xref: /openbmc/linux/drivers/scsi/sd.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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/config.h>
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/sched.h>
40 #include <linux/mm.h>
41 #include <linux/bio.h>
42 #include <linux/genhd.h>
43 #include <linux/hdreg.h>
44 #include <linux/errno.h>
45 #include <linux/idr.h>
46 #include <linux/interrupt.h>
47 #include <linux/init.h>
48 #include <linux/blkdev.h>
49 #include <linux/blkpg.h>
50 #include <linux/kref.h>
51 #include <linux/delay.h>
52 #include <asm/uaccess.h>
53 
54 #include <scsi/scsi.h>
55 #include <scsi/scsi_cmnd.h>
56 #include <scsi/scsi_dbg.h>
57 #include <scsi/scsi_device.h>
58 #include <scsi/scsi_driver.h>
59 #include <scsi/scsi_eh.h>
60 #include <scsi/scsi_host.h>
61 #include <scsi/scsi_ioctl.h>
62 #include <scsi/scsicam.h>
63 
64 #include "scsi_logging.h"
65 
66 /*
67  * More than enough for everybody ;)  The huge number of majors
68  * is a leftover from 16bit dev_t days, we don't really need that
69  * much numberspace.
70  */
71 #define SD_MAJORS	16
72 
73 /*
74  * This is limited by the naming scheme enforced in sd_probe,
75  * add another character to it if you really need more disks.
76  */
77 #define SD_MAX_DISKS	(((26 * 26) + 26 + 1) * 26)
78 
79 /*
80  * Time out in seconds for disks and Magneto-opticals (which are slower).
81  */
82 #define SD_TIMEOUT		(30 * HZ)
83 #define SD_MOD_TIMEOUT		(75 * HZ)
84 
85 /*
86  * Number of allowed retries
87  */
88 #define SD_MAX_RETRIES		5
89 #define SD_PASSTHROUGH_RETRIES	1
90 
91 static void scsi_disk_release(struct kref *kref);
92 
93 struct scsi_disk {
94 	struct scsi_driver *driver;	/* always &sd_template */
95 	struct scsi_device *device;
96 	struct kref	kref;
97 	struct gendisk	*disk;
98 	unsigned int	openers;	/* protected by BKL for now, yuck */
99 	sector_t	capacity;	/* size in 512-byte sectors */
100 	u32		index;
101 	u8		media_present;
102 	u8		write_prot;
103 	unsigned	WCE : 1;	/* state of disk WCE bit */
104 	unsigned	RCD : 1;	/* state of disk RCD bit, unused */
105 };
106 
107 static DEFINE_IDR(sd_index_idr);
108 static DEFINE_SPINLOCK(sd_index_lock);
109 
110 /* This semaphore is used to mediate the 0->1 reference get in the
111  * face of object destruction (i.e. we can't allow a get on an
112  * object after last put) */
113 static DECLARE_MUTEX(sd_ref_sem);
114 
115 static int sd_revalidate_disk(struct gendisk *disk);
116 static void sd_rw_intr(struct scsi_cmnd * SCpnt);
117 
118 static int sd_probe(struct device *);
119 static int sd_remove(struct device *);
120 static void sd_shutdown(struct device *dev);
121 static void sd_rescan(struct device *);
122 static int sd_init_command(struct scsi_cmnd *);
123 static int sd_issue_flush(struct device *, sector_t *);
124 static void sd_end_flush(request_queue_t *, struct request *);
125 static int sd_prepare_flush(request_queue_t *, struct request *);
126 static void sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
127 			     unsigned char *buffer);
128 
129 static struct scsi_driver sd_template = {
130 	.owner			= THIS_MODULE,
131 	.gendrv = {
132 		.name		= "sd",
133 		.probe		= sd_probe,
134 		.remove		= sd_remove,
135 		.shutdown	= sd_shutdown,
136 	},
137 	.rescan			= sd_rescan,
138 	.init_command		= sd_init_command,
139 	.issue_flush		= sd_issue_flush,
140 	.prepare_flush		= sd_prepare_flush,
141 	.end_flush		= sd_end_flush,
142 };
143 
144 /*
145  * Device no to disk mapping:
146  *
147  *       major         disc2     disc  p1
148  *   |............|.............|....|....| <- dev_t
149  *    31        20 19          8 7  4 3  0
150  *
151  * Inside a major, we have 16k disks, however mapped non-
152  * contiguously. The first 16 disks are for major0, the next
153  * ones with major1, ... Disk 256 is for major0 again, disk 272
154  * for major1, ...
155  * As we stay compatible with our numbering scheme, we can reuse
156  * the well-know SCSI majors 8, 65--71, 136--143.
157  */
158 static int sd_major(int major_idx)
159 {
160 	switch (major_idx) {
161 	case 0:
162 		return SCSI_DISK0_MAJOR;
163 	case 1 ... 7:
164 		return SCSI_DISK1_MAJOR + major_idx - 1;
165 	case 8 ... 15:
166 		return SCSI_DISK8_MAJOR + major_idx - 8;
167 	default:
168 		BUG();
169 		return 0;	/* shut up gcc */
170 	}
171 }
172 
173 #define to_scsi_disk(obj) container_of(obj,struct scsi_disk,kref)
174 
175 static inline struct scsi_disk *scsi_disk(struct gendisk *disk)
176 {
177 	return container_of(disk->private_data, struct scsi_disk, driver);
178 }
179 
180 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
181 {
182 	struct scsi_disk *sdkp = NULL;
183 
184 	if (disk->private_data) {
185 		sdkp = scsi_disk(disk);
186 		if (scsi_device_get(sdkp->device) == 0)
187 			kref_get(&sdkp->kref);
188 		else
189 			sdkp = NULL;
190 	}
191 	return sdkp;
192 }
193 
194 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
195 {
196 	struct scsi_disk *sdkp;
197 
198 	down(&sd_ref_sem);
199 	sdkp = __scsi_disk_get(disk);
200 	up(&sd_ref_sem);
201 	return sdkp;
202 }
203 
204 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
205 {
206 	struct scsi_disk *sdkp;
207 
208 	down(&sd_ref_sem);
209 	sdkp = dev_get_drvdata(dev);
210 	if (sdkp)
211 		sdkp = __scsi_disk_get(sdkp->disk);
212 	up(&sd_ref_sem);
213 	return sdkp;
214 }
215 
216 static void scsi_disk_put(struct scsi_disk *sdkp)
217 {
218 	struct scsi_device *sdev = sdkp->device;
219 
220 	down(&sd_ref_sem);
221 	kref_put(&sdkp->kref, scsi_disk_release);
222 	scsi_device_put(sdev);
223 	up(&sd_ref_sem);
224 }
225 
226 /**
227  *	sd_init_command - build a scsi (read or write) command from
228  *	information in the request structure.
229  *	@SCpnt: pointer to mid-level's per scsi command structure that
230  *	contains request and into which the scsi command is written
231  *
232  *	Returns 1 if successful and 0 if error (or cannot be done now).
233  **/
234 static int sd_init_command(struct scsi_cmnd * SCpnt)
235 {
236 	unsigned int this_count, timeout;
237 	struct gendisk *disk;
238 	sector_t block;
239 	struct scsi_device *sdp = SCpnt->device;
240 	struct request *rq = SCpnt->request;
241 
242 	timeout = sdp->timeout;
243 
244 	/*
245 	 * SG_IO from block layer already setup, just copy cdb basically
246 	 */
247 	if (blk_pc_request(rq)) {
248 		if (sizeof(rq->cmd) > sizeof(SCpnt->cmnd))
249 			return 0;
250 
251 		memcpy(SCpnt->cmnd, rq->cmd, sizeof(SCpnt->cmnd));
252 		SCpnt->cmd_len = rq->cmd_len;
253 		if (rq_data_dir(rq) == WRITE)
254 			SCpnt->sc_data_direction = DMA_TO_DEVICE;
255 		else if (rq->data_len)
256 			SCpnt->sc_data_direction = DMA_FROM_DEVICE;
257 		else
258 			SCpnt->sc_data_direction = DMA_NONE;
259 
260 		this_count = rq->data_len;
261 		if (rq->timeout)
262 			timeout = rq->timeout;
263 
264 		SCpnt->transfersize = rq->data_len;
265 		SCpnt->allowed = SD_PASSTHROUGH_RETRIES;
266 		goto queue;
267 	}
268 
269 	/*
270 	 * we only do REQ_CMD and REQ_BLOCK_PC
271 	 */
272 	if (!blk_fs_request(rq))
273 		return 0;
274 
275 	disk = rq->rq_disk;
276 	block = rq->sector;
277 	this_count = SCpnt->request_bufflen >> 9;
278 
279 	SCSI_LOG_HLQUEUE(1, printk("sd_init_command: disk=%s, block=%llu, "
280 			    "count=%d\n", disk->disk_name,
281 			 (unsigned long long)block, this_count));
282 
283 	if (!sdp || !scsi_device_online(sdp) ||
284  	    block + rq->nr_sectors > get_capacity(disk)) {
285 		SCSI_LOG_HLQUEUE(2, printk("Finishing %ld sectors\n",
286 				 rq->nr_sectors));
287 		SCSI_LOG_HLQUEUE(2, printk("Retry with 0x%p\n", SCpnt));
288 		return 0;
289 	}
290 
291 	if (sdp->changed) {
292 		/*
293 		 * quietly refuse to do anything to a changed disc until
294 		 * the changed bit has been reset
295 		 */
296 		/* printk("SCSI disk has been changed. Prohibiting further I/O.\n"); */
297 		return 0;
298 	}
299 	SCSI_LOG_HLQUEUE(2, printk("%s : block=%llu\n",
300 				   disk->disk_name, (unsigned long long)block));
301 
302 	/*
303 	 * If we have a 1K hardware sectorsize, prevent access to single
304 	 * 512 byte sectors.  In theory we could handle this - in fact
305 	 * the scsi cdrom driver must be able to handle this because
306 	 * we typically use 1K blocksizes, and cdroms typically have
307 	 * 2K hardware sectorsizes.  Of course, things are simpler
308 	 * with the cdrom, since it is read-only.  For performance
309 	 * reasons, the filesystems should be able to handle this
310 	 * and not force the scsi disk driver to use bounce buffers
311 	 * for this.
312 	 */
313 	if (sdp->sector_size == 1024) {
314 		if ((block & 1) || (rq->nr_sectors & 1)) {
315 			printk(KERN_ERR "sd: Bad block number requested");
316 			return 0;
317 		} else {
318 			block = block >> 1;
319 			this_count = this_count >> 1;
320 		}
321 	}
322 	if (sdp->sector_size == 2048) {
323 		if ((block & 3) || (rq->nr_sectors & 3)) {
324 			printk(KERN_ERR "sd: Bad block number requested");
325 			return 0;
326 		} else {
327 			block = block >> 2;
328 			this_count = this_count >> 2;
329 		}
330 	}
331 	if (sdp->sector_size == 4096) {
332 		if ((block & 7) || (rq->nr_sectors & 7)) {
333 			printk(KERN_ERR "sd: Bad block number requested");
334 			return 0;
335 		} else {
336 			block = block >> 3;
337 			this_count = this_count >> 3;
338 		}
339 	}
340 	if (rq_data_dir(rq) == WRITE) {
341 		if (!sdp->writeable) {
342 			return 0;
343 		}
344 		SCpnt->cmnd[0] = WRITE_6;
345 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
346 	} else if (rq_data_dir(rq) == READ) {
347 		SCpnt->cmnd[0] = READ_6;
348 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
349 	} else {
350 		printk(KERN_ERR "sd: Unknown command %lx\n", rq->flags);
351 /* overkill 	panic("Unknown sd command %lx\n", rq->flags); */
352 		return 0;
353 	}
354 
355 	SCSI_LOG_HLQUEUE(2, printk("%s : %s %d/%ld 512 byte blocks.\n",
356 		disk->disk_name, (rq_data_dir(rq) == WRITE) ?
357 		"writing" : "reading", this_count, rq->nr_sectors));
358 
359 	SCpnt->cmnd[1] = 0;
360 
361 	if (block > 0xffffffff) {
362 		SCpnt->cmnd[0] += READ_16 - READ_6;
363 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
364 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
365 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
366 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
367 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
368 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
369 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
370 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
371 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
372 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
373 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
374 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
375 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
376 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
377 		   SCpnt->device->use_10_for_rw) {
378 		if (this_count > 0xffff)
379 			this_count = 0xffff;
380 
381 		SCpnt->cmnd[0] += READ_10 - READ_6;
382 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
383 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
384 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
385 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
386 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
387 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
388 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
389 	} else {
390 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
391 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
392 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
393 		SCpnt->cmnd[4] = (unsigned char) this_count;
394 		SCpnt->cmnd[5] = 0;
395 	}
396 	SCpnt->request_bufflen = SCpnt->bufflen =
397 			this_count * sdp->sector_size;
398 
399 	/*
400 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
401 	 * host adapter, it's safe to assume that we can at least transfer
402 	 * this many bytes between each connect / disconnect.
403 	 */
404 	SCpnt->transfersize = sdp->sector_size;
405 	SCpnt->underflow = this_count << 9;
406 	SCpnt->allowed = SD_MAX_RETRIES;
407 
408 queue:
409 	SCpnt->timeout_per_command = timeout;
410 
411 	/*
412 	 * This is the completion routine we use.  This is matched in terms
413 	 * of capability to this function.
414 	 */
415 	SCpnt->done = sd_rw_intr;
416 
417 	/*
418 	 * This indicates that the command is ready from our end to be
419 	 * queued.
420 	 */
421 	return 1;
422 }
423 
424 /**
425  *	sd_open - open a scsi disk device
426  *	@inode: only i_rdev member may be used
427  *	@filp: only f_mode and f_flags may be used
428  *
429  *	Returns 0 if successful. Returns a negated errno value in case
430  *	of error.
431  *
432  *	Note: This can be called from a user context (e.g. fsck(1) )
433  *	or from within the kernel (e.g. as a result of a mount(1) ).
434  *	In the latter case @inode and @filp carry an abridged amount
435  *	of information as noted above.
436  **/
437 static int sd_open(struct inode *inode, struct file *filp)
438 {
439 	struct gendisk *disk = inode->i_bdev->bd_disk;
440 	struct scsi_disk *sdkp;
441 	struct scsi_device *sdev;
442 	int retval;
443 
444 	if (!(sdkp = scsi_disk_get(disk)))
445 		return -ENXIO;
446 
447 
448 	SCSI_LOG_HLQUEUE(3, printk("sd_open: disk=%s\n", disk->disk_name));
449 
450 	sdev = sdkp->device;
451 
452 	/*
453 	 * If the device is in error recovery, wait until it is done.
454 	 * If the device is offline, then disallow any access to it.
455 	 */
456 	retval = -ENXIO;
457 	if (!scsi_block_when_processing_errors(sdev))
458 		goto error_out;
459 
460 	if (sdev->removable || sdkp->write_prot)
461 		check_disk_change(inode->i_bdev);
462 
463 	/*
464 	 * If the drive is empty, just let the open fail.
465 	 */
466 	retval = -ENOMEDIUM;
467 	if (sdev->removable && !sdkp->media_present &&
468 	    !(filp->f_flags & O_NDELAY))
469 		goto error_out;
470 
471 	/*
472 	 * If the device has the write protect tab set, have the open fail
473 	 * if the user expects to be able to write to the thing.
474 	 */
475 	retval = -EROFS;
476 	if (sdkp->write_prot && (filp->f_mode & FMODE_WRITE))
477 		goto error_out;
478 
479 	/*
480 	 * It is possible that the disk changing stuff resulted in
481 	 * the device being taken offline.  If this is the case,
482 	 * report this to the user, and don't pretend that the
483 	 * open actually succeeded.
484 	 */
485 	retval = -ENXIO;
486 	if (!scsi_device_online(sdev))
487 		goto error_out;
488 
489 	if (!sdkp->openers++ && sdev->removable) {
490 		if (scsi_block_when_processing_errors(sdev))
491 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
492 	}
493 
494 	return 0;
495 
496 error_out:
497 	scsi_disk_put(sdkp);
498 	return retval;
499 }
500 
501 /**
502  *	sd_release - invoked when the (last) close(2) is called on this
503  *	scsi disk.
504  *	@inode: only i_rdev member may be used
505  *	@filp: only f_mode and f_flags may be used
506  *
507  *	Returns 0.
508  *
509  *	Note: may block (uninterruptible) if error recovery is underway
510  *	on this disk.
511  **/
512 static int sd_release(struct inode *inode, struct file *filp)
513 {
514 	struct gendisk *disk = inode->i_bdev->bd_disk;
515 	struct scsi_disk *sdkp = scsi_disk(disk);
516 	struct scsi_device *sdev = sdkp->device;
517 
518 	SCSI_LOG_HLQUEUE(3, printk("sd_release: disk=%s\n", disk->disk_name));
519 
520 	if (!--sdkp->openers && sdev->removable) {
521 		if (scsi_block_when_processing_errors(sdev))
522 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
523 	}
524 
525 	/*
526 	 * XXX and what if there are packets in flight and this close()
527 	 * XXX is followed by a "rmmod sd_mod"?
528 	 */
529 	scsi_disk_put(sdkp);
530 	return 0;
531 }
532 
533 static int sd_hdio_getgeo(struct block_device *bdev, struct hd_geometry __user *loc)
534 {
535 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
536 	struct scsi_device *sdp = sdkp->device;
537 	struct Scsi_Host *host = sdp->host;
538 	int diskinfo[4];
539 
540 	/* default to most commonly used values */
541         diskinfo[0] = 0x40;	/* 1 << 6 */
542        	diskinfo[1] = 0x20;	/* 1 << 5 */
543        	diskinfo[2] = sdkp->capacity >> 11;
544 
545 	/* override with calculated, extended default, or driver values */
546 	if (host->hostt->bios_param)
547 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
548 	else
549 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
550 
551 	if (put_user(diskinfo[0], &loc->heads))
552 		return -EFAULT;
553 	if (put_user(diskinfo[1], &loc->sectors))
554 		return -EFAULT;
555 	if (put_user(diskinfo[2], &loc->cylinders))
556 		return -EFAULT;
557 	if (put_user((unsigned)get_start_sect(bdev),
558 	             (unsigned long __user *)&loc->start))
559 		return -EFAULT;
560 	return 0;
561 }
562 
563 /**
564  *	sd_ioctl - process an ioctl
565  *	@inode: only i_rdev/i_bdev members may be used
566  *	@filp: only f_mode and f_flags may be used
567  *	@cmd: ioctl command number
568  *	@arg: this is third argument given to ioctl(2) system call.
569  *	Often contains a pointer.
570  *
571  *	Returns 0 if successful (some ioctls return postive numbers on
572  *	success as well). Returns a negated errno value in case of error.
573  *
574  *	Note: most ioctls are forward onto the block subsystem or further
575  *	down in the scsi subsytem.
576  **/
577 static int sd_ioctl(struct inode * inode, struct file * filp,
578 		    unsigned int cmd, unsigned long arg)
579 {
580 	struct block_device *bdev = inode->i_bdev;
581 	struct gendisk *disk = bdev->bd_disk;
582 	struct scsi_device *sdp = scsi_disk(disk)->device;
583 	void __user *p = (void __user *)arg;
584 	int error;
585 
586 	SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
587 						disk->disk_name, cmd));
588 
589 	/*
590 	 * If we are in the middle of error recovery, don't let anyone
591 	 * else try and use this device.  Also, if error recovery fails, it
592 	 * may try and take the device offline, in which case all further
593 	 * access to the device is prohibited.
594 	 */
595 	error = scsi_nonblockable_ioctl(sdp, cmd, p, filp);
596 	if (!scsi_block_when_processing_errors(sdp) || !error)
597 		return error;
598 
599 	if (cmd == HDIO_GETGEO) {
600 		if (!arg)
601 			return -EINVAL;
602 		return sd_hdio_getgeo(bdev, p);
603 	}
604 
605 	/*
606 	 * Send SCSI addressing ioctls directly to mid level, send other
607 	 * ioctls to block level and then onto mid level if they can't be
608 	 * resolved.
609 	 */
610 	switch (cmd) {
611 		case SCSI_IOCTL_GET_IDLUN:
612 		case SCSI_IOCTL_GET_BUS_NUMBER:
613 			return scsi_ioctl(sdp, cmd, p);
614 		default:
615 			error = scsi_cmd_ioctl(filp, disk, cmd, p);
616 			if (error != -ENOTTY)
617 				return error;
618 	}
619 	return scsi_ioctl(sdp, cmd, p);
620 }
621 
622 static void set_media_not_present(struct scsi_disk *sdkp)
623 {
624 	sdkp->media_present = 0;
625 	sdkp->capacity = 0;
626 	sdkp->device->changed = 1;
627 }
628 
629 /**
630  *	sd_media_changed - check if our medium changed
631  *	@disk: kernel device descriptor
632  *
633  *	Returns 0 if not applicable or no change; 1 if change
634  *
635  *	Note: this function is invoked from the block subsystem.
636  **/
637 static int sd_media_changed(struct gendisk *disk)
638 {
639 	struct scsi_disk *sdkp = scsi_disk(disk);
640 	struct scsi_device *sdp = sdkp->device;
641 	int retval;
642 
643 	SCSI_LOG_HLQUEUE(3, printk("sd_media_changed: disk=%s\n",
644 						disk->disk_name));
645 
646 	if (!sdp->removable)
647 		return 0;
648 
649 	/*
650 	 * If the device is offline, don't send any commands - just pretend as
651 	 * if the command failed.  If the device ever comes back online, we
652 	 * can deal with it then.  It is only because of unrecoverable errors
653 	 * that we would ever take a device offline in the first place.
654 	 */
655 	if (!scsi_device_online(sdp))
656 		goto not_present;
657 
658 	/*
659 	 * Using TEST_UNIT_READY enables differentiation between drive with
660 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
661 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
662 	 *
663 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
664 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
665 	 * sd_revalidate() is called.
666 	 */
667 	retval = -ENODEV;
668 	if (scsi_block_when_processing_errors(sdp))
669 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES);
670 
671 	/*
672 	 * Unable to test, unit probably not ready.   This usually
673 	 * means there is no disc in the drive.  Mark as changed,
674 	 * and we will figure it out later once the drive is
675 	 * available again.
676 	 */
677 	if (retval)
678 		 goto not_present;
679 
680 	/*
681 	 * For removable scsi disk we have to recognise the presence
682 	 * of a disk in the drive. This is kept in the struct scsi_disk
683 	 * struct and tested at open !  Daniel Roche (dan@lectra.fr)
684 	 */
685 	sdkp->media_present = 1;
686 
687 	retval = sdp->changed;
688 	sdp->changed = 0;
689 
690 	return retval;
691 
692 not_present:
693 	set_media_not_present(sdkp);
694 	return 1;
695 }
696 
697 static int sd_sync_cache(struct scsi_device *sdp)
698 {
699 	int retries, res;
700 	struct scsi_sense_hdr sshdr;
701 
702 	if (!scsi_device_online(sdp))
703 		return -ENODEV;
704 
705 
706 	for (retries = 3; retries > 0; --retries) {
707 		unsigned char cmd[10] = { 0 };
708 
709 		cmd[0] = SYNCHRONIZE_CACHE;
710 		/*
711 		 * Leave the rest of the command zero to indicate
712 		 * flush everything.
713 		 */
714 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
715 				       SD_TIMEOUT, SD_MAX_RETRIES);
716 		if (res == 0)
717 			break;
718 	}
719 
720 	if (res) {		printk(KERN_WARNING "FAILED\n  status = %x, message = %02x, "
721 				    "host = %d, driver = %02x\n  ",
722 				    status_byte(res), msg_byte(res),
723 				    host_byte(res), driver_byte(res));
724 			if (driver_byte(res) & DRIVER_SENSE)
725 				scsi_print_sense_hdr("sd", &sshdr);
726 	}
727 
728 	return res;
729 }
730 
731 static int sd_issue_flush(struct device *dev, sector_t *error_sector)
732 {
733 	int ret = 0;
734 	struct scsi_device *sdp = to_scsi_device(dev);
735 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
736 
737 	if (!sdkp)
738                return -ENODEV;
739 
740 	if (sdkp->WCE)
741 		ret = sd_sync_cache(sdp);
742 	scsi_disk_put(sdkp);
743 	return ret;
744 }
745 
746 static void sd_end_flush(request_queue_t *q, struct request *flush_rq)
747 {
748 	struct request *rq = flush_rq->end_io_data;
749 	struct scsi_cmnd *cmd = rq->special;
750 	unsigned int bytes = rq->hard_nr_sectors << 9;
751 
752 	if (!flush_rq->errors) {
753 		spin_unlock(q->queue_lock);
754 		scsi_io_completion(cmd, bytes, 0);
755 		spin_lock(q->queue_lock);
756 	} else if (blk_barrier_postflush(rq)) {
757 		spin_unlock(q->queue_lock);
758 		scsi_io_completion(cmd, 0, bytes);
759 		spin_lock(q->queue_lock);
760 	} else {
761 		/*
762 		 * force journal abort of barriers
763 		 */
764 		end_that_request_first(rq, -EOPNOTSUPP, rq->hard_nr_sectors);
765 		end_that_request_last(rq);
766 	}
767 }
768 
769 static int sd_prepare_flush(request_queue_t *q, struct request *rq)
770 {
771 	struct scsi_device *sdev = q->queuedata;
772 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(&sdev->sdev_gendev);
773 	int ret = 0;
774 
775 	if (sdkp) {
776 		if (sdkp->WCE) {
777 			memset(rq->cmd, 0, sizeof(rq->cmd));
778 			rq->flags |= REQ_BLOCK_PC | REQ_SOFTBARRIER;
779 			rq->timeout = SD_TIMEOUT;
780 			rq->cmd[0] = SYNCHRONIZE_CACHE;
781 			ret = 1;
782 		}
783 		scsi_disk_put(sdkp);
784 	}
785 	return ret;
786 }
787 
788 static void sd_rescan(struct device *dev)
789 {
790 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
791 
792 	if (sdkp) {
793 		sd_revalidate_disk(sdkp->disk);
794 		scsi_disk_put(sdkp);
795 	}
796 }
797 
798 
799 #ifdef CONFIG_COMPAT
800 /*
801  * This gets directly called from VFS. When the ioctl
802  * is not recognized we go back to the other translation paths.
803  */
804 static long sd_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
805 {
806 	struct block_device *bdev = file->f_dentry->d_inode->i_bdev;
807 	struct gendisk *disk = bdev->bd_disk;
808 	struct scsi_device *sdev = scsi_disk(disk)->device;
809 
810 	/*
811 	 * If we are in the middle of error recovery, don't let anyone
812 	 * else try and use this device.  Also, if error recovery fails, it
813 	 * may try and take the device offline, in which case all further
814 	 * access to the device is prohibited.
815 	 */
816 	if (!scsi_block_when_processing_errors(sdev))
817 		return -ENODEV;
818 
819 	if (sdev->host->hostt->compat_ioctl) {
820 		int ret;
821 
822 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
823 
824 		return ret;
825 	}
826 
827 	/*
828 	 * Let the static ioctl translation table take care of it.
829 	 */
830 	return -ENOIOCTLCMD;
831 }
832 #endif
833 
834 static struct block_device_operations sd_fops = {
835 	.owner			= THIS_MODULE,
836 	.open			= sd_open,
837 	.release		= sd_release,
838 	.ioctl			= sd_ioctl,
839 #ifdef CONFIG_COMPAT
840 	.compat_ioctl		= sd_compat_ioctl,
841 #endif
842 	.media_changed		= sd_media_changed,
843 	.revalidate_disk	= sd_revalidate_disk,
844 };
845 
846 /**
847  *	sd_rw_intr - bottom half handler: called when the lower level
848  *	driver has completed (successfully or otherwise) a scsi command.
849  *	@SCpnt: mid-level's per command structure.
850  *
851  *	Note: potentially run from within an ISR. Must not block.
852  **/
853 static void sd_rw_intr(struct scsi_cmnd * SCpnt)
854 {
855 	int result = SCpnt->result;
856 	int this_count = SCpnt->bufflen;
857 	int good_bytes = (result == 0 ? this_count : 0);
858 	sector_t block_sectors = 1;
859 	u64 first_err_block;
860 	sector_t error_sector;
861 	struct scsi_sense_hdr sshdr;
862 	int sense_valid = 0;
863 	int sense_deferred = 0;
864 	int info_valid;
865 
866 	if (result) {
867 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
868 		if (sense_valid)
869 			sense_deferred = scsi_sense_is_deferred(&sshdr);
870 	}
871 
872 #ifdef CONFIG_SCSI_LOGGING
873 	SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: %s: res=0x%x\n",
874 				SCpnt->request->rq_disk->disk_name, result));
875 	if (sense_valid) {
876 		SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: sb[respc,sk,asc,"
877 				"ascq]=%x,%x,%x,%x\n", sshdr.response_code,
878 				sshdr.sense_key, sshdr.asc, sshdr.ascq));
879 	}
880 #endif
881 	/*
882 	   Handle MEDIUM ERRORs that indicate partial success.  Since this is a
883 	   relatively rare error condition, no care is taken to avoid
884 	   unnecessary additional work such as memcpy's that could be avoided.
885 	 */
886 
887 	/*
888 	 * If SG_IO from block layer then set good_bytes to stop retries;
889 	 * else if errors, check them, and if necessary prepare for
890 	 * (partial) retries.
891 	 */
892 	if (blk_pc_request(SCpnt->request))
893 		good_bytes = this_count;
894 	else if (driver_byte(result) != 0 &&
895 		 sense_valid && !sense_deferred) {
896 		switch (sshdr.sense_key) {
897 		case MEDIUM_ERROR:
898 			if (!blk_fs_request(SCpnt->request))
899 				break;
900 			info_valid = scsi_get_sense_info_fld(
901 				SCpnt->sense_buffer, SCSI_SENSE_BUFFERSIZE,
902 				&first_err_block);
903 			/*
904 			 * May want to warn and skip if following cast results
905 			 * in actual truncation (if sector_t < 64 bits)
906 			 */
907 			error_sector = (sector_t)first_err_block;
908 			if (SCpnt->request->bio != NULL)
909 				block_sectors = bio_sectors(SCpnt->request->bio);
910 			switch (SCpnt->device->sector_size) {
911 			case 1024:
912 				error_sector <<= 1;
913 				if (block_sectors < 2)
914 					block_sectors = 2;
915 				break;
916 			case 2048:
917 				error_sector <<= 2;
918 				if (block_sectors < 4)
919 					block_sectors = 4;
920 				break;
921 			case 4096:
922 				error_sector <<=3;
923 				if (block_sectors < 8)
924 					block_sectors = 8;
925 				break;
926 			case 256:
927 				error_sector >>= 1;
928 				break;
929 			default:
930 				break;
931 			}
932 
933 			error_sector &= ~(block_sectors - 1);
934 			good_bytes = (error_sector - SCpnt->request->sector) << 9;
935 			if (good_bytes < 0 || good_bytes >= this_count)
936 				good_bytes = 0;
937 			break;
938 
939 		case RECOVERED_ERROR: /* an error occurred, but it recovered */
940 		case NO_SENSE: /* LLDD got sense data */
941 			/*
942 			 * Inform the user, but make sure that it's not treated
943 			 * as a hard error.
944 			 */
945 			scsi_print_sense("sd", SCpnt);
946 			SCpnt->result = 0;
947 			memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
948 			good_bytes = this_count;
949 			break;
950 
951 		case ILLEGAL_REQUEST:
952 			if (SCpnt->device->use_10_for_rw &&
953 			    (SCpnt->cmnd[0] == READ_10 ||
954 			     SCpnt->cmnd[0] == WRITE_10))
955 				SCpnt->device->use_10_for_rw = 0;
956 			if (SCpnt->device->use_10_for_ms &&
957 			    (SCpnt->cmnd[0] == MODE_SENSE_10 ||
958 			     SCpnt->cmnd[0] == MODE_SELECT_10))
959 				SCpnt->device->use_10_for_ms = 0;
960 			break;
961 
962 		default:
963 			break;
964 		}
965 	}
966 	/*
967 	 * This calls the generic completion function, now that we know
968 	 * how many actual sectors finished, and how many sectors we need
969 	 * to say have failed.
970 	 */
971 	scsi_io_completion(SCpnt, good_bytes, block_sectors << 9);
972 }
973 
974 static int media_not_present(struct scsi_disk *sdkp,
975 			     struct scsi_sense_hdr *sshdr)
976 {
977 
978 	if (!scsi_sense_valid(sshdr))
979 		return 0;
980 	/* not invoked for commands that could return deferred errors */
981 	if (sshdr->sense_key != NOT_READY &&
982 	    sshdr->sense_key != UNIT_ATTENTION)
983 		return 0;
984 	if (sshdr->asc != 0x3A) /* medium not present */
985 		return 0;
986 
987 	set_media_not_present(sdkp);
988 	return 1;
989 }
990 
991 /*
992  * spinup disk - called only in sd_revalidate_disk()
993  */
994 static void
995 sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
996 {
997 	unsigned char cmd[10];
998 	unsigned long spintime_expire = 0;
999 	int retries, spintime;
1000 	unsigned int the_result;
1001 	struct scsi_sense_hdr sshdr;
1002 	int sense_valid = 0;
1003 
1004 	spintime = 0;
1005 
1006 	/* Spin up drives, as required.  Only do this at boot time */
1007 	/* Spinup needs to be done for module loads too. */
1008 	do {
1009 		retries = 0;
1010 
1011 		do {
1012 			cmd[0] = TEST_UNIT_READY;
1013 			memset((void *) &cmd[1], 0, 9);
1014 
1015 			the_result = scsi_execute_req(sdkp->device, cmd,
1016 						      DMA_NONE, NULL, 0,
1017 						      &sshdr, SD_TIMEOUT,
1018 						      SD_MAX_RETRIES);
1019 
1020 			if (the_result)
1021 				sense_valid = scsi_sense_valid(&sshdr);
1022 			retries++;
1023 		} while (retries < 3 &&
1024 			 (!scsi_status_is_good(the_result) ||
1025 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
1026 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1027 
1028 		/*
1029 		 * If the drive has indicated to us that it doesn't have
1030 		 * any media in it, don't bother with any of the rest of
1031 		 * this crap.
1032 		 */
1033 		if (media_not_present(sdkp, &sshdr))
1034 			return;
1035 
1036 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1037 			/* no sense, TUR either succeeded or failed
1038 			 * with a status error */
1039 			if(!spintime && !scsi_status_is_good(the_result))
1040 				printk(KERN_NOTICE "%s: Unit Not Ready, "
1041 				       "error = 0x%x\n", diskname, the_result);
1042 			break;
1043 		}
1044 
1045 		/*
1046 		 * The device does not want the automatic start to be issued.
1047 		 */
1048 		if (sdkp->device->no_start_on_add) {
1049 			break;
1050 		}
1051 
1052 		/*
1053 		 * If manual intervention is required, or this is an
1054 		 * absent USB storage device, a spinup is meaningless.
1055 		 */
1056 		if (sense_valid &&
1057 		    sshdr.sense_key == NOT_READY &&
1058 		    sshdr.asc == 4 && sshdr.ascq == 3) {
1059 			break;		/* manual intervention required */
1060 
1061 		/*
1062 		 * Issue command to spin up drive when not ready
1063 		 */
1064 		} else if (sense_valid && sshdr.sense_key == NOT_READY) {
1065 			if (!spintime) {
1066 				printk(KERN_NOTICE "%s: Spinning up disk...",
1067 				       diskname);
1068 				cmd[0] = START_STOP;
1069 				cmd[1] = 1;	/* Return immediately */
1070 				memset((void *) &cmd[2], 0, 8);
1071 				cmd[4] = 1;	/* Start spin cycle */
1072 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1073 						 NULL, 0, &sshdr,
1074 						 SD_TIMEOUT, SD_MAX_RETRIES);
1075 				spintime_expire = jiffies + 100 * HZ;
1076 				spintime = 1;
1077 			}
1078 			/* Wait 1 second for next try */
1079 			msleep(1000);
1080 			printk(".");
1081 
1082 		/*
1083 		 * Wait for USB flash devices with slow firmware.
1084 		 * Yes, this sense key/ASC combination shouldn't
1085 		 * occur here.  It's characteristic of these devices.
1086 		 */
1087 		} else if (sense_valid &&
1088 				sshdr.sense_key == UNIT_ATTENTION &&
1089 				sshdr.asc == 0x28) {
1090 			if (!spintime) {
1091 				spintime_expire = jiffies + 5 * HZ;
1092 				spintime = 1;
1093 			}
1094 			/* Wait 1 second for next try */
1095 			msleep(1000);
1096 		} else {
1097 			/* we don't understand the sense code, so it's
1098 			 * probably pointless to loop */
1099 			if(!spintime) {
1100 				printk(KERN_NOTICE "%s: Unit Not Ready, "
1101 					"sense:\n", diskname);
1102 				scsi_print_sense_hdr("", &sshdr);
1103 			}
1104 			break;
1105 		}
1106 
1107 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1108 
1109 	if (spintime) {
1110 		if (scsi_status_is_good(the_result))
1111 			printk("ready\n");
1112 		else
1113 			printk("not responding...\n");
1114 	}
1115 }
1116 
1117 /*
1118  * read disk capacity
1119  */
1120 static void
1121 sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
1122 		 unsigned char *buffer)
1123 {
1124 	unsigned char cmd[16];
1125 	int the_result, retries;
1126 	int sector_size = 0;
1127 	int longrc = 0;
1128 	struct scsi_sense_hdr sshdr;
1129 	int sense_valid = 0;
1130 	struct scsi_device *sdp = sdkp->device;
1131 
1132 repeat:
1133 	retries = 3;
1134 	do {
1135 		if (longrc) {
1136 			memset((void *) cmd, 0, 16);
1137 			cmd[0] = SERVICE_ACTION_IN;
1138 			cmd[1] = SAI_READ_CAPACITY_16;
1139 			cmd[13] = 12;
1140 			memset((void *) buffer, 0, 12);
1141 		} else {
1142 			cmd[0] = READ_CAPACITY;
1143 			memset((void *) &cmd[1], 0, 9);
1144 			memset((void *) buffer, 0, 8);
1145 		}
1146 
1147 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1148 					      buffer, longrc ? 12 : 8, &sshdr,
1149 					      SD_TIMEOUT, SD_MAX_RETRIES);
1150 
1151 		if (media_not_present(sdkp, &sshdr))
1152 			return;
1153 
1154 		if (the_result)
1155 			sense_valid = scsi_sense_valid(&sshdr);
1156 		retries--;
1157 
1158 	} while (the_result && retries);
1159 
1160 	if (the_result && !longrc) {
1161 		printk(KERN_NOTICE "%s : READ CAPACITY failed.\n"
1162 		       "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
1163 		       diskname, diskname,
1164 		       status_byte(the_result),
1165 		       msg_byte(the_result),
1166 		       host_byte(the_result),
1167 		       driver_byte(the_result));
1168 
1169 		if (driver_byte(the_result) & DRIVER_SENSE)
1170 			scsi_print_sense_hdr("sd", &sshdr);
1171 		else
1172 			printk("%s : sense not available. \n", diskname);
1173 
1174 		/* Set dirty bit for removable devices if not ready -
1175 		 * sometimes drives will not report this properly. */
1176 		if (sdp->removable &&
1177 		    sense_valid && sshdr.sense_key == NOT_READY)
1178 			sdp->changed = 1;
1179 
1180 		/* Either no media are present but the drive didn't tell us,
1181 		   or they are present but the read capacity command fails */
1182 		/* sdkp->media_present = 0; -- not always correct */
1183 		sdkp->capacity = 0x200000; /* 1 GB - random */
1184 
1185 		return;
1186 	} else if (the_result && longrc) {
1187 		/* READ CAPACITY(16) has been failed */
1188 		printk(KERN_NOTICE "%s : READ CAPACITY(16) failed.\n"
1189 		       "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
1190 		       diskname, diskname,
1191 		       status_byte(the_result),
1192 		       msg_byte(the_result),
1193 		       host_byte(the_result),
1194 		       driver_byte(the_result));
1195 		printk(KERN_NOTICE "%s : use 0xffffffff as device size\n",
1196 		       diskname);
1197 
1198 		sdkp->capacity = 1 + (sector_t) 0xffffffff;
1199 		goto got_data;
1200 	}
1201 
1202 	if (!longrc) {
1203 		sector_size = (buffer[4] << 24) |
1204 			(buffer[5] << 16) | (buffer[6] << 8) | buffer[7];
1205 		if (buffer[0] == 0xff && buffer[1] == 0xff &&
1206 		    buffer[2] == 0xff && buffer[3] == 0xff) {
1207 			if(sizeof(sdkp->capacity) > 4) {
1208 				printk(KERN_NOTICE "%s : very big device. try to use"
1209 				       " READ CAPACITY(16).\n", diskname);
1210 				longrc = 1;
1211 				goto repeat;
1212 			}
1213 			printk(KERN_ERR "%s: too big for this kernel.  Use a "
1214 			       "kernel compiled with support for large block "
1215 			       "devices.\n", diskname);
1216 			sdkp->capacity = 0;
1217 			goto got_data;
1218 		}
1219 		sdkp->capacity = 1 + (((sector_t)buffer[0] << 24) |
1220 			(buffer[1] << 16) |
1221 			(buffer[2] << 8) |
1222 			buffer[3]);
1223 	} else {
1224 		sdkp->capacity = 1 + (((u64)buffer[0] << 56) |
1225 			((u64)buffer[1] << 48) |
1226 			((u64)buffer[2] << 40) |
1227 			((u64)buffer[3] << 32) |
1228 			((sector_t)buffer[4] << 24) |
1229 			((sector_t)buffer[5] << 16) |
1230 			((sector_t)buffer[6] << 8)  |
1231 			(sector_t)buffer[7]);
1232 
1233 		sector_size = (buffer[8] << 24) |
1234 			(buffer[9] << 16) | (buffer[10] << 8) | buffer[11];
1235 	}
1236 
1237 	/* Some devices return the total number of sectors, not the
1238 	 * highest sector number.  Make the necessary adjustment. */
1239 	if (sdp->fix_capacity)
1240 		--sdkp->capacity;
1241 
1242 got_data:
1243 	if (sector_size == 0) {
1244 		sector_size = 512;
1245 		printk(KERN_NOTICE "%s : sector size 0 reported, "
1246 		       "assuming 512.\n", diskname);
1247 	}
1248 
1249 	if (sector_size != 512 &&
1250 	    sector_size != 1024 &&
1251 	    sector_size != 2048 &&
1252 	    sector_size != 4096 &&
1253 	    sector_size != 256) {
1254 		printk(KERN_NOTICE "%s : unsupported sector size "
1255 		       "%d.\n", diskname, sector_size);
1256 		/*
1257 		 * The user might want to re-format the drive with
1258 		 * a supported sectorsize.  Once this happens, it
1259 		 * would be relatively trivial to set the thing up.
1260 		 * For this reason, we leave the thing in the table.
1261 		 */
1262 		sdkp->capacity = 0;
1263 		/*
1264 		 * set a bogus sector size so the normal read/write
1265 		 * logic in the block layer will eventually refuse any
1266 		 * request on this device without tripping over power
1267 		 * of two sector size assumptions
1268 		 */
1269 		sector_size = 512;
1270 	}
1271 	{
1272 		/*
1273 		 * The msdos fs needs to know the hardware sector size
1274 		 * So I have created this table. See ll_rw_blk.c
1275 		 * Jacques Gelinas (Jacques@solucorp.qc.ca)
1276 		 */
1277 		int hard_sector = sector_size;
1278 		sector_t sz = (sdkp->capacity/2) * (hard_sector/256);
1279 		request_queue_t *queue = sdp->request_queue;
1280 		sector_t mb = sz;
1281 
1282 		blk_queue_hardsect_size(queue, hard_sector);
1283 		/* avoid 64-bit division on 32-bit platforms */
1284 		sector_div(sz, 625);
1285 		mb -= sz - 974;
1286 		sector_div(mb, 1950);
1287 
1288 		printk(KERN_NOTICE "SCSI device %s: "
1289 		       "%llu %d-byte hdwr sectors (%llu MB)\n",
1290 		       diskname, (unsigned long long)sdkp->capacity,
1291 		       hard_sector, (unsigned long long)mb);
1292 	}
1293 
1294 	/* Rescale capacity to 512-byte units */
1295 	if (sector_size == 4096)
1296 		sdkp->capacity <<= 3;
1297 	else if (sector_size == 2048)
1298 		sdkp->capacity <<= 2;
1299 	else if (sector_size == 1024)
1300 		sdkp->capacity <<= 1;
1301 	else if (sector_size == 256)
1302 		sdkp->capacity >>= 1;
1303 
1304 	sdkp->device->sector_size = sector_size;
1305 }
1306 
1307 /* called with buffer of length 512 */
1308 static inline int
1309 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1310 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
1311 		 struct scsi_sense_hdr *sshdr)
1312 {
1313 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1314 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
1315 			       sshdr);
1316 }
1317 
1318 /*
1319  * read write protect setting, if possible - called only in sd_revalidate_disk()
1320  * called with buffer of length 512
1321  */
1322 static void
1323 sd_read_write_protect_flag(struct scsi_disk *sdkp, char *diskname,
1324 			   unsigned char *buffer)
1325 {
1326 	int res;
1327 	struct scsi_device *sdp = sdkp->device;
1328 	struct scsi_mode_data data;
1329 
1330 	set_disk_ro(sdkp->disk, 0);
1331 	if (sdp->skip_ms_page_3f) {
1332 		printk(KERN_NOTICE "%s: assuming Write Enabled\n", diskname);
1333 		return;
1334 	}
1335 
1336 	if (sdp->use_192_bytes_for_3f) {
1337 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1338 	} else {
1339 		/*
1340 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1341 		 * We have to start carefully: some devices hang if we ask
1342 		 * for more than is available.
1343 		 */
1344 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1345 
1346 		/*
1347 		 * Second attempt: ask for page 0 When only page 0 is
1348 		 * implemented, a request for page 3F may return Sense Key
1349 		 * 5: Illegal Request, Sense Code 24: Invalid field in
1350 		 * CDB.
1351 		 */
1352 		if (!scsi_status_is_good(res))
1353 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1354 
1355 		/*
1356 		 * Third attempt: ask 255 bytes, as we did earlier.
1357 		 */
1358 		if (!scsi_status_is_good(res))
1359 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1360 					       &data, NULL);
1361 	}
1362 
1363 	if (!scsi_status_is_good(res)) {
1364 		printk(KERN_WARNING
1365 		       "%s: test WP failed, assume Write Enabled\n", diskname);
1366 	} else {
1367 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
1368 		set_disk_ro(sdkp->disk, sdkp->write_prot);
1369 		printk(KERN_NOTICE "%s: Write Protect is %s\n", diskname,
1370 		       sdkp->write_prot ? "on" : "off");
1371 		printk(KERN_DEBUG "%s: Mode Sense: %02x %02x %02x %02x\n",
1372 		       diskname, buffer[0], buffer[1], buffer[2], buffer[3]);
1373 	}
1374 }
1375 
1376 /*
1377  * sd_read_cache_type - called only from sd_revalidate_disk()
1378  * called with buffer of length 512
1379  */
1380 static void
1381 sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
1382 		   unsigned char *buffer)
1383 {
1384 	int len = 0, res;
1385 	struct scsi_device *sdp = sdkp->device;
1386 
1387 	int dbd;
1388 	int modepage;
1389 	struct scsi_mode_data data;
1390 	struct scsi_sense_hdr sshdr;
1391 
1392 	if (sdp->skip_ms_page_8)
1393 		goto defaults;
1394 
1395 	if (sdp->type == TYPE_RBC) {
1396 		modepage = 6;
1397 		dbd = 8;
1398 	} else {
1399 		modepage = 8;
1400 		dbd = 0;
1401 	}
1402 
1403 	/* cautiously ask */
1404 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, 4, &data, &sshdr);
1405 
1406 	if (!scsi_status_is_good(res))
1407 		goto bad_sense;
1408 
1409 	/* that went OK, now ask for the proper length */
1410 	len = data.length;
1411 
1412 	/*
1413 	 * We're only interested in the first three bytes, actually.
1414 	 * But the data cache page is defined for the first 20.
1415 	 */
1416 	if (len < 3)
1417 		goto bad_sense;
1418 	if (len > 20)
1419 		len = 20;
1420 
1421 	/* Take headers and block descriptors into account */
1422 	len += data.header_length + data.block_descriptor_length;
1423 
1424 	/* Get the data */
1425 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
1426 
1427 	if (scsi_status_is_good(res)) {
1428 		const char *types[] = {
1429 			"write through", "none", "write back",
1430 			"write back, no read (daft)"
1431 		};
1432 		int ct = 0;
1433 		int offset = data.header_length + data.block_descriptor_length;
1434 
1435 		if ((buffer[offset] & 0x3f) != modepage) {
1436 			printk(KERN_ERR "%s: got wrong page\n", diskname);
1437 			goto defaults;
1438 		}
1439 
1440 		if (modepage == 8) {
1441 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
1442 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
1443 		} else {
1444 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
1445 			sdkp->RCD = 0;
1446 		}
1447 
1448 		ct =  sdkp->RCD + 2*sdkp->WCE;
1449 
1450 		printk(KERN_NOTICE "SCSI device %s: drive cache: %s\n",
1451 		       diskname, types[ct]);
1452 
1453 		return;
1454 	}
1455 
1456 bad_sense:
1457 	if (scsi_sense_valid(&sshdr) &&
1458 	    sshdr.sense_key == ILLEGAL_REQUEST &&
1459 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
1460 		printk(KERN_NOTICE "%s: cache data unavailable\n",
1461 		       diskname);	/* Invalid field in CDB */
1462 	else
1463 		printk(KERN_ERR "%s: asking for cache data failed\n",
1464 		       diskname);
1465 
1466 defaults:
1467 	printk(KERN_ERR "%s: assuming drive cache: write through\n",
1468 	       diskname);
1469 	sdkp->WCE = 0;
1470 	sdkp->RCD = 0;
1471 }
1472 
1473 /**
1474  *	sd_revalidate_disk - called the first time a new disk is seen,
1475  *	performs disk spin up, read_capacity, etc.
1476  *	@disk: struct gendisk we care about
1477  **/
1478 static int sd_revalidate_disk(struct gendisk *disk)
1479 {
1480 	struct scsi_disk *sdkp = scsi_disk(disk);
1481 	struct scsi_device *sdp = sdkp->device;
1482 	unsigned char *buffer;
1483 
1484 	SCSI_LOG_HLQUEUE(3, printk("sd_revalidate_disk: disk=%s\n", disk->disk_name));
1485 
1486 	/*
1487 	 * If the device is offline, don't try and read capacity or any
1488 	 * of the other niceties.
1489 	 */
1490 	if (!scsi_device_online(sdp))
1491 		goto out;
1492 
1493 	buffer = kmalloc(512, GFP_KERNEL | __GFP_DMA);
1494 	if (!buffer) {
1495 		printk(KERN_WARNING "(sd_revalidate_disk:) Memory allocation "
1496 		       "failure.\n");
1497 		goto out;
1498 	}
1499 
1500 	/* defaults, until the device tells us otherwise */
1501 	sdp->sector_size = 512;
1502 	sdkp->capacity = 0;
1503 	sdkp->media_present = 1;
1504 	sdkp->write_prot = 0;
1505 	sdkp->WCE = 0;
1506 	sdkp->RCD = 0;
1507 
1508 	sd_spinup_disk(sdkp, disk->disk_name);
1509 
1510 	/*
1511 	 * Without media there is no reason to ask; moreover, some devices
1512 	 * react badly if we do.
1513 	 */
1514 	if (sdkp->media_present) {
1515 		sd_read_capacity(sdkp, disk->disk_name, buffer);
1516 		if (sdp->removable)
1517 			sd_read_write_protect_flag(sdkp, disk->disk_name,
1518 						   buffer);
1519 		sd_read_cache_type(sdkp, disk->disk_name, buffer);
1520 	}
1521 
1522 	set_capacity(disk, sdkp->capacity);
1523 	kfree(buffer);
1524 
1525  out:
1526 	return 0;
1527 }
1528 
1529 /**
1530  *	sd_probe - called during driver initialization and whenever a
1531  *	new scsi device is attached to the system. It is called once
1532  *	for each scsi device (not just disks) present.
1533  *	@dev: pointer to device object
1534  *
1535  *	Returns 0 if successful (or not interested in this scsi device
1536  *	(e.g. scanner)); 1 when there is an error.
1537  *
1538  *	Note: this function is invoked from the scsi mid-level.
1539  *	This function sets up the mapping between a given
1540  *	<host,channel,id,lun> (found in sdp) and new device name
1541  *	(e.g. /dev/sda). More precisely it is the block device major
1542  *	and minor number that is chosen here.
1543  *
1544  *	Assume sd_attach is not re-entrant (for time being)
1545  *	Also think about sd_attach() and sd_remove() running coincidentally.
1546  **/
1547 static int sd_probe(struct device *dev)
1548 {
1549 	struct scsi_device *sdp = to_scsi_device(dev);
1550 	struct scsi_disk *sdkp;
1551 	struct gendisk *gd;
1552 	u32 index;
1553 	int error;
1554 
1555 	error = -ENODEV;
1556 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
1557 		goto out;
1558 
1559 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
1560 					"sd_attach\n"));
1561 
1562 	error = -ENOMEM;
1563 	sdkp = kmalloc(sizeof(*sdkp), GFP_KERNEL);
1564 	if (!sdkp)
1565 		goto out;
1566 
1567 	memset (sdkp, 0, sizeof(*sdkp));
1568 	kref_init(&sdkp->kref);
1569 
1570 	gd = alloc_disk(16);
1571 	if (!gd)
1572 		goto out_free;
1573 
1574 	if (!idr_pre_get(&sd_index_idr, GFP_KERNEL))
1575 		goto out_put;
1576 
1577 	spin_lock(&sd_index_lock);
1578 	error = idr_get_new(&sd_index_idr, NULL, &index);
1579 	spin_unlock(&sd_index_lock);
1580 
1581 	if (index >= SD_MAX_DISKS)
1582 		error = -EBUSY;
1583 	if (error)
1584 		goto out_put;
1585 
1586 	get_device(&sdp->sdev_gendev);
1587 	sdkp->device = sdp;
1588 	sdkp->driver = &sd_template;
1589 	sdkp->disk = gd;
1590 	sdkp->index = index;
1591 	sdkp->openers = 0;
1592 
1593 	if (!sdp->timeout) {
1594 		if (sdp->type != TYPE_MOD)
1595 			sdp->timeout = SD_TIMEOUT;
1596 		else
1597 			sdp->timeout = SD_MOD_TIMEOUT;
1598 	}
1599 
1600 	gd->major = sd_major((index & 0xf0) >> 4);
1601 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
1602 	gd->minors = 16;
1603 	gd->fops = &sd_fops;
1604 
1605 	if (index < 26) {
1606 		sprintf(gd->disk_name, "sd%c", 'a' + index % 26);
1607 	} else if (index < (26 + 1) * 26) {
1608 		sprintf(gd->disk_name, "sd%c%c",
1609 			'a' + index / 26 - 1,'a' + index % 26);
1610 	} else {
1611 		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
1612 		const unsigned int m2 = (index / 26 - 1) % 26;
1613 		const unsigned int m3 =  index % 26;
1614 		sprintf(gd->disk_name, "sd%c%c%c",
1615 			'a' + m1, 'a' + m2, 'a' + m3);
1616 	}
1617 
1618 	strcpy(gd->devfs_name, sdp->devfs_name);
1619 
1620 	gd->private_data = &sdkp->driver;
1621 
1622 	sd_revalidate_disk(gd);
1623 
1624 	gd->driverfs_dev = &sdp->sdev_gendev;
1625 	gd->flags = GENHD_FL_DRIVERFS;
1626 	if (sdp->removable)
1627 		gd->flags |= GENHD_FL_REMOVABLE;
1628 	gd->queue = sdkp->device->request_queue;
1629 
1630 	dev_set_drvdata(dev, sdkp);
1631 	add_disk(gd);
1632 
1633 	sdev_printk(KERN_NOTICE, sdp, "Attached scsi %sdisk %s\n",
1634 		    sdp->removable ? "removable " : "", gd->disk_name);
1635 
1636 	return 0;
1637 
1638 out_put:
1639 	put_disk(gd);
1640 out_free:
1641 	kfree(sdkp);
1642 out:
1643 	return error;
1644 }
1645 
1646 /**
1647  *	sd_remove - called whenever a scsi disk (previously recognized by
1648  *	sd_probe) is detached from the system. It is called (potentially
1649  *	multiple times) during sd module unload.
1650  *	@sdp: pointer to mid level scsi device object
1651  *
1652  *	Note: this function is invoked from the scsi mid-level.
1653  *	This function potentially frees up a device name (e.g. /dev/sdc)
1654  *	that could be re-used by a subsequent sd_probe().
1655  *	This function is not called when the built-in sd driver is "exit-ed".
1656  **/
1657 static int sd_remove(struct device *dev)
1658 {
1659 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1660 
1661 	del_gendisk(sdkp->disk);
1662 	sd_shutdown(dev);
1663 
1664 	down(&sd_ref_sem);
1665 	dev_set_drvdata(dev, NULL);
1666 	kref_put(&sdkp->kref, scsi_disk_release);
1667 	up(&sd_ref_sem);
1668 
1669 	return 0;
1670 }
1671 
1672 /**
1673  *	scsi_disk_release - Called to free the scsi_disk structure
1674  *	@kref: pointer to embedded kref
1675  *
1676  *	sd_ref_sem must be held entering this routine.  Because it is
1677  *	called on last put, you should always use the scsi_disk_get()
1678  *	scsi_disk_put() helpers which manipulate the semaphore directly
1679  *	and never do a direct kref_put().
1680  **/
1681 static void scsi_disk_release(struct kref *kref)
1682 {
1683 	struct scsi_disk *sdkp = to_scsi_disk(kref);
1684 	struct gendisk *disk = sdkp->disk;
1685 
1686 	spin_lock(&sd_index_lock);
1687 	idr_remove(&sd_index_idr, sdkp->index);
1688 	spin_unlock(&sd_index_lock);
1689 
1690 	disk->private_data = NULL;
1691 	put_disk(disk);
1692 	put_device(&sdkp->device->sdev_gendev);
1693 
1694 	kfree(sdkp);
1695 }
1696 
1697 /*
1698  * Send a SYNCHRONIZE CACHE instruction down to the device through
1699  * the normal SCSI command structure.  Wait for the command to
1700  * complete.
1701  */
1702 static void sd_shutdown(struct device *dev)
1703 {
1704 	struct scsi_device *sdp = to_scsi_device(dev);
1705 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1706 
1707 	if (!sdkp)
1708 		return;         /* this can happen */
1709 
1710 	if (sdkp->WCE) {
1711 		printk(KERN_NOTICE "Synchronizing SCSI cache for disk %s: \n",
1712 				sdkp->disk->disk_name);
1713 		sd_sync_cache(sdp);
1714 	}
1715 	scsi_disk_put(sdkp);
1716 }
1717 
1718 /**
1719  *	init_sd - entry point for this driver (both when built in or when
1720  *	a module).
1721  *
1722  *	Note: this function registers this driver with the scsi mid-level.
1723  **/
1724 static int __init init_sd(void)
1725 {
1726 	int majors = 0, i;
1727 
1728 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
1729 
1730 	for (i = 0; i < SD_MAJORS; i++)
1731 		if (register_blkdev(sd_major(i), "sd") == 0)
1732 			majors++;
1733 
1734 	if (!majors)
1735 		return -ENODEV;
1736 
1737 	return scsi_register_driver(&sd_template.gendrv);
1738 }
1739 
1740 /**
1741  *	exit_sd - exit point for this driver (when it is a module).
1742  *
1743  *	Note: this function unregisters this driver from the scsi mid-level.
1744  **/
1745 static void __exit exit_sd(void)
1746 {
1747 	int i;
1748 
1749 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
1750 
1751 	scsi_unregister_driver(&sd_template.gendrv);
1752 	for (i = 0; i < SD_MAJORS; i++)
1753 		unregister_blkdev(sd_major(i), "sd");
1754 }
1755 
1756 MODULE_LICENSE("GPL");
1757 MODULE_AUTHOR("Eric Youngdale");
1758 MODULE_DESCRIPTION("SCSI disk (sd) driver");
1759 
1760 module_init(init_sd);
1761 module_exit(exit_sd);
1762