xref: /linux/drivers/scsi/sd.c (revision d8327c784b51b57dac2c26cfad87dce0d68dfd98)
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 <linux/mutex.h>
53 #include <asm/uaccess.h>
54 
55 #include <scsi/scsi.h>
56 #include <scsi/scsi_cmnd.h>
57 #include <scsi/scsi_dbg.h>
58 #include <scsi/scsi_device.h>
59 #include <scsi/scsi_driver.h>
60 #include <scsi/scsi_eh.h>
61 #include <scsi/scsi_host.h>
62 #include <scsi/scsi_ioctl.h>
63 #include <scsi/scsicam.h>
64 
65 #include "scsi_logging.h"
66 
67 /*
68  * More than enough for everybody ;)  The huge number of majors
69  * is a leftover from 16bit dev_t days, we don't really need that
70  * much numberspace.
71  */
72 #define SD_MAJORS	16
73 
74 /*
75  * This is limited by the naming scheme enforced in sd_probe,
76  * add another character to it if you really need more disks.
77  */
78 #define SD_MAX_DISKS	(((26 * 26) + 26 + 1) * 26)
79 
80 /*
81  * Time out in seconds for disks and Magneto-opticals (which are slower).
82  */
83 #define SD_TIMEOUT		(30 * HZ)
84 #define SD_MOD_TIMEOUT		(75 * HZ)
85 
86 /*
87  * Number of allowed retries
88  */
89 #define SD_MAX_RETRIES		5
90 #define SD_PASSTHROUGH_RETRIES	1
91 
92 /*
93  * Size of the initial data buffer for mode and read capacity data
94  */
95 #define SD_BUF_SIZE		512
96 
97 static void scsi_disk_release(struct kref *kref);
98 
99 struct scsi_disk {
100 	struct scsi_driver *driver;	/* always &sd_template */
101 	struct scsi_device *device;
102 	struct kref	kref;
103 	struct gendisk	*disk;
104 	unsigned int	openers;	/* protected by BKL for now, yuck */
105 	sector_t	capacity;	/* size in 512-byte sectors */
106 	u32		index;
107 	u8		media_present;
108 	u8		write_prot;
109 	unsigned	WCE : 1;	/* state of disk WCE bit */
110 	unsigned	RCD : 1;	/* state of disk RCD bit, unused */
111 	unsigned	DPOFUA : 1;	/* state of disk DPOFUA bit */
112 };
113 
114 static DEFINE_IDR(sd_index_idr);
115 static DEFINE_SPINLOCK(sd_index_lock);
116 
117 /* This semaphore is used to mediate the 0->1 reference get in the
118  * face of object destruction (i.e. we can't allow a get on an
119  * object after last put) */
120 static DEFINE_MUTEX(sd_ref_mutex);
121 
122 static int sd_revalidate_disk(struct gendisk *disk);
123 static void sd_rw_intr(struct scsi_cmnd * SCpnt);
124 
125 static int sd_probe(struct device *);
126 static int sd_remove(struct device *);
127 static void sd_shutdown(struct device *dev);
128 static void sd_rescan(struct device *);
129 static int sd_init_command(struct scsi_cmnd *);
130 static int sd_issue_flush(struct device *, sector_t *);
131 static void sd_prepare_flush(request_queue_t *, struct request *);
132 static void sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
133 			     unsigned char *buffer);
134 
135 static struct scsi_driver sd_template = {
136 	.owner			= THIS_MODULE,
137 	.gendrv = {
138 		.name		= "sd",
139 		.probe		= sd_probe,
140 		.remove		= sd_remove,
141 		.shutdown	= sd_shutdown,
142 	},
143 	.rescan			= sd_rescan,
144 	.init_command		= sd_init_command,
145 	.issue_flush		= sd_issue_flush,
146 };
147 
148 /*
149  * Device no to disk mapping:
150  *
151  *       major         disc2     disc  p1
152  *   |............|.............|....|....| <- dev_t
153  *    31        20 19          8 7  4 3  0
154  *
155  * Inside a major, we have 16k disks, however mapped non-
156  * contiguously. The first 16 disks are for major0, the next
157  * ones with major1, ... Disk 256 is for major0 again, disk 272
158  * for major1, ...
159  * As we stay compatible with our numbering scheme, we can reuse
160  * the well-know SCSI majors 8, 65--71, 136--143.
161  */
162 static int sd_major(int major_idx)
163 {
164 	switch (major_idx) {
165 	case 0:
166 		return SCSI_DISK0_MAJOR;
167 	case 1 ... 7:
168 		return SCSI_DISK1_MAJOR + major_idx - 1;
169 	case 8 ... 15:
170 		return SCSI_DISK8_MAJOR + major_idx - 8;
171 	default:
172 		BUG();
173 		return 0;	/* shut up gcc */
174 	}
175 }
176 
177 #define to_scsi_disk(obj) container_of(obj,struct scsi_disk,kref)
178 
179 static inline struct scsi_disk *scsi_disk(struct gendisk *disk)
180 {
181 	return container_of(disk->private_data, struct scsi_disk, driver);
182 }
183 
184 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
185 {
186 	struct scsi_disk *sdkp = NULL;
187 
188 	if (disk->private_data) {
189 		sdkp = scsi_disk(disk);
190 		if (scsi_device_get(sdkp->device) == 0)
191 			kref_get(&sdkp->kref);
192 		else
193 			sdkp = NULL;
194 	}
195 	return sdkp;
196 }
197 
198 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
199 {
200 	struct scsi_disk *sdkp;
201 
202 	mutex_lock(&sd_ref_mutex);
203 	sdkp = __scsi_disk_get(disk);
204 	mutex_unlock(&sd_ref_mutex);
205 	return sdkp;
206 }
207 
208 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
209 {
210 	struct scsi_disk *sdkp;
211 
212 	mutex_lock(&sd_ref_mutex);
213 	sdkp = dev_get_drvdata(dev);
214 	if (sdkp)
215 		sdkp = __scsi_disk_get(sdkp->disk);
216 	mutex_unlock(&sd_ref_mutex);
217 	return sdkp;
218 }
219 
220 static void scsi_disk_put(struct scsi_disk *sdkp)
221 {
222 	struct scsi_device *sdev = sdkp->device;
223 
224 	mutex_lock(&sd_ref_mutex);
225 	kref_put(&sdkp->kref, scsi_disk_release);
226 	scsi_device_put(sdev);
227 	mutex_unlock(&sd_ref_mutex);
228 }
229 
230 /**
231  *	sd_init_command - build a scsi (read or write) command from
232  *	information in the request structure.
233  *	@SCpnt: pointer to mid-level's per scsi command structure that
234  *	contains request and into which the scsi command is written
235  *
236  *	Returns 1 if successful and 0 if error (or cannot be done now).
237  **/
238 static int sd_init_command(struct scsi_cmnd * SCpnt)
239 {
240 	struct scsi_device *sdp = SCpnt->device;
241 	struct request *rq = SCpnt->request;
242 	struct gendisk *disk = rq->rq_disk;
243 	sector_t block = rq->sector;
244 	unsigned int this_count = SCpnt->request_bufflen >> 9;
245 	unsigned int timeout = sdp->timeout;
246 
247 	SCSI_LOG_HLQUEUE(1, printk("sd_init_command: disk=%s, block=%llu, "
248 			    "count=%d\n", disk->disk_name,
249 			 (unsigned long long)block, this_count));
250 
251 	if (!sdp || !scsi_device_online(sdp) ||
252  	    block + rq->nr_sectors > get_capacity(disk)) {
253 		SCSI_LOG_HLQUEUE(2, printk("Finishing %ld sectors\n",
254 				 rq->nr_sectors));
255 		SCSI_LOG_HLQUEUE(2, printk("Retry with 0x%p\n", SCpnt));
256 		return 0;
257 	}
258 
259 	if (sdp->changed) {
260 		/*
261 		 * quietly refuse to do anything to a changed disc until
262 		 * the changed bit has been reset
263 		 */
264 		/* printk("SCSI disk has been changed. Prohibiting further I/O.\n"); */
265 		return 0;
266 	}
267 	SCSI_LOG_HLQUEUE(2, printk("%s : block=%llu\n",
268 				   disk->disk_name, (unsigned long long)block));
269 
270 	/*
271 	 * If we have a 1K hardware sectorsize, prevent access to single
272 	 * 512 byte sectors.  In theory we could handle this - in fact
273 	 * the scsi cdrom driver must be able to handle this because
274 	 * we typically use 1K blocksizes, and cdroms typically have
275 	 * 2K hardware sectorsizes.  Of course, things are simpler
276 	 * with the cdrom, since it is read-only.  For performance
277 	 * reasons, the filesystems should be able to handle this
278 	 * and not force the scsi disk driver to use bounce buffers
279 	 * for this.
280 	 */
281 	if (sdp->sector_size == 1024) {
282 		if ((block & 1) || (rq->nr_sectors & 1)) {
283 			printk(KERN_ERR "sd: Bad block number requested");
284 			return 0;
285 		} else {
286 			block = block >> 1;
287 			this_count = this_count >> 1;
288 		}
289 	}
290 	if (sdp->sector_size == 2048) {
291 		if ((block & 3) || (rq->nr_sectors & 3)) {
292 			printk(KERN_ERR "sd: Bad block number requested");
293 			return 0;
294 		} else {
295 			block = block >> 2;
296 			this_count = this_count >> 2;
297 		}
298 	}
299 	if (sdp->sector_size == 4096) {
300 		if ((block & 7) || (rq->nr_sectors & 7)) {
301 			printk(KERN_ERR "sd: Bad block number requested");
302 			return 0;
303 		} else {
304 			block = block >> 3;
305 			this_count = this_count >> 3;
306 		}
307 	}
308 	if (rq_data_dir(rq) == WRITE) {
309 		if (!sdp->writeable) {
310 			return 0;
311 		}
312 		SCpnt->cmnd[0] = WRITE_6;
313 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
314 	} else if (rq_data_dir(rq) == READ) {
315 		SCpnt->cmnd[0] = READ_6;
316 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
317 	} else {
318 		printk(KERN_ERR "sd: Unknown command %lx\n", rq->flags);
319 /* overkill 	panic("Unknown sd command %lx\n", rq->flags); */
320 		return 0;
321 	}
322 
323 	SCSI_LOG_HLQUEUE(2, printk("%s : %s %d/%ld 512 byte blocks.\n",
324 		disk->disk_name, (rq_data_dir(rq) == WRITE) ?
325 		"writing" : "reading", this_count, rq->nr_sectors));
326 
327 	SCpnt->cmnd[1] = 0;
328 
329 	if (block > 0xffffffff) {
330 		SCpnt->cmnd[0] += READ_16 - READ_6;
331 		SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
332 		SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
333 		SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
334 		SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
335 		SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
336 		SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
337 		SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
338 		SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
339 		SCpnt->cmnd[9] = (unsigned char) block & 0xff;
340 		SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
341 		SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
342 		SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
343 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
344 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
345 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
346 		   SCpnt->device->use_10_for_rw) {
347 		if (this_count > 0xffff)
348 			this_count = 0xffff;
349 
350 		SCpnt->cmnd[0] += READ_10 - READ_6;
351 		SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
352 		SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
353 		SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
354 		SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
355 		SCpnt->cmnd[5] = (unsigned char) block & 0xff;
356 		SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
357 		SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
358 		SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
359 	} else {
360 		if (unlikely(blk_fua_rq(rq))) {
361 			/*
362 			 * This happens only if this drive failed
363 			 * 10byte rw command with ILLEGAL_REQUEST
364 			 * during operation and thus turned off
365 			 * use_10_for_rw.
366 			 */
367 			printk(KERN_ERR "sd: FUA write on READ/WRITE(6) drive\n");
368 			return 0;
369 		}
370 
371 		SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
372 		SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
373 		SCpnt->cmnd[3] = (unsigned char) block & 0xff;
374 		SCpnt->cmnd[4] = (unsigned char) this_count;
375 		SCpnt->cmnd[5] = 0;
376 	}
377 	SCpnt->request_bufflen = SCpnt->bufflen =
378 			this_count * sdp->sector_size;
379 
380 	/*
381 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
382 	 * host adapter, it's safe to assume that we can at least transfer
383 	 * this many bytes between each connect / disconnect.
384 	 */
385 	SCpnt->transfersize = sdp->sector_size;
386 	SCpnt->underflow = this_count << 9;
387 	SCpnt->allowed = SD_MAX_RETRIES;
388 	SCpnt->timeout_per_command = timeout;
389 
390 	/*
391 	 * This is the completion routine we use.  This is matched in terms
392 	 * of capability to this function.
393 	 */
394 	SCpnt->done = sd_rw_intr;
395 
396 	/*
397 	 * This indicates that the command is ready from our end to be
398 	 * queued.
399 	 */
400 	return 1;
401 }
402 
403 /**
404  *	sd_open - open a scsi disk device
405  *	@inode: only i_rdev member may be used
406  *	@filp: only f_mode and f_flags may be used
407  *
408  *	Returns 0 if successful. Returns a negated errno value in case
409  *	of error.
410  *
411  *	Note: This can be called from a user context (e.g. fsck(1) )
412  *	or from within the kernel (e.g. as a result of a mount(1) ).
413  *	In the latter case @inode and @filp carry an abridged amount
414  *	of information as noted above.
415  **/
416 static int sd_open(struct inode *inode, struct file *filp)
417 {
418 	struct gendisk *disk = inode->i_bdev->bd_disk;
419 	struct scsi_disk *sdkp;
420 	struct scsi_device *sdev;
421 	int retval;
422 
423 	if (!(sdkp = scsi_disk_get(disk)))
424 		return -ENXIO;
425 
426 
427 	SCSI_LOG_HLQUEUE(3, printk("sd_open: disk=%s\n", disk->disk_name));
428 
429 	sdev = sdkp->device;
430 
431 	/*
432 	 * If the device is in error recovery, wait until it is done.
433 	 * If the device is offline, then disallow any access to it.
434 	 */
435 	retval = -ENXIO;
436 	if (!scsi_block_when_processing_errors(sdev))
437 		goto error_out;
438 
439 	if (sdev->removable || sdkp->write_prot)
440 		check_disk_change(inode->i_bdev);
441 
442 	/*
443 	 * If the drive is empty, just let the open fail.
444 	 */
445 	retval = -ENOMEDIUM;
446 	if (sdev->removable && !sdkp->media_present &&
447 	    !(filp->f_flags & O_NDELAY))
448 		goto error_out;
449 
450 	/*
451 	 * If the device has the write protect tab set, have the open fail
452 	 * if the user expects to be able to write to the thing.
453 	 */
454 	retval = -EROFS;
455 	if (sdkp->write_prot && (filp->f_mode & FMODE_WRITE))
456 		goto error_out;
457 
458 	/*
459 	 * It is possible that the disk changing stuff resulted in
460 	 * the device being taken offline.  If this is the case,
461 	 * report this to the user, and don't pretend that the
462 	 * open actually succeeded.
463 	 */
464 	retval = -ENXIO;
465 	if (!scsi_device_online(sdev))
466 		goto error_out;
467 
468 	if (!sdkp->openers++ && sdev->removable) {
469 		if (scsi_block_when_processing_errors(sdev))
470 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
471 	}
472 
473 	return 0;
474 
475 error_out:
476 	scsi_disk_put(sdkp);
477 	return retval;
478 }
479 
480 /**
481  *	sd_release - invoked when the (last) close(2) is called on this
482  *	scsi disk.
483  *	@inode: only i_rdev member may be used
484  *	@filp: only f_mode and f_flags may be used
485  *
486  *	Returns 0.
487  *
488  *	Note: may block (uninterruptible) if error recovery is underway
489  *	on this disk.
490  **/
491 static int sd_release(struct inode *inode, struct file *filp)
492 {
493 	struct gendisk *disk = inode->i_bdev->bd_disk;
494 	struct scsi_disk *sdkp = scsi_disk(disk);
495 	struct scsi_device *sdev = sdkp->device;
496 
497 	SCSI_LOG_HLQUEUE(3, printk("sd_release: disk=%s\n", disk->disk_name));
498 
499 	if (!--sdkp->openers && sdev->removable) {
500 		if (scsi_block_when_processing_errors(sdev))
501 			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
502 	}
503 
504 	/*
505 	 * XXX and what if there are packets in flight and this close()
506 	 * XXX is followed by a "rmmod sd_mod"?
507 	 */
508 	scsi_disk_put(sdkp);
509 	return 0;
510 }
511 
512 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
513 {
514 	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
515 	struct scsi_device *sdp = sdkp->device;
516 	struct Scsi_Host *host = sdp->host;
517 	int diskinfo[4];
518 
519 	/* default to most commonly used values */
520         diskinfo[0] = 0x40;	/* 1 << 6 */
521        	diskinfo[1] = 0x20;	/* 1 << 5 */
522        	diskinfo[2] = sdkp->capacity >> 11;
523 
524 	/* override with calculated, extended default, or driver values */
525 	if (host->hostt->bios_param)
526 		host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
527 	else
528 		scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
529 
530 	geo->heads = diskinfo[0];
531 	geo->sectors = diskinfo[1];
532 	geo->cylinders = diskinfo[2];
533 	return 0;
534 }
535 
536 /**
537  *	sd_ioctl - process an ioctl
538  *	@inode: only i_rdev/i_bdev members may be used
539  *	@filp: only f_mode and f_flags may be used
540  *	@cmd: ioctl command number
541  *	@arg: this is third argument given to ioctl(2) system call.
542  *	Often contains a pointer.
543  *
544  *	Returns 0 if successful (some ioctls return postive numbers on
545  *	success as well). Returns a negated errno value in case of error.
546  *
547  *	Note: most ioctls are forward onto the block subsystem or further
548  *	down in the scsi subsytem.
549  **/
550 static int sd_ioctl(struct inode * inode, struct file * filp,
551 		    unsigned int cmd, unsigned long arg)
552 {
553 	struct block_device *bdev = inode->i_bdev;
554 	struct gendisk *disk = bdev->bd_disk;
555 	struct scsi_device *sdp = scsi_disk(disk)->device;
556 	void __user *p = (void __user *)arg;
557 	int error;
558 
559 	SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
560 						disk->disk_name, cmd));
561 
562 	/*
563 	 * If we are in the middle of error recovery, don't let anyone
564 	 * else try and use this device.  Also, if error recovery fails, it
565 	 * may try and take the device offline, in which case all further
566 	 * access to the device is prohibited.
567 	 */
568 	error = scsi_nonblockable_ioctl(sdp, cmd, p, filp);
569 	if (!scsi_block_when_processing_errors(sdp) || !error)
570 		return error;
571 
572 	/*
573 	 * Send SCSI addressing ioctls directly to mid level, send other
574 	 * ioctls to block level and then onto mid level if they can't be
575 	 * resolved.
576 	 */
577 	switch (cmd) {
578 		case SCSI_IOCTL_GET_IDLUN:
579 		case SCSI_IOCTL_GET_BUS_NUMBER:
580 			return scsi_ioctl(sdp, cmd, p);
581 		default:
582 			error = scsi_cmd_ioctl(filp, disk, cmd, p);
583 			if (error != -ENOTTY)
584 				return error;
585 	}
586 	return scsi_ioctl(sdp, cmd, p);
587 }
588 
589 static void set_media_not_present(struct scsi_disk *sdkp)
590 {
591 	sdkp->media_present = 0;
592 	sdkp->capacity = 0;
593 	sdkp->device->changed = 1;
594 }
595 
596 /**
597  *	sd_media_changed - check if our medium changed
598  *	@disk: kernel device descriptor
599  *
600  *	Returns 0 if not applicable or no change; 1 if change
601  *
602  *	Note: this function is invoked from the block subsystem.
603  **/
604 static int sd_media_changed(struct gendisk *disk)
605 {
606 	struct scsi_disk *sdkp = scsi_disk(disk);
607 	struct scsi_device *sdp = sdkp->device;
608 	int retval;
609 
610 	SCSI_LOG_HLQUEUE(3, printk("sd_media_changed: disk=%s\n",
611 						disk->disk_name));
612 
613 	if (!sdp->removable)
614 		return 0;
615 
616 	/*
617 	 * If the device is offline, don't send any commands - just pretend as
618 	 * if the command failed.  If the device ever comes back online, we
619 	 * can deal with it then.  It is only because of unrecoverable errors
620 	 * that we would ever take a device offline in the first place.
621 	 */
622 	if (!scsi_device_online(sdp))
623 		goto not_present;
624 
625 	/*
626 	 * Using TEST_UNIT_READY enables differentiation between drive with
627 	 * no cartridge loaded - NOT READY, drive with changed cartridge -
628 	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
629 	 *
630 	 * Drives that auto spin down. eg iomega jaz 1G, will be started
631 	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
632 	 * sd_revalidate() is called.
633 	 */
634 	retval = -ENODEV;
635 	if (scsi_block_when_processing_errors(sdp))
636 		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES);
637 
638 	/*
639 	 * Unable to test, unit probably not ready.   This usually
640 	 * means there is no disc in the drive.  Mark as changed,
641 	 * and we will figure it out later once the drive is
642 	 * available again.
643 	 */
644 	if (retval)
645 		 goto not_present;
646 
647 	/*
648 	 * For removable scsi disk we have to recognise the presence
649 	 * of a disk in the drive. This is kept in the struct scsi_disk
650 	 * struct and tested at open !  Daniel Roche (dan@lectra.fr)
651 	 */
652 	sdkp->media_present = 1;
653 
654 	retval = sdp->changed;
655 	sdp->changed = 0;
656 
657 	return retval;
658 
659 not_present:
660 	set_media_not_present(sdkp);
661 	return 1;
662 }
663 
664 static int sd_sync_cache(struct scsi_device *sdp)
665 {
666 	int retries, res;
667 	struct scsi_sense_hdr sshdr;
668 
669 	if (!scsi_device_online(sdp))
670 		return -ENODEV;
671 
672 
673 	for (retries = 3; retries > 0; --retries) {
674 		unsigned char cmd[10] = { 0 };
675 
676 		cmd[0] = SYNCHRONIZE_CACHE;
677 		/*
678 		 * Leave the rest of the command zero to indicate
679 		 * flush everything.
680 		 */
681 		res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
682 				       SD_TIMEOUT, SD_MAX_RETRIES);
683 		if (res == 0)
684 			break;
685 	}
686 
687 	if (res) {		printk(KERN_WARNING "FAILED\n  status = %x, message = %02x, "
688 				    "host = %d, driver = %02x\n  ",
689 				    status_byte(res), msg_byte(res),
690 				    host_byte(res), driver_byte(res));
691 			if (driver_byte(res) & DRIVER_SENSE)
692 				scsi_print_sense_hdr("sd", &sshdr);
693 	}
694 
695 	return res;
696 }
697 
698 static int sd_issue_flush(struct device *dev, sector_t *error_sector)
699 {
700 	int ret = 0;
701 	struct scsi_device *sdp = to_scsi_device(dev);
702 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
703 
704 	if (!sdkp)
705                return -ENODEV;
706 
707 	if (sdkp->WCE)
708 		ret = sd_sync_cache(sdp);
709 	scsi_disk_put(sdkp);
710 	return ret;
711 }
712 
713 static void sd_prepare_flush(request_queue_t *q, struct request *rq)
714 {
715 	memset(rq->cmd, 0, sizeof(rq->cmd));
716 	rq->flags |= REQ_BLOCK_PC;
717 	rq->timeout = SD_TIMEOUT;
718 	rq->cmd[0] = SYNCHRONIZE_CACHE;
719 	rq->cmd_len = 10;
720 }
721 
722 static void sd_rescan(struct device *dev)
723 {
724 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
725 
726 	if (sdkp) {
727 		sd_revalidate_disk(sdkp->disk);
728 		scsi_disk_put(sdkp);
729 	}
730 }
731 
732 
733 #ifdef CONFIG_COMPAT
734 /*
735  * This gets directly called from VFS. When the ioctl
736  * is not recognized we go back to the other translation paths.
737  */
738 static long sd_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
739 {
740 	struct block_device *bdev = file->f_dentry->d_inode->i_bdev;
741 	struct gendisk *disk = bdev->bd_disk;
742 	struct scsi_device *sdev = scsi_disk(disk)->device;
743 
744 	/*
745 	 * If we are in the middle of error recovery, don't let anyone
746 	 * else try and use this device.  Also, if error recovery fails, it
747 	 * may try and take the device offline, in which case all further
748 	 * access to the device is prohibited.
749 	 */
750 	if (!scsi_block_when_processing_errors(sdev))
751 		return -ENODEV;
752 
753 	if (sdev->host->hostt->compat_ioctl) {
754 		int ret;
755 
756 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
757 
758 		return ret;
759 	}
760 
761 	/*
762 	 * Let the static ioctl translation table take care of it.
763 	 */
764 	return -ENOIOCTLCMD;
765 }
766 #endif
767 
768 static struct block_device_operations sd_fops = {
769 	.owner			= THIS_MODULE,
770 	.open			= sd_open,
771 	.release		= sd_release,
772 	.ioctl			= sd_ioctl,
773 	.getgeo			= sd_getgeo,
774 #ifdef CONFIG_COMPAT
775 	.compat_ioctl		= sd_compat_ioctl,
776 #endif
777 	.media_changed		= sd_media_changed,
778 	.revalidate_disk	= sd_revalidate_disk,
779 };
780 
781 /**
782  *	sd_rw_intr - bottom half handler: called when the lower level
783  *	driver has completed (successfully or otherwise) a scsi command.
784  *	@SCpnt: mid-level's per command structure.
785  *
786  *	Note: potentially run from within an ISR. Must not block.
787  **/
788 static void sd_rw_intr(struct scsi_cmnd * SCpnt)
789 {
790 	int result = SCpnt->result;
791 	int this_count = SCpnt->bufflen;
792 	int good_bytes = (result == 0 ? this_count : 0);
793 	sector_t block_sectors = 1;
794 	u64 first_err_block;
795 	sector_t error_sector;
796 	struct scsi_sense_hdr sshdr;
797 	int sense_valid = 0;
798 	int sense_deferred = 0;
799 	int info_valid;
800 
801 	if (result) {
802 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
803 		if (sense_valid)
804 			sense_deferred = scsi_sense_is_deferred(&sshdr);
805 	}
806 
807 #ifdef CONFIG_SCSI_LOGGING
808 	SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: %s: res=0x%x\n",
809 				SCpnt->request->rq_disk->disk_name, result));
810 	if (sense_valid) {
811 		SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: sb[respc,sk,asc,"
812 				"ascq]=%x,%x,%x,%x\n", sshdr.response_code,
813 				sshdr.sense_key, sshdr.asc, sshdr.ascq));
814 	}
815 #endif
816 	/*
817 	   Handle MEDIUM ERRORs that indicate partial success.  Since this is a
818 	   relatively rare error condition, no care is taken to avoid
819 	   unnecessary additional work such as memcpy's that could be avoided.
820 	 */
821 	if (driver_byte(result) != 0 &&
822 		 sense_valid && !sense_deferred) {
823 		switch (sshdr.sense_key) {
824 		case MEDIUM_ERROR:
825 			if (!blk_fs_request(SCpnt->request))
826 				break;
827 			info_valid = scsi_get_sense_info_fld(
828 				SCpnt->sense_buffer, SCSI_SENSE_BUFFERSIZE,
829 				&first_err_block);
830 			/*
831 			 * May want to warn and skip if following cast results
832 			 * in actual truncation (if sector_t < 64 bits)
833 			 */
834 			error_sector = (sector_t)first_err_block;
835 			if (SCpnt->request->bio != NULL)
836 				block_sectors = bio_sectors(SCpnt->request->bio);
837 			switch (SCpnt->device->sector_size) {
838 			case 1024:
839 				error_sector <<= 1;
840 				if (block_sectors < 2)
841 					block_sectors = 2;
842 				break;
843 			case 2048:
844 				error_sector <<= 2;
845 				if (block_sectors < 4)
846 					block_sectors = 4;
847 				break;
848 			case 4096:
849 				error_sector <<=3;
850 				if (block_sectors < 8)
851 					block_sectors = 8;
852 				break;
853 			case 256:
854 				error_sector >>= 1;
855 				break;
856 			default:
857 				break;
858 			}
859 
860 			error_sector &= ~(block_sectors - 1);
861 			good_bytes = (error_sector - SCpnt->request->sector) << 9;
862 			if (good_bytes < 0 || good_bytes >= this_count)
863 				good_bytes = 0;
864 			break;
865 
866 		case RECOVERED_ERROR: /* an error occurred, but it recovered */
867 		case NO_SENSE: /* LLDD got sense data */
868 			/*
869 			 * Inform the user, but make sure that it's not treated
870 			 * as a hard error.
871 			 */
872 			scsi_print_sense("sd", SCpnt);
873 			SCpnt->result = 0;
874 			memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
875 			good_bytes = this_count;
876 			break;
877 
878 		case ILLEGAL_REQUEST:
879 			if (SCpnt->device->use_10_for_rw &&
880 			    (SCpnt->cmnd[0] == READ_10 ||
881 			     SCpnt->cmnd[0] == WRITE_10))
882 				SCpnt->device->use_10_for_rw = 0;
883 			if (SCpnt->device->use_10_for_ms &&
884 			    (SCpnt->cmnd[0] == MODE_SENSE_10 ||
885 			     SCpnt->cmnd[0] == MODE_SELECT_10))
886 				SCpnt->device->use_10_for_ms = 0;
887 			break;
888 
889 		default:
890 			break;
891 		}
892 	}
893 	/*
894 	 * This calls the generic completion function, now that we know
895 	 * how many actual sectors finished, and how many sectors we need
896 	 * to say have failed.
897 	 */
898 	scsi_io_completion(SCpnt, good_bytes, block_sectors << 9);
899 }
900 
901 static int media_not_present(struct scsi_disk *sdkp,
902 			     struct scsi_sense_hdr *sshdr)
903 {
904 
905 	if (!scsi_sense_valid(sshdr))
906 		return 0;
907 	/* not invoked for commands that could return deferred errors */
908 	if (sshdr->sense_key != NOT_READY &&
909 	    sshdr->sense_key != UNIT_ATTENTION)
910 		return 0;
911 	if (sshdr->asc != 0x3A) /* medium not present */
912 		return 0;
913 
914 	set_media_not_present(sdkp);
915 	return 1;
916 }
917 
918 /*
919  * spinup disk - called only in sd_revalidate_disk()
920  */
921 static void
922 sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
923 {
924 	unsigned char cmd[10];
925 	unsigned long spintime_expire = 0;
926 	int retries, spintime;
927 	unsigned int the_result;
928 	struct scsi_sense_hdr sshdr;
929 	int sense_valid = 0;
930 
931 	spintime = 0;
932 
933 	/* Spin up drives, as required.  Only do this at boot time */
934 	/* Spinup needs to be done for module loads too. */
935 	do {
936 		retries = 0;
937 
938 		do {
939 			cmd[0] = TEST_UNIT_READY;
940 			memset((void *) &cmd[1], 0, 9);
941 
942 			the_result = scsi_execute_req(sdkp->device, cmd,
943 						      DMA_NONE, NULL, 0,
944 						      &sshdr, SD_TIMEOUT,
945 						      SD_MAX_RETRIES);
946 
947 			if (the_result)
948 				sense_valid = scsi_sense_valid(&sshdr);
949 			retries++;
950 		} while (retries < 3 &&
951 			 (!scsi_status_is_good(the_result) ||
952 			  ((driver_byte(the_result) & DRIVER_SENSE) &&
953 			  sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
954 
955 		/*
956 		 * If the drive has indicated to us that it doesn't have
957 		 * any media in it, don't bother with any of the rest of
958 		 * this crap.
959 		 */
960 		if (media_not_present(sdkp, &sshdr))
961 			return;
962 
963 		if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
964 			/* no sense, TUR either succeeded or failed
965 			 * with a status error */
966 			if(!spintime && !scsi_status_is_good(the_result))
967 				printk(KERN_NOTICE "%s: Unit Not Ready, "
968 				       "error = 0x%x\n", diskname, the_result);
969 			break;
970 		}
971 
972 		/*
973 		 * The device does not want the automatic start to be issued.
974 		 */
975 		if (sdkp->device->no_start_on_add) {
976 			break;
977 		}
978 
979 		/*
980 		 * If manual intervention is required, or this is an
981 		 * absent USB storage device, a spinup is meaningless.
982 		 */
983 		if (sense_valid &&
984 		    sshdr.sense_key == NOT_READY &&
985 		    sshdr.asc == 4 && sshdr.ascq == 3) {
986 			break;		/* manual intervention required */
987 
988 		/*
989 		 * Issue command to spin up drive when not ready
990 		 */
991 		} else if (sense_valid && sshdr.sense_key == NOT_READY) {
992 			if (!spintime) {
993 				printk(KERN_NOTICE "%s: Spinning up disk...",
994 				       diskname);
995 				cmd[0] = START_STOP;
996 				cmd[1] = 1;	/* Return immediately */
997 				memset((void *) &cmd[2], 0, 8);
998 				cmd[4] = 1;	/* Start spin cycle */
999 				scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1000 						 NULL, 0, &sshdr,
1001 						 SD_TIMEOUT, SD_MAX_RETRIES);
1002 				spintime_expire = jiffies + 100 * HZ;
1003 				spintime = 1;
1004 			}
1005 			/* Wait 1 second for next try */
1006 			msleep(1000);
1007 			printk(".");
1008 
1009 		/*
1010 		 * Wait for USB flash devices with slow firmware.
1011 		 * Yes, this sense key/ASC combination shouldn't
1012 		 * occur here.  It's characteristic of these devices.
1013 		 */
1014 		} else if (sense_valid &&
1015 				sshdr.sense_key == UNIT_ATTENTION &&
1016 				sshdr.asc == 0x28) {
1017 			if (!spintime) {
1018 				spintime_expire = jiffies + 5 * HZ;
1019 				spintime = 1;
1020 			}
1021 			/* Wait 1 second for next try */
1022 			msleep(1000);
1023 		} else {
1024 			/* we don't understand the sense code, so it's
1025 			 * probably pointless to loop */
1026 			if(!spintime) {
1027 				printk(KERN_NOTICE "%s: Unit Not Ready, "
1028 					"sense:\n", diskname);
1029 				scsi_print_sense_hdr("", &sshdr);
1030 			}
1031 			break;
1032 		}
1033 
1034 	} while (spintime && time_before_eq(jiffies, spintime_expire));
1035 
1036 	if (spintime) {
1037 		if (scsi_status_is_good(the_result))
1038 			printk("ready\n");
1039 		else
1040 			printk("not responding...\n");
1041 	}
1042 }
1043 
1044 /*
1045  * read disk capacity
1046  */
1047 static void
1048 sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
1049 		 unsigned char *buffer)
1050 {
1051 	unsigned char cmd[16];
1052 	int the_result, retries;
1053 	int sector_size = 0;
1054 	int longrc = 0;
1055 	struct scsi_sense_hdr sshdr;
1056 	int sense_valid = 0;
1057 	struct scsi_device *sdp = sdkp->device;
1058 
1059 repeat:
1060 	retries = 3;
1061 	do {
1062 		if (longrc) {
1063 			memset((void *) cmd, 0, 16);
1064 			cmd[0] = SERVICE_ACTION_IN;
1065 			cmd[1] = SAI_READ_CAPACITY_16;
1066 			cmd[13] = 12;
1067 			memset((void *) buffer, 0, 12);
1068 		} else {
1069 			cmd[0] = READ_CAPACITY;
1070 			memset((void *) &cmd[1], 0, 9);
1071 			memset((void *) buffer, 0, 8);
1072 		}
1073 
1074 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1075 					      buffer, longrc ? 12 : 8, &sshdr,
1076 					      SD_TIMEOUT, SD_MAX_RETRIES);
1077 
1078 		if (media_not_present(sdkp, &sshdr))
1079 			return;
1080 
1081 		if (the_result)
1082 			sense_valid = scsi_sense_valid(&sshdr);
1083 		retries--;
1084 
1085 	} while (the_result && retries);
1086 
1087 	if (the_result && !longrc) {
1088 		printk(KERN_NOTICE "%s : READ CAPACITY failed.\n"
1089 		       "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
1090 		       diskname, diskname,
1091 		       status_byte(the_result),
1092 		       msg_byte(the_result),
1093 		       host_byte(the_result),
1094 		       driver_byte(the_result));
1095 
1096 		if (driver_byte(the_result) & DRIVER_SENSE)
1097 			scsi_print_sense_hdr("sd", &sshdr);
1098 		else
1099 			printk("%s : sense not available. \n", diskname);
1100 
1101 		/* Set dirty bit for removable devices if not ready -
1102 		 * sometimes drives will not report this properly. */
1103 		if (sdp->removable &&
1104 		    sense_valid && sshdr.sense_key == NOT_READY)
1105 			sdp->changed = 1;
1106 
1107 		/* Either no media are present but the drive didn't tell us,
1108 		   or they are present but the read capacity command fails */
1109 		/* sdkp->media_present = 0; -- not always correct */
1110 		sdkp->capacity = 0x200000; /* 1 GB - random */
1111 
1112 		return;
1113 	} else if (the_result && longrc) {
1114 		/* READ CAPACITY(16) has been failed */
1115 		printk(KERN_NOTICE "%s : READ CAPACITY(16) failed.\n"
1116 		       "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
1117 		       diskname, diskname,
1118 		       status_byte(the_result),
1119 		       msg_byte(the_result),
1120 		       host_byte(the_result),
1121 		       driver_byte(the_result));
1122 		printk(KERN_NOTICE "%s : use 0xffffffff as device size\n",
1123 		       diskname);
1124 
1125 		sdkp->capacity = 1 + (sector_t) 0xffffffff;
1126 		goto got_data;
1127 	}
1128 
1129 	if (!longrc) {
1130 		sector_size = (buffer[4] << 24) |
1131 			(buffer[5] << 16) | (buffer[6] << 8) | buffer[7];
1132 		if (buffer[0] == 0xff && buffer[1] == 0xff &&
1133 		    buffer[2] == 0xff && buffer[3] == 0xff) {
1134 			if(sizeof(sdkp->capacity) > 4) {
1135 				printk(KERN_NOTICE "%s : very big device. try to use"
1136 				       " READ CAPACITY(16).\n", diskname);
1137 				longrc = 1;
1138 				goto repeat;
1139 			}
1140 			printk(KERN_ERR "%s: too big for this kernel.  Use a "
1141 			       "kernel compiled with support for large block "
1142 			       "devices.\n", diskname);
1143 			sdkp->capacity = 0;
1144 			goto got_data;
1145 		}
1146 		sdkp->capacity = 1 + (((sector_t)buffer[0] << 24) |
1147 			(buffer[1] << 16) |
1148 			(buffer[2] << 8) |
1149 			buffer[3]);
1150 	} else {
1151 		sdkp->capacity = 1 + (((u64)buffer[0] << 56) |
1152 			((u64)buffer[1] << 48) |
1153 			((u64)buffer[2] << 40) |
1154 			((u64)buffer[3] << 32) |
1155 			((sector_t)buffer[4] << 24) |
1156 			((sector_t)buffer[5] << 16) |
1157 			((sector_t)buffer[6] << 8)  |
1158 			(sector_t)buffer[7]);
1159 
1160 		sector_size = (buffer[8] << 24) |
1161 			(buffer[9] << 16) | (buffer[10] << 8) | buffer[11];
1162 	}
1163 
1164 	/* Some devices return the total number of sectors, not the
1165 	 * highest sector number.  Make the necessary adjustment. */
1166 	if (sdp->fix_capacity)
1167 		--sdkp->capacity;
1168 
1169 got_data:
1170 	if (sector_size == 0) {
1171 		sector_size = 512;
1172 		printk(KERN_NOTICE "%s : sector size 0 reported, "
1173 		       "assuming 512.\n", diskname);
1174 	}
1175 
1176 	if (sector_size != 512 &&
1177 	    sector_size != 1024 &&
1178 	    sector_size != 2048 &&
1179 	    sector_size != 4096 &&
1180 	    sector_size != 256) {
1181 		printk(KERN_NOTICE "%s : unsupported sector size "
1182 		       "%d.\n", diskname, sector_size);
1183 		/*
1184 		 * The user might want to re-format the drive with
1185 		 * a supported sectorsize.  Once this happens, it
1186 		 * would be relatively trivial to set the thing up.
1187 		 * For this reason, we leave the thing in the table.
1188 		 */
1189 		sdkp->capacity = 0;
1190 		/*
1191 		 * set a bogus sector size so the normal read/write
1192 		 * logic in the block layer will eventually refuse any
1193 		 * request on this device without tripping over power
1194 		 * of two sector size assumptions
1195 		 */
1196 		sector_size = 512;
1197 	}
1198 	{
1199 		/*
1200 		 * The msdos fs needs to know the hardware sector size
1201 		 * So I have created this table. See ll_rw_blk.c
1202 		 * Jacques Gelinas (Jacques@solucorp.qc.ca)
1203 		 */
1204 		int hard_sector = sector_size;
1205 		sector_t sz = (sdkp->capacity/2) * (hard_sector/256);
1206 		request_queue_t *queue = sdp->request_queue;
1207 		sector_t mb = sz;
1208 
1209 		blk_queue_hardsect_size(queue, hard_sector);
1210 		/* avoid 64-bit division on 32-bit platforms */
1211 		sector_div(sz, 625);
1212 		mb -= sz - 974;
1213 		sector_div(mb, 1950);
1214 
1215 		printk(KERN_NOTICE "SCSI device %s: "
1216 		       "%llu %d-byte hdwr sectors (%llu MB)\n",
1217 		       diskname, (unsigned long long)sdkp->capacity,
1218 		       hard_sector, (unsigned long long)mb);
1219 	}
1220 
1221 	/* Rescale capacity to 512-byte units */
1222 	if (sector_size == 4096)
1223 		sdkp->capacity <<= 3;
1224 	else if (sector_size == 2048)
1225 		sdkp->capacity <<= 2;
1226 	else if (sector_size == 1024)
1227 		sdkp->capacity <<= 1;
1228 	else if (sector_size == 256)
1229 		sdkp->capacity >>= 1;
1230 
1231 	sdkp->device->sector_size = sector_size;
1232 }
1233 
1234 /* called with buffer of length 512 */
1235 static inline int
1236 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1237 		 unsigned char *buffer, int len, struct scsi_mode_data *data,
1238 		 struct scsi_sense_hdr *sshdr)
1239 {
1240 	return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1241 			       SD_TIMEOUT, SD_MAX_RETRIES, data,
1242 			       sshdr);
1243 }
1244 
1245 /*
1246  * read write protect setting, if possible - called only in sd_revalidate_disk()
1247  * called with buffer of length SD_BUF_SIZE
1248  */
1249 static void
1250 sd_read_write_protect_flag(struct scsi_disk *sdkp, char *diskname,
1251 			   unsigned char *buffer)
1252 {
1253 	int res;
1254 	struct scsi_device *sdp = sdkp->device;
1255 	struct scsi_mode_data data;
1256 
1257 	set_disk_ro(sdkp->disk, 0);
1258 	if (sdp->skip_ms_page_3f) {
1259 		printk(KERN_NOTICE "%s: assuming Write Enabled\n", diskname);
1260 		return;
1261 	}
1262 
1263 	if (sdp->use_192_bytes_for_3f) {
1264 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1265 	} else {
1266 		/*
1267 		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1268 		 * We have to start carefully: some devices hang if we ask
1269 		 * for more than is available.
1270 		 */
1271 		res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1272 
1273 		/*
1274 		 * Second attempt: ask for page 0 When only page 0 is
1275 		 * implemented, a request for page 3F may return Sense Key
1276 		 * 5: Illegal Request, Sense Code 24: Invalid field in
1277 		 * CDB.
1278 		 */
1279 		if (!scsi_status_is_good(res))
1280 			res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1281 
1282 		/*
1283 		 * Third attempt: ask 255 bytes, as we did earlier.
1284 		 */
1285 		if (!scsi_status_is_good(res))
1286 			res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1287 					       &data, NULL);
1288 	}
1289 
1290 	if (!scsi_status_is_good(res)) {
1291 		printk(KERN_WARNING
1292 		       "%s: test WP failed, assume Write Enabled\n", diskname);
1293 	} else {
1294 		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
1295 		set_disk_ro(sdkp->disk, sdkp->write_prot);
1296 		printk(KERN_NOTICE "%s: Write Protect is %s\n", diskname,
1297 		       sdkp->write_prot ? "on" : "off");
1298 		printk(KERN_DEBUG "%s: Mode Sense: %02x %02x %02x %02x\n",
1299 		       diskname, buffer[0], buffer[1], buffer[2], buffer[3]);
1300 	}
1301 }
1302 
1303 /*
1304  * sd_read_cache_type - called only from sd_revalidate_disk()
1305  * called with buffer of length SD_BUF_SIZE
1306  */
1307 static void
1308 sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
1309 		   unsigned char *buffer)
1310 {
1311 	int len = 0, res;
1312 	struct scsi_device *sdp = sdkp->device;
1313 
1314 	int dbd;
1315 	int modepage;
1316 	struct scsi_mode_data data;
1317 	struct scsi_sense_hdr sshdr;
1318 
1319 	if (sdp->skip_ms_page_8)
1320 		goto defaults;
1321 
1322 	if (sdp->type == TYPE_RBC) {
1323 		modepage = 6;
1324 		dbd = 8;
1325 	} else {
1326 		modepage = 8;
1327 		dbd = 0;
1328 	}
1329 
1330 	/* cautiously ask */
1331 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, 4, &data, &sshdr);
1332 
1333 	if (!scsi_status_is_good(res))
1334 		goto bad_sense;
1335 
1336 	/* that went OK, now ask for the proper length */
1337 	len = data.length;
1338 
1339 	/*
1340 	 * We're only interested in the first three bytes, actually.
1341 	 * But the data cache page is defined for the first 20.
1342 	 */
1343 	if (len < 3)
1344 		goto bad_sense;
1345 	if (len > 20)
1346 		len = 20;
1347 
1348 	/* Take headers and block descriptors into account */
1349 	len += data.header_length + data.block_descriptor_length;
1350 	if (len > SD_BUF_SIZE)
1351 		goto bad_sense;
1352 
1353 	/* Get the data */
1354 	res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
1355 
1356 	if (scsi_status_is_good(res)) {
1357 		const char *types[] = {
1358 			"write through", "none", "write back",
1359 			"write back, no read (daft)"
1360 		};
1361 		int ct = 0;
1362 		int offset = data.header_length + data.block_descriptor_length;
1363 
1364 		if (offset >= SD_BUF_SIZE - 2) {
1365 			printk(KERN_ERR "%s: malformed MODE SENSE response",
1366 				diskname);
1367 			goto defaults;
1368 		}
1369 
1370 		if ((buffer[offset] & 0x3f) != modepage) {
1371 			printk(KERN_ERR "%s: got wrong page\n", diskname);
1372 			goto defaults;
1373 		}
1374 
1375 		if (modepage == 8) {
1376 			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
1377 			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
1378 		} else {
1379 			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
1380 			sdkp->RCD = 0;
1381 		}
1382 
1383 		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
1384 		if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
1385 			printk(KERN_NOTICE "SCSI device %s: uses "
1386 			       "READ/WRITE(6), disabling FUA\n", diskname);
1387 			sdkp->DPOFUA = 0;
1388 		}
1389 
1390 		ct =  sdkp->RCD + 2*sdkp->WCE;
1391 
1392 		printk(KERN_NOTICE "SCSI device %s: drive cache: %s%s\n",
1393 		       diskname, types[ct],
1394 		       sdkp->DPOFUA ? " w/ FUA" : "");
1395 
1396 		return;
1397 	}
1398 
1399 bad_sense:
1400 	if (scsi_sense_valid(&sshdr) &&
1401 	    sshdr.sense_key == ILLEGAL_REQUEST &&
1402 	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
1403 		printk(KERN_NOTICE "%s: cache data unavailable\n",
1404 		       diskname);	/* Invalid field in CDB */
1405 	else
1406 		printk(KERN_ERR "%s: asking for cache data failed\n",
1407 		       diskname);
1408 
1409 defaults:
1410 	printk(KERN_ERR "%s: assuming drive cache: write through\n",
1411 	       diskname);
1412 	sdkp->WCE = 0;
1413 	sdkp->RCD = 0;
1414 	sdkp->DPOFUA = 0;
1415 }
1416 
1417 /**
1418  *	sd_revalidate_disk - called the first time a new disk is seen,
1419  *	performs disk spin up, read_capacity, etc.
1420  *	@disk: struct gendisk we care about
1421  **/
1422 static int sd_revalidate_disk(struct gendisk *disk)
1423 {
1424 	struct scsi_disk *sdkp = scsi_disk(disk);
1425 	struct scsi_device *sdp = sdkp->device;
1426 	unsigned char *buffer;
1427 	unsigned ordered;
1428 
1429 	SCSI_LOG_HLQUEUE(3, printk("sd_revalidate_disk: disk=%s\n", disk->disk_name));
1430 
1431 	/*
1432 	 * If the device is offline, don't try and read capacity or any
1433 	 * of the other niceties.
1434 	 */
1435 	if (!scsi_device_online(sdp))
1436 		goto out;
1437 
1438 	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL | __GFP_DMA);
1439 	if (!buffer) {
1440 		printk(KERN_WARNING "(sd_revalidate_disk:) Memory allocation "
1441 		       "failure.\n");
1442 		goto out;
1443 	}
1444 
1445 	/* defaults, until the device tells us otherwise */
1446 	sdp->sector_size = 512;
1447 	sdkp->capacity = 0;
1448 	sdkp->media_present = 1;
1449 	sdkp->write_prot = 0;
1450 	sdkp->WCE = 0;
1451 	sdkp->RCD = 0;
1452 
1453 	sd_spinup_disk(sdkp, disk->disk_name);
1454 
1455 	/*
1456 	 * Without media there is no reason to ask; moreover, some devices
1457 	 * react badly if we do.
1458 	 */
1459 	if (sdkp->media_present) {
1460 		sd_read_capacity(sdkp, disk->disk_name, buffer);
1461 		sd_read_write_protect_flag(sdkp, disk->disk_name, buffer);
1462 		sd_read_cache_type(sdkp, disk->disk_name, buffer);
1463 	}
1464 
1465 	/*
1466 	 * We now have all cache related info, determine how we deal
1467 	 * with ordered requests.  Note that as the current SCSI
1468 	 * dispatch function can alter request order, we cannot use
1469 	 * QUEUE_ORDERED_TAG_* even when ordered tag is supported.
1470 	 */
1471 	if (sdkp->WCE)
1472 		ordered = sdkp->DPOFUA
1473 			? QUEUE_ORDERED_DRAIN_FUA : QUEUE_ORDERED_DRAIN_FLUSH;
1474 	else
1475 		ordered = QUEUE_ORDERED_DRAIN;
1476 
1477 	blk_queue_ordered(sdkp->disk->queue, ordered, sd_prepare_flush);
1478 
1479 	set_capacity(disk, sdkp->capacity);
1480 	kfree(buffer);
1481 
1482  out:
1483 	return 0;
1484 }
1485 
1486 /**
1487  *	sd_probe - called during driver initialization and whenever a
1488  *	new scsi device is attached to the system. It is called once
1489  *	for each scsi device (not just disks) present.
1490  *	@dev: pointer to device object
1491  *
1492  *	Returns 0 if successful (or not interested in this scsi device
1493  *	(e.g. scanner)); 1 when there is an error.
1494  *
1495  *	Note: this function is invoked from the scsi mid-level.
1496  *	This function sets up the mapping between a given
1497  *	<host,channel,id,lun> (found in sdp) and new device name
1498  *	(e.g. /dev/sda). More precisely it is the block device major
1499  *	and minor number that is chosen here.
1500  *
1501  *	Assume sd_attach is not re-entrant (for time being)
1502  *	Also think about sd_attach() and sd_remove() running coincidentally.
1503  **/
1504 static int sd_probe(struct device *dev)
1505 {
1506 	struct scsi_device *sdp = to_scsi_device(dev);
1507 	struct scsi_disk *sdkp;
1508 	struct gendisk *gd;
1509 	u32 index;
1510 	int error;
1511 
1512 	error = -ENODEV;
1513 	if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
1514 		goto out;
1515 
1516 	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
1517 					"sd_attach\n"));
1518 
1519 	error = -ENOMEM;
1520 	sdkp = kmalloc(sizeof(*sdkp), GFP_KERNEL);
1521 	if (!sdkp)
1522 		goto out;
1523 
1524 	memset (sdkp, 0, sizeof(*sdkp));
1525 	kref_init(&sdkp->kref);
1526 
1527 	gd = alloc_disk(16);
1528 	if (!gd)
1529 		goto out_free;
1530 
1531 	if (!idr_pre_get(&sd_index_idr, GFP_KERNEL))
1532 		goto out_put;
1533 
1534 	spin_lock(&sd_index_lock);
1535 	error = idr_get_new(&sd_index_idr, NULL, &index);
1536 	spin_unlock(&sd_index_lock);
1537 
1538 	if (index >= SD_MAX_DISKS)
1539 		error = -EBUSY;
1540 	if (error)
1541 		goto out_put;
1542 
1543 	get_device(&sdp->sdev_gendev);
1544 	sdkp->device = sdp;
1545 	sdkp->driver = &sd_template;
1546 	sdkp->disk = gd;
1547 	sdkp->index = index;
1548 	sdkp->openers = 0;
1549 
1550 	if (!sdp->timeout) {
1551 		if (sdp->type != TYPE_MOD)
1552 			sdp->timeout = SD_TIMEOUT;
1553 		else
1554 			sdp->timeout = SD_MOD_TIMEOUT;
1555 	}
1556 
1557 	gd->major = sd_major((index & 0xf0) >> 4);
1558 	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
1559 	gd->minors = 16;
1560 	gd->fops = &sd_fops;
1561 
1562 	if (index < 26) {
1563 		sprintf(gd->disk_name, "sd%c", 'a' + index % 26);
1564 	} else if (index < (26 + 1) * 26) {
1565 		sprintf(gd->disk_name, "sd%c%c",
1566 			'a' + index / 26 - 1,'a' + index % 26);
1567 	} else {
1568 		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
1569 		const unsigned int m2 = (index / 26 - 1) % 26;
1570 		const unsigned int m3 =  index % 26;
1571 		sprintf(gd->disk_name, "sd%c%c%c",
1572 			'a' + m1, 'a' + m2, 'a' + m3);
1573 	}
1574 
1575 	strcpy(gd->devfs_name, sdp->devfs_name);
1576 
1577 	gd->private_data = &sdkp->driver;
1578 	gd->queue = sdkp->device->request_queue;
1579 
1580 	sd_revalidate_disk(gd);
1581 
1582 	gd->driverfs_dev = &sdp->sdev_gendev;
1583 	gd->flags = GENHD_FL_DRIVERFS;
1584 	if (sdp->removable)
1585 		gd->flags |= GENHD_FL_REMOVABLE;
1586 
1587 	dev_set_drvdata(dev, sdkp);
1588 	add_disk(gd);
1589 
1590 	sdev_printk(KERN_NOTICE, sdp, "Attached scsi %sdisk %s\n",
1591 		    sdp->removable ? "removable " : "", gd->disk_name);
1592 
1593 	return 0;
1594 
1595 out_put:
1596 	put_disk(gd);
1597 out_free:
1598 	kfree(sdkp);
1599 out:
1600 	return error;
1601 }
1602 
1603 /**
1604  *	sd_remove - called whenever a scsi disk (previously recognized by
1605  *	sd_probe) is detached from the system. It is called (potentially
1606  *	multiple times) during sd module unload.
1607  *	@sdp: pointer to mid level scsi device object
1608  *
1609  *	Note: this function is invoked from the scsi mid-level.
1610  *	This function potentially frees up a device name (e.g. /dev/sdc)
1611  *	that could be re-used by a subsequent sd_probe().
1612  *	This function is not called when the built-in sd driver is "exit-ed".
1613  **/
1614 static int sd_remove(struct device *dev)
1615 {
1616 	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1617 
1618 	del_gendisk(sdkp->disk);
1619 	sd_shutdown(dev);
1620 
1621 	mutex_lock(&sd_ref_mutex);
1622 	dev_set_drvdata(dev, NULL);
1623 	kref_put(&sdkp->kref, scsi_disk_release);
1624 	mutex_unlock(&sd_ref_mutex);
1625 
1626 	return 0;
1627 }
1628 
1629 /**
1630  *	scsi_disk_release - Called to free the scsi_disk structure
1631  *	@kref: pointer to embedded kref
1632  *
1633  *	sd_ref_mutex must be held entering this routine.  Because it is
1634  *	called on last put, you should always use the scsi_disk_get()
1635  *	scsi_disk_put() helpers which manipulate the semaphore directly
1636  *	and never do a direct kref_put().
1637  **/
1638 static void scsi_disk_release(struct kref *kref)
1639 {
1640 	struct scsi_disk *sdkp = to_scsi_disk(kref);
1641 	struct gendisk *disk = sdkp->disk;
1642 
1643 	spin_lock(&sd_index_lock);
1644 	idr_remove(&sd_index_idr, sdkp->index);
1645 	spin_unlock(&sd_index_lock);
1646 
1647 	disk->private_data = NULL;
1648 	put_disk(disk);
1649 	put_device(&sdkp->device->sdev_gendev);
1650 
1651 	kfree(sdkp);
1652 }
1653 
1654 /*
1655  * Send a SYNCHRONIZE CACHE instruction down to the device through
1656  * the normal SCSI command structure.  Wait for the command to
1657  * complete.
1658  */
1659 static void sd_shutdown(struct device *dev)
1660 {
1661 	struct scsi_device *sdp = to_scsi_device(dev);
1662 	struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1663 
1664 	if (!sdkp)
1665 		return;         /* this can happen */
1666 
1667 	if (sdkp->WCE) {
1668 		printk(KERN_NOTICE "Synchronizing SCSI cache for disk %s: \n",
1669 				sdkp->disk->disk_name);
1670 		sd_sync_cache(sdp);
1671 	}
1672 	scsi_disk_put(sdkp);
1673 }
1674 
1675 /**
1676  *	init_sd - entry point for this driver (both when built in or when
1677  *	a module).
1678  *
1679  *	Note: this function registers this driver with the scsi mid-level.
1680  **/
1681 static int __init init_sd(void)
1682 {
1683 	int majors = 0, i;
1684 
1685 	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
1686 
1687 	for (i = 0; i < SD_MAJORS; i++)
1688 		if (register_blkdev(sd_major(i), "sd") == 0)
1689 			majors++;
1690 
1691 	if (!majors)
1692 		return -ENODEV;
1693 
1694 	return scsi_register_driver(&sd_template.gendrv);
1695 }
1696 
1697 /**
1698  *	exit_sd - exit point for this driver (when it is a module).
1699  *
1700  *	Note: this function unregisters this driver from the scsi mid-level.
1701  **/
1702 static void __exit exit_sd(void)
1703 {
1704 	int i;
1705 
1706 	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
1707 
1708 	scsi_unregister_driver(&sd_template.gendrv);
1709 	for (i = 0; i < SD_MAJORS; i++)
1710 		unregister_blkdev(sd_major(i), "sd");
1711 }
1712 
1713 MODULE_LICENSE("GPL");
1714 MODULE_AUTHOR("Eric Youngdale");
1715 MODULE_DESCRIPTION("SCSI disk (sd) driver");
1716 
1717 module_init(init_sd);
1718 module_exit(exit_sd);
1719