03163ac5fe9583362491efa831275c77a59e6122
[sfrench/cifs-2.6.git] / drivers / scsi / sd.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *      sd.c Copyright (C) 1992 Drew Eckhardt
4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5  *
6  *      Linux scsi disk driver
7  *              Initial versions: Drew Eckhardt
8  *              Subsequent revisions: Eric Youngdale
9  *      Modification history:
10  *       - Drew Eckhardt <drew@colorado.edu> original
11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple 
12  *         outstanding request, and other enhancements.
13  *         Support loadable low-level scsi drivers.
14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using 
15  *         eight major numbers.
16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in 
18  *         sd_init and cleanups.
19  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
20  *         not being read in sd_open. Fix problem where removable media 
21  *         could be ejected after sd_open.
22  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox 
24  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>: 
25  *         Support 32k/1M disks.
26  *
27  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
28  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
32  *      Note: when the logging level is set by the user, it must be greater
33  *      than the level indicated above to trigger output.       
34  */
35
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
57 #include <linux/pr.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
61
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
71
72 #include "sd.h"
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
75
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
79
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100
101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
102 #define SD_MINORS       16
103 #else
104 #define SD_MINORS       0
105 #endif
106
107 static void sd_config_discard(struct scsi_disk *, unsigned int);
108 static void sd_config_write_same(struct scsi_disk *);
109 static int  sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static int  sd_probe(struct device *);
112 static int  sd_remove(struct device *);
113 static void sd_shutdown(struct device *);
114 static int sd_suspend_system(struct device *);
115 static int sd_suspend_runtime(struct device *);
116 static int sd_resume(struct device *);
117 static void sd_rescan(struct device *);
118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120 static int sd_done(struct scsi_cmnd *);
121 static void sd_eh_reset(struct scsi_cmnd *);
122 static int sd_eh_action(struct scsi_cmnd *, int);
123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124 static void scsi_disk_release(struct device *cdev);
125 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
126 static void sd_print_result(const struct scsi_disk *, const char *, int);
127
128 static DEFINE_IDA(sd_index_ida);
129
130 /* This semaphore is used to mediate the 0->1 reference get in the
131  * face of object destruction (i.e. we can't allow a get on an
132  * object after last put) */
133 static DEFINE_MUTEX(sd_ref_mutex);
134
135 static struct kmem_cache *sd_cdb_cache;
136 static mempool_t *sd_cdb_pool;
137 static mempool_t *sd_page_pool;
138
139 static const char *sd_cache_types[] = {
140         "write through", "none", "write back",
141         "write back, no read (daft)"
142 };
143
144 static void sd_set_flush_flag(struct scsi_disk *sdkp)
145 {
146         bool wc = false, fua = false;
147
148         if (sdkp->WCE) {
149                 wc = true;
150                 if (sdkp->DPOFUA)
151                         fua = true;
152         }
153
154         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
155 }
156
157 static ssize_t
158 cache_type_store(struct device *dev, struct device_attribute *attr,
159                  const char *buf, size_t count)
160 {
161         int ct, rcd, wce, sp;
162         struct scsi_disk *sdkp = to_scsi_disk(dev);
163         struct scsi_device *sdp = sdkp->device;
164         char buffer[64];
165         char *buffer_data;
166         struct scsi_mode_data data;
167         struct scsi_sense_hdr sshdr;
168         static const char temp[] = "temporary ";
169         int len;
170
171         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
172                 /* no cache control on RBC devices; theoretically they
173                  * can do it, but there's probably so many exceptions
174                  * it's not worth the risk */
175                 return -EINVAL;
176
177         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
178                 buf += sizeof(temp) - 1;
179                 sdkp->cache_override = 1;
180         } else {
181                 sdkp->cache_override = 0;
182         }
183
184         ct = sysfs_match_string(sd_cache_types, buf);
185         if (ct < 0)
186                 return -EINVAL;
187
188         rcd = ct & 0x01 ? 1 : 0;
189         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190
191         if (sdkp->cache_override) {
192                 sdkp->WCE = wce;
193                 sdkp->RCD = rcd;
194                 sd_set_flush_flag(sdkp);
195                 return count;
196         }
197
198         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199                             SD_MAX_RETRIES, &data, NULL))
200                 return -EINVAL;
201         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202                   data.block_descriptor_length);
203         buffer_data = buffer + data.header_length +
204                 data.block_descriptor_length;
205         buffer_data[2] &= ~0x05;
206         buffer_data[2] |= wce << 2 | rcd;
207         sp = buffer_data[0] & 0x80 ? 1 : 0;
208         buffer_data[0] &= ~0x80;
209
210         /*
211          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
212          * received mode parameter buffer before doing MODE SELECT.
213          */
214         data.device_specific = 0;
215
216         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
217                              SD_MAX_RETRIES, &data, &sshdr)) {
218                 if (scsi_sense_valid(&sshdr))
219                         sd_print_sense_hdr(sdkp, &sshdr);
220                 return -EINVAL;
221         }
222         revalidate_disk(sdkp->disk);
223         return count;
224 }
225
226 static ssize_t
227 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
228                        char *buf)
229 {
230         struct scsi_disk *sdkp = to_scsi_disk(dev);
231         struct scsi_device *sdp = sdkp->device;
232
233         return sprintf(buf, "%u\n", sdp->manage_start_stop);
234 }
235
236 static ssize_t
237 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
238                         const char *buf, size_t count)
239 {
240         struct scsi_disk *sdkp = to_scsi_disk(dev);
241         struct scsi_device *sdp = sdkp->device;
242         bool v;
243
244         if (!capable(CAP_SYS_ADMIN))
245                 return -EACCES;
246
247         if (kstrtobool(buf, &v))
248                 return -EINVAL;
249
250         sdp->manage_start_stop = v;
251
252         return count;
253 }
254 static DEVICE_ATTR_RW(manage_start_stop);
255
256 static ssize_t
257 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
258 {
259         struct scsi_disk *sdkp = to_scsi_disk(dev);
260
261         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
262 }
263
264 static ssize_t
265 allow_restart_store(struct device *dev, struct device_attribute *attr,
266                     const char *buf, size_t count)
267 {
268         bool v;
269         struct scsi_disk *sdkp = to_scsi_disk(dev);
270         struct scsi_device *sdp = sdkp->device;
271
272         if (!capable(CAP_SYS_ADMIN))
273                 return -EACCES;
274
275         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
276                 return -EINVAL;
277
278         if (kstrtobool(buf, &v))
279                 return -EINVAL;
280
281         sdp->allow_restart = v;
282
283         return count;
284 }
285 static DEVICE_ATTR_RW(allow_restart);
286
287 static ssize_t
288 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
289 {
290         struct scsi_disk *sdkp = to_scsi_disk(dev);
291         int ct = sdkp->RCD + 2*sdkp->WCE;
292
293         return sprintf(buf, "%s\n", sd_cache_types[ct]);
294 }
295 static DEVICE_ATTR_RW(cache_type);
296
297 static ssize_t
298 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
299 {
300         struct scsi_disk *sdkp = to_scsi_disk(dev);
301
302         return sprintf(buf, "%u\n", sdkp->DPOFUA);
303 }
304 static DEVICE_ATTR_RO(FUA);
305
306 static ssize_t
307 protection_type_show(struct device *dev, struct device_attribute *attr,
308                      char *buf)
309 {
310         struct scsi_disk *sdkp = to_scsi_disk(dev);
311
312         return sprintf(buf, "%u\n", sdkp->protection_type);
313 }
314
315 static ssize_t
316 protection_type_store(struct device *dev, struct device_attribute *attr,
317                       const char *buf, size_t count)
318 {
319         struct scsi_disk *sdkp = to_scsi_disk(dev);
320         unsigned int val;
321         int err;
322
323         if (!capable(CAP_SYS_ADMIN))
324                 return -EACCES;
325
326         err = kstrtouint(buf, 10, &val);
327
328         if (err)
329                 return err;
330
331         if (val <= T10_PI_TYPE3_PROTECTION)
332                 sdkp->protection_type = val;
333
334         return count;
335 }
336 static DEVICE_ATTR_RW(protection_type);
337
338 static ssize_t
339 protection_mode_show(struct device *dev, struct device_attribute *attr,
340                      char *buf)
341 {
342         struct scsi_disk *sdkp = to_scsi_disk(dev);
343         struct scsi_device *sdp = sdkp->device;
344         unsigned int dif, dix;
345
346         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
347         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
348
349         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
350                 dif = 0;
351                 dix = 1;
352         }
353
354         if (!dif && !dix)
355                 return sprintf(buf, "none\n");
356
357         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
358 }
359 static DEVICE_ATTR_RO(protection_mode);
360
361 static ssize_t
362 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
363 {
364         struct scsi_disk *sdkp = to_scsi_disk(dev);
365
366         return sprintf(buf, "%u\n", sdkp->ATO);
367 }
368 static DEVICE_ATTR_RO(app_tag_own);
369
370 static ssize_t
371 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
372                        char *buf)
373 {
374         struct scsi_disk *sdkp = to_scsi_disk(dev);
375
376         return sprintf(buf, "%u\n", sdkp->lbpme);
377 }
378 static DEVICE_ATTR_RO(thin_provisioning);
379
380 /* sysfs_match_string() requires dense arrays */
381 static const char *lbp_mode[] = {
382         [SD_LBP_FULL]           = "full",
383         [SD_LBP_UNMAP]          = "unmap",
384         [SD_LBP_WS16]           = "writesame_16",
385         [SD_LBP_WS10]           = "writesame_10",
386         [SD_LBP_ZERO]           = "writesame_zero",
387         [SD_LBP_DISABLE]        = "disabled",
388 };
389
390 static ssize_t
391 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
392                        char *buf)
393 {
394         struct scsi_disk *sdkp = to_scsi_disk(dev);
395
396         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
397 }
398
399 static ssize_t
400 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
401                         const char *buf, size_t count)
402 {
403         struct scsi_disk *sdkp = to_scsi_disk(dev);
404         struct scsi_device *sdp = sdkp->device;
405         int mode;
406
407         if (!capable(CAP_SYS_ADMIN))
408                 return -EACCES;
409
410         if (sd_is_zoned(sdkp)) {
411                 sd_config_discard(sdkp, SD_LBP_DISABLE);
412                 return count;
413         }
414
415         if (sdp->type != TYPE_DISK)
416                 return -EINVAL;
417
418         mode = sysfs_match_string(lbp_mode, buf);
419         if (mode < 0)
420                 return -EINVAL;
421
422         sd_config_discard(sdkp, mode);
423
424         return count;
425 }
426 static DEVICE_ATTR_RW(provisioning_mode);
427
428 /* sysfs_match_string() requires dense arrays */
429 static const char *zeroing_mode[] = {
430         [SD_ZERO_WRITE]         = "write",
431         [SD_ZERO_WS]            = "writesame",
432         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
433         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
434 };
435
436 static ssize_t
437 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
438                   char *buf)
439 {
440         struct scsi_disk *sdkp = to_scsi_disk(dev);
441
442         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
443 }
444
445 static ssize_t
446 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
447                    const char *buf, size_t count)
448 {
449         struct scsi_disk *sdkp = to_scsi_disk(dev);
450         int mode;
451
452         if (!capable(CAP_SYS_ADMIN))
453                 return -EACCES;
454
455         mode = sysfs_match_string(zeroing_mode, buf);
456         if (mode < 0)
457                 return -EINVAL;
458
459         sdkp->zeroing_mode = mode;
460
461         return count;
462 }
463 static DEVICE_ATTR_RW(zeroing_mode);
464
465 static ssize_t
466 max_medium_access_timeouts_show(struct device *dev,
467                                 struct device_attribute *attr, char *buf)
468 {
469         struct scsi_disk *sdkp = to_scsi_disk(dev);
470
471         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
472 }
473
474 static ssize_t
475 max_medium_access_timeouts_store(struct device *dev,
476                                  struct device_attribute *attr, const char *buf,
477                                  size_t count)
478 {
479         struct scsi_disk *sdkp = to_scsi_disk(dev);
480         int err;
481
482         if (!capable(CAP_SYS_ADMIN))
483                 return -EACCES;
484
485         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
486
487         return err ? err : count;
488 }
489 static DEVICE_ATTR_RW(max_medium_access_timeouts);
490
491 static ssize_t
492 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
493                            char *buf)
494 {
495         struct scsi_disk *sdkp = to_scsi_disk(dev);
496
497         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
498 }
499
500 static ssize_t
501 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
502                             const char *buf, size_t count)
503 {
504         struct scsi_disk *sdkp = to_scsi_disk(dev);
505         struct scsi_device *sdp = sdkp->device;
506         unsigned long max;
507         int err;
508
509         if (!capable(CAP_SYS_ADMIN))
510                 return -EACCES;
511
512         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
513                 return -EINVAL;
514
515         err = kstrtoul(buf, 10, &max);
516
517         if (err)
518                 return err;
519
520         if (max == 0)
521                 sdp->no_write_same = 1;
522         else if (max <= SD_MAX_WS16_BLOCKS) {
523                 sdp->no_write_same = 0;
524                 sdkp->max_ws_blocks = max;
525         }
526
527         sd_config_write_same(sdkp);
528
529         return count;
530 }
531 static DEVICE_ATTR_RW(max_write_same_blocks);
532
533 static struct attribute *sd_disk_attrs[] = {
534         &dev_attr_cache_type.attr,
535         &dev_attr_FUA.attr,
536         &dev_attr_allow_restart.attr,
537         &dev_attr_manage_start_stop.attr,
538         &dev_attr_protection_type.attr,
539         &dev_attr_protection_mode.attr,
540         &dev_attr_app_tag_own.attr,
541         &dev_attr_thin_provisioning.attr,
542         &dev_attr_provisioning_mode.attr,
543         &dev_attr_zeroing_mode.attr,
544         &dev_attr_max_write_same_blocks.attr,
545         &dev_attr_max_medium_access_timeouts.attr,
546         NULL,
547 };
548 ATTRIBUTE_GROUPS(sd_disk);
549
550 static struct class sd_disk_class = {
551         .name           = "scsi_disk",
552         .owner          = THIS_MODULE,
553         .dev_release    = scsi_disk_release,
554         .dev_groups     = sd_disk_groups,
555 };
556
557 static const struct dev_pm_ops sd_pm_ops = {
558         .suspend                = sd_suspend_system,
559         .resume                 = sd_resume,
560         .poweroff               = sd_suspend_system,
561         .restore                = sd_resume,
562         .runtime_suspend        = sd_suspend_runtime,
563         .runtime_resume         = sd_resume,
564 };
565
566 static struct scsi_driver sd_template = {
567         .gendrv = {
568                 .name           = "sd",
569                 .owner          = THIS_MODULE,
570                 .probe          = sd_probe,
571                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
572                 .remove         = sd_remove,
573                 .shutdown       = sd_shutdown,
574                 .pm             = &sd_pm_ops,
575         },
576         .rescan                 = sd_rescan,
577         .init_command           = sd_init_command,
578         .uninit_command         = sd_uninit_command,
579         .done                   = sd_done,
580         .eh_action              = sd_eh_action,
581         .eh_reset               = sd_eh_reset,
582 };
583
584 /*
585  * Dummy kobj_map->probe function.
586  * The default ->probe function will call modprobe, which is
587  * pointless as this module is already loaded.
588  */
589 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
590 {
591         return NULL;
592 }
593
594 /*
595  * Device no to disk mapping:
596  * 
597  *       major         disc2     disc  p1
598  *   |............|.............|....|....| <- dev_t
599  *    31        20 19          8 7  4 3  0
600  * 
601  * Inside a major, we have 16k disks, however mapped non-
602  * contiguously. The first 16 disks are for major0, the next
603  * ones with major1, ... Disk 256 is for major0 again, disk 272 
604  * for major1, ... 
605  * As we stay compatible with our numbering scheme, we can reuse 
606  * the well-know SCSI majors 8, 65--71, 136--143.
607  */
608 static int sd_major(int major_idx)
609 {
610         switch (major_idx) {
611         case 0:
612                 return SCSI_DISK0_MAJOR;
613         case 1 ... 7:
614                 return SCSI_DISK1_MAJOR + major_idx - 1;
615         case 8 ... 15:
616                 return SCSI_DISK8_MAJOR + major_idx - 8;
617         default:
618                 BUG();
619                 return 0;       /* shut up gcc */
620         }
621 }
622
623 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
624 {
625         struct scsi_disk *sdkp = NULL;
626
627         mutex_lock(&sd_ref_mutex);
628
629         if (disk->private_data) {
630                 sdkp = scsi_disk(disk);
631                 if (scsi_device_get(sdkp->device) == 0)
632                         get_device(&sdkp->dev);
633                 else
634                         sdkp = NULL;
635         }
636         mutex_unlock(&sd_ref_mutex);
637         return sdkp;
638 }
639
640 static void scsi_disk_put(struct scsi_disk *sdkp)
641 {
642         struct scsi_device *sdev = sdkp->device;
643
644         mutex_lock(&sd_ref_mutex);
645         put_device(&sdkp->dev);
646         scsi_device_put(sdev);
647         mutex_unlock(&sd_ref_mutex);
648 }
649
650 #ifdef CONFIG_BLK_SED_OPAL
651 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
652                 size_t len, bool send)
653 {
654         struct scsi_device *sdev = data;
655         u8 cdb[12] = { 0, };
656         int ret;
657
658         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
659         cdb[1] = secp;
660         put_unaligned_be16(spsp, &cdb[2]);
661         put_unaligned_be32(len, &cdb[6]);
662
663         ret = scsi_execute_req(sdev, cdb,
664                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
665                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
666         return ret <= 0 ? ret : -EIO;
667 }
668 #endif /* CONFIG_BLK_SED_OPAL */
669
670 /*
671  * Look up the DIX operation based on whether the command is read or
672  * write and whether dix and dif are enabled.
673  */
674 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
675 {
676         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
677         static const unsigned int ops[] = {     /* wrt dix dif */
678                 SCSI_PROT_NORMAL,               /*  0   0   0  */
679                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
680                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
681                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
682                 SCSI_PROT_NORMAL,               /*  1   0   0  */
683                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
684                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
685                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
686         };
687
688         return ops[write << 2 | dix << 1 | dif];
689 }
690
691 /*
692  * Returns a mask of the protection flags that are valid for a given DIX
693  * operation.
694  */
695 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
696 {
697         static const unsigned int flag_mask[] = {
698                 [SCSI_PROT_NORMAL]              = 0,
699
700                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
701                                                   SCSI_PROT_GUARD_CHECK |
702                                                   SCSI_PROT_REF_CHECK |
703                                                   SCSI_PROT_REF_INCREMENT,
704
705                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
706                                                   SCSI_PROT_IP_CHECKSUM,
707
708                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
709                                                   SCSI_PROT_GUARD_CHECK |
710                                                   SCSI_PROT_REF_CHECK |
711                                                   SCSI_PROT_REF_INCREMENT |
712                                                   SCSI_PROT_IP_CHECKSUM,
713
714                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
715                                                   SCSI_PROT_REF_INCREMENT,
716
717                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
718                                                   SCSI_PROT_REF_CHECK |
719                                                   SCSI_PROT_REF_INCREMENT |
720                                                   SCSI_PROT_IP_CHECKSUM,
721
722                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
723                                                   SCSI_PROT_GUARD_CHECK |
724                                                   SCSI_PROT_REF_CHECK |
725                                                   SCSI_PROT_REF_INCREMENT |
726                                                   SCSI_PROT_IP_CHECKSUM,
727         };
728
729         return flag_mask[prot_op];
730 }
731
732 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
733                                            unsigned int dix, unsigned int dif)
734 {
735         struct bio *bio = scmd->request->bio;
736         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
737         unsigned int protect = 0;
738
739         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
740                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
741                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
742
743                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
744                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
745         }
746
747         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
748                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
749
750                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
751                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
752         }
753
754         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
755                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
756
757                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
758                         protect = 3 << 5;       /* Disable target PI checking */
759                 else
760                         protect = 1 << 5;       /* Enable target PI checking */
761         }
762
763         scsi_set_prot_op(scmd, prot_op);
764         scsi_set_prot_type(scmd, dif);
765         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
766
767         return protect;
768 }
769
770 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
771 {
772         struct request_queue *q = sdkp->disk->queue;
773         unsigned int logical_block_size = sdkp->device->sector_size;
774         unsigned int max_blocks = 0;
775
776         q->limits.discard_alignment =
777                 sdkp->unmap_alignment * logical_block_size;
778         q->limits.discard_granularity =
779                 max(sdkp->physical_block_size,
780                     sdkp->unmap_granularity * logical_block_size);
781         sdkp->provisioning_mode = mode;
782
783         switch (mode) {
784
785         case SD_LBP_FULL:
786         case SD_LBP_DISABLE:
787                 blk_queue_max_discard_sectors(q, 0);
788                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
789                 return;
790
791         case SD_LBP_UNMAP:
792                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
793                                           (u32)SD_MAX_WS16_BLOCKS);
794                 break;
795
796         case SD_LBP_WS16:
797                 if (sdkp->device->unmap_limit_for_ws)
798                         max_blocks = sdkp->max_unmap_blocks;
799                 else
800                         max_blocks = sdkp->max_ws_blocks;
801
802                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
803                 break;
804
805         case SD_LBP_WS10:
806                 if (sdkp->device->unmap_limit_for_ws)
807                         max_blocks = sdkp->max_unmap_blocks;
808                 else
809                         max_blocks = sdkp->max_ws_blocks;
810
811                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
812                 break;
813
814         case SD_LBP_ZERO:
815                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
816                                           (u32)SD_MAX_WS10_BLOCKS);
817                 break;
818         }
819
820         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
821         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
822 }
823
824 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
825 {
826         struct scsi_device *sdp = cmd->device;
827         struct request *rq = cmd->request;
828         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
829         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
830         unsigned int data_len = 24;
831         char *buf;
832
833         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
834         if (!rq->special_vec.bv_page)
835                 return BLK_STS_RESOURCE;
836         clear_highpage(rq->special_vec.bv_page);
837         rq->special_vec.bv_offset = 0;
838         rq->special_vec.bv_len = data_len;
839         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
840
841         cmd->cmd_len = 10;
842         cmd->cmnd[0] = UNMAP;
843         cmd->cmnd[8] = 24;
844
845         buf = page_address(rq->special_vec.bv_page);
846         put_unaligned_be16(6 + 16, &buf[0]);
847         put_unaligned_be16(16, &buf[2]);
848         put_unaligned_be64(lba, &buf[8]);
849         put_unaligned_be32(nr_blocks, &buf[16]);
850
851         cmd->allowed = SD_MAX_RETRIES;
852         cmd->transfersize = data_len;
853         rq->timeout = SD_TIMEOUT;
854
855         return scsi_init_io(cmd);
856 }
857
858 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
859                 bool unmap)
860 {
861         struct scsi_device *sdp = cmd->device;
862         struct request *rq = cmd->request;
863         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
864         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
865         u32 data_len = sdp->sector_size;
866
867         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
868         if (!rq->special_vec.bv_page)
869                 return BLK_STS_RESOURCE;
870         clear_highpage(rq->special_vec.bv_page);
871         rq->special_vec.bv_offset = 0;
872         rq->special_vec.bv_len = data_len;
873         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
874
875         cmd->cmd_len = 16;
876         cmd->cmnd[0] = WRITE_SAME_16;
877         if (unmap)
878                 cmd->cmnd[1] = 0x8; /* UNMAP */
879         put_unaligned_be64(lba, &cmd->cmnd[2]);
880         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
881
882         cmd->allowed = SD_MAX_RETRIES;
883         cmd->transfersize = data_len;
884         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
885
886         return scsi_init_io(cmd);
887 }
888
889 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
890                 bool unmap)
891 {
892         struct scsi_device *sdp = cmd->device;
893         struct request *rq = cmd->request;
894         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
895         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
896         u32 data_len = sdp->sector_size;
897
898         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
899         if (!rq->special_vec.bv_page)
900                 return BLK_STS_RESOURCE;
901         clear_highpage(rq->special_vec.bv_page);
902         rq->special_vec.bv_offset = 0;
903         rq->special_vec.bv_len = data_len;
904         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905
906         cmd->cmd_len = 10;
907         cmd->cmnd[0] = WRITE_SAME;
908         if (unmap)
909                 cmd->cmnd[1] = 0x8; /* UNMAP */
910         put_unaligned_be32(lba, &cmd->cmnd[2]);
911         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
912
913         cmd->allowed = SD_MAX_RETRIES;
914         cmd->transfersize = data_len;
915         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
916
917         return scsi_init_io(cmd);
918 }
919
920 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
921 {
922         struct request *rq = cmd->request;
923         struct scsi_device *sdp = cmd->device;
924         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
925         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
926         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
927
928         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
929                 switch (sdkp->zeroing_mode) {
930                 case SD_ZERO_WS16_UNMAP:
931                         return sd_setup_write_same16_cmnd(cmd, true);
932                 case SD_ZERO_WS10_UNMAP:
933                         return sd_setup_write_same10_cmnd(cmd, true);
934                 }
935         }
936
937         if (sdp->no_write_same)
938                 return BLK_STS_TARGET;
939
940         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
941                 return sd_setup_write_same16_cmnd(cmd, false);
942
943         return sd_setup_write_same10_cmnd(cmd, false);
944 }
945
946 static void sd_config_write_same(struct scsi_disk *sdkp)
947 {
948         struct request_queue *q = sdkp->disk->queue;
949         unsigned int logical_block_size = sdkp->device->sector_size;
950
951         if (sdkp->device->no_write_same) {
952                 sdkp->max_ws_blocks = 0;
953                 goto out;
954         }
955
956         /* Some devices can not handle block counts above 0xffff despite
957          * supporting WRITE SAME(16). Consequently we default to 64k
958          * blocks per I/O unless the device explicitly advertises a
959          * bigger limit.
960          */
961         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
962                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
963                                                    (u32)SD_MAX_WS16_BLOCKS);
964         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
965                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
966                                                    (u32)SD_MAX_WS10_BLOCKS);
967         else {
968                 sdkp->device->no_write_same = 1;
969                 sdkp->max_ws_blocks = 0;
970         }
971
972         if (sdkp->lbprz && sdkp->lbpws)
973                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
974         else if (sdkp->lbprz && sdkp->lbpws10)
975                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
976         else if (sdkp->max_ws_blocks)
977                 sdkp->zeroing_mode = SD_ZERO_WS;
978         else
979                 sdkp->zeroing_mode = SD_ZERO_WRITE;
980
981         if (sdkp->max_ws_blocks &&
982             sdkp->physical_block_size > logical_block_size) {
983                 /*
984                  * Reporting a maximum number of blocks that is not aligned
985                  * on the device physical size would cause a large write same
986                  * request to be split into physically unaligned chunks by
987                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
988                  * even if the caller of these functions took care to align the
989                  * large request. So make sure the maximum reported is aligned
990                  * to the device physical block size. This is only an optional
991                  * optimization for regular disks, but this is mandatory to
992                  * avoid failure of large write same requests directed at
993                  * sequential write required zones of host-managed ZBC disks.
994                  */
995                 sdkp->max_ws_blocks =
996                         round_down(sdkp->max_ws_blocks,
997                                    bytes_to_logical(sdkp->device,
998                                                     sdkp->physical_block_size));
999         }
1000
1001 out:
1002         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1003                                          (logical_block_size >> 9));
1004         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1005                                          (logical_block_size >> 9));
1006 }
1007
1008 /**
1009  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1010  * @cmd: command to prepare
1011  *
1012  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1013  * the preference indicated by the target device.
1014  **/
1015 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1016 {
1017         struct request *rq = cmd->request;
1018         struct scsi_device *sdp = cmd->device;
1019         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1020         struct bio *bio = rq->bio;
1021         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1022         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023         blk_status_t ret;
1024
1025         if (sdkp->device->no_write_same)
1026                 return BLK_STS_TARGET;
1027
1028         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1029
1030         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1031
1032         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1033                 cmd->cmd_len = 16;
1034                 cmd->cmnd[0] = WRITE_SAME_16;
1035                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1036                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037         } else {
1038                 cmd->cmd_len = 10;
1039                 cmd->cmnd[0] = WRITE_SAME;
1040                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1041                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042         }
1043
1044         cmd->transfersize = sdp->sector_size;
1045         cmd->allowed = SD_MAX_RETRIES;
1046
1047         /*
1048          * For WRITE SAME the data transferred via the DATA OUT buffer is
1049          * different from the amount of data actually written to the target.
1050          *
1051          * We set up __data_len to the amount of data transferred via the
1052          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1053          * to transfer a single sector of data first, but then reset it to
1054          * the amount of data to be written right after so that the I/O path
1055          * knows how much to actually write.
1056          */
1057         rq->__data_len = sdp->sector_size;
1058         ret = scsi_init_io(cmd);
1059         rq->__data_len = blk_rq_bytes(rq);
1060
1061         return ret;
1062 }
1063
1064 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1065 {
1066         struct request *rq = cmd->request;
1067
1068         /* flush requests don't perform I/O, zero the S/G table */
1069         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070
1071         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1072         cmd->cmd_len = 10;
1073         cmd->transfersize = 0;
1074         cmd->allowed = SD_MAX_RETRIES;
1075
1076         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1077         return BLK_STS_OK;
1078 }
1079
1080 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1081                                        sector_t lba, unsigned int nr_blocks,
1082                                        unsigned char flags)
1083 {
1084         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1085         if (unlikely(cmd->cmnd == NULL))
1086                 return BLK_STS_RESOURCE;
1087
1088         cmd->cmd_len = SD_EXT_CDB_SIZE;
1089         memset(cmd->cmnd, 0, cmd->cmd_len);
1090
1091         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1092         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1093         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1094         cmd->cmnd[10] = flags;
1095         put_unaligned_be64(lba, &cmd->cmnd[12]);
1096         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1097         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1098
1099         return BLK_STS_OK;
1100 }
1101
1102 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1103                                        sector_t lba, unsigned int nr_blocks,
1104                                        unsigned char flags)
1105 {
1106         cmd->cmd_len  = 16;
1107         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1108         cmd->cmnd[1]  = flags;
1109         cmd->cmnd[14] = 0;
1110         cmd->cmnd[15] = 0;
1111         put_unaligned_be64(lba, &cmd->cmnd[2]);
1112         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1113
1114         return BLK_STS_OK;
1115 }
1116
1117 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1118                                        sector_t lba, unsigned int nr_blocks,
1119                                        unsigned char flags)
1120 {
1121         cmd->cmd_len = 10;
1122         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1123         cmd->cmnd[1] = flags;
1124         cmd->cmnd[6] = 0;
1125         cmd->cmnd[9] = 0;
1126         put_unaligned_be32(lba, &cmd->cmnd[2]);
1127         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1128
1129         return BLK_STS_OK;
1130 }
1131
1132 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1133                                       sector_t lba, unsigned int nr_blocks,
1134                                       unsigned char flags)
1135 {
1136         /* Avoid that 0 blocks gets translated into 256 blocks. */
1137         if (WARN_ON_ONCE(nr_blocks == 0))
1138                 return BLK_STS_IOERR;
1139
1140         if (unlikely(flags & 0x8)) {
1141                 /*
1142                  * This happens only if this drive failed 10byte rw
1143                  * command with ILLEGAL_REQUEST during operation and
1144                  * thus turned off use_10_for_rw.
1145                  */
1146                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1147                 return BLK_STS_IOERR;
1148         }
1149
1150         cmd->cmd_len = 6;
1151         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1152         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1153         cmd->cmnd[2] = (lba >> 8) & 0xff;
1154         cmd->cmnd[3] = lba & 0xff;
1155         cmd->cmnd[4] = nr_blocks;
1156         cmd->cmnd[5] = 0;
1157
1158         return BLK_STS_OK;
1159 }
1160
1161 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1162 {
1163         struct request *rq = cmd->request;
1164         struct scsi_device *sdp = cmd->device;
1165         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1166         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1167         sector_t threshold;
1168         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1169         bool dif, dix;
1170         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1171         bool write = rq_data_dir(rq) == WRITE;
1172         unsigned char protect, fua;
1173         blk_status_t ret;
1174
1175         ret = scsi_init_io(cmd);
1176         if (ret != BLK_STS_OK)
1177                 return ret;
1178
1179         if (!scsi_device_online(sdp) || sdp->changed) {
1180                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1181                 return BLK_STS_IOERR;
1182         }
1183
1184         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1185                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1186                 return BLK_STS_IOERR;
1187         }
1188
1189         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1190                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1191                 return BLK_STS_IOERR;
1192         }
1193
1194         /*
1195          * Some SD card readers can't handle accesses which touch the
1196          * last one or two logical blocks. Split accesses as needed.
1197          */
1198         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1199
1200         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1201                 if (lba < threshold) {
1202                         /* Access up to the threshold but not beyond */
1203                         nr_blocks = threshold - lba;
1204                 } else {
1205                         /* Access only a single logical block */
1206                         nr_blocks = 1;
1207                 }
1208         }
1209
1210         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1211         dix = scsi_prot_sg_count(cmd);
1212         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1213
1214         if (dif || dix)
1215                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1216         else
1217                 protect = 0;
1218
1219         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1220                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1221                                          protect | fua);
1222         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1223                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1224                                          protect | fua);
1225         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1226                    sdp->use_10_for_rw || protect) {
1227                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1228                                          protect | fua);
1229         } else {
1230                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1231                                         protect | fua);
1232         }
1233
1234         if (unlikely(ret != BLK_STS_OK))
1235                 return ret;
1236
1237         /*
1238          * We shouldn't disconnect in the middle of a sector, so with a dumb
1239          * host adapter, it's safe to assume that we can at least transfer
1240          * this many bytes between each connect / disconnect.
1241          */
1242         cmd->transfersize = sdp->sector_size;
1243         cmd->underflow = nr_blocks << 9;
1244         cmd->allowed = SD_MAX_RETRIES;
1245         cmd->sdb.length = nr_blocks * sdp->sector_size;
1246
1247         SCSI_LOG_HLQUEUE(1,
1248                          scmd_printk(KERN_INFO, cmd,
1249                                      "%s: block=%llu, count=%d\n", __func__,
1250                                      (unsigned long long)blk_rq_pos(rq),
1251                                      blk_rq_sectors(rq)));
1252         SCSI_LOG_HLQUEUE(2,
1253                          scmd_printk(KERN_INFO, cmd,
1254                                      "%s %d/%u 512 byte blocks.\n",
1255                                      write ? "writing" : "reading", nr_blocks,
1256                                      blk_rq_sectors(rq)));
1257
1258         /*
1259          * This indicates that the command is ready from our end to be
1260          * queued.
1261          */
1262         return BLK_STS_OK;
1263 }
1264
1265 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1266 {
1267         struct request *rq = cmd->request;
1268
1269         switch (req_op(rq)) {
1270         case REQ_OP_DISCARD:
1271                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1272                 case SD_LBP_UNMAP:
1273                         return sd_setup_unmap_cmnd(cmd);
1274                 case SD_LBP_WS16:
1275                         return sd_setup_write_same16_cmnd(cmd, true);
1276                 case SD_LBP_WS10:
1277                         return sd_setup_write_same10_cmnd(cmd, true);
1278                 case SD_LBP_ZERO:
1279                         return sd_setup_write_same10_cmnd(cmd, false);
1280                 default:
1281                         return BLK_STS_TARGET;
1282                 }
1283         case REQ_OP_WRITE_ZEROES:
1284                 return sd_setup_write_zeroes_cmnd(cmd);
1285         case REQ_OP_WRITE_SAME:
1286                 return sd_setup_write_same_cmnd(cmd);
1287         case REQ_OP_FLUSH:
1288                 return sd_setup_flush_cmnd(cmd);
1289         case REQ_OP_READ:
1290         case REQ_OP_WRITE:
1291                 return sd_setup_read_write_cmnd(cmd);
1292         case REQ_OP_ZONE_RESET:
1293                 return sd_zbc_setup_reset_cmnd(cmd, false);
1294         case REQ_OP_ZONE_RESET_ALL:
1295                 return sd_zbc_setup_reset_cmnd(cmd, true);
1296         default:
1297                 WARN_ON_ONCE(1);
1298                 return BLK_STS_NOTSUPP;
1299         }
1300 }
1301
1302 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1303 {
1304         struct request *rq = SCpnt->request;
1305         u8 *cmnd;
1306
1307         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1308                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1309
1310         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1311                 cmnd = SCpnt->cmnd;
1312                 SCpnt->cmnd = NULL;
1313                 SCpnt->cmd_len = 0;
1314                 mempool_free(cmnd, sd_cdb_pool);
1315         }
1316 }
1317
1318 /**
1319  *      sd_open - open a scsi disk device
1320  *      @bdev: Block device of the scsi disk to open
1321  *      @mode: FMODE_* mask
1322  *
1323  *      Returns 0 if successful. Returns a negated errno value in case 
1324  *      of error.
1325  *
1326  *      Note: This can be called from a user context (e.g. fsck(1) )
1327  *      or from within the kernel (e.g. as a result of a mount(1) ).
1328  *      In the latter case @inode and @filp carry an abridged amount
1329  *      of information as noted above.
1330  *
1331  *      Locking: called with bdev->bd_mutex held.
1332  **/
1333 static int sd_open(struct block_device *bdev, fmode_t mode)
1334 {
1335         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1336         struct scsi_device *sdev;
1337         int retval;
1338
1339         if (!sdkp)
1340                 return -ENXIO;
1341
1342         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1343
1344         sdev = sdkp->device;
1345
1346         /*
1347          * If the device is in error recovery, wait until it is done.
1348          * If the device is offline, then disallow any access to it.
1349          */
1350         retval = -ENXIO;
1351         if (!scsi_block_when_processing_errors(sdev))
1352                 goto error_out;
1353
1354         if (sdev->removable || sdkp->write_prot)
1355                 check_disk_change(bdev);
1356
1357         /*
1358          * If the drive is empty, just let the open fail.
1359          */
1360         retval = -ENOMEDIUM;
1361         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1362                 goto error_out;
1363
1364         /*
1365          * If the device has the write protect tab set, have the open fail
1366          * if the user expects to be able to write to the thing.
1367          */
1368         retval = -EROFS;
1369         if (sdkp->write_prot && (mode & FMODE_WRITE))
1370                 goto error_out;
1371
1372         /*
1373          * It is possible that the disk changing stuff resulted in
1374          * the device being taken offline.  If this is the case,
1375          * report this to the user, and don't pretend that the
1376          * open actually succeeded.
1377          */
1378         retval = -ENXIO;
1379         if (!scsi_device_online(sdev))
1380                 goto error_out;
1381
1382         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1383                 if (scsi_block_when_processing_errors(sdev))
1384                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1385         }
1386
1387         return 0;
1388
1389 error_out:
1390         scsi_disk_put(sdkp);
1391         return retval;  
1392 }
1393
1394 /**
1395  *      sd_release - invoked when the (last) close(2) is called on this
1396  *      scsi disk.
1397  *      @disk: disk to release
1398  *      @mode: FMODE_* mask
1399  *
1400  *      Returns 0. 
1401  *
1402  *      Note: may block (uninterruptible) if error recovery is underway
1403  *      on this disk.
1404  *
1405  *      Locking: called with bdev->bd_mutex held.
1406  **/
1407 static void sd_release(struct gendisk *disk, fmode_t mode)
1408 {
1409         struct scsi_disk *sdkp = scsi_disk(disk);
1410         struct scsi_device *sdev = sdkp->device;
1411
1412         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1413
1414         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1415                 if (scsi_block_when_processing_errors(sdev))
1416                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1417         }
1418
1419         scsi_disk_put(sdkp);
1420 }
1421
1422 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1423 {
1424         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1425         struct scsi_device *sdp = sdkp->device;
1426         struct Scsi_Host *host = sdp->host;
1427         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1428         int diskinfo[4];
1429
1430         /* default to most commonly used values */
1431         diskinfo[0] = 0x40;     /* 1 << 6 */
1432         diskinfo[1] = 0x20;     /* 1 << 5 */
1433         diskinfo[2] = capacity >> 11;
1434
1435         /* override with calculated, extended default, or driver values */
1436         if (host->hostt->bios_param)
1437                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1438         else
1439                 scsicam_bios_param(bdev, capacity, diskinfo);
1440
1441         geo->heads = diskinfo[0];
1442         geo->sectors = diskinfo[1];
1443         geo->cylinders = diskinfo[2];
1444         return 0;
1445 }
1446
1447 /**
1448  *      sd_ioctl - process an ioctl
1449  *      @bdev: target block device
1450  *      @mode: FMODE_* mask
1451  *      @cmd: ioctl command number
1452  *      @arg: this is third argument given to ioctl(2) system call.
1453  *      Often contains a pointer.
1454  *
1455  *      Returns 0 if successful (some ioctls return positive numbers on
1456  *      success as well). Returns a negated errno value in case of error.
1457  *
1458  *      Note: most ioctls are forward onto the block subsystem or further
1459  *      down in the scsi subsystem.
1460  **/
1461 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1462                     unsigned int cmd, unsigned long arg)
1463 {
1464         struct gendisk *disk = bdev->bd_disk;
1465         struct scsi_disk *sdkp = scsi_disk(disk);
1466         struct scsi_device *sdp = sdkp->device;
1467         void __user *p = (void __user *)arg;
1468         int error;
1469     
1470         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1471                                     "cmd=0x%x\n", disk->disk_name, cmd));
1472
1473         error = scsi_verify_blk_ioctl(bdev, cmd);
1474         if (error < 0)
1475                 return error;
1476
1477         /*
1478          * If we are in the middle of error recovery, don't let anyone
1479          * else try and use this device.  Also, if error recovery fails, it
1480          * may try and take the device offline, in which case all further
1481          * access to the device is prohibited.
1482          */
1483         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1484                         (mode & FMODE_NDELAY) != 0);
1485         if (error)
1486                 goto out;
1487
1488         if (is_sed_ioctl(cmd))
1489                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1490
1491         /*
1492          * Send SCSI addressing ioctls directly to mid level, send other
1493          * ioctls to block level and then onto mid level if they can't be
1494          * resolved.
1495          */
1496         switch (cmd) {
1497                 case SCSI_IOCTL_GET_IDLUN:
1498                 case SCSI_IOCTL_GET_BUS_NUMBER:
1499                         error = scsi_ioctl(sdp, cmd, p);
1500                         break;
1501                 default:
1502                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1503                         if (error != -ENOTTY)
1504                                 break;
1505                         error = scsi_ioctl(sdp, cmd, p);
1506                         break;
1507         }
1508 out:
1509         return error;
1510 }
1511
1512 static void set_media_not_present(struct scsi_disk *sdkp)
1513 {
1514         if (sdkp->media_present)
1515                 sdkp->device->changed = 1;
1516
1517         if (sdkp->device->removable) {
1518                 sdkp->media_present = 0;
1519                 sdkp->capacity = 0;
1520         }
1521 }
1522
1523 static int media_not_present(struct scsi_disk *sdkp,
1524                              struct scsi_sense_hdr *sshdr)
1525 {
1526         if (!scsi_sense_valid(sshdr))
1527                 return 0;
1528
1529         /* not invoked for commands that could return deferred errors */
1530         switch (sshdr->sense_key) {
1531         case UNIT_ATTENTION:
1532         case NOT_READY:
1533                 /* medium not present */
1534                 if (sshdr->asc == 0x3A) {
1535                         set_media_not_present(sdkp);
1536                         return 1;
1537                 }
1538         }
1539         return 0;
1540 }
1541
1542 /**
1543  *      sd_check_events - check media events
1544  *      @disk: kernel device descriptor
1545  *      @clearing: disk events currently being cleared
1546  *
1547  *      Returns mask of DISK_EVENT_*.
1548  *
1549  *      Note: this function is invoked from the block subsystem.
1550  **/
1551 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1552 {
1553         struct scsi_disk *sdkp = scsi_disk_get(disk);
1554         struct scsi_device *sdp;
1555         int retval;
1556
1557         if (!sdkp)
1558                 return 0;
1559
1560         sdp = sdkp->device;
1561         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1562
1563         /*
1564          * If the device is offline, don't send any commands - just pretend as
1565          * if the command failed.  If the device ever comes back online, we
1566          * can deal with it then.  It is only because of unrecoverable errors
1567          * that we would ever take a device offline in the first place.
1568          */
1569         if (!scsi_device_online(sdp)) {
1570                 set_media_not_present(sdkp);
1571                 goto out;
1572         }
1573
1574         /*
1575          * Using TEST_UNIT_READY enables differentiation between drive with
1576          * no cartridge loaded - NOT READY, drive with changed cartridge -
1577          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1578          *
1579          * Drives that auto spin down. eg iomega jaz 1G, will be started
1580          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1581          * sd_revalidate() is called.
1582          */
1583         if (scsi_block_when_processing_errors(sdp)) {
1584                 struct scsi_sense_hdr sshdr = { 0, };
1585
1586                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1587                                               &sshdr);
1588
1589                 /* failed to execute TUR, assume media not present */
1590                 if (host_byte(retval)) {
1591                         set_media_not_present(sdkp);
1592                         goto out;
1593                 }
1594
1595                 if (media_not_present(sdkp, &sshdr))
1596                         goto out;
1597         }
1598
1599         /*
1600          * For removable scsi disk we have to recognise the presence
1601          * of a disk in the drive.
1602          */
1603         if (!sdkp->media_present)
1604                 sdp->changed = 1;
1605         sdkp->media_present = 1;
1606 out:
1607         /*
1608          * sdp->changed is set under the following conditions:
1609          *
1610          *      Medium present state has changed in either direction.
1611          *      Device has indicated UNIT_ATTENTION.
1612          */
1613         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1614         sdp->changed = 0;
1615         scsi_disk_put(sdkp);
1616         return retval;
1617 }
1618
1619 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1620 {
1621         int retries, res;
1622         struct scsi_device *sdp = sdkp->device;
1623         const int timeout = sdp->request_queue->rq_timeout
1624                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1625         struct scsi_sense_hdr my_sshdr;
1626
1627         if (!scsi_device_online(sdp))
1628                 return -ENODEV;
1629
1630         /* caller might not be interested in sense, but we need it */
1631         if (!sshdr)
1632                 sshdr = &my_sshdr;
1633
1634         for (retries = 3; retries > 0; --retries) {
1635                 unsigned char cmd[10] = { 0 };
1636
1637                 cmd[0] = SYNCHRONIZE_CACHE;
1638                 /*
1639                  * Leave the rest of the command zero to indicate
1640                  * flush everything.
1641                  */
1642                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1643                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1644                 if (res == 0)
1645                         break;
1646         }
1647
1648         if (res) {
1649                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1650
1651                 if (driver_byte(res) == DRIVER_SENSE)
1652                         sd_print_sense_hdr(sdkp, sshdr);
1653
1654                 /* we need to evaluate the error return  */
1655                 if (scsi_sense_valid(sshdr) &&
1656                         (sshdr->asc == 0x3a ||  /* medium not present */
1657                          sshdr->asc == 0x20 ||  /* invalid command */
1658                          (sshdr->asc == 0x74 && sshdr->ascq == 0x71)))  /* drive is password locked */
1659                                 /* this is no error here */
1660                                 return 0;
1661
1662                 switch (host_byte(res)) {
1663                 /* ignore errors due to racing a disconnection */
1664                 case DID_BAD_TARGET:
1665                 case DID_NO_CONNECT:
1666                         return 0;
1667                 /* signal the upper layer it might try again */
1668                 case DID_BUS_BUSY:
1669                 case DID_IMM_RETRY:
1670                 case DID_REQUEUE:
1671                 case DID_SOFT_ERROR:
1672                         return -EBUSY;
1673                 default:
1674                         return -EIO;
1675                 }
1676         }
1677         return 0;
1678 }
1679
1680 static void sd_rescan(struct device *dev)
1681 {
1682         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1683
1684         revalidate_disk(sdkp->disk);
1685 }
1686
1687
1688 #ifdef CONFIG_COMPAT
1689 /* 
1690  * This gets directly called from VFS. When the ioctl 
1691  * is not recognized we go back to the other translation paths. 
1692  */
1693 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1694                            unsigned int cmd, unsigned long arg)
1695 {
1696         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1697         int error;
1698
1699         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1700                         (mode & FMODE_NDELAY) != 0);
1701         if (error)
1702                 return error;
1703                
1704         /* 
1705          * Let the static ioctl translation table take care of it.
1706          */
1707         if (!sdev->host->hostt->compat_ioctl)
1708                 return -ENOIOCTLCMD; 
1709         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1710 }
1711 #endif
1712
1713 static char sd_pr_type(enum pr_type type)
1714 {
1715         switch (type) {
1716         case PR_WRITE_EXCLUSIVE:
1717                 return 0x01;
1718         case PR_EXCLUSIVE_ACCESS:
1719                 return 0x03;
1720         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1721                 return 0x05;
1722         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1723                 return 0x06;
1724         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1725                 return 0x07;
1726         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1727                 return 0x08;
1728         default:
1729                 return 0;
1730         }
1731 };
1732
1733 static int sd_pr_command(struct block_device *bdev, u8 sa,
1734                 u64 key, u64 sa_key, u8 type, u8 flags)
1735 {
1736         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1737         struct scsi_sense_hdr sshdr;
1738         int result;
1739         u8 cmd[16] = { 0, };
1740         u8 data[24] = { 0, };
1741
1742         cmd[0] = PERSISTENT_RESERVE_OUT;
1743         cmd[1] = sa;
1744         cmd[2] = type;
1745         put_unaligned_be32(sizeof(data), &cmd[5]);
1746
1747         put_unaligned_be64(key, &data[0]);
1748         put_unaligned_be64(sa_key, &data[8]);
1749         data[20] = flags;
1750
1751         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1752                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1753
1754         if (driver_byte(result) == DRIVER_SENSE &&
1755             scsi_sense_valid(&sshdr)) {
1756                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1757                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1758         }
1759
1760         return result;
1761 }
1762
1763 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1764                 u32 flags)
1765 {
1766         if (flags & ~PR_FL_IGNORE_KEY)
1767                 return -EOPNOTSUPP;
1768         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1769                         old_key, new_key, 0,
1770                         (1 << 0) /* APTPL */);
1771 }
1772
1773 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1774                 u32 flags)
1775 {
1776         if (flags)
1777                 return -EOPNOTSUPP;
1778         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1779 }
1780
1781 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1782 {
1783         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1784 }
1785
1786 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1787                 enum pr_type type, bool abort)
1788 {
1789         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1790                              sd_pr_type(type), 0);
1791 }
1792
1793 static int sd_pr_clear(struct block_device *bdev, u64 key)
1794 {
1795         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1796 }
1797
1798 static const struct pr_ops sd_pr_ops = {
1799         .pr_register    = sd_pr_register,
1800         .pr_reserve     = sd_pr_reserve,
1801         .pr_release     = sd_pr_release,
1802         .pr_preempt     = sd_pr_preempt,
1803         .pr_clear       = sd_pr_clear,
1804 };
1805
1806 static const struct block_device_operations sd_fops = {
1807         .owner                  = THIS_MODULE,
1808         .open                   = sd_open,
1809         .release                = sd_release,
1810         .ioctl                  = sd_ioctl,
1811         .getgeo                 = sd_getgeo,
1812 #ifdef CONFIG_COMPAT
1813         .compat_ioctl           = sd_compat_ioctl,
1814 #endif
1815         .check_events           = sd_check_events,
1816         .revalidate_disk        = sd_revalidate_disk,
1817         .unlock_native_capacity = sd_unlock_native_capacity,
1818         .report_zones           = sd_zbc_report_zones,
1819         .pr_ops                 = &sd_pr_ops,
1820 };
1821
1822 /**
1823  *      sd_eh_reset - reset error handling callback
1824  *      @scmd:          sd-issued command that has failed
1825  *
1826  *      This function is called by the SCSI midlayer before starting
1827  *      SCSI EH. When counting medium access failures we have to be
1828  *      careful to register it only only once per device and SCSI EH run;
1829  *      there might be several timed out commands which will cause the
1830  *      'max_medium_access_timeouts' counter to trigger after the first
1831  *      SCSI EH run already and set the device to offline.
1832  *      So this function resets the internal counter before starting SCSI EH.
1833  **/
1834 static void sd_eh_reset(struct scsi_cmnd *scmd)
1835 {
1836         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1837
1838         /* New SCSI EH run, reset gate variable */
1839         sdkp->ignore_medium_access_errors = false;
1840 }
1841
1842 /**
1843  *      sd_eh_action - error handling callback
1844  *      @scmd:          sd-issued command that has failed
1845  *      @eh_disp:       The recovery disposition suggested by the midlayer
1846  *
1847  *      This function is called by the SCSI midlayer upon completion of an
1848  *      error test command (currently TEST UNIT READY). The result of sending
1849  *      the eh command is passed in eh_disp.  We're looking for devices that
1850  *      fail medium access commands but are OK with non access commands like
1851  *      test unit ready (so wrongly see the device as having a successful
1852  *      recovery)
1853  **/
1854 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1855 {
1856         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1857         struct scsi_device *sdev = scmd->device;
1858
1859         if (!scsi_device_online(sdev) ||
1860             !scsi_medium_access_command(scmd) ||
1861             host_byte(scmd->result) != DID_TIME_OUT ||
1862             eh_disp != SUCCESS)
1863                 return eh_disp;
1864
1865         /*
1866          * The device has timed out executing a medium access command.
1867          * However, the TEST UNIT READY command sent during error
1868          * handling completed successfully. Either the device is in the
1869          * process of recovering or has it suffered an internal failure
1870          * that prevents access to the storage medium.
1871          */
1872         if (!sdkp->ignore_medium_access_errors) {
1873                 sdkp->medium_access_timed_out++;
1874                 sdkp->ignore_medium_access_errors = true;
1875         }
1876
1877         /*
1878          * If the device keeps failing read/write commands but TEST UNIT
1879          * READY always completes successfully we assume that medium
1880          * access is no longer possible and take the device offline.
1881          */
1882         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1883                 scmd_printk(KERN_ERR, scmd,
1884                             "Medium access timeout failure. Offlining disk!\n");
1885                 mutex_lock(&sdev->state_mutex);
1886                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1887                 mutex_unlock(&sdev->state_mutex);
1888
1889                 return SUCCESS;
1890         }
1891
1892         return eh_disp;
1893 }
1894
1895 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1896 {
1897         struct request *req = scmd->request;
1898         struct scsi_device *sdev = scmd->device;
1899         unsigned int transferred, good_bytes;
1900         u64 start_lba, end_lba, bad_lba;
1901
1902         /*
1903          * Some commands have a payload smaller than the device logical
1904          * block size (e.g. INQUIRY on a 4K disk).
1905          */
1906         if (scsi_bufflen(scmd) <= sdev->sector_size)
1907                 return 0;
1908
1909         /* Check if we have a 'bad_lba' information */
1910         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1911                                      SCSI_SENSE_BUFFERSIZE,
1912                                      &bad_lba))
1913                 return 0;
1914
1915         /*
1916          * If the bad lba was reported incorrectly, we have no idea where
1917          * the error is.
1918          */
1919         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1920         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1921         if (bad_lba < start_lba || bad_lba >= end_lba)
1922                 return 0;
1923
1924         /*
1925          * resid is optional but mostly filled in.  When it's unused,
1926          * its value is zero, so we assume the whole buffer transferred
1927          */
1928         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1929
1930         /* This computation should always be done in terms of the
1931          * resolution of the device's medium.
1932          */
1933         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1934
1935         return min(good_bytes, transferred);
1936 }
1937
1938 /**
1939  *      sd_done - bottom half handler: called when the lower level
1940  *      driver has completed (successfully or otherwise) a scsi command.
1941  *      @SCpnt: mid-level's per command structure.
1942  *
1943  *      Note: potentially run from within an ISR. Must not block.
1944  **/
1945 static int sd_done(struct scsi_cmnd *SCpnt)
1946 {
1947         int result = SCpnt->result;
1948         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1949         unsigned int sector_size = SCpnt->device->sector_size;
1950         unsigned int resid;
1951         struct scsi_sense_hdr sshdr;
1952         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1953         struct request *req = SCpnt->request;
1954         int sense_valid = 0;
1955         int sense_deferred = 0;
1956
1957         switch (req_op(req)) {
1958         case REQ_OP_DISCARD:
1959         case REQ_OP_WRITE_ZEROES:
1960         case REQ_OP_WRITE_SAME:
1961         case REQ_OP_ZONE_RESET:
1962         case REQ_OP_ZONE_RESET_ALL:
1963                 if (!result) {
1964                         good_bytes = blk_rq_bytes(req);
1965                         scsi_set_resid(SCpnt, 0);
1966                 } else {
1967                         good_bytes = 0;
1968                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1969                 }
1970                 break;
1971         default:
1972                 /*
1973                  * In case of bogus fw or device, we could end up having
1974                  * an unaligned partial completion. Check this here and force
1975                  * alignment.
1976                  */
1977                 resid = scsi_get_resid(SCpnt);
1978                 if (resid & (sector_size - 1)) {
1979                         sd_printk(KERN_INFO, sdkp,
1980                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1981                                 resid, sector_size);
1982                         scsi_print_command(SCpnt);
1983                         resid = min(scsi_bufflen(SCpnt),
1984                                     round_up(resid, sector_size));
1985                         scsi_set_resid(SCpnt, resid);
1986                 }
1987         }
1988
1989         if (result) {
1990                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1991                 if (sense_valid)
1992                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1993         }
1994         sdkp->medium_access_timed_out = 0;
1995
1996         if (driver_byte(result) != DRIVER_SENSE &&
1997             (!sense_valid || sense_deferred))
1998                 goto out;
1999
2000         switch (sshdr.sense_key) {
2001         case HARDWARE_ERROR:
2002         case MEDIUM_ERROR:
2003                 good_bytes = sd_completed_bytes(SCpnt);
2004                 break;
2005         case RECOVERED_ERROR:
2006                 good_bytes = scsi_bufflen(SCpnt);
2007                 break;
2008         case NO_SENSE:
2009                 /* This indicates a false check condition, so ignore it.  An
2010                  * unknown amount of data was transferred so treat it as an
2011                  * error.
2012                  */
2013                 SCpnt->result = 0;
2014                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2015                 break;
2016         case ABORTED_COMMAND:
2017                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2018                         good_bytes = sd_completed_bytes(SCpnt);
2019                 break;
2020         case ILLEGAL_REQUEST:
2021                 switch (sshdr.asc) {
2022                 case 0x10:      /* DIX: Host detected corruption */
2023                         good_bytes = sd_completed_bytes(SCpnt);
2024                         break;
2025                 case 0x20:      /* INVALID COMMAND OPCODE */
2026                 case 0x24:      /* INVALID FIELD IN CDB */
2027                         switch (SCpnt->cmnd[0]) {
2028                         case UNMAP:
2029                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2030                                 break;
2031                         case WRITE_SAME_16:
2032                         case WRITE_SAME:
2033                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2034                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2035                                 } else {
2036                                         sdkp->device->no_write_same = 1;
2037                                         sd_config_write_same(sdkp);
2038                                         req->rq_flags |= RQF_QUIET;
2039                                 }
2040                                 break;
2041                         }
2042                 }
2043                 break;
2044         default:
2045                 break;
2046         }
2047
2048  out:
2049         if (sd_is_zoned(sdkp))
2050                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2051
2052         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2053                                            "sd_done: completed %d of %d bytes\n",
2054                                            good_bytes, scsi_bufflen(SCpnt)));
2055
2056         return good_bytes;
2057 }
2058
2059 /*
2060  * spinup disk - called only in sd_revalidate_disk()
2061  */
2062 static void
2063 sd_spinup_disk(struct scsi_disk *sdkp)
2064 {
2065         unsigned char cmd[10];
2066         unsigned long spintime_expire = 0;
2067         int retries, spintime;
2068         unsigned int the_result;
2069         struct scsi_sense_hdr sshdr;
2070         int sense_valid = 0;
2071
2072         spintime = 0;
2073
2074         /* Spin up drives, as required.  Only do this at boot time */
2075         /* Spinup needs to be done for module loads too. */
2076         do {
2077                 retries = 0;
2078
2079                 do {
2080                         cmd[0] = TEST_UNIT_READY;
2081                         memset((void *) &cmd[1], 0, 9);
2082
2083                         the_result = scsi_execute_req(sdkp->device, cmd,
2084                                                       DMA_NONE, NULL, 0,
2085                                                       &sshdr, SD_TIMEOUT,
2086                                                       SD_MAX_RETRIES, NULL);
2087
2088                         /*
2089                          * If the drive has indicated to us that it
2090                          * doesn't have any media in it, don't bother
2091                          * with any more polling.
2092                          */
2093                         if (media_not_present(sdkp, &sshdr))
2094                                 return;
2095
2096                         if (the_result)
2097                                 sense_valid = scsi_sense_valid(&sshdr);
2098                         retries++;
2099                 } while (retries < 3 && 
2100                          (!scsi_status_is_good(the_result) ||
2101                           ((driver_byte(the_result) == DRIVER_SENSE) &&
2102                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2103
2104                 if (driver_byte(the_result) != DRIVER_SENSE) {
2105                         /* no sense, TUR either succeeded or failed
2106                          * with a status error */
2107                         if(!spintime && !scsi_status_is_good(the_result)) {
2108                                 sd_print_result(sdkp, "Test Unit Ready failed",
2109                                                 the_result);
2110                         }
2111                         break;
2112                 }
2113
2114                 /*
2115                  * The device does not want the automatic start to be issued.
2116                  */
2117                 if (sdkp->device->no_start_on_add)
2118                         break;
2119
2120                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2121                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2122                                 break;  /* manual intervention required */
2123                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2124                                 break;  /* standby */
2125                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2126                                 break;  /* unavailable */
2127                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2128                                 break;  /* sanitize in progress */
2129                         /*
2130                          * Issue command to spin up drive when not ready
2131                          */
2132                         if (!spintime) {
2133                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2134                                 cmd[0] = START_STOP;
2135                                 cmd[1] = 1;     /* Return immediately */
2136                                 memset((void *) &cmd[2], 0, 8);
2137                                 cmd[4] = 1;     /* Start spin cycle */
2138                                 if (sdkp->device->start_stop_pwr_cond)
2139                                         cmd[4] |= 1 << 4;
2140                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2141                                                  NULL, 0, &sshdr,
2142                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2143                                                  NULL);
2144                                 spintime_expire = jiffies + 100 * HZ;
2145                                 spintime = 1;
2146                         }
2147                         /* Wait 1 second for next try */
2148                         msleep(1000);
2149                         printk(KERN_CONT ".");
2150
2151                 /*
2152                  * Wait for USB flash devices with slow firmware.
2153                  * Yes, this sense key/ASC combination shouldn't
2154                  * occur here.  It's characteristic of these devices.
2155                  */
2156                 } else if (sense_valid &&
2157                                 sshdr.sense_key == UNIT_ATTENTION &&
2158                                 sshdr.asc == 0x28) {
2159                         if (!spintime) {
2160                                 spintime_expire = jiffies + 5 * HZ;
2161                                 spintime = 1;
2162                         }
2163                         /* Wait 1 second for next try */
2164                         msleep(1000);
2165                 } else {
2166                         /* we don't understand the sense code, so it's
2167                          * probably pointless to loop */
2168                         if(!spintime) {
2169                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2170                                 sd_print_sense_hdr(sdkp, &sshdr);
2171                         }
2172                         break;
2173                 }
2174                                 
2175         } while (spintime && time_before_eq(jiffies, spintime_expire));
2176
2177         if (spintime) {
2178                 if (scsi_status_is_good(the_result))
2179                         printk(KERN_CONT "ready\n");
2180                 else
2181                         printk(KERN_CONT "not responding...\n");
2182         }
2183 }
2184
2185 /*
2186  * Determine whether disk supports Data Integrity Field.
2187  */
2188 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2189 {
2190         struct scsi_device *sdp = sdkp->device;
2191         u8 type;
2192         int ret = 0;
2193
2194         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2195                 return ret;
2196
2197         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2198
2199         if (type > T10_PI_TYPE3_PROTECTION)
2200                 ret = -ENODEV;
2201         else if (scsi_host_dif_capable(sdp->host, type))
2202                 ret = 1;
2203
2204         if (sdkp->first_scan || type != sdkp->protection_type)
2205                 switch (ret) {
2206                 case -ENODEV:
2207                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2208                                   " protection type %u. Disabling disk!\n",
2209                                   type);
2210                         break;
2211                 case 1:
2212                         sd_printk(KERN_NOTICE, sdkp,
2213                                   "Enabling DIF Type %u protection\n", type);
2214                         break;
2215                 case 0:
2216                         sd_printk(KERN_NOTICE, sdkp,
2217                                   "Disabling DIF Type %u protection\n", type);
2218                         break;
2219                 }
2220
2221         sdkp->protection_type = type;
2222
2223         return ret;
2224 }
2225
2226 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2227                         struct scsi_sense_hdr *sshdr, int sense_valid,
2228                         int the_result)
2229 {
2230         if (driver_byte(the_result) == DRIVER_SENSE)
2231                 sd_print_sense_hdr(sdkp, sshdr);
2232         else
2233                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2234
2235         /*
2236          * Set dirty bit for removable devices if not ready -
2237          * sometimes drives will not report this properly.
2238          */
2239         if (sdp->removable &&
2240             sense_valid && sshdr->sense_key == NOT_READY)
2241                 set_media_not_present(sdkp);
2242
2243         /*
2244          * We used to set media_present to 0 here to indicate no media
2245          * in the drive, but some drives fail read capacity even with
2246          * media present, so we can't do that.
2247          */
2248         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2249 }
2250
2251 #define RC16_LEN 32
2252 #if RC16_LEN > SD_BUF_SIZE
2253 #error RC16_LEN must not be more than SD_BUF_SIZE
2254 #endif
2255
2256 #define READ_CAPACITY_RETRIES_ON_RESET  10
2257
2258 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2259                                                 unsigned char *buffer)
2260 {
2261         unsigned char cmd[16];
2262         struct scsi_sense_hdr sshdr;
2263         int sense_valid = 0;
2264         int the_result;
2265         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2266         unsigned int alignment;
2267         unsigned long long lba;
2268         unsigned sector_size;
2269
2270         if (sdp->no_read_capacity_16)
2271                 return -EINVAL;
2272
2273         do {
2274                 memset(cmd, 0, 16);
2275                 cmd[0] = SERVICE_ACTION_IN_16;
2276                 cmd[1] = SAI_READ_CAPACITY_16;
2277                 cmd[13] = RC16_LEN;
2278                 memset(buffer, 0, RC16_LEN);
2279
2280                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2281                                         buffer, RC16_LEN, &sshdr,
2282                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2283
2284                 if (media_not_present(sdkp, &sshdr))
2285                         return -ENODEV;
2286
2287                 if (the_result) {
2288                         sense_valid = scsi_sense_valid(&sshdr);
2289                         if (sense_valid &&
2290                             sshdr.sense_key == ILLEGAL_REQUEST &&
2291                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2292                             sshdr.ascq == 0x00)
2293                                 /* Invalid Command Operation Code or
2294                                  * Invalid Field in CDB, just retry
2295                                  * silently with RC10 */
2296                                 return -EINVAL;
2297                         if (sense_valid &&
2298                             sshdr.sense_key == UNIT_ATTENTION &&
2299                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2300                                 /* Device reset might occur several times,
2301                                  * give it one more chance */
2302                                 if (--reset_retries > 0)
2303                                         continue;
2304                 }
2305                 retries--;
2306
2307         } while (the_result && retries);
2308
2309         if (the_result) {
2310                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2311                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2312                 return -EINVAL;
2313         }
2314
2315         sector_size = get_unaligned_be32(&buffer[8]);
2316         lba = get_unaligned_be64(&buffer[0]);
2317
2318         if (sd_read_protection_type(sdkp, buffer) < 0) {
2319                 sdkp->capacity = 0;
2320                 return -ENODEV;
2321         }
2322
2323         /* Logical blocks per physical block exponent */
2324         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2325
2326         /* RC basis */
2327         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2328
2329         /* Lowest aligned logical block */
2330         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2331         blk_queue_alignment_offset(sdp->request_queue, alignment);
2332         if (alignment && sdkp->first_scan)
2333                 sd_printk(KERN_NOTICE, sdkp,
2334                           "physical block alignment offset: %u\n", alignment);
2335
2336         if (buffer[14] & 0x80) { /* LBPME */
2337                 sdkp->lbpme = 1;
2338
2339                 if (buffer[14] & 0x40) /* LBPRZ */
2340                         sdkp->lbprz = 1;
2341
2342                 sd_config_discard(sdkp, SD_LBP_WS16);
2343         }
2344
2345         sdkp->capacity = lba + 1;
2346         return sector_size;
2347 }
2348
2349 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2350                                                 unsigned char *buffer)
2351 {
2352         unsigned char cmd[16];
2353         struct scsi_sense_hdr sshdr;
2354         int sense_valid = 0;
2355         int the_result;
2356         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2357         sector_t lba;
2358         unsigned sector_size;
2359
2360         do {
2361                 cmd[0] = READ_CAPACITY;
2362                 memset(&cmd[1], 0, 9);
2363                 memset(buffer, 0, 8);
2364
2365                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2366                                         buffer, 8, &sshdr,
2367                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2368
2369                 if (media_not_present(sdkp, &sshdr))
2370                         return -ENODEV;
2371
2372                 if (the_result) {
2373                         sense_valid = scsi_sense_valid(&sshdr);
2374                         if (sense_valid &&
2375                             sshdr.sense_key == UNIT_ATTENTION &&
2376                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2377                                 /* Device reset might occur several times,
2378                                  * give it one more chance */
2379                                 if (--reset_retries > 0)
2380                                         continue;
2381                 }
2382                 retries--;
2383
2384         } while (the_result && retries);
2385
2386         if (the_result) {
2387                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2388                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2389                 return -EINVAL;
2390         }
2391
2392         sector_size = get_unaligned_be32(&buffer[4]);
2393         lba = get_unaligned_be32(&buffer[0]);
2394
2395         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2396                 /* Some buggy (usb cardreader) devices return an lba of
2397                    0xffffffff when the want to report a size of 0 (with
2398                    which they really mean no media is present) */
2399                 sdkp->capacity = 0;
2400                 sdkp->physical_block_size = sector_size;
2401                 return sector_size;
2402         }
2403
2404         sdkp->capacity = lba + 1;
2405         sdkp->physical_block_size = sector_size;
2406         return sector_size;
2407 }
2408
2409 static int sd_try_rc16_first(struct scsi_device *sdp)
2410 {
2411         if (sdp->host->max_cmd_len < 16)
2412                 return 0;
2413         if (sdp->try_rc_10_first)
2414                 return 0;
2415         if (sdp->scsi_level > SCSI_SPC_2)
2416                 return 1;
2417         if (scsi_device_protection(sdp))
2418                 return 1;
2419         return 0;
2420 }
2421
2422 /*
2423  * read disk capacity
2424  */
2425 static void
2426 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2427 {
2428         int sector_size;
2429         struct scsi_device *sdp = sdkp->device;
2430
2431         if (sd_try_rc16_first(sdp)) {
2432                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2433                 if (sector_size == -EOVERFLOW)
2434                         goto got_data;
2435                 if (sector_size == -ENODEV)
2436                         return;
2437                 if (sector_size < 0)
2438                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2439                 if (sector_size < 0)
2440                         return;
2441         } else {
2442                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2443                 if (sector_size == -EOVERFLOW)
2444                         goto got_data;
2445                 if (sector_size < 0)
2446                         return;
2447                 if ((sizeof(sdkp->capacity) > 4) &&
2448                     (sdkp->capacity > 0xffffffffULL)) {
2449                         int old_sector_size = sector_size;
2450                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2451                                         "Trying to use READ CAPACITY(16).\n");
2452                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2453                         if (sector_size < 0) {
2454                                 sd_printk(KERN_NOTICE, sdkp,
2455                                         "Using 0xffffffff as device size\n");
2456                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2457                                 sector_size = old_sector_size;
2458                                 goto got_data;
2459                         }
2460                         /* Remember that READ CAPACITY(16) succeeded */
2461                         sdp->try_rc_10_first = 0;
2462                 }
2463         }
2464
2465         /* Some devices are known to return the total number of blocks,
2466          * not the highest block number.  Some devices have versions
2467          * which do this and others which do not.  Some devices we might
2468          * suspect of doing this but we don't know for certain.
2469          *
2470          * If we know the reported capacity is wrong, decrement it.  If
2471          * we can only guess, then assume the number of blocks is even
2472          * (usually true but not always) and err on the side of lowering
2473          * the capacity.
2474          */
2475         if (sdp->fix_capacity ||
2476             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2477                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2478                                 "from its reported value: %llu\n",
2479                                 (unsigned long long) sdkp->capacity);
2480                 --sdkp->capacity;
2481         }
2482
2483 got_data:
2484         if (sector_size == 0) {
2485                 sector_size = 512;
2486                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2487                           "assuming 512.\n");
2488         }
2489
2490         if (sector_size != 512 &&
2491             sector_size != 1024 &&
2492             sector_size != 2048 &&
2493             sector_size != 4096) {
2494                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2495                           sector_size);
2496                 /*
2497                  * The user might want to re-format the drive with
2498                  * a supported sectorsize.  Once this happens, it
2499                  * would be relatively trivial to set the thing up.
2500                  * For this reason, we leave the thing in the table.
2501                  */
2502                 sdkp->capacity = 0;
2503                 /*
2504                  * set a bogus sector size so the normal read/write
2505                  * logic in the block layer will eventually refuse any
2506                  * request on this device without tripping over power
2507                  * of two sector size assumptions
2508                  */
2509                 sector_size = 512;
2510         }
2511         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2512         blk_queue_physical_block_size(sdp->request_queue,
2513                                       sdkp->physical_block_size);
2514         sdkp->device->sector_size = sector_size;
2515
2516         if (sdkp->capacity > 0xffffffff)
2517                 sdp->use_16_for_rw = 1;
2518
2519 }
2520
2521 /*
2522  * Print disk capacity
2523  */
2524 static void
2525 sd_print_capacity(struct scsi_disk *sdkp,
2526                   sector_t old_capacity)
2527 {
2528         int sector_size = sdkp->device->sector_size;
2529         char cap_str_2[10], cap_str_10[10];
2530
2531         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2532                 return;
2533
2534         string_get_size(sdkp->capacity, sector_size,
2535                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2536         string_get_size(sdkp->capacity, sector_size,
2537                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2538
2539         sd_printk(KERN_NOTICE, sdkp,
2540                   "%llu %d-byte logical blocks: (%s/%s)\n",
2541                   (unsigned long long)sdkp->capacity,
2542                   sector_size, cap_str_10, cap_str_2);
2543
2544         if (sdkp->physical_block_size != sector_size)
2545                 sd_printk(KERN_NOTICE, sdkp,
2546                           "%u-byte physical blocks\n",
2547                           sdkp->physical_block_size);
2548
2549         sd_zbc_print_zones(sdkp);
2550 }
2551
2552 /* called with buffer of length 512 */
2553 static inline int
2554 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2555                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2556                  struct scsi_sense_hdr *sshdr)
2557 {
2558         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2559                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2560                                sshdr);
2561 }
2562
2563 /*
2564  * read write protect setting, if possible - called only in sd_revalidate_disk()
2565  * called with buffer of length SD_BUF_SIZE
2566  */
2567 static void
2568 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2569 {
2570         int res;
2571         struct scsi_device *sdp = sdkp->device;
2572         struct scsi_mode_data data;
2573         int old_wp = sdkp->write_prot;
2574
2575         set_disk_ro(sdkp->disk, 0);
2576         if (sdp->skip_ms_page_3f) {
2577                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2578                 return;
2579         }
2580
2581         if (sdp->use_192_bytes_for_3f) {
2582                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2583         } else {
2584                 /*
2585                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2586                  * We have to start carefully: some devices hang if we ask
2587                  * for more than is available.
2588                  */
2589                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2590
2591                 /*
2592                  * Second attempt: ask for page 0 When only page 0 is
2593                  * implemented, a request for page 3F may return Sense Key
2594                  * 5: Illegal Request, Sense Code 24: Invalid field in
2595                  * CDB.
2596                  */
2597                 if (!scsi_status_is_good(res))
2598                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2599
2600                 /*
2601                  * Third attempt: ask 255 bytes, as we did earlier.
2602                  */
2603                 if (!scsi_status_is_good(res))
2604                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2605                                                &data, NULL);
2606         }
2607
2608         if (!scsi_status_is_good(res)) {
2609                 sd_first_printk(KERN_WARNING, sdkp,
2610                           "Test WP failed, assume Write Enabled\n");
2611         } else {
2612                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2613                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2614                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2615                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2616                                   sdkp->write_prot ? "on" : "off");
2617                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2618                 }
2619         }
2620 }
2621
2622 /*
2623  * sd_read_cache_type - called only from sd_revalidate_disk()
2624  * called with buffer of length SD_BUF_SIZE
2625  */
2626 static void
2627 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2628 {
2629         int len = 0, res;
2630         struct scsi_device *sdp = sdkp->device;
2631
2632         int dbd;
2633         int modepage;
2634         int first_len;
2635         struct scsi_mode_data data;
2636         struct scsi_sense_hdr sshdr;
2637         int old_wce = sdkp->WCE;
2638         int old_rcd = sdkp->RCD;
2639         int old_dpofua = sdkp->DPOFUA;
2640
2641
2642         if (sdkp->cache_override)
2643                 return;
2644
2645         first_len = 4;
2646         if (sdp->skip_ms_page_8) {
2647                 if (sdp->type == TYPE_RBC)
2648                         goto defaults;
2649                 else {
2650                         if (sdp->skip_ms_page_3f)
2651                                 goto defaults;
2652                         modepage = 0x3F;
2653                         if (sdp->use_192_bytes_for_3f)
2654                                 first_len = 192;
2655                         dbd = 0;
2656                 }
2657         } else if (sdp->type == TYPE_RBC) {
2658                 modepage = 6;
2659                 dbd = 8;
2660         } else {
2661                 modepage = 8;
2662                 dbd = 0;
2663         }
2664
2665         /* cautiously ask */
2666         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2667                         &data, &sshdr);
2668
2669         if (!scsi_status_is_good(res))
2670                 goto bad_sense;
2671
2672         if (!data.header_length) {
2673                 modepage = 6;
2674                 first_len = 0;
2675                 sd_first_printk(KERN_ERR, sdkp,
2676                                 "Missing header in MODE_SENSE response\n");
2677         }
2678
2679         /* that went OK, now ask for the proper length */
2680         len = data.length;
2681
2682         /*
2683          * We're only interested in the first three bytes, actually.
2684          * But the data cache page is defined for the first 20.
2685          */
2686         if (len < 3)
2687                 goto bad_sense;
2688         else if (len > SD_BUF_SIZE) {
2689                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2690                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2691                 len = SD_BUF_SIZE;
2692         }
2693         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2694                 len = 192;
2695
2696         /* Get the data */
2697         if (len > first_len)
2698                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2699                                 &data, &sshdr);
2700
2701         if (scsi_status_is_good(res)) {
2702                 int offset = data.header_length + data.block_descriptor_length;
2703
2704                 while (offset < len) {
2705                         u8 page_code = buffer[offset] & 0x3F;
2706                         u8 spf       = buffer[offset] & 0x40;
2707
2708                         if (page_code == 8 || page_code == 6) {
2709                                 /* We're interested only in the first 3 bytes.
2710                                  */
2711                                 if (len - offset <= 2) {
2712                                         sd_first_printk(KERN_ERR, sdkp,
2713                                                 "Incomplete mode parameter "
2714                                                         "data\n");
2715                                         goto defaults;
2716                                 } else {
2717                                         modepage = page_code;
2718                                         goto Page_found;
2719                                 }
2720                         } else {
2721                                 /* Go to the next page */
2722                                 if (spf && len - offset > 3)
2723                                         offset += 4 + (buffer[offset+2] << 8) +
2724                                                 buffer[offset+3];
2725                                 else if (!spf && len - offset > 1)
2726                                         offset += 2 + buffer[offset+1];
2727                                 else {
2728                                         sd_first_printk(KERN_ERR, sdkp,
2729                                                         "Incomplete mode "
2730                                                         "parameter data\n");
2731                                         goto defaults;
2732                                 }
2733                         }
2734                 }
2735
2736                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2737                 goto defaults;
2738
2739         Page_found:
2740                 if (modepage == 8) {
2741                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2742                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2743                 } else {
2744                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2745                         sdkp->RCD = 0;
2746                 }
2747
2748                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2749                 if (sdp->broken_fua) {
2750                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2751                         sdkp->DPOFUA = 0;
2752                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2753                            !sdkp->device->use_16_for_rw) {
2754                         sd_first_printk(KERN_NOTICE, sdkp,
2755                                   "Uses READ/WRITE(6), disabling FUA\n");
2756                         sdkp->DPOFUA = 0;
2757                 }
2758
2759                 /* No cache flush allowed for write protected devices */
2760                 if (sdkp->WCE && sdkp->write_prot)
2761                         sdkp->WCE = 0;
2762
2763                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2764                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2765                         sd_printk(KERN_NOTICE, sdkp,
2766                                   "Write cache: %s, read cache: %s, %s\n",
2767                                   sdkp->WCE ? "enabled" : "disabled",
2768                                   sdkp->RCD ? "disabled" : "enabled",
2769                                   sdkp->DPOFUA ? "supports DPO and FUA"
2770                                   : "doesn't support DPO or FUA");
2771
2772                 return;
2773         }
2774
2775 bad_sense:
2776         if (scsi_sense_valid(&sshdr) &&
2777             sshdr.sense_key == ILLEGAL_REQUEST &&
2778             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2779                 /* Invalid field in CDB */
2780                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2781         else
2782                 sd_first_printk(KERN_ERR, sdkp,
2783                                 "Asking for cache data failed\n");
2784
2785 defaults:
2786         if (sdp->wce_default_on) {
2787                 sd_first_printk(KERN_NOTICE, sdkp,
2788                                 "Assuming drive cache: write back\n");
2789                 sdkp->WCE = 1;
2790         } else {
2791                 sd_first_printk(KERN_ERR, sdkp,
2792                                 "Assuming drive cache: write through\n");
2793                 sdkp->WCE = 0;
2794         }
2795         sdkp->RCD = 0;
2796         sdkp->DPOFUA = 0;
2797 }
2798
2799 /*
2800  * The ATO bit indicates whether the DIF application tag is available
2801  * for use by the operating system.
2802  */
2803 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2804 {
2805         int res, offset;
2806         struct scsi_device *sdp = sdkp->device;
2807         struct scsi_mode_data data;
2808         struct scsi_sense_hdr sshdr;
2809
2810         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2811                 return;
2812
2813         if (sdkp->protection_type == 0)
2814                 return;
2815
2816         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2817                               SD_MAX_RETRIES, &data, &sshdr);
2818
2819         if (!scsi_status_is_good(res) || !data.header_length ||
2820             data.length < 6) {
2821                 sd_first_printk(KERN_WARNING, sdkp,
2822                           "getting Control mode page failed, assume no ATO\n");
2823
2824                 if (scsi_sense_valid(&sshdr))
2825                         sd_print_sense_hdr(sdkp, &sshdr);
2826
2827                 return;
2828         }
2829
2830         offset = data.header_length + data.block_descriptor_length;
2831
2832         if ((buffer[offset] & 0x3f) != 0x0a) {
2833                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2834                 return;
2835         }
2836
2837         if ((buffer[offset + 5] & 0x80) == 0)
2838                 return;
2839
2840         sdkp->ATO = 1;
2841
2842         return;
2843 }
2844
2845 /**
2846  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2847  * @sdkp: disk to query
2848  */
2849 static void sd_read_block_limits(struct scsi_disk *sdkp)
2850 {
2851         unsigned int sector_sz = sdkp->device->sector_size;
2852         const int vpd_len = 64;
2853         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2854
2855         if (!buffer ||
2856             /* Block Limits VPD */
2857             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2858                 goto out;
2859
2860         blk_queue_io_min(sdkp->disk->queue,
2861                          get_unaligned_be16(&buffer[6]) * sector_sz);
2862
2863         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2864         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2865
2866         if (buffer[3] == 0x3c) {
2867                 unsigned int lba_count, desc_count;
2868
2869                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2870
2871                 if (!sdkp->lbpme)
2872                         goto out;
2873
2874                 lba_count = get_unaligned_be32(&buffer[20]);
2875                 desc_count = get_unaligned_be32(&buffer[24]);
2876
2877                 if (lba_count && desc_count)
2878                         sdkp->max_unmap_blocks = lba_count;
2879
2880                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2881
2882                 if (buffer[32] & 0x80)
2883                         sdkp->unmap_alignment =
2884                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2885
2886                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2887
2888                         if (sdkp->max_unmap_blocks)
2889                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2890                         else
2891                                 sd_config_discard(sdkp, SD_LBP_WS16);
2892
2893                 } else {        /* LBP VPD page tells us what to use */
2894                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2895                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2896                         else if (sdkp->lbpws)
2897                                 sd_config_discard(sdkp, SD_LBP_WS16);
2898                         else if (sdkp->lbpws10)
2899                                 sd_config_discard(sdkp, SD_LBP_WS10);
2900                         else
2901                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2902                 }
2903         }
2904
2905  out:
2906         kfree(buffer);
2907 }
2908
2909 /**
2910  * sd_read_block_characteristics - Query block dev. characteristics
2911  * @sdkp: disk to query
2912  */
2913 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2914 {
2915         struct request_queue *q = sdkp->disk->queue;
2916         unsigned char *buffer;
2917         u16 rot;
2918         const int vpd_len = 64;
2919
2920         buffer = kmalloc(vpd_len, GFP_KERNEL);
2921
2922         if (!buffer ||
2923             /* Block Device Characteristics VPD */
2924             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2925                 goto out;
2926
2927         rot = get_unaligned_be16(&buffer[4]);
2928
2929         if (rot == 1) {
2930                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2931                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2932         }
2933
2934         if (sdkp->device->type == TYPE_ZBC) {
2935                 /* Host-managed */
2936                 q->limits.zoned = BLK_ZONED_HM;
2937         } else {
2938                 sdkp->zoned = (buffer[8] >> 4) & 3;
2939                 if (sdkp->zoned == 1)
2940                         /* Host-aware */
2941                         q->limits.zoned = BLK_ZONED_HA;
2942                 else
2943                         /*
2944                          * Treat drive-managed devices as
2945                          * regular block devices.
2946                          */
2947                         q->limits.zoned = BLK_ZONED_NONE;
2948         }
2949         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2950                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2951                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2952
2953  out:
2954         kfree(buffer);
2955 }
2956
2957 /**
2958  * sd_read_block_provisioning - Query provisioning VPD page
2959  * @sdkp: disk to query
2960  */
2961 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2962 {
2963         unsigned char *buffer;
2964         const int vpd_len = 8;
2965
2966         if (sdkp->lbpme == 0)
2967                 return;
2968
2969         buffer = kmalloc(vpd_len, GFP_KERNEL);
2970
2971         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2972                 goto out;
2973
2974         sdkp->lbpvpd    = 1;
2975         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
2976         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2977         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2978
2979  out:
2980         kfree(buffer);
2981 }
2982
2983 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2984 {
2985         struct scsi_device *sdev = sdkp->device;
2986
2987         if (sdev->host->no_write_same) {
2988                 sdev->no_write_same = 1;
2989
2990                 return;
2991         }
2992
2993         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2994                 /* too large values might cause issues with arcmsr */
2995                 int vpd_buf_len = 64;
2996
2997                 sdev->no_report_opcodes = 1;
2998
2999                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3000                  * CODES is unsupported and the device has an ATA
3001                  * Information VPD page (SAT).
3002                  */
3003                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3004                         sdev->no_write_same = 1;
3005         }
3006
3007         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3008                 sdkp->ws16 = 1;
3009
3010         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3011                 sdkp->ws10 = 1;
3012 }
3013
3014 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3015 {
3016         struct scsi_device *sdev = sdkp->device;
3017
3018         if (!sdev->security_supported)
3019                 return;
3020
3021         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3022                         SECURITY_PROTOCOL_IN) == 1 &&
3023             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3024                         SECURITY_PROTOCOL_OUT) == 1)
3025                 sdkp->security = 1;
3026 }
3027
3028 /*
3029  * Determine the device's preferred I/O size for reads and writes
3030  * unless the reported value is unreasonably small, large, not a
3031  * multiple of the physical block size, or simply garbage.
3032  */
3033 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3034                                       unsigned int dev_max)
3035 {
3036         struct scsi_device *sdp = sdkp->device;
3037         unsigned int opt_xfer_bytes =
3038                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3039
3040         if (sdkp->opt_xfer_blocks == 0)
3041                 return false;
3042
3043         if (sdkp->opt_xfer_blocks > dev_max) {
3044                 sd_first_printk(KERN_WARNING, sdkp,
3045                                 "Optimal transfer size %u logical blocks " \
3046                                 "> dev_max (%u logical blocks)\n",
3047                                 sdkp->opt_xfer_blocks, dev_max);
3048                 return false;
3049         }
3050
3051         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3052                 sd_first_printk(KERN_WARNING, sdkp,
3053                                 "Optimal transfer size %u logical blocks " \
3054                                 "> sd driver limit (%u logical blocks)\n",
3055                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3056                 return false;
3057         }
3058
3059         if (opt_xfer_bytes < PAGE_SIZE) {
3060                 sd_first_printk(KERN_WARNING, sdkp,
3061                                 "Optimal transfer size %u bytes < " \
3062                                 "PAGE_SIZE (%u bytes)\n",
3063                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3064                 return false;
3065         }
3066
3067         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3068                 sd_first_printk(KERN_WARNING, sdkp,
3069                                 "Optimal transfer size %u bytes not a " \
3070                                 "multiple of physical block size (%u bytes)\n",
3071                                 opt_xfer_bytes, sdkp->physical_block_size);
3072                 return false;
3073         }
3074
3075         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3076                         opt_xfer_bytes);
3077         return true;
3078 }
3079
3080 /**
3081  *      sd_revalidate_disk - called the first time a new disk is seen,
3082  *      performs disk spin up, read_capacity, etc.
3083  *      @disk: struct gendisk we care about
3084  **/
3085 static int sd_revalidate_disk(struct gendisk *disk)
3086 {
3087         struct scsi_disk *sdkp = scsi_disk(disk);
3088         struct scsi_device *sdp = sdkp->device;
3089         struct request_queue *q = sdkp->disk->queue;
3090         sector_t old_capacity = sdkp->capacity;
3091         unsigned char *buffer;
3092         unsigned int dev_max, rw_max;
3093
3094         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3095                                       "sd_revalidate_disk\n"));
3096
3097         /*
3098          * If the device is offline, don't try and read capacity or any
3099          * of the other niceties.
3100          */
3101         if (!scsi_device_online(sdp))
3102                 goto out;
3103
3104         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3105         if (!buffer) {
3106                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3107                           "allocation failure.\n");
3108                 goto out;
3109         }
3110
3111         sd_spinup_disk(sdkp);
3112
3113         /*
3114          * Without media there is no reason to ask; moreover, some devices
3115          * react badly if we do.
3116          */
3117         if (sdkp->media_present) {
3118                 sd_read_capacity(sdkp, buffer);
3119
3120                 /*
3121                  * set the default to rotational.  All non-rotational devices
3122                  * support the block characteristics VPD page, which will
3123                  * cause this to be updated correctly and any device which
3124                  * doesn't support it should be treated as rotational.
3125                  */
3126                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3127                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3128
3129                 if (scsi_device_supports_vpd(sdp)) {
3130                         sd_read_block_provisioning(sdkp);
3131                         sd_read_block_limits(sdkp);
3132                         sd_read_block_characteristics(sdkp);
3133                         sd_zbc_read_zones(sdkp, buffer);
3134                 }
3135
3136                 sd_print_capacity(sdkp, old_capacity);
3137
3138                 sd_read_write_protect_flag(sdkp, buffer);
3139                 sd_read_cache_type(sdkp, buffer);
3140                 sd_read_app_tag_own(sdkp, buffer);
3141                 sd_read_write_same(sdkp, buffer);
3142                 sd_read_security(sdkp, buffer);
3143         }
3144
3145         /*
3146          * We now have all cache related info, determine how we deal
3147          * with flush requests.
3148          */
3149         sd_set_flush_flag(sdkp);
3150
3151         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3152         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3153
3154         /* Some devices report a maximum block count for READ/WRITE requests. */
3155         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3156         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3157
3158         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3159                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3160                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3161         } else
3162                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3163                                       (sector_t)BLK_DEF_MAX_SECTORS);
3164
3165         /* Do not exceed controller limit */
3166         rw_max = min(rw_max, queue_max_hw_sectors(q));
3167
3168         /*
3169          * Only update max_sectors if previously unset or if the current value
3170          * exceeds the capabilities of the hardware.
3171          */
3172         if (sdkp->first_scan ||
3173             q->limits.max_sectors > q->limits.max_dev_sectors ||
3174             q->limits.max_sectors > q->limits.max_hw_sectors)
3175                 q->limits.max_sectors = rw_max;
3176
3177         sdkp->first_scan = 0;
3178
3179         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3180         sd_config_write_same(sdkp);
3181         kfree(buffer);
3182
3183  out:
3184         return 0;
3185 }
3186
3187 /**
3188  *      sd_unlock_native_capacity - unlock native capacity
3189  *      @disk: struct gendisk to set capacity for
3190  *
3191  *      Block layer calls this function if it detects that partitions
3192  *      on @disk reach beyond the end of the device.  If the SCSI host
3193  *      implements ->unlock_native_capacity() method, it's invoked to
3194  *      give it a chance to adjust the device capacity.
3195  *
3196  *      CONTEXT:
3197  *      Defined by block layer.  Might sleep.
3198  */
3199 static void sd_unlock_native_capacity(struct gendisk *disk)
3200 {
3201         struct scsi_device *sdev = scsi_disk(disk)->device;
3202
3203         if (sdev->host->hostt->unlock_native_capacity)
3204                 sdev->host->hostt->unlock_native_capacity(sdev);
3205 }
3206
3207 /**
3208  *      sd_format_disk_name - format disk name
3209  *      @prefix: name prefix - ie. "sd" for SCSI disks
3210  *      @index: index of the disk to format name for
3211  *      @buf: output buffer
3212  *      @buflen: length of the output buffer
3213  *
3214  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3215  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3216  *      which is followed by sdaaa.
3217  *
3218  *      This is basically 26 base counting with one extra 'nil' entry
3219  *      at the beginning from the second digit on and can be
3220  *      determined using similar method as 26 base conversion with the
3221  *      index shifted -1 after each digit is computed.
3222  *
3223  *      CONTEXT:
3224  *      Don't care.
3225  *
3226  *      RETURNS:
3227  *      0 on success, -errno on failure.
3228  */
3229 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3230 {
3231         const int base = 'z' - 'a' + 1;
3232         char *begin = buf + strlen(prefix);
3233         char *end = buf + buflen;
3234         char *p;
3235         int unit;
3236
3237         p = end - 1;
3238         *p = '\0';
3239         unit = base;