularity = granularity * sector_sz;
2054 
2055 		if (buffer[32] & 0x80)
2056 			q->limits.discard_alignment =
2057 				get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2058 	}
2059 
2060  out:
2061 	kfree(buffer);
2062 }
2063 
2064 /**
2065  * sd_read_block_characteristics - Query block dev. characteristics
2066  * @disk: disk to query
2067  */
2068 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2069 {
2070 	unsigned char *buffer;
2071 	u16 rot;
2072 	const int vpd_len = 64;
2073 
2074 	buffer = kmalloc(vpd_len, GFP_KERNEL);
2075 
2076 	if (!buffer ||
2077 	    /* Block Device Characteristics VPD */
2078 	    scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2079 		goto out;
2080 
2081 	rot = get_unaligned_be16(&buffer[4]);
2082 
2083 	if (rot == 1)
2084 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2085 
2086  out:
2087 	kfree(buffer);
2088 }
2089 
2090 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2091 {
2092 	/*
2093 	 * Although VPD inquiries can go to SCSI-2 type devices,
2094 	 * some USB ones crash on receiving them, and the pages
2095 	 * we currently ask for are for SPC-3 and beyond
2096 	 */
2097 	if (sdp->scsi_level > SCSI_SPC_2)
2098 		return 1;
2099 	return 0;
2100 }
2101 
2102 /**
2103  *	sd_revalidate_disk - called the first time a new disk is seen,
2104  *	performs disk spin up, read_capacity, etc.
2105  *	@disk: struct gendisk we care about
2106  **/
2107 static int sd_revalidate_disk(struct gendisk *disk)
2108 {
2109 	struct scsi_disk *sdkp = scsi_disk(disk);
2110 	struct scsi_device *sdp = sdkp->device;
2111 	unsigned char *buffer;
2112 	unsigned ordered;
2113 
2114 	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2115 				      "sd_revalidate_disk\n"));
2116 
2117 	/*
2118 	 * If the device is offline, don't try and read capacity or any
2119 	 * of the other niceties.
2120 	 */
2121 	if (!scsi_device_online(sdp))
2122 		goto out;
2123 
2124 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2125 	if (!buffer) {
2126 		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2127 			  "allocation failure.\n");
2128 		goto out;
2129 	}
2130 
2131 	sd_spinup_disk(sdkp);
2132 
2133 	/*
2134 	 * Without media there is no reason to ask; moreover, some devices
2135 	 * react badly if we do.
2136 	 */
2137 	if (sdkp->media_present) {
2138 		sd_read_capacity(sdkp, buffer);
2139 
2140 		if (sd_try_extended_inquiry(sdp)) {
2141 			sd_read_block_limits(sdkp);
2142 			sd_read_block_characteristics(sdkp);
2143 		}
2144 
2145 		sd_read_write_protect_flag(sdkp, buffer);
2146 		sd_read_cache_type(sdkp, buffer);
2147 		sd_read_app_tag_own(sdkp, buffer);
2148 	}
2149 
2150 	sdkp->first_scan = 0;
2151 
2152 	/*
2153 	 * We now have all cache related info, determine how we deal
2154 	 * with ordered requests.  Note that as the current SCSI
2155 	 * dispatch function can alter request order, we cannot use
2156 	 * QUEUE_ORDERED_TAG_* even when ordered tag is supported.
2157 	 */
2158 	if (sdkp->WCE)
2159 		ordered = sdkp->DPOFUA
2160 			? QUEUE_ORDERED_DRAIN_FUA : QUEUE_ORDERED_DRAIN_FLUSH;
2161 	else
2162 		ordered = QUEUE_ORDERED_DRAIN;
2163 
2164 	blk_queue_ordered(sdkp->disk->queue, ordered);
2165 
2166 	set_capacity(disk, sdkp->capacity);
2167 	kfree(buffer);
2168 
2169  out:
2170 	return 0;
2171 }
2172 
2173 /**
2174  *	sd_unlock_native_capacity - unlock native capacity
2175  *	@disk: struct gendisk to set capacity for
2176  *
2177  *	Block layer calls this function if it detects that partitions
2178  *	on @disk reach beyond the end of the device.  If the SCSI host
2179  *	implements ->unlock_native_capacity() method, it's invoked to
2180  *	give it a chance to adjust the device capacity.
2181  *
2182  *	CONTEXT:
2183  *	Defined by block layer.  Might sleep.
2184  */
2185 static void sd_unlock_native_capacity(struct gendisk *disk)
2186 {
2187 	struct scsi_device *sdev = scsi_disk(disk)->device;
2188 
2189 	if (sdev->host->hostt->unlock_native_capacity)
2190 		sdev->host->hostt->unlock_native_capacity(sdev);
2191 }
2192 
2193 /**
2194  *	sd_format_disk_name - format disk name
2195  *	@prefix: name prefix - ie. "sd" for SCSI disks
2196  *	@index: index of the disk to format name for
2197  *	@buf: output buffer
2198  *	@buflen: length of the output buffer
2199  *
2200  *	SCSI disk names starts at sda.  The 26th device is sdz and the
2201  *	27th is sdaa.  The last one for two lettered suffix is sdzz
2202  *	which is followed by sdaaa.
2203  *
2204  *	This is basically 26 base counting with one extra 'nil' entry
2205  *	at the beginning from the second digit on and can be
2206  *	determined using similar method as 26 base conversion with the
2207  *	index shifted -1 after each digit is computed.
2208  *
2209  *	CONTEXT:
2210  *	Don't care.
2211  *
2212  *	RETURNS:
2213  *	0 on success, -errno on failure.
2214  */
2215 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2216 {
2217 	const int base = 'z' - 'a' + 1;
2218 	char *begin = buf + strlen(prefix);
2219 	char *end = buf + buflen;
2220 	char *p;
2221 	int unit;
2222 
2223 	p = end - 1;
2224 	*p = '\0';
2225 	unit = base;
2226 	do {
2227 		if (p == begin)
2228 			return -EINVAL;
2229 		*--p = 'a' + (index % unit);
2230 		index = (index / unit) - 1;
2231 	} while (index >= 0);
2232 
2233 	memmove(begin, p, end - p);
2234 	memcpy(buf, prefix, strlen(prefix));
2235 
2236 	return 0;
2237 }
2238 
2239 /*
2240  * The asynchronous part of sd_probe
2241  */
2242 static void sd_probe_async(void *data, async_cookie_t cookie)
2243 {
2244 	struct scsi_disk *sdkp = data;
2245 	struct scsi_device *sdp;
2246 	struct gendisk *gd;
2247 	u32 index;
2248 	struct device *dev;
2249 
2250 	sdp = sdkp->device;
2251 	gd = sdkp->disk;
2252 	index = sdkp->index;
2253 	dev = &sdp->sdev_gendev;
2254 
2255 	if (index < SD_MAX_DISKS) {
2256 		gd->major = sd_major((index & 0xf0) >> 4);
2257 		gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2258 		gd->minors = SD_MINORS;
2259 	}
2260 	gd->fops = &sd_fops;
2261 	gd->private_data = &sdkp->driver;
2262 	gd->queue = sdkp->device->request_queue;
2263 
2264 	/* defaults, until the device tells us otherwise */
2265 	sdp->sector_size = 512;
2266 	sdkp->capacity = 0;
2267 	sdkp->media_present = 1;
2268 	sdkp->write_prot = 0;
2269 	sdkp->WCE = 0;
2270 	sdkp->RCD = 0;
2271 	sdkp->ATO = 0;
2272 	sdkp->first_scan = 1;
2273 
2274 	sd_revalidate_disk(gd);
2275 
2276 	blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2277 	blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2278 
2279 	gd->driverfs_dev = &sdp->sdev_gendev;
2280 	gd->flags = GENHD_FL_EXT_DEVT;
2281 	if (sdp->removable)
2282 		gd->flags |= GENHD_FL_REMOVABLE;
2283 
2284 	add_disk(gd);
2285 	sd_dif_config_host(sdkp);
2286 
2287 	sd_revalidate_disk(gd);
2288 
2289 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2290 		  sdp->removable ? "removable " : "");
2291 	scsi_autopm_put_device(sdp);
2292 	put_device(&sdkp->dev);
2293 }
2294 
2295 /**
2296  *	sd_probe - called during driver initialization and whenever a
2297  *	new scsi device is attached to the system. It is called once
2298  *	for each scsi device (not just disks) present.
2299  *	@dev: pointer to device object
2300  *
2301  *	Returns 0 if successful (or not interested in this scsi device
2302  *	(e.g. scanner)); 1 when there is an error.
2303  *
2304  *	Note: this function is invoked from the scsi mid-level.
2305  *	This function sets up the mapping between a given
2306  *	<host,channel,id,lun> (found in sdp) and new device name
2307  *	(e.g. /dev/sda). More precisely it is the block device major
2308  *	and minor number that is chosen here.
2309  *
2310  *	Assume sd_attach is not re-entrant (for time being)
2311  *	Also think about sd_attach() and sd_remove() running coincidentally.
2312  **/
2313 static int sd_probe(struct device *dev)
2314 {
2315 	struct scsi_device *sdp = to_scsi_device(dev);
2316 	struct scsi_disk *sdkp;
2317 	struct gendisk *gd;
2318 	int index;
2319 	int error;
2320 
2321 	error = -ENODEV;
2322 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2323 		goto out;
2324 
2325 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2326 					"sd_attach\n"));
2327 
2328 	error = -ENOMEM;
2329 	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2330 	if (!sdkp)
2331 		goto out;
2332 
2333 	gd = alloc_disk(SD_MINORS);
2334 	if (!gd)
2335 		goto out_free;
2336 
2337 	do {
2338 		if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2339 			goto out_put;
2340 
2341 		spin_lock(&sd_index_lock);
2342 		error = ida_get_new(&sd_index_ida, &index);
2343 		spin_unlock(&sd_index_lock);
2344 	} while (error == -EAGAIN);
2345 
2346 	if (error)
2347 		goto out_put;
2348 
2349 	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2350 	if (error)
2351 		goto out_free_index;
2352 
2353 	sdkp->device = sdp;
2354 	sdkp->driver = &sd_template;
2355 	sdkp->disk = gd;
2356 	sdkp->index = index;
2357 	atomic_set(&sdkp->openers, 0);
2358 	sdkp->previous_state = 1;
2359 
2360 	if (!sdp->request_queue->rq_timeout) {
2361 		if (sdp->type != TYPE_MOD)
2362 			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2363 		else
2364 			blk_queue_rq_timeout(sdp->request_queue,
2365 					     SD_MOD_TIMEOUT);
2366 	}
2367 
2368 	device_initialize(&sdkp->dev);
2369 	sdkp->dev.parent = dev;
2370 	sdkp->dev.class = &sd_disk_class;
2371 	dev_set_name(&sdkp->dev, dev_name(dev));
2372 
2373 	if (device_add(&sdkp->dev))
2374 		goto out_free_index;
2375 
2376 	get_device(dev);
2377 	dev_set_drvdata(dev, sdkp);
2378 
2379 	get_device(&sdkp->dev);	/* prevent release before async_schedule */
2380 	async_schedule(sd_probe_async, sdkp);
2381 
2382 	return 0;
2383 
2384  out_free_index:
2385 	spin_lock(&sd_index_lock);
2386 	ida_remove(&sd_index_ida, index);
2387 	spin_unlock(&sd_index_lock);
2388  out_put:
2389 	put_disk(gd);
2390  out_free:
2391 	kfree(sdkp);
2392  out:
2393 	return error;
2394 }
2395 
2396 /**
2397  *	sd_remove - called whenever a scsi disk (previously recognized by
2398  *	sd_probe) is detached from the system. It is called (potentially
2399  *	multiple times) during sd module unload.
2400  *	@sdp: pointer to mid level scsi device object
2401  *
2402  *	Note: this function is invoked from the scsi mid-level.
2403  *	This function potentially frees up a device name (e.g. /dev/sdc)
2404  *	that could be re-used by a subsequent sd_probe().
2405  *	This function is not called when the built-in sd driver is "exit-ed".
2406  **/
2407 static int sd_remove(struct device *dev)
2408 {
2409 	struct scsi_disk *sdkp;
2410 
2411 	sdkp = dev_get_drvdata(dev);
2412 	scsi_autopm_get_device(sdkp->device);
2413 
2414 	async_synchronize_full();
2415 	blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2416 	blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2417 	device_del(&sdkp->dev);
2418 	del_gendisk(sdkp->disk);
2419 	sd_shutdown(dev);
2420 
2421 	mutex_lock(&sd_ref_mutex);
2422 	dev_set_drvdata(dev, NULL);
2423 	put_device(&sdkp->dev);
2424 	mutex_unlock(&sd_ref_mutex);
2425 
2426 	return 0;
2427 }
2428 
2429 /**
2430  *	scsi_disk_release - Called to free the scsi_disk structure
2431  *	@dev: pointer to embedded class device
2432  *
2433  *	sd_ref_mutex must be held entering this routine.  Because it is
2434  *	called on last put, you should always use the scsi_disk_get()
2435  *	scsi_disk_put() helpers which manipulate the semaphore directly
2436  *	and never do a direct put_device.
2437  **/
2438 static void scsi_disk_release(struct device *dev)
2439 {
2440 	struct scsi_disk *sdkp = to_scsi_disk(dev);
2441 	struct gendisk *disk = sdkp->disk;
2442 
2443 	spin_lock(&sd_index_lock);
2444 	ida_remove(&sd_index_ida, sdkp->index);
2445 	spin_unlock(&sd_index_lock);
2446 
2447 	disk->private_data = NULL;
2448 	put_disk(disk);
2449 	put_device(&sdkp->device->sdev_gendev);
2450 
2451 	kfree(sdkp);
2452 }
2453 
2454 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2455 {
2456 	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
2457 	struct scsi_sense_hdr sshdr;
2458 	struct scsi_device *sdp = sdkp->device;
2459 	int res;
2460 
2461 	if (start)
2462 		cmd[4] |= 1;	/* START */
2463 
2464 	if (sdp->start_stop_pwr_cond)
2465 		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
2466 
2467 	if (!scsi_device_online(sdp))
2468 		return -ENODEV;
2469 
2470 	res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2471 			       SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2472 	if (res) {
2473 		sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2474 		sd_print_result(sdkp, res);
2475 		if (driver_byte(res) & DRIVER_SENSE)
2476 			sd_print_sense_hdr(sdkp, &sshdr);
2477 	}
2478 
2479 	return res;
2480 }
2481 
2482 /*
2483  * Send a SYNCHRONIZE CACHE instruction down to the device through
2484  * the normal SCSI command structure.  Wait for the command to
2485  * complete.
2486  */
2487 static void sd_shutdown(struct device *dev)
2488 {
2489 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2490 
2491 	if (!sdkp)
2492 		return;         /* this can happen */
2493 
2494 	if (sdkp->WCE) {
2495 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2496 		sd_sync_cache(sdkp);
2497 	}
2498 
2499 	if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2500 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2501 		sd_start_stop_device(sdkp, 0);
2502 	}
2503 
2504 	scsi_disk_put(sdkp);
2505 }
2506 
2507 static int sd_suspend(struct device *dev, pm_message_t mesg)
2508 {
2509 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2510 	int ret = 0;
2511 
2512 	if (!sdkp)
2513 		return 0;	/* this can happen */
2514 
2515 	if (sdkp->WCE) {
2516 		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2517 		ret = sd_sync_cache(sdkp);
2518 		if (ret)
2519 			goto done;
2520 	}
2521 
2522 	if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2523 		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2524 		ret = sd_start_stop_device(sdkp, 0);
2525 	}
2526 
2527 done:
2528 	scsi_disk_put(sdkp);
2529 	return ret;
2530 }
2531 
2532 static int sd_resume(struct device *dev)
2533 {
2534 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2535 	int ret = 0;
2536 
2537 	if (!sdkp->device->manage_start_stop)
2538 		goto done;
2539 
2540 	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2541 	ret = sd_start_stop_device(sdkp, 1);
2542 
2543 done:
2544 	scsi_disk_put(sdkp);
2545 	return ret;
2546 }
2547 
2548 /**
2549  *	init_sd - entry point for this driver (both when built in or when
2550  *	a module).
2551  *
2552  *	Note: this function registers this driver with the scsi mid-level.
2553  **/
2554 static int __init init_sd(void)
2555 {
2556 	int majors = 0, i, err;
2557 
2558 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2559 
2560 	for (i = 0; i < SD_MAJORS; i++)
2561 		if (register_blkdev(sd_major(i), "sd") == 0)
2562 			majors++;
2563 
2564 	if (!majors)
2565 		return -ENODEV;
2566 
2567 	err = class_register(&sd_disk_class);
2568 	if (err)
2569 		goto err_out;
2570 
2571 	err = scsi_register_driver(&sd_template.gendrv);
2572 	if (err)
2573 		goto err_out_class;
2574 
2575 	sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2576 					 0, 0, NULL);
2577 	if (!sd_cdb_cache) {
2578 		printk(KERN_ERR "sd: can't init extended cdb cache\n");
2579 		goto err_out_class;
2580 	}
2581 
2582 	sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2583 	if (!sd_cdb_pool) {
2584 		printk(KERN_ERR "sd: can't init extended cdb pool\n");
2585 		goto err_out_cache;
2586 	}
2587 
2588 	return 0;
2589 
2590 err_out_cache:
2591 	kmem_cache_destroy(sd_cdb_cache);
2592 
2593 err_out_class:
2594 	class_unregister(&sd_disk_class);
2595 err_out:
2596 	for (i = 0; i < SD_MAJORS; i++)
2597 		unregister_blkdev(sd_major(i), "sd");
2598 	return err;
2599 }
2600 
2601 /**
2602  *	exit_sd - exit point for this driver (when it is a module).
2603  *
2604  *	Note: this function unregisters this driver from the scsi mid-level.
2605  **/
2606 static void __exit exit_sd(void)
2607 {
2608 	int i;
2609 
2610 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
2611 
2612 	mempool_destroy(sd_cdb_pool);
2613 	kmem_cache_destroy(sd_cdb_cache);
2614 
2615 	scsi_unregister_driver(&sd_template.gendrv);
2616 	class_unregister(&sd_disk_class);
2617 
2618 	for (i = 0; i < SD_MAJORS; i++)
2619 		unregister_blkdev(sd_major(i), "sd");
2620 }
2621 
2622 module_init(init_sd);
2623 module_exit(exit_sd);
2624 
2625 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
2626 			       struct scsi_sense_hdr *sshdr)
2627 {
2628 	sd_printk(KERN_INFO, sdkp, "");
2629 	scsi_show_sense_hdr(sshdr);
2630 	sd_printk(KERN_INFO, sdkp, "");
2631 	scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
2632 }
2633 
2634 static void sd_print_result(struct scsi_disk *sdkp, int result)
2635 {
2636 	sd_printk(KERN_INFO, sdkp, "");
2637 	scsi_show_result(result);
2638 }
2639 
2640