scsi: sd: Keep disk read-only when re-reading partition
[sfrench/cifs-2.6.git] / drivers / scsi / sd.c
1 /*
2  *      sd.c Copyright (C) 1992 Drew Eckhardt
3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4  *
5  *      Linux scsi disk driver
6  *              Initial versions: Drew Eckhardt
7  *              Subsequent revisions: Eric Youngdale
8  *      Modification history:
9  *       - Drew Eckhardt <drew@colorado.edu> original
10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple 
11  *         outstanding request, and other enhancements.
12  *         Support loadable low-level scsi drivers.
13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using 
14  *         eight major numbers.
15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in 
17  *         sd_init and cleanups.
18  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
19  *         not being read in sd_open. Fix problem where removable media 
20  *         could be ejected after sd_open.
21  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox 
23  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>: 
24  *         Support 32k/1M disks.
25  *
26  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
27  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
31  *      Note: when the logging level is set by the user, it must be greater
32  *      than the level indicated above to trigger output.       
33  */
34
35 #include <linux/module.h>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/sed-opal.h>
54 #include <linux/pm_runtime.h>
55 #include <linux/pr.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
59
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
69
70 #include "sd.h"
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
73
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
77
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
98
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
100 #define SD_MINORS       16
101 #else
102 #define SD_MINORS       0
103 #endif
104
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int  sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int  sd_probe(struct device *);
110 static int  sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
125
126 static DEFINE_SPINLOCK(sd_index_lock);
127 static DEFINE_IDA(sd_index_ida);
128
129 /* This semaphore is used to mediate the 0->1 reference get in the
130  * face of object destruction (i.e. we can't allow a get on an
131  * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
133
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
136
137 static const char *sd_cache_types[] = {
138         "write through", "none", "write back",
139         "write back, no read (daft)"
140 };
141
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
143 {
144         bool wc = false, fua = false;
145
146         if (sdkp->WCE) {
147                 wc = true;
148                 if (sdkp->DPOFUA)
149                         fua = true;
150         }
151
152         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
153 }
154
155 static ssize_t
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157                  const char *buf, size_t count)
158 {
159         int ct, rcd, wce, sp;
160         struct scsi_disk *sdkp = to_scsi_disk(dev);
161         struct scsi_device *sdp = sdkp->device;
162         char buffer[64];
163         char *buffer_data;
164         struct scsi_mode_data data;
165         struct scsi_sense_hdr sshdr;
166         static const char temp[] = "temporary ";
167         int len;
168
169         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170                 /* no cache control on RBC devices; theoretically they
171                  * can do it, but there's probably so many exceptions
172                  * it's not worth the risk */
173                 return -EINVAL;
174
175         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176                 buf += sizeof(temp) - 1;
177                 sdkp->cache_override = 1;
178         } else {
179                 sdkp->cache_override = 0;
180         }
181
182         ct = sysfs_match_string(sd_cache_types, buf);
183         if (ct < 0)
184                 return -EINVAL;
185
186         rcd = ct & 0x01 ? 1 : 0;
187         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
188
189         if (sdkp->cache_override) {
190                 sdkp->WCE = wce;
191                 sdkp->RCD = rcd;
192                 sd_set_flush_flag(sdkp);
193                 return count;
194         }
195
196         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197                             SD_MAX_RETRIES, &data, NULL))
198                 return -EINVAL;
199         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200                   data.block_descriptor_length);
201         buffer_data = buffer + data.header_length +
202                 data.block_descriptor_length;
203         buffer_data[2] &= ~0x05;
204         buffer_data[2] |= wce << 2 | rcd;
205         sp = buffer_data[0] & 0x80 ? 1 : 0;
206         buffer_data[0] &= ~0x80;
207
208         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209                              SD_MAX_RETRIES, &data, &sshdr)) {
210                 if (scsi_sense_valid(&sshdr))
211                         sd_print_sense_hdr(sdkp, &sshdr);
212                 return -EINVAL;
213         }
214         revalidate_disk(sdkp->disk);
215         return count;
216 }
217
218 static ssize_t
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
220                        char *buf)
221 {
222         struct scsi_disk *sdkp = to_scsi_disk(dev);
223         struct scsi_device *sdp = sdkp->device;
224
225         return sprintf(buf, "%u\n", sdp->manage_start_stop);
226 }
227
228 static ssize_t
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230                         const char *buf, size_t count)
231 {
232         struct scsi_disk *sdkp = to_scsi_disk(dev);
233         struct scsi_device *sdp = sdkp->device;
234         bool v;
235
236         if (!capable(CAP_SYS_ADMIN))
237                 return -EACCES;
238
239         if (kstrtobool(buf, &v))
240                 return -EINVAL;
241
242         sdp->manage_start_stop = v;
243
244         return count;
245 }
246 static DEVICE_ATTR_RW(manage_start_stop);
247
248 static ssize_t
249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
250 {
251         struct scsi_disk *sdkp = to_scsi_disk(dev);
252
253         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
254 }
255
256 static ssize_t
257 allow_restart_store(struct device *dev, struct device_attribute *attr,
258                     const char *buf, size_t count)
259 {
260         bool v;
261         struct scsi_disk *sdkp = to_scsi_disk(dev);
262         struct scsi_device *sdp = sdkp->device;
263
264         if (!capable(CAP_SYS_ADMIN))
265                 return -EACCES;
266
267         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
268                 return -EINVAL;
269
270         if (kstrtobool(buf, &v))
271                 return -EINVAL;
272
273         sdp->allow_restart = v;
274
275         return count;
276 }
277 static DEVICE_ATTR_RW(allow_restart);
278
279 static ssize_t
280 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
281 {
282         struct scsi_disk *sdkp = to_scsi_disk(dev);
283         int ct = sdkp->RCD + 2*sdkp->WCE;
284
285         return sprintf(buf, "%s\n", sd_cache_types[ct]);
286 }
287 static DEVICE_ATTR_RW(cache_type);
288
289 static ssize_t
290 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
291 {
292         struct scsi_disk *sdkp = to_scsi_disk(dev);
293
294         return sprintf(buf, "%u\n", sdkp->DPOFUA);
295 }
296 static DEVICE_ATTR_RO(FUA);
297
298 static ssize_t
299 protection_type_show(struct device *dev, struct device_attribute *attr,
300                      char *buf)
301 {
302         struct scsi_disk *sdkp = to_scsi_disk(dev);
303
304         return sprintf(buf, "%u\n", sdkp->protection_type);
305 }
306
307 static ssize_t
308 protection_type_store(struct device *dev, struct device_attribute *attr,
309                       const char *buf, size_t count)
310 {
311         struct scsi_disk *sdkp = to_scsi_disk(dev);
312         unsigned int val;
313         int err;
314
315         if (!capable(CAP_SYS_ADMIN))
316                 return -EACCES;
317
318         err = kstrtouint(buf, 10, &val);
319
320         if (err)
321                 return err;
322
323         if (val <= T10_PI_TYPE3_PROTECTION)
324                 sdkp->protection_type = val;
325
326         return count;
327 }
328 static DEVICE_ATTR_RW(protection_type);
329
330 static ssize_t
331 protection_mode_show(struct device *dev, struct device_attribute *attr,
332                      char *buf)
333 {
334         struct scsi_disk *sdkp = to_scsi_disk(dev);
335         struct scsi_device *sdp = sdkp->device;
336         unsigned int dif, dix;
337
338         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
339         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
340
341         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
342                 dif = 0;
343                 dix = 1;
344         }
345
346         if (!dif && !dix)
347                 return sprintf(buf, "none\n");
348
349         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
350 }
351 static DEVICE_ATTR_RO(protection_mode);
352
353 static ssize_t
354 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
355 {
356         struct scsi_disk *sdkp = to_scsi_disk(dev);
357
358         return sprintf(buf, "%u\n", sdkp->ATO);
359 }
360 static DEVICE_ATTR_RO(app_tag_own);
361
362 static ssize_t
363 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
364                        char *buf)
365 {
366         struct scsi_disk *sdkp = to_scsi_disk(dev);
367
368         return sprintf(buf, "%u\n", sdkp->lbpme);
369 }
370 static DEVICE_ATTR_RO(thin_provisioning);
371
372 /* sysfs_match_string() requires dense arrays */
373 static const char *lbp_mode[] = {
374         [SD_LBP_FULL]           = "full",
375         [SD_LBP_UNMAP]          = "unmap",
376         [SD_LBP_WS16]           = "writesame_16",
377         [SD_LBP_WS10]           = "writesame_10",
378         [SD_LBP_ZERO]           = "writesame_zero",
379         [SD_LBP_DISABLE]        = "disabled",
380 };
381
382 static ssize_t
383 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
384                        char *buf)
385 {
386         struct scsi_disk *sdkp = to_scsi_disk(dev);
387
388         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
389 }
390
391 static ssize_t
392 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
393                         const char *buf, size_t count)
394 {
395         struct scsi_disk *sdkp = to_scsi_disk(dev);
396         struct scsi_device *sdp = sdkp->device;
397         int mode;
398
399         if (!capable(CAP_SYS_ADMIN))
400                 return -EACCES;
401
402         if (sd_is_zoned(sdkp)) {
403                 sd_config_discard(sdkp, SD_LBP_DISABLE);
404                 return count;
405         }
406
407         if (sdp->type != TYPE_DISK)
408                 return -EINVAL;
409
410         mode = sysfs_match_string(lbp_mode, buf);
411         if (mode < 0)
412                 return -EINVAL;
413
414         sd_config_discard(sdkp, mode);
415
416         return count;
417 }
418 static DEVICE_ATTR_RW(provisioning_mode);
419
420 /* sysfs_match_string() requires dense arrays */
421 static const char *zeroing_mode[] = {
422         [SD_ZERO_WRITE]         = "write",
423         [SD_ZERO_WS]            = "writesame",
424         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
425         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
426 };
427
428 static ssize_t
429 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
430                   char *buf)
431 {
432         struct scsi_disk *sdkp = to_scsi_disk(dev);
433
434         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
435 }
436
437 static ssize_t
438 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
439                    const char *buf, size_t count)
440 {
441         struct scsi_disk *sdkp = to_scsi_disk(dev);
442         int mode;
443
444         if (!capable(CAP_SYS_ADMIN))
445                 return -EACCES;
446
447         mode = sysfs_match_string(zeroing_mode, buf);
448         if (mode < 0)
449                 return -EINVAL;
450
451         sdkp->zeroing_mode = mode;
452
453         return count;
454 }
455 static DEVICE_ATTR_RW(zeroing_mode);
456
457 static ssize_t
458 max_medium_access_timeouts_show(struct device *dev,
459                                 struct device_attribute *attr, char *buf)
460 {
461         struct scsi_disk *sdkp = to_scsi_disk(dev);
462
463         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
464 }
465
466 static ssize_t
467 max_medium_access_timeouts_store(struct device *dev,
468                                  struct device_attribute *attr, const char *buf,
469                                  size_t count)
470 {
471         struct scsi_disk *sdkp = to_scsi_disk(dev);
472         int err;
473
474         if (!capable(CAP_SYS_ADMIN))
475                 return -EACCES;
476
477         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
478
479         return err ? err : count;
480 }
481 static DEVICE_ATTR_RW(max_medium_access_timeouts);
482
483 static ssize_t
484 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
485                            char *buf)
486 {
487         struct scsi_disk *sdkp = to_scsi_disk(dev);
488
489         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
490 }
491
492 static ssize_t
493 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
494                             const char *buf, size_t count)
495 {
496         struct scsi_disk *sdkp = to_scsi_disk(dev);
497         struct scsi_device *sdp = sdkp->device;
498         unsigned long max;
499         int err;
500
501         if (!capable(CAP_SYS_ADMIN))
502                 return -EACCES;
503
504         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
505                 return -EINVAL;
506
507         err = kstrtoul(buf, 10, &max);
508
509         if (err)
510                 return err;
511
512         if (max == 0)
513                 sdp->no_write_same = 1;
514         else if (max <= SD_MAX_WS16_BLOCKS) {
515                 sdp->no_write_same = 0;
516                 sdkp->max_ws_blocks = max;
517         }
518
519         sd_config_write_same(sdkp);
520
521         return count;
522 }
523 static DEVICE_ATTR_RW(max_write_same_blocks);
524
525 static struct attribute *sd_disk_attrs[] = {
526         &dev_attr_cache_type.attr,
527         &dev_attr_FUA.attr,
528         &dev_attr_allow_restart.attr,
529         &dev_attr_manage_start_stop.attr,
530         &dev_attr_protection_type.attr,
531         &dev_attr_protection_mode.attr,
532         &dev_attr_app_tag_own.attr,
533         &dev_attr_thin_provisioning.attr,
534         &dev_attr_provisioning_mode.attr,
535         &dev_attr_zeroing_mode.attr,
536         &dev_attr_max_write_same_blocks.attr,
537         &dev_attr_max_medium_access_timeouts.attr,
538         NULL,
539 };
540 ATTRIBUTE_GROUPS(sd_disk);
541
542 static struct class sd_disk_class = {
543         .name           = "scsi_disk",
544         .owner          = THIS_MODULE,
545         .dev_release    = scsi_disk_release,
546         .dev_groups     = sd_disk_groups,
547 };
548
549 static const struct dev_pm_ops sd_pm_ops = {
550         .suspend                = sd_suspend_system,
551         .resume                 = sd_resume,
552         .poweroff               = sd_suspend_system,
553         .restore                = sd_resume,
554         .runtime_suspend        = sd_suspend_runtime,
555         .runtime_resume         = sd_resume,
556 };
557
558 static struct scsi_driver sd_template = {
559         .gendrv = {
560                 .name           = "sd",
561                 .owner          = THIS_MODULE,
562                 .probe          = sd_probe,
563                 .remove         = sd_remove,
564                 .shutdown       = sd_shutdown,
565                 .pm             = &sd_pm_ops,
566         },
567         .rescan                 = sd_rescan,
568         .init_command           = sd_init_command,
569         .uninit_command         = sd_uninit_command,
570         .done                   = sd_done,
571         .eh_action              = sd_eh_action,
572         .eh_reset               = sd_eh_reset,
573 };
574
575 /*
576  * Dummy kobj_map->probe function.
577  * The default ->probe function will call modprobe, which is
578  * pointless as this module is already loaded.
579  */
580 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
581 {
582         return NULL;
583 }
584
585 /*
586  * Device no to disk mapping:
587  * 
588  *       major         disc2     disc  p1
589  *   |............|.............|....|....| <- dev_t
590  *    31        20 19          8 7  4 3  0
591  * 
592  * Inside a major, we have 16k disks, however mapped non-
593  * contiguously. The first 16 disks are for major0, the next
594  * ones with major1, ... Disk 256 is for major0 again, disk 272 
595  * for major1, ... 
596  * As we stay compatible with our numbering scheme, we can reuse 
597  * the well-know SCSI majors 8, 65--71, 136--143.
598  */
599 static int sd_major(int major_idx)
600 {
601         switch (major_idx) {
602         case 0:
603                 return SCSI_DISK0_MAJOR;
604         case 1 ... 7:
605                 return SCSI_DISK1_MAJOR + major_idx - 1;
606         case 8 ... 15:
607                 return SCSI_DISK8_MAJOR + major_idx - 8;
608         default:
609                 BUG();
610                 return 0;       /* shut up gcc */
611         }
612 }
613
614 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
615 {
616         struct scsi_disk *sdkp = NULL;
617
618         mutex_lock(&sd_ref_mutex);
619
620         if (disk->private_data) {
621                 sdkp = scsi_disk(disk);
622                 if (scsi_device_get(sdkp->device) == 0)
623                         get_device(&sdkp->dev);
624                 else
625                         sdkp = NULL;
626         }
627         mutex_unlock(&sd_ref_mutex);
628         return sdkp;
629 }
630
631 static void scsi_disk_put(struct scsi_disk *sdkp)
632 {
633         struct scsi_device *sdev = sdkp->device;
634
635         mutex_lock(&sd_ref_mutex);
636         put_device(&sdkp->dev);
637         scsi_device_put(sdev);
638         mutex_unlock(&sd_ref_mutex);
639 }
640
641 #ifdef CONFIG_BLK_SED_OPAL
642 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
643                 size_t len, bool send)
644 {
645         struct scsi_device *sdev = data;
646         u8 cdb[12] = { 0, };
647         int ret;
648
649         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
650         cdb[1] = secp;
651         put_unaligned_be16(spsp, &cdb[2]);
652         put_unaligned_be32(len, &cdb[6]);
653
654         ret = scsi_execute_req(sdev, cdb,
655                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
656                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
657         return ret <= 0 ? ret : -EIO;
658 }
659 #endif /* CONFIG_BLK_SED_OPAL */
660
661 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
662                                            unsigned int dix, unsigned int dif)
663 {
664         struct bio *bio = scmd->request->bio;
665         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
666         unsigned int protect = 0;
667
668         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
669                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
670                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
671
672                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
673                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
674         }
675
676         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
677                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
678
679                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
680                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
681         }
682
683         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
684                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
685
686                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
687                         protect = 3 << 5;       /* Disable target PI checking */
688                 else
689                         protect = 1 << 5;       /* Enable target PI checking */
690         }
691
692         scsi_set_prot_op(scmd, prot_op);
693         scsi_set_prot_type(scmd, dif);
694         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
695
696         return protect;
697 }
698
699 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
700 {
701         struct request_queue *q = sdkp->disk->queue;
702         unsigned int logical_block_size = sdkp->device->sector_size;
703         unsigned int max_blocks = 0;
704
705         q->limits.discard_alignment =
706                 sdkp->unmap_alignment * logical_block_size;
707         q->limits.discard_granularity =
708                 max(sdkp->physical_block_size,
709                     sdkp->unmap_granularity * logical_block_size);
710         sdkp->provisioning_mode = mode;
711
712         switch (mode) {
713
714         case SD_LBP_FULL:
715         case SD_LBP_DISABLE:
716                 blk_queue_max_discard_sectors(q, 0);
717                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
718                 return;
719
720         case SD_LBP_UNMAP:
721                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
722                                           (u32)SD_MAX_WS16_BLOCKS);
723                 break;
724
725         case SD_LBP_WS16:
726                 if (sdkp->device->unmap_limit_for_ws)
727                         max_blocks = sdkp->max_unmap_blocks;
728                 else
729                         max_blocks = sdkp->max_ws_blocks;
730
731                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
732                 break;
733
734         case SD_LBP_WS10:
735                 if (sdkp->device->unmap_limit_for_ws)
736                         max_blocks = sdkp->max_unmap_blocks;
737                 else
738                         max_blocks = sdkp->max_ws_blocks;
739
740                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
741                 break;
742
743         case SD_LBP_ZERO:
744                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
745                                           (u32)SD_MAX_WS10_BLOCKS);
746                 break;
747         }
748
749         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
750         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
751 }
752
753 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
754 {
755         struct scsi_device *sdp = cmd->device;
756         struct request *rq = cmd->request;
757         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
758         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
759         unsigned int data_len = 24;
760         char *buf;
761
762         rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
763         if (!rq->special_vec.bv_page)
764                 return BLKPREP_DEFER;
765         rq->special_vec.bv_offset = 0;
766         rq->special_vec.bv_len = data_len;
767         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
768
769         cmd->cmd_len = 10;
770         cmd->cmnd[0] = UNMAP;
771         cmd->cmnd[8] = 24;
772
773         buf = page_address(rq->special_vec.bv_page);
774         put_unaligned_be16(6 + 16, &buf[0]);
775         put_unaligned_be16(16, &buf[2]);
776         put_unaligned_be64(sector, &buf[8]);
777         put_unaligned_be32(nr_sectors, &buf[16]);
778
779         cmd->allowed = SD_MAX_RETRIES;
780         cmd->transfersize = data_len;
781         rq->timeout = SD_TIMEOUT;
782         scsi_req(rq)->resid_len = data_len;
783
784         return scsi_init_io(cmd);
785 }
786
787 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
788 {
789         struct scsi_device *sdp = cmd->device;
790         struct request *rq = cmd->request;
791         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
792         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
793         u32 data_len = sdp->sector_size;
794
795         rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
796         if (!rq->special_vec.bv_page)
797                 return BLKPREP_DEFER;
798         rq->special_vec.bv_offset = 0;
799         rq->special_vec.bv_len = data_len;
800         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
801
802         cmd->cmd_len = 16;
803         cmd->cmnd[0] = WRITE_SAME_16;
804         if (unmap)
805                 cmd->cmnd[1] = 0x8; /* UNMAP */
806         put_unaligned_be64(sector, &cmd->cmnd[2]);
807         put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
808
809         cmd->allowed = SD_MAX_RETRIES;
810         cmd->transfersize = data_len;
811         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
812         scsi_req(rq)->resid_len = data_len;
813
814         return scsi_init_io(cmd);
815 }
816
817 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
818 {
819         struct scsi_device *sdp = cmd->device;
820         struct request *rq = cmd->request;
821         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
822         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
823         u32 data_len = sdp->sector_size;
824
825         rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
826         if (!rq->special_vec.bv_page)
827                 return BLKPREP_DEFER;
828         rq->special_vec.bv_offset = 0;
829         rq->special_vec.bv_len = data_len;
830         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
831
832         cmd->cmd_len = 10;
833         cmd->cmnd[0] = WRITE_SAME;
834         if (unmap)
835                 cmd->cmnd[1] = 0x8; /* UNMAP */
836         put_unaligned_be32(sector, &cmd->cmnd[2]);
837         put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
838
839         cmd->allowed = SD_MAX_RETRIES;
840         cmd->transfersize = data_len;
841         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
842         scsi_req(rq)->resid_len = data_len;
843
844         return scsi_init_io(cmd);
845 }
846
847 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
848 {
849         struct request *rq = cmd->request;
850         struct scsi_device *sdp = cmd->device;
851         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
852         u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
853         u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
854         int ret;
855
856         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
857                 switch (sdkp->zeroing_mode) {
858                 case SD_ZERO_WS16_UNMAP:
859                         ret = sd_setup_write_same16_cmnd(cmd, true);
860                         goto out;
861                 case SD_ZERO_WS10_UNMAP:
862                         ret = sd_setup_write_same10_cmnd(cmd, true);
863                         goto out;
864                 }
865         }
866
867         if (sdp->no_write_same)
868                 return BLKPREP_INVALID;
869
870         if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
871                 ret = sd_setup_write_same16_cmnd(cmd, false);
872         else
873                 ret = sd_setup_write_same10_cmnd(cmd, false);
874
875 out:
876         if (sd_is_zoned(sdkp) && ret == BLKPREP_OK)
877                 return sd_zbc_write_lock_zone(cmd);
878
879         return ret;
880 }
881
882 static void sd_config_write_same(struct scsi_disk *sdkp)
883 {
884         struct request_queue *q = sdkp->disk->queue;
885         unsigned int logical_block_size = sdkp->device->sector_size;
886
887         if (sdkp->device->no_write_same) {
888                 sdkp->max_ws_blocks = 0;
889                 goto out;
890         }
891
892         /* Some devices can not handle block counts above 0xffff despite
893          * supporting WRITE SAME(16). Consequently we default to 64k
894          * blocks per I/O unless the device explicitly advertises a
895          * bigger limit.
896          */
897         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
898                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
899                                                    (u32)SD_MAX_WS16_BLOCKS);
900         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
901                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
902                                                    (u32)SD_MAX_WS10_BLOCKS);
903         else {
904                 sdkp->device->no_write_same = 1;
905                 sdkp->max_ws_blocks = 0;
906         }
907
908         if (sdkp->lbprz && sdkp->lbpws)
909                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
910         else if (sdkp->lbprz && sdkp->lbpws10)
911                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
912         else if (sdkp->max_ws_blocks)
913                 sdkp->zeroing_mode = SD_ZERO_WS;
914         else
915                 sdkp->zeroing_mode = SD_ZERO_WRITE;
916
917         if (sdkp->max_ws_blocks &&
918             sdkp->physical_block_size > logical_block_size) {
919                 /*
920                  * Reporting a maximum number of blocks that is not aligned
921                  * on the device physical size would cause a large write same
922                  * request to be split into physically unaligned chunks by
923                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
924                  * even if the caller of these functions took care to align the
925                  * large request. So make sure the maximum reported is aligned
926                  * to the device physical block size. This is only an optional
927                  * optimization for regular disks, but this is mandatory to
928                  * avoid failure of large write same requests directed at
929                  * sequential write required zones of host-managed ZBC disks.
930                  */
931                 sdkp->max_ws_blocks =
932                         round_down(sdkp->max_ws_blocks,
933                                    bytes_to_logical(sdkp->device,
934                                                     sdkp->physical_block_size));
935         }
936
937 out:
938         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
939                                          (logical_block_size >> 9));
940         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
941                                          (logical_block_size >> 9));
942 }
943
944 /**
945  * sd_setup_write_same_cmnd - write the same data to multiple blocks
946  * @cmd: command to prepare
947  *
948  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
949  * the preference indicated by the target device.
950  **/
951 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
952 {
953         struct request *rq = cmd->request;
954         struct scsi_device *sdp = cmd->device;
955         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
956         struct bio *bio = rq->bio;
957         sector_t sector = blk_rq_pos(rq);
958         unsigned int nr_sectors = blk_rq_sectors(rq);
959         unsigned int nr_bytes = blk_rq_bytes(rq);
960         int ret;
961
962         if (sdkp->device->no_write_same)
963                 return BLKPREP_INVALID;
964
965         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
966
967         if (sd_is_zoned(sdkp)) {
968                 ret = sd_zbc_write_lock_zone(cmd);
969                 if (ret != BLKPREP_OK)
970                         return ret;
971         }
972
973         sector >>= ilog2(sdp->sector_size) - 9;
974         nr_sectors >>= ilog2(sdp->sector_size) - 9;
975
976         rq->timeout = SD_WRITE_SAME_TIMEOUT;
977
978         if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
979                 cmd->cmd_len = 16;
980                 cmd->cmnd[0] = WRITE_SAME_16;
981                 put_unaligned_be64(sector, &cmd->cmnd[2]);
982                 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
983         } else {
984                 cmd->cmd_len = 10;
985                 cmd->cmnd[0] = WRITE_SAME;
986                 put_unaligned_be32(sector, &cmd->cmnd[2]);
987                 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
988         }
989
990         cmd->transfersize = sdp->sector_size;
991         cmd->allowed = SD_MAX_RETRIES;
992
993         /*
994          * For WRITE SAME the data transferred via the DATA OUT buffer is
995          * different from the amount of data actually written to the target.
996          *
997          * We set up __data_len to the amount of data transferred via the
998          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
999          * to transfer a single sector of data first, but then reset it to
1000          * the amount of data to be written right after so that the I/O path
1001          * knows how much to actually write.
1002          */
1003         rq->__data_len = sdp->sector_size;
1004         ret = scsi_init_io(cmd);
1005         rq->__data_len = nr_bytes;
1006
1007         if (sd_is_zoned(sdkp) && ret != BLKPREP_OK)
1008                 sd_zbc_write_unlock_zone(cmd);
1009
1010         return ret;
1011 }
1012
1013 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1014 {
1015         struct request *rq = cmd->request;
1016
1017         /* flush requests don't perform I/O, zero the S/G table */
1018         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1019
1020         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1021         cmd->cmd_len = 10;
1022         cmd->transfersize = 0;
1023         cmd->allowed = SD_MAX_RETRIES;
1024
1025         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1026         return BLKPREP_OK;
1027 }
1028
1029 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1030 {
1031         struct request *rq = SCpnt->request;
1032         struct scsi_device *sdp = SCpnt->device;
1033         struct gendisk *disk = rq->rq_disk;
1034         struct scsi_disk *sdkp = scsi_disk(disk);
1035         sector_t block = blk_rq_pos(rq);
1036         sector_t threshold;
1037         unsigned int this_count = blk_rq_sectors(rq);
1038         unsigned int dif, dix;
1039         bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
1040         int ret;
1041         unsigned char protect;
1042
1043         if (zoned_write) {
1044                 ret = sd_zbc_write_lock_zone(SCpnt);
1045                 if (ret != BLKPREP_OK)
1046                         return ret;
1047         }
1048
1049         ret = scsi_init_io(SCpnt);
1050         if (ret != BLKPREP_OK)
1051                 goto out;
1052         WARN_ON_ONCE(SCpnt != rq->special);
1053
1054         /* from here on until we're complete, any goto out
1055          * is used for a killable error condition */
1056         ret = BLKPREP_KILL;
1057
1058         SCSI_LOG_HLQUEUE(1,
1059                 scmd_printk(KERN_INFO, SCpnt,
1060                         "%s: block=%llu, count=%d\n",
1061                         __func__, (unsigned long long)block, this_count));
1062
1063         if (!sdp || !scsi_device_online(sdp) ||
1064             block + blk_rq_sectors(rq) > get_capacity(disk)) {
1065                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1066                                                 "Finishing %u sectors\n",
1067                                                 blk_rq_sectors(rq)));
1068                 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1069                                                 "Retry with 0x%p\n", SCpnt));
1070                 goto out;
1071         }
1072
1073         if (sdp->changed) {
1074                 /*
1075                  * quietly refuse to do anything to a changed disc until 
1076                  * the changed bit has been reset
1077                  */
1078                 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1079                 goto out;
1080         }
1081
1082         /*
1083          * Some SD card readers can't handle multi-sector accesses which touch
1084          * the last one or two hardware sectors.  Split accesses as needed.
1085          */
1086         threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1087                 (sdp->sector_size / 512);
1088
1089         if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1090                 if (block < threshold) {
1091                         /* Access up to the threshold but not beyond */
1092                         this_count = threshold - block;
1093                 } else {
1094                         /* Access only a single hardware sector */
1095                         this_count = sdp->sector_size / 512;
1096                 }
1097         }
1098
1099         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1100                                         (unsigned long long)block));
1101
1102         /*
1103          * If we have a 1K hardware sectorsize, prevent access to single
1104          * 512 byte sectors.  In theory we could handle this - in fact
1105          * the scsi cdrom driver must be able to handle this because
1106          * we typically use 1K blocksizes, and cdroms typically have
1107          * 2K hardware sectorsizes.  Of course, things are simpler
1108          * with the cdrom, since it is read-only.  For performance
1109          * reasons, the filesystems should be able to handle this
1110          * and not force the scsi disk driver to use bounce buffers
1111          * for this.
1112          */
1113         if (sdp->sector_size == 1024) {
1114                 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1115                         scmd_printk(KERN_ERR, SCpnt,
1116                                     "Bad block number requested\n");
1117                         goto out;
1118                 } else {
1119                         block = block >> 1;
1120                         this_count = this_count >> 1;
1121                 }
1122         }
1123         if (sdp->sector_size == 2048) {
1124                 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1125                         scmd_printk(KERN_ERR, SCpnt,
1126                                     "Bad block number requested\n");
1127                         goto out;
1128                 } else {
1129                         block = block >> 2;
1130                         this_count = this_count >> 2;
1131                 }
1132         }
1133         if (sdp->sector_size == 4096) {
1134                 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1135                         scmd_printk(KERN_ERR, SCpnt,
1136                                     "Bad block number requested\n");
1137                         goto out;
1138                 } else {
1139                         block = block >> 3;
1140                         this_count = this_count >> 3;
1141                 }
1142         }
1143         if (rq_data_dir(rq) == WRITE) {
1144                 SCpnt->cmnd[0] = WRITE_6;
1145
1146                 if (blk_integrity_rq(rq))
1147                         sd_dif_prepare(SCpnt);
1148
1149         } else if (rq_data_dir(rq) == READ) {
1150                 SCpnt->cmnd[0] = READ_6;
1151         } else {
1152                 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1153                 goto out;
1154         }
1155
1156         SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1157                                         "%s %d/%u 512 byte blocks.\n",
1158                                         (rq_data_dir(rq) == WRITE) ?
1159                                         "writing" : "reading", this_count,
1160                                         blk_rq_sectors(rq)));
1161
1162         dix = scsi_prot_sg_count(SCpnt);
1163         dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1164
1165         if (dif || dix)
1166                 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1167         else
1168                 protect = 0;
1169
1170         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1171                 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1172
1173                 if (unlikely(SCpnt->cmnd == NULL)) {
1174                         ret = BLKPREP_DEFER;
1175                         goto out;
1176                 }
1177
1178                 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1179                 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1180                 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1181                 SCpnt->cmnd[7] = 0x18;
1182                 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1183                 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1184
1185                 /* LBA */
1186                 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1187                 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1188                 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1189                 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1190                 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1191                 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1192                 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1193                 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1194
1195                 /* Expected Indirect LBA */
1196                 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1197                 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1198                 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1199                 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1200
1201                 /* Transfer length */
1202                 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1203                 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1204                 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1205                 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1206         } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1207                 SCpnt->cmnd[0] += READ_16 - READ_6;
1208                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1209                 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1210                 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1211                 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1212                 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1213                 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1214                 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1215                 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1216                 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1217                 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1218                 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1219                 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1220                 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1221                 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1222         } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1223                    scsi_device_protection(SCpnt->device) ||
1224                    SCpnt->device->use_10_for_rw) {
1225                 SCpnt->cmnd[0] += READ_10 - READ_6;
1226                 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1227                 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1228                 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1229                 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1230                 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1231                 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1232                 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1233                 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1234         } else {
1235                 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1236                         /*
1237                          * This happens only if this drive failed
1238                          * 10byte rw command with ILLEGAL_REQUEST
1239                          * during operation and thus turned off
1240                          * use_10_for_rw.
1241                          */
1242                         scmd_printk(KERN_ERR, SCpnt,
1243                                     "FUA write on READ/WRITE(6) drive\n");
1244                         goto out;
1245                 }
1246
1247                 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1248                 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1249                 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1250                 SCpnt->cmnd[4] = (unsigned char) this_count;
1251                 SCpnt->cmnd[5] = 0;
1252         }
1253         SCpnt->sdb.length = this_count * sdp->sector_size;
1254
1255         /*
1256          * We shouldn't disconnect in the middle of a sector, so with a dumb
1257          * host adapter, it's safe to assume that we can at least transfer
1258          * this many bytes between each connect / disconnect.
1259          */
1260         SCpnt->transfersize = sdp->sector_size;
1261         SCpnt->underflow = this_count << 9;
1262         SCpnt->allowed = SD_MAX_RETRIES;
1263
1264         /*
1265          * This indicates that the command is ready from our end to be
1266          * queued.
1267          */
1268         ret = BLKPREP_OK;
1269  out:
1270         if (zoned_write && ret != BLKPREP_OK)
1271                 sd_zbc_write_unlock_zone(SCpnt);
1272
1273         return ret;
1274 }
1275
1276 static int sd_init_command(struct scsi_cmnd *cmd)
1277 {
1278         struct request *rq = cmd->request;
1279
1280         switch (req_op(rq)) {
1281         case REQ_OP_DISCARD:
1282                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1283                 case SD_LBP_UNMAP:
1284                         return sd_setup_unmap_cmnd(cmd);
1285                 case SD_LBP_WS16:
1286                         return sd_setup_write_same16_cmnd(cmd, true);
1287                 case SD_LBP_WS10:
1288                         return sd_setup_write_same10_cmnd(cmd, true);
1289                 case SD_LBP_ZERO:
1290                         return sd_setup_write_same10_cmnd(cmd, false);
1291                 default:
1292                         return BLKPREP_INVALID;
1293                 }
1294         case REQ_OP_WRITE_ZEROES:
1295                 return sd_setup_write_zeroes_cmnd(cmd);
1296         case REQ_OP_WRITE_SAME:
1297                 return sd_setup_write_same_cmnd(cmd);
1298         case REQ_OP_FLUSH:
1299                 return sd_setup_flush_cmnd(cmd);
1300         case REQ_OP_READ:
1301         case REQ_OP_WRITE:
1302                 return sd_setup_read_write_cmnd(cmd);
1303         case REQ_OP_ZONE_REPORT:
1304                 return sd_zbc_setup_report_cmnd(cmd);
1305         case REQ_OP_ZONE_RESET:
1306                 return sd_zbc_setup_reset_cmnd(cmd);
1307         default:
1308                 BUG();
1309         }
1310 }
1311
1312 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1313 {
1314         struct request *rq = SCpnt->request;
1315
1316         if (SCpnt->flags & SCMD_ZONE_WRITE_LOCK)
1317                 sd_zbc_write_unlock_zone(SCpnt);
1318
1319         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1320                 __free_page(rq->special_vec.bv_page);
1321
1322         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1323                 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1324                 SCpnt->cmnd = NULL;
1325                 SCpnt->cmd_len = 0;
1326         }
1327 }
1328
1329 /**
1330  *      sd_open - open a scsi disk device
1331  *      @bdev: Block device of the scsi disk to open
1332  *      @mode: FMODE_* mask
1333  *
1334  *      Returns 0 if successful. Returns a negated errno value in case 
1335  *      of error.
1336  *
1337  *      Note: This can be called from a user context (e.g. fsck(1) )
1338  *      or from within the kernel (e.g. as a result of a mount(1) ).
1339  *      In the latter case @inode and @filp carry an abridged amount
1340  *      of information as noted above.
1341  *
1342  *      Locking: called with bdev->bd_mutex held.
1343  **/
1344 static int sd_open(struct block_device *bdev, fmode_t mode)
1345 {
1346         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1347         struct scsi_device *sdev;
1348         int retval;
1349
1350         if (!sdkp)
1351                 return -ENXIO;
1352
1353         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1354
1355         sdev = sdkp->device;
1356
1357         /*
1358          * If the device is in error recovery, wait until it is done.
1359          * If the device is offline, then disallow any access to it.
1360          */
1361         retval = -ENXIO;
1362         if (!scsi_block_when_processing_errors(sdev))
1363                 goto error_out;
1364
1365         if (sdev->removable || sdkp->write_prot)
1366                 check_disk_change(bdev);
1367
1368         /*
1369          * If the drive is empty, just let the open fail.
1370          */
1371         retval = -ENOMEDIUM;
1372         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1373                 goto error_out;
1374
1375         /*
1376          * If the device has the write protect tab set, have the open fail
1377          * if the user expects to be able to write to the thing.
1378          */
1379         retval = -EROFS;
1380         if (sdkp->write_prot && (mode & FMODE_WRITE))
1381                 goto error_out;
1382
1383         /*
1384          * It is possible that the disk changing stuff resulted in
1385          * the device being taken offline.  If this is the case,
1386          * report this to the user, and don't pretend that the
1387          * open actually succeeded.
1388          */
1389         retval = -ENXIO;
1390         if (!scsi_device_online(sdev))
1391                 goto error_out;
1392
1393         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1394                 if (scsi_block_when_processing_errors(sdev))
1395                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1396         }
1397
1398         return 0;
1399
1400 error_out:
1401         scsi_disk_put(sdkp);
1402         return retval;  
1403 }
1404
1405 /**
1406  *      sd_release - invoked when the (last) close(2) is called on this
1407  *      scsi disk.
1408  *      @disk: disk to release
1409  *      @mode: FMODE_* mask
1410  *
1411  *      Returns 0. 
1412  *
1413  *      Note: may block (uninterruptible) if error recovery is underway
1414  *      on this disk.
1415  *
1416  *      Locking: called with bdev->bd_mutex held.
1417  **/
1418 static void sd_release(struct gendisk *disk, fmode_t mode)
1419 {
1420         struct scsi_disk *sdkp = scsi_disk(disk);
1421         struct scsi_device *sdev = sdkp->device;
1422
1423         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1424
1425         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1426                 if (scsi_block_when_processing_errors(sdev))
1427                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1428         }
1429
1430         /*
1431          * XXX and what if there are packets in flight and this close()
1432          * XXX is followed by a "rmmod sd_mod"?
1433          */
1434
1435         scsi_disk_put(sdkp);
1436 }
1437
1438 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1439 {
1440         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1441         struct scsi_device *sdp = sdkp->device;
1442         struct Scsi_Host *host = sdp->host;
1443         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1444         int diskinfo[4];
1445
1446         /* default to most commonly used values */
1447         diskinfo[0] = 0x40;     /* 1 << 6 */
1448         diskinfo[1] = 0x20;     /* 1 << 5 */
1449         diskinfo[2] = capacity >> 11;
1450
1451         /* override with calculated, extended default, or driver values */
1452         if (host->hostt->bios_param)
1453                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1454         else
1455                 scsicam_bios_param(bdev, capacity, diskinfo);
1456
1457         geo->heads = diskinfo[0];
1458         geo->sectors = diskinfo[1];
1459         geo->cylinders = diskinfo[2];
1460         return 0;
1461 }
1462
1463 /**
1464  *      sd_ioctl - process an ioctl
1465  *      @bdev: target block device
1466  *      @mode: FMODE_* mask
1467  *      @cmd: ioctl command number
1468  *      @arg: this is third argument given to ioctl(2) system call.
1469  *      Often contains a pointer.
1470  *
1471  *      Returns 0 if successful (some ioctls return positive numbers on
1472  *      success as well). Returns a negated errno value in case of error.
1473  *
1474  *      Note: most ioctls are forward onto the block subsystem or further
1475  *      down in the scsi subsystem.
1476  **/
1477 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1478                     unsigned int cmd, unsigned long arg)
1479 {
1480         struct gendisk *disk = bdev->bd_disk;
1481         struct scsi_disk *sdkp = scsi_disk(disk);
1482         struct scsi_device *sdp = sdkp->device;
1483         void __user *p = (void __user *)arg;
1484         int error;
1485     
1486         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1487                                     "cmd=0x%x\n", disk->disk_name, cmd));
1488
1489         error = scsi_verify_blk_ioctl(bdev, cmd);
1490         if (error < 0)
1491                 return error;
1492
1493         /*
1494          * If we are in the middle of error recovery, don't let anyone
1495          * else try and use this device.  Also, if error recovery fails, it
1496          * may try and take the device offline, in which case all further
1497          * access to the device is prohibited.
1498          */
1499         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1500                         (mode & FMODE_NDELAY) != 0);
1501         if (error)
1502                 goto out;
1503
1504         if (is_sed_ioctl(cmd))
1505                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1506
1507         /*
1508          * Send SCSI addressing ioctls directly to mid level, send other
1509          * ioctls to block level and then onto mid level if they can't be
1510          * resolved.
1511          */
1512         switch (cmd) {
1513                 case SCSI_IOCTL_GET_IDLUN:
1514                 case SCSI_IOCTL_GET_BUS_NUMBER:
1515                         error = scsi_ioctl(sdp, cmd, p);
1516                         break;
1517                 default:
1518                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1519                         if (error != -ENOTTY)
1520                                 break;
1521                         error = scsi_ioctl(sdp, cmd, p);
1522                         break;
1523         }
1524 out:
1525         return error;
1526 }
1527
1528 static void set_media_not_present(struct scsi_disk *sdkp)
1529 {
1530         if (sdkp->media_present)
1531                 sdkp->device->changed = 1;
1532
1533         if (sdkp->device->removable) {
1534                 sdkp->media_present = 0;
1535                 sdkp->capacity = 0;
1536         }
1537 }
1538
1539 static int media_not_present(struct scsi_disk *sdkp,
1540                              struct scsi_sense_hdr *sshdr)
1541 {
1542         if (!scsi_sense_valid(sshdr))
1543                 return 0;
1544
1545         /* not invoked for commands that could return deferred errors */
1546         switch (sshdr->sense_key) {
1547         case UNIT_ATTENTION:
1548         case NOT_READY:
1549                 /* medium not present */
1550                 if (sshdr->asc == 0x3A) {
1551                         set_media_not_present(sdkp);
1552                         return 1;
1553                 }
1554         }
1555         return 0;
1556 }
1557
1558 /**
1559  *      sd_check_events - check media events
1560  *      @disk: kernel device descriptor
1561  *      @clearing: disk events currently being cleared
1562  *
1563  *      Returns mask of DISK_EVENT_*.
1564  *
1565  *      Note: this function is invoked from the block subsystem.
1566  **/
1567 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1568 {
1569         struct scsi_disk *sdkp = scsi_disk_get(disk);
1570         struct scsi_device *sdp;
1571         int retval;
1572
1573         if (!sdkp)
1574                 return 0;
1575
1576         sdp = sdkp->device;
1577         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1578
1579         /*
1580          * If the device is offline, don't send any commands - just pretend as
1581          * if the command failed.  If the device ever comes back online, we
1582          * can deal with it then.  It is only because of unrecoverable errors
1583          * that we would ever take a device offline in the first place.
1584          */
1585         if (!scsi_device_online(sdp)) {
1586                 set_media_not_present(sdkp);
1587                 goto out;
1588         }
1589
1590         /*
1591          * Using TEST_UNIT_READY enables differentiation between drive with
1592          * no cartridge loaded - NOT READY, drive with changed cartridge -
1593          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1594          *
1595          * Drives that auto spin down. eg iomega jaz 1G, will be started
1596          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1597          * sd_revalidate() is called.
1598          */
1599         if (scsi_block_when_processing_errors(sdp)) {
1600                 struct scsi_sense_hdr sshdr = { 0, };
1601
1602                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1603                                               &sshdr);
1604
1605                 /* failed to execute TUR, assume media not present */
1606                 if (host_byte(retval)) {
1607                         set_media_not_present(sdkp);
1608                         goto out;
1609                 }
1610
1611                 if (media_not_present(sdkp, &sshdr))
1612                         goto out;
1613         }
1614
1615         /*
1616          * For removable scsi disk we have to recognise the presence
1617          * of a disk in the drive.
1618          */
1619         if (!sdkp->media_present)
1620                 sdp->changed = 1;
1621         sdkp->media_present = 1;
1622 out:
1623         /*
1624          * sdp->changed is set under the following conditions:
1625          *
1626          *      Medium present state has changed in either direction.
1627          *      Device has indicated UNIT_ATTENTION.
1628          */
1629         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1630         sdp->changed = 0;
1631         scsi_disk_put(sdkp);
1632         return retval;
1633 }
1634
1635 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1636 {
1637         int retries, res;
1638         struct scsi_device *sdp = sdkp->device;
1639         const int timeout = sdp->request_queue->rq_timeout
1640                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1641         struct scsi_sense_hdr my_sshdr;
1642
1643         if (!scsi_device_online(sdp))
1644                 return -ENODEV;
1645
1646         /* caller might not be interested in sense, but we need it */
1647         if (!sshdr)
1648                 sshdr = &my_sshdr;
1649
1650         for (retries = 3; retries > 0; --retries) {
1651                 unsigned char cmd[10] = { 0 };
1652
1653                 cmd[0] = SYNCHRONIZE_CACHE;
1654                 /*
1655                  * Leave the rest of the command zero to indicate
1656                  * flush everything.
1657                  */
1658                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1659                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1660                 if (res == 0)
1661                         break;
1662         }
1663
1664         if (res) {
1665                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1666
1667                 if (driver_byte(res) & DRIVER_SENSE)
1668                         sd_print_sense_hdr(sdkp, sshdr);
1669
1670                 /* we need to evaluate the error return  */
1671                 if (scsi_sense_valid(sshdr) &&
1672                         (sshdr->asc == 0x3a ||  /* medium not present */
1673                          sshdr->asc == 0x20))   /* invalid command */
1674                                 /* this is no error here */
1675                                 return 0;
1676
1677                 switch (host_byte(res)) {
1678                 /* ignore errors due to racing a disconnection */
1679                 case DID_BAD_TARGET:
1680                 case DID_NO_CONNECT:
1681                         return 0;
1682                 /* signal the upper layer it might try again */
1683                 case DID_BUS_BUSY:
1684                 case DID_IMM_RETRY:
1685                 case DID_REQUEUE:
1686                 case DID_SOFT_ERROR:
1687                         return -EBUSY;
1688                 default:
1689                         return -EIO;
1690                 }
1691         }
1692         return 0;
1693 }
1694
1695 static void sd_rescan(struct device *dev)
1696 {
1697         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1698
1699         revalidate_disk(sdkp->disk);
1700 }
1701
1702
1703 #ifdef CONFIG_COMPAT
1704 /* 
1705  * This gets directly called from VFS. When the ioctl 
1706  * is not recognized we go back to the other translation paths. 
1707  */
1708 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1709                            unsigned int cmd, unsigned long arg)
1710 {
1711         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1712         int error;
1713
1714         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1715                         (mode & FMODE_NDELAY) != 0);
1716         if (error)
1717                 return error;
1718                
1719         /* 
1720          * Let the static ioctl translation table take care of it.
1721          */
1722         if (!sdev->host->hostt->compat_ioctl)
1723                 return -ENOIOCTLCMD; 
1724         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1725 }
1726 #endif
1727
1728 static char sd_pr_type(enum pr_type type)
1729 {
1730         switch (type) {
1731         case PR_WRITE_EXCLUSIVE:
1732                 return 0x01;
1733         case PR_EXCLUSIVE_ACCESS:
1734                 return 0x03;
1735         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1736                 return 0x05;
1737         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1738                 return 0x06;
1739         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1740                 return 0x07;
1741         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1742                 return 0x08;
1743         default:
1744                 return 0;
1745         }
1746 };
1747
1748 static int sd_pr_command(struct block_device *bdev, u8 sa,
1749                 u64 key, u64 sa_key, u8 type, u8 flags)
1750 {
1751         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1752         struct scsi_sense_hdr sshdr;
1753         int result;
1754         u8 cmd[16] = { 0, };
1755         u8 data[24] = { 0, };
1756
1757         cmd[0] = PERSISTENT_RESERVE_OUT;
1758         cmd[1] = sa;
1759         cmd[2] = type;
1760         put_unaligned_be32(sizeof(data), &cmd[5]);
1761
1762         put_unaligned_be64(key, &data[0]);
1763         put_unaligned_be64(sa_key, &data[8]);
1764         data[20] = flags;
1765
1766         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1767                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1768
1769         if ((driver_byte(result) & DRIVER_SENSE) &&
1770             (scsi_sense_valid(&sshdr))) {
1771                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1772                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1773         }
1774
1775         return result;
1776 }
1777
1778 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1779                 u32 flags)
1780 {
1781         if (flags & ~PR_FL_IGNORE_KEY)
1782                 return -EOPNOTSUPP;
1783         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1784                         old_key, new_key, 0,
1785                         (1 << 0) /* APTPL */);
1786 }
1787
1788 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1789                 u32 flags)
1790 {
1791         if (flags)
1792                 return -EOPNOTSUPP;
1793         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1794 }
1795
1796 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1797 {
1798         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1799 }
1800
1801 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1802                 enum pr_type type, bool abort)
1803 {
1804         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1805                              sd_pr_type(type), 0);
1806 }
1807
1808 static int sd_pr_clear(struct block_device *bdev, u64 key)
1809 {
1810         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1811 }
1812
1813 static const struct pr_ops sd_pr_ops = {
1814         .pr_register    = sd_pr_register,
1815         .pr_reserve     = sd_pr_reserve,
1816         .pr_release     = sd_pr_release,
1817         .pr_preempt     = sd_pr_preempt,
1818         .pr_clear       = sd_pr_clear,
1819 };
1820
1821 static const struct block_device_operations sd_fops = {
1822         .owner                  = THIS_MODULE,
1823         .open                   = sd_open,
1824         .release                = sd_release,
1825         .ioctl                  = sd_ioctl,
1826         .getgeo                 = sd_getgeo,
1827 #ifdef CONFIG_COMPAT
1828         .compat_ioctl           = sd_compat_ioctl,
1829 #endif
1830         .check_events           = sd_check_events,
1831         .revalidate_disk        = sd_revalidate_disk,
1832         .unlock_native_capacity = sd_unlock_native_capacity,
1833         .pr_ops                 = &sd_pr_ops,
1834 };
1835
1836 /**
1837  *      sd_eh_reset - reset error handling callback
1838  *      @scmd:          sd-issued command that has failed
1839  *
1840  *      This function is called by the SCSI midlayer before starting
1841  *      SCSI EH. When counting medium access failures we have to be
1842  *      careful to register it only only once per device and SCSI EH run;
1843  *      there might be several timed out commands which will cause the
1844  *      'max_medium_access_timeouts' counter to trigger after the first
1845  *      SCSI EH run already and set the device to offline.
1846  *      So this function resets the internal counter before starting SCSI EH.
1847  **/
1848 static void sd_eh_reset(struct scsi_cmnd *scmd)
1849 {
1850         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1851
1852         /* New SCSI EH run, reset gate variable */
1853         sdkp->ignore_medium_access_errors = false;
1854 }
1855
1856 /**
1857  *      sd_eh_action - error handling callback
1858  *      @scmd:          sd-issued command that has failed
1859  *      @eh_disp:       The recovery disposition suggested by the midlayer
1860  *
1861  *      This function is called by the SCSI midlayer upon completion of an
1862  *      error test command (currently TEST UNIT READY). The result of sending
1863  *      the eh command is passed in eh_disp.  We're looking for devices that
1864  *      fail medium access commands but are OK with non access commands like
1865  *      test unit ready (so wrongly see the device as having a successful
1866  *      recovery)
1867  **/
1868 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1869 {
1870         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1871         struct scsi_device *sdev = scmd->device;
1872
1873         if (!scsi_device_online(sdev) ||
1874             !scsi_medium_access_command(scmd) ||
1875             host_byte(scmd->result) != DID_TIME_OUT ||
1876             eh_disp != SUCCESS)
1877                 return eh_disp;
1878
1879         /*
1880          * The device has timed out executing a medium access command.
1881          * However, the TEST UNIT READY command sent during error
1882          * handling completed successfully. Either the device is in the
1883          * process of recovering or has it suffered an internal failure
1884          * that prevents access to the storage medium.
1885          */
1886         if (!sdkp->ignore_medium_access_errors) {
1887                 sdkp->medium_access_timed_out++;
1888                 sdkp->ignore_medium_access_errors = true;
1889         }
1890
1891         /*
1892          * If the device keeps failing read/write commands but TEST UNIT
1893          * READY always completes successfully we assume that medium
1894          * access is no longer possible and take the device offline.
1895          */
1896         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1897                 scmd_printk(KERN_ERR, scmd,
1898                             "Medium access timeout failure. Offlining disk!\n");
1899                 mutex_lock(&sdev->state_mutex);
1900                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1901                 mutex_unlock(&sdev->state_mutex);
1902
1903                 return SUCCESS;
1904         }
1905
1906         return eh_disp;
1907 }
1908
1909 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1910 {
1911         struct request *req = scmd->request;
1912         struct scsi_device *sdev = scmd->device;
1913         unsigned int transferred, good_bytes;
1914         u64 start_lba, end_lba, bad_lba;
1915
1916         /*
1917          * Some commands have a payload smaller than the device logical
1918          * block size (e.g. INQUIRY on a 4K disk).
1919          */
1920         if (scsi_bufflen(scmd) <= sdev->sector_size)
1921                 return 0;
1922
1923         /* Check if we have a 'bad_lba' information */
1924         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1925                                      SCSI_SENSE_BUFFERSIZE,
1926                                      &bad_lba))
1927                 return 0;
1928
1929         /*
1930          * If the bad lba was reported incorrectly, we have no idea where
1931          * the error is.
1932          */
1933         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1934         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1935         if (bad_lba < start_lba || bad_lba >= end_lba)
1936                 return 0;
1937
1938         /*
1939          * resid is optional but mostly filled in.  When it's unused,
1940          * its value is zero, so we assume the whole buffer transferred
1941          */
1942         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1943
1944         /* This computation should always be done in terms of the
1945          * resolution of the device's medium.
1946          */
1947         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1948
1949         return min(good_bytes, transferred);
1950 }
1951
1952 /**
1953  *      sd_done - bottom half handler: called when the lower level
1954  *      driver has completed (successfully or otherwise) a scsi command.
1955  *      @SCpnt: mid-level's per command structure.
1956  *
1957  *      Note: potentially run from within an ISR. Must not block.
1958  **/
1959 static int sd_done(struct scsi_cmnd *SCpnt)
1960 {
1961         int result = SCpnt->result;
1962         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1963         unsigned int sector_size = SCpnt->device->sector_size;
1964         unsigned int resid;
1965         struct scsi_sense_hdr sshdr;
1966         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1967         struct request *req = SCpnt->request;
1968         int sense_valid = 0;
1969         int sense_deferred = 0;
1970
1971         switch (req_op(req)) {
1972         case REQ_OP_DISCARD:
1973         case REQ_OP_WRITE_ZEROES:
1974         case REQ_OP_WRITE_SAME:
1975         case REQ_OP_ZONE_RESET:
1976                 if (!result) {
1977                         good_bytes = blk_rq_bytes(req);
1978                         scsi_set_resid(SCpnt, 0);
1979                 } else {
1980                         good_bytes = 0;
1981                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1982                 }
1983                 break;
1984         case REQ_OP_ZONE_REPORT:
1985                 if (!result) {
1986                         good_bytes = scsi_bufflen(SCpnt)
1987                                 - scsi_get_resid(SCpnt);
1988                         scsi_set_resid(SCpnt, 0);
1989                 } else {
1990                         good_bytes = 0;
1991                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1992                 }
1993                 break;
1994         default:
1995                 /*
1996                  * In case of bogus fw or device, we could end up having
1997                  * an unaligned partial completion. Check this here and force
1998                  * alignment.
1999                  */
2000                 resid = scsi_get_resid(SCpnt);
2001                 if (resid & (sector_size - 1)) {
2002                         sd_printk(KERN_INFO, sdkp,
2003                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2004                                 resid, sector_size);
2005                         resid = min(scsi_bufflen(SCpnt),
2006                                     round_up(resid, sector_size));
2007                         scsi_set_resid(SCpnt, resid);
2008                 }
2009         }
2010
2011         if (result) {
2012                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2013                 if (sense_valid)
2014                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2015         }
2016         sdkp->medium_access_timed_out = 0;
2017
2018         if (driver_byte(result) != DRIVER_SENSE &&
2019             (!sense_valid || sense_deferred))
2020                 goto out;
2021
2022         switch (sshdr.sense_key) {
2023         case HARDWARE_ERROR:
2024         case MEDIUM_ERROR:
2025                 good_bytes = sd_completed_bytes(SCpnt);
2026                 break;
2027         case RECOVERED_ERROR:
2028                 good_bytes = scsi_bufflen(SCpnt);
2029                 break;
2030         case NO_SENSE:
2031                 /* This indicates a false check condition, so ignore it.  An
2032                  * unknown amount of data was transferred so treat it as an
2033                  * error.
2034                  */
2035                 SCpnt->result = 0;
2036                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2037                 break;
2038         case ABORTED_COMMAND:
2039                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2040                         good_bytes = sd_completed_bytes(SCpnt);
2041                 break;
2042         case ILLEGAL_REQUEST:
2043                 switch (sshdr.asc) {
2044                 case 0x10:      /* DIX: Host detected corruption */
2045                         good_bytes = sd_completed_bytes(SCpnt);
2046                         break;
2047                 case 0x20:      /* INVALID COMMAND OPCODE */
2048                 case 0x24:      /* INVALID FIELD IN CDB */
2049                         switch (SCpnt->cmnd[0]) {
2050                         case UNMAP:
2051                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2052                                 break;
2053                         case WRITE_SAME_16:
2054                         case WRITE_SAME:
2055                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2056                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2057                                 } else {
2058                                         sdkp->device->no_write_same = 1;
2059                                         sd_config_write_same(sdkp);
2060                                         req->__data_len = blk_rq_bytes(req);
2061                                         req->rq_flags |= RQF_QUIET;
2062                                 }
2063                                 break;
2064                         }
2065                 }
2066                 break;
2067         default:
2068                 break;
2069         }
2070
2071  out:
2072         if (sd_is_zoned(sdkp))
2073                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2074
2075         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2076                                            "sd_done: completed %d of %d bytes\n",
2077                                            good_bytes, scsi_bufflen(SCpnt)));
2078
2079         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2080                 sd_dif_complete(SCpnt, good_bytes);
2081
2082         return good_bytes;
2083 }
2084
2085 /*
2086  * spinup disk - called only in sd_revalidate_disk()
2087  */
2088 static void
2089 sd_spinup_disk(struct scsi_disk *sdkp)
2090 {
2091         unsigned char cmd[10];
2092         unsigned long spintime_expire = 0;
2093         int retries, spintime;
2094         unsigned int the_result;
2095         struct scsi_sense_hdr sshdr;
2096         int sense_valid = 0;
2097
2098         spintime = 0;
2099
2100         /* Spin up drives, as required.  Only do this at boot time */
2101         /* Spinup needs to be done for module loads too. */
2102         do {
2103                 retries = 0;
2104
2105                 do {
2106                         cmd[0] = TEST_UNIT_READY;
2107                         memset((void *) &cmd[1], 0, 9);
2108
2109                         the_result = scsi_execute_req(sdkp->device, cmd,
2110                                                       DMA_NONE, NULL, 0,
2111                                                       &sshdr, SD_TIMEOUT,
2112                                                       SD_MAX_RETRIES, NULL);
2113
2114                         /*
2115                          * If the drive has indicated to us that it
2116                          * doesn't have any media in it, don't bother
2117                          * with any more polling.
2118                          */
2119                         if (media_not_present(sdkp, &sshdr))
2120                                 return;
2121
2122                         if (the_result)
2123                                 sense_valid = scsi_sense_valid(&sshdr);
2124                         retries++;
2125                 } while (retries < 3 && 
2126                          (!scsi_status_is_good(the_result) ||
2127                           ((driver_byte(the_result) & DRIVER_SENSE) &&
2128                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2129
2130                 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2131                         /* no sense, TUR either succeeded or failed
2132                          * with a status error */
2133                         if(!spintime && !scsi_status_is_good(the_result)) {
2134                                 sd_print_result(sdkp, "Test Unit Ready failed",
2135                                                 the_result);
2136                         }
2137                         break;
2138                 }
2139
2140                 /*
2141                  * The device does not want the automatic start to be issued.
2142                  */
2143                 if (sdkp->device->no_start_on_add)
2144                         break;
2145
2146                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2147                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2148                                 break;  /* manual intervention required */
2149                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2150                                 break;  /* standby */
2151                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2152                                 break;  /* unavailable */
2153                         /*
2154                          * Issue command to spin up drive when not ready
2155                          */
2156                         if (!spintime) {
2157                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2158                                 cmd[0] = START_STOP;
2159                                 cmd[1] = 1;     /* Return immediately */
2160                                 memset((void *) &cmd[2], 0, 8);
2161                                 cmd[4] = 1;     /* Start spin cycle */
2162                                 if (sdkp->device->start_stop_pwr_cond)
2163                                         cmd[4] |= 1 << 4;
2164                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2165                                                  NULL, 0, &sshdr,
2166                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2167                                                  NULL);
2168                                 spintime_expire = jiffies + 100 * HZ;
2169                                 spintime = 1;
2170                         }
2171                         /* Wait 1 second for next try */
2172                         msleep(1000);
2173                         printk(KERN_CONT ".");
2174
2175                 /*
2176                  * Wait for USB flash devices with slow firmware.
2177                  * Yes, this sense key/ASC combination shouldn't
2178                  * occur here.  It's characteristic of these devices.
2179                  */
2180                 } else if (sense_valid &&
2181                                 sshdr.sense_key == UNIT_ATTENTION &&
2182                                 sshdr.asc == 0x28) {
2183                         if (!spintime) {
2184                                 spintime_expire = jiffies + 5 * HZ;
2185                                 spintime = 1;
2186                         }
2187                         /* Wait 1 second for next try */
2188                         msleep(1000);
2189                 } else {
2190                         /* we don't understand the sense code, so it's
2191                          * probably pointless to loop */
2192                         if(!spintime) {
2193                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2194                                 sd_print_sense_hdr(sdkp, &sshdr);
2195                         }
2196                         break;
2197                 }
2198                                 
2199         } while (spintime && time_before_eq(jiffies, spintime_expire));
2200
2201         if (spintime) {
2202                 if (scsi_status_is_good(the_result))
2203                         printk(KERN_CONT "ready\n");
2204                 else
2205                         printk(KERN_CONT "not responding...\n");
2206         }
2207 }
2208
2209 /*
2210  * Determine whether disk supports Data Integrity Field.
2211  */
2212 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2213 {
2214         struct scsi_device *sdp = sdkp->device;
2215         u8 type;
2216         int ret = 0;
2217
2218         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2219                 return ret;
2220
2221         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2222
2223         if (type > T10_PI_TYPE3_PROTECTION)
2224                 ret = -ENODEV;
2225         else if (scsi_host_dif_capable(sdp->host, type))
2226                 ret = 1;
2227
2228         if (sdkp->first_scan || type != sdkp->protection_type)
2229                 switch (ret) {
2230                 case -ENODEV:
2231                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2232                                   " protection type %u. Disabling disk!\n",
2233                                   type);
2234                         break;
2235                 case 1:
2236                         sd_printk(KERN_NOTICE, sdkp,
2237                                   "Enabling DIF Type %u protection\n", type);
2238                         break;
2239                 case 0:
2240                         sd_printk(KERN_NOTICE, sdkp,
2241                                   "Disabling DIF Type %u protection\n", type);
2242                         break;
2243                 }
2244
2245         sdkp->protection_type = type;
2246
2247         return ret;
2248 }
2249
2250 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2251                         struct scsi_sense_hdr *sshdr, int sense_valid,
2252                         int the_result)
2253 {
2254         if (driver_byte(the_result) & DRIVER_SENSE)
2255                 sd_print_sense_hdr(sdkp, sshdr);
2256         else
2257                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2258
2259         /*
2260          * Set dirty bit for removable devices if not ready -
2261          * sometimes drives will not report this properly.
2262          */
2263         if (sdp->removable &&
2264             sense_valid && sshdr->sense_key == NOT_READY)
2265                 set_media_not_present(sdkp);
2266
2267         /*
2268          * We used to set media_present to 0 here to indicate no media
2269          * in the drive, but some drives fail read capacity even with
2270          * media present, so we can't do that.
2271          */
2272         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2273 }
2274
2275 #define RC16_LEN 32
2276 #if RC16_LEN > SD_BUF_SIZE
2277 #error RC16_LEN must not be more than SD_BUF_SIZE
2278 #endif
2279
2280 #define READ_CAPACITY_RETRIES_ON_RESET  10
2281
2282 /*
2283  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2284  * and the reported logical block size is bigger than 512 bytes. Note
2285  * that last_sector is a u64 and therefore logical_to_sectors() is not
2286  * applicable.
2287  */
2288 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2289 {
2290         u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2291
2292         if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2293                 return false;
2294
2295         return true;
2296 }
2297
2298 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2299                                                 unsigned char *buffer)
2300 {
2301         unsigned char cmd[16];
2302         struct scsi_sense_hdr sshdr;
2303         int sense_valid = 0;
2304         int the_result;
2305         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2306         unsigned int alignment;
2307         unsigned long long lba;
2308         unsigned sector_size;
2309
2310         if (sdp->no_read_capacity_16)
2311                 return -EINVAL;
2312
2313         do {
2314                 memset(cmd, 0, 16);
2315                 cmd[0] = SERVICE_ACTION_IN_16;
2316                 cmd[1] = SAI_READ_CAPACITY_16;
2317                 cmd[13] = RC16_LEN;
2318                 memset(buffer, 0, RC16_LEN);
2319
2320                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2321                                         buffer, RC16_LEN, &sshdr,
2322                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2323
2324                 if (media_not_present(sdkp, &sshdr))
2325                         return -ENODEV;
2326
2327                 if (the_result) {
2328                         sense_valid = scsi_sense_valid(&sshdr);
2329                         if (sense_valid &&
2330                             sshdr.sense_key == ILLEGAL_REQUEST &&
2331                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2332                             sshdr.ascq == 0x00)
2333                                 /* Invalid Command Operation Code or
2334                                  * Invalid Field in CDB, just retry
2335                                  * silently with RC10 */
2336                                 return -EINVAL;
2337                         if (sense_valid &&
2338                             sshdr.sense_key == UNIT_ATTENTION &&
2339                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2340                                 /* Device reset might occur several times,
2341                                  * give it one more chance */
2342                                 if (--reset_retries > 0)
2343                                         continue;
2344                 }
2345                 retries--;
2346
2347         } while (the_result && retries);
2348
2349         if (the_result) {
2350                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2351                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2352                 return -EINVAL;
2353         }
2354
2355         sector_size = get_unaligned_be32(&buffer[8]);
2356         lba = get_unaligned_be64(&buffer[0]);
2357
2358         if (sd_read_protection_type(sdkp, buffer) < 0) {
2359                 sdkp->capacity = 0;
2360                 return -ENODEV;
2361         }
2362
2363         if (!sd_addressable_capacity(lba, sector_size)) {
2364                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2365                         "kernel compiled with support for large block "
2366                         "devices.\n");
2367                 sdkp->capacity = 0;
2368                 return -EOVERFLOW;
2369         }
2370
2371         /* Logical blocks per physical block exponent */
2372         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2373
2374         /* RC basis */
2375         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2376
2377         /* Lowest aligned logical block */
2378         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2379         blk_queue_alignment_offset(sdp->request_queue, alignment);
2380         if (alignment && sdkp->first_scan)
2381                 sd_printk(KERN_NOTICE, sdkp,
2382                           "physical block alignment offset: %u\n", alignment);
2383
2384         if (buffer[14] & 0x80) { /* LBPME */
2385                 sdkp->lbpme = 1;
2386
2387                 if (buffer[14] & 0x40) /* LBPRZ */
2388                         sdkp->lbprz = 1;
2389
2390                 sd_config_discard(sdkp, SD_LBP_WS16);
2391         }
2392
2393         sdkp->capacity = lba + 1;
2394         return sector_size;
2395 }
2396
2397 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2398                                                 unsigned char *buffer)
2399 {
2400         unsigned char cmd[16];
2401         struct scsi_sense_hdr sshdr;
2402         int sense_valid = 0;
2403         int the_result;
2404         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2405         sector_t lba;
2406         unsigned sector_size;
2407
2408         do {
2409                 cmd[0] = READ_CAPACITY;
2410                 memset(&cmd[1], 0, 9);
2411                 memset(buffer, 0, 8);
2412
2413                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2414                                         buffer, 8, &sshdr,
2415                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2416
2417                 if (media_not_present(sdkp, &sshdr))
2418                         return -ENODEV;
2419
2420                 if (the_result) {
2421                         sense_valid = scsi_sense_valid(&sshdr);
2422                         if (sense_valid &&
2423                             sshdr.sense_key == UNIT_ATTENTION &&
2424                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2425                                 /* Device reset might occur several times,
2426                                  * give it one more chance */
2427                                 if (--reset_retries > 0)
2428                                         continue;
2429                 }
2430                 retries--;
2431
2432         } while (the_result && retries);
2433
2434         if (the_result) {
2435                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2436                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2437                 return -EINVAL;
2438         }
2439
2440         sector_size = get_unaligned_be32(&buffer[4]);
2441         lba = get_unaligned_be32(&buffer[0]);
2442
2443         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2444                 /* Some buggy (usb cardreader) devices return an lba of
2445                    0xffffffff when the want to report a size of 0 (with
2446                    which they really mean no media is present) */
2447                 sdkp->capacity = 0;
2448                 sdkp->physical_block_size = sector_size;
2449                 return sector_size;
2450         }
2451
2452         if (!sd_addressable_capacity(lba, sector_size)) {
2453                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2454                         "kernel compiled with support for large block "
2455                         "devices.\n");
2456                 sdkp->capacity = 0;
2457                 return -EOVERFLOW;
2458         }
2459
2460         sdkp->capacity = lba + 1;
2461         sdkp->physical_block_size = sector_size;
2462         return sector_size;
2463 }
2464
2465 static int sd_try_rc16_first(struct scsi_device *sdp)
2466 {
2467         if (sdp->host->max_cmd_len < 16)
2468                 return 0;
2469         if (sdp->try_rc_10_first)
2470                 return 0;
2471         if (sdp->scsi_level > SCSI_SPC_2)
2472                 return 1;
2473         if (scsi_device_protection(sdp))
2474                 return 1;
2475         return 0;
2476 }
2477
2478 /*
2479  * read disk capacity
2480  */
2481 static void
2482 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2483 {
2484         int sector_size;
2485         struct scsi_device *sdp = sdkp->device;
2486
2487         if (sd_try_rc16_first(sdp)) {
2488                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2489                 if (sector_size == -EOVERFLOW)
2490                         goto got_data;
2491                 if (sector_size == -ENODEV)
2492                         return;
2493                 if (sector_size < 0)
2494                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2495                 if (sector_size < 0)
2496                         return;
2497         } else {
2498                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2499                 if (sector_size == -EOVERFLOW)
2500                         goto got_data;
2501                 if (sector_size < 0)
2502                         return;
2503                 if ((sizeof(sdkp->capacity) > 4) &&
2504                     (sdkp->capacity > 0xffffffffULL)) {
2505                         int old_sector_size = sector_size;
2506                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2507                                         "Trying to use READ CAPACITY(16).\n");
2508                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2509                         if (sector_size < 0) {
2510                                 sd_printk(KERN_NOTICE, sdkp,
2511                                         "Using 0xffffffff as device size\n");
2512                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2513                                 sector_size = old_sector_size;
2514                                 goto got_data;
2515                         }
2516                 }
2517         }
2518
2519         /* Some devices are known to return the total number of blocks,
2520          * not the highest block number.  Some devices have versions
2521          * which do this and others which do not.  Some devices we might
2522          * suspect of doing this but we don't know for certain.
2523          *
2524          * If we know the reported capacity is wrong, decrement it.  If
2525          * we can only guess, then assume the number of blocks is even
2526          * (usually true but not always) and err on the side of lowering
2527          * the capacity.
2528          */
2529         if (sdp->fix_capacity ||
2530             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2531                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2532                                 "from its reported value: %llu\n",
2533                                 (unsigned long long) sdkp->capacity);
2534                 --sdkp->capacity;
2535         }
2536
2537 got_data:
2538         if (sector_size == 0) {
2539                 sector_size = 512;
2540                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2541                           "assuming 512.\n");
2542         }
2543
2544         if (sector_size != 512 &&
2545             sector_size != 1024 &&
2546             sector_size != 2048 &&
2547             sector_size != 4096) {
2548                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2549                           sector_size);
2550                 /*
2551                  * The user might want to re-format the drive with
2552                  * a supported sectorsize.  Once this happens, it
2553                  * would be relatively trivial to set the thing up.
2554                  * For this reason, we leave the thing in the table.
2555                  */
2556                 sdkp->capacity = 0;
2557                 /*
2558                  * set a bogus sector size so the normal read/write
2559                  * logic in the block layer will eventually refuse any
2560                  * request on this device without tripping over power
2561                  * of two sector size assumptions
2562                  */
2563                 sector_size = 512;
2564         }
2565         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2566         blk_queue_physical_block_size(sdp->request_queue,
2567                                       sdkp->physical_block_size);
2568         sdkp->device->sector_size = sector_size;
2569
2570         if (sdkp->capacity > 0xffffffff)
2571                 sdp->use_16_for_rw = 1;
2572
2573 }
2574
2575 /*
2576  * Print disk capacity
2577  */
2578 static void
2579 sd_print_capacity(struct scsi_disk *sdkp,
2580                   sector_t old_capacity)
2581 {
2582         int sector_size = sdkp->device->sector_size;
2583         char cap_str_2[10], cap_str_10[10];
2584
2585         string_get_size(sdkp->capacity, sector_size,
2586                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2587         string_get_size(sdkp->capacity, sector_size,
2588                         STRING_UNITS_10, cap_str_10,
2589                         sizeof(cap_str_10));
2590
2591         if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2592                 sd_printk(KERN_NOTICE, sdkp,
2593                           "%llu %d-byte logical blocks: (%s/%s)\n",
2594                           (unsigned long long)sdkp->capacity,
2595                           sector_size, cap_str_10, cap_str_2);
2596
2597                 if (sdkp->physical_block_size != sector_size)
2598                         sd_printk(KERN_NOTICE, sdkp,
2599                                   "%u-byte physical blocks\n",
2600                                   sdkp->physical_block_size);
2601
2602                 sd_zbc_print_zones(sdkp);
2603         }
2604 }
2605
2606 /* called with buffer of length 512 */
2607 static inline int
2608 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2609                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2610                  struct scsi_sense_hdr *sshdr)
2611 {
2612         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2613                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2614                                sshdr);
2615 }
2616
2617 /*
2618  * read write protect setting, if possible - called only in sd_revalidate_disk()
2619  * called with buffer of length SD_BUF_SIZE
2620  */
2621 static void
2622 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2623 {
2624         int res;
2625         struct scsi_device *sdp = sdkp->device;
2626         struct scsi_mode_data data;
2627         int disk_ro = get_disk_ro(sdkp->disk);
2628         int old_wp = sdkp->write_prot;
2629
2630         set_disk_ro(sdkp->disk, 0);
2631         if (sdp->skip_ms_page_3f) {
2632                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2633                 return;
2634         }
2635
2636         if (sdp->use_192_bytes_for_3f) {
2637                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2638         } else {
2639                 /*
2640                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2641                  * We have to start carefully: some devices hang if we ask
2642                  * for more than is available.
2643                  */
2644                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2645
2646                 /*
2647                  * Second attempt: ask for page 0 When only page 0 is
2648                  * implemented, a request for page 3F may return Sense Key
2649                  * 5: Illegal Request, Sense Code 24: Invalid field in
2650                  * CDB.
2651                  */
2652                 if (!scsi_status_is_good(res))
2653                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2654
2655                 /*
2656                  * Third attempt: ask 255 bytes, as we did earlier.
2657                  */
2658                 if (!scsi_status_is_good(res))
2659                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2660                                                &data, NULL);
2661         }
2662
2663         if (!scsi_status_is_good(res)) {
2664                 sd_first_printk(KERN_WARNING, sdkp,
2665                           "Test WP failed, assume Write Enabled\n");
2666         } else {
2667                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2668                 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2669                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2670                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2671                                   sdkp->write_prot ? "on" : "off");
2672                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2673                 }
2674         }
2675 }
2676
2677 /*
2678  * sd_read_cache_type - called only from sd_revalidate_disk()
2679  * called with buffer of length SD_BUF_SIZE
2680  */
2681 static void
2682 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2683 {
2684         int len = 0, res;
2685         struct scsi_device *sdp = sdkp->device;
2686
2687         int dbd;
2688         int modepage;
2689         int first_len;
2690         struct scsi_mode_data data;
2691         struct scsi_sense_hdr sshdr;
2692         int old_wce = sdkp->WCE;
2693         int old_rcd = sdkp->RCD;
2694         int old_dpofua = sdkp->DPOFUA;
2695
2696
2697         if (sdkp->cache_override)
2698                 return;
2699
2700         first_len = 4;
2701         if (sdp->skip_ms_page_8) {
2702                 if (sdp->type == TYPE_RBC)
2703                         goto defaults;
2704                 else {
2705                         if (sdp->skip_ms_page_3f)
2706                                 goto defaults;
2707                         modepage = 0x3F;
2708                         if (sdp->use_192_bytes_for_3f)
2709                                 first_len = 192;
2710                         dbd = 0;
2711                 }
2712         } else if (sdp->type == TYPE_RBC) {
2713                 modepage = 6;
2714                 dbd = 8;
2715         } else {
2716                 modepage = 8;
2717                 dbd = 0;
2718         }
2719
2720         /* cautiously ask */
2721         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2722                         &data, &sshdr);
2723
2724         if (!scsi_status_is_good(res))
2725                 goto bad_sense;
2726
2727         if (!data.header_length) {
2728                 modepage = 6;
2729                 first_len = 0;
2730                 sd_first_printk(KERN_ERR, sdkp,
2731                                 "Missing header in MODE_SENSE response\n");
2732         }
2733
2734         /* that went OK, now ask for the proper length */
2735         len = data.length;
2736
2737         /*
2738          * We're only interested in the first three bytes, actually.
2739          * But the data cache page is defined for the first 20.
2740          */
2741         if (len < 3)
2742                 goto bad_sense;
2743         else if (len > SD_BUF_SIZE) {
2744                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2745                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2746                 len = SD_BUF_SIZE;
2747         }
2748         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2749                 len = 192;
2750
2751         /* Get the data */
2752         if (len > first_len)
2753                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2754                                 &data, &sshdr);
2755
2756         if (scsi_status_is_good(res)) {
2757                 int offset = data.header_length + data.block_descriptor_length;
2758
2759                 while (offset < len) {
2760                         u8 page_code = buffer[offset] & 0x3F;
2761                         u8 spf       = buffer[offset] & 0x40;
2762
2763                         if (page_code == 8 || page_code == 6) {
2764                                 /* We're interested only in the first 3 bytes.
2765                                  */
2766                                 if (len - offset <= 2) {
2767                                         sd_first_printk(KERN_ERR, sdkp,
2768                                                 "Incomplete mode parameter "
2769                                                         "data\n");
2770                                         goto defaults;
2771                                 } else {
2772                                         modepage = page_code;
2773                                         goto Page_found;
2774                                 }
2775                         } else {
2776                                 /* Go to the next page */
2777                                 if (spf && len - offset > 3)
2778                                         offset += 4 + (buffer[offset+2] << 8) +
2779                                                 buffer[offset+3];
2780                                 else if (!spf && len - offset > 1)
2781                                         offset += 2 + buffer[offset+1];
2782                                 else {
2783                                         sd_first_printk(KERN_ERR, sdkp,
2784                                                         "Incomplete mode "
2785                                                         "parameter data\n");
2786                                         goto defaults;
2787                                 }
2788                         }
2789                 }
2790
2791                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2792                 goto defaults;
2793
2794         Page_found:
2795                 if (modepage == 8) {
2796                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2797                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2798                 } else {
2799                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2800                         sdkp->RCD = 0;
2801                 }
2802
2803                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2804                 if (sdp->broken_fua) {
2805                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2806                         sdkp->DPOFUA = 0;
2807                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2808                            !sdkp->device->use_16_for_rw) {
2809                         sd_first_printk(KERN_NOTICE, sdkp,
2810                                   "Uses READ/WRITE(6), disabling FUA\n");
2811                         sdkp->DPOFUA = 0;
2812                 }
2813
2814                 /* No cache flush allowed for write protected devices */
2815                 if (sdkp->WCE && sdkp->write_prot)
2816                         sdkp->WCE = 0;
2817
2818                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2819                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2820                         sd_printk(KERN_NOTICE, sdkp,
2821                                   "Write cache: %s, read cache: %s, %s\n",
2822                                   sdkp->WCE ? "enabled" : "disabled",
2823                                   sdkp->RCD ? "disabled" : "enabled",
2824                                   sdkp->DPOFUA ? "supports DPO and FUA"
2825                                   : "doesn't support DPO or FUA");
2826
2827                 return;
2828         }
2829
2830 bad_sense:
2831         if (scsi_sense_valid(&sshdr) &&
2832             sshdr.sense_key == ILLEGAL_REQUEST &&
2833             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2834                 /* Invalid field in CDB */
2835                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2836         else
2837                 sd_first_printk(KERN_ERR, sdkp,
2838                                 "Asking for cache data failed\n");
2839
2840 defaults:
2841         if (sdp->wce_default_on) {
2842                 sd_first_printk(KERN_NOTICE, sdkp,
2843                                 "Assuming drive cache: write back\n");
2844                 sdkp->WCE = 1;
2845         } else {
2846                 sd_first_printk(KERN_ERR, sdkp,
2847                                 "Assuming drive cache: write through\n");
2848                 sdkp->WCE = 0;
2849         }
2850         sdkp->RCD = 0;
2851         sdkp->DPOFUA = 0;
2852 }
2853
2854 /*
2855  * The ATO bit indicates whether the DIF application tag is available
2856  * for use by the operating system.
2857  */
2858 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2859 {
2860         int res, offset;
2861         struct scsi_device *sdp = sdkp->device;
2862         struct scsi_mode_data data;
2863         struct scsi_sense_hdr sshdr;
2864
2865         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2866                 return;
2867
2868         if (sdkp->protection_type == 0)
2869                 return;
2870
2871         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2872                               SD_MAX_RETRIES, &data, &sshdr);
2873
2874         if (!scsi_status_is_good(res) || !data.header_length ||
2875             data.length < 6) {
2876                 sd_first_printk(KERN_WARNING, sdkp,
2877                           "getting Control mode page failed, assume no ATO\n");
2878
2879                 if (scsi_sense_valid(&sshdr))
2880                         sd_print_sense_hdr(sdkp, &sshdr);
2881
2882                 return;
2883         }
2884
2885         offset = data.header_length + data.block_descriptor_length;
2886
2887         if ((buffer[offset] & 0x3f) != 0x0a) {
2888                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2889                 return;
2890         }
2891
2892         if ((buffer[offset + 5] & 0x80) == 0)
2893                 return;
2894
2895         sdkp->ATO = 1;
2896
2897         return;
2898 }
2899
2900 /**
2901  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2902  * @sdkp: disk to query
2903  */
2904 static void sd_read_block_limits(struct scsi_disk *sdkp)
2905 {
2906         unsigned int sector_sz = sdkp->device->sector_size;
2907         const int vpd_len = 64;
2908         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2909
2910         if (!buffer ||
2911             /* Block Limits VPD */
2912             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2913                 goto out;
2914
2915         blk_queue_io_min(sdkp->disk->queue,
2916                          get_unaligned_be16(&buffer[6]) * sector_sz);
2917
2918         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2919         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2920
2921         if (buffer[3] == 0x3c) {
2922                 unsigned int lba_count, desc_count;
2923
2924                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2925
2926                 if (!sdkp->lbpme)
2927                         goto out;
2928
2929                 lba_count = get_unaligned_be32(&buffer[20]);
2930                 desc_count = get_unaligned_be32(&buffer[24]);
2931
2932                 if (lba_count && desc_count)
2933                         sdkp->max_unmap_blocks = lba_count;
2934
2935                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2936
2937                 if (buffer[32] & 0x80)
2938                         sdkp->unmap_alignment =
2939                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2940
2941                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2942
2943                         if (sdkp->max_unmap_blocks)
2944                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2945                         else
2946                                 sd_config_discard(sdkp, SD_LBP_WS16);
2947
2948                 } else {        /* LBP VPD page tells us what to use */
2949                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2950                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2951                         else if (sdkp->lbpws)
2952                                 sd_config_discard(sdkp, SD_LBP_WS16);
2953                         else if (sdkp->lbpws10)
2954                                 sd_config_discard(sdkp, SD_LBP_WS10);
2955                         else
2956                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2957                 }
2958         }
2959
2960  out:
2961         kfree(buffer);
2962 }
2963
2964 /**
2965  * sd_read_block_characteristics - Query block dev. characteristics
2966  * @sdkp: disk to query
2967  */
2968 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2969 {
2970         struct request_queue *q = sdkp->disk->queue;
2971         unsigned char *buffer;
2972         u16 rot;
2973         const int vpd_len = 64;
2974
2975         buffer = kmalloc(vpd_len, GFP_KERNEL);
2976
2977         if (!buffer ||
2978             /* Block Device Characteristics VPD */
2979             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2980                 goto out;
2981
2982         rot = get_unaligned_be16(&buffer[4]);
2983
2984         if (rot == 1) {
2985                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2986                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2987         }
2988
2989         if (sdkp->device->type == TYPE_ZBC) {
2990                 /* Host-managed */
2991                 q->limits.zoned = BLK_ZONED_HM;
2992         } else {
2993                 sdkp->zoned = (buffer[8] >> 4) & 3;
2994                 if (sdkp->zoned == 1)
2995                         /* Host-aware */
2996                         q->limits.zoned = BLK_ZONED_HA;
2997                 else
2998                         /*
2999                          * Treat drive-managed devices as
3000                          * regular block devices.
3001                          */
3002                         q->limits.zoned = BLK_ZONED_NONE;
3003         }
3004         if (blk_queue_is_zoned(q) && sdkp->first_scan)
3005                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3006                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3007
3008  out:
3009         kfree(buffer);
3010 }
3011
3012 /**
3013  * sd_read_block_provisioning - Query provisioning VPD page
3014  * @sdkp: disk to query
3015  */
3016 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3017 {
3018         unsigned char *buffer;
3019         const int vpd_len = 8;
3020
3021         if (sdkp->lbpme == 0)
3022                 return;
3023
3024         buffer = kmalloc(vpd_len, GFP_KERNEL);
3025
3026         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3027                 goto out;
3028
3029         sdkp->lbpvpd    = 1;
3030         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3031         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3032         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3033
3034  out:
3035         kfree(buffer);
3036 }
3037
3038 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3039 {
3040         struct scsi_device *sdev = sdkp->device;
3041
3042         if (sdev->host->no_write_same) {
3043                 sdev->no_write_same = 1;
3044
3045                 return;
3046         }
3047
3048         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3049                 /* too large values might cause issues with arcmsr */
3050                 int vpd_buf_len = 64;
3051
3052                 sdev->no_report_opcodes = 1;
3053
3054                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3055                  * CODES is unsupported and the device has an ATA
3056                  * Information VPD page (SAT).
3057                  */
3058                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3059                         sdev->no_write_same = 1;
3060         }
3061
3062         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3063                 sdkp->ws16 = 1;
3064
3065         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3066                 sdkp->ws10 = 1;
3067 }
3068
3069 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3070 {
3071         struct scsi_device *sdev = sdkp->device;
3072
3073         if (!sdev->security_supported)
3074                 return;
3075
3076         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3077                         SECURITY_PROTOCOL_IN) == 1 &&
3078             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3079                         SECURITY_PROTOCOL_OUT) == 1)
3080                 sdkp->security = 1;
3081 }
3082
3083 /**
3084  *      sd_revalidate_disk - called the first time a new disk is seen,
3085  *      performs disk spin up, read_capacity, etc.
3086  *      @disk: struct gendisk we care about
3087  **/
3088 static int sd_revalidate_disk(struct gendisk *disk)
3089 {
3090         struct scsi_disk *sdkp = scsi_disk(disk);
3091         struct scsi_device *sdp = sdkp->device;
3092         struct request_queue *q = sdkp->disk->queue;
3093         sector_t old_capacity = sdkp->capacity;
3094         unsigned char *buffer;
3095         unsigned int dev_max, rw_max;
3096
3097         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3098                                       "sd_revalidate_disk\n"));
3099
3100         /*
3101          * If the device is offline, don't try and read capacity or any
3102          * of the other niceties.
3103          */
3104         if (!scsi_device_online(sdp))
3105                 goto out;
3106
3107         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3108         if (!buffer) {
3109                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3110                           "allocation failure.\n");
3111                 goto out;
3112         }
3113
3114         sd_spinup_disk(sdkp);
3115
3116         /*
3117          * Without media there is no reason to ask; moreover, some devices
3118          * react badly if we do.
3119          */
3120         if (sdkp->media_present) {
3121                 sd_read_capacity(sdkp, buffer);
3122
3123                 if (scsi_device_supports_vpd(sdp)) {
3124                         sd_read_block_provisioning(sdkp);
3125                         sd_read_block_limits(sdkp);
3126                         sd_read_block_characteristics(sdkp);
3127                         sd_zbc_read_zones(sdkp, buffer);
3128                 }
3129
3130                 sd_print_capacity(sdkp, old_capacity);
3131
3132                 sd_read_write_protect_flag(sdkp, buffer);
3133                 sd_read_cache_type(sdkp, buffer);
3134                 sd_read_app_tag_own(sdkp, buffer);
3135                 sd_read_write_same(sdkp, buffer);
3136                 sd_read_security(sdkp, buffer);
3137         }
3138
3139         /*
3140          * We now have all cache related info, determine how we deal
3141          * with flush requests.
3142          */
3143         sd_set_flush_flag(sdkp);
3144
3145         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3146         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3147
3148         /* Some devices report a maximum block count for READ/WRITE requests. */
3149         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3150         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3151
3152         /*
3153          * Determine the device's preferred I/O size for reads and writes
3154          * unless the reported value is unreasonably small, large, or
3155          * garbage.
3156          */
3157         if (sdkp->opt_xfer_blocks &&
3158             sdkp->opt_xfer_blocks <= dev_max &&
3159             sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3160             logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3161                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3162                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3163         } else
3164                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3165                                       (sector_t)BLK_DEF_MAX_SECTORS);
3166
3167         /* Do not exceed controller limit */
3168         rw_max = min(rw_max, queue_max_hw_sectors(q));
3169
3170         /*
3171          * Only update max_sectors if previously unset or if the current value
3172          * exceeds the capabilities of the hardware.
3173          */
3174         if (sdkp->first_scan ||
3175             q->limits.max_sectors > q->limits.max_dev_sectors ||
3176             q->limits.max_sectors > q->limits.max_hw_sectors)
3177                 q->limits.max_sectors = rw_max;
3178
3179         sdkp->first_scan = 0;
3180
3181         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3182         sd_config_write_same(sdkp);
3183         kfree(buffer);
3184
3185  out:
3186         return 0;
3187 }
3188
3189 /**
3190  *      sd_unlock_native_capacity - unlock native capacity
3191  *      @disk: struct gendisk to set capacity for
3192  *
3193  *      Block layer calls this function if it detects that partitions
3194  *      on @disk reach beyond the end of the device.  If the SCSI host
3195  *      implements ->unlock_native_capacity() method, it's invoked to
3196  *      give it a chance to adjust the device capacity.
3197  *
3198  *      CONTEXT:
3199  *      Defined by block layer.  Might sleep.
3200  */
3201 static void sd_unlock_native_capacity(struct gendisk *disk)
3202 {
3203         struct scsi_device *sdev = scsi_disk(disk)->device;
3204
3205         if (sdev->host->hostt->unlock_native_capacity)
3206                 sdev->host->hostt->unlock_native_capacity(sdev);
3207 }
3208
3209 /**
3210  *      sd_format_disk_name - format disk name
3211  *      @prefix: name prefix - ie. "sd" for SCSI disks
3212  *      @index: index of the disk to format name for
3213  *      @buf: output buffer
3214  *      @buflen: length of the output buffer
3215  *
3216  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3217  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3218  *      which is followed by sdaaa.
3219  *
3220  *      This is basically 26 base counting with one extra 'nil' entry
3221  *      at the beginning from the second digit on and can be
3222  *      determined using similar method as 26 base conversion with the
3223  *      index shifted -1 after each digit is computed.
3224  *
3225  *      CONTEXT:
3226  *      Don't care.
3227 &n