projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'pm-6.16-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
[linux-2.6-block.git] / drivers / scsi / sd.c
... / ...
CommitLineData
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/bio-integrity.h>
37#include <linux/module.h>
38#include <linux/fs.h>
39#include <linux/kernel.h>
40#include <linux/mm.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 <linux/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_devinfo.h>
67#include <scsi/scsi_driver.h>
68#include <scsi/scsi_eh.h>
69#include <scsi/scsi_host.h>
70#include <scsi/scsi_ioctl.h>
71#include <scsi/scsicam.h>
72#include <scsi/scsi_common.h>
73
74#include "sd.h"
75#include "scsi_priv.h"
76#include "scsi_logging.h"
77
78MODULE_AUTHOR("Eric Youngdale");
79MODULE_DESCRIPTION("SCSI disk (sd) driver");
80MODULE_LICENSE("GPL");
81
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
94MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
95MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
96MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
97MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
98MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
99MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
100MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
101MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
102
103#define SD_MINORS 16
104
105static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
106 unsigned int mode);
107static void sd_config_write_same(struct scsi_disk *sdkp,
108 struct queue_limits *lim);
109static int sd_revalidate_disk(struct gendisk *);
110static void sd_unlock_native_capacity(struct gendisk *disk);
111static void sd_shutdown(struct device *);
112static void scsi_disk_release(struct device *cdev);
113
114static DEFINE_IDA(sd_index_ida);
115
116static mempool_t *sd_page_pool;
117static struct lock_class_key sd_bio_compl_lkclass;
118
119static const char *sd_cache_types[] = {
120 "write through", "none", "write back",
121 "write back, no read (daft)"
122};
123
124static void sd_set_flush_flag(struct scsi_disk *sdkp,
125 struct queue_limits *lim)
126{
127 if (sdkp->WCE) {
128 lim->features |= BLK_FEAT_WRITE_CACHE;
129 if (sdkp->DPOFUA)
130 lim->features |= BLK_FEAT_FUA;
131 else
132 lim->features &= ~BLK_FEAT_FUA;
133 } else {
134 lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA);
135 }
136}
137
138static ssize_t
139cache_type_store(struct device *dev, struct device_attribute *attr,
140 const char *buf, size_t count)
141{
142 int ct, rcd, wce, sp;
143 struct scsi_disk *sdkp = to_scsi_disk(dev);
144 struct scsi_device *sdp = sdkp->device;
145 char buffer[64];
146 char *buffer_data;
147 struct scsi_mode_data data;
148 struct scsi_sense_hdr sshdr;
149 static const char temp[] = "temporary ";
150 int len, ret;
151
152 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
153 /* no cache control on RBC devices; theoretically they
154 * can do it, but there's probably so many exceptions
155 * it's not worth the risk */
156 return -EINVAL;
157
158 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
159 buf += sizeof(temp) - 1;
160 sdkp->cache_override = 1;
161 } else {
162 sdkp->cache_override = 0;
163 }
164
165 ct = sysfs_match_string(sd_cache_types, buf);
166 if (ct < 0)
167 return -EINVAL;
168
169 rcd = ct & 0x01 ? 1 : 0;
170 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
171
172 if (sdkp->cache_override) {
173 struct queue_limits lim;
174
175 sdkp->WCE = wce;
176 sdkp->RCD = rcd;
177
178 lim = queue_limits_start_update(sdkp->disk->queue);
179 sd_set_flush_flag(sdkp, &lim);
180 ret = queue_limits_commit_update_frozen(sdkp->disk->queue,
181 &lim);
182 if (ret)
183 return ret;
184 return count;
185 }
186
187 if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
188 sdkp->max_retries, &data, NULL))
189 return -EINVAL;
190 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
191 data.block_descriptor_length);
192 buffer_data = buffer + data.header_length +
193 data.block_descriptor_length;
194 buffer_data[2] &= ~0x05;
195 buffer_data[2] |= wce << 2 | rcd;
196 sp = buffer_data[0] & 0x80 ? 1 : 0;
197 buffer_data[0] &= ~0x80;
198
199 /*
200 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
201 * received mode parameter buffer before doing MODE SELECT.
202 */
203 data.device_specific = 0;
204
205 ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
206 sdkp->max_retries, &data, &sshdr);
207 if (ret) {
208 if (ret > 0 && scsi_sense_valid(&sshdr))
209 sd_print_sense_hdr(sdkp, &sshdr);
210 return -EINVAL;
211 }
212 sd_revalidate_disk(sdkp->disk);
213 return count;
214}
215
216static ssize_t
217manage_start_stop_show(struct device *dev,
218 struct device_attribute *attr, char *buf)
219{
220 struct scsi_disk *sdkp = to_scsi_disk(dev);
221 struct scsi_device *sdp = sdkp->device;
222
223 return sysfs_emit(buf, "%u\n",
224 sdp->manage_system_start_stop &&
225 sdp->manage_runtime_start_stop &&
226 sdp->manage_shutdown);
227}
228static DEVICE_ATTR_RO(manage_start_stop);
229
230static ssize_t
231manage_system_start_stop_show(struct device *dev,
232 struct device_attribute *attr, char *buf)
233{
234 struct scsi_disk *sdkp = to_scsi_disk(dev);
235 struct scsi_device *sdp = sdkp->device;
236
237 return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
238}
239
240static ssize_t
241manage_system_start_stop_store(struct device *dev,
242 struct device_attribute *attr,
243 const char *buf, size_t count)
244{
245 struct scsi_disk *sdkp = to_scsi_disk(dev);
246 struct scsi_device *sdp = sdkp->device;
247 bool v;
248
249 if (!capable(CAP_SYS_ADMIN))
250 return -EACCES;
251
252 if (kstrtobool(buf, &v))
253 return -EINVAL;
254
255 sdp->manage_system_start_stop = v;
256
257 return count;
258}
259static DEVICE_ATTR_RW(manage_system_start_stop);
260
261static ssize_t
262manage_runtime_start_stop_show(struct device *dev,
263 struct device_attribute *attr, char *buf)
264{
265 struct scsi_disk *sdkp = to_scsi_disk(dev);
266 struct scsi_device *sdp = sdkp->device;
267
268 return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
269}
270
271static ssize_t
272manage_runtime_start_stop_store(struct device *dev,
273 struct device_attribute *attr,
274 const char *buf, size_t count)
275{
276 struct scsi_disk *sdkp = to_scsi_disk(dev);
277 struct scsi_device *sdp = sdkp->device;
278 bool v;
279
280 if (!capable(CAP_SYS_ADMIN))
281 return -EACCES;
282
283 if (kstrtobool(buf, &v))
284 return -EINVAL;
285
286 sdp->manage_runtime_start_stop = v;
287
288 return count;
289}
290static DEVICE_ATTR_RW(manage_runtime_start_stop);
291
292static ssize_t manage_shutdown_show(struct device *dev,
293 struct device_attribute *attr, char *buf)
294{
295 struct scsi_disk *sdkp = to_scsi_disk(dev);
296 struct scsi_device *sdp = sdkp->device;
297
298 return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
299}
300
301static ssize_t manage_shutdown_store(struct device *dev,
302 struct device_attribute *attr,
303 const char *buf, size_t count)
304{
305 struct scsi_disk *sdkp = to_scsi_disk(dev);
306 struct scsi_device *sdp = sdkp->device;
307 bool v;
308
309 if (!capable(CAP_SYS_ADMIN))
310 return -EACCES;
311
312 if (kstrtobool(buf, &v))
313 return -EINVAL;
314
315 sdp->manage_shutdown = v;
316
317 return count;
318}
319static DEVICE_ATTR_RW(manage_shutdown);
320
321static ssize_t
322allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
323{
324 struct scsi_disk *sdkp = to_scsi_disk(dev);
325
326 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
327}
328
329static ssize_t
330allow_restart_store(struct device *dev, struct device_attribute *attr,
331 const char *buf, size_t count)
332{
333 bool v;
334 struct scsi_disk *sdkp = to_scsi_disk(dev);
335 struct scsi_device *sdp = sdkp->device;
336
337 if (!capable(CAP_SYS_ADMIN))
338 return -EACCES;
339
340 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
341 return -EINVAL;
342
343 if (kstrtobool(buf, &v))
344 return -EINVAL;
345
346 sdp->allow_restart = v;
347
348 return count;
349}
350static DEVICE_ATTR_RW(allow_restart);
351
352static ssize_t
353cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
354{
355 struct scsi_disk *sdkp = to_scsi_disk(dev);
356 int ct = sdkp->RCD + 2*sdkp->WCE;
357
358 return sprintf(buf, "%s\n", sd_cache_types[ct]);
359}
360static DEVICE_ATTR_RW(cache_type);
361
362static ssize_t
363FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
364{
365 struct scsi_disk *sdkp = to_scsi_disk(dev);
366
367 return sprintf(buf, "%u\n", sdkp->DPOFUA);
368}
369static DEVICE_ATTR_RO(FUA);
370
371static ssize_t
372protection_type_show(struct device *dev, struct device_attribute *attr,
373 char *buf)
374{
375 struct scsi_disk *sdkp = to_scsi_disk(dev);
376
377 return sprintf(buf, "%u\n", sdkp->protection_type);
378}
379
380static ssize_t
381protection_type_store(struct device *dev, struct device_attribute *attr,
382 const char *buf, size_t count)
383{
384 struct scsi_disk *sdkp = to_scsi_disk(dev);
385 unsigned int val;
386 int err;
387
388 if (!capable(CAP_SYS_ADMIN))
389 return -EACCES;
390
391 err = kstrtouint(buf, 10, &val);
392
393 if (err)
394 return err;
395
396 if (val <= T10_PI_TYPE3_PROTECTION)
397 sdkp->protection_type = val;
398
399 return count;
400}
401static DEVICE_ATTR_RW(protection_type);
402
403static ssize_t
404protection_mode_show(struct device *dev, struct device_attribute *attr,
405 char *buf)
406{
407 struct scsi_disk *sdkp = to_scsi_disk(dev);
408 struct scsi_device *sdp = sdkp->device;
409 unsigned int dif, dix;
410
411 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
412 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
413
414 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
415 dif = 0;
416 dix = 1;
417 }
418
419 if (!dif && !dix)
420 return sprintf(buf, "none\n");
421
422 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
423}
424static DEVICE_ATTR_RO(protection_mode);
425
426static ssize_t
427app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
428{
429 struct scsi_disk *sdkp = to_scsi_disk(dev);
430
431 return sprintf(buf, "%u\n", sdkp->ATO);
432}
433static DEVICE_ATTR_RO(app_tag_own);
434
435static ssize_t
436thin_provisioning_show(struct device *dev, struct device_attribute *attr,
437 char *buf)
438{
439 struct scsi_disk *sdkp = to_scsi_disk(dev);
440
441 return sprintf(buf, "%u\n", sdkp->lbpme);
442}
443static DEVICE_ATTR_RO(thin_provisioning);
444
445/* sysfs_match_string() requires dense arrays */
446static const char *lbp_mode[] = {
447 [SD_LBP_FULL] = "full",
448 [SD_LBP_UNMAP] = "unmap",
449 [SD_LBP_WS16] = "writesame_16",
450 [SD_LBP_WS10] = "writesame_10",
451 [SD_LBP_ZERO] = "writesame_zero",
452 [SD_LBP_DISABLE] = "disabled",
453};
454
455static ssize_t
456provisioning_mode_show(struct device *dev, struct device_attribute *attr,
457 char *buf)
458{
459 struct scsi_disk *sdkp = to_scsi_disk(dev);
460
461 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
462}
463
464static ssize_t
465provisioning_mode_store(struct device *dev, struct device_attribute *attr,
466 const char *buf, size_t count)
467{
468 struct scsi_disk *sdkp = to_scsi_disk(dev);
469 struct scsi_device *sdp = sdkp->device;
470 struct queue_limits lim;
471 int mode, err;
472
473 if (!capable(CAP_SYS_ADMIN))
474 return -EACCES;
475
476 if (sdp->type != TYPE_DISK)
477 return -EINVAL;
478
479 mode = sysfs_match_string(lbp_mode, buf);
480 if (mode < 0)
481 return -EINVAL;
482
483 lim = queue_limits_start_update(sdkp->disk->queue);
484 sd_config_discard(sdkp, &lim, mode);
485 err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
486 if (err)
487 return err;
488 return count;
489}
490static DEVICE_ATTR_RW(provisioning_mode);
491
492/* sysfs_match_string() requires dense arrays */
493static const char *zeroing_mode[] = {
494 [SD_ZERO_WRITE] = "write",
495 [SD_ZERO_WS] = "writesame",
496 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
497 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
498};
499
500static ssize_t
501zeroing_mode_show(struct device *dev, struct device_attribute *attr,
502 char *buf)
503{
504 struct scsi_disk *sdkp = to_scsi_disk(dev);
505
506 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
507}
508
509static ssize_t
510zeroing_mode_store(struct device *dev, struct device_attribute *attr,
511 const char *buf, size_t count)
512{
513 struct scsi_disk *sdkp = to_scsi_disk(dev);
514 int mode;
515
516 if (!capable(CAP_SYS_ADMIN))
517 return -EACCES;
518
519 mode = sysfs_match_string(zeroing_mode, buf);
520 if (mode < 0)
521 return -EINVAL;
522
523 sdkp->zeroing_mode = mode;
524
525 return count;
526}
527static DEVICE_ATTR_RW(zeroing_mode);
528
529static ssize_t
530max_medium_access_timeouts_show(struct device *dev,
531 struct device_attribute *attr, char *buf)
532{
533 struct scsi_disk *sdkp = to_scsi_disk(dev);
534
535 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
536}
537
538static ssize_t
539max_medium_access_timeouts_store(struct device *dev,
540 struct device_attribute *attr, const char *buf,
541 size_t count)
542{
543 struct scsi_disk *sdkp = to_scsi_disk(dev);
544 int err;
545
546 if (!capable(CAP_SYS_ADMIN))
547 return -EACCES;
548
549 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
550
551 return err ? err : count;
552}
553static DEVICE_ATTR_RW(max_medium_access_timeouts);
554
555static ssize_t
556max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
557 char *buf)
558{
559 struct scsi_disk *sdkp = to_scsi_disk(dev);
560
561 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
562}
563
564static ssize_t
565max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
566 const char *buf, size_t count)
567{
568 struct scsi_disk *sdkp = to_scsi_disk(dev);
569 struct scsi_device *sdp = sdkp->device;
570 struct queue_limits lim;
571 unsigned long max;
572 int err;
573
574 if (!capable(CAP_SYS_ADMIN))
575 return -EACCES;
576
577 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
578 return -EINVAL;
579
580 err = kstrtoul(buf, 10, &max);
581
582 if (err)
583 return err;
584
585 if (max == 0)
586 sdp->no_write_same = 1;
587 else if (max <= SD_MAX_WS16_BLOCKS) {
588 sdp->no_write_same = 0;
589 sdkp->max_ws_blocks = max;
590 }
591
592 lim = queue_limits_start_update(sdkp->disk->queue);
593 sd_config_write_same(sdkp, &lim);
594 err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
595 if (err)
596 return err;
597 return count;
598}
599static DEVICE_ATTR_RW(max_write_same_blocks);
600
601static ssize_t
602zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
603{
604 struct scsi_disk *sdkp = to_scsi_disk(dev);
605
606 if (sdkp->device->type == TYPE_ZBC)
607 return sprintf(buf, "host-managed\n");
608 if (sdkp->zoned == 1)
609 return sprintf(buf, "host-aware\n");
610 if (sdkp->zoned == 2)
611 return sprintf(buf, "drive-managed\n");
612 return sprintf(buf, "none\n");
613}
614static DEVICE_ATTR_RO(zoned_cap);
615
616static ssize_t
617max_retries_store(struct device *dev, struct device_attribute *attr,
618 const char *buf, size_t count)
619{
620 struct scsi_disk *sdkp = to_scsi_disk(dev);
621 struct scsi_device *sdev = sdkp->device;
622 int retries, err;
623
624 err = kstrtoint(buf, 10, &retries);
625 if (err)
626 return err;
627
628 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
629 sdkp->max_retries = retries;
630 return count;
631 }
632
633 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
634 SD_MAX_RETRIES);
635 return -EINVAL;
636}
637
638static ssize_t
639max_retries_show(struct device *dev, struct device_attribute *attr,
640 char *buf)
641{
642 struct scsi_disk *sdkp = to_scsi_disk(dev);
643
644 return sprintf(buf, "%d\n", sdkp->max_retries);
645}
646
647static DEVICE_ATTR_RW(max_retries);
648
649static struct attribute *sd_disk_attrs[] = {
650 &dev_attr_cache_type.attr,
651 &dev_attr_FUA.attr,
652 &dev_attr_allow_restart.attr,
653 &dev_attr_manage_start_stop.attr,
654 &dev_attr_manage_system_start_stop.attr,
655 &dev_attr_manage_runtime_start_stop.attr,
656 &dev_attr_manage_shutdown.attr,
657 &dev_attr_protection_type.attr,
658 &dev_attr_protection_mode.attr,
659 &dev_attr_app_tag_own.attr,
660 &dev_attr_thin_provisioning.attr,
661 &dev_attr_provisioning_mode.attr,
662 &dev_attr_zeroing_mode.attr,
663 &dev_attr_max_write_same_blocks.attr,
664 &dev_attr_max_medium_access_timeouts.attr,
665 &dev_attr_zoned_cap.attr,
666 &dev_attr_max_retries.attr,
667 NULL,
668};
669ATTRIBUTE_GROUPS(sd_disk);
670
671static struct class sd_disk_class = {
672 .name = "scsi_disk",
673 .dev_release = scsi_disk_release,
674 .dev_groups = sd_disk_groups,
675};
676
677/*
678 * Don't request a new module, as that could deadlock in multipath
679 * environment.
680 */
681static void sd_default_probe(dev_t devt)
682{
683}
684
685/*
686 * Device no to disk mapping:
687 *
688 * major disc2 disc p1
689 * |............|.............|....|....| <- dev_t
690 * 31 20 19 8 7 4 3 0
691 *
692 * Inside a major, we have 16k disks, however mapped non-
693 * contiguously. The first 16 disks are for major0, the next
694 * ones with major1, ... Disk 256 is for major0 again, disk 272
695 * for major1, ...
696 * As we stay compatible with our numbering scheme, we can reuse
697 * the well-know SCSI majors 8, 65--71, 136--143.
698 */
699static int sd_major(int major_idx)
700{
701 switch (major_idx) {
702 case 0:
703 return SCSI_DISK0_MAJOR;
704 case 1 ... 7:
705 return SCSI_DISK1_MAJOR + major_idx - 1;
706 case 8 ... 15:
707 return SCSI_DISK8_MAJOR + major_idx - 8;
708 default:
709 BUG();
710 return 0; /* shut up gcc */
711 }
712}
713
714#ifdef CONFIG_BLK_SED_OPAL
715static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
716 size_t len, bool send)
717{
718 struct scsi_disk *sdkp = data;
719 struct scsi_device *sdev = sdkp->device;
720 u8 cdb[12] = { 0, };
721 const struct scsi_exec_args exec_args = {
722 .req_flags = BLK_MQ_REQ_PM,
723 };
724 int ret;
725
726 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
727 cdb[1] = secp;
728 put_unaligned_be16(spsp, &cdb[2]);
729 put_unaligned_be32(len, &cdb[6]);
730
731 ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
732 buffer, len, SD_TIMEOUT, sdkp->max_retries,
733 &exec_args);
734 return ret <= 0 ? ret : -EIO;
735}
736#endif /* CONFIG_BLK_SED_OPAL */
737
738/*
739 * Look up the DIX operation based on whether the command is read or
740 * write and whether dix and dif are enabled.
741 */
742static unsigned int sd_prot_op(bool write, bool dix, bool dif)
743{
744 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
745 static const unsigned int ops[] = { /* wrt dix dif */
746 SCSI_PROT_NORMAL, /* 0 0 0 */
747 SCSI_PROT_READ_STRIP, /* 0 0 1 */
748 SCSI_PROT_READ_INSERT, /* 0 1 0 */
749 SCSI_PROT_READ_PASS, /* 0 1 1 */
750 SCSI_PROT_NORMAL, /* 1 0 0 */
751 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
752 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
753 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
754 };
755
756 return ops[write << 2 | dix << 1 | dif];
757}
758
759/*
760 * Returns a mask of the protection flags that are valid for a given DIX
761 * operation.
762 */
763static unsigned int sd_prot_flag_mask(unsigned int prot_op)
764{
765 static const unsigned int flag_mask[] = {
766 [SCSI_PROT_NORMAL] = 0,
767
768 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
769 SCSI_PROT_GUARD_CHECK |
770 SCSI_PROT_REF_CHECK |
771 SCSI_PROT_REF_INCREMENT,
772
773 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
774 SCSI_PROT_IP_CHECKSUM,
775
776 [SCSI_PROT_READ_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 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
783 SCSI_PROT_REF_INCREMENT,
784
785 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
786 SCSI_PROT_REF_CHECK |
787 SCSI_PROT_REF_INCREMENT |
788 SCSI_PROT_IP_CHECKSUM,
789
790 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
791 SCSI_PROT_GUARD_CHECK |
792 SCSI_PROT_REF_CHECK |
793 SCSI_PROT_REF_INCREMENT |
794 SCSI_PROT_IP_CHECKSUM,
795 };
796
797 return flag_mask[prot_op];
798}
799
800static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
801 unsigned int dix, unsigned int dif)
802{
803 struct request *rq = scsi_cmd_to_rq(scmd);
804 struct bio *bio = rq->bio;
805 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
806 unsigned int protect = 0;
807
808 if (dix) { /* DIX Type 0, 1, 2, 3 */
809 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
810 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
811
812 if (bio_integrity_flagged(bio, BIP_CHECK_GUARD))
813 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
814 }
815
816 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
817 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
818
819 if (bio_integrity_flagged(bio, BIP_CHECK_REFTAG))
820 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
821 }
822
823 if (dif) { /* DIX/DIF Type 1, 2, 3 */
824 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
825
826 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
827 protect = 3 << 5; /* Disable target PI checking */
828 else
829 protect = 1 << 5; /* Enable target PI checking */
830 }
831
832 scsi_set_prot_op(scmd, prot_op);
833 scsi_set_prot_type(scmd, dif);
834 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
835
836 return protect;
837}
838
839static void sd_disable_discard(struct scsi_disk *sdkp)
840{
841 sdkp->provisioning_mode = SD_LBP_DISABLE;
842 blk_queue_disable_discard(sdkp->disk->queue);
843}
844
845static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
846 unsigned int mode)
847{
848 unsigned int logical_block_size = sdkp->device->sector_size;
849 unsigned int max_blocks = 0;
850
851 lim->discard_alignment = sdkp->unmap_alignment * logical_block_size;
852 lim->discard_granularity = max(sdkp->physical_block_size,
853 sdkp->unmap_granularity * logical_block_size);
854 sdkp->provisioning_mode = mode;
855
856 switch (mode) {
857
858 case SD_LBP_FULL:
859 case SD_LBP_DISABLE:
860 break;
861
862 case SD_LBP_UNMAP:
863 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
864 (u32)SD_MAX_WS16_BLOCKS);
865 break;
866
867 case SD_LBP_WS16:
868 if (sdkp->device->unmap_limit_for_ws)
869 max_blocks = sdkp->max_unmap_blocks;
870 else
871 max_blocks = sdkp->max_ws_blocks;
872
873 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
874 break;
875
876 case SD_LBP_WS10:
877 if (sdkp->device->unmap_limit_for_ws)
878 max_blocks = sdkp->max_unmap_blocks;
879 else
880 max_blocks = sdkp->max_ws_blocks;
881
882 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
883 break;
884
885 case SD_LBP_ZERO:
886 max_blocks = min_not_zero(sdkp->max_ws_blocks,
887 (u32)SD_MAX_WS10_BLOCKS);
888 break;
889 }
890
891 lim->max_hw_discard_sectors = max_blocks *
892 (logical_block_size >> SECTOR_SHIFT);
893}
894
895static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
896{
897 struct page *page;
898
899 page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
900 if (!page)
901 return NULL;
902 clear_highpage(page);
903 bvec_set_page(&rq->special_vec, page, data_len, 0);
904 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905 return bvec_virt(&rq->special_vec);
906}
907
908static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
909{
910 struct scsi_device *sdp = cmd->device;
911 struct request *rq = scsi_cmd_to_rq(cmd);
912 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
913 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
914 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
915 unsigned int data_len = 24;
916 char *buf;
917
918 buf = sd_set_special_bvec(rq, data_len);
919 if (!buf)
920 return BLK_STS_RESOURCE;
921
922 cmd->cmd_len = 10;
923 cmd->cmnd[0] = UNMAP;
924 cmd->cmnd[8] = 24;
925
926 put_unaligned_be16(6 + 16, &buf[0]);
927 put_unaligned_be16(16, &buf[2]);
928 put_unaligned_be64(lba, &buf[8]);
929 put_unaligned_be32(nr_blocks, &buf[16]);
930
931 cmd->allowed = sdkp->max_retries;
932 cmd->transfersize = data_len;
933 rq->timeout = SD_TIMEOUT;
934
935 return scsi_alloc_sgtables(cmd);
936}
937
938static void sd_config_atomic(struct scsi_disk *sdkp, struct queue_limits *lim)
939{
940 unsigned int logical_block_size = sdkp->device->sector_size,
941 physical_block_size_sectors, max_atomic, unit_min, unit_max;
942
943 if ((!sdkp->max_atomic && !sdkp->max_atomic_with_boundary) ||
944 sdkp->protection_type == T10_PI_TYPE2_PROTECTION)
945 return;
946
947 physical_block_size_sectors = sdkp->physical_block_size /
948 sdkp->device->sector_size;
949
950 unit_min = rounddown_pow_of_two(sdkp->atomic_granularity ?
951 sdkp->atomic_granularity :
952 physical_block_size_sectors);
953
954 /*
955 * Only use atomic boundary when we have the odd scenario of
956 * sdkp->max_atomic == 0, which the spec does permit.
957 */
958 if (sdkp->max_atomic) {
959 max_atomic = sdkp->max_atomic;
960 unit_max = rounddown_pow_of_two(sdkp->max_atomic);
961 sdkp->use_atomic_write_boundary = 0;
962 } else {
963 max_atomic = sdkp->max_atomic_with_boundary;
964 unit_max = rounddown_pow_of_two(sdkp->max_atomic_boundary);
965 sdkp->use_atomic_write_boundary = 1;
966 }
967
968 /*
969 * Ensure compliance with granularity and alignment. For now, keep it
970 * simple and just don't support atomic writes for values mismatched
971 * with max_{boundary}atomic, physical block size, and
972 * atomic_granularity itself.
973 *
974 * We're really being distrustful by checking unit_max also...
975 */
976 if (sdkp->atomic_granularity > 1) {
977 if (unit_min > 1 && unit_min % sdkp->atomic_granularity)
978 return;
979 if (unit_max > 1 && unit_max % sdkp->atomic_granularity)
980 return;
981 }
982
983 if (sdkp->atomic_alignment > 1) {
984 if (unit_min > 1 && unit_min % sdkp->atomic_alignment)
985 return;
986 if (unit_max > 1 && unit_max % sdkp->atomic_alignment)
987 return;
988 }
989
990 lim->atomic_write_hw_max = max_atomic * logical_block_size;
991 lim->atomic_write_hw_boundary = 0;
992 lim->atomic_write_hw_unit_min = unit_min * logical_block_size;
993 lim->atomic_write_hw_unit_max = unit_max * logical_block_size;
994 lim->features |= BLK_FEAT_ATOMIC_WRITES;
995}
996
997static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
998 bool unmap)
999{
1000 struct scsi_device *sdp = cmd->device;
1001 struct request *rq = scsi_cmd_to_rq(cmd);
1002 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1003 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1004 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1005 u32 data_len = sdp->sector_size;
1006
1007 if (!sd_set_special_bvec(rq, data_len))
1008 return BLK_STS_RESOURCE;
1009
1010 cmd->cmd_len = 16;
1011 cmd->cmnd[0] = WRITE_SAME_16;
1012 if (unmap)
1013 cmd->cmnd[1] = 0x8; /* UNMAP */
1014 put_unaligned_be64(lba, &cmd->cmnd[2]);
1015 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1016
1017 cmd->allowed = sdkp->max_retries;
1018 cmd->transfersize = data_len;
1019 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1020
1021 return scsi_alloc_sgtables(cmd);
1022}
1023
1024static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
1025 bool unmap)
1026{
1027 struct scsi_device *sdp = cmd->device;
1028 struct request *rq = scsi_cmd_to_rq(cmd);
1029 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1030 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1031 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1032 u32 data_len = sdp->sector_size;
1033
1034 if (!sd_set_special_bvec(rq, data_len))
1035 return BLK_STS_RESOURCE;
1036
1037 cmd->cmd_len = 10;
1038 cmd->cmnd[0] = WRITE_SAME;
1039 if (unmap)
1040 cmd->cmnd[1] = 0x8; /* UNMAP */
1041 put_unaligned_be32(lba, &cmd->cmnd[2]);
1042 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1043
1044 cmd->allowed = sdkp->max_retries;
1045 cmd->transfersize = data_len;
1046 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1047
1048 return scsi_alloc_sgtables(cmd);
1049}
1050
1051static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
1052{
1053 struct request *rq = scsi_cmd_to_rq(cmd);
1054 struct scsi_device *sdp = cmd->device;
1055 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1056 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1057 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1058
1059 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
1060 switch (sdkp->zeroing_mode) {
1061 case SD_ZERO_WS16_UNMAP:
1062 return sd_setup_write_same16_cmnd(cmd, true);
1063 case SD_ZERO_WS10_UNMAP:
1064 return sd_setup_write_same10_cmnd(cmd, true);
1065 }
1066 }
1067
1068 if (sdp->no_write_same) {
1069 rq->rq_flags |= RQF_QUIET;
1070 return BLK_STS_TARGET;
1071 }
1072
1073 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
1074 return sd_setup_write_same16_cmnd(cmd, false);
1075
1076 return sd_setup_write_same10_cmnd(cmd, false);
1077}
1078
1079static void sd_disable_write_same(struct scsi_disk *sdkp)
1080{
1081 sdkp->device->no_write_same = 1;
1082 sdkp->max_ws_blocks = 0;
1083 blk_queue_disable_write_zeroes(sdkp->disk->queue);
1084}
1085
1086static void sd_config_write_same(struct scsi_disk *sdkp,
1087 struct queue_limits *lim)
1088{
1089 unsigned int logical_block_size = sdkp->device->sector_size;
1090
1091 if (sdkp->device->no_write_same) {
1092 sdkp->max_ws_blocks = 0;
1093 goto out;
1094 }
1095
1096 /* Some devices can not handle block counts above 0xffff despite
1097 * supporting WRITE SAME(16). Consequently we default to 64k
1098 * blocks per I/O unless the device explicitly advertises a
1099 * bigger limit.
1100 */
1101 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1102 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1103 (u32)SD_MAX_WS16_BLOCKS);
1104 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1105 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1106 (u32)SD_MAX_WS10_BLOCKS);
1107 else {
1108 sdkp->device->no_write_same = 1;
1109 sdkp->max_ws_blocks = 0;
1110 }
1111
1112 if (sdkp->lbprz && sdkp->lbpws)
1113 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1114 else if (sdkp->lbprz && sdkp->lbpws10)
1115 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1116 else if (sdkp->max_ws_blocks)
1117 sdkp->zeroing_mode = SD_ZERO_WS;
1118 else
1119 sdkp->zeroing_mode = SD_ZERO_WRITE;
1120
1121 if (sdkp->max_ws_blocks &&
1122 sdkp->physical_block_size > logical_block_size) {
1123 /*
1124 * Reporting a maximum number of blocks that is not aligned
1125 * on the device physical size would cause a large write same
1126 * request to be split into physically unaligned chunks by
1127 * __blkdev_issue_write_zeroes() even if the caller of this
1128 * functions took care to align the large request. So make sure
1129 * the maximum reported is aligned to the device physical block
1130 * size. This is only an optional optimization for regular
1131 * disks, but this is mandatory to avoid failure of large write
1132 * same requests directed at sequential write required zones of
1133 * host-managed ZBC disks.
1134 */
1135 sdkp->max_ws_blocks =
1136 round_down(sdkp->max_ws_blocks,
1137 bytes_to_logical(sdkp->device,
1138 sdkp->physical_block_size));
1139 }
1140
1141out:
1142 lim->max_write_zeroes_sectors =
1143 sdkp->max_ws_blocks * (logical_block_size >> SECTOR_SHIFT);
1144}
1145
1146static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1147{
1148 struct request *rq = scsi_cmd_to_rq(cmd);
1149 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1150
1151 /* flush requests don't perform I/O, zero the S/G table */
1152 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1153
1154 if (cmd->device->use_16_for_sync) {
1155 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1156 cmd->cmd_len = 16;
1157 } else {
1158 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1159 cmd->cmd_len = 10;
1160 }
1161 cmd->transfersize = 0;
1162 cmd->allowed = sdkp->max_retries;
1163
1164 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1165 return BLK_STS_OK;
1166}
1167
1168/**
1169 * sd_group_number() - Compute the GROUP NUMBER field
1170 * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1171 * field.
1172 *
1173 * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1174 * 0: no relative lifetime.
1175 * 1: shortest relative lifetime.
1176 * 2: second shortest relative lifetime.
1177 * 3 - 0x3d: intermediate relative lifetimes.
1178 * 0x3e: second longest relative lifetime.
1179 * 0x3f: longest relative lifetime.
1180 */
1181static u8 sd_group_number(struct scsi_cmnd *cmd)
1182{
1183 const struct request *rq = scsi_cmd_to_rq(cmd);
1184 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1185
1186 if (!sdkp->rscs)
1187 return 0;
1188
1189 return min3((u32)rq->bio->bi_write_hint,
1190 (u32)sdkp->permanent_stream_count, 0x3fu);
1191}
1192
1193static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1194 sector_t lba, unsigned int nr_blocks,
1195 unsigned char flags, unsigned int dld)
1196{
1197 cmd->cmd_len = SD_EXT_CDB_SIZE;
1198 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1199 cmd->cmnd[6] = sd_group_number(cmd);
1200 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1201 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1202 cmd->cmnd[10] = flags;
1203 cmd->cmnd[11] = dld & 0x07;
1204 put_unaligned_be64(lba, &cmd->cmnd[12]);
1205 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1206 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1207
1208 return BLK_STS_OK;
1209}
1210
1211static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1212 sector_t lba, unsigned int nr_blocks,
1213 unsigned char flags, unsigned int dld)
1214{
1215 cmd->cmd_len = 16;
1216 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1217 cmd->cmnd[1] = flags | ((dld >> 2) & 0x01);
1218 cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1219 cmd->cmnd[15] = 0;
1220 put_unaligned_be64(lba, &cmd->cmnd[2]);
1221 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1222
1223 return BLK_STS_OK;
1224}
1225
1226static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1227 sector_t lba, unsigned int nr_blocks,
1228 unsigned char flags)
1229{
1230 cmd->cmd_len = 10;
1231 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1232 cmd->cmnd[1] = flags;
1233 cmd->cmnd[6] = sd_group_number(cmd);
1234 cmd->cmnd[9] = 0;
1235 put_unaligned_be32(lba, &cmd->cmnd[2]);
1236 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1237
1238 return BLK_STS_OK;
1239}
1240
1241static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1242 sector_t lba, unsigned int nr_blocks,
1243 unsigned char flags)
1244{
1245 /* Avoid that 0 blocks gets translated into 256 blocks. */
1246 if (WARN_ON_ONCE(nr_blocks == 0))
1247 return BLK_STS_IOERR;
1248
1249 if (unlikely(flags & 0x8)) {
1250 /*
1251 * This happens only if this drive failed 10byte rw
1252 * command with ILLEGAL_REQUEST during operation and
1253 * thus turned off use_10_for_rw.
1254 */
1255 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1256 return BLK_STS_IOERR;
1257 }
1258
1259 cmd->cmd_len = 6;
1260 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1261 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1262 cmd->cmnd[2] = (lba >> 8) & 0xff;
1263 cmd->cmnd[3] = lba & 0xff;
1264 cmd->cmnd[4] = nr_blocks;
1265 cmd->cmnd[5] = 0;
1266
1267 return BLK_STS_OK;
1268}
1269
1270/*
1271 * Check if a command has a duration limit set. If it does, and the target
1272 * device supports CDL and the feature is enabled, return the limit
1273 * descriptor index to use. Return 0 (no limit) otherwise.
1274 */
1275static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1276{
1277 struct scsi_device *sdp = sdkp->device;
1278 int hint;
1279
1280 if (!sdp->cdl_supported || !sdp->cdl_enable)
1281 return 0;
1282
1283 /*
1284 * Use "no limit" if the request ioprio does not specify a duration
1285 * limit hint.
1286 */
1287 hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1288 if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1289 hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1290 return 0;
1291
1292 return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1293}
1294
1295static blk_status_t sd_setup_atomic_cmnd(struct scsi_cmnd *cmd,
1296 sector_t lba, unsigned int nr_blocks,
1297 bool boundary, unsigned char flags)
1298{
1299 cmd->cmd_len = 16;
1300 cmd->cmnd[0] = WRITE_ATOMIC_16;
1301 cmd->cmnd[1] = flags;
1302 put_unaligned_be64(lba, &cmd->cmnd[2]);
1303 put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1304 if (boundary)
1305 put_unaligned_be16(nr_blocks, &cmd->cmnd[10]);
1306 else
1307 put_unaligned_be16(0, &cmd->cmnd[10]);
1308 put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1309 cmd->cmnd[14] = 0;
1310 cmd->cmnd[15] = 0;
1311
1312 return BLK_STS_OK;
1313}
1314
1315static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1316{
1317 struct request *rq = scsi_cmd_to_rq(cmd);
1318 struct scsi_device *sdp = cmd->device;
1319 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1320 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1321 sector_t threshold;
1322 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1323 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1324 bool write = rq_data_dir(rq) == WRITE;
1325 unsigned char protect, fua;
1326 unsigned int dld;
1327 blk_status_t ret;
1328 unsigned int dif;
1329 bool dix;
1330
1331 ret = scsi_alloc_sgtables(cmd);
1332 if (ret != BLK_STS_OK)
1333 return ret;
1334
1335 ret = BLK_STS_IOERR;
1336 if (!scsi_device_online(sdp) || sdp->changed) {
1337 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1338 goto fail;
1339 }
1340
1341 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1342 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1343 goto fail;
1344 }
1345
1346 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1347 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1348 goto fail;
1349 }
1350
1351 /*
1352 * Some SD card readers can't handle accesses which touch the
1353 * last one or two logical blocks. Split accesses as needed.
1354 */
1355 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1356
1357 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1358 if (lba < threshold) {
1359 /* Access up to the threshold but not beyond */
1360 nr_blocks = threshold - lba;
1361 } else {
1362 /* Access only a single logical block */
1363 nr_blocks = 1;
1364 }
1365 }
1366
1367 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1368 dix = scsi_prot_sg_count(cmd);
1369 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1370 dld = sd_cdl_dld(sdkp, cmd);
1371
1372 if (dif || dix)
1373 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1374 else
1375 protect = 0;
1376
1377 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1378 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1379 protect | fua, dld);
1380 } else if (rq->cmd_flags & REQ_ATOMIC) {
1381 ret = sd_setup_atomic_cmnd(cmd, lba, nr_blocks,
1382 sdkp->use_atomic_write_boundary,
1383 protect | fua);
1384 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1385 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1386 protect | fua, dld);
1387 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1388 sdp->use_10_for_rw || protect || rq->bio->bi_write_hint) {
1389 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1390 protect | fua);
1391 } else {
1392 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1393 protect | fua);
1394 }
1395
1396 if (unlikely(ret != BLK_STS_OK))
1397 goto fail;
1398
1399 /*
1400 * We shouldn't disconnect in the middle of a sector, so with a dumb
1401 * host adapter, it's safe to assume that we can at least transfer
1402 * this many bytes between each connect / disconnect.
1403 */
1404 cmd->transfersize = sdp->sector_size;
1405 cmd->underflow = nr_blocks << 9;
1406 cmd->allowed = sdkp->max_retries;
1407 cmd->sdb.length = nr_blocks * sdp->sector_size;
1408
1409 SCSI_LOG_HLQUEUE(1,
1410 scmd_printk(KERN_INFO, cmd,
1411 "%s: block=%llu, count=%d\n", __func__,
1412 (unsigned long long)blk_rq_pos(rq),
1413 blk_rq_sectors(rq)));
1414 SCSI_LOG_HLQUEUE(2,
1415 scmd_printk(KERN_INFO, cmd,
1416 "%s %d/%u 512 byte blocks.\n",
1417 write ? "writing" : "reading", nr_blocks,
1418 blk_rq_sectors(rq)));
1419
1420 /*
1421 * This indicates that the command is ready from our end to be queued.
1422 */
1423 return BLK_STS_OK;
1424fail:
1425 scsi_free_sgtables(cmd);
1426 return ret;
1427}
1428
1429static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1430{
1431 struct request *rq = scsi_cmd_to_rq(cmd);
1432
1433 switch (req_op(rq)) {
1434 case REQ_OP_DISCARD:
1435 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1436 case SD_LBP_UNMAP:
1437 return sd_setup_unmap_cmnd(cmd);
1438 case SD_LBP_WS16:
1439 return sd_setup_write_same16_cmnd(cmd, true);
1440 case SD_LBP_WS10:
1441 return sd_setup_write_same10_cmnd(cmd, true);
1442 case SD_LBP_ZERO:
1443 return sd_setup_write_same10_cmnd(cmd, false);
1444 default:
1445 return BLK_STS_TARGET;
1446 }
1447 case REQ_OP_WRITE_ZEROES:
1448 return sd_setup_write_zeroes_cmnd(cmd);
1449 case REQ_OP_FLUSH:
1450 return sd_setup_flush_cmnd(cmd);
1451 case REQ_OP_READ:
1452 case REQ_OP_WRITE:
1453 return sd_setup_read_write_cmnd(cmd);
1454 case REQ_OP_ZONE_RESET:
1455 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1456 false);
1457 case REQ_OP_ZONE_RESET_ALL:
1458 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1459 true);
1460 case REQ_OP_ZONE_OPEN:
1461 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1462 case REQ_OP_ZONE_CLOSE:
1463 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1464 case REQ_OP_ZONE_FINISH:
1465 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1466 default:
1467 WARN_ON_ONCE(1);
1468 return BLK_STS_NOTSUPP;
1469 }
1470}
1471
1472static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1473{
1474 struct request *rq = scsi_cmd_to_rq(SCpnt);
1475
1476 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1477 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1478}
1479
1480static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1481{
1482 if (sdkp->device->removable || sdkp->write_prot) {
1483 if (disk_check_media_change(disk))
1484 return true;
1485 }
1486
1487 /*
1488 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1489 * nothing to do with partitions, BLKRRPART is used to force a full
1490 * revalidate after things like a format for historical reasons.
1491 */
1492 return test_bit(GD_NEED_PART_SCAN, &disk->state);
1493}
1494
1495/**
1496 * sd_open - open a scsi disk device
1497 * @disk: disk to open
1498 * @mode: open mode
1499 *
1500 * Returns 0 if successful. Returns a negated errno value in case
1501 * of error.
1502 *
1503 * Note: This can be called from a user context (e.g. fsck(1) )
1504 * or from within the kernel (e.g. as a result of a mount(1) ).
1505 * In the latter case @inode and @filp carry an abridged amount
1506 * of information as noted above.
1507 *
1508 * Locking: called with disk->open_mutex held.
1509 **/
1510static int sd_open(struct gendisk *disk, blk_mode_t mode)
1511{
1512 struct scsi_disk *sdkp = scsi_disk(disk);
1513 struct scsi_device *sdev = sdkp->device;
1514 int retval;
1515
1516 if (scsi_device_get(sdev))
1517 return -ENXIO;
1518
1519 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1520
1521 /*
1522 * If the device is in error recovery, wait until it is done.
1523 * If the device is offline, then disallow any access to it.
1524 */
1525 retval = -ENXIO;
1526 if (!scsi_block_when_processing_errors(sdev))
1527 goto error_out;
1528
1529 if (sd_need_revalidate(disk, sdkp))
1530 sd_revalidate_disk(disk);
1531
1532 /*
1533 * If the drive is empty, just let the open fail.
1534 */
1535 retval = -ENOMEDIUM;
1536 if (sdev->removable && !sdkp->media_present &&
1537 !(mode & BLK_OPEN_NDELAY))
1538 goto error_out;
1539
1540 /*
1541 * If the device has the write protect tab set, have the open fail
1542 * if the user expects to be able to write to the thing.
1543 */
1544 retval = -EROFS;
1545 if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1546 goto error_out;
1547
1548 /*
1549 * It is possible that the disk changing stuff resulted in
1550 * the device being taken offline. If this is the case,
1551 * report this to the user, and don't pretend that the
1552 * open actually succeeded.
1553 */
1554 retval = -ENXIO;
1555 if (!scsi_device_online(sdev))
1556 goto error_out;
1557
1558 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1559 if (scsi_block_when_processing_errors(sdev))
1560 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1561 }
1562
1563 return 0;
1564
1565error_out:
1566 scsi_device_put(sdev);
1567 return retval;
1568}
1569
1570/**
1571 * sd_release - invoked when the (last) close(2) is called on this
1572 * scsi disk.
1573 * @disk: disk to release
1574 *
1575 * Returns 0.
1576 *
1577 * Note: may block (uninterruptible) if error recovery is underway
1578 * on this disk.
1579 *
1580 * Locking: called with disk->open_mutex held.
1581 **/
1582static void sd_release(struct gendisk *disk)
1583{
1584 struct scsi_disk *sdkp = scsi_disk(disk);
1585 struct scsi_device *sdev = sdkp->device;
1586
1587 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1588
1589 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1590 if (scsi_block_when_processing_errors(sdev))
1591 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1592 }
1593
1594 scsi_device_put(sdev);
1595}
1596
1597static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1598{
1599 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1600 struct scsi_device *sdp = sdkp->device;
1601 struct Scsi_Host *host = sdp->host;
1602 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1603 int diskinfo[4];
1604
1605 /* default to most commonly used values */
1606 diskinfo[0] = 0x40; /* 1 << 6 */
1607 diskinfo[1] = 0x20; /* 1 << 5 */
1608 diskinfo[2] = capacity >> 11;
1609
1610 /* override with calculated, extended default, or driver values */
1611 if (host->hostt->bios_param)
1612 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1613 else
1614 scsicam_bios_param(bdev, capacity, diskinfo);
1615
1616 geo->heads = diskinfo[0];
1617 geo->sectors = diskinfo[1];
1618 geo->cylinders = diskinfo[2];
1619 return 0;
1620}
1621
1622/**
1623 * sd_ioctl - process an ioctl
1624 * @bdev: target block device
1625 * @mode: open mode
1626 * @cmd: ioctl command number
1627 * @arg: this is third argument given to ioctl(2) system call.
1628 * Often contains a pointer.
1629 *
1630 * Returns 0 if successful (some ioctls return positive numbers on
1631 * success as well). Returns a negated errno value in case of error.
1632 *
1633 * Note: most ioctls are forward onto the block subsystem or further
1634 * down in the scsi subsystem.
1635 **/
1636static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1637 unsigned int cmd, unsigned long arg)
1638{
1639 struct gendisk *disk = bdev->bd_disk;
1640 struct scsi_disk *sdkp = scsi_disk(disk);
1641 struct scsi_device *sdp = sdkp->device;
1642 void __user *p = (void __user *)arg;
1643 int error;
1644
1645 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1646 "cmd=0x%x\n", disk->disk_name, cmd));
1647
1648 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1649 return -ENOIOCTLCMD;
1650
1651 /*
1652 * If we are in the middle of error recovery, don't let anyone
1653 * else try and use this device. Also, if error recovery fails, it
1654 * may try and take the device offline, in which case all further
1655 * access to the device is prohibited.
1656 */
1657 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1658 (mode & BLK_OPEN_NDELAY));
1659 if (error)
1660 return error;
1661
1662 if (is_sed_ioctl(cmd))
1663 return sed_ioctl(sdkp->opal_dev, cmd, p);
1664 return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1665}
1666
1667static void set_media_not_present(struct scsi_disk *sdkp)
1668{
1669 if (sdkp->media_present)
1670 sdkp->device->changed = 1;
1671
1672 if (sdkp->device->removable) {
1673 sdkp->media_present = 0;
1674 sdkp->capacity = 0;
1675 }
1676}
1677
1678static int media_not_present(struct scsi_disk *sdkp,
1679 struct scsi_sense_hdr *sshdr)
1680{
1681 if (!scsi_sense_valid(sshdr))
1682 return 0;
1683
1684 /* not invoked for commands that could return deferred errors */
1685 switch (sshdr->sense_key) {
1686 case UNIT_ATTENTION:
1687 case NOT_READY:
1688 /* medium not present */
1689 if (sshdr->asc == 0x3A) {
1690 set_media_not_present(sdkp);
1691 return 1;
1692 }
1693 }
1694 return 0;
1695}
1696
1697/**
1698 * sd_check_events - check media events
1699 * @disk: kernel device descriptor
1700 * @clearing: disk events currently being cleared
1701 *
1702 * Returns mask of DISK_EVENT_*.
1703 *
1704 * Note: this function is invoked from the block subsystem.
1705 **/
1706static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1707{
1708 struct scsi_disk *sdkp = disk->private_data;
1709 struct scsi_device *sdp;
1710 int retval;
1711 bool disk_changed;
1712
1713 if (!sdkp)
1714 return 0;
1715
1716 sdp = sdkp->device;
1717 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1718
1719 /*
1720 * If the device is offline, don't send any commands - just pretend as
1721 * if the command failed. If the device ever comes back online, we
1722 * can deal with it then. It is only because of unrecoverable errors
1723 * that we would ever take a device offline in the first place.
1724 */
1725 if (!scsi_device_online(sdp)) {
1726 set_media_not_present(sdkp);
1727 goto out;
1728 }
1729
1730 /*
1731 * Using TEST_UNIT_READY enables differentiation between drive with
1732 * no cartridge loaded - NOT READY, drive with changed cartridge -
1733 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1734 *
1735 * Drives that auto spin down. eg iomega jaz 1G, will be started
1736 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1737 * sd_revalidate() is called.
1738 */
1739 if (scsi_block_when_processing_errors(sdp)) {
1740 struct scsi_sense_hdr sshdr = { 0, };
1741
1742 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1743 &sshdr);
1744
1745 /* failed to execute TUR, assume media not present */
1746 if (retval < 0 || host_byte(retval)) {
1747 set_media_not_present(sdkp);
1748 goto out;
1749 }
1750
1751 if (media_not_present(sdkp, &sshdr))
1752 goto out;
1753 }
1754
1755 /*
1756 * For removable scsi disk we have to recognise the presence
1757 * of a disk in the drive.
1758 */
1759 if (!sdkp->media_present)
1760 sdp->changed = 1;
1761 sdkp->media_present = 1;
1762out:
1763 /*
1764 * sdp->changed is set under the following conditions:
1765 *
1766 * Medium present state has changed in either direction.
1767 * Device has indicated UNIT_ATTENTION.
1768 */
1769 disk_changed = sdp->changed;
1770 sdp->changed = 0;
1771 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1772}
1773
1774static int sd_sync_cache(struct scsi_disk *sdkp)
1775{
1776 int res;
1777 struct scsi_device *sdp = sdkp->device;
1778 const int timeout = sdp->request_queue->rq_timeout
1779 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1780 /* Leave the rest of the command zero to indicate flush everything. */
1781 const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1782 SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1783 struct scsi_sense_hdr sshdr;
1784 struct scsi_failure failure_defs[] = {
1785 {
1786 .allowed = 3,
1787 .result = SCMD_FAILURE_RESULT_ANY,
1788 },
1789 {}
1790 };
1791 struct scsi_failures failures = {
1792 .failure_definitions = failure_defs,
1793 };
1794 const struct scsi_exec_args exec_args = {
1795 .req_flags = BLK_MQ_REQ_PM,
1796 .sshdr = &sshdr,
1797 .failures = &failures,
1798 };
1799
1800 if (!scsi_device_online(sdp))
1801 return -ENODEV;
1802
1803 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1804 sdkp->max_retries, &exec_args);
1805 if (res) {
1806 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1807
1808 if (res < 0)
1809 return res;
1810
1811 if (scsi_status_is_check_condition(res) &&
1812 scsi_sense_valid(&sshdr)) {
1813 sd_print_sense_hdr(sdkp, &sshdr);
1814
1815 /* we need to evaluate the error return */
1816 if (sshdr.asc == 0x3a || /* medium not present */
1817 sshdr.asc == 0x20 || /* invalid command */
1818 (sshdr.asc == 0x74 && sshdr.ascq == 0x71)) /* drive is password locked */
1819 /* this is no error here */
1820 return 0;
1821
1822 /*
1823 * If a format is in progress or if the drive does not
1824 * support sync, there is not much we can do because
1825 * this is called during shutdown or suspend so just
1826 * return success so those operations can proceed.
1827 */
1828 if ((sshdr.asc == 0x04 && sshdr.ascq == 0x04) ||
1829 sshdr.sense_key == ILLEGAL_REQUEST)
1830 return 0;
1831 }
1832
1833 switch (host_byte(res)) {
1834 /* ignore errors due to racing a disconnection */
1835 case DID_BAD_TARGET:
1836 case DID_NO_CONNECT:
1837 return 0;
1838 /* signal the upper layer it might try again */
1839 case DID_BUS_BUSY:
1840 case DID_IMM_RETRY:
1841 case DID_REQUEUE:
1842 case DID_SOFT_ERROR:
1843 return -EBUSY;
1844 default:
1845 return -EIO;
1846 }
1847 }
1848 return 0;
1849}
1850
1851static void sd_rescan(struct device *dev)
1852{
1853 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1854
1855 sd_revalidate_disk(sdkp->disk);
1856}
1857
1858static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1859 enum blk_unique_id type)
1860{
1861 struct scsi_device *sdev = scsi_disk(disk)->device;
1862 const struct scsi_vpd *vpd;
1863 const unsigned char *d;
1864 int ret = -ENXIO, len;
1865
1866 rcu_read_lock();
1867 vpd = rcu_dereference(sdev->vpd_pg83);
1868 if (!vpd)
1869 goto out_unlock;
1870
1871 ret = -EINVAL;
1872 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1873 /* we only care about designators with LU association */
1874 if (((d[1] >> 4) & 0x3) != 0x00)
1875 continue;
1876 if ((d[1] & 0xf) != type)
1877 continue;
1878
1879 /*
1880 * Only exit early if a 16-byte descriptor was found. Otherwise
1881 * keep looking as one with more entropy might still show up.
1882 */
1883 len = d[3];
1884 if (len != 8 && len != 12 && len != 16)
1885 continue;
1886 ret = len;
1887 memcpy(id, d + 4, len);
1888 if (len == 16)
1889 break;
1890 }
1891out_unlock:
1892 rcu_read_unlock();
1893 return ret;
1894}
1895
1896static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1897{
1898 switch (host_byte(result)) {
1899 case DID_TRANSPORT_MARGINAL:
1900 case DID_TRANSPORT_DISRUPTED:
1901 case DID_BUS_BUSY:
1902 return PR_STS_RETRY_PATH_FAILURE;
1903 case DID_NO_CONNECT:
1904 return PR_STS_PATH_FAILED;
1905 case DID_TRANSPORT_FAILFAST:
1906 return PR_STS_PATH_FAST_FAILED;
1907 }
1908
1909 switch (status_byte(result)) {
1910 case SAM_STAT_RESERVATION_CONFLICT:
1911 return PR_STS_RESERVATION_CONFLICT;
1912 case SAM_STAT_CHECK_CONDITION:
1913 if (!scsi_sense_valid(sshdr))
1914 return PR_STS_IOERR;
1915
1916 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1917 (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1918 return -EINVAL;
1919
1920 fallthrough;
1921 default:
1922 return PR_STS_IOERR;
1923 }
1924}
1925
1926static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1927 unsigned char *data, int data_len)
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 u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1933 struct scsi_failure failure_defs[] = {
1934 {
1935 .sense = UNIT_ATTENTION,
1936 .asc = SCMD_FAILURE_ASC_ANY,
1937 .ascq = SCMD_FAILURE_ASCQ_ANY,
1938 .allowed = 5,
1939 .result = SAM_STAT_CHECK_CONDITION,
1940 },
1941 {}
1942 };
1943 struct scsi_failures failures = {
1944 .failure_definitions = failure_defs,
1945 };
1946 const struct scsi_exec_args exec_args = {
1947 .sshdr = &sshdr,
1948 .failures = &failures,
1949 };
1950 int result;
1951
1952 put_unaligned_be16(data_len, &cmd[7]);
1953
1954 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1955 SD_TIMEOUT, sdkp->max_retries, &exec_args);
1956 if (scsi_status_is_check_condition(result) &&
1957 scsi_sense_valid(&sshdr)) {
1958 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1959 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1960 }
1961
1962 if (result <= 0)
1963 return result;
1964
1965 return sd_scsi_to_pr_err(&sshdr, result);
1966}
1967
1968static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1969{
1970 int result, i, data_offset, num_copy_keys;
1971 u32 num_keys = keys_info->num_keys;
1972 int data_len = num_keys * 8 + 8;
1973 u8 *data;
1974
1975 data = kzalloc(data_len, GFP_KERNEL);
1976 if (!data)
1977 return -ENOMEM;
1978
1979 result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1980 if (result)
1981 goto free_data;
1982
1983 keys_info->generation = get_unaligned_be32(&data[0]);
1984 keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1985
1986 data_offset = 8;
1987 num_copy_keys = min(num_keys, keys_info->num_keys);
1988
1989 for (i = 0; i < num_copy_keys; i++) {
1990 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1991 data_offset += 8;
1992 }
1993
1994free_data:
1995 kfree(data);
1996 return result;
1997}
1998
1999static int sd_pr_read_reservation(struct block_device *bdev,
2000 struct pr_held_reservation *rsv)
2001{
2002 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2003 struct scsi_device *sdev = sdkp->device;
2004 u8 data[24] = { };
2005 int result, len;
2006
2007 result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
2008 if (result)
2009 return result;
2010
2011 len = get_unaligned_be32(&data[4]);
2012 if (!len)
2013 return 0;
2014
2015 /* Make sure we have at least the key and type */
2016 if (len < 14) {
2017 sdev_printk(KERN_INFO, sdev,
2018 "READ RESERVATION failed due to short return buffer of %d bytes\n",
2019 len);
2020 return -EINVAL;
2021 }
2022
2023 rsv->generation = get_unaligned_be32(&data[0]);
2024 rsv->key = get_unaligned_be64(&data[8]);
2025 rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
2026 return 0;
2027}
2028
2029static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
2030 u64 sa_key, enum scsi_pr_type type, u8 flags)
2031{
2032 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2033 struct scsi_device *sdev = sdkp->device;
2034 struct scsi_sense_hdr sshdr;
2035 struct scsi_failure failure_defs[] = {
2036 {
2037 .sense = UNIT_ATTENTION,
2038 .asc = SCMD_FAILURE_ASC_ANY,
2039 .ascq = SCMD_FAILURE_ASCQ_ANY,
2040 .allowed = 5,
2041 .result = SAM_STAT_CHECK_CONDITION,
2042 },
2043 {}
2044 };
2045 struct scsi_failures failures = {
2046 .failure_definitions = failure_defs,
2047 };
2048 const struct scsi_exec_args exec_args = {
2049 .sshdr = &sshdr,
2050 .failures = &failures,
2051 };
2052 int result;
2053 u8 cmd[16] = { 0, };
2054 u8 data[24] = { 0, };
2055
2056 cmd[0] = PERSISTENT_RESERVE_OUT;
2057 cmd[1] = sa;
2058 cmd[2] = type;
2059 put_unaligned_be32(sizeof(data), &cmd[5]);
2060
2061 put_unaligned_be64(key, &data[0]);
2062 put_unaligned_be64(sa_key, &data[8]);
2063 data[20] = flags;
2064
2065 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
2066 sizeof(data), SD_TIMEOUT, sdkp->max_retries,
2067 &exec_args);
2068
2069 if (scsi_status_is_check_condition(result) &&
2070 scsi_sense_valid(&sshdr)) {
2071 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
2072 scsi_print_sense_hdr(sdev, NULL, &sshdr);
2073 }
2074
2075 if (result <= 0)
2076 return result;
2077
2078 return sd_scsi_to_pr_err(&sshdr, result);
2079}
2080
2081static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2082 u32 flags)
2083{
2084 if (flags & ~PR_FL_IGNORE_KEY)
2085 return -EOPNOTSUPP;
2086 return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
2087 old_key, new_key, 0,
2088 (1 << 0) /* APTPL */);
2089}
2090
2091static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2092 u32 flags)
2093{
2094 if (flags)
2095 return -EOPNOTSUPP;
2096 return sd_pr_out_command(bdev, 0x01, key, 0,
2097 block_pr_type_to_scsi(type), 0);
2098}
2099
2100static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2101{
2102 return sd_pr_out_command(bdev, 0x02, key, 0,
2103 block_pr_type_to_scsi(type), 0);
2104}
2105
2106static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2107 enum pr_type type, bool abort)
2108{
2109 return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2110 block_pr_type_to_scsi(type), 0);
2111}
2112
2113static int sd_pr_clear(struct block_device *bdev, u64 key)
2114{
2115 return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2116}
2117
2118static const struct pr_ops sd_pr_ops = {
2119 .pr_register = sd_pr_register,
2120 .pr_reserve = sd_pr_reserve,
2121 .pr_release = sd_pr_release,
2122 .pr_preempt = sd_pr_preempt,
2123 .pr_clear = sd_pr_clear,
2124 .pr_read_keys = sd_pr_read_keys,
2125 .pr_read_reservation = sd_pr_read_reservation,
2126};
2127
2128static void scsi_disk_free_disk(struct gendisk *disk)
2129{
2130 struct scsi_disk *sdkp = scsi_disk(disk);
2131
2132 put_device(&sdkp->disk_dev);
2133}
2134
2135static const struct block_device_operations sd_fops = {
2136 .owner = THIS_MODULE,
2137 .open = sd_open,
2138 .release = sd_release,
2139 .ioctl = sd_ioctl,
2140 .getgeo = sd_getgeo,
2141 .compat_ioctl = blkdev_compat_ptr_ioctl,
2142 .check_events = sd_check_events,
2143 .unlock_native_capacity = sd_unlock_native_capacity,
2144 .report_zones = sd_zbc_report_zones,
2145 .get_unique_id = sd_get_unique_id,
2146 .free_disk = scsi_disk_free_disk,
2147 .pr_ops = &sd_pr_ops,
2148};
2149
2150/**
2151 * sd_eh_reset - reset error handling callback
2152 * @scmd: sd-issued command that has failed
2153 *
2154 * This function is called by the SCSI midlayer before starting
2155 * SCSI EH. When counting medium access failures we have to be
2156 * careful to register it only only once per device and SCSI EH run;
2157 * there might be several timed out commands which will cause the
2158 * 'max_medium_access_timeouts' counter to trigger after the first
2159 * SCSI EH run already and set the device to offline.
2160 * So this function resets the internal counter before starting SCSI EH.
2161 **/
2162static void sd_eh_reset(struct scsi_cmnd *scmd)
2163{
2164 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2165
2166 /* New SCSI EH run, reset gate variable */
2167 sdkp->ignore_medium_access_errors = false;
2168}
2169
2170/**
2171 * sd_eh_action - error handling callback
2172 * @scmd: sd-issued command that has failed
2173 * @eh_disp: The recovery disposition suggested by the midlayer
2174 *
2175 * This function is called by the SCSI midlayer upon completion of an
2176 * error test command (currently TEST UNIT READY). The result of sending
2177 * the eh command is passed in eh_disp. We're looking for devices that
2178 * fail medium access commands but are OK with non access commands like
2179 * test unit ready (so wrongly see the device as having a successful
2180 * recovery)
2181 **/
2182static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2183{
2184 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2185 struct scsi_device *sdev = scmd->device;
2186
2187 if (!scsi_device_online(sdev) ||
2188 !scsi_medium_access_command(scmd) ||
2189 host_byte(scmd->result) != DID_TIME_OUT ||
2190 eh_disp != SUCCESS)
2191 return eh_disp;
2192
2193 /*
2194 * The device has timed out executing a medium access command.
2195 * However, the TEST UNIT READY command sent during error
2196 * handling completed successfully. Either the device is in the
2197 * process of recovering or has it suffered an internal failure
2198 * that prevents access to the storage medium.
2199 */
2200 if (!sdkp->ignore_medium_access_errors) {
2201 sdkp->medium_access_timed_out++;
2202 sdkp->ignore_medium_access_errors = true;
2203 }
2204
2205 /*
2206 * If the device keeps failing read/write commands but TEST UNIT
2207 * READY always completes successfully we assume that medium
2208 * access is no longer possible and take the device offline.
2209 */
2210 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2211 scmd_printk(KERN_ERR, scmd,
2212 "Medium access timeout failure. Offlining disk!\n");
2213 mutex_lock(&sdev->state_mutex);
2214 scsi_device_set_state(sdev, SDEV_OFFLINE);
2215 mutex_unlock(&sdev->state_mutex);
2216
2217 return SUCCESS;
2218 }
2219
2220 return eh_disp;
2221}
2222
2223static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2224{
2225 struct request *req = scsi_cmd_to_rq(scmd);
2226 struct scsi_device *sdev = scmd->device;
2227 unsigned int transferred, good_bytes;
2228 u64 start_lba, end_lba, bad_lba;
2229
2230 /*
2231 * Some commands have a payload smaller than the device logical
2232 * block size (e.g. INQUIRY on a 4K disk).
2233 */
2234 if (scsi_bufflen(scmd) <= sdev->sector_size)
2235 return 0;
2236
2237 /* Check if we have a 'bad_lba' information */
2238 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2239 SCSI_SENSE_BUFFERSIZE,
2240 &bad_lba))
2241 return 0;
2242
2243 /*
2244 * If the bad lba was reported incorrectly, we have no idea where
2245 * the error is.
2246 */
2247 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2248 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2249 if (bad_lba < start_lba || bad_lba >= end_lba)
2250 return 0;
2251
2252 /*
2253 * resid is optional but mostly filled in. When it's unused,
2254 * its value is zero, so we assume the whole buffer transferred
2255 */
2256 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2257
2258 /* This computation should always be done in terms of the
2259 * resolution of the device's medium.
2260 */
2261 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2262
2263 return min(good_bytes, transferred);
2264}
2265
2266/**
2267 * sd_done - bottom half handler: called when the lower level
2268 * driver has completed (successfully or otherwise) a scsi command.
2269 * @SCpnt: mid-level's per command structure.
2270 *
2271 * Note: potentially run from within an ISR. Must not block.
2272 **/
2273static int sd_done(struct scsi_cmnd *SCpnt)
2274{
2275 int result = SCpnt->result;
2276 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2277 unsigned int sector_size = SCpnt->device->sector_size;
2278 unsigned int resid;
2279 struct scsi_sense_hdr sshdr;
2280 struct request *req = scsi_cmd_to_rq(SCpnt);
2281 struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2282 int sense_valid = 0;
2283 int sense_deferred = 0;
2284
2285 switch (req_op(req)) {
2286 case REQ_OP_DISCARD:
2287 case REQ_OP_WRITE_ZEROES:
2288 case REQ_OP_ZONE_RESET:
2289 case REQ_OP_ZONE_RESET_ALL:
2290 case REQ_OP_ZONE_OPEN:
2291 case REQ_OP_ZONE_CLOSE:
2292 case REQ_OP_ZONE_FINISH:
2293 if (!result) {
2294 good_bytes = blk_rq_bytes(req);
2295 scsi_set_resid(SCpnt, 0);
2296 } else {
2297 good_bytes = 0;
2298 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2299 }
2300 break;
2301 default:
2302 /*
2303 * In case of bogus fw or device, we could end up having
2304 * an unaligned partial completion. Check this here and force
2305 * alignment.
2306 */
2307 resid = scsi_get_resid(SCpnt);
2308 if (resid & (sector_size - 1)) {
2309 sd_printk(KERN_INFO, sdkp,
2310 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2311 resid, sector_size);
2312 scsi_print_command(SCpnt);
2313 resid = min(scsi_bufflen(SCpnt),
2314 round_up(resid, sector_size));
2315 scsi_set_resid(SCpnt, resid);
2316 }
2317 }
2318
2319 if (result) {
2320 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2321 if (sense_valid)
2322 sense_deferred = scsi_sense_is_deferred(&sshdr);
2323 }
2324 sdkp->medium_access_timed_out = 0;
2325
2326 if (!scsi_status_is_check_condition(result) &&
2327 (!sense_valid || sense_deferred))
2328 goto out;
2329
2330 switch (sshdr.sense_key) {
2331 case HARDWARE_ERROR:
2332 case MEDIUM_ERROR:
2333 good_bytes = sd_completed_bytes(SCpnt);
2334 break;
2335 case RECOVERED_ERROR:
2336 good_bytes = scsi_bufflen(SCpnt);
2337 break;
2338 case NO_SENSE:
2339 /* This indicates a false check condition, so ignore it. An
2340 * unknown amount of data was transferred so treat it as an
2341 * error.
2342 */
2343 SCpnt->result = 0;
2344 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2345 break;
2346 case ABORTED_COMMAND:
2347 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2348 good_bytes = sd_completed_bytes(SCpnt);
2349 break;
2350 case ILLEGAL_REQUEST:
2351 switch (sshdr.asc) {
2352 case 0x10: /* DIX: Host detected corruption */
2353 good_bytes = sd_completed_bytes(SCpnt);
2354 break;
2355 case 0x20: /* INVALID COMMAND OPCODE */
2356 case 0x24: /* INVALID FIELD IN CDB */
2357 switch (SCpnt->cmnd[0]) {
2358 case UNMAP:
2359 sd_disable_discard(sdkp);
2360 break;
2361 case WRITE_SAME_16:
2362 case WRITE_SAME:
2363 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2364 sd_disable_discard(sdkp);
2365 } else {
2366 sd_disable_write_same(sdkp);
2367 req->rq_flags |= RQF_QUIET;
2368 }
2369 break;
2370 }
2371 }
2372 break;
2373 default:
2374 break;
2375 }
2376
2377 out:
2378 if (sdkp->device->type == TYPE_ZBC)
2379 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2380
2381 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2382 "sd_done: completed %d of %d bytes\n",
2383 good_bytes, scsi_bufflen(SCpnt)));
2384
2385 return good_bytes;
2386}
2387
2388/*
2389 * spinup disk - called only in sd_revalidate_disk()
2390 */
2391static void
2392sd_spinup_disk(struct scsi_disk *sdkp)
2393{
2394 static const u8 cmd[10] = { TEST_UNIT_READY };
2395 unsigned long spintime_expire = 0;
2396 int spintime, sense_valid = 0;
2397 unsigned int the_result;
2398 struct scsi_sense_hdr sshdr;
2399 struct scsi_failure failure_defs[] = {
2400 /* Do not retry Medium Not Present */
2401 {
2402 .sense = UNIT_ATTENTION,
2403 .asc = 0x3A,
2404 .ascq = SCMD_FAILURE_ASCQ_ANY,
2405 .result = SAM_STAT_CHECK_CONDITION,
2406 },
2407 {
2408 .sense = NOT_READY,
2409 .asc = 0x3A,
2410 .ascq = SCMD_FAILURE_ASCQ_ANY,
2411 .result = SAM_STAT_CHECK_CONDITION,
2412 },
2413 /* Retry when scsi_status_is_good would return false 3 times */
2414 {
2415 .result = SCMD_FAILURE_STAT_ANY,
2416 .allowed = 3,
2417 },
2418 {}
2419 };
2420 struct scsi_failures failures = {
2421 .failure_definitions = failure_defs,
2422 };
2423 const struct scsi_exec_args exec_args = {
2424 .sshdr = &sshdr,
2425 .failures = &failures,
2426 };
2427
2428 spintime = 0;
2429
2430 /* Spin up drives, as required. Only do this at boot time */
2431 /* Spinup needs to be done for module loads too. */
2432 do {
2433 bool media_was_present = sdkp->media_present;
2434
2435 scsi_failures_reset_retries(&failures);
2436
2437 the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2438 NULL, 0, SD_TIMEOUT,
2439 sdkp->max_retries, &exec_args);
2440
2441
2442 if (the_result > 0) {
2443 /*
2444 * If the drive has indicated to us that it doesn't
2445 * have any media in it, don't bother with any more
2446 * polling.
2447 */
2448 if (media_not_present(sdkp, &sshdr)) {
2449 if (media_was_present)
2450 sd_printk(KERN_NOTICE, sdkp,
2451 "Media removed, stopped polling\n");
2452 return;
2453 }
2454 sense_valid = scsi_sense_valid(&sshdr);
2455 }
2456
2457 if (!scsi_status_is_check_condition(the_result)) {
2458 /* no sense, TUR either succeeded or failed
2459 * with a status error */
2460 if(!spintime && !scsi_status_is_good(the_result)) {
2461 sd_print_result(sdkp, "Test Unit Ready failed",
2462 the_result);
2463 }
2464 break;
2465 }
2466
2467 /*
2468 * The device does not want the automatic start to be issued.
2469 */
2470 if (sdkp->device->no_start_on_add)
2471 break;
2472
2473 if (sense_valid && sshdr.sense_key == NOT_READY) {
2474 if (sshdr.asc == 4 && sshdr.ascq == 3)
2475 break; /* manual intervention required */
2476 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2477 break; /* standby */
2478 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2479 break; /* unavailable */
2480 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2481 break; /* sanitize in progress */
2482 if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2483 break; /* depopulation in progress */
2484 if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2485 break; /* depopulation restoration in progress */
2486 /*
2487 * Issue command to spin up drive when not ready
2488 */
2489 if (!spintime) {
2490 /* Return immediately and start spin cycle */
2491 const u8 start_cmd[10] = {
2492 [0] = START_STOP,
2493 [1] = 1,
2494 [4] = sdkp->device->start_stop_pwr_cond ?
2495 0x11 : 1,
2496 };
2497
2498 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2499 scsi_execute_cmd(sdkp->device, start_cmd,
2500 REQ_OP_DRV_IN, NULL, 0,
2501 SD_TIMEOUT, sdkp->max_retries,
2502 &exec_args);
2503 spintime_expire = jiffies + 100 * HZ;
2504 spintime = 1;
2505 }
2506 /* Wait 1 second for next try */
2507 msleep(1000);
2508 printk(KERN_CONT ".");
2509
2510 /*
2511 * Wait for USB flash devices with slow firmware.
2512 * Yes, this sense key/ASC combination shouldn't
2513 * occur here. It's characteristic of these devices.
2514 */
2515 } else if (sense_valid &&
2516 sshdr.sense_key == UNIT_ATTENTION &&
2517 sshdr.asc == 0x28) {
2518 if (!spintime) {
2519 spintime_expire = jiffies + 5 * HZ;
2520 spintime = 1;
2521 }
2522 /* Wait 1 second for next try */
2523 msleep(1000);
2524 } else {
2525 /* we don't understand the sense code, so it's
2526 * probably pointless to loop */
2527 if(!spintime) {
2528 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2529 sd_print_sense_hdr(sdkp, &sshdr);
2530 }
2531 break;
2532 }
2533
2534 } while (spintime && time_before_eq(jiffies, spintime_expire));
2535
2536 if (spintime) {
2537 if (scsi_status_is_good(the_result))
2538 printk(KERN_CONT "ready\n");
2539 else
2540 printk(KERN_CONT "not responding...\n");
2541 }
2542}
2543
2544/*
2545 * Determine whether disk supports Data Integrity Field.
2546 */
2547static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2548{
2549 struct scsi_device *sdp = sdkp->device;
2550 u8 type;
2551
2552 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2553 sdkp->protection_type = 0;
2554 return 0;
2555 }
2556
2557 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2558
2559 if (type > T10_PI_TYPE3_PROTECTION) {
2560 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2561 " protection type %u. Disabling disk!\n",
2562 type);
2563 sdkp->protection_type = 0;
2564 return -ENODEV;
2565 }
2566
2567 sdkp->protection_type = type;
2568
2569 return 0;
2570}
2571
2572static void sd_config_protection(struct scsi_disk *sdkp,
2573 struct queue_limits *lim)
2574{
2575 struct scsi_device *sdp = sdkp->device;
2576
2577 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
2578 sd_dif_config_host(sdkp, lim);
2579
2580 if (!sdkp->protection_type)
2581 return;
2582
2583 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2584 sd_first_printk(KERN_NOTICE, sdkp,
2585 "Disabling DIF Type %u protection\n",
2586 sdkp->protection_type);
2587 sdkp->protection_type = 0;
2588 }
2589
2590 sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2591 sdkp->protection_type);
2592}
2593
2594static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2595 struct scsi_sense_hdr *sshdr, int sense_valid,
2596 int the_result)
2597{
2598 if (sense_valid)
2599 sd_print_sense_hdr(sdkp, sshdr);
2600 else
2601 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2602
2603 /*
2604 * Set dirty bit for removable devices if not ready -
2605 * sometimes drives will not report this properly.
2606 */
2607 if (sdp->removable &&
2608 sense_valid && sshdr->sense_key == NOT_READY)
2609 set_media_not_present(sdkp);
2610
2611 /*
2612 * We used to set media_present to 0 here to indicate no media
2613 * in the drive, but some drives fail read capacity even with
2614 * media present, so we can't do that.
2615 */
2616 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2617}
2618
2619#define RC16_LEN 32
2620#if RC16_LEN > SD_BUF_SIZE
2621#error RC16_LEN must not be more than SD_BUF_SIZE
2622#endif
2623
2624#define READ_CAPACITY_RETRIES_ON_RESET 10
2625
2626static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2627 struct queue_limits *lim, unsigned char *buffer)
2628{
2629 unsigned char cmd[16];
2630 struct scsi_sense_hdr sshdr;
2631 const struct scsi_exec_args exec_args = {
2632 .sshdr = &sshdr,
2633 };
2634 int sense_valid = 0;
2635 int the_result;
2636 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2637 unsigned int alignment;
2638 unsigned long long lba;
2639 unsigned sector_size;
2640
2641 if (sdp->no_read_capacity_16)
2642 return -EINVAL;
2643
2644 do {
2645 memset(cmd, 0, 16);
2646 cmd[0] = SERVICE_ACTION_IN_16;
2647 cmd[1] = SAI_READ_CAPACITY_16;
2648 cmd[13] = RC16_LEN;
2649 memset(buffer, 0, RC16_LEN);
2650
2651 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2652 buffer, RC16_LEN, SD_TIMEOUT,
2653 sdkp->max_retries, &exec_args);
2654 if (the_result > 0) {
2655 if (media_not_present(sdkp, &sshdr))
2656 return -ENODEV;
2657
2658 sense_valid = scsi_sense_valid(&sshdr);
2659 if (sense_valid &&
2660 sshdr.sense_key == ILLEGAL_REQUEST &&
2661 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2662 sshdr.ascq == 0x00)
2663 /* Invalid Command Operation Code or
2664 * Invalid Field in CDB, just retry
2665 * silently with RC10 */
2666 return -EINVAL;
2667 if (sense_valid &&
2668 sshdr.sense_key == UNIT_ATTENTION &&
2669 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2670 /* Device reset might occur several times,
2671 * give it one more chance */
2672 if (--reset_retries > 0)
2673 continue;
2674 }
2675 retries--;
2676
2677 } while (the_result && retries);
2678
2679 if (the_result) {
2680 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2681 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2682 return -EINVAL;
2683 }
2684
2685 sector_size = get_unaligned_be32(&buffer[8]);
2686 lba = get_unaligned_be64(&buffer[0]);
2687
2688 if (sd_read_protection_type(sdkp, buffer) < 0) {
2689 sdkp->capacity = 0;
2690 return -ENODEV;
2691 }
2692
2693 /* Logical blocks per physical block exponent */
2694 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2695
2696 /* RC basis */
2697 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2698
2699 /* Lowest aligned logical block */
2700 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2701 lim->alignment_offset = alignment;
2702 if (alignment && sdkp->first_scan)
2703 sd_printk(KERN_NOTICE, sdkp,
2704 "physical block alignment offset: %u\n", alignment);
2705
2706 if (buffer[14] & 0x80) { /* LBPME */
2707 sdkp->lbpme = 1;
2708
2709 if (buffer[14] & 0x40) /* LBPRZ */
2710 sdkp->lbprz = 1;
2711 }
2712
2713 sdkp->capacity = lba + 1;
2714 return sector_size;
2715}
2716
2717static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2718 unsigned char *buffer)
2719{
2720 static const u8 cmd[10] = { READ_CAPACITY };
2721 struct scsi_sense_hdr sshdr;
2722 struct scsi_failure failure_defs[] = {
2723 /* Do not retry Medium Not Present */
2724 {
2725 .sense = UNIT_ATTENTION,
2726 .asc = 0x3A,
2727 .result = SAM_STAT_CHECK_CONDITION,
2728 },
2729 {
2730 .sense = NOT_READY,
2731 .asc = 0x3A,
2732 .result = SAM_STAT_CHECK_CONDITION,
2733 },
2734 /* Device reset might occur several times so retry a lot */
2735 {
2736 .sense = UNIT_ATTENTION,
2737 .asc = 0x29,
2738 .allowed = READ_CAPACITY_RETRIES_ON_RESET,
2739 .result = SAM_STAT_CHECK_CONDITION,
2740 },
2741 /* Any other error not listed above retry 3 times */
2742 {
2743 .result = SCMD_FAILURE_RESULT_ANY,
2744 .allowed = 3,
2745 },
2746 {}
2747 };
2748 struct scsi_failures failures = {
2749 .failure_definitions = failure_defs,
2750 };
2751 const struct scsi_exec_args exec_args = {
2752 .sshdr = &sshdr,
2753 .failures = &failures,
2754 };
2755 int sense_valid = 0;
2756 int the_result;
2757 sector_t lba;
2758 unsigned sector_size;
2759
2760 memset(buffer, 0, 8);
2761
2762 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2763 8, SD_TIMEOUT, sdkp->max_retries,
2764 &exec_args);
2765
2766 if (the_result > 0) {
2767 sense_valid = scsi_sense_valid(&sshdr);
2768
2769 if (media_not_present(sdkp, &sshdr))
2770 return -ENODEV;
2771 }
2772
2773 if (the_result) {
2774 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2775 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2776 return -EINVAL;
2777 }
2778
2779 sector_size = get_unaligned_be32(&buffer[4]);
2780 lba = get_unaligned_be32(&buffer[0]);
2781
2782 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2783 /* Some buggy (usb cardreader) devices return an lba of
2784 0xffffffff when the want to report a size of 0 (with
2785 which they really mean no media is present) */
2786 sdkp->capacity = 0;
2787 sdkp->physical_block_size = sector_size;
2788 return sector_size;
2789 }
2790
2791 sdkp->capacity = lba + 1;
2792 sdkp->physical_block_size = sector_size;
2793 return sector_size;
2794}
2795
2796static int sd_try_rc16_first(struct scsi_device *sdp)
2797{
2798 if (sdp->host->max_cmd_len < 16)
2799 return 0;
2800 if (sdp->try_rc_10_first)
2801 return 0;
2802 if (sdp->scsi_level > SCSI_SPC_2)
2803 return 1;
2804 if (scsi_device_protection(sdp))
2805 return 1;
2806 return 0;
2807}
2808
2809/*
2810 * read disk capacity
2811 */
2812static void
2813sd_read_capacity(struct scsi_disk *sdkp, struct queue_limits *lim,
2814 unsigned char *buffer)
2815{
2816 int sector_size;
2817 struct scsi_device *sdp = sdkp->device;
2818
2819 if (sd_try_rc16_first(sdp)) {
2820 sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2821 if (sector_size == -EOVERFLOW)
2822 goto got_data;
2823 if (sector_size == -ENODEV)
2824 return;
2825 if (sector_size < 0)
2826 sector_size = read_capacity_10(sdkp, sdp, buffer);
2827 if (sector_size < 0)
2828 return;
2829 } else {
2830 sector_size = read_capacity_10(sdkp, sdp, buffer);
2831 if (sector_size == -EOVERFLOW)
2832 goto got_data;
2833 if (sector_size < 0)
2834 return;
2835 if ((sizeof(sdkp->capacity) > 4) &&
2836 (sdkp->capacity > 0xffffffffULL)) {
2837 int old_sector_size = sector_size;
2838 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2839 "Trying to use READ CAPACITY(16).\n");
2840 sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2841 if (sector_size < 0) {
2842 sd_printk(KERN_NOTICE, sdkp,
2843 "Using 0xffffffff as device size\n");
2844 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2845 sector_size = old_sector_size;
2846 goto got_data;
2847 }
2848 /* Remember that READ CAPACITY(16) succeeded */
2849 sdp->try_rc_10_first = 0;
2850 }
2851 }
2852
2853 /* Some devices are known to return the total number of blocks,
2854 * not the highest block number. Some devices have versions
2855 * which do this and others which do not. Some devices we might
2856 * suspect of doing this but we don't know for certain.
2857 *
2858 * If we know the reported capacity is wrong, decrement it. If
2859 * we can only guess, then assume the number of blocks is even
2860 * (usually true but not always) and err on the side of lowering
2861 * the capacity.
2862 */
2863 if (sdp->fix_capacity ||
2864 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2865 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2866 "from its reported value: %llu\n",
2867 (unsigned long long) sdkp->capacity);
2868 --sdkp->capacity;
2869 }
2870
2871got_data:
2872 if (sector_size == 0) {
2873 sector_size = 512;
2874 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2875 "assuming 512.\n");
2876 }
2877
2878 if (sector_size != 512 &&
2879 sector_size != 1024 &&
2880 sector_size != 2048 &&
2881 sector_size != 4096) {
2882 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2883 sector_size);
2884 /*
2885 * The user might want to re-format the drive with
2886 * a supported sectorsize. Once this happens, it
2887 * would be relatively trivial to set the thing up.
2888 * For this reason, we leave the thing in the table.
2889 */
2890 sdkp->capacity = 0;
2891 /*
2892 * set a bogus sector size so the normal read/write
2893 * logic in the block layer will eventually refuse any
2894 * request on this device without tripping over power
2895 * of two sector size assumptions
2896 */
2897 sector_size = 512;
2898 }
2899 lim->logical_block_size = sector_size;
2900 lim->physical_block_size = sdkp->physical_block_size;
2901 sdkp->device->sector_size = sector_size;
2902
2903 if (sdkp->capacity > 0xffffffff)
2904 sdp->use_16_for_rw = 1;
2905
2906}
2907
2908/*
2909 * Print disk capacity
2910 */
2911static void
2912sd_print_capacity(struct scsi_disk *sdkp,
2913 sector_t old_capacity)
2914{
2915 int sector_size = sdkp->device->sector_size;
2916 char cap_str_2[10], cap_str_10[10];
2917
2918 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2919 return;
2920
2921 string_get_size(sdkp->capacity, sector_size,
2922 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2923 string_get_size(sdkp->capacity, sector_size,
2924 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2925
2926 sd_printk(KERN_NOTICE, sdkp,
2927 "%llu %d-byte logical blocks: (%s/%s)\n",
2928 (unsigned long long)sdkp->capacity,
2929 sector_size, cap_str_10, cap_str_2);
2930
2931 if (sdkp->physical_block_size != sector_size)
2932 sd_printk(KERN_NOTICE, sdkp,
2933 "%u-byte physical blocks\n",
2934 sdkp->physical_block_size);
2935}
2936
2937/* called with buffer of length 512 */
2938static inline int
2939sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2940 unsigned char *buffer, int len, struct scsi_mode_data *data,
2941 struct scsi_sense_hdr *sshdr)
2942{
2943 /*
2944 * If we must use MODE SENSE(10), make sure that the buffer length
2945 * is at least 8 bytes so that the mode sense header fits.
2946 */
2947 if (sdkp->device->use_10_for_ms && len < 8)
2948 len = 8;
2949
2950 return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2951 SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2952}
2953
2954/*
2955 * read write protect setting, if possible - called only in sd_revalidate_disk()
2956 * called with buffer of length SD_BUF_SIZE
2957 */
2958static void
2959sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2960{
2961 int res;
2962 struct scsi_device *sdp = sdkp->device;
2963 struct scsi_mode_data data;
2964 int old_wp = sdkp->write_prot;
2965
2966 set_disk_ro(sdkp->disk, 0);
2967 if (sdp->skip_ms_page_3f) {
2968 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2969 return;
2970 }
2971
2972 if (sdp->use_192_bytes_for_3f) {
2973 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2974 } else {
2975 /*
2976 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2977 * We have to start carefully: some devices hang if we ask
2978 * for more than is available.
2979 */
2980 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2981
2982 /*
2983 * Second attempt: ask for page 0 When only page 0 is
2984 * implemented, a request for page 3F may return Sense Key
2985 * 5: Illegal Request, Sense Code 24: Invalid field in
2986 * CDB.
2987 */
2988 if (res < 0)
2989 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2990
2991 /*
2992 * Third attempt: ask 255 bytes, as we did earlier.
2993 */
2994 if (res < 0)
2995 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2996 &data, NULL);
2997 }
2998
2999 if (res < 0) {
3000 sd_first_printk(KERN_WARNING, sdkp,
3001 "Test WP failed, assume Write Enabled\n");
3002 } else {
3003 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
3004 set_disk_ro(sdkp->disk, sdkp->write_prot);
3005 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
3006 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
3007 sdkp->write_prot ? "on" : "off");
3008 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
3009 }
3010 }
3011}
3012
3013/*
3014 * sd_read_cache_type - called only from sd_revalidate_disk()
3015 * called with buffer of length SD_BUF_SIZE
3016 */
3017static void
3018sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
3019{
3020 int len = 0, res;
3021 struct scsi_device *sdp = sdkp->device;
3022
3023 int dbd;
3024 int modepage;
3025 int first_len;
3026 struct scsi_mode_data data;
3027 struct scsi_sense_hdr sshdr;
3028 int old_wce = sdkp->WCE;
3029 int old_rcd = sdkp->RCD;
3030 int old_dpofua = sdkp->DPOFUA;
3031
3032
3033 if (sdkp->cache_override)
3034 return;
3035
3036 first_len = 4;
3037 if (sdp->skip_ms_page_8) {
3038 if (sdp->type == TYPE_RBC)
3039 goto defaults;
3040 else {
3041 if (sdp->skip_ms_page_3f)
3042 goto defaults;
3043 modepage = 0x3F;
3044 if (sdp->use_192_bytes_for_3f)
3045 first_len = 192;
3046 dbd = 0;
3047 }
3048 } else if (sdp->type == TYPE_RBC) {
3049 modepage = 6;
3050 dbd = 8;
3051 } else {
3052 modepage = 8;
3053 dbd = 0;
3054 }
3055
3056 /* cautiously ask */
3057 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
3058 &data, &sshdr);
3059
3060 if (res < 0)
3061 goto bad_sense;
3062
3063 if (!data.header_length) {
3064 modepage = 6;
3065 first_len = 0;
3066 sd_first_printk(KERN_ERR, sdkp,
3067 "Missing header in MODE_SENSE response\n");
3068 }
3069
3070 /* that went OK, now ask for the proper length */
3071 len = data.length;
3072
3073 /*
3074 * We're only interested in the first three bytes, actually.
3075 * But the data cache page is defined for the first 20.
3076 */
3077 if (len < 3)
3078 goto bad_sense;
3079 else if (len > SD_BUF_SIZE) {
3080 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
3081 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
3082 len = SD_BUF_SIZE;
3083 }
3084 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
3085 len = 192;
3086
3087 /* Get the data */
3088 if (len > first_len)
3089 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
3090 &data, &sshdr);
3091
3092 if (!res) {
3093 int offset = data.header_length + data.block_descriptor_length;
3094
3095 while (offset < len) {
3096 u8 page_code = buffer[offset] & 0x3F;
3097 u8 spf = buffer[offset] & 0x40;
3098
3099 if (page_code == 8 || page_code == 6) {
3100 /* We're interested only in the first 3 bytes.
3101 */
3102 if (len - offset <= 2) {
3103 sd_first_printk(KERN_ERR, sdkp,
3104 "Incomplete mode parameter "
3105 "data\n");
3106 goto defaults;
3107 } else {
3108 modepage = page_code;
3109 goto Page_found;
3110 }
3111 } else {
3112 /* Go to the next page */
3113 if (spf && len - offset > 3)
3114 offset += 4 + (buffer[offset+2] << 8) +
3115 buffer[offset+3];
3116 else if (!spf && len - offset > 1)
3117 offset += 2 + buffer[offset+1];
3118 else {
3119 sd_first_printk(KERN_ERR, sdkp,
3120 "Incomplete mode "
3121 "parameter data\n");
3122 goto defaults;
3123 }
3124 }
3125 }
3126
3127 sd_first_printk(KERN_WARNING, sdkp,
3128 "No Caching mode page found\n");
3129 goto defaults;
3130
3131 Page_found:
3132 if (modepage == 8) {
3133 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3134 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3135 } else {
3136 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3137 sdkp->RCD = 0;
3138 }
3139
3140 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3141 if (sdp->broken_fua) {
3142 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3143 sdkp->DPOFUA = 0;
3144 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3145 !sdkp->device->use_16_for_rw) {
3146 sd_first_printk(KERN_NOTICE, sdkp,
3147 "Uses READ/WRITE(6), disabling FUA\n");
3148 sdkp->DPOFUA = 0;
3149 }
3150
3151 /* No cache flush allowed for write protected devices */
3152 if (sdkp->WCE && sdkp->write_prot)
3153 sdkp->WCE = 0;
3154
3155 if (sdkp->first_scan || old_wce != sdkp->WCE ||
3156 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3157 sd_printk(KERN_NOTICE, sdkp,
3158 "Write cache: %s, read cache: %s, %s\n",
3159 sdkp->WCE ? "enabled" : "disabled",
3160 sdkp->RCD ? "disabled" : "enabled",
3161 sdkp->DPOFUA ? "supports DPO and FUA"
3162 : "doesn't support DPO or FUA");
3163
3164 return;
3165 }
3166
3167bad_sense:
3168 if (res == -EIO && scsi_sense_valid(&sshdr) &&
3169 sshdr.sense_key == ILLEGAL_REQUEST &&
3170 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3171 /* Invalid field in CDB */
3172 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3173 else
3174 sd_first_printk(KERN_ERR, sdkp,
3175 "Asking for cache data failed\n");
3176
3177defaults:
3178 if (sdp->wce_default_on) {
3179 sd_first_printk(KERN_NOTICE, sdkp,
3180 "Assuming drive cache: write back\n");
3181 sdkp->WCE = 1;
3182 } else {
3183 sd_first_printk(KERN_WARNING, sdkp,
3184 "Assuming drive cache: write through\n");
3185 sdkp->WCE = 0;
3186 }
3187 sdkp->RCD = 0;
3188 sdkp->DPOFUA = 0;
3189}
3190
3191static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3192{
3193 u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3194 struct {
3195 struct scsi_stream_status_header h;
3196 struct scsi_stream_status s;
3197 } buf;
3198 struct scsi_device *sdev = sdkp->device;
3199 struct scsi_sense_hdr sshdr;
3200 const struct scsi_exec_args exec_args = {
3201 .sshdr = &sshdr,
3202 };
3203 int res;
3204
3205 put_unaligned_be16(stream_id, &cdb[4]);
3206 put_unaligned_be32(sizeof(buf), &cdb[10]);
3207
3208 res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3209 SD_TIMEOUT, sdkp->max_retries, &exec_args);
3210 if (res < 0)
3211 return false;
3212 if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3213 sd_print_sense_hdr(sdkp, &sshdr);
3214 if (res)
3215 return false;
3216 if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3217 return false;
3218 return buf.s.perm;
3219}
3220
3221static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3222{
3223 struct scsi_device *sdp = sdkp->device;
3224 const struct scsi_io_group_descriptor *desc, *start, *end;
3225 u16 permanent_stream_count_old;
3226 struct scsi_sense_hdr sshdr;
3227 struct scsi_mode_data data;
3228 int res;
3229
3230 if (sdp->sdev_bflags & BLIST_SKIP_IO_HINTS)
3231 return;
3232
3233 res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3234 /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3235 sdkp->max_retries, &data, &sshdr);
3236 if (res < 0)
3237 return;
3238 start = (void *)buffer + data.header_length + 16;
3239 end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3240 sizeof(*end));
3241 /*
3242 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3243 * should assign the lowest numbered stream identifiers to permanent
3244 * streams.
3245 */
3246 for (desc = start; desc < end; desc++)
3247 if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3248 break;
3249 permanent_stream_count_old = sdkp->permanent_stream_count;
3250 sdkp->permanent_stream_count = desc - start;
3251 if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3252 sd_printk(KERN_INFO, sdkp,
3253 "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3254 sdkp->permanent_stream_count);
3255 else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3256 sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3257 sdkp->permanent_stream_count);
3258}
3259
3260/*
3261 * The ATO bit indicates whether the DIF application tag is available
3262 * for use by the operating system.
3263 */
3264static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3265{
3266 int res, offset;
3267 struct scsi_device *sdp = sdkp->device;
3268 struct scsi_mode_data data;
3269 struct scsi_sense_hdr sshdr;
3270
3271 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3272 return;
3273
3274 if (sdkp->protection_type == 0)
3275 return;
3276
3277 res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3278 sdkp->max_retries, &data, &sshdr);
3279
3280 if (res < 0 || !data.header_length ||
3281 data.length < 6) {
3282 sd_first_printk(KERN_WARNING, sdkp,
3283 "getting Control mode page failed, assume no ATO\n");
3284
3285 if (res == -EIO && scsi_sense_valid(&sshdr))
3286 sd_print_sense_hdr(sdkp, &sshdr);
3287
3288 return;
3289 }
3290
3291 offset = data.header_length + data.block_descriptor_length;
3292
3293 if ((buffer[offset] & 0x3f) != 0x0a) {
3294 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3295 return;
3296 }
3297
3298 if ((buffer[offset + 5] & 0x80) == 0)
3299 return;
3300
3301 sdkp->ATO = 1;
3302
3303 return;
3304}
3305
3306static unsigned int sd_discard_mode(struct scsi_disk *sdkp)
3307{
3308 if (!sdkp->lbpme)
3309 return SD_LBP_FULL;
3310
3311 if (!sdkp->lbpvpd) {
3312 /* LBP VPD page not provided */
3313 if (sdkp->max_unmap_blocks)
3314 return SD_LBP_UNMAP;
3315 return SD_LBP_WS16;
3316 }
3317
3318 /* LBP VPD page tells us what to use */
3319 if (sdkp->lbpu && sdkp->max_unmap_blocks)
3320 return SD_LBP_UNMAP;
3321 if (sdkp->lbpws)
3322 return SD_LBP_WS16;
3323 if (sdkp->lbpws10)
3324 return SD_LBP_WS10;
3325 return SD_LBP_DISABLE;
3326}
3327
3328/*
3329 * Query disk device for preferred I/O sizes.
3330 */
3331static void sd_read_block_limits(struct scsi_disk *sdkp,
3332 struct queue_limits *lim)
3333{
3334 struct scsi_vpd *vpd;
3335
3336 rcu_read_lock();
3337
3338 vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3339 if (!vpd || vpd->len < 16)
3340 goto out;
3341
3342 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3343 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3344 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3345
3346 if (vpd->len >= 64) {
3347 unsigned int lba_count, desc_count;
3348
3349 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3350
3351 if (!sdkp->lbpme)
3352 goto config_atomic;
3353
3354 lba_count = get_unaligned_be32(&vpd->data[20]);
3355 desc_count = get_unaligned_be32(&vpd->data[24]);
3356
3357 if (lba_count && desc_count)
3358 sdkp->max_unmap_blocks = lba_count;
3359
3360 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3361
3362 if (vpd->data[32] & 0x80)
3363 sdkp->unmap_alignment =
3364 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3365
3366config_atomic:
3367 sdkp->max_atomic = get_unaligned_be32(&vpd->data[44]);
3368 sdkp->atomic_alignment = get_unaligned_be32(&vpd->data[48]);
3369 sdkp->atomic_granularity = get_unaligned_be32(&vpd->data[52]);
3370 sdkp->max_atomic_with_boundary = get_unaligned_be32(&vpd->data[56]);
3371 sdkp->max_atomic_boundary = get_unaligned_be32(&vpd->data[60]);
3372
3373 sd_config_atomic(sdkp, lim);
3374 }
3375
3376 out:
3377 rcu_read_unlock();
3378}
3379
3380/* Parse the Block Limits Extension VPD page (0xb7) */
3381static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3382{
3383 struct scsi_vpd *vpd;
3384
3385 rcu_read_lock();
3386 vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3387 if (vpd && vpd->len >= 6)
3388 sdkp->rscs = vpd->data[5] & 1;
3389 rcu_read_unlock();
3390}
3391
3392/* Query block device characteristics */
3393static void sd_read_block_characteristics(struct scsi_disk *sdkp,
3394 struct queue_limits *lim)
3395{
3396 struct scsi_vpd *vpd;
3397 u16 rot;
3398
3399 rcu_read_lock();
3400 vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3401
3402 if (!vpd || vpd->len <= 8) {
3403 rcu_read_unlock();
3404 return;
3405 }
3406
3407 rot = get_unaligned_be16(&vpd->data[4]);
3408 sdkp->zoned = (vpd->data[8] >> 4) & 3;
3409 rcu_read_unlock();
3410
3411 if (rot == 1)
3412 lim->features &= ~(BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3413
3414 if (!sdkp->first_scan)
3415 return;
3416
3417 if (sdkp->device->type == TYPE_ZBC)
3418 sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3419 else if (sdkp->zoned == 1)
3420 sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3421 else if (sdkp->zoned == 2)
3422 sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3423}
3424
3425/**
3426 * sd_read_block_provisioning - Query provisioning VPD page
3427 * @sdkp: disk to query
3428 */
3429static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3430{
3431 struct scsi_vpd *vpd;
3432
3433 if (sdkp->lbpme == 0)
3434 return;
3435
3436 rcu_read_lock();
3437 vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3438
3439 if (!vpd || vpd->len < 8) {
3440 rcu_read_unlock();
3441 return;
3442 }
3443
3444 sdkp->lbpvpd = 1;
3445 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */
3446 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3447 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3448 rcu_read_unlock();
3449}
3450
3451static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3452{
3453 struct scsi_device *sdev = sdkp->device;
3454
3455 if (sdev->host->no_write_same) {
3456 sdev->no_write_same = 1;
3457
3458 return;
3459 }
3460
3461 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3462 struct scsi_vpd *vpd;
3463
3464 sdev->no_report_opcodes = 1;
3465
3466 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3467 * CODES is unsupported and the device has an ATA
3468 * Information VPD page (SAT).
3469 */
3470 rcu_read_lock();
3471 vpd = rcu_dereference(sdev->vpd_pg89);
3472 if (vpd)
3473 sdev->no_write_same = 1;
3474 rcu_read_unlock();
3475 }
3476
3477 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3478 sdkp->ws16 = 1;
3479
3480 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3481 sdkp->ws10 = 1;
3482}
3483
3484static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3485{
3486 struct scsi_device *sdev = sdkp->device;
3487
3488 if (!sdev->security_supported)
3489 return;
3490
3491 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3492 SECURITY_PROTOCOL_IN, 0) == 1 &&
3493 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3494 SECURITY_PROTOCOL_OUT, 0) == 1)
3495 sdkp->security = 1;
3496}
3497
3498static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3499{
3500 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3501}
3502
3503/**
3504 * sd_read_cpr - Query concurrent positioning ranges
3505 * @sdkp: disk to query
3506 */
3507static void sd_read_cpr(struct scsi_disk *sdkp)
3508{
3509 struct blk_independent_access_ranges *iars = NULL;
3510 unsigned char *buffer = NULL;
3511 unsigned int nr_cpr = 0;
3512 int i, vpd_len, buf_len = SD_BUF_SIZE;
3513 u8 *desc;
3514
3515 /*
3516 * We need to have the capacity set first for the block layer to be
3517 * able to check the ranges.
3518 */
3519 if (sdkp->first_scan)
3520 return;
3521
3522 if (!sdkp->capacity)
3523 goto out;
3524
3525 /*
3526 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3527 * leading to a maximum page size of 64 + 256*32 bytes.
3528 */
3529 buf_len = 64 + 256*32;
3530 buffer = kmalloc(buf_len, GFP_KERNEL);
3531 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3532 goto out;
3533
3534 /* We must have at least a 64B header and one 32B range descriptor */
3535 vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3536 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3537 sd_printk(KERN_ERR, sdkp,
3538 "Invalid Concurrent Positioning Ranges VPD page\n");
3539 goto out;
3540 }
3541
3542 nr_cpr = (vpd_len - 64) / 32;
3543 if (nr_cpr == 1) {
3544 nr_cpr = 0;
3545 goto out;
3546 }
3547
3548 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3549 if (!iars) {
3550 nr_cpr = 0;
3551 goto out;
3552 }
3553
3554 desc = &buffer[64];
3555 for (i = 0; i < nr_cpr; i++, desc += 32) {
3556 if (desc[0] != i) {
3557 sd_printk(KERN_ERR, sdkp,
3558 "Invalid Concurrent Positioning Range number\n");
3559 nr_cpr = 0;
3560 break;
3561 }
3562
3563 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3564 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3565 }
3566
3567out:
3568 disk_set_independent_access_ranges(sdkp->disk, iars);
3569 if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3570 sd_printk(KERN_NOTICE, sdkp,
3571 "%u concurrent positioning ranges\n", nr_cpr);
3572 sdkp->nr_actuators = nr_cpr;
3573 }
3574
3575 kfree(buffer);
3576}
3577
3578static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3579{
3580 struct scsi_device *sdp = sdkp->device;
3581 unsigned int min_xfer_bytes =
3582 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3583
3584 if (sdkp->min_xfer_blocks == 0)
3585 return false;
3586
3587 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3588 sd_first_printk(KERN_WARNING, sdkp,
3589 "Preferred minimum I/O size %u bytes not a " \
3590 "multiple of physical block size (%u bytes)\n",
3591 min_xfer_bytes, sdkp->physical_block_size);
3592 sdkp->min_xfer_blocks = 0;
3593 return false;
3594 }
3595
3596 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3597 min_xfer_bytes);
3598 return true;
3599}
3600
3601/*
3602 * Determine the device's preferred I/O size for reads and writes
3603 * unless the reported value is unreasonably small, large, not a
3604 * multiple of the physical block size, or simply garbage.
3605 */
3606static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3607 unsigned int dev_max)
3608{
3609 struct scsi_device *sdp = sdkp->device;
3610 unsigned int opt_xfer_bytes =
3611 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3612 unsigned int min_xfer_bytes =
3613 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3614
3615 if (sdkp->opt_xfer_blocks == 0)
3616 return false;
3617
3618 if (sdkp->opt_xfer_blocks > dev_max) {
3619 sd_first_printk(KERN_WARNING, sdkp,
3620 "Optimal transfer size %u logical blocks " \
3621 "> dev_max (%u logical blocks)\n",
3622 sdkp->opt_xfer_blocks, dev_max);
3623 return false;
3624 }
3625
3626 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3627 sd_first_printk(KERN_WARNING, sdkp,
3628 "Optimal transfer size %u logical blocks " \
3629 "> sd driver limit (%u logical blocks)\n",
3630 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3631 return false;
3632 }
3633
3634 if (opt_xfer_bytes < PAGE_SIZE) {
3635 sd_first_printk(KERN_WARNING, sdkp,
3636 "Optimal transfer size %u bytes < " \
3637 "PAGE_SIZE (%u bytes)\n",
3638 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3639 return false;
3640 }
3641
3642 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3643 sd_first_printk(KERN_WARNING, sdkp,
3644 "Optimal transfer size %u bytes not a " \
3645 "multiple of preferred minimum block " \
3646 "size (%u bytes)\n",
3647 opt_xfer_bytes, min_xfer_bytes);
3648 return false;
3649 }
3650
3651 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3652 sd_first_printk(KERN_WARNING, sdkp,
3653 "Optimal transfer size %u bytes not a " \
3654 "multiple of physical block size (%u bytes)\n",
3655 opt_xfer_bytes, sdkp->physical_block_size);
3656 return false;
3657 }
3658
3659 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3660 opt_xfer_bytes);
3661 return true;
3662}
3663
3664static void sd_read_block_zero(struct scsi_disk *sdkp)
3665{
3666 struct scsi_device *sdev = sdkp->device;
3667 unsigned int buf_len = sdev->sector_size;
3668 u8 *buffer, cmd[16] = { };
3669
3670 buffer = kmalloc(buf_len, GFP_KERNEL);
3671 if (!buffer)
3672 return;
3673
3674 if (sdev->use_16_for_rw) {
3675 cmd[0] = READ_16;
3676 put_unaligned_be64(0, &cmd[2]); /* Logical block address 0 */
3677 put_unaligned_be32(1, &cmd[10]);/* Transfer 1 logical block */
3678 } else {
3679 cmd[0] = READ_10;
3680 put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3681 put_unaligned_be16(1, &cmd[7]); /* Transfer 1 logical block */
3682 }
3683
3684 scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3685 SD_TIMEOUT, sdkp->max_retries, NULL);
3686 kfree(buffer);
3687}
3688
3689/**
3690 * sd_revalidate_disk - called the first time a new disk is seen,
3691 * performs disk spin up, read_capacity, etc.
3692 * @disk: struct gendisk we care about
3693 **/
3694static int sd_revalidate_disk(struct gendisk *disk)
3695{
3696 struct scsi_disk *sdkp = scsi_disk(disk);
3697 struct scsi_device *sdp = sdkp->device;
3698 sector_t old_capacity = sdkp->capacity;
3699 struct queue_limits lim;
3700 unsigned char *buffer;
3701 unsigned int dev_max;
3702 int err;
3703
3704 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3705 "sd_revalidate_disk\n"));
3706
3707 /*
3708 * If the device is offline, don't try and read capacity or any
3709 * of the other niceties.
3710 */
3711 if (!scsi_device_online(sdp))
3712 goto out;
3713
3714 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3715 if (!buffer) {
3716 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3717 "allocation failure.\n");
3718 goto out;
3719 }
3720
3721 sd_spinup_disk(sdkp);
3722
3723 lim = queue_limits_start_update(sdkp->disk->queue);
3724
3725 /*
3726 * Without media there is no reason to ask; moreover, some devices
3727 * react badly if we do.
3728 */
3729 if (sdkp->media_present) {
3730 sd_read_capacity(sdkp, &lim, buffer);
3731 /*
3732 * Some USB/UAS devices return generic values for mode pages
3733 * until the media has been accessed. Trigger a READ operation
3734 * to force the device to populate mode pages.
3735 */
3736 if (sdp->read_before_ms)
3737 sd_read_block_zero(sdkp);
3738 /*
3739 * set the default to rotational. All non-rotational devices
3740 * support the block characteristics VPD page, which will
3741 * cause this to be updated correctly and any device which
3742 * doesn't support it should be treated as rotational.
3743 */
3744 lim.features |= (BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3745
3746 if (scsi_device_supports_vpd(sdp)) {
3747 sd_read_block_provisioning(sdkp);
3748 sd_read_block_limits(sdkp, &lim);
3749 sd_read_block_limits_ext(sdkp);
3750 sd_read_block_characteristics(sdkp, &lim);
3751 sd_zbc_read_zones(sdkp, &lim, buffer);
3752 }
3753
3754 sd_config_discard(sdkp, &lim, sd_discard_mode(sdkp));
3755
3756 sd_print_capacity(sdkp, old_capacity);
3757
3758 sd_read_write_protect_flag(sdkp, buffer);
3759 sd_read_cache_type(sdkp, buffer);
3760 sd_read_io_hints(sdkp, buffer);
3761 sd_read_app_tag_own(sdkp, buffer);
3762 sd_read_write_same(sdkp, buffer);
3763 sd_read_security(sdkp, buffer);
3764 sd_config_protection(sdkp, &lim);
3765 }
3766
3767 /*
3768 * We now have all cache related info, determine how we deal
3769 * with flush requests.
3770 */
3771 sd_set_flush_flag(sdkp, &lim);
3772
3773 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3774 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3775
3776 /* Some devices report a maximum block count for READ/WRITE requests. */
3777 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3778 lim.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3779
3780 if (sd_validate_min_xfer_size(sdkp))
3781 lim.io_min = logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3782 else
3783 lim.io_min = 0;
3784
3785 /*
3786 * Limit default to SCSI host optimal sector limit if set. There may be
3787 * an impact on performance for when the size of a request exceeds this
3788 * host limit.
3789 */
3790 lim.io_opt = sdp->host->opt_sectors << SECTOR_SHIFT;
3791 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3792 lim.io_opt = min_not_zero(lim.io_opt,
3793 logical_to_bytes(sdp, sdkp->opt_xfer_blocks));
3794 }
3795
3796 sdkp->first_scan = 0;
3797
3798 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3799 sd_config_write_same(sdkp, &lim);
3800 kfree(buffer);
3801
3802 err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
3803 if (err)
3804 return err;
3805
3806 /*
3807 * Query concurrent positioning ranges after
3808 * queue_limits_commit_update() unlocked q->limits_lock to avoid
3809 * deadlock with q->sysfs_dir_lock and q->sysfs_lock.
3810 */
3811 if (sdkp->media_present && scsi_device_supports_vpd(sdp))
3812 sd_read_cpr(sdkp);
3813
3814 /*
3815 * For a zoned drive, revalidating the zones can be done only once
3816 * the gendisk capacity is set. So if this fails, set back the gendisk
3817 * capacity to 0.
3818 */
3819 if (sd_zbc_revalidate_zones(sdkp))
3820 set_capacity_and_notify(disk, 0);
3821
3822 out:
3823 return 0;
3824}
3825
3826/**
3827 * sd_unlock_native_capacity - unlock native capacity
3828 * @disk: struct gendisk to set capacity for
3829 *
3830 * Block layer calls this function if it detects that partitions
3831 * on @disk reach beyond the end of the device. If the SCSI host
3832 * implements ->unlock_native_capacity() method, it's invoked to
3833 * give it a chance to adjust the device capacity.
3834 *
3835 * CONTEXT:
3836 * Defined by block layer. Might sleep.
3837 */
3838static void sd_unlock_native_capacity(struct gendisk *disk)
3839{
3840 struct scsi_device *sdev = scsi_disk(disk)->device;
3841
3842 if (sdev->host->hostt->unlock_native_capacity)
3843 sdev->host->hostt->unlock_native_capacity(sdev);
3844}
3845
3846/**
3847 * sd_format_disk_name - format disk name
3848 * @prefix: name prefix - ie. "sd" for SCSI disks
3849 * @index: index of the disk to format name for
3850 * @buf: output buffer
3851 * @buflen: length of the output buffer
3852 *
3853 * SCSI disk names starts at sda. The 26th device is sdz and the
3854 * 27th is sdaa. The last one for two lettered suffix is sdzz
3855 * which is followed by sdaaa.
3856 *
3857 * This is basically 26 base counting with one extra 'nil' entry
3858 * at the beginning from the second digit on and can be
3859 * determined using similar method as 26 base conversion with the
3860 * index shifted -1 after each digit is computed.
3861 *
3862 * CONTEXT:
3863 * Don't care.
3864 *
3865 * RETURNS:
3866 * 0 on success, -errno on failure.
3867 */
3868static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3869{
3870 const int base = 'z' - 'a' + 1;
3871 char *begin = buf + strlen(prefix);
3872 char *end = buf + buflen;
3873 char *p;
3874 int unit;
3875
3876 p = end - 1;
3877 *p = '\0';
3878 unit = base;
3879 do {
3880 if (p == begin)
3881 return -EINVAL;
3882 *--p = 'a' + (index % unit);
3883 index = (index / unit) - 1;
3884 } while (index >= 0);
3885
3886 memmove(begin, p, end - p);
3887 memcpy(buf, prefix, strlen(prefix));
3888
3889 return 0;
3890}
3891
3892/**
3893 * sd_probe - called during driver initialization and whenever a
3894 * new scsi device is attached to the system. It is called once
3895 * for each scsi device (not just disks) present.
3896 * @dev: pointer to device object
3897 *
3898 * Returns 0 if successful (or not interested in this scsi device
3899 * (e.g. scanner)); 1 when there is an error.
3900 *
3901 * Note: this function is invoked from the scsi mid-level.
3902 * This function sets up the mapping between a given
3903 * <host,channel,id,lun> (found in sdp) and new device name
3904 * (e.g. /dev/sda). More precisely it is the block device major
3905 * and minor number that is chosen here.
3906 *
3907 * Assume sd_probe is not re-entrant (for time being)
3908 * Also think about sd_probe() and sd_remove() running coincidentally.
3909 **/
3910static int sd_probe(struct device *dev)
3911{
3912 struct scsi_device *sdp = to_scsi_device(dev);
3913 struct scsi_disk *sdkp;
3914 struct gendisk *gd;
3915 int index;
3916 int error;
3917
3918 scsi_autopm_get_device(sdp);
3919 error = -ENODEV;
3920 if (sdp->type != TYPE_DISK &&
3921 sdp->type != TYPE_ZBC &&
3922 sdp->type != TYPE_MOD &&
3923 sdp->type != TYPE_RBC)
3924 goto out;
3925
3926 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3927 sdev_printk(KERN_WARNING, sdp,
3928 "Unsupported ZBC host-managed device.\n");
3929 goto out;
3930 }
3931
3932 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3933 "sd_probe\n"));
3934
3935 error = -ENOMEM;
3936 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3937 if (!sdkp)
3938 goto out;
3939
3940 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3941 &sd_bio_compl_lkclass);
3942 if (!gd)
3943 goto out_free;
3944
3945 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3946 if (index < 0) {
3947 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3948 goto out_put;
3949 }
3950
3951 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3952 if (error) {
3953 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3954 goto out_free_index;
3955 }
3956
3957 sdkp->device = sdp;
3958 sdkp->disk = gd;
3959 sdkp->index = index;
3960 sdkp->max_retries = SD_MAX_RETRIES;
3961 atomic_set(&sdkp->openers, 0);
3962 atomic_set(&sdkp->device->ioerr_cnt, 0);
3963
3964 if (!sdp->request_queue->rq_timeout) {
3965 if (sdp->type != TYPE_MOD)
3966 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3967 else
3968 blk_queue_rq_timeout(sdp->request_queue,
3969 SD_MOD_TIMEOUT);
3970 }
3971
3972 device_initialize(&sdkp->disk_dev);
3973 sdkp->disk_dev.parent = get_device(dev);
3974 sdkp->disk_dev.class = &sd_disk_class;
3975 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3976
3977 error = device_add(&sdkp->disk_dev);
3978 if (error) {
3979 put_device(&sdkp->disk_dev);
3980 goto out;
3981 }
3982
3983 dev_set_drvdata(dev, sdkp);
3984
3985 gd->major = sd_major((index & 0xf0) >> 4);
3986 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3987 gd->minors = SD_MINORS;
3988
3989 gd->fops = &sd_fops;
3990 gd->private_data = sdkp;
3991
3992 /* defaults, until the device tells us otherwise */
3993 sdp->sector_size = 512;
3994 sdkp->capacity = 0;
3995 sdkp->media_present = 1;
3996 sdkp->write_prot = 0;
3997 sdkp->cache_override = 0;
3998 sdkp->WCE = 0;
3999 sdkp->RCD = 0;
4000 sdkp->ATO = 0;
4001 sdkp->first_scan = 1;
4002 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
4003
4004 sd_revalidate_disk(gd);
4005
4006 if (sdp->removable) {
4007 gd->flags |= GENHD_FL_REMOVABLE;
4008 gd->events |= DISK_EVENT_MEDIA_CHANGE;
4009 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
4010 }
4011
4012 blk_pm_runtime_init(sdp->request_queue, dev);
4013 if (sdp->rpm_autosuspend) {
4014 pm_runtime_set_autosuspend_delay(dev,
4015 sdp->host->rpm_autosuspend_delay);
4016 }
4017
4018 error = device_add_disk(dev, gd, NULL);
4019 if (error) {
4020 device_unregister(&sdkp->disk_dev);
4021 put_disk(gd);
4022 goto out;
4023 }
4024
4025 if (sdkp->security) {
4026 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
4027 if (sdkp->opal_dev)
4028 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
4029 }
4030
4031 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
4032 sdp->removable ? "removable " : "");
4033 scsi_autopm_put_device(sdp);
4034
4035 return 0;
4036
4037 out_free_index:
4038 ida_free(&sd_index_ida, index);
4039 out_put:
4040 put_disk(gd);
4041 out_free:
4042 kfree(sdkp);
4043 out:
4044 scsi_autopm_put_device(sdp);
4045 return error;
4046}
4047
4048/**
4049 * sd_remove - called whenever a scsi disk (previously recognized by
4050 * sd_probe) is detached from the system. It is called (potentially
4051 * multiple times) during sd module unload.
4052 * @dev: pointer to device object
4053 *
4054 * Note: this function is invoked from the scsi mid-level.
4055 * This function potentially frees up a device name (e.g. /dev/sdc)
4056 * that could be re-used by a subsequent sd_probe().
4057 * This function is not called when the built-in sd driver is "exit-ed".
4058 **/
4059static int sd_remove(struct device *dev)
4060{
4061 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4062
4063 scsi_autopm_get_device(sdkp->device);
4064
4065 device_del(&sdkp->disk_dev);
4066 del_gendisk(sdkp->disk);
4067 if (!sdkp->suspended)
4068 sd_shutdown(dev);
4069
4070 put_disk(sdkp->disk);
4071 return 0;
4072}
4073
4074static void scsi_disk_release(struct device *dev)
4075{
4076 struct scsi_disk *sdkp = to_scsi_disk(dev);
4077
4078 ida_free(&sd_index_ida, sdkp->index);
4079 put_device(&sdkp->device->sdev_gendev);
4080 free_opal_dev(sdkp->opal_dev);
4081
4082 kfree(sdkp);
4083}
4084
4085static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
4086{
4087 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
4088 struct scsi_sense_hdr sshdr;
4089 struct scsi_failure failure_defs[] = {
4090 {
4091 /* Power on, reset, or bus device reset occurred */
4092 .sense = UNIT_ATTENTION,
4093 .asc = 0x29,
4094 .ascq = 0,
4095 .result = SAM_STAT_CHECK_CONDITION,
4096 },
4097 {
4098 /* Power on occurred */
4099 .sense = UNIT_ATTENTION,
4100 .asc = 0x29,
4101 .ascq = 1,
4102 .result = SAM_STAT_CHECK_CONDITION,
4103 },
4104 {
4105 /* SCSI bus reset */
4106 .sense = UNIT_ATTENTION,
4107 .asc = 0x29,
4108 .ascq = 2,
4109 .result = SAM_STAT_CHECK_CONDITION,
4110 },
4111 {}
4112 };
4113 struct scsi_failures failures = {
4114 .total_allowed = 3,
4115 .failure_definitions = failure_defs,
4116 };
4117 const struct scsi_exec_args exec_args = {
4118 .sshdr = &sshdr,
4119 .req_flags = BLK_MQ_REQ_PM,
4120 .failures = &failures,
4121 };
4122 struct scsi_device *sdp = sdkp->device;
4123 int res;
4124
4125 if (start)
4126 cmd[4] |= 1; /* START */
4127
4128 if (sdp->start_stop_pwr_cond)
4129 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
4130
4131 if (!scsi_device_online(sdp))
4132 return -ENODEV;
4133
4134 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4135 sdkp->max_retries, &exec_args);
4136 if (res) {
4137 sd_print_result(sdkp, "Start/Stop Unit failed", res);
4138 if (res > 0 && scsi_sense_valid(&sshdr)) {
4139 sd_print_sense_hdr(sdkp, &sshdr);
4140 /* 0x3a is medium not present */
4141 if (sshdr.asc == 0x3a)
4142 res = 0;
4143 }
4144 }
4145
4146 /* SCSI error codes must not go to the generic layer */
4147 if (res)
4148 return -EIO;
4149
4150 return 0;
4151}
4152
4153/*
4154 * Send a SYNCHRONIZE CACHE instruction down to the device through
4155 * the normal SCSI command structure. Wait for the command to
4156 * complete.
4157 */
4158static void sd_shutdown(struct device *dev)
4159{
4160 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4161
4162 if (!sdkp)
4163 return; /* this can happen */
4164
4165 if (pm_runtime_suspended(dev))
4166 return;
4167
4168 if (sdkp->WCE && sdkp->media_present) {
4169 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4170 sd_sync_cache(sdkp);
4171 }
4172
4173 if ((system_state != SYSTEM_RESTART &&
4174 sdkp->device->manage_system_start_stop) ||
4175 (system_state == SYSTEM_POWER_OFF &&
4176 sdkp->device->manage_shutdown)) {
4177 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4178 sd_start_stop_device(sdkp, 0);
4179 }
4180}
4181
4182static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4183{
4184 return (sdev->manage_system_start_stop && !runtime) ||
4185 (sdev->manage_runtime_start_stop && runtime);
4186}
4187
4188static int sd_suspend_common(struct device *dev, bool runtime)
4189{
4190 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4191 int ret = 0;
4192
4193 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
4194 return 0;
4195
4196 if (sdkp->WCE && sdkp->media_present) {
4197 if (!sdkp->device->silence_suspend)
4198 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4199 ret = sd_sync_cache(sdkp);
4200 /* ignore OFFLINE device */
4201 if (ret == -ENODEV)
4202 return 0;
4203
4204 if (ret)
4205 return ret;
4206 }
4207
4208 if (sd_do_start_stop(sdkp->device, runtime)) {
4209 if (!sdkp->device->silence_suspend)
4210 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4211 /* an error is not worth aborting a system sleep */
4212 ret = sd_start_stop_device(sdkp, 0);
4213 if (!runtime)
4214 ret = 0;
4215 }
4216
4217 if (!ret)
4218 sdkp->suspended = true;
4219
4220 return ret;
4221}
4222
4223static int sd_suspend_system(struct device *dev)
4224{
4225 if (pm_runtime_suspended(dev))
4226 return 0;
4227
4228 return sd_suspend_common(dev, false);
4229}
4230
4231static int sd_suspend_runtime(struct device *dev)
4232{
4233 return sd_suspend_common(dev, true);
4234}
4235
4236static int sd_resume(struct device *dev)
4237{
4238 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4239
4240 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4241
4242 if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4243 sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4244 return -EIO;
4245 }
4246
4247 return 0;
4248}
4249
4250static int sd_resume_common(struct device *dev, bool runtime)
4251{
4252 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4253 int ret;
4254
4255 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4256 return 0;
4257
4258 if (!sd_do_start_stop(sdkp->device, runtime)) {
4259 sdkp->suspended = false;
4260 return 0;
4261 }
4262
4263 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4264 ret = sd_start_stop_device(sdkp, 1);
4265 if (!ret) {
4266 sd_resume(dev);
4267 sdkp->suspended = false;
4268 }
4269
4270 return ret;
4271}
4272
4273static int sd_resume_system(struct device *dev)
4274{
4275 if (pm_runtime_suspended(dev)) {
4276 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4277 struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4278
4279 if (sdp && sdp->force_runtime_start_on_system_start)
4280 pm_request_resume(dev);
4281
4282 return 0;
4283 }
4284
4285 return sd_resume_common(dev, false);
4286}
4287
4288static int sd_resume_runtime(struct device *dev)
4289{
4290 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4291 struct scsi_device *sdp;
4292
4293 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4294 return 0;
4295
4296 sdp = sdkp->device;
4297
4298 if (sdp->ignore_media_change) {
4299 /* clear the device's sense data */
4300 static const u8 cmd[10] = { REQUEST_SENSE };
4301 const struct scsi_exec_args exec_args = {
4302 .req_flags = BLK_MQ_REQ_PM,
4303 };
4304
4305 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4306 sdp->request_queue->rq_timeout, 1,
4307 &exec_args))
4308 sd_printk(KERN_NOTICE, sdkp,
4309 "Failed to clear sense data\n");
4310 }
4311
4312 return sd_resume_common(dev, true);
4313}
4314
4315static const struct dev_pm_ops sd_pm_ops = {
4316 .suspend = sd_suspend_system,
4317 .resume = sd_resume_system,
4318 .poweroff = sd_suspend_system,
4319 .restore = sd_resume_system,
4320 .runtime_suspend = sd_suspend_runtime,
4321 .runtime_resume = sd_resume_runtime,
4322};
4323
4324static struct scsi_driver sd_template = {
4325 .gendrv = {
4326 .name = "sd",
4327 .probe = sd_probe,
4328 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
4329 .remove = sd_remove,
4330 .shutdown = sd_shutdown,
4331 .pm = &sd_pm_ops,
4332 },
4333 .rescan = sd_rescan,
4334 .resume = sd_resume,
4335 .init_command = sd_init_command,
4336 .uninit_command = sd_uninit_command,
4337 .done = sd_done,
4338 .eh_action = sd_eh_action,
4339 .eh_reset = sd_eh_reset,
4340};
4341
4342/**
4343 * init_sd - entry point for this driver (both when built in or when
4344 * a module).
4345 *
4346 * Note: this function registers this driver with the scsi mid-level.
4347 **/
4348static int __init init_sd(void)
4349{
4350 int majors = 0, i, err;
4351
4352 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4353
4354 for (i = 0; i < SD_MAJORS; i++) {
4355 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4356 continue;
4357 majors++;
4358 }
4359
4360 if (!majors)
4361 return -ENODEV;
4362
4363 err = class_register(&sd_disk_class);
4364 if (err)
4365 goto err_out;
4366
4367 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4368 if (!sd_page_pool) {
4369 printk(KERN_ERR "sd: can't init discard page pool\n");
4370 err = -ENOMEM;
4371 goto err_out_class;
4372 }
4373
4374 err = scsi_register_driver(&sd_template.gendrv);
4375 if (err)
4376 goto err_out_driver;
4377
4378 return 0;
4379
4380err_out_driver:
4381 mempool_destroy(sd_page_pool);
4382err_out_class:
4383 class_unregister(&sd_disk_class);
4384err_out:
4385 for (i = 0; i < SD_MAJORS; i++)
4386 unregister_blkdev(sd_major(i), "sd");
4387 return err;
4388}
4389
4390/**
4391 * exit_sd - exit point for this driver (when it is a module).
4392 *
4393 * Note: this function unregisters this driver from the scsi mid-level.
4394 **/
4395static void __exit exit_sd(void)
4396{
4397 int i;
4398
4399 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4400
4401 scsi_unregister_driver(&sd_template.gendrv);
4402 mempool_destroy(sd_page_pool);
4403
4404 class_unregister(&sd_disk_class);
4405
4406 for (i = 0; i < SD_MAJORS; i++)
4407 unregister_blkdev(sd_major(i), "sd");
4408}
4409
4410module_init(init_sd);
4411module_exit(exit_sd);
4412
4413void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4414{
4415 scsi_print_sense_hdr(sdkp->device,
4416 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4417}
4418
4419void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4420{
4421 const char *hb_string = scsi_hostbyte_string(result);
4422
4423 if (hb_string)
4424 sd_printk(KERN_INFO, sdkp,
4425 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4426 hb_string ? hb_string : "invalid",
4427 "DRIVER_OK");
4428 else
4429 sd_printk(KERN_INFO, sdkp,
4430 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4431 msg, host_byte(result), "DRIVER_OK");
4432}
Linux 6.x block layer and io_uring tree(s)