]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - drivers/scsi/sd.c
Merge tag 'omap-for-v3.7-rc5/fixes-signed' of git://git.kernel.org/pub/scm/linux...
[~shefty/rdma-dev.git] / drivers / scsi / sd.c
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *      Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple 
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using 
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in 
17  *         sd_init and cleanups.
18  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *         not being read in sd_open. Fix problem where removable media 
20  *         could be ejected after sd_open.
21  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox 
23  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>: 
24  *         Support 32k/1M disks.
25  *
26  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *      Note: when the logging level is set by the user, it must be greater
32  *      than the level indicated above to trigger output.       
33  */
34
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <asm/uaccess.h>
55 #include <asm/unaligned.h>
56
57 #include <scsi/scsi.h>
58 #include <scsi/scsi_cmnd.h>
59 #include <scsi/scsi_dbg.h>
60 #include <scsi/scsi_device.h>
61 #include <scsi/scsi_driver.h>
62 #include <scsi/scsi_eh.h>
63 #include <scsi/scsi_host.h>
64 #include <scsi/scsi_ioctl.h>
65 #include <scsi/scsicam.h>
66
67 #include "sd.h"
68 #include "scsi_priv.h"
69 #include "scsi_logging.h"
70
71 MODULE_AUTHOR("Eric Youngdale");
72 MODULE_DESCRIPTION("SCSI disk (sd) driver");
73 MODULE_LICENSE("GPL");
74
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
94
95 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
96 #define SD_MINORS       16
97 #else
98 #define SD_MINORS       0
99 #endif
100
101 static void sd_config_discard(struct scsi_disk *, unsigned int);
102 static int  sd_revalidate_disk(struct gendisk *);
103 static void sd_unlock_native_capacity(struct gendisk *disk);
104 static int  sd_probe(struct device *);
105 static int  sd_remove(struct device *);
106 static void sd_shutdown(struct device *);
107 static int sd_suspend(struct device *, pm_message_t state);
108 static int sd_resume(struct device *);
109 static void sd_rescan(struct device *);
110 static int sd_done(struct scsi_cmnd *);
111 static int sd_eh_action(struct scsi_cmnd *, unsigned char *, int, int);
112 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
113 static void scsi_disk_release(struct device *cdev);
114 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
115 static void sd_print_result(struct scsi_disk *, int);
116
117 static DEFINE_SPINLOCK(sd_index_lock);
118 static DEFINE_IDA(sd_index_ida);
119
120 /* This semaphore is used to mediate the 0->1 reference get in the
121  * face of object destruction (i.e. we can't allow a get on an
122  * object after last put) */
123 static DEFINE_MUTEX(sd_ref_mutex);
124
125 static struct kmem_cache *sd_cdb_cache;
126 static mempool_t *sd_cdb_pool;
127
128 static const char *sd_cache_types[] = {
129         "write through", "none", "write back",
130         "write back, no read (daft)"
131 };
132
133 static ssize_t
134 sd_store_cache_type(struct device *dev, struct device_attribute *attr,
135                     const char *buf, size_t count)
136 {
137         int i, ct = -1, rcd, wce, sp;
138         struct scsi_disk *sdkp = to_scsi_disk(dev);
139         struct scsi_device *sdp = sdkp->device;
140         char buffer[64];
141         char *buffer_data;
142         struct scsi_mode_data data;
143         struct scsi_sense_hdr sshdr;
144         int len;
145
146         if (sdp->type != TYPE_DISK)
147                 /* no cache control on RBC devices; theoretically they
148                  * can do it, but there's probably so many exceptions
149                  * it's not worth the risk */
150                 return -EINVAL;
151
152         for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
153                 len = strlen(sd_cache_types[i]);
154                 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
155                     buf[len] == '\n') {
156                         ct = i;
157                         break;
158                 }
159         }
160         if (ct < 0)
161                 return -EINVAL;
162         rcd = ct & 0x01 ? 1 : 0;
163         wce = ct & 0x02 ? 1 : 0;
164         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
165                             SD_MAX_RETRIES, &data, NULL))
166                 return -EINVAL;
167         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
168                   data.block_descriptor_length);
169         buffer_data = buffer + data.header_length +
170                 data.block_descriptor_length;
171         buffer_data[2] &= ~0x05;
172         buffer_data[2] |= wce << 2 | rcd;
173         sp = buffer_data[0] & 0x80 ? 1 : 0;
174
175         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
176                              SD_MAX_RETRIES, &data, &sshdr)) {
177                 if (scsi_sense_valid(&sshdr))
178                         sd_print_sense_hdr(sdkp, &sshdr);
179                 return -EINVAL;
180         }
181         revalidate_disk(sdkp->disk);
182         return count;
183 }
184
185 static ssize_t
186 sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
187                            const char *buf, size_t count)
188 {
189         struct scsi_disk *sdkp = to_scsi_disk(dev);
190         struct scsi_device *sdp = sdkp->device;
191
192         if (!capable(CAP_SYS_ADMIN))
193                 return -EACCES;
194
195         sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
196
197         return count;
198 }
199
200 static ssize_t
201 sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
202                        const char *buf, size_t count)
203 {
204         struct scsi_disk *sdkp = to_scsi_disk(dev);
205         struct scsi_device *sdp = sdkp->device;
206
207         if (!capable(CAP_SYS_ADMIN))
208                 return -EACCES;
209
210         if (sdp->type != TYPE_DISK)
211                 return -EINVAL;
212
213         sdp->allow_restart = simple_strtoul(buf, NULL, 10);
214
215         return count;
216 }
217
218 static ssize_t
219 sd_show_cache_type(struct device *dev, struct device_attribute *attr,
220                    char *buf)
221 {
222         struct scsi_disk *sdkp = to_scsi_disk(dev);
223         int ct = sdkp->RCD + 2*sdkp->WCE;
224
225         return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
226 }
227
228 static ssize_t
229 sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
230 {
231         struct scsi_disk *sdkp = to_scsi_disk(dev);
232
233         return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
234 }
235
236 static ssize_t
237 sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
238                           char *buf)
239 {
240         struct scsi_disk *sdkp = to_scsi_disk(dev);
241         struct scsi_device *sdp = sdkp->device;
242
243         return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
244 }
245
246 static ssize_t
247 sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
248                       char *buf)
249 {
250         struct scsi_disk *sdkp = to_scsi_disk(dev);
251
252         return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
253 }
254
255 static ssize_t
256 sd_show_protection_type(struct device *dev, struct device_attribute *attr,
257                         char *buf)
258 {
259         struct scsi_disk *sdkp = to_scsi_disk(dev);
260
261         return snprintf(buf, 20, "%u\n", sdkp->protection_type);
262 }
263
264 static ssize_t
265 sd_store_protection_type(struct device *dev, struct device_attribute *attr,
266                          const char *buf, size_t count)
267 {
268         struct scsi_disk *sdkp = to_scsi_disk(dev);
269         unsigned int val;
270         int err;
271
272         if (!capable(CAP_SYS_ADMIN))
273                 return -EACCES;
274
275         err = kstrtouint(buf, 10, &val);
276
277         if (err)
278                 return err;
279
280         if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
281                 sdkp->protection_type = val;
282
283         return count;
284 }
285
286 static ssize_t
287 sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
288                         char *buf)
289 {
290         struct scsi_disk *sdkp = to_scsi_disk(dev);
291         struct scsi_device *sdp = sdkp->device;
292         unsigned int dif, dix;
293
294         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
295         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
296
297         if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
298                 dif = 0;
299                 dix = 1;
300         }
301
302         if (!dif && !dix)
303                 return snprintf(buf, 20, "none\n");
304
305         return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
306 }
307
308 static ssize_t
309 sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
310                     char *buf)
311 {
312         struct scsi_disk *sdkp = to_scsi_disk(dev);
313
314         return snprintf(buf, 20, "%u\n", sdkp->ATO);
315 }
316
317 static ssize_t
318 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
319                           char *buf)
320 {
321         struct scsi_disk *sdkp = to_scsi_disk(dev);
322
323         return snprintf(buf, 20, "%u\n", sdkp->lbpme);
324 }
325
326 static const char *lbp_mode[] = {
327         [SD_LBP_FULL]           = "full",
328         [SD_LBP_UNMAP]          = "unmap",
329         [SD_LBP_WS16]           = "writesame_16",
330         [SD_LBP_WS10]           = "writesame_10",
331         [SD_LBP_ZERO]           = "writesame_zero",
332         [SD_LBP_DISABLE]        = "disabled",
333 };
334
335 static ssize_t
336 sd_show_provisioning_mode(struct device *dev, struct device_attribute *attr,
337                           char *buf)
338 {
339         struct scsi_disk *sdkp = to_scsi_disk(dev);
340
341         return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
342 }
343
344 static ssize_t
345 sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
346                            const char *buf, size_t count)
347 {
348         struct scsi_disk *sdkp = to_scsi_disk(dev);
349         struct scsi_device *sdp = sdkp->device;
350
351         if (!capable(CAP_SYS_ADMIN))
352                 return -EACCES;
353
354         if (sdp->type != TYPE_DISK)
355                 return -EINVAL;
356
357         if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
358                 sd_config_discard(sdkp, SD_LBP_UNMAP);
359         else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
360                 sd_config_discard(sdkp, SD_LBP_WS16);
361         else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
362                 sd_config_discard(sdkp, SD_LBP_WS10);
363         else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
364                 sd_config_discard(sdkp, SD_LBP_ZERO);
365         else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
366                 sd_config_discard(sdkp, SD_LBP_DISABLE);
367         else
368                 return -EINVAL;
369
370         return count;
371 }
372
373 static ssize_t
374 sd_show_max_medium_access_timeouts(struct device *dev,
375                                    struct device_attribute *attr, char *buf)
376 {
377         struct scsi_disk *sdkp = to_scsi_disk(dev);
378
379         return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
380 }
381
382 static ssize_t
383 sd_store_max_medium_access_timeouts(struct device *dev,
384                                     struct device_attribute *attr,
385                                     const char *buf, size_t count)
386 {
387         struct scsi_disk *sdkp = to_scsi_disk(dev);
388         int err;
389
390         if (!capable(CAP_SYS_ADMIN))
391                 return -EACCES;
392
393         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
394
395         return err ? err : count;
396 }
397
398 static struct device_attribute sd_disk_attrs[] = {
399         __ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
400                sd_store_cache_type),
401         __ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
402         __ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
403                sd_store_allow_restart),
404         __ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
405                sd_store_manage_start_stop),
406         __ATTR(protection_type, S_IRUGO|S_IWUSR, sd_show_protection_type,
407                sd_store_protection_type),
408         __ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
409         __ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
410         __ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
411         __ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
412                sd_store_provisioning_mode),
413         __ATTR(max_medium_access_timeouts, S_IRUGO|S_IWUSR,
414                sd_show_max_medium_access_timeouts,
415                sd_store_max_medium_access_timeouts),
416         __ATTR_NULL,
417 };
418
419 static struct class sd_disk_class = {
420         .name           = "scsi_disk",
421         .owner          = THIS_MODULE,
422         .dev_release    = scsi_disk_release,
423         .dev_attrs      = sd_disk_attrs,
424 };
425
426 static struct scsi_driver sd_template = {
427         .owner                  = THIS_MODULE,
428         .gendrv = {
429                 .name           = "sd",
430                 .probe          = sd_probe,
431                 .remove         = sd_remove,
432                 .suspend        = sd_suspend,
433                 .resume         = sd_resume,
434                 .shutdown       = sd_shutdown,
435         },
436         .rescan                 = sd_rescan,
437         .done                   = sd_done,
438         .eh_action              = sd_eh_action,
439 };
440
441 /*
442  * Device no to disk mapping:
443  * 
444  *       major         disc2     disc  p1
445  *   |............|.............|....|....| <- dev_t
446  *    31        20 19          8 7  4 3  0
447  * 
448  * Inside a major, we have 16k disks, however mapped non-
449  * contiguously. The first 16 disks are for major0, the next
450  * ones with major1, ... Disk 256 is for major0 again, disk 272 
451  * for major1, ... 
452  * As we stay compatible with our numbering scheme, we can reuse 
453  * the well-know SCSI majors 8, 65--71, 136--143.
454  */
455 static int sd_major(int major_idx)
456 {
457         switch (major_idx) {
458         case 0:
459                 return SCSI_DISK0_MAJOR;
460         case 1 ... 7:
461                 return SCSI_DISK1_MAJOR + major_idx - 1;
462         case 8 ... 15:
463                 return SCSI_DISK8_MAJOR + major_idx - 8;
464         default:
465                 BUG();
466                 return 0;       /* shut up gcc */
467         }
468 }
469
470 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
471 {
472         struct scsi_disk *sdkp = NULL;
473
474         if (disk->private_data) {
475                 sdkp = scsi_disk(disk);
476                 if (scsi_device_get(sdkp->device) == 0)
477                         get_device(&sdkp->dev);
478                 else
479                         sdkp = NULL;
480         }
481         return sdkp;
482 }
483
484 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
485 {
486         struct scsi_disk *sdkp;
487
488         mutex_lock(&sd_ref_mutex);
489         sdkp = __scsi_disk_get(disk);
490         mutex_unlock(&sd_ref_mutex);
491         return sdkp;
492 }
493
494 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
495 {
496         struct scsi_disk *sdkp;
497
498         mutex_lock(&sd_ref_mutex);
499         sdkp = dev_get_drvdata(dev);
500         if (sdkp)
501                 sdkp = __scsi_disk_get(sdkp->disk);
502         mutex_unlock(&sd_ref_mutex);
503         return sdkp;
504 }
505
506 static void scsi_disk_put(struct scsi_disk *sdkp)
507 {
508         struct scsi_device *sdev = sdkp->device;
509
510         mutex_lock(&sd_ref_mutex);
511         put_device(&sdkp->dev);
512         scsi_device_put(sdev);
513         mutex_unlock(&sd_ref_mutex);
514 }
515
516 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
517 {
518         unsigned int prot_op = SCSI_PROT_NORMAL;
519         unsigned int dix = scsi_prot_sg_count(scmd);
520
521         if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
522                 if (dif && dix)
523                         prot_op = SCSI_PROT_READ_PASS;
524                 else if (dif && !dix)
525                         prot_op = SCSI_PROT_READ_STRIP;
526                 else if (!dif && dix)
527                         prot_op = SCSI_PROT_READ_INSERT;
528         } else {
529                 if (dif && dix)
530                         prot_op = SCSI_PROT_WRITE_PASS;
531                 else if (dif && !dix)
532                         prot_op = SCSI_PROT_WRITE_INSERT;
533                 else if (!dif && dix)
534                         prot_op = SCSI_PROT_WRITE_STRIP;
535         }
536
537         scsi_set_prot_op(scmd, prot_op);
538         scsi_set_prot_type(scmd, dif);
539 }
540
541 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
542 {
543         struct request_queue *q = sdkp->disk->queue;
544         unsigned int logical_block_size = sdkp->device->sector_size;
545         unsigned int max_blocks = 0;
546
547         q->limits.discard_zeroes_data = sdkp->lbprz;
548         q->limits.discard_alignment = sdkp->unmap_alignment *
549                 logical_block_size;
550         q->limits.discard_granularity =
551                 max(sdkp->physical_block_size,
552                     sdkp->unmap_granularity * logical_block_size);
553
554         sdkp->provisioning_mode = mode;
555
556         switch (mode) {
557
558         case SD_LBP_DISABLE:
559                 q->limits.max_discard_sectors = 0;
560                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
561                 return;
562
563         case SD_LBP_UNMAP:
564                 max_blocks = min_not_zero(sdkp->max_unmap_blocks, 0xffffffff);
565                 break;
566
567         case SD_LBP_WS16:
568                 max_blocks = min_not_zero(sdkp->max_ws_blocks, 0xffffffff);
569                 break;
570
571         case SD_LBP_WS10:
572                 max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
573                 break;
574
575         case SD_LBP_ZERO:
576                 max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
577                 q->limits.discard_zeroes_data = 1;
578                 break;
579         }
580
581         q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
582         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
583 }
584
585 /**
586  * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
587  * @sdp: scsi device to operate one
588  * @rq: Request to prepare
589  *
590  * Will issue either UNMAP or WRITE SAME(16) depending on preference
591  * indicated by target device.
592  **/
593 static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
594 {
595         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
596         struct bio *bio = rq->bio;
597         sector_t sector = bio->bi_sector;
598         unsigned int nr_sectors = bio_sectors(bio);
599         unsigned int len;
600         int ret;
601         char *buf;
602         struct page *page;
603
604         if (sdkp->device->sector_size == 4096) {
605                 sector >>= 3;
606                 nr_sectors >>= 3;
607         }
608
609         rq->timeout = SD_TIMEOUT;
610
611         memset(rq->cmd, 0, rq->cmd_len);
612
613         page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
614         if (!page)
615                 return BLKPREP_DEFER;
616
617         switch (sdkp->provisioning_mode) {
618         case SD_LBP_UNMAP:
619                 buf = page_address(page);
620
621                 rq->cmd_len = 10;
622                 rq->cmd[0] = UNMAP;
623                 rq->cmd[8] = 24;
624
625                 put_unaligned_be16(6 + 16, &buf[0]);
626                 put_unaligned_be16(16, &buf[2]);
627                 put_unaligned_be64(sector, &buf[8]);
628                 put_unaligned_be32(nr_sectors, &buf[16]);
629
630                 len = 24;
631                 break;
632
633         case SD_LBP_WS16:
634                 rq->cmd_len = 16;
635                 rq->cmd[0] = WRITE_SAME_16;
636                 rq->cmd[1] = 0x8; /* UNMAP */
637                 put_unaligned_be64(sector, &rq->cmd[2]);
638                 put_unaligned_be32(nr_sectors, &rq->cmd[10]);
639
640                 len = sdkp->device->sector_size;
641                 break;
642
643         case SD_LBP_WS10:
644         case SD_LBP_ZERO:
645                 rq->cmd_len = 10;
646                 rq->cmd[0] = WRITE_SAME;
647                 if (sdkp->provisioning_mode == SD_LBP_WS10)
648                         rq->cmd[1] = 0x8; /* UNMAP */
649                 put_unaligned_be32(sector, &rq->cmd[2]);
650                 put_unaligned_be16(nr_sectors, &rq->cmd[7]);
651
652                 len = sdkp->device->sector_size;
653                 break;
654
655         default:
656                 ret = BLKPREP_KILL;
657                 goto out;
658         }
659
660         blk_add_request_payload(rq, page, len);
661         ret = scsi_setup_blk_pc_cmnd(sdp, rq);
662         rq->buffer = page_address(page);
663
664 out:
665         if (ret != BLKPREP_OK) {
666                 __free_page(page);
667                 rq->buffer = NULL;
668         }
669         return ret;
670 }
671
672 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
673 {
674         rq->timeout = SD_FLUSH_TIMEOUT;
675         rq->retries = SD_MAX_RETRIES;
676         rq->cmd[0] = SYNCHRONIZE_CACHE;
677         rq->cmd_len = 10;
678
679         return scsi_setup_blk_pc_cmnd(sdp, rq);
680 }
681
682 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
683 {
684         if (rq->cmd_flags & REQ_DISCARD) {
685                 free_page((unsigned long)rq->buffer);
686                 rq->buffer = NULL;
687         }
688 }
689
690 /**
691  *      sd_prep_fn - build a scsi (read or write) command from
692  *      information in the request structure.
693  *      @SCpnt: pointer to mid-level's per scsi command structure that
694  *      contains request and into which the scsi command is written
695  *
696  *      Returns 1 if successful and 0 if error (or cannot be done now).
697  **/
698 static int sd_prep_fn(struct request_queue *q, struct request *rq)
699 {
700         struct scsi_cmnd *SCpnt;
701         struct scsi_device *sdp = q->queuedata;
702         struct gendisk *disk = rq->rq_disk;
703         struct scsi_disk *sdkp;
704         sector_t block = blk_rq_pos(rq);
705         sector_t threshold;
706         unsigned int this_count = blk_rq_sectors(rq);
707         int ret, host_dif;
708         unsigned char protect;
709
710         /*
711          * Discard request come in as REQ_TYPE_FS but we turn them into
712          * block PC requests to make life easier.
713          */
714         if (rq->cmd_flags & REQ_DISCARD) {
715                 ret = scsi_setup_discard_cmnd(sdp, rq);
716                 goto out;
717         } else if (rq->cmd_flags & REQ_FLUSH) {
718                 ret = scsi_setup_flush_cmnd(sdp, rq);
719                 goto out;
720         } else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
721                 ret = scsi_setup_blk_pc_cmnd(sdp, rq);
722                 goto out;
723         } else if (rq->cmd_type != REQ_TYPE_FS) {
724                 ret = BLKPREP_KILL;
725                 goto out;
726         }
727         ret = scsi_setup_fs_cmnd(sdp, rq);
728         if (ret != BLKPREP_OK)
729                 goto out;
730         SCpnt = rq->special;
731         sdkp = scsi_disk(disk);
732
733         /* from here on until we're complete, any goto out
734          * is used for a killable error condition */
735         ret = BLKPREP_KILL;
736
737         SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
738                                         "sd_prep_fn: block=%llu, "
739                                         "count=%d\n",
740                                         (unsigned long long)block,
741                                         this_count));
742
743         if (!sdp || !scsi_device_online(sdp) ||
744             block + blk_rq_sectors(rq) > get_capacity(disk)) {
745                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
746                                                 "Finishing %u sectors\n",
747                                                 blk_rq_sectors(rq)));
748                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
749                                                 "Retry with 0x%p\n", SCpnt));
750                 goto out;
751         }
752
753         if (sdp->changed) {
754                 /*
755                  * quietly refuse to do anything to a changed disc until 
756                  * the changed bit has been reset
757                  */
758                 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
759                 goto out;
760         }
761
762         /*
763          * Some SD card readers can't handle multi-sector accesses which touch
764          * the last one or two hardware sectors.  Split accesses as needed.
765          */
766         threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
767                 (sdp->sector_size / 512);
768
769         if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
770                 if (block < threshold) {
771                         /* Access up to the threshold but not beyond */
772                         this_count = threshold - block;
773                 } else {
774                         /* Access only a single hardware sector */
775                         this_count = sdp->sector_size / 512;
776                 }
777         }
778
779         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
780                                         (unsigned long long)block));
781
782         /*
783          * If we have a 1K hardware sectorsize, prevent access to single
784          * 512 byte sectors.  In theory we could handle this - in fact
785          * the scsi cdrom driver must be able to handle this because
786          * we typically use 1K blocksizes, and cdroms typically have
787          * 2K hardware sectorsizes.  Of course, things are simpler
788          * with the cdrom, since it is read-only.  For performance
789          * reasons, the filesystems should be able to handle this
790          * and not force the scsi disk driver to use bounce buffers
791          * for this.
792          */
793         if (sdp->sector_size == 1024) {
794                 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
795                         scmd_printk(KERN_ERR, SCpnt,
796                                     "Bad block number requested\n");
797                         goto out;
798                 } else {
799                         block = block >> 1;
800                         this_count = this_count >> 1;
801                 }
802         }
803         if (sdp->sector_size == 2048) {
804                 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
805                         scmd_printk(KERN_ERR, SCpnt,
806                                     "Bad block number requested\n");
807                         goto out;
808                 } else {
809                         block = block >> 2;
810                         this_count = this_count >> 2;
811                 }
812         }
813         if (sdp->sector_size == 4096) {
814                 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
815                         scmd_printk(KERN_ERR, SCpnt,
816                                     "Bad block number requested\n");
817                         goto out;
818                 } else {
819                         block = block >> 3;
820                         this_count = this_count >> 3;
821                 }
822         }
823         if (rq_data_dir(rq) == WRITE) {
824                 if (!sdp->writeable) {
825                         goto out;
826                 }
827                 SCpnt->cmnd[0] = WRITE_6;
828                 SCpnt->sc_data_direction = DMA_TO_DEVICE;
829
830                 if (blk_integrity_rq(rq))
831                         sd_dif_prepare(rq, block, sdp->sector_size);
832
833         } else if (rq_data_dir(rq) == READ) {
834                 SCpnt->cmnd[0] = READ_6;
835                 SCpnt->sc_data_direction = DMA_FROM_DEVICE;
836         } else {
837                 scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
838                 goto out;
839         }
840
841         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
842                                         "%s %d/%u 512 byte blocks.\n",
843                                         (rq_data_dir(rq) == WRITE) ?
844                                         "writing" : "reading", this_count,
845                                         blk_rq_sectors(rq)));
846
847         /* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
848         host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
849         if (host_dif)
850                 protect = 1 << 5;
851         else
852                 protect = 0;
853
854         if (host_dif == SD_DIF_TYPE2_PROTECTION) {
855                 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
856
857                 if (unlikely(SCpnt->cmnd == NULL)) {
858                         ret = BLKPREP_DEFER;
859                         goto out;
860                 }
861
862                 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
863                 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
864                 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
865                 SCpnt->cmnd[7] = 0x18;
866                 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
867                 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
868
869                 /* LBA */
870                 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
871                 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
872                 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
873                 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
874                 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
875                 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
876                 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
877                 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
878
879                 /* Expected Indirect LBA */
880                 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
881                 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
882                 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
883                 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
884
885                 /* Transfer length */
886                 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
887                 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
888                 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
889                 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
890         } else if (block > 0xffffffff) {
891                 SCpnt->cmnd[0] += READ_16 - READ_6;
892                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
893                 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
894                 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
895                 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
896                 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
897                 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
898                 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
899                 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
900                 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
901                 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
902                 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
903                 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
904                 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
905                 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
906         } else if ((this_count > 0xff) || (block > 0x1fffff) ||
907                    scsi_device_protection(SCpnt->device) ||
908                    SCpnt->device->use_10_for_rw) {
909                 if (this_count > 0xffff)
910                         this_count = 0xffff;
911
912                 SCpnt->cmnd[0] += READ_10 - READ_6;
913                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
914                 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
915                 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
916                 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
917                 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
918                 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
919                 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
920                 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
921         } else {
922                 if (unlikely(rq->cmd_flags & REQ_FUA)) {
923                         /*
924                          * This happens only if this drive failed
925                          * 10byte rw command with ILLEGAL_REQUEST
926                          * during operation and thus turned off
927                          * use_10_for_rw.
928                          */
929                         scmd_printk(KERN_ERR, SCpnt,
930                                     "FUA write on READ/WRITE(6) drive\n");
931                         goto out;
932                 }
933
934                 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
935                 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
936                 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
937                 SCpnt->cmnd[4] = (unsigned char) this_count;
938                 SCpnt->cmnd[5] = 0;
939         }
940         SCpnt->sdb.length = this_count * sdp->sector_size;
941
942         /* If DIF or DIX is enabled, tell HBA how to handle request */
943         if (host_dif || scsi_prot_sg_count(SCpnt))
944                 sd_prot_op(SCpnt, host_dif);
945
946         /*
947          * We shouldn't disconnect in the middle of a sector, so with a dumb
948          * host adapter, it's safe to assume that we can at least transfer
949          * this many bytes between each connect / disconnect.
950          */
951         SCpnt->transfersize = sdp->sector_size;
952         SCpnt->underflow = this_count << 9;
953         SCpnt->allowed = SD_MAX_RETRIES;
954
955         /*
956          * This indicates that the command is ready from our end to be
957          * queued.
958          */
959         ret = BLKPREP_OK;
960  out:
961         return scsi_prep_return(q, rq, ret);
962 }
963
964 /**
965  *      sd_open - open a scsi disk device
966  *      @inode: only i_rdev member may be used
967  *      @filp: only f_mode and f_flags may be used
968  *
969  *      Returns 0 if successful. Returns a negated errno value in case 
970  *      of error.
971  *
972  *      Note: This can be called from a user context (e.g. fsck(1) )
973  *      or from within the kernel (e.g. as a result of a mount(1) ).
974  *      In the latter case @inode and @filp carry an abridged amount
975  *      of information as noted above.
976  *
977  *      Locking: called with bdev->bd_mutex held.
978  **/
979 static int sd_open(struct block_device *bdev, fmode_t mode)
980 {
981         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
982         struct scsi_device *sdev;
983         int retval;
984
985         if (!sdkp)
986                 return -ENXIO;
987
988         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
989
990         sdev = sdkp->device;
991
992         retval = scsi_autopm_get_device(sdev);
993         if (retval)
994                 goto error_autopm;
995
996         /*
997          * If the device is in error recovery, wait until it is done.
998          * If the device is offline, then disallow any access to it.
999          */
1000         retval = -ENXIO;
1001         if (!scsi_block_when_processing_errors(sdev))
1002                 goto error_out;
1003
1004         if (sdev->removable || sdkp->write_prot)
1005                 check_disk_change(bdev);
1006
1007         /*
1008          * If the drive is empty, just let the open fail.
1009          */
1010         retval = -ENOMEDIUM;
1011         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1012                 goto error_out;
1013
1014         /*
1015          * If the device has the write protect tab set, have the open fail
1016          * if the user expects to be able to write to the thing.
1017          */
1018         retval = -EROFS;
1019         if (sdkp->write_prot && (mode & FMODE_WRITE))
1020                 goto error_out;
1021
1022         /*
1023          * It is possible that the disk changing stuff resulted in
1024          * the device being taken offline.  If this is the case,
1025          * report this to the user, and don't pretend that the
1026          * open actually succeeded.
1027          */
1028         retval = -ENXIO;
1029         if (!scsi_device_online(sdev))
1030                 goto error_out;
1031
1032         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1033                 if (scsi_block_when_processing_errors(sdev))
1034                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1035         }
1036
1037         return 0;
1038
1039 error_out:
1040         scsi_autopm_put_device(sdev);
1041 error_autopm:
1042         scsi_disk_put(sdkp);
1043         return retval;  
1044 }
1045
1046 /**
1047  *      sd_release - invoked when the (last) close(2) is called on this
1048  *      scsi disk.
1049  *      @inode: only i_rdev member may be used
1050  *      @filp: only f_mode and f_flags may be used
1051  *
1052  *      Returns 0. 
1053  *
1054  *      Note: may block (uninterruptible) if error recovery is underway
1055  *      on this disk.
1056  *
1057  *      Locking: called with bdev->bd_mutex held.
1058  **/
1059 static int sd_release(struct gendisk *disk, fmode_t mode)
1060 {
1061         struct scsi_disk *sdkp = scsi_disk(disk);
1062         struct scsi_device *sdev = sdkp->device;
1063
1064         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1065
1066         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1067                 if (scsi_block_when_processing_errors(sdev))
1068                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1069         }
1070
1071         /*
1072          * XXX and what if there are packets in flight and this close()
1073          * XXX is followed by a "rmmod sd_mod"?
1074          */
1075
1076         scsi_autopm_put_device(sdev);
1077         scsi_disk_put(sdkp);
1078         return 0;
1079 }
1080
1081 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1082 {
1083         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1084         struct scsi_device *sdp = sdkp->device;
1085         struct Scsi_Host *host = sdp->host;
1086         int diskinfo[4];
1087
1088         /* default to most commonly used values */
1089         diskinfo[0] = 0x40;     /* 1 << 6 */
1090         diskinfo[1] = 0x20;     /* 1 << 5 */
1091         diskinfo[2] = sdkp->capacity >> 11;
1092         
1093         /* override with calculated, extended default, or driver values */
1094         if (host->hostt->bios_param)
1095                 host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1096         else
1097                 scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1098
1099         geo->heads = diskinfo[0];
1100         geo->sectors = diskinfo[1];
1101         geo->cylinders = diskinfo[2];
1102         return 0;
1103 }
1104
1105 /**
1106  *      sd_ioctl - process an ioctl
1107  *      @inode: only i_rdev/i_bdev members may be used
1108  *      @filp: only f_mode and f_flags may be used
1109  *      @cmd: ioctl command number
1110  *      @arg: this is third argument given to ioctl(2) system call.
1111  *      Often contains a pointer.
1112  *
1113  *      Returns 0 if successful (some ioctls return positive numbers on
1114  *      success as well). Returns a negated errno value in case of error.
1115  *
1116  *      Note: most ioctls are forward onto the block subsystem or further
1117  *      down in the scsi subsystem.
1118  **/
1119 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1120                     unsigned int cmd, unsigned long arg)
1121 {
1122         struct gendisk *disk = bdev->bd_disk;
1123         struct scsi_disk *sdkp = scsi_disk(disk);
1124         struct scsi_device *sdp = sdkp->device;
1125         void __user *p = (void __user *)arg;
1126         int error;
1127     
1128         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1129                                     "cmd=0x%x\n", disk->disk_name, cmd));
1130
1131         error = scsi_verify_blk_ioctl(bdev, cmd);
1132         if (error < 0)
1133                 return error;
1134
1135         /*
1136          * If we are in the middle of error recovery, don't let anyone
1137          * else try and use this device.  Also, if error recovery fails, it
1138          * may try and take the device offline, in which case all further
1139          * access to the device is prohibited.
1140          */
1141         error = scsi_nonblockable_ioctl(sdp, cmd, p,
1142                                         (mode & FMODE_NDELAY) != 0);
1143         if (!scsi_block_when_processing_errors(sdp) || !error)
1144                 goto out;
1145
1146         /*
1147          * Send SCSI addressing ioctls directly to mid level, send other
1148          * ioctls to block level and then onto mid level if they can't be
1149          * resolved.
1150          */
1151         switch (cmd) {
1152                 case SCSI_IOCTL_GET_IDLUN:
1153                 case SCSI_IOCTL_GET_BUS_NUMBER:
1154                         error = scsi_ioctl(sdp, cmd, p);
1155                         break;
1156                 default:
1157                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1158                         if (error != -ENOTTY)
1159                                 break;
1160                         error = scsi_ioctl(sdp, cmd, p);
1161                         break;
1162         }
1163 out:
1164         return error;
1165 }
1166
1167 static void set_media_not_present(struct scsi_disk *sdkp)
1168 {
1169         if (sdkp->media_present)
1170                 sdkp->device->changed = 1;
1171
1172         if (sdkp->device->removable) {
1173                 sdkp->media_present = 0;
1174                 sdkp->capacity = 0;
1175         }
1176 }
1177
1178 static int media_not_present(struct scsi_disk *sdkp,
1179                              struct scsi_sense_hdr *sshdr)
1180 {
1181         if (!scsi_sense_valid(sshdr))
1182                 return 0;
1183
1184         /* not invoked for commands that could return deferred errors */
1185         switch (sshdr->sense_key) {
1186         case UNIT_ATTENTION:
1187         case NOT_READY:
1188                 /* medium not present */
1189                 if (sshdr->asc == 0x3A) {
1190                         set_media_not_present(sdkp);
1191                         return 1;
1192                 }
1193         }
1194         return 0;
1195 }
1196
1197 /**
1198  *      sd_check_events - check media events
1199  *      @disk: kernel device descriptor
1200  *      @clearing: disk events currently being cleared
1201  *
1202  *      Returns mask of DISK_EVENT_*.
1203  *
1204  *      Note: this function is invoked from the block subsystem.
1205  **/
1206 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1207 {
1208         struct scsi_disk *sdkp = scsi_disk(disk);
1209         struct scsi_device *sdp = sdkp->device;
1210         struct scsi_sense_hdr *sshdr = NULL;
1211         int retval;
1212
1213         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1214
1215         /*
1216          * If the device is offline, don't send any commands - just pretend as
1217          * if the command failed.  If the device ever comes back online, we
1218          * can deal with it then.  It is only because of unrecoverable errors
1219          * that we would ever take a device offline in the first place.
1220          */
1221         if (!scsi_device_online(sdp)) {
1222                 set_media_not_present(sdkp);
1223                 goto out;
1224         }
1225
1226         /*
1227          * Using TEST_UNIT_READY enables differentiation between drive with
1228          * no cartridge loaded - NOT READY, drive with changed cartridge -
1229          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1230          *
1231          * Drives that auto spin down. eg iomega jaz 1G, will be started
1232          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1233          * sd_revalidate() is called.
1234          */
1235         retval = -ENODEV;
1236
1237         if (scsi_block_when_processing_errors(sdp)) {
1238                 retval = scsi_autopm_get_device(sdp);
1239                 if (retval)
1240                         goto out;
1241
1242                 sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1243                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1244                                               sshdr);
1245                 scsi_autopm_put_device(sdp);
1246         }
1247
1248         /* failed to execute TUR, assume media not present */
1249         if (host_byte(retval)) {
1250                 set_media_not_present(sdkp);
1251                 goto out;
1252         }
1253
1254         if (media_not_present(sdkp, sshdr))
1255                 goto out;
1256
1257         /*
1258          * For removable scsi disk we have to recognise the presence
1259          * of a disk in the drive.
1260          */
1261         if (!sdkp->media_present)
1262                 sdp->changed = 1;
1263         sdkp->media_present = 1;
1264 out:
1265         /*
1266          * sdp->changed is set under the following conditions:
1267          *
1268          *      Medium present state has changed in either direction.
1269          *      Device has indicated UNIT_ATTENTION.
1270          */
1271         kfree(sshdr);
1272         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1273         sdp->changed = 0;
1274         return retval;
1275 }
1276
1277 static int sd_sync_cache(struct scsi_disk *sdkp)
1278 {
1279         int retries, res;
1280         struct scsi_device *sdp = sdkp->device;
1281         struct scsi_sense_hdr sshdr;
1282
1283         if (!scsi_device_online(sdp))
1284                 return -ENODEV;
1285
1286
1287         for (retries = 3; retries > 0; --retries) {
1288                 unsigned char cmd[10] = { 0 };
1289
1290                 cmd[0] = SYNCHRONIZE_CACHE;
1291                 /*
1292                  * Leave the rest of the command zero to indicate
1293                  * flush everything.
1294                  */
1295                 res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1296                                        SD_FLUSH_TIMEOUT, SD_MAX_RETRIES, NULL);
1297                 if (res == 0)
1298                         break;
1299         }
1300
1301         if (res) {
1302                 sd_print_result(sdkp, res);
1303                 if (driver_byte(res) & DRIVER_SENSE)
1304                         sd_print_sense_hdr(sdkp, &sshdr);
1305         }
1306
1307         if (res)
1308                 return -EIO;
1309         return 0;
1310 }
1311
1312 static void sd_rescan(struct device *dev)
1313 {
1314         struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1315
1316         if (sdkp) {
1317                 revalidate_disk(sdkp->disk);
1318                 scsi_disk_put(sdkp);
1319         }
1320 }
1321
1322
1323 #ifdef CONFIG_COMPAT
1324 /* 
1325  * This gets directly called from VFS. When the ioctl 
1326  * is not recognized we go back to the other translation paths. 
1327  */
1328 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1329                            unsigned int cmd, unsigned long arg)
1330 {
1331         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1332         int ret;
1333
1334         ret = scsi_verify_blk_ioctl(bdev, cmd);
1335         if (ret < 0)
1336                 return ret;
1337
1338         /*
1339          * If we are in the middle of error recovery, don't let anyone
1340          * else try and use this device.  Also, if error recovery fails, it
1341          * may try and take the device offline, in which case all further
1342          * access to the device is prohibited.
1343          */
1344         if (!scsi_block_when_processing_errors(sdev))
1345                 return -ENODEV;
1346                
1347         if (sdev->host->hostt->compat_ioctl) {
1348                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1349
1350                 return ret;
1351         }
1352
1353         /* 
1354          * Let the static ioctl translation table take care of it.
1355          */
1356         return -ENOIOCTLCMD; 
1357 }
1358 #endif
1359
1360 static const struct block_device_operations sd_fops = {
1361         .owner                  = THIS_MODULE,
1362         .open                   = sd_open,
1363         .release                = sd_release,
1364         .ioctl                  = sd_ioctl,
1365         .getgeo                 = sd_getgeo,
1366 #ifdef CONFIG_COMPAT
1367         .compat_ioctl           = sd_compat_ioctl,
1368 #endif
1369         .check_events           = sd_check_events,
1370         .revalidate_disk        = sd_revalidate_disk,
1371         .unlock_native_capacity = sd_unlock_native_capacity,
1372 };
1373
1374 /**
1375  *      sd_eh_action - error handling callback
1376  *      @scmd:          sd-issued command that has failed
1377  *      @eh_cmnd:       The command that was sent during error handling
1378  *      @eh_cmnd_len:   Length of eh_cmnd in bytes
1379  *      @eh_disp:       The recovery disposition suggested by the midlayer
1380  *
1381  *      This function is called by the SCSI midlayer upon completion of
1382  *      an error handling command (TEST UNIT READY, START STOP UNIT,
1383  *      etc.) The command sent to the device by the error handler is
1384  *      stored in eh_cmnd. The result of sending the eh command is
1385  *      passed in eh_disp.
1386  **/
1387 static int sd_eh_action(struct scsi_cmnd *scmd, unsigned char *eh_cmnd,
1388                         int eh_cmnd_len, int eh_disp)
1389 {
1390         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1391
1392         if (!scsi_device_online(scmd->device) ||
1393             !scsi_medium_access_command(scmd))
1394                 return eh_disp;
1395
1396         /*
1397          * The device has timed out executing a medium access command.
1398          * However, the TEST UNIT READY command sent during error
1399          * handling completed successfully. Either the device is in the
1400          * process of recovering or has it suffered an internal failure
1401          * that prevents access to the storage medium.
1402          */
1403         if (host_byte(scmd->result) == DID_TIME_OUT && eh_disp == SUCCESS &&
1404             eh_cmnd_len && eh_cmnd[0] == TEST_UNIT_READY)
1405                 sdkp->medium_access_timed_out++;
1406
1407         /*
1408          * If the device keeps failing read/write commands but TEST UNIT
1409          * READY always completes successfully we assume that medium
1410          * access is no longer possible and take the device offline.
1411          */
1412         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1413                 scmd_printk(KERN_ERR, scmd,
1414                             "Medium access timeout failure. Offlining disk!\n");
1415                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1416
1417                 return FAILED;
1418         }
1419
1420         return eh_disp;
1421 }
1422
1423 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1424 {
1425         u64 start_lba = blk_rq_pos(scmd->request);
1426         u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1427         u64 bad_lba;
1428         int info_valid;
1429         /*
1430          * resid is optional but mostly filled in.  When it's unused,
1431          * its value is zero, so we assume the whole buffer transferred
1432          */
1433         unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1434         unsigned int good_bytes;
1435
1436         if (scmd->request->cmd_type != REQ_TYPE_FS)
1437                 return 0;
1438
1439         info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1440                                              SCSI_SENSE_BUFFERSIZE,
1441                                              &bad_lba);
1442         if (!info_valid)
1443                 return 0;
1444
1445         if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1446                 return 0;
1447
1448         if (scmd->device->sector_size < 512) {
1449                 /* only legitimate sector_size here is 256 */
1450                 start_lba <<= 1;
1451                 end_lba <<= 1;
1452         } else {
1453                 /* be careful ... don't want any overflows */
1454                 u64 factor = scmd->device->sector_size / 512;
1455                 do_div(start_lba, factor);
1456                 do_div(end_lba, factor);
1457         }
1458
1459         /* The bad lba was reported incorrectly, we have no idea where
1460          * the error is.
1461          */
1462         if (bad_lba < start_lba  || bad_lba >= end_lba)
1463                 return 0;
1464
1465         /* This computation should always be done in terms of
1466          * the resolution of the device's medium.
1467          */
1468         good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1469         return min(good_bytes, transferred);
1470 }
1471
1472 /**
1473  *      sd_done - bottom half handler: called when the lower level
1474  *      driver has completed (successfully or otherwise) a scsi command.
1475  *      @SCpnt: mid-level's per command structure.
1476  *
1477  *      Note: potentially run from within an ISR. Must not block.
1478  **/
1479 static int sd_done(struct scsi_cmnd *SCpnt)
1480 {
1481         int result = SCpnt->result;
1482         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1483         struct scsi_sense_hdr sshdr;
1484         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1485         int sense_valid = 0;
1486         int sense_deferred = 0;
1487         unsigned char op = SCpnt->cmnd[0];
1488
1489         if ((SCpnt->request->cmd_flags & REQ_DISCARD) && !result)
1490                 scsi_set_resid(SCpnt, 0);
1491
1492         if (result) {
1493                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1494                 if (sense_valid)
1495                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1496         }
1497 #ifdef CONFIG_SCSI_LOGGING
1498         SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1499         if (sense_valid) {
1500                 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1501                                                    "sd_done: sb[respc,sk,asc,"
1502                                                    "ascq]=%x,%x,%x,%x\n",
1503                                                    sshdr.response_code,
1504                                                    sshdr.sense_key, sshdr.asc,
1505                                                    sshdr.ascq));
1506         }
1507 #endif
1508         if (driver_byte(result) != DRIVER_SENSE &&
1509             (!sense_valid || sense_deferred))
1510                 goto out;
1511
1512         sdkp->medium_access_timed_out = 0;
1513
1514         switch (sshdr.sense_key) {
1515         case HARDWARE_ERROR:
1516         case MEDIUM_ERROR:
1517                 good_bytes = sd_completed_bytes(SCpnt);
1518                 break;
1519         case RECOVERED_ERROR:
1520                 good_bytes = scsi_bufflen(SCpnt);
1521                 break;
1522         case NO_SENSE:
1523                 /* This indicates a false check condition, so ignore it.  An
1524                  * unknown amount of data was transferred so treat it as an
1525                  * error.
1526                  */
1527                 scsi_print_sense("sd", SCpnt);
1528                 SCpnt->result = 0;
1529                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1530                 break;
1531         case ABORTED_COMMAND:
1532                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
1533                         good_bytes = sd_completed_bytes(SCpnt);
1534                 break;
1535         case ILLEGAL_REQUEST:
1536                 if (sshdr.asc == 0x10)  /* DIX: Host detected corruption */
1537                         good_bytes = sd_completed_bytes(SCpnt);
1538                 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1539                 if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1540                     (op == UNMAP || op == WRITE_SAME_16 || op == WRITE_SAME))
1541                         sd_config_discard(sdkp, SD_LBP_DISABLE);
1542                 break;
1543         default:
1544                 break;
1545         }
1546  out:
1547         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1548                 sd_dif_complete(SCpnt, good_bytes);
1549
1550         if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1551             == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1552
1553                 /* We have to print a failed command here as the
1554                  * extended CDB gets freed before scsi_io_completion()
1555                  * is called.
1556                  */
1557                 if (result)
1558                         scsi_print_command(SCpnt);
1559
1560                 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1561                 SCpnt->cmnd = NULL;
1562                 SCpnt->cmd_len = 0;
1563         }
1564
1565         return good_bytes;
1566 }
1567
1568 /*
1569  * spinup disk - called only in sd_revalidate_disk()
1570  */
1571 static void
1572 sd_spinup_disk(struct scsi_disk *sdkp)
1573 {
1574         unsigned char cmd[10];
1575         unsigned long spintime_expire = 0;
1576         int retries, spintime;
1577         unsigned int the_result;
1578         struct scsi_sense_hdr sshdr;
1579         int sense_valid = 0;
1580
1581         spintime = 0;
1582
1583         /* Spin up drives, as required.  Only do this at boot time */
1584         /* Spinup needs to be done for module loads too. */
1585         do {
1586                 retries = 0;
1587
1588                 do {
1589                         cmd[0] = TEST_UNIT_READY;
1590                         memset((void *) &cmd[1], 0, 9);
1591
1592                         the_result = scsi_execute_req(sdkp->device, cmd,
1593                                                       DMA_NONE, NULL, 0,
1594                                                       &sshdr, SD_TIMEOUT,
1595                                                       SD_MAX_RETRIES, NULL);
1596
1597                         /*
1598                          * If the drive has indicated to us that it
1599                          * doesn't have any media in it, don't bother
1600                          * with any more polling.
1601                          */
1602                         if (media_not_present(sdkp, &sshdr))
1603                                 return;
1604
1605                         if (the_result)
1606                                 sense_valid = scsi_sense_valid(&sshdr);
1607                         retries++;
1608                 } while (retries < 3 && 
1609                          (!scsi_status_is_good(the_result) ||
1610                           ((driver_byte(the_result) & DRIVER_SENSE) &&
1611                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1612
1613                 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1614                         /* no sense, TUR either succeeded or failed
1615                          * with a status error */
1616                         if(!spintime && !scsi_status_is_good(the_result)) {
1617                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1618                                 sd_print_result(sdkp, the_result);
1619                         }
1620                         break;
1621                 }
1622                                         
1623                 /*
1624                  * The device does not want the automatic start to be issued.
1625                  */
1626                 if (sdkp->device->no_start_on_add)
1627                         break;
1628
1629                 if (sense_valid && sshdr.sense_key == NOT_READY) {
1630                         if (sshdr.asc == 4 && sshdr.ascq == 3)
1631                                 break;  /* manual intervention required */
1632                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1633                                 break;  /* standby */
1634                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1635                                 break;  /* unavailable */
1636                         /*
1637                          * Issue command to spin up drive when not ready
1638                          */
1639                         if (!spintime) {
1640                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1641                                 cmd[0] = START_STOP;
1642                                 cmd[1] = 1;     /* Return immediately */
1643                                 memset((void *) &cmd[2], 0, 8);
1644                                 cmd[4] = 1;     /* Start spin cycle */
1645                                 if (sdkp->device->start_stop_pwr_cond)
1646                                         cmd[4] |= 1 << 4;
1647                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1648                                                  NULL, 0, &sshdr,
1649                                                  SD_TIMEOUT, SD_MAX_RETRIES,
1650                                                  NULL);
1651                                 spintime_expire = jiffies + 100 * HZ;
1652                                 spintime = 1;
1653                         }
1654                         /* Wait 1 second for next try */
1655                         msleep(1000);
1656                         printk(".");
1657
1658                 /*
1659                  * Wait for USB flash devices with slow firmware.
1660                  * Yes, this sense key/ASC combination shouldn't
1661                  * occur here.  It's characteristic of these devices.
1662                  */
1663                 } else if (sense_valid &&
1664                                 sshdr.sense_key == UNIT_ATTENTION &&
1665                                 sshdr.asc == 0x28) {
1666                         if (!spintime) {
1667                                 spintime_expire = jiffies + 5 * HZ;
1668                                 spintime = 1;
1669                         }
1670                         /* Wait 1 second for next try */
1671                         msleep(1000);
1672                 } else {
1673                         /* we don't understand the sense code, so it's
1674                          * probably pointless to loop */
1675                         if(!spintime) {
1676                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1677                                 sd_print_sense_hdr(sdkp, &sshdr);
1678                         }
1679                         break;
1680                 }
1681                                 
1682         } while (spintime && time_before_eq(jiffies, spintime_expire));
1683
1684         if (spintime) {
1685                 if (scsi_status_is_good(the_result))
1686                         printk("ready\n");
1687                 else
1688                         printk("not responding...\n");
1689         }
1690 }
1691
1692
1693 /*
1694  * Determine whether disk supports Data Integrity Field.
1695  */
1696 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1697 {
1698         struct scsi_device *sdp = sdkp->device;
1699         u8 type;
1700         int ret = 0;
1701
1702         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1703                 return ret;
1704
1705         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1706
1707         if (type > SD_DIF_TYPE3_PROTECTION)
1708                 ret = -ENODEV;
1709         else if (scsi_host_dif_capable(sdp->host, type))
1710                 ret = 1;
1711
1712         if (sdkp->first_scan || type != sdkp->protection_type)
1713                 switch (ret) {
1714                 case -ENODEV:
1715                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
1716                                   " protection type %u. Disabling disk!\n",
1717                                   type);
1718                         break;
1719                 case 1:
1720                         sd_printk(KERN_NOTICE, sdkp,
1721                                   "Enabling DIF Type %u protection\n", type);
1722                         break;
1723                 case 0:
1724                         sd_printk(KERN_NOTICE, sdkp,
1725                                   "Disabling DIF Type %u protection\n", type);
1726                         break;
1727                 }
1728
1729         sdkp->protection_type = type;
1730
1731         return ret;
1732 }
1733
1734 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1735                         struct scsi_sense_hdr *sshdr, int sense_valid,
1736                         int the_result)
1737 {
1738         sd_print_result(sdkp, the_result);
1739         if (driver_byte(the_result) & DRIVER_SENSE)
1740                 sd_print_sense_hdr(sdkp, sshdr);
1741         else
1742                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1743
1744         /*
1745          * Set dirty bit for removable devices if not ready -
1746          * sometimes drives will not report this properly.
1747          */
1748         if (sdp->removable &&
1749             sense_valid && sshdr->sense_key == NOT_READY)
1750                 set_media_not_present(sdkp);
1751
1752         /*
1753          * We used to set media_present to 0 here to indicate no media
1754          * in the drive, but some drives fail read capacity even with
1755          * media present, so we can't do that.
1756          */
1757         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1758 }
1759
1760 #define RC16_LEN 32
1761 #if RC16_LEN > SD_BUF_SIZE
1762 #error RC16_LEN must not be more than SD_BUF_SIZE
1763 #endif
1764
1765 #define READ_CAPACITY_RETRIES_ON_RESET  10
1766
1767 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1768                                                 unsigned char *buffer)
1769 {
1770         unsigned char cmd[16];
1771         struct scsi_sense_hdr sshdr;
1772         int sense_valid = 0;
1773         int the_result;
1774         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1775         unsigned int alignment;
1776         unsigned long long lba;
1777         unsigned sector_size;
1778
1779         if (sdp->no_read_capacity_16)
1780                 return -EINVAL;
1781
1782         do {
1783                 memset(cmd, 0, 16);
1784                 cmd[0] = SERVICE_ACTION_IN;
1785                 cmd[1] = SAI_READ_CAPACITY_16;
1786                 cmd[13] = RC16_LEN;
1787                 memset(buffer, 0, RC16_LEN);
1788
1789                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1790                                         buffer, RC16_LEN, &sshdr,
1791                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1792
1793                 if (media_not_present(sdkp, &sshdr))
1794                         return -ENODEV;
1795
1796                 if (the_result) {
1797                         sense_valid = scsi_sense_valid(&sshdr);
1798                         if (sense_valid &&
1799                             sshdr.sense_key == ILLEGAL_REQUEST &&
1800                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1801                             sshdr.ascq == 0x00)
1802                                 /* Invalid Command Operation Code or
1803                                  * Invalid Field in CDB, just retry
1804                                  * silently with RC10 */
1805                                 return -EINVAL;
1806                         if (sense_valid &&
1807                             sshdr.sense_key == UNIT_ATTENTION &&
1808                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1809                                 /* Device reset might occur several times,
1810                                  * give it one more chance */
1811                                 if (--reset_retries > 0)
1812                                         continue;
1813                 }
1814                 retries--;
1815
1816         } while (the_result && retries);
1817
1818         if (the_result) {
1819                 sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1820                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1821                 return -EINVAL;
1822         }
1823
1824         sector_size = get_unaligned_be32(&buffer[8]);
1825         lba = get_unaligned_be64(&buffer[0]);
1826
1827         if (sd_read_protection_type(sdkp, buffer) < 0) {
1828                 sdkp->capacity = 0;
1829                 return -ENODEV;
1830         }
1831
1832         if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1833                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1834                         "kernel compiled with support for large block "
1835                         "devices.\n");
1836                 sdkp->capacity = 0;
1837                 return -EOVERFLOW;
1838         }
1839
1840         /* Logical blocks per physical block exponent */
1841         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
1842
1843         /* Lowest aligned logical block */
1844         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1845         blk_queue_alignment_offset(sdp->request_queue, alignment);
1846         if (alignment && sdkp->first_scan)
1847                 sd_printk(KERN_NOTICE, sdkp,
1848                           "physical block alignment offset: %u\n", alignment);
1849
1850         if (buffer[14] & 0x80) { /* LBPME */
1851                 sdkp->lbpme = 1;
1852
1853                 if (buffer[14] & 0x40) /* LBPRZ */
1854                         sdkp->lbprz = 1;
1855
1856                 sd_config_discard(sdkp, SD_LBP_WS16);
1857         }
1858
1859         sdkp->capacity = lba + 1;
1860         return sector_size;
1861 }
1862
1863 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1864                                                 unsigned char *buffer)
1865 {
1866         unsigned char cmd[16];
1867         struct scsi_sense_hdr sshdr;
1868         int sense_valid = 0;
1869         int the_result;
1870         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1871         sector_t lba;
1872         unsigned sector_size;
1873
1874         do {
1875                 cmd[0] = READ_CAPACITY;
1876                 memset(&cmd[1], 0, 9);
1877                 memset(buffer, 0, 8);
1878
1879                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1880                                         buffer, 8, &sshdr,
1881                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1882
1883                 if (media_not_present(sdkp, &sshdr))
1884                         return -ENODEV;
1885
1886                 if (the_result) {
1887                         sense_valid = scsi_sense_valid(&sshdr);
1888                         if (sense_valid &&
1889                             sshdr.sense_key == UNIT_ATTENTION &&
1890                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1891                                 /* Device reset might occur several times,
1892                                  * give it one more chance */
1893                                 if (--reset_retries > 0)
1894                                         continue;
1895                 }
1896                 retries--;
1897
1898         } while (the_result && retries);
1899
1900         if (the_result) {
1901                 sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1902                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1903                 return -EINVAL;
1904         }
1905
1906         sector_size = get_unaligned_be32(&buffer[4]);
1907         lba = get_unaligned_be32(&buffer[0]);
1908
1909         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
1910                 /* Some buggy (usb cardreader) devices return an lba of
1911                    0xffffffff when the want to report a size of 0 (with
1912                    which they really mean no media is present) */
1913                 sdkp->capacity = 0;
1914                 sdkp->physical_block_size = sector_size;
1915                 return sector_size;
1916         }
1917
1918         if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1919                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1920                         "kernel compiled with support for large block "
1921                         "devices.\n");
1922                 sdkp->capacity = 0;
1923                 return -EOVERFLOW;
1924         }
1925
1926         sdkp->capacity = lba + 1;
1927         sdkp->physical_block_size = sector_size;
1928         return sector_size;
1929 }
1930
1931 static int sd_try_rc16_first(struct scsi_device *sdp)
1932 {
1933         if (sdp->host->max_cmd_len < 16)
1934                 return 0;
1935         if (sdp->try_rc_10_first)
1936                 return 0;
1937         if (sdp->scsi_level > SCSI_SPC_2)
1938                 return 1;
1939         if (scsi_device_protection(sdp))
1940                 return 1;
1941         return 0;
1942 }
1943
1944 /*
1945  * read disk capacity
1946  */
1947 static void
1948 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1949 {
1950         int sector_size;
1951         struct scsi_device *sdp = sdkp->device;
1952         sector_t old_capacity = sdkp->capacity;
1953
1954         if (sd_try_rc16_first(sdp)) {
1955                 sector_size = read_capacity_16(sdkp, sdp, buffer);
1956                 if (sector_size == -EOVERFLOW)
1957                         goto got_data;
1958                 if (sector_size == -ENODEV)
1959                         return;
1960                 if (sector_size < 0)
1961                         sector_size = read_capacity_10(sdkp, sdp, buffer);
1962                 if (sector_size < 0)
1963                         return;
1964         } else {
1965                 sector_size = read_capacity_10(sdkp, sdp, buffer);
1966                 if (sector_size == -EOVERFLOW)
1967                         goto got_data;
1968                 if (sector_size < 0)
1969                         return;
1970                 if ((sizeof(sdkp->capacity) > 4) &&
1971                     (sdkp->capacity > 0xffffffffULL)) {
1972                         int old_sector_size = sector_size;
1973                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1974                                         "Trying to use READ CAPACITY(16).\n");
1975                         sector_size = read_capacity_16(sdkp, sdp, buffer);
1976                         if (sector_size < 0) {
1977                                 sd_printk(KERN_NOTICE, sdkp,
1978                                         "Using 0xffffffff as device size\n");
1979                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
1980                                 sector_size = old_sector_size;
1981                                 goto got_data;
1982                         }
1983                 }
1984         }
1985
1986         /* Some devices are known to return the total number of blocks,
1987          * not the highest block number.  Some devices have versions
1988          * which do this and others which do not.  Some devices we might
1989          * suspect of doing this but we don't know for certain.
1990          *
1991          * If we know the reported capacity is wrong, decrement it.  If
1992          * we can only guess, then assume the number of blocks is even
1993          * (usually true but not always) and err on the side of lowering
1994          * the capacity.
1995          */
1996         if (sdp->fix_capacity ||
1997             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1998                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1999                                 "from its reported value: %llu\n",
2000                                 (unsigned long long) sdkp->capacity);
2001                 --sdkp->capacity;
2002         }
2003
2004 got_data:
2005         if (sector_size == 0) {
2006                 sector_size = 512;
2007                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2008                           "assuming 512.\n");
2009         }
2010
2011         if (sector_size != 512 &&
2012             sector_size != 1024 &&
2013             sector_size != 2048 &&
2014             sector_size != 4096 &&
2015             sector_size != 256) {
2016                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2017                           sector_size);
2018                 /*
2019                  * The user might want to re-format the drive with
2020                  * a supported sectorsize.  Once this happens, it
2021                  * would be relatively trivial to set the thing up.
2022                  * For this reason, we leave the thing in the table.
2023                  */
2024                 sdkp->capacity = 0;
2025                 /*
2026                  * set a bogus sector size so the normal read/write
2027                  * logic in the block layer will eventually refuse any
2028                  * request on this device without tripping over power
2029                  * of two sector size assumptions
2030                  */
2031                 sector_size = 512;
2032         }
2033         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2034
2035         {
2036                 char cap_str_2[10], cap_str_10[10];
2037                 u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
2038
2039                 string_get_size(sz, STRING_UNITS_2, cap_str_2,
2040                                 sizeof(cap_str_2));
2041                 string_get_size(sz, STRING_UNITS_10, cap_str_10,
2042                                 sizeof(cap_str_10));
2043
2044                 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2045                         sd_printk(KERN_NOTICE, sdkp,
2046                                   "%llu %d-byte logical blocks: (%s/%s)\n",
2047                                   (unsigned long long)sdkp->capacity,
2048                                   sector_size, cap_str_10, cap_str_2);
2049
2050                         if (sdkp->physical_block_size != sector_size)
2051                                 sd_printk(KERN_NOTICE, sdkp,
2052                                           "%u-byte physical blocks\n",
2053                                           sdkp->physical_block_size);
2054                 }
2055         }
2056
2057         /* Rescale capacity to 512-byte units */
2058         if (sector_size == 4096)
2059                 sdkp->capacity <<= 3;
2060         else if (sector_size == 2048)
2061                 sdkp->capacity <<= 2;
2062         else if (sector_size == 1024)
2063                 sdkp->capacity <<= 1;
2064         else if (sector_size == 256)
2065                 sdkp->capacity >>= 1;
2066
2067         blk_queue_physical_block_size(sdp->request_queue,
2068                                       sdkp->physical_block_size);
2069         sdkp->device->sector_size = sector_size;
2070 }
2071
2072 /* called with buffer of length 512 */
2073 static inline int
2074 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2075                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2076                  struct scsi_sense_hdr *sshdr)
2077 {
2078         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2079                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2080                                sshdr);
2081 }
2082
2083 /*
2084  * read write protect setting, if possible - called only in sd_revalidate_disk()
2085  * called with buffer of length SD_BUF_SIZE
2086  */
2087 static void
2088 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2089 {
2090         int res;
2091         struct scsi_device *sdp = sdkp->device;
2092         struct scsi_mode_data data;
2093         int old_wp = sdkp->write_prot;
2094
2095         set_disk_ro(sdkp->disk, 0);
2096         if (sdp->skip_ms_page_3f) {
2097                 sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2098                 return;
2099         }
2100
2101         if (sdp->use_192_bytes_for_3f) {
2102                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2103         } else {
2104                 /*
2105                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2106                  * We have to start carefully: some devices hang if we ask
2107                  * for more than is available.
2108                  */
2109                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2110
2111                 /*
2112                  * Second attempt: ask for page 0 When only page 0 is
2113                  * implemented, a request for page 3F may return Sense Key
2114                  * 5: Illegal Request, Sense Code 24: Invalid field in
2115                  * CDB.
2116                  */
2117                 if (!scsi_status_is_good(res))
2118                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2119
2120                 /*
2121                  * Third attempt: ask 255 bytes, as we did earlier.
2122                  */
2123                 if (!scsi_status_is_good(res))
2124                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2125                                                &data, NULL);
2126         }
2127
2128         if (!scsi_status_is_good(res)) {
2129                 sd_printk(KERN_WARNING, sdkp,
2130                           "Test WP failed, assume Write Enabled\n");
2131         } else {
2132                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2133                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2134                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2135                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2136                                   sdkp->write_prot ? "on" : "off");
2137                         sd_printk(KERN_DEBUG, sdkp,
2138                                   "Mode Sense: %02x %02x %02x %02x\n",
2139                                   buffer[0], buffer[1], buffer[2], buffer[3]);
2140                 }
2141         }
2142 }
2143
2144 /*
2145  * sd_read_cache_type - called only from sd_revalidate_disk()
2146  * called with buffer of length SD_BUF_SIZE
2147  */
2148 static void
2149 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2150 {
2151         int len = 0, res;
2152         struct scsi_device *sdp = sdkp->device;
2153
2154         int dbd;
2155         int modepage;
2156         int first_len;
2157         struct scsi_mode_data data;
2158         struct scsi_sense_hdr sshdr;
2159         int old_wce = sdkp->WCE;
2160         int old_rcd = sdkp->RCD;
2161         int old_dpofua = sdkp->DPOFUA;
2162
2163         first_len = 4;
2164         if (sdp->skip_ms_page_8) {
2165                 if (sdp->type == TYPE_RBC)
2166                         goto defaults;
2167                 else {
2168                         if (sdp->skip_ms_page_3f)
2169                                 goto defaults;
2170                         modepage = 0x3F;
2171                         if (sdp->use_192_bytes_for_3f)
2172                                 first_len = 192;
2173                         dbd = 0;
2174                 }
2175         } else if (sdp->type == TYPE_RBC) {
2176                 modepage = 6;
2177                 dbd = 8;
2178         } else {
2179                 modepage = 8;
2180                 dbd = 0;
2181         }
2182
2183         /* cautiously ask */
2184         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2185                         &data, &sshdr);
2186
2187         if (!scsi_status_is_good(res))
2188                 goto bad_sense;
2189
2190         if (!data.header_length) {
2191                 modepage = 6;
2192                 first_len = 0;
2193                 sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
2194         }
2195
2196         /* that went OK, now ask for the proper length */
2197         len = data.length;
2198
2199         /*
2200          * We're only interested in the first three bytes, actually.
2201          * But the data cache page is defined for the first 20.
2202          */
2203         if (len < 3)
2204                 goto bad_sense;
2205         else if (len > SD_BUF_SIZE) {
2206                 sd_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2207                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2208                 len = SD_BUF_SIZE;
2209         }
2210         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2211                 len = 192;
2212
2213         /* Get the data */
2214         if (len > first_len)
2215                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2216                                 &data, &sshdr);
2217
2218         if (scsi_status_is_good(res)) {
2219                 int offset = data.header_length + data.block_descriptor_length;
2220
2221                 while (offset < len) {
2222                         u8 page_code = buffer[offset] & 0x3F;
2223                         u8 spf       = buffer[offset] & 0x40;
2224
2225                         if (page_code == 8 || page_code == 6) {
2226                                 /* We're interested only in the first 3 bytes.
2227                                  */
2228                                 if (len - offset <= 2) {
2229                                         sd_printk(KERN_ERR, sdkp, "Incomplete "
2230                                                   "mode parameter data\n");
2231                                         goto defaults;
2232                                 } else {
2233                                         modepage = page_code;
2234                                         goto Page_found;
2235                                 }
2236                         } else {
2237                                 /* Go to the next page */
2238                                 if (spf && len - offset > 3)
2239                                         offset += 4 + (buffer[offset+2] << 8) +
2240                                                 buffer[offset+3];
2241                                 else if (!spf && len - offset > 1)
2242                                         offset += 2 + buffer[offset+1];
2243                                 else {
2244                                         sd_printk(KERN_ERR, sdkp, "Incomplete "
2245                                                   "mode parameter data\n");
2246                                         goto defaults;
2247                                 }
2248                         }
2249                 }
2250
2251                 if (modepage == 0x3F) {
2252                         sd_printk(KERN_ERR, sdkp, "No Caching mode page "
2253                                   "present\n");
2254                         goto defaults;
2255                 } else if ((buffer[offset] & 0x3f) != modepage) {
2256                         sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
2257                         goto defaults;
2258                 }
2259         Page_found:
2260                 if (modepage == 8) {
2261                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2262                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2263                 } else {
2264                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2265                         sdkp->RCD = 0;
2266                 }
2267
2268                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2269                 if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2270                         sd_printk(KERN_NOTICE, sdkp,
2271                                   "Uses READ/WRITE(6), disabling FUA\n");
2272                         sdkp->DPOFUA = 0;
2273                 }
2274
2275                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2276                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2277                         sd_printk(KERN_NOTICE, sdkp,
2278                                   "Write cache: %s, read cache: %s, %s\n",
2279                                   sdkp->WCE ? "enabled" : "disabled",
2280                                   sdkp->RCD ? "disabled" : "enabled",
2281                                   sdkp->DPOFUA ? "supports DPO and FUA"
2282                                   : "doesn't support DPO or FUA");
2283
2284                 return;
2285         }
2286
2287 bad_sense:
2288         if (scsi_sense_valid(&sshdr) &&
2289             sshdr.sense_key == ILLEGAL_REQUEST &&
2290             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2291                 /* Invalid field in CDB */
2292                 sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2293         else
2294                 sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
2295
2296 defaults:
2297         if (sdp->wce_default_on) {
2298                 sd_printk(KERN_NOTICE, sdkp, "Assuming drive cache: write back\n");
2299                 sdkp->WCE = 1;
2300         } else {
2301                 sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
2302                 sdkp->WCE = 0;
2303         }
2304         sdkp->RCD = 0;
2305         sdkp->DPOFUA = 0;
2306 }
2307
2308 /*
2309  * The ATO bit indicates whether the DIF application tag is available
2310  * for use by the operating system.
2311  */
2312 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2313 {
2314         int res, offset;
2315         struct scsi_device *sdp = sdkp->device;
2316         struct scsi_mode_data data;
2317         struct scsi_sense_hdr sshdr;
2318
2319         if (sdp->type != TYPE_DISK)
2320                 return;
2321
2322         if (sdkp->protection_type == 0)
2323                 return;
2324
2325         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2326                               SD_MAX_RETRIES, &data, &sshdr);
2327
2328         if (!scsi_status_is_good(res) || !data.header_length ||
2329             data.length < 6) {
2330                 sd_printk(KERN_WARNING, sdkp,
2331                           "getting Control mode page failed, assume no ATO\n");
2332
2333                 if (scsi_sense_valid(&sshdr))
2334                         sd_print_sense_hdr(sdkp, &sshdr);
2335
2336                 return;
2337         }
2338
2339         offset = data.header_length + data.block_descriptor_length;
2340
2341         if ((buffer[offset] & 0x3f) != 0x0a) {
2342                 sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2343                 return;
2344         }
2345
2346         if ((buffer[offset + 5] & 0x80) == 0)
2347                 return;
2348
2349         sdkp->ATO = 1;
2350
2351         return;
2352 }
2353
2354 /**
2355  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2356  * @disk: disk to query
2357  */
2358 static void sd_read_block_limits(struct scsi_disk *sdkp)
2359 {
2360         unsigned int sector_sz = sdkp->device->sector_size;
2361         const int vpd_len = 64;
2362         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2363
2364         if (!buffer ||
2365             /* Block Limits VPD */
2366             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2367                 goto out;
2368
2369         blk_queue_io_min(sdkp->disk->queue,
2370                          get_unaligned_be16(&buffer[6]) * sector_sz);
2371         blk_queue_io_opt(sdkp->disk->queue,
2372                          get_unaligned_be32(&buffer[12]) * sector_sz);
2373
2374         if (buffer[3] == 0x3c) {
2375                 unsigned int lba_count, desc_count;
2376
2377                 sdkp->max_ws_blocks =
2378                         (u32) min_not_zero(get_unaligned_be64(&buffer[36]),
2379                                            (u64)0xffffffff);
2380
2381                 if (!sdkp->lbpme)
2382                         goto out;
2383
2384                 lba_count = get_unaligned_be32(&buffer[20]);
2385                 desc_count = get_unaligned_be32(&buffer[24]);
2386
2387                 if (lba_count && desc_count)
2388                         sdkp->max_unmap_blocks = lba_count;
2389
2390                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2391
2392                 if (buffer[32] & 0x80)
2393                         sdkp->unmap_alignment =
2394                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2395
2396                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2397
2398                         if (sdkp->max_unmap_blocks)
2399                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2400                         else
2401                                 sd_config_discard(sdkp, SD_LBP_WS16);
2402
2403                 } else {        /* LBP VPD page tells us what to use */
2404
2405                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2406                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2407                         else if (sdkp->lbpws)
2408                                 sd_config_discard(sdkp, SD_LBP_WS16);
2409                         else if (sdkp->lbpws10)
2410                                 sd_config_discard(sdkp, SD_LBP_WS10);
2411                         else
2412                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2413                 }
2414         }
2415
2416  out:
2417         kfree(buffer);
2418 }
2419
2420 /**
2421  * sd_read_block_characteristics - Query block dev. characteristics
2422  * @disk: disk to query
2423  */
2424 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2425 {
2426         unsigned char *buffer;
2427         u16 rot;
2428         const int vpd_len = 64;
2429
2430         buffer = kmalloc(vpd_len, GFP_KERNEL);
2431
2432         if (!buffer ||
2433             /* Block Device Characteristics VPD */
2434             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2435                 goto out;
2436
2437         rot = get_unaligned_be16(&buffer[4]);
2438
2439         if (rot == 1)
2440                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2441
2442  out:
2443         kfree(buffer);
2444 }
2445
2446 /**
2447  * sd_read_block_provisioning - Query provisioning VPD page
2448  * @disk: disk to query
2449  */
2450 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2451 {
2452         unsigned char *buffer;
2453         const int vpd_len = 8;
2454
2455         if (sdkp->lbpme == 0)
2456                 return;
2457
2458         buffer = kmalloc(vpd_len, GFP_KERNEL);
2459
2460         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2461                 goto out;
2462
2463         sdkp->lbpvpd    = 1;
2464         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
2465         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2466         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2467
2468  out:
2469         kfree(buffer);
2470 }
2471
2472 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2473 {
2474         /*
2475          * Although VPD inquiries can go to SCSI-2 type devices,
2476          * some USB ones crash on receiving them, and the pages
2477          * we currently ask for are for SPC-3 and beyond
2478          */
2479         if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2480                 return 1;
2481         return 0;
2482 }
2483
2484 /**
2485  *      sd_revalidate_disk - called the first time a new disk is seen,
2486  *      performs disk spin up, read_capacity, etc.
2487  *      @disk: struct gendisk we care about
2488  **/
2489 static int sd_revalidate_disk(struct gendisk *disk)
2490 {
2491         struct scsi_disk *sdkp = scsi_disk(disk);
2492         struct scsi_device *sdp = sdkp->device;
2493         unsigned char *buffer;
2494         unsigned flush = 0;
2495
2496         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2497                                       "sd_revalidate_disk\n"));
2498
2499         /*
2500          * If the device is offline, don't try and read capacity or any
2501          * of the other niceties.
2502          */
2503         if (!scsi_device_online(sdp))
2504                 goto out;
2505
2506         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2507         if (!buffer) {
2508                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2509                           "allocation failure.\n");
2510                 goto out;
2511         }
2512
2513         sd_spinup_disk(sdkp);
2514
2515         /*
2516          * Without media there is no reason to ask; moreover, some devices
2517          * react badly if we do.
2518          */
2519         if (sdkp->media_present) {
2520                 sd_read_capacity(sdkp, buffer);
2521
2522                 if (sd_try_extended_inquiry(sdp)) {
2523                         sd_read_block_provisioning(sdkp);
2524                         sd_read_block_limits(sdkp);
2525                         sd_read_block_characteristics(sdkp);
2526                 }
2527
2528                 sd_read_write_protect_flag(sdkp, buffer);
2529                 sd_read_cache_type(sdkp, buffer);
2530                 sd_read_app_tag_own(sdkp, buffer);
2531         }
2532
2533         sdkp->first_scan = 0;
2534
2535         /*
2536          * We now have all cache related info, determine how we deal
2537          * with flush requests.
2538          */
2539         if (sdkp->WCE) {
2540                 flush |= REQ_FLUSH;
2541                 if (sdkp->DPOFUA)
2542                         flush |= REQ_FUA;
2543         }
2544
2545         blk_queue_flush(sdkp->disk->queue, flush);
2546
2547         set_capacity(disk, sdkp->capacity);
2548         kfree(buffer);
2549
2550  out:
2551         return 0;
2552 }
2553
2554 /**
2555  *      sd_unlock_native_capacity - unlock native capacity
2556  *      @disk: struct gendisk to set capacity for
2557  *
2558  *      Block layer calls this function if it detects that partitions
2559  *      on @disk reach beyond the end of the device.  If the SCSI host
2560  *      implements ->unlock_native_capacity() method, it's invoked to
2561  *      give it a chance to adjust the device capacity.
2562  *
2563  *      CONTEXT:
2564  *      Defined by block layer.  Might sleep.
2565  */
2566 static void sd_unlock_native_capacity(struct gendisk *disk)
2567 {
2568         struct scsi_device *sdev = scsi_disk(disk)->device;
2569
2570         if (sdev->host->hostt->unlock_native_capacity)
2571                 sdev->host->hostt->unlock_native_capacity(sdev);
2572 }
2573
2574 /**
2575  *      sd_format_disk_name - format disk name
2576  *      @prefix: name prefix - ie. "sd" for SCSI disks
2577  *      @index: index of the disk to format name for
2578  *      @buf: output buffer
2579  *      @buflen: length of the output buffer
2580  *
2581  *      SCSI disk names starts at sda.  The 26th device is sdz and the
2582  *      27th is sdaa.  The last one for two lettered suffix is sdzz
2583  *      which is followed by sdaaa.
2584  *
2585  *      This is basically 26 base counting with one extra 'nil' entry
2586  *      at the beginning from the second digit on and can be
2587  *      determined using similar method as 26 base conversion with the
2588  *      index shifted -1 after each digit is computed.
2589  *
2590  *      CONTEXT:
2591  *      Don't care.
2592  *
2593  *      RETURNS:
2594  *      0 on success, -errno on failure.
2595  */
2596 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2597 {
2598         const int base = 'z' - 'a' + 1;
2599         char *begin = buf + strlen(prefix);
2600         char *end = buf + buflen;
2601         char *p;
2602         int unit;
2603
2604         p = end - 1;
2605         *p = '\0';
2606         unit = base;
2607         do {
2608                 if (p == begin)
2609                         return -EINVAL;
2610                 *--p = 'a' + (index % unit);
2611                 index = (index / unit) - 1;
2612         } while (index >= 0);
2613
2614         memmove(begin, p, end - p);
2615         memcpy(buf, prefix, strlen(prefix));
2616
2617         return 0;
2618 }
2619
2620 /*
2621  * The asynchronous part of sd_probe
2622  */
2623 static void sd_probe_async(void *data, async_cookie_t cookie)
2624 {
2625         struct scsi_disk *sdkp = data;
2626         struct scsi_device *sdp;
2627         struct gendisk *gd;
2628         u32 index;
2629         struct device *dev;
2630
2631         sdp = sdkp->device;
2632         gd = sdkp->disk;
2633         index = sdkp->index;
2634         dev = &sdp->sdev_gendev;
2635
2636         gd->major = sd_major((index & 0xf0) >> 4);
2637         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2638         gd->minors = SD_MINORS;
2639
2640         gd->fops = &sd_fops;
2641         gd->private_data = &sdkp->driver;
2642         gd->queue = sdkp->device->request_queue;
2643
2644         /* defaults, until the device tells us otherwise */
2645         sdp->sector_size = 512;
2646         sdkp->capacity = 0;
2647         sdkp->media_present = 1;
2648         sdkp->write_prot = 0;
2649         sdkp->WCE = 0;
2650         sdkp->RCD = 0;
2651         sdkp->ATO = 0;
2652         sdkp->first_scan = 1;
2653         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2654
2655         sd_revalidate_disk(gd);
2656
2657         blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2658         blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2659
2660         gd->driverfs_dev = &sdp->sdev_gendev;
2661         gd->flags = GENHD_FL_EXT_DEVT;
2662         if (sdp->removable) {
2663                 gd->flags |= GENHD_FL_REMOVABLE;
2664                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
2665         }
2666
2667         add_disk(gd);
2668         if (sdkp->capacity)
2669                 sd_dif_config_host(sdkp);
2670
2671         sd_revalidate_disk(gd);
2672
2673         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2674                   sdp->removable ? "removable " : "");
2675         scsi_autopm_put_device(sdp);
2676         put_device(&sdkp->dev);
2677 }
2678
2679 /**
2680  *      sd_probe - called during driver initialization and whenever a
2681  *      new scsi device is attached to the system. It is called once
2682  *      for each scsi device (not just disks) present.
2683  *      @dev: pointer to device object
2684  *
2685  *      Returns 0 if successful (or not interested in this scsi device 
2686  *      (e.g. scanner)); 1 when there is an error.
2687  *
2688  *      Note: this function is invoked from the scsi mid-level.
2689  *      This function sets up the mapping between a given 
2690  *      <host,channel,id,lun> (found in sdp) and new device name 
2691  *      (e.g. /dev/sda). More precisely it is the block device major 
2692  *      and minor number that is chosen here.
2693  *
2694  *      Assume sd_probe is not re-entrant (for time being)
2695  *      Also think about sd_probe() and sd_remove() running coincidentally.
2696  **/
2697 static int sd_probe(struct device *dev)
2698 {
2699         struct scsi_device *sdp = to_scsi_device(dev);
2700         struct scsi_disk *sdkp;
2701         struct gendisk *gd;
2702         int index;
2703         int error;
2704
2705         error = -ENODEV;
2706         if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2707                 goto out;
2708
2709         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2710                                         "sd_probe\n"));
2711
2712         error = -ENOMEM;
2713         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2714         if (!sdkp)
2715                 goto out;
2716
2717         gd = alloc_disk(SD_MINORS);
2718         if (!gd)
2719                 goto out_free;
2720
2721         do {
2722                 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2723                         goto out_put;
2724
2725                 spin_lock(&sd_index_lock);
2726                 error = ida_get_new(&sd_index_ida, &index);
2727                 spin_unlock(&sd_index_lock);
2728         } while (error == -EAGAIN);
2729
2730         if (error) {
2731                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2732                 goto out_put;
2733         }
2734
2735         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2736         if (error) {
2737                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2738                 goto out_free_index;
2739         }
2740
2741         sdkp->device = sdp;
2742         sdkp->driver = &sd_template;
2743         sdkp->disk = gd;
2744         sdkp->index = index;
2745         atomic_set(&sdkp->openers, 0);
2746         atomic_set(&sdkp->device->ioerr_cnt, 0);
2747
2748         if (!sdp->request_queue->rq_timeout) {
2749                 if (sdp->type != TYPE_MOD)
2750                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2751                 else
2752                         blk_queue_rq_timeout(sdp->request_queue,
2753                                              SD_MOD_TIMEOUT);
2754         }
2755
2756         device_initialize(&sdkp->dev);
2757         sdkp->dev.parent = dev;
2758         sdkp->dev.class = &sd_disk_class;
2759         dev_set_name(&sdkp->dev, dev_name(dev));
2760
2761         if (device_add(&sdkp->dev))
2762                 goto out_free_index;
2763
2764         get_device(dev);
2765         dev_set_drvdata(dev, sdkp);
2766
2767         get_device(&sdkp->dev); /* prevent release before async_schedule */
2768         async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
2769
2770         return 0;
2771
2772  out_free_index:
2773         spin_lock(&sd_index_lock);
2774         ida_remove(&sd_index_ida, index);
2775         spin_unlock(&sd_index_lock);
2776  out_put:
2777         put_disk(gd);
2778  out_free:
2779         kfree(sdkp);
2780  out:
2781         return error;
2782 }
2783
2784 /**
2785  *      sd_remove - called whenever a scsi disk (previously recognized by
2786  *      sd_probe) is detached from the system. It is called (potentially
2787  *      multiple times) during sd module unload.
2788  *      @sdp: pointer to mid level scsi device object
2789  *
2790  *      Note: this function is invoked from the scsi mid-level.
2791  *      This function potentially frees up a device name (e.g. /dev/sdc)
2792  *      that could be re-used by a subsequent sd_probe().
2793  *      This function is not called when the built-in sd driver is "exit-ed".
2794  **/
2795 static int sd_remove(struct device *dev)
2796 {
2797         struct scsi_disk *sdkp;
2798
2799         sdkp = dev_get_drvdata(dev);
2800         scsi_autopm_get_device(sdkp->device);
2801
2802         async_synchronize_full_domain(&scsi_sd_probe_domain);
2803         blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2804         blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2805         device_del(&sdkp->dev);
2806         del_gendisk(sdkp->disk);
2807         sd_shutdown(dev);
2808
2809         mutex_lock(&sd_ref_mutex);
2810         dev_set_drvdata(dev, NULL);
2811         put_device(&sdkp->dev);
2812         mutex_unlock(&sd_ref_mutex);
2813
2814         return 0;
2815 }
2816
2817 /**
2818  *      scsi_disk_release - Called to free the scsi_disk structure
2819  *      @dev: pointer to embedded class device
2820  *
2821  *      sd_ref_mutex must be held entering this routine.  Because it is
2822  *      called on last put, you should always use the scsi_disk_get()
2823  *      scsi_disk_put() helpers which manipulate the semaphore directly
2824  *      and never do a direct put_device.
2825  **/
2826 static void scsi_disk_release(struct device *dev)
2827 {
2828         struct scsi_disk *sdkp = to_scsi_disk(dev);
2829         struct gendisk *disk = sdkp->disk;
2830         
2831         spin_lock(&sd_index_lock);
2832         ida_remove(&sd_index_ida, sdkp->index);
2833         spin_unlock(&sd_index_lock);
2834
2835         disk->private_data = NULL;
2836         put_disk(disk);
2837         put_device(&sdkp->device->sdev_gendev);
2838
2839         kfree(sdkp);
2840 }
2841
2842 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2843 {
2844         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
2845         struct scsi_sense_hdr sshdr;
2846         struct scsi_device *sdp = sdkp->device;
2847         int res;
2848
2849         if (start)
2850                 cmd[4] |= 1;    /* START */
2851
2852         if (sdp->start_stop_pwr_cond)
2853                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
2854
2855         if (!scsi_device_online(sdp))
2856                 return -ENODEV;
2857
2858         res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2859                                SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2860         if (res) {
2861                 sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2862                 sd_print_result(sdkp, res);
2863                 if (driver_byte(res) & DRIVER_SENSE)
2864                         sd_print_sense_hdr(sdkp, &sshdr);
2865         }
2866
2867         return res;
2868 }
2869
2870 /*
2871  * Send a SYNCHRONIZE CACHE instruction down to the device through
2872  * the normal SCSI command structure.  Wait for the command to
2873  * complete.
2874  */
2875 static void sd_shutdown(struct device *dev)
2876 {
2877         struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2878
2879         if (!sdkp)
2880                 return;         /* this can happen */
2881
2882         if (pm_runtime_suspended(dev))
2883                 goto exit;
2884
2885         if (sdkp->WCE) {
2886                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2887                 sd_sync_cache(sdkp);
2888         }
2889
2890         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2891                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2892                 sd_start_stop_device(sdkp, 0);
2893         }
2894
2895 exit:
2896         scsi_disk_put(sdkp);
2897 }
2898
2899 static int sd_suspend(struct device *dev, pm_message_t mesg)
2900 {
2901         struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2902         int ret = 0;
2903
2904         if (!sdkp)
2905                 return 0;       /* this can happen */
2906
2907         if (sdkp->WCE) {
2908                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2909                 ret = sd_sync_cache(sdkp);
2910                 if (ret)
2911                         goto done;
2912         }
2913
2914         if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2915                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2916                 ret = sd_start_stop_device(sdkp, 0);
2917         }
2918
2919 done:
2920         scsi_disk_put(sdkp);
2921         return ret;
2922 }
2923
2924 static int sd_resume(struct device *dev)
2925 {
2926         struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2927         int ret = 0;
2928
2929         if (!sdkp->device->manage_start_stop)
2930                 goto done;
2931
2932         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2933         ret = sd_start_stop_device(sdkp, 1);
2934
2935 done:
2936         scsi_disk_put(sdkp);
2937         return ret;
2938 }
2939
2940 /**
2941  *      init_sd - entry point for this driver (both when built in or when
2942  *      a module).
2943  *
2944  *      Note: this function registers this driver with the scsi mid-level.
2945  **/
2946 static int __init init_sd(void)
2947 {
2948         int majors = 0, i, err;
2949
2950         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2951
2952         for (i = 0; i < SD_MAJORS; i++)
2953                 if (register_blkdev(sd_major(i), "sd") == 0)
2954                         majors++;
2955
2956         if (!majors)
2957                 return -ENODEV;
2958
2959         err = class_register(&sd_disk_class);
2960         if (err)
2961                 goto err_out;
2962
2963         err = scsi_register_driver(&sd_template.gendrv);
2964         if (err)
2965                 goto err_out_class;
2966
2967         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2968                                          0, 0, NULL);
2969         if (!sd_cdb_cache) {
2970                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
2971                 goto err_out_class;
2972         }
2973
2974         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2975         if (!sd_cdb_pool) {
2976                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
2977                 goto err_out_cache;
2978         }
2979
2980         return 0;
2981
2982 err_out_cache:
2983         kmem_cache_destroy(sd_cdb_cache);
2984
2985 err_out_class:
2986         class_unregister(&sd_disk_class);
2987 err_out:
2988         for (i = 0; i < SD_MAJORS; i++)
2989                 unregister_blkdev(sd_major(i), "sd");
2990         return err;
2991 }
2992
2993 /**
2994  *      exit_sd - exit point for this driver (when it is a module).
2995  *
2996  *      Note: this function unregisters this driver from the scsi mid-level.
2997  **/
2998 static void __exit exit_sd(void)
2999 {
3000         int i;
3001
3002         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3003
3004         mempool_destroy(sd_cdb_pool);
3005         kmem_cache_destroy(sd_cdb_cache);
3006
3007         scsi_unregister_driver(&sd_template.gendrv);
3008         class_unregister(&sd_disk_class);
3009
3010         for (i = 0; i < SD_MAJORS; i++)
3011                 unregister_blkdev(sd_major(i), "sd");
3012 }
3013
3014 module_init(init_sd);
3015 module_exit(exit_sd);
3016
3017 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3018                                struct scsi_sense_hdr *sshdr)
3019 {
3020         sd_printk(KERN_INFO, sdkp, " ");
3021         scsi_show_sense_hdr(sshdr);
3022         sd_printk(KERN_INFO, sdkp, " ");
3023         scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
3024 }
3025
3026 static void sd_print_result(struct scsi_disk *sdkp, int result)
3027 {
3028         sd_printk(KERN_INFO, sdkp, " ");
3029         scsi_show_result(result);
3030 }
3031