Merge tag 'scsi-postmerge' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb...
[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/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/sed-opal.h>
54 #include <linux/pm_runtime.h>
55 #include <linux/pr.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
59
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
69
70 #include "sd.h"
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
73
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
77
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
98
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
100 #define SD_MINORS       16
101 #else
102 #define SD_MINORS       0
103 #endif
104
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int  sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int  sd_probe(struct device *);
110 static int  sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
125
126 static DEFINE_SPINLOCK(sd_index_lock);
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
137 static const char *sd_cache_types[] = {
138         "write through", "none", "write back",
139         "write back, no read (daft)"
140 };
141
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
143 {
144         bool wc = false, fua = false;
145
146         if (sdkp->WCE) {
147                 wc = true;
148                 if (sdkp->DPOFUA)
149                         fua = true;
150         }
151
152         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
153 }
154
155 static ssize_t
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157                  const char *buf, size_t count)
158 {
159         int ct, rcd, wce, sp;
160         struct scsi_disk *sdkp = to_scsi_disk(dev);
161         struct scsi_device *sdp = sdkp->device;
162         char buffer[64];
163         char *buffer_data;
164         struct scsi_mode_data data;
165         struct scsi_sense_hdr sshdr;
166         static const char temp[] = "temporary ";
167         int len;
168
169         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170                 /* no cache control on RBC devices; theoretically they
171                  * can do it, but there's probably so many exceptions
172                  * it's not worth the risk */
173                 return -EINVAL;
174
175         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176                 buf += sizeof(temp) - 1;
177                 sdkp->cache_override = 1;
178         } else {
179                 sdkp->cache_override = 0;
180         }
181
182         ct = sysfs_match_string(sd_cache_types, buf);
183         if (ct < 0)
184                 return -EINVAL;
185
186         rcd = ct & 0x01 ? 1 : 0;
187         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
188
189         if (sdkp->cache_override) {
190                 sdkp->WCE = wce;
191                 sdkp->RCD = rcd;
192                 sd_set_flush_flag(sdkp);
193                 return count;
194         }
195
196         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197                             SD_MAX_RETRIES, &data, NULL))
198                 return -EINVAL;
199         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200                   data.block_descriptor_length);
201         buffer_data = buffer + data.header_length +
202                 data.block_descriptor_length;
203         buffer_data[2] &= ~0x05;
204         buffer_data[2] |= wce << 2 | rcd;
205         sp = buffer_data[0] & 0x80 ? 1 : 0;
206         buffer_data[0] &= ~0x80;
207
208         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209                              SD_MAX_RETRIES, &data, &sshdr)) {
210                 if (scsi_sense_valid(&sshdr))
211                         sd_print_sense_hdr(sdkp, &sshdr);
212                 return -EINVAL;
213         }
214         revalidate_disk(sdkp->disk);
215         return count;
216 }
217
218 static ssize_t
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
220                        char *buf)
221 {
222         struct scsi_disk *sdkp = to_scsi_disk(dev);
223         struct scsi_device *sdp = sdkp->device;
224
225         return sprintf(buf, "%u\n", sdp->manage_start_stop);
226 }
227
228 static ssize_t
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230                         const char *buf, size_t count)
231 {
232         struct scsi_disk *sdkp = to_scsi_disk(dev);
233         struct scsi_device *sdp = sdkp->device;
234         bool v;
235
236         if (!capable(CAP_SYS_ADMIN))
237                 return -EACCES;
238
239         if (kstrtobool(buf, &v))
240                 return -EINVAL;
241
242         sdp->manage_start_stop = v;
243
244         return count;
245 }
246 static DEVICE_ATTR_RW(manage_start_stop);
247
248 static ssize_t
249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
250 {
251         struct scsi_disk *sdkp = to_scsi_disk(dev);
252
253         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
254 }
255
256 static ssize_t
257 allow_restart_store(struct device *dev, struct device_attribute *attr,
258                     const char *buf, size_t count)
259 {
260         bool v;
261         struct scsi_disk *sdkp = to_scsi_disk(dev);
262         struct scsi_device *sdp = sdkp->device;
263
264         if (!capable(CAP_SYS_ADMIN))
265                 return -EACCES;
266
267         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
268                 return -EINVAL;
269
270         if (kstrtobool(buf, &v))
271                 return -EINVAL;
272
273         sdp->allow_restart = v;
274
275         return count;
276 }
277 static DEVICE_ATTR_RW(allow_restart);
278
279 static ssize_t
280 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
281 {
282         struct scsi_disk *sdkp = to_scsi_disk(dev);
283         int ct = sdkp->RCD + 2*sdkp->WCE;
284
285         return sprintf(buf, "%s\n", sd_cache_types[ct]);
286 }
287 static DEVICE_ATTR_RW(cache_type);
288
289 static ssize_t
290 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
291 {
292         struct scsi_disk *sdkp = to_scsi_disk(dev);
293
294         return sprintf(buf, "%u\n", sdkp->DPOFUA);
295 }
296 static DEVICE_ATTR_RO(FUA);
297
298 static ssize_t
299 protection_type_show(struct device *dev, struct device_attribute *attr,
300                      char *buf)
301 {
302         struct scsi_disk *sdkp = to_scsi_disk(dev);
303
304         return sprintf(buf, "%u\n", sdkp->protection_type);
305 }
306
307 static ssize_t
308 protection_type_store(struct device *dev, struct device_attribute *attr,
309                       const char *buf, size_t count)
310 {
311         struct scsi_disk *sdkp = to_scsi_disk(dev);
312         unsigned int val;
313         int err;
314
315         if (!capable(CAP_SYS_ADMIN))
316                 return -EACCES;
317
318         err = kstrtouint(buf, 10, &val);
319
320         if (err)
321                 return err;
322
323         if (val <= T10_PI_TYPE3_PROTECTION)
324                 sdkp->protection_type = val;
325
326         return count;
327 }
328 static DEVICE_ATTR_RW(protection_type);
329
330 static ssize_t
331 protection_mode_show(struct device *dev, struct device_attribute *attr,
332                      char *buf)
333 {
334         struct scsi_disk *sdkp = to_scsi_disk(dev);
335         struct scsi_device *sdp = sdkp->device;
336         unsigned int dif, dix;
337
338         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
339         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
340
341         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
342                 dif = 0;
343                 dix = 1;
344         }
345
346         if (!dif && !dix)
347                 return sprintf(buf, "none\n");
348
349         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
350 }
351 static DEVICE_ATTR_RO(protection_mode);
352
353 static ssize_t
354 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
355 {
356         struct scsi_disk *sdkp = to_scsi_disk(dev);
357
358         return sprintf(buf, "%u\n", sdkp->ATO);
359 }
360 static DEVICE_ATTR_RO(app_tag_own);
361
362 static ssize_t
363 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
364                        char *buf)
365 {
366         struct scsi_disk *sdkp = to_scsi_disk(dev);
367
368         return sprintf(buf, "%u\n", sdkp->lbpme);
369 }
370 static DEVICE_ATTR_RO(thin_provisioning);
371
372 /* sysfs_match_string() requires dense arrays */
373 static const char *lbp_mode[] = {
374         [SD_LBP_FULL]           = "full",
375         [SD_LBP_UNMAP]          = "unmap",
376         [SD_LBP_WS16]           = "writesame_16",
377         [SD_LBP_WS10]           = "writesame_10",
378         [SD_LBP_ZERO]           = "writesame_zero",
379         [SD_LBP_DISABLE]        = "disabled",
380 };
381
382 static ssize_t
383 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
384                        char *buf)
385 {
386         struct scsi_disk *sdkp = to_scsi_disk(dev);
387
388         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
389 }
390
391 static ssize_t
392 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
393                         const char *buf, size_t count)
394 {
395         struct scsi_disk *sdkp = to_scsi_disk(dev);
396         struct scsi_device *sdp = sdkp->device;
397         int mode;
398
399         if (!capable(CAP_SYS_ADMIN))
400                 return -EACCES;
401
402         if (sd_is_zoned(sdkp)) {
403                 sd_config_discard(sdkp, SD_LBP_DISABLE);
404                 return count;
405         }
406
407         if (sdp->type != TYPE_DISK)
408                 return -EINVAL;
409
410         mode = sysfs_match_string(lbp_mode, buf);
411         if (mode < 0)
412                 return -EINVAL;
413
414         sd_config_discard(sdkp, mode);
415
416         return count;
417 }
418 static DEVICE_ATTR_RW(provisioning_mode);
419
420 /* sysfs_match_string() requires dense arrays */
421 static const char *zeroing_mode[] = {
422         [SD_ZERO_WRITE]         = "write",
423         [SD_ZERO_WS]            = "writesame",
424         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
425         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
426 };
427
428 static ssize_t
429 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
430                   char *buf)
431 {
432         struct scsi_disk *sdkp = to_scsi_disk(dev);
433
434         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
435 }
436
437 static ssize_t
438 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
439                    const char *buf, size_t count)
440 {
441         struct scsi_disk *sdkp = to_scsi_disk(dev);
442         int mode;
443
444         if (!capable(CAP_SYS_ADMIN))
445                 return -EACCES;
446
447         mode = sysfs_match_string(zeroing_mode, buf);
448         if (mode < 0)
449                 return -EINVAL;
450
451         sdkp->zeroing_mode = mode;
452
453         return count;
454 }
455 static DEVICE_ATTR_RW(zeroing_mode);
456
457 static ssize_t
458 max_medium_access_timeouts_show(struct device *dev,
459                                 struct device_attribute *attr, char *buf)
460 {
461         struct scsi_disk *sdkp = to_scsi_disk(dev);
462
463         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
464 }
465
466 static ssize_t
467 max_medium_access_timeouts_store(struct device *dev,
468                                  struct device_attribute *attr, const char *buf,
469                                  size_t count)
470 {
471         struct scsi_disk *sdkp = to_scsi_disk(dev);
472         int err;
473
474         if (!capable(CAP_SYS_ADMIN))
475                 return -EACCES;
476
477         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
478
479         return err ? err : count;
480 }
481 static DEVICE_ATTR_RW(max_medium_access_timeouts);
482
483 static ssize_t
484 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
485                            char *buf)
486 {
487         struct scsi_disk *sdkp = to_scsi_disk(dev);
488
489         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
490 }
491
492 static ssize_t
493 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
494                             const char *buf, size_t count)
495 {
496         struct scsi_disk *sdkp = to_scsi_disk(dev);
497         struct scsi_device *sdp = sdkp->device;
498         unsigned long max;
499         int err;
500
501         if (!capable(CAP_SYS_ADMIN))
502                 return -EACCES;
503
504         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
505                 return -EINVAL;
506
507         err = kstrtoul(buf, 10, &max);
508
509         if (err)
510                 return err;
511
512         if (max == 0)
513                 sdp->no_write_same = 1;
514         else if (max <= SD_MAX_WS16_BLOCKS) {
515                 sdp->no_write_same = 0;
516                 sdkp->max_ws_blocks = max;
517         }
518
519         sd_config_write_same(sdkp);
520
521         return count;
522 }
523 static DEVICE_ATTR_RW(max_write_same_blocks);
524
525 static struct attribute *sd_disk_attrs[] = {
526         &dev_attr_cache_type.attr,
527         &dev_attr_FUA.attr,
528         &dev_attr_allow_restart.attr,
529         &dev_attr_manage_start_stop.attr,
530         &dev_attr_protection_type.attr,
531         &dev_attr_protection_mode.attr,
532         &dev_attr_app_tag_own.attr,
533         &dev_attr_thin_provisioning.attr,
534         &dev_attr_provisioning_mode.attr,
535         &dev_attr_zeroing_mode.attr,
536         &dev_attr_max_write_same_blocks.attr,
537         &dev_attr_max_medium_access_timeouts.attr,
538         NULL,
539 };
540 ATTRIBUTE_GROUPS(sd_disk);
541
542 static struct class sd_disk_class = {
543         .name           = "scsi_disk",
544         .owner          = THIS_MODULE,
545         .dev_release    = scsi_disk_release,
546         .dev_groups     = sd_disk_groups,
547 };
548
549 static const struct dev_pm_ops sd_pm_ops = {
550         .suspend                = sd_suspend_system,
551         .resume                 = sd_resume,
552         .poweroff               = sd_suspend_system,
553         .restore                = sd_resume,
554         .runtime_suspend        = sd_suspend_runtime,
555         .runtime_resume         = sd_resume,
556 };
557
558 static struct scsi_driver sd_template = {
559         .gendrv = {
560                 .name           = "sd",
561                 .owner          = THIS_MODULE,
562                 .probe          = sd_probe,
563                 .remove         = sd_remove,
564                 .shutdown       = sd_shutdown,
565                 .pm             = &sd_pm_ops,
566         },
567         .rescan                 = sd_rescan,
568         .init_command           = sd_init_command,
569         .uninit_command         = sd_uninit_command,
570         .done                   = sd_done,
571         .eh_action              = sd_eh_action,
572         .eh_reset               = sd_eh_reset,
573 };
574
575 /*
576  * Dummy kobj_map->probe function.
577  * The default ->probe function will call modprobe, which is
578  * pointless as this module is already loaded.
579  */
580 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
581 {
582         return NULL;
583 }
584
585 /*
586  * Device no to disk mapping:
587  * 
588  *       major         disc2     disc  p1
589  *   |............|.............|....|....| <- dev_t
590  *    31        20 19          8 7  4 3  0
591  * 
592  * Inside a major, we have 16k disks, however mapped non-
593  * contiguously. The first 16 disks are for major0, the next
594  * ones with major1, ... Disk 256 is for major0 again, disk 272 
595  * for major1, ... 
596  * As we stay compatible with our numbering scheme, we can reuse 
597  * the well-know SCSI majors 8, 65--71, 136--143.
598  */
599 static int sd_major(int major_idx)
600 {
601         switch (major_idx) {
602         case 0:
603                 return SCSI_DISK0_MAJOR;
604         case 1 ... 7:
605                 return SCSI_DISK1_MAJOR + major_idx - 1;
606         case 8 ... 15:
607                 return SCSI_DISK8_MAJOR + major_idx - 8;
608         default:
609                 BUG();
610                 return 0;       /* shut up gcc */
611         }
612 }
613
614 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
615 {
616         struct scsi_disk *sdkp = NULL;
617
618         mutex_lock(&sd_ref_mutex);
619
620         if (disk->private_data) {
621                 sdkp = scsi_disk(disk);
622                 if (scsi_device_get(sdkp->device) == 0)
623                         get_device(&sdkp->dev);
624                 else
625                         sdkp = NULL;
626         }
627         mutex_unlock(&sd_ref_mutex);
628         return sdkp;
629 }
630
631 static void scsi_disk_put(struct scsi_disk *sdkp)
632 {
633         struct scsi_device *sdev = sdkp->device;
634
635         mutex_lock(&sd_ref_mutex);
636         put_device(&sdkp->dev);
637         scsi_device_put(sdev);
638         mutex_unlock(&sd_ref_mutex);
639 }
640
641 #ifdef CONFIG_BLK_SED_OPAL
642 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
643                 size_t len, bool send)
644 {
645         struct scsi_device *sdev = data;
646         u8 cdb[12] = { 0, };
647         int ret;
648
649         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
650         cdb[1] = secp;
651         put_unaligned_be16(spsp, &cdb[2]);
652         put_unaligned_be32(len, &cdb[6]);
653
654         ret = scsi_execute_req(sdev, cdb,
655                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
656                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
657         return ret <= 0 ? ret : -EIO;
658 }
659 #endif /* CONFIG_BLK_SED_OPAL */
660
661 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
662                                            unsigned int dix, unsigned int dif)
663 {
664         struct bio *bio = scmd->request->bio;
665         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
666         unsigned int protect = 0;
667
668         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
669                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
670                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
671
672                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
673                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
674         }
675
676         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
677                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
678
679                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
680                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
681         }
682
683         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
684                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
685
686                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
687                         protect = 3 << 5;       /* Disable target PI checking */
688                 else
689                         protect = 1 << 5;       /* Enable target PI checking */
690         }
691
692         scsi_set_prot_op(scmd, prot_op);
693         scsi_set_prot_type(scmd, dif);
694         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
695
696         return protect;
697 }
698
699 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
700 {
701         struct request_queue *q = sdkp->disk->queue;
702         unsigned int logical_block_size = sdkp->device->sector_size;
703         unsigned int max_blocks = 0;
704
705         q->limits.discard_alignment =
706                 sdkp->unmap_alignment * logical_block_size;
707         q->limits.discard_granularity =
708                 max(sdkp->physical_block_size,
709                     sdkp->unmap_granularity * logical_block_size);
710         sdkp->provisioning_mode = mode;
711
712         switch (mode) {
713
714         case SD_LBP_FULL:
715         case SD_LBP_DISABLE:
716                 blk_queue_max_discard_sectors(q, 0);
717                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
718                 return;
719
720         case SD_LBP_UNMAP:
721                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
722                                           (u32)SD_MAX_WS16_BLOCKS);
723                 break;
724
725         case SD_LBP_WS16:
726                 if (sdkp->device->unmap_limit_for_ws)
727                         max_blocks = sdkp->max_unmap_blocks;
728                 else
729                         max_blocks = sdkp->max_ws_blocks;
730
731                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
732                 break;
733
734         case SD_LBP_WS10:
735                 if (sdkp->device->unmap_limit_for_ws)
736                         max_blocks = sdkp->max_unmap_blocks;
737                 else
738                         max_blocks = sdkp->max_ws_blocks;
739
740                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
741                 break;
742
743         case SD_LBP_ZERO:
744                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
745                                           (u32)SD_MAX_WS10_BLOCKS);
746                 break;
747         }
748
749         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
750         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
751 }
752
753 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
754 {
755         struct scsi_device *sdp = cmd->device;
756         struct request *rq = cmd->request;
757         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
758         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
759         unsigned int data_len = 24;
760         char *buf;
761
762         rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
763         if (!rq->special_vec.bv_page)
764                 return BLKPREP_DEFER;
765         rq->special_vec.bv_offset = 0;
766         rq->special_vec.bv_len = data_len;
767         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
768
769         cmd->cmd_len = 10;
770         cmd->cmnd[0] = UNMAP;
771         cmd->cmnd[8] = 24;
772
773         buf = page_address(rq->special_vec.bv_page);
774         put_unaligned_be16(6 + 16, &buf[0]);
775         put_unaligned_be16(16, &buf[2]);
776         put_unaligned_be64(sector, &buf[8]);
777         put_unaligned_be32(nr_sectors, &buf[16]);
778
779         cmd->allowed = SD_MAX_RETRIES;
780         cmd->transfersize = data_len;
781         rq->timeout = SD_TIMEOUT;
782         scsi_req(rq)->resid_len = data_len;
783
784         return scsi_init_io(cmd);
785 }
786
787 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
788 {
789         struct scsi_device *sdp = cmd->device;
790         struct request *rq = cmd->request;
791         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
792         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
793         u32 data_len = sdp->sector_size;
794
795         rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
796         if (!rq->special_vec.bv_page)
797                 return BLKPREP_DEFER;
798         rq->special_vec.bv_offset = 0;
799         rq->special_vec.bv_len = data_len;
800         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
801
802         cmd->cmd_len = 16;
803         cmd->cmnd[0] = WRITE_SAME_16;
804         if (unmap)
805                 cmd->cmnd[1] = 0x8; /* UNMAP */
806         put_unaligned_be64(sector, &cmd->cmnd[2]);
807         put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
808
809         cmd->allowed = SD_MAX_RETRIES;
810         cmd->transfersize = data_len;
811         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
812         scsi_req(rq)->resid_len = data_len;
813
814         return scsi_init_io(cmd);
815 }
816
817 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
818 {
819         struct scsi_device *sdp = cmd->device;
820         struct request *rq = cmd->request;
821         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
822         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
823         u32 data_len = sdp->sector_size;
824
825         rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
826         if (!rq->special_vec.bv_page)
827                 return BLKPREP_DEFER;
828         rq->special_vec.bv_offset = 0;
829         rq->special_vec.bv_len = data_len;
830         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
831
832         cmd->cmd_len = 10;
833         cmd->cmnd[0] = WRITE_SAME;
834         if (unmap)
835                 cmd->cmnd[1] = 0x8; /* UNMAP */
836         put_unaligned_be32(sector, &cmd->cmnd[2]);
837         put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
838
839         cmd->allowed = SD_MAX_RETRIES;
840         cmd->transfersize = data_len;
841         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
842         scsi_req(rq)->resid_len = data_len;
843
844         return scsi_init_io(cmd);
845 }
846
847 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
848 {
849         struct request *rq = cmd->request;
850         struct scsi_device *sdp = cmd->device;
851         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
852         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
853         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
854
855         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
856                 switch (sdkp->zeroing_mode) {
857                 case SD_ZERO_WS16_UNMAP:
858                         return sd_setup_write_same16_cmnd(cmd, true);
859                 case SD_ZERO_WS10_UNMAP:
860                         return sd_setup_write_same10_cmnd(cmd, true);
861                 }
862         }
863
864         if (sdp->no_write_same)
865                 return BLKPREP_INVALID;
866
867         if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
868                 return sd_setup_write_same16_cmnd(cmd, false);
869
870         return sd_setup_write_same10_cmnd(cmd, false);
871 }
872
873 static void sd_config_write_same(struct scsi_disk *sdkp)
874 {
875         struct request_queue *q = sdkp->disk->queue;
876         unsigned int logical_block_size = sdkp->device->sector_size;
877
878         if (sdkp->device->no_write_same) {
879                 sdkp->max_ws_blocks = 0;
880                 goto out;
881         }
882
883         /* Some devices can not handle block counts above 0xffff despite
884          * supporting WRITE SAME(16). Consequently we default to 64k
885          * blocks per I/O unless the device explicitly advertises a
886          * bigger limit.
887          */
888         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
889                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
890                                                    (u32)SD_MAX_WS16_BLOCKS);
891         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
892                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
893                                                    (u32)SD_MAX_WS10_BLOCKS);
894         else {
895                 sdkp->device->no_write_same = 1;
896                 sdkp->max_ws_blocks = 0;
897         }
898
899         if (sdkp->lbprz && sdkp->lbpws)
900                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
901         else if (sdkp->lbprz && sdkp->lbpws10)
902                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
903         else if (sdkp->max_ws_blocks)
904                 sdkp->zeroing_mode = SD_ZERO_WS;
905         else
906                 sdkp->zeroing_mode = SD_ZERO_WRITE;
907
908         if (sdkp->max_ws_blocks &&
909             sdkp->physical_block_size > logical_block_size) {
910                 /*
911                  * Reporting a maximum number of blocks that is not aligned
912                  * on the device physical size would cause a large write same
913                  * request to be split into physically unaligned chunks by
914                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
915                  * even if the caller of these functions took care to align the
916                  * large request. So make sure the maximum reported is aligned
917                  * to the device physical block size. This is only an optional
918                  * optimization for regular disks, but this is mandatory to
919                  * avoid failure of large write same requests directed at
920                  * sequential write required zones of host-managed ZBC disks.
921                  */
922                 sdkp->max_ws_blocks =
923                         round_down(sdkp->max_ws_blocks,
924                                    bytes_to_logical(sdkp->device,
925                                                     sdkp->physical_block_size));
926         }
927
928 out:
929         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
930                                          (logical_block_size >> 9));
931         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
932                                          (logical_block_size >> 9));
933 }
934
935 /**
936  * sd_setup_write_same_cmnd - write the same data to multiple blocks
937  * @cmd: command to prepare
938  *
939  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
940  * the preference indicated by the target device.
941  **/
942 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
943 {
944         struct request *rq = cmd->request;
945         struct scsi_device *sdp = cmd->device;
946         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
947         struct bio *bio = rq->bio;
948         sector_t sector = blk_rq_pos(rq);
949         unsigned int nr_sectors = blk_rq_sectors(rq);
950         unsigned int nr_bytes = blk_rq_bytes(rq);
951         int ret;
952
953         if (sdkp->device->no_write_same)
954                 return BLKPREP_INVALID;
955
956         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
957
958         sector >>= ilog2(sdp->sector_size) - 9;
959         nr_sectors >>= ilog2(sdp->sector_size) - 9;
960
961         rq->timeout = SD_WRITE_SAME_TIMEOUT;
962
963         if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
964                 cmd->cmd_len = 16;
965                 cmd->cmnd[0] = WRITE_SAME_16;
966                 put_unaligned_be64(sector, &cmd->cmnd[2]);
967                 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
968         } else {
969                 cmd->cmd_len = 10;
970                 cmd->cmnd[0] = WRITE_SAME;
971                 put_unaligned_be32(sector, &cmd->cmnd[2]);
972                 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
973         }
974
975         cmd->transfersize = sdp->sector_size;
976         cmd->allowed = SD_MAX_RETRIES;
977
978         /*
979          * For WRITE SAME the data transferred via the DATA OUT buffer is
980          * different from the amount of data actually written to the target.
981          *
982          * We set up __data_len to the amount of data transferred via the
983          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
984          * to transfer a single sector of data first, but then reset it to
985          * the amount of data to be written right after so that the I/O path
986          * knows how much to actually write.
987          */
988         rq->__data_len = sdp->sector_size;
989         ret = scsi_init_io(cmd);
990         rq->__data_len = nr_bytes;
991
992         return ret;
993 }
994
995 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
996 {
997         struct request *rq = cmd->request;
998
999         /* flush requests don't perform I/O, zero the S/G table */
1000         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1001
1002         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1003         cmd->cmd_len = 10;
1004         cmd->transfersize = 0;
1005         cmd->allowed = SD_MAX_RETRIES;
1006
1007         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1008         return BLKPREP_OK;
1009 }
1010
1011 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1012 {
1013         struct request *rq = SCpnt->request;
1014         struct scsi_device *sdp = SCpnt->device;
1015         struct gendisk *disk = rq->rq_disk;
1016         struct scsi_disk *sdkp = scsi_disk(disk);
1017         sector_t block = blk_rq_pos(rq);
1018         sector_t threshold;
1019         unsigned int this_count = blk_rq_sectors(rq);
1020         unsigned int dif, dix;
1021         int ret;
1022         unsigned char protect;
1023
1024         ret = scsi_init_io(SCpnt);
1025         if (ret != BLKPREP_OK)
1026                 return ret;
1027         WARN_ON_ONCE(SCpnt != rq->special);
1028
1029         /* from here on until we're complete, any goto out
1030          * is used for a killable error condition */
1031         ret = BLKPREP_KILL;
1032
1033         SCSI_LOG_HLQUEUE(1,
1034                 scmd_printk(KERN_INFO, SCpnt,
1035                         "%s: block=%llu, count=%d\n",
1036                         __func__, (unsigned long long)block, this_count));
1037
1038         if (!sdp || !scsi_device_online(sdp) ||
1039             block + blk_rq_sectors(rq) > get_capacity(disk)) {
1040                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1041                                                 "Finishing %u sectors\n",
1042                                                 blk_rq_sectors(rq)));
1043                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1044                                                 "Retry with 0x%p\n", SCpnt));
1045                 goto out;
1046         }
1047
1048         if (sdp->changed) {
1049                 /*
1050                  * quietly refuse to do anything to a changed disc until 
1051                  * the changed bit has been reset
1052                  */
1053                 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1054                 goto out;
1055         }
1056
1057         /*
1058          * Some SD card readers can't handle multi-sector accesses which touch
1059          * the last one or two hardware sectors.  Split accesses as needed.
1060          */
1061         threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1062                 (sdp->sector_size / 512);
1063
1064         if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1065                 if (block < threshold) {
1066                         /* Access up to the threshold but not beyond */
1067                         this_count = threshold - block;
1068                 } else {
1069                         /* Access only a single hardware sector */
1070                         this_count = sdp->sector_size / 512;
1071                 }
1072         }
1073
1074         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1075                                         (unsigned long long)block));
1076
1077         /*
1078          * If we have a 1K hardware sectorsize, prevent access to single
1079          * 512 byte sectors.  In theory we could handle this - in fact
1080          * the scsi cdrom driver must be able to handle this because
1081          * we typically use 1K blocksizes, and cdroms typically have
1082          * 2K hardware sectorsizes.  Of course, things are simpler
1083          * with the cdrom, since it is read-only.  For performance
1084          * reasons, the filesystems should be able to handle this
1085          * and not force the scsi disk driver to use bounce buffers
1086          * for this.
1087          */
1088         if (sdp->sector_size == 1024) {
1089                 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1090                         scmd_printk(KERN_ERR, SCpnt,
1091                                     "Bad block number requested\n");
1092                         goto out;
1093                 } else {
1094                         block = block >> 1;
1095                         this_count = this_count >> 1;
1096                 }
1097         }
1098         if (sdp->sector_size == 2048) {
1099                 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1100                         scmd_printk(KERN_ERR, SCpnt,
1101                                     "Bad block number requested\n");
1102                         goto out;
1103                 } else {
1104                         block = block >> 2;
1105                         this_count = this_count >> 2;
1106                 }
1107         }
1108         if (sdp->sector_size == 4096) {
1109                 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1110                         scmd_printk(KERN_ERR, SCpnt,
1111                                     "Bad block number requested\n");
1112                         goto out;
1113                 } else {
1114                         block = block >> 3;
1115                         this_count = this_count >> 3;
1116                 }
1117         }
1118         if (rq_data_dir(rq) == WRITE) {
1119                 SCpnt->cmnd[0] = WRITE_6;
1120
1121                 if (blk_integrity_rq(rq))
1122                         sd_dif_prepare(SCpnt);
1123
1124         } else if (rq_data_dir(rq) == READ) {
1125                 SCpnt->cmnd[0] = READ_6;
1126         } else {
1127                 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1128                 goto out;
1129         }
1130
1131         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1132                                         "%s %d/%u 512 byte blocks.\n",
1133                                         (rq_data_dir(rq) == WRITE) ?
1134                                         "writing" : "reading", this_count,
1135                                         blk_rq_sectors(rq)));
1136
1137         dix = scsi_prot_sg_count(SCpnt);
1138         dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1139
1140         if (dif || dix)
1141                 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1142         else
1143                 protect = 0;
1144
1145         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1146                 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1147
1148                 if (unlikely(SCpnt->cmnd == NULL)) {
1149                         ret = BLKPREP_DEFER;
1150                         goto out;
1151                 }
1152
1153                 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1154                 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1155                 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1156                 SCpnt->cmnd[7] = 0x18;
1157                 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1158                 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1159
1160                 /* LBA */
1161                 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1162                 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1163                 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1164                 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1165                 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1166                 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1167                 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1168                 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1169
1170                 /* Expected Indirect LBA */
1171                 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1172                 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1173                 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1174                 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1175
1176                 /* Transfer length */
1177                 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1178                 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1179                 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1180                 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1181         } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1182                 SCpnt->cmnd[0] += READ_16 - READ_6;
1183                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1184                 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1185                 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1186                 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1187                 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1188                 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1189                 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1190                 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1191                 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1192                 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1193                 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1194                 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1195                 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1196                 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1197         } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1198                    scsi_device_protection(SCpnt->device) ||
1199                    SCpnt->device->use_10_for_rw) {
1200                 SCpnt->cmnd[0] += READ_10 - READ_6;
1201                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1202                 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1203                 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1204                 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1205                 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1206                 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1207                 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1208                 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1209         } else {
1210                 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1211                         /*
1212                          * This happens only if this drive failed
1213                          * 10byte rw command with ILLEGAL_REQUEST
1214                          * during operation and thus turned off
1215                          * use_10_for_rw.
1216                          */
1217                         scmd_printk(KERN_ERR, SCpnt,
1218                                     "FUA write on READ/WRITE(6) drive\n");
1219                         goto out;
1220                 }
1221
1222                 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1223                 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1224                 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1225                 SCpnt->cmnd[4] = (unsigned char) this_count;
1226                 SCpnt->cmnd[5] = 0;
1227         }
1228         SCpnt->sdb.length = this_count * sdp->sector_size;
1229
1230         /*
1231          * We shouldn't disconnect in the middle of a sector, so with a dumb
1232          * host adapter, it's safe to assume that we can at least transfer
1233          * this many bytes between each connect / disconnect.
1234          */
1235         SCpnt->transfersize = sdp->sector_size;
1236         SCpnt->underflow = this_count << 9;
1237         SCpnt->allowed = SD_MAX_RETRIES;
1238
1239         /*
1240          * This indicates that the command is ready from our end to be
1241          * queued.
1242          */
1243         ret = BLKPREP_OK;
1244  out:
1245         return ret;
1246 }
1247
1248 static int sd_init_command(struct scsi_cmnd *cmd)
1249 {
1250         struct request *rq = cmd->request;
1251
1252         switch (req_op(rq)) {
1253         case REQ_OP_DISCARD:
1254                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1255                 case SD_LBP_UNMAP:
1256                         return sd_setup_unmap_cmnd(cmd);
1257                 case SD_LBP_WS16:
1258                         return sd_setup_write_same16_cmnd(cmd, true);
1259                 case SD_LBP_WS10:
1260                         return sd_setup_write_same10_cmnd(cmd, true);
1261                 case SD_LBP_ZERO:
1262                         return sd_setup_write_same10_cmnd(cmd, false);
1263                 default:
1264                         return BLKPREP_INVALID;
1265                 }
1266         case REQ_OP_WRITE_ZEROES:
1267                 return sd_setup_write_zeroes_cmnd(cmd);
1268         case REQ_OP_WRITE_SAME:
1269                 return sd_setup_write_same_cmnd(cmd);
1270         case REQ_OP_FLUSH:
1271                 return sd_setup_flush_cmnd(cmd);
1272         case REQ_OP_READ:
1273         case REQ_OP_WRITE:
1274                 return sd_setup_read_write_cmnd(cmd);
1275         case REQ_OP_ZONE_REPORT:
1276                 return sd_zbc_setup_report_cmnd(cmd);
1277         case REQ_OP_ZONE_RESET:
1278                 return sd_zbc_setup_reset_cmnd(cmd);
1279         default:
1280                 BUG();
1281         }
1282 }
1283
1284 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1285 {
1286         struct request *rq = SCpnt->request;
1287         u8 *cmnd;
1288
1289         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1290                 __free_page(rq->special_vec.bv_page);
1291
1292         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1293                 cmnd = SCpnt->cmnd;
1294                 SCpnt->cmnd = NULL;
1295                 SCpnt->cmd_len = 0;
1296                 mempool_free(cmnd, sd_cdb_pool);
1297         }
1298 }
1299
1300 /**
1301  *      sd_open - open a scsi disk device
1302  *      @bdev: Block device of the scsi disk to open
1303  *      @mode: FMODE_* mask
1304  *
1305  *      Returns 0 if successful. Returns a negated errno value in case 
1306  *      of error.
1307  *
1308  *      Note: This can be called from a user context (e.g. fsck(1) )
1309  *      or from within the kernel (e.g. as a result of a mount(1) ).
1310  *      In the latter case @inode and @filp carry an abridged amount
1311  *      of information as noted above.
1312  *
1313  *      Locking: called with bdev->bd_mutex held.
1314  **/
1315 static int sd_open(struct block_device *bdev, fmode_t mode)
1316 {
1317         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1318         struct scsi_device *sdev;
1319         int retval;
1320
1321         if (!sdkp)
1322                 return -ENXIO;
1323
1324         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1325
1326         sdev = sdkp->device;
1327
1328         /*
1329          * If the device is in error recovery, wait until it is done.
1330          * If the device is offline, then disallow any access to it.
1331          */
1332         retval = -ENXIO;
1333         if (!scsi_block_when_processing_errors(sdev))
1334                 goto error_out;
1335
1336         if (sdev->removable || sdkp->write_prot)
1337                 check_disk_change(bdev);
1338
1339         /*
1340          * If the drive is empty, just let the open fail.
1341          */
1342         retval = -ENOMEDIUM;
1343         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1344                 goto error_out;
1345
1346         /*
1347          * If the device has the write protect tab set, have the open fail
1348          * if the user expects to be able to write to the thing.
1349          */
1350         retval = -EROFS;
1351         if (sdkp->write_prot && (mode & FMODE_WRITE))
1352                 goto error_out;
1353
1354         /*
1355          * It is possible that the disk changing stuff resulted in
1356          * the device being taken offline.  If this is the case,
1357          * report this to the user, and don't pretend that the
1358          * open actually succeeded.
1359          */
1360         retval = -ENXIO;
1361         if (!scsi_device_online(sdev))
1362                 goto error_out;
1363
1364         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1365                 if (scsi_block_when_processing_errors(sdev))
1366                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1367         }
1368
1369         return 0;
1370
1371 error_out:
1372         scsi_disk_put(sdkp);
1373         return retval;  
1374 }
1375
1376 /**
1377  *      sd_release - invoked when the (last) close(2) is called on this
1378  *      scsi disk.
1379  *      @disk: disk to release
1380  *      @mode: FMODE_* mask
1381  *
1382  *      Returns 0. 
1383  *
1384  *      Note: may block (uninterruptible) if error recovery is underway
1385  *      on this disk.
1386  *
1387  *      Locking: called with bdev->bd_mutex held.
1388  **/
1389 static void sd_release(struct gendisk *disk, fmode_t mode)
1390 {
1391         struct scsi_disk *sdkp = scsi_disk(disk);
1392         struct scsi_device *sdev = sdkp->device;
1393
1394         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1395
1396         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1397                 if (scsi_block_when_processing_errors(sdev))
1398                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1399         }
1400
1401         /*
1402          * XXX and what if there are packets in flight and this close()
1403          * XXX is followed by a "rmmod sd_mod"?
1404          */
1405
1406         scsi_disk_put(sdkp);
1407 }
1408
1409 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1410 {
1411         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1412         struct scsi_device *sdp = sdkp->device;
1413         struct Scsi_Host *host = sdp->host;
1414         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1415         int diskinfo[4];
1416
1417         /* default to most commonly used values */
1418         diskinfo[0] = 0x40;     /* 1 << 6 */
1419         diskinfo[1] = 0x20;     /* 1 << 5 */
1420         diskinfo[2] = capacity >> 11;
1421
1422         /* override with calculated, extended default, or driver values */
1423         if (host->hostt->bios_param)
1424                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1425         else
1426                 scsicam_bios_param(bdev, capacity, diskinfo);
1427
1428         geo->heads = diskinfo[0];
1429         geo->sectors = diskinfo[1];
1430         geo->cylinders = diskinfo[2];
1431         return 0;
1432 }
1433
1434 /**
1435  *      sd_ioctl - process an ioctl
1436  *      @bdev: target block device
1437  *      @mode: FMODE_* mask
1438  *      @cmd: ioctl command number
1439  *      @arg: this is third argument given to ioctl(2) system call.
1440  *      Often contains a pointer.
1441  *
1442  *      Returns 0 if successful (some ioctls return positive numbers on
1443  *      success as well). Returns a negated errno value in case of error.
1444  *
1445  *      Note: most ioctls are forward onto the block subsystem or further
1446  *      down in the scsi subsystem.
1447  **/
1448 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1449                     unsigned int cmd, unsigned long arg)
1450 {
1451         struct gendisk *disk = bdev->bd_disk;
1452         struct scsi_disk *sdkp = scsi_disk(disk);
1453         struct scsi_device *sdp = sdkp->device;
1454         void __user *p = (void __user *)arg;
1455         int error;
1456     
1457         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1458                                     "cmd=0x%x\n", disk->disk_name, cmd));
1459
1460         error = scsi_verify_blk_ioctl(bdev, cmd);
1461         if (error < 0)
1462                 return error;
1463
1464         /*
1465          * If we are in the middle of error recovery, don't let anyone
1466          * else try and use this device.  Also, if error recovery fails, it
1467          * may try and take the device offline, in which case all further
1468          * access to the device is prohibited.
1469          */
1470         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1471                         (mode & FMODE_NDELAY) != 0);
1472         if (error)
1473                 goto out;
1474
1475         if (is_sed_ioctl(cmd))
1476                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1477
1478         /*
1479          * Send SCSI addressing ioctls directly to mid level, send other
1480          * ioctls to block level and then onto mid level if they can't be
1481          * resolved.
1482          */
1483         switch (cmd) {
1484                 case SCSI_IOCTL_GET_IDLUN:
1485                 case SCSI_IOCTL_GET_BUS_NUMBER:
1486                         error = scsi_ioctl(sdp, cmd, p);
1487                         break;
1488                 default:
1489                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1490                         if (error != -ENOTTY)
1491                                 break;
1492                         error = scsi_ioctl(sdp, cmd, p);
1493                         break;
1494         }
1495 out:
1496         return error;
1497 }
1498
1499 static void set_media_not_present(struct scsi_disk *sdkp)
1500 {
1501         if (sdkp->media_present)
1502                 sdkp->device->changed = 1;
1503
1504         if (sdkp->device->removable) {
1505                 sdkp->media_present = 0;
1506                 sdkp->capacity = 0;
1507         }
1508 }
1509
1510 static int media_not_present(struct scsi_disk *sdkp,
1511                              struct scsi_sense_hdr *sshdr)
1512 {
1513         if (!scsi_sense_valid(sshdr))
1514                 return 0;
1515
1516         /* not invoked for commands that could return deferred errors */
1517         switch (sshdr->sense_key) {
1518         case UNIT_ATTENTION:
1519         case NOT_READY:
1520                 /* medium not present */
1521                 if (sshdr->asc == 0x3A) {
1522                         set_media_not_present(sdkp);
1523                         return 1;
1524                 }
1525         }
1526         return 0;
1527 }
1528
1529 /**
1530  *      sd_check_events - check media events
1531  *      @disk: kernel device descriptor
1532  *      @clearing: disk events currently being cleared
1533  *
1534  *      Returns mask of DISK_EVENT_*.
1535  *
1536  *      Note: this function is invoked from the block subsystem.
1537  **/
1538 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1539 {
1540         struct scsi_disk *sdkp = scsi_disk_get(disk);
1541         struct scsi_device *sdp;
1542         int retval;
1543
1544         if (!sdkp)
1545                 return 0;
1546
1547         sdp = sdkp->device;
1548         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1549
1550         /*
1551          * If the device is offline, don't send any commands - just pretend as
1552          * if the command failed.  If the device ever comes back online, we
1553          * can deal with it then.  It is only because of unrecoverable errors
1554          * that we would ever take a device offline in the first place.
1555          */
1556         if (!scsi_device_online(sdp)) {
1557                 set_media_not_present(sdkp);
1558                 goto out;
1559         }
1560
1561         /*
1562          * Using TEST_UNIT_READY enables differentiation between drive with
1563          * no cartridge loaded - NOT READY, drive with changed cartridge -
1564          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1565          *
1566          * Drives that auto spin down. eg iomega jaz 1G, will be started
1567          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1568          * sd_revalidate() is called.
1569          */
1570         if (scsi_block_when_processing_errors(sdp)) {
1571                 struct scsi_sense_hdr sshdr = { 0, };
1572
1573                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1574                                               &sshdr);
1575
1576                 /* failed to execute TUR, assume media not present */
1577                 if (host_byte(retval)) {
1578                         set_media_not_present(sdkp);
1579                         goto out;
1580                 }
1581
1582                 if (media_not_present(sdkp, &sshdr))
1583                         goto out;
1584         }
1585
1586         /*
1587          * For removable scsi disk we have to recognise the presence
1588          * of a disk in the drive.
1589          */
1590         if (!sdkp->media_present)
1591                 sdp->changed = 1;
1592         sdkp->media_present = 1;
1593 out:
1594         /*
1595          * sdp->changed is set under the following conditions:
1596          *
1597          *      Medium present state has changed in either direction.
1598          *      Device has indicated UNIT_ATTENTION.
1599          */
1600         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1601         sdp->changed = 0;
1602         scsi_disk_put(sdkp);
1603         return retval;
1604 }
1605
1606 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1607 {
1608         int retries, res;
1609         struct scsi_device *sdp = sdkp->device;
1610         const int timeout = sdp->request_queue->rq_timeout
1611                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1612         struct scsi_sense_hdr my_sshdr;
1613
1614         if (!scsi_device_online(sdp))
1615                 return -ENODEV;
1616
1617         /* caller might not be interested in sense, but we need it */
1618         if (!sshdr)
1619                 sshdr = &my_sshdr;
1620
1621         for (retries = 3; retries > 0; --retries) {
1622                 unsigned char cmd[10] = { 0 };
1623
1624                 cmd[0] = SYNCHRONIZE_CACHE;
1625                 /*
1626                  * Leave the rest of the command zero to indicate
1627                  * flush everything.
1628                  */
1629                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1630                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1631                 if (res == 0)
1632                         break;
1633         }
1634
1635         if (res) {
1636                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1637
1638                 if (driver_byte(res) & DRIVER_SENSE)
1639                         sd_print_sense_hdr(sdkp, sshdr);
1640
1641                 /* we need to evaluate the error return  */
1642                 if (scsi_sense_valid(sshdr) &&
1643                         (sshdr->asc == 0x3a ||  /* medium not present */
1644                          sshdr->asc == 0x20))   /* invalid command */
1645                                 /* this is no error here */
1646                                 return 0;
1647
1648                 switch (host_byte(res)) {
1649                 /* ignore errors due to racing a disconnection */
1650                 case DID_BAD_TARGET:
1651                 case DID_NO_CONNECT:
1652                         return 0;
1653                 /* signal the upper layer it might try again */
1654                 case DID_BUS_BUSY:
1655                 case DID_IMM_RETRY:
1656                 case DID_REQUEUE:
1657                 case DID_SOFT_ERROR:
1658                         return -EBUSY;
1659                 default:
1660                         return -EIO;
1661                 }
1662         }
1663         return 0;
1664 }
1665
1666 static void sd_rescan(struct device *dev)
1667 {
1668         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1669
1670         revalidate_disk(sdkp->disk);
1671 }
1672
1673
1674 #ifdef CONFIG_COMPAT
1675 /* 
1676  * This gets directly called from VFS. When the ioctl 
1677  * is not recognized we go back to the other translation paths. 
1678  */
1679 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1680                            unsigned int cmd, unsigned long arg)
1681 {
1682         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1683         int error;
1684
1685         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1686                         (mode & FMODE_NDELAY) != 0);
1687         if (error)
1688                 return error;
1689                
1690         /* 
1691          * Let the static ioctl translation table take care of it.
1692          */
1693         if (!sdev->host->hostt->compat_ioctl)
1694                 return -ENOIOCTLCMD; 
1695         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1696 }
1697 #endif
1698
1699 static char sd_pr_type(enum pr_type type)
1700 {
1701         switch (type) {
1702         case PR_WRITE_EXCLUSIVE:
1703                 return 0x01;
1704         case PR_EXCLUSIVE_ACCESS:
1705                 return 0x03;
1706         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1707                 return 0x05;
1708         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1709                 return 0x06;
1710         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1711                 return 0x07;
1712         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1713                 return 0x08;
1714         default:
1715                 return 0;
1716         }
1717 };
1718
1719 static int sd_pr_command(struct block_device *bdev, u8 sa,
1720                 u64 key, u64 sa_key, u8 type, u8 flags)
1721 {
1722         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1723         struct scsi_sense_hdr sshdr;
1724         int result;
1725         u8 cmd[16] = { 0, };
1726         u8 data[24] = { 0, };
1727
1728         cmd[0] = PERSISTENT_RESERVE_OUT;
1729         cmd[1] = sa;
1730         cmd[2] = type;
1731         put_unaligned_be32(sizeof(data), &cmd[5]);
1732
1733         put_unaligned_be64(key, &data[0]);
1734         put_unaligned_be64(sa_key, &data[8]);
1735         data[20] = flags;
1736
1737         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1738                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1739
1740         if ((driver_byte(result) & DRIVER_SENSE) &&
1741             (scsi_sense_valid(&sshdr))) {
1742                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1743                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1744         }
1745
1746         return result;
1747 }
1748
1749 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1750                 u32 flags)
1751 {
1752         if (flags & ~PR_FL_IGNORE_KEY)
1753                 return -EOPNOTSUPP;
1754         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1755                         old_key, new_key, 0,
1756                         (1 << 0) /* APTPL */);
1757 }
1758
1759 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1760                 u32 flags)
1761 {
1762         if (flags)
1763                 return -EOPNOTSUPP;
1764         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1765 }
1766
1767 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1768 {
1769         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1770 }
1771
1772 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1773                 enum pr_type type, bool abort)
1774 {
1775         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1776                              sd_pr_type(type), 0);
1777 }
1778
1779 static int sd_pr_clear(struct block_device *bdev, u64 key)
1780 {
1781         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1782 }
1783
1784 static const struct pr_ops sd_pr_ops = {
1785         .pr_register    = sd_pr_register,
1786         .pr_reserve     = sd_pr_reserve,
1787         .pr_release     = sd_pr_release,
1788         .pr_preempt     = sd_pr_preempt,
1789         .pr_clear       = sd_pr_clear,
1790 };
1791
1792 static const struct block_device_operations sd_fops = {
1793         .owner                  = THIS_MODULE,
1794         .open                   = sd_open,
1795         .release                = sd_release,
1796         .ioctl                  = sd_ioctl,
1797         .getgeo                 = sd_getgeo,
1798 #ifdef CONFIG_COMPAT
1799         .compat_ioctl           = sd_compat_ioctl,
1800 #endif
1801         .check_events           = sd_check_events,
1802         .revalidate_disk        = sd_revalidate_disk,
1803         .unlock_native_capacity = sd_unlock_native_capacity,
1804         .pr_ops                 = &sd_pr_ops,
1805 };
1806
1807 /**
1808  *      sd_eh_reset - reset error handling callback
1809  *      @scmd:          sd-issued command that has failed
1810  *
1811  *      This function is called by the SCSI midlayer before starting
1812  *      SCSI EH. When counting medium access failures we have to be
1813  *      careful to register it only only once per device and SCSI EH run;
1814  *      there might be several timed out commands which will cause the
1815  *      'max_medium_access_timeouts' counter to trigger after the first
1816  *      SCSI EH run already and set the device to offline.
1817  *      So this function resets the internal counter before starting SCSI EH.
1818  **/
1819 static void sd_eh_reset(struct scsi_cmnd *scmd)
1820 {
1821         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1822
1823         /* New SCSI EH run, reset gate variable */
1824         sdkp->ignore_medium_access_errors = false;
1825 }
1826
1827 /**
1828  *      sd_eh_action - error handling callback
1829  *      @scmd:          sd-issued command that has failed
1830  *      @eh_disp:       The recovery disposition suggested by the midlayer
1831  *
1832  *      This function is called by the SCSI midlayer upon completion of an
1833  *      error test command (currently TEST UNIT READY). The result of sending
1834  *      the eh command is passed in eh_disp.  We're looking for devices that
1835  *      fail medium access commands but are OK with non access commands like
1836  *      test unit ready (so wrongly see the device as having a successful
1837  *      recovery)
1838  **/
1839 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1840 {
1841         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1842         struct scsi_device *sdev = scmd->device;
1843
1844         if (!scsi_device_online(sdev) ||
1845             !scsi_medium_access_command(scmd) ||
1846             host_byte(scmd->result) != DID_TIME_OUT ||
1847             eh_disp != SUCCESS)
1848                 return eh_disp;
1849
1850         /*
1851          * The device has timed out executing a medium access command.
1852          * However, the TEST UNIT READY command sent during error
1853          * handling completed successfully. Either the device is in the
1854          * process of recovering or has it suffered an internal failure
1855          * that prevents access to the storage medium.
1856          */
1857         if (!sdkp->ignore_medium_access_errors) {
1858                 sdkp->medium_access_timed_out++;
1859                 sdkp->ignore_medium_access_errors = true;
1860         }
1861
1862         /*
1863          * If the device keeps failing read/write commands but TEST UNIT
1864          * READY always completes successfully we assume that medium
1865          * access is no longer possible and take the device offline.
1866          */
1867         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1868                 scmd_printk(KERN_ERR, scmd,
1869                             "Medium access timeout failure. Offlining disk!\n");
1870                 mutex_lock(&sdev->state_mutex);
1871                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1872                 mutex_unlock(&sdev->state_mutex);
1873
1874                 return SUCCESS;
1875         }
1876
1877         return eh_disp;
1878 }
1879
1880 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1881 {
1882         struct request *req = scmd->request;
1883         struct scsi_device *sdev = scmd->device;
1884         unsigned int transferred, good_bytes;
1885         u64 start_lba, end_lba, bad_lba;
1886
1887         /*
1888          * Some commands have a payload smaller than the device logical
1889          * block size (e.g. INQUIRY on a 4K disk).
1890          */
1891         if (scsi_bufflen(scmd) <= sdev->sector_size)
1892                 return 0;
1893
1894         /* Check if we have a 'bad_lba' information */
1895         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1896                                      SCSI_SENSE_BUFFERSIZE,
1897                                      &bad_lba))
1898                 return 0;
1899
1900         /*
1901          * If the bad lba was reported incorrectly, we have no idea where
1902          * the error is.
1903          */
1904         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1905         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1906         if (bad_lba < start_lba || bad_lba >= end_lba)
1907                 return 0;
1908
1909         /*
1910          * resid is optional but mostly filled in.  When it's unused,
1911          * its value is zero, so we assume the whole buffer transferred
1912          */
1913         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1914
1915         /* This computation should always be done in terms of the
1916          * resolution of the device's medium.
1917          */
1918         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1919
1920         return min(good_bytes, transferred);
1921 }
1922
1923 /**
1924  *      sd_done - bottom half handler: called when the lower level
1925  *      driver has completed (successfully or otherwise) a scsi command.
1926  *      @SCpnt: mid-level's per command structure.
1927  *
1928  *      Note: potentially run from within an ISR. Must not block.
1929  **/
1930 static int sd_done(struct scsi_cmnd *SCpnt)
1931 {
1932         int result = SCpnt->result;
1933         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1934         unsigned int sector_size = SCpnt->device->sector_size;
1935         unsigned int resid;
1936         struct scsi_sense_hdr sshdr;
1937         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1938         struct request *req = SCpnt->request;
1939         int sense_valid = 0;
1940         int sense_deferred = 0;
1941
1942         switch (req_op(req)) {
1943         case REQ_OP_DISCARD:
1944         case REQ_OP_WRITE_ZEROES:
1945         case REQ_OP_WRITE_SAME:
1946         case REQ_OP_ZONE_RESET:
1947                 if (!result) {
1948                         good_bytes = blk_rq_bytes(req);
1949                         scsi_set_resid(SCpnt, 0);
1950                 } else {
1951                         good_bytes = 0;
1952                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1953                 }
1954                 break;
1955         case REQ_OP_ZONE_REPORT:
1956                 if (!result) {
1957                         good_bytes = scsi_bufflen(SCpnt)
1958                                 - scsi_get_resid(SCpnt);
1959                         scsi_set_resid(SCpnt, 0);
1960                 } else {
1961                         good_bytes = 0;
1962                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1963                 }
1964                 break;
1965         default:
1966                 /*
1967                  * In case of bogus fw or device, we could end up having
1968                  * an unaligned partial completion. Check this here and force
1969                  * alignment.
1970                  */
1971                 resid = scsi_get_resid(SCpnt);
1972                 if (resid & (sector_size - 1)) {
1973                         sd_printk(KERN_INFO, sdkp,
1974                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1975                                 resid, sector_size);
1976                         resid = min(scsi_bufflen(SCpnt),
1977                                     round_up(resid, sector_size));
1978                         scsi_set_resid(SCpnt, resid);
1979                 }
1980         }
1981
1982         if (result) {
1983                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1984                 if (sense_valid)
1985                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1986         }
1987         sdkp->medium_access_timed_out = 0;
1988
1989         if (driver_byte(result) != DRIVER_SENSE &&
1990             (!sense_valid || sense_deferred))
1991                 goto out;
1992
1993         switch (sshdr.sense_key) {
1994         case HARDWARE_ERROR:
1995         case MEDIUM_ERROR:
1996                 good_bytes = sd_completed_bytes(SCpnt);
1997                 break;
1998         case RECOVERED_ERROR:
1999                 good_bytes = scsi_bufflen(SCpnt);
2000                 break;
2001         case NO_SENSE:
2002                 /* This indicates a false check condition, so ignore it.  An
2003                  * unknown amount of data was transferred so treat it as an
2004                  * error.
2005                  */
2006                 SCpnt->result = 0;
2007                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2008                 break;
2009         case ABORTED_COMMAND:
2010                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2011                         good_bytes = sd_completed_bytes(SCpnt);
2012                 break;
2013         case ILLEGAL_REQUEST:
2014                 switch (sshdr.asc) {
2015                 case 0x10:      /* DIX: Host detected corruption */
2016                         good_bytes = sd_completed_bytes(SCpnt);
2017                         break;
2018                 case 0x20:      /* INVALID COMMAND OPCODE */
2019                 case 0x24:      /* INVALID FIELD IN CDB */
2020                         switch (SCpnt->cmnd[0]) {
2021                         case UNMAP:
2022                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2023                                 break;
2024                         case WRITE_SAME_16:
2025                         case WRITE_SAME:
2026                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2027                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2028                                 } else {
2029                                         sdkp->device->no_write_same = 1;
2030                                         sd_config_write_same(sdkp);
2031                                         req->__data_len = blk_rq_bytes(req);
2032                                         req->rq_flags |= RQF_QUIET;
2033                                 }
2034                                 break;
2035                         }
2036                 }
2037                 break;
2038         default:
2039                 break;
2040         }
2041
2042  out:
2043         if (sd_is_zoned(sdkp))
2044                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2045
2046         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2047                                            "sd_done: completed %d of %d bytes\n",
2048                                            good_bytes, scsi_bufflen(SCpnt)));
2049
2050         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2051                 sd_dif_complete(SCpnt, good_bytes);
2052
2053         return good_bytes;
2054 }
2055
2056 /*
2057  * spinup disk - called only in sd_revalidate_disk()
2058  */
2059 static void
2060 sd_spinup_disk(struct scsi_disk *sdkp)
2061 {
2062         unsigned char cmd[10];
2063         unsigned long spintime_expire = 0;
2064         int retries, spintime;
2065         unsigned int the_result;
2066         struct scsi_sense_hdr sshdr;
2067         int sense_valid = 0;
2068
2069         spintime = 0;
2070
2071         /* Spin up drives, as required.  Only do this at boot time */
2072         /* Spinup needs to be done for module loads too. */
2073         do {
2074                 retries = 0;
2075
2076                 do {
2077                         cmd[0] = TEST_UNIT_READY;
2078                         memset((void *) &cmd[1], 0, 9);
2079
2080                         the_result = scsi_execute_req(sdkp->device, cmd,
2081                                                       DMA_NONE, NULL, 0,
2082                                                       &sshdr, SD_TIMEOUT,
2083                                                       SD_MAX_RETRIES, NULL);
2084
2085                         /*
2086                          * If the drive has indicated to us that it
2087                          * doesn't have any media in it, don't bother
2088                          * with any more polling.
2089                          */
2090                         if (media_not_present(sdkp, &sshdr))
2091                                 return;
2092
2093                         if (the_result)
2094                                 sense_valid = scsi_sense_valid(&sshdr);
2095                         retries++;
2096                 } while (retries < 3 && 
2097                          (!scsi_status_is_good(the_result) ||
2098                           ((driver_byte(the_result) & DRIVER_SENSE) &&
2099                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2100
2101                 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2102                         /* no sense, TUR either succeeded or failed
2103                          * with a status error */
2104                         if(!spintime && !scsi_status_is_good(the_result)) {
2105                                 sd_print_result(sdkp, "Test Unit Ready failed",
2106                                                 the_result);
2107                         }
2108                         break;
2109                 }
2110
2111                 /*
2112                  * The device does not want the automatic start to be issued.
2113                  */
2114                 if (sdkp->device->no_start_on_add)
2115                         break;
2116
2117                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2118                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2119                                 break;  /* manual intervention required */
2120                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2121                                 break;  /* standby */
2122                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2123                                 break;  /* unavailable */
2124                         /*
2125                          * Issue command to spin up drive when not ready
2126                          */
2127                         if (!spintime) {
2128                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2129                                 cmd[0] = START_STOP;
2130                                 cmd[1] = 1;     /* Return immediately */
2131                                 memset((void *) &cmd[2], 0, 8);
2132                                 cmd[4] = 1;     /* Start spin cycle */
2133                                 if (sdkp->device->start_stop_pwr_cond)
2134                                         cmd[4] |= 1 << 4;
2135                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2136                                                  NULL, 0, &sshdr,
2137                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2138                                                  NULL);
2139                                 spintime_expire = jiffies + 100 * HZ;
2140                                 spintime = 1;
2141                         }
2142                         /* Wait 1 second for next try */
2143                         msleep(1000);
2144                         printk(KERN_CONT ".");
2145
2146                 /*
2147                  * Wait for USB flash devices with slow firmware.
2148                  * Yes, this sense key/ASC combination shouldn't
2149                  * occur here.  It's characteristic of these devices.
2150                  */
2151                 } else if (sense_valid &&
2152                                 sshdr.sense_key == UNIT_ATTENTION &&
2153                                 sshdr.asc == 0x28) {
2154                         if (!spintime) {
2155                                 spintime_expire = jiffies + 5 * HZ;
2156                                 spintime = 1;
2157                         }
2158                         /* Wait 1 second for next try */
2159                         msleep(1000);
2160                 } else {
2161                         /* we don't understand the sense code, so it's
2162                          * probably pointless to loop */
2163                         if(!spintime) {
2164                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2165                                 sd_print_sense_hdr(sdkp, &sshdr);
2166                         }
2167                         break;
2168                 }
2169                                 
2170         } while (spintime && time_before_eq(jiffies, spintime_expire));
2171
2172         if (spintime) {
2173                 if (scsi_status_is_good(the_result))
2174                         printk(KERN_CONT "ready\n");
2175                 else
2176                         printk(KERN_CONT "not responding...\n");
2177         }
2178 }
2179
2180 /*
2181  * Determine whether disk supports Data Integrity Field.
2182  */
2183 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2184 {
2185         struct scsi_device *sdp = sdkp->device;
2186         u8 type;
2187         int ret = 0;
2188
2189         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2190                 return ret;
2191
2192         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2193
2194         if (type > T10_PI_TYPE3_PROTECTION)
2195                 ret = -ENODEV;
2196         else if (scsi_host_dif_capable(sdp->host, type))
2197                 ret = 1;
2198
2199         if (sdkp->first_scan || type != sdkp->protection_type)
2200                 switch (ret) {
2201                 case -ENODEV:
2202                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2203                                   " protection type %u. Disabling disk!\n",
2204                                   type);
2205                         break;
2206                 case 1:
2207                         sd_printk(KERN_NOTICE, sdkp,
2208                                   "Enabling DIF Type %u protection\n", type);
2209                         break;
2210                 case 0:
2211                         sd_printk(KERN_NOTICE, sdkp,
2212                                   "Disabling DIF Type %u protection\n", type);
2213                         break;
2214                 }
2215
2216         sdkp->protection_type = type;
2217
2218         return ret;
2219 }
2220
2221 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2222                         struct scsi_sense_hdr *sshdr, int sense_valid,
2223                         int the_result)
2224 {
2225         if (driver_byte(the_result) & DRIVER_SENSE)
2226                 sd_print_sense_hdr(sdkp, sshdr);
2227         else
2228                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2229
2230         /*
2231          * Set dirty bit for removable devices if not ready -
2232          * sometimes drives will not report this properly.
2233          */
2234         if (sdp->removable &&
2235             sense_valid && sshdr->sense_key == NOT_READY)
2236                 set_media_not_present(sdkp);
2237
2238         /*
2239          * We used to set media_present to 0 here to indicate no media
2240          * in the drive, but some drives fail read capacity even with
2241          * media present, so we can't do that.
2242          */
2243         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2244 }
2245
2246 #define RC16_LEN 32
2247 #if RC16_LEN > SD_BUF_SIZE
2248 #error RC16_LEN must not be more than SD_BUF_SIZE
2249 #endif
2250
2251 #define READ_CAPACITY_RETRIES_ON_RESET  10
2252
2253 /*
2254  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2255  * and the reported logical block size is bigger than 512 bytes. Note
2256  * that last_sector is a u64 and therefore logical_to_sectors() is not
2257  * applicable.
2258  */
2259 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2260 {
2261         u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2262
2263         if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2264                 return false;
2265
2266         return true;
2267 }
2268
2269 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2270                                                 unsigned char *buffer)
2271 {
2272         unsigned char cmd[16];
2273         struct scsi_sense_hdr sshdr;
2274         int sense_valid = 0;
2275         int the_result;
2276         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2277         unsigned int alignment;
2278         unsigned long long lba;
2279         unsigned sector_size;
2280
2281         if (sdp->no_read_capacity_16)
2282                 return -EINVAL;
2283
2284         do {
2285                 memset(cmd, 0, 16);
2286                 cmd[0] = SERVICE_ACTION_IN_16;
2287                 cmd[1] = SAI_READ_CAPACITY_16;
2288                 cmd[13] = RC16_LEN;
2289                 memset(buffer, 0, RC16_LEN);
2290
2291                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2292                                         buffer, RC16_LEN, &sshdr,
2293                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2294
2295                 if (media_not_present(sdkp, &sshdr))
2296                         return -ENODEV;
2297
2298                 if (the_result) {
2299                         sense_valid = scsi_sense_valid(&sshdr);
2300                         if (sense_valid &&
2301                             sshdr.sense_key == ILLEGAL_REQUEST &&
2302                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2303                             sshdr.ascq == 0x00)
2304                                 /* Invalid Command Operation Code or
2305                                  * Invalid Field in CDB, just retry
2306                                  * silently with RC10 */
2307                                 return -EINVAL;
2308                         if (sense_valid &&
2309                             sshdr.sense_key == UNIT_ATTENTION &&
2310                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2311                                 /* Device reset might occur several times,
2312                                  * give it one more chance */
2313                                 if (--reset_retries > 0)
2314                                         continue;
2315                 }
2316                 retries--;
2317
2318         } while (the_result && retries);
2319
2320         if (the_result) {
2321                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2322                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2323                 return -EINVAL;
2324         }
2325
2326         sector_size = get_unaligned_be32(&buffer[8]);
2327         lba = get_unaligned_be64(&buffer[0]);
2328
2329         if (sd_read_protection_type(sdkp, buffer) < 0) {
2330                 sdkp->capacity = 0;
2331                 return -ENODEV;
2332         }
2333
2334         if (!sd_addressable_capacity(lba, sector_size)) {
2335                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2336                         "kernel compiled with support for large block "
2337                         "devices.\n");
2338                 sdkp->capacity = 0;
2339                 return -EOVERFLOW;
2340         }
2341
2342         /* Logical blocks per physical block exponent */
2343         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2344
2345         /* RC basis */
2346         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2347
2348         /* Lowest aligned logical block */
2349         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2350         blk_queue_alignment_offset(sdp->request_queue, alignment);
2351         if (alignment && sdkp->first_scan)
2352                 sd_printk(KERN_NOTICE, sdkp,
2353                           "physical block alignment offset: %u\n", alignment);
2354
2355         if (buffer[14] & 0x80) { /* LBPME */
2356                 sdkp->lbpme = 1;
2357
2358                 if (buffer[14] & 0x40) /* LBPRZ */
2359                         sdkp->lbprz = 1;
2360
2361                 sd_config_discard(sdkp, SD_LBP_WS16);
2362         }
2363
2364         sdkp->capacity = lba + 1;
2365         return sector_size;
2366 }
2367
2368 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2369                                                 unsigned char *buffer)
2370 {
2371         unsigned char cmd[16];
2372         struct scsi_sense_hdr sshdr;
2373         int sense_valid = 0;
2374         int the_result;
2375         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2376         sector_t lba;
2377         unsigned sector_size;
2378
2379         do {
2380                 cmd[0] = READ_CAPACITY;
2381                 memset(&cmd[1], 0, 9);
2382                 memset(buffer, 0, 8);
2383
2384                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2385                                         buffer, 8, &sshdr,
2386                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2387
2388                 if (media_not_present(sdkp, &sshdr))
2389                         return -ENODEV;
2390
2391                 if (the_result) {
2392                         sense_valid = scsi_sense_valid(&sshdr);
2393                         if (sense_valid &&
2394                             sshdr.sense_key == UNIT_ATTENTION &&
2395                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2396                                 /* Device reset might occur several times,
2397                                  * give it one more chance */
2398                                 if (--reset_retries > 0)
2399                                         continue;
2400                 }
2401                 retries--;
2402
2403         } while (the_result && retries);
2404
2405         if (the_result) {
2406                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2407                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2408                 return -EINVAL;
2409         }
2410
2411         sector_size = get_unaligned_be32(&buffer[4]);
2412         lba = get_unaligned_be32(&buffer[0]);
2413
2414         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2415                 /* Some buggy (usb cardreader) devices return an lba of
2416                    0xffffffff when the want to report a size of 0 (with
2417                    which they really mean no media is present) */
2418                 sdkp->capacity = 0;
2419                 sdkp->physical_block_size = sector_size;
2420                 return sector_size;
2421         }
2422
2423         if (!sd_addressable_capacity(lba, sector_size)) {
2424                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2425                         "kernel compiled with support for large block "
2426                         "devices.\n");
2427                 sdkp->capacity = 0;
2428                 return -EOVERFLOW;
2429         }
2430
2431         sdkp->capacity = lba + 1;
2432         sdkp->physical_block_size = sector_size;
2433         return sector_size;
2434 }
2435
2436 static int sd_try_rc16_first(struct scsi_device *sdp)
2437 {
2438         if (sdp->host->max_cmd_len < 16)
2439                 return 0;
2440         if (sdp->try_rc_10_first)
2441                 return 0;
2442         if (sdp->scsi_level > SCSI_SPC_2)
2443                 return 1;
2444         if (scsi_device_protection(sdp))
2445                 return 1;
2446         return 0;
2447 }
2448
2449 /*
2450  * read disk capacity
2451  */
2452 static void
2453 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2454 {
2455         int sector_size;
2456         struct scsi_device *sdp = sdkp->device;
2457
2458         if (sd_try_rc16_first(sdp)) {
2459                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2460                 if (sector_size == -EOVERFLOW)
2461                         goto got_data;
2462                 if (sector_size == -ENODEV)
2463                         return;
2464                 if (sector_size < 0)
2465                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2466                 if (sector_size < 0)
2467                         return;
2468         } else {
2469                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2470                 if (sector_size == -EOVERFLOW)
2471                         goto got_data;
2472                 if (sector_size < 0)
2473                         return;
2474                 if ((sizeof(sdkp->capacity) > 4) &&
2475                     (sdkp->capacity > 0xffffffffULL)) {
2476                         int old_sector_size = sector_size;
2477                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2478                                         "Trying to use READ CAPACITY(16).\n");
2479                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2480                         if (sector_size < 0) {
2481                                 sd_printk(KERN_NOTICE, sdkp,
2482                                         "Using 0xffffffff as device size\n");
2483                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2484                                 sector_size = old_sector_size;
2485                                 goto got_data;
2486                         }
2487                 }
2488         }
2489
2490         /* Some devices are known to return the total number of blocks,
2491          * not the highest block number.  Some devices have versions
2492          * which do this and others which do not.  Some devices we might
2493          * suspect of doing this but we don't know for certain.
2494          *
2495          * If we know the reported capacity is wrong, decrement it.  If
2496          * we can only guess, then assume the number of blocks is even
2497          * (usually true but not always) and err on the side of lowering
2498          * the capacity.
2499          */
2500         if (sdp->fix_capacity ||
2501             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2502                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2503                                 "from its reported value: %llu\n",
2504                                 (unsigned long long) sdkp->capacity);
2505                 --sdkp->capacity;
2506         }
2507
2508 got_data:
2509         if (sector_size == 0) {
2510                 sector_size = 512;
2511                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2512                           "assuming 512.\n");
2513         }
2514
2515         if (sector_size != 512 &&
2516             sector_size != 1024 &&
2517             sector_size != 2048 &&
2518             sector_size != 4096) {
2519                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2520                           sector_size);
2521                 /*
2522                  * The user might want to re-format the drive with
2523                  * a supported sectorsize.  Once this happens, it
2524                  * would be relatively trivial to set the thing up.
2525                  * For this reason, we leave the thing in the table.
2526                  */
2527                 sdkp->capacity = 0;
2528                 /*
2529                  * set a bogus sector size so the normal read/write
2530                  * logic in the block layer will eventually refuse any
2531                  * request on this device without tripping over power
2532                  * of two sector size assumptions
2533                  */
2534                 sector_size = 512;
2535         }
2536         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2537         blk_queue_physical_block_size(sdp->request_queue,
2538                                       sdkp->physical_block_size);
2539         sdkp->device->sector_size = sector_size;
2540
2541         if (sdkp->capacity > 0xffffffff)
2542                 sdp->use_16_for_rw = 1;
2543
2544 }
2545
2546 /*
2547  * Print disk capacity
2548  */
2549 static void
2550 sd_print_capacity(struct scsi_disk *sdkp,
2551                   sector_t old_capacity)
2552 {
2553         int sector_size = sdkp->device->sector_size;
2554         char cap_str_2[10], cap_str_10[10];
2555
2556         string_get_size(sdkp->capacity, sector_size,
2557                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2558         string_get_size(sdkp->capacity, sector_size,
2559                         STRING_UNITS_10, cap_str_10,
2560                         sizeof(cap_str_10));
2561
2562         if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2563                 sd_printk(KERN_NOTICE, sdkp,
2564                           "%llu %d-byte logical blocks: (%s/%s)\n",
2565                           (unsigned long long)sdkp->capacity,
2566                           sector_size, cap_str_10, cap_str_2);
2567
2568                 if (sdkp->physical_block_size != sector_size)
2569                         sd_printk(KERN_NOTICE, sdkp,
2570                                   "%u-byte physical blocks\n",
2571                                   sdkp->physical_block_size);
2572
2573                 sd_zbc_print_zones(sdkp);
2574         }
2575 }
2576
2577 /* called with buffer of length 512 */
2578 static inline int
2579 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2580                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2581                  struct scsi_sense_hdr *sshdr)
2582 {
2583         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2584                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2585                                sshdr);
2586 }
2587
2588 /*
2589  * read write protect setting, if possible - called only in sd_revalidate_disk()
2590  * called with buffer of length SD_BUF_SIZE
2591  */
2592 static void
2593 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2594 {
2595         int res;
2596         struct scsi_device *sdp = sdkp->device;
2597         struct scsi_mode_data data;
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);
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                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2956                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2957         }
2958
2959         if (sdkp->device->type == TYPE_ZBC) {
2960                 /* Host-managed */
2961                 q->limits.zoned = BLK_ZONED_HM;
2962         } else {
2963                 sdkp->zoned = (buffer[8] >> 4) & 3;
2964                 if (sdkp->zoned == 1)
2965                         /* Host-aware */
2966                         q->limits.zoned = BLK_ZONED_HA;
2967                 else
2968                         /*
2969                          * Treat drive-managed devices as
2970                          * regular block devices.
2971                          */
2972                         q->limits.zoned = BLK_ZONED_NONE;
2973         }
2974         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2975                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2976                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2977
2978  out:
2979         kfree(buffer);
2980 }
2981
2982 /**
2983  * sd_read_block_provisioning - Query provisioning VPD page
2984  * @sdkp: disk to query
2985  */
2986 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2987 {
2988         unsigned char *buffer;
2989         const int vpd_len = 8;
2990
2991         if (sdkp->lbpme == 0)
2992                 return;
2993
2994         buffer = kmalloc(vpd_len, GFP_KERNEL);
2995
2996         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2997                 goto out;
2998
2999         sdkp->lbpvpd    = 1;
3000         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3001         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3002         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3003
3004  out:
3005         kfree(buffer);
3006 }
3007
3008 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3009 {
3010         struct scsi_device *sdev = sdkp->device;
3011
3012         if (sdev->host->no_write_same) {
3013                 sdev->no_write_same = 1;
3014
3015                 return;
3016         }
3017
3018         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3019                 /* too large values might cause issues with arcmsr */
3020                 int vpd_buf_len = 64;
3021
3022                 sdev->no_report_opcodes = 1;
3023
3024                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3025                  * CODES is unsupported and the device has an ATA
3026                  * Information VPD page (SAT).
3027                  */
3028                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3029                         sdev->no_write_same = 1;
3030         }
3031
3032         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3033                 sdkp->ws16 = 1;
3034
3035         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3036                 sdkp->ws10 = 1;
3037 }
3038
3039 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3040 {
3041         struct scsi_device *sdev = sdkp->device;
3042
3043         if (!sdev->security_supported)
3044                 return;
3045
3046         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3047                         SECURITY_PROTOCOL_IN) == 1 &&
3048             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3049                         SECURITY_PROTOCOL_OUT) == 1)
3050                 sdkp->security = 1;
3051 }
3052
3053 /**
3054  *      sd_revalidate_disk - called the first time a new disk is seen,
3055  *      performs disk spin up, read_capacity, etc.
3056  *      @disk: struct gendisk we care about
3057  **/
3058 static int sd_revalidate_disk(struct gendisk *disk)
3059 {
3060         struct scsi_disk *sdkp = scsi_disk(disk);
3061         struct scsi_device *sdp = sdkp->device;
3062         struct request_queue *q = sdkp->disk->queue;
3063         sector_t old_capacity = sdkp->capacity;
3064         unsigned char *buffer;
3065         unsigned int dev_max, rw_max;
3066
3067         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3068                                       "sd_revalidate_disk\n"));
3069
3070         /*
3071          * If the device is offline, don't try and read capacity or any
3072          * of the other niceties.
3073          */
3074         if (!scsi_device_online(sdp))
3075                 goto out;
3076
3077         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3078         if (!buffer) {
3079                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3080                           "allocation failure.\n");
3081                 goto out;
3082         }
3083
3084         sd_spinup_disk(sdkp);
3085
3086         /*
3087          * Without media there is no reason to ask; moreover, some devices
3088          * react badly if we do.
3089          */
3090         if (sdkp->media_present) {
3091                 sd_read_capacity(sdkp, buffer);
3092
3093                 if (scsi_device_supports_vpd(sdp)) {
3094                         sd_read_block_provisioning(sdkp);
3095                         sd_read_block_limits(sdkp);
3096                         sd_read_block_characteristics(sdkp);
3097                         sd_zbc_read_zones(sdkp, buffer);
3098                 }
3099
3100                 sd_print_capacity(sdkp, old_capacity);
3101
3102                 sd_read_write_protect_flag(sdkp, buffer);
3103                 sd_read_cache_type(sdkp, buffer);
3104                 sd_read_app_tag_own(sdkp, buffer);
3105                 sd_read_write_same(sdkp, buffer);
3106                 sd_read_security(sdkp, buffer);
3107         }
3108
3109         /*
3110          * We now have all cache related info, determine how we deal
3111          * with flush requests.
3112          */
3113         sd_set_flush_flag(sdkp);
3114
3115         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3116         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3117
3118         /* Some devices report a maximum block count for READ/WRITE requests. */
3119         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3120         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3121
3122         /*
3123          * Determine the device's preferred I/O size for reads and writes
3124          * unless the reported value is unreasonably small, large, or
3125          * garbage.
3126          */
3127         if (sdkp->opt_xfer_blocks &&
3128             sdkp->opt_xfer_blocks <= dev_max &&
3129             sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3130             logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3131                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3132                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3133         } else
3134                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3135                                       (sector_t)BLK_DEF_MAX_SECTORS);
3136
3137         /* Do not exceed controller limit */
3138         rw_max = min(rw_max, queue_max_hw_sectors(q));
3139
3140         /*
3141          * Only update max_sectors if previously unset or if the current value
3142          * exceeds the capabilities of the hardware.
3143          */
3144         if (sdkp->first_scan ||
3145             q->limits.max_sectors > q->limits.max_dev_sectors ||
3146             q->limits.max_sectors > q->limits.max_hw_sectors)
3147                 q->limits.max_sectors = rw_max;
3148
3149         sdkp->first_scan = 0;
3150
3151         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3152         sd_config_write_same(sdkp);
3153         kfree(buffer);
3154
3155  out:
3156         return 0;
3157 }
3158
3159 /**
3160  *      sd_unlock_native_capacity - unlock native capacity
3161  *      @disk: struct gendisk to set capacity for
3162  *
3163  *      Block layer calls this function if it detects that partitions
3164  *      on @disk reach beyond the end of the device.  If the SCSI host
3165  *      implements ->unlock_native_capacity() method, it's invoked to
3166  *      give it a chance to adjust the device capacity.
3167  *
3168  *      CONTEXT:
3169  *      Defined by block layer.  Might sleep.
3170  */
3171 static void sd_unlock_native_capacity(struct gendisk *disk)
3172 {
3173         struct scsi_device *sdev = scsi_disk(disk)->device;
3174
3175         if (sdev->host->hostt->unlock_native_capacity)
3176                 sdev->host->hostt->unlock_native_capacity(sdev);
3177 }
3178
3179 /**
3180  *      sd_format_disk_name - format disk name
3181  *      @prefix: name prefix - ie. "sd" for SCSI disks
3182  *      @index: index of the disk to format name for
3183  *      @buf: output buffer
3184  *      @buflen: length of the output buffer
3185  *
3186  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3187  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3188  *      which is followed by sdaaa.
3189  *
3190  *      This is basically 26 base counting with one extra 'nil' entry
3191  *      at the beginning from the second digit on and can be
3192  *      determined using similar method as 26 base conversion with the
3193  *      index shifted -1 after each digit is computed.
3194  *
3195  *      CONTEXT:
3196  *      Don't care.
3197  *
3198  *      RETURNS:
3199  *      0 on success, -errno on failure.
3200  */
3201 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3202 {
3203         const int base = 'z' - 'a' + 1;
3204         char *begin = buf + strlen(prefix);
3205         char *end = buf + buflen;
3206         char *p;
3207         int unit;
3208
3209         p = end - 1;
3210         *p = '\0';
3211         unit = base;
3212         do {
3213                 if (p == begin)
3214                         return -EINVAL;
3215                 *--p = 'a' + (index % unit);
3216                 index = (index / unit) - 1;
3217         } while (index >= 0);
3218
3219         memmove(begin, p, end - p);
3220         memcpy(buf, prefix, strlen(prefix));
3221
3222         return 0;
3223 }
3224
3225 /*
3226  * The asynchronous part of sd_probe
3227  */
3228 static void sd_probe_async(void *data, async_cookie_t cookie)
3229 {