xref: /linux/drivers/scsi/sd.c (revision 87c2ce3b9305b9b723faeedf6e32ef703ec9b33a)
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 	unsigned	DPOFUA : 1;	/* state of disk DPOFUA bit */
106 };
107 
108 static DEFINE_IDR(sd_index_idr);
109 static DEFINE_SPINLOCK(sd_index_lock);
110 
111 /* This semaphore is used to mediate the 0->1 reference get in the
112  * face of object destruction (i.e. we can't allow a get on an
113  * object after last put) */
114 static DECLARE_MUTEX(sd_ref_sem);
115 
116 static int sd_revalidate_disk(struct gendisk *disk);
117 static void sd_rw_intr(struct scsi_cmnd * SCpnt);
118 
119 static int sd_probe(struct device *);
120 static int sd_remove(struct device *);
121 static void sd_shutdown(struct device *dev);
122 static void sd_rescan(struct device *);
123 static int sd_init_command(struct scsi_cmnd *);
124 static int sd_issue_flush(struct device *, sector_t *);
125 static void 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 };
141 
142 /*
143  * Device no to disk mapping:
144  *
145  *       major         disc2     disc  p1
146  *   |............|.............|....|....| <- dev_t
147  *    31        20 19          8 7  4 3  0
148  *
149  * Inside a major, we have 16k disks, however mapped non-
150  * contiguously. The first 16 disks are for major0, the next
151  * ones with major1, ... Disk 256 is for major0 again, disk 272
152  * for major1, ...
153  * As we stay compatible with our numbering scheme, we can reuse
154  * the well-know SCSI majors 8, 65--71, 136--143.
155  */
156 static int sd_major(int major_idx)
157 {
158 	switch (major_idx) {
159 	case 0:
160 		return SCSI_DISK0_MAJOR;
161 	case 1 ... 7:
162 		return SCSI_DISK1_MAJOR + major_idx - 1;
163 	case 8 ... 15:
164 		return SCSI_DISK8_MAJOR + major_idx - 8;
165 	default:
166 		BUG();
167 		return 0;	/* shut up gcc */
168 	}
169 }
170 
171 #define to_scsi_disk(obj) container_of(obj,struct scsi_disk,kref)
172 
173 static inline struct scsi_disk *scsi_disk(struct gendisk *disk)
174 {
175 	return container_of(disk->private_data, struct scsi_disk, driver);
176 }
177 
178 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
179 {
180 	struct scsi_disk *sdkp = NULL;
181 
182 	if (disk->private_data) {
183 		sdkp = scsi_disk(disk);
184 		if (scsi_device_get(sdkp->device) == 0)
185 			kref_get(&sdkp->kref);
186 		else
187 			sdkp = NULL;
188 	}
189 	return sdkp;
190 }
191 
192 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
193 {
194 	struct scsi_disk *sdkp;
195 
196 	down(&sd_ref_sem);
197 	sdkp = __scsi_disk_get(disk);
198 	up(&sd_ref_sem);
199 	return sdkp;
200 }
201 
202 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
203 {
204 	struct scsi_disk *sdkp;
205 
206 	down(&sd_ref_sem);
207 	sdkp = dev_get_drvdata(dev);
208 	if (sdkp)
209 		sdkp = __scsi_disk_get(sdkp->disk);
210 	up(&sd_ref_sem);
211 	return sdkp;
212 }
213 
214 static void scsi_disk_put(struct scsi_disk *sdkp)
215 {
216 	struct scsi_device *sdev = sdkp->device;
217 
218 	down(&sd_ref_sem);
219 	kref_put(&sdkp->kref, scsi_disk_release);
220 	scsi_device_put(sdev);
221 	up(&sd_ref_sem);
222 }
223 
224 /**
225  *	sd_init_command - build a scsi (read or write) command from
226  *	information in the request structure.
227  *	@SCpnt: pointer to mid-level's per scsi command structure that
228  *	contains request and into which the scsi command is written
229  *
230  *	Returns 1 if successful and 0 if error (or cannot be done now).
231  **/
232 static int sd_init_command(struct scsi_cmnd * SCpnt)
233 {
234 	unsigned int this_count, timeout;
235 	struct gendisk *disk;
236 	sector_t block;
237 	struct scsi_device *sdp = SCpnt->device;
238 	struct request *rq = SCpnt->request;
239 
240 	timeout = sdp->timeout;
241 
242 	/*
243 	 * SG_IO from block layer already setup, just copy cdb basically
244 	 */
245 	if (blk_pc_request(rq)) {
246 		scsi_setup_blk_pc_cmnd(SCpnt);
247 		if (rq->timeout)
248 			timeout = rq->timeout;
249 
250 		goto queue;
251 	}
252 
253 	/*
254 	 * we only do REQ_CMD and REQ_BLOCK_PC
255 	 */
256 	if (!blk_fs_request(rq))
257 		return 0;
258 
259 	disk = rq->rq_disk;
260 	block = rq->sector;
261 	this_count = SCpnt->request_bufflen >> 9;
262 
263 	SCSI_LOG_HLQUEUE(1, printk("sd_init_command: disk=%s, block=%llu, "
264 			    "count=%d\n", disk->disk_name,
265 			 (unsigned long long)block, this_count));
266 
267 	if (!sdp || !scsi_device_online(sdp) ||
268  	    block + rq->nr_sectors > get_capacity(disk)) {
269 		SCSI_LOG_HLQUEUE(2, printk("Finishing %ld sectors\n",
270 				 rq->nr_sectors));
271 		SCSI_LOG_HLQUEUE(2, printk("Retry with 0x%p\n", SCpnt));
272 		return 0;
273 	}
274 
275 	if (sdp->changed) {
276 		/*
277 		 * quietly refuse to do anything to a changed disc until
278 		 * the changed bit has been reset
279 		 */
280 		/* printk("SCSI disk has been changed. Prohibiting further I/O.\n"); */
281 		return 0;
282 	}
283 	SCSI_LOG_HLQUEUE(2, printk("%s : block=%llu\n",
284 				   disk->disk_name, (unsigned long long)block));
285 
286 	/*
287 	 * If we have a 1K hardware sectorsize, prevent access to single
288 	 * 512 byte sectors.  In theory we could handle this - in fact
289 	 * the scsi cdrom driver must be able to handle this because
290 	 * we typically use 1K blocksizes, and cdroms typically have
291 	 * 2K hardware sectorsizes.  Of course, things are simpler
292 	 * with the cdrom, since it is read-only.  For performance
293 	 * reasons, the filesystems should be able to handle this
294 	 * and not force the scsi disk driver to use bounce buffers
295 	 * for this.
296 	 */
297 	if (sdp->sector_size == 1024) {
298 		if ((block & 1) || (rq->nr_sectors & 1)) {
299 			printk(KERN_ERR "sd: Bad block number requested");
300 			return 0;
301 		} else {
302 			block = block >> 1;
303 			this_count = this_count >> 1;
304 		}
305 	}
306 	if (sdp->sector_size == 2048) {
307 		if ((block & 3) || (rq->nr_sectors & 3)) {
308 			printk(KERN_ERR "sd: Bad block number requested");
309 			return 0;
310 		} else {
311 			block = block >> 2;
312 			this_count = this_count >> 2;
313 		}
314 	}
315 	if (sdp->sector_size == 4096) {
316 		if ((block & 7) || (rq->nr_sectors & 7)) {
317 			printk(KERN_ERR "sd: Bad block number requested");
318 			return 0;
319 		} else {
320 			block = block >> 3;
321 			this_count = this_count >> 3;
322 		}
323 	}
324 	if (rq_data_dir(rq) == WRITE) {
325 		if (!sdp->writeable) {
326 			return 0;
327 		}
328 		SCpnt->cmnd[0] = WRITE_6;
329 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
330 	} else if (rq_data_dir(rq) == READ) {
331 		SCpnt->cmnd[0] = READ_6;
332 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
333 	} else {
334 		printk(KERN_ERR "sd: Unknown command %lx\n", rq->flags);
335 /* overkill 	panic("Unknown sd command %lx\n", rq->flags); */
336 		return 0;
337 	}
338 
339 	SCSI_LOG_HLQUEUE(2, printk("%s : %s %d/%ld 512 byte blocks.\n",
340 		disk->disk_name, (rq_data_dir(rq) == WRITE) ?
341 		"writing" : "reading", this_count, rq->nr_sectors));
342 
343 	SCpnt->cmnd[1] = 0;
344 
345 	if (block > 0xffffffff) {
346 		SCpnt->cmnd[0] += READ_16 - READ_6;
347 		SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
348 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
349 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
350 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
351 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
352 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
353 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
354 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
355 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
356 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
357 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
358 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
359 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
360 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
361 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
362 		   SCpnt->device->use_10_for_rw) {
363 		if (this_count > 0xffff)
364 			this_count = 0xffff;
365 
366 		SCpnt->cmnd[0] += READ_10 - READ_6;
367 		SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
368 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
369 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
370 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
371 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
372 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
373 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
374 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
375 	} else {
376 		if (unlikely(blk_fua_rq(rq))) {
377 			/*
378 			 * This happens only if this drive failed
379 			 * 10byte rw command with ILLEGAL_REQUEST
380 			 * during operation and thus turned off
381 			 * use_10_for_rw.
382 			 */
383 			printk(KERN_ERR "sd: FUA write on READ/WRITE(6) drive\n");
384 			return 0;
385 		}
386 
387 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
388 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
389 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
390 		SCpnt->cmnd[4] = (unsigned char) this_count;
391 		SCpnt->cmnd[5] = 0;
392 	}
393 	SCpnt->request_bufflen = SCpnt->bufflen =
394 			this_count * sdp->sector_size;
395 
396 	/*
397 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
398 	 * host adapter, it's safe to assume that we can at least transfer
399 	 * this many bytes between each connect / disconnect.
400 	 */
401 	SCpnt->transfersize = sdp->sector_size;
402 	SCpnt->underflow = this_count << 9;
403 	SCpnt->allowed = SD_MAX_RETRIES;
404 
405 queue:
406 	SCpnt->timeout_per_command = timeout;
407 
408 	/*
409 	 * This is the completion routine we use.  This is matched in terms
410 	 * of capability to this function.
411 	 */
412 	SCpnt->done = sd_rw_intr;
413 
414 	/*
415 	 * This indicates that the command is ready from our end to be
416 	 * queued.
417 	 */
418 	return 1;
419 }
420 
421 /**
422  *	sd_open - open a scsi disk device
423  *	@inode: only i_rdev member may be used
424  *	@filp: only f_mode and f_flags may be used
425  *
426  *	Returns 0 if successful. Returns a negated errno value in case
427  *	of error.
428  *
429  *	Note: This can be called from a user context (e.g. fsck(1) )
430  *	or from within the kernel (e.g. as a result of a mount(1) ).
431  *	In the latter case @inode and @filp carry an abridged amount
432  *	of information as noted above.
433  **/
434 static int sd_open(struct inode *inode, struct file *filp)
435 {
436 	struct gendisk *disk = inode->i_bdev->bd_disk;
437 	struct scsi_disk *sdkp;
438 	struct scsi_device *sdev;
439 	int retval;
440 
441 	if (!(sdkp = scsi_disk_get(disk)))
442 		return -ENXIO;
443 
444 
445 	SCSI_LOG_HLQUEUE(3, printk("sd_open: disk=%s\n", disk->disk_name));
446 
447 	sdev = sdkp->device;
448 
449 	/*
450 	 * If the device is in error recovery, wait until it is done.
451 	 * If the device is offline, then disallow any access to it.
452 	 */
453 	retval = -ENXIO;
454 	if (!scsi_block_when_processing_errors(sdev))
455 		goto error_out;
456 
457 	if (sdev->removable || sdkp->write_prot)
458 		check_disk_change(inode->i_bdev);
459 
460 	/*
461 	 * If the drive is empty, just let the open fail.
462 	 */
463 	retval = -ENOMEDIUM;
464 	if (sdev->removable && !sdkp->media_present &&
465 	    !(filp->f_flags & O_NDELAY))
466 		goto error_out;
467 
468 	/*
469 	 * If the device has the write protect tab set, have the open fail
470 	 * if the user expects to be able to write to the thing.
471 	 */
472 	retval = -EROFS;
473 	if (sdkp->write_prot && (filp->f_mode & FMODE_WRITE))
474 		goto error_out;
475 
476 	/*
477 	 * It is possible that the disk changing stuff resulted in
478 	 * the device being taken offline.  If this is the case,
479 	 * report this to the user, and don't pretend that the
480 	 * open actually succeeded.
481 	 */
482 	retval = -ENXIO;
483 	if (!scsi_device_online(sdev))
484 		goto error_out;
485 
486 	if (!sdkp->openers++ && sdev->removable) {
487 		if (scsi_block_when_processing_errors(sdev))
488 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
489 	}
490 
491 	return 0;
492 
493 error_out:
494 	scsi_disk_put(sdkp);
495 	return retval;
496 }
497 
498 /**
499  *	sd_release - invoked when the (last) close(2) is called on this
500  *	scsi disk.
501  *	@inode: only i_rdev member may be used
502  *	@filp: only f_mode and f_flags may be used
503  *
504  *	Returns 0.
505  *
506  *	Note: may block (uninterruptible) if error recovery is underway
507  *	on this disk.
508  **/
509 static int sd_release(struct inode *inode, struct file *filp)
510 {
511 	struct gendisk *disk = inode->i_bdev->bd_disk;
512 	struct scsi_disk *sdkp = scsi_disk(disk);
513 	struct scsi_device *sdev = sdkp->device;
514 
515 	SCSI_LOG_HLQUEUE(3, printk("sd_release: disk=%s\n", disk->disk_name));
516 
517 	if (!--sdkp->openers && sdev->removable) {
518 		if (scsi_block_when_processing_errors(sdev))
519 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
520 	}
521 
522 	/*
523 	 * XXX and what if there are packets in flight and this close()
524 	 * XXX is followed by a "rmmod sd_mod"?
525 	 */
526 	scsi_disk_put(sdkp);
527 	return 0;
528 }
529 
530 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
531 {
532 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
533 	struct scsi_device *sdp = sdkp->device;
534 	struct Scsi_Host *host = sdp->host;
535 	int diskinfo[4];
536 
537 	/* default to most commonly used values */
538         diskinfo[0] = 0x40;	/* 1 << 6 */
539        	diskinfo[1] = 0x20;	/* 1 << 5 */
540        	diskinfo[2] = sdkp->capacity >> 11;
541 
542 	/* override with calculated, extended default, or driver values */
543 	if (host->hostt->bios_param)
544 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
545 	else
546 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
547 
548 	geo->heads = diskinfo[0];
549 	geo->sectors = diskinfo[1];
550 	geo->cylinders = diskinfo[2];
551 	return 0;
552 }
553 
554 /**
555  *	sd_ioctl - process an ioctl
556  *	@inode: only i_rdev/i_bdev members may be used
557  *	@filp: only f_mode and f_flags may be used
558  *	@cmd: ioctl command number
559  *	@arg: this is third argument given to ioctl(2) system call.
560  *	Often contains a pointer.
561  *
562  *	Returns 0 if successful (some ioctls return postive numbers on
563  *	success as well). Returns a negated errno value in case of error.
564  *
565  *	Note: most ioctls are forward onto the block subsystem or further
566  *	down in the scsi subsytem.
567  **/
568 static int sd_ioctl(struct inode * inode, struct file * filp,
569 		    unsigned int cmd, unsigned long arg)
570 {
571 	struct block_device *bdev = inode->i_bdev;
572 	struct gendisk *disk = bdev->bd_disk;
573 	struct scsi_device *sdp = scsi_disk(disk)->device;
574 	void __user *p = (void __user *)arg;
575 	int error;
576 
577 	SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
578 						disk->disk_name, cmd));
579 
580 	/*
581 	 * If we are in the middle of error recovery, don't let anyone
582 	 * else try and use this device.  Also, if error recovery fails, it
583 	 * may try and take the device offline, in which case all further
584 	 * access to the device is prohibited.
585 	 */
586 	error = scsi_nonblockable_ioctl(sdp, cmd, p, filp);
587 	if (!scsi_block_when_processing_errors(sdp) || !error)
588 		return error;
589 
590 	/*
591 	 * Send SCSI addressing ioctls directly to mid level, send other
592 	 * ioctls to block level and then onto mid level if they can't be
593 	 * resolved.
594 	 */
595 	switch (cmd) {
596 		case SCSI_IOCTL_GET_IDLUN:
597 		case SCSI_IOCTL_GET_BUS_NUMBER:
598 			return scsi_ioctl(sdp, cmd, p);
599 		default:
600 			error = scsi_cmd_ioctl(filp, disk, cmd, p);
601 			if (error != -ENOTTY)
602 				return error;
603 	}
604 	return scsi_ioctl(sdp, cmd, p);
605 }
606 
607 static void set_media_not_present(struct scsi_disk *sdkp)
608 {
609 	sdkp->media_present = 0;
610 	sdkp->capacity = 0;
611 	sdkp->device->changed = 1;
612 }
613 
614 /**
615  *	sd_media_changed - check if our medium changed
616  *	@disk: kernel device descriptor
617  *
618  *	Returns 0 if not applicable or no change; 1 if change
619  *
620  *	Note: this function is invoked from the block subsystem.
621  **/
622 static int sd_media_changed(struct gendisk *disk)
623 {
624 	struct scsi_disk *sdkp = scsi_disk(disk);
625 	struct scsi_device *sdp = sdkp->device;
626 	int retval;
627 
628 	SCSI_LOG_HLQUEUE(3, printk("sd_media_changed: disk=%s\n",
629 						disk->disk_name));
630 
631 	if (!sdp->removable)
632 		return 0;
633 
634 	/*
635 	 * If the device is offline, don't send any commands - just pretend as
636 	 * if the command failed.  If the device ever comes back online, we
637 	 * can deal with it then.  It is only because of unrecoverable errors
638 	 * that we would ever take a device offline in the first place.
639 	 */
640 	if (!scsi_device_online(sdp))
641 		goto not_present;
642 
643 	/*
644 	 * Using TEST_UNIT_READY enables differentiation between drive with
645 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
646 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
647 	 *
648 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
649 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
650 	 * sd_revalidate() is called.
651 	 */
652 	retval = -ENODEV;
653 	if (scsi_block_when_processing_errors(sdp))
654 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES);
655 
656 	/*
657 	 * Unable to test, unit probably not ready.   This usually
658 	 * means there is no disc in the drive.  Mark as changed,
659 	 * and we will figure it out later once the drive is
660 	 * available again.
661 	 */
662 	if (retval)
663 		 goto not_present;
664 
665 	/*
666 	 * For removable scsi disk we have to recognise the presence
667 	 * of a disk in the drive. This is kept in the struct scsi_disk
668 	 * struct and tested at open !  Daniel Roche (dan@lectra.fr)
669 	 */
670 	sdkp->media_present = 1;
671 
672 	retval = sdp->changed;
673 	sdp->changed = 0;
674 
675 	return retval;
676 
677 not_present:
678 	set_media_not_present(sdkp);
679 	return 1;
680 }
681 
682 static int sd_sync_cache(struct scsi_device *sdp)
683 {
684 	int retries, res;
685 	struct scsi_sense_hdr sshdr;
686 
687 	if (!scsi_device_online(sdp))
688 		return -ENODEV;
689 
690 
691 	for (retries = 3; retries > 0; --retries) {
692 		unsigned char cmd[10] = { 0 };
693 
694 		cmd[0] = SYNCHRONIZE_CACHE;
695 		/*
696 		 * Leave the rest of the command zero to indicate
697 		 * flush everything.
698 		 */
699 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
700 				       SD_TIMEOUT, SD_MAX_RETRIES);
701 		if (res == 0)
702 			break;
703 	}
704 
705 	if (res) {		printk(KERN_WARNING "FAILED\n  status = %x, message = %02x, "
706 				    "host = %d, driver = %02x\n  ",
707 				    status_byte(res), msg_byte(res),
708 				    host_byte(res), driver_byte(res));
709 			if (driver_byte(res) & DRIVER_SENSE)
710 				scsi_print_sense_hdr("sd", &sshdr);
711 	}
712 
713 	return res;
714 }
715 
716 static int sd_issue_flush(struct device *dev, sector_t *error_sector)
717 {
718 	int ret = 0;
719 	struct scsi_device *sdp = to_scsi_device(dev);
720 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
721 
722 	if (!sdkp)
723                return -ENODEV;
724 
725 	if (sdkp->WCE)
726 		ret = sd_sync_cache(sdp);
727 	scsi_disk_put(sdkp);
728 	return ret;
729 }
730 
731 static void sd_prepare_flush(request_queue_t *q, struct request *rq)
732 {
733 	memset(rq->cmd, 0, sizeof(rq->cmd));
734 	rq->flags |= REQ_BLOCK_PC;
735 	rq->timeout = SD_TIMEOUT;
736 	rq->cmd[0] = SYNCHRONIZE_CACHE;
737 	rq->cmd_len = 10;
738 }
739 
740 static void sd_rescan(struct device *dev)
741 {
742 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
743 
744 	if (sdkp) {
745 		sd_revalidate_disk(sdkp->disk);
746 		scsi_disk_put(sdkp);
747 	}
748 }
749 
750 
751 #ifdef CONFIG_COMPAT
752 /*
753  * This gets directly called from VFS. When the ioctl
754  * is not recognized we go back to the other translation paths.
755  */
756 static long sd_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
757 {
758 	struct block_device *bdev = file->f_dentry->d_inode->i_bdev;
759 	struct gendisk *disk = bdev->bd_disk;
760 	struct scsi_device *sdev = scsi_disk(disk)->device;
761 
762 	/*
763 	 * If we are in the middle of error recovery, don't let anyone
764 	 * else try and use this device.  Also, if error recovery fails, it
765 	 * may try and take the device offline, in which case all further
766 	 * access to the device is prohibited.
767 	 */
768 	if (!scsi_block_when_processing_errors(sdev))
769 		return -ENODEV;
770 
771 	if (sdev->host->hostt->compat_ioctl) {
772 		int ret;
773 
774 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
775 
776 		return ret;
777 	}
778 
779 	/*
780 	 * Let the static ioctl translation table take care of it.
781 	 */
782 	return -ENOIOCTLCMD;
783 }
784 #endif
785 
786 static struct block_device_operations sd_fops = {
787 	.owner			= THIS_MODULE,
788 	.open			= sd_open,
789 	.release		= sd_release,
790 	.ioctl			= sd_ioctl,
791 	.getgeo			= sd_getgeo,
792 #ifdef CONFIG_COMPAT
793 	.compat_ioctl		= sd_compat_ioctl,
794 #endif
795 	.media_changed		= sd_media_changed,
796 	.revalidate_disk	= sd_revalidate_disk,
797 };
798 
799 /**
800  *	sd_rw_intr - bottom half handler: called when the lower level
801  *	driver has completed (successfully or otherwise) a scsi command.
802  *	@SCpnt: mid-level's per command structure.
803  *
804  *	Note: potentially run from within an ISR. Must not block.
805  **/
806 static void sd_rw_intr(struct scsi_cmnd * SCpnt)
807 {
808 	int result = SCpnt->result;
809 	int this_count = SCpnt->bufflen;
810 	int good_bytes = (result == 0 ? this_count : 0);
811 	sector_t block_sectors = 1;
812 	u64 first_err_block;
813 	sector_t error_sector;
814 	struct scsi_sense_hdr sshdr;
815 	int sense_valid = 0;
816 	int sense_deferred = 0;
817 	int info_valid;
818 
819 	if (result) {
820 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
821 		if (sense_valid)
822 			sense_deferred = scsi_sense_is_deferred(&sshdr);
823 	}
824 
825 #ifdef CONFIG_SCSI_LOGGING
826 	SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: %s: res=0x%x\n",
827 				SCpnt->request->rq_disk->disk_name, result));
828 	if (sense_valid) {
829 		SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: sb[respc,sk,asc,"
830 				"ascq]=%x,%x,%x,%x\n", sshdr.response_code,
831 				sshdr.sense_key, sshdr.asc, sshdr.ascq));
832 	}
833 #endif
834 	/*
835 	   Handle MEDIUM ERRORs that indicate partial success.  Since this is a
836 	   relatively rare error condition, no care is taken to avoid
837 	   unnecessary additional work such as memcpy's that could be avoided.
838 	 */
839 
840 	/*
841 	 * If SG_IO from block layer then set good_bytes to stop retries;
842 	 * else if errors, check them, and if necessary prepare for
843 	 * (partial) retries.
844 	 */
845 	if (blk_pc_request(SCpnt->request))
846 		good_bytes = this_count;
847 	else if (driver_byte(result) != 0 &&
848 		 sense_valid && !sense_deferred) {
849 		switch (sshdr.sense_key) {
850 		case MEDIUM_ERROR:
851 			if (!blk_fs_request(SCpnt->request))
852 				break;
853 			info_valid = scsi_get_sense_info_fld(
854 				SCpnt->sense_buffer, SCSI_SENSE_BUFFERSIZE,
855 				&first_err_block);
856 			/*
857 			 * May want to warn and skip if following cast results
858 			 * in actual truncation (if sector_t < 64 bits)
859 			 */
860 			error_sector = (sector_t)first_err_block;
861 			if (SCpnt->request->bio != NULL)
862 				block_sectors = bio_sectors(SCpnt->request->bio);
863 			switch (SCpnt->device->sector_size) {
864 			case 1024:
865 				error_sector <<= 1;
866 				if (block_sectors < 2)
867 					block_sectors = 2;
868 				break;
869 			case 2048:
870 				error_sector <<= 2;
871 				if (block_sectors < 4)
872 					block_sectors = 4;
873 				break;
874 			case 4096:
875 				error_sector <<=3;
876 				if (block_sectors < 8)
877 					block_sectors = 8;
878 				break;
879 			case 256:
880 				error_sector >>= 1;
881 				break;
882 			default:
883 				break;
884 			}
885 
886 			error_sector &= ~(block_sectors - 1);
887 			good_bytes = (error_sector - SCpnt->request->sector) << 9;
888 			if (good_bytes < 0 || good_bytes >= this_count)
889 				good_bytes = 0;
890 			break;
891 
892 		case RECOVERED_ERROR: /* an error occurred, but it recovered */
893 		case NO_SENSE: /* LLDD got sense data */
894 			/*
895 			 * Inform the user, but make sure that it's not treated
896 			 * as a hard error.
897 			 */
898 			scsi_print_sense("sd", SCpnt);
899 			SCpnt->result = 0;
900 			memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
901 			good_bytes = this_count;
902 			break;
903 
904 		case ILLEGAL_REQUEST:
905 			if (SCpnt->device->use_10_for_rw &&
906 			    (SCpnt->cmnd[0] == READ_10 ||
907 			     SCpnt->cmnd[0] == WRITE_10))
908 				SCpnt->device->use_10_for_rw = 0;
909 			if (SCpnt->device->use_10_for_ms &&
910 			    (SCpnt->cmnd[0] == MODE_SENSE_10 ||
911 			     SCpnt->cmnd[0] == MODE_SELECT_10))
912 				SCpnt->device->use_10_for_ms = 0;
913 			break;
914 
915 		default:
916 			break;
917 		}
918 	}
919 	/*
920 	 * This calls the generic completion function, now that we know
921 	 * how many actual sectors finished, and how many sectors we need
922 	 * to say have failed.
923 	 */
924 	scsi_io_completion(SCpnt, good_bytes, block_sectors << 9);
925 }
926 
927 static int media_not_present(struct scsi_disk *sdkp,
928 			     struct scsi_sense_hdr *sshdr)
929 {
930 
931 	if (!scsi_sense_valid(sshdr))
932 		return 0;
933 	/* not invoked for commands that could return deferred errors */
934 	if (sshdr->sense_key != NOT_READY &&
935 	    sshdr->sense_key != UNIT_ATTENTION)
936 		return 0;
937 	if (sshdr->asc != 0x3A) /* medium not present */
938 		return 0;
939 
940 	set_media_not_present(sdkp);
941 	return 1;
942 }
943 
944 /*
945  * spinup disk - called only in sd_revalidate_disk()
946  */
947 static void
948 sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
949 {
950 	unsigned char cmd[10];
951 	unsigned long spintime_expire = 0;
952 	int retries, spintime;
953 	unsigned int the_result;
954 	struct scsi_sense_hdr sshdr;
955 	int sense_valid = 0;
956 
957 	spintime = 0;
958 
959 	/* Spin up drives, as required.  Only do this at boot time */
960 	/* Spinup needs to be done for module loads too. */
961 	do {
962 		retries = 0;
963 
964 		do {
965 			cmd[0] = TEST_UNIT_READY;
966 			memset((void *) &cmd[1], 0, 9);
967 
968 			the_result = scsi_execute_req(sdkp->device, cmd,
969 						      DMA_NONE, NULL, 0,
970 						      &sshdr, SD_TIMEOUT,
971 						      SD_MAX_RETRIES);
972 
973 			if (the_result)
974 				sense_valid = scsi_sense_valid(&sshdr);
975 			retries++;
976 		} while (retries < 3 &&
977 			 (!scsi_status_is_good(the_result) ||
978 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
979 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
980 
981 		/*
982 		 * If the drive has indicated to us that it doesn't have
983 		 * any media in it, don't bother with any of the rest of
984 		 * this crap.
985 		 */
986 		if (media_not_present(sdkp, &sshdr))
987 			return;
988 
989 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
990 			/* no sense, TUR either succeeded or failed
991 			 * with a status error */
992 			if(!spintime && !scsi_status_is_good(the_result))
993 				printk(KERN_NOTICE "%s: Unit Not Ready, "
994 				       "error = 0x%x\n", diskname, the_result);
995 			break;
996 		}
997 
998 		/*
999 		 * The device does not want the automatic start to be issued.
1000 		 */
1001 		if (sdkp->device->no_start_on_add) {
1002 			break;
1003 		}
1004 
1005 		/*
1006 		 * If manual intervention is required, or this is an
1007 		 * absent USB storage device, a spinup is meaningless.
1008 		 */
1009 		if (sense_valid &&
1010 		    sshdr.sense_key == NOT_READY &&
1011 		    sshdr.asc == 4 && sshdr.ascq == 3) {
1012 			break;		/* manual intervention required */
1013 
1014 		/*
1015 		 * Issue command to spin up drive when not ready
1016 		 */
1017 		} else if (sense_valid && sshdr.sense_key == NOT_READY) {
1018 			if (!spintime) {
1019 				printk(KERN_NOTICE "%s: Spinning up disk...",
1020 				       diskname);
1021 				cmd[0] = START_STOP;
1022 				cmd[1] = 1;	/* Return immediately */
1023 				memset((void *) &cmd[2], 0, 8);
1024 				cmd[4] = 1;	/* Start spin cycle */
1025 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1026 						 NULL, 0, &sshdr,
1027 						 SD_TIMEOUT, SD_MAX_RETRIES);
1028 				spintime_expire = jiffies + 100 * HZ;
1029 				spintime = 1;
1030 			}
1031 			/* Wait 1 second for next try */
1032 			msleep(1000);
1033 			printk(".");
1034 
1035 		/*
1036 		 * Wait for USB flash devices with slow firmware.
1037 		 * Yes, this sense key/ASC combination shouldn't
1038 		 * occur here.  It's characteristic of these devices.
1039 		 */
1040 		} else if (sense_valid &&
1041 				sshdr.sense_key == UNIT_ATTENTION &&
1042 				sshdr.asc == 0x28) {
1043 			if (!spintime) {
1044 				spintime_expire = jiffies + 5 * HZ;
1045 				spintime = 1;
1046 			}
1047 			/* Wait 1 second for next try */
1048 			msleep(1000);
1049 		} else {
1050 			/* we don't understand the sense code, so it's
1051 			 * probably pointless to loop */
1052 			if(!spintime) {
1053 				printk(KERN_NOTICE "%s: Unit Not Ready, "
1054 					"sense:\n", diskname);
1055 				scsi_print_sense_hdr("", &sshdr);
1056 			}
1057 			break;
1058 		}
1059 
1060 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1061 
1062 	if (spintime) {
1063 		if (scsi_status_is_good(the_result))
1064 			printk("ready\n");
1065 		else
1066 			printk("not responding...\n");
1067 	}
1068 }
1069 
1070 /*
1071  * read disk capacity
1072  */
1073 static void
1074 sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
1075 		 unsigned char *buffer)
1076 {
1077 	unsigned char cmd[16];
1078 	int the_result, retries;
1079 	int sector_size = 0;
1080 	int longrc = 0;
1081 	struct scsi_sense_hdr sshdr;
1082 	int sense_valid = 0;
1083 	struct scsi_device *sdp = sdkp->device;
1084 
1085 repeat:
1086 	retries = 3;
1087 	do {
1088 		if (longrc) {
1089 			memset((void *) cmd, 0, 16);
1090 			cmd[0] = SERVICE_ACTION_IN;
1091 			cmd[1] = SAI_READ_CAPACITY_16;
1092 			cmd[13] = 12;
1093 			memset((void *) buffer, 0, 12);
1094 		} else {
1095 			cmd[0] = READ_CAPACITY;
1096 			memset((void *) &cmd[1], 0, 9);
1097 			memset((void *) buffer, 0, 8);
1098 		}
1099 
1100 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1101 					      buffer, longrc ? 12 : 8, &sshdr,
1102 					      SD_TIMEOUT, SD_MAX_RETRIES);
1103 
1104 		if (media_not_present(sdkp, &sshdr))
1105 			return;
1106 
1107 		if (the_result)
1108 			sense_valid = scsi_sense_valid(&sshdr);
1109 		retries--;
1110 
1111 	} while (the_result && retries);
1112 
1113 	if (the_result && !longrc) {
1114 		printk(KERN_NOTICE "%s : READ CAPACITY failed.\n"
1115 		       "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
1116 		       diskname, diskname,
1117 		       status_byte(the_result),
1118 		       msg_byte(the_result),
1119 		       host_byte(the_result),
1120 		       driver_byte(the_result));
1121 
1122 		if (driver_byte(the_result) & DRIVER_SENSE)
1123 			scsi_print_sense_hdr("sd", &sshdr);
1124 		else
1125 			printk("%s : sense not available. \n", diskname);
1126 
1127 		/* Set dirty bit for removable devices if not ready -
1128 		 * sometimes drives will not report this properly. */
1129 		if (sdp->removable &&
1130 		    sense_valid && sshdr.sense_key == NOT_READY)
1131 			sdp->changed = 1;
1132 
1133 		/* Either no media are present but the drive didn't tell us,
1134 		   or they are present but the read capacity command fails */
1135 		/* sdkp->media_present = 0; -- not always correct */
1136 		sdkp->capacity = 0x200000; /* 1 GB - random */
1137 
1138 		return;
1139 	} else if (the_result && longrc) {
1140 		/* READ CAPACITY(16) has been failed */
1141 		printk(KERN_NOTICE "%s : READ CAPACITY(16) failed.\n"
1142 		       "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
1143 		       diskname, diskname,
1144 		       status_byte(the_result),
1145 		       msg_byte(the_result),
1146 		       host_byte(the_result),
1147 		       driver_byte(the_result));
1148 		printk(KERN_NOTICE "%s : use 0xffffffff as device size\n",
1149 		       diskname);
1150 
1151 		sdkp->capacity = 1 + (sector_t) 0xffffffff;
1152 		goto got_data;
1153 	}
1154 
1155 	if (!longrc) {
1156 		sector_size = (buffer[4] << 24) |
1157 			(buffer[5] << 16) | (buffer[6] << 8) | buffer[7];
1158 		if (buffer[0] == 0xff && buffer[1] == 0xff &&
1159 		    buffer[2] == 0xff && buffer[3] == 0xff) {
1160 			if(sizeof(sdkp->capacity) > 4) {
1161 				printk(KERN_NOTICE "%s : very big device. try to use"
1162 				       " READ CAPACITY(16).\n", diskname);
1163 				longrc = 1;
1164 				goto repeat;
1165 			}
1166 			printk(KERN_ERR "%s: too big for this kernel.  Use a "
1167 			       "kernel compiled with support for large block "
1168 			       "devices.\n", diskname);
1169 			sdkp->capacity = 0;
1170 			goto got_data;
1171 		}
1172 		sdkp->capacity = 1 + (((sector_t)buffer[0] << 24) |
1173 			(buffer[1] << 16) |
1174 			(buffer[2] << 8) |
1175 			buffer[3]);
1176 	} else {
1177 		sdkp->capacity = 1 + (((u64)buffer[0] << 56) |
1178 			((u64)buffer[1] << 48) |
1179 			((u64)buffer[2] << 40) |
1180 			((u64)buffer[3] << 32) |
1181 			((sector_t)buffer[4] << 24) |
1182 			((sector_t)buffer[5] << 16) |
1183 			((sector_t)buffer[6] << 8)  |
1184 			(sector_t)buffer[7]);
1185 
1186 		sector_size = (buffer[8] << 24) |
1187 			(buffer[9] << 16) | (buffer[10] << 8) | buffer[11];
1188 	}
1189 
1190 	/* Some devices return the total number of sectors, not the
1191 	 * highest sector number.  Make the necessary adjustment. */
1192 	if (sdp->fix_capacity)
1193 		--sdkp->capacity;
1194 
1195 got_data:
1196 	if (sector_size == 0) {
1197 		sector_size = 512;
1198 		printk(KERN_NOTICE "%s : sector size 0 reported, "
1199 		       "assuming 512.\n", diskname);
1200 	}
1201 
1202 	if (sector_size != 512 &&
1203 	    sector_size != 1024 &&
1204 	    sector_size != 2048 &&
1205 	    sector_size != 4096 &&
1206 	    sector_size != 256) {
1207 		printk(KERN_NOTICE "%s : unsupported sector size "
1208 		       "%d.\n", diskname, sector_size);
1209 		/*
1210 		 * The user might want to re-format the drive with
1211 		 * a supported sectorsize.  Once this happens, it
1212 		 * would be relatively trivial to set the thing up.
1213 		 * For this reason, we leave the thing in the table.
1214 		 */
1215 		sdkp->capacity = 0;
1216 		/*
1217 		 * set a bogus sector size so the normal read/write
1218 		 * logic in the block layer will eventually refuse any
1219 		 * request on this device without tripping over power
1220 		 * of two sector size assumptions
1221 		 */
1222 		sector_size = 512;
1223 	}
1224 	{
1225 		/*
1226 		 * The msdos fs needs to know the hardware sector size
1227 		 * So I have created this table. See ll_rw_blk.c
1228 		 * Jacques Gelinas (Jacques@solucorp.qc.ca)
1229 		 */
1230 		int hard_sector = sector_size;
1231 		sector_t sz = (sdkp->capacity/2) * (hard_sector/256);
1232 		request_queue_t *queue = sdp->request_queue;
1233 		sector_t mb = sz;
1234 
1235 		blk_queue_hardsect_size(queue, hard_sector);
1236 		/* avoid 64-bit division on 32-bit platforms */
1237 		sector_div(sz, 625);
1238 		mb -= sz - 974;
1239 		sector_div(mb, 1950);
1240 
1241 		printk(KERN_NOTICE "SCSI device %s: "
1242 		       "%llu %d-byte hdwr sectors (%llu MB)\n",
1243 		       diskname, (unsigned long long)sdkp->capacity,
1244 		       hard_sector, (unsigned long long)mb);
1245 	}
1246 
1247 	/* Rescale capacity to 512-byte units */
1248 	if (sector_size == 4096)
1249 		sdkp->capacity <<= 3;
1250 	else if (sector_size == 2048)
1251 		sdkp->capacity <<= 2;
1252 	else if (sector_size == 1024)
1253 		sdkp->capacity <<= 1;
1254 	else if (sector_size == 256)
1255 		sdkp->capacity >>= 1;
1256 
1257 	sdkp->device->sector_size = sector_size;
1258 }
1259 
1260 /* called with buffer of length 512 */
1261 static inline int
1262 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1263 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
1264 		 struct scsi_sense_hdr *sshdr)
1265 {
1266 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1267 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
1268 			       sshdr);
1269 }
1270 
1271 /*
1272  * read write protect setting, if possible - called only in sd_revalidate_disk()
1273  * called with buffer of length 512
1274  */
1275 static void
1276 sd_read_write_protect_flag(struct scsi_disk *sdkp, char *diskname,
1277 			   unsigned char *buffer)
1278 {
1279 	int res;
1280 	struct scsi_device *sdp = sdkp->device;
1281 	struct scsi_mode_data data;
1282 
1283 	set_disk_ro(sdkp->disk, 0);
1284 	if (sdp->skip_ms_page_3f) {
1285 		printk(KERN_NOTICE "%s: assuming Write Enabled\n", diskname);
1286 		return;
1287 	}
1288 
1289 	if (sdp->use_192_bytes_for_3f) {
1290 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1291 	} else {
1292 		/*
1293 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1294 		 * We have to start carefully: some devices hang if we ask
1295 		 * for more than is available.
1296 		 */
1297 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1298 
1299 		/*
1300 		 * Second attempt: ask for page 0 When only page 0 is
1301 		 * implemented, a request for page 3F may return Sense Key
1302 		 * 5: Illegal Request, Sense Code 24: Invalid field in
1303 		 * CDB.
1304 		 */
1305 		if (!scsi_status_is_good(res))
1306 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1307 
1308 		/*
1309 		 * Third attempt: ask 255 bytes, as we did earlier.
1310 		 */
1311 		if (!scsi_status_is_good(res))
1312 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1313 					       &data, NULL);
1314 	}
1315 
1316 	if (!scsi_status_is_good(res)) {
1317 		printk(KERN_WARNING
1318 		       "%s: test WP failed, assume Write Enabled\n", diskname);
1319 	} else {
1320 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
1321 		set_disk_ro(sdkp->disk, sdkp->write_prot);
1322 		printk(KERN_NOTICE "%s: Write Protect is %s\n", diskname,
1323 		       sdkp->write_prot ? "on" : "off");
1324 		printk(KERN_DEBUG "%s: Mode Sense: %02x %02x %02x %02x\n",
1325 		       diskname, buffer[0], buffer[1], buffer[2], buffer[3]);
1326 	}
1327 }
1328 
1329 /*
1330  * sd_read_cache_type - called only from sd_revalidate_disk()
1331  * called with buffer of length 512
1332  */
1333 static void
1334 sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
1335 		   unsigned char *buffer)
1336 {
1337 	int len = 0, res;
1338 	struct scsi_device *sdp = sdkp->device;
1339 
1340 	int dbd;
1341 	int modepage;
1342 	struct scsi_mode_data data;
1343 	struct scsi_sense_hdr sshdr;
1344 
1345 	if (sdp->skip_ms_page_8)
1346 		goto defaults;
1347 
1348 	if (sdp->type == TYPE_RBC) {
1349 		modepage = 6;
1350 		dbd = 8;
1351 	} else {
1352 		modepage = 8;
1353 		dbd = 0;
1354 	}
1355 
1356 	/* cautiously ask */
1357 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, 4, &data, &sshdr);
1358 
1359 	if (!scsi_status_is_good(res))
1360 		goto bad_sense;
1361 
1362 	/* that went OK, now ask for the proper length */
1363 	len = data.length;
1364 
1365 	/*
1366 	 * We're only interested in the first three bytes, actually.
1367 	 * But the data cache page is defined for the first 20.
1368 	 */
1369 	if (len < 3)
1370 		goto bad_sense;
1371 	if (len > 20)
1372 		len = 20;
1373 
1374 	/* Take headers and block descriptors into account */
1375 	len += data.header_length + data.block_descriptor_length;
1376 
1377 	/* Get the data */
1378 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
1379 
1380 	if (scsi_status_is_good(res)) {
1381 		const char *types[] = {
1382 			"write through", "none", "write back",
1383 			"write back, no read (daft)"
1384 		};
1385 		int ct = 0;
1386 		int offset = data.header_length + data.block_descriptor_length;
1387 
1388 		if ((buffer[offset] & 0x3f) != modepage) {
1389 			printk(KERN_ERR "%s: got wrong page\n", diskname);
1390 			goto defaults;
1391 		}
1392 
1393 		if (modepage == 8) {
1394 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
1395 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
1396 		} else {
1397 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
1398 			sdkp->RCD = 0;
1399 		}
1400 
1401 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
1402 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
1403 			printk(KERN_NOTICE "SCSI device %s: uses "
1404 			       "READ/WRITE(6), disabling FUA\n", diskname);
1405 			sdkp->DPOFUA = 0;
1406 		}
1407 
1408 		ct =  sdkp->RCD + 2*sdkp->WCE;
1409 
1410 		printk(KERN_NOTICE "SCSI device %s: drive cache: %s%s\n",
1411 		       diskname, types[ct],
1412 		       sdkp->DPOFUA ? " w/ FUA" : "");
1413 
1414 		return;
1415 	}
1416 
1417 bad_sense:
1418 	if (scsi_sense_valid(&sshdr) &&
1419 	    sshdr.sense_key == ILLEGAL_REQUEST &&
1420 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
1421 		printk(KERN_NOTICE "%s: cache data unavailable\n",
1422 		       diskname);	/* Invalid field in CDB */
1423 	else
1424 		printk(KERN_ERR "%s: asking for cache data failed\n",
1425 		       diskname);
1426 
1427 defaults:
1428 	printk(KERN_ERR "%s: assuming drive cache: write through\n",
1429 	       diskname);
1430 	sdkp->WCE = 0;
1431 	sdkp->RCD = 0;
1432 }
1433 
1434 /**
1435  *	sd_revalidate_disk - called the first time a new disk is seen,
1436  *	performs disk spin up, read_capacity, etc.
1437  *	@disk: struct gendisk we care about
1438  **/
1439 static int sd_revalidate_disk(struct gendisk *disk)
1440 {
1441 	struct scsi_disk *sdkp = scsi_disk(disk);
1442 	struct scsi_device *sdp = sdkp->device;
1443 	unsigned char *buffer;
1444 	unsigned ordered;
1445 
1446 	SCSI_LOG_HLQUEUE(3, printk("sd_revalidate_disk: disk=%s\n", disk->disk_name));
1447 
1448 	/*
1449 	 * If the device is offline, don't try and read capacity or any
1450 	 * of the other niceties.
1451 	 */
1452 	if (!scsi_device_online(sdp))
1453 		goto out;
1454 
1455 	buffer = kmalloc(512, GFP_KERNEL | __GFP_DMA);
1456 	if (!buffer) {
1457 		printk(KERN_WARNING "(sd_revalidate_disk:) Memory allocation "
1458 		       "failure.\n");
1459 		goto out;
1460 	}
1461 
1462 	/* defaults, until the device tells us otherwise */
1463 	sdp->sector_size = 512;
1464 	sdkp->capacity = 0;
1465 	sdkp->media_present = 1;
1466 	sdkp->write_prot = 0;
1467 	sdkp->WCE = 0;
1468 	sdkp->RCD = 0;
1469 
1470 	sd_spinup_disk(sdkp, disk->disk_name);
1471 
1472 	/*
1473 	 * Without media there is no reason to ask; moreover, some devices
1474 	 * react badly if we do.
1475 	 */
1476 	if (sdkp->media_present) {
1477 		sd_read_capacity(sdkp, disk->disk_name, buffer);
1478 		sd_read_write_protect_flag(sdkp, disk->disk_name, buffer);
1479 		sd_read_cache_type(sdkp, disk->disk_name, buffer);
1480 	}
1481 
1482 	/*
1483 	 * We now have all cache related info, determine how we deal
1484 	 * with ordered requests.  Note that as the current SCSI
1485 	 * dispatch function can alter request order, we cannot use
1486 	 * QUEUE_ORDERED_TAG_* even when ordered tag is supported.
1487 	 */
1488 	if (sdkp->WCE)
1489 		ordered = sdkp->DPOFUA
1490 			? QUEUE_ORDERED_DRAIN_FUA : QUEUE_ORDERED_DRAIN_FLUSH;
1491 	else
1492 		ordered = QUEUE_ORDERED_DRAIN;
1493 
1494 	blk_queue_ordered(sdkp->disk->queue, ordered, sd_prepare_flush);
1495 
1496 	set_capacity(disk, sdkp->capacity);
1497 	kfree(buffer);
1498 
1499  out:
1500 	return 0;
1501 }
1502 
1503 /**
1504  *	sd_probe - called during driver initialization and whenever a
1505  *	new scsi device is attached to the system. It is called once
1506  *	for each scsi device (not just disks) present.
1507  *	@dev: pointer to device object
1508  *
1509  *	Returns 0 if successful (or not interested in this scsi device
1510  *	(e.g. scanner)); 1 when there is an error.
1511  *
1512  *	Note: this function is invoked from the scsi mid-level.
1513  *	This function sets up the mapping between a given
1514  *	<host,channel,id,lun> (found in sdp) and new device name
1515  *	(e.g. /dev/sda). More precisely it is the block device major
1516  *	and minor number that is chosen here.
1517  *
1518  *	Assume sd_attach is not re-entrant (for time being)
1519  *	Also think about sd_attach() and sd_remove() running coincidentally.
1520  **/
1521 static int sd_probe(struct device *dev)
1522 {
1523 	struct scsi_device *sdp = to_scsi_device(dev);
1524 	struct scsi_disk *sdkp;
1525 	struct gendisk *gd;
1526 	u32 index;
1527 	int error;
1528 
1529 	error = -ENODEV;
1530 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
1531 		goto out;
1532 
1533 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
1534 					"sd_attach\n"));
1535 
1536 	error = -ENOMEM;
1537 	sdkp = kmalloc(sizeof(*sdkp), GFP_KERNEL);
1538 	if (!sdkp)
1539 		goto out;
1540 
1541 	memset (sdkp, 0, sizeof(*sdkp));
1542 	kref_init(&sdkp->kref);
1543 
1544 	gd = alloc_disk(16);
1545 	if (!gd)
1546 		goto out_free;
1547 
1548 	if (!idr_pre_get(&sd_index_idr, GFP_KERNEL))
1549 		goto out_put;
1550 
1551 	spin_lock(&sd_index_lock);
1552 	error = idr_get_new(&sd_index_idr, NULL, &index);
1553 	spin_unlock(&sd_index_lock);
1554 
1555 	if (index >= SD_MAX_DISKS)
1556 		error = -EBUSY;
1557 	if (error)
1558 		goto out_put;
1559 
1560 	get_device(&sdp->sdev_gendev);
1561 	sdkp->device = sdp;
1562 	sdkp->driver = &sd_template;
1563 	sdkp->disk = gd;
1564 	sdkp->index = index;
1565 	sdkp->openers = 0;
1566 
1567 	if (!sdp->timeout) {
1568 		if (sdp->type != TYPE_MOD)
1569 			sdp->timeout = SD_TIMEOUT;
1570 		else
1571 			sdp->timeout = SD_MOD_TIMEOUT;
1572 	}
1573 
1574 	gd->major = sd_major((index & 0xf0) >> 4);
1575 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
1576 	gd->minors = 16;
1577 	gd->fops = &sd_fops;
1578 
1579 	if (index < 26) {
1580 		sprintf(gd->disk_name, "sd%c", 'a' + index % 26);
1581 	} else if (index < (26 + 1) * 26) {
1582 		sprintf(gd->disk_name, "sd%c%c",
1583 			'a' + index / 26 - 1,'a' + index % 26);
1584 	} else {
1585 		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
1586 		const unsigned int m2 = (index / 26 - 1) % 26;
1587 		const unsigned int m3 =  index % 26;
1588 		sprintf(gd->disk_name, "sd%c%c%c",
1589 			'a' + m1, 'a' + m2, 'a' + m3);
1590 	}
1591 
1592 	strcpy(gd->devfs_name, sdp->devfs_name);
1593 
1594 	gd->private_data = &sdkp->driver;
1595 	gd->queue = sdkp->device->request_queue;
1596 
1597 	sd_revalidate_disk(gd);
1598 
1599 	gd->driverfs_dev = &sdp->sdev_gendev;
1600 	gd->flags = GENHD_FL_DRIVERFS;
1601 	if (sdp->removable)
1602 		gd->flags |= GENHD_FL_REMOVABLE;
1603 
1604 	dev_set_drvdata(dev, sdkp);
1605 	add_disk(gd);
1606 
1607 	sdev_printk(KERN_NOTICE, sdp, "Attached scsi %sdisk %s\n",
1608 		    sdp->removable ? "removable " : "", gd->disk_name);
1609 
1610 	return 0;
1611 
1612 out_put:
1613 	put_disk(gd);
1614 out_free:
1615 	kfree(sdkp);
1616 out:
1617 	return error;
1618 }
1619 
1620 /**
1621  *	sd_remove - called whenever a scsi disk (previously recognized by
1622  *	sd_probe) is detached from the system. It is called (potentially
1623  *	multiple times) during sd module unload.
1624  *	@sdp: pointer to mid level scsi device object
1625  *
1626  *	Note: this function is invoked from the scsi mid-level.
1627  *	This function potentially frees up a device name (e.g. /dev/sdc)
1628  *	that could be re-used by a subsequent sd_probe().
1629  *	This function is not called when the built-in sd driver is "exit-ed".
1630  **/
1631 static int sd_remove(struct device *dev)
1632 {
1633 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1634 
1635 	del_gendisk(sdkp->disk);
1636 	sd_shutdown(dev);
1637 
1638 	down(&sd_ref_sem);
1639 	dev_set_drvdata(dev, NULL);
1640 	kref_put(&sdkp->kref, scsi_disk_release);
1641 	up(&sd_ref_sem);
1642 
1643 	return 0;
1644 }
1645 
1646 /**
1647  *	scsi_disk_release - Called to free the scsi_disk structure
1648  *	@kref: pointer to embedded kref
1649  *
1650  *	sd_ref_sem must be held entering this routine.  Because it is
1651  *	called on last put, you should always use the scsi_disk_get()
1652  *	scsi_disk_put() helpers which manipulate the semaphore directly
1653  *	and never do a direct kref_put().
1654  **/
1655 static void scsi_disk_release(struct kref *kref)
1656 {
1657 	struct scsi_disk *sdkp = to_scsi_disk(kref);
1658 	struct gendisk *disk = sdkp->disk;
1659 
1660 	spin_lock(&sd_index_lock);
1661 	idr_remove(&sd_index_idr, sdkp->index);
1662 	spin_unlock(&sd_index_lock);
1663 
1664 	disk->private_data = NULL;
1665 	put_disk(disk);
1666 	put_device(&sdkp->device->sdev_gendev);
1667 
1668 	kfree(sdkp);
1669 }
1670 
1671 /*
1672  * Send a SYNCHRONIZE CACHE instruction down to the device through
1673  * the normal SCSI command structure.  Wait for the command to
1674  * complete.
1675  */
1676 static void sd_shutdown(struct device *dev)
1677 {
1678 	struct scsi_device *sdp = to_scsi_device(dev);
1679 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1680 
1681 	if (!sdkp)
1682 		return;         /* this can happen */
1683 
1684 	if (sdkp->WCE) {
1685 		printk(KERN_NOTICE "Synchronizing SCSI cache for disk %s: \n",
1686 				sdkp->disk->disk_name);
1687 		sd_sync_cache(sdp);
1688 	}
1689 	scsi_disk_put(sdkp);
1690 }
1691 
1692 /**
1693  *	init_sd - entry point for this driver (both when built in or when
1694  *	a module).
1695  *
1696  *	Note: this function registers this driver with the scsi mid-level.
1697  **/
1698 static int __init init_sd(void)
1699 {
1700 	int majors = 0, i;
1701 
1702 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
1703 
1704 	for (i = 0; i < SD_MAJORS; i++)
1705 		if (register_blkdev(sd_major(i), "sd") == 0)
1706 			majors++;
1707 
1708 	if (!majors)
1709 		return -ENODEV;
1710 
1711 	return scsi_register_driver(&sd_template.gendrv);
1712 }
1713 
1714 /**
1715  *	exit_sd - exit point for this driver (when it is a module).
1716  *
1717  *	Note: this function unregisters this driver from the scsi mid-level.
1718  **/
1719 static void __exit exit_sd(void)
1720 {
1721 	int i;
1722 
1723 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
1724 
1725 	scsi_unregister_driver(&sd_template.gendrv);
1726 	for (i = 0; i < SD_MAJORS; i++)
1727 		unregister_blkdev(sd_major(i), "sd");
1728 }
1729 
1730 MODULE_LICENSE("GPL");
1731 MODULE_AUTHOR("Eric Youngdale");
1732 MODULE_DESCRIPTION("SCSI disk (sd) driver");
1733 
1734 module_init(init_sd);
1735 module_exit(exit_sd);
1736