Loading...
1/*
2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4 *
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24 * Support 32k/1M disks.
25 *
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
33 */
34
35#include <linux/module.h>
36#include <linux/fs.h>
37#include <linux/kernel.h>
38#include <linux/mm.h>
39#include <linux/bio.h>
40#include <linux/genhd.h>
41#include <linux/hdreg.h>
42#include <linux/errno.h>
43#include <linux/idr.h>
44#include <linux/interrupt.h>
45#include <linux/init.h>
46#include <linux/blkdev.h>
47#include <linux/blkpg.h>
48#include <linux/delay.h>
49#include <linux/mutex.h>
50#include <linux/string_helpers.h>
51#include <linux/async.h>
52#include <linux/slab.h>
53#include <linux/pm_runtime.h>
54#include <linux/pr.h>
55#include <linux/t10-pi.h>
56#include <linux/uaccess.h>
57#include <asm/unaligned.h>
58
59#include <scsi/scsi.h>
60#include <scsi/scsi_cmnd.h>
61#include <scsi/scsi_dbg.h>
62#include <scsi/scsi_device.h>
63#include <scsi/scsi_driver.h>
64#include <scsi/scsi_eh.h>
65#include <scsi/scsi_host.h>
66#include <scsi/scsi_ioctl.h>
67#include <scsi/scsicam.h>
68
69#include "sd.h"
70#include "scsi_priv.h"
71#include "scsi_logging.h"
72
73MODULE_AUTHOR("Eric Youngdale");
74MODULE_DESCRIPTION("SCSI disk (sd) driver");
75MODULE_LICENSE("GPL");
76
77MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
78MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
79MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
80MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
93MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
94MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
95MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
96MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
97
98#if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
99#define SD_MINORS 16
100#else
101#define SD_MINORS 0
102#endif
103
104static void sd_config_discard(struct scsi_disk *, unsigned int);
105static void sd_config_write_same(struct scsi_disk *);
106static int sd_revalidate_disk(struct gendisk *);
107static void sd_unlock_native_capacity(struct gendisk *disk);
108static int sd_probe(struct device *);
109static int sd_remove(struct device *);
110static void sd_shutdown(struct device *);
111static int sd_suspend_system(struct device *);
112static int sd_suspend_runtime(struct device *);
113static int sd_resume(struct device *);
114static void sd_rescan(struct device *);
115static int sd_init_command(struct scsi_cmnd *SCpnt);
116static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117static int sd_done(struct scsi_cmnd *);
118static int sd_eh_action(struct scsi_cmnd *, int);
119static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
120static void scsi_disk_release(struct device *cdev);
121static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
122static void sd_print_result(const struct scsi_disk *, const char *, int);
123
124static DEFINE_SPINLOCK(sd_index_lock);
125static DEFINE_IDA(sd_index_ida);
126
127/* This semaphore is used to mediate the 0->1 reference get in the
128 * face of object destruction (i.e. we can't allow a get on an
129 * object after last put) */
130static DEFINE_MUTEX(sd_ref_mutex);
131
132static struct kmem_cache *sd_cdb_cache;
133static mempool_t *sd_cdb_pool;
134
135static const char *sd_cache_types[] = {
136 "write through", "none", "write back",
137 "write back, no read (daft)"
138};
139
140static void sd_set_flush_flag(struct scsi_disk *sdkp)
141{
142 bool wc = false, fua = false;
143
144 if (sdkp->WCE) {
145 wc = true;
146 if (sdkp->DPOFUA)
147 fua = true;
148 }
149
150 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
151}
152
153static ssize_t
154cache_type_store(struct device *dev, struct device_attribute *attr,
155 const char *buf, size_t count)
156{
157 int i, ct = -1, rcd, wce, sp;
158 struct scsi_disk *sdkp = to_scsi_disk(dev);
159 struct scsi_device *sdp = sdkp->device;
160 char buffer[64];
161 char *buffer_data;
162 struct scsi_mode_data data;
163 struct scsi_sense_hdr sshdr;
164 static const char temp[] = "temporary ";
165 int len;
166
167 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
168 /* no cache control on RBC devices; theoretically they
169 * can do it, but there's probably so many exceptions
170 * it's not worth the risk */
171 return -EINVAL;
172
173 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
174 buf += sizeof(temp) - 1;
175 sdkp->cache_override = 1;
176 } else {
177 sdkp->cache_override = 0;
178 }
179
180 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
181 len = strlen(sd_cache_types[i]);
182 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
183 buf[len] == '\n') {
184 ct = i;
185 break;
186 }
187 }
188 if (ct < 0)
189 return -EINVAL;
190 rcd = ct & 0x01 ? 1 : 0;
191 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
192
193 if (sdkp->cache_override) {
194 sdkp->WCE = wce;
195 sdkp->RCD = rcd;
196 sd_set_flush_flag(sdkp);
197 return count;
198 }
199
200 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
201 SD_MAX_RETRIES, &data, NULL))
202 return -EINVAL;
203 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
204 data.block_descriptor_length);
205 buffer_data = buffer + data.header_length +
206 data.block_descriptor_length;
207 buffer_data[2] &= ~0x05;
208 buffer_data[2] |= wce << 2 | rcd;
209 sp = buffer_data[0] & 0x80 ? 1 : 0;
210 buffer_data[0] &= ~0x80;
211
212 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
213 SD_MAX_RETRIES, &data, &sshdr)) {
214 if (scsi_sense_valid(&sshdr))
215 sd_print_sense_hdr(sdkp, &sshdr);
216 return -EINVAL;
217 }
218 revalidate_disk(sdkp->disk);
219 return count;
220}
221
222static ssize_t
223manage_start_stop_show(struct device *dev, struct device_attribute *attr,
224 char *buf)
225{
226 struct scsi_disk *sdkp = to_scsi_disk(dev);
227 struct scsi_device *sdp = sdkp->device;
228
229 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
230}
231
232static ssize_t
233manage_start_stop_store(struct device *dev, struct device_attribute *attr,
234 const char *buf, size_t count)
235{
236 struct scsi_disk *sdkp = to_scsi_disk(dev);
237 struct scsi_device *sdp = sdkp->device;
238
239 if (!capable(CAP_SYS_ADMIN))
240 return -EACCES;
241
242 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
243
244 return count;
245}
246static DEVICE_ATTR_RW(manage_start_stop);
247
248static ssize_t
249allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
250{
251 struct scsi_disk *sdkp = to_scsi_disk(dev);
252
253 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
254}
255
256static ssize_t
257allow_restart_store(struct device *dev, struct device_attribute *attr,
258 const char *buf, size_t count)
259{
260 struct scsi_disk *sdkp = to_scsi_disk(dev);
261 struct scsi_device *sdp = sdkp->device;
262
263 if (!capable(CAP_SYS_ADMIN))
264 return -EACCES;
265
266 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
267 return -EINVAL;
268
269 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
270
271 return count;
272}
273static DEVICE_ATTR_RW(allow_restart);
274
275static ssize_t
276cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
277{
278 struct scsi_disk *sdkp = to_scsi_disk(dev);
279 int ct = sdkp->RCD + 2*sdkp->WCE;
280
281 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
282}
283static DEVICE_ATTR_RW(cache_type);
284
285static ssize_t
286FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
287{
288 struct scsi_disk *sdkp = to_scsi_disk(dev);
289
290 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
291}
292static DEVICE_ATTR_RO(FUA);
293
294static ssize_t
295protection_type_show(struct device *dev, struct device_attribute *attr,
296 char *buf)
297{
298 struct scsi_disk *sdkp = to_scsi_disk(dev);
299
300 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
301}
302
303static ssize_t
304protection_type_store(struct device *dev, struct device_attribute *attr,
305 const char *buf, size_t count)
306{
307 struct scsi_disk *sdkp = to_scsi_disk(dev);
308 unsigned int val;
309 int err;
310
311 if (!capable(CAP_SYS_ADMIN))
312 return -EACCES;
313
314 err = kstrtouint(buf, 10, &val);
315
316 if (err)
317 return err;
318
319 if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
320 sdkp->protection_type = val;
321
322 return count;
323}
324static DEVICE_ATTR_RW(protection_type);
325
326static ssize_t
327protection_mode_show(struct device *dev, struct device_attribute *attr,
328 char *buf)
329{
330 struct scsi_disk *sdkp = to_scsi_disk(dev);
331 struct scsi_device *sdp = sdkp->device;
332 unsigned int dif, dix;
333
334 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
335 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
336
337 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
338 dif = 0;
339 dix = 1;
340 }
341
342 if (!dif && !dix)
343 return snprintf(buf, 20, "none\n");
344
345 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
346}
347static DEVICE_ATTR_RO(protection_mode);
348
349static ssize_t
350app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
351{
352 struct scsi_disk *sdkp = to_scsi_disk(dev);
353
354 return snprintf(buf, 20, "%u\n", sdkp->ATO);
355}
356static DEVICE_ATTR_RO(app_tag_own);
357
358static ssize_t
359thin_provisioning_show(struct device *dev, struct device_attribute *attr,
360 char *buf)
361{
362 struct scsi_disk *sdkp = to_scsi_disk(dev);
363
364 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
365}
366static DEVICE_ATTR_RO(thin_provisioning);
367
368static const char *lbp_mode[] = {
369 [SD_LBP_FULL] = "full",
370 [SD_LBP_UNMAP] = "unmap",
371 [SD_LBP_WS16] = "writesame_16",
372 [SD_LBP_WS10] = "writesame_10",
373 [SD_LBP_ZERO] = "writesame_zero",
374 [SD_LBP_DISABLE] = "disabled",
375};
376
377static ssize_t
378provisioning_mode_show(struct device *dev, struct device_attribute *attr,
379 char *buf)
380{
381 struct scsi_disk *sdkp = to_scsi_disk(dev);
382
383 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
384}
385
386static ssize_t
387provisioning_mode_store(struct device *dev, struct device_attribute *attr,
388 const char *buf, size_t count)
389{
390 struct scsi_disk *sdkp = to_scsi_disk(dev);
391 struct scsi_device *sdp = sdkp->device;
392
393 if (!capable(CAP_SYS_ADMIN))
394 return -EACCES;
395
396 if (sd_is_zoned(sdkp)) {
397 sd_config_discard(sdkp, SD_LBP_DISABLE);
398 return count;
399 }
400
401 if (sdp->type != TYPE_DISK)
402 return -EINVAL;
403
404 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
405 sd_config_discard(sdkp, SD_LBP_UNMAP);
406 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
407 sd_config_discard(sdkp, SD_LBP_WS16);
408 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
409 sd_config_discard(sdkp, SD_LBP_WS10);
410 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
411 sd_config_discard(sdkp, SD_LBP_ZERO);
412 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
413 sd_config_discard(sdkp, SD_LBP_DISABLE);
414 else
415 return -EINVAL;
416
417 return count;
418}
419static DEVICE_ATTR_RW(provisioning_mode);
420
421static ssize_t
422max_medium_access_timeouts_show(struct device *dev,
423 struct device_attribute *attr, char *buf)
424{
425 struct scsi_disk *sdkp = to_scsi_disk(dev);
426
427 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
428}
429
430static ssize_t
431max_medium_access_timeouts_store(struct device *dev,
432 struct device_attribute *attr, const char *buf,
433 size_t count)
434{
435 struct scsi_disk *sdkp = to_scsi_disk(dev);
436 int err;
437
438 if (!capable(CAP_SYS_ADMIN))
439 return -EACCES;
440
441 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
442
443 return err ? err : count;
444}
445static DEVICE_ATTR_RW(max_medium_access_timeouts);
446
447static ssize_t
448max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
449 char *buf)
450{
451 struct scsi_disk *sdkp = to_scsi_disk(dev);
452
453 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
454}
455
456static ssize_t
457max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
458 const char *buf, size_t count)
459{
460 struct scsi_disk *sdkp = to_scsi_disk(dev);
461 struct scsi_device *sdp = sdkp->device;
462 unsigned long max;
463 int err;
464
465 if (!capable(CAP_SYS_ADMIN))
466 return -EACCES;
467
468 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
469 return -EINVAL;
470
471 err = kstrtoul(buf, 10, &max);
472
473 if (err)
474 return err;
475
476 if (max == 0)
477 sdp->no_write_same = 1;
478 else if (max <= SD_MAX_WS16_BLOCKS) {
479 sdp->no_write_same = 0;
480 sdkp->max_ws_blocks = max;
481 }
482
483 sd_config_write_same(sdkp);
484
485 return count;
486}
487static DEVICE_ATTR_RW(max_write_same_blocks);
488
489static struct attribute *sd_disk_attrs[] = {
490 &dev_attr_cache_type.attr,
491 &dev_attr_FUA.attr,
492 &dev_attr_allow_restart.attr,
493 &dev_attr_manage_start_stop.attr,
494 &dev_attr_protection_type.attr,
495 &dev_attr_protection_mode.attr,
496 &dev_attr_app_tag_own.attr,
497 &dev_attr_thin_provisioning.attr,
498 &dev_attr_provisioning_mode.attr,
499 &dev_attr_max_write_same_blocks.attr,
500 &dev_attr_max_medium_access_timeouts.attr,
501 NULL,
502};
503ATTRIBUTE_GROUPS(sd_disk);
504
505static struct class sd_disk_class = {
506 .name = "scsi_disk",
507 .owner = THIS_MODULE,
508 .dev_release = scsi_disk_release,
509 .dev_groups = sd_disk_groups,
510};
511
512static const struct dev_pm_ops sd_pm_ops = {
513 .suspend = sd_suspend_system,
514 .resume = sd_resume,
515 .poweroff = sd_suspend_system,
516 .restore = sd_resume,
517 .runtime_suspend = sd_suspend_runtime,
518 .runtime_resume = sd_resume,
519};
520
521static struct scsi_driver sd_template = {
522 .gendrv = {
523 .name = "sd",
524 .owner = THIS_MODULE,
525 .probe = sd_probe,
526 .remove = sd_remove,
527 .shutdown = sd_shutdown,
528 .pm = &sd_pm_ops,
529 },
530 .rescan = sd_rescan,
531 .init_command = sd_init_command,
532 .uninit_command = sd_uninit_command,
533 .done = sd_done,
534 .eh_action = sd_eh_action,
535};
536
537/*
538 * Dummy kobj_map->probe function.
539 * The default ->probe function will call modprobe, which is
540 * pointless as this module is already loaded.
541 */
542static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
543{
544 return NULL;
545}
546
547/*
548 * Device no to disk mapping:
549 *
550 * major disc2 disc p1
551 * |............|.............|....|....| <- dev_t
552 * 31 20 19 8 7 4 3 0
553 *
554 * Inside a major, we have 16k disks, however mapped non-
555 * contiguously. The first 16 disks are for major0, the next
556 * ones with major1, ... Disk 256 is for major0 again, disk 272
557 * for major1, ...
558 * As we stay compatible with our numbering scheme, we can reuse
559 * the well-know SCSI majors 8, 65--71, 136--143.
560 */
561static int sd_major(int major_idx)
562{
563 switch (major_idx) {
564 case 0:
565 return SCSI_DISK0_MAJOR;
566 case 1 ... 7:
567 return SCSI_DISK1_MAJOR + major_idx - 1;
568 case 8 ... 15:
569 return SCSI_DISK8_MAJOR + major_idx - 8;
570 default:
571 BUG();
572 return 0; /* shut up gcc */
573 }
574}
575
576static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
577{
578 struct scsi_disk *sdkp = NULL;
579
580 mutex_lock(&sd_ref_mutex);
581
582 if (disk->private_data) {
583 sdkp = scsi_disk(disk);
584 if (scsi_device_get(sdkp->device) == 0)
585 get_device(&sdkp->dev);
586 else
587 sdkp = NULL;
588 }
589 mutex_unlock(&sd_ref_mutex);
590 return sdkp;
591}
592
593static void scsi_disk_put(struct scsi_disk *sdkp)
594{
595 struct scsi_device *sdev = sdkp->device;
596
597 mutex_lock(&sd_ref_mutex);
598 put_device(&sdkp->dev);
599 scsi_device_put(sdev);
600 mutex_unlock(&sd_ref_mutex);
601}
602
603static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
604 unsigned int dix, unsigned int dif)
605{
606 struct bio *bio = scmd->request->bio;
607 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
608 unsigned int protect = 0;
609
610 if (dix) { /* DIX Type 0, 1, 2, 3 */
611 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
612 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
613
614 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
615 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
616 }
617
618 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
619 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
620
621 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
622 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
623 }
624
625 if (dif) { /* DIX/DIF Type 1, 2, 3 */
626 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
627
628 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
629 protect = 3 << 5; /* Disable target PI checking */
630 else
631 protect = 1 << 5; /* Enable target PI checking */
632 }
633
634 scsi_set_prot_op(scmd, prot_op);
635 scsi_set_prot_type(scmd, dif);
636 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
637
638 return protect;
639}
640
641static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
642{
643 struct request_queue *q = sdkp->disk->queue;
644 unsigned int logical_block_size = sdkp->device->sector_size;
645 unsigned int max_blocks = 0;
646
647 q->limits.discard_zeroes_data = 0;
648
649 /*
650 * When LBPRZ is reported, discard alignment and granularity
651 * must be fixed to the logical block size. Otherwise the block
652 * layer will drop misaligned portions of the request which can
653 * lead to data corruption. If LBPRZ is not set, we honor the
654 * device preference.
655 */
656 if (sdkp->lbprz) {
657 q->limits.discard_alignment = 0;
658 q->limits.discard_granularity = logical_block_size;
659 } else {
660 q->limits.discard_alignment = sdkp->unmap_alignment *
661 logical_block_size;
662 q->limits.discard_granularity =
663 max(sdkp->physical_block_size,
664 sdkp->unmap_granularity * logical_block_size);
665 }
666
667 sdkp->provisioning_mode = mode;
668
669 switch (mode) {
670
671 case SD_LBP_DISABLE:
672 blk_queue_max_discard_sectors(q, 0);
673 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
674 return;
675
676 case SD_LBP_UNMAP:
677 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
678 (u32)SD_MAX_WS16_BLOCKS);
679 break;
680
681 case SD_LBP_WS16:
682 max_blocks = min_not_zero(sdkp->max_ws_blocks,
683 (u32)SD_MAX_WS16_BLOCKS);
684 q->limits.discard_zeroes_data = sdkp->lbprz;
685 break;
686
687 case SD_LBP_WS10:
688 max_blocks = min_not_zero(sdkp->max_ws_blocks,
689 (u32)SD_MAX_WS10_BLOCKS);
690 q->limits.discard_zeroes_data = sdkp->lbprz;
691 break;
692
693 case SD_LBP_ZERO:
694 max_blocks = min_not_zero(sdkp->max_ws_blocks,
695 (u32)SD_MAX_WS10_BLOCKS);
696 q->limits.discard_zeroes_data = 1;
697 break;
698 }
699
700 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
701 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
702}
703
704/**
705 * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
706 * @sdp: scsi device to operate one
707 * @rq: Request to prepare
708 *
709 * Will issue either UNMAP or WRITE SAME(16) depending on preference
710 * indicated by target device.
711 **/
712static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
713{
714 struct request *rq = cmd->request;
715 struct scsi_device *sdp = cmd->device;
716 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
717 sector_t sector = blk_rq_pos(rq);
718 unsigned int nr_sectors = blk_rq_sectors(rq);
719 unsigned int len;
720 int ret;
721 char *buf;
722 struct page *page;
723
724 sector >>= ilog2(sdp->sector_size) - 9;
725 nr_sectors >>= ilog2(sdp->sector_size) - 9;
726
727 page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
728 if (!page)
729 return BLKPREP_DEFER;
730
731 switch (sdkp->provisioning_mode) {
732 case SD_LBP_UNMAP:
733 buf = page_address(page);
734
735 cmd->cmd_len = 10;
736 cmd->cmnd[0] = UNMAP;
737 cmd->cmnd[8] = 24;
738
739 put_unaligned_be16(6 + 16, &buf[0]);
740 put_unaligned_be16(16, &buf[2]);
741 put_unaligned_be64(sector, &buf[8]);
742 put_unaligned_be32(nr_sectors, &buf[16]);
743
744 len = 24;
745 break;
746
747 case SD_LBP_WS16:
748 cmd->cmd_len = 16;
749 cmd->cmnd[0] = WRITE_SAME_16;
750 cmd->cmnd[1] = 0x8; /* UNMAP */
751 put_unaligned_be64(sector, &cmd->cmnd[2]);
752 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
753
754 len = sdkp->device->sector_size;
755 break;
756
757 case SD_LBP_WS10:
758 case SD_LBP_ZERO:
759 cmd->cmd_len = 10;
760 cmd->cmnd[0] = WRITE_SAME;
761 if (sdkp->provisioning_mode == SD_LBP_WS10)
762 cmd->cmnd[1] = 0x8; /* UNMAP */
763 put_unaligned_be32(sector, &cmd->cmnd[2]);
764 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
765
766 len = sdkp->device->sector_size;
767 break;
768
769 default:
770 ret = BLKPREP_INVALID;
771 goto out;
772 }
773
774 rq->timeout = SD_TIMEOUT;
775
776 cmd->transfersize = len;
777 cmd->allowed = SD_MAX_RETRIES;
778
779 rq->special_vec.bv_page = page;
780 rq->special_vec.bv_offset = 0;
781 rq->special_vec.bv_len = len;
782
783 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
784 rq->resid_len = len;
785
786 ret = scsi_init_io(cmd);
787out:
788 if (ret != BLKPREP_OK)
789 __free_page(page);
790 return ret;
791}
792
793static void sd_config_write_same(struct scsi_disk *sdkp)
794{
795 struct request_queue *q = sdkp->disk->queue;
796 unsigned int logical_block_size = sdkp->device->sector_size;
797
798 if (sdkp->device->no_write_same) {
799 sdkp->max_ws_blocks = 0;
800 goto out;
801 }
802
803 /* Some devices can not handle block counts above 0xffff despite
804 * supporting WRITE SAME(16). Consequently we default to 64k
805 * blocks per I/O unless the device explicitly advertises a
806 * bigger limit.
807 */
808 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
809 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
810 (u32)SD_MAX_WS16_BLOCKS);
811 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
812 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
813 (u32)SD_MAX_WS10_BLOCKS);
814 else {
815 sdkp->device->no_write_same = 1;
816 sdkp->max_ws_blocks = 0;
817 }
818
819out:
820 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
821 (logical_block_size >> 9));
822}
823
824/**
825 * sd_setup_write_same_cmnd - write the same data to multiple blocks
826 * @cmd: command to prepare
827 *
828 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
829 * preference indicated by target device.
830 **/
831static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
832{
833 struct request *rq = cmd->request;
834 struct scsi_device *sdp = cmd->device;
835 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
836 struct bio *bio = rq->bio;
837 sector_t sector = blk_rq_pos(rq);
838 unsigned int nr_sectors = blk_rq_sectors(rq);
839 unsigned int nr_bytes = blk_rq_bytes(rq);
840 int ret;
841
842 if (sdkp->device->no_write_same)
843 return BLKPREP_INVALID;
844
845 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
846
847 if (sd_is_zoned(sdkp)) {
848 ret = sd_zbc_setup_write_cmnd(cmd);
849 if (ret != BLKPREP_OK)
850 return ret;
851 }
852
853 sector >>= ilog2(sdp->sector_size) - 9;
854 nr_sectors >>= ilog2(sdp->sector_size) - 9;
855
856 rq->timeout = SD_WRITE_SAME_TIMEOUT;
857
858 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
859 cmd->cmd_len = 16;
860 cmd->cmnd[0] = WRITE_SAME_16;
861 put_unaligned_be64(sector, &cmd->cmnd[2]);
862 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
863 } else {
864 cmd->cmd_len = 10;
865 cmd->cmnd[0] = WRITE_SAME;
866 put_unaligned_be32(sector, &cmd->cmnd[2]);
867 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
868 }
869
870 cmd->transfersize = sdp->sector_size;
871 cmd->allowed = SD_MAX_RETRIES;
872
873 /*
874 * For WRITE SAME the data transferred via the DATA OUT buffer is
875 * different from the amount of data actually written to the target.
876 *
877 * We set up __data_len to the amount of data transferred via the
878 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
879 * to transfer a single sector of data first, but then reset it to
880 * the amount of data to be written right after so that the I/O path
881 * knows how much to actually write.
882 */
883 rq->__data_len = sdp->sector_size;
884 ret = scsi_init_io(cmd);
885 rq->__data_len = nr_bytes;
886 return ret;
887}
888
889static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
890{
891 struct request *rq = cmd->request;
892
893 /* flush requests don't perform I/O, zero the S/G table */
894 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
895
896 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
897 cmd->cmd_len = 10;
898 cmd->transfersize = 0;
899 cmd->allowed = SD_MAX_RETRIES;
900
901 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
902 return BLKPREP_OK;
903}
904
905static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
906{
907 struct request *rq = SCpnt->request;
908 struct scsi_device *sdp = SCpnt->device;
909 struct gendisk *disk = rq->rq_disk;
910 struct scsi_disk *sdkp = scsi_disk(disk);
911 sector_t block = blk_rq_pos(rq);
912 sector_t threshold;
913 unsigned int this_count = blk_rq_sectors(rq);
914 unsigned int dif, dix;
915 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
916 int ret;
917 unsigned char protect;
918
919 if (zoned_write) {
920 ret = sd_zbc_setup_write_cmnd(SCpnt);
921 if (ret != BLKPREP_OK)
922 return ret;
923 }
924
925 ret = scsi_init_io(SCpnt);
926 if (ret != BLKPREP_OK)
927 goto out;
928 SCpnt = rq->special;
929
930 /* from here on until we're complete, any goto out
931 * is used for a killable error condition */
932 ret = BLKPREP_KILL;
933
934 SCSI_LOG_HLQUEUE(1,
935 scmd_printk(KERN_INFO, SCpnt,
936 "%s: block=%llu, count=%d\n",
937 __func__, (unsigned long long)block, this_count));
938
939 if (!sdp || !scsi_device_online(sdp) ||
940 block + blk_rq_sectors(rq) > get_capacity(disk)) {
941 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
942 "Finishing %u sectors\n",
943 blk_rq_sectors(rq)));
944 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
945 "Retry with 0x%p\n", SCpnt));
946 goto out;
947 }
948
949 if (sdp->changed) {
950 /*
951 * quietly refuse to do anything to a changed disc until
952 * the changed bit has been reset
953 */
954 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
955 goto out;
956 }
957
958 /*
959 * Some SD card readers can't handle multi-sector accesses which touch
960 * the last one or two hardware sectors. Split accesses as needed.
961 */
962 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
963 (sdp->sector_size / 512);
964
965 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
966 if (block < threshold) {
967 /* Access up to the threshold but not beyond */
968 this_count = threshold - block;
969 } else {
970 /* Access only a single hardware sector */
971 this_count = sdp->sector_size / 512;
972 }
973 }
974
975 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
976 (unsigned long long)block));
977
978 /*
979 * If we have a 1K hardware sectorsize, prevent access to single
980 * 512 byte sectors. In theory we could handle this - in fact
981 * the scsi cdrom driver must be able to handle this because
982 * we typically use 1K blocksizes, and cdroms typically have
983 * 2K hardware sectorsizes. Of course, things are simpler
984 * with the cdrom, since it is read-only. For performance
985 * reasons, the filesystems should be able to handle this
986 * and not force the scsi disk driver to use bounce buffers
987 * for this.
988 */
989 if (sdp->sector_size == 1024) {
990 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
991 scmd_printk(KERN_ERR, SCpnt,
992 "Bad block number requested\n");
993 goto out;
994 } else {
995 block = block >> 1;
996 this_count = this_count >> 1;
997 }
998 }
999 if (sdp->sector_size == 2048) {
1000 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1001 scmd_printk(KERN_ERR, SCpnt,
1002 "Bad block number requested\n");
1003 goto out;
1004 } else {
1005 block = block >> 2;
1006 this_count = this_count >> 2;
1007 }
1008 }
1009 if (sdp->sector_size == 4096) {
1010 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1011 scmd_printk(KERN_ERR, SCpnt,
1012 "Bad block number requested\n");
1013 goto out;
1014 } else {
1015 block = block >> 3;
1016 this_count = this_count >> 3;
1017 }
1018 }
1019 if (rq_data_dir(rq) == WRITE) {
1020 SCpnt->cmnd[0] = WRITE_6;
1021
1022 if (blk_integrity_rq(rq))
1023 sd_dif_prepare(SCpnt);
1024
1025 } else if (rq_data_dir(rq) == READ) {
1026 SCpnt->cmnd[0] = READ_6;
1027 } else {
1028 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1029 goto out;
1030 }
1031
1032 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1033 "%s %d/%u 512 byte blocks.\n",
1034 (rq_data_dir(rq) == WRITE) ?
1035 "writing" : "reading", this_count,
1036 blk_rq_sectors(rq)));
1037
1038 dix = scsi_prot_sg_count(SCpnt);
1039 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1040
1041 if (dif || dix)
1042 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1043 else
1044 protect = 0;
1045
1046 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1047 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1048
1049 if (unlikely(SCpnt->cmnd == NULL)) {
1050 ret = BLKPREP_DEFER;
1051 goto out;
1052 }
1053
1054 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1055 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1056 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1057 SCpnt->cmnd[7] = 0x18;
1058 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1059 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1060
1061 /* LBA */
1062 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1063 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1064 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1065 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1066 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1067 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1068 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1069 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1070
1071 /* Expected Indirect LBA */
1072 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1073 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1074 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1075 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1076
1077 /* Transfer length */
1078 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1079 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1080 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1081 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1082 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1083 SCpnt->cmnd[0] += READ_16 - READ_6;
1084 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1085 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1086 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1087 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1088 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1089 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1090 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1091 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1092 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1093 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1094 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1095 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1096 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1097 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1098 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1099 scsi_device_protection(SCpnt->device) ||
1100 SCpnt->device->use_10_for_rw) {
1101 SCpnt->cmnd[0] += READ_10 - READ_6;
1102 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1103 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1104 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1105 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1106 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1107 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1108 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1109 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1110 } else {
1111 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1112 /*
1113 * This happens only if this drive failed
1114 * 10byte rw command with ILLEGAL_REQUEST
1115 * during operation and thus turned off
1116 * use_10_for_rw.
1117 */
1118 scmd_printk(KERN_ERR, SCpnt,
1119 "FUA write on READ/WRITE(6) drive\n");
1120 goto out;
1121 }
1122
1123 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1124 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1125 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1126 SCpnt->cmnd[4] = (unsigned char) this_count;
1127 SCpnt->cmnd[5] = 0;
1128 }
1129 SCpnt->sdb.length = this_count * sdp->sector_size;
1130
1131 /*
1132 * We shouldn't disconnect in the middle of a sector, so with a dumb
1133 * host adapter, it's safe to assume that we can at least transfer
1134 * this many bytes between each connect / disconnect.
1135 */
1136 SCpnt->transfersize = sdp->sector_size;
1137 SCpnt->underflow = this_count << 9;
1138 SCpnt->allowed = SD_MAX_RETRIES;
1139
1140 /*
1141 * This indicates that the command is ready from our end to be
1142 * queued.
1143 */
1144 ret = BLKPREP_OK;
1145 out:
1146 if (zoned_write && ret != BLKPREP_OK)
1147 sd_zbc_cancel_write_cmnd(SCpnt);
1148
1149 return ret;
1150}
1151
1152static int sd_init_command(struct scsi_cmnd *cmd)
1153{
1154 struct request *rq = cmd->request;
1155
1156 switch (req_op(rq)) {
1157 case REQ_OP_DISCARD:
1158 return sd_setup_discard_cmnd(cmd);
1159 case REQ_OP_WRITE_SAME:
1160 return sd_setup_write_same_cmnd(cmd);
1161 case REQ_OP_FLUSH:
1162 return sd_setup_flush_cmnd(cmd);
1163 case REQ_OP_READ:
1164 case REQ_OP_WRITE:
1165 return sd_setup_read_write_cmnd(cmd);
1166 case REQ_OP_ZONE_REPORT:
1167 return sd_zbc_setup_report_cmnd(cmd);
1168 case REQ_OP_ZONE_RESET:
1169 return sd_zbc_setup_reset_cmnd(cmd);
1170 default:
1171 BUG();
1172 }
1173}
1174
1175static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1176{
1177 struct request *rq = SCpnt->request;
1178
1179 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1180 __free_page(rq->special_vec.bv_page);
1181
1182 if (SCpnt->cmnd != rq->cmd) {
1183 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1184 SCpnt->cmnd = NULL;
1185 SCpnt->cmd_len = 0;
1186 }
1187}
1188
1189/**
1190 * sd_open - open a scsi disk device
1191 * @inode: only i_rdev member may be used
1192 * @filp: only f_mode and f_flags may be used
1193 *
1194 * Returns 0 if successful. Returns a negated errno value in case
1195 * of error.
1196 *
1197 * Note: This can be called from a user context (e.g. fsck(1) )
1198 * or from within the kernel (e.g. as a result of a mount(1) ).
1199 * In the latter case @inode and @filp carry an abridged amount
1200 * of information as noted above.
1201 *
1202 * Locking: called with bdev->bd_mutex held.
1203 **/
1204static int sd_open(struct block_device *bdev, fmode_t mode)
1205{
1206 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1207 struct scsi_device *sdev;
1208 int retval;
1209
1210 if (!sdkp)
1211 return -ENXIO;
1212
1213 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1214
1215 sdev = sdkp->device;
1216
1217 /*
1218 * If the device is in error recovery, wait until it is done.
1219 * If the device is offline, then disallow any access to it.
1220 */
1221 retval = -ENXIO;
1222 if (!scsi_block_when_processing_errors(sdev))
1223 goto error_out;
1224
1225 if (sdev->removable || sdkp->write_prot)
1226 check_disk_change(bdev);
1227
1228 /*
1229 * If the drive is empty, just let the open fail.
1230 */
1231 retval = -ENOMEDIUM;
1232 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1233 goto error_out;
1234
1235 /*
1236 * If the device has the write protect tab set, have the open fail
1237 * if the user expects to be able to write to the thing.
1238 */
1239 retval = -EROFS;
1240 if (sdkp->write_prot && (mode & FMODE_WRITE))
1241 goto error_out;
1242
1243 /*
1244 * It is possible that the disk changing stuff resulted in
1245 * the device being taken offline. If this is the case,
1246 * report this to the user, and don't pretend that the
1247 * open actually succeeded.
1248 */
1249 retval = -ENXIO;
1250 if (!scsi_device_online(sdev))
1251 goto error_out;
1252
1253 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1254 if (scsi_block_when_processing_errors(sdev))
1255 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1256 }
1257
1258 return 0;
1259
1260error_out:
1261 scsi_disk_put(sdkp);
1262 return retval;
1263}
1264
1265/**
1266 * sd_release - invoked when the (last) close(2) is called on this
1267 * scsi disk.
1268 * @inode: only i_rdev member may be used
1269 * @filp: only f_mode and f_flags may be used
1270 *
1271 * Returns 0.
1272 *
1273 * Note: may block (uninterruptible) if error recovery is underway
1274 * on this disk.
1275 *
1276 * Locking: called with bdev->bd_mutex held.
1277 **/
1278static void sd_release(struct gendisk *disk, fmode_t mode)
1279{
1280 struct scsi_disk *sdkp = scsi_disk(disk);
1281 struct scsi_device *sdev = sdkp->device;
1282
1283 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1284
1285 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1286 if (scsi_block_when_processing_errors(sdev))
1287 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1288 }
1289
1290 /*
1291 * XXX and what if there are packets in flight and this close()
1292 * XXX is followed by a "rmmod sd_mod"?
1293 */
1294
1295 scsi_disk_put(sdkp);
1296}
1297
1298static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1299{
1300 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1301 struct scsi_device *sdp = sdkp->device;
1302 struct Scsi_Host *host = sdp->host;
1303 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1304 int diskinfo[4];
1305
1306 /* default to most commonly used values */
1307 diskinfo[0] = 0x40; /* 1 << 6 */
1308 diskinfo[1] = 0x20; /* 1 << 5 */
1309 diskinfo[2] = capacity >> 11;
1310
1311 /* override with calculated, extended default, or driver values */
1312 if (host->hostt->bios_param)
1313 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1314 else
1315 scsicam_bios_param(bdev, capacity, diskinfo);
1316
1317 geo->heads = diskinfo[0];
1318 geo->sectors = diskinfo[1];
1319 geo->cylinders = diskinfo[2];
1320 return 0;
1321}
1322
1323/**
1324 * sd_ioctl - process an ioctl
1325 * @inode: only i_rdev/i_bdev members may be used
1326 * @filp: only f_mode and f_flags may be used
1327 * @cmd: ioctl command number
1328 * @arg: this is third argument given to ioctl(2) system call.
1329 * Often contains a pointer.
1330 *
1331 * Returns 0 if successful (some ioctls return positive numbers on
1332 * success as well). Returns a negated errno value in case of error.
1333 *
1334 * Note: most ioctls are forward onto the block subsystem or further
1335 * down in the scsi subsystem.
1336 **/
1337static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1338 unsigned int cmd, unsigned long arg)
1339{
1340 struct gendisk *disk = bdev->bd_disk;
1341 struct scsi_disk *sdkp = scsi_disk(disk);
1342 struct scsi_device *sdp = sdkp->device;
1343 void __user *p = (void __user *)arg;
1344 int error;
1345
1346 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1347 "cmd=0x%x\n", disk->disk_name, cmd));
1348
1349 error = scsi_verify_blk_ioctl(bdev, cmd);
1350 if (error < 0)
1351 return error;
1352
1353 /*
1354 * If we are in the middle of error recovery, don't let anyone
1355 * else try and use this device. Also, if error recovery fails, it
1356 * may try and take the device offline, in which case all further
1357 * access to the device is prohibited.
1358 */
1359 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1360 (mode & FMODE_NDELAY) != 0);
1361 if (error)
1362 goto out;
1363
1364 /*
1365 * Send SCSI addressing ioctls directly to mid level, send other
1366 * ioctls to block level and then onto mid level if they can't be
1367 * resolved.
1368 */
1369 switch (cmd) {
1370 case SCSI_IOCTL_GET_IDLUN:
1371 case SCSI_IOCTL_GET_BUS_NUMBER:
1372 error = scsi_ioctl(sdp, cmd, p);
1373 break;
1374 default:
1375 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1376 if (error != -ENOTTY)
1377 break;
1378 error = scsi_ioctl(sdp, cmd, p);
1379 break;
1380 }
1381out:
1382 return error;
1383}
1384
1385static void set_media_not_present(struct scsi_disk *sdkp)
1386{
1387 if (sdkp->media_present)
1388 sdkp->device->changed = 1;
1389
1390 if (sdkp->device->removable) {
1391 sdkp->media_present = 0;
1392 sdkp->capacity = 0;
1393 }
1394}
1395
1396static int media_not_present(struct scsi_disk *sdkp,
1397 struct scsi_sense_hdr *sshdr)
1398{
1399 if (!scsi_sense_valid(sshdr))
1400 return 0;
1401
1402 /* not invoked for commands that could return deferred errors */
1403 switch (sshdr->sense_key) {
1404 case UNIT_ATTENTION:
1405 case NOT_READY:
1406 /* medium not present */
1407 if (sshdr->asc == 0x3A) {
1408 set_media_not_present(sdkp);
1409 return 1;
1410 }
1411 }
1412 return 0;
1413}
1414
1415/**
1416 * sd_check_events - check media events
1417 * @disk: kernel device descriptor
1418 * @clearing: disk events currently being cleared
1419 *
1420 * Returns mask of DISK_EVENT_*.
1421 *
1422 * Note: this function is invoked from the block subsystem.
1423 **/
1424static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1425{
1426 struct scsi_disk *sdkp = scsi_disk_get(disk);
1427 struct scsi_device *sdp;
1428 struct scsi_sense_hdr *sshdr = NULL;
1429 int retval;
1430
1431 if (!sdkp)
1432 return 0;
1433
1434 sdp = sdkp->device;
1435 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1436
1437 /*
1438 * If the device is offline, don't send any commands - just pretend as
1439 * if the command failed. If the device ever comes back online, we
1440 * can deal with it then. It is only because of unrecoverable errors
1441 * that we would ever take a device offline in the first place.
1442 */
1443 if (!scsi_device_online(sdp)) {
1444 set_media_not_present(sdkp);
1445 goto out;
1446 }
1447
1448 /*
1449 * Using TEST_UNIT_READY enables differentiation between drive with
1450 * no cartridge loaded - NOT READY, drive with changed cartridge -
1451 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1452 *
1453 * Drives that auto spin down. eg iomega jaz 1G, will be started
1454 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1455 * sd_revalidate() is called.
1456 */
1457 retval = -ENODEV;
1458
1459 if (scsi_block_when_processing_errors(sdp)) {
1460 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1461 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1462 sshdr);
1463 }
1464
1465 /* failed to execute TUR, assume media not present */
1466 if (host_byte(retval)) {
1467 set_media_not_present(sdkp);
1468 goto out;
1469 }
1470
1471 if (media_not_present(sdkp, sshdr))
1472 goto out;
1473
1474 /*
1475 * For removable scsi disk we have to recognise the presence
1476 * of a disk in the drive.
1477 */
1478 if (!sdkp->media_present)
1479 sdp->changed = 1;
1480 sdkp->media_present = 1;
1481out:
1482 /*
1483 * sdp->changed is set under the following conditions:
1484 *
1485 * Medium present state has changed in either direction.
1486 * Device has indicated UNIT_ATTENTION.
1487 */
1488 kfree(sshdr);
1489 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1490 sdp->changed = 0;
1491 scsi_disk_put(sdkp);
1492 return retval;
1493}
1494
1495static int sd_sync_cache(struct scsi_disk *sdkp)
1496{
1497 int retries, res;
1498 struct scsi_device *sdp = sdkp->device;
1499 const int timeout = sdp->request_queue->rq_timeout
1500 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1501 struct scsi_sense_hdr sshdr;
1502
1503 if (!scsi_device_online(sdp))
1504 return -ENODEV;
1505
1506 for (retries = 3; retries > 0; --retries) {
1507 unsigned char cmd[10] = { 0 };
1508
1509 cmd[0] = SYNCHRONIZE_CACHE;
1510 /*
1511 * Leave the rest of the command zero to indicate
1512 * flush everything.
1513 */
1514 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1515 &sshdr, timeout, SD_MAX_RETRIES,
1516 NULL, 0, RQF_PM);
1517 if (res == 0)
1518 break;
1519 }
1520
1521 if (res) {
1522 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1523
1524 if (driver_byte(res) & DRIVER_SENSE)
1525 sd_print_sense_hdr(sdkp, &sshdr);
1526 /* we need to evaluate the error return */
1527 if (scsi_sense_valid(&sshdr) &&
1528 (sshdr.asc == 0x3a || /* medium not present */
1529 sshdr.asc == 0x20)) /* invalid command */
1530 /* this is no error here */
1531 return 0;
1532
1533 switch (host_byte(res)) {
1534 /* ignore errors due to racing a disconnection */
1535 case DID_BAD_TARGET:
1536 case DID_NO_CONNECT:
1537 return 0;
1538 /* signal the upper layer it might try again */
1539 case DID_BUS_BUSY:
1540 case DID_IMM_RETRY:
1541 case DID_REQUEUE:
1542 case DID_SOFT_ERROR:
1543 return -EBUSY;
1544 default:
1545 return -EIO;
1546 }
1547 }
1548 return 0;
1549}
1550
1551static void sd_rescan(struct device *dev)
1552{
1553 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1554
1555 revalidate_disk(sdkp->disk);
1556}
1557
1558
1559#ifdef CONFIG_COMPAT
1560/*
1561 * This gets directly called from VFS. When the ioctl
1562 * is not recognized we go back to the other translation paths.
1563 */
1564static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1565 unsigned int cmd, unsigned long arg)
1566{
1567 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1568 int error;
1569
1570 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1571 (mode & FMODE_NDELAY) != 0);
1572 if (error)
1573 return error;
1574
1575 /*
1576 * Let the static ioctl translation table take care of it.
1577 */
1578 if (!sdev->host->hostt->compat_ioctl)
1579 return -ENOIOCTLCMD;
1580 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1581}
1582#endif
1583
1584static char sd_pr_type(enum pr_type type)
1585{
1586 switch (type) {
1587 case PR_WRITE_EXCLUSIVE:
1588 return 0x01;
1589 case PR_EXCLUSIVE_ACCESS:
1590 return 0x03;
1591 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1592 return 0x05;
1593 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1594 return 0x06;
1595 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1596 return 0x07;
1597 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1598 return 0x08;
1599 default:
1600 return 0;
1601 }
1602};
1603
1604static int sd_pr_command(struct block_device *bdev, u8 sa,
1605 u64 key, u64 sa_key, u8 type, u8 flags)
1606{
1607 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1608 struct scsi_sense_hdr sshdr;
1609 int result;
1610 u8 cmd[16] = { 0, };
1611 u8 data[24] = { 0, };
1612
1613 cmd[0] = PERSISTENT_RESERVE_OUT;
1614 cmd[1] = sa;
1615 cmd[2] = type;
1616 put_unaligned_be32(sizeof(data), &cmd[5]);
1617
1618 put_unaligned_be64(key, &data[0]);
1619 put_unaligned_be64(sa_key, &data[8]);
1620 data[20] = flags;
1621
1622 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1623 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1624
1625 if ((driver_byte(result) & DRIVER_SENSE) &&
1626 (scsi_sense_valid(&sshdr))) {
1627 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1628 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1629 }
1630
1631 return result;
1632}
1633
1634static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1635 u32 flags)
1636{
1637 if (flags & ~PR_FL_IGNORE_KEY)
1638 return -EOPNOTSUPP;
1639 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1640 old_key, new_key, 0,
1641 (1 << 0) /* APTPL */);
1642}
1643
1644static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1645 u32 flags)
1646{
1647 if (flags)
1648 return -EOPNOTSUPP;
1649 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1650}
1651
1652static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1653{
1654 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1655}
1656
1657static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1658 enum pr_type type, bool abort)
1659{
1660 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1661 sd_pr_type(type), 0);
1662}
1663
1664static int sd_pr_clear(struct block_device *bdev, u64 key)
1665{
1666 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1667}
1668
1669static const struct pr_ops sd_pr_ops = {
1670 .pr_register = sd_pr_register,
1671 .pr_reserve = sd_pr_reserve,
1672 .pr_release = sd_pr_release,
1673 .pr_preempt = sd_pr_preempt,
1674 .pr_clear = sd_pr_clear,
1675};
1676
1677static const struct block_device_operations sd_fops = {
1678 .owner = THIS_MODULE,
1679 .open = sd_open,
1680 .release = sd_release,
1681 .ioctl = sd_ioctl,
1682 .getgeo = sd_getgeo,
1683#ifdef CONFIG_COMPAT
1684 .compat_ioctl = sd_compat_ioctl,
1685#endif
1686 .check_events = sd_check_events,
1687 .revalidate_disk = sd_revalidate_disk,
1688 .unlock_native_capacity = sd_unlock_native_capacity,
1689 .pr_ops = &sd_pr_ops,
1690};
1691
1692/**
1693 * sd_eh_action - error handling callback
1694 * @scmd: sd-issued command that has failed
1695 * @eh_disp: The recovery disposition suggested by the midlayer
1696 *
1697 * This function is called by the SCSI midlayer upon completion of an
1698 * error test command (currently TEST UNIT READY). The result of sending
1699 * the eh command is passed in eh_disp. We're looking for devices that
1700 * fail medium access commands but are OK with non access commands like
1701 * test unit ready (so wrongly see the device as having a successful
1702 * recovery)
1703 **/
1704static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1705{
1706 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1707
1708 if (!scsi_device_online(scmd->device) ||
1709 !scsi_medium_access_command(scmd) ||
1710 host_byte(scmd->result) != DID_TIME_OUT ||
1711 eh_disp != SUCCESS)
1712 return eh_disp;
1713
1714 /*
1715 * The device has timed out executing a medium access command.
1716 * However, the TEST UNIT READY command sent during error
1717 * handling completed successfully. Either the device is in the
1718 * process of recovering or has it suffered an internal failure
1719 * that prevents access to the storage medium.
1720 */
1721 sdkp->medium_access_timed_out++;
1722
1723 /*
1724 * If the device keeps failing read/write commands but TEST UNIT
1725 * READY always completes successfully we assume that medium
1726 * access is no longer possible and take the device offline.
1727 */
1728 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1729 scmd_printk(KERN_ERR, scmd,
1730 "Medium access timeout failure. Offlining disk!\n");
1731 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1732
1733 return FAILED;
1734 }
1735
1736 return eh_disp;
1737}
1738
1739static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1740{
1741 u64 start_lba = blk_rq_pos(scmd->request);
1742 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1743 u64 factor = scmd->device->sector_size / 512;
1744 u64 bad_lba;
1745 int info_valid;
1746 /*
1747 * resid is optional but mostly filled in. When it's unused,
1748 * its value is zero, so we assume the whole buffer transferred
1749 */
1750 unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1751 unsigned int good_bytes;
1752
1753 if (scmd->request->cmd_type != REQ_TYPE_FS)
1754 return 0;
1755
1756 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1757 SCSI_SENSE_BUFFERSIZE,
1758 &bad_lba);
1759 if (!info_valid)
1760 return 0;
1761
1762 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1763 return 0;
1764
1765 /* be careful ... don't want any overflows */
1766 do_div(start_lba, factor);
1767 do_div(end_lba, factor);
1768
1769 /* The bad lba was reported incorrectly, we have no idea where
1770 * the error is.
1771 */
1772 if (bad_lba < start_lba || bad_lba >= end_lba)
1773 return 0;
1774
1775 /* This computation should always be done in terms of
1776 * the resolution of the device's medium.
1777 */
1778 good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1779 return min(good_bytes, transferred);
1780}
1781
1782/**
1783 * sd_done - bottom half handler: called when the lower level
1784 * driver has completed (successfully or otherwise) a scsi command.
1785 * @SCpnt: mid-level's per command structure.
1786 *
1787 * Note: potentially run from within an ISR. Must not block.
1788 **/
1789static int sd_done(struct scsi_cmnd *SCpnt)
1790{
1791 int result = SCpnt->result;
1792 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1793 unsigned int sector_size = SCpnt->device->sector_size;
1794 unsigned int resid;
1795 struct scsi_sense_hdr sshdr;
1796 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1797 struct request *req = SCpnt->request;
1798 int sense_valid = 0;
1799 int sense_deferred = 0;
1800 unsigned char op = SCpnt->cmnd[0];
1801 unsigned char unmap = SCpnt->cmnd[1] & 8;
1802
1803 switch (req_op(req)) {
1804 case REQ_OP_DISCARD:
1805 case REQ_OP_WRITE_SAME:
1806 case REQ_OP_ZONE_RESET:
1807 if (!result) {
1808 good_bytes = blk_rq_bytes(req);
1809 scsi_set_resid(SCpnt, 0);
1810 } else {
1811 good_bytes = 0;
1812 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1813 }
1814 break;
1815 case REQ_OP_ZONE_REPORT:
1816 if (!result) {
1817 good_bytes = scsi_bufflen(SCpnt)
1818 - scsi_get_resid(SCpnt);
1819 scsi_set_resid(SCpnt, 0);
1820 } else {
1821 good_bytes = 0;
1822 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1823 }
1824 break;
1825 default:
1826 /*
1827 * In case of bogus fw or device, we could end up having
1828 * an unaligned partial completion. Check this here and force
1829 * alignment.
1830 */
1831 resid = scsi_get_resid(SCpnt);
1832 if (resid & (sector_size - 1)) {
1833 sd_printk(KERN_INFO, sdkp,
1834 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1835 resid, sector_size);
1836 resid = min(scsi_bufflen(SCpnt),
1837 round_up(resid, sector_size));
1838 scsi_set_resid(SCpnt, resid);
1839 }
1840 }
1841
1842 if (result) {
1843 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1844 if (sense_valid)
1845 sense_deferred = scsi_sense_is_deferred(&sshdr);
1846 }
1847 sdkp->medium_access_timed_out = 0;
1848
1849 if (driver_byte(result) != DRIVER_SENSE &&
1850 (!sense_valid || sense_deferred))
1851 goto out;
1852
1853 switch (sshdr.sense_key) {
1854 case HARDWARE_ERROR:
1855 case MEDIUM_ERROR:
1856 good_bytes = sd_completed_bytes(SCpnt);
1857 break;
1858 case RECOVERED_ERROR:
1859 good_bytes = scsi_bufflen(SCpnt);
1860 break;
1861 case NO_SENSE:
1862 /* This indicates a false check condition, so ignore it. An
1863 * unknown amount of data was transferred so treat it as an
1864 * error.
1865 */
1866 SCpnt->result = 0;
1867 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1868 break;
1869 case ABORTED_COMMAND:
1870 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1871 good_bytes = sd_completed_bytes(SCpnt);
1872 break;
1873 case ILLEGAL_REQUEST:
1874 if (sshdr.asc == 0x10) /* DIX: Host detected corruption */
1875 good_bytes = sd_completed_bytes(SCpnt);
1876 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1877 if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1878 switch (op) {
1879 case UNMAP:
1880 sd_config_discard(sdkp, SD_LBP_DISABLE);
1881 break;
1882 case WRITE_SAME_16:
1883 case WRITE_SAME:
1884 if (unmap)
1885 sd_config_discard(sdkp, SD_LBP_DISABLE);
1886 else {
1887 sdkp->device->no_write_same = 1;
1888 sd_config_write_same(sdkp);
1889
1890 good_bytes = 0;
1891 req->__data_len = blk_rq_bytes(req);
1892 req->rq_flags |= RQF_QUIET;
1893 }
1894 }
1895 }
1896 break;
1897 default:
1898 break;
1899 }
1900
1901 out:
1902 if (sd_is_zoned(sdkp))
1903 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
1904
1905 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1906 "sd_done: completed %d of %d bytes\n",
1907 good_bytes, scsi_bufflen(SCpnt)));
1908
1909 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1910 sd_dif_complete(SCpnt, good_bytes);
1911
1912 return good_bytes;
1913}
1914
1915/*
1916 * spinup disk - called only in sd_revalidate_disk()
1917 */
1918static void
1919sd_spinup_disk(struct scsi_disk *sdkp)
1920{
1921 unsigned char cmd[10];
1922 unsigned long spintime_expire = 0;
1923 int retries, spintime;
1924 unsigned int the_result;
1925 struct scsi_sense_hdr sshdr;
1926 int sense_valid = 0;
1927
1928 spintime = 0;
1929
1930 /* Spin up drives, as required. Only do this at boot time */
1931 /* Spinup needs to be done for module loads too. */
1932 do {
1933 retries = 0;
1934
1935 do {
1936 cmd[0] = TEST_UNIT_READY;
1937 memset((void *) &cmd[1], 0, 9);
1938
1939 the_result = scsi_execute_req(sdkp->device, cmd,
1940 DMA_NONE, NULL, 0,
1941 &sshdr, SD_TIMEOUT,
1942 SD_MAX_RETRIES, NULL);
1943
1944 /*
1945 * If the drive has indicated to us that it
1946 * doesn't have any media in it, don't bother
1947 * with any more polling.
1948 */
1949 if (media_not_present(sdkp, &sshdr))
1950 return;
1951
1952 if (the_result)
1953 sense_valid = scsi_sense_valid(&sshdr);
1954 retries++;
1955 } while (retries < 3 &&
1956 (!scsi_status_is_good(the_result) ||
1957 ((driver_byte(the_result) & DRIVER_SENSE) &&
1958 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1959
1960 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1961 /* no sense, TUR either succeeded or failed
1962 * with a status error */
1963 if(!spintime && !scsi_status_is_good(the_result)) {
1964 sd_print_result(sdkp, "Test Unit Ready failed",
1965 the_result);
1966 }
1967 break;
1968 }
1969
1970 /*
1971 * The device does not want the automatic start to be issued.
1972 */
1973 if (sdkp->device->no_start_on_add)
1974 break;
1975
1976 if (sense_valid && sshdr.sense_key == NOT_READY) {
1977 if (sshdr.asc == 4 && sshdr.ascq == 3)
1978 break; /* manual intervention required */
1979 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1980 break; /* standby */
1981 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1982 break; /* unavailable */
1983 /*
1984 * Issue command to spin up drive when not ready
1985 */
1986 if (!spintime) {
1987 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1988 cmd[0] = START_STOP;
1989 cmd[1] = 1; /* Return immediately */
1990 memset((void *) &cmd[2], 0, 8);
1991 cmd[4] = 1; /* Start spin cycle */
1992 if (sdkp->device->start_stop_pwr_cond)
1993 cmd[4] |= 1 << 4;
1994 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1995 NULL, 0, &sshdr,
1996 SD_TIMEOUT, SD_MAX_RETRIES,
1997 NULL);
1998 spintime_expire = jiffies + 100 * HZ;
1999 spintime = 1;
2000 }
2001 /* Wait 1 second for next try */
2002 msleep(1000);
2003 printk(".");
2004
2005 /*
2006 * Wait for USB flash devices with slow firmware.
2007 * Yes, this sense key/ASC combination shouldn't
2008 * occur here. It's characteristic of these devices.
2009 */
2010 } else if (sense_valid &&
2011 sshdr.sense_key == UNIT_ATTENTION &&
2012 sshdr.asc == 0x28) {
2013 if (!spintime) {
2014 spintime_expire = jiffies + 5 * HZ;
2015 spintime = 1;
2016 }
2017 /* Wait 1 second for next try */
2018 msleep(1000);
2019 } else {
2020 /* we don't understand the sense code, so it's
2021 * probably pointless to loop */
2022 if(!spintime) {
2023 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2024 sd_print_sense_hdr(sdkp, &sshdr);
2025 }
2026 break;
2027 }
2028
2029 } while (spintime && time_before_eq(jiffies, spintime_expire));
2030
2031 if (spintime) {
2032 if (scsi_status_is_good(the_result))
2033 printk("ready\n");
2034 else
2035 printk("not responding...\n");
2036 }
2037}
2038
2039/*
2040 * Determine whether disk supports Data Integrity Field.
2041 */
2042static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2043{
2044 struct scsi_device *sdp = sdkp->device;
2045 u8 type;
2046 int ret = 0;
2047
2048 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2049 return ret;
2050
2051 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2052
2053 if (type > T10_PI_TYPE3_PROTECTION)
2054 ret = -ENODEV;
2055 else if (scsi_host_dif_capable(sdp->host, type))
2056 ret = 1;
2057
2058 if (sdkp->first_scan || type != sdkp->protection_type)
2059 switch (ret) {
2060 case -ENODEV:
2061 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2062 " protection type %u. Disabling disk!\n",
2063 type);
2064 break;
2065 case 1:
2066 sd_printk(KERN_NOTICE, sdkp,
2067 "Enabling DIF Type %u protection\n", type);
2068 break;
2069 case 0:
2070 sd_printk(KERN_NOTICE, sdkp,
2071 "Disabling DIF Type %u protection\n", type);
2072 break;
2073 }
2074
2075 sdkp->protection_type = type;
2076
2077 return ret;
2078}
2079
2080static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2081 struct scsi_sense_hdr *sshdr, int sense_valid,
2082 int the_result)
2083{
2084 if (driver_byte(the_result) & DRIVER_SENSE)
2085 sd_print_sense_hdr(sdkp, sshdr);
2086 else
2087 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2088
2089 /*
2090 * Set dirty bit for removable devices if not ready -
2091 * sometimes drives will not report this properly.
2092 */
2093 if (sdp->removable &&
2094 sense_valid && sshdr->sense_key == NOT_READY)
2095 set_media_not_present(sdkp);
2096
2097 /*
2098 * We used to set media_present to 0 here to indicate no media
2099 * in the drive, but some drives fail read capacity even with
2100 * media present, so we can't do that.
2101 */
2102 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2103}
2104
2105#define RC16_LEN 32
2106#if RC16_LEN > SD_BUF_SIZE
2107#error RC16_LEN must not be more than SD_BUF_SIZE
2108#endif
2109
2110#define READ_CAPACITY_RETRIES_ON_RESET 10
2111
2112static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2113 unsigned char *buffer)
2114{
2115 unsigned char cmd[16];
2116 struct scsi_sense_hdr sshdr;
2117 int sense_valid = 0;
2118 int the_result;
2119 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2120 unsigned int alignment;
2121 unsigned long long lba;
2122 unsigned sector_size;
2123
2124 if (sdp->no_read_capacity_16)
2125 return -EINVAL;
2126
2127 do {
2128 memset(cmd, 0, 16);
2129 cmd[0] = SERVICE_ACTION_IN_16;
2130 cmd[1] = SAI_READ_CAPACITY_16;
2131 cmd[13] = RC16_LEN;
2132 memset(buffer, 0, RC16_LEN);
2133
2134 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2135 buffer, RC16_LEN, &sshdr,
2136 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2137
2138 if (media_not_present(sdkp, &sshdr))
2139 return -ENODEV;
2140
2141 if (the_result) {
2142 sense_valid = scsi_sense_valid(&sshdr);
2143 if (sense_valid &&
2144 sshdr.sense_key == ILLEGAL_REQUEST &&
2145 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2146 sshdr.ascq == 0x00)
2147 /* Invalid Command Operation Code or
2148 * Invalid Field in CDB, just retry
2149 * silently with RC10 */
2150 return -EINVAL;
2151 if (sense_valid &&
2152 sshdr.sense_key == UNIT_ATTENTION &&
2153 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2154 /* Device reset might occur several times,
2155 * give it one more chance */
2156 if (--reset_retries > 0)
2157 continue;
2158 }
2159 retries--;
2160
2161 } while (the_result && retries);
2162
2163 if (the_result) {
2164 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2165 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2166 return -EINVAL;
2167 }
2168
2169 sector_size = get_unaligned_be32(&buffer[8]);
2170 lba = get_unaligned_be64(&buffer[0]);
2171
2172 if (sd_read_protection_type(sdkp, buffer) < 0) {
2173 sdkp->capacity = 0;
2174 return -ENODEV;
2175 }
2176
2177 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2178 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2179 "kernel compiled with support for large block "
2180 "devices.\n");
2181 sdkp->capacity = 0;
2182 return -EOVERFLOW;
2183 }
2184
2185 /* Logical blocks per physical block exponent */
2186 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2187
2188 /* RC basis */
2189 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2190
2191 /* Lowest aligned logical block */
2192 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2193 blk_queue_alignment_offset(sdp->request_queue, alignment);
2194 if (alignment && sdkp->first_scan)
2195 sd_printk(KERN_NOTICE, sdkp,
2196 "physical block alignment offset: %u\n", alignment);
2197
2198 if (buffer[14] & 0x80) { /* LBPME */
2199 sdkp->lbpme = 1;
2200
2201 if (buffer[14] & 0x40) /* LBPRZ */
2202 sdkp->lbprz = 1;
2203
2204 sd_config_discard(sdkp, SD_LBP_WS16);
2205 }
2206
2207 sdkp->capacity = lba + 1;
2208 return sector_size;
2209}
2210
2211static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2212 unsigned char *buffer)
2213{
2214 unsigned char cmd[16];
2215 struct scsi_sense_hdr sshdr;
2216 int sense_valid = 0;
2217 int the_result;
2218 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2219 sector_t lba;
2220 unsigned sector_size;
2221
2222 do {
2223 cmd[0] = READ_CAPACITY;
2224 memset(&cmd[1], 0, 9);
2225 memset(buffer, 0, 8);
2226
2227 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2228 buffer, 8, &sshdr,
2229 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2230
2231 if (media_not_present(sdkp, &sshdr))
2232 return -ENODEV;
2233
2234 if (the_result) {
2235 sense_valid = scsi_sense_valid(&sshdr);
2236 if (sense_valid &&
2237 sshdr.sense_key == UNIT_ATTENTION &&
2238 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2239 /* Device reset might occur several times,
2240 * give it one more chance */
2241 if (--reset_retries > 0)
2242 continue;
2243 }
2244 retries--;
2245
2246 } while (the_result && retries);
2247
2248 if (the_result) {
2249 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2250 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2251 return -EINVAL;
2252 }
2253
2254 sector_size = get_unaligned_be32(&buffer[4]);
2255 lba = get_unaligned_be32(&buffer[0]);
2256
2257 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2258 /* Some buggy (usb cardreader) devices return an lba of
2259 0xffffffff when the want to report a size of 0 (with
2260 which they really mean no media is present) */
2261 sdkp->capacity = 0;
2262 sdkp->physical_block_size = sector_size;
2263 return sector_size;
2264 }
2265
2266 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2267 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2268 "kernel compiled with support for large block "
2269 "devices.\n");
2270 sdkp->capacity = 0;
2271 return -EOVERFLOW;
2272 }
2273
2274 sdkp->capacity = lba + 1;
2275 sdkp->physical_block_size = sector_size;
2276 return sector_size;
2277}
2278
2279static int sd_try_rc16_first(struct scsi_device *sdp)
2280{
2281 if (sdp->host->max_cmd_len < 16)
2282 return 0;
2283 if (sdp->try_rc_10_first)
2284 return 0;
2285 if (sdp->scsi_level > SCSI_SPC_2)
2286 return 1;
2287 if (scsi_device_protection(sdp))
2288 return 1;
2289 return 0;
2290}
2291
2292/*
2293 * read disk capacity
2294 */
2295static void
2296sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2297{
2298 int sector_size;
2299 struct scsi_device *sdp = sdkp->device;
2300
2301 if (sd_try_rc16_first(sdp)) {
2302 sector_size = read_capacity_16(sdkp, sdp, buffer);
2303 if (sector_size == -EOVERFLOW)
2304 goto got_data;
2305 if (sector_size == -ENODEV)
2306 return;
2307 if (sector_size < 0)
2308 sector_size = read_capacity_10(sdkp, sdp, buffer);
2309 if (sector_size < 0)
2310 return;
2311 } else {
2312 sector_size = read_capacity_10(sdkp, sdp, buffer);
2313 if (sector_size == -EOVERFLOW)
2314 goto got_data;
2315 if (sector_size < 0)
2316 return;
2317 if ((sizeof(sdkp->capacity) > 4) &&
2318 (sdkp->capacity > 0xffffffffULL)) {
2319 int old_sector_size = sector_size;
2320 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2321 "Trying to use READ CAPACITY(16).\n");
2322 sector_size = read_capacity_16(sdkp, sdp, buffer);
2323 if (sector_size < 0) {
2324 sd_printk(KERN_NOTICE, sdkp,
2325 "Using 0xffffffff as device size\n");
2326 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2327 sector_size = old_sector_size;
2328 goto got_data;
2329 }
2330 }
2331 }
2332
2333 /* Some devices are known to return the total number of blocks,
2334 * not the highest block number. Some devices have versions
2335 * which do this and others which do not. Some devices we might
2336 * suspect of doing this but we don't know for certain.
2337 *
2338 * If we know the reported capacity is wrong, decrement it. If
2339 * we can only guess, then assume the number of blocks is even
2340 * (usually true but not always) and err on the side of lowering
2341 * the capacity.
2342 */
2343 if (sdp->fix_capacity ||
2344 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2345 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2346 "from its reported value: %llu\n",
2347 (unsigned long long) sdkp->capacity);
2348 --sdkp->capacity;
2349 }
2350
2351got_data:
2352 if (sector_size == 0) {
2353 sector_size = 512;
2354 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2355 "assuming 512.\n");
2356 }
2357
2358 if (sector_size != 512 &&
2359 sector_size != 1024 &&
2360 sector_size != 2048 &&
2361 sector_size != 4096) {
2362 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2363 sector_size);
2364 /*
2365 * The user might want to re-format the drive with
2366 * a supported sectorsize. Once this happens, it
2367 * would be relatively trivial to set the thing up.
2368 * For this reason, we leave the thing in the table.
2369 */
2370 sdkp->capacity = 0;
2371 /*
2372 * set a bogus sector size so the normal read/write
2373 * logic in the block layer will eventually refuse any
2374 * request on this device without tripping over power
2375 * of two sector size assumptions
2376 */
2377 sector_size = 512;
2378 }
2379 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2380 blk_queue_physical_block_size(sdp->request_queue,
2381 sdkp->physical_block_size);
2382 sdkp->device->sector_size = sector_size;
2383
2384 if (sdkp->capacity > 0xffffffff)
2385 sdp->use_16_for_rw = 1;
2386
2387}
2388
2389/*
2390 * Print disk capacity
2391 */
2392static void
2393sd_print_capacity(struct scsi_disk *sdkp,
2394 sector_t old_capacity)
2395{
2396 int sector_size = sdkp->device->sector_size;
2397 char cap_str_2[10], cap_str_10[10];
2398
2399 string_get_size(sdkp->capacity, sector_size,
2400 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2401 string_get_size(sdkp->capacity, sector_size,
2402 STRING_UNITS_10, cap_str_10,
2403 sizeof(cap_str_10));
2404
2405 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2406 sd_printk(KERN_NOTICE, sdkp,
2407 "%llu %d-byte logical blocks: (%s/%s)\n",
2408 (unsigned long long)sdkp->capacity,
2409 sector_size, cap_str_10, cap_str_2);
2410
2411 if (sdkp->physical_block_size != sector_size)
2412 sd_printk(KERN_NOTICE, sdkp,
2413 "%u-byte physical blocks\n",
2414 sdkp->physical_block_size);
2415
2416 sd_zbc_print_zones(sdkp);
2417 }
2418}
2419
2420/* called with buffer of length 512 */
2421static inline int
2422sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2423 unsigned char *buffer, int len, struct scsi_mode_data *data,
2424 struct scsi_sense_hdr *sshdr)
2425{
2426 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2427 SD_TIMEOUT, SD_MAX_RETRIES, data,
2428 sshdr);
2429}
2430
2431/*
2432 * read write protect setting, if possible - called only in sd_revalidate_disk()
2433 * called with buffer of length SD_BUF_SIZE
2434 */
2435static void
2436sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2437{
2438 int res;
2439 struct scsi_device *sdp = sdkp->device;
2440 struct scsi_mode_data data;
2441 int old_wp = sdkp->write_prot;
2442
2443 set_disk_ro(sdkp->disk, 0);
2444 if (sdp->skip_ms_page_3f) {
2445 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2446 return;
2447 }
2448
2449 if (sdp->use_192_bytes_for_3f) {
2450 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2451 } else {
2452 /*
2453 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2454 * We have to start carefully: some devices hang if we ask
2455 * for more than is available.
2456 */
2457 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2458
2459 /*
2460 * Second attempt: ask for page 0 When only page 0 is
2461 * implemented, a request for page 3F may return Sense Key
2462 * 5: Illegal Request, Sense Code 24: Invalid field in
2463 * CDB.
2464 */
2465 if (!scsi_status_is_good(res))
2466 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2467
2468 /*
2469 * Third attempt: ask 255 bytes, as we did earlier.
2470 */
2471 if (!scsi_status_is_good(res))
2472 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2473 &data, NULL);
2474 }
2475
2476 if (!scsi_status_is_good(res)) {
2477 sd_first_printk(KERN_WARNING, sdkp,
2478 "Test WP failed, assume Write Enabled\n");
2479 } else {
2480 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2481 set_disk_ro(sdkp->disk, sdkp->write_prot);
2482 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2483 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2484 sdkp->write_prot ? "on" : "off");
2485 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2486 }
2487 }
2488}
2489
2490/*
2491 * sd_read_cache_type - called only from sd_revalidate_disk()
2492 * called with buffer of length SD_BUF_SIZE
2493 */
2494static void
2495sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2496{
2497 int len = 0, res;
2498 struct scsi_device *sdp = sdkp->device;
2499
2500 int dbd;
2501 int modepage;
2502 int first_len;
2503 struct scsi_mode_data data;
2504 struct scsi_sense_hdr sshdr;
2505 int old_wce = sdkp->WCE;
2506 int old_rcd = sdkp->RCD;
2507 int old_dpofua = sdkp->DPOFUA;
2508
2509
2510 if (sdkp->cache_override)
2511 return;
2512
2513 first_len = 4;
2514 if (sdp->skip_ms_page_8) {
2515 if (sdp->type == TYPE_RBC)
2516 goto defaults;
2517 else {
2518 if (sdp->skip_ms_page_3f)
2519 goto defaults;
2520 modepage = 0x3F;
2521 if (sdp->use_192_bytes_for_3f)
2522 first_len = 192;
2523 dbd = 0;
2524 }
2525 } else if (sdp->type == TYPE_RBC) {
2526 modepage = 6;
2527 dbd = 8;
2528 } else {
2529 modepage = 8;
2530 dbd = 0;
2531 }
2532
2533 /* cautiously ask */
2534 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2535 &data, &sshdr);
2536
2537 if (!scsi_status_is_good(res))
2538 goto bad_sense;
2539
2540 if (!data.header_length) {
2541 modepage = 6;
2542 first_len = 0;
2543 sd_first_printk(KERN_ERR, sdkp,
2544 "Missing header in MODE_SENSE response\n");
2545 }
2546
2547 /* that went OK, now ask for the proper length */
2548 len = data.length;
2549
2550 /*
2551 * We're only interested in the first three bytes, actually.
2552 * But the data cache page is defined for the first 20.
2553 */
2554 if (len < 3)
2555 goto bad_sense;
2556 else if (len > SD_BUF_SIZE) {
2557 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2558 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2559 len = SD_BUF_SIZE;
2560 }
2561 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2562 len = 192;
2563
2564 /* Get the data */
2565 if (len > first_len)
2566 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2567 &data, &sshdr);
2568
2569 if (scsi_status_is_good(res)) {
2570 int offset = data.header_length + data.block_descriptor_length;
2571
2572 while (offset < len) {
2573 u8 page_code = buffer[offset] & 0x3F;
2574 u8 spf = buffer[offset] & 0x40;
2575
2576 if (page_code == 8 || page_code == 6) {
2577 /* We're interested only in the first 3 bytes.
2578 */
2579 if (len - offset <= 2) {
2580 sd_first_printk(KERN_ERR, sdkp,
2581 "Incomplete mode parameter "
2582 "data\n");
2583 goto defaults;
2584 } else {
2585 modepage = page_code;
2586 goto Page_found;
2587 }
2588 } else {
2589 /* Go to the next page */
2590 if (spf && len - offset > 3)
2591 offset += 4 + (buffer[offset+2] << 8) +
2592 buffer[offset+3];
2593 else if (!spf && len - offset > 1)
2594 offset += 2 + buffer[offset+1];
2595 else {
2596 sd_first_printk(KERN_ERR, sdkp,
2597 "Incomplete mode "
2598 "parameter data\n");
2599 goto defaults;
2600 }
2601 }
2602 }
2603
2604 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2605 goto defaults;
2606
2607 Page_found:
2608 if (modepage == 8) {
2609 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2610 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2611 } else {
2612 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2613 sdkp->RCD = 0;
2614 }
2615
2616 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2617 if (sdp->broken_fua) {
2618 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2619 sdkp->DPOFUA = 0;
2620 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2621 !sdkp->device->use_16_for_rw) {
2622 sd_first_printk(KERN_NOTICE, sdkp,
2623 "Uses READ/WRITE(6), disabling FUA\n");
2624 sdkp->DPOFUA = 0;
2625 }
2626
2627 /* No cache flush allowed for write protected devices */
2628 if (sdkp->WCE && sdkp->write_prot)
2629 sdkp->WCE = 0;
2630
2631 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2632 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2633 sd_printk(KERN_NOTICE, sdkp,
2634 "Write cache: %s, read cache: %s, %s\n",
2635 sdkp->WCE ? "enabled" : "disabled",
2636 sdkp->RCD ? "disabled" : "enabled",
2637 sdkp->DPOFUA ? "supports DPO and FUA"
2638 : "doesn't support DPO or FUA");
2639
2640 return;
2641 }
2642
2643bad_sense:
2644 if (scsi_sense_valid(&sshdr) &&
2645 sshdr.sense_key == ILLEGAL_REQUEST &&
2646 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2647 /* Invalid field in CDB */
2648 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2649 else
2650 sd_first_printk(KERN_ERR, sdkp,
2651 "Asking for cache data failed\n");
2652
2653defaults:
2654 if (sdp->wce_default_on) {
2655 sd_first_printk(KERN_NOTICE, sdkp,
2656 "Assuming drive cache: write back\n");
2657 sdkp->WCE = 1;
2658 } else {
2659 sd_first_printk(KERN_ERR, sdkp,
2660 "Assuming drive cache: write through\n");
2661 sdkp->WCE = 0;
2662 }
2663 sdkp->RCD = 0;
2664 sdkp->DPOFUA = 0;
2665}
2666
2667/*
2668 * The ATO bit indicates whether the DIF application tag is available
2669 * for use by the operating system.
2670 */
2671static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2672{
2673 int res, offset;
2674 struct scsi_device *sdp = sdkp->device;
2675 struct scsi_mode_data data;
2676 struct scsi_sense_hdr sshdr;
2677
2678 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2679 return;
2680
2681 if (sdkp->protection_type == 0)
2682 return;
2683
2684 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2685 SD_MAX_RETRIES, &data, &sshdr);
2686
2687 if (!scsi_status_is_good(res) || !data.header_length ||
2688 data.length < 6) {
2689 sd_first_printk(KERN_WARNING, sdkp,
2690 "getting Control mode page failed, assume no ATO\n");
2691
2692 if (scsi_sense_valid(&sshdr))
2693 sd_print_sense_hdr(sdkp, &sshdr);
2694
2695 return;
2696 }
2697
2698 offset = data.header_length + data.block_descriptor_length;
2699
2700 if ((buffer[offset] & 0x3f) != 0x0a) {
2701 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2702 return;
2703 }
2704
2705 if ((buffer[offset + 5] & 0x80) == 0)
2706 return;
2707
2708 sdkp->ATO = 1;
2709
2710 return;
2711}
2712
2713/**
2714 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2715 * @disk: disk to query
2716 */
2717static void sd_read_block_limits(struct scsi_disk *sdkp)
2718{
2719 unsigned int sector_sz = sdkp->device->sector_size;
2720 const int vpd_len = 64;
2721 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2722
2723 if (!buffer ||
2724 /* Block Limits VPD */
2725 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2726 goto out;
2727
2728 blk_queue_io_min(sdkp->disk->queue,
2729 get_unaligned_be16(&buffer[6]) * sector_sz);
2730
2731 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2732 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2733
2734 if (buffer[3] == 0x3c) {
2735 unsigned int lba_count, desc_count;
2736
2737 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2738
2739 if (!sdkp->lbpme)
2740 goto out;
2741
2742 lba_count = get_unaligned_be32(&buffer[20]);
2743 desc_count = get_unaligned_be32(&buffer[24]);
2744
2745 if (lba_count && desc_count)
2746 sdkp->max_unmap_blocks = lba_count;
2747
2748 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2749
2750 if (buffer[32] & 0x80)
2751 sdkp->unmap_alignment =
2752 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2753
2754 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2755
2756 if (sdkp->max_unmap_blocks)
2757 sd_config_discard(sdkp, SD_LBP_UNMAP);
2758 else
2759 sd_config_discard(sdkp, SD_LBP_WS16);
2760
2761 } else { /* LBP VPD page tells us what to use */
2762 if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2763 sd_config_discard(sdkp, SD_LBP_UNMAP);
2764 else if (sdkp->lbpws)
2765 sd_config_discard(sdkp, SD_LBP_WS16);
2766 else if (sdkp->lbpws10)
2767 sd_config_discard(sdkp, SD_LBP_WS10);
2768 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2769 sd_config_discard(sdkp, SD_LBP_UNMAP);
2770 else
2771 sd_config_discard(sdkp, SD_LBP_DISABLE);
2772 }
2773 }
2774
2775 out:
2776 kfree(buffer);
2777}
2778
2779/**
2780 * sd_read_block_characteristics - Query block dev. characteristics
2781 * @disk: disk to query
2782 */
2783static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2784{
2785 struct request_queue *q = sdkp->disk->queue;
2786 unsigned char *buffer;
2787 u16 rot;
2788 const int vpd_len = 64;
2789
2790 buffer = kmalloc(vpd_len, GFP_KERNEL);
2791
2792 if (!buffer ||
2793 /* Block Device Characteristics VPD */
2794 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2795 goto out;
2796
2797 rot = get_unaligned_be16(&buffer[4]);
2798
2799 if (rot == 1) {
2800 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2801 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2802 }
2803
2804 if (sdkp->device->type == TYPE_ZBC) {
2805 /* Host-managed */
2806 q->limits.zoned = BLK_ZONED_HM;
2807 } else {
2808 sdkp->zoned = (buffer[8] >> 4) & 3;
2809 if (sdkp->zoned == 1)
2810 /* Host-aware */
2811 q->limits.zoned = BLK_ZONED_HA;
2812 else
2813 /*
2814 * Treat drive-managed devices as
2815 * regular block devices.
2816 */
2817 q->limits.zoned = BLK_ZONED_NONE;
2818 }
2819 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2820 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2821 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2822
2823 out:
2824 kfree(buffer);
2825}
2826
2827/**
2828 * sd_read_block_provisioning - Query provisioning VPD page
2829 * @disk: disk to query
2830 */
2831static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2832{
2833 unsigned char *buffer;
2834 const int vpd_len = 8;
2835
2836 if (sdkp->lbpme == 0)
2837 return;
2838
2839 buffer = kmalloc(vpd_len, GFP_KERNEL);
2840
2841 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2842 goto out;
2843
2844 sdkp->lbpvpd = 1;
2845 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2846 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2847 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2848
2849 out:
2850 kfree(buffer);
2851}
2852
2853static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2854{
2855 struct scsi_device *sdev = sdkp->device;
2856
2857 if (sdev->host->no_write_same) {
2858 sdev->no_write_same = 1;
2859
2860 return;
2861 }
2862
2863 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2864 /* too large values might cause issues with arcmsr */
2865 int vpd_buf_len = 64;
2866
2867 sdev->no_report_opcodes = 1;
2868
2869 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2870 * CODES is unsupported and the device has an ATA
2871 * Information VPD page (SAT).
2872 */
2873 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2874 sdev->no_write_same = 1;
2875 }
2876
2877 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2878 sdkp->ws16 = 1;
2879
2880 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2881 sdkp->ws10 = 1;
2882}
2883
2884/**
2885 * sd_revalidate_disk - called the first time a new disk is seen,
2886 * performs disk spin up, read_capacity, etc.
2887 * @disk: struct gendisk we care about
2888 **/
2889static int sd_revalidate_disk(struct gendisk *disk)
2890{
2891 struct scsi_disk *sdkp = scsi_disk(disk);
2892 struct scsi_device *sdp = sdkp->device;
2893 struct request_queue *q = sdkp->disk->queue;
2894 sector_t old_capacity = sdkp->capacity;
2895 unsigned char *buffer;
2896 unsigned int dev_max, rw_max;
2897
2898 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2899 "sd_revalidate_disk\n"));
2900
2901 /*
2902 * If the device is offline, don't try and read capacity or any
2903 * of the other niceties.
2904 */
2905 if (!scsi_device_online(sdp))
2906 goto out;
2907
2908 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2909 if (!buffer) {
2910 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2911 "allocation failure.\n");
2912 goto out;
2913 }
2914
2915 sd_spinup_disk(sdkp);
2916
2917 /*
2918 * Without media there is no reason to ask; moreover, some devices
2919 * react badly if we do.
2920 */
2921 if (sdkp->media_present) {
2922 sd_read_capacity(sdkp, buffer);
2923
2924 if (scsi_device_supports_vpd(sdp)) {
2925 sd_read_block_provisioning(sdkp);
2926 sd_read_block_limits(sdkp);
2927 sd_read_block_characteristics(sdkp);
2928 sd_zbc_read_zones(sdkp, buffer);
2929 }
2930
2931 sd_print_capacity(sdkp, old_capacity);
2932
2933 sd_read_write_protect_flag(sdkp, buffer);
2934 sd_read_cache_type(sdkp, buffer);
2935 sd_read_app_tag_own(sdkp, buffer);
2936 sd_read_write_same(sdkp, buffer);
2937 }
2938
2939 sdkp->first_scan = 0;
2940
2941 /*
2942 * We now have all cache related info, determine how we deal
2943 * with flush requests.
2944 */
2945 sd_set_flush_flag(sdkp);
2946
2947 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2948 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2949
2950 /* Some devices report a maximum block count for READ/WRITE requests. */
2951 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2952 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2953
2954 /*
2955 * Use the device's preferred I/O size for reads and writes
2956 * unless the reported value is unreasonably small, large, or
2957 * garbage.
2958 */
2959 if (sdkp->opt_xfer_blocks &&
2960 sdkp->opt_xfer_blocks <= dev_max &&
2961 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2962 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2963 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2964 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2965 } else
2966 rw_max = BLK_DEF_MAX_SECTORS;
2967
2968 /* Combine with controller limits */
2969 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2970
2971 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2972 sd_config_write_same(sdkp);
2973 kfree(buffer);
2974
2975 out:
2976 return 0;
2977}
2978
2979/**
2980 * sd_unlock_native_capacity - unlock native capacity
2981 * @disk: struct gendisk to set capacity for
2982 *
2983 * Block layer calls this function if it detects that partitions
2984 * on @disk reach beyond the end of the device. If the SCSI host
2985 * implements ->unlock_native_capacity() method, it's invoked to
2986 * give it a chance to adjust the device capacity.
2987 *
2988 * CONTEXT:
2989 * Defined by block layer. Might sleep.
2990 */
2991static void sd_unlock_native_capacity(struct gendisk *disk)
2992{
2993 struct scsi_device *sdev = scsi_disk(disk)->device;
2994
2995 if (sdev->host->hostt->unlock_native_capacity)
2996 sdev->host->hostt->unlock_native_capacity(sdev);
2997}
2998
2999/**
3000 * sd_format_disk_name - format disk name
3001 * @prefix: name prefix - ie. "sd" for SCSI disks
3002 * @index: index of the disk to format name for
3003 * @buf: output buffer
3004 * @buflen: length of the output buffer
3005 *
3006 * SCSI disk names starts at sda. The 26th device is sdz and the
3007 * 27th is sdaa. The last one for two lettered suffix is sdzz
3008 * which is followed by sdaaa.
3009 *
3010 * This is basically 26 base counting with one extra 'nil' entry
3011 * at the beginning from the second digit on and can be
3012 * determined using similar method as 26 base conversion with the
3013 * index shifted -1 after each digit is computed.
3014 *
3015 * CONTEXT:
3016 * Don't care.
3017 *
3018 * RETURNS:
3019 * 0 on success, -errno on failure.
3020 */
3021static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3022{
3023 const int base = 'z' - 'a' + 1;
3024 char *begin = buf + strlen(prefix);
3025 char *end = buf + buflen;
3026 char *p;
3027 int unit;
3028
3029 p = end - 1;
3030 *p = '\0';
3031 unit = base;
3032 do {
3033 if (p == begin)
3034 return -EINVAL;
3035 *--p = 'a' + (index % unit);
3036 index = (index / unit) - 1;
3037 } while (index >= 0);
3038
3039 memmove(begin, p, end - p);
3040 memcpy(buf, prefix, strlen(prefix));
3041
3042 return 0;
3043}
3044
3045/*
3046 * The asynchronous part of sd_probe
3047 */
3048static void sd_probe_async(void *data, async_cookie_t cookie)
3049{
3050 struct scsi_disk *sdkp = data;
3051 struct scsi_device *sdp;
3052 struct gendisk *gd;
3053 u32 index;
3054 struct device *dev;
3055
3056 sdp = sdkp->device;
3057 gd = sdkp->disk;
3058 index = sdkp->index;
3059 dev = &sdp->sdev_gendev;
3060
3061 gd->major = sd_major((index & 0xf0) >> 4);
3062 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3063 gd->minors = SD_MINORS;
3064
3065 gd->fops = &sd_fops;
3066 gd->private_data = &sdkp->driver;
3067 gd->queue = sdkp->device->request_queue;
3068
3069 /* defaults, until the device tells us otherwise */
3070 sdp->sector_size = 512;
3071 sdkp->capacity = 0;
3072 sdkp->media_present = 1;
3073 sdkp->write_prot = 0;
3074 sdkp->cache_override = 0;
3075 sdkp->WCE = 0;
3076 sdkp->RCD = 0;
3077 sdkp->ATO = 0;
3078 sdkp->first_scan = 1;
3079 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3080
3081 sd_revalidate_disk(gd);
3082
3083 gd->flags = GENHD_FL_EXT_DEVT;
3084 if (sdp->removable) {
3085 gd->flags |= GENHD_FL_REMOVABLE;
3086 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3087 }
3088
3089 blk_pm_runtime_init(sdp->request_queue, dev);
3090 device_add_disk(dev, gd);
3091 if (sdkp->capacity)
3092 sd_dif_config_host(sdkp);
3093
3094 sd_revalidate_disk(gd);
3095
3096 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3097 sdp->removable ? "removable " : "");
3098 scsi_autopm_put_device(sdp);
3099 put_device(&sdkp->dev);
3100}
3101
3102/**
3103 * sd_probe - called during driver initialization and whenever a
3104 * new scsi device is attached to the system. It is called once
3105 * for each scsi device (not just disks) present.
3106 * @dev: pointer to device object
3107 *
3108 * Returns 0 if successful (or not interested in this scsi device
3109 * (e.g. scanner)); 1 when there is an error.
3110 *
3111 * Note: this function is invoked from the scsi mid-level.
3112 * This function sets up the mapping between a given
3113 * <host,channel,id,lun> (found in sdp) and new device name
3114 * (e.g. /dev/sda). More precisely it is the block device major
3115 * and minor number that is chosen here.
3116 *
3117 * Assume sd_probe is not re-entrant (for time being)
3118 * Also think about sd_probe() and sd_remove() running coincidentally.
3119 **/
3120static int sd_probe(struct device *dev)
3121{
3122 struct scsi_device *sdp = to_scsi_device(dev);
3123 struct scsi_disk *sdkp;
3124 struct gendisk *gd;
3125 int index;
3126 int error;
3127
3128 scsi_autopm_get_device(sdp);
3129 error = -ENODEV;
3130 if (sdp->type != TYPE_DISK &&
3131 sdp->type != TYPE_ZBC &&
3132 sdp->type != TYPE_MOD &&
3133 sdp->type != TYPE_RBC)
3134 goto out;
3135
3136#ifndef CONFIG_BLK_DEV_ZONED
3137 if (sdp->type == TYPE_ZBC)
3138 goto out;
3139#endif
3140 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3141 "sd_probe\n"));
3142
3143 error = -ENOMEM;
3144 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3145 if (!sdkp)
3146 goto out;
3147
3148 gd = alloc_disk(SD_MINORS);
3149 if (!gd)
3150 goto out_free;
3151
3152 do {
3153 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3154 goto out_put;
3155
3156 spin_lock(&sd_index_lock);
3157 error = ida_get_new(&sd_index_ida, &index);
3158 spin_unlock(&sd_index_lock);
3159 } while (error == -EAGAIN);
3160
3161 if (error) {
3162 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3163 goto out_put;
3164 }
3165
3166 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3167 if (error) {
3168 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3169 goto out_free_index;
3170 }
3171
3172 sdkp->device = sdp;
3173 sdkp->driver = &sd_template;
3174 sdkp->disk = gd;
3175 sdkp->index = index;
3176 atomic_set(&sdkp->openers, 0);
3177 atomic_set(&sdkp->device->ioerr_cnt, 0);
3178
3179 if (!sdp->request_queue->rq_timeout) {
3180 if (sdp->type != TYPE_MOD)
3181 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3182 else
3183 blk_queue_rq_timeout(sdp->request_queue,
3184 SD_MOD_TIMEOUT);
3185 }
3186
3187 device_initialize(&sdkp->dev);
3188 sdkp->dev.parent = dev;
3189 sdkp->dev.class = &sd_disk_class;
3190 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3191
3192 error = device_add(&sdkp->dev);
3193 if (error)
3194 goto out_free_index;
3195
3196 get_device(dev);
3197 dev_set_drvdata(dev, sdkp);
3198
3199 get_device(&sdkp->dev); /* prevent release before async_schedule */
3200 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3201
3202 return 0;
3203
3204 out_free_index:
3205 spin_lock(&sd_index_lock);
3206 ida_remove(&sd_index_ida, index);
3207 spin_unlock(&sd_index_lock);
3208 out_put:
3209 put_disk(gd);
3210 out_free:
3211 kfree(sdkp);
3212 out:
3213 scsi_autopm_put_device(sdp);
3214 return error;
3215}
3216
3217/**
3218 * sd_remove - called whenever a scsi disk (previously recognized by
3219 * sd_probe) is detached from the system. It is called (potentially
3220 * multiple times) during sd module unload.
3221 * @sdp: pointer to mid level scsi device object
3222 *
3223 * Note: this function is invoked from the scsi mid-level.
3224 * This function potentially frees up a device name (e.g. /dev/sdc)
3225 * that could be re-used by a subsequent sd_probe().
3226 * This function is not called when the built-in sd driver is "exit-ed".
3227 **/
3228static int sd_remove(struct device *dev)
3229{
3230 struct scsi_disk *sdkp;
3231 dev_t devt;
3232
3233 sdkp = dev_get_drvdata(dev);
3234 devt = disk_devt(sdkp->disk);
3235 scsi_autopm_get_device(sdkp->device);
3236
3237 async_synchronize_full_domain(&scsi_sd_pm_domain);
3238 async_synchronize_full_domain(&scsi_sd_probe_domain);
3239 device_del(&sdkp->dev);
3240 del_gendisk(sdkp->disk);
3241 sd_shutdown(dev);
3242
3243 sd_zbc_remove(sdkp);
3244
3245 blk_register_region(devt, SD_MINORS, NULL,
3246 sd_default_probe, NULL, NULL);
3247
3248 mutex_lock(&sd_ref_mutex);
3249 dev_set_drvdata(dev, NULL);
3250 put_device(&sdkp->dev);
3251 mutex_unlock(&sd_ref_mutex);
3252
3253 return 0;
3254}
3255
3256/**
3257 * scsi_disk_release - Called to free the scsi_disk structure
3258 * @dev: pointer to embedded class device
3259 *
3260 * sd_ref_mutex must be held entering this routine. Because it is
3261 * called on last put, you should always use the scsi_disk_get()
3262 * scsi_disk_put() helpers which manipulate the semaphore directly
3263 * and never do a direct put_device.
3264 **/
3265static void scsi_disk_release(struct device *dev)
3266{
3267 struct scsi_disk *sdkp = to_scsi_disk(dev);
3268 struct gendisk *disk = sdkp->disk;
3269
3270 spin_lock(&sd_index_lock);
3271 ida_remove(&sd_index_ida, sdkp->index);
3272 spin_unlock(&sd_index_lock);
3273
3274 disk->private_data = NULL;
3275 put_disk(disk);
3276 put_device(&sdkp->device->sdev_gendev);
3277
3278 kfree(sdkp);
3279}
3280
3281static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3282{
3283 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3284 struct scsi_sense_hdr sshdr;
3285 struct scsi_device *sdp = sdkp->device;
3286 int res;
3287
3288 if (start)
3289 cmd[4] |= 1; /* START */
3290
3291 if (sdp->start_stop_pwr_cond)
3292 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3293
3294 if (!scsi_device_online(sdp))
3295 return -ENODEV;
3296
3297 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3298 SD_TIMEOUT, SD_MAX_RETRIES, NULL, 0, RQF_PM);
3299 if (res) {
3300 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3301 if (driver_byte(res) & DRIVER_SENSE)
3302 sd_print_sense_hdr(sdkp, &sshdr);
3303 if (scsi_sense_valid(&sshdr) &&
3304 /* 0x3a is medium not present */
3305 sshdr.asc == 0x3a)
3306 res = 0;
3307 }
3308
3309 /* SCSI error codes must not go to the generic layer */
3310 if (res)
3311 return -EIO;
3312
3313 return 0;
3314}
3315
3316/*
3317 * Send a SYNCHRONIZE CACHE instruction down to the device through
3318 * the normal SCSI command structure. Wait for the command to
3319 * complete.
3320 */
3321static void sd_shutdown(struct device *dev)
3322{
3323 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3324
3325 if (!sdkp)
3326 return; /* this can happen */
3327
3328 if (pm_runtime_suspended(dev))
3329 return;
3330
3331 if (sdkp->WCE && sdkp->media_present) {
3332 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3333 sd_sync_cache(sdkp);
3334 }
3335
3336 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3337 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3338 sd_start_stop_device(sdkp, 0);
3339 }
3340}
3341
3342static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3343{
3344 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3345 int ret = 0;
3346
3347 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3348 return 0;
3349
3350 if (sdkp->WCE && sdkp->media_present) {
3351 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3352 ret = sd_sync_cache(sdkp);
3353 if (ret) {
3354 /* ignore OFFLINE device */
3355 if (ret == -ENODEV)
3356 ret = 0;
3357 goto done;
3358 }
3359 }
3360
3361 if (sdkp->device->manage_start_stop) {
3362 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3363 /* an error is not worth aborting a system sleep */
3364 ret = sd_start_stop_device(sdkp, 0);
3365 if (ignore_stop_errors)
3366 ret = 0;
3367 }
3368
3369done:
3370 return ret;
3371}
3372
3373static int sd_suspend_system(struct device *dev)
3374{
3375 return sd_suspend_common(dev, true);
3376}
3377
3378static int sd_suspend_runtime(struct device *dev)
3379{
3380 return sd_suspend_common(dev, false);
3381}
3382
3383static int sd_resume(struct device *dev)
3384{
3385 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3386
3387 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3388 return 0;
3389
3390 if (!sdkp->device->manage_start_stop)
3391 return 0;
3392
3393 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3394 return sd_start_stop_device(sdkp, 1);
3395}
3396
3397/**
3398 * init_sd - entry point for this driver (both when built in or when
3399 * a module).
3400 *
3401 * Note: this function registers this driver with the scsi mid-level.
3402 **/
3403static int __init init_sd(void)
3404{
3405 int majors = 0, i, err;
3406
3407 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3408
3409 for (i = 0; i < SD_MAJORS; i++) {
3410 if (register_blkdev(sd_major(i), "sd") != 0)
3411 continue;
3412 majors++;
3413 blk_register_region(sd_major(i), SD_MINORS, NULL,
3414 sd_default_probe, NULL, NULL);
3415 }
3416
3417 if (!majors)
3418 return -ENODEV;
3419
3420 err = class_register(&sd_disk_class);
3421 if (err)
3422 goto err_out;
3423
3424 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3425 0, 0, NULL);
3426 if (!sd_cdb_cache) {
3427 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3428 err = -ENOMEM;
3429 goto err_out_class;
3430 }
3431
3432 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3433 if (!sd_cdb_pool) {
3434 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3435 err = -ENOMEM;
3436 goto err_out_cache;
3437 }
3438
3439 err = scsi_register_driver(&sd_template.gendrv);
3440 if (err)
3441 goto err_out_driver;
3442
3443 return 0;
3444
3445err_out_driver:
3446 mempool_destroy(sd_cdb_pool);
3447
3448err_out_cache:
3449 kmem_cache_destroy(sd_cdb_cache);
3450
3451err_out_class:
3452 class_unregister(&sd_disk_class);
3453err_out:
3454 for (i = 0; i < SD_MAJORS; i++)
3455 unregister_blkdev(sd_major(i), "sd");
3456 return err;
3457}
3458
3459/**
3460 * exit_sd - exit point for this driver (when it is a module).
3461 *
3462 * Note: this function unregisters this driver from the scsi mid-level.
3463 **/
3464static void __exit exit_sd(void)
3465{
3466 int i;
3467
3468 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3469
3470 scsi_unregister_driver(&sd_template.gendrv);
3471 mempool_destroy(sd_cdb_pool);
3472 kmem_cache_destroy(sd_cdb_cache);
3473
3474 class_unregister(&sd_disk_class);
3475
3476 for (i = 0; i < SD_MAJORS; i++) {
3477 blk_unregister_region(sd_major(i), SD_MINORS);
3478 unregister_blkdev(sd_major(i), "sd");
3479 }
3480}
3481
3482module_init(init_sd);
3483module_exit(exit_sd);
3484
3485static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3486 struct scsi_sense_hdr *sshdr)
3487{
3488 scsi_print_sense_hdr(sdkp->device,
3489 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3490}
3491
3492static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3493 int result)
3494{
3495 const char *hb_string = scsi_hostbyte_string(result);
3496 const char *db_string = scsi_driverbyte_string(result);
3497
3498 if (hb_string || db_string)
3499 sd_printk(KERN_INFO, sdkp,
3500 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3501 hb_string ? hb_string : "invalid",
3502 db_string ? db_string : "invalid");
3503 else
3504 sd_printk(KERN_INFO, sdkp,
3505 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3506 msg, host_byte(result), driver_byte(result));
3507}
3508
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 *
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
26 *
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
34 */
35
36#include <linux/bio-integrity.h>
37#include <linux/module.h>
38#include <linux/fs.h>
39#include <linux/kernel.h>
40#include <linux/mm.h>
41#include <linux/hdreg.h>
42#include <linux/errno.h>
43#include <linux/idr.h>
44#include <linux/interrupt.h>
45#include <linux/init.h>
46#include <linux/blkdev.h>
47#include <linux/blkpg.h>
48#include <linux/blk-pm.h>
49#include <linux/delay.h>
50#include <linux/rw_hint.h>
51#include <linux/major.h>
52#include <linux/mutex.h>
53#include <linux/string_helpers.h>
54#include <linux/slab.h>
55#include <linux/sed-opal.h>
56#include <linux/pm_runtime.h>
57#include <linux/pr.h>
58#include <linux/t10-pi.h>
59#include <linux/uaccess.h>
60#include <linux/unaligned.h>
61
62#include <scsi/scsi.h>
63#include <scsi/scsi_cmnd.h>
64#include <scsi/scsi_dbg.h>
65#include <scsi/scsi_device.h>
66#include <scsi/scsi_devinfo.h>
67#include <scsi/scsi_driver.h>
68#include <scsi/scsi_eh.h>
69#include <scsi/scsi_host.h>
70#include <scsi/scsi_ioctl.h>
71#include <scsi/scsicam.h>
72#include <scsi/scsi_common.h>
73
74#include "sd.h"
75#include "scsi_priv.h"
76#include "scsi_logging.h"
77
78MODULE_AUTHOR("Eric Youngdale");
79MODULE_DESCRIPTION("SCSI disk (sd) driver");
80MODULE_LICENSE("GPL");
81
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
94MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
95MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
96MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
97MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
98MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
99MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
100MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
101MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
102
103#define SD_MINORS 16
104
105static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
106 unsigned int mode);
107static void sd_config_write_same(struct scsi_disk *sdkp,
108 struct queue_limits *lim);
109static int sd_revalidate_disk(struct gendisk *);
110static void sd_unlock_native_capacity(struct gendisk *disk);
111static void sd_shutdown(struct device *);
112static void scsi_disk_release(struct device *cdev);
113
114static DEFINE_IDA(sd_index_ida);
115
116static mempool_t *sd_page_pool;
117static struct lock_class_key sd_bio_compl_lkclass;
118
119static const char *sd_cache_types[] = {
120 "write through", "none", "write back",
121 "write back, no read (daft)"
122};
123
124static void sd_set_flush_flag(struct scsi_disk *sdkp,
125 struct queue_limits *lim)
126{
127 if (sdkp->WCE) {
128 lim->features |= BLK_FEAT_WRITE_CACHE;
129 if (sdkp->DPOFUA)
130 lim->features |= BLK_FEAT_FUA;
131 else
132 lim->features &= ~BLK_FEAT_FUA;
133 } else {
134 lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA);
135 }
136}
137
138static ssize_t
139cache_type_store(struct device *dev, struct device_attribute *attr,
140 const char *buf, size_t count)
141{
142 int ct, rcd, wce, sp;
143 struct scsi_disk *sdkp = to_scsi_disk(dev);
144 struct scsi_device *sdp = sdkp->device;
145 char buffer[64];
146 char *buffer_data;
147 struct scsi_mode_data data;
148 struct scsi_sense_hdr sshdr;
149 static const char temp[] = "temporary ";
150 int len, ret;
151
152 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
153 /* no cache control on RBC devices; theoretically they
154 * can do it, but there's probably so many exceptions
155 * it's not worth the risk */
156 return -EINVAL;
157
158 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
159 buf += sizeof(temp) - 1;
160 sdkp->cache_override = 1;
161 } else {
162 sdkp->cache_override = 0;
163 }
164
165 ct = sysfs_match_string(sd_cache_types, buf);
166 if (ct < 0)
167 return -EINVAL;
168
169 rcd = ct & 0x01 ? 1 : 0;
170 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
171
172 if (sdkp->cache_override) {
173 struct queue_limits lim;
174
175 sdkp->WCE = wce;
176 sdkp->RCD = rcd;
177
178 lim = queue_limits_start_update(sdkp->disk->queue);
179 sd_set_flush_flag(sdkp, &lim);
180 ret = queue_limits_commit_update_frozen(sdkp->disk->queue,
181 &lim);
182 if (ret)
183 return ret;
184 return count;
185 }
186
187 if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
188 sdkp->max_retries, &data, NULL))
189 return -EINVAL;
190 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
191 data.block_descriptor_length);
192 buffer_data = buffer + data.header_length +
193 data.block_descriptor_length;
194 buffer_data[2] &= ~0x05;
195 buffer_data[2] |= wce << 2 | rcd;
196 sp = buffer_data[0] & 0x80 ? 1 : 0;
197 buffer_data[0] &= ~0x80;
198
199 /*
200 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
201 * received mode parameter buffer before doing MODE SELECT.
202 */
203 data.device_specific = 0;
204
205 ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
206 sdkp->max_retries, &data, &sshdr);
207 if (ret) {
208 if (ret > 0 && scsi_sense_valid(&sshdr))
209 sd_print_sense_hdr(sdkp, &sshdr);
210 return -EINVAL;
211 }
212 sd_revalidate_disk(sdkp->disk);
213 return count;
214}
215
216static ssize_t
217manage_start_stop_show(struct device *dev,
218 struct device_attribute *attr, char *buf)
219{
220 struct scsi_disk *sdkp = to_scsi_disk(dev);
221 struct scsi_device *sdp = sdkp->device;
222
223 return sysfs_emit(buf, "%u\n",
224 sdp->manage_system_start_stop &&
225 sdp->manage_runtime_start_stop &&
226 sdp->manage_shutdown);
227}
228static DEVICE_ATTR_RO(manage_start_stop);
229
230static ssize_t
231manage_system_start_stop_show(struct device *dev,
232 struct device_attribute *attr, char *buf)
233{
234 struct scsi_disk *sdkp = to_scsi_disk(dev);
235 struct scsi_device *sdp = sdkp->device;
236
237 return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
238}
239
240static ssize_t
241manage_system_start_stop_store(struct device *dev,
242 struct device_attribute *attr,
243 const char *buf, size_t count)
244{
245 struct scsi_disk *sdkp = to_scsi_disk(dev);
246 struct scsi_device *sdp = sdkp->device;
247 bool v;
248
249 if (!capable(CAP_SYS_ADMIN))
250 return -EACCES;
251
252 if (kstrtobool(buf, &v))
253 return -EINVAL;
254
255 sdp->manage_system_start_stop = v;
256
257 return count;
258}
259static DEVICE_ATTR_RW(manage_system_start_stop);
260
261static ssize_t
262manage_runtime_start_stop_show(struct device *dev,
263 struct device_attribute *attr, char *buf)
264{
265 struct scsi_disk *sdkp = to_scsi_disk(dev);
266 struct scsi_device *sdp = sdkp->device;
267
268 return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
269}
270
271static ssize_t
272manage_runtime_start_stop_store(struct device *dev,
273 struct device_attribute *attr,
274 const char *buf, size_t count)
275{
276 struct scsi_disk *sdkp = to_scsi_disk(dev);
277 struct scsi_device *sdp = sdkp->device;
278 bool v;
279
280 if (!capable(CAP_SYS_ADMIN))
281 return -EACCES;
282
283 if (kstrtobool(buf, &v))
284 return -EINVAL;
285
286 sdp->manage_runtime_start_stop = v;
287
288 return count;
289}
290static DEVICE_ATTR_RW(manage_runtime_start_stop);
291
292static ssize_t manage_shutdown_show(struct device *dev,
293 struct device_attribute *attr, char *buf)
294{
295 struct scsi_disk *sdkp = to_scsi_disk(dev);
296 struct scsi_device *sdp = sdkp->device;
297
298 return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
299}
300
301static ssize_t manage_shutdown_store(struct device *dev,
302 struct device_attribute *attr,
303 const char *buf, size_t count)
304{
305 struct scsi_disk *sdkp = to_scsi_disk(dev);
306 struct scsi_device *sdp = sdkp->device;
307 bool v;
308
309 if (!capable(CAP_SYS_ADMIN))
310 return -EACCES;
311
312 if (kstrtobool(buf, &v))
313 return -EINVAL;
314
315 sdp->manage_shutdown = v;
316
317 return count;
318}
319static DEVICE_ATTR_RW(manage_shutdown);
320
321static ssize_t
322allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
323{
324 struct scsi_disk *sdkp = to_scsi_disk(dev);
325
326 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
327}
328
329static ssize_t
330allow_restart_store(struct device *dev, struct device_attribute *attr,
331 const char *buf, size_t count)
332{
333 bool v;
334 struct scsi_disk *sdkp = to_scsi_disk(dev);
335 struct scsi_device *sdp = sdkp->device;
336
337 if (!capable(CAP_SYS_ADMIN))
338 return -EACCES;
339
340 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
341 return -EINVAL;
342
343 if (kstrtobool(buf, &v))
344 return -EINVAL;
345
346 sdp->allow_restart = v;
347
348 return count;
349}
350static DEVICE_ATTR_RW(allow_restart);
351
352static ssize_t
353cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
354{
355 struct scsi_disk *sdkp = to_scsi_disk(dev);
356 int ct = sdkp->RCD + 2*sdkp->WCE;
357
358 return sprintf(buf, "%s\n", sd_cache_types[ct]);
359}
360static DEVICE_ATTR_RW(cache_type);
361
362static ssize_t
363FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
364{
365 struct scsi_disk *sdkp = to_scsi_disk(dev);
366
367 return sprintf(buf, "%u\n", sdkp->DPOFUA);
368}
369static DEVICE_ATTR_RO(FUA);
370
371static ssize_t
372protection_type_show(struct device *dev, struct device_attribute *attr,
373 char *buf)
374{
375 struct scsi_disk *sdkp = to_scsi_disk(dev);
376
377 return sprintf(buf, "%u\n", sdkp->protection_type);
378}
379
380static ssize_t
381protection_type_store(struct device *dev, struct device_attribute *attr,
382 const char *buf, size_t count)
383{
384 struct scsi_disk *sdkp = to_scsi_disk(dev);
385 unsigned int val;
386 int err;
387
388 if (!capable(CAP_SYS_ADMIN))
389 return -EACCES;
390
391 err = kstrtouint(buf, 10, &val);
392
393 if (err)
394 return err;
395
396 if (val <= T10_PI_TYPE3_PROTECTION)
397 sdkp->protection_type = val;
398
399 return count;
400}
401static DEVICE_ATTR_RW(protection_type);
402
403static ssize_t
404protection_mode_show(struct device *dev, struct device_attribute *attr,
405 char *buf)
406{
407 struct scsi_disk *sdkp = to_scsi_disk(dev);
408 struct scsi_device *sdp = sdkp->device;
409 unsigned int dif, dix;
410
411 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
412 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
413
414 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
415 dif = 0;
416 dix = 1;
417 }
418
419 if (!dif && !dix)
420 return sprintf(buf, "none\n");
421
422 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
423}
424static DEVICE_ATTR_RO(protection_mode);
425
426static ssize_t
427app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
428{
429 struct scsi_disk *sdkp = to_scsi_disk(dev);
430
431 return sprintf(buf, "%u\n", sdkp->ATO);
432}
433static DEVICE_ATTR_RO(app_tag_own);
434
435static ssize_t
436thin_provisioning_show(struct device *dev, struct device_attribute *attr,
437 char *buf)
438{
439 struct scsi_disk *sdkp = to_scsi_disk(dev);
440
441 return sprintf(buf, "%u\n", sdkp->lbpme);
442}
443static DEVICE_ATTR_RO(thin_provisioning);
444
445/* sysfs_match_string() requires dense arrays */
446static const char *lbp_mode[] = {
447 [SD_LBP_FULL] = "full",
448 [SD_LBP_UNMAP] = "unmap",
449 [SD_LBP_WS16] = "writesame_16",
450 [SD_LBP_WS10] = "writesame_10",
451 [SD_LBP_ZERO] = "writesame_zero",
452 [SD_LBP_DISABLE] = "disabled",
453};
454
455static ssize_t
456provisioning_mode_show(struct device *dev, struct device_attribute *attr,
457 char *buf)
458{
459 struct scsi_disk *sdkp = to_scsi_disk(dev);
460
461 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
462}
463
464static ssize_t
465provisioning_mode_store(struct device *dev, struct device_attribute *attr,
466 const char *buf, size_t count)
467{
468 struct scsi_disk *sdkp = to_scsi_disk(dev);
469 struct scsi_device *sdp = sdkp->device;
470 struct queue_limits lim;
471 int mode, err;
472
473 if (!capable(CAP_SYS_ADMIN))
474 return -EACCES;
475
476 if (sdp->type != TYPE_DISK)
477 return -EINVAL;
478
479 mode = sysfs_match_string(lbp_mode, buf);
480 if (mode < 0)
481 return -EINVAL;
482
483 lim = queue_limits_start_update(sdkp->disk->queue);
484 sd_config_discard(sdkp, &lim, mode);
485 err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
486 if (err)
487 return err;
488 return count;
489}
490static DEVICE_ATTR_RW(provisioning_mode);
491
492/* sysfs_match_string() requires dense arrays */
493static const char *zeroing_mode[] = {
494 [SD_ZERO_WRITE] = "write",
495 [SD_ZERO_WS] = "writesame",
496 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
497 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
498};
499
500static ssize_t
501zeroing_mode_show(struct device *dev, struct device_attribute *attr,
502 char *buf)
503{
504 struct scsi_disk *sdkp = to_scsi_disk(dev);
505
506 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
507}
508
509static ssize_t
510zeroing_mode_store(struct device *dev, struct device_attribute *attr,
511 const char *buf, size_t count)
512{
513 struct scsi_disk *sdkp = to_scsi_disk(dev);
514 int mode;
515
516 if (!capable(CAP_SYS_ADMIN))
517 return -EACCES;
518
519 mode = sysfs_match_string(zeroing_mode, buf);
520 if (mode < 0)
521 return -EINVAL;
522
523 sdkp->zeroing_mode = mode;
524
525 return count;
526}
527static DEVICE_ATTR_RW(zeroing_mode);
528
529static ssize_t
530max_medium_access_timeouts_show(struct device *dev,
531 struct device_attribute *attr, char *buf)
532{
533 struct scsi_disk *sdkp = to_scsi_disk(dev);
534
535 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
536}
537
538static ssize_t
539max_medium_access_timeouts_store(struct device *dev,
540 struct device_attribute *attr, const char *buf,
541 size_t count)
542{
543 struct scsi_disk *sdkp = to_scsi_disk(dev);
544 int err;
545
546 if (!capable(CAP_SYS_ADMIN))
547 return -EACCES;
548
549 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
550
551 return err ? err : count;
552}
553static DEVICE_ATTR_RW(max_medium_access_timeouts);
554
555static ssize_t
556max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
557 char *buf)
558{
559 struct scsi_disk *sdkp = to_scsi_disk(dev);
560
561 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
562}
563
564static ssize_t
565max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
566 const char *buf, size_t count)
567{
568 struct scsi_disk *sdkp = to_scsi_disk(dev);
569 struct scsi_device *sdp = sdkp->device;
570 struct queue_limits lim;
571 unsigned long max;
572 int err;
573
574 if (!capable(CAP_SYS_ADMIN))
575 return -EACCES;
576
577 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
578 return -EINVAL;
579
580 err = kstrtoul(buf, 10, &max);
581
582 if (err)
583 return err;
584
585 if (max == 0)
586 sdp->no_write_same = 1;
587 else if (max <= SD_MAX_WS16_BLOCKS) {
588 sdp->no_write_same = 0;
589 sdkp->max_ws_blocks = max;
590 }
591
592 lim = queue_limits_start_update(sdkp->disk->queue);
593 sd_config_write_same(sdkp, &lim);
594 err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
595 if (err)
596 return err;
597 return count;
598}
599static DEVICE_ATTR_RW(max_write_same_blocks);
600
601static ssize_t
602zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
603{
604 struct scsi_disk *sdkp = to_scsi_disk(dev);
605
606 if (sdkp->device->type == TYPE_ZBC)
607 return sprintf(buf, "host-managed\n");
608 if (sdkp->zoned == 1)
609 return sprintf(buf, "host-aware\n");
610 if (sdkp->zoned == 2)
611 return sprintf(buf, "drive-managed\n");
612 return sprintf(buf, "none\n");
613}
614static DEVICE_ATTR_RO(zoned_cap);
615
616static ssize_t
617max_retries_store(struct device *dev, struct device_attribute *attr,
618 const char *buf, size_t count)
619{
620 struct scsi_disk *sdkp = to_scsi_disk(dev);
621 struct scsi_device *sdev = sdkp->device;
622 int retries, err;
623
624 err = kstrtoint(buf, 10, &retries);
625 if (err)
626 return err;
627
628 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
629 sdkp->max_retries = retries;
630 return count;
631 }
632
633 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
634 SD_MAX_RETRIES);
635 return -EINVAL;
636}
637
638static ssize_t
639max_retries_show(struct device *dev, struct device_attribute *attr,
640 char *buf)
641{
642 struct scsi_disk *sdkp = to_scsi_disk(dev);
643
644 return sprintf(buf, "%d\n", sdkp->max_retries);
645}
646
647static DEVICE_ATTR_RW(max_retries);
648
649static struct attribute *sd_disk_attrs[] = {
650 &dev_attr_cache_type.attr,
651 &dev_attr_FUA.attr,
652 &dev_attr_allow_restart.attr,
653 &dev_attr_manage_start_stop.attr,
654 &dev_attr_manage_system_start_stop.attr,
655 &dev_attr_manage_runtime_start_stop.attr,
656 &dev_attr_manage_shutdown.attr,
657 &dev_attr_protection_type.attr,
658 &dev_attr_protection_mode.attr,
659 &dev_attr_app_tag_own.attr,
660 &dev_attr_thin_provisioning.attr,
661 &dev_attr_provisioning_mode.attr,
662 &dev_attr_zeroing_mode.attr,
663 &dev_attr_max_write_same_blocks.attr,
664 &dev_attr_max_medium_access_timeouts.attr,
665 &dev_attr_zoned_cap.attr,
666 &dev_attr_max_retries.attr,
667 NULL,
668};
669ATTRIBUTE_GROUPS(sd_disk);
670
671static struct class sd_disk_class = {
672 .name = "scsi_disk",
673 .dev_release = scsi_disk_release,
674 .dev_groups = sd_disk_groups,
675};
676
677/*
678 * Don't request a new module, as that could deadlock in multipath
679 * environment.
680 */
681static void sd_default_probe(dev_t devt)
682{
683}
684
685/*
686 * Device no to disk mapping:
687 *
688 * major disc2 disc p1
689 * |............|.............|....|....| <- dev_t
690 * 31 20 19 8 7 4 3 0
691 *
692 * Inside a major, we have 16k disks, however mapped non-
693 * contiguously. The first 16 disks are for major0, the next
694 * ones with major1, ... Disk 256 is for major0 again, disk 272
695 * for major1, ...
696 * As we stay compatible with our numbering scheme, we can reuse
697 * the well-know SCSI majors 8, 65--71, 136--143.
698 */
699static int sd_major(int major_idx)
700{
701 switch (major_idx) {
702 case 0:
703 return SCSI_DISK0_MAJOR;
704 case 1 ... 7:
705 return SCSI_DISK1_MAJOR + major_idx - 1;
706 case 8 ... 15:
707 return SCSI_DISK8_MAJOR + major_idx - 8;
708 default:
709 BUG();
710 return 0; /* shut up gcc */
711 }
712}
713
714#ifdef CONFIG_BLK_SED_OPAL
715static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
716 size_t len, bool send)
717{
718 struct scsi_disk *sdkp = data;
719 struct scsi_device *sdev = sdkp->device;
720 u8 cdb[12] = { 0, };
721 const struct scsi_exec_args exec_args = {
722 .req_flags = BLK_MQ_REQ_PM,
723 };
724 int ret;
725
726 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
727 cdb[1] = secp;
728 put_unaligned_be16(spsp, &cdb[2]);
729 put_unaligned_be32(len, &cdb[6]);
730
731 ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
732 buffer, len, SD_TIMEOUT, sdkp->max_retries,
733 &exec_args);
734 return ret <= 0 ? ret : -EIO;
735}
736#endif /* CONFIG_BLK_SED_OPAL */
737
738/*
739 * Look up the DIX operation based on whether the command is read or
740 * write and whether dix and dif are enabled.
741 */
742static unsigned int sd_prot_op(bool write, bool dix, bool dif)
743{
744 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
745 static const unsigned int ops[] = { /* wrt dix dif */
746 SCSI_PROT_NORMAL, /* 0 0 0 */
747 SCSI_PROT_READ_STRIP, /* 0 0 1 */
748 SCSI_PROT_READ_INSERT, /* 0 1 0 */
749 SCSI_PROT_READ_PASS, /* 0 1 1 */
750 SCSI_PROT_NORMAL, /* 1 0 0 */
751 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
752 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
753 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
754 };
755
756 return ops[write << 2 | dix << 1 | dif];
757}
758
759/*
760 * Returns a mask of the protection flags that are valid for a given DIX
761 * operation.
762 */
763static unsigned int sd_prot_flag_mask(unsigned int prot_op)
764{
765 static const unsigned int flag_mask[] = {
766 [SCSI_PROT_NORMAL] = 0,
767
768 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
769 SCSI_PROT_GUARD_CHECK |
770 SCSI_PROT_REF_CHECK |
771 SCSI_PROT_REF_INCREMENT,
772
773 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
774 SCSI_PROT_IP_CHECKSUM,
775
776 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
777 SCSI_PROT_GUARD_CHECK |
778 SCSI_PROT_REF_CHECK |
779 SCSI_PROT_REF_INCREMENT |
780 SCSI_PROT_IP_CHECKSUM,
781
782 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
783 SCSI_PROT_REF_INCREMENT,
784
785 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
786 SCSI_PROT_REF_CHECK |
787 SCSI_PROT_REF_INCREMENT |
788 SCSI_PROT_IP_CHECKSUM,
789
790 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
791 SCSI_PROT_GUARD_CHECK |
792 SCSI_PROT_REF_CHECK |
793 SCSI_PROT_REF_INCREMENT |
794 SCSI_PROT_IP_CHECKSUM,
795 };
796
797 return flag_mask[prot_op];
798}
799
800static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
801 unsigned int dix, unsigned int dif)
802{
803 struct request *rq = scsi_cmd_to_rq(scmd);
804 struct bio *bio = rq->bio;
805 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
806 unsigned int protect = 0;
807
808 if (dix) { /* DIX Type 0, 1, 2, 3 */
809 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
810 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
811
812 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
813 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
814 }
815
816 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
817 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
818
819 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
820 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
821 }
822
823 if (dif) { /* DIX/DIF Type 1, 2, 3 */
824 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
825
826 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
827 protect = 3 << 5; /* Disable target PI checking */
828 else
829 protect = 1 << 5; /* Enable target PI checking */
830 }
831
832 scsi_set_prot_op(scmd, prot_op);
833 scsi_set_prot_type(scmd, dif);
834 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
835
836 return protect;
837}
838
839static void sd_disable_discard(struct scsi_disk *sdkp)
840{
841 sdkp->provisioning_mode = SD_LBP_DISABLE;
842 blk_queue_disable_discard(sdkp->disk->queue);
843}
844
845static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
846 unsigned int mode)
847{
848 unsigned int logical_block_size = sdkp->device->sector_size;
849 unsigned int max_blocks = 0;
850
851 lim->discard_alignment = sdkp->unmap_alignment * logical_block_size;
852 lim->discard_granularity = max(sdkp->physical_block_size,
853 sdkp->unmap_granularity * logical_block_size);
854 sdkp->provisioning_mode = mode;
855
856 switch (mode) {
857
858 case SD_LBP_FULL:
859 case SD_LBP_DISABLE:
860 break;
861
862 case SD_LBP_UNMAP:
863 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
864 (u32)SD_MAX_WS16_BLOCKS);
865 break;
866
867 case SD_LBP_WS16:
868 if (sdkp->device->unmap_limit_for_ws)
869 max_blocks = sdkp->max_unmap_blocks;
870 else
871 max_blocks = sdkp->max_ws_blocks;
872
873 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
874 break;
875
876 case SD_LBP_WS10:
877 if (sdkp->device->unmap_limit_for_ws)
878 max_blocks = sdkp->max_unmap_blocks;
879 else
880 max_blocks = sdkp->max_ws_blocks;
881
882 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
883 break;
884
885 case SD_LBP_ZERO:
886 max_blocks = min_not_zero(sdkp->max_ws_blocks,
887 (u32)SD_MAX_WS10_BLOCKS);
888 break;
889 }
890
891 lim->max_hw_discard_sectors = max_blocks *
892 (logical_block_size >> SECTOR_SHIFT);
893}
894
895static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
896{
897 struct page *page;
898
899 page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
900 if (!page)
901 return NULL;
902 clear_highpage(page);
903 bvec_set_page(&rq->special_vec, page, data_len, 0);
904 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905 return bvec_virt(&rq->special_vec);
906}
907
908static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
909{
910 struct scsi_device *sdp = cmd->device;
911 struct request *rq = scsi_cmd_to_rq(cmd);
912 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
913 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
914 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
915 unsigned int data_len = 24;
916 char *buf;
917
918 buf = sd_set_special_bvec(rq, data_len);
919 if (!buf)
920 return BLK_STS_RESOURCE;
921
922 cmd->cmd_len = 10;
923 cmd->cmnd[0] = UNMAP;
924 cmd->cmnd[8] = 24;
925
926 put_unaligned_be16(6 + 16, &buf[0]);
927 put_unaligned_be16(16, &buf[2]);
928 put_unaligned_be64(lba, &buf[8]);
929 put_unaligned_be32(nr_blocks, &buf[16]);
930
931 cmd->allowed = sdkp->max_retries;
932 cmd->transfersize = data_len;
933 rq->timeout = SD_TIMEOUT;
934
935 return scsi_alloc_sgtables(cmd);
936}
937
938static void sd_config_atomic(struct scsi_disk *sdkp, struct queue_limits *lim)
939{
940 unsigned int logical_block_size = sdkp->device->sector_size,
941 physical_block_size_sectors, max_atomic, unit_min, unit_max;
942
943 if ((!sdkp->max_atomic && !sdkp->max_atomic_with_boundary) ||
944 sdkp->protection_type == T10_PI_TYPE2_PROTECTION)
945 return;
946
947 physical_block_size_sectors = sdkp->physical_block_size /
948 sdkp->device->sector_size;
949
950 unit_min = rounddown_pow_of_two(sdkp->atomic_granularity ?
951 sdkp->atomic_granularity :
952 physical_block_size_sectors);
953
954 /*
955 * Only use atomic boundary when we have the odd scenario of
956 * sdkp->max_atomic == 0, which the spec does permit.
957 */
958 if (sdkp->max_atomic) {
959 max_atomic = sdkp->max_atomic;
960 unit_max = rounddown_pow_of_two(sdkp->max_atomic);
961 sdkp->use_atomic_write_boundary = 0;
962 } else {
963 max_atomic = sdkp->max_atomic_with_boundary;
964 unit_max = rounddown_pow_of_two(sdkp->max_atomic_boundary);
965 sdkp->use_atomic_write_boundary = 1;
966 }
967
968 /*
969 * Ensure compliance with granularity and alignment. For now, keep it
970 * simple and just don't support atomic writes for values mismatched
971 * with max_{boundary}atomic, physical block size, and
972 * atomic_granularity itself.
973 *
974 * We're really being distrustful by checking unit_max also...
975 */
976 if (sdkp->atomic_granularity > 1) {
977 if (unit_min > 1 && unit_min % sdkp->atomic_granularity)
978 return;
979 if (unit_max > 1 && unit_max % sdkp->atomic_granularity)
980 return;
981 }
982
983 if (sdkp->atomic_alignment > 1) {
984 if (unit_min > 1 && unit_min % sdkp->atomic_alignment)
985 return;
986 if (unit_max > 1 && unit_max % sdkp->atomic_alignment)
987 return;
988 }
989
990 lim->atomic_write_hw_max = max_atomic * logical_block_size;
991 lim->atomic_write_hw_boundary = 0;
992 lim->atomic_write_hw_unit_min = unit_min * logical_block_size;
993 lim->atomic_write_hw_unit_max = unit_max * logical_block_size;
994}
995
996static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
997 bool unmap)
998{
999 struct scsi_device *sdp = cmd->device;
1000 struct request *rq = scsi_cmd_to_rq(cmd);
1001 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1002 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1003 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1004 u32 data_len = sdp->sector_size;
1005
1006 if (!sd_set_special_bvec(rq, data_len))
1007 return BLK_STS_RESOURCE;
1008
1009 cmd->cmd_len = 16;
1010 cmd->cmnd[0] = WRITE_SAME_16;
1011 if (unmap)
1012 cmd->cmnd[1] = 0x8; /* UNMAP */
1013 put_unaligned_be64(lba, &cmd->cmnd[2]);
1014 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1015
1016 cmd->allowed = sdkp->max_retries;
1017 cmd->transfersize = data_len;
1018 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1019
1020 return scsi_alloc_sgtables(cmd);
1021}
1022
1023static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
1024 bool unmap)
1025{
1026 struct scsi_device *sdp = cmd->device;
1027 struct request *rq = scsi_cmd_to_rq(cmd);
1028 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1029 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1030 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1031 u32 data_len = sdp->sector_size;
1032
1033 if (!sd_set_special_bvec(rq, data_len))
1034 return BLK_STS_RESOURCE;
1035
1036 cmd->cmd_len = 10;
1037 cmd->cmnd[0] = WRITE_SAME;
1038 if (unmap)
1039 cmd->cmnd[1] = 0x8; /* UNMAP */
1040 put_unaligned_be32(lba, &cmd->cmnd[2]);
1041 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042
1043 cmd->allowed = sdkp->max_retries;
1044 cmd->transfersize = data_len;
1045 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1046
1047 return scsi_alloc_sgtables(cmd);
1048}
1049
1050static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
1051{
1052 struct request *rq = scsi_cmd_to_rq(cmd);
1053 struct scsi_device *sdp = cmd->device;
1054 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1055 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1056 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1057
1058 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
1059 switch (sdkp->zeroing_mode) {
1060 case SD_ZERO_WS16_UNMAP:
1061 return sd_setup_write_same16_cmnd(cmd, true);
1062 case SD_ZERO_WS10_UNMAP:
1063 return sd_setup_write_same10_cmnd(cmd, true);
1064 }
1065 }
1066
1067 if (sdp->no_write_same) {
1068 rq->rq_flags |= RQF_QUIET;
1069 return BLK_STS_TARGET;
1070 }
1071
1072 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
1073 return sd_setup_write_same16_cmnd(cmd, false);
1074
1075 return sd_setup_write_same10_cmnd(cmd, false);
1076}
1077
1078static void sd_disable_write_same(struct scsi_disk *sdkp)
1079{
1080 sdkp->device->no_write_same = 1;
1081 sdkp->max_ws_blocks = 0;
1082 blk_queue_disable_write_zeroes(sdkp->disk->queue);
1083}
1084
1085static void sd_config_write_same(struct scsi_disk *sdkp,
1086 struct queue_limits *lim)
1087{
1088 unsigned int logical_block_size = sdkp->device->sector_size;
1089
1090 if (sdkp->device->no_write_same) {
1091 sdkp->max_ws_blocks = 0;
1092 goto out;
1093 }
1094
1095 /* Some devices can not handle block counts above 0xffff despite
1096 * supporting WRITE SAME(16). Consequently we default to 64k
1097 * blocks per I/O unless the device explicitly advertises a
1098 * bigger limit.
1099 */
1100 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1101 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1102 (u32)SD_MAX_WS16_BLOCKS);
1103 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1104 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1105 (u32)SD_MAX_WS10_BLOCKS);
1106 else {
1107 sdkp->device->no_write_same = 1;
1108 sdkp->max_ws_blocks = 0;
1109 }
1110
1111 if (sdkp->lbprz && sdkp->lbpws)
1112 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1113 else if (sdkp->lbprz && sdkp->lbpws10)
1114 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1115 else if (sdkp->max_ws_blocks)
1116 sdkp->zeroing_mode = SD_ZERO_WS;
1117 else
1118 sdkp->zeroing_mode = SD_ZERO_WRITE;
1119
1120 if (sdkp->max_ws_blocks &&
1121 sdkp->physical_block_size > logical_block_size) {
1122 /*
1123 * Reporting a maximum number of blocks that is not aligned
1124 * on the device physical size would cause a large write same
1125 * request to be split into physically unaligned chunks by
1126 * __blkdev_issue_write_zeroes() even if the caller of this
1127 * functions took care to align the large request. So make sure
1128 * the maximum reported is aligned to the device physical block
1129 * size. This is only an optional optimization for regular
1130 * disks, but this is mandatory to avoid failure of large write
1131 * same requests directed at sequential write required zones of
1132 * host-managed ZBC disks.
1133 */
1134 sdkp->max_ws_blocks =
1135 round_down(sdkp->max_ws_blocks,
1136 bytes_to_logical(sdkp->device,
1137 sdkp->physical_block_size));
1138 }
1139
1140out:
1141 lim->max_write_zeroes_sectors =
1142 sdkp->max_ws_blocks * (logical_block_size >> SECTOR_SHIFT);
1143}
1144
1145static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1146{
1147 struct request *rq = scsi_cmd_to_rq(cmd);
1148 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1149
1150 /* flush requests don't perform I/O, zero the S/G table */
1151 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1152
1153 if (cmd->device->use_16_for_sync) {
1154 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1155 cmd->cmd_len = 16;
1156 } else {
1157 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1158 cmd->cmd_len = 10;
1159 }
1160 cmd->transfersize = 0;
1161 cmd->allowed = sdkp->max_retries;
1162
1163 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1164 return BLK_STS_OK;
1165}
1166
1167/**
1168 * sd_group_number() - Compute the GROUP NUMBER field
1169 * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1170 * field.
1171 *
1172 * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1173 * 0: no relative lifetime.
1174 * 1: shortest relative lifetime.
1175 * 2: second shortest relative lifetime.
1176 * 3 - 0x3d: intermediate relative lifetimes.
1177 * 0x3e: second longest relative lifetime.
1178 * 0x3f: longest relative lifetime.
1179 */
1180static u8 sd_group_number(struct scsi_cmnd *cmd)
1181{
1182 const struct request *rq = scsi_cmd_to_rq(cmd);
1183 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1184
1185 if (!sdkp->rscs)
1186 return 0;
1187
1188 return min3((u32)rq->bio->bi_write_hint,
1189 (u32)sdkp->permanent_stream_count, 0x3fu);
1190}
1191
1192static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1193 sector_t lba, unsigned int nr_blocks,
1194 unsigned char flags, unsigned int dld)
1195{
1196 cmd->cmd_len = SD_EXT_CDB_SIZE;
1197 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1198 cmd->cmnd[6] = sd_group_number(cmd);
1199 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1200 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1201 cmd->cmnd[10] = flags;
1202 cmd->cmnd[11] = dld & 0x07;
1203 put_unaligned_be64(lba, &cmd->cmnd[12]);
1204 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1205 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1206
1207 return BLK_STS_OK;
1208}
1209
1210static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1211 sector_t lba, unsigned int nr_blocks,
1212 unsigned char flags, unsigned int dld)
1213{
1214 cmd->cmd_len = 16;
1215 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1216 cmd->cmnd[1] = flags | ((dld >> 2) & 0x01);
1217 cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1218 cmd->cmnd[15] = 0;
1219 put_unaligned_be64(lba, &cmd->cmnd[2]);
1220 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1221
1222 return BLK_STS_OK;
1223}
1224
1225static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1226 sector_t lba, unsigned int nr_blocks,
1227 unsigned char flags)
1228{
1229 cmd->cmd_len = 10;
1230 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1231 cmd->cmnd[1] = flags;
1232 cmd->cmnd[6] = sd_group_number(cmd);
1233 cmd->cmnd[9] = 0;
1234 put_unaligned_be32(lba, &cmd->cmnd[2]);
1235 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1236
1237 return BLK_STS_OK;
1238}
1239
1240static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1241 sector_t lba, unsigned int nr_blocks,
1242 unsigned char flags)
1243{
1244 /* Avoid that 0 blocks gets translated into 256 blocks. */
1245 if (WARN_ON_ONCE(nr_blocks == 0))
1246 return BLK_STS_IOERR;
1247
1248 if (unlikely(flags & 0x8)) {
1249 /*
1250 * This happens only if this drive failed 10byte rw
1251 * command with ILLEGAL_REQUEST during operation and
1252 * thus turned off use_10_for_rw.
1253 */
1254 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1255 return BLK_STS_IOERR;
1256 }
1257
1258 cmd->cmd_len = 6;
1259 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1260 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1261 cmd->cmnd[2] = (lba >> 8) & 0xff;
1262 cmd->cmnd[3] = lba & 0xff;
1263 cmd->cmnd[4] = nr_blocks;
1264 cmd->cmnd[5] = 0;
1265
1266 return BLK_STS_OK;
1267}
1268
1269/*
1270 * Check if a command has a duration limit set. If it does, and the target
1271 * device supports CDL and the feature is enabled, return the limit
1272 * descriptor index to use. Return 0 (no limit) otherwise.
1273 */
1274static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1275{
1276 struct scsi_device *sdp = sdkp->device;
1277 int hint;
1278
1279 if (!sdp->cdl_supported || !sdp->cdl_enable)
1280 return 0;
1281
1282 /*
1283 * Use "no limit" if the request ioprio does not specify a duration
1284 * limit hint.
1285 */
1286 hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1287 if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1288 hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1289 return 0;
1290
1291 return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1292}
1293
1294static blk_status_t sd_setup_atomic_cmnd(struct scsi_cmnd *cmd,
1295 sector_t lba, unsigned int nr_blocks,
1296 bool boundary, unsigned char flags)
1297{
1298 cmd->cmd_len = 16;
1299 cmd->cmnd[0] = WRITE_ATOMIC_16;
1300 cmd->cmnd[1] = flags;
1301 put_unaligned_be64(lba, &cmd->cmnd[2]);
1302 put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1303 if (boundary)
1304 put_unaligned_be16(nr_blocks, &cmd->cmnd[10]);
1305 else
1306 put_unaligned_be16(0, &cmd->cmnd[10]);
1307 put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1308 cmd->cmnd[14] = 0;
1309 cmd->cmnd[15] = 0;
1310
1311 return BLK_STS_OK;
1312}
1313
1314static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1315{
1316 struct request *rq = scsi_cmd_to_rq(cmd);
1317 struct scsi_device *sdp = cmd->device;
1318 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1319 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1320 sector_t threshold;
1321 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1322 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1323 bool write = rq_data_dir(rq) == WRITE;
1324 unsigned char protect, fua;
1325 unsigned int dld;
1326 blk_status_t ret;
1327 unsigned int dif;
1328 bool dix;
1329
1330 ret = scsi_alloc_sgtables(cmd);
1331 if (ret != BLK_STS_OK)
1332 return ret;
1333
1334 ret = BLK_STS_IOERR;
1335 if (!scsi_device_online(sdp) || sdp->changed) {
1336 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1337 goto fail;
1338 }
1339
1340 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1341 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1342 goto fail;
1343 }
1344
1345 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1346 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1347 goto fail;
1348 }
1349
1350 /*
1351 * Some SD card readers can't handle accesses which touch the
1352 * last one or two logical blocks. Split accesses as needed.
1353 */
1354 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1355
1356 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1357 if (lba < threshold) {
1358 /* Access up to the threshold but not beyond */
1359 nr_blocks = threshold - lba;
1360 } else {
1361 /* Access only a single logical block */
1362 nr_blocks = 1;
1363 }
1364 }
1365
1366 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1367 dix = scsi_prot_sg_count(cmd);
1368 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1369 dld = sd_cdl_dld(sdkp, cmd);
1370
1371 if (dif || dix)
1372 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1373 else
1374 protect = 0;
1375
1376 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1377 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1378 protect | fua, dld);
1379 } else if (rq->cmd_flags & REQ_ATOMIC) {
1380 ret = sd_setup_atomic_cmnd(cmd, lba, nr_blocks,
1381 sdkp->use_atomic_write_boundary,
1382 protect | fua);
1383 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1384 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1385 protect | fua, dld);
1386 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1387 sdp->use_10_for_rw || protect || rq->bio->bi_write_hint) {
1388 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1389 protect | fua);
1390 } else {
1391 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1392 protect | fua);
1393 }
1394
1395 if (unlikely(ret != BLK_STS_OK))
1396 goto fail;
1397
1398 /*
1399 * We shouldn't disconnect in the middle of a sector, so with a dumb
1400 * host adapter, it's safe to assume that we can at least transfer
1401 * this many bytes between each connect / disconnect.
1402 */
1403 cmd->transfersize = sdp->sector_size;
1404 cmd->underflow = nr_blocks << 9;
1405 cmd->allowed = sdkp->max_retries;
1406 cmd->sdb.length = nr_blocks * sdp->sector_size;
1407
1408 SCSI_LOG_HLQUEUE(1,
1409 scmd_printk(KERN_INFO, cmd,
1410 "%s: block=%llu, count=%d\n", __func__,
1411 (unsigned long long)blk_rq_pos(rq),
1412 blk_rq_sectors(rq)));
1413 SCSI_LOG_HLQUEUE(2,
1414 scmd_printk(KERN_INFO, cmd,
1415 "%s %d/%u 512 byte blocks.\n",
1416 write ? "writing" : "reading", nr_blocks,
1417 blk_rq_sectors(rq)));
1418
1419 /*
1420 * This indicates that the command is ready from our end to be queued.
1421 */
1422 return BLK_STS_OK;
1423fail:
1424 scsi_free_sgtables(cmd);
1425 return ret;
1426}
1427
1428static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1429{
1430 struct request *rq = scsi_cmd_to_rq(cmd);
1431
1432 switch (req_op(rq)) {
1433 case REQ_OP_DISCARD:
1434 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1435 case SD_LBP_UNMAP:
1436 return sd_setup_unmap_cmnd(cmd);
1437 case SD_LBP_WS16:
1438 return sd_setup_write_same16_cmnd(cmd, true);
1439 case SD_LBP_WS10:
1440 return sd_setup_write_same10_cmnd(cmd, true);
1441 case SD_LBP_ZERO:
1442 return sd_setup_write_same10_cmnd(cmd, false);
1443 default:
1444 return BLK_STS_TARGET;
1445 }
1446 case REQ_OP_WRITE_ZEROES:
1447 return sd_setup_write_zeroes_cmnd(cmd);
1448 case REQ_OP_FLUSH:
1449 return sd_setup_flush_cmnd(cmd);
1450 case REQ_OP_READ:
1451 case REQ_OP_WRITE:
1452 return sd_setup_read_write_cmnd(cmd);
1453 case REQ_OP_ZONE_RESET:
1454 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1455 false);
1456 case REQ_OP_ZONE_RESET_ALL:
1457 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1458 true);
1459 case REQ_OP_ZONE_OPEN:
1460 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1461 case REQ_OP_ZONE_CLOSE:
1462 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1463 case REQ_OP_ZONE_FINISH:
1464 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1465 default:
1466 WARN_ON_ONCE(1);
1467 return BLK_STS_NOTSUPP;
1468 }
1469}
1470
1471static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1472{
1473 struct request *rq = scsi_cmd_to_rq(SCpnt);
1474
1475 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1476 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1477}
1478
1479static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1480{
1481 if (sdkp->device->removable || sdkp->write_prot) {
1482 if (disk_check_media_change(disk))
1483 return true;
1484 }
1485
1486 /*
1487 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1488 * nothing to do with partitions, BLKRRPART is used to force a full
1489 * revalidate after things like a format for historical reasons.
1490 */
1491 return test_bit(GD_NEED_PART_SCAN, &disk->state);
1492}
1493
1494/**
1495 * sd_open - open a scsi disk device
1496 * @disk: disk to open
1497 * @mode: open mode
1498 *
1499 * Returns 0 if successful. Returns a negated errno value in case
1500 * of error.
1501 *
1502 * Note: This can be called from a user context (e.g. fsck(1) )
1503 * or from within the kernel (e.g. as a result of a mount(1) ).
1504 * In the latter case @inode and @filp carry an abridged amount
1505 * of information as noted above.
1506 *
1507 * Locking: called with disk->open_mutex held.
1508 **/
1509static int sd_open(struct gendisk *disk, blk_mode_t mode)
1510{
1511 struct scsi_disk *sdkp = scsi_disk(disk);
1512 struct scsi_device *sdev = sdkp->device;
1513 int retval;
1514
1515 if (scsi_device_get(sdev))
1516 return -ENXIO;
1517
1518 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1519
1520 /*
1521 * If the device is in error recovery, wait until it is done.
1522 * If the device is offline, then disallow any access to it.
1523 */
1524 retval = -ENXIO;
1525 if (!scsi_block_when_processing_errors(sdev))
1526 goto error_out;
1527
1528 if (sd_need_revalidate(disk, sdkp))
1529 sd_revalidate_disk(disk);
1530
1531 /*
1532 * If the drive is empty, just let the open fail.
1533 */
1534 retval = -ENOMEDIUM;
1535 if (sdev->removable && !sdkp->media_present &&
1536 !(mode & BLK_OPEN_NDELAY))
1537 goto error_out;
1538
1539 /*
1540 * If the device has the write protect tab set, have the open fail
1541 * if the user expects to be able to write to the thing.
1542 */
1543 retval = -EROFS;
1544 if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1545 goto error_out;
1546
1547 /*
1548 * It is possible that the disk changing stuff resulted in
1549 * the device being taken offline. If this is the case,
1550 * report this to the user, and don't pretend that the
1551 * open actually succeeded.
1552 */
1553 retval = -ENXIO;
1554 if (!scsi_device_online(sdev))
1555 goto error_out;
1556
1557 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1558 if (scsi_block_when_processing_errors(sdev))
1559 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1560 }
1561
1562 return 0;
1563
1564error_out:
1565 scsi_device_put(sdev);
1566 return retval;
1567}
1568
1569/**
1570 * sd_release - invoked when the (last) close(2) is called on this
1571 * scsi disk.
1572 * @disk: disk to release
1573 *
1574 * Returns 0.
1575 *
1576 * Note: may block (uninterruptible) if error recovery is underway
1577 * on this disk.
1578 *
1579 * Locking: called with disk->open_mutex held.
1580 **/
1581static void sd_release(struct gendisk *disk)
1582{
1583 struct scsi_disk *sdkp = scsi_disk(disk);
1584 struct scsi_device *sdev = sdkp->device;
1585
1586 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1587
1588 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1589 if (scsi_block_when_processing_errors(sdev))
1590 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1591 }
1592
1593 scsi_device_put(sdev);
1594}
1595
1596static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1597{
1598 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1599 struct scsi_device *sdp = sdkp->device;
1600 struct Scsi_Host *host = sdp->host;
1601 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1602 int diskinfo[4];
1603
1604 /* default to most commonly used values */
1605 diskinfo[0] = 0x40; /* 1 << 6 */
1606 diskinfo[1] = 0x20; /* 1 << 5 */
1607 diskinfo[2] = capacity >> 11;
1608
1609 /* override with calculated, extended default, or driver values */
1610 if (host->hostt->bios_param)
1611 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1612 else
1613 scsicam_bios_param(bdev, capacity, diskinfo);
1614
1615 geo->heads = diskinfo[0];
1616 geo->sectors = diskinfo[1];
1617 geo->cylinders = diskinfo[2];
1618 return 0;
1619}
1620
1621/**
1622 * sd_ioctl - process an ioctl
1623 * @bdev: target block device
1624 * @mode: open mode
1625 * @cmd: ioctl command number
1626 * @arg: this is third argument given to ioctl(2) system call.
1627 * Often contains a pointer.
1628 *
1629 * Returns 0 if successful (some ioctls return positive numbers on
1630 * success as well). Returns a negated errno value in case of error.
1631 *
1632 * Note: most ioctls are forward onto the block subsystem or further
1633 * down in the scsi subsystem.
1634 **/
1635static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1636 unsigned int cmd, unsigned long arg)
1637{
1638 struct gendisk *disk = bdev->bd_disk;
1639 struct scsi_disk *sdkp = scsi_disk(disk);
1640 struct scsi_device *sdp = sdkp->device;
1641 void __user *p = (void __user *)arg;
1642 int error;
1643
1644 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1645 "cmd=0x%x\n", disk->disk_name, cmd));
1646
1647 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1648 return -ENOIOCTLCMD;
1649
1650 /*
1651 * If we are in the middle of error recovery, don't let anyone
1652 * else try and use this device. Also, if error recovery fails, it
1653 * may try and take the device offline, in which case all further
1654 * access to the device is prohibited.
1655 */
1656 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1657 (mode & BLK_OPEN_NDELAY));
1658 if (error)
1659 return error;
1660
1661 if (is_sed_ioctl(cmd))
1662 return sed_ioctl(sdkp->opal_dev, cmd, p);
1663 return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1664}
1665
1666static void set_media_not_present(struct scsi_disk *sdkp)
1667{
1668 if (sdkp->media_present)
1669 sdkp->device->changed = 1;
1670
1671 if (sdkp->device->removable) {
1672 sdkp->media_present = 0;
1673 sdkp->capacity = 0;
1674 }
1675}
1676
1677static int media_not_present(struct scsi_disk *sdkp,
1678 struct scsi_sense_hdr *sshdr)
1679{
1680 if (!scsi_sense_valid(sshdr))
1681 return 0;
1682
1683 /* not invoked for commands that could return deferred errors */
1684 switch (sshdr->sense_key) {
1685 case UNIT_ATTENTION:
1686 case NOT_READY:
1687 /* medium not present */
1688 if (sshdr->asc == 0x3A) {
1689 set_media_not_present(sdkp);
1690 return 1;
1691 }
1692 }
1693 return 0;
1694}
1695
1696/**
1697 * sd_check_events - check media events
1698 * @disk: kernel device descriptor
1699 * @clearing: disk events currently being cleared
1700 *
1701 * Returns mask of DISK_EVENT_*.
1702 *
1703 * Note: this function is invoked from the block subsystem.
1704 **/
1705static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1706{
1707 struct scsi_disk *sdkp = disk->private_data;
1708 struct scsi_device *sdp;
1709 int retval;
1710 bool disk_changed;
1711
1712 if (!sdkp)
1713 return 0;
1714
1715 sdp = sdkp->device;
1716 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1717
1718 /*
1719 * If the device is offline, don't send any commands - just pretend as
1720 * if the command failed. If the device ever comes back online, we
1721 * can deal with it then. It is only because of unrecoverable errors
1722 * that we would ever take a device offline in the first place.
1723 */
1724 if (!scsi_device_online(sdp)) {
1725 set_media_not_present(sdkp);
1726 goto out;
1727 }
1728
1729 /*
1730 * Using TEST_UNIT_READY enables differentiation between drive with
1731 * no cartridge loaded - NOT READY, drive with changed cartridge -
1732 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1733 *
1734 * Drives that auto spin down. eg iomega jaz 1G, will be started
1735 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1736 * sd_revalidate() is called.
1737 */
1738 if (scsi_block_when_processing_errors(sdp)) {
1739 struct scsi_sense_hdr sshdr = { 0, };
1740
1741 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1742 &sshdr);
1743
1744 /* failed to execute TUR, assume media not present */
1745 if (retval < 0 || host_byte(retval)) {
1746 set_media_not_present(sdkp);
1747 goto out;
1748 }
1749
1750 if (media_not_present(sdkp, &sshdr))
1751 goto out;
1752 }
1753
1754 /*
1755 * For removable scsi disk we have to recognise the presence
1756 * of a disk in the drive.
1757 */
1758 if (!sdkp->media_present)
1759 sdp->changed = 1;
1760 sdkp->media_present = 1;
1761out:
1762 /*
1763 * sdp->changed is set under the following conditions:
1764 *
1765 * Medium present state has changed in either direction.
1766 * Device has indicated UNIT_ATTENTION.
1767 */
1768 disk_changed = sdp->changed;
1769 sdp->changed = 0;
1770 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1771}
1772
1773static int sd_sync_cache(struct scsi_disk *sdkp)
1774{
1775 int res;
1776 struct scsi_device *sdp = sdkp->device;
1777 const int timeout = sdp->request_queue->rq_timeout
1778 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1779 /* Leave the rest of the command zero to indicate flush everything. */
1780 const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1781 SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1782 struct scsi_sense_hdr sshdr;
1783 struct scsi_failure failure_defs[] = {
1784 {
1785 .allowed = 3,
1786 .result = SCMD_FAILURE_RESULT_ANY,
1787 },
1788 {}
1789 };
1790 struct scsi_failures failures = {
1791 .failure_definitions = failure_defs,
1792 };
1793 const struct scsi_exec_args exec_args = {
1794 .req_flags = BLK_MQ_REQ_PM,
1795 .sshdr = &sshdr,
1796 .failures = &failures,
1797 };
1798
1799 if (!scsi_device_online(sdp))
1800 return -ENODEV;
1801
1802 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1803 sdkp->max_retries, &exec_args);
1804 if (res) {
1805 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1806
1807 if (res < 0)
1808 return res;
1809
1810 if (scsi_status_is_check_condition(res) &&
1811 scsi_sense_valid(&sshdr)) {
1812 sd_print_sense_hdr(sdkp, &sshdr);
1813
1814 /* we need to evaluate the error return */
1815 if (sshdr.asc == 0x3a || /* medium not present */
1816 sshdr.asc == 0x20 || /* invalid command */
1817 (sshdr.asc == 0x74 && sshdr.ascq == 0x71)) /* drive is password locked */
1818 /* this is no error here */
1819 return 0;
1820
1821 /*
1822 * If a format is in progress or if the drive does not
1823 * support sync, there is not much we can do because
1824 * this is called during shutdown or suspend so just
1825 * return success so those operations can proceed.
1826 */
1827 if ((sshdr.asc == 0x04 && sshdr.ascq == 0x04) ||
1828 sshdr.sense_key == ILLEGAL_REQUEST)
1829 return 0;
1830 }
1831
1832 switch (host_byte(res)) {
1833 /* ignore errors due to racing a disconnection */
1834 case DID_BAD_TARGET:
1835 case DID_NO_CONNECT:
1836 return 0;
1837 /* signal the upper layer it might try again */
1838 case DID_BUS_BUSY:
1839 case DID_IMM_RETRY:
1840 case DID_REQUEUE:
1841 case DID_SOFT_ERROR:
1842 return -EBUSY;
1843 default:
1844 return -EIO;
1845 }
1846 }
1847 return 0;
1848}
1849
1850static void sd_rescan(struct device *dev)
1851{
1852 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1853
1854 sd_revalidate_disk(sdkp->disk);
1855}
1856
1857static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1858 enum blk_unique_id type)
1859{
1860 struct scsi_device *sdev = scsi_disk(disk)->device;
1861 const struct scsi_vpd *vpd;
1862 const unsigned char *d;
1863 int ret = -ENXIO, len;
1864
1865 rcu_read_lock();
1866 vpd = rcu_dereference(sdev->vpd_pg83);
1867 if (!vpd)
1868 goto out_unlock;
1869
1870 ret = -EINVAL;
1871 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1872 /* we only care about designators with LU association */
1873 if (((d[1] >> 4) & 0x3) != 0x00)
1874 continue;
1875 if ((d[1] & 0xf) != type)
1876 continue;
1877
1878 /*
1879 * Only exit early if a 16-byte descriptor was found. Otherwise
1880 * keep looking as one with more entropy might still show up.
1881 */
1882 len = d[3];
1883 if (len != 8 && len != 12 && len != 16)
1884 continue;
1885 ret = len;
1886 memcpy(id, d + 4, len);
1887 if (len == 16)
1888 break;
1889 }
1890out_unlock:
1891 rcu_read_unlock();
1892 return ret;
1893}
1894
1895static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1896{
1897 switch (host_byte(result)) {
1898 case DID_TRANSPORT_MARGINAL:
1899 case DID_TRANSPORT_DISRUPTED:
1900 case DID_BUS_BUSY:
1901 return PR_STS_RETRY_PATH_FAILURE;
1902 case DID_NO_CONNECT:
1903 return PR_STS_PATH_FAILED;
1904 case DID_TRANSPORT_FAILFAST:
1905 return PR_STS_PATH_FAST_FAILED;
1906 }
1907
1908 switch (status_byte(result)) {
1909 case SAM_STAT_RESERVATION_CONFLICT:
1910 return PR_STS_RESERVATION_CONFLICT;
1911 case SAM_STAT_CHECK_CONDITION:
1912 if (!scsi_sense_valid(sshdr))
1913 return PR_STS_IOERR;
1914
1915 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1916 (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1917 return -EINVAL;
1918
1919 fallthrough;
1920 default:
1921 return PR_STS_IOERR;
1922 }
1923}
1924
1925static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1926 unsigned char *data, int data_len)
1927{
1928 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1929 struct scsi_device *sdev = sdkp->device;
1930 struct scsi_sense_hdr sshdr;
1931 u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1932 struct scsi_failure failure_defs[] = {
1933 {
1934 .sense = UNIT_ATTENTION,
1935 .asc = SCMD_FAILURE_ASC_ANY,
1936 .ascq = SCMD_FAILURE_ASCQ_ANY,
1937 .allowed = 5,
1938 .result = SAM_STAT_CHECK_CONDITION,
1939 },
1940 {}
1941 };
1942 struct scsi_failures failures = {
1943 .failure_definitions = failure_defs,
1944 };
1945 const struct scsi_exec_args exec_args = {
1946 .sshdr = &sshdr,
1947 .failures = &failures,
1948 };
1949 int result;
1950
1951 put_unaligned_be16(data_len, &cmd[7]);
1952
1953 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1954 SD_TIMEOUT, sdkp->max_retries, &exec_args);
1955 if (scsi_status_is_check_condition(result) &&
1956 scsi_sense_valid(&sshdr)) {
1957 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1958 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1959 }
1960
1961 if (result <= 0)
1962 return result;
1963
1964 return sd_scsi_to_pr_err(&sshdr, result);
1965}
1966
1967static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1968{
1969 int result, i, data_offset, num_copy_keys;
1970 u32 num_keys = keys_info->num_keys;
1971 int data_len = num_keys * 8 + 8;
1972 u8 *data;
1973
1974 data = kzalloc(data_len, GFP_KERNEL);
1975 if (!data)
1976 return -ENOMEM;
1977
1978 result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1979 if (result)
1980 goto free_data;
1981
1982 keys_info->generation = get_unaligned_be32(&data[0]);
1983 keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1984
1985 data_offset = 8;
1986 num_copy_keys = min(num_keys, keys_info->num_keys);
1987
1988 for (i = 0; i < num_copy_keys; i++) {
1989 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1990 data_offset += 8;
1991 }
1992
1993free_data:
1994 kfree(data);
1995 return result;
1996}
1997
1998static int sd_pr_read_reservation(struct block_device *bdev,
1999 struct pr_held_reservation *rsv)
2000{
2001 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2002 struct scsi_device *sdev = sdkp->device;
2003 u8 data[24] = { };
2004 int result, len;
2005
2006 result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
2007 if (result)
2008 return result;
2009
2010 len = get_unaligned_be32(&data[4]);
2011 if (!len)
2012 return 0;
2013
2014 /* Make sure we have at least the key and type */
2015 if (len < 14) {
2016 sdev_printk(KERN_INFO, sdev,
2017 "READ RESERVATION failed due to short return buffer of %d bytes\n",
2018 len);
2019 return -EINVAL;
2020 }
2021
2022 rsv->generation = get_unaligned_be32(&data[0]);
2023 rsv->key = get_unaligned_be64(&data[8]);
2024 rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
2025 return 0;
2026}
2027
2028static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
2029 u64 sa_key, enum scsi_pr_type type, u8 flags)
2030{
2031 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2032 struct scsi_device *sdev = sdkp->device;
2033 struct scsi_sense_hdr sshdr;
2034 struct scsi_failure failure_defs[] = {
2035 {
2036 .sense = UNIT_ATTENTION,
2037 .asc = SCMD_FAILURE_ASC_ANY,
2038 .ascq = SCMD_FAILURE_ASCQ_ANY,
2039 .allowed = 5,
2040 .result = SAM_STAT_CHECK_CONDITION,
2041 },
2042 {}
2043 };
2044 struct scsi_failures failures = {
2045 .failure_definitions = failure_defs,
2046 };
2047 const struct scsi_exec_args exec_args = {
2048 .sshdr = &sshdr,
2049 .failures = &failures,
2050 };
2051 int result;
2052 u8 cmd[16] = { 0, };
2053 u8 data[24] = { 0, };
2054
2055 cmd[0] = PERSISTENT_RESERVE_OUT;
2056 cmd[1] = sa;
2057 cmd[2] = type;
2058 put_unaligned_be32(sizeof(data), &cmd[5]);
2059
2060 put_unaligned_be64(key, &data[0]);
2061 put_unaligned_be64(sa_key, &data[8]);
2062 data[20] = flags;
2063
2064 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
2065 sizeof(data), SD_TIMEOUT, sdkp->max_retries,
2066 &exec_args);
2067
2068 if (scsi_status_is_check_condition(result) &&
2069 scsi_sense_valid(&sshdr)) {
2070 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
2071 scsi_print_sense_hdr(sdev, NULL, &sshdr);
2072 }
2073
2074 if (result <= 0)
2075 return result;
2076
2077 return sd_scsi_to_pr_err(&sshdr, result);
2078}
2079
2080static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2081 u32 flags)
2082{
2083 if (flags & ~PR_FL_IGNORE_KEY)
2084 return -EOPNOTSUPP;
2085 return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
2086 old_key, new_key, 0,
2087 (1 << 0) /* APTPL */);
2088}
2089
2090static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2091 u32 flags)
2092{
2093 if (flags)
2094 return -EOPNOTSUPP;
2095 return sd_pr_out_command(bdev, 0x01, key, 0,
2096 block_pr_type_to_scsi(type), 0);
2097}
2098
2099static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2100{
2101 return sd_pr_out_command(bdev, 0x02, key, 0,
2102 block_pr_type_to_scsi(type), 0);
2103}
2104
2105static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2106 enum pr_type type, bool abort)
2107{
2108 return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2109 block_pr_type_to_scsi(type), 0);
2110}
2111
2112static int sd_pr_clear(struct block_device *bdev, u64 key)
2113{
2114 return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2115}
2116
2117static const struct pr_ops sd_pr_ops = {
2118 .pr_register = sd_pr_register,
2119 .pr_reserve = sd_pr_reserve,
2120 .pr_release = sd_pr_release,
2121 .pr_preempt = sd_pr_preempt,
2122 .pr_clear = sd_pr_clear,
2123 .pr_read_keys = sd_pr_read_keys,
2124 .pr_read_reservation = sd_pr_read_reservation,
2125};
2126
2127static void scsi_disk_free_disk(struct gendisk *disk)
2128{
2129 struct scsi_disk *sdkp = scsi_disk(disk);
2130
2131 put_device(&sdkp->disk_dev);
2132}
2133
2134static const struct block_device_operations sd_fops = {
2135 .owner = THIS_MODULE,
2136 .open = sd_open,
2137 .release = sd_release,
2138 .ioctl = sd_ioctl,
2139 .getgeo = sd_getgeo,
2140 .compat_ioctl = blkdev_compat_ptr_ioctl,
2141 .check_events = sd_check_events,
2142 .unlock_native_capacity = sd_unlock_native_capacity,
2143 .report_zones = sd_zbc_report_zones,
2144 .get_unique_id = sd_get_unique_id,
2145 .free_disk = scsi_disk_free_disk,
2146 .pr_ops = &sd_pr_ops,
2147};
2148
2149/**
2150 * sd_eh_reset - reset error handling callback
2151 * @scmd: sd-issued command that has failed
2152 *
2153 * This function is called by the SCSI midlayer before starting
2154 * SCSI EH. When counting medium access failures we have to be
2155 * careful to register it only only once per device and SCSI EH run;
2156 * there might be several timed out commands which will cause the
2157 * 'max_medium_access_timeouts' counter to trigger after the first
2158 * SCSI EH run already and set the device to offline.
2159 * So this function resets the internal counter before starting SCSI EH.
2160 **/
2161static void sd_eh_reset(struct scsi_cmnd *scmd)
2162{
2163 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2164
2165 /* New SCSI EH run, reset gate variable */
2166 sdkp->ignore_medium_access_errors = false;
2167}
2168
2169/**
2170 * sd_eh_action - error handling callback
2171 * @scmd: sd-issued command that has failed
2172 * @eh_disp: The recovery disposition suggested by the midlayer
2173 *
2174 * This function is called by the SCSI midlayer upon completion of an
2175 * error test command (currently TEST UNIT READY). The result of sending
2176 * the eh command is passed in eh_disp. We're looking for devices that
2177 * fail medium access commands but are OK with non access commands like
2178 * test unit ready (so wrongly see the device as having a successful
2179 * recovery)
2180 **/
2181static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2182{
2183 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2184 struct scsi_device *sdev = scmd->device;
2185
2186 if (!scsi_device_online(sdev) ||
2187 !scsi_medium_access_command(scmd) ||
2188 host_byte(scmd->result) != DID_TIME_OUT ||
2189 eh_disp != SUCCESS)
2190 return eh_disp;
2191
2192 /*
2193 * The device has timed out executing a medium access command.
2194 * However, the TEST UNIT READY command sent during error
2195 * handling completed successfully. Either the device is in the
2196 * process of recovering or has it suffered an internal failure
2197 * that prevents access to the storage medium.
2198 */
2199 if (!sdkp->ignore_medium_access_errors) {
2200 sdkp->medium_access_timed_out++;
2201 sdkp->ignore_medium_access_errors = true;
2202 }
2203
2204 /*
2205 * If the device keeps failing read/write commands but TEST UNIT
2206 * READY always completes successfully we assume that medium
2207 * access is no longer possible and take the device offline.
2208 */
2209 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2210 scmd_printk(KERN_ERR, scmd,
2211 "Medium access timeout failure. Offlining disk!\n");
2212 mutex_lock(&sdev->state_mutex);
2213 scsi_device_set_state(sdev, SDEV_OFFLINE);
2214 mutex_unlock(&sdev->state_mutex);
2215
2216 return SUCCESS;
2217 }
2218
2219 return eh_disp;
2220}
2221
2222static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2223{
2224 struct request *req = scsi_cmd_to_rq(scmd);
2225 struct scsi_device *sdev = scmd->device;
2226 unsigned int transferred, good_bytes;
2227 u64 start_lba, end_lba, bad_lba;
2228
2229 /*
2230 * Some commands have a payload smaller than the device logical
2231 * block size (e.g. INQUIRY on a 4K disk).
2232 */
2233 if (scsi_bufflen(scmd) <= sdev->sector_size)
2234 return 0;
2235
2236 /* Check if we have a 'bad_lba' information */
2237 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2238 SCSI_SENSE_BUFFERSIZE,
2239 &bad_lba))
2240 return 0;
2241
2242 /*
2243 * If the bad lba was reported incorrectly, we have no idea where
2244 * the error is.
2245 */
2246 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2247 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2248 if (bad_lba < start_lba || bad_lba >= end_lba)
2249 return 0;
2250
2251 /*
2252 * resid is optional but mostly filled in. When it's unused,
2253 * its value is zero, so we assume the whole buffer transferred
2254 */
2255 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2256
2257 /* This computation should always be done in terms of the
2258 * resolution of the device's medium.
2259 */
2260 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2261
2262 return min(good_bytes, transferred);
2263}
2264
2265/**
2266 * sd_done - bottom half handler: called when the lower level
2267 * driver has completed (successfully or otherwise) a scsi command.
2268 * @SCpnt: mid-level's per command structure.
2269 *
2270 * Note: potentially run from within an ISR. Must not block.
2271 **/
2272static int sd_done(struct scsi_cmnd *SCpnt)
2273{
2274 int result = SCpnt->result;
2275 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2276 unsigned int sector_size = SCpnt->device->sector_size;
2277 unsigned int resid;
2278 struct scsi_sense_hdr sshdr;
2279 struct request *req = scsi_cmd_to_rq(SCpnt);
2280 struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2281 int sense_valid = 0;
2282 int sense_deferred = 0;
2283
2284 switch (req_op(req)) {
2285 case REQ_OP_DISCARD:
2286 case REQ_OP_WRITE_ZEROES:
2287 case REQ_OP_ZONE_RESET:
2288 case REQ_OP_ZONE_RESET_ALL:
2289 case REQ_OP_ZONE_OPEN:
2290 case REQ_OP_ZONE_CLOSE:
2291 case REQ_OP_ZONE_FINISH:
2292 if (!result) {
2293 good_bytes = blk_rq_bytes(req);
2294 scsi_set_resid(SCpnt, 0);
2295 } else {
2296 good_bytes = 0;
2297 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2298 }
2299 break;
2300 default:
2301 /*
2302 * In case of bogus fw or device, we could end up having
2303 * an unaligned partial completion. Check this here and force
2304 * alignment.
2305 */
2306 resid = scsi_get_resid(SCpnt);
2307 if (resid & (sector_size - 1)) {
2308 sd_printk(KERN_INFO, sdkp,
2309 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2310 resid, sector_size);
2311 scsi_print_command(SCpnt);
2312 resid = min(scsi_bufflen(SCpnt),
2313 round_up(resid, sector_size));
2314 scsi_set_resid(SCpnt, resid);
2315 }
2316 }
2317
2318 if (result) {
2319 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2320 if (sense_valid)
2321 sense_deferred = scsi_sense_is_deferred(&sshdr);
2322 }
2323 sdkp->medium_access_timed_out = 0;
2324
2325 if (!scsi_status_is_check_condition(result) &&
2326 (!sense_valid || sense_deferred))
2327 goto out;
2328
2329 switch (sshdr.sense_key) {
2330 case HARDWARE_ERROR:
2331 case MEDIUM_ERROR:
2332 good_bytes = sd_completed_bytes(SCpnt);
2333 break;
2334 case RECOVERED_ERROR:
2335 good_bytes = scsi_bufflen(SCpnt);
2336 break;
2337 case NO_SENSE:
2338 /* This indicates a false check condition, so ignore it. An
2339 * unknown amount of data was transferred so treat it as an
2340 * error.
2341 */
2342 SCpnt->result = 0;
2343 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2344 break;
2345 case ABORTED_COMMAND:
2346 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2347 good_bytes = sd_completed_bytes(SCpnt);
2348 break;
2349 case ILLEGAL_REQUEST:
2350 switch (sshdr.asc) {
2351 case 0x10: /* DIX: Host detected corruption */
2352 good_bytes = sd_completed_bytes(SCpnt);
2353 break;
2354 case 0x20: /* INVALID COMMAND OPCODE */
2355 case 0x24: /* INVALID FIELD IN CDB */
2356 switch (SCpnt->cmnd[0]) {
2357 case UNMAP:
2358 sd_disable_discard(sdkp);
2359 break;
2360 case WRITE_SAME_16:
2361 case WRITE_SAME:
2362 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2363 sd_disable_discard(sdkp);
2364 } else {
2365 sd_disable_write_same(sdkp);
2366 req->rq_flags |= RQF_QUIET;
2367 }
2368 break;
2369 }
2370 }
2371 break;
2372 default:
2373 break;
2374 }
2375
2376 out:
2377 if (sdkp->device->type == TYPE_ZBC)
2378 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2379
2380 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2381 "sd_done: completed %d of %d bytes\n",
2382 good_bytes, scsi_bufflen(SCpnt)));
2383
2384 return good_bytes;
2385}
2386
2387/*
2388 * spinup disk - called only in sd_revalidate_disk()
2389 */
2390static void
2391sd_spinup_disk(struct scsi_disk *sdkp)
2392{
2393 static const u8 cmd[10] = { TEST_UNIT_READY };
2394 unsigned long spintime_expire = 0;
2395 int spintime, sense_valid = 0;
2396 unsigned int the_result;
2397 struct scsi_sense_hdr sshdr;
2398 struct scsi_failure failure_defs[] = {
2399 /* Do not retry Medium Not Present */
2400 {
2401 .sense = UNIT_ATTENTION,
2402 .asc = 0x3A,
2403 .ascq = SCMD_FAILURE_ASCQ_ANY,
2404 .result = SAM_STAT_CHECK_CONDITION,
2405 },
2406 {
2407 .sense = NOT_READY,
2408 .asc = 0x3A,
2409 .ascq = SCMD_FAILURE_ASCQ_ANY,
2410 .result = SAM_STAT_CHECK_CONDITION,
2411 },
2412 /* Retry when scsi_status_is_good would return false 3 times */
2413 {
2414 .result = SCMD_FAILURE_STAT_ANY,
2415 .allowed = 3,
2416 },
2417 {}
2418 };
2419 struct scsi_failures failures = {
2420 .failure_definitions = failure_defs,
2421 };
2422 const struct scsi_exec_args exec_args = {
2423 .sshdr = &sshdr,
2424 .failures = &failures,
2425 };
2426
2427 spintime = 0;
2428
2429 /* Spin up drives, as required. Only do this at boot time */
2430 /* Spinup needs to be done for module loads too. */
2431 do {
2432 bool media_was_present = sdkp->media_present;
2433
2434 scsi_failures_reset_retries(&failures);
2435
2436 the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2437 NULL, 0, SD_TIMEOUT,
2438 sdkp->max_retries, &exec_args);
2439
2440
2441 if (the_result > 0) {
2442 /*
2443 * If the drive has indicated to us that it doesn't
2444 * have any media in it, don't bother with any more
2445 * polling.
2446 */
2447 if (media_not_present(sdkp, &sshdr)) {
2448 if (media_was_present)
2449 sd_printk(KERN_NOTICE, sdkp,
2450 "Media removed, stopped polling\n");
2451 return;
2452 }
2453 sense_valid = scsi_sense_valid(&sshdr);
2454 }
2455
2456 if (!scsi_status_is_check_condition(the_result)) {
2457 /* no sense, TUR either succeeded or failed
2458 * with a status error */
2459 if(!spintime && !scsi_status_is_good(the_result)) {
2460 sd_print_result(sdkp, "Test Unit Ready failed",
2461 the_result);
2462 }
2463 break;
2464 }
2465
2466 /*
2467 * The device does not want the automatic start to be issued.
2468 */
2469 if (sdkp->device->no_start_on_add)
2470 break;
2471
2472 if (sense_valid && sshdr.sense_key == NOT_READY) {
2473 if (sshdr.asc == 4 && sshdr.ascq == 3)
2474 break; /* manual intervention required */
2475 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2476 break; /* standby */
2477 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2478 break; /* unavailable */
2479 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2480 break; /* sanitize in progress */
2481 if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2482 break; /* depopulation in progress */
2483 if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2484 break; /* depopulation restoration in progress */
2485 /*
2486 * Issue command to spin up drive when not ready
2487 */
2488 if (!spintime) {
2489 /* Return immediately and start spin cycle */
2490 const u8 start_cmd[10] = {
2491 [0] = START_STOP,
2492 [1] = 1,
2493 [4] = sdkp->device->start_stop_pwr_cond ?
2494 0x11 : 1,
2495 };
2496
2497 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2498 scsi_execute_cmd(sdkp->device, start_cmd,
2499 REQ_OP_DRV_IN, NULL, 0,
2500 SD_TIMEOUT, sdkp->max_retries,
2501 &exec_args);
2502 spintime_expire = jiffies + 100 * HZ;
2503 spintime = 1;
2504 }
2505 /* Wait 1 second for next try */
2506 msleep(1000);
2507 printk(KERN_CONT ".");
2508
2509 /*
2510 * Wait for USB flash devices with slow firmware.
2511 * Yes, this sense key/ASC combination shouldn't
2512 * occur here. It's characteristic of these devices.
2513 */
2514 } else if (sense_valid &&
2515 sshdr.sense_key == UNIT_ATTENTION &&
2516 sshdr.asc == 0x28) {
2517 if (!spintime) {
2518 spintime_expire = jiffies + 5 * HZ;
2519 spintime = 1;
2520 }
2521 /* Wait 1 second for next try */
2522 msleep(1000);
2523 } else {
2524 /* we don't understand the sense code, so it's
2525 * probably pointless to loop */
2526 if(!spintime) {
2527 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2528 sd_print_sense_hdr(sdkp, &sshdr);
2529 }
2530 break;
2531 }
2532
2533 } while (spintime && time_before_eq(jiffies, spintime_expire));
2534
2535 if (spintime) {
2536 if (scsi_status_is_good(the_result))
2537 printk(KERN_CONT "ready\n");
2538 else
2539 printk(KERN_CONT "not responding...\n");
2540 }
2541}
2542
2543/*
2544 * Determine whether disk supports Data Integrity Field.
2545 */
2546static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2547{
2548 struct scsi_device *sdp = sdkp->device;
2549 u8 type;
2550
2551 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2552 sdkp->protection_type = 0;
2553 return 0;
2554 }
2555
2556 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2557
2558 if (type > T10_PI_TYPE3_PROTECTION) {
2559 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2560 " protection type %u. Disabling disk!\n",
2561 type);
2562 sdkp->protection_type = 0;
2563 return -ENODEV;
2564 }
2565
2566 sdkp->protection_type = type;
2567
2568 return 0;
2569}
2570
2571static void sd_config_protection(struct scsi_disk *sdkp,
2572 struct queue_limits *lim)
2573{
2574 struct scsi_device *sdp = sdkp->device;
2575
2576 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
2577 sd_dif_config_host(sdkp, lim);
2578
2579 if (!sdkp->protection_type)
2580 return;
2581
2582 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2583 sd_first_printk(KERN_NOTICE, sdkp,
2584 "Disabling DIF Type %u protection\n",
2585 sdkp->protection_type);
2586 sdkp->protection_type = 0;
2587 }
2588
2589 sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2590 sdkp->protection_type);
2591}
2592
2593static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2594 struct scsi_sense_hdr *sshdr, int sense_valid,
2595 int the_result)
2596{
2597 if (sense_valid)
2598 sd_print_sense_hdr(sdkp, sshdr);
2599 else
2600 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2601
2602 /*
2603 * Set dirty bit for removable devices if not ready -
2604 * sometimes drives will not report this properly.
2605 */
2606 if (sdp->removable &&
2607 sense_valid && sshdr->sense_key == NOT_READY)
2608 set_media_not_present(sdkp);
2609
2610 /*
2611 * We used to set media_present to 0 here to indicate no media
2612 * in the drive, but some drives fail read capacity even with
2613 * media present, so we can't do that.
2614 */
2615 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2616}
2617
2618#define RC16_LEN 32
2619#if RC16_LEN > SD_BUF_SIZE
2620#error RC16_LEN must not be more than SD_BUF_SIZE
2621#endif
2622
2623#define READ_CAPACITY_RETRIES_ON_RESET 10
2624
2625static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2626 struct queue_limits *lim, unsigned char *buffer)
2627{
2628 unsigned char cmd[16];
2629 struct scsi_sense_hdr sshdr;
2630 const struct scsi_exec_args exec_args = {
2631 .sshdr = &sshdr,
2632 };
2633 int sense_valid = 0;
2634 int the_result;
2635 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2636 unsigned int alignment;
2637 unsigned long long lba;
2638 unsigned sector_size;
2639
2640 if (sdp->no_read_capacity_16)
2641 return -EINVAL;
2642
2643 do {
2644 memset(cmd, 0, 16);
2645 cmd[0] = SERVICE_ACTION_IN_16;
2646 cmd[1] = SAI_READ_CAPACITY_16;
2647 cmd[13] = RC16_LEN;
2648 memset(buffer, 0, RC16_LEN);
2649
2650 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2651 buffer, RC16_LEN, SD_TIMEOUT,
2652 sdkp->max_retries, &exec_args);
2653 if (the_result > 0) {
2654 if (media_not_present(sdkp, &sshdr))
2655 return -ENODEV;
2656
2657 sense_valid = scsi_sense_valid(&sshdr);
2658 if (sense_valid &&
2659 sshdr.sense_key == ILLEGAL_REQUEST &&
2660 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2661 sshdr.ascq == 0x00)
2662 /* Invalid Command Operation Code or
2663 * Invalid Field in CDB, just retry
2664 * silently with RC10 */
2665 return -EINVAL;
2666 if (sense_valid &&
2667 sshdr.sense_key == UNIT_ATTENTION &&
2668 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2669 /* Device reset might occur several times,
2670 * give it one more chance */
2671 if (--reset_retries > 0)
2672 continue;
2673 }
2674 retries--;
2675
2676 } while (the_result && retries);
2677
2678 if (the_result) {
2679 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2680 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2681 return -EINVAL;
2682 }
2683
2684 sector_size = get_unaligned_be32(&buffer[8]);
2685 lba = get_unaligned_be64(&buffer[0]);
2686
2687 if (sd_read_protection_type(sdkp, buffer) < 0) {
2688 sdkp->capacity = 0;
2689 return -ENODEV;
2690 }
2691
2692 /* Logical blocks per physical block exponent */
2693 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2694
2695 /* RC basis */
2696 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2697
2698 /* Lowest aligned logical block */
2699 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2700 lim->alignment_offset = alignment;
2701 if (alignment && sdkp->first_scan)
2702 sd_printk(KERN_NOTICE, sdkp,
2703 "physical block alignment offset: %u\n", alignment);
2704
2705 if (buffer[14] & 0x80) { /* LBPME */
2706 sdkp->lbpme = 1;
2707
2708 if (buffer[14] & 0x40) /* LBPRZ */
2709 sdkp->lbprz = 1;
2710 }
2711
2712 sdkp->capacity = lba + 1;
2713 return sector_size;
2714}
2715
2716static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2717 unsigned char *buffer)
2718{
2719 static const u8 cmd[10] = { READ_CAPACITY };
2720 struct scsi_sense_hdr sshdr;
2721 struct scsi_failure failure_defs[] = {
2722 /* Do not retry Medium Not Present */
2723 {
2724 .sense = UNIT_ATTENTION,
2725 .asc = 0x3A,
2726 .result = SAM_STAT_CHECK_CONDITION,
2727 },
2728 {
2729 .sense = NOT_READY,
2730 .asc = 0x3A,
2731 .result = SAM_STAT_CHECK_CONDITION,
2732 },
2733 /* Device reset might occur several times so retry a lot */
2734 {
2735 .sense = UNIT_ATTENTION,
2736 .asc = 0x29,
2737 .allowed = READ_CAPACITY_RETRIES_ON_RESET,
2738 .result = SAM_STAT_CHECK_CONDITION,
2739 },
2740 /* Any other error not listed above retry 3 times */
2741 {
2742 .result = SCMD_FAILURE_RESULT_ANY,
2743 .allowed = 3,
2744 },
2745 {}
2746 };
2747 struct scsi_failures failures = {
2748 .failure_definitions = failure_defs,
2749 };
2750 const struct scsi_exec_args exec_args = {
2751 .sshdr = &sshdr,
2752 .failures = &failures,
2753 };
2754 int sense_valid = 0;
2755 int the_result;
2756 sector_t lba;
2757 unsigned sector_size;
2758
2759 memset(buffer, 0, 8);
2760
2761 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2762 8, SD_TIMEOUT, sdkp->max_retries,
2763 &exec_args);
2764
2765 if (the_result > 0) {
2766 sense_valid = scsi_sense_valid(&sshdr);
2767
2768 if (media_not_present(sdkp, &sshdr))
2769 return -ENODEV;
2770 }
2771
2772 if (the_result) {
2773 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2774 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2775 return -EINVAL;
2776 }
2777
2778 sector_size = get_unaligned_be32(&buffer[4]);
2779 lba = get_unaligned_be32(&buffer[0]);
2780
2781 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2782 /* Some buggy (usb cardreader) devices return an lba of
2783 0xffffffff when the want to report a size of 0 (with
2784 which they really mean no media is present) */
2785 sdkp->capacity = 0;
2786 sdkp->physical_block_size = sector_size;
2787 return sector_size;
2788 }
2789
2790 sdkp->capacity = lba + 1;
2791 sdkp->physical_block_size = sector_size;
2792 return sector_size;
2793}
2794
2795static int sd_try_rc16_first(struct scsi_device *sdp)
2796{
2797 if (sdp->host->max_cmd_len < 16)
2798 return 0;
2799 if (sdp->try_rc_10_first)
2800 return 0;
2801 if (sdp->scsi_level > SCSI_SPC_2)
2802 return 1;
2803 if (scsi_device_protection(sdp))
2804 return 1;
2805 return 0;
2806}
2807
2808/*
2809 * read disk capacity
2810 */
2811static void
2812sd_read_capacity(struct scsi_disk *sdkp, struct queue_limits *lim,
2813 unsigned char *buffer)
2814{
2815 int sector_size;
2816 struct scsi_device *sdp = sdkp->device;
2817
2818 if (sd_try_rc16_first(sdp)) {
2819 sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2820 if (sector_size == -EOVERFLOW)
2821 goto got_data;
2822 if (sector_size == -ENODEV)
2823 return;
2824 if (sector_size < 0)
2825 sector_size = read_capacity_10(sdkp, sdp, buffer);
2826 if (sector_size < 0)
2827 return;
2828 } else {
2829 sector_size = read_capacity_10(sdkp, sdp, buffer);
2830 if (sector_size == -EOVERFLOW)
2831 goto got_data;
2832 if (sector_size < 0)
2833 return;
2834 if ((sizeof(sdkp->capacity) > 4) &&
2835 (sdkp->capacity > 0xffffffffULL)) {
2836 int old_sector_size = sector_size;
2837 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2838 "Trying to use READ CAPACITY(16).\n");
2839 sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2840 if (sector_size < 0) {
2841 sd_printk(KERN_NOTICE, sdkp,
2842 "Using 0xffffffff as device size\n");
2843 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2844 sector_size = old_sector_size;
2845 goto got_data;
2846 }
2847 /* Remember that READ CAPACITY(16) succeeded */
2848 sdp->try_rc_10_first = 0;
2849 }
2850 }
2851
2852 /* Some devices are known to return the total number of blocks,
2853 * not the highest block number. Some devices have versions
2854 * which do this and others which do not. Some devices we might
2855 * suspect of doing this but we don't know for certain.
2856 *
2857 * If we know the reported capacity is wrong, decrement it. If
2858 * we can only guess, then assume the number of blocks is even
2859 * (usually true but not always) and err on the side of lowering
2860 * the capacity.
2861 */
2862 if (sdp->fix_capacity ||
2863 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2864 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2865 "from its reported value: %llu\n",
2866 (unsigned long long) sdkp->capacity);
2867 --sdkp->capacity;
2868 }
2869
2870got_data:
2871 if (sector_size == 0) {
2872 sector_size = 512;
2873 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2874 "assuming 512.\n");
2875 }
2876
2877 if (sector_size != 512 &&
2878 sector_size != 1024 &&
2879 sector_size != 2048 &&
2880 sector_size != 4096) {
2881 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2882 sector_size);
2883 /*
2884 * The user might want to re-format the drive with
2885 * a supported sectorsize. Once this happens, it
2886 * would be relatively trivial to set the thing up.
2887 * For this reason, we leave the thing in the table.
2888 */
2889 sdkp->capacity = 0;
2890 /*
2891 * set a bogus sector size so the normal read/write
2892 * logic in the block layer will eventually refuse any
2893 * request on this device without tripping over power
2894 * of two sector size assumptions
2895 */
2896 sector_size = 512;
2897 }
2898 lim->logical_block_size = sector_size;
2899 lim->physical_block_size = sdkp->physical_block_size;
2900 sdkp->device->sector_size = sector_size;
2901
2902 if (sdkp->capacity > 0xffffffff)
2903 sdp->use_16_for_rw = 1;
2904
2905}
2906
2907/*
2908 * Print disk capacity
2909 */
2910static void
2911sd_print_capacity(struct scsi_disk *sdkp,
2912 sector_t old_capacity)
2913{
2914 int sector_size = sdkp->device->sector_size;
2915 char cap_str_2[10], cap_str_10[10];
2916
2917 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2918 return;
2919
2920 string_get_size(sdkp->capacity, sector_size,
2921 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2922 string_get_size(sdkp->capacity, sector_size,
2923 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2924
2925 sd_printk(KERN_NOTICE, sdkp,
2926 "%llu %d-byte logical blocks: (%s/%s)\n",
2927 (unsigned long long)sdkp->capacity,
2928 sector_size, cap_str_10, cap_str_2);
2929
2930 if (sdkp->physical_block_size != sector_size)
2931 sd_printk(KERN_NOTICE, sdkp,
2932 "%u-byte physical blocks\n",
2933 sdkp->physical_block_size);
2934}
2935
2936/* called with buffer of length 512 */
2937static inline int
2938sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2939 unsigned char *buffer, int len, struct scsi_mode_data *data,
2940 struct scsi_sense_hdr *sshdr)
2941{
2942 /*
2943 * If we must use MODE SENSE(10), make sure that the buffer length
2944 * is at least 8 bytes so that the mode sense header fits.
2945 */
2946 if (sdkp->device->use_10_for_ms && len < 8)
2947 len = 8;
2948
2949 return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2950 SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2951}
2952
2953/*
2954 * read write protect setting, if possible - called only in sd_revalidate_disk()
2955 * called with buffer of length SD_BUF_SIZE
2956 */
2957static void
2958sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2959{
2960 int res;
2961 struct scsi_device *sdp = sdkp->device;
2962 struct scsi_mode_data data;
2963 int old_wp = sdkp->write_prot;
2964
2965 set_disk_ro(sdkp->disk, 0);
2966 if (sdp->skip_ms_page_3f) {
2967 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2968 return;
2969 }
2970
2971 if (sdp->use_192_bytes_for_3f) {
2972 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2973 } else {
2974 /*
2975 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2976 * We have to start carefully: some devices hang if we ask
2977 * for more than is available.
2978 */
2979 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2980
2981 /*
2982 * Second attempt: ask for page 0 When only page 0 is
2983 * implemented, a request for page 3F may return Sense Key
2984 * 5: Illegal Request, Sense Code 24: Invalid field in
2985 * CDB.
2986 */
2987 if (res < 0)
2988 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2989
2990 /*
2991 * Third attempt: ask 255 bytes, as we did earlier.
2992 */
2993 if (res < 0)
2994 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2995 &data, NULL);
2996 }
2997
2998 if (res < 0) {
2999 sd_first_printk(KERN_WARNING, sdkp,
3000 "Test WP failed, assume Write Enabled\n");
3001 } else {
3002 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
3003 set_disk_ro(sdkp->disk, sdkp->write_prot);
3004 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
3005 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
3006 sdkp->write_prot ? "on" : "off");
3007 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
3008 }
3009 }
3010}
3011
3012/*
3013 * sd_read_cache_type - called only from sd_revalidate_disk()
3014 * called with buffer of length SD_BUF_SIZE
3015 */
3016static void
3017sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
3018{
3019 int len = 0, res;
3020 struct scsi_device *sdp = sdkp->device;
3021
3022 int dbd;
3023 int modepage;
3024 int first_len;
3025 struct scsi_mode_data data;
3026 struct scsi_sense_hdr sshdr;
3027 int old_wce = sdkp->WCE;
3028 int old_rcd = sdkp->RCD;
3029 int old_dpofua = sdkp->DPOFUA;
3030
3031
3032 if (sdkp->cache_override)
3033 return;
3034
3035 first_len = 4;
3036 if (sdp->skip_ms_page_8) {
3037 if (sdp->type == TYPE_RBC)
3038 goto defaults;
3039 else {
3040 if (sdp->skip_ms_page_3f)
3041 goto defaults;
3042 modepage = 0x3F;
3043 if (sdp->use_192_bytes_for_3f)
3044 first_len = 192;
3045 dbd = 0;
3046 }
3047 } else if (sdp->type == TYPE_RBC) {
3048 modepage = 6;
3049 dbd = 8;
3050 } else {
3051 modepage = 8;
3052 dbd = 0;
3053 }
3054
3055 /* cautiously ask */
3056 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
3057 &data, &sshdr);
3058
3059 if (res < 0)
3060 goto bad_sense;
3061
3062 if (!data.header_length) {
3063 modepage = 6;
3064 first_len = 0;
3065 sd_first_printk(KERN_ERR, sdkp,
3066 "Missing header in MODE_SENSE response\n");
3067 }
3068
3069 /* that went OK, now ask for the proper length */
3070 len = data.length;
3071
3072 /*
3073 * We're only interested in the first three bytes, actually.
3074 * But the data cache page is defined for the first 20.
3075 */
3076 if (len < 3)
3077 goto bad_sense;
3078 else if (len > SD_BUF_SIZE) {
3079 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
3080 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
3081 len = SD_BUF_SIZE;
3082 }
3083 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
3084 len = 192;
3085
3086 /* Get the data */
3087 if (len > first_len)
3088 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
3089 &data, &sshdr);
3090
3091 if (!res) {
3092 int offset = data.header_length + data.block_descriptor_length;
3093
3094 while (offset < len) {
3095 u8 page_code = buffer[offset] & 0x3F;
3096 u8 spf = buffer[offset] & 0x40;
3097
3098 if (page_code == 8 || page_code == 6) {
3099 /* We're interested only in the first 3 bytes.
3100 */
3101 if (len - offset <= 2) {
3102 sd_first_printk(KERN_ERR, sdkp,
3103 "Incomplete mode parameter "
3104 "data\n");
3105 goto defaults;
3106 } else {
3107 modepage = page_code;
3108 goto Page_found;
3109 }
3110 } else {
3111 /* Go to the next page */
3112 if (spf && len - offset > 3)
3113 offset += 4 + (buffer[offset+2] << 8) +
3114 buffer[offset+3];
3115 else if (!spf && len - offset > 1)
3116 offset += 2 + buffer[offset+1];
3117 else {
3118 sd_first_printk(KERN_ERR, sdkp,
3119 "Incomplete mode "
3120 "parameter data\n");
3121 goto defaults;
3122 }
3123 }
3124 }
3125
3126 sd_first_printk(KERN_WARNING, sdkp,
3127 "No Caching mode page found\n");
3128 goto defaults;
3129
3130 Page_found:
3131 if (modepage == 8) {
3132 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3133 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3134 } else {
3135 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3136 sdkp->RCD = 0;
3137 }
3138
3139 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3140 if (sdp->broken_fua) {
3141 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3142 sdkp->DPOFUA = 0;
3143 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3144 !sdkp->device->use_16_for_rw) {
3145 sd_first_printk(KERN_NOTICE, sdkp,
3146 "Uses READ/WRITE(6), disabling FUA\n");
3147 sdkp->DPOFUA = 0;
3148 }
3149
3150 /* No cache flush allowed for write protected devices */
3151 if (sdkp->WCE && sdkp->write_prot)
3152 sdkp->WCE = 0;
3153
3154 if (sdkp->first_scan || old_wce != sdkp->WCE ||
3155 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3156 sd_printk(KERN_NOTICE, sdkp,
3157 "Write cache: %s, read cache: %s, %s\n",
3158 sdkp->WCE ? "enabled" : "disabled",
3159 sdkp->RCD ? "disabled" : "enabled",
3160 sdkp->DPOFUA ? "supports DPO and FUA"
3161 : "doesn't support DPO or FUA");
3162
3163 return;
3164 }
3165
3166bad_sense:
3167 if (res == -EIO && scsi_sense_valid(&sshdr) &&
3168 sshdr.sense_key == ILLEGAL_REQUEST &&
3169 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3170 /* Invalid field in CDB */
3171 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3172 else
3173 sd_first_printk(KERN_ERR, sdkp,
3174 "Asking for cache data failed\n");
3175
3176defaults:
3177 if (sdp->wce_default_on) {
3178 sd_first_printk(KERN_NOTICE, sdkp,
3179 "Assuming drive cache: write back\n");
3180 sdkp->WCE = 1;
3181 } else {
3182 sd_first_printk(KERN_WARNING, sdkp,
3183 "Assuming drive cache: write through\n");
3184 sdkp->WCE = 0;
3185 }
3186 sdkp->RCD = 0;
3187 sdkp->DPOFUA = 0;
3188}
3189
3190static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3191{
3192 u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3193 struct {
3194 struct scsi_stream_status_header h;
3195 struct scsi_stream_status s;
3196 } buf;
3197 struct scsi_device *sdev = sdkp->device;
3198 struct scsi_sense_hdr sshdr;
3199 const struct scsi_exec_args exec_args = {
3200 .sshdr = &sshdr,
3201 };
3202 int res;
3203
3204 put_unaligned_be16(stream_id, &cdb[4]);
3205 put_unaligned_be32(sizeof(buf), &cdb[10]);
3206
3207 res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3208 SD_TIMEOUT, sdkp->max_retries, &exec_args);
3209 if (res < 0)
3210 return false;
3211 if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3212 sd_print_sense_hdr(sdkp, &sshdr);
3213 if (res)
3214 return false;
3215 if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3216 return false;
3217 return buf.h.stream_status[0].perm;
3218}
3219
3220static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3221{
3222 struct scsi_device *sdp = sdkp->device;
3223 const struct scsi_io_group_descriptor *desc, *start, *end;
3224 u16 permanent_stream_count_old;
3225 struct scsi_sense_hdr sshdr;
3226 struct scsi_mode_data data;
3227 int res;
3228
3229 if (sdp->sdev_bflags & BLIST_SKIP_IO_HINTS)
3230 return;
3231
3232 res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3233 /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3234 sdkp->max_retries, &data, &sshdr);
3235 if (res < 0)
3236 return;
3237 start = (void *)buffer + data.header_length + 16;
3238 end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3239 sizeof(*end));
3240 /*
3241 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3242 * should assign the lowest numbered stream identifiers to permanent
3243 * streams.
3244 */
3245 for (desc = start; desc < end; desc++)
3246 if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3247 break;
3248 permanent_stream_count_old = sdkp->permanent_stream_count;
3249 sdkp->permanent_stream_count = desc - start;
3250 if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3251 sd_printk(KERN_INFO, sdkp,
3252 "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3253 sdkp->permanent_stream_count);
3254 else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3255 sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3256 sdkp->permanent_stream_count);
3257}
3258
3259/*
3260 * The ATO bit indicates whether the DIF application tag is available
3261 * for use by the operating system.
3262 */
3263static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3264{
3265 int res, offset;
3266 struct scsi_device *sdp = sdkp->device;
3267 struct scsi_mode_data data;
3268 struct scsi_sense_hdr sshdr;
3269
3270 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3271 return;
3272
3273 if (sdkp->protection_type == 0)
3274 return;
3275
3276 res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3277 sdkp->max_retries, &data, &sshdr);
3278
3279 if (res < 0 || !data.header_length ||
3280 data.length < 6) {
3281 sd_first_printk(KERN_WARNING, sdkp,
3282 "getting Control mode page failed, assume no ATO\n");
3283
3284 if (res == -EIO && scsi_sense_valid(&sshdr))
3285 sd_print_sense_hdr(sdkp, &sshdr);
3286
3287 return;
3288 }
3289
3290 offset = data.header_length + data.block_descriptor_length;
3291
3292 if ((buffer[offset] & 0x3f) != 0x0a) {
3293 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3294 return;
3295 }
3296
3297 if ((buffer[offset + 5] & 0x80) == 0)
3298 return;
3299
3300 sdkp->ATO = 1;
3301
3302 return;
3303}
3304
3305static unsigned int sd_discard_mode(struct scsi_disk *sdkp)
3306{
3307 if (!sdkp->lbpme)
3308 return SD_LBP_FULL;
3309
3310 if (!sdkp->lbpvpd) {
3311 /* LBP VPD page not provided */
3312 if (sdkp->max_unmap_blocks)
3313 return SD_LBP_UNMAP;
3314 return SD_LBP_WS16;
3315 }
3316
3317 /* LBP VPD page tells us what to use */
3318 if (sdkp->lbpu && sdkp->max_unmap_blocks)
3319 return SD_LBP_UNMAP;
3320 if (sdkp->lbpws)
3321 return SD_LBP_WS16;
3322 if (sdkp->lbpws10)
3323 return SD_LBP_WS10;
3324 return SD_LBP_DISABLE;
3325}
3326
3327/*
3328 * Query disk device for preferred I/O sizes.
3329 */
3330static void sd_read_block_limits(struct scsi_disk *sdkp,
3331 struct queue_limits *lim)
3332{
3333 struct scsi_vpd *vpd;
3334
3335 rcu_read_lock();
3336
3337 vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3338 if (!vpd || vpd->len < 16)
3339 goto out;
3340
3341 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3342 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3343 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3344
3345 if (vpd->len >= 64) {
3346 unsigned int lba_count, desc_count;
3347
3348 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3349
3350 if (!sdkp->lbpme)
3351 goto config_atomic;
3352
3353 lba_count = get_unaligned_be32(&vpd->data[20]);
3354 desc_count = get_unaligned_be32(&vpd->data[24]);
3355
3356 if (lba_count && desc_count)
3357 sdkp->max_unmap_blocks = lba_count;
3358
3359 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3360
3361 if (vpd->data[32] & 0x80)
3362 sdkp->unmap_alignment =
3363 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3364
3365config_atomic:
3366 sdkp->max_atomic = get_unaligned_be32(&vpd->data[44]);
3367 sdkp->atomic_alignment = get_unaligned_be32(&vpd->data[48]);
3368 sdkp->atomic_granularity = get_unaligned_be32(&vpd->data[52]);
3369 sdkp->max_atomic_with_boundary = get_unaligned_be32(&vpd->data[56]);
3370 sdkp->max_atomic_boundary = get_unaligned_be32(&vpd->data[60]);
3371
3372 sd_config_atomic(sdkp, lim);
3373 }
3374
3375 out:
3376 rcu_read_unlock();
3377}
3378
3379/* Parse the Block Limits Extension VPD page (0xb7) */
3380static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3381{
3382 struct scsi_vpd *vpd;
3383
3384 rcu_read_lock();
3385 vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3386 if (vpd && vpd->len >= 2)
3387 sdkp->rscs = vpd->data[5] & 1;
3388 rcu_read_unlock();
3389}
3390
3391/* Query block device characteristics */
3392static void sd_read_block_characteristics(struct scsi_disk *sdkp,
3393 struct queue_limits *lim)
3394{
3395 struct scsi_vpd *vpd;
3396 u16 rot;
3397
3398 rcu_read_lock();
3399 vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3400
3401 if (!vpd || vpd->len <= 8) {
3402 rcu_read_unlock();
3403 return;
3404 }
3405
3406 rot = get_unaligned_be16(&vpd->data[4]);
3407 sdkp->zoned = (vpd->data[8] >> 4) & 3;
3408 rcu_read_unlock();
3409
3410 if (rot == 1)
3411 lim->features &= ~(BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3412
3413 if (!sdkp->first_scan)
3414 return;
3415
3416 if (sdkp->device->type == TYPE_ZBC)
3417 sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3418 else if (sdkp->zoned == 1)
3419 sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3420 else if (sdkp->zoned == 2)
3421 sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3422}
3423
3424/**
3425 * sd_read_block_provisioning - Query provisioning VPD page
3426 * @sdkp: disk to query
3427 */
3428static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3429{
3430 struct scsi_vpd *vpd;
3431
3432 if (sdkp->lbpme == 0)
3433 return;
3434
3435 rcu_read_lock();
3436 vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3437
3438 if (!vpd || vpd->len < 8) {
3439 rcu_read_unlock();
3440 return;
3441 }
3442
3443 sdkp->lbpvpd = 1;
3444 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */
3445 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3446 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3447 rcu_read_unlock();
3448}
3449
3450static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3451{
3452 struct scsi_device *sdev = sdkp->device;
3453
3454 if (sdev->host->no_write_same) {
3455 sdev->no_write_same = 1;
3456
3457 return;
3458 }
3459
3460 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3461 struct scsi_vpd *vpd;
3462
3463 sdev->no_report_opcodes = 1;
3464
3465 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3466 * CODES is unsupported and the device has an ATA
3467 * Information VPD page (SAT).
3468 */
3469 rcu_read_lock();
3470 vpd = rcu_dereference(sdev->vpd_pg89);
3471 if (vpd)
3472 sdev->no_write_same = 1;
3473 rcu_read_unlock();
3474 }
3475
3476 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3477 sdkp->ws16 = 1;
3478
3479 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3480 sdkp->ws10 = 1;
3481}
3482
3483static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3484{
3485 struct scsi_device *sdev = sdkp->device;
3486
3487 if (!sdev->security_supported)
3488 return;
3489
3490 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3491 SECURITY_PROTOCOL_IN, 0) == 1 &&
3492 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3493 SECURITY_PROTOCOL_OUT, 0) == 1)
3494 sdkp->security = 1;
3495}
3496
3497static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3498{
3499 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3500}
3501
3502/**
3503 * sd_read_cpr - Query concurrent positioning ranges
3504 * @sdkp: disk to query
3505 */
3506static void sd_read_cpr(struct scsi_disk *sdkp)
3507{
3508 struct blk_independent_access_ranges *iars = NULL;
3509 unsigned char *buffer = NULL;
3510 unsigned int nr_cpr = 0;
3511 int i, vpd_len, buf_len = SD_BUF_SIZE;
3512 u8 *desc;
3513
3514 /*
3515 * We need to have the capacity set first for the block layer to be
3516 * able to check the ranges.
3517 */
3518 if (sdkp->first_scan)
3519 return;
3520
3521 if (!sdkp->capacity)
3522 goto out;
3523
3524 /*
3525 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3526 * leading to a maximum page size of 64 + 256*32 bytes.
3527 */
3528 buf_len = 64 + 256*32;
3529 buffer = kmalloc(buf_len, GFP_KERNEL);
3530 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3531 goto out;
3532
3533 /* We must have at least a 64B header and one 32B range descriptor */
3534 vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3535 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3536 sd_printk(KERN_ERR, sdkp,
3537 "Invalid Concurrent Positioning Ranges VPD page\n");
3538 goto out;
3539 }
3540
3541 nr_cpr = (vpd_len - 64) / 32;
3542 if (nr_cpr == 1) {
3543 nr_cpr = 0;
3544 goto out;
3545 }
3546
3547 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3548 if (!iars) {
3549 nr_cpr = 0;
3550 goto out;
3551 }
3552
3553 desc = &buffer[64];
3554 for (i = 0; i < nr_cpr; i++, desc += 32) {
3555 if (desc[0] != i) {
3556 sd_printk(KERN_ERR, sdkp,
3557 "Invalid Concurrent Positioning Range number\n");
3558 nr_cpr = 0;
3559 break;
3560 }
3561
3562 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3563 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3564 }
3565
3566out:
3567 disk_set_independent_access_ranges(sdkp->disk, iars);
3568 if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3569 sd_printk(KERN_NOTICE, sdkp,
3570 "%u concurrent positioning ranges\n", nr_cpr);
3571 sdkp->nr_actuators = nr_cpr;
3572 }
3573
3574 kfree(buffer);
3575}
3576
3577static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3578{
3579 struct scsi_device *sdp = sdkp->device;
3580 unsigned int min_xfer_bytes =
3581 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3582
3583 if (sdkp->min_xfer_blocks == 0)
3584 return false;
3585
3586 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3587 sd_first_printk(KERN_WARNING, sdkp,
3588 "Preferred minimum I/O size %u bytes not a " \
3589 "multiple of physical block size (%u bytes)\n",
3590 min_xfer_bytes, sdkp->physical_block_size);
3591 sdkp->min_xfer_blocks = 0;
3592 return false;
3593 }
3594
3595 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3596 min_xfer_bytes);
3597 return true;
3598}
3599
3600/*
3601 * Determine the device's preferred I/O size for reads and writes
3602 * unless the reported value is unreasonably small, large, not a
3603 * multiple of the physical block size, or simply garbage.
3604 */
3605static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3606 unsigned int dev_max)
3607{
3608 struct scsi_device *sdp = sdkp->device;
3609 unsigned int opt_xfer_bytes =
3610 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3611 unsigned int min_xfer_bytes =
3612 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3613
3614 if (sdkp->opt_xfer_blocks == 0)
3615 return false;
3616
3617 if (sdkp->opt_xfer_blocks > dev_max) {
3618 sd_first_printk(KERN_WARNING, sdkp,
3619 "Optimal transfer size %u logical blocks " \
3620 "> dev_max (%u logical blocks)\n",
3621 sdkp->opt_xfer_blocks, dev_max);
3622 return false;
3623 }
3624
3625 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3626 sd_first_printk(KERN_WARNING, sdkp,
3627 "Optimal transfer size %u logical blocks " \
3628 "> sd driver limit (%u logical blocks)\n",
3629 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3630 return false;
3631 }
3632
3633 if (opt_xfer_bytes < PAGE_SIZE) {
3634 sd_first_printk(KERN_WARNING, sdkp,
3635 "Optimal transfer size %u bytes < " \
3636 "PAGE_SIZE (%u bytes)\n",
3637 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3638 return false;
3639 }
3640
3641 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3642 sd_first_printk(KERN_WARNING, sdkp,
3643 "Optimal transfer size %u bytes not a " \
3644 "multiple of preferred minimum block " \
3645 "size (%u bytes)\n",
3646 opt_xfer_bytes, min_xfer_bytes);
3647 return false;
3648 }
3649
3650 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3651 sd_first_printk(KERN_WARNING, sdkp,
3652 "Optimal transfer size %u bytes not a " \
3653 "multiple of physical block size (%u bytes)\n",
3654 opt_xfer_bytes, sdkp->physical_block_size);
3655 return false;
3656 }
3657
3658 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3659 opt_xfer_bytes);
3660 return true;
3661}
3662
3663static void sd_read_block_zero(struct scsi_disk *sdkp)
3664{
3665 struct scsi_device *sdev = sdkp->device;
3666 unsigned int buf_len = sdev->sector_size;
3667 u8 *buffer, cmd[16] = { };
3668
3669 buffer = kmalloc(buf_len, GFP_KERNEL);
3670 if (!buffer)
3671 return;
3672
3673 if (sdev->use_16_for_rw) {
3674 cmd[0] = READ_16;
3675 put_unaligned_be64(0, &cmd[2]); /* Logical block address 0 */
3676 put_unaligned_be32(1, &cmd[10]);/* Transfer 1 logical block */
3677 } else {
3678 cmd[0] = READ_10;
3679 put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3680 put_unaligned_be16(1, &cmd[7]); /* Transfer 1 logical block */
3681 }
3682
3683 scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3684 SD_TIMEOUT, sdkp->max_retries, NULL);
3685 kfree(buffer);
3686}
3687
3688/**
3689 * sd_revalidate_disk - called the first time a new disk is seen,
3690 * performs disk spin up, read_capacity, etc.
3691 * @disk: struct gendisk we care about
3692 **/
3693static int sd_revalidate_disk(struct gendisk *disk)
3694{
3695 struct scsi_disk *sdkp = scsi_disk(disk);
3696 struct scsi_device *sdp = sdkp->device;
3697 sector_t old_capacity = sdkp->capacity;
3698 struct queue_limits lim;
3699 unsigned char *buffer;
3700 unsigned int dev_max;
3701 int err;
3702
3703 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3704 "sd_revalidate_disk\n"));
3705
3706 /*
3707 * If the device is offline, don't try and read capacity or any
3708 * of the other niceties.
3709 */
3710 if (!scsi_device_online(sdp))
3711 goto out;
3712
3713 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3714 if (!buffer) {
3715 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3716 "allocation failure.\n");
3717 goto out;
3718 }
3719
3720 sd_spinup_disk(sdkp);
3721
3722 lim = queue_limits_start_update(sdkp->disk->queue);
3723
3724 /*
3725 * Without media there is no reason to ask; moreover, some devices
3726 * react badly if we do.
3727 */
3728 if (sdkp->media_present) {
3729 sd_read_capacity(sdkp, &lim, buffer);
3730 /*
3731 * Some USB/UAS devices return generic values for mode pages
3732 * until the media has been accessed. Trigger a READ operation
3733 * to force the device to populate mode pages.
3734 */
3735 if (sdp->read_before_ms)
3736 sd_read_block_zero(sdkp);
3737 /*
3738 * set the default to rotational. All non-rotational devices
3739 * support the block characteristics VPD page, which will
3740 * cause this to be updated correctly and any device which
3741 * doesn't support it should be treated as rotational.
3742 */
3743 lim.features |= (BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3744
3745 if (scsi_device_supports_vpd(sdp)) {
3746 sd_read_block_provisioning(sdkp);
3747 sd_read_block_limits(sdkp, &lim);
3748 sd_read_block_limits_ext(sdkp);
3749 sd_read_block_characteristics(sdkp, &lim);
3750 sd_zbc_read_zones(sdkp, &lim, buffer);
3751 }
3752
3753 sd_config_discard(sdkp, &lim, sd_discard_mode(sdkp));
3754
3755 sd_print_capacity(sdkp, old_capacity);
3756
3757 sd_read_write_protect_flag(sdkp, buffer);
3758 sd_read_cache_type(sdkp, buffer);
3759 sd_read_io_hints(sdkp, buffer);
3760 sd_read_app_tag_own(sdkp, buffer);
3761 sd_read_write_same(sdkp, buffer);
3762 sd_read_security(sdkp, buffer);
3763 sd_config_protection(sdkp, &lim);
3764 }
3765
3766 /*
3767 * We now have all cache related info, determine how we deal
3768 * with flush requests.
3769 */
3770 sd_set_flush_flag(sdkp, &lim);
3771
3772 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3773 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3774
3775 /* Some devices report a maximum block count for READ/WRITE requests. */
3776 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3777 lim.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3778
3779 if (sd_validate_min_xfer_size(sdkp))
3780 lim.io_min = logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3781 else
3782 lim.io_min = 0;
3783
3784 /*
3785 * Limit default to SCSI host optimal sector limit if set. There may be
3786 * an impact on performance for when the size of a request exceeds this
3787 * host limit.
3788 */
3789 lim.io_opt = sdp->host->opt_sectors << SECTOR_SHIFT;
3790 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3791 lim.io_opt = min_not_zero(lim.io_opt,
3792 logical_to_bytes(sdp, sdkp->opt_xfer_blocks));
3793 }
3794
3795 sdkp->first_scan = 0;
3796
3797 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3798 sd_config_write_same(sdkp, &lim);
3799 kfree(buffer);
3800
3801 err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
3802 if (err)
3803 return err;
3804
3805 /*
3806 * Query concurrent positioning ranges after
3807 * queue_limits_commit_update() unlocked q->limits_lock to avoid
3808 * deadlock with q->sysfs_dir_lock and q->sysfs_lock.
3809 */
3810 if (sdkp->media_present && scsi_device_supports_vpd(sdp))
3811 sd_read_cpr(sdkp);
3812
3813 /*
3814 * For a zoned drive, revalidating the zones can be done only once
3815 * the gendisk capacity is set. So if this fails, set back the gendisk
3816 * capacity to 0.
3817 */
3818 if (sd_zbc_revalidate_zones(sdkp))
3819 set_capacity_and_notify(disk, 0);
3820
3821 out:
3822 return 0;
3823}
3824
3825/**
3826 * sd_unlock_native_capacity - unlock native capacity
3827 * @disk: struct gendisk to set capacity for
3828 *
3829 * Block layer calls this function if it detects that partitions
3830 * on @disk reach beyond the end of the device. If the SCSI host
3831 * implements ->unlock_native_capacity() method, it's invoked to
3832 * give it a chance to adjust the device capacity.
3833 *
3834 * CONTEXT:
3835 * Defined by block layer. Might sleep.
3836 */
3837static void sd_unlock_native_capacity(struct gendisk *disk)
3838{
3839 struct scsi_device *sdev = scsi_disk(disk)->device;
3840
3841 if (sdev->host->hostt->unlock_native_capacity)
3842 sdev->host->hostt->unlock_native_capacity(sdev);
3843}
3844
3845/**
3846 * sd_format_disk_name - format disk name
3847 * @prefix: name prefix - ie. "sd" for SCSI disks
3848 * @index: index of the disk to format name for
3849 * @buf: output buffer
3850 * @buflen: length of the output buffer
3851 *
3852 * SCSI disk names starts at sda. The 26th device is sdz and the
3853 * 27th is sdaa. The last one for two lettered suffix is sdzz
3854 * which is followed by sdaaa.
3855 *
3856 * This is basically 26 base counting with one extra 'nil' entry
3857 * at the beginning from the second digit on and can be
3858 * determined using similar method as 26 base conversion with the
3859 * index shifted -1 after each digit is computed.
3860 *
3861 * CONTEXT:
3862 * Don't care.
3863 *
3864 * RETURNS:
3865 * 0 on success, -errno on failure.
3866 */
3867static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3868{
3869 const int base = 'z' - 'a' + 1;
3870 char *begin = buf + strlen(prefix);
3871 char *end = buf + buflen;
3872 char *p;
3873 int unit;
3874
3875 p = end - 1;
3876 *p = '\0';
3877 unit = base;
3878 do {
3879 if (p == begin)
3880 return -EINVAL;
3881 *--p = 'a' + (index % unit);
3882 index = (index / unit) - 1;
3883 } while (index >= 0);
3884
3885 memmove(begin, p, end - p);
3886 memcpy(buf, prefix, strlen(prefix));
3887
3888 return 0;
3889}
3890
3891/**
3892 * sd_probe - called during driver initialization and whenever a
3893 * new scsi device is attached to the system. It is called once
3894 * for each scsi device (not just disks) present.
3895 * @dev: pointer to device object
3896 *
3897 * Returns 0 if successful (or not interested in this scsi device
3898 * (e.g. scanner)); 1 when there is an error.
3899 *
3900 * Note: this function is invoked from the scsi mid-level.
3901 * This function sets up the mapping between a given
3902 * <host,channel,id,lun> (found in sdp) and new device name
3903 * (e.g. /dev/sda). More precisely it is the block device major
3904 * and minor number that is chosen here.
3905 *
3906 * Assume sd_probe is not re-entrant (for time being)
3907 * Also think about sd_probe() and sd_remove() running coincidentally.
3908 **/
3909static int sd_probe(struct device *dev)
3910{
3911 struct scsi_device *sdp = to_scsi_device(dev);
3912 struct scsi_disk *sdkp;
3913 struct gendisk *gd;
3914 int index;
3915 int error;
3916
3917 scsi_autopm_get_device(sdp);
3918 error = -ENODEV;
3919 if (sdp->type != TYPE_DISK &&
3920 sdp->type != TYPE_ZBC &&
3921 sdp->type != TYPE_MOD &&
3922 sdp->type != TYPE_RBC)
3923 goto out;
3924
3925 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3926 sdev_printk(KERN_WARNING, sdp,
3927 "Unsupported ZBC host-managed device.\n");
3928 goto out;
3929 }
3930
3931 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3932 "sd_probe\n"));
3933
3934 error = -ENOMEM;
3935 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3936 if (!sdkp)
3937 goto out;
3938
3939 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3940 &sd_bio_compl_lkclass);
3941 if (!gd)
3942 goto out_free;
3943
3944 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3945 if (index < 0) {
3946 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3947 goto out_put;
3948 }
3949
3950 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3951 if (error) {
3952 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3953 goto out_free_index;
3954 }
3955
3956 sdkp->device = sdp;
3957 sdkp->disk = gd;
3958 sdkp->index = index;
3959 sdkp->max_retries = SD_MAX_RETRIES;
3960 atomic_set(&sdkp->openers, 0);
3961 atomic_set(&sdkp->device->ioerr_cnt, 0);
3962
3963 if (!sdp->request_queue->rq_timeout) {
3964 if (sdp->type != TYPE_MOD)
3965 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3966 else
3967 blk_queue_rq_timeout(sdp->request_queue,
3968 SD_MOD_TIMEOUT);
3969 }
3970
3971 device_initialize(&sdkp->disk_dev);
3972 sdkp->disk_dev.parent = get_device(dev);
3973 sdkp->disk_dev.class = &sd_disk_class;
3974 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3975
3976 error = device_add(&sdkp->disk_dev);
3977 if (error) {
3978 put_device(&sdkp->disk_dev);
3979 goto out;
3980 }
3981
3982 dev_set_drvdata(dev, sdkp);
3983
3984 gd->major = sd_major((index & 0xf0) >> 4);
3985 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3986 gd->minors = SD_MINORS;
3987
3988 gd->fops = &sd_fops;
3989 gd->private_data = sdkp;
3990
3991 /* defaults, until the device tells us otherwise */
3992 sdp->sector_size = 512;
3993 sdkp->capacity = 0;
3994 sdkp->media_present = 1;
3995 sdkp->write_prot = 0;
3996 sdkp->cache_override = 0;
3997 sdkp->WCE = 0;
3998 sdkp->RCD = 0;
3999 sdkp->ATO = 0;
4000 sdkp->first_scan = 1;
4001 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
4002
4003 sd_revalidate_disk(gd);
4004
4005 if (sdp->removable) {
4006 gd->flags |= GENHD_FL_REMOVABLE;
4007 gd->events |= DISK_EVENT_MEDIA_CHANGE;
4008 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
4009 }
4010
4011 blk_pm_runtime_init(sdp->request_queue, dev);
4012 if (sdp->rpm_autosuspend) {
4013 pm_runtime_set_autosuspend_delay(dev,
4014 sdp->host->rpm_autosuspend_delay);
4015 }
4016
4017 error = device_add_disk(dev, gd, NULL);
4018 if (error) {
4019 device_unregister(&sdkp->disk_dev);
4020 put_disk(gd);
4021 goto out;
4022 }
4023
4024 if (sdkp->security) {
4025 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
4026 if (sdkp->opal_dev)
4027 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
4028 }
4029
4030 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
4031 sdp->removable ? "removable " : "");
4032 scsi_autopm_put_device(sdp);
4033
4034 return 0;
4035
4036 out_free_index:
4037 ida_free(&sd_index_ida, index);
4038 out_put:
4039 put_disk(gd);
4040 out_free:
4041 kfree(sdkp);
4042 out:
4043 scsi_autopm_put_device(sdp);
4044 return error;
4045}
4046
4047/**
4048 * sd_remove - called whenever a scsi disk (previously recognized by
4049 * sd_probe) is detached from the system. It is called (potentially
4050 * multiple times) during sd module unload.
4051 * @dev: pointer to device object
4052 *
4053 * Note: this function is invoked from the scsi mid-level.
4054 * This function potentially frees up a device name (e.g. /dev/sdc)
4055 * that could be re-used by a subsequent sd_probe().
4056 * This function is not called when the built-in sd driver is "exit-ed".
4057 **/
4058static int sd_remove(struct device *dev)
4059{
4060 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4061
4062 scsi_autopm_get_device(sdkp->device);
4063
4064 device_del(&sdkp->disk_dev);
4065 del_gendisk(sdkp->disk);
4066 if (!sdkp->suspended)
4067 sd_shutdown(dev);
4068
4069 put_disk(sdkp->disk);
4070 return 0;
4071}
4072
4073static void scsi_disk_release(struct device *dev)
4074{
4075 struct scsi_disk *sdkp = to_scsi_disk(dev);
4076
4077 ida_free(&sd_index_ida, sdkp->index);
4078 put_device(&sdkp->device->sdev_gendev);
4079 free_opal_dev(sdkp->opal_dev);
4080
4081 kfree(sdkp);
4082}
4083
4084static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
4085{
4086 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
4087 struct scsi_sense_hdr sshdr;
4088 struct scsi_failure failure_defs[] = {
4089 {
4090 /* Power on, reset, or bus device reset occurred */
4091 .sense = UNIT_ATTENTION,
4092 .asc = 0x29,
4093 .ascq = 0,
4094 .result = SAM_STAT_CHECK_CONDITION,
4095 },
4096 {
4097 /* Power on occurred */
4098 .sense = UNIT_ATTENTION,
4099 .asc = 0x29,
4100 .ascq = 1,
4101 .result = SAM_STAT_CHECK_CONDITION,
4102 },
4103 {
4104 /* SCSI bus reset */
4105 .sense = UNIT_ATTENTION,
4106 .asc = 0x29,
4107 .ascq = 2,
4108 .result = SAM_STAT_CHECK_CONDITION,
4109 },
4110 {}
4111 };
4112 struct scsi_failures failures = {
4113 .total_allowed = 3,
4114 .failure_definitions = failure_defs,
4115 };
4116 const struct scsi_exec_args exec_args = {
4117 .sshdr = &sshdr,
4118 .req_flags = BLK_MQ_REQ_PM,
4119 .failures = &failures,
4120 };
4121 struct scsi_device *sdp = sdkp->device;
4122 int res;
4123
4124 if (start)
4125 cmd[4] |= 1; /* START */
4126
4127 if (sdp->start_stop_pwr_cond)
4128 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
4129
4130 if (!scsi_device_online(sdp))
4131 return -ENODEV;
4132
4133 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4134 sdkp->max_retries, &exec_args);
4135 if (res) {
4136 sd_print_result(sdkp, "Start/Stop Unit failed", res);
4137 if (res > 0 && scsi_sense_valid(&sshdr)) {
4138 sd_print_sense_hdr(sdkp, &sshdr);
4139 /* 0x3a is medium not present */
4140 if (sshdr.asc == 0x3a)
4141 res = 0;
4142 }
4143 }
4144
4145 /* SCSI error codes must not go to the generic layer */
4146 if (res)
4147 return -EIO;
4148
4149 return 0;
4150}
4151
4152/*
4153 * Send a SYNCHRONIZE CACHE instruction down to the device through
4154 * the normal SCSI command structure. Wait for the command to
4155 * complete.
4156 */
4157static void sd_shutdown(struct device *dev)
4158{
4159 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4160
4161 if (!sdkp)
4162 return; /* this can happen */
4163
4164 if (pm_runtime_suspended(dev))
4165 return;
4166
4167 if (sdkp->WCE && sdkp->media_present) {
4168 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4169 sd_sync_cache(sdkp);
4170 }
4171
4172 if ((system_state != SYSTEM_RESTART &&
4173 sdkp->device->manage_system_start_stop) ||
4174 (system_state == SYSTEM_POWER_OFF &&
4175 sdkp->device->manage_shutdown)) {
4176 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4177 sd_start_stop_device(sdkp, 0);
4178 }
4179}
4180
4181static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4182{
4183 return (sdev->manage_system_start_stop && !runtime) ||
4184 (sdev->manage_runtime_start_stop && runtime);
4185}
4186
4187static int sd_suspend_common(struct device *dev, bool runtime)
4188{
4189 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4190 int ret = 0;
4191
4192 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
4193 return 0;
4194
4195 if (sdkp->WCE && sdkp->media_present) {
4196 if (!sdkp->device->silence_suspend)
4197 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4198 ret = sd_sync_cache(sdkp);
4199 /* ignore OFFLINE device */
4200 if (ret == -ENODEV)
4201 return 0;
4202
4203 if (ret)
4204 return ret;
4205 }
4206
4207 if (sd_do_start_stop(sdkp->device, runtime)) {
4208 if (!sdkp->device->silence_suspend)
4209 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4210 /* an error is not worth aborting a system sleep */
4211 ret = sd_start_stop_device(sdkp, 0);
4212 if (!runtime)
4213 ret = 0;
4214 }
4215
4216 if (!ret)
4217 sdkp->suspended = true;
4218
4219 return ret;
4220}
4221
4222static int sd_suspend_system(struct device *dev)
4223{
4224 if (pm_runtime_suspended(dev))
4225 return 0;
4226
4227 return sd_suspend_common(dev, false);
4228}
4229
4230static int sd_suspend_runtime(struct device *dev)
4231{
4232 return sd_suspend_common(dev, true);
4233}
4234
4235static int sd_resume(struct device *dev)
4236{
4237 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4238
4239 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4240
4241 if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4242 sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4243 return -EIO;
4244 }
4245
4246 return 0;
4247}
4248
4249static int sd_resume_common(struct device *dev, bool runtime)
4250{
4251 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4252 int ret;
4253
4254 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4255 return 0;
4256
4257 if (!sd_do_start_stop(sdkp->device, runtime)) {
4258 sdkp->suspended = false;
4259 return 0;
4260 }
4261
4262 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4263 ret = sd_start_stop_device(sdkp, 1);
4264 if (!ret) {
4265 sd_resume(dev);
4266 sdkp->suspended = false;
4267 }
4268
4269 return ret;
4270}
4271
4272static int sd_resume_system(struct device *dev)
4273{
4274 if (pm_runtime_suspended(dev)) {
4275 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4276 struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4277
4278 if (sdp && sdp->force_runtime_start_on_system_start)
4279 pm_request_resume(dev);
4280
4281 return 0;
4282 }
4283
4284 return sd_resume_common(dev, false);
4285}
4286
4287static int sd_resume_runtime(struct device *dev)
4288{
4289 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4290 struct scsi_device *sdp;
4291
4292 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4293 return 0;
4294
4295 sdp = sdkp->device;
4296
4297 if (sdp->ignore_media_change) {
4298 /* clear the device's sense data */
4299 static const u8 cmd[10] = { REQUEST_SENSE };
4300 const struct scsi_exec_args exec_args = {
4301 .req_flags = BLK_MQ_REQ_PM,
4302 };
4303
4304 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4305 sdp->request_queue->rq_timeout, 1,
4306 &exec_args))
4307 sd_printk(KERN_NOTICE, sdkp,
4308 "Failed to clear sense data\n");
4309 }
4310
4311 return sd_resume_common(dev, true);
4312}
4313
4314static const struct dev_pm_ops sd_pm_ops = {
4315 .suspend = sd_suspend_system,
4316 .resume = sd_resume_system,
4317 .poweroff = sd_suspend_system,
4318 .restore = sd_resume_system,
4319 .runtime_suspend = sd_suspend_runtime,
4320 .runtime_resume = sd_resume_runtime,
4321};
4322
4323static struct scsi_driver sd_template = {
4324 .gendrv = {
4325 .name = "sd",
4326 .probe = sd_probe,
4327 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
4328 .remove = sd_remove,
4329 .shutdown = sd_shutdown,
4330 .pm = &sd_pm_ops,
4331 },
4332 .rescan = sd_rescan,
4333 .resume = sd_resume,
4334 .init_command = sd_init_command,
4335 .uninit_command = sd_uninit_command,
4336 .done = sd_done,
4337 .eh_action = sd_eh_action,
4338 .eh_reset = sd_eh_reset,
4339};
4340
4341/**
4342 * init_sd - entry point for this driver (both when built in or when
4343 * a module).
4344 *
4345 * Note: this function registers this driver with the scsi mid-level.
4346 **/
4347static int __init init_sd(void)
4348{
4349 int majors = 0, i, err;
4350
4351 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4352
4353 for (i = 0; i < SD_MAJORS; i++) {
4354 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4355 continue;
4356 majors++;
4357 }
4358
4359 if (!majors)
4360 return -ENODEV;
4361
4362 err = class_register(&sd_disk_class);
4363 if (err)
4364 goto err_out;
4365
4366 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4367 if (!sd_page_pool) {
4368 printk(KERN_ERR "sd: can't init discard page pool\n");
4369 err = -ENOMEM;
4370 goto err_out_class;
4371 }
4372
4373 err = scsi_register_driver(&sd_template.gendrv);
4374 if (err)
4375 goto err_out_driver;
4376
4377 return 0;
4378
4379err_out_driver:
4380 mempool_destroy(sd_page_pool);
4381err_out_class:
4382 class_unregister(&sd_disk_class);
4383err_out:
4384 for (i = 0; i < SD_MAJORS; i++)
4385 unregister_blkdev(sd_major(i), "sd");
4386 return err;
4387}
4388
4389/**
4390 * exit_sd - exit point for this driver (when it is a module).
4391 *
4392 * Note: this function unregisters this driver from the scsi mid-level.
4393 **/
4394static void __exit exit_sd(void)
4395{
4396 int i;
4397
4398 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4399
4400 scsi_unregister_driver(&sd_template.gendrv);
4401 mempool_destroy(sd_page_pool);
4402
4403 class_unregister(&sd_disk_class);
4404
4405 for (i = 0; i < SD_MAJORS; i++)
4406 unregister_blkdev(sd_major(i), "sd");
4407}
4408
4409module_init(init_sd);
4410module_exit(exit_sd);
4411
4412void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4413{
4414 scsi_print_sense_hdr(sdkp->device,
4415 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4416}
4417
4418void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4419{
4420 const char *hb_string = scsi_hostbyte_string(result);
4421
4422 if (hb_string)
4423 sd_printk(KERN_INFO, sdkp,
4424 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4425 hb_string ? hb_string : "invalid",
4426 "DRIVER_OK");
4427 else
4428 sd_printk(KERN_INFO, sdkp,
4429 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4430 msg, host_byte(result), "DRIVER_OK");
4431}