Merge tag 'y2038-for-4.21' of ssh://gitolite.kernel.org:/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.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/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/mutex.h>
51 #include <linux/string_helpers.h>
52 #include <linux/async.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
56 #include <linux/pr.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
60
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
70
71 #include "sd.h"
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
74
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
78
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99
100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
101 #define SD_MINORS       16
102 #else
103 #define SD_MINORS       0
104 #endif
105
106 static void sd_config_discard(struct scsi_disk *, unsigned int);
107 static void sd_config_write_same(struct scsi_disk *);
108 static int  sd_revalidate_disk(struct gendisk *);
109 static void sd_unlock_native_capacity(struct gendisk *disk);
110 static int  sd_probe(struct device *);
111 static int  sd_remove(struct device *);
112 static void sd_shutdown(struct device *);
113 static int sd_suspend_system(struct device *);
114 static int sd_suspend_runtime(struct device *);
115 static int sd_resume(struct device *);
116 static void sd_rescan(struct device *);
117 static int sd_init_command(struct scsi_cmnd *SCpnt);
118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
119 static int sd_done(struct scsi_cmnd *);
120 static void sd_eh_reset(struct scsi_cmnd *);
121 static int sd_eh_action(struct scsi_cmnd *, int);
122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
123 static void scsi_disk_release(struct device *cdev);
124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
125 static void sd_print_result(const struct scsi_disk *, const char *, int);
126
127 static DEFINE_IDA(sd_index_ida);
128
129 /* This semaphore is used to mediate the 0->1 reference get in the
130  * face of object destruction (i.e. we can't allow a get on an
131  * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
133
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
136 static mempool_t *sd_page_pool;
137
138 static const char *sd_cache_types[] = {
139         "write through", "none", "write back",
140         "write back, no read (daft)"
141 };
142
143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 {
145         bool wc = false, fua = false;
146
147         if (sdkp->WCE) {
148                 wc = true;
149                 if (sdkp->DPOFUA)
150                         fua = true;
151         }
152
153         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
154 }
155
156 static ssize_t
157 cache_type_store(struct device *dev, struct device_attribute *attr,
158                  const char *buf, size_t count)
159 {
160         int ct, rcd, wce, sp;
161         struct scsi_disk *sdkp = to_scsi_disk(dev);
162         struct scsi_device *sdp = sdkp->device;
163         char buffer[64];
164         char *buffer_data;
165         struct scsi_mode_data data;
166         struct scsi_sense_hdr sshdr;
167         static const char temp[] = "temporary ";
168         int len;
169
170         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
171                 /* no cache control on RBC devices; theoretically they
172                  * can do it, but there's probably so many exceptions
173                  * it's not worth the risk */
174                 return -EINVAL;
175
176         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
177                 buf += sizeof(temp) - 1;
178                 sdkp->cache_override = 1;
179         } else {
180                 sdkp->cache_override = 0;
181         }
182
183         ct = sysfs_match_string(sd_cache_types, buf);
184         if (ct < 0)
185                 return -EINVAL;
186
187         rcd = ct & 0x01 ? 1 : 0;
188         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
189
190         if (sdkp->cache_override) {
191                 sdkp->WCE = wce;
192                 sdkp->RCD = rcd;
193                 sd_set_flush_flag(sdkp);
194                 return count;
195         }
196
197         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198                             SD_MAX_RETRIES, &data, NULL))
199                 return -EINVAL;
200         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201                   data.block_descriptor_length);
202         buffer_data = buffer + data.header_length +
203                 data.block_descriptor_length;
204         buffer_data[2] &= ~0x05;
205         buffer_data[2] |= wce << 2 | rcd;
206         sp = buffer_data[0] & 0x80 ? 1 : 0;
207         buffer_data[0] &= ~0x80;
208
209         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
210                              SD_MAX_RETRIES, &data, &sshdr)) {
211                 if (scsi_sense_valid(&sshdr))
212                         sd_print_sense_hdr(sdkp, &sshdr);
213                 return -EINVAL;
214         }
215         revalidate_disk(sdkp->disk);
216         return count;
217 }
218
219 static ssize_t
220 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
221                        char *buf)
222 {
223         struct scsi_disk *sdkp = to_scsi_disk(dev);
224         struct scsi_device *sdp = sdkp->device;
225
226         return sprintf(buf, "%u\n", sdp->manage_start_stop);
227 }
228
229 static ssize_t
230 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
231                         const char *buf, size_t count)
232 {
233         struct scsi_disk *sdkp = to_scsi_disk(dev);
234         struct scsi_device *sdp = sdkp->device;
235         bool v;
236
237         if (!capable(CAP_SYS_ADMIN))
238                 return -EACCES;
239
240         if (kstrtobool(buf, &v))
241                 return -EINVAL;
242
243         sdp->manage_start_stop = v;
244
245         return count;
246 }
247 static DEVICE_ATTR_RW(manage_start_stop);
248
249 static ssize_t
250 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
251 {
252         struct scsi_disk *sdkp = to_scsi_disk(dev);
253
254         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
255 }
256
257 static ssize_t
258 allow_restart_store(struct device *dev, struct device_attribute *attr,
259                     const char *buf, size_t count)
260 {
261         bool v;
262         struct scsi_disk *sdkp = to_scsi_disk(dev);
263         struct scsi_device *sdp = sdkp->device;
264
265         if (!capable(CAP_SYS_ADMIN))
266                 return -EACCES;
267
268         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
269                 return -EINVAL;
270
271         if (kstrtobool(buf, &v))
272                 return -EINVAL;
273
274         sdp->allow_restart = v;
275
276         return count;
277 }
278 static DEVICE_ATTR_RW(allow_restart);
279
280 static ssize_t
281 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
282 {
283         struct scsi_disk *sdkp = to_scsi_disk(dev);
284         int ct = sdkp->RCD + 2*sdkp->WCE;
285
286         return sprintf(buf, "%s\n", sd_cache_types[ct]);
287 }
288 static DEVICE_ATTR_RW(cache_type);
289
290 static ssize_t
291 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
292 {
293         struct scsi_disk *sdkp = to_scsi_disk(dev);
294
295         return sprintf(buf, "%u\n", sdkp->DPOFUA);
296 }
297 static DEVICE_ATTR_RO(FUA);
298
299 static ssize_t
300 protection_type_show(struct device *dev, struct device_attribute *attr,
301                      char *buf)
302 {
303         struct scsi_disk *sdkp = to_scsi_disk(dev);
304
305         return sprintf(buf, "%u\n", sdkp->protection_type);
306 }
307
308 static ssize_t
309 protection_type_store(struct device *dev, struct device_attribute *attr,
310                       const char *buf, size_t count)
311 {
312         struct scsi_disk *sdkp = to_scsi_disk(dev);
313         unsigned int val;
314         int err;
315
316         if (!capable(CAP_SYS_ADMIN))
317                 return -EACCES;
318
319         err = kstrtouint(buf, 10, &val);
320
321         if (err)
322                 return err;
323
324         if (val <= T10_PI_TYPE3_PROTECTION)
325                 sdkp->protection_type = val;
326
327         return count;
328 }
329 static DEVICE_ATTR_RW(protection_type);
330
331 static ssize_t
332 protection_mode_show(struct device *dev, struct device_attribute *attr,
333                      char *buf)
334 {
335         struct scsi_disk *sdkp = to_scsi_disk(dev);
336         struct scsi_device *sdp = sdkp->device;
337         unsigned int dif, dix;
338
339         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
340         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
341
342         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
343                 dif = 0;
344                 dix = 1;
345         }
346
347         if (!dif && !dix)
348                 return sprintf(buf, "none\n");
349
350         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
351 }
352 static DEVICE_ATTR_RO(protection_mode);
353
354 static ssize_t
355 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
356 {
357         struct scsi_disk *sdkp = to_scsi_disk(dev);
358
359         return sprintf(buf, "%u\n", sdkp->ATO);
360 }
361 static DEVICE_ATTR_RO(app_tag_own);
362
363 static ssize_t
364 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
365                        char *buf)
366 {
367         struct scsi_disk *sdkp = to_scsi_disk(dev);
368
369         return sprintf(buf, "%u\n", sdkp->lbpme);
370 }
371 static DEVICE_ATTR_RO(thin_provisioning);
372
373 /* sysfs_match_string() requires dense arrays */
374 static const char *lbp_mode[] = {
375         [SD_LBP_FULL]           = "full",
376         [SD_LBP_UNMAP]          = "unmap",
377         [SD_LBP_WS16]           = "writesame_16",
378         [SD_LBP_WS10]           = "writesame_10",
379         [SD_LBP_ZERO]           = "writesame_zero",
380         [SD_LBP_DISABLE]        = "disabled",
381 };
382
383 static ssize_t
384 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
385                        char *buf)
386 {
387         struct scsi_disk *sdkp = to_scsi_disk(dev);
388
389         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
390 }
391
392 static ssize_t
393 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
394                         const char *buf, size_t count)
395 {
396         struct scsi_disk *sdkp = to_scsi_disk(dev);
397         struct scsi_device *sdp = sdkp->device;
398         int mode;
399
400         if (!capable(CAP_SYS_ADMIN))
401                 return -EACCES;
402
403         if (sd_is_zoned(sdkp)) {
404                 sd_config_discard(sdkp, SD_LBP_DISABLE);
405                 return count;
406         }
407
408         if (sdp->type != TYPE_DISK)
409                 return -EINVAL;
410
411         mode = sysfs_match_string(lbp_mode, buf);
412         if (mode < 0)
413                 return -EINVAL;
414
415         sd_config_discard(sdkp, mode);
416
417         return count;
418 }
419 static DEVICE_ATTR_RW(provisioning_mode);
420
421 /* sysfs_match_string() requires dense arrays */
422 static const char *zeroing_mode[] = {
423         [SD_ZERO_WRITE]         = "write",
424         [SD_ZERO_WS]            = "writesame",
425         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
426         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
427 };
428
429 static ssize_t
430 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
431                   char *buf)
432 {
433         struct scsi_disk *sdkp = to_scsi_disk(dev);
434
435         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
436 }
437
438 static ssize_t
439 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
440                    const char *buf, size_t count)
441 {
442         struct scsi_disk *sdkp = to_scsi_disk(dev);
443         int mode;
444
445         if (!capable(CAP_SYS_ADMIN))
446                 return -EACCES;
447
448         mode = sysfs_match_string(zeroing_mode, buf);
449         if (mode < 0)
450                 return -EINVAL;
451
452         sdkp->zeroing_mode = mode;
453
454         return count;
455 }
456 static DEVICE_ATTR_RW(zeroing_mode);
457
458 static ssize_t
459 max_medium_access_timeouts_show(struct device *dev,
460                                 struct device_attribute *attr, char *buf)
461 {
462         struct scsi_disk *sdkp = to_scsi_disk(dev);
463
464         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
465 }
466
467 static ssize_t
468 max_medium_access_timeouts_store(struct device *dev,
469                                  struct device_attribute *attr, const char *buf,
470                                  size_t count)
471 {
472         struct scsi_disk *sdkp = to_scsi_disk(dev);
473         int err;
474
475         if (!capable(CAP_SYS_ADMIN))
476                 return -EACCES;
477
478         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
479
480         return err ? err : count;
481 }
482 static DEVICE_ATTR_RW(max_medium_access_timeouts);
483
484 static ssize_t
485 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
486                            char *buf)
487 {
488         struct scsi_disk *sdkp = to_scsi_disk(dev);
489
490         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
491 }
492
493 static ssize_t
494 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
495                             const char *buf, size_t count)
496 {
497         struct scsi_disk *sdkp = to_scsi_disk(dev);
498         struct scsi_device *sdp = sdkp->device;
499         unsigned long max;
500         int err;
501
502         if (!capable(CAP_SYS_ADMIN))
503                 return -EACCES;
504
505         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
506                 return -EINVAL;
507
508         err = kstrtoul(buf, 10, &max);
509
510         if (err)
511                 return err;
512
513         if (max == 0)
514                 sdp->no_write_same = 1;
515         else if (max <= SD_MAX_WS16_BLOCKS) {
516                 sdp->no_write_same = 0;
517                 sdkp->max_ws_blocks = max;
518         }
519
520         sd_config_write_same(sdkp);
521
522         return count;
523 }
524 static DEVICE_ATTR_RW(max_write_same_blocks);
525
526 static struct attribute *sd_disk_attrs[] = {
527         &dev_attr_cache_type.attr,
528         &dev_attr_FUA.attr,
529         &dev_attr_allow_restart.attr,
530         &dev_attr_manage_start_stop.attr,
531         &dev_attr_protection_type.attr,
532         &dev_attr_protection_mode.attr,
533         &dev_attr_app_tag_own.attr,
534         &dev_attr_thin_provisioning.attr,
535         &dev_attr_provisioning_mode.attr,
536         &dev_attr_zeroing_mode.attr,
537         &dev_attr_max_write_same_blocks.attr,
538         &dev_attr_max_medium_access_timeouts.attr,
539         NULL,
540 };
541 ATTRIBUTE_GROUPS(sd_disk);
542
543 static struct class sd_disk_class = {
544         .name           = "scsi_disk",
545         .owner          = THIS_MODULE,
546         .dev_release    = scsi_disk_release,
547         .dev_groups     = sd_disk_groups,
548 };
549
550 static const struct dev_pm_ops sd_pm_ops = {
551         .suspend                = sd_suspend_system,
552         .resume                 = sd_resume,
553         .poweroff               = sd_suspend_system,
554         .restore                = sd_resume,
555         .runtime_suspend        = sd_suspend_runtime,
556         .runtime_resume         = sd_resume,
557 };
558
559 static struct scsi_driver sd_template = {
560         .gendrv = {
561                 .name           = "sd",
562                 .owner          = THIS_MODULE,
563                 .probe          = sd_probe,
564                 .remove         = sd_remove,
565                 .shutdown       = sd_shutdown,
566                 .pm             = &sd_pm_ops,
567         },
568         .rescan                 = sd_rescan,
569         .init_command           = sd_init_command,
570         .uninit_command         = sd_uninit_command,
571         .done                   = sd_done,
572         .eh_action              = sd_eh_action,
573         .eh_reset               = sd_eh_reset,
574 };
575
576 /*
577  * Dummy kobj_map->probe function.
578  * The default ->probe function will call modprobe, which is
579  * pointless as this module is already loaded.
580  */
581 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
582 {
583         return NULL;
584 }
585
586 /*
587  * Device no to disk mapping:
588  * 
589  *       major         disc2     disc  p1
590  *   |............|.............|....|....| <- dev_t
591  *    31        20 19          8 7  4 3  0
592  * 
593  * Inside a major, we have 16k disks, however mapped non-
594  * contiguously. The first 16 disks are for major0, the next
595  * ones with major1, ... Disk 256 is for major0 again, disk 272 
596  * for major1, ... 
597  * As we stay compatible with our numbering scheme, we can reuse 
598  * the well-know SCSI majors 8, 65--71, 136--143.
599  */
600 static int sd_major(int major_idx)
601 {
602         switch (major_idx) {
603         case 0:
604                 return SCSI_DISK0_MAJOR;
605         case 1 ... 7:
606                 return SCSI_DISK1_MAJOR + major_idx - 1;
607         case 8 ... 15:
608                 return SCSI_DISK8_MAJOR + major_idx - 8;
609         default:
610                 BUG();
611                 return 0;       /* shut up gcc */
612         }
613 }
614
615 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
616 {
617         struct scsi_disk *sdkp = NULL;
618
619         mutex_lock(&sd_ref_mutex);
620
621         if (disk->private_data) {
622                 sdkp = scsi_disk(disk);
623                 if (scsi_device_get(sdkp->device) == 0)
624                         get_device(&sdkp->dev);
625                 else
626                         sdkp = NULL;
627         }
628         mutex_unlock(&sd_ref_mutex);
629         return sdkp;
630 }
631
632 static void scsi_disk_put(struct scsi_disk *sdkp)
633 {
634         struct scsi_device *sdev = sdkp->device;
635
636         mutex_lock(&sd_ref_mutex);
637         put_device(&sdkp->dev);
638         scsi_device_put(sdev);
639         mutex_unlock(&sd_ref_mutex);
640 }
641
642 #ifdef CONFIG_BLK_SED_OPAL
643 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
644                 size_t len, bool send)
645 {
646         struct scsi_device *sdev = data;
647         u8 cdb[12] = { 0, };
648         int ret;
649
650         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
651         cdb[1] = secp;
652         put_unaligned_be16(spsp, &cdb[2]);
653         put_unaligned_be32(len, &cdb[6]);
654
655         ret = scsi_execute_req(sdev, cdb,
656                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
657                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
658         return ret <= 0 ? ret : -EIO;
659 }
660 #endif /* CONFIG_BLK_SED_OPAL */
661
662 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
663                                            unsigned int dix, unsigned int dif)
664 {
665         struct bio *bio = scmd->request->bio;
666         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
667         unsigned int protect = 0;
668
669         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
670                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
671                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
672
673                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
674                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
675         }
676
677         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
678                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
679
680                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
681                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
682         }
683
684         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
685                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
686
687                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
688                         protect = 3 << 5;       /* Disable target PI checking */
689                 else
690                         protect = 1 << 5;       /* Enable target PI checking */
691         }
692
693         scsi_set_prot_op(scmd, prot_op);
694         scsi_set_prot_type(scmd, dif);
695         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
696
697         return protect;
698 }
699
700 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
701 {
702         struct request_queue *q = sdkp->disk->queue;
703         unsigned int logical_block_size = sdkp->device->sector_size;
704         unsigned int max_blocks = 0;
705
706         q->limits.discard_alignment =
707                 sdkp->unmap_alignment * logical_block_size;
708         q->limits.discard_granularity =
709                 max(sdkp->physical_block_size,
710                     sdkp->unmap_granularity * logical_block_size);
711         sdkp->provisioning_mode = mode;
712
713         switch (mode) {
714
715         case SD_LBP_FULL:
716         case SD_LBP_DISABLE:
717                 blk_queue_max_discard_sectors(q, 0);
718                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
719                 return;
720
721         case SD_LBP_UNMAP:
722                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
723                                           (u32)SD_MAX_WS16_BLOCKS);
724                 break;
725
726         case SD_LBP_WS16:
727                 if (sdkp->device->unmap_limit_for_ws)
728                         max_blocks = sdkp->max_unmap_blocks;
729                 else
730                         max_blocks = sdkp->max_ws_blocks;
731
732                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
733                 break;
734
735         case SD_LBP_WS10:
736                 if (sdkp->device->unmap_limit_for_ws)
737                         max_blocks = sdkp->max_unmap_blocks;
738                 else
739                         max_blocks = sdkp->max_ws_blocks;
740
741                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
742                 break;
743
744         case SD_LBP_ZERO:
745                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
746                                           (u32)SD_MAX_WS10_BLOCKS);
747                 break;
748         }
749
750         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
751         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
752 }
753
754 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
755 {
756         struct scsi_device *sdp = cmd->device;
757         struct request *rq = cmd->request;
758         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
759         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
760         unsigned int data_len = 24;
761         char *buf;
762
763         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
764         if (!rq->special_vec.bv_page)
765                 return BLKPREP_DEFER;
766         clear_highpage(rq->special_vec.bv_page);
767         rq->special_vec.bv_offset = 0;
768         rq->special_vec.bv_len = data_len;
769         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
770
771         cmd->cmd_len = 10;
772         cmd->cmnd[0] = UNMAP;
773         cmd->cmnd[8] = 24;
774
775         buf = page_address(rq->special_vec.bv_page);
776         put_unaligned_be16(6 + 16, &buf[0]);
777         put_unaligned_be16(16, &buf[2]);
778         put_unaligned_be64(sector, &buf[8]);
779         put_unaligned_be32(nr_sectors, &buf[16]);
780
781         cmd->allowed = SD_MAX_RETRIES;
782         cmd->transfersize = data_len;
783         rq->timeout = SD_TIMEOUT;
784         scsi_req(rq)->resid_len = data_len;
785
786         return scsi_init_io(cmd);
787 }
788
789 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
790 {
791         struct scsi_device *sdp = cmd->device;
792         struct request *rq = cmd->request;
793         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
794         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
795         u32 data_len = sdp->sector_size;
796
797         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
798         if (!rq->special_vec.bv_page)
799                 return BLKPREP_DEFER;
800         clear_highpage(rq->special_vec.bv_page);
801         rq->special_vec.bv_offset = 0;
802         rq->special_vec.bv_len = data_len;
803         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
804
805         cmd->cmd_len = 16;
806         cmd->cmnd[0] = WRITE_SAME_16;
807         if (unmap)
808                 cmd->cmnd[1] = 0x8; /* UNMAP */
809         put_unaligned_be64(sector, &cmd->cmnd[2]);
810         put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
811
812         cmd->allowed = SD_MAX_RETRIES;
813         cmd->transfersize = data_len;
814         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
815         scsi_req(rq)->resid_len = data_len;
816
817         return scsi_init_io(cmd);
818 }
819
820 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
821 {
822         struct scsi_device *sdp = cmd->device;
823         struct request *rq = cmd->request;
824         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
825         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
826         u32 data_len = sdp->sector_size;
827
828         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
829         if (!rq->special_vec.bv_page)
830                 return BLKPREP_DEFER;
831         clear_highpage(rq->special_vec.bv_page);
832         rq->special_vec.bv_offset = 0;
833         rq->special_vec.bv_len = data_len;
834         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
835
836         cmd->cmd_len = 10;
837         cmd->cmnd[0] = WRITE_SAME;
838         if (unmap)
839                 cmd->cmnd[1] = 0x8; /* UNMAP */
840         put_unaligned_be32(sector, &cmd->cmnd[2]);
841         put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
842
843         cmd->allowed = SD_MAX_RETRIES;
844         cmd->transfersize = data_len;
845         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
846         scsi_req(rq)->resid_len = data_len;
847
848         return scsi_init_io(cmd);
849 }
850
851 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
852 {
853         struct request *rq = cmd->request;
854         struct scsi_device *sdp = cmd->device;
855         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
856         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
857         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
858
859         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
860                 switch (sdkp->zeroing_mode) {
861                 case SD_ZERO_WS16_UNMAP:
862                         return sd_setup_write_same16_cmnd(cmd, true);
863                 case SD_ZERO_WS10_UNMAP:
864                         return sd_setup_write_same10_cmnd(cmd, true);
865                 }
866         }
867
868         if (sdp->no_write_same)
869                 return BLKPREP_INVALID;
870
871         if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
872                 return sd_setup_write_same16_cmnd(cmd, false);
873
874         return sd_setup_write_same10_cmnd(cmd, false);
875 }
876
877 static void sd_config_write_same(struct scsi_disk *sdkp)
878 {
879         struct request_queue *q = sdkp->disk->queue;
880         unsigned int logical_block_size = sdkp->device->sector_size;
881
882         if (sdkp->device->no_write_same) {
883                 sdkp->max_ws_blocks = 0;
884                 goto out;
885         }
886
887         /* Some devices can not handle block counts above 0xffff despite
888          * supporting WRITE SAME(16). Consequently we default to 64k
889          * blocks per I/O unless the device explicitly advertises a
890          * bigger limit.
891          */
892         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
893                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
894                                                    (u32)SD_MAX_WS16_BLOCKS);
895         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
896                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
897                                                    (u32)SD_MAX_WS10_BLOCKS);
898         else {
899                 sdkp->device->no_write_same = 1;
900                 sdkp->max_ws_blocks = 0;
901         }
902
903         if (sdkp->lbprz && sdkp->lbpws)
904                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
905         else if (sdkp->lbprz && sdkp->lbpws10)
906                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
907         else if (sdkp->max_ws_blocks)
908                 sdkp->zeroing_mode = SD_ZERO_WS;
909         else
910                 sdkp->zeroing_mode = SD_ZERO_WRITE;
911
912         if (sdkp->max_ws_blocks &&
913             sdkp->physical_block_size > logical_block_size) {
914                 /*
915                  * Reporting a maximum number of blocks that is not aligned
916                  * on the device physical size would cause a large write same
917                  * request to be split into physically unaligned chunks by
918                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
919                  * even if the caller of these functions took care to align the
920                  * large request. So make sure the maximum reported is aligned
921                  * to the device physical block size. This is only an optional
922                  * optimization for regular disks, but this is mandatory to
923                  * avoid failure of large write same requests directed at
924                  * sequential write required zones of host-managed ZBC disks.
925                  */
926                 sdkp->max_ws_blocks =
927                         round_down(sdkp->max_ws_blocks,
928                                    bytes_to_logical(sdkp->device,
929                                                     sdkp->physical_block_size));
930         }
931
932 out:
933         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
934                                          (logical_block_size >> 9));
935         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
936                                          (logical_block_size >> 9));
937 }
938
939 /**
940  * sd_setup_write_same_cmnd - write the same data to multiple blocks
941  * @cmd: command to prepare
942  *
943  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
944  * the preference indicated by the target device.
945  **/
946 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
947 {
948         struct request *rq = cmd->request;
949         struct scsi_device *sdp = cmd->device;
950         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
951         struct bio *bio = rq->bio;
952         sector_t sector = blk_rq_pos(rq);
953         unsigned int nr_sectors = blk_rq_sectors(rq);
954         unsigned int nr_bytes = blk_rq_bytes(rq);
955         int ret;
956
957         if (sdkp->device->no_write_same)
958                 return BLKPREP_INVALID;
959
960         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
961
962         sector >>= ilog2(sdp->sector_size) - 9;
963         nr_sectors >>= ilog2(sdp->sector_size) - 9;
964
965         rq->timeout = SD_WRITE_SAME_TIMEOUT;
966
967         if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
968                 cmd->cmd_len = 16;
969                 cmd->cmnd[0] = WRITE_SAME_16;
970                 put_unaligned_be64(sector, &cmd->cmnd[2]);
971                 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
972         } else {
973                 cmd->cmd_len = 10;
974                 cmd->cmnd[0] = WRITE_SAME;
975                 put_unaligned_be32(sector, &cmd->cmnd[2]);
976                 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
977         }
978
979         cmd->transfersize = sdp->sector_size;
980         cmd->allowed = SD_MAX_RETRIES;
981
982         /*
983          * For WRITE SAME the data transferred via the DATA OUT buffer is
984          * different from the amount of data actually written to the target.
985          *
986          * We set up __data_len to the amount of data transferred via the
987          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
988          * to transfer a single sector of data first, but then reset it to
989          * the amount of data to be written right after so that the I/O path
990          * knows how much to actually write.
991          */
992         rq->__data_len = sdp->sector_size;
993         ret = scsi_init_io(cmd);
994         rq->__data_len = nr_bytes;
995
996         return ret;
997 }
998
999 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1000 {
1001         struct request *rq = cmd->request;
1002
1003         /* flush requests don't perform I/O, zero the S/G table */
1004         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1005
1006         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1007         cmd->cmd_len = 10;
1008         cmd->transfersize = 0;
1009         cmd->allowed = SD_MAX_RETRIES;
1010
1011         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1012         return BLKPREP_OK;
1013 }
1014
1015 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1016 {
1017         struct request *rq = SCpnt->request;
1018         struct scsi_device *sdp = SCpnt->device;
1019         struct gendisk *disk = rq->rq_disk;
1020         struct scsi_disk *sdkp = scsi_disk(disk);
1021         sector_t block = blk_rq_pos(rq);
1022         sector_t threshold;
1023         unsigned int this_count = blk_rq_sectors(rq);
1024         unsigned int dif, dix;
1025         int ret;
1026         unsigned char protect;
1027
1028         ret = scsi_init_io(SCpnt);
1029         if (ret != BLKPREP_OK)
1030                 return ret;
1031         WARN_ON_ONCE(SCpnt != rq->special);
1032
1033         /* from here on until we're complete, any goto out
1034          * is used for a killable error condition */
1035         ret = BLKPREP_KILL;
1036
1037         SCSI_LOG_HLQUEUE(1,
1038                 scmd_printk(KERN_INFO, SCpnt,
1039                         "%s: block=%llu, count=%d\n",
1040                         __func__, (unsigned long long)block, this_count));
1041
1042         if (!sdp || !scsi_device_online(sdp) ||
1043             block + blk_rq_sectors(rq) > get_capacity(disk)) {
1044                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1045                                                 "Finishing %u sectors\n",
1046                                                 blk_rq_sectors(rq)));
1047                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1048                                                 "Retry with 0x%p\n", SCpnt));
1049                 goto out;
1050         }
1051
1052         if (sdp->changed) {
1053                 /*
1054                  * quietly refuse to do anything to a changed disc until 
1055                  * the changed bit has been reset
1056                  */
1057                 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1058                 goto out;
1059         }
1060
1061         /*
1062          * Some SD card readers can't handle multi-sector accesses which touch
1063          * the last one or two hardware sectors.  Split accesses as needed.
1064          */
1065         threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1066                 (sdp->sector_size / 512);
1067
1068         if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1069                 if (block < threshold) {
1070                         /* Access up to the threshold but not beyond */
1071                         this_count = threshold - block;
1072                 } else {
1073                         /* Access only a single hardware sector */
1074                         this_count = sdp->sector_size / 512;
1075                 }
1076         }
1077
1078         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1079                                         (unsigned long long)block));
1080
1081         /*
1082          * If we have a 1K hardware sectorsize, prevent access to single
1083          * 512 byte sectors.  In theory we could handle this - in fact
1084          * the scsi cdrom driver must be able to handle this because
1085          * we typically use 1K blocksizes, and cdroms typically have
1086          * 2K hardware sectorsizes.  Of course, things are simpler
1087          * with the cdrom, since it is read-only.  For performance
1088          * reasons, the filesystems should be able to handle this
1089          * and not force the scsi disk driver to use bounce buffers
1090          * for this.
1091          */
1092         if (sdp->sector_size == 1024) {
1093                 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1094                         scmd_printk(KERN_ERR, SCpnt,
1095                                     "Bad block number requested\n");
1096                         goto out;
1097                 } else {
1098                         block = block >> 1;
1099                         this_count = this_count >> 1;
1100                 }
1101         }
1102         if (sdp->sector_size == 2048) {
1103                 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1104                         scmd_printk(KERN_ERR, SCpnt,
1105                                     "Bad block number requested\n");
1106                         goto out;
1107                 } else {
1108                         block = block >> 2;
1109                         this_count = this_count >> 2;
1110                 }
1111         }
1112         if (sdp->sector_size == 4096) {
1113                 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1114                         scmd_printk(KERN_ERR, SCpnt,
1115                                     "Bad block number requested\n");
1116                         goto out;
1117                 } else {
1118                         block = block >> 3;
1119                         this_count = this_count >> 3;
1120                 }
1121         }
1122         if (rq_data_dir(rq) == WRITE) {
1123                 SCpnt->cmnd[0] = WRITE_6;
1124
1125                 if (blk_integrity_rq(rq))
1126                         t10_pi_prepare(SCpnt->request, sdkp->protection_type);
1127
1128         } else if (rq_data_dir(rq) == READ) {
1129                 SCpnt->cmnd[0] = READ_6;
1130         } else {
1131                 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1132                 goto out;
1133         }
1134
1135         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1136                                         "%s %d/%u 512 byte blocks.\n",
1137                                         (rq_data_dir(rq) == WRITE) ?
1138                                         "writing" : "reading", this_count,
1139                                         blk_rq_sectors(rq)));
1140
1141         dix = scsi_prot_sg_count(SCpnt);
1142         dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1143
1144         if (dif || dix)
1145                 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1146         else
1147                 protect = 0;
1148
1149         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1150                 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1151
1152                 if (unlikely(SCpnt->cmnd == NULL)) {
1153                         ret = BLKPREP_DEFER;
1154                         goto out;
1155                 }
1156
1157                 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1158                 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1159                 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1160                 SCpnt->cmnd[7] = 0x18;
1161                 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1162                 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1163
1164                 /* LBA */
1165                 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1166                 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1167                 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1168                 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1169                 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1170                 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1171                 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1172                 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1173
1174                 /* Expected Indirect LBA */
1175                 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1176                 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1177                 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1178                 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1179
1180                 /* Transfer length */
1181                 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1182                 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1183                 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1184                 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1185         } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1186                 SCpnt->cmnd[0] += READ_16 - READ_6;
1187                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1188                 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1189                 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1190                 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1191                 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1192                 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1193                 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1194                 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1195                 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1196                 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1197                 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1198                 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1199                 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1200                 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1201         } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1202                    scsi_device_protection(SCpnt->device) ||
1203                    SCpnt->device->use_10_for_rw) {
1204                 SCpnt->cmnd[0] += READ_10 - READ_6;
1205                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1206                 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1207                 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1208                 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1209                 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1210                 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1211                 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1212                 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1213         } else {
1214                 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1215                         /*
1216                          * This happens only if this drive failed
1217                          * 10byte rw command with ILLEGAL_REQUEST
1218                          * during operation and thus turned off
1219                          * use_10_for_rw.
1220                          */
1221                         scmd_printk(KERN_ERR, SCpnt,
1222                                     "FUA write on READ/WRITE(6) drive\n");
1223                         goto out;
1224                 }
1225
1226                 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1227                 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1228                 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1229                 SCpnt->cmnd[4] = (unsigned char) this_count;
1230                 SCpnt->cmnd[5] = 0;
1231         }
1232         SCpnt->sdb.length = this_count * sdp->sector_size;
1233
1234         /*
1235          * We shouldn't disconnect in the middle of a sector, so with a dumb
1236          * host adapter, it's safe to assume that we can at least transfer
1237          * this many bytes between each connect / disconnect.
1238          */
1239         SCpnt->transfersize = sdp->sector_size;
1240         SCpnt->underflow = this_count << 9;
1241         SCpnt->allowed = SD_MAX_RETRIES;
1242
1243         /*
1244          * This indicates that the command is ready from our end to be
1245          * queued.
1246          */
1247         ret = BLKPREP_OK;
1248  out:
1249         return ret;
1250 }
1251
1252 static int sd_init_command(struct scsi_cmnd *cmd)
1253 {
1254         struct request *rq = cmd->request;
1255
1256         switch (req_op(rq)) {
1257         case REQ_OP_DISCARD:
1258                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1259                 case SD_LBP_UNMAP:
1260                         return sd_setup_unmap_cmnd(cmd);
1261                 case SD_LBP_WS16:
1262                         return sd_setup_write_same16_cmnd(cmd, true);
1263                 case SD_LBP_WS10:
1264                         return sd_setup_write_same10_cmnd(cmd, true);
1265                 case SD_LBP_ZERO:
1266                         return sd_setup_write_same10_cmnd(cmd, false);
1267                 default:
1268                         return BLKPREP_INVALID;
1269                 }
1270         case REQ_OP_WRITE_ZEROES:
1271                 return sd_setup_write_zeroes_cmnd(cmd);
1272         case REQ_OP_WRITE_SAME:
1273                 return sd_setup_write_same_cmnd(cmd);
1274         case REQ_OP_FLUSH:
1275                 return sd_setup_flush_cmnd(cmd);
1276         case REQ_OP_READ:
1277         case REQ_OP_WRITE:
1278                 return sd_setup_read_write_cmnd(cmd);
1279         case REQ_OP_ZONE_RESET:
1280                 return sd_zbc_setup_reset_cmnd(cmd);
1281         default:
1282                 WARN_ON_ONCE(1);
1283                 return BLKPREP_KILL;
1284         }
1285 }
1286
1287 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1288 {
1289         struct request *rq = SCpnt->request;
1290         u8 *cmnd;
1291
1292         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1293                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1294
1295         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1296                 cmnd = SCpnt->cmnd;
1297                 SCpnt->cmnd = NULL;
1298                 SCpnt->cmd_len = 0;
1299                 mempool_free(cmnd, sd_cdb_pool);
1300         }
1301 }
1302
1303 /**
1304  *      sd_open - open a scsi disk device
1305  *      @bdev: Block device of the scsi disk to open
1306  *      @mode: FMODE_* mask
1307  *
1308  *      Returns 0 if successful. Returns a negated errno value in case 
1309  *      of error.
1310  *
1311  *      Note: This can be called from a user context (e.g. fsck(1) )
1312  *      or from within the kernel (e.g. as a result of a mount(1) ).
1313  *      In the latter case @inode and @filp carry an abridged amount
1314  *      of information as noted above.
1315  *
1316  *      Locking: called with bdev->bd_mutex held.
1317  **/
1318 static int sd_open(struct block_device *bdev, fmode_t mode)
1319 {
1320         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1321         struct scsi_device *sdev;
1322         int retval;
1323
1324         if (!sdkp)
1325                 return -ENXIO;
1326
1327         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1328
1329         sdev = sdkp->device;
1330
1331         /*
1332          * If the device is in error recovery, wait until it is done.
1333          * If the device is offline, then disallow any access to it.
1334          */
1335         retval = -ENXIO;
1336         if (!scsi_block_when_processing_errors(sdev))
1337                 goto error_out;
1338
1339         if (sdev->removable || sdkp->write_prot)
1340                 check_disk_change(bdev);
1341
1342         /*
1343          * If the drive is empty, just let the open fail.
1344          */
1345         retval = -ENOMEDIUM;
1346         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1347                 goto error_out;
1348
1349         /*
1350          * If the device has the write protect tab set, have the open fail
1351          * if the user expects to be able to write to the thing.
1352          */
1353         retval = -EROFS;
1354         if (sdkp->write_prot && (mode & FMODE_WRITE))
1355                 goto error_out;
1356
1357         /*
1358          * It is possible that the disk changing stuff resulted in
1359          * the device being taken offline.  If this is the case,
1360          * report this to the user, and don't pretend that the
1361          * open actually succeeded.
1362          */
1363         retval = -ENXIO;
1364         if (!scsi_device_online(sdev))
1365                 goto error_out;
1366
1367         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1368                 if (scsi_block_when_processing_errors(sdev))
1369                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1370         }
1371
1372         return 0;
1373
1374 error_out:
1375         scsi_disk_put(sdkp);
1376         return retval;  
1377 }
1378
1379 /**
1380  *      sd_release - invoked when the (last) close(2) is called on this
1381  *      scsi disk.
1382  *      @disk: disk to release
1383  *      @mode: FMODE_* mask
1384  *
1385  *      Returns 0. 
1386  *
1387  *      Note: may block (uninterruptible) if error recovery is underway
1388  *      on this disk.
1389  *
1390  *      Locking: called with bdev->bd_mutex held.
1391  **/
1392 static void sd_release(struct gendisk *disk, fmode_t mode)
1393 {
1394         struct scsi_disk *sdkp = scsi_disk(disk);
1395         struct scsi_device *sdev = sdkp->device;
1396
1397         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1398
1399         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1400                 if (scsi_block_when_processing_errors(sdev))
1401                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1402         }
1403
1404         /*
1405          * XXX and what if there are packets in flight and this close()
1406          * XXX is followed by a "rmmod sd_mod"?
1407          */
1408
1409         scsi_disk_put(sdkp);
1410 }
1411
1412 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1413 {
1414         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1415         struct scsi_device *sdp = sdkp->device;
1416         struct Scsi_Host *host = sdp->host;
1417         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1418         int diskinfo[4];
1419
1420         /* default to most commonly used values */
1421         diskinfo[0] = 0x40;     /* 1 << 6 */
1422         diskinfo[1] = 0x20;     /* 1 << 5 */
1423         diskinfo[2] = capacity >> 11;
1424
1425         /* override with calculated, extended default, or driver values */
1426         if (host->hostt->bios_param)
1427                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1428         else
1429                 scsicam_bios_param(bdev, capacity, diskinfo);
1430
1431         geo->heads = diskinfo[0];
1432         geo->sectors = diskinfo[1];
1433         geo->cylinders = diskinfo[2];
1434         return 0;
1435 }
1436
1437 /**
1438  *      sd_ioctl - process an ioctl
1439  *      @bdev: target block device
1440  *      @mode: FMODE_* mask
1441  *      @cmd: ioctl command number
1442  *      @arg: this is third argument given to ioctl(2) system call.
1443  *      Often contains a pointer.
1444  *
1445  *      Returns 0 if successful (some ioctls return positive numbers on
1446  *      success as well). Returns a negated errno value in case of error.
1447  *
1448  *      Note: most ioctls are forward onto the block subsystem or further
1449  *      down in the scsi subsystem.
1450  **/
1451 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1452                     unsigned int cmd, unsigned long arg)
1453 {
1454         struct gendisk *disk = bdev->bd_disk;
1455         struct scsi_disk *sdkp = scsi_disk(disk);
1456         struct scsi_device *sdp = sdkp->device;
1457         void __user *p = (void __user *)arg;
1458         int error;
1459     
1460         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1461                                     "cmd=0x%x\n", disk->disk_name, cmd));
1462
1463         error = scsi_verify_blk_ioctl(bdev, cmd);
1464         if (error < 0)
1465                 return error;
1466
1467         /*
1468          * If we are in the middle of error recovery, don't let anyone
1469          * else try and use this device.  Also, if error recovery fails, it
1470          * may try and take the device offline, in which case all further
1471          * access to the device is prohibited.
1472          */
1473         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1474                         (mode & FMODE_NDELAY) != 0);
1475         if (error)
1476                 goto out;
1477
1478         if (is_sed_ioctl(cmd))
1479                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1480
1481         /*
1482          * Send SCSI addressing ioctls directly to mid level, send other
1483          * ioctls to block level and then onto mid level if they can't be
1484          * resolved.
1485          */
1486         switch (cmd) {
1487                 case SCSI_IOCTL_GET_IDLUN:
1488                 case SCSI_IOCTL_GET_BUS_NUMBER:
1489                         error = scsi_ioctl(sdp, cmd, p);
1490                         break;
1491                 default:
1492                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1493                         if (error != -ENOTTY)
1494                                 break;
1495                         error = scsi_ioctl(sdp, cmd, p);
1496                         break;
1497         }
1498 out:
1499         return error;
1500 }
1501
1502 static void set_media_not_present(struct scsi_disk *sdkp)
1503 {
1504         if (sdkp->media_present)
1505                 sdkp->device->changed = 1;
1506
1507         if (sdkp->device->removable) {
1508                 sdkp->media_present = 0;
1509                 sdkp->capacity = 0;
1510         }
1511 }
1512
1513 static int media_not_present(struct scsi_disk *sdkp,
1514                              struct scsi_sense_hdr *sshdr)
1515 {
1516         if (!scsi_sense_valid(sshdr))
1517                 return 0;
1518
1519         /* not invoked for commands that could return deferred errors */
1520         switch (sshdr->sense_key) {
1521         case UNIT_ATTENTION:
1522         case NOT_READY:
1523                 /* medium not present */
1524                 if (sshdr->asc == 0x3A) {
1525                         set_media_not_present(sdkp);
1526                         return 1;
1527                 }
1528         }
1529         return 0;
1530 }
1531
1532 /**
1533  *      sd_check_events - check media events
1534  *      @disk: kernel device descriptor
1535  *      @clearing: disk events currently being cleared
1536  *
1537  *      Returns mask of DISK_EVENT_*.
1538  *
1539  *      Note: this function is invoked from the block subsystem.
1540  **/
1541 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1542 {
1543         struct scsi_disk *sdkp = scsi_disk_get(disk);
1544         struct scsi_device *sdp;
1545         int retval;
1546
1547         if (!sdkp)
1548                 return 0;
1549
1550         sdp = sdkp->device;
1551         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1552
1553         /*
1554          * If the device is offline, don't send any commands - just pretend as
1555          * if the command failed.  If the device ever comes back online, we
1556          * can deal with it then.  It is only because of unrecoverable errors
1557          * that we would ever take a device offline in the first place.
1558          */
1559         if (!scsi_device_online(sdp)) {
1560                 set_media_not_present(sdkp);
1561                 goto out;
1562         }
1563
1564         /*
1565          * Using TEST_UNIT_READY enables differentiation between drive with
1566          * no cartridge loaded - NOT READY, drive with changed cartridge -
1567          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1568          *
1569          * Drives that auto spin down. eg iomega jaz 1G, will be started
1570          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1571          * sd_revalidate() is called.
1572          */
1573         if (scsi_block_when_processing_errors(sdp)) {
1574                 struct scsi_sense_hdr sshdr = { 0, };
1575
1576                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1577                                               &sshdr);
1578
1579                 /* failed to execute TUR, assume media not present */
1580                 if (host_byte(retval)) {
1581                         set_media_not_present(sdkp);
1582                         goto out;
1583                 }
1584
1585                 if (media_not_present(sdkp, &sshdr))
1586                         goto out;
1587         }
1588
1589         /*
1590          * For removable scsi disk we have to recognise the presence
1591          * of a disk in the drive.
1592          */
1593         if (!sdkp->media_present)
1594                 sdp->changed = 1;
1595         sdkp->media_present = 1;
1596 out:
1597         /*
1598          * sdp->changed is set under the following conditions:
1599          *
1600          *      Medium present state has changed in either direction.
1601          *      Device has indicated UNIT_ATTENTION.
1602          */
1603         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1604         sdp->changed = 0;
1605         scsi_disk_put(sdkp);
1606         return retval;
1607 }
1608
1609 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1610 {
1611         int retries, res;
1612         struct scsi_device *sdp = sdkp->device;
1613         const int timeout = sdp->request_queue->rq_timeout
1614                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1615         struct scsi_sense_hdr my_sshdr;
1616
1617         if (!scsi_device_online(sdp))
1618                 return -ENODEV;
1619
1620         /* caller might not be interested in sense, but we need it */
1621         if (!sshdr)
1622                 sshdr = &my_sshdr;
1623
1624         for (retries = 3; retries > 0; --retries) {
1625                 unsigned char cmd[10] = { 0 };
1626
1627                 cmd[0] = SYNCHRONIZE_CACHE;
1628                 /*
1629                  * Leave the rest of the command zero to indicate
1630                  * flush everything.
1631                  */
1632                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1633                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1634                 if (res == 0)
1635                         break;
1636         }
1637
1638         if (res) {
1639                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1640
1641                 if (driver_byte(res) == DRIVER_SENSE)
1642                         sd_print_sense_hdr(sdkp, sshdr);
1643
1644                 /* we need to evaluate the error return  */
1645                 if (scsi_sense_valid(sshdr) &&
1646                         (sshdr->asc == 0x3a ||  /* medium not present */
1647                          sshdr->asc == 0x20))   /* invalid command */
1648                                 /* this is no error here */
1649                                 return 0;
1650
1651                 switch (host_byte(res)) {
1652                 /* ignore errors due to racing a disconnection */
1653                 case DID_BAD_TARGET:
1654                 case DID_NO_CONNECT:
1655                         return 0;
1656                 /* signal the upper layer it might try again */
1657                 case DID_BUS_BUSY:
1658                 case DID_IMM_RETRY:
1659                 case DID_REQUEUE:
1660                 case DID_SOFT_ERROR:
1661                         return -EBUSY;
1662                 default:
1663                         return -EIO;
1664                 }
1665         }
1666         return 0;
1667 }
1668
1669 static void sd_rescan(struct device *dev)
1670 {
1671         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1672
1673         revalidate_disk(sdkp->disk);
1674 }
1675
1676
1677 #ifdef CONFIG_COMPAT
1678 /* 
1679  * This gets directly called from VFS. When the ioctl 
1680  * is not recognized we go back to the other translation paths. 
1681  */
1682 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1683                            unsigned int cmd, unsigned long arg)
1684 {
1685         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1686         int error;
1687
1688         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1689                         (mode & FMODE_NDELAY) != 0);
1690         if (error)
1691                 return error;
1692                
1693         /* 
1694          * Let the static ioctl translation table take care of it.
1695          */
1696         if (!sdev->host->hostt->compat_ioctl)
1697                 return -ENOIOCTLCMD; 
1698         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1699 }
1700 #endif
1701
1702 static char sd_pr_type(enum pr_type type)
1703 {
1704         switch (type) {
1705         case PR_WRITE_EXCLUSIVE:
1706                 return 0x01;
1707         case PR_EXCLUSIVE_ACCESS:
1708                 return 0x03;
1709         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1710                 return 0x05;
1711         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1712                 return 0x06;
1713         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1714                 return 0x07;
1715         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1716                 return 0x08;
1717         default:
1718                 return 0;
1719         }
1720 };
1721
1722 static int sd_pr_command(struct block_device *bdev, u8 sa,
1723                 u64 key, u64 sa_key, u8 type, u8 flags)
1724 {
1725         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1726         struct scsi_sense_hdr sshdr;
1727         int result;
1728         u8 cmd[16] = { 0, };
1729         u8 data[24] = { 0, };
1730
1731         cmd[0] = PERSISTENT_RESERVE_OUT;
1732         cmd[1] = sa;
1733         cmd[2] = type;
1734         put_unaligned_be32(sizeof(data), &cmd[5]);
1735
1736         put_unaligned_be64(key, &data[0]);
1737         put_unaligned_be64(sa_key, &data[8]);
1738         data[20] = flags;
1739
1740         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1741                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1742
1743         if (driver_byte(result) == DRIVER_SENSE &&
1744             scsi_sense_valid(&sshdr)) {
1745                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1746                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1747         }
1748
1749         return result;
1750 }
1751
1752 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1753                 u32 flags)
1754 {
1755         if (flags & ~PR_FL_IGNORE_KEY)
1756                 return -EOPNOTSUPP;
1757         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1758                         old_key, new_key, 0,
1759                         (1 << 0) /* APTPL */);
1760 }
1761
1762 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1763                 u32 flags)
1764 {
1765         if (flags)
1766                 return -EOPNOTSUPP;
1767         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1768 }
1769
1770 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1771 {
1772         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1773 }
1774
1775 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1776                 enum pr_type type, bool abort)
1777 {
1778         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1779                              sd_pr_type(type), 0);
1780 }
1781
1782 static int sd_pr_clear(struct block_device *bdev, u64 key)
1783 {
1784         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1785 }
1786
1787 static const struct pr_ops sd_pr_ops = {
1788         .pr_register    = sd_pr_register,
1789         .pr_reserve     = sd_pr_reserve,
1790         .pr_release     = sd_pr_release,
1791         .pr_preempt     = sd_pr_preempt,
1792         .pr_clear       = sd_pr_clear,
1793 };
1794
1795 static const struct block_device_operations sd_fops = {
1796         .owner                  = THIS_MODULE,
1797         .open                   = sd_open,
1798         .release                = sd_release,
1799         .ioctl                  = sd_ioctl,
1800         .getgeo                 = sd_getgeo,
1801 #ifdef CONFIG_COMPAT
1802         .compat_ioctl           = sd_compat_ioctl,
1803 #endif
1804         .check_events           = sd_check_events,
1805         .revalidate_disk        = sd_revalidate_disk,
1806         .unlock_native_capacity = sd_unlock_native_capacity,
1807         .report_zones           = sd_zbc_report_zones,
1808         .pr_ops                 = &sd_pr_ops,
1809 };
1810
1811 /**
1812  *      sd_eh_reset - reset error handling callback
1813  *      @scmd:          sd-issued command that has failed
1814  *
1815  *      This function is called by the SCSI midlayer before starting
1816  *      SCSI EH. When counting medium access failures we have to be
1817  *      careful to register it only only once per device and SCSI EH run;
1818  *      there might be several timed out commands which will cause the
1819  *      'max_medium_access_timeouts' counter to trigger after the first
1820  *      SCSI EH run already and set the device to offline.
1821  *      So this function resets the internal counter before starting SCSI EH.
1822  **/
1823 static void sd_eh_reset(struct scsi_cmnd *scmd)
1824 {
1825         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1826
1827         /* New SCSI EH run, reset gate variable */
1828         sdkp->ignore_medium_access_errors = false;
1829 }
1830
1831 /**
1832  *      sd_eh_action - error handling callback
1833  *      @scmd:          sd-issued command that has failed
1834  *      @eh_disp:       The recovery disposition suggested by the midlayer
1835  *
1836  *      This function is called by the SCSI midlayer upon completion of an
1837  *      error test command (currently TEST UNIT READY). The result of sending
1838  *      the eh command is passed in eh_disp.  We're looking for devices that
1839  *      fail medium access commands but are OK with non access commands like
1840  *      test unit ready (so wrongly see the device as having a successful
1841  *      recovery)
1842  **/
1843 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1844 {
1845         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1846         struct scsi_device *sdev = scmd->device;
1847
1848         if (!scsi_device_online(sdev) ||
1849             !scsi_medium_access_command(scmd) ||
1850             host_byte(scmd->result) != DID_TIME_OUT ||
1851             eh_disp != SUCCESS)
1852                 return eh_disp;
1853
1854         /*
1855          * The device has timed out executing a medium access command.
1856          * However, the TEST UNIT READY command sent during error
1857          * handling completed successfully. Either the device is in the
1858          * process of recovering or has it suffered an internal failure
1859          * that prevents access to the storage medium.
1860          */
1861         if (!sdkp->ignore_medium_access_errors) {
1862                 sdkp->medium_access_timed_out++;
1863                 sdkp->ignore_medium_access_errors = true;
1864         }
1865
1866         /*
1867          * If the device keeps failing read/write commands but TEST UNIT
1868          * READY always completes successfully we assume that medium
1869          * access is no longer possible and take the device offline.
1870          */
1871         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1872                 scmd_printk(KERN_ERR, scmd,
1873                             "Medium access timeout failure. Offlining disk!\n");
1874                 mutex_lock(&sdev->state_mutex);
1875                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1876                 mutex_unlock(&sdev->state_mutex);
1877
1878                 return SUCCESS;
1879         }
1880
1881         return eh_disp;
1882 }
1883
1884 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1885 {
1886         struct request *req = scmd->request;
1887         struct scsi_device *sdev = scmd->device;
1888         unsigned int transferred, good_bytes;
1889         u64 start_lba, end_lba, bad_lba;
1890
1891         /*
1892          * Some commands have a payload smaller than the device logical
1893          * block size (e.g. INQUIRY on a 4K disk).
1894          */
1895         if (scsi_bufflen(scmd) <= sdev->sector_size)
1896                 return 0;
1897
1898         /* Check if we have a 'bad_lba' information */
1899         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1900                                      SCSI_SENSE_BUFFERSIZE,
1901                                      &bad_lba))
1902                 return 0;
1903
1904         /*
1905          * If the bad lba was reported incorrectly, we have no idea where
1906          * the error is.
1907          */
1908         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1909         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1910         if (bad_lba < start_lba || bad_lba >= end_lba)
1911                 return 0;
1912
1913         /*
1914          * resid is optional but mostly filled in.  When it's unused,
1915          * its value is zero, so we assume the whole buffer transferred
1916          */
1917         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1918
1919         /* This computation should always be done in terms of the
1920          * resolution of the device's medium.
1921          */
1922         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1923
1924         return min(good_bytes, transferred);
1925 }
1926
1927 /**
1928  *      sd_done - bottom half handler: called when the lower level
1929  *      driver has completed (successfully or otherwise) a scsi command.
1930  *      @SCpnt: mid-level's per command structure.
1931  *
1932  *      Note: potentially run from within an ISR. Must not block.
1933  **/
1934 static int sd_done(struct scsi_cmnd *SCpnt)
1935 {
1936         int result = SCpnt->result;
1937         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1938         unsigned int sector_size = SCpnt->device->sector_size;
1939         unsigned int resid;
1940         struct scsi_sense_hdr sshdr;
1941         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1942         struct request *req = SCpnt->request;
1943         int sense_valid = 0;
1944         int sense_deferred = 0;
1945
1946         switch (req_op(req)) {
1947         case REQ_OP_DISCARD:
1948         case REQ_OP_WRITE_ZEROES:
1949         case REQ_OP_WRITE_SAME:
1950         case REQ_OP_ZONE_RESET:
1951                 if (!result) {
1952                         good_bytes = blk_rq_bytes(req);
1953                         scsi_set_resid(SCpnt, 0);
1954                 } else {
1955                         good_bytes = 0;
1956                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1957                 }
1958                 break;
1959         default:
1960                 /*
1961                  * In case of bogus fw or device, we could end up having
1962                  * an unaligned partial completion. Check this here and force
1963                  * alignment.
1964                  */
1965                 resid = scsi_get_resid(SCpnt);
1966                 if (resid & (sector_size - 1)) {
1967                         sd_printk(KERN_INFO, sdkp,
1968                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1969                                 resid, sector_size);
1970                         resid = min(scsi_bufflen(SCpnt),
1971                                     round_up(resid, sector_size));
1972                         scsi_set_resid(SCpnt, resid);
1973                 }
1974         }
1975
1976         if (result) {
1977                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1978                 if (sense_valid)
1979                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1980         }
1981         sdkp->medium_access_timed_out = 0;
1982
1983         if (driver_byte(result) != DRIVER_SENSE &&
1984             (!sense_valid || sense_deferred))
1985                 goto out;
1986
1987         switch (sshdr.sense_key) {
1988         case HARDWARE_ERROR:
1989         case MEDIUM_ERROR:
1990                 good_bytes = sd_completed_bytes(SCpnt);
1991                 break;
1992         case RECOVERED_ERROR:
1993                 good_bytes = scsi_bufflen(SCpnt);
1994                 break;
1995         case NO_SENSE:
1996                 /* This indicates a false check condition, so ignore it.  An
1997                  * unknown amount of data was transferred so treat it as an
1998                  * error.
1999                  */
2000                 SCpnt->result = 0;
2001                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2002                 break;
2003         case ABORTED_COMMAND:
2004                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2005                         good_bytes = sd_completed_bytes(SCpnt);
2006                 break;
2007         case ILLEGAL_REQUEST:
2008                 switch (sshdr.asc) {
2009                 case 0x10:      /* DIX: Host detected corruption */
2010                         good_bytes = sd_completed_bytes(SCpnt);
2011                         break;
2012                 case 0x20:      /* INVALID COMMAND OPCODE */
2013                 case 0x24:      /* INVALID FIELD IN CDB */
2014                         switch (SCpnt->cmnd[0]) {
2015                         case UNMAP:
2016                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2017                                 break;
2018                         case WRITE_SAME_16:
2019                         case WRITE_SAME:
2020                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2021                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2022                                 } else {
2023                                         sdkp->device->no_write_same = 1;
2024                                         sd_config_write_same(sdkp);
2025                                         req->rq_flags |= RQF_QUIET;
2026                                 }
2027                                 break;
2028                         }
2029                 }
2030                 break;
2031         default:
2032                 break;
2033         }
2034
2035  out:
2036         if (sd_is_zoned(sdkp))
2037                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2038
2039         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2040                                            "sd_done: completed %d of %d bytes\n",
2041                                            good_bytes, scsi_bufflen(SCpnt)));
2042
2043         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2044             good_bytes)
2045                 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2046                                 good_bytes / scsi_prot_interval(SCpnt));
2047
2048         return good_bytes;
2049 }
2050
2051 /*
2052  * spinup disk - called only in sd_revalidate_disk()
2053  */
2054 static void
2055 sd_spinup_disk(struct scsi_disk *sdkp)
2056 {
2057         unsigned char cmd[10];
2058         unsigned long spintime_expire = 0;
2059         int retries, spintime;
2060         unsigned int the_result;
2061         struct scsi_sense_hdr sshdr;
2062         int sense_valid = 0;
2063
2064         spintime = 0;
2065
2066         /* Spin up drives, as required.  Only do this at boot time */
2067         /* Spinup needs to be done for module loads too. */
2068         do {
2069                 retries = 0;
2070
2071                 do {
2072                         cmd[0] = TEST_UNIT_READY;
2073                         memset((void *) &cmd[1], 0, 9);
2074
2075                         the_result = scsi_execute_req(sdkp->device, cmd,
2076                                                       DMA_NONE, NULL, 0,
2077                                                       &sshdr, SD_TIMEOUT,
2078                                                       SD_MAX_RETRIES, NULL);
2079
2080                         /*
2081                          * If the drive has indicated to us that it
2082                          * doesn't have any media in it, don't bother
2083                          * with any more polling.
2084                          */
2085                         if (media_not_present(sdkp, &sshdr))
2086                                 return;
2087
2088                         if (the_result)
2089                                 sense_valid = scsi_sense_valid(&sshdr);
2090                         retries++;
2091                 } while (retries < 3 && 
2092                          (!scsi_status_is_good(the_result) ||
2093                           ((driver_byte(the_result) == DRIVER_SENSE) &&
2094                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2095
2096                 if (driver_byte(the_result) != DRIVER_SENSE) {
2097                         /* no sense, TUR either succeeded or failed
2098                          * with a status error */
2099                         if(!spintime && !scsi_status_is_good(the_result)) {
2100                                 sd_print_result(sdkp, "Test Unit Ready failed",
2101                                                 the_result);
2102                         }
2103                         break;
2104                 }
2105
2106                 /*
2107                  * The device does not want the automatic start to be issued.
2108                  */
2109                 if (sdkp->device->no_start_on_add)
2110                         break;
2111
2112                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2113                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2114                                 break;  /* manual intervention required */
2115                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2116                                 break;  /* standby */
2117                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2118                                 break;  /* unavailable */
2119                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2120                                 break;  /* sanitize in progress */
2121                         /*
2122                          * Issue command to spin up drive when not ready
2123                          */
2124                         if (!spintime) {
2125                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2126                                 cmd[0] = START_STOP;
2127                                 cmd[1] = 1;     /* Return immediately */
2128                                 memset((void *) &cmd[2], 0, 8);
2129                                 cmd[4] = 1;     /* Start spin cycle */
2130                                 if (sdkp->device->start_stop_pwr_cond)
2131                                         cmd[4] |= 1 << 4;
2132                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2133                                                  NULL, 0, &sshdr,
2134                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2135                                                  NULL);
2136                                 spintime_expire = jiffies + 100 * HZ;
2137                                 spintime = 1;
2138                         }
2139                         /* Wait 1 second for next try */
2140                         msleep(1000);
2141                         printk(KERN_CONT ".");
2142
2143                 /*
2144                  * Wait for USB flash devices with slow firmware.
2145                  * Yes, this sense key/ASC combination shouldn't
2146                  * occur here.  It's characteristic of these devices.
2147                  */
2148                 } else if (sense_valid &&
2149                                 sshdr.sense_key == UNIT_ATTENTION &&
2150                                 sshdr.asc == 0x28) {
2151                         if (!spintime) {
2152                                 spintime_expire = jiffies + 5 * HZ;
2153                                 spintime = 1;
2154                         }
2155                         /* Wait 1 second for next try */
2156                         msleep(1000);
2157                 } else {
2158                         /* we don't understand the sense code, so it's
2159                          * probably pointless to loop */
2160                         if(!spintime) {
2161                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2162                                 sd_print_sense_hdr(sdkp, &sshdr);
2163                         }
2164                         break;
2165                 }
2166                                 
2167         } while (spintime && time_before_eq(jiffies, spintime_expire));
2168
2169         if (spintime) {
2170                 if (scsi_status_is_good(the_result))
2171                         printk(KERN_CONT "ready\n");
2172                 else
2173                         printk(KERN_CONT "not responding...\n");
2174         }
2175 }
2176
2177 /*
2178  * Determine whether disk supports Data Integrity Field.
2179  */
2180 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2181 {
2182         struct scsi_device *sdp = sdkp->device;
2183         u8 type;
2184         int ret = 0;
2185
2186         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2187                 return ret;
2188
2189         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2190
2191         if (type > T10_PI_TYPE3_PROTECTION)
2192                 ret = -ENODEV;
2193         else if (scsi_host_dif_capable(sdp->host, type))
2194                 ret = 1;
2195
2196         if (sdkp->first_scan || type != sdkp->protection_type)
2197                 switch (ret) {
2198                 case -ENODEV:
2199                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2200                                   " protection type %u. Disabling disk!\n",
2201                                   type);
2202                         break;
2203                 case 1:
2204                         sd_printk(KERN_NOTICE, sdkp,
2205                                   "Enabling DIF Type %u protection\n", type);
2206                         break;
2207                 case 0:
2208                         sd_printk(KERN_NOTICE, sdkp,
2209                                   "Disabling DIF Type %u protection\n", type);
2210                         break;
2211                 }
2212
2213         sdkp->protection_type = type;
2214
2215         return ret;
2216 }
2217
2218 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2219                         struct scsi_sense_hdr *sshdr, int sense_valid,
2220                         int the_result)
2221 {
2222         if (driver_byte(the_result) == DRIVER_SENSE)
2223                 sd_print_sense_hdr(sdkp, sshdr);
2224         else
2225                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2226
2227         /*
2228          * Set dirty bit for removable devices if not ready -
2229          * sometimes drives will not report this properly.
2230          */
2231         if (sdp->removable &&
2232             sense_valid && sshdr->sense_key == NOT_READY)
2233                 set_media_not_present(sdkp);
2234
2235         /*
2236          * We used to set media_present to 0 here to indicate no media
2237          * in the drive, but some drives fail read capacity even with
2238          * media present, so we can't do that.
2239          */
2240         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2241 }
2242
2243 #define RC16_LEN 32
2244 #if RC16_LEN > SD_BUF_SIZE
2245 #error RC16_LEN must not be more than SD_BUF_SIZE
2246 #endif
2247
2248 #define READ_CAPACITY_RETRIES_ON_RESET  10
2249
2250 /*
2251  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2252  * and the reported logical block size is bigger than 512 bytes. Note
2253  * that last_sector is a u64 and therefore logical_to_sectors() is not
2254  * applicable.
2255  */
2256 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2257 {
2258         u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2259
2260         if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2261                 return false;
2262
2263         return true;
2264 }
2265
2266 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2267                                                 unsigned char *buffer)
2268 {
2269         unsigned char cmd[16];
2270         struct scsi_sense_hdr sshdr;
2271         int sense_valid = 0;
2272         int the_result;
2273         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2274         unsigned int alignment;
2275         unsigned long long lba;
2276         unsigned sector_size;
2277
2278         if (sdp->no_read_capacity_16)
2279                 return -EINVAL;
2280
2281         do {
2282                 memset(cmd, 0, 16);
2283                 cmd[0] = SERVICE_ACTION_IN_16;
2284                 cmd[1] = SAI_READ_CAPACITY_16;
2285                 cmd[13] = RC16_LEN;
2286                 memset(buffer, 0, RC16_LEN);
2287
2288                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2289                                         buffer, RC16_LEN, &sshdr,
2290                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2291
2292                 if (media_not_present(sdkp, &sshdr))
2293                         return -ENODEV;
2294
2295                 if (the_result) {
2296                         sense_valid = scsi_sense_valid(&sshdr);
2297                         if (sense_valid &&
2298                             sshdr.sense_key == ILLEGAL_REQUEST &&
2299                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2300                             sshdr.ascq == 0x00)
2301                                 /* Invalid Command Operation Code or
2302                                  * Invalid Field in CDB, just retry
2303                                  * silently with RC10 */
2304                                 return -EINVAL;
2305                         if (sense_valid &&
2306                             sshdr.sense_key == UNIT_ATTENTION &&
2307                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2308                                 /* Device reset might occur several times,
2309                                  * give it one more chance */
2310                                 if (--reset_retries > 0)
2311                                         continue;
2312                 }
2313                 retries--;
2314
2315         } while (the_result && retries);
2316
2317         if (the_result) {
2318                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2319                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2320                 return -EINVAL;
2321         }
2322
2323         sector_size = get_unaligned_be32(&buffer[8]);
2324         lba = get_unaligned_be64(&buffer[0]);
2325
2326         if (sd_read_protection_type(sdkp, buffer) < 0) {
2327                 sdkp->capacity = 0;
2328                 return -ENODEV;
2329         }
2330
2331         if (!sd_addressable_capacity(lba, sector_size)) {
2332                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2333                         "kernel compiled with support for large block "
2334                         "devices.\n");
2335                 sdkp->capacity = 0;
2336                 return -EOVERFLOW;
2337         }
2338
2339         /* Logical blocks per physical block exponent */
2340         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2341
2342         /* RC basis */
2343         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2344
2345         /* Lowest aligned logical block */
2346         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2347         blk_queue_alignment_offset(sdp->request_queue, alignment);
2348         if (alignment && sdkp->first_scan)
2349                 sd_printk(KERN_NOTICE, sdkp,
2350                           "physical block alignment offset: %u\n", alignment);
2351
2352         if (buffer[14] & 0x80) { /* LBPME */
2353                 sdkp->lbpme = 1;
2354
2355                 if (buffer[14] & 0x40) /* LBPRZ */
2356                         sdkp->lbprz = 1;
2357
2358                 sd_config_discard(sdkp, SD_LBP_WS16);
2359         }
2360
2361         sdkp->capacity = lba + 1;
2362         return sector_size;
2363 }
2364
2365 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2366                                                 unsigned char *buffer)
2367 {
2368         unsigned char cmd[16];
2369         struct scsi_sense_hdr sshdr;
2370         int sense_valid = 0;
2371         int the_result;
2372         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2373         sector_t lba;
2374         unsigned sector_size;
2375
2376         do {
2377                 cmd[0] = READ_CAPACITY;
2378                 memset(&cmd[1], 0, 9);
2379                 memset(buffer, 0, 8);
2380
2381                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2382                                         buffer, 8, &sshdr,
2383                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2384
2385                 if (media_not_present(sdkp, &sshdr))
2386                         return -ENODEV;
2387
2388                 if (the_result) {
2389                         sense_valid = scsi_sense_valid(&sshdr);
2390                         if (sense_valid &&
2391                             sshdr.sense_key == UNIT_ATTENTION &&
2392                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2393                                 /* Device reset might occur several times,
2394                                  * give it one more chance */
2395                                 if (--reset_retries > 0)
2396                                         continue;
2397                 }
2398                 retries--;
2399
2400         } while (the_result && retries);
2401
2402         if (the_result) {
2403                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2404                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2405                 return -EINVAL;
2406         }
2407
2408         sector_size = get_unaligned_be32(&buffer[4]);
2409         lba = get_unaligned_be32(&buffer[0]);
2410
2411         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2412                 /* Some buggy (usb cardreader) devices return an lba of
2413                    0xffffffff when the want to report a size of 0 (with
2414                    which they really mean no media is present) */
2415                 sdkp->capacity = 0;
2416                 sdkp->physical_block_size = sector_size;
2417                 return sector_size;
2418         }
2419
2420         if (!sd_addressable_capacity(lba, sector_size)) {
2421                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2422                         "kernel compiled with support for large block "
2423                         "devices.\n");
2424                 sdkp->capacity = 0;
2425                 return -EOVERFLOW;
2426         }
2427
2428         sdkp->capacity = lba + 1;
2429         sdkp->physical_block_size = sector_size;
2430         return sector_size;
2431 }
2432
2433 static int sd_try_rc16_first(struct scsi_device *sdp)
2434 {
2435         if (sdp->host->max_cmd_len < 16)
2436                 return 0;
2437         if (sdp->try_rc_10_first)
2438                 return 0;
2439         if (sdp->scsi_level > SCSI_SPC_2)
2440                 return 1;
2441         if (scsi_device_protection(sdp))
2442                 return 1;
2443         return 0;
2444 }
2445
2446 /*
2447  * read disk capacity
2448  */
2449 static void
2450 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2451 {
2452         int sector_size;
2453         struct scsi_device *sdp = sdkp->device;
2454
2455         if (sd_try_rc16_first(sdp)) {
2456                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2457                 if (sector_size == -EOVERFLOW)
2458                         goto got_data;
2459                 if (sector_size == -ENODEV)
2460                         return;
2461                 if (sector_size < 0)
2462                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2463                 if (sector_size < 0)
2464                         return;
2465         } else {
2466                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2467                 if (sector_size == -EOVERFLOW)
2468                         goto got_data;
2469                 if (sector_size < 0)
2470                         return;
2471                 if ((sizeof(sdkp->capacity) > 4) &&
2472                     (sdkp->capacity > 0xffffffffULL)) {
2473                         int old_sector_size = sector_size;
2474                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2475                                         "Trying to use READ CAPACITY(16).\n");
2476                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2477                         if (sector_size < 0) {
2478                                 sd_printk(KERN_NOTICE, sdkp,
2479                                         "Using 0xffffffff as device size\n");
2480                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2481                                 sector_size = old_sector_size;
2482                                 goto got_data;
2483                         }
2484                         /* Remember that READ CAPACITY(16) succeeded */
2485                         sdp->try_rc_10_first = 0;
2486                 }
2487         }
2488
2489         /* Some devices are known to return the total number of blocks,
2490          * not the highest block number.  Some devices have versions
2491          * which do this and others which do not.  Some devices we might
2492          * suspect of doing this but we don't know for certain.
2493          *
2494          * If we know the reported capacity is wrong, decrement it.  If
2495          * we can only guess, then assume the number of blocks is even
2496          * (usually true but not always) and err on the side of lowering
2497          * the capacity.
2498          */
2499         if (sdp->fix_capacity ||
2500             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2501                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2502                                 "from its reported value: %llu\n",
2503                                 (unsigned long long) sdkp->capacity);
2504                 --sdkp->capacity;
2505         }
2506
2507 got_data:
2508         if (sector_size == 0) {
2509                 sector_size = 512;
2510                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2511                           "assuming 512.\n");
2512         }
2513
2514         if (sector_size != 512 &&
2515             sector_size != 1024 &&
2516             sector_size != 2048 &&
2517             sector_size != 4096) {
2518                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2519                           sector_size);
2520                 /*
2521                  * The user might want to re-format the drive with
2522                  * a supported sectorsize.  Once this happens, it
2523                  * would be relatively trivial to set the thing up.
2524                  * For this reason, we leave the thing in the table.
2525                  */
2526                 sdkp->capacity = 0;
2527                 /*
2528                  * set a bogus sector size so the normal read/write
2529                  * logic in the block layer will eventually refuse any
2530                  * request on this device without tripping over power
2531                  * of two sector size assumptions
2532                  */
2533                 sector_size = 512;
2534         }
2535         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2536         blk_queue_physical_block_size(sdp->request_queue,
2537                                       sdkp->physical_block_size);
2538         sdkp->device->sector_size = sector_size;
2539
2540         if (sdkp->capacity > 0xffffffff)
2541                 sdp->use_16_for_rw = 1;
2542
2543 }
2544
2545 /*
2546  * Print disk capacity
2547  */
2548 static void
2549 sd_print_capacity(struct scsi_disk *sdkp,
2550                   sector_t old_capacity)
2551 {
2552         int sector_size = sdkp->device->sector_size;
2553         char cap_str_2[10], cap_str_10[10];
2554
2555         string_get_size(sdkp->capacity, sector_size,
2556                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2557         string_get_size(sdkp->capacity, sector_size,
2558                         STRING_UNITS_10, cap_str_10,
2559                         sizeof(cap_str_10));
2560
2561         if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2562                 sd_printk(KERN_NOTICE, sdkp,
2563                           "%llu %d-byte logical blocks: (%s/%s)\n",
2564                           (unsigned long long)sdkp->capacity,
2565                           sector_size, cap_str_10, cap_str_2);
2566
2567                 if (sdkp->physical_block_size != sector_size)
2568                         sd_printk(KERN_NOTICE, sdkp,
2569                                   "%u-byte physical blocks\n",
2570                                   sdkp->physical_block_size);
2571
2572                 sd_zbc_print_zones(sdkp);
2573         }
2574 }
2575
2576 /* called with buffer of length 512 */
2577 static inline int
2578 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2579                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2580                  struct scsi_sense_hdr *sshdr)
2581 {
2582         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2583                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2584                                sshdr);
2585 }
2586
2587 /*
2588  * read write protect setting, if possible - called only in sd_revalidate_disk()
2589  * called with buffer of length SD_BUF_SIZE
2590  */
2591 static void
2592 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2593 {
2594         int res;
2595         struct scsi_device *sdp = sdkp->device;
2596         struct scsi_mode_data data;
2597         int disk_ro = get_disk_ro(sdkp->disk);
2598         int old_wp = sdkp->write_prot;
2599
2600         set_disk_ro(sdkp->disk, 0);
2601         if (sdp->skip_ms_page_3f) {
2602                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2603                 return;
2604         }
2605
2606         if (sdp->use_192_bytes_for_3f) {
2607                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2608         } else {
2609                 /*
2610                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2611                  * We have to start carefully: some devices hang if we ask
2612                  * for more than is available.
2613                  */
2614                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2615
2616                 /*
2617                  * Second attempt: ask for page 0 When only page 0 is
2618                  * implemented, a request for page 3F may return Sense Key
2619                  * 5: Illegal Request, Sense Code 24: Invalid field in
2620                  * CDB.
2621                  */
2622                 if (!scsi_status_is_good(res))
2623                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2624
2625                 /*
2626                  * Third attempt: ask 255 bytes, as we did earlier.
2627                  */
2628                 if (!scsi_status_is_good(res))
2629                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2630                                                &data, NULL);
2631         }
2632
2633         if (!scsi_status_is_good(res)) {
2634                 sd_first_printk(KERN_WARNING, sdkp,
2635                           "Test WP failed, assume Write Enabled\n");
2636         } else {
2637                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2638                 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2639                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2640                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2641                                   sdkp->write_prot ? "on" : "off");
2642                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2643                 }
2644         }
2645 }
2646
2647 /*
2648  * sd_read_cache_type - called only from sd_revalidate_disk()
2649  * called with buffer of length SD_BUF_SIZE
2650  */
2651 static void
2652 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2653 {
2654         int len = 0, res;
2655         struct scsi_device *sdp = sdkp->device;
2656
2657         int dbd;
2658         int modepage;
2659         int first_len;
2660         struct scsi_mode_data data;
2661         struct scsi_sense_hdr sshdr;
2662         int old_wce = sdkp->WCE;
2663         int old_rcd = sdkp->RCD;
2664         int old_dpofua = sdkp->DPOFUA;
2665
2666
2667         if (sdkp->cache_override)
2668                 return;
2669
2670         first_len = 4;
2671         if (sdp->skip_ms_page_8) {
2672                 if (sdp->type == TYPE_RBC)
2673                         goto defaults;
2674                 else {
2675                         if (sdp->skip_ms_page_3f)
2676                                 goto defaults;
2677                         modepage = 0x3F;
2678                         if (sdp->use_192_bytes_for_3f)
2679                                 first_len = 192;
2680                         dbd = 0;
2681                 }
2682         } else if (sdp->type == TYPE_RBC) {
2683                 modepage = 6;
2684                 dbd = 8;
2685         } else {
2686                 modepage = 8;
2687                 dbd = 0;
2688         }
2689
2690         /* cautiously ask */
2691         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2692                         &data, &sshdr);
2693
2694         if (!scsi_status_is_good(res))
2695                 goto bad_sense;
2696
2697         if (!data.header_length) {
2698                 modepage = 6;
2699                 first_len = 0;
2700                 sd_first_printk(KERN_ERR, sdkp,
2701                                 "Missing header in MODE_SENSE response\n");
2702         }
2703
2704         /* that went OK, now ask for the proper length */
2705         len = data.length;
2706
2707         /*
2708          * We're only interested in the first three bytes, actually.
2709          * But the data cache page is defined for the first 20.
2710          */
2711         if (len < 3)
2712                 goto bad_sense;
2713         else if (len > SD_BUF_SIZE) {
2714                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2715                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2716                 len = SD_BUF_SIZE;
2717         }
2718         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2719                 len = 192;
2720
2721         /* Get the data */
2722         if (len > first_len)
2723                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2724                                 &data, &sshdr);
2725
2726         if (scsi_status_is_good(res)) {
2727                 int offset = data.header_length + data.block_descriptor_length;
2728
2729                 while (offset < len) {
2730                         u8 page_code = buffer[offset] & 0x3F;
2731                         u8 spf       = buffer[offset] & 0x40;
2732
2733                         if (page_code == 8 || page_code == 6) {
2734                                 /* We're interested only in the first 3 bytes.
2735                                  */
2736                                 if (len - offset <= 2) {
2737                                         sd_first_printk(KERN_ERR, sdkp,
2738                                                 "Incomplete mode parameter "
2739                                                         "data\n");
2740                                         goto defaults;
2741                                 } else {
2742                                         modepage = page_code;
2743                                         goto Page_found;
2744                                 }
2745                         } else {
2746                                 /* Go to the next page */
2747                                 if (spf && len - offset > 3)
2748                                         offset += 4 + (buffer[offset+2] << 8) +
2749                                                 buffer[offset+3];
2750                                 else if (!spf && len - offset > 1)
2751                                         offset += 2 + buffer[offset+1];
2752                                 else {
2753                                         sd_first_printk(KERN_ERR, sdkp,
2754                                                         "Incomplete mode "
2755                                                         "parameter data\n");
2756                                         goto defaults;
2757                                 }
2758                         }
2759                 }
2760
2761                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2762                 goto defaults;
2763
2764         Page_found:
2765                 if (modepage == 8) {
2766                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2767                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2768                 } else {
2769                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2770                         sdkp->RCD = 0;
2771                 }
2772
2773                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2774                 if (sdp->broken_fua) {
2775                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2776                         sdkp->DPOFUA = 0;
2777                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2778                            !sdkp->device->use_16_for_rw) {
2779                         sd_first_printk(KERN_NOTICE, sdkp,
2780                                   "Uses READ/WRITE(6), disabling FUA\n");
2781                         sdkp->DPOFUA = 0;
2782                 }
2783
2784                 /* No cache flush allowed for write protected devices */
2785                 if (sdkp->WCE && sdkp->write_prot)
2786                         sdkp->WCE = 0;
2787
2788                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2789                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2790                         sd_printk(KERN_NOTICE, sdkp,
2791                                   "Write cache: %s, read cache: %s, %s\n",
2792                                   sdkp->WCE ? "enabled" : "disabled",
2793                                   sdkp->RCD ? "disabled" : "enabled",
2794                                   sdkp->DPOFUA ? "supports DPO and FUA"
2795                                   : "doesn't support DPO or FUA");
2796
2797                 return;
2798         }
2799
2800 bad_sense:
2801         if (scsi_sense_valid(&sshdr) &&
2802             sshdr.sense_key == ILLEGAL_REQUEST &&
2803             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2804                 /* Invalid field in CDB */
2805                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2806         else
2807                 sd_first_printk(KERN_ERR, sdkp,
2808                                 "Asking for cache data failed\n");
2809
2810 defaults:
2811         if (sdp->wce_default_on) {
2812                 sd_first_printk(KERN_NOTICE, sdkp,
2813                                 "Assuming drive cache: write back\n");
2814                 sdkp->WCE = 1;
2815         } else {
2816                 sd_first_printk(KERN_ERR, sdkp,
2817                                 "Assuming drive cache: write through\n");
2818                 sdkp->WCE = 0;
2819         }
2820         sdkp->RCD = 0;
2821         sdkp->DPOFUA = 0;
2822 }
2823
2824 /*
2825  * The ATO bit indicates whether the DIF application tag is available
2826  * for use by the operating system.
2827  */
2828 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2829 {
2830         int res, offset;
2831         struct scsi_device *sdp = sdkp->device;
2832         struct scsi_mode_data data;
2833         struct scsi_sense_hdr sshdr;
2834
2835         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2836                 return;
2837
2838         if (sdkp->protection_type == 0)
2839                 return;
2840
2841         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2842                               SD_MAX_RETRIES, &data, &sshdr);
2843
2844         if (!scsi_status_is_good(res) || !data.header_length ||
2845             data.length < 6) {
2846                 sd_first_printk(KERN_WARNING, sdkp,
2847                           "getting Control mode page failed, assume no ATO\n");
2848
2849                 if (scsi_sense_valid(&sshdr))
2850                         sd_print_sense_hdr(sdkp, &sshdr);
2851
2852                 return;
2853         }
2854
2855         offset = data.header_length + data.block_descriptor_length;
2856
2857         if ((buffer[offset] & 0x3f) != 0x0a) {
2858                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2859                 return;
2860         }
2861
2862         if ((buffer[offset + 5] & 0x80) == 0)
2863                 return;
2864
2865         sdkp->ATO = 1;
2866
2867         return;
2868 }
2869
2870 /**
2871  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2872  * @sdkp: disk to query
2873  */
2874 static void sd_read_block_limits(struct scsi_disk *sdkp)
2875 {
2876         unsigned int sector_sz = sdkp->device->sector_size;
2877         const int vpd_len = 64;
2878         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2879
2880         if (!buffer ||
2881             /* Block Limits VPD */
2882             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2883                 goto out;
2884
2885         blk_queue_io_min(sdkp->disk->queue,
2886                          get_unaligned_be16(&buffer[6]) * sector_sz);
2887
2888         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2889         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2890
2891         if (buffer[3] == 0x3c) {
2892                 unsigned int lba_count, desc_count;
2893
2894                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2895
2896                 if (!sdkp->lbpme)
2897                         goto out;
2898
2899                 lba_count = get_unaligned_be32(&buffer[20]);
2900                 desc_count = get_unaligned_be32(&buffer[24]);
2901
2902                 if (lba_count && desc_count)
2903                         sdkp->max_unmap_blocks = lba_count;
2904
2905                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2906
2907                 if (buffer[32] & 0x80)
2908                         sdkp->unmap_alignment =
2909                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2910
2911                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2912
2913                         if (sdkp->max_unmap_blocks)
2914                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2915                         else
2916                                 sd_config_discard(sdkp, SD_LBP_WS16);
2917
2918                 } else {        /* LBP VPD page tells us what to use */
2919                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2920                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2921                         else if (sdkp->lbpws)
2922                                 sd_config_discard(sdkp, SD_LBP_WS16);
2923                         else if (sdkp->lbpws10)
2924                                 sd_config_discard(sdkp, SD_LBP_WS10);
2925                         else
2926                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2927                 }
2928         }
2929
2930  out:
2931         kfree(buffer);
2932 }
2933
2934 /**
2935  * sd_read_block_characteristics - Query block dev. characteristics
2936  * @sdkp: disk to query
2937  */
2938 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2939 {
2940         struct request_queue *q = sdkp->disk->queue;
2941         unsigned char *buffer;
2942         u16 rot;
2943         const int vpd_len = 64;
2944
2945         buffer = kmalloc(vpd_len, GFP_KERNEL);
2946
2947         if (!buffer ||
2948             /* Block Device Characteristics VPD */
2949             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2950                 goto out;
2951
2952         rot = get_unaligned_be16(&buffer[4]);
2953
2954         if (rot == 1) {
2955                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2956                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2957         } else {
2958                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2959                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
2960         }
2961
2962         if (sdkp->device->type == TYPE_ZBC) {
2963                 /* Host-managed */
2964                 q->limits.zoned = BLK_ZONED_HM;
2965         } else {
2966                 sdkp->zoned = (buffer[8] >> 4) & 3;
2967                 if (sdkp->zoned == 1)
2968                         /* Host-aware */
2969                         q->limits.zoned = BLK_ZONED_HA;
2970                 else
2971                         /*
2972                          * Treat drive-managed devices as
2973                          * regular block devices.
2974                          */
2975                         q->limits.zoned = BLK_ZONED_NONE;
2976         }
2977         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2978                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2979                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2980
2981  out:
2982         kfree(buffer);
2983 }
2984
2985 /**
2986  * sd_read_block_provisioning - Query provisioning VPD page
2987  * @sdkp: disk to query
2988  */
2989 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2990 {
2991         unsigned char *buffer;
2992         const int vpd_len = 8;
2993
2994         if (sdkp->lbpme == 0)
2995                 return;
2996
2997         buffer = kmalloc(vpd_len, GFP_KERNEL);
2998
2999         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3000                 goto out;
3001
3002         sdkp->lbpvpd    = 1;
3003         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3004         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3005         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3006
3007  out:
3008         kfree(buffer);
3009 }
3010
3011 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3012 {
3013         struct scsi_device *sdev = sdkp->device;
3014
3015         if (sdev->host->no_write_same) {
3016                 sdev->no_write_same = 1;
3017
3018                 return;
3019         }
3020
3021         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3022                 /* too large values might cause issues with arcmsr */
3023                 int vpd_buf_len = 64;
3024
3025                 sdev->no_report_opcodes = 1;
3026
3027                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3028                  * CODES is unsupported and the device has an ATA
3029                  * Information VPD page (SAT).
3030                  */
3031                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3032                         sdev->no_write_same = 1;
3033         }
3034
3035         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3036                 sdkp->ws16 = 1;
3037
3038         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3039                 sdkp->ws10 = 1;
3040 }
3041
3042 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3043 {
3044         struct scsi_device *sdev = sdkp->device;
3045
3046         if (!sdev->security_supported)
3047                 return;
3048
3049         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3050                         SECURITY_PROTOCOL_IN) == 1 &&
3051             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3052                         SECURITY_PROTOCOL_OUT) == 1)
3053                 sdkp->security = 1;
3054 }
3055
3056 /**
3057  *      sd_revalidate_disk - called the first time a new disk is seen,
3058  *      performs disk spin up, read_capacity, etc.
3059  *      @disk: struct gendisk we care about
3060  **/
3061 static int sd_revalidate_disk(struct gendisk *disk)
3062 {
3063         struct scsi_disk *sdkp = scsi_disk(disk);
3064         struct scsi_device *sdp = sdkp->device;
3065         struct request_queue *q = sdkp->disk->queue;
3066         sector_t old_capacity = sdkp->capacity;
3067         unsigned char *buffer;
3068         unsigned int dev_max, rw_max;
3069
3070         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3071                                       "sd_revalidate_disk\n"));
3072
3073         /*
3074          * If the device is offline, don't try and read capacity or any
3075          * of the other niceties.
3076          */
3077         if (!scsi_device_online(sdp))
3078                 goto out;
3079
3080         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3081         if (!buffer) {
3082                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3083                           "allocation failure.\n");
3084                 goto out;
3085         }
3086
3087         sd_spinup_disk(sdkp);
3088
3089         /*
3090          * Without media there is no reason to ask; moreover, some devices
3091          * react badly if we do.
3092          */
3093         if (sdkp->media_present) {
3094                 sd_read_capacity(sdkp, buffer);
3095
3096                 if (scsi_device_supports_vpd(sdp)) {
3097                         sd_read_block_provisioning(sdkp);
3098                         sd_read_block_limits(sdkp);
3099                         sd_read_block_characteristics(sdkp);
3100                         sd_zbc_read_zones(sdkp, buffer);
3101                 }
3102
3103                 sd_print_capacity(sdkp, old_capacity);
3104
3105                 sd_read_write_protect_flag(sdkp, buffer);
3106                 sd_read_cache_type(sdkp, buffer);
3107                 sd_read_app_tag_own(sdkp, buffer);
3108                 sd_read_write_same(sdkp, buffer);
3109                 sd_read_security(sdkp, buffer);
3110         }
3111
3112         /*
3113          * We now have all cache related info, determine how we deal
3114          * with flush requests.
3115          */
3116         sd_set_flush_flag(sdkp);
3117
3118         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3119         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3120
3121         /* Some devices report a maximum block count for READ/WRITE requests. */
3122         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3123         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3124
3125         /*
3126          * Determine the device's preferred I/O size for reads and writes
3127          * unless the reported value is unreasonably small, large, or
3128          * garbage.
3129          */
3130         if (sdkp->opt_xfer_blocks &&
3131             sdkp->opt_xfer_blocks <= dev_max &&
3132             sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3133             logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3134                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3135                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3136         } else
3137                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3138                                       (sector_t)BLK_DEF_MAX_SECTORS);
3139
3140         /* Do not exceed controller limit */
3141         rw_max = min(rw_max, queue_max_hw_sectors(q));
3142
3143         /*
3144          * Only update max_sectors if previously unset or if the current value
3145          * exceeds the capabilities of the hardware.
3146          */
3147         if (sdkp->first_scan ||
3148             q->limits.max_sectors > q->limits.max_dev_sectors ||
3149             q->limits.max_sectors > q->limits.max_hw_sectors)
3150                 q->limits.max_sectors = rw_max;
3151
3152         sdkp->first_scan = 0;
3153
3154         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3155         sd_config_write_same(sdkp);
3156         kfree(buffer);
3157
3158  out:
3159         return 0;
3160 }
3161
3162 /**
3163  *      sd_unlock_native_capacity - unlock native capacity
3164  *      @disk: struct gendisk to set capacity for
3165  *
3166  *      Block layer calls this function if it detects that partitions
3167  *      on @disk reach beyond the end of the device.  If the SCSI host
3168  *      implements ->unlock_native_capacity() method, it's invoked to
3169  *      give it a chance to adjust the device capacity.
3170  *
3171  *      CONTEXT:
3172  *      Defined by block layer.  Might sleep.
3173  */
3174 static void sd_unlock_native_capacity(struct gendisk *disk)
3175 {
3176         struct scsi_device *sdev = scsi_disk(disk)->device;
3177
3178         if (sdev->host->hostt->unlock_native_capacity)
3179                 sdev->host->hostt->unlock_native_capacity(sdev);
3180 }
3181
3182 /**
3183  *      sd_format_disk_name - format disk name
3184  *      @prefix: name prefix - ie. "sd" for SCSI disks
3185  *      @index: index of the disk to format name for
3186  *      @buf: output buffer
3187  *      @buflen: length of the output buffer
3188  *
3189  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3190  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3191  *      which is followed by sdaaa.
3192  *
3193  *      This is basically 26 base counting with one extra 'nil' entry
3194  *      at the beginning from the second digit on and can be
3195  *      determined using similar method as 26 base conversion with the
3196  *      index shifted -1 after each digit is computed.
3197  *
3198  *      CONTEXT:
3199  *      Don't care.
3200  *
3201  *      RETURNS:
3202  *      0 on success, -errno on failure.
3203  */
3204 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3205 {
3206         const int base = 'z' - 'a' + 1;
3207         char *begin = buf + strlen(prefix);
3208         char *end = buf + buflen;
3209         char *p;
3210         int unit;
3211
3212         p = end - 1;
3213         *p = '\0';
3214         unit = base;
3215         do {
3216                 if (p == begin)
3217                         return -EINVAL;
3218                 *--p = 'a' + (index % unit);
3219                 index = (index / unit) - 1;
3220         } while (index >= 0);
3221
3222         memmove(begin, p, end - p);
3223         memcpy(buf, prefix, strlen(prefix));
3224
3225         return 0;
3226 }
3227
3228 /*