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/*
2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4 *
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24 * Support 32k/1M disks.
25 *
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
33 */
34
35#include <linux/module.h>
36#include <linux/fs.h>
37#include <linux/kernel.h>
38#include <linux/mm.h>
39#include <linux/bio.h>
40#include <linux/genhd.h>
41#include <linux/hdreg.h>
42#include <linux/errno.h>
43#include <linux/idr.h>
44#include <linux/interrupt.h>
45#include <linux/init.h>
46#include <linux/blkdev.h>
47#include <linux/blkpg.h>
48#include <linux/delay.h>
49#include <linux/mutex.h>
50#include <linux/string_helpers.h>
51#include <linux/async.h>
52#include <linux/slab.h>
53#include <asm/uaccess.h>
54#include <asm/unaligned.h>
55
56#include <scsi/scsi.h>
57#include <scsi/scsi_cmnd.h>
58#include <scsi/scsi_dbg.h>
59#include <scsi/scsi_device.h>
60#include <scsi/scsi_driver.h>
61#include <scsi/scsi_eh.h>
62#include <scsi/scsi_host.h>
63#include <scsi/scsi_ioctl.h>
64#include <scsi/scsicam.h>
65
66#include "sd.h"
67#include "scsi_logging.h"
68
69MODULE_AUTHOR("Eric Youngdale");
70MODULE_DESCRIPTION("SCSI disk (sd) driver");
71MODULE_LICENSE("GPL");
72
73MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
74MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
75MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
76MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
77MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
78MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
79MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
80MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
89MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
90MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
91MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
92
93#if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
94#define SD_MINORS 16
95#else
96#define SD_MINORS 0
97#endif
98
99static void sd_config_discard(struct scsi_disk *, unsigned int);
100static int sd_revalidate_disk(struct gendisk *);
101static void sd_unlock_native_capacity(struct gendisk *disk);
102static int sd_probe(struct device *);
103static int sd_remove(struct device *);
104static void sd_shutdown(struct device *);
105static int sd_suspend(struct device *, pm_message_t state);
106static int sd_resume(struct device *);
107static void sd_rescan(struct device *);
108static int sd_done(struct scsi_cmnd *);
109static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
110static void scsi_disk_release(struct device *cdev);
111static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
112static void sd_print_result(struct scsi_disk *, int);
113
114static DEFINE_SPINLOCK(sd_index_lock);
115static DEFINE_IDA(sd_index_ida);
116
117/* This semaphore is used to mediate the 0->1 reference get in the
118 * face of object destruction (i.e. we can't allow a get on an
119 * object after last put) */
120static DEFINE_MUTEX(sd_ref_mutex);
121
122static struct kmem_cache *sd_cdb_cache;
123static mempool_t *sd_cdb_pool;
124
125static const char *sd_cache_types[] = {
126 "write through", "none", "write back",
127 "write back, no read (daft)"
128};
129
130static ssize_t
131sd_store_cache_type(struct device *dev, struct device_attribute *attr,
132 const char *buf, size_t count)
133{
134 int i, ct = -1, rcd, wce, sp;
135 struct scsi_disk *sdkp = to_scsi_disk(dev);
136 struct scsi_device *sdp = sdkp->device;
137 char buffer[64];
138 char *buffer_data;
139 struct scsi_mode_data data;
140 struct scsi_sense_hdr sshdr;
141 int len;
142
143 if (sdp->type != TYPE_DISK)
144 /* no cache control on RBC devices; theoretically they
145 * can do it, but there's probably so many exceptions
146 * it's not worth the risk */
147 return -EINVAL;
148
149 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
150 len = strlen(sd_cache_types[i]);
151 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
152 buf[len] == '\n') {
153 ct = i;
154 break;
155 }
156 }
157 if (ct < 0)
158 return -EINVAL;
159 rcd = ct & 0x01 ? 1 : 0;
160 wce = ct & 0x02 ? 1 : 0;
161 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
162 SD_MAX_RETRIES, &data, NULL))
163 return -EINVAL;
164 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
165 data.block_descriptor_length);
166 buffer_data = buffer + data.header_length +
167 data.block_descriptor_length;
168 buffer_data[2] &= ~0x05;
169 buffer_data[2] |= wce << 2 | rcd;
170 sp = buffer_data[0] & 0x80 ? 1 : 0;
171
172 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
173 SD_MAX_RETRIES, &data, &sshdr)) {
174 if (scsi_sense_valid(&sshdr))
175 sd_print_sense_hdr(sdkp, &sshdr);
176 return -EINVAL;
177 }
178 revalidate_disk(sdkp->disk);
179 return count;
180}
181
182static ssize_t
183sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
184 const char *buf, size_t count)
185{
186 struct scsi_disk *sdkp = to_scsi_disk(dev);
187 struct scsi_device *sdp = sdkp->device;
188
189 if (!capable(CAP_SYS_ADMIN))
190 return -EACCES;
191
192 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
193
194 return count;
195}
196
197static ssize_t
198sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
199 const char *buf, size_t count)
200{
201 struct scsi_disk *sdkp = to_scsi_disk(dev);
202 struct scsi_device *sdp = sdkp->device;
203
204 if (!capable(CAP_SYS_ADMIN))
205 return -EACCES;
206
207 if (sdp->type != TYPE_DISK)
208 return -EINVAL;
209
210 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
211
212 return count;
213}
214
215static ssize_t
216sd_show_cache_type(struct device *dev, struct device_attribute *attr,
217 char *buf)
218{
219 struct scsi_disk *sdkp = to_scsi_disk(dev);
220 int ct = sdkp->RCD + 2*sdkp->WCE;
221
222 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
223}
224
225static ssize_t
226sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
227{
228 struct scsi_disk *sdkp = to_scsi_disk(dev);
229
230 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
231}
232
233static ssize_t
234sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
235 char *buf)
236{
237 struct scsi_disk *sdkp = to_scsi_disk(dev);
238 struct scsi_device *sdp = sdkp->device;
239
240 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
241}
242
243static ssize_t
244sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
245 char *buf)
246{
247 struct scsi_disk *sdkp = to_scsi_disk(dev);
248
249 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
250}
251
252static ssize_t
253sd_show_protection_type(struct device *dev, struct device_attribute *attr,
254 char *buf)
255{
256 struct scsi_disk *sdkp = to_scsi_disk(dev);
257
258 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
259}
260
261static ssize_t
262sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
263 char *buf)
264{
265 struct scsi_disk *sdkp = to_scsi_disk(dev);
266 struct scsi_device *sdp = sdkp->device;
267 unsigned int dif, dix;
268
269 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
270 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
271
272 if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
273 dif = 0;
274 dix = 1;
275 }
276
277 if (!dif && !dix)
278 return snprintf(buf, 20, "none\n");
279
280 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
281}
282
283static ssize_t
284sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
285 char *buf)
286{
287 struct scsi_disk *sdkp = to_scsi_disk(dev);
288
289 return snprintf(buf, 20, "%u\n", sdkp->ATO);
290}
291
292static ssize_t
293sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
294 char *buf)
295{
296 struct scsi_disk *sdkp = to_scsi_disk(dev);
297
298 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
299}
300
301static const char *lbp_mode[] = {
302 [SD_LBP_FULL] = "full",
303 [SD_LBP_UNMAP] = "unmap",
304 [SD_LBP_WS16] = "writesame_16",
305 [SD_LBP_WS10] = "writesame_10",
306 [SD_LBP_ZERO] = "writesame_zero",
307 [SD_LBP_DISABLE] = "disabled",
308};
309
310static ssize_t
311sd_show_provisioning_mode(struct device *dev, struct device_attribute *attr,
312 char *buf)
313{
314 struct scsi_disk *sdkp = to_scsi_disk(dev);
315
316 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
317}
318
319static ssize_t
320sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
321 const char *buf, size_t count)
322{
323 struct scsi_disk *sdkp = to_scsi_disk(dev);
324 struct scsi_device *sdp = sdkp->device;
325
326 if (!capable(CAP_SYS_ADMIN))
327 return -EACCES;
328
329 if (sdp->type != TYPE_DISK)
330 return -EINVAL;
331
332 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
333 sd_config_discard(sdkp, SD_LBP_UNMAP);
334 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
335 sd_config_discard(sdkp, SD_LBP_WS16);
336 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
337 sd_config_discard(sdkp, SD_LBP_WS10);
338 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
339 sd_config_discard(sdkp, SD_LBP_ZERO);
340 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
341 sd_config_discard(sdkp, SD_LBP_DISABLE);
342 else
343 return -EINVAL;
344
345 return count;
346}
347
348static struct device_attribute sd_disk_attrs[] = {
349 __ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
350 sd_store_cache_type),
351 __ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
352 __ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
353 sd_store_allow_restart),
354 __ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
355 sd_store_manage_start_stop),
356 __ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
357 __ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
358 __ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
359 __ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
360 __ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
361 sd_store_provisioning_mode),
362 __ATTR_NULL,
363};
364
365static struct class sd_disk_class = {
366 .name = "scsi_disk",
367 .owner = THIS_MODULE,
368 .dev_release = scsi_disk_release,
369 .dev_attrs = sd_disk_attrs,
370};
371
372static struct scsi_driver sd_template = {
373 .owner = THIS_MODULE,
374 .gendrv = {
375 .name = "sd",
376 .probe = sd_probe,
377 .remove = sd_remove,
378 .suspend = sd_suspend,
379 .resume = sd_resume,
380 .shutdown = sd_shutdown,
381 },
382 .rescan = sd_rescan,
383 .done = sd_done,
384};
385
386/*
387 * Device no to disk mapping:
388 *
389 * major disc2 disc p1
390 * |............|.............|....|....| <- dev_t
391 * 31 20 19 8 7 4 3 0
392 *
393 * Inside a major, we have 16k disks, however mapped non-
394 * contiguously. The first 16 disks are for major0, the next
395 * ones with major1, ... Disk 256 is for major0 again, disk 272
396 * for major1, ...
397 * As we stay compatible with our numbering scheme, we can reuse
398 * the well-know SCSI majors 8, 65--71, 136--143.
399 */
400static int sd_major(int major_idx)
401{
402 switch (major_idx) {
403 case 0:
404 return SCSI_DISK0_MAJOR;
405 case 1 ... 7:
406 return SCSI_DISK1_MAJOR + major_idx - 1;
407 case 8 ... 15:
408 return SCSI_DISK8_MAJOR + major_idx - 8;
409 default:
410 BUG();
411 return 0; /* shut up gcc */
412 }
413}
414
415static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
416{
417 struct scsi_disk *sdkp = NULL;
418
419 if (disk->private_data) {
420 sdkp = scsi_disk(disk);
421 if (scsi_device_get(sdkp->device) == 0)
422 get_device(&sdkp->dev);
423 else
424 sdkp = NULL;
425 }
426 return sdkp;
427}
428
429static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
430{
431 struct scsi_disk *sdkp;
432
433 mutex_lock(&sd_ref_mutex);
434 sdkp = __scsi_disk_get(disk);
435 mutex_unlock(&sd_ref_mutex);
436 return sdkp;
437}
438
439static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
440{
441 struct scsi_disk *sdkp;
442
443 mutex_lock(&sd_ref_mutex);
444 sdkp = dev_get_drvdata(dev);
445 if (sdkp)
446 sdkp = __scsi_disk_get(sdkp->disk);
447 mutex_unlock(&sd_ref_mutex);
448 return sdkp;
449}
450
451static void scsi_disk_put(struct scsi_disk *sdkp)
452{
453 struct scsi_device *sdev = sdkp->device;
454
455 mutex_lock(&sd_ref_mutex);
456 put_device(&sdkp->dev);
457 scsi_device_put(sdev);
458 mutex_unlock(&sd_ref_mutex);
459}
460
461static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
462{
463 unsigned int prot_op = SCSI_PROT_NORMAL;
464 unsigned int dix = scsi_prot_sg_count(scmd);
465
466 if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
467 if (dif && dix)
468 prot_op = SCSI_PROT_READ_PASS;
469 else if (dif && !dix)
470 prot_op = SCSI_PROT_READ_STRIP;
471 else if (!dif && dix)
472 prot_op = SCSI_PROT_READ_INSERT;
473 } else {
474 if (dif && dix)
475 prot_op = SCSI_PROT_WRITE_PASS;
476 else if (dif && !dix)
477 prot_op = SCSI_PROT_WRITE_INSERT;
478 else if (!dif && dix)
479 prot_op = SCSI_PROT_WRITE_STRIP;
480 }
481
482 scsi_set_prot_op(scmd, prot_op);
483 scsi_set_prot_type(scmd, dif);
484}
485
486static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
487{
488 struct request_queue *q = sdkp->disk->queue;
489 unsigned int logical_block_size = sdkp->device->sector_size;
490 unsigned int max_blocks = 0;
491
492 q->limits.discard_zeroes_data = sdkp->lbprz;
493 q->limits.discard_alignment = sdkp->unmap_alignment *
494 logical_block_size;
495 q->limits.discard_granularity =
496 max(sdkp->physical_block_size,
497 sdkp->unmap_granularity * logical_block_size);
498
499 switch (mode) {
500
501 case SD_LBP_DISABLE:
502 q->limits.max_discard_sectors = 0;
503 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
504 return;
505
506 case SD_LBP_UNMAP:
507 max_blocks = min_not_zero(sdkp->max_unmap_blocks, 0xffffffff);
508 break;
509
510 case SD_LBP_WS16:
511 max_blocks = min_not_zero(sdkp->max_ws_blocks, 0xffffffff);
512 break;
513
514 case SD_LBP_WS10:
515 max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
516 break;
517
518 case SD_LBP_ZERO:
519 max_blocks = min_not_zero(sdkp->max_ws_blocks, (u32)0xffff);
520 q->limits.discard_zeroes_data = 1;
521 break;
522 }
523
524 q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
525 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
526
527 sdkp->provisioning_mode = mode;
528}
529
530/**
531 * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
532 * @sdp: scsi device to operate one
533 * @rq: Request to prepare
534 *
535 * Will issue either UNMAP or WRITE SAME(16) depending on preference
536 * indicated by target device.
537 **/
538static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
539{
540 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
541 struct bio *bio = rq->bio;
542 sector_t sector = bio->bi_sector;
543 unsigned int nr_sectors = bio_sectors(bio);
544 unsigned int len;
545 int ret;
546 char *buf;
547 struct page *page;
548
549 if (sdkp->device->sector_size == 4096) {
550 sector >>= 3;
551 nr_sectors >>= 3;
552 }
553
554 rq->timeout = SD_TIMEOUT;
555
556 memset(rq->cmd, 0, rq->cmd_len);
557
558 page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
559 if (!page)
560 return BLKPREP_DEFER;
561
562 switch (sdkp->provisioning_mode) {
563 case SD_LBP_UNMAP:
564 buf = page_address(page);
565
566 rq->cmd_len = 10;
567 rq->cmd[0] = UNMAP;
568 rq->cmd[8] = 24;
569
570 put_unaligned_be16(6 + 16, &buf[0]);
571 put_unaligned_be16(16, &buf[2]);
572 put_unaligned_be64(sector, &buf[8]);
573 put_unaligned_be32(nr_sectors, &buf[16]);
574
575 len = 24;
576 break;
577
578 case SD_LBP_WS16:
579 rq->cmd_len = 16;
580 rq->cmd[0] = WRITE_SAME_16;
581 rq->cmd[1] = 0x8; /* UNMAP */
582 put_unaligned_be64(sector, &rq->cmd[2]);
583 put_unaligned_be32(nr_sectors, &rq->cmd[10]);
584
585 len = sdkp->device->sector_size;
586 break;
587
588 case SD_LBP_WS10:
589 case SD_LBP_ZERO:
590 rq->cmd_len = 10;
591 rq->cmd[0] = WRITE_SAME;
592 if (sdkp->provisioning_mode == SD_LBP_WS10)
593 rq->cmd[1] = 0x8; /* UNMAP */
594 put_unaligned_be32(sector, &rq->cmd[2]);
595 put_unaligned_be16(nr_sectors, &rq->cmd[7]);
596
597 len = sdkp->device->sector_size;
598 break;
599
600 default:
601 ret = BLKPREP_KILL;
602 goto out;
603 }
604
605 blk_add_request_payload(rq, page, len);
606 ret = scsi_setup_blk_pc_cmnd(sdp, rq);
607 rq->buffer = page_address(page);
608
609out:
610 if (ret != BLKPREP_OK) {
611 __free_page(page);
612 rq->buffer = NULL;
613 }
614 return ret;
615}
616
617static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
618{
619 rq->timeout = SD_FLUSH_TIMEOUT;
620 rq->retries = SD_MAX_RETRIES;
621 rq->cmd[0] = SYNCHRONIZE_CACHE;
622 rq->cmd_len = 10;
623
624 return scsi_setup_blk_pc_cmnd(sdp, rq);
625}
626
627static void sd_unprep_fn(struct request_queue *q, struct request *rq)
628{
629 if (rq->cmd_flags & REQ_DISCARD) {
630 free_page((unsigned long)rq->buffer);
631 rq->buffer = NULL;
632 }
633}
634
635/**
636 * sd_init_command - build a scsi (read or write) command from
637 * information in the request structure.
638 * @SCpnt: pointer to mid-level's per scsi command structure that
639 * contains request and into which the scsi command is written
640 *
641 * Returns 1 if successful and 0 if error (or cannot be done now).
642 **/
643static int sd_prep_fn(struct request_queue *q, struct request *rq)
644{
645 struct scsi_cmnd *SCpnt;
646 struct scsi_device *sdp = q->queuedata;
647 struct gendisk *disk = rq->rq_disk;
648 struct scsi_disk *sdkp;
649 sector_t block = blk_rq_pos(rq);
650 sector_t threshold;
651 unsigned int this_count = blk_rq_sectors(rq);
652 int ret, host_dif;
653 unsigned char protect;
654
655 /*
656 * Discard request come in as REQ_TYPE_FS but we turn them into
657 * block PC requests to make life easier.
658 */
659 if (rq->cmd_flags & REQ_DISCARD) {
660 ret = scsi_setup_discard_cmnd(sdp, rq);
661 goto out;
662 } else if (rq->cmd_flags & REQ_FLUSH) {
663 ret = scsi_setup_flush_cmnd(sdp, rq);
664 goto out;
665 } else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
666 ret = scsi_setup_blk_pc_cmnd(sdp, rq);
667 goto out;
668 } else if (rq->cmd_type != REQ_TYPE_FS) {
669 ret = BLKPREP_KILL;
670 goto out;
671 }
672 ret = scsi_setup_fs_cmnd(sdp, rq);
673 if (ret != BLKPREP_OK)
674 goto out;
675 SCpnt = rq->special;
676 sdkp = scsi_disk(disk);
677
678 /* from here on until we're complete, any goto out
679 * is used for a killable error condition */
680 ret = BLKPREP_KILL;
681
682 SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
683 "sd_init_command: block=%llu, "
684 "count=%d\n",
685 (unsigned long long)block,
686 this_count));
687
688 if (!sdp || !scsi_device_online(sdp) ||
689 block + blk_rq_sectors(rq) > get_capacity(disk)) {
690 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
691 "Finishing %u sectors\n",
692 blk_rq_sectors(rq)));
693 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
694 "Retry with 0x%p\n", SCpnt));
695 goto out;
696 }
697
698 if (sdp->changed) {
699 /*
700 * quietly refuse to do anything to a changed disc until
701 * the changed bit has been reset
702 */
703 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
704 goto out;
705 }
706
707 /*
708 * Some SD card readers can't handle multi-sector accesses which touch
709 * the last one or two hardware sectors. Split accesses as needed.
710 */
711 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
712 (sdp->sector_size / 512);
713
714 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
715 if (block < threshold) {
716 /* Access up to the threshold but not beyond */
717 this_count = threshold - block;
718 } else {
719 /* Access only a single hardware sector */
720 this_count = sdp->sector_size / 512;
721 }
722 }
723
724 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
725 (unsigned long long)block));
726
727 /*
728 * If we have a 1K hardware sectorsize, prevent access to single
729 * 512 byte sectors. In theory we could handle this - in fact
730 * the scsi cdrom driver must be able to handle this because
731 * we typically use 1K blocksizes, and cdroms typically have
732 * 2K hardware sectorsizes. Of course, things are simpler
733 * with the cdrom, since it is read-only. For performance
734 * reasons, the filesystems should be able to handle this
735 * and not force the scsi disk driver to use bounce buffers
736 * for this.
737 */
738 if (sdp->sector_size == 1024) {
739 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
740 scmd_printk(KERN_ERR, SCpnt,
741 "Bad block number requested\n");
742 goto out;
743 } else {
744 block = block >> 1;
745 this_count = this_count >> 1;
746 }
747 }
748 if (sdp->sector_size == 2048) {
749 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
750 scmd_printk(KERN_ERR, SCpnt,
751 "Bad block number requested\n");
752 goto out;
753 } else {
754 block = block >> 2;
755 this_count = this_count >> 2;
756 }
757 }
758 if (sdp->sector_size == 4096) {
759 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
760 scmd_printk(KERN_ERR, SCpnt,
761 "Bad block number requested\n");
762 goto out;
763 } else {
764 block = block >> 3;
765 this_count = this_count >> 3;
766 }
767 }
768 if (rq_data_dir(rq) == WRITE) {
769 if (!sdp->writeable) {
770 goto out;
771 }
772 SCpnt->cmnd[0] = WRITE_6;
773 SCpnt->sc_data_direction = DMA_TO_DEVICE;
774
775 if (blk_integrity_rq(rq) &&
776 sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
777 goto out;
778
779 } else if (rq_data_dir(rq) == READ) {
780 SCpnt->cmnd[0] = READ_6;
781 SCpnt->sc_data_direction = DMA_FROM_DEVICE;
782 } else {
783 scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
784 goto out;
785 }
786
787 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
788 "%s %d/%u 512 byte blocks.\n",
789 (rq_data_dir(rq) == WRITE) ?
790 "writing" : "reading", this_count,
791 blk_rq_sectors(rq)));
792
793 /* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
794 host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
795 if (host_dif)
796 protect = 1 << 5;
797 else
798 protect = 0;
799
800 if (host_dif == SD_DIF_TYPE2_PROTECTION) {
801 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
802
803 if (unlikely(SCpnt->cmnd == NULL)) {
804 ret = BLKPREP_DEFER;
805 goto out;
806 }
807
808 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
809 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
810 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
811 SCpnt->cmnd[7] = 0x18;
812 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
813 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
814
815 /* LBA */
816 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
817 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
818 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
819 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
820 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
821 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
822 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
823 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
824
825 /* Expected Indirect LBA */
826 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
827 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
828 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
829 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
830
831 /* Transfer length */
832 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
833 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
834 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
835 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
836 } else if (block > 0xffffffff) {
837 SCpnt->cmnd[0] += READ_16 - READ_6;
838 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
839 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
840 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
841 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
842 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
843 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
844 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
845 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
846 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
847 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
848 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
849 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
850 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
851 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
852 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
853 scsi_device_protection(SCpnt->device) ||
854 SCpnt->device->use_10_for_rw) {
855 if (this_count > 0xffff)
856 this_count = 0xffff;
857
858 SCpnt->cmnd[0] += READ_10 - READ_6;
859 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
860 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
861 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
862 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
863 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
864 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
865 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
866 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
867 } else {
868 if (unlikely(rq->cmd_flags & REQ_FUA)) {
869 /*
870 * This happens only if this drive failed
871 * 10byte rw command with ILLEGAL_REQUEST
872 * during operation and thus turned off
873 * use_10_for_rw.
874 */
875 scmd_printk(KERN_ERR, SCpnt,
876 "FUA write on READ/WRITE(6) drive\n");
877 goto out;
878 }
879
880 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
881 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
882 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
883 SCpnt->cmnd[4] = (unsigned char) this_count;
884 SCpnt->cmnd[5] = 0;
885 }
886 SCpnt->sdb.length = this_count * sdp->sector_size;
887
888 /* If DIF or DIX is enabled, tell HBA how to handle request */
889 if (host_dif || scsi_prot_sg_count(SCpnt))
890 sd_prot_op(SCpnt, host_dif);
891
892 /*
893 * We shouldn't disconnect in the middle of a sector, so with a dumb
894 * host adapter, it's safe to assume that we can at least transfer
895 * this many bytes between each connect / disconnect.
896 */
897 SCpnt->transfersize = sdp->sector_size;
898 SCpnt->underflow = this_count << 9;
899 SCpnt->allowed = SD_MAX_RETRIES;
900
901 /*
902 * This indicates that the command is ready from our end to be
903 * queued.
904 */
905 ret = BLKPREP_OK;
906 out:
907 return scsi_prep_return(q, rq, ret);
908}
909
910/**
911 * sd_open - open a scsi disk device
912 * @inode: only i_rdev member may be used
913 * @filp: only f_mode and f_flags may be used
914 *
915 * Returns 0 if successful. Returns a negated errno value in case
916 * of error.
917 *
918 * Note: This can be called from a user context (e.g. fsck(1) )
919 * or from within the kernel (e.g. as a result of a mount(1) ).
920 * In the latter case @inode and @filp carry an abridged amount
921 * of information as noted above.
922 *
923 * Locking: called with bdev->bd_mutex held.
924 **/
925static int sd_open(struct block_device *bdev, fmode_t mode)
926{
927 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
928 struct scsi_device *sdev;
929 int retval;
930
931 if (!sdkp)
932 return -ENXIO;
933
934 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
935
936 sdev = sdkp->device;
937
938 retval = scsi_autopm_get_device(sdev);
939 if (retval)
940 goto error_autopm;
941
942 /*
943 * If the device is in error recovery, wait until it is done.
944 * If the device is offline, then disallow any access to it.
945 */
946 retval = -ENXIO;
947 if (!scsi_block_when_processing_errors(sdev))
948 goto error_out;
949
950 if (sdev->removable || sdkp->write_prot)
951 check_disk_change(bdev);
952
953 /*
954 * If the drive is empty, just let the open fail.
955 */
956 retval = -ENOMEDIUM;
957 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
958 goto error_out;
959
960 /*
961 * If the device has the write protect tab set, have the open fail
962 * if the user expects to be able to write to the thing.
963 */
964 retval = -EROFS;
965 if (sdkp->write_prot && (mode & FMODE_WRITE))
966 goto error_out;
967
968 /*
969 * It is possible that the disk changing stuff resulted in
970 * the device being taken offline. If this is the case,
971 * report this to the user, and don't pretend that the
972 * open actually succeeded.
973 */
974 retval = -ENXIO;
975 if (!scsi_device_online(sdev))
976 goto error_out;
977
978 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
979 if (scsi_block_when_processing_errors(sdev))
980 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
981 }
982
983 return 0;
984
985error_out:
986 scsi_autopm_put_device(sdev);
987error_autopm:
988 scsi_disk_put(sdkp);
989 return retval;
990}
991
992/**
993 * sd_release - invoked when the (last) close(2) is called on this
994 * scsi disk.
995 * @inode: only i_rdev member may be used
996 * @filp: only f_mode and f_flags may be used
997 *
998 * Returns 0.
999 *
1000 * Note: may block (uninterruptible) if error recovery is underway
1001 * on this disk.
1002 *
1003 * Locking: called with bdev->bd_mutex held.
1004 **/
1005static int sd_release(struct gendisk *disk, fmode_t mode)
1006{
1007 struct scsi_disk *sdkp = scsi_disk(disk);
1008 struct scsi_device *sdev = sdkp->device;
1009
1010 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1011
1012 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1013 if (scsi_block_when_processing_errors(sdev))
1014 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1015 }
1016
1017 /*
1018 * XXX and what if there are packets in flight and this close()
1019 * XXX is followed by a "rmmod sd_mod"?
1020 */
1021
1022 scsi_autopm_put_device(sdev);
1023 scsi_disk_put(sdkp);
1024 return 0;
1025}
1026
1027static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1028{
1029 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1030 struct scsi_device *sdp = sdkp->device;
1031 struct Scsi_Host *host = sdp->host;
1032 int diskinfo[4];
1033
1034 /* default to most commonly used values */
1035 diskinfo[0] = 0x40; /* 1 << 6 */
1036 diskinfo[1] = 0x20; /* 1 << 5 */
1037 diskinfo[2] = sdkp->capacity >> 11;
1038
1039 /* override with calculated, extended default, or driver values */
1040 if (host->hostt->bios_param)
1041 host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1042 else
1043 scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1044
1045 geo->heads = diskinfo[0];
1046 geo->sectors = diskinfo[1];
1047 geo->cylinders = diskinfo[2];
1048 return 0;
1049}
1050
1051/**
1052 * sd_ioctl - process an ioctl
1053 * @inode: only i_rdev/i_bdev members may be used
1054 * @filp: only f_mode and f_flags may be used
1055 * @cmd: ioctl command number
1056 * @arg: this is third argument given to ioctl(2) system call.
1057 * Often contains a pointer.
1058 *
1059 * Returns 0 if successful (some ioctls return positive numbers on
1060 * success as well). Returns a negated errno value in case of error.
1061 *
1062 * Note: most ioctls are forward onto the block subsystem or further
1063 * down in the scsi subsystem.
1064 **/
1065static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1066 unsigned int cmd, unsigned long arg)
1067{
1068 struct gendisk *disk = bdev->bd_disk;
1069 struct scsi_device *sdp = scsi_disk(disk)->device;
1070 void __user *p = (void __user *)arg;
1071 int error;
1072
1073 SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
1074 disk->disk_name, cmd));
1075
1076 /*
1077 * If we are in the middle of error recovery, don't let anyone
1078 * else try and use this device. Also, if error recovery fails, it
1079 * may try and take the device offline, in which case all further
1080 * access to the device is prohibited.
1081 */
1082 error = scsi_nonblockable_ioctl(sdp, cmd, p,
1083 (mode & FMODE_NDELAY) != 0);
1084 if (!scsi_block_when_processing_errors(sdp) || !error)
1085 goto out;
1086
1087 /*
1088 * Send SCSI addressing ioctls directly to mid level, send other
1089 * ioctls to block level and then onto mid level if they can't be
1090 * resolved.
1091 */
1092 switch (cmd) {
1093 case SCSI_IOCTL_GET_IDLUN:
1094 case SCSI_IOCTL_GET_BUS_NUMBER:
1095 error = scsi_ioctl(sdp, cmd, p);
1096 break;
1097 default:
1098 error = scsi_cmd_ioctl(disk->queue, disk, mode, cmd, p);
1099 if (error != -ENOTTY)
1100 break;
1101 error = scsi_ioctl(sdp, cmd, p);
1102 break;
1103 }
1104out:
1105 return error;
1106}
1107
1108static void set_media_not_present(struct scsi_disk *sdkp)
1109{
1110 if (sdkp->media_present)
1111 sdkp->device->changed = 1;
1112
1113 if (sdkp->device->removable) {
1114 sdkp->media_present = 0;
1115 sdkp->capacity = 0;
1116 }
1117}
1118
1119static int media_not_present(struct scsi_disk *sdkp,
1120 struct scsi_sense_hdr *sshdr)
1121{
1122 if (!scsi_sense_valid(sshdr))
1123 return 0;
1124
1125 /* not invoked for commands that could return deferred errors */
1126 switch (sshdr->sense_key) {
1127 case UNIT_ATTENTION:
1128 case NOT_READY:
1129 /* medium not present */
1130 if (sshdr->asc == 0x3A) {
1131 set_media_not_present(sdkp);
1132 return 1;
1133 }
1134 }
1135 return 0;
1136}
1137
1138/**
1139 * sd_check_events - check media events
1140 * @disk: kernel device descriptor
1141 * @clearing: disk events currently being cleared
1142 *
1143 * Returns mask of DISK_EVENT_*.
1144 *
1145 * Note: this function is invoked from the block subsystem.
1146 **/
1147static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1148{
1149 struct scsi_disk *sdkp = scsi_disk(disk);
1150 struct scsi_device *sdp = sdkp->device;
1151 struct scsi_sense_hdr *sshdr = NULL;
1152 int retval;
1153
1154 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1155
1156 /*
1157 * If the device is offline, don't send any commands - just pretend as
1158 * if the command failed. If the device ever comes back online, we
1159 * can deal with it then. It is only because of unrecoverable errors
1160 * that we would ever take a device offline in the first place.
1161 */
1162 if (!scsi_device_online(sdp)) {
1163 set_media_not_present(sdkp);
1164 goto out;
1165 }
1166
1167 /*
1168 * Using TEST_UNIT_READY enables differentiation between drive with
1169 * no cartridge loaded - NOT READY, drive with changed cartridge -
1170 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1171 *
1172 * Drives that auto spin down. eg iomega jaz 1G, will be started
1173 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1174 * sd_revalidate() is called.
1175 */
1176 retval = -ENODEV;
1177
1178 if (scsi_block_when_processing_errors(sdp)) {
1179 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1180 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1181 sshdr);
1182 }
1183
1184 /* failed to execute TUR, assume media not present */
1185 if (host_byte(retval)) {
1186 set_media_not_present(sdkp);
1187 goto out;
1188 }
1189
1190 if (media_not_present(sdkp, sshdr))
1191 goto out;
1192
1193 /*
1194 * For removable scsi disk we have to recognise the presence
1195 * of a disk in the drive.
1196 */
1197 if (!sdkp->media_present)
1198 sdp->changed = 1;
1199 sdkp->media_present = 1;
1200out:
1201 /*
1202 * sdp->changed is set under the following conditions:
1203 *
1204 * Medium present state has changed in either direction.
1205 * Device has indicated UNIT_ATTENTION.
1206 */
1207 kfree(sshdr);
1208 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1209 sdp->changed = 0;
1210 return retval;
1211}
1212
1213static int sd_sync_cache(struct scsi_disk *sdkp)
1214{
1215 int retries, res;
1216 struct scsi_device *sdp = sdkp->device;
1217 struct scsi_sense_hdr sshdr;
1218
1219 if (!scsi_device_online(sdp))
1220 return -ENODEV;
1221
1222
1223 for (retries = 3; retries > 0; --retries) {
1224 unsigned char cmd[10] = { 0 };
1225
1226 cmd[0] = SYNCHRONIZE_CACHE;
1227 /*
1228 * Leave the rest of the command zero to indicate
1229 * flush everything.
1230 */
1231 res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1232 SD_FLUSH_TIMEOUT, SD_MAX_RETRIES, NULL);
1233 if (res == 0)
1234 break;
1235 }
1236
1237 if (res) {
1238 sd_print_result(sdkp, res);
1239 if (driver_byte(res) & DRIVER_SENSE)
1240 sd_print_sense_hdr(sdkp, &sshdr);
1241 }
1242
1243 if (res)
1244 return -EIO;
1245 return 0;
1246}
1247
1248static void sd_rescan(struct device *dev)
1249{
1250 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1251
1252 if (sdkp) {
1253 revalidate_disk(sdkp->disk);
1254 scsi_disk_put(sdkp);
1255 }
1256}
1257
1258
1259#ifdef CONFIG_COMPAT
1260/*
1261 * This gets directly called from VFS. When the ioctl
1262 * is not recognized we go back to the other translation paths.
1263 */
1264static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1265 unsigned int cmd, unsigned long arg)
1266{
1267 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1268
1269 /*
1270 * If we are in the middle of error recovery, don't let anyone
1271 * else try and use this device. Also, if error recovery fails, it
1272 * may try and take the device offline, in which case all further
1273 * access to the device is prohibited.
1274 */
1275 if (!scsi_block_when_processing_errors(sdev))
1276 return -ENODEV;
1277
1278 if (sdev->host->hostt->compat_ioctl) {
1279 int ret;
1280
1281 ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1282
1283 return ret;
1284 }
1285
1286 /*
1287 * Let the static ioctl translation table take care of it.
1288 */
1289 return -ENOIOCTLCMD;
1290}
1291#endif
1292
1293static const struct block_device_operations sd_fops = {
1294 .owner = THIS_MODULE,
1295 .open = sd_open,
1296 .release = sd_release,
1297 .ioctl = sd_ioctl,
1298 .getgeo = sd_getgeo,
1299#ifdef CONFIG_COMPAT
1300 .compat_ioctl = sd_compat_ioctl,
1301#endif
1302 .check_events = sd_check_events,
1303 .revalidate_disk = sd_revalidate_disk,
1304 .unlock_native_capacity = sd_unlock_native_capacity,
1305};
1306
1307static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1308{
1309 u64 start_lba = blk_rq_pos(scmd->request);
1310 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1311 u64 bad_lba;
1312 int info_valid;
1313 /*
1314 * resid is optional but mostly filled in. When it's unused,
1315 * its value is zero, so we assume the whole buffer transferred
1316 */
1317 unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1318 unsigned int good_bytes;
1319
1320 if (scmd->request->cmd_type != REQ_TYPE_FS)
1321 return 0;
1322
1323 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1324 SCSI_SENSE_BUFFERSIZE,
1325 &bad_lba);
1326 if (!info_valid)
1327 return 0;
1328
1329 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1330 return 0;
1331
1332 if (scmd->device->sector_size < 512) {
1333 /* only legitimate sector_size here is 256 */
1334 start_lba <<= 1;
1335 end_lba <<= 1;
1336 } else {
1337 /* be careful ... don't want any overflows */
1338 u64 factor = scmd->device->sector_size / 512;
1339 do_div(start_lba, factor);
1340 do_div(end_lba, factor);
1341 }
1342
1343 /* The bad lba was reported incorrectly, we have no idea where
1344 * the error is.
1345 */
1346 if (bad_lba < start_lba || bad_lba >= end_lba)
1347 return 0;
1348
1349 /* This computation should always be done in terms of
1350 * the resolution of the device's medium.
1351 */
1352 good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1353 return min(good_bytes, transferred);
1354}
1355
1356/**
1357 * sd_done - bottom half handler: called when the lower level
1358 * driver has completed (successfully or otherwise) a scsi command.
1359 * @SCpnt: mid-level's per command structure.
1360 *
1361 * Note: potentially run from within an ISR. Must not block.
1362 **/
1363static int sd_done(struct scsi_cmnd *SCpnt)
1364{
1365 int result = SCpnt->result;
1366 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1367 struct scsi_sense_hdr sshdr;
1368 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1369 int sense_valid = 0;
1370 int sense_deferred = 0;
1371 unsigned char op = SCpnt->cmnd[0];
1372
1373 if ((SCpnt->request->cmd_flags & REQ_DISCARD) && !result)
1374 scsi_set_resid(SCpnt, 0);
1375
1376 if (result) {
1377 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1378 if (sense_valid)
1379 sense_deferred = scsi_sense_is_deferred(&sshdr);
1380 }
1381#ifdef CONFIG_SCSI_LOGGING
1382 SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1383 if (sense_valid) {
1384 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1385 "sd_done: sb[respc,sk,asc,"
1386 "ascq]=%x,%x,%x,%x\n",
1387 sshdr.response_code,
1388 sshdr.sense_key, sshdr.asc,
1389 sshdr.ascq));
1390 }
1391#endif
1392 if (driver_byte(result) != DRIVER_SENSE &&
1393 (!sense_valid || sense_deferred))
1394 goto out;
1395
1396 switch (sshdr.sense_key) {
1397 case HARDWARE_ERROR:
1398 case MEDIUM_ERROR:
1399 good_bytes = sd_completed_bytes(SCpnt);
1400 break;
1401 case RECOVERED_ERROR:
1402 good_bytes = scsi_bufflen(SCpnt);
1403 break;
1404 case NO_SENSE:
1405 /* This indicates a false check condition, so ignore it. An
1406 * unknown amount of data was transferred so treat it as an
1407 * error.
1408 */
1409 scsi_print_sense("sd", SCpnt);
1410 SCpnt->result = 0;
1411 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1412 break;
1413 case ABORTED_COMMAND:
1414 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1415 good_bytes = sd_completed_bytes(SCpnt);
1416 break;
1417 case ILLEGAL_REQUEST:
1418 if (sshdr.asc == 0x10) /* DIX: Host detected corruption */
1419 good_bytes = sd_completed_bytes(SCpnt);
1420 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1421 if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1422 (op == UNMAP || op == WRITE_SAME_16 || op == WRITE_SAME))
1423 sd_config_discard(sdkp, SD_LBP_DISABLE);
1424 break;
1425 default:
1426 break;
1427 }
1428 out:
1429 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1430 sd_dif_complete(SCpnt, good_bytes);
1431
1432 if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1433 == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1434
1435 /* We have to print a failed command here as the
1436 * extended CDB gets freed before scsi_io_completion()
1437 * is called.
1438 */
1439 if (result)
1440 scsi_print_command(SCpnt);
1441
1442 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1443 SCpnt->cmnd = NULL;
1444 SCpnt->cmd_len = 0;
1445 }
1446
1447 return good_bytes;
1448}
1449
1450/*
1451 * spinup disk - called only in sd_revalidate_disk()
1452 */
1453static void
1454sd_spinup_disk(struct scsi_disk *sdkp)
1455{
1456 unsigned char cmd[10];
1457 unsigned long spintime_expire = 0;
1458 int retries, spintime;
1459 unsigned int the_result;
1460 struct scsi_sense_hdr sshdr;
1461 int sense_valid = 0;
1462
1463 spintime = 0;
1464
1465 /* Spin up drives, as required. Only do this at boot time */
1466 /* Spinup needs to be done for module loads too. */
1467 do {
1468 retries = 0;
1469
1470 do {
1471 cmd[0] = TEST_UNIT_READY;
1472 memset((void *) &cmd[1], 0, 9);
1473
1474 the_result = scsi_execute_req(sdkp->device, cmd,
1475 DMA_NONE, NULL, 0,
1476 &sshdr, SD_TIMEOUT,
1477 SD_MAX_RETRIES, NULL);
1478
1479 /*
1480 * If the drive has indicated to us that it
1481 * doesn't have any media in it, don't bother
1482 * with any more polling.
1483 */
1484 if (media_not_present(sdkp, &sshdr))
1485 return;
1486
1487 if (the_result)
1488 sense_valid = scsi_sense_valid(&sshdr);
1489 retries++;
1490 } while (retries < 3 &&
1491 (!scsi_status_is_good(the_result) ||
1492 ((driver_byte(the_result) & DRIVER_SENSE) &&
1493 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1494
1495 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1496 /* no sense, TUR either succeeded or failed
1497 * with a status error */
1498 if(!spintime && !scsi_status_is_good(the_result)) {
1499 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1500 sd_print_result(sdkp, the_result);
1501 }
1502 break;
1503 }
1504
1505 /*
1506 * The device does not want the automatic start to be issued.
1507 */
1508 if (sdkp->device->no_start_on_add)
1509 break;
1510
1511 if (sense_valid && sshdr.sense_key == NOT_READY) {
1512 if (sshdr.asc == 4 && sshdr.ascq == 3)
1513 break; /* manual intervention required */
1514 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1515 break; /* standby */
1516 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1517 break; /* unavailable */
1518 /*
1519 * Issue command to spin up drive when not ready
1520 */
1521 if (!spintime) {
1522 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1523 cmd[0] = START_STOP;
1524 cmd[1] = 1; /* Return immediately */
1525 memset((void *) &cmd[2], 0, 8);
1526 cmd[4] = 1; /* Start spin cycle */
1527 if (sdkp->device->start_stop_pwr_cond)
1528 cmd[4] |= 1 << 4;
1529 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1530 NULL, 0, &sshdr,
1531 SD_TIMEOUT, SD_MAX_RETRIES,
1532 NULL);
1533 spintime_expire = jiffies + 100 * HZ;
1534 spintime = 1;
1535 }
1536 /* Wait 1 second for next try */
1537 msleep(1000);
1538 printk(".");
1539
1540 /*
1541 * Wait for USB flash devices with slow firmware.
1542 * Yes, this sense key/ASC combination shouldn't
1543 * occur here. It's characteristic of these devices.
1544 */
1545 } else if (sense_valid &&
1546 sshdr.sense_key == UNIT_ATTENTION &&
1547 sshdr.asc == 0x28) {
1548 if (!spintime) {
1549 spintime_expire = jiffies + 5 * HZ;
1550 spintime = 1;
1551 }
1552 /* Wait 1 second for next try */
1553 msleep(1000);
1554 } else {
1555 /* we don't understand the sense code, so it's
1556 * probably pointless to loop */
1557 if(!spintime) {
1558 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1559 sd_print_sense_hdr(sdkp, &sshdr);
1560 }
1561 break;
1562 }
1563
1564 } while (spintime && time_before_eq(jiffies, spintime_expire));
1565
1566 if (spintime) {
1567 if (scsi_status_is_good(the_result))
1568 printk("ready\n");
1569 else
1570 printk("not responding...\n");
1571 }
1572}
1573
1574
1575/*
1576 * Determine whether disk supports Data Integrity Field.
1577 */
1578static void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1579{
1580 struct scsi_device *sdp = sdkp->device;
1581 u8 type;
1582
1583 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1584 return;
1585
1586 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1587
1588 if (type == sdkp->protection_type || !sdkp->first_scan)
1589 return;
1590
1591 sdkp->protection_type = type;
1592
1593 if (type > SD_DIF_TYPE3_PROTECTION) {
1594 sd_printk(KERN_ERR, sdkp, "formatted with unsupported " \
1595 "protection type %u. Disabling disk!\n", type);
1596 sdkp->capacity = 0;
1597 return;
1598 }
1599
1600 if (scsi_host_dif_capable(sdp->host, type))
1601 sd_printk(KERN_NOTICE, sdkp,
1602 "Enabling DIF Type %u protection\n", type);
1603 else
1604 sd_printk(KERN_NOTICE, sdkp,
1605 "Disabling DIF Type %u protection\n", type);
1606}
1607
1608static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1609 struct scsi_sense_hdr *sshdr, int sense_valid,
1610 int the_result)
1611{
1612 sd_print_result(sdkp, the_result);
1613 if (driver_byte(the_result) & DRIVER_SENSE)
1614 sd_print_sense_hdr(sdkp, sshdr);
1615 else
1616 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1617
1618 /*
1619 * Set dirty bit for removable devices if not ready -
1620 * sometimes drives will not report this properly.
1621 */
1622 if (sdp->removable &&
1623 sense_valid && sshdr->sense_key == NOT_READY)
1624 set_media_not_present(sdkp);
1625
1626 /*
1627 * We used to set media_present to 0 here to indicate no media
1628 * in the drive, but some drives fail read capacity even with
1629 * media present, so we can't do that.
1630 */
1631 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1632}
1633
1634#define RC16_LEN 32
1635#if RC16_LEN > SD_BUF_SIZE
1636#error RC16_LEN must not be more than SD_BUF_SIZE
1637#endif
1638
1639#define READ_CAPACITY_RETRIES_ON_RESET 10
1640
1641static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1642 unsigned char *buffer)
1643{
1644 unsigned char cmd[16];
1645 struct scsi_sense_hdr sshdr;
1646 int sense_valid = 0;
1647 int the_result;
1648 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1649 unsigned int alignment;
1650 unsigned long long lba;
1651 unsigned sector_size;
1652
1653 if (sdp->no_read_capacity_16)
1654 return -EINVAL;
1655
1656 do {
1657 memset(cmd, 0, 16);
1658 cmd[0] = SERVICE_ACTION_IN;
1659 cmd[1] = SAI_READ_CAPACITY_16;
1660 cmd[13] = RC16_LEN;
1661 memset(buffer, 0, RC16_LEN);
1662
1663 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1664 buffer, RC16_LEN, &sshdr,
1665 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1666
1667 if (media_not_present(sdkp, &sshdr))
1668 return -ENODEV;
1669
1670 if (the_result) {
1671 sense_valid = scsi_sense_valid(&sshdr);
1672 if (sense_valid &&
1673 sshdr.sense_key == ILLEGAL_REQUEST &&
1674 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1675 sshdr.ascq == 0x00)
1676 /* Invalid Command Operation Code or
1677 * Invalid Field in CDB, just retry
1678 * silently with RC10 */
1679 return -EINVAL;
1680 if (sense_valid &&
1681 sshdr.sense_key == UNIT_ATTENTION &&
1682 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1683 /* Device reset might occur several times,
1684 * give it one more chance */
1685 if (--reset_retries > 0)
1686 continue;
1687 }
1688 retries--;
1689
1690 } while (the_result && retries);
1691
1692 if (the_result) {
1693 sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1694 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1695 return -EINVAL;
1696 }
1697
1698 sector_size = get_unaligned_be32(&buffer[8]);
1699 lba = get_unaligned_be64(&buffer[0]);
1700
1701 sd_read_protection_type(sdkp, buffer);
1702
1703 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1704 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1705 "kernel compiled with support for large block "
1706 "devices.\n");
1707 sdkp->capacity = 0;
1708 return -EOVERFLOW;
1709 }
1710
1711 /* Logical blocks per physical block exponent */
1712 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
1713
1714 /* Lowest aligned logical block */
1715 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1716 blk_queue_alignment_offset(sdp->request_queue, alignment);
1717 if (alignment && sdkp->first_scan)
1718 sd_printk(KERN_NOTICE, sdkp,
1719 "physical block alignment offset: %u\n", alignment);
1720
1721 if (buffer[14] & 0x80) { /* LBPME */
1722 sdkp->lbpme = 1;
1723
1724 if (buffer[14] & 0x40) /* LBPRZ */
1725 sdkp->lbprz = 1;
1726
1727 sd_config_discard(sdkp, SD_LBP_WS16);
1728 }
1729
1730 sdkp->capacity = lba + 1;
1731 return sector_size;
1732}
1733
1734static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1735 unsigned char *buffer)
1736{
1737 unsigned char cmd[16];
1738 struct scsi_sense_hdr sshdr;
1739 int sense_valid = 0;
1740 int the_result;
1741 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1742 sector_t lba;
1743 unsigned sector_size;
1744
1745 do {
1746 cmd[0] = READ_CAPACITY;
1747 memset(&cmd[1], 0, 9);
1748 memset(buffer, 0, 8);
1749
1750 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1751 buffer, 8, &sshdr,
1752 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1753
1754 if (media_not_present(sdkp, &sshdr))
1755 return -ENODEV;
1756
1757 if (the_result) {
1758 sense_valid = scsi_sense_valid(&sshdr);
1759 if (sense_valid &&
1760 sshdr.sense_key == UNIT_ATTENTION &&
1761 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1762 /* Device reset might occur several times,
1763 * give it one more chance */
1764 if (--reset_retries > 0)
1765 continue;
1766 }
1767 retries--;
1768
1769 } while (the_result && retries);
1770
1771 if (the_result) {
1772 sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1773 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1774 return -EINVAL;
1775 }
1776
1777 sector_size = get_unaligned_be32(&buffer[4]);
1778 lba = get_unaligned_be32(&buffer[0]);
1779
1780 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
1781 /* Some buggy (usb cardreader) devices return an lba of
1782 0xffffffff when the want to report a size of 0 (with
1783 which they really mean no media is present) */
1784 sdkp->capacity = 0;
1785 sdkp->physical_block_size = sector_size;
1786 return sector_size;
1787 }
1788
1789 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1790 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1791 "kernel compiled with support for large block "
1792 "devices.\n");
1793 sdkp->capacity = 0;
1794 return -EOVERFLOW;
1795 }
1796
1797 sdkp->capacity = lba + 1;
1798 sdkp->physical_block_size = sector_size;
1799 return sector_size;
1800}
1801
1802static int sd_try_rc16_first(struct scsi_device *sdp)
1803{
1804 if (sdp->host->max_cmd_len < 16)
1805 return 0;
1806 if (sdp->scsi_level > SCSI_SPC_2)
1807 return 1;
1808 if (scsi_device_protection(sdp))
1809 return 1;
1810 return 0;
1811}
1812
1813/*
1814 * read disk capacity
1815 */
1816static void
1817sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1818{
1819 int sector_size;
1820 struct scsi_device *sdp = sdkp->device;
1821 sector_t old_capacity = sdkp->capacity;
1822
1823 if (sd_try_rc16_first(sdp)) {
1824 sector_size = read_capacity_16(sdkp, sdp, buffer);
1825 if (sector_size == -EOVERFLOW)
1826 goto got_data;
1827 if (sector_size == -ENODEV)
1828 return;
1829 if (sector_size < 0)
1830 sector_size = read_capacity_10(sdkp, sdp, buffer);
1831 if (sector_size < 0)
1832 return;
1833 } else {
1834 sector_size = read_capacity_10(sdkp, sdp, buffer);
1835 if (sector_size == -EOVERFLOW)
1836 goto got_data;
1837 if (sector_size < 0)
1838 return;
1839 if ((sizeof(sdkp->capacity) > 4) &&
1840 (sdkp->capacity > 0xffffffffULL)) {
1841 int old_sector_size = sector_size;
1842 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1843 "Trying to use READ CAPACITY(16).\n");
1844 sector_size = read_capacity_16(sdkp, sdp, buffer);
1845 if (sector_size < 0) {
1846 sd_printk(KERN_NOTICE, sdkp,
1847 "Using 0xffffffff as device size\n");
1848 sdkp->capacity = 1 + (sector_t) 0xffffffff;
1849 sector_size = old_sector_size;
1850 goto got_data;
1851 }
1852 }
1853 }
1854
1855 /* Some devices are known to return the total number of blocks,
1856 * not the highest block number. Some devices have versions
1857 * which do this and others which do not. Some devices we might
1858 * suspect of doing this but we don't know for certain.
1859 *
1860 * If we know the reported capacity is wrong, decrement it. If
1861 * we can only guess, then assume the number of blocks is even
1862 * (usually true but not always) and err on the side of lowering
1863 * the capacity.
1864 */
1865 if (sdp->fix_capacity ||
1866 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1867 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1868 "from its reported value: %llu\n",
1869 (unsigned long long) sdkp->capacity);
1870 --sdkp->capacity;
1871 }
1872
1873got_data:
1874 if (sector_size == 0) {
1875 sector_size = 512;
1876 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
1877 "assuming 512.\n");
1878 }
1879
1880 if (sector_size != 512 &&
1881 sector_size != 1024 &&
1882 sector_size != 2048 &&
1883 sector_size != 4096 &&
1884 sector_size != 256) {
1885 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
1886 sector_size);
1887 /*
1888 * The user might want to re-format the drive with
1889 * a supported sectorsize. Once this happens, it
1890 * would be relatively trivial to set the thing up.
1891 * For this reason, we leave the thing in the table.
1892 */
1893 sdkp->capacity = 0;
1894 /*
1895 * set a bogus sector size so the normal read/write
1896 * logic in the block layer will eventually refuse any
1897 * request on this device without tripping over power
1898 * of two sector size assumptions
1899 */
1900 sector_size = 512;
1901 }
1902 blk_queue_logical_block_size(sdp->request_queue, sector_size);
1903
1904 {
1905 char cap_str_2[10], cap_str_10[10];
1906 u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
1907
1908 string_get_size(sz, STRING_UNITS_2, cap_str_2,
1909 sizeof(cap_str_2));
1910 string_get_size(sz, STRING_UNITS_10, cap_str_10,
1911 sizeof(cap_str_10));
1912
1913 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
1914 sd_printk(KERN_NOTICE, sdkp,
1915 "%llu %d-byte logical blocks: (%s/%s)\n",
1916 (unsigned long long)sdkp->capacity,
1917 sector_size, cap_str_10, cap_str_2);
1918
1919 if (sdkp->physical_block_size != sector_size)
1920 sd_printk(KERN_NOTICE, sdkp,
1921 "%u-byte physical blocks\n",
1922 sdkp->physical_block_size);
1923 }
1924 }
1925
1926 /* Rescale capacity to 512-byte units */
1927 if (sector_size == 4096)
1928 sdkp->capacity <<= 3;
1929 else if (sector_size == 2048)
1930 sdkp->capacity <<= 2;
1931 else if (sector_size == 1024)
1932 sdkp->capacity <<= 1;
1933 else if (sector_size == 256)
1934 sdkp->capacity >>= 1;
1935
1936 blk_queue_physical_block_size(sdp->request_queue,
1937 sdkp->physical_block_size);
1938 sdkp->device->sector_size = sector_size;
1939}
1940
1941/* called with buffer of length 512 */
1942static inline int
1943sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1944 unsigned char *buffer, int len, struct scsi_mode_data *data,
1945 struct scsi_sense_hdr *sshdr)
1946{
1947 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1948 SD_TIMEOUT, SD_MAX_RETRIES, data,
1949 sshdr);
1950}
1951
1952/*
1953 * read write protect setting, if possible - called only in sd_revalidate_disk()
1954 * called with buffer of length SD_BUF_SIZE
1955 */
1956static void
1957sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
1958{
1959 int res;
1960 struct scsi_device *sdp = sdkp->device;
1961 struct scsi_mode_data data;
1962 int old_wp = sdkp->write_prot;
1963
1964 set_disk_ro(sdkp->disk, 0);
1965 if (sdp->skip_ms_page_3f) {
1966 sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
1967 return;
1968 }
1969
1970 if (sdp->use_192_bytes_for_3f) {
1971 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1972 } else {
1973 /*
1974 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1975 * We have to start carefully: some devices hang if we ask
1976 * for more than is available.
1977 */
1978 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1979
1980 /*
1981 * Second attempt: ask for page 0 When only page 0 is
1982 * implemented, a request for page 3F may return Sense Key
1983 * 5: Illegal Request, Sense Code 24: Invalid field in
1984 * CDB.
1985 */
1986 if (!scsi_status_is_good(res))
1987 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1988
1989 /*
1990 * Third attempt: ask 255 bytes, as we did earlier.
1991 */
1992 if (!scsi_status_is_good(res))
1993 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1994 &data, NULL);
1995 }
1996
1997 if (!scsi_status_is_good(res)) {
1998 sd_printk(KERN_WARNING, sdkp,
1999 "Test WP failed, assume Write Enabled\n");
2000 } else {
2001 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2002 set_disk_ro(sdkp->disk, sdkp->write_prot);
2003 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2004 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2005 sdkp->write_prot ? "on" : "off");
2006 sd_printk(KERN_DEBUG, sdkp,
2007 "Mode Sense: %02x %02x %02x %02x\n",
2008 buffer[0], buffer[1], buffer[2], buffer[3]);
2009 }
2010 }
2011}
2012
2013/*
2014 * sd_read_cache_type - called only from sd_revalidate_disk()
2015 * called with buffer of length SD_BUF_SIZE
2016 */
2017static void
2018sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2019{
2020 int len = 0, res;
2021 struct scsi_device *sdp = sdkp->device;
2022
2023 int dbd;
2024 int modepage;
2025 int first_len;
2026 struct scsi_mode_data data;
2027 struct scsi_sense_hdr sshdr;
2028 int old_wce = sdkp->WCE;
2029 int old_rcd = sdkp->RCD;
2030 int old_dpofua = sdkp->DPOFUA;
2031
2032 first_len = 4;
2033 if (sdp->skip_ms_page_8) {
2034 if (sdp->type == TYPE_RBC)
2035 goto defaults;
2036 else {
2037 if (sdp->skip_ms_page_3f)
2038 goto defaults;
2039 modepage = 0x3F;
2040 if (sdp->use_192_bytes_for_3f)
2041 first_len = 192;
2042 dbd = 0;
2043 }
2044 } else if (sdp->type == TYPE_RBC) {
2045 modepage = 6;
2046 dbd = 8;
2047 } else {
2048 modepage = 8;
2049 dbd = 0;
2050 }
2051
2052 /* cautiously ask */
2053 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2054 &data, &sshdr);
2055
2056 if (!scsi_status_is_good(res))
2057 goto bad_sense;
2058
2059 if (!data.header_length) {
2060 modepage = 6;
2061 first_len = 0;
2062 sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
2063 }
2064
2065 /* that went OK, now ask for the proper length */
2066 len = data.length;
2067
2068 /*
2069 * We're only interested in the first three bytes, actually.
2070 * But the data cache page is defined for the first 20.
2071 */
2072 if (len < 3)
2073 goto bad_sense;
2074 else if (len > SD_BUF_SIZE) {
2075 sd_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2076 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2077 len = SD_BUF_SIZE;
2078 }
2079 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2080 len = 192;
2081
2082 /* Get the data */
2083 if (len > first_len)
2084 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2085 &data, &sshdr);
2086
2087 if (scsi_status_is_good(res)) {
2088 int offset = data.header_length + data.block_descriptor_length;
2089
2090 while (offset < len) {
2091 u8 page_code = buffer[offset] & 0x3F;
2092 u8 spf = buffer[offset] & 0x40;
2093
2094 if (page_code == 8 || page_code == 6) {
2095 /* We're interested only in the first 3 bytes.
2096 */
2097 if (len - offset <= 2) {
2098 sd_printk(KERN_ERR, sdkp, "Incomplete "
2099 "mode parameter data\n");
2100 goto defaults;
2101 } else {
2102 modepage = page_code;
2103 goto Page_found;
2104 }
2105 } else {
2106 /* Go to the next page */
2107 if (spf && len - offset > 3)
2108 offset += 4 + (buffer[offset+2] << 8) +
2109 buffer[offset+3];
2110 else if (!spf && len - offset > 1)
2111 offset += 2 + buffer[offset+1];
2112 else {
2113 sd_printk(KERN_ERR, sdkp, "Incomplete "
2114 "mode parameter data\n");
2115 goto defaults;
2116 }
2117 }
2118 }
2119
2120 if (modepage == 0x3F) {
2121 sd_printk(KERN_ERR, sdkp, "No Caching mode page "
2122 "present\n");
2123 goto defaults;
2124 } else if ((buffer[offset] & 0x3f) != modepage) {
2125 sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
2126 goto defaults;
2127 }
2128 Page_found:
2129 if (modepage == 8) {
2130 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2131 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2132 } else {
2133 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2134 sdkp->RCD = 0;
2135 }
2136
2137 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2138 if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2139 sd_printk(KERN_NOTICE, sdkp,
2140 "Uses READ/WRITE(6), disabling FUA\n");
2141 sdkp->DPOFUA = 0;
2142 }
2143
2144 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2145 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2146 sd_printk(KERN_NOTICE, sdkp,
2147 "Write cache: %s, read cache: %s, %s\n",
2148 sdkp->WCE ? "enabled" : "disabled",
2149 sdkp->RCD ? "disabled" : "enabled",
2150 sdkp->DPOFUA ? "supports DPO and FUA"
2151 : "doesn't support DPO or FUA");
2152
2153 return;
2154 }
2155
2156bad_sense:
2157 if (scsi_sense_valid(&sshdr) &&
2158 sshdr.sense_key == ILLEGAL_REQUEST &&
2159 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2160 /* Invalid field in CDB */
2161 sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2162 else
2163 sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
2164
2165defaults:
2166 sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
2167 sdkp->WCE = 0;
2168 sdkp->RCD = 0;
2169 sdkp->DPOFUA = 0;
2170}
2171
2172/*
2173 * The ATO bit indicates whether the DIF application tag is available
2174 * for use by the operating system.
2175 */
2176static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2177{
2178 int res, offset;
2179 struct scsi_device *sdp = sdkp->device;
2180 struct scsi_mode_data data;
2181 struct scsi_sense_hdr sshdr;
2182
2183 if (sdp->type != TYPE_DISK)
2184 return;
2185
2186 if (sdkp->protection_type == 0)
2187 return;
2188
2189 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2190 SD_MAX_RETRIES, &data, &sshdr);
2191
2192 if (!scsi_status_is_good(res) || !data.header_length ||
2193 data.length < 6) {
2194 sd_printk(KERN_WARNING, sdkp,
2195 "getting Control mode page failed, assume no ATO\n");
2196
2197 if (scsi_sense_valid(&sshdr))
2198 sd_print_sense_hdr(sdkp, &sshdr);
2199
2200 return;
2201 }
2202
2203 offset = data.header_length + data.block_descriptor_length;
2204
2205 if ((buffer[offset] & 0x3f) != 0x0a) {
2206 sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2207 return;
2208 }
2209
2210 if ((buffer[offset + 5] & 0x80) == 0)
2211 return;
2212
2213 sdkp->ATO = 1;
2214
2215 return;
2216}
2217
2218/**
2219 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2220 * @disk: disk to query
2221 */
2222static void sd_read_block_limits(struct scsi_disk *sdkp)
2223{
2224 unsigned int sector_sz = sdkp->device->sector_size;
2225 const int vpd_len = 64;
2226 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2227
2228 if (!buffer ||
2229 /* Block Limits VPD */
2230 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2231 goto out;
2232
2233 blk_queue_io_min(sdkp->disk->queue,
2234 get_unaligned_be16(&buffer[6]) * sector_sz);
2235 blk_queue_io_opt(sdkp->disk->queue,
2236 get_unaligned_be32(&buffer[12]) * sector_sz);
2237
2238 if (buffer[3] == 0x3c) {
2239 unsigned int lba_count, desc_count;
2240
2241 sdkp->max_ws_blocks =
2242 (u32) min_not_zero(get_unaligned_be64(&buffer[36]),
2243 (u64)0xffffffff);
2244
2245 if (!sdkp->lbpme)
2246 goto out;
2247
2248 lba_count = get_unaligned_be32(&buffer[20]);
2249 desc_count = get_unaligned_be32(&buffer[24]);
2250
2251 if (lba_count && desc_count)
2252 sdkp->max_unmap_blocks = lba_count;
2253
2254 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2255
2256 if (buffer[32] & 0x80)
2257 sdkp->unmap_alignment =
2258 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2259
2260 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2261
2262 if (sdkp->max_unmap_blocks)
2263 sd_config_discard(sdkp, SD_LBP_UNMAP);
2264 else
2265 sd_config_discard(sdkp, SD_LBP_WS16);
2266
2267 } else { /* LBP VPD page tells us what to use */
2268
2269 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2270 sd_config_discard(sdkp, SD_LBP_UNMAP);
2271 else if (sdkp->lbpws)
2272 sd_config_discard(sdkp, SD_LBP_WS16);
2273 else if (sdkp->lbpws10)
2274 sd_config_discard(sdkp, SD_LBP_WS10);
2275 else
2276 sd_config_discard(sdkp, SD_LBP_DISABLE);
2277 }
2278 }
2279
2280 out:
2281 kfree(buffer);
2282}
2283
2284/**
2285 * sd_read_block_characteristics - Query block dev. characteristics
2286 * @disk: disk to query
2287 */
2288static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2289{
2290 unsigned char *buffer;
2291 u16 rot;
2292 const int vpd_len = 64;
2293
2294 buffer = kmalloc(vpd_len, GFP_KERNEL);
2295
2296 if (!buffer ||
2297 /* Block Device Characteristics VPD */
2298 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2299 goto out;
2300
2301 rot = get_unaligned_be16(&buffer[4]);
2302
2303 if (rot == 1)
2304 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2305
2306 out:
2307 kfree(buffer);
2308}
2309
2310/**
2311 * sd_read_block_provisioning - Query provisioning VPD page
2312 * @disk: disk to query
2313 */
2314static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2315{
2316 unsigned char *buffer;
2317 const int vpd_len = 8;
2318
2319 if (sdkp->lbpme == 0)
2320 return;
2321
2322 buffer = kmalloc(vpd_len, GFP_KERNEL);
2323
2324 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2325 goto out;
2326
2327 sdkp->lbpvpd = 1;
2328 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2329 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2330 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2331
2332 out:
2333 kfree(buffer);
2334}
2335
2336static int sd_try_extended_inquiry(struct scsi_device *sdp)
2337{
2338 /*
2339 * Although VPD inquiries can go to SCSI-2 type devices,
2340 * some USB ones crash on receiving them, and the pages
2341 * we currently ask for are for SPC-3 and beyond
2342 */
2343 if (sdp->scsi_level > SCSI_SPC_2)
2344 return 1;
2345 return 0;
2346}
2347
2348/**
2349 * sd_revalidate_disk - called the first time a new disk is seen,
2350 * performs disk spin up, read_capacity, etc.
2351 * @disk: struct gendisk we care about
2352 **/
2353static int sd_revalidate_disk(struct gendisk *disk)
2354{
2355 struct scsi_disk *sdkp = scsi_disk(disk);
2356 struct scsi_device *sdp = sdkp->device;
2357 unsigned char *buffer;
2358 unsigned flush = 0;
2359
2360 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2361 "sd_revalidate_disk\n"));
2362
2363 /*
2364 * If the device is offline, don't try and read capacity or any
2365 * of the other niceties.
2366 */
2367 if (!scsi_device_online(sdp))
2368 goto out;
2369
2370 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2371 if (!buffer) {
2372 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2373 "allocation failure.\n");
2374 goto out;
2375 }
2376
2377 sd_spinup_disk(sdkp);
2378
2379 /*
2380 * Without media there is no reason to ask; moreover, some devices
2381 * react badly if we do.
2382 */
2383 if (sdkp->media_present) {
2384 sd_read_capacity(sdkp, buffer);
2385
2386 if (sd_try_extended_inquiry(sdp)) {
2387 sd_read_block_provisioning(sdkp);
2388 sd_read_block_limits(sdkp);
2389 sd_read_block_characteristics(sdkp);
2390 }
2391
2392 sd_read_write_protect_flag(sdkp, buffer);
2393 sd_read_cache_type(sdkp, buffer);
2394 sd_read_app_tag_own(sdkp, buffer);
2395 }
2396
2397 sdkp->first_scan = 0;
2398
2399 /*
2400 * We now have all cache related info, determine how we deal
2401 * with flush requests.
2402 */
2403 if (sdkp->WCE) {
2404 flush |= REQ_FLUSH;
2405 if (sdkp->DPOFUA)
2406 flush |= REQ_FUA;
2407 }
2408
2409 blk_queue_flush(sdkp->disk->queue, flush);
2410
2411 set_capacity(disk, sdkp->capacity);
2412 kfree(buffer);
2413
2414 out:
2415 return 0;
2416}
2417
2418/**
2419 * sd_unlock_native_capacity - unlock native capacity
2420 * @disk: struct gendisk to set capacity for
2421 *
2422 * Block layer calls this function if it detects that partitions
2423 * on @disk reach beyond the end of the device. If the SCSI host
2424 * implements ->unlock_native_capacity() method, it's invoked to
2425 * give it a chance to adjust the device capacity.
2426 *
2427 * CONTEXT:
2428 * Defined by block layer. Might sleep.
2429 */
2430static void sd_unlock_native_capacity(struct gendisk *disk)
2431{
2432 struct scsi_device *sdev = scsi_disk(disk)->device;
2433
2434 if (sdev->host->hostt->unlock_native_capacity)
2435 sdev->host->hostt->unlock_native_capacity(sdev);
2436}
2437
2438/**
2439 * sd_format_disk_name - format disk name
2440 * @prefix: name prefix - ie. "sd" for SCSI disks
2441 * @index: index of the disk to format name for
2442 * @buf: output buffer
2443 * @buflen: length of the output buffer
2444 *
2445 * SCSI disk names starts at sda. The 26th device is sdz and the
2446 * 27th is sdaa. The last one for two lettered suffix is sdzz
2447 * which is followed by sdaaa.
2448 *
2449 * This is basically 26 base counting with one extra 'nil' entry
2450 * at the beginning from the second digit on and can be
2451 * determined using similar method as 26 base conversion with the
2452 * index shifted -1 after each digit is computed.
2453 *
2454 * CONTEXT:
2455 * Don't care.
2456 *
2457 * RETURNS:
2458 * 0 on success, -errno on failure.
2459 */
2460static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2461{
2462 const int base = 'z' - 'a' + 1;
2463 char *begin = buf + strlen(prefix);
2464 char *end = buf + buflen;
2465 char *p;
2466 int unit;
2467
2468 p = end - 1;
2469 *p = '\0';
2470 unit = base;
2471 do {
2472 if (p == begin)
2473 return -EINVAL;
2474 *--p = 'a' + (index % unit);
2475 index = (index / unit) - 1;
2476 } while (index >= 0);
2477
2478 memmove(begin, p, end - p);
2479 memcpy(buf, prefix, strlen(prefix));
2480
2481 return 0;
2482}
2483
2484/*
2485 * The asynchronous part of sd_probe
2486 */
2487static void sd_probe_async(void *data, async_cookie_t cookie)
2488{
2489 struct scsi_disk *sdkp = data;
2490 struct scsi_device *sdp;
2491 struct gendisk *gd;
2492 u32 index;
2493 struct device *dev;
2494
2495 sdp = sdkp->device;
2496 gd = sdkp->disk;
2497 index = sdkp->index;
2498 dev = &sdp->sdev_gendev;
2499
2500 gd->major = sd_major((index & 0xf0) >> 4);
2501 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2502 gd->minors = SD_MINORS;
2503
2504 gd->fops = &sd_fops;
2505 gd->private_data = &sdkp->driver;
2506 gd->queue = sdkp->device->request_queue;
2507
2508 /* defaults, until the device tells us otherwise */
2509 sdp->sector_size = 512;
2510 sdkp->capacity = 0;
2511 sdkp->media_present = 1;
2512 sdkp->write_prot = 0;
2513 sdkp->WCE = 0;
2514 sdkp->RCD = 0;
2515 sdkp->ATO = 0;
2516 sdkp->first_scan = 1;
2517
2518 sd_revalidate_disk(gd);
2519
2520 blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2521 blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2522
2523 gd->driverfs_dev = &sdp->sdev_gendev;
2524 gd->flags = GENHD_FL_EXT_DEVT;
2525 if (sdp->removable) {
2526 gd->flags |= GENHD_FL_REMOVABLE;
2527 gd->events |= DISK_EVENT_MEDIA_CHANGE;
2528 }
2529
2530 add_disk(gd);
2531 sd_dif_config_host(sdkp);
2532
2533 sd_revalidate_disk(gd);
2534
2535 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2536 sdp->removable ? "removable " : "");
2537 scsi_autopm_put_device(sdp);
2538 put_device(&sdkp->dev);
2539}
2540
2541/**
2542 * sd_probe - called during driver initialization and whenever a
2543 * new scsi device is attached to the system. It is called once
2544 * for each scsi device (not just disks) present.
2545 * @dev: pointer to device object
2546 *
2547 * Returns 0 if successful (or not interested in this scsi device
2548 * (e.g. scanner)); 1 when there is an error.
2549 *
2550 * Note: this function is invoked from the scsi mid-level.
2551 * This function sets up the mapping between a given
2552 * <host,channel,id,lun> (found in sdp) and new device name
2553 * (e.g. /dev/sda). More precisely it is the block device major
2554 * and minor number that is chosen here.
2555 *
2556 * Assume sd_attach is not re-entrant (for time being)
2557 * Also think about sd_attach() and sd_remove() running coincidentally.
2558 **/
2559static int sd_probe(struct device *dev)
2560{
2561 struct scsi_device *sdp = to_scsi_device(dev);
2562 struct scsi_disk *sdkp;
2563 struct gendisk *gd;
2564 int index;
2565 int error;
2566
2567 error = -ENODEV;
2568 if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2569 goto out;
2570
2571 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2572 "sd_attach\n"));
2573
2574 error = -ENOMEM;
2575 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2576 if (!sdkp)
2577 goto out;
2578
2579 gd = alloc_disk(SD_MINORS);
2580 if (!gd)
2581 goto out_free;
2582
2583 do {
2584 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2585 goto out_put;
2586
2587 spin_lock(&sd_index_lock);
2588 error = ida_get_new(&sd_index_ida, &index);
2589 spin_unlock(&sd_index_lock);
2590 } while (error == -EAGAIN);
2591
2592 if (error)
2593 goto out_put;
2594
2595 if (index >= SD_MAX_DISKS) {
2596 error = -ENODEV;
2597 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name space exhausted.\n");
2598 goto out_free_index;
2599 }
2600
2601 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2602 if (error)
2603 goto out_free_index;
2604
2605 sdkp->device = sdp;
2606 sdkp->driver = &sd_template;
2607 sdkp->disk = gd;
2608 sdkp->index = index;
2609 atomic_set(&sdkp->openers, 0);
2610
2611 if (!sdp->request_queue->rq_timeout) {
2612 if (sdp->type != TYPE_MOD)
2613 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2614 else
2615 blk_queue_rq_timeout(sdp->request_queue,
2616 SD_MOD_TIMEOUT);
2617 }
2618
2619 device_initialize(&sdkp->dev);
2620 sdkp->dev.parent = dev;
2621 sdkp->dev.class = &sd_disk_class;
2622 dev_set_name(&sdkp->dev, dev_name(dev));
2623
2624 if (device_add(&sdkp->dev))
2625 goto out_free_index;
2626
2627 get_device(dev);
2628 dev_set_drvdata(dev, sdkp);
2629
2630 get_device(&sdkp->dev); /* prevent release before async_schedule */
2631 async_schedule(sd_probe_async, sdkp);
2632
2633 return 0;
2634
2635 out_free_index:
2636 spin_lock(&sd_index_lock);
2637 ida_remove(&sd_index_ida, index);
2638 spin_unlock(&sd_index_lock);
2639 out_put:
2640 put_disk(gd);
2641 out_free:
2642 kfree(sdkp);
2643 out:
2644 return error;
2645}
2646
2647/**
2648 * sd_remove - called whenever a scsi disk (previously recognized by
2649 * sd_probe) is detached from the system. It is called (potentially
2650 * multiple times) during sd module unload.
2651 * @sdp: pointer to mid level scsi device object
2652 *
2653 * Note: this function is invoked from the scsi mid-level.
2654 * This function potentially frees up a device name (e.g. /dev/sdc)
2655 * that could be re-used by a subsequent sd_probe().
2656 * This function is not called when the built-in sd driver is "exit-ed".
2657 **/
2658static int sd_remove(struct device *dev)
2659{
2660 struct scsi_disk *sdkp;
2661
2662 sdkp = dev_get_drvdata(dev);
2663 scsi_autopm_get_device(sdkp->device);
2664
2665 async_synchronize_full();
2666 blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2667 blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2668 device_del(&sdkp->dev);
2669 del_gendisk(sdkp->disk);
2670 sd_shutdown(dev);
2671
2672 mutex_lock(&sd_ref_mutex);
2673 dev_set_drvdata(dev, NULL);
2674 put_device(&sdkp->dev);
2675 mutex_unlock(&sd_ref_mutex);
2676
2677 return 0;
2678}
2679
2680/**
2681 * scsi_disk_release - Called to free the scsi_disk structure
2682 * @dev: pointer to embedded class device
2683 *
2684 * sd_ref_mutex must be held entering this routine. Because it is
2685 * called on last put, you should always use the scsi_disk_get()
2686 * scsi_disk_put() helpers which manipulate the semaphore directly
2687 * and never do a direct put_device.
2688 **/
2689static void scsi_disk_release(struct device *dev)
2690{
2691 struct scsi_disk *sdkp = to_scsi_disk(dev);
2692 struct gendisk *disk = sdkp->disk;
2693
2694 spin_lock(&sd_index_lock);
2695 ida_remove(&sd_index_ida, sdkp->index);
2696 spin_unlock(&sd_index_lock);
2697
2698 disk->private_data = NULL;
2699 put_disk(disk);
2700 put_device(&sdkp->device->sdev_gendev);
2701
2702 kfree(sdkp);
2703}
2704
2705static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2706{
2707 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
2708 struct scsi_sense_hdr sshdr;
2709 struct scsi_device *sdp = sdkp->device;
2710 int res;
2711
2712 if (start)
2713 cmd[4] |= 1; /* START */
2714
2715 if (sdp->start_stop_pwr_cond)
2716 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
2717
2718 if (!scsi_device_online(sdp))
2719 return -ENODEV;
2720
2721 res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2722 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2723 if (res) {
2724 sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2725 sd_print_result(sdkp, res);
2726 if (driver_byte(res) & DRIVER_SENSE)
2727 sd_print_sense_hdr(sdkp, &sshdr);
2728 }
2729
2730 return res;
2731}
2732
2733/*
2734 * Send a SYNCHRONIZE CACHE instruction down to the device through
2735 * the normal SCSI command structure. Wait for the command to
2736 * complete.
2737 */
2738static void sd_shutdown(struct device *dev)
2739{
2740 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2741
2742 if (!sdkp)
2743 return; /* this can happen */
2744
2745 if (sdkp->WCE) {
2746 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2747 sd_sync_cache(sdkp);
2748 }
2749
2750 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2751 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2752 sd_start_stop_device(sdkp, 0);
2753 }
2754
2755 scsi_disk_put(sdkp);
2756}
2757
2758static int sd_suspend(struct device *dev, pm_message_t mesg)
2759{
2760 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2761 int ret = 0;
2762
2763 if (!sdkp)
2764 return 0; /* this can happen */
2765
2766 if (sdkp->WCE) {
2767 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2768 ret = sd_sync_cache(sdkp);
2769 if (ret)
2770 goto done;
2771 }
2772
2773 if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2774 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2775 ret = sd_start_stop_device(sdkp, 0);
2776 }
2777
2778done:
2779 scsi_disk_put(sdkp);
2780 return ret;
2781}
2782
2783static int sd_resume(struct device *dev)
2784{
2785 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2786 int ret = 0;
2787
2788 if (!sdkp->device->manage_start_stop)
2789 goto done;
2790
2791 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2792 ret = sd_start_stop_device(sdkp, 1);
2793
2794done:
2795 scsi_disk_put(sdkp);
2796 return ret;
2797}
2798
2799/**
2800 * init_sd - entry point for this driver (both when built in or when
2801 * a module).
2802 *
2803 * Note: this function registers this driver with the scsi mid-level.
2804 **/
2805static int __init init_sd(void)
2806{
2807 int majors = 0, i, err;
2808
2809 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2810
2811 for (i = 0; i < SD_MAJORS; i++)
2812 if (register_blkdev(sd_major(i), "sd") == 0)
2813 majors++;
2814
2815 if (!majors)
2816 return -ENODEV;
2817
2818 err = class_register(&sd_disk_class);
2819 if (err)
2820 goto err_out;
2821
2822 err = scsi_register_driver(&sd_template.gendrv);
2823 if (err)
2824 goto err_out_class;
2825
2826 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2827 0, 0, NULL);
2828 if (!sd_cdb_cache) {
2829 printk(KERN_ERR "sd: can't init extended cdb cache\n");
2830 goto err_out_class;
2831 }
2832
2833 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2834 if (!sd_cdb_pool) {
2835 printk(KERN_ERR "sd: can't init extended cdb pool\n");
2836 goto err_out_cache;
2837 }
2838
2839 return 0;
2840
2841err_out_cache:
2842 kmem_cache_destroy(sd_cdb_cache);
2843
2844err_out_class:
2845 class_unregister(&sd_disk_class);
2846err_out:
2847 for (i = 0; i < SD_MAJORS; i++)
2848 unregister_blkdev(sd_major(i), "sd");
2849 return err;
2850}
2851
2852/**
2853 * exit_sd - exit point for this driver (when it is a module).
2854 *
2855 * Note: this function unregisters this driver from the scsi mid-level.
2856 **/
2857static void __exit exit_sd(void)
2858{
2859 int i;
2860
2861 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
2862
2863 mempool_destroy(sd_cdb_pool);
2864 kmem_cache_destroy(sd_cdb_cache);
2865
2866 scsi_unregister_driver(&sd_template.gendrv);
2867 class_unregister(&sd_disk_class);
2868
2869 for (i = 0; i < SD_MAJORS; i++)
2870 unregister_blkdev(sd_major(i), "sd");
2871}
2872
2873module_init(init_sd);
2874module_exit(exit_sd);
2875
2876static void sd_print_sense_hdr(struct scsi_disk *sdkp,
2877 struct scsi_sense_hdr *sshdr)
2878{
2879 sd_printk(KERN_INFO, sdkp, " ");
2880 scsi_show_sense_hdr(sshdr);
2881 sd_printk(KERN_INFO, sdkp, " ");
2882 scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
2883}
2884
2885static void sd_print_result(struct scsi_disk *sdkp, int result)
2886{
2887 sd_printk(KERN_INFO, sdkp, " ");
2888 scsi_show_result(result);
2889}
2890