Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 *
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
26 *
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
34 */
35
36#include <linux/module.h>
37#include <linux/fs.h>
38#include <linux/kernel.h>
39#include <linux/mm.h>
40#include <linux/bio.h>
41#include <linux/hdreg.h>
42#include <linux/errno.h>
43#include <linux/idr.h>
44#include <linux/interrupt.h>
45#include <linux/init.h>
46#include <linux/blkdev.h>
47#include <linux/blkpg.h>
48#include <linux/blk-pm.h>
49#include <linux/delay.h>
50#include <linux/rw_hint.h>
51#include <linux/major.h>
52#include <linux/mutex.h>
53#include <linux/string_helpers.h>
54#include <linux/slab.h>
55#include <linux/sed-opal.h>
56#include <linux/pm_runtime.h>
57#include <linux/pr.h>
58#include <linux/t10-pi.h>
59#include <linux/uaccess.h>
60#include <asm/unaligned.h>
61
62#include <scsi/scsi.h>
63#include <scsi/scsi_cmnd.h>
64#include <scsi/scsi_dbg.h>
65#include <scsi/scsi_device.h>
66#include <scsi/scsi_driver.h>
67#include <scsi/scsi_eh.h>
68#include <scsi/scsi_host.h>
69#include <scsi/scsi_ioctl.h>
70#include <scsi/scsicam.h>
71#include <scsi/scsi_common.h>
72
73#include "sd.h"
74#include "scsi_priv.h"
75#include "scsi_logging.h"
76
77MODULE_AUTHOR("Eric Youngdale");
78MODULE_DESCRIPTION("SCSI disk (sd) driver");
79MODULE_LICENSE("GPL");
80
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
94MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
95MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
96MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
97MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
98MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
99MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
100MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
101
102#define SD_MINORS 16
103
104static void sd_config_discard(struct scsi_disk *, unsigned int);
105static void sd_config_write_same(struct scsi_disk *);
106static int sd_revalidate_disk(struct gendisk *);
107static void sd_unlock_native_capacity(struct gendisk *disk);
108static void sd_shutdown(struct device *);
109static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
110static void scsi_disk_release(struct device *cdev);
111
112static DEFINE_IDA(sd_index_ida);
113
114static mempool_t *sd_page_pool;
115static struct lock_class_key sd_bio_compl_lkclass;
116
117static const char *sd_cache_types[] = {
118 "write through", "none", "write back",
119 "write back, no read (daft)"
120};
121
122static void sd_set_flush_flag(struct scsi_disk *sdkp)
123{
124 bool wc = false, fua = false;
125
126 if (sdkp->WCE) {
127 wc = true;
128 if (sdkp->DPOFUA)
129 fua = true;
130 }
131
132 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
133}
134
135static ssize_t
136cache_type_store(struct device *dev, struct device_attribute *attr,
137 const char *buf, size_t count)
138{
139 int ct, rcd, wce, sp;
140 struct scsi_disk *sdkp = to_scsi_disk(dev);
141 struct scsi_device *sdp = sdkp->device;
142 char buffer[64];
143 char *buffer_data;
144 struct scsi_mode_data data;
145 struct scsi_sense_hdr sshdr;
146 static const char temp[] = "temporary ";
147 int len, ret;
148
149 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
150 /* no cache control on RBC devices; theoretically they
151 * can do it, but there's probably so many exceptions
152 * it's not worth the risk */
153 return -EINVAL;
154
155 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
156 buf += sizeof(temp) - 1;
157 sdkp->cache_override = 1;
158 } else {
159 sdkp->cache_override = 0;
160 }
161
162 ct = sysfs_match_string(sd_cache_types, buf);
163 if (ct < 0)
164 return -EINVAL;
165
166 rcd = ct & 0x01 ? 1 : 0;
167 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
168
169 if (sdkp->cache_override) {
170 sdkp->WCE = wce;
171 sdkp->RCD = rcd;
172 sd_set_flush_flag(sdkp);
173 return count;
174 }
175
176 if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
177 sdkp->max_retries, &data, NULL))
178 return -EINVAL;
179 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
180 data.block_descriptor_length);
181 buffer_data = buffer + data.header_length +
182 data.block_descriptor_length;
183 buffer_data[2] &= ~0x05;
184 buffer_data[2] |= wce << 2 | rcd;
185 sp = buffer_data[0] & 0x80 ? 1 : 0;
186 buffer_data[0] &= ~0x80;
187
188 /*
189 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
190 * received mode parameter buffer before doing MODE SELECT.
191 */
192 data.device_specific = 0;
193
194 ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
195 sdkp->max_retries, &data, &sshdr);
196 if (ret) {
197 if (ret > 0 && scsi_sense_valid(&sshdr))
198 sd_print_sense_hdr(sdkp, &sshdr);
199 return -EINVAL;
200 }
201 sd_revalidate_disk(sdkp->disk);
202 return count;
203}
204
205static ssize_t
206manage_start_stop_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
208{
209 struct scsi_disk *sdkp = to_scsi_disk(dev);
210 struct scsi_device *sdp = sdkp->device;
211
212 return sysfs_emit(buf, "%u\n",
213 sdp->manage_system_start_stop &&
214 sdp->manage_runtime_start_stop &&
215 sdp->manage_shutdown);
216}
217static DEVICE_ATTR_RO(manage_start_stop);
218
219static ssize_t
220manage_system_start_stop_show(struct device *dev,
221 struct device_attribute *attr, char *buf)
222{
223 struct scsi_disk *sdkp = to_scsi_disk(dev);
224 struct scsi_device *sdp = sdkp->device;
225
226 return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
227}
228
229static ssize_t
230manage_system_start_stop_store(struct device *dev,
231 struct device_attribute *attr,
232 const char *buf, size_t count)
233{
234 struct scsi_disk *sdkp = to_scsi_disk(dev);
235 struct scsi_device *sdp = sdkp->device;
236 bool v;
237
238 if (!capable(CAP_SYS_ADMIN))
239 return -EACCES;
240
241 if (kstrtobool(buf, &v))
242 return -EINVAL;
243
244 sdp->manage_system_start_stop = v;
245
246 return count;
247}
248static DEVICE_ATTR_RW(manage_system_start_stop);
249
250static ssize_t
251manage_runtime_start_stop_show(struct device *dev,
252 struct device_attribute *attr, char *buf)
253{
254 struct scsi_disk *sdkp = to_scsi_disk(dev);
255 struct scsi_device *sdp = sdkp->device;
256
257 return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
258}
259
260static ssize_t
261manage_runtime_start_stop_store(struct device *dev,
262 struct device_attribute *attr,
263 const char *buf, size_t count)
264{
265 struct scsi_disk *sdkp = to_scsi_disk(dev);
266 struct scsi_device *sdp = sdkp->device;
267 bool v;
268
269 if (!capable(CAP_SYS_ADMIN))
270 return -EACCES;
271
272 if (kstrtobool(buf, &v))
273 return -EINVAL;
274
275 sdp->manage_runtime_start_stop = v;
276
277 return count;
278}
279static DEVICE_ATTR_RW(manage_runtime_start_stop);
280
281static ssize_t manage_shutdown_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 struct scsi_disk *sdkp = to_scsi_disk(dev);
285 struct scsi_device *sdp = sdkp->device;
286
287 return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
288}
289
290static ssize_t manage_shutdown_store(struct device *dev,
291 struct device_attribute *attr,
292 const char *buf, size_t count)
293{
294 struct scsi_disk *sdkp = to_scsi_disk(dev);
295 struct scsi_device *sdp = sdkp->device;
296 bool v;
297
298 if (!capable(CAP_SYS_ADMIN))
299 return -EACCES;
300
301 if (kstrtobool(buf, &v))
302 return -EINVAL;
303
304 sdp->manage_shutdown = v;
305
306 return count;
307}
308static DEVICE_ATTR_RW(manage_shutdown);
309
310static ssize_t
311allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
312{
313 struct scsi_disk *sdkp = to_scsi_disk(dev);
314
315 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
316}
317
318static ssize_t
319allow_restart_store(struct device *dev, struct device_attribute *attr,
320 const char *buf, size_t count)
321{
322 bool v;
323 struct scsi_disk *sdkp = to_scsi_disk(dev);
324 struct scsi_device *sdp = sdkp->device;
325
326 if (!capable(CAP_SYS_ADMIN))
327 return -EACCES;
328
329 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
330 return -EINVAL;
331
332 if (kstrtobool(buf, &v))
333 return -EINVAL;
334
335 sdp->allow_restart = v;
336
337 return count;
338}
339static DEVICE_ATTR_RW(allow_restart);
340
341static ssize_t
342cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
343{
344 struct scsi_disk *sdkp = to_scsi_disk(dev);
345 int ct = sdkp->RCD + 2*sdkp->WCE;
346
347 return sprintf(buf, "%s\n", sd_cache_types[ct]);
348}
349static DEVICE_ATTR_RW(cache_type);
350
351static ssize_t
352FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
353{
354 struct scsi_disk *sdkp = to_scsi_disk(dev);
355
356 return sprintf(buf, "%u\n", sdkp->DPOFUA);
357}
358static DEVICE_ATTR_RO(FUA);
359
360static ssize_t
361protection_type_show(struct device *dev, struct device_attribute *attr,
362 char *buf)
363{
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
365
366 return sprintf(buf, "%u\n", sdkp->protection_type);
367}
368
369static ssize_t
370protection_type_store(struct device *dev, struct device_attribute *attr,
371 const char *buf, size_t count)
372{
373 struct scsi_disk *sdkp = to_scsi_disk(dev);
374 unsigned int val;
375 int err;
376
377 if (!capable(CAP_SYS_ADMIN))
378 return -EACCES;
379
380 err = kstrtouint(buf, 10, &val);
381
382 if (err)
383 return err;
384
385 if (val <= T10_PI_TYPE3_PROTECTION)
386 sdkp->protection_type = val;
387
388 return count;
389}
390static DEVICE_ATTR_RW(protection_type);
391
392static ssize_t
393protection_mode_show(struct device *dev, struct device_attribute *attr,
394 char *buf)
395{
396 struct scsi_disk *sdkp = to_scsi_disk(dev);
397 struct scsi_device *sdp = sdkp->device;
398 unsigned int dif, dix;
399
400 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
401 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
402
403 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
404 dif = 0;
405 dix = 1;
406 }
407
408 if (!dif && !dix)
409 return sprintf(buf, "none\n");
410
411 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
412}
413static DEVICE_ATTR_RO(protection_mode);
414
415static ssize_t
416app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
417{
418 struct scsi_disk *sdkp = to_scsi_disk(dev);
419
420 return sprintf(buf, "%u\n", sdkp->ATO);
421}
422static DEVICE_ATTR_RO(app_tag_own);
423
424static ssize_t
425thin_provisioning_show(struct device *dev, struct device_attribute *attr,
426 char *buf)
427{
428 struct scsi_disk *sdkp = to_scsi_disk(dev);
429
430 return sprintf(buf, "%u\n", sdkp->lbpme);
431}
432static DEVICE_ATTR_RO(thin_provisioning);
433
434/* sysfs_match_string() requires dense arrays */
435static const char *lbp_mode[] = {
436 [SD_LBP_FULL] = "full",
437 [SD_LBP_UNMAP] = "unmap",
438 [SD_LBP_WS16] = "writesame_16",
439 [SD_LBP_WS10] = "writesame_10",
440 [SD_LBP_ZERO] = "writesame_zero",
441 [SD_LBP_DISABLE] = "disabled",
442};
443
444static ssize_t
445provisioning_mode_show(struct device *dev, struct device_attribute *attr,
446 char *buf)
447{
448 struct scsi_disk *sdkp = to_scsi_disk(dev);
449
450 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
451}
452
453static ssize_t
454provisioning_mode_store(struct device *dev, struct device_attribute *attr,
455 const char *buf, size_t count)
456{
457 struct scsi_disk *sdkp = to_scsi_disk(dev);
458 struct scsi_device *sdp = sdkp->device;
459 int mode;
460
461 if (!capable(CAP_SYS_ADMIN))
462 return -EACCES;
463
464 if (sd_is_zoned(sdkp)) {
465 sd_config_discard(sdkp, SD_LBP_DISABLE);
466 return count;
467 }
468
469 if (sdp->type != TYPE_DISK)
470 return -EINVAL;
471
472 mode = sysfs_match_string(lbp_mode, buf);
473 if (mode < 0)
474 return -EINVAL;
475
476 sd_config_discard(sdkp, mode);
477
478 return count;
479}
480static DEVICE_ATTR_RW(provisioning_mode);
481
482/* sysfs_match_string() requires dense arrays */
483static const char *zeroing_mode[] = {
484 [SD_ZERO_WRITE] = "write",
485 [SD_ZERO_WS] = "writesame",
486 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
487 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
488};
489
490static ssize_t
491zeroing_mode_show(struct device *dev, struct device_attribute *attr,
492 char *buf)
493{
494 struct scsi_disk *sdkp = to_scsi_disk(dev);
495
496 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
497}
498
499static ssize_t
500zeroing_mode_store(struct device *dev, struct device_attribute *attr,
501 const char *buf, size_t count)
502{
503 struct scsi_disk *sdkp = to_scsi_disk(dev);
504 int mode;
505
506 if (!capable(CAP_SYS_ADMIN))
507 return -EACCES;
508
509 mode = sysfs_match_string(zeroing_mode, buf);
510 if (mode < 0)
511 return -EINVAL;
512
513 sdkp->zeroing_mode = mode;
514
515 return count;
516}
517static DEVICE_ATTR_RW(zeroing_mode);
518
519static ssize_t
520max_medium_access_timeouts_show(struct device *dev,
521 struct device_attribute *attr, char *buf)
522{
523 struct scsi_disk *sdkp = to_scsi_disk(dev);
524
525 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
526}
527
528static ssize_t
529max_medium_access_timeouts_store(struct device *dev,
530 struct device_attribute *attr, const char *buf,
531 size_t count)
532{
533 struct scsi_disk *sdkp = to_scsi_disk(dev);
534 int err;
535
536 if (!capable(CAP_SYS_ADMIN))
537 return -EACCES;
538
539 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
540
541 return err ? err : count;
542}
543static DEVICE_ATTR_RW(max_medium_access_timeouts);
544
545static ssize_t
546max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
547 char *buf)
548{
549 struct scsi_disk *sdkp = to_scsi_disk(dev);
550
551 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
552}
553
554static ssize_t
555max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
556 const char *buf, size_t count)
557{
558 struct scsi_disk *sdkp = to_scsi_disk(dev);
559 struct scsi_device *sdp = sdkp->device;
560 unsigned long max;
561 int err;
562
563 if (!capable(CAP_SYS_ADMIN))
564 return -EACCES;
565
566 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
567 return -EINVAL;
568
569 err = kstrtoul(buf, 10, &max);
570
571 if (err)
572 return err;
573
574 if (max == 0)
575 sdp->no_write_same = 1;
576 else if (max <= SD_MAX_WS16_BLOCKS) {
577 sdp->no_write_same = 0;
578 sdkp->max_ws_blocks = max;
579 }
580
581 sd_config_write_same(sdkp);
582
583 return count;
584}
585static DEVICE_ATTR_RW(max_write_same_blocks);
586
587static ssize_t
588zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
589{
590 struct scsi_disk *sdkp = to_scsi_disk(dev);
591
592 if (sdkp->device->type == TYPE_ZBC)
593 return sprintf(buf, "host-managed\n");
594 if (sdkp->zoned == 1)
595 return sprintf(buf, "host-aware\n");
596 if (sdkp->zoned == 2)
597 return sprintf(buf, "drive-managed\n");
598 return sprintf(buf, "none\n");
599}
600static DEVICE_ATTR_RO(zoned_cap);
601
602static ssize_t
603max_retries_store(struct device *dev, struct device_attribute *attr,
604 const char *buf, size_t count)
605{
606 struct scsi_disk *sdkp = to_scsi_disk(dev);
607 struct scsi_device *sdev = sdkp->device;
608 int retries, err;
609
610 err = kstrtoint(buf, 10, &retries);
611 if (err)
612 return err;
613
614 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
615 sdkp->max_retries = retries;
616 return count;
617 }
618
619 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
620 SD_MAX_RETRIES);
621 return -EINVAL;
622}
623
624static ssize_t
625max_retries_show(struct device *dev, struct device_attribute *attr,
626 char *buf)
627{
628 struct scsi_disk *sdkp = to_scsi_disk(dev);
629
630 return sprintf(buf, "%d\n", sdkp->max_retries);
631}
632
633static DEVICE_ATTR_RW(max_retries);
634
635static struct attribute *sd_disk_attrs[] = {
636 &dev_attr_cache_type.attr,
637 &dev_attr_FUA.attr,
638 &dev_attr_allow_restart.attr,
639 &dev_attr_manage_start_stop.attr,
640 &dev_attr_manage_system_start_stop.attr,
641 &dev_attr_manage_runtime_start_stop.attr,
642 &dev_attr_manage_shutdown.attr,
643 &dev_attr_protection_type.attr,
644 &dev_attr_protection_mode.attr,
645 &dev_attr_app_tag_own.attr,
646 &dev_attr_thin_provisioning.attr,
647 &dev_attr_provisioning_mode.attr,
648 &dev_attr_zeroing_mode.attr,
649 &dev_attr_max_write_same_blocks.attr,
650 &dev_attr_max_medium_access_timeouts.attr,
651 &dev_attr_zoned_cap.attr,
652 &dev_attr_max_retries.attr,
653 NULL,
654};
655ATTRIBUTE_GROUPS(sd_disk);
656
657static struct class sd_disk_class = {
658 .name = "scsi_disk",
659 .dev_release = scsi_disk_release,
660 .dev_groups = sd_disk_groups,
661};
662
663/*
664 * Don't request a new module, as that could deadlock in multipath
665 * environment.
666 */
667static void sd_default_probe(dev_t devt)
668{
669}
670
671/*
672 * Device no to disk mapping:
673 *
674 * major disc2 disc p1
675 * |............|.............|....|....| <- dev_t
676 * 31 20 19 8 7 4 3 0
677 *
678 * Inside a major, we have 16k disks, however mapped non-
679 * contiguously. The first 16 disks are for major0, the next
680 * ones with major1, ... Disk 256 is for major0 again, disk 272
681 * for major1, ...
682 * As we stay compatible with our numbering scheme, we can reuse
683 * the well-know SCSI majors 8, 65--71, 136--143.
684 */
685static int sd_major(int major_idx)
686{
687 switch (major_idx) {
688 case 0:
689 return SCSI_DISK0_MAJOR;
690 case 1 ... 7:
691 return SCSI_DISK1_MAJOR + major_idx - 1;
692 case 8 ... 15:
693 return SCSI_DISK8_MAJOR + major_idx - 8;
694 default:
695 BUG();
696 return 0; /* shut up gcc */
697 }
698}
699
700#ifdef CONFIG_BLK_SED_OPAL
701static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
702 size_t len, bool send)
703{
704 struct scsi_disk *sdkp = data;
705 struct scsi_device *sdev = sdkp->device;
706 u8 cdb[12] = { 0, };
707 const struct scsi_exec_args exec_args = {
708 .req_flags = BLK_MQ_REQ_PM,
709 };
710 int ret;
711
712 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
713 cdb[1] = secp;
714 put_unaligned_be16(spsp, &cdb[2]);
715 put_unaligned_be32(len, &cdb[6]);
716
717 ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
718 buffer, len, SD_TIMEOUT, sdkp->max_retries,
719 &exec_args);
720 return ret <= 0 ? ret : -EIO;
721}
722#endif /* CONFIG_BLK_SED_OPAL */
723
724/*
725 * Look up the DIX operation based on whether the command is read or
726 * write and whether dix and dif are enabled.
727 */
728static unsigned int sd_prot_op(bool write, bool dix, bool dif)
729{
730 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
731 static const unsigned int ops[] = { /* wrt dix dif */
732 SCSI_PROT_NORMAL, /* 0 0 0 */
733 SCSI_PROT_READ_STRIP, /* 0 0 1 */
734 SCSI_PROT_READ_INSERT, /* 0 1 0 */
735 SCSI_PROT_READ_PASS, /* 0 1 1 */
736 SCSI_PROT_NORMAL, /* 1 0 0 */
737 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
738 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
739 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
740 };
741
742 return ops[write << 2 | dix << 1 | dif];
743}
744
745/*
746 * Returns a mask of the protection flags that are valid for a given DIX
747 * operation.
748 */
749static unsigned int sd_prot_flag_mask(unsigned int prot_op)
750{
751 static const unsigned int flag_mask[] = {
752 [SCSI_PROT_NORMAL] = 0,
753
754 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
755 SCSI_PROT_GUARD_CHECK |
756 SCSI_PROT_REF_CHECK |
757 SCSI_PROT_REF_INCREMENT,
758
759 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
760 SCSI_PROT_IP_CHECKSUM,
761
762 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
763 SCSI_PROT_GUARD_CHECK |
764 SCSI_PROT_REF_CHECK |
765 SCSI_PROT_REF_INCREMENT |
766 SCSI_PROT_IP_CHECKSUM,
767
768 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
769 SCSI_PROT_REF_INCREMENT,
770
771 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
772 SCSI_PROT_REF_CHECK |
773 SCSI_PROT_REF_INCREMENT |
774 SCSI_PROT_IP_CHECKSUM,
775
776 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
777 SCSI_PROT_GUARD_CHECK |
778 SCSI_PROT_REF_CHECK |
779 SCSI_PROT_REF_INCREMENT |
780 SCSI_PROT_IP_CHECKSUM,
781 };
782
783 return flag_mask[prot_op];
784}
785
786static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
787 unsigned int dix, unsigned int dif)
788{
789 struct request *rq = scsi_cmd_to_rq(scmd);
790 struct bio *bio = rq->bio;
791 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
792 unsigned int protect = 0;
793
794 if (dix) { /* DIX Type 0, 1, 2, 3 */
795 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
796 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
797
798 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
799 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
800 }
801
802 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
803 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
804
805 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
806 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
807 }
808
809 if (dif) { /* DIX/DIF Type 1, 2, 3 */
810 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
811
812 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
813 protect = 3 << 5; /* Disable target PI checking */
814 else
815 protect = 1 << 5; /* Enable target PI checking */
816 }
817
818 scsi_set_prot_op(scmd, prot_op);
819 scsi_set_prot_type(scmd, dif);
820 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
821
822 return protect;
823}
824
825static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
826{
827 struct request_queue *q = sdkp->disk->queue;
828 unsigned int logical_block_size = sdkp->device->sector_size;
829 unsigned int max_blocks = 0;
830
831 q->limits.discard_alignment =
832 sdkp->unmap_alignment * logical_block_size;
833 q->limits.discard_granularity =
834 max(sdkp->physical_block_size,
835 sdkp->unmap_granularity * logical_block_size);
836 sdkp->provisioning_mode = mode;
837
838 switch (mode) {
839
840 case SD_LBP_FULL:
841 case SD_LBP_DISABLE:
842 blk_queue_max_discard_sectors(q, 0);
843 return;
844
845 case SD_LBP_UNMAP:
846 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
847 (u32)SD_MAX_WS16_BLOCKS);
848 break;
849
850 case SD_LBP_WS16:
851 if (sdkp->device->unmap_limit_for_ws)
852 max_blocks = sdkp->max_unmap_blocks;
853 else
854 max_blocks = sdkp->max_ws_blocks;
855
856 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
857 break;
858
859 case SD_LBP_WS10:
860 if (sdkp->device->unmap_limit_for_ws)
861 max_blocks = sdkp->max_unmap_blocks;
862 else
863 max_blocks = sdkp->max_ws_blocks;
864
865 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
866 break;
867
868 case SD_LBP_ZERO:
869 max_blocks = min_not_zero(sdkp->max_ws_blocks,
870 (u32)SD_MAX_WS10_BLOCKS);
871 break;
872 }
873
874 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
875}
876
877static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
878{
879 struct page *page;
880
881 page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 if (!page)
883 return NULL;
884 clear_highpage(page);
885 bvec_set_page(&rq->special_vec, page, data_len, 0);
886 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
887 return bvec_virt(&rq->special_vec);
888}
889
890static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
891{
892 struct scsi_device *sdp = cmd->device;
893 struct request *rq = scsi_cmd_to_rq(cmd);
894 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
895 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
896 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
897 unsigned int data_len = 24;
898 char *buf;
899
900 buf = sd_set_special_bvec(rq, data_len);
901 if (!buf)
902 return BLK_STS_RESOURCE;
903
904 cmd->cmd_len = 10;
905 cmd->cmnd[0] = UNMAP;
906 cmd->cmnd[8] = 24;
907
908 put_unaligned_be16(6 + 16, &buf[0]);
909 put_unaligned_be16(16, &buf[2]);
910 put_unaligned_be64(lba, &buf[8]);
911 put_unaligned_be32(nr_blocks, &buf[16]);
912
913 cmd->allowed = sdkp->max_retries;
914 cmd->transfersize = data_len;
915 rq->timeout = SD_TIMEOUT;
916
917 return scsi_alloc_sgtables(cmd);
918}
919
920static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
921 bool unmap)
922{
923 struct scsi_device *sdp = cmd->device;
924 struct request *rq = scsi_cmd_to_rq(cmd);
925 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
926 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
927 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
928 u32 data_len = sdp->sector_size;
929
930 if (!sd_set_special_bvec(rq, data_len))
931 return BLK_STS_RESOURCE;
932
933 cmd->cmd_len = 16;
934 cmd->cmnd[0] = WRITE_SAME_16;
935 if (unmap)
936 cmd->cmnd[1] = 0x8; /* UNMAP */
937 put_unaligned_be64(lba, &cmd->cmnd[2]);
938 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
939
940 cmd->allowed = sdkp->max_retries;
941 cmd->transfersize = data_len;
942 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
943
944 return scsi_alloc_sgtables(cmd);
945}
946
947static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
948 bool unmap)
949{
950 struct scsi_device *sdp = cmd->device;
951 struct request *rq = scsi_cmd_to_rq(cmd);
952 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
953 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
954 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
955 u32 data_len = sdp->sector_size;
956
957 if (!sd_set_special_bvec(rq, data_len))
958 return BLK_STS_RESOURCE;
959
960 cmd->cmd_len = 10;
961 cmd->cmnd[0] = WRITE_SAME;
962 if (unmap)
963 cmd->cmnd[1] = 0x8; /* UNMAP */
964 put_unaligned_be32(lba, &cmd->cmnd[2]);
965 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
966
967 cmd->allowed = sdkp->max_retries;
968 cmd->transfersize = data_len;
969 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
970
971 return scsi_alloc_sgtables(cmd);
972}
973
974static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
975{
976 struct request *rq = scsi_cmd_to_rq(cmd);
977 struct scsi_device *sdp = cmd->device;
978 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
979 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
980 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
981
982 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
983 switch (sdkp->zeroing_mode) {
984 case SD_ZERO_WS16_UNMAP:
985 return sd_setup_write_same16_cmnd(cmd, true);
986 case SD_ZERO_WS10_UNMAP:
987 return sd_setup_write_same10_cmnd(cmd, true);
988 }
989 }
990
991 if (sdp->no_write_same) {
992 rq->rq_flags |= RQF_QUIET;
993 return BLK_STS_TARGET;
994 }
995
996 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
997 return sd_setup_write_same16_cmnd(cmd, false);
998
999 return sd_setup_write_same10_cmnd(cmd, false);
1000}
1001
1002static void sd_config_write_same(struct scsi_disk *sdkp)
1003{
1004 struct request_queue *q = sdkp->disk->queue;
1005 unsigned int logical_block_size = sdkp->device->sector_size;
1006
1007 if (sdkp->device->no_write_same) {
1008 sdkp->max_ws_blocks = 0;
1009 goto out;
1010 }
1011
1012 /* Some devices can not handle block counts above 0xffff despite
1013 * supporting WRITE SAME(16). Consequently we default to 64k
1014 * blocks per I/O unless the device explicitly advertises a
1015 * bigger limit.
1016 */
1017 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1018 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1019 (u32)SD_MAX_WS16_BLOCKS);
1020 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1021 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1022 (u32)SD_MAX_WS10_BLOCKS);
1023 else {
1024 sdkp->device->no_write_same = 1;
1025 sdkp->max_ws_blocks = 0;
1026 }
1027
1028 if (sdkp->lbprz && sdkp->lbpws)
1029 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1030 else if (sdkp->lbprz && sdkp->lbpws10)
1031 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1032 else if (sdkp->max_ws_blocks)
1033 sdkp->zeroing_mode = SD_ZERO_WS;
1034 else
1035 sdkp->zeroing_mode = SD_ZERO_WRITE;
1036
1037 if (sdkp->max_ws_blocks &&
1038 sdkp->physical_block_size > logical_block_size) {
1039 /*
1040 * Reporting a maximum number of blocks that is not aligned
1041 * on the device physical size would cause a large write same
1042 * request to be split into physically unaligned chunks by
1043 * __blkdev_issue_write_zeroes() even if the caller of this
1044 * functions took care to align the large request. So make sure
1045 * the maximum reported is aligned to the device physical block
1046 * size. This is only an optional optimization for regular
1047 * disks, but this is mandatory to avoid failure of large write
1048 * same requests directed at sequential write required zones of
1049 * host-managed ZBC disks.
1050 */
1051 sdkp->max_ws_blocks =
1052 round_down(sdkp->max_ws_blocks,
1053 bytes_to_logical(sdkp->device,
1054 sdkp->physical_block_size));
1055 }
1056
1057out:
1058 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1059 (logical_block_size >> 9));
1060}
1061
1062static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1063{
1064 struct request *rq = scsi_cmd_to_rq(cmd);
1065 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1066
1067 /* flush requests don't perform I/O, zero the S/G table */
1068 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1069
1070 if (cmd->device->use_16_for_sync) {
1071 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1072 cmd->cmd_len = 16;
1073 } else {
1074 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1075 cmd->cmd_len = 10;
1076 }
1077 cmd->transfersize = 0;
1078 cmd->allowed = sdkp->max_retries;
1079
1080 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1081 return BLK_STS_OK;
1082}
1083
1084/**
1085 * sd_group_number() - Compute the GROUP NUMBER field
1086 * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1087 * field.
1088 *
1089 * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1090 * 0: no relative lifetime.
1091 * 1: shortest relative lifetime.
1092 * 2: second shortest relative lifetime.
1093 * 3 - 0x3d: intermediate relative lifetimes.
1094 * 0x3e: second longest relative lifetime.
1095 * 0x3f: longest relative lifetime.
1096 */
1097static u8 sd_group_number(struct scsi_cmnd *cmd)
1098{
1099 const struct request *rq = scsi_cmd_to_rq(cmd);
1100 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1101
1102 if (!sdkp->rscs)
1103 return 0;
1104
1105 return min3((u32)rq->write_hint, (u32)sdkp->permanent_stream_count,
1106 0x3fu);
1107}
1108
1109static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1110 sector_t lba, unsigned int nr_blocks,
1111 unsigned char flags, unsigned int dld)
1112{
1113 cmd->cmd_len = SD_EXT_CDB_SIZE;
1114 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1115 cmd->cmnd[6] = sd_group_number(cmd);
1116 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1117 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1118 cmd->cmnd[10] = flags;
1119 cmd->cmnd[11] = dld & 0x07;
1120 put_unaligned_be64(lba, &cmd->cmnd[12]);
1121 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1122 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1123
1124 return BLK_STS_OK;
1125}
1126
1127static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1128 sector_t lba, unsigned int nr_blocks,
1129 unsigned char flags, unsigned int dld)
1130{
1131 cmd->cmd_len = 16;
1132 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1133 cmd->cmnd[1] = flags | ((dld >> 2) & 0x01);
1134 cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1135 cmd->cmnd[15] = 0;
1136 put_unaligned_be64(lba, &cmd->cmnd[2]);
1137 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1138
1139 return BLK_STS_OK;
1140}
1141
1142static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1143 sector_t lba, unsigned int nr_blocks,
1144 unsigned char flags)
1145{
1146 cmd->cmd_len = 10;
1147 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1148 cmd->cmnd[1] = flags;
1149 cmd->cmnd[6] = sd_group_number(cmd);
1150 cmd->cmnd[9] = 0;
1151 put_unaligned_be32(lba, &cmd->cmnd[2]);
1152 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1153
1154 return BLK_STS_OK;
1155}
1156
1157static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1158 sector_t lba, unsigned int nr_blocks,
1159 unsigned char flags)
1160{
1161 /* Avoid that 0 blocks gets translated into 256 blocks. */
1162 if (WARN_ON_ONCE(nr_blocks == 0))
1163 return BLK_STS_IOERR;
1164
1165 if (unlikely(flags & 0x8)) {
1166 /*
1167 * This happens only if this drive failed 10byte rw
1168 * command with ILLEGAL_REQUEST during operation and
1169 * thus turned off use_10_for_rw.
1170 */
1171 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1172 return BLK_STS_IOERR;
1173 }
1174
1175 cmd->cmd_len = 6;
1176 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1177 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1178 cmd->cmnd[2] = (lba >> 8) & 0xff;
1179 cmd->cmnd[3] = lba & 0xff;
1180 cmd->cmnd[4] = nr_blocks;
1181 cmd->cmnd[5] = 0;
1182
1183 return BLK_STS_OK;
1184}
1185
1186/*
1187 * Check if a command has a duration limit set. If it does, and the target
1188 * device supports CDL and the feature is enabled, return the limit
1189 * descriptor index to use. Return 0 (no limit) otherwise.
1190 */
1191static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1192{
1193 struct scsi_device *sdp = sdkp->device;
1194 int hint;
1195
1196 if (!sdp->cdl_supported || !sdp->cdl_enable)
1197 return 0;
1198
1199 /*
1200 * Use "no limit" if the request ioprio does not specify a duration
1201 * limit hint.
1202 */
1203 hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1204 if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1205 hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1206 return 0;
1207
1208 return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1209}
1210
1211static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1212{
1213 struct request *rq = scsi_cmd_to_rq(cmd);
1214 struct scsi_device *sdp = cmd->device;
1215 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1216 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1217 sector_t threshold;
1218 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1219 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1220 bool write = rq_data_dir(rq) == WRITE;
1221 unsigned char protect, fua;
1222 unsigned int dld;
1223 blk_status_t ret;
1224 unsigned int dif;
1225 bool dix;
1226
1227 ret = scsi_alloc_sgtables(cmd);
1228 if (ret != BLK_STS_OK)
1229 return ret;
1230
1231 ret = BLK_STS_IOERR;
1232 if (!scsi_device_online(sdp) || sdp->changed) {
1233 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1234 goto fail;
1235 }
1236
1237 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1238 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1239 goto fail;
1240 }
1241
1242 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1243 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1244 goto fail;
1245 }
1246
1247 /*
1248 * Some SD card readers can't handle accesses which touch the
1249 * last one or two logical blocks. Split accesses as needed.
1250 */
1251 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1252
1253 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1254 if (lba < threshold) {
1255 /* Access up to the threshold but not beyond */
1256 nr_blocks = threshold - lba;
1257 } else {
1258 /* Access only a single logical block */
1259 nr_blocks = 1;
1260 }
1261 }
1262
1263 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1264 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1265 if (ret)
1266 goto fail;
1267 }
1268
1269 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1270 dix = scsi_prot_sg_count(cmd);
1271 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1272 dld = sd_cdl_dld(sdkp, cmd);
1273
1274 if (dif || dix)
1275 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1276 else
1277 protect = 0;
1278
1279 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1280 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1281 protect | fua, dld);
1282 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1283 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1284 protect | fua, dld);
1285 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1286 sdp->use_10_for_rw || protect || rq->write_hint) {
1287 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1288 protect | fua);
1289 } else {
1290 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1291 protect | fua);
1292 }
1293
1294 if (unlikely(ret != BLK_STS_OK))
1295 goto fail;
1296
1297 /*
1298 * We shouldn't disconnect in the middle of a sector, so with a dumb
1299 * host adapter, it's safe to assume that we can at least transfer
1300 * this many bytes between each connect / disconnect.
1301 */
1302 cmd->transfersize = sdp->sector_size;
1303 cmd->underflow = nr_blocks << 9;
1304 cmd->allowed = sdkp->max_retries;
1305 cmd->sdb.length = nr_blocks * sdp->sector_size;
1306
1307 SCSI_LOG_HLQUEUE(1,
1308 scmd_printk(KERN_INFO, cmd,
1309 "%s: block=%llu, count=%d\n", __func__,
1310 (unsigned long long)blk_rq_pos(rq),
1311 blk_rq_sectors(rq)));
1312 SCSI_LOG_HLQUEUE(2,
1313 scmd_printk(KERN_INFO, cmd,
1314 "%s %d/%u 512 byte blocks.\n",
1315 write ? "writing" : "reading", nr_blocks,
1316 blk_rq_sectors(rq)));
1317
1318 /*
1319 * This indicates that the command is ready from our end to be queued.
1320 */
1321 return BLK_STS_OK;
1322fail:
1323 scsi_free_sgtables(cmd);
1324 return ret;
1325}
1326
1327static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1328{
1329 struct request *rq = scsi_cmd_to_rq(cmd);
1330
1331 switch (req_op(rq)) {
1332 case REQ_OP_DISCARD:
1333 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1334 case SD_LBP_UNMAP:
1335 return sd_setup_unmap_cmnd(cmd);
1336 case SD_LBP_WS16:
1337 return sd_setup_write_same16_cmnd(cmd, true);
1338 case SD_LBP_WS10:
1339 return sd_setup_write_same10_cmnd(cmd, true);
1340 case SD_LBP_ZERO:
1341 return sd_setup_write_same10_cmnd(cmd, false);
1342 default:
1343 return BLK_STS_TARGET;
1344 }
1345 case REQ_OP_WRITE_ZEROES:
1346 return sd_setup_write_zeroes_cmnd(cmd);
1347 case REQ_OP_FLUSH:
1348 return sd_setup_flush_cmnd(cmd);
1349 case REQ_OP_READ:
1350 case REQ_OP_WRITE:
1351 case REQ_OP_ZONE_APPEND:
1352 return sd_setup_read_write_cmnd(cmd);
1353 case REQ_OP_ZONE_RESET:
1354 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1355 false);
1356 case REQ_OP_ZONE_RESET_ALL:
1357 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358 true);
1359 case REQ_OP_ZONE_OPEN:
1360 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1361 case REQ_OP_ZONE_CLOSE:
1362 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1363 case REQ_OP_ZONE_FINISH:
1364 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1365 default:
1366 WARN_ON_ONCE(1);
1367 return BLK_STS_NOTSUPP;
1368 }
1369}
1370
1371static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1372{
1373 struct request *rq = scsi_cmd_to_rq(SCpnt);
1374
1375 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1376 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1377}
1378
1379static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1380{
1381 if (sdkp->device->removable || sdkp->write_prot) {
1382 if (disk_check_media_change(disk))
1383 return true;
1384 }
1385
1386 /*
1387 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1388 * nothing to do with partitions, BLKRRPART is used to force a full
1389 * revalidate after things like a format for historical reasons.
1390 */
1391 return test_bit(GD_NEED_PART_SCAN, &disk->state);
1392}
1393
1394/**
1395 * sd_open - open a scsi disk device
1396 * @disk: disk to open
1397 * @mode: open mode
1398 *
1399 * Returns 0 if successful. Returns a negated errno value in case
1400 * of error.
1401 *
1402 * Note: This can be called from a user context (e.g. fsck(1) )
1403 * or from within the kernel (e.g. as a result of a mount(1) ).
1404 * In the latter case @inode and @filp carry an abridged amount
1405 * of information as noted above.
1406 *
1407 * Locking: called with disk->open_mutex held.
1408 **/
1409static int sd_open(struct gendisk *disk, blk_mode_t mode)
1410{
1411 struct scsi_disk *sdkp = scsi_disk(disk);
1412 struct scsi_device *sdev = sdkp->device;
1413 int retval;
1414
1415 if (scsi_device_get(sdev))
1416 return -ENXIO;
1417
1418 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1419
1420 /*
1421 * If the device is in error recovery, wait until it is done.
1422 * If the device is offline, then disallow any access to it.
1423 */
1424 retval = -ENXIO;
1425 if (!scsi_block_when_processing_errors(sdev))
1426 goto error_out;
1427
1428 if (sd_need_revalidate(disk, sdkp))
1429 sd_revalidate_disk(disk);
1430
1431 /*
1432 * If the drive is empty, just let the open fail.
1433 */
1434 retval = -ENOMEDIUM;
1435 if (sdev->removable && !sdkp->media_present &&
1436 !(mode & BLK_OPEN_NDELAY))
1437 goto error_out;
1438
1439 /*
1440 * If the device has the write protect tab set, have the open fail
1441 * if the user expects to be able to write to the thing.
1442 */
1443 retval = -EROFS;
1444 if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1445 goto error_out;
1446
1447 /*
1448 * It is possible that the disk changing stuff resulted in
1449 * the device being taken offline. If this is the case,
1450 * report this to the user, and don't pretend that the
1451 * open actually succeeded.
1452 */
1453 retval = -ENXIO;
1454 if (!scsi_device_online(sdev))
1455 goto error_out;
1456
1457 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1458 if (scsi_block_when_processing_errors(sdev))
1459 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1460 }
1461
1462 return 0;
1463
1464error_out:
1465 scsi_device_put(sdev);
1466 return retval;
1467}
1468
1469/**
1470 * sd_release - invoked when the (last) close(2) is called on this
1471 * scsi disk.
1472 * @disk: disk to release
1473 *
1474 * Returns 0.
1475 *
1476 * Note: may block (uninterruptible) if error recovery is underway
1477 * on this disk.
1478 *
1479 * Locking: called with disk->open_mutex held.
1480 **/
1481static void sd_release(struct gendisk *disk)
1482{
1483 struct scsi_disk *sdkp = scsi_disk(disk);
1484 struct scsi_device *sdev = sdkp->device;
1485
1486 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1487
1488 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1489 if (scsi_block_when_processing_errors(sdev))
1490 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1491 }
1492
1493 scsi_device_put(sdev);
1494}
1495
1496static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1497{
1498 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1499 struct scsi_device *sdp = sdkp->device;
1500 struct Scsi_Host *host = sdp->host;
1501 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1502 int diskinfo[4];
1503
1504 /* default to most commonly used values */
1505 diskinfo[0] = 0x40; /* 1 << 6 */
1506 diskinfo[1] = 0x20; /* 1 << 5 */
1507 diskinfo[2] = capacity >> 11;
1508
1509 /* override with calculated, extended default, or driver values */
1510 if (host->hostt->bios_param)
1511 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1512 else
1513 scsicam_bios_param(bdev, capacity, diskinfo);
1514
1515 geo->heads = diskinfo[0];
1516 geo->sectors = diskinfo[1];
1517 geo->cylinders = diskinfo[2];
1518 return 0;
1519}
1520
1521/**
1522 * sd_ioctl - process an ioctl
1523 * @bdev: target block device
1524 * @mode: open mode
1525 * @cmd: ioctl command number
1526 * @arg: this is third argument given to ioctl(2) system call.
1527 * Often contains a pointer.
1528 *
1529 * Returns 0 if successful (some ioctls return positive numbers on
1530 * success as well). Returns a negated errno value in case of error.
1531 *
1532 * Note: most ioctls are forward onto the block subsystem or further
1533 * down in the scsi subsystem.
1534 **/
1535static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1536 unsigned int cmd, unsigned long arg)
1537{
1538 struct gendisk *disk = bdev->bd_disk;
1539 struct scsi_disk *sdkp = scsi_disk(disk);
1540 struct scsi_device *sdp = sdkp->device;
1541 void __user *p = (void __user *)arg;
1542 int error;
1543
1544 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1545 "cmd=0x%x\n", disk->disk_name, cmd));
1546
1547 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1548 return -ENOIOCTLCMD;
1549
1550 /*
1551 * If we are in the middle of error recovery, don't let anyone
1552 * else try and use this device. Also, if error recovery fails, it
1553 * may try and take the device offline, in which case all further
1554 * access to the device is prohibited.
1555 */
1556 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1557 (mode & BLK_OPEN_NDELAY));
1558 if (error)
1559 return error;
1560
1561 if (is_sed_ioctl(cmd))
1562 return sed_ioctl(sdkp->opal_dev, cmd, p);
1563 return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1564}
1565
1566static void set_media_not_present(struct scsi_disk *sdkp)
1567{
1568 if (sdkp->media_present)
1569 sdkp->device->changed = 1;
1570
1571 if (sdkp->device->removable) {
1572 sdkp->media_present = 0;
1573 sdkp->capacity = 0;
1574 }
1575}
1576
1577static int media_not_present(struct scsi_disk *sdkp,
1578 struct scsi_sense_hdr *sshdr)
1579{
1580 if (!scsi_sense_valid(sshdr))
1581 return 0;
1582
1583 /* not invoked for commands that could return deferred errors */
1584 switch (sshdr->sense_key) {
1585 case UNIT_ATTENTION:
1586 case NOT_READY:
1587 /* medium not present */
1588 if (sshdr->asc == 0x3A) {
1589 set_media_not_present(sdkp);
1590 return 1;
1591 }
1592 }
1593 return 0;
1594}
1595
1596/**
1597 * sd_check_events - check media events
1598 * @disk: kernel device descriptor
1599 * @clearing: disk events currently being cleared
1600 *
1601 * Returns mask of DISK_EVENT_*.
1602 *
1603 * Note: this function is invoked from the block subsystem.
1604 **/
1605static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1606{
1607 struct scsi_disk *sdkp = disk->private_data;
1608 struct scsi_device *sdp;
1609 int retval;
1610 bool disk_changed;
1611
1612 if (!sdkp)
1613 return 0;
1614
1615 sdp = sdkp->device;
1616 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1617
1618 /*
1619 * If the device is offline, don't send any commands - just pretend as
1620 * if the command failed. If the device ever comes back online, we
1621 * can deal with it then. It is only because of unrecoverable errors
1622 * that we would ever take a device offline in the first place.
1623 */
1624 if (!scsi_device_online(sdp)) {
1625 set_media_not_present(sdkp);
1626 goto out;
1627 }
1628
1629 /*
1630 * Using TEST_UNIT_READY enables differentiation between drive with
1631 * no cartridge loaded - NOT READY, drive with changed cartridge -
1632 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1633 *
1634 * Drives that auto spin down. eg iomega jaz 1G, will be started
1635 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1636 * sd_revalidate() is called.
1637 */
1638 if (scsi_block_when_processing_errors(sdp)) {
1639 struct scsi_sense_hdr sshdr = { 0, };
1640
1641 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1642 &sshdr);
1643
1644 /* failed to execute TUR, assume media not present */
1645 if (retval < 0 || host_byte(retval)) {
1646 set_media_not_present(sdkp);
1647 goto out;
1648 }
1649
1650 if (media_not_present(sdkp, &sshdr))
1651 goto out;
1652 }
1653
1654 /*
1655 * For removable scsi disk we have to recognise the presence
1656 * of a disk in the drive.
1657 */
1658 if (!sdkp->media_present)
1659 sdp->changed = 1;
1660 sdkp->media_present = 1;
1661out:
1662 /*
1663 * sdp->changed is set under the following conditions:
1664 *
1665 * Medium present state has changed in either direction.
1666 * Device has indicated UNIT_ATTENTION.
1667 */
1668 disk_changed = sdp->changed;
1669 sdp->changed = 0;
1670 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1671}
1672
1673static int sd_sync_cache(struct scsi_disk *sdkp)
1674{
1675 int res;
1676 struct scsi_device *sdp = sdkp->device;
1677 const int timeout = sdp->request_queue->rq_timeout
1678 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1679 /* Leave the rest of the command zero to indicate flush everything. */
1680 const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1681 SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1682 struct scsi_sense_hdr sshdr;
1683 struct scsi_failure failure_defs[] = {
1684 {
1685 .allowed = 3,
1686 .result = SCMD_FAILURE_RESULT_ANY,
1687 },
1688 {}
1689 };
1690 struct scsi_failures failures = {
1691 .failure_definitions = failure_defs,
1692 };
1693 const struct scsi_exec_args exec_args = {
1694 .req_flags = BLK_MQ_REQ_PM,
1695 .sshdr = &sshdr,
1696 .failures = &failures,
1697 };
1698
1699 if (!scsi_device_online(sdp))
1700 return -ENODEV;
1701
1702 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1703 sdkp->max_retries, &exec_args);
1704 if (res) {
1705 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1706
1707 if (res < 0)
1708 return res;
1709
1710 if (scsi_status_is_check_condition(res) &&
1711 scsi_sense_valid(&sshdr)) {
1712 sd_print_sense_hdr(sdkp, &sshdr);
1713
1714 /* we need to evaluate the error return */
1715 if (sshdr.asc == 0x3a || /* medium not present */
1716 sshdr.asc == 0x20 || /* invalid command */
1717 (sshdr.asc == 0x74 && sshdr.ascq == 0x71)) /* drive is password locked */
1718 /* this is no error here */
1719 return 0;
1720 /*
1721 * This drive doesn't support sync and there's not much
1722 * we can do because this is called during shutdown
1723 * or suspend so just return success so those operations
1724 * can proceed.
1725 */
1726 if (sshdr.sense_key == ILLEGAL_REQUEST)
1727 return 0;
1728 }
1729
1730 switch (host_byte(res)) {
1731 /* ignore errors due to racing a disconnection */
1732 case DID_BAD_TARGET:
1733 case DID_NO_CONNECT:
1734 return 0;
1735 /* signal the upper layer it might try again */
1736 case DID_BUS_BUSY:
1737 case DID_IMM_RETRY:
1738 case DID_REQUEUE:
1739 case DID_SOFT_ERROR:
1740 return -EBUSY;
1741 default:
1742 return -EIO;
1743 }
1744 }
1745 return 0;
1746}
1747
1748static void sd_rescan(struct device *dev)
1749{
1750 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1751
1752 sd_revalidate_disk(sdkp->disk);
1753}
1754
1755static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1756 enum blk_unique_id type)
1757{
1758 struct scsi_device *sdev = scsi_disk(disk)->device;
1759 const struct scsi_vpd *vpd;
1760 const unsigned char *d;
1761 int ret = -ENXIO, len;
1762
1763 rcu_read_lock();
1764 vpd = rcu_dereference(sdev->vpd_pg83);
1765 if (!vpd)
1766 goto out_unlock;
1767
1768 ret = -EINVAL;
1769 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1770 /* we only care about designators with LU association */
1771 if (((d[1] >> 4) & 0x3) != 0x00)
1772 continue;
1773 if ((d[1] & 0xf) != type)
1774 continue;
1775
1776 /*
1777 * Only exit early if a 16-byte descriptor was found. Otherwise
1778 * keep looking as one with more entropy might still show up.
1779 */
1780 len = d[3];
1781 if (len != 8 && len != 12 && len != 16)
1782 continue;
1783 ret = len;
1784 memcpy(id, d + 4, len);
1785 if (len == 16)
1786 break;
1787 }
1788out_unlock:
1789 rcu_read_unlock();
1790 return ret;
1791}
1792
1793static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1794{
1795 switch (host_byte(result)) {
1796 case DID_TRANSPORT_MARGINAL:
1797 case DID_TRANSPORT_DISRUPTED:
1798 case DID_BUS_BUSY:
1799 return PR_STS_RETRY_PATH_FAILURE;
1800 case DID_NO_CONNECT:
1801 return PR_STS_PATH_FAILED;
1802 case DID_TRANSPORT_FAILFAST:
1803 return PR_STS_PATH_FAST_FAILED;
1804 }
1805
1806 switch (status_byte(result)) {
1807 case SAM_STAT_RESERVATION_CONFLICT:
1808 return PR_STS_RESERVATION_CONFLICT;
1809 case SAM_STAT_CHECK_CONDITION:
1810 if (!scsi_sense_valid(sshdr))
1811 return PR_STS_IOERR;
1812
1813 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1814 (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1815 return -EINVAL;
1816
1817 fallthrough;
1818 default:
1819 return PR_STS_IOERR;
1820 }
1821}
1822
1823static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1824 unsigned char *data, int data_len)
1825{
1826 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1827 struct scsi_device *sdev = sdkp->device;
1828 struct scsi_sense_hdr sshdr;
1829 u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1830 struct scsi_failure failure_defs[] = {
1831 {
1832 .sense = UNIT_ATTENTION,
1833 .asc = SCMD_FAILURE_ASC_ANY,
1834 .ascq = SCMD_FAILURE_ASCQ_ANY,
1835 .allowed = 5,
1836 .result = SAM_STAT_CHECK_CONDITION,
1837 },
1838 {}
1839 };
1840 struct scsi_failures failures = {
1841 .failure_definitions = failure_defs,
1842 };
1843 const struct scsi_exec_args exec_args = {
1844 .sshdr = &sshdr,
1845 .failures = &failures,
1846 };
1847 int result;
1848
1849 put_unaligned_be16(data_len, &cmd[7]);
1850
1851 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1852 SD_TIMEOUT, sdkp->max_retries, &exec_args);
1853 if (scsi_status_is_check_condition(result) &&
1854 scsi_sense_valid(&sshdr)) {
1855 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1856 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1857 }
1858
1859 if (result <= 0)
1860 return result;
1861
1862 return sd_scsi_to_pr_err(&sshdr, result);
1863}
1864
1865static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1866{
1867 int result, i, data_offset, num_copy_keys;
1868 u32 num_keys = keys_info->num_keys;
1869 int data_len = num_keys * 8 + 8;
1870 u8 *data;
1871
1872 data = kzalloc(data_len, GFP_KERNEL);
1873 if (!data)
1874 return -ENOMEM;
1875
1876 result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1877 if (result)
1878 goto free_data;
1879
1880 keys_info->generation = get_unaligned_be32(&data[0]);
1881 keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1882
1883 data_offset = 8;
1884 num_copy_keys = min(num_keys, keys_info->num_keys);
1885
1886 for (i = 0; i < num_copy_keys; i++) {
1887 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1888 data_offset += 8;
1889 }
1890
1891free_data:
1892 kfree(data);
1893 return result;
1894}
1895
1896static int sd_pr_read_reservation(struct block_device *bdev,
1897 struct pr_held_reservation *rsv)
1898{
1899 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1900 struct scsi_device *sdev = sdkp->device;
1901 u8 data[24] = { };
1902 int result, len;
1903
1904 result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1905 if (result)
1906 return result;
1907
1908 len = get_unaligned_be32(&data[4]);
1909 if (!len)
1910 return 0;
1911
1912 /* Make sure we have at least the key and type */
1913 if (len < 14) {
1914 sdev_printk(KERN_INFO, sdev,
1915 "READ RESERVATION failed due to short return buffer of %d bytes\n",
1916 len);
1917 return -EINVAL;
1918 }
1919
1920 rsv->generation = get_unaligned_be32(&data[0]);
1921 rsv->key = get_unaligned_be64(&data[8]);
1922 rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1923 return 0;
1924}
1925
1926static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1927 u64 sa_key, enum scsi_pr_type type, u8 flags)
1928{
1929 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1930 struct scsi_device *sdev = sdkp->device;
1931 struct scsi_sense_hdr sshdr;
1932 struct scsi_failure failure_defs[] = {
1933 {
1934 .sense = UNIT_ATTENTION,
1935 .asc = SCMD_FAILURE_ASC_ANY,
1936 .ascq = SCMD_FAILURE_ASCQ_ANY,
1937 .allowed = 5,
1938 .result = SAM_STAT_CHECK_CONDITION,
1939 },
1940 {}
1941 };
1942 struct scsi_failures failures = {
1943 .failure_definitions = failure_defs,
1944 };
1945 const struct scsi_exec_args exec_args = {
1946 .sshdr = &sshdr,
1947 .failures = &failures,
1948 };
1949 int result;
1950 u8 cmd[16] = { 0, };
1951 u8 data[24] = { 0, };
1952
1953 cmd[0] = PERSISTENT_RESERVE_OUT;
1954 cmd[1] = sa;
1955 cmd[2] = type;
1956 put_unaligned_be32(sizeof(data), &cmd[5]);
1957
1958 put_unaligned_be64(key, &data[0]);
1959 put_unaligned_be64(sa_key, &data[8]);
1960 data[20] = flags;
1961
1962 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1963 sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1964 &exec_args);
1965
1966 if (scsi_status_is_check_condition(result) &&
1967 scsi_sense_valid(&sshdr)) {
1968 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1969 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1970 }
1971
1972 if (result <= 0)
1973 return result;
1974
1975 return sd_scsi_to_pr_err(&sshdr, result);
1976}
1977
1978static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1979 u32 flags)
1980{
1981 if (flags & ~PR_FL_IGNORE_KEY)
1982 return -EOPNOTSUPP;
1983 return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1984 old_key, new_key, 0,
1985 (1 << 0) /* APTPL */);
1986}
1987
1988static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1989 u32 flags)
1990{
1991 if (flags)
1992 return -EOPNOTSUPP;
1993 return sd_pr_out_command(bdev, 0x01, key, 0,
1994 block_pr_type_to_scsi(type), 0);
1995}
1996
1997static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1998{
1999 return sd_pr_out_command(bdev, 0x02, key, 0,
2000 block_pr_type_to_scsi(type), 0);
2001}
2002
2003static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2004 enum pr_type type, bool abort)
2005{
2006 return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2007 block_pr_type_to_scsi(type), 0);
2008}
2009
2010static int sd_pr_clear(struct block_device *bdev, u64 key)
2011{
2012 return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2013}
2014
2015static const struct pr_ops sd_pr_ops = {
2016 .pr_register = sd_pr_register,
2017 .pr_reserve = sd_pr_reserve,
2018 .pr_release = sd_pr_release,
2019 .pr_preempt = sd_pr_preempt,
2020 .pr_clear = sd_pr_clear,
2021 .pr_read_keys = sd_pr_read_keys,
2022 .pr_read_reservation = sd_pr_read_reservation,
2023};
2024
2025static void scsi_disk_free_disk(struct gendisk *disk)
2026{
2027 struct scsi_disk *sdkp = scsi_disk(disk);
2028
2029 put_device(&sdkp->disk_dev);
2030}
2031
2032static const struct block_device_operations sd_fops = {
2033 .owner = THIS_MODULE,
2034 .open = sd_open,
2035 .release = sd_release,
2036 .ioctl = sd_ioctl,
2037 .getgeo = sd_getgeo,
2038 .compat_ioctl = blkdev_compat_ptr_ioctl,
2039 .check_events = sd_check_events,
2040 .unlock_native_capacity = sd_unlock_native_capacity,
2041 .report_zones = sd_zbc_report_zones,
2042 .get_unique_id = sd_get_unique_id,
2043 .free_disk = scsi_disk_free_disk,
2044 .pr_ops = &sd_pr_ops,
2045};
2046
2047/**
2048 * sd_eh_reset - reset error handling callback
2049 * @scmd: sd-issued command that has failed
2050 *
2051 * This function is called by the SCSI midlayer before starting
2052 * SCSI EH. When counting medium access failures we have to be
2053 * careful to register it only only once per device and SCSI EH run;
2054 * there might be several timed out commands which will cause the
2055 * 'max_medium_access_timeouts' counter to trigger after the first
2056 * SCSI EH run already and set the device to offline.
2057 * So this function resets the internal counter before starting SCSI EH.
2058 **/
2059static void sd_eh_reset(struct scsi_cmnd *scmd)
2060{
2061 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2062
2063 /* New SCSI EH run, reset gate variable */
2064 sdkp->ignore_medium_access_errors = false;
2065}
2066
2067/**
2068 * sd_eh_action - error handling callback
2069 * @scmd: sd-issued command that has failed
2070 * @eh_disp: The recovery disposition suggested by the midlayer
2071 *
2072 * This function is called by the SCSI midlayer upon completion of an
2073 * error test command (currently TEST UNIT READY). The result of sending
2074 * the eh command is passed in eh_disp. We're looking for devices that
2075 * fail medium access commands but are OK with non access commands like
2076 * test unit ready (so wrongly see the device as having a successful
2077 * recovery)
2078 **/
2079static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2080{
2081 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2082 struct scsi_device *sdev = scmd->device;
2083
2084 if (!scsi_device_online(sdev) ||
2085 !scsi_medium_access_command(scmd) ||
2086 host_byte(scmd->result) != DID_TIME_OUT ||
2087 eh_disp != SUCCESS)
2088 return eh_disp;
2089
2090 /*
2091 * The device has timed out executing a medium access command.
2092 * However, the TEST UNIT READY command sent during error
2093 * handling completed successfully. Either the device is in the
2094 * process of recovering or has it suffered an internal failure
2095 * that prevents access to the storage medium.
2096 */
2097 if (!sdkp->ignore_medium_access_errors) {
2098 sdkp->medium_access_timed_out++;
2099 sdkp->ignore_medium_access_errors = true;
2100 }
2101
2102 /*
2103 * If the device keeps failing read/write commands but TEST UNIT
2104 * READY always completes successfully we assume that medium
2105 * access is no longer possible and take the device offline.
2106 */
2107 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2108 scmd_printk(KERN_ERR, scmd,
2109 "Medium access timeout failure. Offlining disk!\n");
2110 mutex_lock(&sdev->state_mutex);
2111 scsi_device_set_state(sdev, SDEV_OFFLINE);
2112 mutex_unlock(&sdev->state_mutex);
2113
2114 return SUCCESS;
2115 }
2116
2117 return eh_disp;
2118}
2119
2120static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2121{
2122 struct request *req = scsi_cmd_to_rq(scmd);
2123 struct scsi_device *sdev = scmd->device;
2124 unsigned int transferred, good_bytes;
2125 u64 start_lba, end_lba, bad_lba;
2126
2127 /*
2128 * Some commands have a payload smaller than the device logical
2129 * block size (e.g. INQUIRY on a 4K disk).
2130 */
2131 if (scsi_bufflen(scmd) <= sdev->sector_size)
2132 return 0;
2133
2134 /* Check if we have a 'bad_lba' information */
2135 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2136 SCSI_SENSE_BUFFERSIZE,
2137 &bad_lba))
2138 return 0;
2139
2140 /*
2141 * If the bad lba was reported incorrectly, we have no idea where
2142 * the error is.
2143 */
2144 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2145 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2146 if (bad_lba < start_lba || bad_lba >= end_lba)
2147 return 0;
2148
2149 /*
2150 * resid is optional but mostly filled in. When it's unused,
2151 * its value is zero, so we assume the whole buffer transferred
2152 */
2153 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2154
2155 /* This computation should always be done in terms of the
2156 * resolution of the device's medium.
2157 */
2158 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2159
2160 return min(good_bytes, transferred);
2161}
2162
2163/**
2164 * sd_done - bottom half handler: called when the lower level
2165 * driver has completed (successfully or otherwise) a scsi command.
2166 * @SCpnt: mid-level's per command structure.
2167 *
2168 * Note: potentially run from within an ISR. Must not block.
2169 **/
2170static int sd_done(struct scsi_cmnd *SCpnt)
2171{
2172 int result = SCpnt->result;
2173 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2174 unsigned int sector_size = SCpnt->device->sector_size;
2175 unsigned int resid;
2176 struct scsi_sense_hdr sshdr;
2177 struct request *req = scsi_cmd_to_rq(SCpnt);
2178 struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2179 int sense_valid = 0;
2180 int sense_deferred = 0;
2181
2182 switch (req_op(req)) {
2183 case REQ_OP_DISCARD:
2184 case REQ_OP_WRITE_ZEROES:
2185 case REQ_OP_ZONE_RESET:
2186 case REQ_OP_ZONE_RESET_ALL:
2187 case REQ_OP_ZONE_OPEN:
2188 case REQ_OP_ZONE_CLOSE:
2189 case REQ_OP_ZONE_FINISH:
2190 if (!result) {
2191 good_bytes = blk_rq_bytes(req);
2192 scsi_set_resid(SCpnt, 0);
2193 } else {
2194 good_bytes = 0;
2195 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2196 }
2197 break;
2198 default:
2199 /*
2200 * In case of bogus fw or device, we could end up having
2201 * an unaligned partial completion. Check this here and force
2202 * alignment.
2203 */
2204 resid = scsi_get_resid(SCpnt);
2205 if (resid & (sector_size - 1)) {
2206 sd_printk(KERN_INFO, sdkp,
2207 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2208 resid, sector_size);
2209 scsi_print_command(SCpnt);
2210 resid = min(scsi_bufflen(SCpnt),
2211 round_up(resid, sector_size));
2212 scsi_set_resid(SCpnt, resid);
2213 }
2214 }
2215
2216 if (result) {
2217 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2218 if (sense_valid)
2219 sense_deferred = scsi_sense_is_deferred(&sshdr);
2220 }
2221 sdkp->medium_access_timed_out = 0;
2222
2223 if (!scsi_status_is_check_condition(result) &&
2224 (!sense_valid || sense_deferred))
2225 goto out;
2226
2227 switch (sshdr.sense_key) {
2228 case HARDWARE_ERROR:
2229 case MEDIUM_ERROR:
2230 good_bytes = sd_completed_bytes(SCpnt);
2231 break;
2232 case RECOVERED_ERROR:
2233 good_bytes = scsi_bufflen(SCpnt);
2234 break;
2235 case NO_SENSE:
2236 /* This indicates a false check condition, so ignore it. An
2237 * unknown amount of data was transferred so treat it as an
2238 * error.
2239 */
2240 SCpnt->result = 0;
2241 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2242 break;
2243 case ABORTED_COMMAND:
2244 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2245 good_bytes = sd_completed_bytes(SCpnt);
2246 break;
2247 case ILLEGAL_REQUEST:
2248 switch (sshdr.asc) {
2249 case 0x10: /* DIX: Host detected corruption */
2250 good_bytes = sd_completed_bytes(SCpnt);
2251 break;
2252 case 0x20: /* INVALID COMMAND OPCODE */
2253 case 0x24: /* INVALID FIELD IN CDB */
2254 switch (SCpnt->cmnd[0]) {
2255 case UNMAP:
2256 sd_config_discard(sdkp, SD_LBP_DISABLE);
2257 break;
2258 case WRITE_SAME_16:
2259 case WRITE_SAME:
2260 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2261 sd_config_discard(sdkp, SD_LBP_DISABLE);
2262 } else {
2263 sdkp->device->no_write_same = 1;
2264 sd_config_write_same(sdkp);
2265 req->rq_flags |= RQF_QUIET;
2266 }
2267 break;
2268 }
2269 }
2270 break;
2271 default:
2272 break;
2273 }
2274
2275 out:
2276 if (sd_is_zoned(sdkp))
2277 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2278
2279 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2280 "sd_done: completed %d of %d bytes\n",
2281 good_bytes, scsi_bufflen(SCpnt)));
2282
2283 return good_bytes;
2284}
2285
2286/*
2287 * spinup disk - called only in sd_revalidate_disk()
2288 */
2289static void
2290sd_spinup_disk(struct scsi_disk *sdkp)
2291{
2292 static const u8 cmd[10] = { TEST_UNIT_READY };
2293 unsigned long spintime_expire = 0;
2294 int spintime, sense_valid = 0;
2295 unsigned int the_result;
2296 struct scsi_sense_hdr sshdr;
2297 struct scsi_failure failure_defs[] = {
2298 /* Do not retry Medium Not Present */
2299 {
2300 .sense = UNIT_ATTENTION,
2301 .asc = 0x3A,
2302 .ascq = SCMD_FAILURE_ASCQ_ANY,
2303 .result = SAM_STAT_CHECK_CONDITION,
2304 },
2305 {
2306 .sense = NOT_READY,
2307 .asc = 0x3A,
2308 .ascq = SCMD_FAILURE_ASCQ_ANY,
2309 .result = SAM_STAT_CHECK_CONDITION,
2310 },
2311 /* Retry when scsi_status_is_good would return false 3 times */
2312 {
2313 .result = SCMD_FAILURE_STAT_ANY,
2314 .allowed = 3,
2315 },
2316 {}
2317 };
2318 struct scsi_failures failures = {
2319 .failure_definitions = failure_defs,
2320 };
2321 const struct scsi_exec_args exec_args = {
2322 .sshdr = &sshdr,
2323 .failures = &failures,
2324 };
2325
2326 spintime = 0;
2327
2328 /* Spin up drives, as required. Only do this at boot time */
2329 /* Spinup needs to be done for module loads too. */
2330 do {
2331 bool media_was_present = sdkp->media_present;
2332
2333 scsi_failures_reset_retries(&failures);
2334
2335 the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2336 NULL, 0, SD_TIMEOUT,
2337 sdkp->max_retries, &exec_args);
2338
2339
2340 if (the_result > 0) {
2341 /*
2342 * If the drive has indicated to us that it doesn't
2343 * have any media in it, don't bother with any more
2344 * polling.
2345 */
2346 if (media_not_present(sdkp, &sshdr)) {
2347 if (media_was_present)
2348 sd_printk(KERN_NOTICE, sdkp,
2349 "Media removed, stopped polling\n");
2350 return;
2351 }
2352 sense_valid = scsi_sense_valid(&sshdr);
2353 }
2354
2355 if (!scsi_status_is_check_condition(the_result)) {
2356 /* no sense, TUR either succeeded or failed
2357 * with a status error */
2358 if(!spintime && !scsi_status_is_good(the_result)) {
2359 sd_print_result(sdkp, "Test Unit Ready failed",
2360 the_result);
2361 }
2362 break;
2363 }
2364
2365 /*
2366 * The device does not want the automatic start to be issued.
2367 */
2368 if (sdkp->device->no_start_on_add)
2369 break;
2370
2371 if (sense_valid && sshdr.sense_key == NOT_READY) {
2372 if (sshdr.asc == 4 && sshdr.ascq == 3)
2373 break; /* manual intervention required */
2374 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2375 break; /* standby */
2376 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2377 break; /* unavailable */
2378 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2379 break; /* sanitize in progress */
2380 if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2381 break; /* depopulation in progress */
2382 if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2383 break; /* depopulation restoration in progress */
2384 /*
2385 * Issue command to spin up drive when not ready
2386 */
2387 if (!spintime) {
2388 /* Return immediately and start spin cycle */
2389 const u8 start_cmd[10] = {
2390 [0] = START_STOP,
2391 [1] = 1,
2392 [4] = sdkp->device->start_stop_pwr_cond ?
2393 0x11 : 1,
2394 };
2395
2396 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2397 scsi_execute_cmd(sdkp->device, start_cmd,
2398 REQ_OP_DRV_IN, NULL, 0,
2399 SD_TIMEOUT, sdkp->max_retries,
2400 &exec_args);
2401 spintime_expire = jiffies + 100 * HZ;
2402 spintime = 1;
2403 }
2404 /* Wait 1 second for next try */
2405 msleep(1000);
2406 printk(KERN_CONT ".");
2407
2408 /*
2409 * Wait for USB flash devices with slow firmware.
2410 * Yes, this sense key/ASC combination shouldn't
2411 * occur here. It's characteristic of these devices.
2412 */
2413 } else if (sense_valid &&
2414 sshdr.sense_key == UNIT_ATTENTION &&
2415 sshdr.asc == 0x28) {
2416 if (!spintime) {
2417 spintime_expire = jiffies + 5 * HZ;
2418 spintime = 1;
2419 }
2420 /* Wait 1 second for next try */
2421 msleep(1000);
2422 } else {
2423 /* we don't understand the sense code, so it's
2424 * probably pointless to loop */
2425 if(!spintime) {
2426 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2427 sd_print_sense_hdr(sdkp, &sshdr);
2428 }
2429 break;
2430 }
2431
2432 } while (spintime && time_before_eq(jiffies, spintime_expire));
2433
2434 if (spintime) {
2435 if (scsi_status_is_good(the_result))
2436 printk(KERN_CONT "ready\n");
2437 else
2438 printk(KERN_CONT "not responding...\n");
2439 }
2440}
2441
2442/*
2443 * Determine whether disk supports Data Integrity Field.
2444 */
2445static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2446{
2447 struct scsi_device *sdp = sdkp->device;
2448 u8 type;
2449
2450 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2451 sdkp->protection_type = 0;
2452 return 0;
2453 }
2454
2455 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2456
2457 if (type > T10_PI_TYPE3_PROTECTION) {
2458 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2459 " protection type %u. Disabling disk!\n",
2460 type);
2461 sdkp->protection_type = 0;
2462 return -ENODEV;
2463 }
2464
2465 sdkp->protection_type = type;
2466
2467 return 0;
2468}
2469
2470static void sd_config_protection(struct scsi_disk *sdkp)
2471{
2472 struct scsi_device *sdp = sdkp->device;
2473
2474 sd_dif_config_host(sdkp);
2475
2476 if (!sdkp->protection_type)
2477 return;
2478
2479 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2480 sd_first_printk(KERN_NOTICE, sdkp,
2481 "Disabling DIF Type %u protection\n",
2482 sdkp->protection_type);
2483 sdkp->protection_type = 0;
2484 }
2485
2486 sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2487 sdkp->protection_type);
2488}
2489
2490static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2491 struct scsi_sense_hdr *sshdr, int sense_valid,
2492 int the_result)
2493{
2494 if (sense_valid)
2495 sd_print_sense_hdr(sdkp, sshdr);
2496 else
2497 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2498
2499 /*
2500 * Set dirty bit for removable devices if not ready -
2501 * sometimes drives will not report this properly.
2502 */
2503 if (sdp->removable &&
2504 sense_valid && sshdr->sense_key == NOT_READY)
2505 set_media_not_present(sdkp);
2506
2507 /*
2508 * We used to set media_present to 0 here to indicate no media
2509 * in the drive, but some drives fail read capacity even with
2510 * media present, so we can't do that.
2511 */
2512 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2513}
2514
2515#define RC16_LEN 32
2516#if RC16_LEN > SD_BUF_SIZE
2517#error RC16_LEN must not be more than SD_BUF_SIZE
2518#endif
2519
2520#define READ_CAPACITY_RETRIES_ON_RESET 10
2521
2522static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2523 unsigned char *buffer)
2524{
2525 unsigned char cmd[16];
2526 struct scsi_sense_hdr sshdr;
2527 const struct scsi_exec_args exec_args = {
2528 .sshdr = &sshdr,
2529 };
2530 int sense_valid = 0;
2531 int the_result;
2532 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2533 unsigned int alignment;
2534 unsigned long long lba;
2535 unsigned sector_size;
2536
2537 if (sdp->no_read_capacity_16)
2538 return -EINVAL;
2539
2540 do {
2541 memset(cmd, 0, 16);
2542 cmd[0] = SERVICE_ACTION_IN_16;
2543 cmd[1] = SAI_READ_CAPACITY_16;
2544 cmd[13] = RC16_LEN;
2545 memset(buffer, 0, RC16_LEN);
2546
2547 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2548 buffer, RC16_LEN, SD_TIMEOUT,
2549 sdkp->max_retries, &exec_args);
2550 if (the_result > 0) {
2551 if (media_not_present(sdkp, &sshdr))
2552 return -ENODEV;
2553
2554 sense_valid = scsi_sense_valid(&sshdr);
2555 if (sense_valid &&
2556 sshdr.sense_key == ILLEGAL_REQUEST &&
2557 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2558 sshdr.ascq == 0x00)
2559 /* Invalid Command Operation Code or
2560 * Invalid Field in CDB, just retry
2561 * silently with RC10 */
2562 return -EINVAL;
2563 if (sense_valid &&
2564 sshdr.sense_key == UNIT_ATTENTION &&
2565 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2566 /* Device reset might occur several times,
2567 * give it one more chance */
2568 if (--reset_retries > 0)
2569 continue;
2570 }
2571 retries--;
2572
2573 } while (the_result && retries);
2574
2575 if (the_result) {
2576 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2577 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2578 return -EINVAL;
2579 }
2580
2581 sector_size = get_unaligned_be32(&buffer[8]);
2582 lba = get_unaligned_be64(&buffer[0]);
2583
2584 if (sd_read_protection_type(sdkp, buffer) < 0) {
2585 sdkp->capacity = 0;
2586 return -ENODEV;
2587 }
2588
2589 /* Logical blocks per physical block exponent */
2590 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2591
2592 /* RC basis */
2593 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2594
2595 /* Lowest aligned logical block */
2596 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2597 blk_queue_alignment_offset(sdp->request_queue, alignment);
2598 if (alignment && sdkp->first_scan)
2599 sd_printk(KERN_NOTICE, sdkp,
2600 "physical block alignment offset: %u\n", alignment);
2601
2602 if (buffer[14] & 0x80) { /* LBPME */
2603 sdkp->lbpme = 1;
2604
2605 if (buffer[14] & 0x40) /* LBPRZ */
2606 sdkp->lbprz = 1;
2607
2608 sd_config_discard(sdkp, SD_LBP_WS16);
2609 }
2610
2611 sdkp->capacity = lba + 1;
2612 return sector_size;
2613}
2614
2615static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2616 unsigned char *buffer)
2617{
2618 static const u8 cmd[10] = { READ_CAPACITY };
2619 struct scsi_sense_hdr sshdr;
2620 struct scsi_failure failure_defs[] = {
2621 /* Do not retry Medium Not Present */
2622 {
2623 .sense = UNIT_ATTENTION,
2624 .asc = 0x3A,
2625 .result = SAM_STAT_CHECK_CONDITION,
2626 },
2627 {
2628 .sense = NOT_READY,
2629 .asc = 0x3A,
2630 .result = SAM_STAT_CHECK_CONDITION,
2631 },
2632 /* Device reset might occur several times so retry a lot */
2633 {
2634 .sense = UNIT_ATTENTION,
2635 .asc = 0x29,
2636 .allowed = READ_CAPACITY_RETRIES_ON_RESET,
2637 .result = SAM_STAT_CHECK_CONDITION,
2638 },
2639 /* Any other error not listed above retry 3 times */
2640 {
2641 .result = SCMD_FAILURE_RESULT_ANY,
2642 .allowed = 3,
2643 },
2644 {}
2645 };
2646 struct scsi_failures failures = {
2647 .failure_definitions = failure_defs,
2648 };
2649 const struct scsi_exec_args exec_args = {
2650 .sshdr = &sshdr,
2651 .failures = &failures,
2652 };
2653 int sense_valid = 0;
2654 int the_result;
2655 sector_t lba;
2656 unsigned sector_size;
2657
2658 memset(buffer, 0, 8);
2659
2660 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2661 8, SD_TIMEOUT, sdkp->max_retries,
2662 &exec_args);
2663
2664 if (the_result > 0) {
2665 sense_valid = scsi_sense_valid(&sshdr);
2666
2667 if (media_not_present(sdkp, &sshdr))
2668 return -ENODEV;
2669 }
2670
2671 if (the_result) {
2672 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2673 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2674 return -EINVAL;
2675 }
2676
2677 sector_size = get_unaligned_be32(&buffer[4]);
2678 lba = get_unaligned_be32(&buffer[0]);
2679
2680 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2681 /* Some buggy (usb cardreader) devices return an lba of
2682 0xffffffff when the want to report a size of 0 (with
2683 which they really mean no media is present) */
2684 sdkp->capacity = 0;
2685 sdkp->physical_block_size = sector_size;
2686 return sector_size;
2687 }
2688
2689 sdkp->capacity = lba + 1;
2690 sdkp->physical_block_size = sector_size;
2691 return sector_size;
2692}
2693
2694static int sd_try_rc16_first(struct scsi_device *sdp)
2695{
2696 if (sdp->host->max_cmd_len < 16)
2697 return 0;
2698 if (sdp->try_rc_10_first)
2699 return 0;
2700 if (sdp->scsi_level > SCSI_SPC_2)
2701 return 1;
2702 if (scsi_device_protection(sdp))
2703 return 1;
2704 return 0;
2705}
2706
2707/*
2708 * read disk capacity
2709 */
2710static void
2711sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2712{
2713 int sector_size;
2714 struct scsi_device *sdp = sdkp->device;
2715
2716 if (sd_try_rc16_first(sdp)) {
2717 sector_size = read_capacity_16(sdkp, sdp, buffer);
2718 if (sector_size == -EOVERFLOW)
2719 goto got_data;
2720 if (sector_size == -ENODEV)
2721 return;
2722 if (sector_size < 0)
2723 sector_size = read_capacity_10(sdkp, sdp, buffer);
2724 if (sector_size < 0)
2725 return;
2726 } else {
2727 sector_size = read_capacity_10(sdkp, sdp, buffer);
2728 if (sector_size == -EOVERFLOW)
2729 goto got_data;
2730 if (sector_size < 0)
2731 return;
2732 if ((sizeof(sdkp->capacity) > 4) &&
2733 (sdkp->capacity > 0xffffffffULL)) {
2734 int old_sector_size = sector_size;
2735 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2736 "Trying to use READ CAPACITY(16).\n");
2737 sector_size = read_capacity_16(sdkp, sdp, buffer);
2738 if (sector_size < 0) {
2739 sd_printk(KERN_NOTICE, sdkp,
2740 "Using 0xffffffff as device size\n");
2741 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2742 sector_size = old_sector_size;
2743 goto got_data;
2744 }
2745 /* Remember that READ CAPACITY(16) succeeded */
2746 sdp->try_rc_10_first = 0;
2747 }
2748 }
2749
2750 /* Some devices are known to return the total number of blocks,
2751 * not the highest block number. Some devices have versions
2752 * which do this and others which do not. Some devices we might
2753 * suspect of doing this but we don't know for certain.
2754 *
2755 * If we know the reported capacity is wrong, decrement it. If
2756 * we can only guess, then assume the number of blocks is even
2757 * (usually true but not always) and err on the side of lowering
2758 * the capacity.
2759 */
2760 if (sdp->fix_capacity ||
2761 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2762 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2763 "from its reported value: %llu\n",
2764 (unsigned long long) sdkp->capacity);
2765 --sdkp->capacity;
2766 }
2767
2768got_data:
2769 if (sector_size == 0) {
2770 sector_size = 512;
2771 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2772 "assuming 512.\n");
2773 }
2774
2775 if (sector_size != 512 &&
2776 sector_size != 1024 &&
2777 sector_size != 2048 &&
2778 sector_size != 4096) {
2779 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2780 sector_size);
2781 /*
2782 * The user might want to re-format the drive with
2783 * a supported sectorsize. Once this happens, it
2784 * would be relatively trivial to set the thing up.
2785 * For this reason, we leave the thing in the table.
2786 */
2787 sdkp->capacity = 0;
2788 /*
2789 * set a bogus sector size so the normal read/write
2790 * logic in the block layer will eventually refuse any
2791 * request on this device without tripping over power
2792 * of two sector size assumptions
2793 */
2794 sector_size = 512;
2795 }
2796 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2797 blk_queue_physical_block_size(sdp->request_queue,
2798 sdkp->physical_block_size);
2799 sdkp->device->sector_size = sector_size;
2800
2801 if (sdkp->capacity > 0xffffffff)
2802 sdp->use_16_for_rw = 1;
2803
2804}
2805
2806/*
2807 * Print disk capacity
2808 */
2809static void
2810sd_print_capacity(struct scsi_disk *sdkp,
2811 sector_t old_capacity)
2812{
2813 int sector_size = sdkp->device->sector_size;
2814 char cap_str_2[10], cap_str_10[10];
2815
2816 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2817 return;
2818
2819 string_get_size(sdkp->capacity, sector_size,
2820 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2821 string_get_size(sdkp->capacity, sector_size,
2822 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2823
2824 sd_printk(KERN_NOTICE, sdkp,
2825 "%llu %d-byte logical blocks: (%s/%s)\n",
2826 (unsigned long long)sdkp->capacity,
2827 sector_size, cap_str_10, cap_str_2);
2828
2829 if (sdkp->physical_block_size != sector_size)
2830 sd_printk(KERN_NOTICE, sdkp,
2831 "%u-byte physical blocks\n",
2832 sdkp->physical_block_size);
2833}
2834
2835/* called with buffer of length 512 */
2836static inline int
2837sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2838 unsigned char *buffer, int len, struct scsi_mode_data *data,
2839 struct scsi_sense_hdr *sshdr)
2840{
2841 /*
2842 * If we must use MODE SENSE(10), make sure that the buffer length
2843 * is at least 8 bytes so that the mode sense header fits.
2844 */
2845 if (sdkp->device->use_10_for_ms && len < 8)
2846 len = 8;
2847
2848 return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2849 SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2850}
2851
2852/*
2853 * read write protect setting, if possible - called only in sd_revalidate_disk()
2854 * called with buffer of length SD_BUF_SIZE
2855 */
2856static void
2857sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2858{
2859 int res;
2860 struct scsi_device *sdp = sdkp->device;
2861 struct scsi_mode_data data;
2862 int old_wp = sdkp->write_prot;
2863
2864 set_disk_ro(sdkp->disk, 0);
2865 if (sdp->skip_ms_page_3f) {
2866 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2867 return;
2868 }
2869
2870 if (sdp->use_192_bytes_for_3f) {
2871 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2872 } else {
2873 /*
2874 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2875 * We have to start carefully: some devices hang if we ask
2876 * for more than is available.
2877 */
2878 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2879
2880 /*
2881 * Second attempt: ask for page 0 When only page 0 is
2882 * implemented, a request for page 3F may return Sense Key
2883 * 5: Illegal Request, Sense Code 24: Invalid field in
2884 * CDB.
2885 */
2886 if (res < 0)
2887 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2888
2889 /*
2890 * Third attempt: ask 255 bytes, as we did earlier.
2891 */
2892 if (res < 0)
2893 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2894 &data, NULL);
2895 }
2896
2897 if (res < 0) {
2898 sd_first_printk(KERN_WARNING, sdkp,
2899 "Test WP failed, assume Write Enabled\n");
2900 } else {
2901 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2902 set_disk_ro(sdkp->disk, sdkp->write_prot);
2903 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2904 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2905 sdkp->write_prot ? "on" : "off");
2906 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2907 }
2908 }
2909}
2910
2911/*
2912 * sd_read_cache_type - called only from sd_revalidate_disk()
2913 * called with buffer of length SD_BUF_SIZE
2914 */
2915static void
2916sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2917{
2918 int len = 0, res;
2919 struct scsi_device *sdp = sdkp->device;
2920
2921 int dbd;
2922 int modepage;
2923 int first_len;
2924 struct scsi_mode_data data;
2925 struct scsi_sense_hdr sshdr;
2926 int old_wce = sdkp->WCE;
2927 int old_rcd = sdkp->RCD;
2928 int old_dpofua = sdkp->DPOFUA;
2929
2930
2931 if (sdkp->cache_override)
2932 return;
2933
2934 first_len = 4;
2935 if (sdp->skip_ms_page_8) {
2936 if (sdp->type == TYPE_RBC)
2937 goto defaults;
2938 else {
2939 if (sdp->skip_ms_page_3f)
2940 goto defaults;
2941 modepage = 0x3F;
2942 if (sdp->use_192_bytes_for_3f)
2943 first_len = 192;
2944 dbd = 0;
2945 }
2946 } else if (sdp->type == TYPE_RBC) {
2947 modepage = 6;
2948 dbd = 8;
2949 } else {
2950 modepage = 8;
2951 dbd = 0;
2952 }
2953
2954 /* cautiously ask */
2955 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2956 &data, &sshdr);
2957
2958 if (res < 0)
2959 goto bad_sense;
2960
2961 if (!data.header_length) {
2962 modepage = 6;
2963 first_len = 0;
2964 sd_first_printk(KERN_ERR, sdkp,
2965 "Missing header in MODE_SENSE response\n");
2966 }
2967
2968 /* that went OK, now ask for the proper length */
2969 len = data.length;
2970
2971 /*
2972 * We're only interested in the first three bytes, actually.
2973 * But the data cache page is defined for the first 20.
2974 */
2975 if (len < 3)
2976 goto bad_sense;
2977 else if (len > SD_BUF_SIZE) {
2978 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2979 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2980 len = SD_BUF_SIZE;
2981 }
2982 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2983 len = 192;
2984
2985 /* Get the data */
2986 if (len > first_len)
2987 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2988 &data, &sshdr);
2989
2990 if (!res) {
2991 int offset = data.header_length + data.block_descriptor_length;
2992
2993 while (offset < len) {
2994 u8 page_code = buffer[offset] & 0x3F;
2995 u8 spf = buffer[offset] & 0x40;
2996
2997 if (page_code == 8 || page_code == 6) {
2998 /* We're interested only in the first 3 bytes.
2999 */
3000 if (len - offset <= 2) {
3001 sd_first_printk(KERN_ERR, sdkp,
3002 "Incomplete mode parameter "
3003 "data\n");
3004 goto defaults;
3005 } else {
3006 modepage = page_code;
3007 goto Page_found;
3008 }
3009 } else {
3010 /* Go to the next page */
3011 if (spf && len - offset > 3)
3012 offset += 4 + (buffer[offset+2] << 8) +
3013 buffer[offset+3];
3014 else if (!spf && len - offset > 1)
3015 offset += 2 + buffer[offset+1];
3016 else {
3017 sd_first_printk(KERN_ERR, sdkp,
3018 "Incomplete mode "
3019 "parameter data\n");
3020 goto defaults;
3021 }
3022 }
3023 }
3024
3025 sd_first_printk(KERN_WARNING, sdkp,
3026 "No Caching mode page found\n");
3027 goto defaults;
3028
3029 Page_found:
3030 if (modepage == 8) {
3031 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3032 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3033 } else {
3034 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3035 sdkp->RCD = 0;
3036 }
3037
3038 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3039 if (sdp->broken_fua) {
3040 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3041 sdkp->DPOFUA = 0;
3042 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3043 !sdkp->device->use_16_for_rw) {
3044 sd_first_printk(KERN_NOTICE, sdkp,
3045 "Uses READ/WRITE(6), disabling FUA\n");
3046 sdkp->DPOFUA = 0;
3047 }
3048
3049 /* No cache flush allowed for write protected devices */
3050 if (sdkp->WCE && sdkp->write_prot)
3051 sdkp->WCE = 0;
3052
3053 if (sdkp->first_scan || old_wce != sdkp->WCE ||
3054 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3055 sd_printk(KERN_NOTICE, sdkp,
3056 "Write cache: %s, read cache: %s, %s\n",
3057 sdkp->WCE ? "enabled" : "disabled",
3058 sdkp->RCD ? "disabled" : "enabled",
3059 sdkp->DPOFUA ? "supports DPO and FUA"
3060 : "doesn't support DPO or FUA");
3061
3062 return;
3063 }
3064
3065bad_sense:
3066 if (res == -EIO && scsi_sense_valid(&sshdr) &&
3067 sshdr.sense_key == ILLEGAL_REQUEST &&
3068 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3069 /* Invalid field in CDB */
3070 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3071 else
3072 sd_first_printk(KERN_ERR, sdkp,
3073 "Asking for cache data failed\n");
3074
3075defaults:
3076 if (sdp->wce_default_on) {
3077 sd_first_printk(KERN_NOTICE, sdkp,
3078 "Assuming drive cache: write back\n");
3079 sdkp->WCE = 1;
3080 } else {
3081 sd_first_printk(KERN_WARNING, sdkp,
3082 "Assuming drive cache: write through\n");
3083 sdkp->WCE = 0;
3084 }
3085 sdkp->RCD = 0;
3086 sdkp->DPOFUA = 0;
3087}
3088
3089static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3090{
3091 u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3092 struct {
3093 struct scsi_stream_status_header h;
3094 struct scsi_stream_status s;
3095 } buf;
3096 struct scsi_device *sdev = sdkp->device;
3097 struct scsi_sense_hdr sshdr;
3098 const struct scsi_exec_args exec_args = {
3099 .sshdr = &sshdr,
3100 };
3101 int res;
3102
3103 put_unaligned_be16(stream_id, &cdb[4]);
3104 put_unaligned_be32(sizeof(buf), &cdb[10]);
3105
3106 res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3107 SD_TIMEOUT, sdkp->max_retries, &exec_args);
3108 if (res < 0)
3109 return false;
3110 if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3111 sd_print_sense_hdr(sdkp, &sshdr);
3112 if (res)
3113 return false;
3114 if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3115 return false;
3116 return buf.h.stream_status[0].perm;
3117}
3118
3119static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3120{
3121 struct scsi_device *sdp = sdkp->device;
3122 const struct scsi_io_group_descriptor *desc, *start, *end;
3123 u16 permanent_stream_count_old;
3124 struct scsi_sense_hdr sshdr;
3125 struct scsi_mode_data data;
3126 int res;
3127
3128 res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3129 /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3130 sdkp->max_retries, &data, &sshdr);
3131 if (res < 0)
3132 return;
3133 start = (void *)buffer + data.header_length + 16;
3134 end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3135 sizeof(*end));
3136 /*
3137 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3138 * should assign the lowest numbered stream identifiers to permanent
3139 * streams.
3140 */
3141 for (desc = start; desc < end; desc++)
3142 if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3143 break;
3144 permanent_stream_count_old = sdkp->permanent_stream_count;
3145 sdkp->permanent_stream_count = desc - start;
3146 if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3147 sd_printk(KERN_INFO, sdkp,
3148 "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3149 sdkp->permanent_stream_count);
3150 else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3151 sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3152 sdkp->permanent_stream_count);
3153}
3154
3155/*
3156 * The ATO bit indicates whether the DIF application tag is available
3157 * for use by the operating system.
3158 */
3159static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3160{
3161 int res, offset;
3162 struct scsi_device *sdp = sdkp->device;
3163 struct scsi_mode_data data;
3164 struct scsi_sense_hdr sshdr;
3165
3166 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3167 return;
3168
3169 if (sdkp->protection_type == 0)
3170 return;
3171
3172 res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3173 sdkp->max_retries, &data, &sshdr);
3174
3175 if (res < 0 || !data.header_length ||
3176 data.length < 6) {
3177 sd_first_printk(KERN_WARNING, sdkp,
3178 "getting Control mode page failed, assume no ATO\n");
3179
3180 if (res == -EIO && scsi_sense_valid(&sshdr))
3181 sd_print_sense_hdr(sdkp, &sshdr);
3182
3183 return;
3184 }
3185
3186 offset = data.header_length + data.block_descriptor_length;
3187
3188 if ((buffer[offset] & 0x3f) != 0x0a) {
3189 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3190 return;
3191 }
3192
3193 if ((buffer[offset + 5] & 0x80) == 0)
3194 return;
3195
3196 sdkp->ATO = 1;
3197
3198 return;
3199}
3200
3201/**
3202 * sd_read_block_limits - Query disk device for preferred I/O sizes.
3203 * @sdkp: disk to query
3204 */
3205static void sd_read_block_limits(struct scsi_disk *sdkp)
3206{
3207 struct scsi_vpd *vpd;
3208
3209 rcu_read_lock();
3210
3211 vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3212 if (!vpd || vpd->len < 16)
3213 goto out;
3214
3215 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3216 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3217 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3218
3219 if (vpd->len >= 64) {
3220 unsigned int lba_count, desc_count;
3221
3222 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3223
3224 if (!sdkp->lbpme)
3225 goto out;
3226
3227 lba_count = get_unaligned_be32(&vpd->data[20]);
3228 desc_count = get_unaligned_be32(&vpd->data[24]);
3229
3230 if (lba_count && desc_count)
3231 sdkp->max_unmap_blocks = lba_count;
3232
3233 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3234
3235 if (vpd->data[32] & 0x80)
3236 sdkp->unmap_alignment =
3237 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3238
3239 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3240
3241 if (sdkp->max_unmap_blocks)
3242 sd_config_discard(sdkp, SD_LBP_UNMAP);
3243 else
3244 sd_config_discard(sdkp, SD_LBP_WS16);
3245
3246 } else { /* LBP VPD page tells us what to use */
3247 if (sdkp->lbpu && sdkp->max_unmap_blocks)
3248 sd_config_discard(sdkp, SD_LBP_UNMAP);
3249 else if (sdkp->lbpws)
3250 sd_config_discard(sdkp, SD_LBP_WS16);
3251 else if (sdkp->lbpws10)
3252 sd_config_discard(sdkp, SD_LBP_WS10);
3253 else
3254 sd_config_discard(sdkp, SD_LBP_DISABLE);
3255 }
3256 }
3257
3258 out:
3259 rcu_read_unlock();
3260}
3261
3262/* Parse the Block Limits Extension VPD page (0xb7) */
3263static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3264{
3265 struct scsi_vpd *vpd;
3266
3267 rcu_read_lock();
3268 vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3269 if (vpd && vpd->len >= 2)
3270 sdkp->rscs = vpd->data[5] & 1;
3271 rcu_read_unlock();
3272}
3273
3274/**
3275 * sd_read_block_characteristics - Query block dev. characteristics
3276 * @sdkp: disk to query
3277 */
3278static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3279{
3280 struct request_queue *q = sdkp->disk->queue;
3281 struct scsi_vpd *vpd;
3282 u16 rot;
3283
3284 rcu_read_lock();
3285 vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3286
3287 if (!vpd || vpd->len < 8) {
3288 rcu_read_unlock();
3289 return;
3290 }
3291
3292 rot = get_unaligned_be16(&vpd->data[4]);
3293 sdkp->zoned = (vpd->data[8] >> 4) & 3;
3294 rcu_read_unlock();
3295
3296 if (rot == 1) {
3297 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3298 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3299 }
3300
3301
3302#ifdef CONFIG_BLK_DEV_ZONED /* sd_probe rejects ZBD devices early otherwise */
3303 if (sdkp->device->type == TYPE_ZBC) {
3304 /*
3305 * Host-managed.
3306 */
3307 disk_set_zoned(sdkp->disk);
3308
3309 /*
3310 * Per ZBC and ZAC specifications, writes in sequential write
3311 * required zones of host-managed devices must be aligned to
3312 * the device physical block size.
3313 */
3314 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3315 } else {
3316 /*
3317 * Host-aware devices are treated as conventional.
3318 */
3319 WARN_ON_ONCE(blk_queue_is_zoned(q));
3320 }
3321#endif /* CONFIG_BLK_DEV_ZONED */
3322
3323 if (!sdkp->first_scan)
3324 return;
3325
3326 if (blk_queue_is_zoned(q))
3327 sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3328 else if (sdkp->zoned == 1)
3329 sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3330 else if (sdkp->zoned == 2)
3331 sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3332}
3333
3334/**
3335 * sd_read_block_provisioning - Query provisioning VPD page
3336 * @sdkp: disk to query
3337 */
3338static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3339{
3340 struct scsi_vpd *vpd;
3341
3342 if (sdkp->lbpme == 0)
3343 return;
3344
3345 rcu_read_lock();
3346 vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3347
3348 if (!vpd || vpd->len < 8) {
3349 rcu_read_unlock();
3350 return;
3351 }
3352
3353 sdkp->lbpvpd = 1;
3354 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */
3355 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3356 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3357 rcu_read_unlock();
3358}
3359
3360static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3361{
3362 struct scsi_device *sdev = sdkp->device;
3363
3364 if (sdev->host->no_write_same) {
3365 sdev->no_write_same = 1;
3366
3367 return;
3368 }
3369
3370 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3371 struct scsi_vpd *vpd;
3372
3373 sdev->no_report_opcodes = 1;
3374
3375 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3376 * CODES is unsupported and the device has an ATA
3377 * Information VPD page (SAT).
3378 */
3379 rcu_read_lock();
3380 vpd = rcu_dereference(sdev->vpd_pg89);
3381 if (vpd)
3382 sdev->no_write_same = 1;
3383 rcu_read_unlock();
3384 }
3385
3386 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3387 sdkp->ws16 = 1;
3388
3389 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3390 sdkp->ws10 = 1;
3391}
3392
3393static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3394{
3395 struct scsi_device *sdev = sdkp->device;
3396
3397 if (!sdev->security_supported)
3398 return;
3399
3400 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3401 SECURITY_PROTOCOL_IN, 0) == 1 &&
3402 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3403 SECURITY_PROTOCOL_OUT, 0) == 1)
3404 sdkp->security = 1;
3405}
3406
3407static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3408{
3409 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3410}
3411
3412/**
3413 * sd_read_cpr - Query concurrent positioning ranges
3414 * @sdkp: disk to query
3415 */
3416static void sd_read_cpr(struct scsi_disk *sdkp)
3417{
3418 struct blk_independent_access_ranges *iars = NULL;
3419 unsigned char *buffer = NULL;
3420 unsigned int nr_cpr = 0;
3421 int i, vpd_len, buf_len = SD_BUF_SIZE;
3422 u8 *desc;
3423
3424 /*
3425 * We need to have the capacity set first for the block layer to be
3426 * able to check the ranges.
3427 */
3428 if (sdkp->first_scan)
3429 return;
3430
3431 if (!sdkp->capacity)
3432 goto out;
3433
3434 /*
3435 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3436 * leading to a maximum page size of 64 + 256*32 bytes.
3437 */
3438 buf_len = 64 + 256*32;
3439 buffer = kmalloc(buf_len, GFP_KERNEL);
3440 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3441 goto out;
3442
3443 /* We must have at least a 64B header and one 32B range descriptor */
3444 vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3445 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3446 sd_printk(KERN_ERR, sdkp,
3447 "Invalid Concurrent Positioning Ranges VPD page\n");
3448 goto out;
3449 }
3450
3451 nr_cpr = (vpd_len - 64) / 32;
3452 if (nr_cpr == 1) {
3453 nr_cpr = 0;
3454 goto out;
3455 }
3456
3457 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3458 if (!iars) {
3459 nr_cpr = 0;
3460 goto out;
3461 }
3462
3463 desc = &buffer[64];
3464 for (i = 0; i < nr_cpr; i++, desc += 32) {
3465 if (desc[0] != i) {
3466 sd_printk(KERN_ERR, sdkp,
3467 "Invalid Concurrent Positioning Range number\n");
3468 nr_cpr = 0;
3469 break;
3470 }
3471
3472 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3473 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3474 }
3475
3476out:
3477 disk_set_independent_access_ranges(sdkp->disk, iars);
3478 if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3479 sd_printk(KERN_NOTICE, sdkp,
3480 "%u concurrent positioning ranges\n", nr_cpr);
3481 sdkp->nr_actuators = nr_cpr;
3482 }
3483
3484 kfree(buffer);
3485}
3486
3487static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3488{
3489 struct scsi_device *sdp = sdkp->device;
3490 unsigned int min_xfer_bytes =
3491 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3492
3493 if (sdkp->min_xfer_blocks == 0)
3494 return false;
3495
3496 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3497 sd_first_printk(KERN_WARNING, sdkp,
3498 "Preferred minimum I/O size %u bytes not a " \
3499 "multiple of physical block size (%u bytes)\n",
3500 min_xfer_bytes, sdkp->physical_block_size);
3501 sdkp->min_xfer_blocks = 0;
3502 return false;
3503 }
3504
3505 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3506 min_xfer_bytes);
3507 return true;
3508}
3509
3510/*
3511 * Determine the device's preferred I/O size for reads and writes
3512 * unless the reported value is unreasonably small, large, not a
3513 * multiple of the physical block size, or simply garbage.
3514 */
3515static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3516 unsigned int dev_max)
3517{
3518 struct scsi_device *sdp = sdkp->device;
3519 unsigned int opt_xfer_bytes =
3520 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3521 unsigned int min_xfer_bytes =
3522 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3523
3524 if (sdkp->opt_xfer_blocks == 0)
3525 return false;
3526
3527 if (sdkp->opt_xfer_blocks > dev_max) {
3528 sd_first_printk(KERN_WARNING, sdkp,
3529 "Optimal transfer size %u logical blocks " \
3530 "> dev_max (%u logical blocks)\n",
3531 sdkp->opt_xfer_blocks, dev_max);
3532 return false;
3533 }
3534
3535 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3536 sd_first_printk(KERN_WARNING, sdkp,
3537 "Optimal transfer size %u logical blocks " \
3538 "> sd driver limit (%u logical blocks)\n",
3539 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3540 return false;
3541 }
3542
3543 if (opt_xfer_bytes < PAGE_SIZE) {
3544 sd_first_printk(KERN_WARNING, sdkp,
3545 "Optimal transfer size %u bytes < " \
3546 "PAGE_SIZE (%u bytes)\n",
3547 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3548 return false;
3549 }
3550
3551 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3552 sd_first_printk(KERN_WARNING, sdkp,
3553 "Optimal transfer size %u bytes not a " \
3554 "multiple of preferred minimum block " \
3555 "size (%u bytes)\n",
3556 opt_xfer_bytes, min_xfer_bytes);
3557 return false;
3558 }
3559
3560 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3561 sd_first_printk(KERN_WARNING, sdkp,
3562 "Optimal transfer size %u bytes not a " \
3563 "multiple of physical block size (%u bytes)\n",
3564 opt_xfer_bytes, sdkp->physical_block_size);
3565 return false;
3566 }
3567
3568 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3569 opt_xfer_bytes);
3570 return true;
3571}
3572
3573static void sd_read_block_zero(struct scsi_disk *sdkp)
3574{
3575 unsigned int buf_len = sdkp->device->sector_size;
3576 char *buffer, cmd[10] = { };
3577
3578 buffer = kmalloc(buf_len, GFP_KERNEL);
3579 if (!buffer)
3580 return;
3581
3582 cmd[0] = READ_10;
3583 put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3584 put_unaligned_be16(1, &cmd[7]); /* Transfer 1 logical block */
3585
3586 scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3587 SD_TIMEOUT, sdkp->max_retries, NULL);
3588 kfree(buffer);
3589}
3590
3591/**
3592 * sd_revalidate_disk - called the first time a new disk is seen,
3593 * performs disk spin up, read_capacity, etc.
3594 * @disk: struct gendisk we care about
3595 **/
3596static int sd_revalidate_disk(struct gendisk *disk)
3597{
3598 struct scsi_disk *sdkp = scsi_disk(disk);
3599 struct scsi_device *sdp = sdkp->device;
3600 struct request_queue *q = sdkp->disk->queue;
3601 sector_t old_capacity = sdkp->capacity;
3602 unsigned char *buffer;
3603 unsigned int dev_max, rw_max;
3604
3605 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3606 "sd_revalidate_disk\n"));
3607
3608 /*
3609 * If the device is offline, don't try and read capacity or any
3610 * of the other niceties.
3611 */
3612 if (!scsi_device_online(sdp))
3613 goto out;
3614
3615 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3616 if (!buffer) {
3617 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3618 "allocation failure.\n");
3619 goto out;
3620 }
3621
3622 sd_spinup_disk(sdkp);
3623
3624 /*
3625 * Without media there is no reason to ask; moreover, some devices
3626 * react badly if we do.
3627 */
3628 if (sdkp->media_present) {
3629 sd_read_capacity(sdkp, buffer);
3630 /*
3631 * Some USB/UAS devices return generic values for mode pages
3632 * until the media has been accessed. Trigger a READ operation
3633 * to force the device to populate mode pages.
3634 */
3635 if (sdp->read_before_ms)
3636 sd_read_block_zero(sdkp);
3637 /*
3638 * set the default to rotational. All non-rotational devices
3639 * support the block characteristics VPD page, which will
3640 * cause this to be updated correctly and any device which
3641 * doesn't support it should be treated as rotational.
3642 */
3643 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3644 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3645
3646 if (scsi_device_supports_vpd(sdp)) {
3647 sd_read_block_provisioning(sdkp);
3648 sd_read_block_limits(sdkp);
3649 sd_read_block_limits_ext(sdkp);
3650 sd_read_block_characteristics(sdkp);
3651 sd_zbc_read_zones(sdkp, buffer);
3652 sd_read_cpr(sdkp);
3653 }
3654
3655 sd_print_capacity(sdkp, old_capacity);
3656
3657 sd_read_write_protect_flag(sdkp, buffer);
3658 sd_read_cache_type(sdkp, buffer);
3659 sd_read_io_hints(sdkp, buffer);
3660 sd_read_app_tag_own(sdkp, buffer);
3661 sd_read_write_same(sdkp, buffer);
3662 sd_read_security(sdkp, buffer);
3663 sd_config_protection(sdkp);
3664 }
3665
3666 /*
3667 * We now have all cache related info, determine how we deal
3668 * with flush requests.
3669 */
3670 sd_set_flush_flag(sdkp);
3671
3672 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3673 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3674
3675 /* Some devices report a maximum block count for READ/WRITE requests. */
3676 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3677 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3678
3679 if (sd_validate_min_xfer_size(sdkp))
3680 blk_queue_io_min(sdkp->disk->queue,
3681 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3682 else
3683 blk_queue_io_min(sdkp->disk->queue, 0);
3684
3685 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3686 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3687 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3688 } else {
3689 q->limits.io_opt = 0;
3690 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3691 (sector_t)BLK_DEF_MAX_SECTORS_CAP);
3692 }
3693
3694 /*
3695 * Limit default to SCSI host optimal sector limit if set. There may be
3696 * an impact on performance for when the size of a request exceeds this
3697 * host limit.
3698 */
3699 rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3700
3701 /* Do not exceed controller limit */
3702 rw_max = min(rw_max, queue_max_hw_sectors(q));
3703
3704 /*
3705 * Only update max_sectors if previously unset or if the current value
3706 * exceeds the capabilities of the hardware.
3707 */
3708 if (sdkp->first_scan ||
3709 q->limits.max_sectors > q->limits.max_dev_sectors ||
3710 q->limits.max_sectors > q->limits.max_hw_sectors) {
3711 q->limits.max_sectors = rw_max;
3712 q->limits.max_user_sectors = rw_max;
3713 }
3714
3715 sdkp->first_scan = 0;
3716
3717 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3718 sd_config_write_same(sdkp);
3719 kfree(buffer);
3720
3721 /*
3722 * For a zoned drive, revalidating the zones can be done only once
3723 * the gendisk capacity is set. So if this fails, set back the gendisk
3724 * capacity to 0.
3725 */
3726 if (sd_zbc_revalidate_zones(sdkp))
3727 set_capacity_and_notify(disk, 0);
3728
3729 out:
3730 return 0;
3731}
3732
3733/**
3734 * sd_unlock_native_capacity - unlock native capacity
3735 * @disk: struct gendisk to set capacity for
3736 *
3737 * Block layer calls this function if it detects that partitions
3738 * on @disk reach beyond the end of the device. If the SCSI host
3739 * implements ->unlock_native_capacity() method, it's invoked to
3740 * give it a chance to adjust the device capacity.
3741 *
3742 * CONTEXT:
3743 * Defined by block layer. Might sleep.
3744 */
3745static void sd_unlock_native_capacity(struct gendisk *disk)
3746{
3747 struct scsi_device *sdev = scsi_disk(disk)->device;
3748
3749 if (sdev->host->hostt->unlock_native_capacity)
3750 sdev->host->hostt->unlock_native_capacity(sdev);
3751}
3752
3753/**
3754 * sd_format_disk_name - format disk name
3755 * @prefix: name prefix - ie. "sd" for SCSI disks
3756 * @index: index of the disk to format name for
3757 * @buf: output buffer
3758 * @buflen: length of the output buffer
3759 *
3760 * SCSI disk names starts at sda. The 26th device is sdz and the
3761 * 27th is sdaa. The last one for two lettered suffix is sdzz
3762 * which is followed by sdaaa.
3763 *
3764 * This is basically 26 base counting with one extra 'nil' entry
3765 * at the beginning from the second digit on and can be
3766 * determined using similar method as 26 base conversion with the
3767 * index shifted -1 after each digit is computed.
3768 *
3769 * CONTEXT:
3770 * Don't care.
3771 *
3772 * RETURNS:
3773 * 0 on success, -errno on failure.
3774 */
3775static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3776{
3777 const int base = 'z' - 'a' + 1;
3778 char *begin = buf + strlen(prefix);
3779 char *end = buf + buflen;
3780 char *p;
3781 int unit;
3782
3783 p = end - 1;
3784 *p = '\0';
3785 unit = base;
3786 do {
3787 if (p == begin)
3788 return -EINVAL;
3789 *--p = 'a' + (index % unit);
3790 index = (index / unit) - 1;
3791 } while (index >= 0);
3792
3793 memmove(begin, p, end - p);
3794 memcpy(buf, prefix, strlen(prefix));
3795
3796 return 0;
3797}
3798
3799/**
3800 * sd_probe - called during driver initialization and whenever a
3801 * new scsi device is attached to the system. It is called once
3802 * for each scsi device (not just disks) present.
3803 * @dev: pointer to device object
3804 *
3805 * Returns 0 if successful (or not interested in this scsi device
3806 * (e.g. scanner)); 1 when there is an error.
3807 *
3808 * Note: this function is invoked from the scsi mid-level.
3809 * This function sets up the mapping between a given
3810 * <host,channel,id,lun> (found in sdp) and new device name
3811 * (e.g. /dev/sda). More precisely it is the block device major
3812 * and minor number that is chosen here.
3813 *
3814 * Assume sd_probe is not re-entrant (for time being)
3815 * Also think about sd_probe() and sd_remove() running coincidentally.
3816 **/
3817static int sd_probe(struct device *dev)
3818{
3819 struct scsi_device *sdp = to_scsi_device(dev);
3820 struct scsi_disk *sdkp;
3821 struct gendisk *gd;
3822 int index;
3823 int error;
3824
3825 scsi_autopm_get_device(sdp);
3826 error = -ENODEV;
3827 if (sdp->type != TYPE_DISK &&
3828 sdp->type != TYPE_ZBC &&
3829 sdp->type != TYPE_MOD &&
3830 sdp->type != TYPE_RBC)
3831 goto out;
3832
3833 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3834 sdev_printk(KERN_WARNING, sdp,
3835 "Unsupported ZBC host-managed device.\n");
3836 goto out;
3837 }
3838
3839 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3840 "sd_probe\n"));
3841
3842 error = -ENOMEM;
3843 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3844 if (!sdkp)
3845 goto out;
3846
3847 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3848 &sd_bio_compl_lkclass);
3849 if (!gd)
3850 goto out_free;
3851
3852 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3853 if (index < 0) {
3854 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3855 goto out_put;
3856 }
3857
3858 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3859 if (error) {
3860 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3861 goto out_free_index;
3862 }
3863
3864 sdkp->device = sdp;
3865 sdkp->disk = gd;
3866 sdkp->index = index;
3867 sdkp->max_retries = SD_MAX_RETRIES;
3868 atomic_set(&sdkp->openers, 0);
3869 atomic_set(&sdkp->device->ioerr_cnt, 0);
3870
3871 if (!sdp->request_queue->rq_timeout) {
3872 if (sdp->type != TYPE_MOD)
3873 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3874 else
3875 blk_queue_rq_timeout(sdp->request_queue,
3876 SD_MOD_TIMEOUT);
3877 }
3878
3879 device_initialize(&sdkp->disk_dev);
3880 sdkp->disk_dev.parent = get_device(dev);
3881 sdkp->disk_dev.class = &sd_disk_class;
3882 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3883
3884 error = device_add(&sdkp->disk_dev);
3885 if (error) {
3886 put_device(&sdkp->disk_dev);
3887 goto out;
3888 }
3889
3890 dev_set_drvdata(dev, sdkp);
3891
3892 gd->major = sd_major((index & 0xf0) >> 4);
3893 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3894 gd->minors = SD_MINORS;
3895
3896 gd->fops = &sd_fops;
3897 gd->private_data = sdkp;
3898
3899 /* defaults, until the device tells us otherwise */
3900 sdp->sector_size = 512;
3901 sdkp->capacity = 0;
3902 sdkp->media_present = 1;
3903 sdkp->write_prot = 0;
3904 sdkp->cache_override = 0;
3905 sdkp->WCE = 0;
3906 sdkp->RCD = 0;
3907 sdkp->ATO = 0;
3908 sdkp->first_scan = 1;
3909 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3910
3911 sd_revalidate_disk(gd);
3912
3913 if (sdp->removable) {
3914 gd->flags |= GENHD_FL_REMOVABLE;
3915 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3916 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3917 }
3918
3919 blk_pm_runtime_init(sdp->request_queue, dev);
3920 if (sdp->rpm_autosuspend) {
3921 pm_runtime_set_autosuspend_delay(dev,
3922 sdp->host->rpm_autosuspend_delay);
3923 }
3924
3925 error = device_add_disk(dev, gd, NULL);
3926 if (error) {
3927 device_unregister(&sdkp->disk_dev);
3928 put_disk(gd);
3929 goto out;
3930 }
3931
3932 if (sdkp->security) {
3933 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3934 if (sdkp->opal_dev)
3935 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3936 }
3937
3938 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3939 sdp->removable ? "removable " : "");
3940 scsi_autopm_put_device(sdp);
3941
3942 return 0;
3943
3944 out_free_index:
3945 ida_free(&sd_index_ida, index);
3946 out_put:
3947 put_disk(gd);
3948 out_free:
3949 kfree(sdkp);
3950 out:
3951 scsi_autopm_put_device(sdp);
3952 return error;
3953}
3954
3955/**
3956 * sd_remove - called whenever a scsi disk (previously recognized by
3957 * sd_probe) is detached from the system. It is called (potentially
3958 * multiple times) during sd module unload.
3959 * @dev: pointer to device object
3960 *
3961 * Note: this function is invoked from the scsi mid-level.
3962 * This function potentially frees up a device name (e.g. /dev/sdc)
3963 * that could be re-used by a subsequent sd_probe().
3964 * This function is not called when the built-in sd driver is "exit-ed".
3965 **/
3966static int sd_remove(struct device *dev)
3967{
3968 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3969
3970 scsi_autopm_get_device(sdkp->device);
3971
3972 device_del(&sdkp->disk_dev);
3973 del_gendisk(sdkp->disk);
3974 if (!sdkp->suspended)
3975 sd_shutdown(dev);
3976
3977 put_disk(sdkp->disk);
3978 return 0;
3979}
3980
3981static void scsi_disk_release(struct device *dev)
3982{
3983 struct scsi_disk *sdkp = to_scsi_disk(dev);
3984
3985 ida_free(&sd_index_ida, sdkp->index);
3986 sd_zbc_free_zone_info(sdkp);
3987 put_device(&sdkp->device->sdev_gendev);
3988 free_opal_dev(sdkp->opal_dev);
3989
3990 kfree(sdkp);
3991}
3992
3993static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3994{
3995 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3996 struct scsi_sense_hdr sshdr;
3997 const struct scsi_exec_args exec_args = {
3998 .sshdr = &sshdr,
3999 .req_flags = BLK_MQ_REQ_PM,
4000 };
4001 struct scsi_device *sdp = sdkp->device;
4002 int res;
4003
4004 if (start)
4005 cmd[4] |= 1; /* START */
4006
4007 if (sdp->start_stop_pwr_cond)
4008 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
4009
4010 if (!scsi_device_online(sdp))
4011 return -ENODEV;
4012
4013 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4014 sdkp->max_retries, &exec_args);
4015 if (res) {
4016 sd_print_result(sdkp, "Start/Stop Unit failed", res);
4017 if (res > 0 && scsi_sense_valid(&sshdr)) {
4018 sd_print_sense_hdr(sdkp, &sshdr);
4019 /* 0x3a is medium not present */
4020 if (sshdr.asc == 0x3a)
4021 res = 0;
4022 }
4023 }
4024
4025 /* SCSI error codes must not go to the generic layer */
4026 if (res)
4027 return -EIO;
4028
4029 return 0;
4030}
4031
4032/*
4033 * Send a SYNCHRONIZE CACHE instruction down to the device through
4034 * the normal SCSI command structure. Wait for the command to
4035 * complete.
4036 */
4037static void sd_shutdown(struct device *dev)
4038{
4039 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4040
4041 if (!sdkp)
4042 return; /* this can happen */
4043
4044 if (pm_runtime_suspended(dev))
4045 return;
4046
4047 if (sdkp->WCE && sdkp->media_present) {
4048 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4049 sd_sync_cache(sdkp);
4050 }
4051
4052 if ((system_state != SYSTEM_RESTART &&
4053 sdkp->device->manage_system_start_stop) ||
4054 (system_state == SYSTEM_POWER_OFF &&
4055 sdkp->device->manage_shutdown)) {
4056 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4057 sd_start_stop_device(sdkp, 0);
4058 }
4059}
4060
4061static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4062{
4063 return (sdev->manage_system_start_stop && !runtime) ||
4064 (sdev->manage_runtime_start_stop && runtime);
4065}
4066
4067static int sd_suspend_common(struct device *dev, bool runtime)
4068{
4069 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4070 int ret = 0;
4071
4072 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
4073 return 0;
4074
4075 if (sdkp->WCE && sdkp->media_present) {
4076 if (!sdkp->device->silence_suspend)
4077 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4078 ret = sd_sync_cache(sdkp);
4079 /* ignore OFFLINE device */
4080 if (ret == -ENODEV)
4081 return 0;
4082
4083 if (ret)
4084 return ret;
4085 }
4086
4087 if (sd_do_start_stop(sdkp->device, runtime)) {
4088 if (!sdkp->device->silence_suspend)
4089 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4090 /* an error is not worth aborting a system sleep */
4091 ret = sd_start_stop_device(sdkp, 0);
4092 if (!runtime)
4093 ret = 0;
4094 }
4095
4096 if (!ret)
4097 sdkp->suspended = true;
4098
4099 return ret;
4100}
4101
4102static int sd_suspend_system(struct device *dev)
4103{
4104 if (pm_runtime_suspended(dev))
4105 return 0;
4106
4107 return sd_suspend_common(dev, false);
4108}
4109
4110static int sd_suspend_runtime(struct device *dev)
4111{
4112 return sd_suspend_common(dev, true);
4113}
4114
4115static int sd_resume(struct device *dev)
4116{
4117 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4118
4119 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4120
4121 if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4122 sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4123 return -EIO;
4124 }
4125
4126 return 0;
4127}
4128
4129static int sd_resume_common(struct device *dev, bool runtime)
4130{
4131 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4132 int ret;
4133
4134 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4135 return 0;
4136
4137 if (!sd_do_start_stop(sdkp->device, runtime)) {
4138 sdkp->suspended = false;
4139 return 0;
4140 }
4141
4142 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4143 ret = sd_start_stop_device(sdkp, 1);
4144 if (!ret) {
4145 sd_resume(dev);
4146 sdkp->suspended = false;
4147 }
4148
4149 return ret;
4150}
4151
4152static int sd_resume_system(struct device *dev)
4153{
4154 if (pm_runtime_suspended(dev)) {
4155 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4156 struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4157
4158 if (sdp && sdp->force_runtime_start_on_system_start)
4159 pm_request_resume(dev);
4160
4161 return 0;
4162 }
4163
4164 return sd_resume_common(dev, false);
4165}
4166
4167static int sd_resume_runtime(struct device *dev)
4168{
4169 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4170 struct scsi_device *sdp;
4171
4172 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4173 return 0;
4174
4175 sdp = sdkp->device;
4176
4177 if (sdp->ignore_media_change) {
4178 /* clear the device's sense data */
4179 static const u8 cmd[10] = { REQUEST_SENSE };
4180 const struct scsi_exec_args exec_args = {
4181 .req_flags = BLK_MQ_REQ_PM,
4182 };
4183
4184 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4185 sdp->request_queue->rq_timeout, 1,
4186 &exec_args))
4187 sd_printk(KERN_NOTICE, sdkp,
4188 "Failed to clear sense data\n");
4189 }
4190
4191 return sd_resume_common(dev, true);
4192}
4193
4194static const struct dev_pm_ops sd_pm_ops = {
4195 .suspend = sd_suspend_system,
4196 .resume = sd_resume_system,
4197 .poweroff = sd_suspend_system,
4198 .restore = sd_resume_system,
4199 .runtime_suspend = sd_suspend_runtime,
4200 .runtime_resume = sd_resume_runtime,
4201};
4202
4203static struct scsi_driver sd_template = {
4204 .gendrv = {
4205 .name = "sd",
4206 .owner = THIS_MODULE,
4207 .probe = sd_probe,
4208 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
4209 .remove = sd_remove,
4210 .shutdown = sd_shutdown,
4211 .pm = &sd_pm_ops,
4212 },
4213 .rescan = sd_rescan,
4214 .resume = sd_resume,
4215 .init_command = sd_init_command,
4216 .uninit_command = sd_uninit_command,
4217 .done = sd_done,
4218 .eh_action = sd_eh_action,
4219 .eh_reset = sd_eh_reset,
4220};
4221
4222/**
4223 * init_sd - entry point for this driver (both when built in or when
4224 * a module).
4225 *
4226 * Note: this function registers this driver with the scsi mid-level.
4227 **/
4228static int __init init_sd(void)
4229{
4230 int majors = 0, i, err;
4231
4232 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4233
4234 for (i = 0; i < SD_MAJORS; i++) {
4235 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4236 continue;
4237 majors++;
4238 }
4239
4240 if (!majors)
4241 return -ENODEV;
4242
4243 err = class_register(&sd_disk_class);
4244 if (err)
4245 goto err_out;
4246
4247 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4248 if (!sd_page_pool) {
4249 printk(KERN_ERR "sd: can't init discard page pool\n");
4250 err = -ENOMEM;
4251 goto err_out_class;
4252 }
4253
4254 err = scsi_register_driver(&sd_template.gendrv);
4255 if (err)
4256 goto err_out_driver;
4257
4258 return 0;
4259
4260err_out_driver:
4261 mempool_destroy(sd_page_pool);
4262err_out_class:
4263 class_unregister(&sd_disk_class);
4264err_out:
4265 for (i = 0; i < SD_MAJORS; i++)
4266 unregister_blkdev(sd_major(i), "sd");
4267 return err;
4268}
4269
4270/**
4271 * exit_sd - exit point for this driver (when it is a module).
4272 *
4273 * Note: this function unregisters this driver from the scsi mid-level.
4274 **/
4275static void __exit exit_sd(void)
4276{
4277 int i;
4278
4279 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4280
4281 scsi_unregister_driver(&sd_template.gendrv);
4282 mempool_destroy(sd_page_pool);
4283
4284 class_unregister(&sd_disk_class);
4285
4286 for (i = 0; i < SD_MAJORS; i++)
4287 unregister_blkdev(sd_major(i), "sd");
4288}
4289
4290module_init(init_sd);
4291module_exit(exit_sd);
4292
4293void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4294{
4295 scsi_print_sense_hdr(sdkp->device,
4296 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4297}
4298
4299void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4300{
4301 const char *hb_string = scsi_hostbyte_string(result);
4302
4303 if (hb_string)
4304 sd_printk(KERN_INFO, sdkp,
4305 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4306 hb_string ? hb_string : "invalid",
4307 "DRIVER_OK");
4308 else
4309 sd_printk(KERN_INFO, sdkp,
4310 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4311 msg, host_byte(result), "DRIVER_OK");
4312}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 *
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
26 *
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
34 */
35
36#include <linux/module.h>
37#include <linux/fs.h>
38#include <linux/kernel.h>
39#include <linux/mm.h>
40#include <linux/bio.h>
41#include <linux/genhd.h>
42#include <linux/hdreg.h>
43#include <linux/errno.h>
44#include <linux/idr.h>
45#include <linux/interrupt.h>
46#include <linux/init.h>
47#include <linux/blkdev.h>
48#include <linux/blkpg.h>
49#include <linux/blk-pm.h>
50#include <linux/delay.h>
51#include <linux/mutex.h>
52#include <linux/string_helpers.h>
53#include <linux/async.h>
54#include <linux/slab.h>
55#include <linux/sed-opal.h>
56#include <linux/pm_runtime.h>
57#include <linux/pr.h>
58#include <linux/t10-pi.h>
59#include <linux/uaccess.h>
60#include <asm/unaligned.h>
61
62#include <scsi/scsi.h>
63#include <scsi/scsi_cmnd.h>
64#include <scsi/scsi_dbg.h>
65#include <scsi/scsi_device.h>
66#include <scsi/scsi_driver.h>
67#include <scsi/scsi_eh.h>
68#include <scsi/scsi_host.h>
69#include <scsi/scsi_ioctl.h>
70#include <scsi/scsicam.h>
71
72#include "sd.h"
73#include "scsi_priv.h"
74#include "scsi_logging.h"
75
76MODULE_AUTHOR("Eric Youngdale");
77MODULE_DESCRIPTION("SCSI disk (sd) driver");
78MODULE_LICENSE("GPL");
79
80MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100
101#if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
102#define SD_MINORS 16
103#else
104#define SD_MINORS 0
105#endif
106
107static void sd_config_discard(struct scsi_disk *, unsigned int);
108static void sd_config_write_same(struct scsi_disk *);
109static int sd_revalidate_disk(struct gendisk *);
110static void sd_unlock_native_capacity(struct gendisk *disk);
111static int sd_probe(struct device *);
112static int sd_remove(struct device *);
113static void sd_shutdown(struct device *);
114static int sd_suspend_system(struct device *);
115static int sd_suspend_runtime(struct device *);
116static int sd_resume(struct device *);
117static void sd_rescan(struct device *);
118static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120static int sd_done(struct scsi_cmnd *);
121static void sd_eh_reset(struct scsi_cmnd *);
122static int sd_eh_action(struct scsi_cmnd *, int);
123static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124static void scsi_disk_release(struct device *cdev);
125static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
126static void sd_print_result(const struct scsi_disk *, const char *, int);
127
128static DEFINE_IDA(sd_index_ida);
129
130/* This semaphore is used to mediate the 0->1 reference get in the
131 * face of object destruction (i.e. we can't allow a get on an
132 * object after last put) */
133static DEFINE_MUTEX(sd_ref_mutex);
134
135static struct kmem_cache *sd_cdb_cache;
136static mempool_t *sd_cdb_pool;
137static mempool_t *sd_page_pool;
138
139static const char *sd_cache_types[] = {
140 "write through", "none", "write back",
141 "write back, no read (daft)"
142};
143
144static void sd_set_flush_flag(struct scsi_disk *sdkp)
145{
146 bool wc = false, fua = false;
147
148 if (sdkp->WCE) {
149 wc = true;
150 if (sdkp->DPOFUA)
151 fua = true;
152 }
153
154 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
155}
156
157static ssize_t
158cache_type_store(struct device *dev, struct device_attribute *attr,
159 const char *buf, size_t count)
160{
161 int ct, rcd, wce, sp;
162 struct scsi_disk *sdkp = to_scsi_disk(dev);
163 struct scsi_device *sdp = sdkp->device;
164 char buffer[64];
165 char *buffer_data;
166 struct scsi_mode_data data;
167 struct scsi_sense_hdr sshdr;
168 static const char temp[] = "temporary ";
169 int len;
170
171 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
172 /* no cache control on RBC devices; theoretically they
173 * can do it, but there's probably so many exceptions
174 * it's not worth the risk */
175 return -EINVAL;
176
177 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
178 buf += sizeof(temp) - 1;
179 sdkp->cache_override = 1;
180 } else {
181 sdkp->cache_override = 0;
182 }
183
184 ct = sysfs_match_string(sd_cache_types, buf);
185 if (ct < 0)
186 return -EINVAL;
187
188 rcd = ct & 0x01 ? 1 : 0;
189 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190
191 if (sdkp->cache_override) {
192 sdkp->WCE = wce;
193 sdkp->RCD = rcd;
194 sd_set_flush_flag(sdkp);
195 return count;
196 }
197
198 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 SD_MAX_RETRIES, &data, NULL))
200 return -EINVAL;
201 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 data.block_descriptor_length);
203 buffer_data = buffer + data.header_length +
204 data.block_descriptor_length;
205 buffer_data[2] &= ~0x05;
206 buffer_data[2] |= wce << 2 | rcd;
207 sp = buffer_data[0] & 0x80 ? 1 : 0;
208 buffer_data[0] &= ~0x80;
209
210 /*
211 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
212 * received mode parameter buffer before doing MODE SELECT.
213 */
214 data.device_specific = 0;
215
216 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
217 SD_MAX_RETRIES, &data, &sshdr)) {
218 if (scsi_sense_valid(&sshdr))
219 sd_print_sense_hdr(sdkp, &sshdr);
220 return -EINVAL;
221 }
222 revalidate_disk(sdkp->disk);
223 return count;
224}
225
226static ssize_t
227manage_start_stop_show(struct device *dev, struct device_attribute *attr,
228 char *buf)
229{
230 struct scsi_disk *sdkp = to_scsi_disk(dev);
231 struct scsi_device *sdp = sdkp->device;
232
233 return sprintf(buf, "%u\n", sdp->manage_start_stop);
234}
235
236static ssize_t
237manage_start_stop_store(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t count)
239{
240 struct scsi_disk *sdkp = to_scsi_disk(dev);
241 struct scsi_device *sdp = sdkp->device;
242 bool v;
243
244 if (!capable(CAP_SYS_ADMIN))
245 return -EACCES;
246
247 if (kstrtobool(buf, &v))
248 return -EINVAL;
249
250 sdp->manage_start_stop = v;
251
252 return count;
253}
254static DEVICE_ATTR_RW(manage_start_stop);
255
256static ssize_t
257allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
258{
259 struct scsi_disk *sdkp = to_scsi_disk(dev);
260
261 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
262}
263
264static ssize_t
265allow_restart_store(struct device *dev, struct device_attribute *attr,
266 const char *buf, size_t count)
267{
268 bool v;
269 struct scsi_disk *sdkp = to_scsi_disk(dev);
270 struct scsi_device *sdp = sdkp->device;
271
272 if (!capable(CAP_SYS_ADMIN))
273 return -EACCES;
274
275 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
276 return -EINVAL;
277
278 if (kstrtobool(buf, &v))
279 return -EINVAL;
280
281 sdp->allow_restart = v;
282
283 return count;
284}
285static DEVICE_ATTR_RW(allow_restart);
286
287static ssize_t
288cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
289{
290 struct scsi_disk *sdkp = to_scsi_disk(dev);
291 int ct = sdkp->RCD + 2*sdkp->WCE;
292
293 return sprintf(buf, "%s\n", sd_cache_types[ct]);
294}
295static DEVICE_ATTR_RW(cache_type);
296
297static ssize_t
298FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
299{
300 struct scsi_disk *sdkp = to_scsi_disk(dev);
301
302 return sprintf(buf, "%u\n", sdkp->DPOFUA);
303}
304static DEVICE_ATTR_RO(FUA);
305
306static ssize_t
307protection_type_show(struct device *dev, struct device_attribute *attr,
308 char *buf)
309{
310 struct scsi_disk *sdkp = to_scsi_disk(dev);
311
312 return sprintf(buf, "%u\n", sdkp->protection_type);
313}
314
315static ssize_t
316protection_type_store(struct device *dev, struct device_attribute *attr,
317 const char *buf, size_t count)
318{
319 struct scsi_disk *sdkp = to_scsi_disk(dev);
320 unsigned int val;
321 int err;
322
323 if (!capable(CAP_SYS_ADMIN))
324 return -EACCES;
325
326 err = kstrtouint(buf, 10, &val);
327
328 if (err)
329 return err;
330
331 if (val <= T10_PI_TYPE3_PROTECTION)
332 sdkp->protection_type = val;
333
334 return count;
335}
336static DEVICE_ATTR_RW(protection_type);
337
338static ssize_t
339protection_mode_show(struct device *dev, struct device_attribute *attr,
340 char *buf)
341{
342 struct scsi_disk *sdkp = to_scsi_disk(dev);
343 struct scsi_device *sdp = sdkp->device;
344 unsigned int dif, dix;
345
346 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
347 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
348
349 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
350 dif = 0;
351 dix = 1;
352 }
353
354 if (!dif && !dix)
355 return sprintf(buf, "none\n");
356
357 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
358}
359static DEVICE_ATTR_RO(protection_mode);
360
361static ssize_t
362app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
363{
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
365
366 return sprintf(buf, "%u\n", sdkp->ATO);
367}
368static DEVICE_ATTR_RO(app_tag_own);
369
370static ssize_t
371thin_provisioning_show(struct device *dev, struct device_attribute *attr,
372 char *buf)
373{
374 struct scsi_disk *sdkp = to_scsi_disk(dev);
375
376 return sprintf(buf, "%u\n", sdkp->lbpme);
377}
378static DEVICE_ATTR_RO(thin_provisioning);
379
380/* sysfs_match_string() requires dense arrays */
381static const char *lbp_mode[] = {
382 [SD_LBP_FULL] = "full",
383 [SD_LBP_UNMAP] = "unmap",
384 [SD_LBP_WS16] = "writesame_16",
385 [SD_LBP_WS10] = "writesame_10",
386 [SD_LBP_ZERO] = "writesame_zero",
387 [SD_LBP_DISABLE] = "disabled",
388};
389
390static ssize_t
391provisioning_mode_show(struct device *dev, struct device_attribute *attr,
392 char *buf)
393{
394 struct scsi_disk *sdkp = to_scsi_disk(dev);
395
396 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
397}
398
399static ssize_t
400provisioning_mode_store(struct device *dev, struct device_attribute *attr,
401 const char *buf, size_t count)
402{
403 struct scsi_disk *sdkp = to_scsi_disk(dev);
404 struct scsi_device *sdp = sdkp->device;
405 int mode;
406
407 if (!capable(CAP_SYS_ADMIN))
408 return -EACCES;
409
410 if (sd_is_zoned(sdkp)) {
411 sd_config_discard(sdkp, SD_LBP_DISABLE);
412 return count;
413 }
414
415 if (sdp->type != TYPE_DISK)
416 return -EINVAL;
417
418 mode = sysfs_match_string(lbp_mode, buf);
419 if (mode < 0)
420 return -EINVAL;
421
422 sd_config_discard(sdkp, mode);
423
424 return count;
425}
426static DEVICE_ATTR_RW(provisioning_mode);
427
428/* sysfs_match_string() requires dense arrays */
429static const char *zeroing_mode[] = {
430 [SD_ZERO_WRITE] = "write",
431 [SD_ZERO_WS] = "writesame",
432 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
433 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
434};
435
436static ssize_t
437zeroing_mode_show(struct device *dev, struct device_attribute *attr,
438 char *buf)
439{
440 struct scsi_disk *sdkp = to_scsi_disk(dev);
441
442 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
443}
444
445static ssize_t
446zeroing_mode_store(struct device *dev, struct device_attribute *attr,
447 const char *buf, size_t count)
448{
449 struct scsi_disk *sdkp = to_scsi_disk(dev);
450 int mode;
451
452 if (!capable(CAP_SYS_ADMIN))
453 return -EACCES;
454
455 mode = sysfs_match_string(zeroing_mode, buf);
456 if (mode < 0)
457 return -EINVAL;
458
459 sdkp->zeroing_mode = mode;
460
461 return count;
462}
463static DEVICE_ATTR_RW(zeroing_mode);
464
465static ssize_t
466max_medium_access_timeouts_show(struct device *dev,
467 struct device_attribute *attr, char *buf)
468{
469 struct scsi_disk *sdkp = to_scsi_disk(dev);
470
471 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
472}
473
474static ssize_t
475max_medium_access_timeouts_store(struct device *dev,
476 struct device_attribute *attr, const char *buf,
477 size_t count)
478{
479 struct scsi_disk *sdkp = to_scsi_disk(dev);
480 int err;
481
482 if (!capable(CAP_SYS_ADMIN))
483 return -EACCES;
484
485 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
486
487 return err ? err : count;
488}
489static DEVICE_ATTR_RW(max_medium_access_timeouts);
490
491static ssize_t
492max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
493 char *buf)
494{
495 struct scsi_disk *sdkp = to_scsi_disk(dev);
496
497 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
498}
499
500static ssize_t
501max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
502 const char *buf, size_t count)
503{
504 struct scsi_disk *sdkp = to_scsi_disk(dev);
505 struct scsi_device *sdp = sdkp->device;
506 unsigned long max;
507 int err;
508
509 if (!capable(CAP_SYS_ADMIN))
510 return -EACCES;
511
512 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
513 return -EINVAL;
514
515 err = kstrtoul(buf, 10, &max);
516
517 if (err)
518 return err;
519
520 if (max == 0)
521 sdp->no_write_same = 1;
522 else if (max <= SD_MAX_WS16_BLOCKS) {
523 sdp->no_write_same = 0;
524 sdkp->max_ws_blocks = max;
525 }
526
527 sd_config_write_same(sdkp);
528
529 return count;
530}
531static DEVICE_ATTR_RW(max_write_same_blocks);
532
533static struct attribute *sd_disk_attrs[] = {
534 &dev_attr_cache_type.attr,
535 &dev_attr_FUA.attr,
536 &dev_attr_allow_restart.attr,
537 &dev_attr_manage_start_stop.attr,
538 &dev_attr_protection_type.attr,
539 &dev_attr_protection_mode.attr,
540 &dev_attr_app_tag_own.attr,
541 &dev_attr_thin_provisioning.attr,
542 &dev_attr_provisioning_mode.attr,
543 &dev_attr_zeroing_mode.attr,
544 &dev_attr_max_write_same_blocks.attr,
545 &dev_attr_max_medium_access_timeouts.attr,
546 NULL,
547};
548ATTRIBUTE_GROUPS(sd_disk);
549
550static struct class sd_disk_class = {
551 .name = "scsi_disk",
552 .owner = THIS_MODULE,
553 .dev_release = scsi_disk_release,
554 .dev_groups = sd_disk_groups,
555};
556
557static const struct dev_pm_ops sd_pm_ops = {
558 .suspend = sd_suspend_system,
559 .resume = sd_resume,
560 .poweroff = sd_suspend_system,
561 .restore = sd_resume,
562 .runtime_suspend = sd_suspend_runtime,
563 .runtime_resume = sd_resume,
564};
565
566static struct scsi_driver sd_template = {
567 .gendrv = {
568 .name = "sd",
569 .owner = THIS_MODULE,
570 .probe = sd_probe,
571 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
572 .remove = sd_remove,
573 .shutdown = sd_shutdown,
574 .pm = &sd_pm_ops,
575 },
576 .rescan = sd_rescan,
577 .init_command = sd_init_command,
578 .uninit_command = sd_uninit_command,
579 .done = sd_done,
580 .eh_action = sd_eh_action,
581 .eh_reset = sd_eh_reset,
582};
583
584/*
585 * Dummy kobj_map->probe function.
586 * The default ->probe function will call modprobe, which is
587 * pointless as this module is already loaded.
588 */
589static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
590{
591 return NULL;
592}
593
594/*
595 * Device no to disk mapping:
596 *
597 * major disc2 disc p1
598 * |............|.............|....|....| <- dev_t
599 * 31 20 19 8 7 4 3 0
600 *
601 * Inside a major, we have 16k disks, however mapped non-
602 * contiguously. The first 16 disks are for major0, the next
603 * ones with major1, ... Disk 256 is for major0 again, disk 272
604 * for major1, ...
605 * As we stay compatible with our numbering scheme, we can reuse
606 * the well-know SCSI majors 8, 65--71, 136--143.
607 */
608static int sd_major(int major_idx)
609{
610 switch (major_idx) {
611 case 0:
612 return SCSI_DISK0_MAJOR;
613 case 1 ... 7:
614 return SCSI_DISK1_MAJOR + major_idx - 1;
615 case 8 ... 15:
616 return SCSI_DISK8_MAJOR + major_idx - 8;
617 default:
618 BUG();
619 return 0; /* shut up gcc */
620 }
621}
622
623static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
624{
625 struct scsi_disk *sdkp = NULL;
626
627 mutex_lock(&sd_ref_mutex);
628
629 if (disk->private_data) {
630 sdkp = scsi_disk(disk);
631 if (scsi_device_get(sdkp->device) == 0)
632 get_device(&sdkp->dev);
633 else
634 sdkp = NULL;
635 }
636 mutex_unlock(&sd_ref_mutex);
637 return sdkp;
638}
639
640static void scsi_disk_put(struct scsi_disk *sdkp)
641{
642 struct scsi_device *sdev = sdkp->device;
643
644 mutex_lock(&sd_ref_mutex);
645 put_device(&sdkp->dev);
646 scsi_device_put(sdev);
647 mutex_unlock(&sd_ref_mutex);
648}
649
650#ifdef CONFIG_BLK_SED_OPAL
651static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
652 size_t len, bool send)
653{
654 struct scsi_device *sdev = data;
655 u8 cdb[12] = { 0, };
656 int ret;
657
658 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
659 cdb[1] = secp;
660 put_unaligned_be16(spsp, &cdb[2]);
661 put_unaligned_be32(len, &cdb[6]);
662
663 ret = scsi_execute_req(sdev, cdb,
664 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
665 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
666 return ret <= 0 ? ret : -EIO;
667}
668#endif /* CONFIG_BLK_SED_OPAL */
669
670/*
671 * Look up the DIX operation based on whether the command is read or
672 * write and whether dix and dif are enabled.
673 */
674static unsigned int sd_prot_op(bool write, bool dix, bool dif)
675{
676 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
677 static const unsigned int ops[] = { /* wrt dix dif */
678 SCSI_PROT_NORMAL, /* 0 0 0 */
679 SCSI_PROT_READ_STRIP, /* 0 0 1 */
680 SCSI_PROT_READ_INSERT, /* 0 1 0 */
681 SCSI_PROT_READ_PASS, /* 0 1 1 */
682 SCSI_PROT_NORMAL, /* 1 0 0 */
683 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
684 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
685 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
686 };
687
688 return ops[write << 2 | dix << 1 | dif];
689}
690
691/*
692 * Returns a mask of the protection flags that are valid for a given DIX
693 * operation.
694 */
695static unsigned int sd_prot_flag_mask(unsigned int prot_op)
696{
697 static const unsigned int flag_mask[] = {
698 [SCSI_PROT_NORMAL] = 0,
699
700 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
701 SCSI_PROT_GUARD_CHECK |
702 SCSI_PROT_REF_CHECK |
703 SCSI_PROT_REF_INCREMENT,
704
705 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
706 SCSI_PROT_IP_CHECKSUM,
707
708 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
709 SCSI_PROT_GUARD_CHECK |
710 SCSI_PROT_REF_CHECK |
711 SCSI_PROT_REF_INCREMENT |
712 SCSI_PROT_IP_CHECKSUM,
713
714 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
715 SCSI_PROT_REF_INCREMENT,
716
717 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
718 SCSI_PROT_REF_CHECK |
719 SCSI_PROT_REF_INCREMENT |
720 SCSI_PROT_IP_CHECKSUM,
721
722 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
723 SCSI_PROT_GUARD_CHECK |
724 SCSI_PROT_REF_CHECK |
725 SCSI_PROT_REF_INCREMENT |
726 SCSI_PROT_IP_CHECKSUM,
727 };
728
729 return flag_mask[prot_op];
730}
731
732static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
733 unsigned int dix, unsigned int dif)
734{
735 struct bio *bio = scmd->request->bio;
736 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
737 unsigned int protect = 0;
738
739 if (dix) { /* DIX Type 0, 1, 2, 3 */
740 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
741 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
742
743 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
744 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
745 }
746
747 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
748 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
749
750 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
751 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
752 }
753
754 if (dif) { /* DIX/DIF Type 1, 2, 3 */
755 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
756
757 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
758 protect = 3 << 5; /* Disable target PI checking */
759 else
760 protect = 1 << 5; /* Enable target PI checking */
761 }
762
763 scsi_set_prot_op(scmd, prot_op);
764 scsi_set_prot_type(scmd, dif);
765 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
766
767 return protect;
768}
769
770static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
771{
772 struct request_queue *q = sdkp->disk->queue;
773 unsigned int logical_block_size = sdkp->device->sector_size;
774 unsigned int max_blocks = 0;
775
776 q->limits.discard_alignment =
777 sdkp->unmap_alignment * logical_block_size;
778 q->limits.discard_granularity =
779 max(sdkp->physical_block_size,
780 sdkp->unmap_granularity * logical_block_size);
781 sdkp->provisioning_mode = mode;
782
783 switch (mode) {
784
785 case SD_LBP_FULL:
786 case SD_LBP_DISABLE:
787 blk_queue_max_discard_sectors(q, 0);
788 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
789 return;
790
791 case SD_LBP_UNMAP:
792 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
793 (u32)SD_MAX_WS16_BLOCKS);
794 break;
795
796 case SD_LBP_WS16:
797 if (sdkp->device->unmap_limit_for_ws)
798 max_blocks = sdkp->max_unmap_blocks;
799 else
800 max_blocks = sdkp->max_ws_blocks;
801
802 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
803 break;
804
805 case SD_LBP_WS10:
806 if (sdkp->device->unmap_limit_for_ws)
807 max_blocks = sdkp->max_unmap_blocks;
808 else
809 max_blocks = sdkp->max_ws_blocks;
810
811 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
812 break;
813
814 case SD_LBP_ZERO:
815 max_blocks = min_not_zero(sdkp->max_ws_blocks,
816 (u32)SD_MAX_WS10_BLOCKS);
817 break;
818 }
819
820 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
821 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
822}
823
824static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
825{
826 struct scsi_device *sdp = cmd->device;
827 struct request *rq = cmd->request;
828 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
829 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
830 unsigned int data_len = 24;
831 char *buf;
832
833 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
834 if (!rq->special_vec.bv_page)
835 return BLK_STS_RESOURCE;
836 clear_highpage(rq->special_vec.bv_page);
837 rq->special_vec.bv_offset = 0;
838 rq->special_vec.bv_len = data_len;
839 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
840
841 cmd->cmd_len = 10;
842 cmd->cmnd[0] = UNMAP;
843 cmd->cmnd[8] = 24;
844
845 buf = page_address(rq->special_vec.bv_page);
846 put_unaligned_be16(6 + 16, &buf[0]);
847 put_unaligned_be16(16, &buf[2]);
848 put_unaligned_be64(lba, &buf[8]);
849 put_unaligned_be32(nr_blocks, &buf[16]);
850
851 cmd->allowed = SD_MAX_RETRIES;
852 cmd->transfersize = data_len;
853 rq->timeout = SD_TIMEOUT;
854
855 return scsi_init_io(cmd);
856}
857
858static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
859 bool unmap)
860{
861 struct scsi_device *sdp = cmd->device;
862 struct request *rq = cmd->request;
863 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
864 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
865 u32 data_len = sdp->sector_size;
866
867 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
868 if (!rq->special_vec.bv_page)
869 return BLK_STS_RESOURCE;
870 clear_highpage(rq->special_vec.bv_page);
871 rq->special_vec.bv_offset = 0;
872 rq->special_vec.bv_len = data_len;
873 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
874
875 cmd->cmd_len = 16;
876 cmd->cmnd[0] = WRITE_SAME_16;
877 if (unmap)
878 cmd->cmnd[1] = 0x8; /* UNMAP */
879 put_unaligned_be64(lba, &cmd->cmnd[2]);
880 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
881
882 cmd->allowed = SD_MAX_RETRIES;
883 cmd->transfersize = data_len;
884 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
885
886 return scsi_init_io(cmd);
887}
888
889static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
890 bool unmap)
891{
892 struct scsi_device *sdp = cmd->device;
893 struct request *rq = cmd->request;
894 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
895 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
896 u32 data_len = sdp->sector_size;
897
898 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
899 if (!rq->special_vec.bv_page)
900 return BLK_STS_RESOURCE;
901 clear_highpage(rq->special_vec.bv_page);
902 rq->special_vec.bv_offset = 0;
903 rq->special_vec.bv_len = data_len;
904 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905
906 cmd->cmd_len = 10;
907 cmd->cmnd[0] = WRITE_SAME;
908 if (unmap)
909 cmd->cmnd[1] = 0x8; /* UNMAP */
910 put_unaligned_be32(lba, &cmd->cmnd[2]);
911 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
912
913 cmd->allowed = SD_MAX_RETRIES;
914 cmd->transfersize = data_len;
915 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
916
917 return scsi_init_io(cmd);
918}
919
920static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
921{
922 struct request *rq = cmd->request;
923 struct scsi_device *sdp = cmd->device;
924 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
925 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
926 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
927
928 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
929 switch (sdkp->zeroing_mode) {
930 case SD_ZERO_WS16_UNMAP:
931 return sd_setup_write_same16_cmnd(cmd, true);
932 case SD_ZERO_WS10_UNMAP:
933 return sd_setup_write_same10_cmnd(cmd, true);
934 }
935 }
936
937 if (sdp->no_write_same)
938 return BLK_STS_TARGET;
939
940 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
941 return sd_setup_write_same16_cmnd(cmd, false);
942
943 return sd_setup_write_same10_cmnd(cmd, false);
944}
945
946static void sd_config_write_same(struct scsi_disk *sdkp)
947{
948 struct request_queue *q = sdkp->disk->queue;
949 unsigned int logical_block_size = sdkp->device->sector_size;
950
951 if (sdkp->device->no_write_same) {
952 sdkp->max_ws_blocks = 0;
953 goto out;
954 }
955
956 /* Some devices can not handle block counts above 0xffff despite
957 * supporting WRITE SAME(16). Consequently we default to 64k
958 * blocks per I/O unless the device explicitly advertises a
959 * bigger limit.
960 */
961 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
962 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
963 (u32)SD_MAX_WS16_BLOCKS);
964 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
965 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
966 (u32)SD_MAX_WS10_BLOCKS);
967 else {
968 sdkp->device->no_write_same = 1;
969 sdkp->max_ws_blocks = 0;
970 }
971
972 if (sdkp->lbprz && sdkp->lbpws)
973 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
974 else if (sdkp->lbprz && sdkp->lbpws10)
975 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
976 else if (sdkp->max_ws_blocks)
977 sdkp->zeroing_mode = SD_ZERO_WS;
978 else
979 sdkp->zeroing_mode = SD_ZERO_WRITE;
980
981 if (sdkp->max_ws_blocks &&
982 sdkp->physical_block_size > logical_block_size) {
983 /*
984 * Reporting a maximum number of blocks that is not aligned
985 * on the device physical size would cause a large write same
986 * request to be split into physically unaligned chunks by
987 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
988 * even if the caller of these functions took care to align the
989 * large request. So make sure the maximum reported is aligned
990 * to the device physical block size. This is only an optional
991 * optimization for regular disks, but this is mandatory to
992 * avoid failure of large write same requests directed at
993 * sequential write required zones of host-managed ZBC disks.
994 */
995 sdkp->max_ws_blocks =
996 round_down(sdkp->max_ws_blocks,
997 bytes_to_logical(sdkp->device,
998 sdkp->physical_block_size));
999 }
1000
1001out:
1002 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1003 (logical_block_size >> 9));
1004 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1005 (logical_block_size >> 9));
1006}
1007
1008/**
1009 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1010 * @cmd: command to prepare
1011 *
1012 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1013 * the preference indicated by the target device.
1014 **/
1015static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1016{
1017 struct request *rq = cmd->request;
1018 struct scsi_device *sdp = cmd->device;
1019 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1020 struct bio *bio = rq->bio;
1021 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1022 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023 blk_status_t ret;
1024
1025 if (sdkp->device->no_write_same)
1026 return BLK_STS_TARGET;
1027
1028 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1029
1030 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1031
1032 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1033 cmd->cmd_len = 16;
1034 cmd->cmnd[0] = WRITE_SAME_16;
1035 put_unaligned_be64(lba, &cmd->cmnd[2]);
1036 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037 } else {
1038 cmd->cmd_len = 10;
1039 cmd->cmnd[0] = WRITE_SAME;
1040 put_unaligned_be32(lba, &cmd->cmnd[2]);
1041 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042 }
1043
1044 cmd->transfersize = sdp->sector_size;
1045 cmd->allowed = SD_MAX_RETRIES;
1046
1047 /*
1048 * For WRITE SAME the data transferred via the DATA OUT buffer is
1049 * different from the amount of data actually written to the target.
1050 *
1051 * We set up __data_len to the amount of data transferred via the
1052 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1053 * to transfer a single sector of data first, but then reset it to
1054 * the amount of data to be written right after so that the I/O path
1055 * knows how much to actually write.
1056 */
1057 rq->__data_len = sdp->sector_size;
1058 ret = scsi_init_io(cmd);
1059 rq->__data_len = blk_rq_bytes(rq);
1060
1061 return ret;
1062}
1063
1064static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1065{
1066 struct request *rq = cmd->request;
1067
1068 /* flush requests don't perform I/O, zero the S/G table */
1069 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070
1071 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1072 cmd->cmd_len = 10;
1073 cmd->transfersize = 0;
1074 cmd->allowed = SD_MAX_RETRIES;
1075
1076 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1077 return BLK_STS_OK;
1078}
1079
1080static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1081 sector_t lba, unsigned int nr_blocks,
1082 unsigned char flags)
1083{
1084 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1085 if (unlikely(cmd->cmnd == NULL))
1086 return BLK_STS_RESOURCE;
1087
1088 cmd->cmd_len = SD_EXT_CDB_SIZE;
1089 memset(cmd->cmnd, 0, cmd->cmd_len);
1090
1091 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1092 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1093 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1094 cmd->cmnd[10] = flags;
1095 put_unaligned_be64(lba, &cmd->cmnd[12]);
1096 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1097 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1098
1099 return BLK_STS_OK;
1100}
1101
1102static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1103 sector_t lba, unsigned int nr_blocks,
1104 unsigned char flags)
1105{
1106 cmd->cmd_len = 16;
1107 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1108 cmd->cmnd[1] = flags;
1109 cmd->cmnd[14] = 0;
1110 cmd->cmnd[15] = 0;
1111 put_unaligned_be64(lba, &cmd->cmnd[2]);
1112 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1113
1114 return BLK_STS_OK;
1115}
1116
1117static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1118 sector_t lba, unsigned int nr_blocks,
1119 unsigned char flags)
1120{
1121 cmd->cmd_len = 10;
1122 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1123 cmd->cmnd[1] = flags;
1124 cmd->cmnd[6] = 0;
1125 cmd->cmnd[9] = 0;
1126 put_unaligned_be32(lba, &cmd->cmnd[2]);
1127 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1128
1129 return BLK_STS_OK;
1130}
1131
1132static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1133 sector_t lba, unsigned int nr_blocks,
1134 unsigned char flags)
1135{
1136 /* Avoid that 0 blocks gets translated into 256 blocks. */
1137 if (WARN_ON_ONCE(nr_blocks == 0))
1138 return BLK_STS_IOERR;
1139
1140 if (unlikely(flags & 0x8)) {
1141 /*
1142 * This happens only if this drive failed 10byte rw
1143 * command with ILLEGAL_REQUEST during operation and
1144 * thus turned off use_10_for_rw.
1145 */
1146 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1147 return BLK_STS_IOERR;
1148 }
1149
1150 cmd->cmd_len = 6;
1151 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1152 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1153 cmd->cmnd[2] = (lba >> 8) & 0xff;
1154 cmd->cmnd[3] = lba & 0xff;
1155 cmd->cmnd[4] = nr_blocks;
1156 cmd->cmnd[5] = 0;
1157
1158 return BLK_STS_OK;
1159}
1160
1161static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1162{
1163 struct request *rq = cmd->request;
1164 struct scsi_device *sdp = cmd->device;
1165 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1166 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1167 sector_t threshold;
1168 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1169 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170 bool write = rq_data_dir(rq) == WRITE;
1171 unsigned char protect, fua;
1172 blk_status_t ret;
1173 unsigned int dif;
1174 bool dix;
1175
1176 ret = scsi_init_io(cmd);
1177 if (ret != BLK_STS_OK)
1178 return ret;
1179
1180 if (!scsi_device_online(sdp) || sdp->changed) {
1181 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1182 return BLK_STS_IOERR;
1183 }
1184
1185 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1186 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1187 return BLK_STS_IOERR;
1188 }
1189
1190 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1191 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1192 return BLK_STS_IOERR;
1193 }
1194
1195 /*
1196 * Some SD card readers can't handle accesses which touch the
1197 * last one or two logical blocks. Split accesses as needed.
1198 */
1199 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1200
1201 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1202 if (lba < threshold) {
1203 /* Access up to the threshold but not beyond */
1204 nr_blocks = threshold - lba;
1205 } else {
1206 /* Access only a single logical block */
1207 nr_blocks = 1;
1208 }
1209 }
1210
1211 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1212 dix = scsi_prot_sg_count(cmd);
1213 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1214
1215 if (dif || dix)
1216 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1217 else
1218 protect = 0;
1219
1220 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1221 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1222 protect | fua);
1223 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1224 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1225 protect | fua);
1226 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1227 sdp->use_10_for_rw || protect) {
1228 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1229 protect | fua);
1230 } else {
1231 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1232 protect | fua);
1233 }
1234
1235 if (unlikely(ret != BLK_STS_OK))
1236 return ret;
1237
1238 /*
1239 * We shouldn't disconnect in the middle of a sector, so with a dumb
1240 * host adapter, it's safe to assume that we can at least transfer
1241 * this many bytes between each connect / disconnect.
1242 */
1243 cmd->transfersize = sdp->sector_size;
1244 cmd->underflow = nr_blocks << 9;
1245 cmd->allowed = SD_MAX_RETRIES;
1246 cmd->sdb.length = nr_blocks * sdp->sector_size;
1247
1248 SCSI_LOG_HLQUEUE(1,
1249 scmd_printk(KERN_INFO, cmd,
1250 "%s: block=%llu, count=%d\n", __func__,
1251 (unsigned long long)blk_rq_pos(rq),
1252 blk_rq_sectors(rq)));
1253 SCSI_LOG_HLQUEUE(2,
1254 scmd_printk(KERN_INFO, cmd,
1255 "%s %d/%u 512 byte blocks.\n",
1256 write ? "writing" : "reading", nr_blocks,
1257 blk_rq_sectors(rq)));
1258
1259 /*
1260 * This indicates that the command is ready from our end to be
1261 * queued.
1262 */
1263 return BLK_STS_OK;
1264}
1265
1266static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1267{
1268 struct request *rq = cmd->request;
1269
1270 switch (req_op(rq)) {
1271 case REQ_OP_DISCARD:
1272 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1273 case SD_LBP_UNMAP:
1274 return sd_setup_unmap_cmnd(cmd);
1275 case SD_LBP_WS16:
1276 return sd_setup_write_same16_cmnd(cmd, true);
1277 case SD_LBP_WS10:
1278 return sd_setup_write_same10_cmnd(cmd, true);
1279 case SD_LBP_ZERO:
1280 return sd_setup_write_same10_cmnd(cmd, false);
1281 default:
1282 return BLK_STS_TARGET;
1283 }
1284 case REQ_OP_WRITE_ZEROES:
1285 return sd_setup_write_zeroes_cmnd(cmd);
1286 case REQ_OP_WRITE_SAME:
1287 return sd_setup_write_same_cmnd(cmd);
1288 case REQ_OP_FLUSH:
1289 return sd_setup_flush_cmnd(cmd);
1290 case REQ_OP_READ:
1291 case REQ_OP_WRITE:
1292 return sd_setup_read_write_cmnd(cmd);
1293 case REQ_OP_ZONE_RESET:
1294 return sd_zbc_setup_reset_cmnd(cmd, false);
1295 case REQ_OP_ZONE_RESET_ALL:
1296 return sd_zbc_setup_reset_cmnd(cmd, true);
1297 default:
1298 WARN_ON_ONCE(1);
1299 return BLK_STS_NOTSUPP;
1300 }
1301}
1302
1303static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1304{
1305 struct request *rq = SCpnt->request;
1306 u8 *cmnd;
1307
1308 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1309 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1310
1311 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1312 cmnd = SCpnt->cmnd;
1313 SCpnt->cmnd = NULL;
1314 SCpnt->cmd_len = 0;
1315 mempool_free(cmnd, sd_cdb_pool);
1316 }
1317}
1318
1319/**
1320 * sd_open - open a scsi disk device
1321 * @bdev: Block device of the scsi disk to open
1322 * @mode: FMODE_* mask
1323 *
1324 * Returns 0 if successful. Returns a negated errno value in case
1325 * of error.
1326 *
1327 * Note: This can be called from a user context (e.g. fsck(1) )
1328 * or from within the kernel (e.g. as a result of a mount(1) ).
1329 * In the latter case @inode and @filp carry an abridged amount
1330 * of information as noted above.
1331 *
1332 * Locking: called with bdev->bd_mutex held.
1333 **/
1334static int sd_open(struct block_device *bdev, fmode_t mode)
1335{
1336 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1337 struct scsi_device *sdev;
1338 int retval;
1339
1340 if (!sdkp)
1341 return -ENXIO;
1342
1343 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1344
1345 sdev = sdkp->device;
1346
1347 /*
1348 * If the device is in error recovery, wait until it is done.
1349 * If the device is offline, then disallow any access to it.
1350 */
1351 retval = -ENXIO;
1352 if (!scsi_block_when_processing_errors(sdev))
1353 goto error_out;
1354
1355 if (sdev->removable || sdkp->write_prot)
1356 check_disk_change(bdev);
1357
1358 /*
1359 * If the drive is empty, just let the open fail.
1360 */
1361 retval = -ENOMEDIUM;
1362 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1363 goto error_out;
1364
1365 /*
1366 * If the device has the write protect tab set, have the open fail
1367 * if the user expects to be able to write to the thing.
1368 */
1369 retval = -EROFS;
1370 if (sdkp->write_prot && (mode & FMODE_WRITE))
1371 goto error_out;
1372
1373 /*
1374 * It is possible that the disk changing stuff resulted in
1375 * the device being taken offline. If this is the case,
1376 * report this to the user, and don't pretend that the
1377 * open actually succeeded.
1378 */
1379 retval = -ENXIO;
1380 if (!scsi_device_online(sdev))
1381 goto error_out;
1382
1383 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1384 if (scsi_block_when_processing_errors(sdev))
1385 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1386 }
1387
1388 return 0;
1389
1390error_out:
1391 scsi_disk_put(sdkp);
1392 return retval;
1393}
1394
1395/**
1396 * sd_release - invoked when the (last) close(2) is called on this
1397 * scsi disk.
1398 * @disk: disk to release
1399 * @mode: FMODE_* mask
1400 *
1401 * Returns 0.
1402 *
1403 * Note: may block (uninterruptible) if error recovery is underway
1404 * on this disk.
1405 *
1406 * Locking: called with bdev->bd_mutex held.
1407 **/
1408static void sd_release(struct gendisk *disk, fmode_t mode)
1409{
1410 struct scsi_disk *sdkp = scsi_disk(disk);
1411 struct scsi_device *sdev = sdkp->device;
1412
1413 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1414
1415 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1416 if (scsi_block_when_processing_errors(sdev))
1417 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1418 }
1419
1420 scsi_disk_put(sdkp);
1421}
1422
1423static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1424{
1425 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1426 struct scsi_device *sdp = sdkp->device;
1427 struct Scsi_Host *host = sdp->host;
1428 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1429 int diskinfo[4];
1430
1431 /* default to most commonly used values */
1432 diskinfo[0] = 0x40; /* 1 << 6 */
1433 diskinfo[1] = 0x20; /* 1 << 5 */
1434 diskinfo[2] = capacity >> 11;
1435
1436 /* override with calculated, extended default, or driver values */
1437 if (host->hostt->bios_param)
1438 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1439 else
1440 scsicam_bios_param(bdev, capacity, diskinfo);
1441
1442 geo->heads = diskinfo[0];
1443 geo->sectors = diskinfo[1];
1444 geo->cylinders = diskinfo[2];
1445 return 0;
1446}
1447
1448/**
1449 * sd_ioctl - process an ioctl
1450 * @bdev: target block device
1451 * @mode: FMODE_* mask
1452 * @cmd: ioctl command number
1453 * @arg: this is third argument given to ioctl(2) system call.
1454 * Often contains a pointer.
1455 *
1456 * Returns 0 if successful (some ioctls return positive numbers on
1457 * success as well). Returns a negated errno value in case of error.
1458 *
1459 * Note: most ioctls are forward onto the block subsystem or further
1460 * down in the scsi subsystem.
1461 **/
1462static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1463 unsigned int cmd, unsigned long arg)
1464{
1465 struct gendisk *disk = bdev->bd_disk;
1466 struct scsi_disk *sdkp = scsi_disk(disk);
1467 struct scsi_device *sdp = sdkp->device;
1468 void __user *p = (void __user *)arg;
1469 int error;
1470
1471 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1472 "cmd=0x%x\n", disk->disk_name, cmd));
1473
1474 error = scsi_verify_blk_ioctl(bdev, cmd);
1475 if (error < 0)
1476 return error;
1477
1478 /*
1479 * If we are in the middle of error recovery, don't let anyone
1480 * else try and use this device. Also, if error recovery fails, it
1481 * may try and take the device offline, in which case all further
1482 * access to the device is prohibited.
1483 */
1484 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1485 (mode & FMODE_NDELAY) != 0);
1486 if (error)
1487 goto out;
1488
1489 if (is_sed_ioctl(cmd))
1490 return sed_ioctl(sdkp->opal_dev, cmd, p);
1491
1492 /*
1493 * Send SCSI addressing ioctls directly to mid level, send other
1494 * ioctls to block level and then onto mid level if they can't be
1495 * resolved.
1496 */
1497 switch (cmd) {
1498 case SCSI_IOCTL_GET_IDLUN:
1499 case SCSI_IOCTL_GET_BUS_NUMBER:
1500 error = scsi_ioctl(sdp, cmd, p);
1501 break;
1502 default:
1503 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1504 if (error != -ENOTTY)
1505 break;
1506 error = scsi_ioctl(sdp, cmd, p);
1507 break;
1508 }
1509out:
1510 return error;
1511}
1512
1513static void set_media_not_present(struct scsi_disk *sdkp)
1514{
1515 if (sdkp->media_present)
1516 sdkp->device->changed = 1;
1517
1518 if (sdkp->device->removable) {
1519 sdkp->media_present = 0;
1520 sdkp->capacity = 0;
1521 }
1522}
1523
1524static int media_not_present(struct scsi_disk *sdkp,
1525 struct scsi_sense_hdr *sshdr)
1526{
1527 if (!scsi_sense_valid(sshdr))
1528 return 0;
1529
1530 /* not invoked for commands that could return deferred errors */
1531 switch (sshdr->sense_key) {
1532 case UNIT_ATTENTION:
1533 case NOT_READY:
1534 /* medium not present */
1535 if (sshdr->asc == 0x3A) {
1536 set_media_not_present(sdkp);
1537 return 1;
1538 }
1539 }
1540 return 0;
1541}
1542
1543/**
1544 * sd_check_events - check media events
1545 * @disk: kernel device descriptor
1546 * @clearing: disk events currently being cleared
1547 *
1548 * Returns mask of DISK_EVENT_*.
1549 *
1550 * Note: this function is invoked from the block subsystem.
1551 **/
1552static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1553{
1554 struct scsi_disk *sdkp = scsi_disk_get(disk);
1555 struct scsi_device *sdp;
1556 int retval;
1557
1558 if (!sdkp)
1559 return 0;
1560
1561 sdp = sdkp->device;
1562 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1563
1564 /*
1565 * If the device is offline, don't send any commands - just pretend as
1566 * if the command failed. If the device ever comes back online, we
1567 * can deal with it then. It is only because of unrecoverable errors
1568 * that we would ever take a device offline in the first place.
1569 */
1570 if (!scsi_device_online(sdp)) {
1571 set_media_not_present(sdkp);
1572 goto out;
1573 }
1574
1575 /*
1576 * Using TEST_UNIT_READY enables differentiation between drive with
1577 * no cartridge loaded - NOT READY, drive with changed cartridge -
1578 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1579 *
1580 * Drives that auto spin down. eg iomega jaz 1G, will be started
1581 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1582 * sd_revalidate() is called.
1583 */
1584 if (scsi_block_when_processing_errors(sdp)) {
1585 struct scsi_sense_hdr sshdr = { 0, };
1586
1587 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1588 &sshdr);
1589
1590 /* failed to execute TUR, assume media not present */
1591 if (host_byte(retval)) {
1592 set_media_not_present(sdkp);
1593 goto out;
1594 }
1595
1596 if (media_not_present(sdkp, &sshdr))
1597 goto out;
1598 }
1599
1600 /*
1601 * For removable scsi disk we have to recognise the presence
1602 * of a disk in the drive.
1603 */
1604 if (!sdkp->media_present)
1605 sdp->changed = 1;
1606 sdkp->media_present = 1;
1607out:
1608 /*
1609 * sdp->changed is set under the following conditions:
1610 *
1611 * Medium present state has changed in either direction.
1612 * Device has indicated UNIT_ATTENTION.
1613 */
1614 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1615 sdp->changed = 0;
1616 scsi_disk_put(sdkp);
1617 return retval;
1618}
1619
1620static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1621{
1622 int retries, res;
1623 struct scsi_device *sdp = sdkp->device;
1624 const int timeout = sdp->request_queue->rq_timeout
1625 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1626 struct scsi_sense_hdr my_sshdr;
1627
1628 if (!scsi_device_online(sdp))
1629 return -ENODEV;
1630
1631 /* caller might not be interested in sense, but we need it */
1632 if (!sshdr)
1633 sshdr = &my_sshdr;
1634
1635 for (retries = 3; retries > 0; --retries) {
1636 unsigned char cmd[10] = { 0 };
1637
1638 cmd[0] = SYNCHRONIZE_CACHE;
1639 /*
1640 * Leave the rest of the command zero to indicate
1641 * flush everything.
1642 */
1643 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1644 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1645 if (res == 0)
1646 break;
1647 }
1648
1649 if (res) {
1650 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1651
1652 if (driver_byte(res) == DRIVER_SENSE)
1653 sd_print_sense_hdr(sdkp, sshdr);
1654
1655 /* we need to evaluate the error return */
1656 if (scsi_sense_valid(sshdr) &&
1657 (sshdr->asc == 0x3a || /* medium not present */
1658 sshdr->asc == 0x20 || /* invalid command */
1659 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1660 /* this is no error here */
1661 return 0;
1662
1663 switch (host_byte(res)) {
1664 /* ignore errors due to racing a disconnection */
1665 case DID_BAD_TARGET:
1666 case DID_NO_CONNECT:
1667 return 0;
1668 /* signal the upper layer it might try again */
1669 case DID_BUS_BUSY:
1670 case DID_IMM_RETRY:
1671 case DID_REQUEUE:
1672 case DID_SOFT_ERROR:
1673 return -EBUSY;
1674 default:
1675 return -EIO;
1676 }
1677 }
1678 return 0;
1679}
1680
1681static void sd_rescan(struct device *dev)
1682{
1683 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1684
1685 revalidate_disk(sdkp->disk);
1686}
1687
1688
1689#ifdef CONFIG_COMPAT
1690/*
1691 * This gets directly called from VFS. When the ioctl
1692 * is not recognized we go back to the other translation paths.
1693 */
1694static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1695 unsigned int cmd, unsigned long arg)
1696{
1697 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1698 int error;
1699
1700 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1701 (mode & FMODE_NDELAY) != 0);
1702 if (error)
1703 return error;
1704
1705 /*
1706 * Let the static ioctl translation table take care of it.
1707 */
1708 if (!sdev->host->hostt->compat_ioctl)
1709 return -ENOIOCTLCMD;
1710 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1711}
1712#endif
1713
1714static char sd_pr_type(enum pr_type type)
1715{
1716 switch (type) {
1717 case PR_WRITE_EXCLUSIVE:
1718 return 0x01;
1719 case PR_EXCLUSIVE_ACCESS:
1720 return 0x03;
1721 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1722 return 0x05;
1723 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1724 return 0x06;
1725 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1726 return 0x07;
1727 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1728 return 0x08;
1729 default:
1730 return 0;
1731 }
1732};
1733
1734static int sd_pr_command(struct block_device *bdev, u8 sa,
1735 u64 key, u64 sa_key, u8 type, u8 flags)
1736{
1737 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1738 struct scsi_sense_hdr sshdr;
1739 int result;
1740 u8 cmd[16] = { 0, };
1741 u8 data[24] = { 0, };
1742
1743 cmd[0] = PERSISTENT_RESERVE_OUT;
1744 cmd[1] = sa;
1745 cmd[2] = type;
1746 put_unaligned_be32(sizeof(data), &cmd[5]);
1747
1748 put_unaligned_be64(key, &data[0]);
1749 put_unaligned_be64(sa_key, &data[8]);
1750 data[20] = flags;
1751
1752 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1753 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1754
1755 if (driver_byte(result) == DRIVER_SENSE &&
1756 scsi_sense_valid(&sshdr)) {
1757 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1758 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1759 }
1760
1761 return result;
1762}
1763
1764static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1765 u32 flags)
1766{
1767 if (flags & ~PR_FL_IGNORE_KEY)
1768 return -EOPNOTSUPP;
1769 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1770 old_key, new_key, 0,
1771 (1 << 0) /* APTPL */);
1772}
1773
1774static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1775 u32 flags)
1776{
1777 if (flags)
1778 return -EOPNOTSUPP;
1779 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1780}
1781
1782static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1783{
1784 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1785}
1786
1787static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1788 enum pr_type type, bool abort)
1789{
1790 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1791 sd_pr_type(type), 0);
1792}
1793
1794static int sd_pr_clear(struct block_device *bdev, u64 key)
1795{
1796 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1797}
1798
1799static const struct pr_ops sd_pr_ops = {
1800 .pr_register = sd_pr_register,
1801 .pr_reserve = sd_pr_reserve,
1802 .pr_release = sd_pr_release,
1803 .pr_preempt = sd_pr_preempt,
1804 .pr_clear = sd_pr_clear,
1805};
1806
1807static const struct block_device_operations sd_fops = {
1808 .owner = THIS_MODULE,
1809 .open = sd_open,
1810 .release = sd_release,
1811 .ioctl = sd_ioctl,
1812 .getgeo = sd_getgeo,
1813#ifdef CONFIG_COMPAT
1814 .compat_ioctl = sd_compat_ioctl,
1815#endif
1816 .check_events = sd_check_events,
1817 .revalidate_disk = sd_revalidate_disk,
1818 .unlock_native_capacity = sd_unlock_native_capacity,
1819 .report_zones = sd_zbc_report_zones,
1820 .pr_ops = &sd_pr_ops,
1821};
1822
1823/**
1824 * sd_eh_reset - reset error handling callback
1825 * @scmd: sd-issued command that has failed
1826 *
1827 * This function is called by the SCSI midlayer before starting
1828 * SCSI EH. When counting medium access failures we have to be
1829 * careful to register it only only once per device and SCSI EH run;
1830 * there might be several timed out commands which will cause the
1831 * 'max_medium_access_timeouts' counter to trigger after the first
1832 * SCSI EH run already and set the device to offline.
1833 * So this function resets the internal counter before starting SCSI EH.
1834 **/
1835static void sd_eh_reset(struct scsi_cmnd *scmd)
1836{
1837 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1838
1839 /* New SCSI EH run, reset gate variable */
1840 sdkp->ignore_medium_access_errors = false;
1841}
1842
1843/**
1844 * sd_eh_action - error handling callback
1845 * @scmd: sd-issued command that has failed
1846 * @eh_disp: The recovery disposition suggested by the midlayer
1847 *
1848 * This function is called by the SCSI midlayer upon completion of an
1849 * error test command (currently TEST UNIT READY). The result of sending
1850 * the eh command is passed in eh_disp. We're looking for devices that
1851 * fail medium access commands but are OK with non access commands like
1852 * test unit ready (so wrongly see the device as having a successful
1853 * recovery)
1854 **/
1855static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1856{
1857 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1858 struct scsi_device *sdev = scmd->device;
1859
1860 if (!scsi_device_online(sdev) ||
1861 !scsi_medium_access_command(scmd) ||
1862 host_byte(scmd->result) != DID_TIME_OUT ||
1863 eh_disp != SUCCESS)
1864 return eh_disp;
1865
1866 /*
1867 * The device has timed out executing a medium access command.
1868 * However, the TEST UNIT READY command sent during error
1869 * handling completed successfully. Either the device is in the
1870 * process of recovering or has it suffered an internal failure
1871 * that prevents access to the storage medium.
1872 */
1873 if (!sdkp->ignore_medium_access_errors) {
1874 sdkp->medium_access_timed_out++;
1875 sdkp->ignore_medium_access_errors = true;
1876 }
1877
1878 /*
1879 * If the device keeps failing read/write commands but TEST UNIT
1880 * READY always completes successfully we assume that medium
1881 * access is no longer possible and take the device offline.
1882 */
1883 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1884 scmd_printk(KERN_ERR, scmd,
1885 "Medium access timeout failure. Offlining disk!\n");
1886 mutex_lock(&sdev->state_mutex);
1887 scsi_device_set_state(sdev, SDEV_OFFLINE);
1888 mutex_unlock(&sdev->state_mutex);
1889
1890 return SUCCESS;
1891 }
1892
1893 return eh_disp;
1894}
1895
1896static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1897{
1898 struct request *req = scmd->request;
1899 struct scsi_device *sdev = scmd->device;
1900 unsigned int transferred, good_bytes;
1901 u64 start_lba, end_lba, bad_lba;
1902
1903 /*
1904 * Some commands have a payload smaller than the device logical
1905 * block size (e.g. INQUIRY on a 4K disk).
1906 */
1907 if (scsi_bufflen(scmd) <= sdev->sector_size)
1908 return 0;
1909
1910 /* Check if we have a 'bad_lba' information */
1911 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1912 SCSI_SENSE_BUFFERSIZE,
1913 &bad_lba))
1914 return 0;
1915
1916 /*
1917 * If the bad lba was reported incorrectly, we have no idea where
1918 * the error is.
1919 */
1920 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1921 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1922 if (bad_lba < start_lba || bad_lba >= end_lba)
1923 return 0;
1924
1925 /*
1926 * resid is optional but mostly filled in. When it's unused,
1927 * its value is zero, so we assume the whole buffer transferred
1928 */
1929 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1930
1931 /* This computation should always be done in terms of the
1932 * resolution of the device's medium.
1933 */
1934 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1935
1936 return min(good_bytes, transferred);
1937}
1938
1939/**
1940 * sd_done - bottom half handler: called when the lower level
1941 * driver has completed (successfully or otherwise) a scsi command.
1942 * @SCpnt: mid-level's per command structure.
1943 *
1944 * Note: potentially run from within an ISR. Must not block.
1945 **/
1946static int sd_done(struct scsi_cmnd *SCpnt)
1947{
1948 int result = SCpnt->result;
1949 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1950 unsigned int sector_size = SCpnt->device->sector_size;
1951 unsigned int resid;
1952 struct scsi_sense_hdr sshdr;
1953 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1954 struct request *req = SCpnt->request;
1955 int sense_valid = 0;
1956 int sense_deferred = 0;
1957
1958 switch (req_op(req)) {
1959 case REQ_OP_DISCARD:
1960 case REQ_OP_WRITE_ZEROES:
1961 case REQ_OP_WRITE_SAME:
1962 case REQ_OP_ZONE_RESET:
1963 case REQ_OP_ZONE_RESET_ALL:
1964 if (!result) {
1965 good_bytes = blk_rq_bytes(req);
1966 scsi_set_resid(SCpnt, 0);
1967 } else {
1968 good_bytes = 0;
1969 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1970 }
1971 break;
1972 default:
1973 /*
1974 * In case of bogus fw or device, we could end up having
1975 * an unaligned partial completion. Check this here and force
1976 * alignment.
1977 */
1978 resid = scsi_get_resid(SCpnt);
1979 if (resid & (sector_size - 1)) {
1980 sd_printk(KERN_INFO, sdkp,
1981 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1982 resid, sector_size);
1983 scsi_print_command(SCpnt);
1984 resid = min(scsi_bufflen(SCpnt),
1985 round_up(resid, sector_size));
1986 scsi_set_resid(SCpnt, resid);
1987 }
1988 }
1989
1990 if (result) {
1991 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1992 if (sense_valid)
1993 sense_deferred = scsi_sense_is_deferred(&sshdr);
1994 }
1995 sdkp->medium_access_timed_out = 0;
1996
1997 if (driver_byte(result) != DRIVER_SENSE &&
1998 (!sense_valid || sense_deferred))
1999 goto out;
2000
2001 switch (sshdr.sense_key) {
2002 case HARDWARE_ERROR:
2003 case MEDIUM_ERROR:
2004 good_bytes = sd_completed_bytes(SCpnt);
2005 break;
2006 case RECOVERED_ERROR:
2007 good_bytes = scsi_bufflen(SCpnt);
2008 break;
2009 case NO_SENSE:
2010 /* This indicates a false check condition, so ignore it. An
2011 * unknown amount of data was transferred so treat it as an
2012 * error.
2013 */
2014 SCpnt->result = 0;
2015 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2016 break;
2017 case ABORTED_COMMAND:
2018 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2019 good_bytes = sd_completed_bytes(SCpnt);
2020 break;
2021 case ILLEGAL_REQUEST:
2022 switch (sshdr.asc) {
2023 case 0x10: /* DIX: Host detected corruption */
2024 good_bytes = sd_completed_bytes(SCpnt);
2025 break;
2026 case 0x20: /* INVALID COMMAND OPCODE */
2027 case 0x24: /* INVALID FIELD IN CDB */
2028 switch (SCpnt->cmnd[0]) {
2029 case UNMAP:
2030 sd_config_discard(sdkp, SD_LBP_DISABLE);
2031 break;
2032 case WRITE_SAME_16:
2033 case WRITE_SAME:
2034 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2035 sd_config_discard(sdkp, SD_LBP_DISABLE);
2036 } else {
2037 sdkp->device->no_write_same = 1;
2038 sd_config_write_same(sdkp);
2039 req->rq_flags |= RQF_QUIET;
2040 }
2041 break;
2042 }
2043 }
2044 break;
2045 default:
2046 break;
2047 }
2048
2049 out:
2050 if (sd_is_zoned(sdkp))
2051 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2052
2053 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2054 "sd_done: completed %d of %d bytes\n",
2055 good_bytes, scsi_bufflen(SCpnt)));
2056
2057 return good_bytes;
2058}
2059
2060/*
2061 * spinup disk - called only in sd_revalidate_disk()
2062 */
2063static void
2064sd_spinup_disk(struct scsi_disk *sdkp)
2065{
2066 unsigned char cmd[10];
2067 unsigned long spintime_expire = 0;
2068 int retries, spintime;
2069 unsigned int the_result;
2070 struct scsi_sense_hdr sshdr;
2071 int sense_valid = 0;
2072
2073 spintime = 0;
2074
2075 /* Spin up drives, as required. Only do this at boot time */
2076 /* Spinup needs to be done for module loads too. */
2077 do {
2078 retries = 0;
2079
2080 do {
2081 cmd[0] = TEST_UNIT_READY;
2082 memset((void *) &cmd[1], 0, 9);
2083
2084 the_result = scsi_execute_req(sdkp->device, cmd,
2085 DMA_NONE, NULL, 0,
2086 &sshdr, SD_TIMEOUT,
2087 SD_MAX_RETRIES, NULL);
2088
2089 /*
2090 * If the drive has indicated to us that it
2091 * doesn't have any media in it, don't bother
2092 * with any more polling.
2093 */
2094 if (media_not_present(sdkp, &sshdr))
2095 return;
2096
2097 if (the_result)
2098 sense_valid = scsi_sense_valid(&sshdr);
2099 retries++;
2100 } while (retries < 3 &&
2101 (!scsi_status_is_good(the_result) ||
2102 ((driver_byte(the_result) == DRIVER_SENSE) &&
2103 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2104
2105 if (driver_byte(the_result) != DRIVER_SENSE) {
2106 /* no sense, TUR either succeeded or failed
2107 * with a status error */
2108 if(!spintime && !scsi_status_is_good(the_result)) {
2109 sd_print_result(sdkp, "Test Unit Ready failed",
2110 the_result);
2111 }
2112 break;
2113 }
2114
2115 /*
2116 * The device does not want the automatic start to be issued.
2117 */
2118 if (sdkp->device->no_start_on_add)
2119 break;
2120
2121 if (sense_valid && sshdr.sense_key == NOT_READY) {
2122 if (sshdr.asc == 4 && sshdr.ascq == 3)
2123 break; /* manual intervention required */
2124 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2125 break; /* standby */
2126 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2127 break; /* unavailable */
2128 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2129 break; /* sanitize in progress */
2130 /*
2131 * Issue command to spin up drive when not ready
2132 */
2133 if (!spintime) {
2134 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2135 cmd[0] = START_STOP;
2136 cmd[1] = 1; /* Return immediately */
2137 memset((void *) &cmd[2], 0, 8);
2138 cmd[4] = 1; /* Start spin cycle */
2139 if (sdkp->device->start_stop_pwr_cond)
2140 cmd[4] |= 1 << 4;
2141 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2142 NULL, 0, &sshdr,
2143 SD_TIMEOUT, SD_MAX_RETRIES,
2144 NULL);
2145 spintime_expire = jiffies + 100 * HZ;
2146 spintime = 1;
2147 }
2148 /* Wait 1 second for next try */
2149 msleep(1000);
2150 printk(KERN_CONT ".");
2151
2152 /*
2153 * Wait for USB flash devices with slow firmware.
2154 * Yes, this sense key/ASC combination shouldn't
2155 * occur here. It's characteristic of these devices.
2156 */
2157 } else if (sense_valid &&
2158 sshdr.sense_key == UNIT_ATTENTION &&
2159 sshdr.asc == 0x28) {
2160 if (!spintime) {
2161 spintime_expire = jiffies + 5 * HZ;
2162 spintime = 1;
2163 }
2164 /* Wait 1 second for next try */
2165 msleep(1000);
2166 } else {
2167 /* we don't understand the sense code, so it's
2168 * probably pointless to loop */
2169 if(!spintime) {
2170 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2171 sd_print_sense_hdr(sdkp, &sshdr);
2172 }
2173 break;
2174 }
2175
2176 } while (spintime && time_before_eq(jiffies, spintime_expire));
2177
2178 if (spintime) {
2179 if (scsi_status_is_good(the_result))
2180 printk(KERN_CONT "ready\n");
2181 else
2182 printk(KERN_CONT "not responding...\n");
2183 }
2184}
2185
2186/*
2187 * Determine whether disk supports Data Integrity Field.
2188 */
2189static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2190{
2191 struct scsi_device *sdp = sdkp->device;
2192 u8 type;
2193 int ret = 0;
2194
2195 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2196 return ret;
2197
2198 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2199
2200 if (type > T10_PI_TYPE3_PROTECTION)
2201 ret = -ENODEV;
2202 else if (scsi_host_dif_capable(sdp->host, type))
2203 ret = 1;
2204
2205 if (sdkp->first_scan || type != sdkp->protection_type)
2206 switch (ret) {
2207 case -ENODEV:
2208 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2209 " protection type %u. Disabling disk!\n",
2210 type);
2211 break;
2212 case 1:
2213 sd_printk(KERN_NOTICE, sdkp,
2214 "Enabling DIF Type %u protection\n", type);
2215 break;
2216 case 0:
2217 sd_printk(KERN_NOTICE, sdkp,
2218 "Disabling DIF Type %u protection\n", type);
2219 break;
2220 }
2221
2222 sdkp->protection_type = type;
2223
2224 return ret;
2225}
2226
2227static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2228 struct scsi_sense_hdr *sshdr, int sense_valid,
2229 int the_result)
2230{
2231 if (driver_byte(the_result) == DRIVER_SENSE)
2232 sd_print_sense_hdr(sdkp, sshdr);
2233 else
2234 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2235
2236 /*
2237 * Set dirty bit for removable devices if not ready -
2238 * sometimes drives will not report this properly.
2239 */
2240 if (sdp->removable &&
2241 sense_valid && sshdr->sense_key == NOT_READY)
2242 set_media_not_present(sdkp);
2243
2244 /*
2245 * We used to set media_present to 0 here to indicate no media
2246 * in the drive, but some drives fail read capacity even with
2247 * media present, so we can't do that.
2248 */
2249 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2250}
2251
2252#define RC16_LEN 32
2253#if RC16_LEN > SD_BUF_SIZE
2254#error RC16_LEN must not be more than SD_BUF_SIZE
2255#endif
2256
2257#define READ_CAPACITY_RETRIES_ON_RESET 10
2258
2259static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2260 unsigned char *buffer)
2261{
2262 unsigned char cmd[16];
2263 struct scsi_sense_hdr sshdr;
2264 int sense_valid = 0;
2265 int the_result;
2266 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2267 unsigned int alignment;
2268 unsigned long long lba;
2269 unsigned sector_size;
2270
2271 if (sdp->no_read_capacity_16)
2272 return -EINVAL;
2273
2274 do {
2275 memset(cmd, 0, 16);
2276 cmd[0] = SERVICE_ACTION_IN_16;
2277 cmd[1] = SAI_READ_CAPACITY_16;
2278 cmd[13] = RC16_LEN;
2279 memset(buffer, 0, RC16_LEN);
2280
2281 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2282 buffer, RC16_LEN, &sshdr,
2283 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2284
2285 if (media_not_present(sdkp, &sshdr))
2286 return -ENODEV;
2287
2288 if (the_result) {
2289 sense_valid = scsi_sense_valid(&sshdr);
2290 if (sense_valid &&
2291 sshdr.sense_key == ILLEGAL_REQUEST &&
2292 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2293 sshdr.ascq == 0x00)
2294 /* Invalid Command Operation Code or
2295 * Invalid Field in CDB, just retry
2296 * silently with RC10 */
2297 return -EINVAL;
2298 if (sense_valid &&
2299 sshdr.sense_key == UNIT_ATTENTION &&
2300 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2301 /* Device reset might occur several times,
2302 * give it one more chance */
2303 if (--reset_retries > 0)
2304 continue;
2305 }
2306 retries--;
2307
2308 } while (the_result && retries);
2309
2310 if (the_result) {
2311 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2312 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2313 return -EINVAL;
2314 }
2315
2316 sector_size = get_unaligned_be32(&buffer[8]);
2317 lba = get_unaligned_be64(&buffer[0]);
2318
2319 if (sd_read_protection_type(sdkp, buffer) < 0) {
2320 sdkp->capacity = 0;
2321 return -ENODEV;
2322 }
2323
2324 /* Logical blocks per physical block exponent */
2325 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2326
2327 /* RC basis */
2328 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2329
2330 /* Lowest aligned logical block */
2331 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2332 blk_queue_alignment_offset(sdp->request_queue, alignment);
2333 if (alignment && sdkp->first_scan)
2334 sd_printk(KERN_NOTICE, sdkp,
2335 "physical block alignment offset: %u\n", alignment);
2336
2337 if (buffer[14] & 0x80) { /* LBPME */
2338 sdkp->lbpme = 1;
2339
2340 if (buffer[14] & 0x40) /* LBPRZ */
2341 sdkp->lbprz = 1;
2342
2343 sd_config_discard(sdkp, SD_LBP_WS16);
2344 }
2345
2346 sdkp->capacity = lba + 1;
2347 return sector_size;
2348}
2349
2350static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2351 unsigned char *buffer)
2352{
2353 unsigned char cmd[16];
2354 struct scsi_sense_hdr sshdr;
2355 int sense_valid = 0;
2356 int the_result;
2357 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2358 sector_t lba;
2359 unsigned sector_size;
2360
2361 do {
2362 cmd[0] = READ_CAPACITY;
2363 memset(&cmd[1], 0, 9);
2364 memset(buffer, 0, 8);
2365
2366 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2367 buffer, 8, &sshdr,
2368 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2369
2370 if (media_not_present(sdkp, &sshdr))
2371 return -ENODEV;
2372
2373 if (the_result) {
2374 sense_valid = scsi_sense_valid(&sshdr);
2375 if (sense_valid &&
2376 sshdr.sense_key == UNIT_ATTENTION &&
2377 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2378 /* Device reset might occur several times,
2379 * give it one more chance */
2380 if (--reset_retries > 0)
2381 continue;
2382 }
2383 retries--;
2384
2385 } while (the_result && retries);
2386
2387 if (the_result) {
2388 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2389 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2390 return -EINVAL;
2391 }
2392
2393 sector_size = get_unaligned_be32(&buffer[4]);
2394 lba = get_unaligned_be32(&buffer[0]);
2395
2396 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2397 /* Some buggy (usb cardreader) devices return an lba of
2398 0xffffffff when the want to report a size of 0 (with
2399 which they really mean no media is present) */
2400 sdkp->capacity = 0;
2401 sdkp->physical_block_size = sector_size;
2402 return sector_size;
2403 }
2404
2405 sdkp->capacity = lba + 1;
2406 sdkp->physical_block_size = sector_size;
2407 return sector_size;
2408}
2409
2410static int sd_try_rc16_first(struct scsi_device *sdp)
2411{
2412 if (sdp->host->max_cmd_len < 16)
2413 return 0;
2414 if (sdp->try_rc_10_first)
2415 return 0;
2416 if (sdp->scsi_level > SCSI_SPC_2)
2417 return 1;
2418 if (scsi_device_protection(sdp))
2419 return 1;
2420 return 0;
2421}
2422
2423/*
2424 * read disk capacity
2425 */
2426static void
2427sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2428{
2429 int sector_size;
2430 struct scsi_device *sdp = sdkp->device;
2431
2432 if (sd_try_rc16_first(sdp)) {
2433 sector_size = read_capacity_16(sdkp, sdp, buffer);
2434 if (sector_size == -EOVERFLOW)
2435 goto got_data;
2436 if (sector_size == -ENODEV)
2437 return;
2438 if (sector_size < 0)
2439 sector_size = read_capacity_10(sdkp, sdp, buffer);
2440 if (sector_size < 0)
2441 return;
2442 } else {
2443 sector_size = read_capacity_10(sdkp, sdp, buffer);
2444 if (sector_size == -EOVERFLOW)
2445 goto got_data;
2446 if (sector_size < 0)
2447 return;
2448 if ((sizeof(sdkp->capacity) > 4) &&
2449 (sdkp->capacity > 0xffffffffULL)) {
2450 int old_sector_size = sector_size;
2451 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2452 "Trying to use READ CAPACITY(16).\n");
2453 sector_size = read_capacity_16(sdkp, sdp, buffer);
2454 if (sector_size < 0) {
2455 sd_printk(KERN_NOTICE, sdkp,
2456 "Using 0xffffffff as device size\n");
2457 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2458 sector_size = old_sector_size;
2459 goto got_data;
2460 }
2461 /* Remember that READ CAPACITY(16) succeeded */
2462 sdp->try_rc_10_first = 0;
2463 }
2464 }
2465
2466 /* Some devices are known to return the total number of blocks,
2467 * not the highest block number. Some devices have versions
2468 * which do this and others which do not. Some devices we might
2469 * suspect of doing this but we don't know for certain.
2470 *
2471 * If we know the reported capacity is wrong, decrement it. If
2472 * we can only guess, then assume the number of blocks is even
2473 * (usually true but not always) and err on the side of lowering
2474 * the capacity.
2475 */
2476 if (sdp->fix_capacity ||
2477 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2478 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2479 "from its reported value: %llu\n",
2480 (unsigned long long) sdkp->capacity);
2481 --sdkp->capacity;
2482 }
2483
2484got_data:
2485 if (sector_size == 0) {
2486 sector_size = 512;
2487 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2488 "assuming 512.\n");
2489 }
2490
2491 if (sector_size != 512 &&
2492 sector_size != 1024 &&
2493 sector_size != 2048 &&
2494 sector_size != 4096) {
2495 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2496 sector_size);
2497 /*
2498 * The user might want to re-format the drive with
2499 * a supported sectorsize. Once this happens, it
2500 * would be relatively trivial to set the thing up.
2501 * For this reason, we leave the thing in the table.
2502 */
2503 sdkp->capacity = 0;
2504 /*
2505 * set a bogus sector size so the normal read/write
2506 * logic in the block layer will eventually refuse any
2507 * request on this device without tripping over power
2508 * of two sector size assumptions
2509 */
2510 sector_size = 512;
2511 }
2512 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2513 blk_queue_physical_block_size(sdp->request_queue,
2514 sdkp->physical_block_size);
2515 sdkp->device->sector_size = sector_size;
2516
2517 if (sdkp->capacity > 0xffffffff)
2518 sdp->use_16_for_rw = 1;
2519
2520}
2521
2522/*
2523 * Print disk capacity
2524 */
2525static void
2526sd_print_capacity(struct scsi_disk *sdkp,
2527 sector_t old_capacity)
2528{
2529 int sector_size = sdkp->device->sector_size;
2530 char cap_str_2[10], cap_str_10[10];
2531
2532 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2533 return;
2534
2535 string_get_size(sdkp->capacity, sector_size,
2536 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2537 string_get_size(sdkp->capacity, sector_size,
2538 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2539
2540 sd_printk(KERN_NOTICE, sdkp,
2541 "%llu %d-byte logical blocks: (%s/%s)\n",
2542 (unsigned long long)sdkp->capacity,
2543 sector_size, cap_str_10, cap_str_2);
2544
2545 if (sdkp->physical_block_size != sector_size)
2546 sd_printk(KERN_NOTICE, sdkp,
2547 "%u-byte physical blocks\n",
2548 sdkp->physical_block_size);
2549
2550 sd_zbc_print_zones(sdkp);
2551}
2552
2553/* called with buffer of length 512 */
2554static inline int
2555sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2556 unsigned char *buffer, int len, struct scsi_mode_data *data,
2557 struct scsi_sense_hdr *sshdr)
2558{
2559 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2560 SD_TIMEOUT, SD_MAX_RETRIES, data,
2561 sshdr);
2562}
2563
2564/*
2565 * read write protect setting, if possible - called only in sd_revalidate_disk()
2566 * called with buffer of length SD_BUF_SIZE
2567 */
2568static void
2569sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2570{
2571 int res;
2572 struct scsi_device *sdp = sdkp->device;
2573 struct scsi_mode_data data;
2574 int old_wp = sdkp->write_prot;
2575
2576 set_disk_ro(sdkp->disk, 0);
2577 if (sdp->skip_ms_page_3f) {
2578 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2579 return;
2580 }
2581
2582 if (sdp->use_192_bytes_for_3f) {
2583 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2584 } else {
2585 /*
2586 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2587 * We have to start carefully: some devices hang if we ask
2588 * for more than is available.
2589 */
2590 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2591
2592 /*
2593 * Second attempt: ask for page 0 When only page 0 is
2594 * implemented, a request for page 3F may return Sense Key
2595 * 5: Illegal Request, Sense Code 24: Invalid field in
2596 * CDB.
2597 */
2598 if (!scsi_status_is_good(res))
2599 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2600
2601 /*
2602 * Third attempt: ask 255 bytes, as we did earlier.
2603 */
2604 if (!scsi_status_is_good(res))
2605 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2606 &data, NULL);
2607 }
2608
2609 if (!scsi_status_is_good(res)) {
2610 sd_first_printk(KERN_WARNING, sdkp,
2611 "Test WP failed, assume Write Enabled\n");
2612 } else {
2613 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2614 set_disk_ro(sdkp->disk, sdkp->write_prot);
2615 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2616 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2617 sdkp->write_prot ? "on" : "off");
2618 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2619 }
2620 }
2621}
2622
2623/*
2624 * sd_read_cache_type - called only from sd_revalidate_disk()
2625 * called with buffer of length SD_BUF_SIZE
2626 */
2627static void
2628sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2629{
2630 int len = 0, res;
2631 struct scsi_device *sdp = sdkp->device;
2632
2633 int dbd;
2634 int modepage;
2635 int first_len;
2636 struct scsi_mode_data data;
2637 struct scsi_sense_hdr sshdr;
2638 int old_wce = sdkp->WCE;
2639 int old_rcd = sdkp->RCD;
2640 int old_dpofua = sdkp->DPOFUA;
2641
2642
2643 if (sdkp->cache_override)
2644 return;
2645
2646 first_len = 4;
2647 if (sdp->skip_ms_page_8) {
2648 if (sdp->type == TYPE_RBC)
2649 goto defaults;
2650 else {
2651 if (sdp->skip_ms_page_3f)
2652 goto defaults;
2653 modepage = 0x3F;
2654 if (sdp->use_192_bytes_for_3f)
2655 first_len = 192;
2656 dbd = 0;
2657 }
2658 } else if (sdp->type == TYPE_RBC) {
2659 modepage = 6;
2660 dbd = 8;
2661 } else {
2662 modepage = 8;
2663 dbd = 0;
2664 }
2665
2666 /* cautiously ask */
2667 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2668 &data, &sshdr);
2669
2670 if (!scsi_status_is_good(res))
2671 goto bad_sense;
2672
2673 if (!data.header_length) {
2674 modepage = 6;
2675 first_len = 0;
2676 sd_first_printk(KERN_ERR, sdkp,
2677 "Missing header in MODE_SENSE response\n");
2678 }
2679
2680 /* that went OK, now ask for the proper length */
2681 len = data.length;
2682
2683 /*
2684 * We're only interested in the first three bytes, actually.
2685 * But the data cache page is defined for the first 20.
2686 */
2687 if (len < 3)
2688 goto bad_sense;
2689 else if (len > SD_BUF_SIZE) {
2690 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2691 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2692 len = SD_BUF_SIZE;
2693 }
2694 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2695 len = 192;
2696
2697 /* Get the data */
2698 if (len > first_len)
2699 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2700 &data, &sshdr);
2701
2702 if (scsi_status_is_good(res)) {
2703 int offset = data.header_length + data.block_descriptor_length;
2704
2705 while (offset < len) {
2706 u8 page_code = buffer[offset] & 0x3F;
2707 u8 spf = buffer[offset] & 0x40;
2708
2709 if (page_code == 8 || page_code == 6) {
2710 /* We're interested only in the first 3 bytes.
2711 */
2712 if (len - offset <= 2) {
2713 sd_first_printk(KERN_ERR, sdkp,
2714 "Incomplete mode parameter "
2715 "data\n");
2716 goto defaults;
2717 } else {
2718 modepage = page_code;
2719 goto Page_found;
2720 }
2721 } else {
2722 /* Go to the next page */
2723 if (spf && len - offset > 3)
2724 offset += 4 + (buffer[offset+2] << 8) +
2725 buffer[offset+3];
2726 else if (!spf && len - offset > 1)
2727 offset += 2 + buffer[offset+1];
2728 else {
2729 sd_first_printk(KERN_ERR, sdkp,
2730 "Incomplete mode "
2731 "parameter data\n");
2732 goto defaults;
2733 }
2734 }
2735 }
2736
2737 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2738 goto defaults;
2739
2740 Page_found:
2741 if (modepage == 8) {
2742 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2743 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2744 } else {
2745 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2746 sdkp->RCD = 0;
2747 }
2748
2749 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2750 if (sdp->broken_fua) {
2751 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2752 sdkp->DPOFUA = 0;
2753 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2754 !sdkp->device->use_16_for_rw) {
2755 sd_first_printk(KERN_NOTICE, sdkp,
2756 "Uses READ/WRITE(6), disabling FUA\n");
2757 sdkp->DPOFUA = 0;
2758 }
2759
2760 /* No cache flush allowed for write protected devices */
2761 if (sdkp->WCE && sdkp->write_prot)
2762 sdkp->WCE = 0;
2763
2764 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2765 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2766 sd_printk(KERN_NOTICE, sdkp,
2767 "Write cache: %s, read cache: %s, %s\n",
2768 sdkp->WCE ? "enabled" : "disabled",
2769 sdkp->RCD ? "disabled" : "enabled",
2770 sdkp->DPOFUA ? "supports DPO and FUA"
2771 : "doesn't support DPO or FUA");
2772
2773 return;
2774 }
2775
2776bad_sense:
2777 if (scsi_sense_valid(&sshdr) &&
2778 sshdr.sense_key == ILLEGAL_REQUEST &&
2779 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2780 /* Invalid field in CDB */
2781 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2782 else
2783 sd_first_printk(KERN_ERR, sdkp,
2784 "Asking for cache data failed\n");
2785
2786defaults:
2787 if (sdp->wce_default_on) {
2788 sd_first_printk(KERN_NOTICE, sdkp,
2789 "Assuming drive cache: write back\n");
2790 sdkp->WCE = 1;
2791 } else {
2792 sd_first_printk(KERN_ERR, sdkp,
2793 "Assuming drive cache: write through\n");
2794 sdkp->WCE = 0;
2795 }
2796 sdkp->RCD = 0;
2797 sdkp->DPOFUA = 0;
2798}
2799
2800/*
2801 * The ATO bit indicates whether the DIF application tag is available
2802 * for use by the operating system.
2803 */
2804static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2805{
2806 int res, offset;
2807 struct scsi_device *sdp = sdkp->device;
2808 struct scsi_mode_data data;
2809 struct scsi_sense_hdr sshdr;
2810
2811 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2812 return;
2813
2814 if (sdkp->protection_type == 0)
2815 return;
2816
2817 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2818 SD_MAX_RETRIES, &data, &sshdr);
2819
2820 if (!scsi_status_is_good(res) || !data.header_length ||
2821 data.length < 6) {
2822 sd_first_printk(KERN_WARNING, sdkp,
2823 "getting Control mode page failed, assume no ATO\n");
2824
2825 if (scsi_sense_valid(&sshdr))
2826 sd_print_sense_hdr(sdkp, &sshdr);
2827
2828 return;
2829 }
2830
2831 offset = data.header_length + data.block_descriptor_length;
2832
2833 if ((buffer[offset] & 0x3f) != 0x0a) {
2834 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2835 return;
2836 }
2837
2838 if ((buffer[offset + 5] & 0x80) == 0)
2839 return;
2840
2841 sdkp->ATO = 1;
2842
2843 return;
2844}
2845
2846/**
2847 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2848 * @sdkp: disk to query
2849 */
2850static void sd_read_block_limits(struct scsi_disk *sdkp)
2851{
2852 unsigned int sector_sz = sdkp->device->sector_size;
2853 const int vpd_len = 64;
2854 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2855
2856 if (!buffer ||
2857 /* Block Limits VPD */
2858 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2859 goto out;
2860
2861 blk_queue_io_min(sdkp->disk->queue,
2862 get_unaligned_be16(&buffer[6]) * sector_sz);
2863
2864 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2865 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2866
2867 if (buffer[3] == 0x3c) {
2868 unsigned int lba_count, desc_count;
2869
2870 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2871
2872 if (!sdkp->lbpme)
2873 goto out;
2874
2875 lba_count = get_unaligned_be32(&buffer[20]);
2876 desc_count = get_unaligned_be32(&buffer[24]);
2877
2878 if (lba_count && desc_count)
2879 sdkp->max_unmap_blocks = lba_count;
2880
2881 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2882
2883 if (buffer[32] & 0x80)
2884 sdkp->unmap_alignment =
2885 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2886
2887 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2888
2889 if (sdkp->max_unmap_blocks)
2890 sd_config_discard(sdkp, SD_LBP_UNMAP);
2891 else
2892 sd_config_discard(sdkp, SD_LBP_WS16);
2893
2894 } else { /* LBP VPD page tells us what to use */
2895 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2896 sd_config_discard(sdkp, SD_LBP_UNMAP);
2897 else if (sdkp->lbpws)
2898 sd_config_discard(sdkp, SD_LBP_WS16);
2899 else if (sdkp->lbpws10)
2900 sd_config_discard(sdkp, SD_LBP_WS10);
2901 else
2902 sd_config_discard(sdkp, SD_LBP_DISABLE);
2903 }
2904 }
2905
2906 out:
2907 kfree(buffer);
2908}
2909
2910/**
2911 * sd_read_block_characteristics - Query block dev. characteristics
2912 * @sdkp: disk to query
2913 */
2914static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2915{
2916 struct request_queue *q = sdkp->disk->queue;
2917 unsigned char *buffer;
2918 u16 rot;
2919 const int vpd_len = 64;
2920
2921 buffer = kmalloc(vpd_len, GFP_KERNEL);
2922
2923 if (!buffer ||
2924 /* Block Device Characteristics VPD */
2925 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2926 goto out;
2927
2928 rot = get_unaligned_be16(&buffer[4]);
2929
2930 if (rot == 1) {
2931 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2932 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2933 }
2934
2935 if (sdkp->device->type == TYPE_ZBC) {
2936 /* Host-managed */
2937 q->limits.zoned = BLK_ZONED_HM;
2938 } else {
2939 sdkp->zoned = (buffer[8] >> 4) & 3;
2940 if (sdkp->zoned == 1)
2941 /* Host-aware */
2942 q->limits.zoned = BLK_ZONED_HA;
2943 else
2944 /*
2945 * Treat drive-managed devices as
2946 * regular block devices.
2947 */
2948 q->limits.zoned = BLK_ZONED_NONE;
2949 }
2950 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2951 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2952 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2953
2954 out:
2955 kfree(buffer);
2956}
2957
2958/**
2959 * sd_read_block_provisioning - Query provisioning VPD page
2960 * @sdkp: disk to query
2961 */
2962static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2963{
2964 unsigned char *buffer;
2965 const int vpd_len = 8;
2966
2967 if (sdkp->lbpme == 0)
2968 return;
2969
2970 buffer = kmalloc(vpd_len, GFP_KERNEL);
2971
2972 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2973 goto out;
2974
2975 sdkp->lbpvpd = 1;
2976 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2977 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2978 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2979
2980 out:
2981 kfree(buffer);
2982}
2983
2984static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2985{
2986 struct scsi_device *sdev = sdkp->device;
2987
2988 if (sdev->host->no_write_same) {
2989 sdev->no_write_same = 1;
2990
2991 return;
2992 }
2993
2994 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2995 /* too large values might cause issues with arcmsr */
2996 int vpd_buf_len = 64;
2997
2998 sdev->no_report_opcodes = 1;
2999
3000 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3001 * CODES is unsupported and the device has an ATA
3002 * Information VPD page (SAT).
3003 */
3004 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3005 sdev->no_write_same = 1;
3006 }
3007
3008 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3009 sdkp->ws16 = 1;
3010
3011 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3012 sdkp->ws10 = 1;
3013}
3014
3015static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3016{
3017 struct scsi_device *sdev = sdkp->device;
3018
3019 if (!sdev->security_supported)
3020 return;
3021
3022 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3023 SECURITY_PROTOCOL_IN) == 1 &&
3024 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3025 SECURITY_PROTOCOL_OUT) == 1)
3026 sdkp->security = 1;
3027}
3028
3029/*
3030 * Determine the device's preferred I/O size for reads and writes
3031 * unless the reported value is unreasonably small, large, not a
3032 * multiple of the physical block size, or simply garbage.
3033 */
3034static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3035 unsigned int dev_max)
3036{
3037 struct scsi_device *sdp = sdkp->device;
3038 unsigned int opt_xfer_bytes =
3039 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3040
3041 if (sdkp->opt_xfer_blocks == 0)
3042 return false;
3043
3044 if (sdkp->opt_xfer_blocks > dev_max) {
3045 sd_first_printk(KERN_WARNING, sdkp,
3046 "Optimal transfer size %u logical blocks " \
3047 "> dev_max (%u logical blocks)\n",
3048 sdkp->opt_xfer_blocks, dev_max);
3049 return false;
3050 }
3051
3052 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3053 sd_first_printk(KERN_WARNING, sdkp,
3054 "Optimal transfer size %u logical blocks " \
3055 "> sd driver limit (%u logical blocks)\n",
3056 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3057 return false;
3058 }
3059
3060 if (opt_xfer_bytes < PAGE_SIZE) {
3061 sd_first_printk(KERN_WARNING, sdkp,
3062 "Optimal transfer size %u bytes < " \
3063 "PAGE_SIZE (%u bytes)\n",
3064 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3065 return false;
3066 }
3067
3068 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3069 sd_first_printk(KERN_WARNING, sdkp,
3070 "Optimal transfer size %u bytes not a " \
3071 "multiple of physical block size (%u bytes)\n",
3072 opt_xfer_bytes, sdkp->physical_block_size);
3073 return false;
3074 }
3075
3076 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3077 opt_xfer_bytes);
3078 return true;
3079}
3080
3081/**
3082 * sd_revalidate_disk - called the first time a new disk is seen,
3083 * performs disk spin up, read_capacity, etc.
3084 * @disk: struct gendisk we care about
3085 **/
3086static int sd_revalidate_disk(struct gendisk *disk)
3087{
3088 struct scsi_disk *sdkp = scsi_disk(disk);
3089 struct scsi_device *sdp = sdkp->device;
3090 struct request_queue *q = sdkp->disk->queue;
3091 sector_t old_capacity = sdkp->capacity;
3092 unsigned char *buffer;
3093 unsigned int dev_max, rw_max;
3094
3095 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3096 "sd_revalidate_disk\n"));
3097
3098 /*
3099 * If the device is offline, don't try and read capacity or any
3100 * of the other niceties.
3101 */
3102 if (!scsi_device_online(sdp))
3103 goto out;
3104
3105 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3106 if (!buffer) {
3107 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3108 "allocation failure.\n");
3109 goto out;
3110 }
3111
3112 sd_spinup_disk(sdkp);
3113
3114 /*
3115 * Without media there is no reason to ask; moreover, some devices
3116 * react badly if we do.
3117 */
3118 if (sdkp->media_present) {
3119 sd_read_capacity(sdkp, buffer);
3120
3121 /*
3122 * set the default to rotational. All non-rotational devices
3123 * support the block characteristics VPD page, which will
3124 * cause this to be updated correctly and any device which
3125 * doesn't support it should be treated as rotational.
3126 */
3127 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3128 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3129
3130 if (scsi_device_supports_vpd(sdp)) {
3131 sd_read_block_provisioning(sdkp);
3132 sd_read_block_limits(sdkp);
3133 sd_read_block_characteristics(sdkp);
3134 sd_zbc_read_zones(sdkp, buffer);
3135 }
3136
3137 sd_print_capacity(sdkp, old_capacity);
3138
3139 sd_read_write_protect_flag(sdkp, buffer);
3140 sd_read_cache_type(sdkp, buffer);
3141 sd_read_app_tag_own(sdkp, buffer);
3142 sd_read_write_same(sdkp, buffer);
3143 sd_read_security(sdkp, buffer);
3144 }
3145
3146 /*
3147 * We now have all cache related info, determine how we deal
3148 * with flush requests.
3149 */
3150 sd_set_flush_flag(sdkp);
3151
3152 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3153 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3154
3155 /* Some devices report a maximum block count for READ/WRITE requests. */
3156 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3157 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3158
3159 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3160 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3161 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3162 } else
3163 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3164 (sector_t)BLK_DEF_MAX_SECTORS);
3165
3166 /* Do not exceed controller limit */
3167 rw_max = min(rw_max, queue_max_hw_sectors(q));
3168
3169 /*
3170 * Only update max_sectors if previously unset or if the current value
3171 * exceeds the capabilities of the hardware.
3172 */
3173 if (sdkp->first_scan ||
3174 q->limits.max_sectors > q->limits.max_dev_sectors ||
3175 q->limits.max_sectors > q->limits.max_hw_sectors)
3176 q->limits.max_sectors = rw_max;
3177
3178 sdkp->first_scan = 0;
3179
3180 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3181 sd_config_write_same(sdkp);
3182 kfree(buffer);
3183
3184 out:
3185 return 0;
3186}
3187
3188/**
3189 * sd_unlock_native_capacity - unlock native capacity
3190 * @disk: struct gendisk to set capacity for
3191 *
3192 * Block layer calls this function if it detects that partitions
3193 * on @disk reach beyond the end of the device. If the SCSI host
3194 * implements ->unlock_native_capacity() method, it's invoked to
3195 * give it a chance to adjust the device capacity.
3196 *
3197 * CONTEXT:
3198 * Defined by block layer. Might sleep.
3199 */
3200static void sd_unlock_native_capacity(struct gendisk *disk)
3201{
3202 struct scsi_device *sdev = scsi_disk(disk)->device;
3203
3204 if (sdev->host->hostt->unlock_native_capacity)
3205 sdev->host->hostt->unlock_native_capacity(sdev);
3206}
3207
3208/**
3209 * sd_format_disk_name - format disk name
3210 * @prefix: name prefix - ie. "sd" for SCSI disks
3211 * @index: index of the disk to format name for
3212 * @buf: output buffer
3213 * @buflen: length of the output buffer
3214 *
3215 * SCSI disk names starts at sda. The 26th device is sdz and the
3216 * 27th is sdaa. The last one for two lettered suffix is sdzz
3217 * which is followed by sdaaa.
3218 *
3219 * This is basically 26 base counting with one extra 'nil' entry
3220 * at the beginning from the second digit on and can be
3221 * determined using similar method as 26 base conversion with the
3222 * index shifted -1 after each digit is computed.
3223 *
3224 * CONTEXT:
3225 * Don't care.
3226 *
3227 * RETURNS:
3228 * 0 on success, -errno on failure.
3229 */
3230static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3231{
3232 const int base = 'z' - 'a' + 1;
3233 char *begin = buf + strlen(prefix);
3234 char *end = buf + buflen;
3235 char *p;
3236 int unit;
3237
3238 p = end - 1;
3239 *p = '\0';
3240 unit = base;
3241 do {
3242 if (p == begin)
3243 return -EINVAL;
3244 *--p = 'a' + (index % unit);
3245 index = (index / unit) - 1;
3246 } while (index >= 0);
3247
3248 memmove(begin, p, end - p);
3249 memcpy(buf, prefix, strlen(prefix));
3250
3251 return 0;
3252}
3253
3254/**
3255 * sd_probe - called during driver initialization and whenever a
3256 * new scsi device is attached to the system. It is called once
3257 * for each scsi device (not just disks) present.
3258 * @dev: pointer to device object
3259 *
3260 * Returns 0 if successful (or not interested in this scsi device
3261 * (e.g. scanner)); 1 when there is an error.
3262 *
3263 * Note: this function is invoked from the scsi mid-level.
3264 * This function sets up the mapping between a given
3265 * <host,channel,id,lun> (found in sdp) and new device name
3266 * (e.g. /dev/sda). More precisely it is the block device major
3267 * and minor number that is chosen here.
3268 *
3269 * Assume sd_probe is not re-entrant (for time being)
3270 * Also think about sd_probe() and sd_remove() running coincidentally.
3271 **/
3272static int sd_probe(struct device *dev)
3273{
3274 struct scsi_device *sdp = to_scsi_device(dev);
3275 struct scsi_disk *sdkp;
3276 struct gendisk *gd;
3277 int index;
3278 int error;
3279
3280 scsi_autopm_get_device(sdp);
3281 error = -ENODEV;
3282 if (sdp->type != TYPE_DISK &&
3283 sdp->type != TYPE_ZBC &&
3284 sdp->type != TYPE_MOD &&
3285 sdp->type != TYPE_RBC)
3286 goto out;
3287
3288#ifndef CONFIG_BLK_DEV_ZONED
3289 if (sdp->type == TYPE_ZBC)
3290 goto out;
3291#endif
3292 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3293 "sd_probe\n"));
3294
3295 error = -ENOMEM;
3296 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3297 if (!sdkp)
3298 goto out;
3299
3300 gd = alloc_disk(SD_MINORS);
3301 if (!gd)
3302 goto out_free;
3303
3304 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3305 if (index < 0) {
3306 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3307 goto out_put;
3308 }
3309
3310 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3311 if (error) {
3312 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3313 goto out_free_index;
3314 }
3315
3316 sdkp->device = sdp;
3317 sdkp->driver = &sd_template;
3318 sdkp->disk = gd;
3319 sdkp->index = index;
3320 atomic_set(&sdkp->openers, 0);
3321 atomic_set(&sdkp->device->ioerr_cnt, 0);
3322
3323 if (!sdp->request_queue->rq_timeout) {
3324 if (sdp->type != TYPE_MOD)
3325 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3326 else
3327 blk_queue_rq_timeout(sdp->request_queue,
3328 SD_MOD_TIMEOUT);
3329 }
3330
3331 device_initialize(&sdkp->dev);
3332 sdkp->dev.parent = dev;
3333 sdkp->dev.class = &sd_disk_class;
3334 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3335
3336 error = device_add(&sdkp->dev);
3337 if (error)
3338 goto out_free_index;
3339
3340 get_device(dev);
3341 dev_set_drvdata(dev, sdkp);
3342
3343 gd->major = sd_major((index & 0xf0) >> 4);
3344 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3345
3346 gd->fops = &sd_fops;
3347 gd->private_data = &sdkp->driver;
3348 gd->queue = sdkp->device->request_queue;
3349
3350 /* defaults, until the device tells us otherwise */
3351 sdp->sector_size = 512;
3352 sdkp->capacity = 0;
3353 sdkp->media_present = 1;
3354 sdkp->write_prot = 0;
3355 sdkp->cache_override = 0;
3356 sdkp->WCE = 0;
3357 sdkp->RCD = 0;
3358 sdkp->ATO = 0;
3359 sdkp->first_scan = 1;
3360 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3361
3362 sd_revalidate_disk(gd);
3363
3364 gd->flags = GENHD_FL_EXT_DEVT;
3365 if (sdp->removable) {
3366 gd->flags |= GENHD_FL_REMOVABLE;
3367 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3368 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3369 }
3370
3371 blk_pm_runtime_init(sdp->request_queue, dev);
3372 device_add_disk(dev, gd, NULL);
3373 if (sdkp->capacity)
3374 sd_dif_config_host(sdkp);
3375
3376 sd_revalidate_disk(gd);
3377
3378 if (sdkp->security) {
3379 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3380 if (sdkp->opal_dev)
3381 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3382 }
3383
3384 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3385 sdp->removable ? "removable " : "");
3386 scsi_autopm_put_device(sdp);
3387
3388 return 0;
3389
3390 out_free_index:
3391 ida_free(&sd_index_ida, index);
3392 out_put:
3393 put_disk(gd);
3394 out_free:
3395 kfree(sdkp);
3396 out:
3397 scsi_autopm_put_device(sdp);
3398 return error;
3399}
3400
3401/**
3402 * sd_remove - called whenever a scsi disk (previously recognized by
3403 * sd_probe) is detached from the system. It is called (potentially
3404 * multiple times) during sd module unload.
3405 * @dev: pointer to device object
3406 *
3407 * Note: this function is invoked from the scsi mid-level.
3408 * This function potentially frees up a device name (e.g. /dev/sdc)
3409 * that could be re-used by a subsequent sd_probe().
3410 * This function is not called when the built-in sd driver is "exit-ed".
3411 **/
3412static int sd_remove(struct device *dev)
3413{
3414 struct scsi_disk *sdkp;
3415 dev_t devt;
3416
3417 sdkp = dev_get_drvdata(dev);
3418 devt = disk_devt(sdkp->disk);
3419 scsi_autopm_get_device(sdkp->device);
3420
3421 async_synchronize_full_domain(&scsi_sd_pm_domain);
3422 device_del(&sdkp->dev);
3423 del_gendisk(sdkp->disk);
3424 sd_shutdown(dev);
3425
3426 free_opal_dev(sdkp->opal_dev);
3427
3428 blk_register_region(devt, SD_MINORS, NULL,
3429 sd_default_probe, NULL, NULL);
3430
3431 mutex_lock(&sd_ref_mutex);
3432 dev_set_drvdata(dev, NULL);
3433 put_device(&sdkp->dev);
3434 mutex_unlock(&sd_ref_mutex);
3435
3436 return 0;
3437}
3438
3439/**
3440 * scsi_disk_release - Called to free the scsi_disk structure
3441 * @dev: pointer to embedded class device
3442 *
3443 * sd_ref_mutex must be held entering this routine. Because it is
3444 * called on last put, you should always use the scsi_disk_get()
3445 * scsi_disk_put() helpers which manipulate the semaphore directly
3446 * and never do a direct put_device.
3447 **/
3448static void scsi_disk_release(struct device *dev)
3449{
3450 struct scsi_disk *sdkp = to_scsi_disk(dev);
3451 struct gendisk *disk = sdkp->disk;
3452 struct request_queue *q = disk->queue;
3453
3454 ida_free(&sd_index_ida, sdkp->index);
3455
3456 /*
3457 * Wait until all requests that are in progress have completed.
3458 * This is necessary to avoid that e.g. scsi_end_request() crashes
3459 * due to clearing the disk->private_data pointer. Wait from inside
3460 * scsi_disk_release() instead of from sd_release() to avoid that
3461 * freezing and unfreezing the request queue affects user space I/O
3462 * in case multiple processes open a /dev/sd... node concurrently.
3463 */
3464 blk_mq_freeze_queue(q);
3465 blk_mq_unfreeze_queue(q);
3466
3467 disk->private_data = NULL;
3468 put_disk(disk);
3469 put_device(&sdkp->device->sdev_gendev);
3470
3471 kfree(sdkp);
3472}
3473
3474static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3475{
3476 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3477 struct scsi_sense_hdr sshdr;
3478 struct scsi_device *sdp = sdkp->device;
3479 int res;
3480
3481 if (start)
3482 cmd[4] |= 1; /* START */
3483
3484 if (sdp->start_stop_pwr_cond)
3485 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3486
3487 if (!scsi_device_online(sdp))
3488 return -ENODEV;
3489
3490 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3491 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3492 if (res) {
3493 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3494 if (driver_byte(res) == DRIVER_SENSE)
3495 sd_print_sense_hdr(sdkp, &sshdr);
3496 if (scsi_sense_valid(&sshdr) &&
3497 /* 0x3a is medium not present */
3498 sshdr.asc == 0x3a)
3499 res = 0;
3500 }
3501
3502 /* SCSI error codes must not go to the generic layer */
3503 if (res)
3504 return -EIO;
3505
3506 return 0;
3507}
3508
3509/*
3510 * Send a SYNCHRONIZE CACHE instruction down to the device through
3511 * the normal SCSI command structure. Wait for the command to
3512 * complete.
3513 */
3514static void sd_shutdown(struct device *dev)
3515{
3516 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3517
3518 if (!sdkp)
3519 return; /* this can happen */
3520
3521 if (pm_runtime_suspended(dev))
3522 return;
3523
3524 if (sdkp->WCE && sdkp->media_present) {
3525 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3526 sd_sync_cache(sdkp, NULL);
3527 }
3528
3529 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3530 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3531 sd_start_stop_device(sdkp, 0);
3532 }
3533}
3534
3535static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3536{
3537 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3538 struct scsi_sense_hdr sshdr;
3539 int ret = 0;
3540
3541 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3542 return 0;
3543
3544 if (sdkp->WCE && sdkp->media_present) {
3545 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3546 ret = sd_sync_cache(sdkp, &sshdr);
3547
3548 if (ret) {
3549 /* ignore OFFLINE device */
3550 if (ret == -ENODEV)
3551 return 0;
3552
3553 if (!scsi_sense_valid(&sshdr) ||
3554 sshdr.sense_key != ILLEGAL_REQUEST)
3555 return ret;
3556
3557 /*
3558 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3559 * doesn't support sync. There's not much to do and
3560 * suspend shouldn't fail.
3561 */
3562 ret = 0;
3563 }
3564 }
3565
3566 if (sdkp->device->manage_start_stop) {
3567 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3568 /* an error is not worth aborting a system sleep */
3569 ret = sd_start_stop_device(sdkp, 0);
3570 if (ignore_stop_errors)
3571 ret = 0;
3572 }
3573
3574 return ret;
3575}
3576
3577static int sd_suspend_system(struct device *dev)
3578{
3579 return sd_suspend_common(dev, true);
3580}
3581
3582static int sd_suspend_runtime(struct device *dev)
3583{
3584 return sd_suspend_common(dev, false);
3585}
3586
3587static int sd_resume(struct device *dev)
3588{
3589 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3590 int ret;
3591
3592 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3593 return 0;
3594
3595 if (!sdkp->device->manage_start_stop)
3596 return 0;
3597
3598 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3599 ret = sd_start_stop_device(sdkp, 1);
3600 if (!ret)
3601 opal_unlock_from_suspend(sdkp->opal_dev);
3602 return ret;
3603}
3604
3605/**
3606 * init_sd - entry point for this driver (both when built in or when
3607 * a module).
3608 *
3609 * Note: this function registers this driver with the scsi mid-level.
3610 **/
3611static int __init init_sd(void)
3612{
3613 int majors = 0, i, err;
3614
3615 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3616
3617 for (i = 0; i < SD_MAJORS; i++) {
3618 if (register_blkdev(sd_major(i), "sd") != 0)
3619 continue;
3620 majors++;
3621 blk_register_region(sd_major(i), SD_MINORS, NULL,
3622 sd_default_probe, NULL, NULL);
3623 }
3624
3625 if (!majors)
3626 return -ENODEV;
3627
3628 err = class_register(&sd_disk_class);
3629 if (err)
3630 goto err_out;
3631
3632 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3633 0, 0, NULL);
3634 if (!sd_cdb_cache) {
3635 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3636 err = -ENOMEM;
3637 goto err_out_class;
3638 }
3639
3640 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3641 if (!sd_cdb_pool) {
3642 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3643 err = -ENOMEM;
3644 goto err_out_cache;
3645 }
3646
3647 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3648 if (!sd_page_pool) {
3649 printk(KERN_ERR "sd: can't init discard page pool\n");
3650 err = -ENOMEM;
3651 goto err_out_ppool;
3652 }
3653
3654 err = scsi_register_driver(&sd_template.gendrv);
3655 if (err)
3656 goto err_out_driver;
3657
3658 return 0;
3659
3660err_out_driver:
3661 mempool_destroy(sd_page_pool);
3662
3663err_out_ppool:
3664 mempool_destroy(sd_cdb_pool);
3665
3666err_out_cache:
3667 kmem_cache_destroy(sd_cdb_cache);
3668
3669err_out_class:
3670 class_unregister(&sd_disk_class);
3671err_out:
3672 for (i = 0; i < SD_MAJORS; i++)
3673 unregister_blkdev(sd_major(i), "sd");
3674 return err;
3675}
3676
3677/**
3678 * exit_sd - exit point for this driver (when it is a module).
3679 *
3680 * Note: this function unregisters this driver from the scsi mid-level.
3681 **/
3682static void __exit exit_sd(void)
3683{
3684 int i;
3685
3686 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3687
3688 scsi_unregister_driver(&sd_template.gendrv);
3689 mempool_destroy(sd_cdb_pool);
3690 mempool_destroy(sd_page_pool);
3691 kmem_cache_destroy(sd_cdb_cache);
3692
3693 class_unregister(&sd_disk_class);
3694
3695 for (i = 0; i < SD_MAJORS; i++) {
3696 blk_unregister_region(sd_major(i), SD_MINORS);
3697 unregister_blkdev(sd_major(i), "sd");
3698 }
3699}
3700
3701module_init(init_sd);
3702module_exit(exit_sd);
3703
3704static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3705 struct scsi_sense_hdr *sshdr)
3706{
3707 scsi_print_sense_hdr(sdkp->device,
3708 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3709}
3710
3711static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3712 int result)
3713{
3714 const char *hb_string = scsi_hostbyte_string(result);
3715 const char *db_string = scsi_driverbyte_string(result);
3716
3717 if (hb_string || db_string)
3718 sd_printk(KERN_INFO, sdkp,
3719 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3720 hb_string ? hb_string : "invalid",
3721 db_string ? db_string : "invalid");
3722 else
3723 sd_printk(KERN_INFO, sdkp,
3724 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3725 msg, host_byte(result), driver_byte(result));
3726}
3727