?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dmaengine: imx-sdma: Let the core do the device node validation
[linux/fpc-iii.git] / drivers / scsi / sd.c
blobebc80354714c57de7e5cfda37ce3611e7d238d11
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/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/mutex.h>
51 #include <linux/string_helpers.h>
52 #include <linux/async.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
71 #include "sd.h"
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
74
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
78
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99
100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
101 #define SD_MINORS 16
102 #else
103 #define SD_MINORS 0
104 #endif
105
106 static void sd_config_discard(struct scsi_disk *, unsigned int);
107 static void sd_config_write_same(struct scsi_disk *);
108 static int sd_revalidate_disk(struct gendisk *);
109 static void sd_unlock_native_capacity(struct gendisk *disk);
110 static int sd_probe(struct device *);
111 static int sd_remove(struct device *);
112 static void sd_shutdown(struct device *);
113 static int sd_suspend_system(struct device *);
114 static int sd_suspend_runtime(struct device *);
115 static int sd_resume(struct device *);
116 static void sd_rescan(struct device *);
117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
119 static int sd_done(struct scsi_cmnd *);
120 static void sd_eh_reset(struct scsi_cmnd *);
121 static int sd_eh_action(struct scsi_cmnd *, int);
122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
123 static void scsi_disk_release(struct device *cdev);
124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
125 static void sd_print_result(const struct scsi_disk *, const char *, int);
126
127 static DEFINE_IDA(sd_index_ida);
128
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
133
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
136 static mempool_t *sd_page_pool;
137
138 static const char *sd_cache_types[] = {
139 "write through", "none", "write back",
140 "write back, no read (daft)"
141 };
142
143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 {
145 bool wc = false, fua = false;
146
147 if (sdkp->WCE) {
148 wc = true;
149 if (sdkp->DPOFUA)
150 fua = true;
151 }
152
153 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
154 }
155
156 static ssize_t
157 cache_type_store(struct device *dev, struct device_attribute *attr,
158 const char *buf, size_t count)
159 {
160 int ct, rcd, wce, sp;
161 struct scsi_disk *sdkp = to_scsi_disk(dev);
162 struct scsi_device *sdp = sdkp->device;
163 char buffer[64];
164 char *buffer_data;
165 struct scsi_mode_data data;
166 struct scsi_sense_hdr sshdr;
167 static const char temp[] = "temporary ";
168 int len;
169
170 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
171 /* no cache control on RBC devices; theoretically they
172 * can do it, but there's probably so many exceptions
173 * it's not worth the risk */
174 return -EINVAL;
175
176 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
177 buf += sizeof(temp) - 1;
178 sdkp->cache_override = 1;
179 } else {
180 sdkp->cache_override = 0;
181 }
182
183 ct = sysfs_match_string(sd_cache_types, buf);
184 if (ct < 0)
185 return -EINVAL;
186
187 rcd = ct & 0x01 ? 1 : 0;
188 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
189
190 if (sdkp->cache_override) {
191 sdkp->WCE = wce;
192 sdkp->RCD = rcd;
193 sd_set_flush_flag(sdkp);
194 return count;
195 }
196
197 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198 SD_MAX_RETRIES, &data, NULL))
199 return -EINVAL;
200 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201 data.block_descriptor_length);
202 buffer_data = buffer + data.header_length +
203 data.block_descriptor_length;
204 buffer_data[2] &= ~0x05;
205 buffer_data[2] |= wce << 2 | rcd;
206 sp = buffer_data[0] & 0x80 ? 1 : 0;
207 buffer_data[0] &= ~0x80;
208
209 /*
210 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
211 * received mode parameter buffer before doing MODE SELECT.
212 */
213 data.device_specific = 0;
214
215 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
216 SD_MAX_RETRIES, &data, &sshdr)) {
217 if (scsi_sense_valid(&sshdr))
218 sd_print_sense_hdr(sdkp, &sshdr);
219 return -EINVAL;
220 }
221 revalidate_disk(sdkp->disk);
222 return count;
223 }
224
225 static ssize_t
226 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
227 char *buf)
228 {
229 struct scsi_disk *sdkp = to_scsi_disk(dev);
230 struct scsi_device *sdp = sdkp->device;
231
232 return sprintf(buf, "%u\n", sdp->manage_start_stop);
233 }
234
235 static ssize_t
236 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
237 const char *buf, size_t count)
238 {
239 struct scsi_disk *sdkp = to_scsi_disk(dev);
240 struct scsi_device *sdp = sdkp->device;
241 bool v;
242
243 if (!capable(CAP_SYS_ADMIN))
244 return -EACCES;
245
246 if (kstrtobool(buf, &v))
247 return -EINVAL;
248
249 sdp->manage_start_stop = v;
250
251 return count;
252 }
253 static DEVICE_ATTR_RW(manage_start_stop);
254
255 static ssize_t
256 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
257 {
258 struct scsi_disk *sdkp = to_scsi_disk(dev);
259
260 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
261 }
262
263 static ssize_t
264 allow_restart_store(struct device *dev, struct device_attribute *attr,
265 const char *buf, size_t count)
266 {
267 bool v;
268 struct scsi_disk *sdkp = to_scsi_disk(dev);
269 struct scsi_device *sdp = sdkp->device;
270
271 if (!capable(CAP_SYS_ADMIN))
272 return -EACCES;
273
274 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
275 return -EINVAL;
276
277 if (kstrtobool(buf, &v))
278 return -EINVAL;
279
280 sdp->allow_restart = v;
281
282 return count;
283 }
284 static DEVICE_ATTR_RW(allow_restart);
285
286 static ssize_t
287 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
288 {
289 struct scsi_disk *sdkp = to_scsi_disk(dev);
290 int ct = sdkp->RCD + 2*sdkp->WCE;
291
292 return sprintf(buf, "%s\n", sd_cache_types[ct]);
293 }
294 static DEVICE_ATTR_RW(cache_type);
295
296 static ssize_t
297 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
298 {
299 struct scsi_disk *sdkp = to_scsi_disk(dev);
300
301 return sprintf(buf, "%u\n", sdkp->DPOFUA);
302 }
303 static DEVICE_ATTR_RO(FUA);
304
305 static ssize_t
306 protection_type_show(struct device *dev, struct device_attribute *attr,
307 char *buf)
308 {
309 struct scsi_disk *sdkp = to_scsi_disk(dev);
310
311 return sprintf(buf, "%u\n", sdkp->protection_type);
312 }
313
314 static ssize_t
315 protection_type_store(struct device *dev, struct device_attribute *attr,
316 const char *buf, size_t count)
317 {
318 struct scsi_disk *sdkp = to_scsi_disk(dev);
319 unsigned int val;
320 int err;
321
322 if (!capable(CAP_SYS_ADMIN))
323 return -EACCES;
324
325 err = kstrtouint(buf, 10, &val);
326
327 if (err)
328 return err;
329
330 if (val <= T10_PI_TYPE3_PROTECTION)
331 sdkp->protection_type = val;
332
333 return count;
334 }
335 static DEVICE_ATTR_RW(protection_type);
336
337 static ssize_t
338 protection_mode_show(struct device *dev, struct device_attribute *attr,
339 char *buf)
340 {
341 struct scsi_disk *sdkp = to_scsi_disk(dev);
342 struct scsi_device *sdp = sdkp->device;
343 unsigned int dif, dix;
344
345 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
346 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
347
348 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
349 dif = 0;
350 dix = 1;
351 }
352
353 if (!dif && !dix)
354 return sprintf(buf, "none\n");
355
356 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
357 }
358 static DEVICE_ATTR_RO(protection_mode);
359
360 static ssize_t
361 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
362 {
363 struct scsi_disk *sdkp = to_scsi_disk(dev);
364
365 return sprintf(buf, "%u\n", sdkp->ATO);
366 }
367 static DEVICE_ATTR_RO(app_tag_own);
368
369 static ssize_t
370 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
371 char *buf)
372 {
373 struct scsi_disk *sdkp = to_scsi_disk(dev);
374
375 return sprintf(buf, "%u\n", sdkp->lbpme);
376 }
377 static DEVICE_ATTR_RO(thin_provisioning);
378
379 /* sysfs_match_string() requires dense arrays */
380 static const char *lbp_mode[] = {
381 [SD_LBP_FULL] = "full",
382 [SD_LBP_UNMAP] = "unmap",
383 [SD_LBP_WS16] = "writesame_16",
384 [SD_LBP_WS10] = "writesame_10",
385 [SD_LBP_ZERO] = "writesame_zero",
386 [SD_LBP_DISABLE] = "disabled",
387 };
388
389 static ssize_t
390 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
391 char *buf)
392 {
393 struct scsi_disk *sdkp = to_scsi_disk(dev);
394
395 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
396 }
397
398 static ssize_t
399 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
400 const char *buf, size_t count)
401 {
402 struct scsi_disk *sdkp = to_scsi_disk(dev);
403 struct scsi_device *sdp = sdkp->device;
404 int mode;
405
406 if (!capable(CAP_SYS_ADMIN))
407 return -EACCES;
408
409 if (sd_is_zoned(sdkp)) {
410 sd_config_discard(sdkp, SD_LBP_DISABLE);
411 return count;
412 }
413
414 if (sdp->type != TYPE_DISK)
415 return -EINVAL;
416
417 mode = sysfs_match_string(lbp_mode, buf);
418 if (mode < 0)
419 return -EINVAL;
420
421 sd_config_discard(sdkp, mode);
422
423 return count;
424 }
425 static DEVICE_ATTR_RW(provisioning_mode);
426
427 /* sysfs_match_string() requires dense arrays */
428 static const char *zeroing_mode[] = {
429 [SD_ZERO_WRITE] = "write",
430 [SD_ZERO_WS] = "writesame",
431 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
432 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
433 };
434
435 static ssize_t
436 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
437 char *buf)
438 {
439 struct scsi_disk *sdkp = to_scsi_disk(dev);
440
441 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
442 }
443
444 static ssize_t
445 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
446 const char *buf, size_t count)
447 {
448 struct scsi_disk *sdkp = to_scsi_disk(dev);
449 int mode;
450
451 if (!capable(CAP_SYS_ADMIN))
452 return -EACCES;
453
454 mode = sysfs_match_string(zeroing_mode, buf);
455 if (mode < 0)
456 return -EINVAL;
457
458 sdkp->zeroing_mode = mode;
459
460 return count;
461 }
462 static DEVICE_ATTR_RW(zeroing_mode);
463
464 static ssize_t
465 max_medium_access_timeouts_show(struct device *dev,
466 struct device_attribute *attr, char *buf)
467 {
468 struct scsi_disk *sdkp = to_scsi_disk(dev);
469
470 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
471 }
472
473 static ssize_t
474 max_medium_access_timeouts_store(struct device *dev,
475 struct device_attribute *attr, const char *buf,
476 size_t count)
477 {
478 struct scsi_disk *sdkp = to_scsi_disk(dev);
479 int err;
480
481 if (!capable(CAP_SYS_ADMIN))
482 return -EACCES;
483
484 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
485
486 return err ? err : count;
487 }
488 static DEVICE_ATTR_RW(max_medium_access_timeouts);
489
490 static ssize_t
491 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
492 char *buf)
493 {
494 struct scsi_disk *sdkp = to_scsi_disk(dev);
495
496 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
497 }
498
499 static ssize_t
500 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
501 const char *buf, size_t count)
502 {
503 struct scsi_disk *sdkp = to_scsi_disk(dev);
504 struct scsi_device *sdp = sdkp->device;
505 unsigned long max;
506 int err;
507
508 if (!capable(CAP_SYS_ADMIN))
509 return -EACCES;
510
511 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
512 return -EINVAL;
513
514 err = kstrtoul(buf, 10, &max);
515
516 if (err)
517 return err;
518
519 if (max == 0)
520 sdp->no_write_same = 1;
521 else if (max <= SD_MAX_WS16_BLOCKS) {
522 sdp->no_write_same = 0;
523 sdkp->max_ws_blocks = max;
524 }
525
526 sd_config_write_same(sdkp);
527
528 return count;
529 }
530 static DEVICE_ATTR_RW(max_write_same_blocks);
531
532 static struct attribute *sd_disk_attrs[] = {
533 &dev_attr_cache_type.attr,
534 &dev_attr_FUA.attr,
535 &dev_attr_allow_restart.attr,
536 &dev_attr_manage_start_stop.attr,
537 &dev_attr_protection_type.attr,
538 &dev_attr_protection_mode.attr,
539 &dev_attr_app_tag_own.attr,
540 &dev_attr_thin_provisioning.attr,
541 &dev_attr_provisioning_mode.attr,
542 &dev_attr_zeroing_mode.attr,
543 &dev_attr_max_write_same_blocks.attr,
544 &dev_attr_max_medium_access_timeouts.attr,
545 NULL,
546 };
547 ATTRIBUTE_GROUPS(sd_disk);
548
549 static struct class sd_disk_class = {
550 .name = "scsi_disk",
551 .owner = THIS_MODULE,
552 .dev_release = scsi_disk_release,
553 .dev_groups = sd_disk_groups,
554 };
555
556 static const struct dev_pm_ops sd_pm_ops = {
557 .suspend = sd_suspend_system,
558 .resume = sd_resume,
559 .poweroff = sd_suspend_system,
560 .restore = sd_resume,
561 .runtime_suspend = sd_suspend_runtime,
562 .runtime_resume = sd_resume,
563 };
564
565 static struct scsi_driver sd_template = {
566 .gendrv = {
567 .name = "sd",
568 .owner = THIS_MODULE,
569 .probe = sd_probe,
570 .remove = sd_remove,
571 .shutdown = sd_shutdown,
572 .pm = &sd_pm_ops,
573 },
574 .rescan = sd_rescan,
575 .init_command = sd_init_command,
576 .uninit_command = sd_uninit_command,
577 .done = sd_done,
578 .eh_action = sd_eh_action,
579 .eh_reset = sd_eh_reset,
580 };
581
582 /*
583 * Dummy kobj_map->probe function.
584 * The default ->probe function will call modprobe, which is
585 * pointless as this module is already loaded.
586 */
587 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
588 {
589 return NULL;
590 }
591
592 /*
593 * Device no to disk mapping:
594 *
595 * major disc2 disc p1
596 * |............|.............|....|....| <- dev_t
597 * 31 20 19 8 7 4 3 0
598 *
599 * Inside a major, we have 16k disks, however mapped non-
600 * contiguously. The first 16 disks are for major0, the next
601 * ones with major1, ... Disk 256 is for major0 again, disk 272
602 * for major1, ...
603 * As we stay compatible with our numbering scheme, we can reuse
604 * the well-know SCSI majors 8, 65--71, 136--143.
605 */
606 static int sd_major(int major_idx)
607 {
608 switch (major_idx) {
609 case 0:
610 return SCSI_DISK0_MAJOR;
611 case 1 ... 7:
612 return SCSI_DISK1_MAJOR + major_idx - 1;
613 case 8 ... 15:
614 return SCSI_DISK8_MAJOR + major_idx - 8;
615 default:
616 BUG();
617 return 0; /* shut up gcc */
618 }
619 }
620
621 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
622 {
623 struct scsi_disk *sdkp = NULL;
624
625 mutex_lock(&sd_ref_mutex);
626
627 if (disk->private_data) {
628 sdkp = scsi_disk(disk);
629 if (scsi_device_get(sdkp->device) == 0)
630 get_device(&sdkp->dev);
631 else
632 sdkp = NULL;
633 }
634 mutex_unlock(&sd_ref_mutex);
635 return sdkp;
636 }
637
638 static void scsi_disk_put(struct scsi_disk *sdkp)
639 {
640 struct scsi_device *sdev = sdkp->device;
641
642 mutex_lock(&sd_ref_mutex);
643 put_device(&sdkp->dev);
644 scsi_device_put(sdev);
645 mutex_unlock(&sd_ref_mutex);
646 }
647
648 #ifdef CONFIG_BLK_SED_OPAL
649 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
650 size_t len, bool send)
651 {
652 struct scsi_device *sdev = data;
653 u8 cdb[12] = { 0, };
654 int ret;
655
656 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
657 cdb[1] = secp;
658 put_unaligned_be16(spsp, &cdb[2]);
659 put_unaligned_be32(len, &cdb[6]);
660
661 ret = scsi_execute_req(sdev, cdb,
662 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
663 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
664 return ret <= 0 ? ret : -EIO;
665 }
666 #endif /* CONFIG_BLK_SED_OPAL */
667
668 /*
669 * Look up the DIX operation based on whether the command is read or
670 * write and whether dix and dif are enabled.
671 */
672 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
673 {
674 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
675 static const unsigned int ops[] = { /* wrt dix dif */
676 SCSI_PROT_NORMAL, /* 0 0 0 */
677 SCSI_PROT_READ_STRIP, /* 0 0 1 */
678 SCSI_PROT_READ_INSERT, /* 0 1 0 */
679 SCSI_PROT_READ_PASS, /* 0 1 1 */
680 SCSI_PROT_NORMAL, /* 1 0 0 */
681 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
682 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
683 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
684 };
685
686 return ops[write << 2 | dix << 1 | dif];
687 }
688
689 /*
690 * Returns a mask of the protection flags that are valid for a given DIX
691 * operation.
692 */
693 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
694 {
695 static const unsigned int flag_mask[] = {
696 [SCSI_PROT_NORMAL] = 0,
697
698 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
699 SCSI_PROT_GUARD_CHECK |
700 SCSI_PROT_REF_CHECK |
701 SCSI_PROT_REF_INCREMENT,
702
703 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
704 SCSI_PROT_IP_CHECKSUM,
705
706 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
707 SCSI_PROT_GUARD_CHECK |
708 SCSI_PROT_REF_CHECK |
709 SCSI_PROT_REF_INCREMENT |
710 SCSI_PROT_IP_CHECKSUM,
711
712 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
713 SCSI_PROT_REF_INCREMENT,
714
715 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
716 SCSI_PROT_REF_CHECK |
717 SCSI_PROT_REF_INCREMENT |
718 SCSI_PROT_IP_CHECKSUM,
719
720 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
721 SCSI_PROT_GUARD_CHECK |
722 SCSI_PROT_REF_CHECK |
723 SCSI_PROT_REF_INCREMENT |
724 SCSI_PROT_IP_CHECKSUM,
725 };
726
727 return flag_mask[prot_op];
728 }
729
730 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
731 unsigned int dix, unsigned int dif)
732 {
733 struct bio *bio = scmd->request->bio;
734 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
735 unsigned int protect = 0;
736
737 if (dix) { /* DIX Type 0, 1, 2, 3 */
738 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
739 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
740
741 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
742 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
743 }
744
745 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
746 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
747
748 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
749 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
750 }
751
752 if (dif) { /* DIX/DIF Type 1, 2, 3 */
753 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
754
755 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
756 protect = 3 << 5; /* Disable target PI checking */
757 else
758 protect = 1 << 5; /* Enable target PI checking */
759 }
760
761 scsi_set_prot_op(scmd, prot_op);
762 scsi_set_prot_type(scmd, dif);
763 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
764
765 return protect;
766 }
767
768 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
769 {
770 struct request_queue *q = sdkp->disk->queue;
771 unsigned int logical_block_size = sdkp->device->sector_size;
772 unsigned int max_blocks = 0;
773
774 q->limits.discard_alignment =
775 sdkp->unmap_alignment * logical_block_size;
776 q->limits.discard_granularity =
777 max(sdkp->physical_block_size,
778 sdkp->unmap_granularity * logical_block_size);
779 sdkp->provisioning_mode = mode;
780
781 switch (mode) {
782
783 case SD_LBP_FULL:
784 case SD_LBP_DISABLE:
785 blk_queue_max_discard_sectors(q, 0);
786 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
787 return;
788
789 case SD_LBP_UNMAP:
790 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
791 (u32)SD_MAX_WS16_BLOCKS);
792 break;
793
794 case SD_LBP_WS16:
795 if (sdkp->device->unmap_limit_for_ws)
796 max_blocks = sdkp->max_unmap_blocks;
797 else
798 max_blocks = sdkp->max_ws_blocks;
799
800 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
801 break;
802
803 case SD_LBP_WS10:
804 if (sdkp->device->unmap_limit_for_ws)
805 max_blocks = sdkp->max_unmap_blocks;
806 else
807 max_blocks = sdkp->max_ws_blocks;
808
809 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
810 break;
811
812 case SD_LBP_ZERO:
813 max_blocks = min_not_zero(sdkp->max_ws_blocks,
814 (u32)SD_MAX_WS10_BLOCKS);
815 break;
816 }
817
818 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
819 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
820 }
821
822 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
823 {
824 struct scsi_device *sdp = cmd->device;
825 struct request *rq = cmd->request;
826 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
827 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
828 unsigned int data_len = 24;
829 char *buf;
830
831 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
832 if (!rq->special_vec.bv_page)
833 return BLK_STS_RESOURCE;
834 clear_highpage(rq->special_vec.bv_page);
835 rq->special_vec.bv_offset = 0;
836 rq->special_vec.bv_len = data_len;
837 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
838
839 cmd->cmd_len = 10;
840 cmd->cmnd[0] = UNMAP;
841 cmd->cmnd[8] = 24;
842
843 buf = page_address(rq->special_vec.bv_page);
844 put_unaligned_be16(6 + 16, &buf[0]);
845 put_unaligned_be16(16, &buf[2]);
846 put_unaligned_be64(lba, &buf[8]);
847 put_unaligned_be32(nr_blocks, &buf[16]);
848
849 cmd->allowed = SD_MAX_RETRIES;
850 cmd->transfersize = data_len;
851 rq->timeout = SD_TIMEOUT;
852
853 return scsi_init_io(cmd);
854 }
855
856 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
857 bool unmap)
858 {
859 struct scsi_device *sdp = cmd->device;
860 struct request *rq = cmd->request;
861 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
862 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
863 u32 data_len = sdp->sector_size;
864
865 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
866 if (!rq->special_vec.bv_page)
867 return BLK_STS_RESOURCE;
868 clear_highpage(rq->special_vec.bv_page);
869 rq->special_vec.bv_offset = 0;
870 rq->special_vec.bv_len = data_len;
871 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
872
873 cmd->cmd_len = 16;
874 cmd->cmnd[0] = WRITE_SAME_16;
875 if (unmap)
876 cmd->cmnd[1] = 0x8; /* UNMAP */
877 put_unaligned_be64(lba, &cmd->cmnd[2]);
878 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
879
880 cmd->allowed = SD_MAX_RETRIES;
881 cmd->transfersize = data_len;
882 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
883
884 return scsi_init_io(cmd);
885 }
886
887 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
888 bool unmap)
889 {
890 struct scsi_device *sdp = cmd->device;
891 struct request *rq = cmd->request;
892 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
893 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
894 u32 data_len = sdp->sector_size;
895
896 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
897 if (!rq->special_vec.bv_page)
898 return BLK_STS_RESOURCE;
899 clear_highpage(rq->special_vec.bv_page);
900 rq->special_vec.bv_offset = 0;
901 rq->special_vec.bv_len = data_len;
902 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
903
904 cmd->cmd_len = 10;
905 cmd->cmnd[0] = WRITE_SAME;
906 if (unmap)
907 cmd->cmnd[1] = 0x8; /* UNMAP */
908 put_unaligned_be32(lba, &cmd->cmnd[2]);
909 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
910
911 cmd->allowed = SD_MAX_RETRIES;
912 cmd->transfersize = data_len;
913 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
914
915 return scsi_init_io(cmd);
916 }
917
918 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
919 {
920 struct request *rq = cmd->request;
921 struct scsi_device *sdp = cmd->device;
922 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
923 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
924 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
925
926 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
927 switch (sdkp->zeroing_mode) {
928 case SD_ZERO_WS16_UNMAP:
929 return sd_setup_write_same16_cmnd(cmd, true);
930 case SD_ZERO_WS10_UNMAP:
931 return sd_setup_write_same10_cmnd(cmd, true);
932 }
933 }
934
935 if (sdp->no_write_same)
936 return BLK_STS_TARGET;
937
938 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
939 return sd_setup_write_same16_cmnd(cmd, false);
940
941 return sd_setup_write_same10_cmnd(cmd, false);
942 }
943
944 static void sd_config_write_same(struct scsi_disk *sdkp)
945 {
946 struct request_queue *q = sdkp->disk->queue;
947 unsigned int logical_block_size = sdkp->device->sector_size;
948
949 if (sdkp->device->no_write_same) {
950 sdkp->max_ws_blocks = 0;
951 goto out;
952 }
953
954 /* Some devices can not handle block counts above 0xffff despite
955 * supporting WRITE SAME(16). Consequently we default to 64k
956 * blocks per I/O unless the device explicitly advertises a
957 * bigger limit.
958 */
959 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
960 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
961 (u32)SD_MAX_WS16_BLOCKS);
962 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
963 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
964 (u32)SD_MAX_WS10_BLOCKS);
965 else {
966 sdkp->device->no_write_same = 1;
967 sdkp->max_ws_blocks = 0;
968 }
969
970 if (sdkp->lbprz && sdkp->lbpws)
971 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
972 else if (sdkp->lbprz && sdkp->lbpws10)
973 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
974 else if (sdkp->max_ws_blocks)
975 sdkp->zeroing_mode = SD_ZERO_WS;
976 else
977 sdkp->zeroing_mode = SD_ZERO_WRITE;
978
979 if (sdkp->max_ws_blocks &&
980 sdkp->physical_block_size > logical_block_size) {
981 /*
982 * Reporting a maximum number of blocks that is not aligned
983 * on the device physical size would cause a large write same
984 * request to be split into physically unaligned chunks by
985 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
986 * even if the caller of these functions took care to align the
987 * large request. So make sure the maximum reported is aligned
988 * to the device physical block size. This is only an optional
989 * optimization for regular disks, but this is mandatory to
990 * avoid failure of large write same requests directed at
991 * sequential write required zones of host-managed ZBC disks.
992 */
993 sdkp->max_ws_blocks =
994 round_down(sdkp->max_ws_blocks,
995 bytes_to_logical(sdkp->device,
996 sdkp->physical_block_size));
997 }
998
999 out:
1000 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1001 (logical_block_size >> 9));
1002 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1003 (logical_block_size >> 9));
1004 }
1005
1006 /**
1007 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1008 * @cmd: command to prepare
1009 *
1010 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1011 * the preference indicated by the target device.
1012 **/
1013 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1014 {
1015 struct request *rq = cmd->request;
1016 struct scsi_device *sdp = cmd->device;
1017 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1018 struct bio *bio = rq->bio;
1019 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1020 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1021 blk_status_t ret;
1022
1023 if (sdkp->device->no_write_same)
1024 return BLK_STS_TARGET;
1025
1026 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1027
1028 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1029
1030 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1031 cmd->cmd_len = 16;
1032 cmd->cmnd[0] = WRITE_SAME_16;
1033 put_unaligned_be64(lba, &cmd->cmnd[2]);
1034 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1035 } else {
1036 cmd->cmd_len = 10;
1037 cmd->cmnd[0] = WRITE_SAME;
1038 put_unaligned_be32(lba, &cmd->cmnd[2]);
1039 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1040 }
1041
1042 cmd->transfersize = sdp->sector_size;
1043 cmd->allowed = SD_MAX_RETRIES;
1044
1045 /*
1046 * For WRITE SAME the data transferred via the DATA OUT buffer is
1047 * different from the amount of data actually written to the target.
1048 *
1049 * We set up __data_len to the amount of data transferred via the
1050 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1051 * to transfer a single sector of data first, but then reset it to
1052 * the amount of data to be written right after so that the I/O path
1053 * knows how much to actually write.
1054 */
1055 rq->__data_len = sdp->sector_size;
1056 ret = scsi_init_io(cmd);
1057 rq->__data_len = blk_rq_bytes(rq);
1058
1059 return ret;
1060 }
1061
1062 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1063 {
1064 struct request *rq = cmd->request;
1065
1066 /* flush requests don't perform I/O, zero the S/G table */
1067 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1068
1069 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1070 cmd->cmd_len = 10;
1071 cmd->transfersize = 0;
1072 cmd->allowed = SD_MAX_RETRIES;
1073
1074 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1075 return BLK_STS_OK;
1076 }
1077
1078 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1079 sector_t lba, unsigned int nr_blocks,
1080 unsigned char flags)
1081 {
1082 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1083 if (unlikely(cmd->cmnd == NULL))
1084 return BLK_STS_RESOURCE;
1085
1086 cmd->cmd_len = SD_EXT_CDB_SIZE;
1087 memset(cmd->cmnd, 0, cmd->cmd_len);
1088
1089 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1090 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1091 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1092 cmd->cmnd[10] = flags;
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
1100 static 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)
1103 {
1104 cmd->cmd_len = 16;
1105 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1106 cmd->cmnd[1] = flags;
1107 cmd->cmnd[14] = 0;
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
1115 static 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
1130 static 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 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1160 {
1161 struct request *rq = cmd->request;
1162 struct scsi_device *sdp = cmd->device;
1163 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1164 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1165 sector_t threshold;
1166 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1167 bool dif, dix;
1168 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1169 bool write = rq_data_dir(rq) == WRITE;
1170 unsigned char protect, fua;
1171 blk_status_t ret;
1172
1173 ret = scsi_init_io(cmd);
1174 if (ret != BLK_STS_OK)
1175 return ret;
1176
1177 if (!scsi_device_online(sdp) || sdp->changed) {
1178 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1179 return BLK_STS_IOERR;
1180 }
1181
1182 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1183 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1184 return BLK_STS_IOERR;
1185 }
1186
1187 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1188 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1189 return BLK_STS_IOERR;
1190 }
1191
1192 /*
1193 * Some SD card readers can't handle accesses which touch the
1194 * last one or two logical blocks. Split accesses as needed.
1195 */
1196 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1197
1198 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1199 if (lba < threshold) {
1200 /* Access up to the threshold but not beyond */
1201 nr_blocks = threshold - lba;
1202 } else {
1203 /* Access only a single logical block */
1204 nr_blocks = 1;
1205 }
1206 }
1207
1208 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1209 dix = scsi_prot_sg_count(cmd);
1210 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1211
1212 if (write && dix)
1213 t10_pi_prepare(cmd->request, sdkp->protection_type);
1214
1215 if (dif || dix)
1216 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1217 else
1218 protect = 0;
1219
1220 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1221 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1222 protect | fua);
1223 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1224 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1225 protect | fua);
1226 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1227 sdp->use_10_for_rw || protect) {
1228 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1229 protect | fua);
1230 } else {
1231 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1232 protect | fua);
1233 }
1234
1235 if (unlikely(ret != BLK_STS_OK))
1236 return ret;
1237
1238 /*
1239 * We shouldn't disconnect in the middle of a sector, so with a dumb
1240 * host adapter, it's safe to assume that we can at least transfer
1241 * this many bytes between each connect / disconnect.
1242 */
1243 cmd->transfersize = sdp->sector_size;
1244 cmd->underflow = nr_blocks << 9;
1245 cmd->allowed = SD_MAX_RETRIES;
1246 cmd->sdb.length = nr_blocks * sdp->sector_size;
1247
1248 SCSI_LOG_HLQUEUE(1,
1249 scmd_printk(KERN_INFO, cmd,
1250 "%s: block=%llu, count=%d\n", __func__,
1251 (unsigned long long)blk_rq_pos(rq),
1252 blk_rq_sectors(rq)));
1253 SCSI_LOG_HLQUEUE(2,
1254 scmd_printk(KERN_INFO, cmd,
1255 "%s %d/%u 512 byte blocks.\n",
1256 write ? "writing" : "reading", nr_blocks,
1257 blk_rq_sectors(rq)));
1258
1259 /*
1260 * This indicates that the command is ready from our end to be
1261 * queued.
1262 */
1263 return BLK_STS_OK;
1264 }
1265
1266 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1267 {
1268 struct request *rq = cmd->request;
1269
1270 switch (req_op(rq)) {
1271 case REQ_OP_DISCARD:
1272 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1273 case SD_LBP_UNMAP:
1274 return sd_setup_unmap_cmnd(cmd);
1275 case SD_LBP_WS16:
1276 return sd_setup_write_same16_cmnd(cmd, true);
1277 case SD_LBP_WS10:
1278 return sd_setup_write_same10_cmnd(cmd, true);
1279 case SD_LBP_ZERO:
1280 return sd_setup_write_same10_cmnd(cmd, false);
1281 default:
1282 return BLK_STS_TARGET;
1283 }
1284 case REQ_OP_WRITE_ZEROES:
1285 return sd_setup_write_zeroes_cmnd(cmd);
1286 case REQ_OP_WRITE_SAME:
1287 return sd_setup_write_same_cmnd(cmd);
1288 case REQ_OP_FLUSH:
1289 return sd_setup_flush_cmnd(cmd);
1290 case REQ_OP_READ:
1291 case REQ_OP_WRITE:
1292 return sd_setup_read_write_cmnd(cmd);
1293 case REQ_OP_ZONE_RESET:
1294 return sd_zbc_setup_reset_cmnd(cmd);
1295 default:
1296 WARN_ON_ONCE(1);
1297 return BLK_STS_NOTSUPP;
1298 }
1299 }
1300
1301 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1302 {
1303 struct request *rq = SCpnt->request;
1304 u8 *cmnd;
1305
1306 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1307 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1308
1309 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1310 cmnd = SCpnt->cmnd;
1311 SCpnt->cmnd = NULL;
1312 SCpnt->cmd_len = 0;
1313 mempool_free(cmnd, sd_cdb_pool);
1314 }
1315 }
1316
1317 /**
1318 * sd_open - open a scsi disk device
1319 * @bdev: Block device of the scsi disk to open
1320 * @mode: FMODE_* mask
1321 *
1322 * Returns 0 if successful. Returns a negated errno value in case
1323 * of error.
1324 *
1325 * Note: This can be called from a user context (e.g. fsck(1) )
1326 * or from within the kernel (e.g. as a result of a mount(1) ).
1327 * In the latter case @inode and @filp carry an abridged amount
1328 * of information as noted above.
1329 *
1330 * Locking: called with bdev->bd_mutex held.
1331 **/
1332 static int sd_open(struct block_device *bdev, fmode_t mode)
1333 {
1334 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1335 struct scsi_device *sdev;
1336 int retval;
1337
1338 if (!sdkp)
1339 return -ENXIO;
1340
1341 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1342
1343 sdev = sdkp->device;
1344
1345 /*
1346 * If the device is in error recovery, wait until it is done.
1347 * If the device is offline, then disallow any access to it.
1348 */
1349 retval = -ENXIO;
1350 if (!scsi_block_when_processing_errors(sdev))
1351 goto error_out;
1352
1353 if (sdev->removable || sdkp->write_prot)
1354 check_disk_change(bdev);
1355
1356 /*
1357 * If the drive is empty, just let the open fail.
1358 */
1359 retval = -ENOMEDIUM;
1360 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1361 goto error_out;
1362
1363 /*
1364 * If the device has the write protect tab set, have the open fail
1365 * if the user expects to be able to write to the thing.
1366 */
1367 retval = -EROFS;
1368 if (sdkp->write_prot && (mode & FMODE_WRITE))
1369 goto error_out;
1370
1371 /*
1372 * It is possible that the disk changing stuff resulted in
1373 * the device being taken offline. If this is the case,
1374 * report this to the user, and don't pretend that the
1375 * open actually succeeded.
1376 */
1377 retval = -ENXIO;
1378 if (!scsi_device_online(sdev))
1379 goto error_out;
1380
1381 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1382 if (scsi_block_when_processing_errors(sdev))
1383 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1384 }
1385
1386 return 0;
1387
1388 error_out:
1389 scsi_disk_put(sdkp);
1390 return retval;
1391 }
1392
1393 /**
1394 * sd_release - invoked when the (last) close(2) is called on this
1395 * scsi disk.
1396 * @disk: disk to release
1397 * @mode: FMODE_* mask
1398 *
1399 * Returns 0.
1400 *
1401 * Note: may block (uninterruptible) if error recovery is underway
1402 * on this disk.
1403 *
1404 * Locking: called with bdev->bd_mutex held.
1405 **/
1406 static void sd_release(struct gendisk *disk, fmode_t mode)
1407 {
1408 struct scsi_disk *sdkp = scsi_disk(disk);
1409 struct scsi_device *sdev = sdkp->device;
1410
1411 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1412
1413 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1414 if (scsi_block_when_processing_errors(sdev))
1415 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1416 }
1417
1418 scsi_disk_put(sdkp);
1419 }
1420
1421 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1422 {
1423 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1424 struct scsi_device *sdp = sdkp->device;
1425 struct Scsi_Host *host = sdp->host;
1426 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1427 int diskinfo[4];
1428
1429 /* default to most commonly used values */
1430 diskinfo[0] = 0x40; /* 1 << 6 */
1431 diskinfo[1] = 0x20; /* 1 << 5 */
1432 diskinfo[2] = capacity >> 11;
1433
1434 /* override with calculated, extended default, or driver values */
1435 if (host->hostt->bios_param)
1436 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1437 else
1438 scsicam_bios_param(bdev, capacity, diskinfo);
1439
1440 geo->heads = diskinfo[0];
1441 geo->sectors = diskinfo[1];
1442 geo->cylinders = diskinfo[2];
1443 return 0;
1444 }
1445
1446 /**
1447 * sd_ioctl - process an ioctl
1448 * @bdev: target block device
1449 * @mode: FMODE_* mask
1450 * @cmd: ioctl command number
1451 * @arg: this is third argument given to ioctl(2) system call.
1452 * Often contains a pointer.
1453 *
1454 * Returns 0 if successful (some ioctls return positive numbers on
1455 * success as well). Returns a negated errno value in case of error.
1456 *
1457 * Note: most ioctls are forward onto the block subsystem or further
1458 * down in the scsi subsystem.
1459 **/
1460 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1461 unsigned int cmd, unsigned long arg)
1462 {
1463 struct gendisk *disk = bdev->bd_disk;
1464 struct scsi_disk *sdkp = scsi_disk(disk);
1465 struct scsi_device *sdp = sdkp->device;
1466 void __user *p = (void __user *)arg;
1467 int error;
1468
1469 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1470 "cmd=0x%x\n", disk->disk_name, cmd));
1471
1472 error = scsi_verify_blk_ioctl(bdev, cmd);
1473 if (error < 0)
1474 return error;
1475
1476 /*
1477 * If we are in the middle of error recovery, don't let anyone
1478 * else try and use this device. Also, if error recovery fails, it
1479 * may try and take the device offline, in which case all further
1480 * access to the device is prohibited.
1481 */
1482 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1483 (mode & FMODE_NDELAY) != 0);
1484 if (error)
1485 goto out;
1486
1487 if (is_sed_ioctl(cmd))
1488 return sed_ioctl(sdkp->opal_dev, cmd, p);
1489
1490 /*
1491 * Send SCSI addressing ioctls directly to mid level, send other
1492 * ioctls to block level and then onto mid level if they can't be
1493 * resolved.
1494 */
1495 switch (cmd) {
1496 case SCSI_IOCTL_GET_IDLUN:
1497 case SCSI_IOCTL_GET_BUS_NUMBER:
1498 error = scsi_ioctl(sdp, cmd, p);
1499 break;
1500 default:
1501 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1502 if (error != -ENOTTY)
1503 break;
1504 error = scsi_ioctl(sdp, cmd, p);
1505 break;
1506 }
1507 out:
1508 return error;
1509 }
1510
1511 static void set_media_not_present(struct scsi_disk *sdkp)
1512 {
1513 if (sdkp->media_present)
1514 sdkp->device->changed = 1;
1515
1516 if (sdkp->device->removable) {
1517 sdkp->media_present = 0;
1518 sdkp->capacity = 0;
1519 }
1520 }
1521
1522 static int media_not_present(struct scsi_disk *sdkp,
1523 struct scsi_sense_hdr *sshdr)
1524 {
1525 if (!scsi_sense_valid(sshdr))
1526 return 0;
1527
1528 /* not invoked for commands that could return deferred errors */
1529 switch (sshdr->sense_key) {
1530 case UNIT_ATTENTION:
1531 case NOT_READY:
1532 /* medium not present */
1533 if (sshdr->asc == 0x3A) {
1534 set_media_not_present(sdkp);
1535 return 1;
1536 }
1537 }
1538 return 0;
1539 }
1540
1541 /**
1542 * sd_check_events - check media events
1543 * @disk: kernel device descriptor
1544 * @clearing: disk events currently being cleared
1545 *
1546 * Returns mask of DISK_EVENT_*.
1547 *
1548 * Note: this function is invoked from the block subsystem.
1549 **/
1550 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1551 {
1552 struct scsi_disk *sdkp = scsi_disk_get(disk);
1553 struct scsi_device *sdp;
1554 int retval;
1555
1556 if (!sdkp)
1557 return 0;
1558
1559 sdp = sdkp->device;
1560 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1561
1562 /*
1563 * If the device is offline, don't send any commands - just pretend as
1564 * if the command failed. If the device ever comes back online, we
1565 * can deal with it then. It is only because of unrecoverable errors
1566 * that we would ever take a device offline in the first place.
1567 */
1568 if (!scsi_device_online(sdp)) {
1569 set_media_not_present(sdkp);
1570 goto out;
1571 }
1572
1573 /*
1574 * Using TEST_UNIT_READY enables differentiation between drive with
1575 * no cartridge loaded - NOT READY, drive with changed cartridge -
1576 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1577 *
1578 * Drives that auto spin down. eg iomega jaz 1G, will be started
1579 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1580 * sd_revalidate() is called.
1581 */
1582 if (scsi_block_when_processing_errors(sdp)) {
1583 struct scsi_sense_hdr sshdr = { 0, };
1584
1585 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1586 &sshdr);
1587
1588 /* failed to execute TUR, assume media not present */
1589 if (host_byte(retval)) {
1590 set_media_not_present(sdkp);
1591 goto out;
1592 }
1593
1594 if (media_not_present(sdkp, &sshdr))
1595 goto out;
1596 }
1597
1598 /*
1599 * For removable scsi disk we have to recognise the presence
1600 * of a disk in the drive.
1601 */
1602 if (!sdkp->media_present)
1603 sdp->changed = 1;
1604 sdkp->media_present = 1;
1605 out:
1606 /*
1607 * sdp->changed is set under the following conditions:
1608 *
1609 * Medium present state has changed in either direction.
1610 * Device has indicated UNIT_ATTENTION.
1611 */
1612 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1613 sdp->changed = 0;
1614 scsi_disk_put(sdkp);
1615 return retval;
1616 }
1617
1618 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1619 {
1620 int retries, res;
1621 struct scsi_device *sdp = sdkp->device;
1622 const int timeout = sdp->request_queue->rq_timeout
1623 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1624 struct scsi_sense_hdr my_sshdr;
1625
1626 if (!scsi_device_online(sdp))
1627 return -ENODEV;
1628
1629 /* caller might not be interested in sense, but we need it */
1630 if (!sshdr)
1631 sshdr = &my_sshdr;
1632
1633 for (retries = 3; retries > 0; --retries) {
1634 unsigned char cmd[10] = { 0 };
1635
1636 cmd[0] = SYNCHRONIZE_CACHE;
1637 /*
1638 * Leave the rest of the command zero to indicate
1639 * flush everything.
1640 */
1641 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1642 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1643 if (res == 0)
1644 break;
1645 }
1646
1647 if (res) {
1648 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1649
1650 if (driver_byte(res) == DRIVER_SENSE)
1651 sd_print_sense_hdr(sdkp, sshdr);
1652
1653 /* we need to evaluate the error return */
1654 if (scsi_sense_valid(sshdr) &&
1655 (sshdr->asc == 0x3a || /* medium not present */
1656 sshdr->asc == 0x20)) /* invalid command */
1657 /* this is no error here */
1658 return 0;
1659
1660 switch (host_byte(res)) {
1661 /* ignore errors due to racing a disconnection */
1662 case DID_BAD_TARGET:
1663 case DID_NO_CONNECT:
1664 return 0;
1665 /* signal the upper layer it might try again */
1666 case DID_BUS_BUSY:
1667 case DID_IMM_RETRY:
1668 case DID_REQUEUE:
1669 case DID_SOFT_ERROR:
1670 return -EBUSY;
1671 default:
1672 return -EIO;
1673 }
1674 }
1675 return 0;
1676 }
1677
1678 static void sd_rescan(struct device *dev)
1679 {
1680 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1681
1682 revalidate_disk(sdkp->disk);
1683 }
1684
1685
1686 #ifdef CONFIG_COMPAT
1687 /*
1688 * This gets directly called from VFS. When the ioctl
1689 * is not recognized we go back to the other translation paths.
1690 */
1691 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1692 unsigned int cmd, unsigned long arg)
1693 {
1694 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1695 int error;
1696
1697 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1698 (mode & FMODE_NDELAY) != 0);
1699 if (error)
1700 return error;
1701
1702 /*
1703 * Let the static ioctl translation table take care of it.
1704 */
1705 if (!sdev->host->hostt->compat_ioctl)
1706 return -ENOIOCTLCMD;
1707 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1708 }
1709 #endif
1710
1711 static char sd_pr_type(enum pr_type type)
1712 {
1713 switch (type) {
1714 case PR_WRITE_EXCLUSIVE:
1715 return 0x01;
1716 case PR_EXCLUSIVE_ACCESS:
1717 return 0x03;
1718 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1719 return 0x05;
1720 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1721 return 0x06;
1722 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1723 return 0x07;
1724 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1725 return 0x08;
1726 default:
1727 return 0;
1728 }
1729 };
1730
1731 static int sd_pr_command(struct block_device *bdev, u8 sa,
1732 u64 key, u64 sa_key, u8 type, u8 flags)
1733 {
1734 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1735 struct scsi_sense_hdr sshdr;
1736 int result;
1737 u8 cmd[16] = { 0, };
1738 u8 data[24] = { 0, };
1739
1740 cmd[0] = PERSISTENT_RESERVE_OUT;
1741 cmd[1] = sa;
1742 cmd[2] = type;
1743 put_unaligned_be32(sizeof(data), &cmd[5]);
1744
1745 put_unaligned_be64(key, &data[0]);
1746 put_unaligned_be64(sa_key, &data[8]);
1747 data[20] = flags;
1748
1749 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1750 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1751
1752 if (driver_byte(result) == DRIVER_SENSE &&
1753 scsi_sense_valid(&sshdr)) {
1754 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1755 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1756 }
1757
1758 return result;
1759 }
1760
1761 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1762 u32 flags)
1763 {
1764 if (flags & ~PR_FL_IGNORE_KEY)
1765 return -EOPNOTSUPP;
1766 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1767 old_key, new_key, 0,
1768 (1 << 0) /* APTPL */);
1769 }
1770
1771 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1772 u32 flags)
1773 {
1774 if (flags)
1775 return -EOPNOTSUPP;
1776 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1777 }
1778
1779 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1780 {
1781 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1782 }
1783
1784 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1785 enum pr_type type, bool abort)
1786 {
1787 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1788 sd_pr_type(type), 0);
1789 }
1790
1791 static int sd_pr_clear(struct block_device *bdev, u64 key)
1792 {
1793 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1794 }
1795
1796 static const struct pr_ops sd_pr_ops = {
1797 .pr_register = sd_pr_register,
1798 .pr_reserve = sd_pr_reserve,
1799 .pr_release = sd_pr_release,
1800 .pr_preempt = sd_pr_preempt,
1801 .pr_clear = sd_pr_clear,
1802 };
1803
1804 static const struct block_device_operations sd_fops = {
1805 .owner = THIS_MODULE,
1806 .open = sd_open,
1807 .release = sd_release,
1808 .ioctl = sd_ioctl,
1809 .getgeo = sd_getgeo,
1810 #ifdef CONFIG_COMPAT
1811 .compat_ioctl = sd_compat_ioctl,
1812 #endif
1813 .check_events = sd_check_events,
1814 .revalidate_disk = sd_revalidate_disk,
1815 .unlock_native_capacity = sd_unlock_native_capacity,
1816 .report_zones = sd_zbc_report_zones,
1817 .pr_ops = &sd_pr_ops,
1818 };
1819
1820 /**
1821 * sd_eh_reset - reset error handling callback
1822 * @scmd: sd-issued command that has failed
1823 *
1824 * This function is called by the SCSI midlayer before starting
1825 * SCSI EH. When counting medium access failures we have to be
1826 * careful to register it only only once per device and SCSI EH run;
1827 * there might be several timed out commands which will cause the
1828 * 'max_medium_access_timeouts' counter to trigger after the first
1829 * SCSI EH run already and set the device to offline.
1830 * So this function resets the internal counter before starting SCSI EH.
1831 **/
1832 static void sd_eh_reset(struct scsi_cmnd *scmd)
1833 {
1834 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1835
1836 /* New SCSI EH run, reset gate variable */
1837 sdkp->ignore_medium_access_errors = false;
1838 }
1839
1840 /**
1841 * sd_eh_action - error handling callback
1842 * @scmd: sd-issued command that has failed
1843 * @eh_disp: The recovery disposition suggested by the midlayer
1844 *
1845 * This function is called by the SCSI midlayer upon completion of an
1846 * error test command (currently TEST UNIT READY). The result of sending
1847 * the eh command is passed in eh_disp. We're looking for devices that
1848 * fail medium access commands but are OK with non access commands like
1849 * test unit ready (so wrongly see the device as having a successful
1850 * recovery)
1851 **/
1852 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1853 {
1854 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1855 struct scsi_device *sdev = scmd->device;
1856
1857 if (!scsi_device_online(sdev) ||
1858 !scsi_medium_access_command(scmd) ||
1859 host_byte(scmd->result) != DID_TIME_OUT ||
1860 eh_disp != SUCCESS)
1861 return eh_disp;
1862
1863 /*
1864 * The device has timed out executing a medium access command.
1865 * However, the TEST UNIT READY command sent during error
1866 * handling completed successfully. Either the device is in the
1867 * process of recovering or has it suffered an internal failure
1868 * that prevents access to the storage medium.
1869 */
1870 if (!sdkp->ignore_medium_access_errors) {
1871 sdkp->medium_access_timed_out++;
1872 sdkp->ignore_medium_access_errors = true;
1873 }
1874
1875 /*
1876 * If the device keeps failing read/write commands but TEST UNIT
1877 * READY always completes successfully we assume that medium
1878 * access is no longer possible and take the device offline.
1879 */
1880 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1881 scmd_printk(KERN_ERR, scmd,
1882 "Medium access timeout failure. Offlining disk!\n");
1883 mutex_lock(&sdev->state_mutex);
1884 scsi_device_set_state(sdev, SDEV_OFFLINE);
1885 mutex_unlock(&sdev->state_mutex);
1886
1887 return SUCCESS;
1888 }
1889
1890 return eh_disp;
1891 }
1892
1893 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1894 {
1895 struct request *req = scmd->request;
1896 struct scsi_device *sdev = scmd->device;
1897 unsigned int transferred, good_bytes;
1898 u64 start_lba, end_lba, bad_lba;
1899
1900 /*
1901 * Some commands have a payload smaller than the device logical
1902 * block size (e.g. INQUIRY on a 4K disk).
1903 */
1904 if (scsi_bufflen(scmd) <= sdev->sector_size)
1905 return 0;
1906
1907 /* Check if we have a 'bad_lba' information */
1908 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1909 SCSI_SENSE_BUFFERSIZE,
1910 &bad_lba))
1911 return 0;
1912
1913 /*
1914 * If the bad lba was reported incorrectly, we have no idea where
1915 * the error is.
1916 */
1917 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1918 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1919 if (bad_lba < start_lba || bad_lba >= end_lba)
1920 return 0;
1921
1922 /*
1923 * resid is optional but mostly filled in. When it's unused,
1924 * its value is zero, so we assume the whole buffer transferred
1925 */
1926 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1927
1928 /* This computation should always be done in terms of the
1929 * resolution of the device's medium.
1930 */
1931 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1932
1933 return min(good_bytes, transferred);
1934 }
1935
1936 /**
1937 * sd_done - bottom half handler: called when the lower level
1938 * driver has completed (successfully or otherwise) a scsi command.
1939 * @SCpnt: mid-level's per command structure.
1940 *
1941 * Note: potentially run from within an ISR. Must not block.
1942 **/
1943 static int sd_done(struct scsi_cmnd *SCpnt)
1944 {
1945 int result = SCpnt->result;
1946 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1947 unsigned int sector_size = SCpnt->device->sector_size;
1948 unsigned int resid;
1949 struct scsi_sense_hdr sshdr;
1950 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1951 struct request *req = SCpnt->request;
1952 int sense_valid = 0;
1953 int sense_deferred = 0;
1954
1955 switch (req_op(req)) {
1956 case REQ_OP_DISCARD:
1957 case REQ_OP_WRITE_ZEROES:
1958 case REQ_OP_WRITE_SAME:
1959 case REQ_OP_ZONE_RESET:
1960 if (!result) {
1961 good_bytes = blk_rq_bytes(req);
1962 scsi_set_resid(SCpnt, 0);
1963 } else {
1964 good_bytes = 0;
1965 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1966 }
1967 break;
1968 default:
1969 /*
1970 * In case of bogus fw or device, we could end up having
1971 * an unaligned partial completion. Check this here and force
1972 * alignment.
1973 */
1974 resid = scsi_get_resid(SCpnt);
1975 if (resid & (sector_size - 1)) {
1976 sd_printk(KERN_INFO, sdkp,
1977 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1978 resid, sector_size);
1979 resid = min(scsi_bufflen(SCpnt),
1980 round_up(resid, sector_size));
1981 scsi_set_resid(SCpnt, resid);
1982 }
1983 }
1984
1985 if (result) {
1986 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1987 if (sense_valid)
1988 sense_deferred = scsi_sense_is_deferred(&sshdr);
1989 }
1990 sdkp->medium_access_timed_out = 0;
1991
1992 if (driver_byte(result) != DRIVER_SENSE &&
1993 (!sense_valid || sense_deferred))
1994 goto out;
1995
1996 switch (sshdr.sense_key) {
1997 case HARDWARE_ERROR:
1998 case MEDIUM_ERROR:
1999 good_bytes = sd_completed_bytes(SCpnt);
2000 break;
2001 case RECOVERED_ERROR:
2002 good_bytes = scsi_bufflen(SCpnt);
2003 break;
2004 case NO_SENSE:
2005 /* This indicates a false check condition, so ignore it. An
2006 * unknown amount of data was transferred so treat it as an
2007 * error.
2008 */
2009 SCpnt->result = 0;
2010 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2011 break;
2012 case ABORTED_COMMAND:
2013 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2014 good_bytes = sd_completed_bytes(SCpnt);
2015 break;
2016 case ILLEGAL_REQUEST:
2017 switch (sshdr.asc) {
2018 case 0x10: /* DIX: Host detected corruption */
2019 good_bytes = sd_completed_bytes(SCpnt);
2020 break;
2021 case 0x20: /* INVALID COMMAND OPCODE */
2022 case 0x24: /* INVALID FIELD IN CDB */
2023 switch (SCpnt->cmnd[0]) {
2024 case UNMAP:
2025 sd_config_discard(sdkp, SD_LBP_DISABLE);
2026 break;
2027 case WRITE_SAME_16:
2028 case WRITE_SAME:
2029 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2030 sd_config_discard(sdkp, SD_LBP_DISABLE);
2031 } else {
2032 sdkp->device->no_write_same = 1;
2033 sd_config_write_same(sdkp);
2034 req->rq_flags |= RQF_QUIET;
2035 }
2036 break;
2037 }
2038 }
2039 break;
2040 default:
2041 break;
2042 }
2043
2044 out:
2045 if (sd_is_zoned(sdkp))
2046 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2047
2048 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2049 "sd_done: completed %d of %d bytes\n",
2050 good_bytes, scsi_bufflen(SCpnt)));
2051
2052 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2053 good_bytes)
2054 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2055 good_bytes / scsi_prot_interval(SCpnt));
2056
2057 return good_bytes;
2058 }
2059
2060 /*
2061 * spinup disk - called only in sd_revalidate_disk()
2062 */
2063 static void
2064 sd_spinup_disk(struct scsi_disk *sdkp)
2065 {
2066 unsigned char cmd[10];
2067 unsigned long spintime_expire = 0;
2068 int retries, spintime;
2069 unsigned int the_result;
2070 struct scsi_sense_hdr sshdr;
2071 int sense_valid = 0;
2072
2073 spintime = 0;
2074
2075 /* Spin up drives, as required. Only do this at boot time */
2076 /* Spinup needs to be done for module loads too. */
2077 do {
2078 retries = 0;
2079
2080 do {
2081 cmd[0] = TEST_UNIT_READY;
2082 memset((void *) &cmd[1], 0, 9);
2083
2084 the_result = scsi_execute_req(sdkp->device, cmd,
2085 DMA_NONE, NULL, 0,
2086 &sshdr, SD_TIMEOUT,
2087 SD_MAX_RETRIES, NULL);
2088
2089 /*
2090 * If the drive has indicated to us that it
2091 * doesn't have any media in it, don't bother
2092 * with any more polling.
2093 */
2094 if (media_not_present(sdkp, &sshdr))
2095 return;
2096
2097 if (the_result)
2098 sense_valid = scsi_sense_valid(&sshdr);
2099 retries++;
2100 } while (retries < 3 &&
2101 (!scsi_status_is_good(the_result) ||
2102 ((driver_byte(the_result) == DRIVER_SENSE) &&
2103 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2104
2105 if (driver_byte(the_result) != DRIVER_SENSE) {
2106 /* no sense, TUR either succeeded or failed
2107 * with a status error */
2108 if(!spintime && !scsi_status_is_good(the_result)) {
2109 sd_print_result(sdkp, "Test Unit Ready failed",
2110 the_result);
2111 }
2112 break;
2113 }
2114
2115 /*
2116 * The device does not want the automatic start to be issued.
2117 */
2118 if (sdkp->device->no_start_on_add)
2119 break;
2120
2121 if (sense_valid && sshdr.sense_key == NOT_READY) {
2122 if (sshdr.asc == 4 && sshdr.ascq == 3)
2123 break; /* manual intervention required */
2124 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2125 break; /* standby */
2126 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2127 break; /* unavailable */
2128 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2129 break; /* sanitize in progress */
2130 /*
2131 * Issue command to spin up drive when not ready
2132 */
2133 if (!spintime) {
2134 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2135 cmd[0] = START_STOP;
2136 cmd[1] = 1; /* Return immediately */
2137 memset((void *) &cmd[2], 0, 8);
2138 cmd[4] = 1; /* Start spin cycle */
2139 if (sdkp->device->start_stop_pwr_cond)
2140 cmd[4] |= 1 << 4;
2141 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2142 NULL, 0, &sshdr,
2143 SD_TIMEOUT, SD_MAX_RETRIES,
2144 NULL);
2145 spintime_expire = jiffies + 100 * HZ;
2146 spintime = 1;
2147 }
2148 /* Wait 1 second for next try */
2149 msleep(1000);
2150 printk(KERN_CONT ".");
2151
2152 /*
2153 * Wait for USB flash devices with slow firmware.
2154 * Yes, this sense key/ASC combination shouldn't
2155 * occur here. It's characteristic of these devices.
2156 */
2157 } else if (sense_valid &&
2158 sshdr.sense_key == UNIT_ATTENTION &&
2159 sshdr.asc == 0x28) {
2160 if (!spintime) {
2161 spintime_expire = jiffies + 5 * HZ;
2162 spintime = 1;
2163 }
2164 /* Wait 1 second for next try */
2165 msleep(1000);
2166 } else {
2167 /* we don't understand the sense code, so it's
2168 * probably pointless to loop */
2169 if(!spintime) {
2170 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2171 sd_print_sense_hdr(sdkp, &sshdr);
2172 }
2173 break;
2174 }
2175
2176 } while (spintime && time_before_eq(jiffies, spintime_expire));
2177
2178 if (spintime) {
2179 if (scsi_status_is_good(the_result))
2180 printk(KERN_CONT "ready\n");
2181 else
2182 printk(KERN_CONT "not responding...\n");
2183 }
2184 }
2185
2186 /*
2187 * Determine whether disk supports Data Integrity Field.
2188 */
2189 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2190 {
2191 struct scsi_device *sdp = sdkp->device;
2192 u8 type;
2193 int ret = 0;
2194
2195 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2196 return ret;
2197
2198 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2199
2200 if (type > T10_PI_TYPE3_PROTECTION)
2201 ret = -ENODEV;
2202 else if (scsi_host_dif_capable(sdp->host, type))
2203 ret = 1;
2204
2205 if (sdkp->first_scan || type != sdkp->protection_type)
2206 switch (ret) {
2207 case -ENODEV:
2208 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2209 " protection type %u. Disabling disk!\n",
2210 type);
2211 break;
2212 case 1:
2213 sd_printk(KERN_NOTICE, sdkp,
2214 "Enabling DIF Type %u protection\n", type);
2215 break;
2216 case 0:
2217 sd_printk(KERN_NOTICE, sdkp,
2218 "Disabling DIF Type %u protection\n", type);
2219 break;
2220 }
2221
2222 sdkp->protection_type = type;
2223
2224 return ret;
2225 }
2226
2227 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2228 struct scsi_sense_hdr *sshdr, int sense_valid,
2229 int the_result)
2230 {
2231 if (driver_byte(the_result) == DRIVER_SENSE)
2232 sd_print_sense_hdr(sdkp, sshdr);
2233 else
2234 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2235
2236 /*
2237 * Set dirty bit for removable devices if not ready -
2238 * sometimes drives will not report this properly.
2239 */
2240 if (sdp->removable &&
2241 sense_valid && sshdr->sense_key == NOT_READY)
2242 set_media_not_present(sdkp);
2243
2244 /*
2245 * We used to set media_present to 0 here to indicate no media
2246 * in the drive, but some drives fail read capacity even with
2247 * media present, so we can't do that.
2248 */
2249 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2250 }
2251
2252 #define RC16_LEN 32
2253 #if RC16_LEN > SD_BUF_SIZE
2254 #error RC16_LEN must not be more than SD_BUF_SIZE
2255 #endif
2256
2257 #define READ_CAPACITY_RETRIES_ON_RESET 10
2258
2259 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2260 unsigned char *buffer)
2261 {
2262 unsigned char cmd[16];
2263 struct scsi_sense_hdr sshdr;
2264 int sense_valid = 0;
2265 int the_result;
2266 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2267 unsigned int alignment;
2268 unsigned long long lba;
2269 unsigned sector_size;
2270
2271 if (sdp->no_read_capacity_16)
2272 return -EINVAL;
2273
2274 do {
2275 memset(cmd, 0, 16);
2276 cmd[0] = SERVICE_ACTION_IN_16;
2277 cmd[1] = SAI_READ_CAPACITY_16;
2278 cmd[13] = RC16_LEN;
2279 memset(buffer, 0, RC16_LEN);
2280
2281 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2282 buffer, RC16_LEN, &sshdr,
2283 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2284
2285 if (media_not_present(sdkp, &sshdr))
2286 return -ENODEV;
2287
2288 if (the_result) {
2289 sense_valid = scsi_sense_valid(&sshdr);
2290 if (sense_valid &&
2291 sshdr.sense_key == ILLEGAL_REQUEST &&
2292 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2293 sshdr.ascq == 0x00)
2294 /* Invalid Command Operation Code or
2295 * Invalid Field in CDB, just retry
2296 * silently with RC10 */
2297 return -EINVAL;
2298 if (sense_valid &&
2299 sshdr.sense_key == UNIT_ATTENTION &&
2300 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2301 /* Device reset might occur several times,
2302 * give it one more chance */
2303 if (--reset_retries > 0)
2304 continue;
2305 }
2306 retries--;
2307
2308 } while (the_result && retries);
2309
2310 if (the_result) {
2311 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2312 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2313 return -EINVAL;
2314 }
2315
2316 sector_size = get_unaligned_be32(&buffer[8]);
2317 lba = get_unaligned_be64(&buffer[0]);
2318
2319 if (sd_read_protection_type(sdkp, buffer) < 0) {
2320 sdkp->capacity = 0;
2321 return -ENODEV;
2322 }
2323
2324 /* Logical blocks per physical block exponent */
2325 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2326
2327 /* RC basis */
2328 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2329
2330 /* Lowest aligned logical block */
2331 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2332 blk_queue_alignment_offset(sdp->request_queue, alignment);
2333 if (alignment && sdkp->first_scan)
2334 sd_printk(KERN_NOTICE, sdkp,
2335 "physical block alignment offset: %u\n", alignment);
2336
2337 if (buffer[14] & 0x80) { /* LBPME */
2338 sdkp->lbpme = 1;
2339
2340 if (buffer[14] & 0x40) /* LBPRZ */
2341 sdkp->lbprz = 1;
2342
2343 sd_config_discard(sdkp, SD_LBP_WS16);
2344 }
2345
2346 sdkp->capacity = lba + 1;
2347 return sector_size;
2348 }
2349
2350 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2351 unsigned char *buffer)
2352 {
2353 unsigned char cmd[16];
2354 struct scsi_sense_hdr sshdr;
2355 int sense_valid = 0;
2356 int the_result;
2357 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2358 sector_t lba;
2359 unsigned sector_size;
2360
2361 do {
2362 cmd[0] = READ_CAPACITY;
2363 memset(&cmd[1], 0, 9);
2364 memset(buffer, 0, 8);
2365
2366 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2367 buffer, 8, &sshdr,
2368 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2369
2370 if (media_not_present(sdkp, &sshdr))
2371 return -ENODEV;
2372
2373 if (the_result) {
2374 sense_valid = scsi_sense_valid(&sshdr);
2375 if (sense_valid &&
2376 sshdr.sense_key == UNIT_ATTENTION &&
2377 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2378 /* Device reset might occur several times,
2379 * give it one more chance */
2380 if (--reset_retries > 0)
2381 continue;
2382 }
2383 retries--;
2384
2385 } while (the_result && retries);
2386
2387 if (the_result) {
2388 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2389 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2390 return -EINVAL;
2391 }
2392
2393 sector_size = get_unaligned_be32(&buffer[4]);
2394 lba = get_unaligned_be32(&buffer[0]);
2395
2396 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2397 /* Some buggy (usb cardreader) devices return an lba of
2398 0xffffffff when the want to report a size of 0 (with
2399 which they really mean no media is present) */
2400 sdkp->capacity = 0;
2401 sdkp->physical_block_size = sector_size;
2402 return sector_size;
2403 }
2404
2405 sdkp->capacity = lba + 1;
2406 sdkp->physical_block_size = sector_size;
2407 return sector_size;
2408 }
2409
2410 static int sd_try_rc16_first(struct scsi_device *sdp)
2411 {
2412 if (sdp->host->max_cmd_len < 16)
2413 return 0;
2414 if (sdp->try_rc_10_first)
2415 return 0;
2416 if (sdp->scsi_level > SCSI_SPC_2)
2417 return 1;
2418 if (scsi_device_protection(sdp))
2419 return 1;
2420 return 0;
2421 }
2422
2423 /*
2424 * read disk capacity
2425 */
2426 static void
2427 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2428 {
2429 int sector_size;
2430 struct scsi_device *sdp = sdkp->device;
2431
2432 if (sd_try_rc16_first(sdp)) {
2433 sector_size = read_capacity_16(sdkp, sdp, buffer);
2434 if (sector_size == -EOVERFLOW)
2435 goto got_data;
2436 if (sector_size == -ENODEV)
2437 return;
2438 if (sector_size < 0)
2439 sector_size = read_capacity_10(sdkp, sdp, buffer);
2440 if (sector_size < 0)
2441 return;
2442 } else {
2443 sector_size = read_capacity_10(sdkp, sdp, buffer);
2444 if (sector_size == -EOVERFLOW)
2445 goto got_data;
2446 if (sector_size < 0)
2447 return;
2448 if ((sizeof(sdkp->capacity) > 4) &&
2449 (sdkp->capacity > 0xffffffffULL)) {
2450 int old_sector_size = sector_size;
2451 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2452 "Trying to use READ CAPACITY(16).\n");
2453 sector_size = read_capacity_16(sdkp, sdp, buffer);
2454 if (sector_size < 0) {
2455 sd_printk(KERN_NOTICE, sdkp,
2456 "Using 0xffffffff as device size\n");
2457 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2458 sector_size = old_sector_size;
2459 goto got_data;
2460 }
2461 /* Remember that READ CAPACITY(16) succeeded */
2462 sdp->try_rc_10_first = 0;
2463 }
2464 }
2465
2466 /* Some devices are known to return the total number of blocks,
2467 * not the highest block number. Some devices have versions
2468 * which do this and others which do not. Some devices we might
2469 * suspect of doing this but we don't know for certain.
2470 *
2471 * If we know the reported capacity is wrong, decrement it. If
2472 * we can only guess, then assume the number of blocks is even
2473 * (usually true but not always) and err on the side of lowering
2474 * the capacity.
2475 */
2476 if (sdp->fix_capacity ||
2477 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2478 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2479 "from its reported value: %llu\n",
2480 (unsigned long long) sdkp->capacity);
2481 --sdkp->capacity;
2482 }
2483
2484 got_data:
2485 if (sector_size == 0) {
2486 sector_size = 512;
2487 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2488 "assuming 512.\n");
2489 }
2490
2491 if (sector_size != 512 &&
2492 sector_size != 1024 &&
2493 sector_size != 2048 &&
2494 sector_size != 4096) {
2495 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2496 sector_size);
2497 /*
2498 * The user might want to re-format the drive with
2499 * a supported sectorsize. Once this happens, it
2500 * would be relatively trivial to set the thing up.
2501 * For this reason, we leave the thing in the table.
2502 */
2503 sdkp->capacity = 0;
2504 /*
2505 * set a bogus sector size so the normal read/write
2506 * logic in the block layer will eventually refuse any
2507 * request on this device without tripping over power
2508 * of two sector size assumptions
2509 */
2510 sector_size = 512;
2511 }
2512 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2513 blk_queue_physical_block_size(sdp->request_queue,
2514 sdkp->physical_block_size);
2515 sdkp->device->sector_size = sector_size;
2516
2517 if (sdkp->capacity > 0xffffffff)
2518 sdp->use_16_for_rw = 1;
2519
2520 }
2521
2522 /*
2523 * Print disk capacity
2524 */
2525 static void
2526 sd_print_capacity(struct scsi_disk *sdkp,
2527 sector_t old_capacity)
2528 {
2529 int sector_size = sdkp->device->sector_size;
2530 char cap_str_2[10], cap_str_10[10];
2531
2532 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2533 return;
2534
2535 string_get_size(sdkp->capacity, sector_size,
2536 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2537 string_get_size(sdkp->capacity, sector_size,
2538 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2539
2540 sd_printk(KERN_NOTICE, sdkp,
2541 "%llu %d-byte logical blocks: (%s/%s)\n",
2542 (unsigned long long)sdkp->capacity,
2543 sector_size, cap_str_10, cap_str_2);
2544
2545 if (sdkp->physical_block_size != sector_size)
2546 sd_printk(KERN_NOTICE, sdkp,
2547 "%u-byte physical blocks\n",
2548 sdkp->physical_block_size);
2549
2550 sd_zbc_print_zones(sdkp);
2551 }
2552
2553 /* called with buffer of length 512 */
2554 static inline int
2555 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2556 unsigned char *buffer, int len, struct scsi_mode_data *data,
2557 struct scsi_sense_hdr *sshdr)
2558 {
2559 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2560 SD_TIMEOUT, SD_MAX_RETRIES, data,
2561 sshdr);
2562 }
2563
2564 /*
2565 * read write protect setting, if possible - called only in sd_revalidate_disk()
2566 * called with buffer of length SD_BUF_SIZE
2567 */
2568 static void
2569 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2570 {
2571 int res;
2572 struct scsi_device *sdp = sdkp->device;
2573 struct scsi_mode_data data;
2574 int disk_ro = get_disk_ro(sdkp->disk);
2575 int old_wp = sdkp->write_prot;
2576
2577 set_disk_ro(sdkp->disk, 0);
2578 if (sdp->skip_ms_page_3f) {
2579 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2580 return;
2581 }
2582
2583 if (sdp->use_192_bytes_for_3f) {
2584 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2585 } else {
2586 /*
2587 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2588 * We have to start carefully: some devices hang if we ask
2589 * for more than is available.
2590 */
2591 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2592
2593 /*
2594 * Second attempt: ask for page 0 When only page 0 is
2595 * implemented, a request for page 3F may return Sense Key
2596 * 5: Illegal Request, Sense Code 24: Invalid field in
2597 * CDB.
2598 */
2599 if (!scsi_status_is_good(res))
2600 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2601
2602 /*
2603 * Third attempt: ask 255 bytes, as we did earlier.
2604 */
2605 if (!scsi_status_is_good(res))
2606 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2607 &data, NULL);
2608 }
2609
2610 if (!scsi_status_is_good(res)) {
2611 sd_first_printk(KERN_WARNING, sdkp,
2612 "Test WP failed, assume Write Enabled\n");
2613 } else {
2614 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2615 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2616 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2617 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2618 sdkp->write_prot ? "on" : "off");
2619 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2620 }
2621 }
2622 }
2623
2624 /*
2625 * sd_read_cache_type - called only from sd_revalidate_disk()
2626 * called with buffer of length SD_BUF_SIZE
2627 */
2628 static void
2629 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2630 {
2631 int len = 0, res;
2632 struct scsi_device *sdp = sdkp->device;
2633
2634 int dbd;
2635 int modepage;
2636 int first_len;
2637 struct scsi_mode_data data;
2638 struct scsi_sense_hdr sshdr;
2639 int old_wce = sdkp->WCE;
2640 int old_rcd = sdkp->RCD;
2641 int old_dpofua = sdkp->DPOFUA;
2642
2643
2644 if (sdkp->cache_override)
2645 return;
2646
2647 first_len = 4;
2648 if (sdp->skip_ms_page_8) {
2649 if (sdp->type == TYPE_RBC)
2650 goto defaults;
2651 else {
2652 if (sdp->skip_ms_page_3f)
2653 goto defaults;
2654 modepage = 0x3F;
2655 if (sdp->use_192_bytes_for_3f)
2656 first_len = 192;
2657 dbd = 0;
2658 }
2659 } else if (sdp->type == TYPE_RBC) {
2660 modepage = 6;
2661 dbd = 8;
2662 } else {
2663 modepage = 8;
2664 dbd = 0;
2665 }
2666
2667 /* cautiously ask */
2668 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2669 &data, &sshdr);
2670
2671 if (!scsi_status_is_good(res))
2672 goto bad_sense;
2673
2674 if (!data.header_length) {
2675 modepage = 6;
2676 first_len = 0;
2677 sd_first_printk(KERN_ERR, sdkp,
2678 "Missing header in MODE_SENSE response\n");
2679 }
2680
2681 /* that went OK, now ask for the proper length */
2682 len = data.length;
2683
2684 /*
2685 * We're only interested in the first three bytes, actually.
2686 * But the data cache page is defined for the first 20.
2687 */
2688 if (len < 3)
2689 goto bad_sense;
2690 else if (len > SD_BUF_SIZE) {
2691 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2692 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2693 len = SD_BUF_SIZE;
2694 }
2695 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2696 len = 192;
2697
2698 /* Get the data */
2699 if (len > first_len)
2700 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2701 &data, &sshdr);
2702
2703 if (scsi_status_is_good(res)) {
2704 int offset = data.header_length + data.block_descriptor_length;
2705
2706 while (offset < len) {
2707 u8 page_code = buffer[offset] & 0x3F;
2708 u8 spf = buffer[offset] & 0x40;
2709
2710 if (page_code == 8 || page_code == 6) {
2711 /* We're interested only in the first 3 bytes.
2712 */
2713 if (len - offset <= 2) {
2714 sd_first_printk(KERN_ERR, sdkp,
2715 "Incomplete mode parameter "
2716 "data\n");
2717 goto defaults;
2718 } else {
2719 modepage = page_code;
2720 goto Page_found;
2721 }
2722 } else {
2723 /* Go to the next page */
2724 if (spf && len - offset > 3)
2725 offset += 4 + (buffer[offset+2] << 8) +
2726 buffer[offset+3];
2727 else if (!spf && len - offset > 1)
2728 offset += 2 + buffer[offset+1];
2729 else {
2730 sd_first_printk(KERN_ERR, sdkp,
2731 "Incomplete mode "
2732 "parameter data\n");
2733 goto defaults;
2734 }
2735 }
2736 }
2737
2738 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2739 goto defaults;
2740
2741 Page_found:
2742 if (modepage == 8) {
2743 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2744 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2745 } else {
2746 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2747 sdkp->RCD = 0;
2748 }
2749
2750 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2751 if (sdp->broken_fua) {
2752 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2753 sdkp->DPOFUA = 0;
2754 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2755 !sdkp->device->use_16_for_rw) {
2756 sd_first_printk(KERN_NOTICE, sdkp,
2757 "Uses READ/WRITE(6), disabling FUA\n");
2758 sdkp->DPOFUA = 0;
2759 }
2760
2761 /* No cache flush allowed for write protected devices */
2762 if (sdkp->WCE && sdkp->write_prot)
2763 sdkp->WCE = 0;
2764
2765 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2766 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2767 sd_printk(KERN_NOTICE, sdkp,
2768 "Write cache: %s, read cache: %s, %s\n",
2769 sdkp->WCE ? "enabled" : "disabled",
2770 sdkp->RCD ? "disabled" : "enabled",
2771 sdkp->DPOFUA ? "supports DPO and FUA"
2772 : "doesn't support DPO or FUA");
2773
2774 return;
2775 }
2776
2777 bad_sense:
2778 if (scsi_sense_valid(&sshdr) &&
2779 sshdr.sense_key == ILLEGAL_REQUEST &&
2780 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2781 /* Invalid field in CDB */
2782 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2783 else
2784 sd_first_printk(KERN_ERR, sdkp,
2785 "Asking for cache data failed\n");
2786
2787 defaults:
2788 if (sdp->wce_default_on) {
2789 sd_first_printk(KERN_NOTICE, sdkp,
2790 "Assuming drive cache: write back\n");
2791 sdkp->WCE = 1;
2792 } else {
2793 sd_first_printk(KERN_ERR, sdkp,
2794 "Assuming drive cache: write through\n");
2795 sdkp->WCE = 0;
2796 }
2797 sdkp->RCD = 0;
2798 sdkp->DPOFUA = 0;
2799 }
2800
2801 /*
2802 * The ATO bit indicates whether the DIF application tag is available
2803 * for use by the operating system.
2804 */
2805 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2806 {
2807 int res, offset;
2808 struct scsi_device *sdp = sdkp->device;
2809 struct scsi_mode_data data;
2810 struct scsi_sense_hdr sshdr;
2811
2812 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2813 return;
2814
2815 if (sdkp->protection_type == 0)
2816 return;
2817
2818 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2819 SD_MAX_RETRIES, &data, &sshdr);
2820
2821 if (!scsi_status_is_good(res) || !data.header_length ||
2822 data.length < 6) {
2823 sd_first_printk(KERN_WARNING, sdkp,
2824 "getting Control mode page failed, assume no ATO\n");
2825
2826 if (scsi_sense_valid(&sshdr))
2827 sd_print_sense_hdr(sdkp, &sshdr);
2828
2829 return;
2830 }
2831
2832 offset = data.header_length + data.block_descriptor_length;
2833
2834 if ((buffer[offset] & 0x3f) != 0x0a) {
2835 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2836 return;
2837 }
2838
2839 if ((buffer[offset + 5] & 0x80) == 0)
2840 return;
2841
2842 sdkp->ATO = 1;
2843
2844 return;
2845 }
2846
2847 /**
2848 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2849 * @sdkp: disk to query
2850 */
2851 static void sd_read_block_limits(struct scsi_disk *sdkp)
2852 {
2853 unsigned int sector_sz = sdkp->device->sector_size;
2854 const int vpd_len = 64;
2855 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2856
2857 if (!buffer ||
2858 /* Block Limits VPD */
2859 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2860 goto out;
2861
2862 blk_queue_io_min(sdkp->disk->queue,
2863 get_unaligned_be16(&buffer[6]) * sector_sz);
2864
2865 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2866 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2867
2868 if (buffer[3] == 0x3c) {
2869 unsigned int lba_count, desc_count;
2870
2871 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2872
2873 if (!sdkp->lbpme)
2874 goto out;
2875
2876 lba_count = get_unaligned_be32(&buffer[20]);
2877 desc_count = get_unaligned_be32(&buffer[24]);
2878
2879 if (lba_count && desc_count)
2880 sdkp->max_unmap_blocks = lba_count;
2881
2882 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2883
2884 if (buffer[32] & 0x80)
2885 sdkp->unmap_alignment =
2886 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2887
2888 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2889
2890 if (sdkp->max_unmap_blocks)
2891 sd_config_discard(sdkp, SD_LBP_UNMAP);
2892 else
2893 sd_config_discard(sdkp, SD_LBP_WS16);
2894
2895 } else { /* LBP VPD page tells us what to use */
2896 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2897 sd_config_discard(sdkp, SD_LBP_UNMAP);
2898 else if (sdkp->lbpws)
2899 sd_config_discard(sdkp, SD_LBP_WS16);
2900 else if (sdkp->lbpws10)
2901 sd_config_discard(sdkp, SD_LBP_WS10);
2902 else
2903 sd_config_discard(sdkp, SD_LBP_DISABLE);
2904 }
2905 }
2906
2907 out:
2908 kfree(buffer);
2909 }
2910
2911 /**
2912 * sd_read_block_characteristics - Query block dev. characteristics
2913 * @sdkp: disk to query
2914 */
2915 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2916 {
2917 struct request_queue *q = sdkp->disk->queue;
2918 unsigned char *buffer;
2919 u16 rot;
2920 const int vpd_len = 64;
2921
2922 buffer = kmalloc(vpd_len, GFP_KERNEL);
2923
2924 if (!buffer ||
2925 /* Block Device Characteristics VPD */
2926 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2927 goto out;
2928
2929 rot = get_unaligned_be16(&buffer[4]);
2930
2931 if (rot == 1) {
2932 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2933 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2934 }
2935
2936 if (sdkp->device->type == TYPE_ZBC) {
2937 /* Host-managed */
2938 q->limits.zoned = BLK_ZONED_HM;
2939 } else {
2940 sdkp->zoned = (buffer[8] >> 4) & 3;
2941 if (sdkp->zoned == 1)
2942 /* Host-aware */
2943 q->limits.zoned = BLK_ZONED_HA;
2944 else
2945 /*
2946 * Treat drive-managed devices as
2947 * regular block devices.
2948 */
2949 q->limits.zoned = BLK_ZONED_NONE;
2950 }
2951 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2952 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2953 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2954
2955 out:
2956 kfree(buffer);
2957 }
2958
2959 /**
2960 * sd_read_block_provisioning - Query provisioning VPD page
2961 * @sdkp: disk to query
2962 */
2963 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2964 {
2965 unsigned char *buffer;
2966 const int vpd_len = 8;
2967
2968 if (sdkp->lbpme == 0)
2969 return;
2970
2971 buffer = kmalloc(vpd_len, GFP_KERNEL);
2972
2973 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2974 goto out;
2975
2976 sdkp->lbpvpd = 1;
2977 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2978 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2979 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2980
2981 out:
2982 kfree(buffer);
2983 }
2984
2985 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2986 {
2987 struct scsi_device *sdev = sdkp->device;
2988
2989 if (sdev->host->no_write_same) {
2990 sdev->no_write_same = 1;
2991
2992 return;
2993 }
2994
2995 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2996 /* too large values might cause issues with arcmsr */
2997 int vpd_buf_len = 64;
2998
2999 sdev->no_report_opcodes = 1;
3000
3001 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3002 * CODES is unsupported and the device has an ATA
3003 * Information VPD page (SAT).
3004 */
3005 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3006 sdev->no_write_same = 1;
3007 }
3008
3009 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3010 sdkp->ws16 = 1;
3011
3012 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3013 sdkp->ws10 = 1;
3014 }
3015
3016 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3017 {
3018 struct scsi_device *sdev = sdkp->device;
3019
3020 if (!sdev->security_supported)
3021 return;
3022
3023 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3024 SECURITY_PROTOCOL_IN) == 1 &&
3025 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3026 SECURITY_PROTOCOL_OUT) == 1)
3027 sdkp->security = 1;
3028 }
3029
3030 /*
3031 * Determine the device's preferred I/O size for reads and writes
3032 * unless the reported value is unreasonably small, large, not a
3033 * multiple of the physical block size, or simply garbage.
3034 */
3035 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3036 unsigned int dev_max)
3037 {
3038 struct scsi_device *sdp = sdkp->device;
3039 unsigned int opt_xfer_bytes =
3040 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3041
3042 if (sdkp->opt_xfer_blocks == 0)
3043 return false;
3044
3045 if (sdkp->opt_xfer_blocks > dev_max) {
3046 sd_first_printk(KERN_WARNING, sdkp,
3047 "Optimal transfer size %u logical blocks " \
3048 "> dev_max (%u logical blocks)\n",
3049 sdkp->opt_xfer_blocks, dev_max);
3050 return false;
3051 }
3052
3053 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3054 sd_first_printk(KERN_WARNING, sdkp,
3055 "Optimal transfer size %u logical blocks " \
3056 "> sd driver limit (%u logical blocks)\n",
3057 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3058 return false;
3059 }
3060
3061 if (opt_xfer_bytes < PAGE_SIZE) {
3062 sd_first_printk(KERN_WARNING, sdkp,
3063 "Optimal transfer size %u bytes < " \
3064 "PAGE_SIZE (%u bytes)\n",
3065 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3066 return false;
3067 }
3068
3069 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3070 sd_first_printk(KERN_WARNING, sdkp,
3071 "Optimal transfer size %u bytes not a " \
3072 "multiple of physical block size (%u bytes)\n",
3073 opt_xfer_bytes, sdkp->physical_block_size);
3074 return false;
3075 }
3076
3077 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3078 opt_xfer_bytes);
3079 return true;
3080 }
3081
3082 /**
3083 * sd_revalidate_disk - called the first time a new disk is seen,
3084 * performs disk spin up, read_capacity, etc.
3085 * @disk: struct gendisk we care about
3086 **/
3087 static int sd_revalidate_disk(struct gendisk *disk)
3088 {
3089 struct scsi_disk *sdkp = scsi_disk(disk);
3090 struct scsi_device *sdp = sdkp->device;
3091 struct request_queue *q = sdkp->disk->queue;
3092 sector_t old_capacity = sdkp->capacity;
3093 unsigned char *buffer;
3094 unsigned int dev_max, rw_max;
3095
3096 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3097 "sd_revalidate_disk\n"));
3098
3099 /*
3100 * If the device is offline, don't try and read capacity or any
3101 * of the other niceties.
3102 */
3103 if (!scsi_device_online(sdp))
3104 goto out;
3105
3106 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3107 if (!buffer) {
3108 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3109 "allocation failure.\n");
3110 goto out;
3111 }
3112
3113 sd_spinup_disk(sdkp);
3114
3115 /*
3116 * Without media there is no reason to ask; moreover, some devices
3117 * react badly if we do.
3118 */
3119 if (sdkp->media_present) {
3120 sd_read_capacity(sdkp, buffer);
3121
3122 /*
3123 * set the default to rotational. All non-rotational devices
3124 * support the block characteristics VPD page, which will
3125 * cause this to be updated correctly and any device which
3126 * doesn't support it should be treated as rotational.
3127 */
3128 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3129 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3130
3131 if (scsi_device_supports_vpd(sdp)) {
3132 sd_read_block_provisioning(sdkp);
3133 sd_read_block_limits(sdkp);
3134 sd_read_block_characteristics(sdkp);
3135 sd_zbc_read_zones(sdkp, buffer);
3136 }
3137
3138 sd_print_capacity(sdkp, old_capacity);
3139
3140 sd_read_write_protect_flag(sdkp, buffer);
3141 sd_read_cache_type(sdkp, buffer);
3142 sd_read_app_tag_own(sdkp, buffer);
3143 sd_read_write_same(sdkp, buffer);
3144 sd_read_security(sdkp, buffer);
3145 }
3146
3147 /*
3148 * We now have all cache related info, determine how we deal
3149 * with flush requests.
3150 */
3151 sd_set_flush_flag(sdkp);
3152
3153 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3154 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3155
3156 /* Some devices report a maximum block count for READ/WRITE requests. */
3157 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3158 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3159
3160 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3161 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3162 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3163 } else
3164 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3165 (sector_t)BLK_DEF_MAX_SECTORS);
3166
3167 /* Do not exceed controller limit */
3168 rw_max = min(rw_max, queue_max_hw_sectors(q));
3169
3170 /*
3171 * Only update max_sectors if previously unset or if the current value
3172 * exceeds the capabilities of the hardware.
3173 */
3174 if (sdkp->first_scan ||
3175 q->limits.max_sectors > q->limits.max_dev_sectors ||
3176 q->limits.max_sectors > q->limits.max_hw_sectors)
3177 q->limits.max_sectors = rw_max;
3178
3179 sdkp->first_scan = 0;
3180
3181 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3182 sd_config_write_same(sdkp);
3183 kfree(buffer);
3184
3185 out:
3186 return 0;
3187 }
3188
3189 /**
3190 * sd_unlock_native_capacity - unlock native capacity
3191 * @disk: struct gendisk to set capacity for
3192 *
3193 * Block layer calls this function if it detects that partitions
3194 * on @disk reach beyond the end of the device. If the SCSI host
3195 * implements ->unlock_native_capacity() method, it's invoked to
3196 * give it a chance to adjust the device capacity.
3197 *
3198 * CONTEXT:
3199 * Defined by block layer. Might sleep.
3200 */
3201 static void sd_unlock_native_capacity(struct gendisk *disk)
3202 {
3203 struct scsi_device *sdev = scsi_disk(disk)->device;
3204
3205 if (sdev->host->hostt->unlock_native_capacity)
3206 sdev->host->hostt->unlock_native_capacity(sdev);
3207 }
3208
3209 /**
3210 * sd_format_disk_name - format disk name
3211 * @prefix: name prefix - ie. "sd" for SCSI disks
3212 * @index: index of the disk to format name for
3213 * @buf: output buffer
3214 * @buflen: length of the output buffer
3215 *
3216 * SCSI disk names starts at sda. The 26th device is sdz and the
3217 * 27th is sdaa. The last one for two lettered suffix is sdzz
3218 * which is followed by sdaaa.
3219 *
3220 * This is basically 26 base counting with one extra 'nil' entry
3221 * at the beginning from the second digit on and can be
3222 * determined using similar method as 26 base conversion with the
3223 * index shifted -1 after each digit is computed.
3224 *
3225 * CONTEXT:
3226 * Don't care.
3227 *
3228 * RETURNS:
3229 * 0 on success, -errno on failure.
3230 */
3231 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3232 {
3233 const int base = 'z' - 'a' + 1;
3234 char *begin = buf + strlen(prefix);
3235 char *end = buf + buflen;
3236 char *p;
3237 int unit;
3238
3239 p = end - 1;
3240 *p = '\0';
3241 unit = base;
3242 do {
3243 if (p == begin)
3244 return -EINVAL;
3245 *--p = 'a' + (index % unit);
3246 index = (index / unit) - 1;
3247 } while (index >= 0);
3248
3249 memmove(begin, p, end - p);
3250 memcpy(buf, prefix, strlen(prefix));
3251
3252 return 0;
3253 }
3254
3255 /*
3256 * The asynchronous part of sd_probe
3257 */
3258 static void sd_probe_async(void *data, async_cookie_t cookie)
3259 {
3260 struct scsi_disk *sdkp = data;
3261 struct scsi_device *sdp;
3262 struct gendisk *gd;
3263 u32 index;
3264 struct device *dev;
3265
3266 sdp = sdkp->device;
3267 gd = sdkp->disk;
3268 index = sdkp->index;
3269 dev = &sdp->sdev_gendev;
3270
3271 gd->major = sd_major((index & 0xf0) >> 4);
3272 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3273
3274 gd->fops = &sd_fops;
3275 gd->private_data = &sdkp->driver;
3276 gd->queue = sdkp->device->request_queue;
3277
3278 /* defaults, until the device tells us otherwise */
3279 sdp->sector_size = 512;
3280 sdkp->capacity = 0;
3281 sdkp->media_present = 1;
3282 sdkp->write_prot = 0;
3283 sdkp->cache_override = 0;
3284 sdkp->WCE = 0;
3285 sdkp->RCD = 0;
3286 sdkp->ATO = 0;
3287 sdkp->first_scan = 1;
3288 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3289
3290 sd_revalidate_disk(gd);
3291
3292 gd->flags = GENHD_FL_EXT_DEVT;
3293 if (sdp->removable) {
3294 gd->flags |= GENHD_FL_REMOVABLE;
3295 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3296 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3297 }
3298
3299 blk_pm_runtime_init(sdp->request_queue, dev);
3300 device_add_disk(dev, gd, NULL);
3301 if (sdkp->capacity)
3302 sd_dif_config_host(sdkp);
3303
3304 sd_revalidate_disk(gd);
3305
3306 if (sdkp->security) {
3307 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3308 if (sdkp->opal_dev)
3309 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3310 }
3311
3312 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3313 sdp->removable ? "removable " : "");
3314 scsi_autopm_put_device(sdp);
3315 put_device(&sdkp->dev);
3316 }
3317
3318 /**
3319 * sd_probe - called during driver initialization and whenever a
3320 * new scsi device is attached to the system. It is called once
3321 * for each scsi device (not just disks) present.
3322 * @dev: pointer to device object
3323 *
3324 * Returns 0 if successful (or not interested in this scsi device
3325 * (e.g. scanner)); 1 when there is an error.
3326 *
3327 * Note: this function is invoked from the scsi mid-level.
3328 * This function sets up the mapping between a given
3329 * <host,channel,id,lun> (found in sdp) and new device name
3330 * (e.g. /dev/sda). More precisely it is the block device major
3331 * and minor number that is chosen here.
3332 *
3333 * Assume sd_probe is not re-entrant (for time being)
3334 * Also think about sd_probe() and sd_remove() running coincidentally.
3335 **/
3336 static int sd_probe(struct device *dev)
3337 {
3338 struct scsi_device *sdp = to_scsi_device(dev);
3339 struct scsi_disk *sdkp;
3340 struct gendisk *gd;
3341 int index;
3342 int error;
3343
3344 scsi_autopm_get_device(sdp);
3345 error = -ENODEV;
3346 if (sdp->type != TYPE_DISK &&
3347 sdp->type != TYPE_ZBC &&
3348 sdp->type != TYPE_MOD &&
3349 sdp->type != TYPE_RBC)
3350 goto out;
3351
3352 #ifndef CONFIG_BLK_DEV_ZONED
3353 if (sdp->type == TYPE_ZBC)
3354 goto out;
3355 #endif
3356 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3357 "sd_probe\n"));
3358
3359 error = -ENOMEM;
3360 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3361 if (!sdkp)
3362 goto out;
3363
3364 gd = alloc_disk(SD_MINORS);
3365 if (!gd)
3366 goto out_free;
3367
3368 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3369 if (index < 0) {
3370 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3371 goto out_put;
3372 }
3373
3374 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3375 if (error) {
3376 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3377 goto out_free_index;
3378 }
3379
3380 sdkp->device = sdp;
3381 sdkp->driver = &sd_template;
3382 sdkp->disk = gd;
3383 sdkp->index = index;
3384 atomic_set(&sdkp->openers, 0);
3385 atomic_set(&sdkp->device->ioerr_cnt, 0);
3386
3387 if (!sdp->request_queue->rq_timeout) {
3388 if (sdp->type != TYPE_MOD)
3389 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3390 else
3391 blk_queue_rq_timeout(sdp->request_queue,
3392 SD_MOD_TIMEOUT);
3393 }
3394
3395 device_initialize(&sdkp->dev);
3396 sdkp->dev.parent = dev;
3397 sdkp->dev.class = &sd_disk_class;
3398 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3399
3400 error = device_add(&sdkp->dev);
3401 if (error)
3402 goto out_free_index;
3403
3404 get_device(dev);
3405 dev_set_drvdata(dev, sdkp);
3406
3407 get_device(&sdkp->dev); /* prevent release before async_schedule */
3408 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3409
3410 return 0;
3411
3412 out_free_index:
3413 ida_free(&sd_index_ida, index);
3414 out_put:
3415 put_disk(gd);
3416 out_free:
3417 kfree(sdkp);
3418 out:
3419 scsi_autopm_put_device(sdp);
3420 return error;
3421 }
3422
3423 /**
3424 * sd_remove - called whenever a scsi disk (previously recognized by
3425 * sd_probe) is detached from the system. It is called (potentially
3426 * multiple times) during sd module unload.
3427 * @dev: pointer to device object
3428 *
3429 * Note: this function is invoked from the scsi mid-level.
3430 * This function potentially frees up a device name (e.g. /dev/sdc)
3431 * that could be re-used by a subsequent sd_probe().
3432 * This function is not called when the built-in sd driver is "exit-ed".
3433 **/
3434 static int sd_remove(struct device *dev)
3435 {
3436 struct scsi_disk *sdkp;
3437 dev_t devt;
3438
3439 sdkp = dev_get_drvdata(dev);
3440 devt = disk_devt(sdkp->disk);
3441 scsi_autopm_get_device(sdkp->device);
3442
3443 async_synchronize_full_domain(&scsi_sd_pm_domain);
3444 async_synchronize_full_domain(&scsi_sd_probe_domain);
3445 device_del(&sdkp->dev);
3446 del_gendisk(sdkp->disk);
3447 sd_shutdown(dev);
3448
3449 free_opal_dev(sdkp->opal_dev);
3450
3451 blk_register_region(devt, SD_MINORS, NULL,
3452 sd_default_probe, NULL, NULL);
3453
3454 mutex_lock(&sd_ref_mutex);
3455 dev_set_drvdata(dev, NULL);
3456 put_device(&sdkp->dev);
3457 mutex_unlock(&sd_ref_mutex);
3458
3459 return 0;
3460 }
3461
3462 /**
3463 * scsi_disk_release - Called to free the scsi_disk structure
3464 * @dev: pointer to embedded class device
3465 *
3466 * sd_ref_mutex must be held entering this routine. Because it is
3467 * called on last put, you should always use the scsi_disk_get()
3468 * scsi_disk_put() helpers which manipulate the semaphore directly
3469 * and never do a direct put_device.
3470 **/
3471 static void scsi_disk_release(struct device *dev)
3472 {
3473 struct scsi_disk *sdkp = to_scsi_disk(dev);
3474 struct gendisk *disk = sdkp->disk;
3475 struct request_queue *q = disk->queue;
3476
3477 ida_free(&sd_index_ida, sdkp->index);
3478
3479 /*
3480 * Wait until all requests that are in progress have completed.
3481 * This is necessary to avoid that e.g. scsi_end_request() crashes
3482 * due to clearing the disk->private_data pointer. Wait from inside
3483 * scsi_disk_release() instead of from sd_release() to avoid that
3484 * freezing and unfreezing the request queue affects user space I/O
3485 * in case multiple processes open a /dev/sd... node concurrently.
3486 */
3487 blk_mq_freeze_queue(q);
3488 blk_mq_unfreeze_queue(q);
3489
3490 disk->private_data = NULL;
3491 put_disk(disk);
3492 put_device(&sdkp->device->sdev_gendev);
3493
3494 kfree(sdkp);
3495 }
3496
3497 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3498 {
3499 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3500 struct scsi_sense_hdr sshdr;
3501 struct scsi_device *sdp = sdkp->device;
3502 int res;
3503
3504 if (start)
3505 cmd[4] |= 1; /* START */
3506
3507 if (sdp->start_stop_pwr_cond)
3508 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3509
3510 if (!scsi_device_online(sdp))
3511 return -ENODEV;
3512
3513 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3514 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3515 if (res) {
3516 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3517 if (driver_byte(res) == DRIVER_SENSE)
3518 sd_print_sense_hdr(sdkp, &sshdr);
3519 if (scsi_sense_valid(&sshdr) &&
3520 /* 0x3a is medium not present */
3521 sshdr.asc == 0x3a)
3522 res = 0;
3523 }
3524
3525 /* SCSI error codes must not go to the generic layer */
3526 if (res)
3527 return -EIO;
3528
3529 return 0;
3530 }
3531
3532 /*
3533 * Send a SYNCHRONIZE CACHE instruction down to the device through
3534 * the normal SCSI command structure. Wait for the command to
3535 * complete.
3536 */
3537 static void sd_shutdown(struct device *dev)
3538 {
3539 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3540
3541 if (!sdkp)
3542 return; /* this can happen */
3543
3544 if (pm_runtime_suspended(dev))
3545 return;
3546
3547 if (sdkp->WCE && sdkp->media_present) {
3548 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3549 sd_sync_cache(sdkp, NULL);
3550 }
3551
3552 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3553 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3554 sd_start_stop_device(sdkp, 0);
3555 }
3556 }
3557
3558 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3559 {
3560 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3561 struct scsi_sense_hdr sshdr;
3562 int ret = 0;
3563
3564 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3565 return 0;
3566
3567 if (sdkp->WCE && sdkp->media_present) {
3568 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3569 ret = sd_sync_cache(sdkp, &sshdr);
3570
3571 if (ret) {
3572 /* ignore OFFLINE device */
3573 if (ret == -ENODEV)
3574 return 0;
3575
3576 if (!scsi_sense_valid(&sshdr) ||
3577 sshdr.sense_key != ILLEGAL_REQUEST)
3578 return ret;
3579
3580 /*
3581 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3582 * doesn't support sync. There's not much to do and
3583 * suspend shouldn't fail.
3584 */
3585 ret = 0;
3586 }
3587 }
3588
3589 if (sdkp->device->manage_start_stop) {
3590 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3591 /* an error is not worth aborting a system sleep */
3592 ret = sd_start_stop_device(sdkp, 0);
3593 if (ignore_stop_errors)
3594 ret = 0;
3595 }
3596
3597 return ret;
3598 }
3599
3600 static int sd_suspend_system(struct device *dev)
3601 {
3602 return sd_suspend_common(dev, true);
3603 }
3604
3605 static int sd_suspend_runtime(struct device *dev)
3606 {
3607 return sd_suspend_common(dev, false);
3608 }
3609
3610 static int sd_resume(struct device *dev)
3611 {
3612 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3613 int ret;
3614
3615 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3616 return 0;
3617
3618 if (!sdkp->device->manage_start_stop)
3619 return 0;
3620
3621 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3622 ret = sd_start_stop_device(sdkp, 1);
3623 if (!ret)
3624 opal_unlock_from_suspend(sdkp->opal_dev);
3625 return ret;
3626 }
3627
3628 /**
3629 * init_sd - entry point for this driver (both when built in or when
3630 * a module).
3631 *
3632 * Note: this function registers this driver with the scsi mid-level.
3633 **/
3634 static int __init init_sd(void)
3635 {
3636 int majors = 0, i, err;
3637
3638 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3639
3640 for (i = 0; i < SD_MAJORS; i++) {
3641 if (register_blkdev(sd_major(i), "sd") != 0)
3642 continue;
3643 majors++;
3644 blk_register_region(sd_major(i), SD_MINORS, NULL,
3645 sd_default_probe, NULL, NULL);
3646 }
3647
3648 if (!majors)
3649 return -ENODEV;
3650
3651 err = class_register(&sd_disk_class);
3652 if (err)
3653 goto err_out;
3654
3655 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3656 0, 0, NULL);
3657 if (!sd_cdb_cache) {
3658 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3659 err = -ENOMEM;
3660 goto err_out_class;
3661 }
3662
3663 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3664 if (!sd_cdb_pool) {
3665 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3666 err = -ENOMEM;
3667 goto err_out_cache;
3668 }
3669
3670 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3671 if (!sd_page_pool) {
3672 printk(KERN_ERR "sd: can't init discard page pool\n");
3673 err = -ENOMEM;
3674 goto err_out_ppool;
3675 }
3676
3677 err = scsi_register_driver(&sd_template.gendrv);
3678 if (err)
3679 goto err_out_driver;
3680
3681 return 0;
3682
3683 err_out_driver:
3684 mempool_destroy(sd_page_pool);
3685
3686 err_out_ppool:
3687 mempool_destroy(sd_cdb_pool);
3688
3689 err_out_cache:
3690 kmem_cache_destroy(sd_cdb_cache);
3691
3692 err_out_class:
3693 class_unregister(&sd_disk_class);
3694 err_out:
3695 for (i = 0; i < SD_MAJORS; i++)
3696 unregister_blkdev(sd_major(i), "sd");
3697 return err;
3698 }
3699
3700 /**
3701 * exit_sd - exit point for this driver (when it is a module).
3702 *
3703 * Note: this function unregisters this driver from the scsi mid-level.
3704 **/
3705 static void __exit exit_sd(void)
3706 {
3707 int i;
3708
3709 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3710
3711 scsi_unregister_driver(&sd_template.gendrv);
3712 mempool_destroy(sd_cdb_pool);
3713 mempool_destroy(sd_page_pool);
3714 kmem_cache_destroy(sd_cdb_cache);
3715
3716 class_unregister(&sd_disk_class);
3717
3718 for (i = 0; i < SD_MAJORS; i++) {
3719 blk_unregister_region(sd_major(i), SD_MINORS);
3720 unregister_blkdev(sd_major(i), "sd");
3721 }
3722 }
3723
3724 module_init(init_sd);
3725 module_exit(exit_sd);
3726
3727 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3728 struct scsi_sense_hdr *sshdr)
3729 {
3730 scsi_print_sense_hdr(sdkp->device,
3731 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3732 }
3733
3734 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3735 int result)
3736 {
3737 const char *hb_string = scsi_hostbyte_string(result);
3738 const char *db_string = scsi_driverbyte_string(result);
3739
3740 if (hb_string || db_string)
3741 sd_printk(KERN_INFO, sdkp,
3742 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3743 hb_string ? hb_string : "invalid",
3744 db_string ? db_string : "invalid");
3745 else
3746 sd_printk(KERN_INFO, sdkp,
3747 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3748 msg, host_byte(result), driver_byte(result));
3749 }
3750
Linux with added FPC-III support