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