1 // SPDX-License-Identifier: GPL-2.0-only
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
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.
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.
36 #include <linux/module.h>
38 #include <linux/kernel.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR
);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR
);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR
);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR
);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR
);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR
);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR
);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR
);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR
);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR
);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR
);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR
);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR
);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR
);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR
);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR
);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK
);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD
);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC
);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC
);
103 static void sd_config_discard(struct scsi_disk
*, unsigned int);
104 static void sd_config_write_same(struct scsi_disk
*);
105 static int sd_revalidate_disk(struct gendisk
*);
106 static void sd_unlock_native_capacity(struct gendisk
*disk
);
107 static int sd_probe(struct device
*);
108 static int sd_remove(struct device
*);
109 static void sd_shutdown(struct device
*);
110 static int sd_suspend_system(struct device
*);
111 static int sd_suspend_runtime(struct device
*);
112 static int sd_resume(struct device
*);
113 static int sd_resume_runtime(struct device
*);
114 static void sd_rescan(struct device
*);
115 static blk_status_t
sd_init_command(struct scsi_cmnd
*SCpnt
);
116 static void sd_uninit_command(struct scsi_cmnd
*SCpnt
);
117 static int sd_done(struct scsi_cmnd
*);
118 static void sd_eh_reset(struct scsi_cmnd
*);
119 static int sd_eh_action(struct scsi_cmnd
*, int);
120 static void sd_read_capacity(struct scsi_disk
*sdkp
, unsigned char *buffer
);
121 static void scsi_disk_release(struct device
*cdev
);
123 static DEFINE_IDA(sd_index_ida
);
125 /* This semaphore is used to mediate the 0->1 reference get in the
126 * face of object destruction (i.e. we can't allow a get on an
127 * object after last put) */
128 static DEFINE_MUTEX(sd_ref_mutex
);
130 static struct kmem_cache
*sd_cdb_cache
;
131 static mempool_t
*sd_cdb_pool
;
132 static mempool_t
*sd_page_pool
;
133 static struct lock_class_key sd_bio_compl_lkclass
;
135 static const char *sd_cache_types
[] = {
136 "write through", "none", "write back",
137 "write back, no read (daft)"
140 static void sd_set_flush_flag(struct scsi_disk
*sdkp
)
142 bool wc
= false, fua
= false;
150 blk_queue_write_cache(sdkp
->disk
->queue
, wc
, fua
);
154 cache_type_store(struct device
*dev
, struct device_attribute
*attr
,
155 const char *buf
, size_t count
)
157 int ct
, rcd
, wce
, sp
;
158 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
159 struct scsi_device
*sdp
= sdkp
->device
;
162 struct scsi_mode_data data
;
163 struct scsi_sense_hdr sshdr
;
164 static const char temp
[] = "temporary ";
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 */
173 if (strncmp(buf
, temp
, sizeof(temp
) - 1) == 0) {
174 buf
+= sizeof(temp
) - 1;
175 sdkp
->cache_override
= 1;
177 sdkp
->cache_override
= 0;
180 ct
= sysfs_match_string(sd_cache_types
, buf
);
184 rcd
= ct
& 0x01 ? 1 : 0;
185 wce
= (ct
& 0x02) && !sdkp
->write_prot
? 1 : 0;
187 if (sdkp
->cache_override
) {
190 sd_set_flush_flag(sdkp
);
194 if (scsi_mode_sense(sdp
, 0x08, 8, buffer
, sizeof(buffer
), SD_TIMEOUT
,
195 sdkp
->max_retries
, &data
, NULL
))
197 len
= min_t(size_t, sizeof(buffer
), data
.length
- data
.header_length
-
198 data
.block_descriptor_length
);
199 buffer_data
= buffer
+ data
.header_length
+
200 data
.block_descriptor_length
;
201 buffer_data
[2] &= ~0x05;
202 buffer_data
[2] |= wce
<< 2 | rcd
;
203 sp
= buffer_data
[0] & 0x80 ? 1 : 0;
204 buffer_data
[0] &= ~0x80;
207 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
208 * received mode parameter buffer before doing MODE SELECT.
210 data
.device_specific
= 0;
212 if (scsi_mode_select(sdp
, 1, sp
, 8, buffer_data
, len
, SD_TIMEOUT
,
213 sdkp
->max_retries
, &data
, &sshdr
)) {
214 if (scsi_sense_valid(&sshdr
))
215 sd_print_sense_hdr(sdkp
, &sshdr
);
218 sd_revalidate_disk(sdkp
->disk
);
223 manage_start_stop_show(struct device
*dev
, struct device_attribute
*attr
,
226 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
227 struct scsi_device
*sdp
= sdkp
->device
;
229 return sprintf(buf
, "%u\n", sdp
->manage_start_stop
);
233 manage_start_stop_store(struct device
*dev
, struct device_attribute
*attr
,
234 const char *buf
, size_t count
)
236 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
237 struct scsi_device
*sdp
= sdkp
->device
;
240 if (!capable(CAP_SYS_ADMIN
))
243 if (kstrtobool(buf
, &v
))
246 sdp
->manage_start_stop
= v
;
250 static DEVICE_ATTR_RW(manage_start_stop
);
253 allow_restart_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
255 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
257 return sprintf(buf
, "%u\n", sdkp
->device
->allow_restart
);
261 allow_restart_store(struct device
*dev
, struct device_attribute
*attr
,
262 const char *buf
, size_t count
)
265 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
266 struct scsi_device
*sdp
= sdkp
->device
;
268 if (!capable(CAP_SYS_ADMIN
))
271 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_ZBC
)
274 if (kstrtobool(buf
, &v
))
277 sdp
->allow_restart
= v
;
281 static DEVICE_ATTR_RW(allow_restart
);
284 cache_type_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
286 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
287 int ct
= sdkp
->RCD
+ 2*sdkp
->WCE
;
289 return sprintf(buf
, "%s\n", sd_cache_types
[ct
]);
291 static DEVICE_ATTR_RW(cache_type
);
294 FUA_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
296 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
298 return sprintf(buf
, "%u\n", sdkp
->DPOFUA
);
300 static DEVICE_ATTR_RO(FUA
);
303 protection_type_show(struct device
*dev
, struct device_attribute
*attr
,
306 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
308 return sprintf(buf
, "%u\n", sdkp
->protection_type
);
312 protection_type_store(struct device
*dev
, struct device_attribute
*attr
,
313 const char *buf
, size_t count
)
315 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
319 if (!capable(CAP_SYS_ADMIN
))
322 err
= kstrtouint(buf
, 10, &val
);
327 if (val
<= T10_PI_TYPE3_PROTECTION
)
328 sdkp
->protection_type
= val
;
332 static DEVICE_ATTR_RW(protection_type
);
335 protection_mode_show(struct device
*dev
, struct device_attribute
*attr
,
338 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
339 struct scsi_device
*sdp
= sdkp
->device
;
340 unsigned int dif
, dix
;
342 dif
= scsi_host_dif_capable(sdp
->host
, sdkp
->protection_type
);
343 dix
= scsi_host_dix_capable(sdp
->host
, sdkp
->protection_type
);
345 if (!dix
&& scsi_host_dix_capable(sdp
->host
, T10_PI_TYPE0_PROTECTION
)) {
351 return sprintf(buf
, "none\n");
353 return sprintf(buf
, "%s%u\n", dix
? "dix" : "dif", dif
);
355 static DEVICE_ATTR_RO(protection_mode
);
358 app_tag_own_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
360 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
362 return sprintf(buf
, "%u\n", sdkp
->ATO
);
364 static DEVICE_ATTR_RO(app_tag_own
);
367 thin_provisioning_show(struct device
*dev
, struct device_attribute
*attr
,
370 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
372 return sprintf(buf
, "%u\n", sdkp
->lbpme
);
374 static DEVICE_ATTR_RO(thin_provisioning
);
376 /* sysfs_match_string() requires dense arrays */
377 static const char *lbp_mode
[] = {
378 [SD_LBP_FULL
] = "full",
379 [SD_LBP_UNMAP
] = "unmap",
380 [SD_LBP_WS16
] = "writesame_16",
381 [SD_LBP_WS10
] = "writesame_10",
382 [SD_LBP_ZERO
] = "writesame_zero",
383 [SD_LBP_DISABLE
] = "disabled",
387 provisioning_mode_show(struct device
*dev
, struct device_attribute
*attr
,
390 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
392 return sprintf(buf
, "%s\n", lbp_mode
[sdkp
->provisioning_mode
]);
396 provisioning_mode_store(struct device
*dev
, struct device_attribute
*attr
,
397 const char *buf
, size_t count
)
399 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
400 struct scsi_device
*sdp
= sdkp
->device
;
403 if (!capable(CAP_SYS_ADMIN
))
406 if (sd_is_zoned(sdkp
)) {
407 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
411 if (sdp
->type
!= TYPE_DISK
)
414 mode
= sysfs_match_string(lbp_mode
, buf
);
418 sd_config_discard(sdkp
, mode
);
422 static DEVICE_ATTR_RW(provisioning_mode
);
424 /* sysfs_match_string() requires dense arrays */
425 static const char *zeroing_mode
[] = {
426 [SD_ZERO_WRITE
] = "write",
427 [SD_ZERO_WS
] = "writesame",
428 [SD_ZERO_WS16_UNMAP
] = "writesame_16_unmap",
429 [SD_ZERO_WS10_UNMAP
] = "writesame_10_unmap",
433 zeroing_mode_show(struct device
*dev
, struct device_attribute
*attr
,
436 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
438 return sprintf(buf
, "%s\n", zeroing_mode
[sdkp
->zeroing_mode
]);
442 zeroing_mode_store(struct device
*dev
, struct device_attribute
*attr
,
443 const char *buf
, size_t count
)
445 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
448 if (!capable(CAP_SYS_ADMIN
))
451 mode
= sysfs_match_string(zeroing_mode
, buf
);
455 sdkp
->zeroing_mode
= mode
;
459 static DEVICE_ATTR_RW(zeroing_mode
);
462 max_medium_access_timeouts_show(struct device
*dev
,
463 struct device_attribute
*attr
, char *buf
)
465 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
467 return sprintf(buf
, "%u\n", sdkp
->max_medium_access_timeouts
);
471 max_medium_access_timeouts_store(struct device
*dev
,
472 struct device_attribute
*attr
, const char *buf
,
475 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
478 if (!capable(CAP_SYS_ADMIN
))
481 err
= kstrtouint(buf
, 10, &sdkp
->max_medium_access_timeouts
);
483 return err
? err
: count
;
485 static DEVICE_ATTR_RW(max_medium_access_timeouts
);
488 max_write_same_blocks_show(struct device
*dev
, struct device_attribute
*attr
,
491 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
493 return sprintf(buf
, "%u\n", sdkp
->max_ws_blocks
);
497 max_write_same_blocks_store(struct device
*dev
, struct device_attribute
*attr
,
498 const char *buf
, size_t count
)
500 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
501 struct scsi_device
*sdp
= sdkp
->device
;
505 if (!capable(CAP_SYS_ADMIN
))
508 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_ZBC
)
511 err
= kstrtoul(buf
, 10, &max
);
517 sdp
->no_write_same
= 1;
518 else if (max
<= SD_MAX_WS16_BLOCKS
) {
519 sdp
->no_write_same
= 0;
520 sdkp
->max_ws_blocks
= max
;
523 sd_config_write_same(sdkp
);
527 static DEVICE_ATTR_RW(max_write_same_blocks
);
530 zoned_cap_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
532 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
534 if (sdkp
->device
->type
== TYPE_ZBC
)
535 return sprintf(buf
, "host-managed\n");
536 if (sdkp
->zoned
== 1)
537 return sprintf(buf
, "host-aware\n");
538 if (sdkp
->zoned
== 2)
539 return sprintf(buf
, "drive-managed\n");
540 return sprintf(buf
, "none\n");
542 static DEVICE_ATTR_RO(zoned_cap
);
545 max_retries_store(struct device
*dev
, struct device_attribute
*attr
,
546 const char *buf
, size_t count
)
548 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
549 struct scsi_device
*sdev
= sdkp
->device
;
552 err
= kstrtoint(buf
, 10, &retries
);
556 if (retries
== SCSI_CMD_RETRIES_NO_LIMIT
|| retries
<= SD_MAX_RETRIES
) {
557 sdkp
->max_retries
= retries
;
561 sdev_printk(KERN_ERR
, sdev
, "max_retries must be between -1 and %d\n",
567 max_retries_show(struct device
*dev
, struct device_attribute
*attr
,
570 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
572 return sprintf(buf
, "%d\n", sdkp
->max_retries
);
575 static DEVICE_ATTR_RW(max_retries
);
577 static struct attribute
*sd_disk_attrs
[] = {
578 &dev_attr_cache_type
.attr
,
580 &dev_attr_allow_restart
.attr
,
581 &dev_attr_manage_start_stop
.attr
,
582 &dev_attr_protection_type
.attr
,
583 &dev_attr_protection_mode
.attr
,
584 &dev_attr_app_tag_own
.attr
,
585 &dev_attr_thin_provisioning
.attr
,
586 &dev_attr_provisioning_mode
.attr
,
587 &dev_attr_zeroing_mode
.attr
,
588 &dev_attr_max_write_same_blocks
.attr
,
589 &dev_attr_max_medium_access_timeouts
.attr
,
590 &dev_attr_zoned_cap
.attr
,
591 &dev_attr_max_retries
.attr
,
594 ATTRIBUTE_GROUPS(sd_disk
);
596 static struct class sd_disk_class
= {
598 .owner
= THIS_MODULE
,
599 .dev_release
= scsi_disk_release
,
600 .dev_groups
= sd_disk_groups
,
603 static const struct dev_pm_ops sd_pm_ops
= {
604 .suspend
= sd_suspend_system
,
606 .poweroff
= sd_suspend_system
,
607 .restore
= sd_resume
,
608 .runtime_suspend
= sd_suspend_runtime
,
609 .runtime_resume
= sd_resume_runtime
,
612 static struct scsi_driver sd_template
= {
615 .owner
= THIS_MODULE
,
617 .probe_type
= PROBE_PREFER_ASYNCHRONOUS
,
619 .shutdown
= sd_shutdown
,
623 .init_command
= sd_init_command
,
624 .uninit_command
= sd_uninit_command
,
626 .eh_action
= sd_eh_action
,
627 .eh_reset
= sd_eh_reset
,
631 * Don't request a new module, as that could deadlock in multipath
634 static void sd_default_probe(dev_t devt
)
639 * Device no to disk mapping:
641 * major disc2 disc p1
642 * |............|.............|....|....| <- dev_t
645 * Inside a major, we have 16k disks, however mapped non-
646 * contiguously. The first 16 disks are for major0, the next
647 * ones with major1, ... Disk 256 is for major0 again, disk 272
649 * As we stay compatible with our numbering scheme, we can reuse
650 * the well-know SCSI majors 8, 65--71, 136--143.
652 static int sd_major(int major_idx
)
656 return SCSI_DISK0_MAJOR
;
658 return SCSI_DISK1_MAJOR
+ major_idx
- 1;
660 return SCSI_DISK8_MAJOR
+ major_idx
- 8;
663 return 0; /* shut up gcc */
667 static struct scsi_disk
*scsi_disk_get(struct gendisk
*disk
)
669 struct scsi_disk
*sdkp
= NULL
;
671 mutex_lock(&sd_ref_mutex
);
673 if (disk
->private_data
) {
674 sdkp
= scsi_disk(disk
);
675 if (scsi_device_get(sdkp
->device
) == 0)
676 get_device(&sdkp
->dev
);
680 mutex_unlock(&sd_ref_mutex
);
684 static void scsi_disk_put(struct scsi_disk
*sdkp
)
686 struct scsi_device
*sdev
= sdkp
->device
;
688 mutex_lock(&sd_ref_mutex
);
689 put_device(&sdkp
->dev
);
690 scsi_device_put(sdev
);
691 mutex_unlock(&sd_ref_mutex
);
694 #ifdef CONFIG_BLK_SED_OPAL
695 static int sd_sec_submit(void *data
, u16 spsp
, u8 secp
, void *buffer
,
696 size_t len
, bool send
)
698 struct scsi_disk
*sdkp
= data
;
699 struct scsi_device
*sdev
= sdkp
->device
;
703 cdb
[0] = send
? SECURITY_PROTOCOL_OUT
: SECURITY_PROTOCOL_IN
;
705 put_unaligned_be16(spsp
, &cdb
[2]);
706 put_unaligned_be32(len
, &cdb
[6]);
708 ret
= scsi_execute(sdev
, cdb
, send
? DMA_TO_DEVICE
: DMA_FROM_DEVICE
,
709 buffer
, len
, NULL
, NULL
, SD_TIMEOUT
, sdkp
->max_retries
, 0,
711 return ret
<= 0 ? ret
: -EIO
;
713 #endif /* CONFIG_BLK_SED_OPAL */
716 * Look up the DIX operation based on whether the command is read or
717 * write and whether dix and dif are enabled.
719 static unsigned int sd_prot_op(bool write
, bool dix
, bool dif
)
721 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
722 static const unsigned int ops
[] = { /* wrt dix dif */
723 SCSI_PROT_NORMAL
, /* 0 0 0 */
724 SCSI_PROT_READ_STRIP
, /* 0 0 1 */
725 SCSI_PROT_READ_INSERT
, /* 0 1 0 */
726 SCSI_PROT_READ_PASS
, /* 0 1 1 */
727 SCSI_PROT_NORMAL
, /* 1 0 0 */
728 SCSI_PROT_WRITE_INSERT
, /* 1 0 1 */
729 SCSI_PROT_WRITE_STRIP
, /* 1 1 0 */
730 SCSI_PROT_WRITE_PASS
, /* 1 1 1 */
733 return ops
[write
<< 2 | dix
<< 1 | dif
];
737 * Returns a mask of the protection flags that are valid for a given DIX
740 static unsigned int sd_prot_flag_mask(unsigned int prot_op
)
742 static const unsigned int flag_mask
[] = {
743 [SCSI_PROT_NORMAL
] = 0,
745 [SCSI_PROT_READ_STRIP
] = SCSI_PROT_TRANSFER_PI
|
746 SCSI_PROT_GUARD_CHECK
|
747 SCSI_PROT_REF_CHECK
|
748 SCSI_PROT_REF_INCREMENT
,
750 [SCSI_PROT_READ_INSERT
] = SCSI_PROT_REF_INCREMENT
|
751 SCSI_PROT_IP_CHECKSUM
,
753 [SCSI_PROT_READ_PASS
] = SCSI_PROT_TRANSFER_PI
|
754 SCSI_PROT_GUARD_CHECK
|
755 SCSI_PROT_REF_CHECK
|
756 SCSI_PROT_REF_INCREMENT
|
757 SCSI_PROT_IP_CHECKSUM
,
759 [SCSI_PROT_WRITE_INSERT
] = SCSI_PROT_TRANSFER_PI
|
760 SCSI_PROT_REF_INCREMENT
,
762 [SCSI_PROT_WRITE_STRIP
] = SCSI_PROT_GUARD_CHECK
|
763 SCSI_PROT_REF_CHECK
|
764 SCSI_PROT_REF_INCREMENT
|
765 SCSI_PROT_IP_CHECKSUM
,
767 [SCSI_PROT_WRITE_PASS
] = SCSI_PROT_TRANSFER_PI
|
768 SCSI_PROT_GUARD_CHECK
|
769 SCSI_PROT_REF_CHECK
|
770 SCSI_PROT_REF_INCREMENT
|
771 SCSI_PROT_IP_CHECKSUM
,
774 return flag_mask
[prot_op
];
777 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd
*scmd
,
778 unsigned int dix
, unsigned int dif
)
780 struct request
*rq
= scsi_cmd_to_rq(scmd
);
781 struct bio
*bio
= rq
->bio
;
782 unsigned int prot_op
= sd_prot_op(rq_data_dir(rq
), dix
, dif
);
783 unsigned int protect
= 0;
785 if (dix
) { /* DIX Type 0, 1, 2, 3 */
786 if (bio_integrity_flagged(bio
, BIP_IP_CHECKSUM
))
787 scmd
->prot_flags
|= SCSI_PROT_IP_CHECKSUM
;
789 if (bio_integrity_flagged(bio
, BIP_CTRL_NOCHECK
) == false)
790 scmd
->prot_flags
|= SCSI_PROT_GUARD_CHECK
;
793 if (dif
!= T10_PI_TYPE3_PROTECTION
) { /* DIX/DIF Type 0, 1, 2 */
794 scmd
->prot_flags
|= SCSI_PROT_REF_INCREMENT
;
796 if (bio_integrity_flagged(bio
, BIP_CTRL_NOCHECK
) == false)
797 scmd
->prot_flags
|= SCSI_PROT_REF_CHECK
;
800 if (dif
) { /* DIX/DIF Type 1, 2, 3 */
801 scmd
->prot_flags
|= SCSI_PROT_TRANSFER_PI
;
803 if (bio_integrity_flagged(bio
, BIP_DISK_NOCHECK
))
804 protect
= 3 << 5; /* Disable target PI checking */
806 protect
= 1 << 5; /* Enable target PI checking */
809 scsi_set_prot_op(scmd
, prot_op
);
810 scsi_set_prot_type(scmd
, dif
);
811 scmd
->prot_flags
&= sd_prot_flag_mask(prot_op
);
816 static void sd_config_discard(struct scsi_disk
*sdkp
, unsigned int mode
)
818 struct request_queue
*q
= sdkp
->disk
->queue
;
819 unsigned int logical_block_size
= sdkp
->device
->sector_size
;
820 unsigned int max_blocks
= 0;
822 q
->limits
.discard_alignment
=
823 sdkp
->unmap_alignment
* logical_block_size
;
824 q
->limits
.discard_granularity
=
825 max(sdkp
->physical_block_size
,
826 sdkp
->unmap_granularity
* logical_block_size
);
827 sdkp
->provisioning_mode
= mode
;
833 blk_queue_max_discard_sectors(q
, 0);
834 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, q
);
838 max_blocks
= min_not_zero(sdkp
->max_unmap_blocks
,
839 (u32
)SD_MAX_WS16_BLOCKS
);
843 if (sdkp
->device
->unmap_limit_for_ws
)
844 max_blocks
= sdkp
->max_unmap_blocks
;
846 max_blocks
= sdkp
->max_ws_blocks
;
848 max_blocks
= min_not_zero(max_blocks
, (u32
)SD_MAX_WS16_BLOCKS
);
852 if (sdkp
->device
->unmap_limit_for_ws
)
853 max_blocks
= sdkp
->max_unmap_blocks
;
855 max_blocks
= sdkp
->max_ws_blocks
;
857 max_blocks
= min_not_zero(max_blocks
, (u32
)SD_MAX_WS10_BLOCKS
);
861 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
862 (u32
)SD_MAX_WS10_BLOCKS
);
866 blk_queue_max_discard_sectors(q
, max_blocks
* (logical_block_size
>> 9));
867 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, q
);
870 static blk_status_t
sd_setup_unmap_cmnd(struct scsi_cmnd
*cmd
)
872 struct scsi_device
*sdp
= cmd
->device
;
873 struct request
*rq
= scsi_cmd_to_rq(cmd
);
874 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
875 u64 lba
= sectors_to_logical(sdp
, blk_rq_pos(rq
));
876 u32 nr_blocks
= sectors_to_logical(sdp
, blk_rq_sectors(rq
));
877 unsigned int data_len
= 24;
880 rq
->special_vec
.bv_page
= mempool_alloc(sd_page_pool
, GFP_ATOMIC
);
881 if (!rq
->special_vec
.bv_page
)
882 return BLK_STS_RESOURCE
;
883 clear_highpage(rq
->special_vec
.bv_page
);
884 rq
->special_vec
.bv_offset
= 0;
885 rq
->special_vec
.bv_len
= data_len
;
886 rq
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
889 cmd
->cmnd
[0] = UNMAP
;
892 buf
= bvec_virt(&rq
->special_vec
);
893 put_unaligned_be16(6 + 16, &buf
[0]);
894 put_unaligned_be16(16, &buf
[2]);
895 put_unaligned_be64(lba
, &buf
[8]);
896 put_unaligned_be32(nr_blocks
, &buf
[16]);
898 cmd
->allowed
= sdkp
->max_retries
;
899 cmd
->transfersize
= data_len
;
900 rq
->timeout
= SD_TIMEOUT
;
902 return scsi_alloc_sgtables(cmd
);
905 static blk_status_t
sd_setup_write_same16_cmnd(struct scsi_cmnd
*cmd
,
908 struct scsi_device
*sdp
= cmd
->device
;
909 struct request
*rq
= scsi_cmd_to_rq(cmd
);
910 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
911 u64 lba
= sectors_to_logical(sdp
, blk_rq_pos(rq
));
912 u32 nr_blocks
= sectors_to_logical(sdp
, blk_rq_sectors(rq
));
913 u32 data_len
= sdp
->sector_size
;
915 rq
->special_vec
.bv_page
= mempool_alloc(sd_page_pool
, GFP_ATOMIC
);
916 if (!rq
->special_vec
.bv_page
)
917 return BLK_STS_RESOURCE
;
918 clear_highpage(rq
->special_vec
.bv_page
);
919 rq
->special_vec
.bv_offset
= 0;
920 rq
->special_vec
.bv_len
= data_len
;
921 rq
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
924 cmd
->cmnd
[0] = WRITE_SAME_16
;
926 cmd
->cmnd
[1] = 0x8; /* UNMAP */
927 put_unaligned_be64(lba
, &cmd
->cmnd
[2]);
928 put_unaligned_be32(nr_blocks
, &cmd
->cmnd
[10]);
930 cmd
->allowed
= sdkp
->max_retries
;
931 cmd
->transfersize
= data_len
;
932 rq
->timeout
= unmap
? SD_TIMEOUT
: SD_WRITE_SAME_TIMEOUT
;
934 return scsi_alloc_sgtables(cmd
);
937 static blk_status_t
sd_setup_write_same10_cmnd(struct scsi_cmnd
*cmd
,
940 struct scsi_device
*sdp
= cmd
->device
;
941 struct request
*rq
= scsi_cmd_to_rq(cmd
);
942 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
943 u64 lba
= sectors_to_logical(sdp
, blk_rq_pos(rq
));
944 u32 nr_blocks
= sectors_to_logical(sdp
, blk_rq_sectors(rq
));
945 u32 data_len
= sdp
->sector_size
;
947 rq
->special_vec
.bv_page
= mempool_alloc(sd_page_pool
, GFP_ATOMIC
);
948 if (!rq
->special_vec
.bv_page
)
949 return BLK_STS_RESOURCE
;
950 clear_highpage(rq
->special_vec
.bv_page
);
951 rq
->special_vec
.bv_offset
= 0;
952 rq
->special_vec
.bv_len
= data_len
;
953 rq
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
956 cmd
->cmnd
[0] = WRITE_SAME
;
958 cmd
->cmnd
[1] = 0x8; /* UNMAP */
959 put_unaligned_be32(lba
, &cmd
->cmnd
[2]);
960 put_unaligned_be16(nr_blocks
, &cmd
->cmnd
[7]);
962 cmd
->allowed
= sdkp
->max_retries
;
963 cmd
->transfersize
= data_len
;
964 rq
->timeout
= unmap
? SD_TIMEOUT
: SD_WRITE_SAME_TIMEOUT
;
966 return scsi_alloc_sgtables(cmd
);
969 static blk_status_t
sd_setup_write_zeroes_cmnd(struct scsi_cmnd
*cmd
)
971 struct request
*rq
= scsi_cmd_to_rq(cmd
);
972 struct scsi_device
*sdp
= cmd
->device
;
973 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
974 u64 lba
= sectors_to_logical(sdp
, blk_rq_pos(rq
));
975 u32 nr_blocks
= sectors_to_logical(sdp
, blk_rq_sectors(rq
));
977 if (!(rq
->cmd_flags
& REQ_NOUNMAP
)) {
978 switch (sdkp
->zeroing_mode
) {
979 case SD_ZERO_WS16_UNMAP
:
980 return sd_setup_write_same16_cmnd(cmd
, true);
981 case SD_ZERO_WS10_UNMAP
:
982 return sd_setup_write_same10_cmnd(cmd
, true);
986 if (sdp
->no_write_same
) {
987 rq
->rq_flags
|= RQF_QUIET
;
988 return BLK_STS_TARGET
;
991 if (sdkp
->ws16
|| lba
> 0xffffffff || nr_blocks
> 0xffff)
992 return sd_setup_write_same16_cmnd(cmd
, false);
994 return sd_setup_write_same10_cmnd(cmd
, false);
997 static void sd_config_write_same(struct scsi_disk
*sdkp
)
999 struct request_queue
*q
= sdkp
->disk
->queue
;
1000 unsigned int logical_block_size
= sdkp
->device
->sector_size
;
1002 if (sdkp
->device
->no_write_same
) {
1003 sdkp
->max_ws_blocks
= 0;
1007 /* Some devices can not handle block counts above 0xffff despite
1008 * supporting WRITE SAME(16). Consequently we default to 64k
1009 * blocks per I/O unless the device explicitly advertises a
1012 if (sdkp
->max_ws_blocks
> SD_MAX_WS10_BLOCKS
)
1013 sdkp
->max_ws_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
1014 (u32
)SD_MAX_WS16_BLOCKS
);
1015 else if (sdkp
->ws16
|| sdkp
->ws10
|| sdkp
->device
->no_report_opcodes
)
1016 sdkp
->max_ws_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
1017 (u32
)SD_MAX_WS10_BLOCKS
);
1019 sdkp
->device
->no_write_same
= 1;
1020 sdkp
->max_ws_blocks
= 0;
1023 if (sdkp
->lbprz
&& sdkp
->lbpws
)
1024 sdkp
->zeroing_mode
= SD_ZERO_WS16_UNMAP
;
1025 else if (sdkp
->lbprz
&& sdkp
->lbpws10
)
1026 sdkp
->zeroing_mode
= SD_ZERO_WS10_UNMAP
;
1027 else if (sdkp
->max_ws_blocks
)
1028 sdkp
->zeroing_mode
= SD_ZERO_WS
;
1030 sdkp
->zeroing_mode
= SD_ZERO_WRITE
;
1032 if (sdkp
->max_ws_blocks
&&
1033 sdkp
->physical_block_size
> logical_block_size
) {
1035 * Reporting a maximum number of blocks that is not aligned
1036 * on the device physical size would cause a large write same
1037 * request to be split into physically unaligned chunks by
1038 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1039 * even if the caller of these functions took care to align the
1040 * large request. So make sure the maximum reported is aligned
1041 * to the device physical block size. This is only an optional
1042 * optimization for regular disks, but this is mandatory to
1043 * avoid failure of large write same requests directed at
1044 * sequential write required zones of host-managed ZBC disks.
1046 sdkp
->max_ws_blocks
=
1047 round_down(sdkp
->max_ws_blocks
,
1048 bytes_to_logical(sdkp
->device
,
1049 sdkp
->physical_block_size
));
1053 blk_queue_max_write_same_sectors(q
, sdkp
->max_ws_blocks
*
1054 (logical_block_size
>> 9));
1055 blk_queue_max_write_zeroes_sectors(q
, sdkp
->max_ws_blocks
*
1056 (logical_block_size
>> 9));
1060 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1061 * @cmd: command to prepare
1063 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1064 * the preference indicated by the target device.
1066 static blk_status_t
sd_setup_write_same_cmnd(struct scsi_cmnd
*cmd
)
1068 struct request
*rq
= scsi_cmd_to_rq(cmd
);
1069 struct scsi_device
*sdp
= cmd
->device
;
1070 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
1071 struct bio
*bio
= rq
->bio
;
1072 u64 lba
= sectors_to_logical(sdp
, blk_rq_pos(rq
));
1073 u32 nr_blocks
= sectors_to_logical(sdp
, blk_rq_sectors(rq
));
1076 if (sdkp
->device
->no_write_same
)
1077 return BLK_STS_TARGET
;
1079 BUG_ON(bio_offset(bio
) || bio_iovec(bio
).bv_len
!= sdp
->sector_size
);
1081 rq
->timeout
= SD_WRITE_SAME_TIMEOUT
;
1083 if (sdkp
->ws16
|| lba
> 0xffffffff || nr_blocks
> 0xffff) {
1085 cmd
->cmnd
[0] = WRITE_SAME_16
;
1086 put_unaligned_be64(lba
, &cmd
->cmnd
[2]);
1087 put_unaligned_be32(nr_blocks
, &cmd
->cmnd
[10]);
1090 cmd
->cmnd
[0] = WRITE_SAME
;
1091 put_unaligned_be32(lba
, &cmd
->cmnd
[2]);
1092 put_unaligned_be16(nr_blocks
, &cmd
->cmnd
[7]);
1095 cmd
->transfersize
= sdp
->sector_size
;
1096 cmd
->allowed
= sdkp
->max_retries
;
1099 * For WRITE SAME the data transferred via the DATA OUT buffer is
1100 * different from the amount of data actually written to the target.
1102 * We set up __data_len to the amount of data transferred via the
1103 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1104 * to transfer a single sector of data first, but then reset it to
1105 * the amount of data to be written right after so that the I/O path
1106 * knows how much to actually write.
1108 rq
->__data_len
= sdp
->sector_size
;
1109 ret
= scsi_alloc_sgtables(cmd
);
1110 rq
->__data_len
= blk_rq_bytes(rq
);
1115 static blk_status_t
sd_setup_flush_cmnd(struct scsi_cmnd
*cmd
)
1117 struct request
*rq
= scsi_cmd_to_rq(cmd
);
1118 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
1120 /* flush requests don't perform I/O, zero the S/G table */
1121 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1123 cmd
->cmnd
[0] = SYNCHRONIZE_CACHE
;
1125 cmd
->transfersize
= 0;
1126 cmd
->allowed
= sdkp
->max_retries
;
1128 rq
->timeout
= rq
->q
->rq_timeout
* SD_FLUSH_TIMEOUT_MULTIPLIER
;
1132 static blk_status_t
sd_setup_rw32_cmnd(struct scsi_cmnd
*cmd
, bool write
,
1133 sector_t lba
, unsigned int nr_blocks
,
1134 unsigned char flags
)
1136 cmd
->cmnd
= mempool_alloc(sd_cdb_pool
, GFP_ATOMIC
);
1137 if (unlikely(cmd
->cmnd
== NULL
))
1138 return BLK_STS_RESOURCE
;
1140 cmd
->cmd_len
= SD_EXT_CDB_SIZE
;
1141 memset(cmd
->cmnd
, 0, cmd
->cmd_len
);
1143 cmd
->cmnd
[0] = VARIABLE_LENGTH_CMD
;
1144 cmd
->cmnd
[7] = 0x18; /* Additional CDB len */
1145 cmd
->cmnd
[9] = write
? WRITE_32
: READ_32
;
1146 cmd
->cmnd
[10] = flags
;
1147 put_unaligned_be64(lba
, &cmd
->cmnd
[12]);
1148 put_unaligned_be32(lba
, &cmd
->cmnd
[20]); /* Expected Indirect LBA */
1149 put_unaligned_be32(nr_blocks
, &cmd
->cmnd
[28]);
1154 static blk_status_t
sd_setup_rw16_cmnd(struct scsi_cmnd
*cmd
, bool write
,
1155 sector_t lba
, unsigned int nr_blocks
,
1156 unsigned char flags
)
1159 cmd
->cmnd
[0] = write
? WRITE_16
: READ_16
;
1160 cmd
->cmnd
[1] = flags
;
1163 put_unaligned_be64(lba
, &cmd
->cmnd
[2]);
1164 put_unaligned_be32(nr_blocks
, &cmd
->cmnd
[10]);
1169 static blk_status_t
sd_setup_rw10_cmnd(struct scsi_cmnd
*cmd
, bool write
,
1170 sector_t lba
, unsigned int nr_blocks
,
1171 unsigned char flags
)
1174 cmd
->cmnd
[0] = write
? WRITE_10
: READ_10
;
1175 cmd
->cmnd
[1] = flags
;
1178 put_unaligned_be32(lba
, &cmd
->cmnd
[2]);
1179 put_unaligned_be16(nr_blocks
, &cmd
->cmnd
[7]);
1184 static blk_status_t
sd_setup_rw6_cmnd(struct scsi_cmnd
*cmd
, bool write
,
1185 sector_t lba
, unsigned int nr_blocks
,
1186 unsigned char flags
)
1188 /* Avoid that 0 blocks gets translated into 256 blocks. */
1189 if (WARN_ON_ONCE(nr_blocks
== 0))
1190 return BLK_STS_IOERR
;
1192 if (unlikely(flags
& 0x8)) {
1194 * This happens only if this drive failed 10byte rw
1195 * command with ILLEGAL_REQUEST during operation and
1196 * thus turned off use_10_for_rw.
1198 scmd_printk(KERN_ERR
, cmd
, "FUA write on READ/WRITE(6) drive\n");
1199 return BLK_STS_IOERR
;
1203 cmd
->cmnd
[0] = write
? WRITE_6
: READ_6
;
1204 cmd
->cmnd
[1] = (lba
>> 16) & 0x1f;
1205 cmd
->cmnd
[2] = (lba
>> 8) & 0xff;
1206 cmd
->cmnd
[3] = lba
& 0xff;
1207 cmd
->cmnd
[4] = nr_blocks
;
1213 static blk_status_t
sd_setup_read_write_cmnd(struct scsi_cmnd
*cmd
)
1215 struct request
*rq
= scsi_cmd_to_rq(cmd
);
1216 struct scsi_device
*sdp
= cmd
->device
;
1217 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
1218 sector_t lba
= sectors_to_logical(sdp
, blk_rq_pos(rq
));
1220 unsigned int nr_blocks
= sectors_to_logical(sdp
, blk_rq_sectors(rq
));
1221 unsigned int mask
= logical_to_sectors(sdp
, 1) - 1;
1222 bool write
= rq_data_dir(rq
) == WRITE
;
1223 unsigned char protect
, fua
;
1228 ret
= scsi_alloc_sgtables(cmd
);
1229 if (ret
!= BLK_STS_OK
)
1232 ret
= BLK_STS_IOERR
;
1233 if (!scsi_device_online(sdp
) || sdp
->changed
) {
1234 scmd_printk(KERN_ERR
, cmd
, "device offline or changed\n");
1238 if (blk_rq_pos(rq
) + blk_rq_sectors(rq
) > get_capacity(rq
->rq_disk
)) {
1239 scmd_printk(KERN_ERR
, cmd
, "access beyond end of device\n");
1243 if ((blk_rq_pos(rq
) & mask
) || (blk_rq_sectors(rq
) & mask
)) {
1244 scmd_printk(KERN_ERR
, cmd
, "request not aligned to the logical block size\n");
1249 * Some SD card readers can't handle accesses which touch the
1250 * last one or two logical blocks. Split accesses as needed.
1252 threshold
= sdkp
->capacity
- SD_LAST_BUGGY_SECTORS
;
1254 if (unlikely(sdp
->last_sector_bug
&& lba
+ nr_blocks
> threshold
)) {
1255 if (lba
< threshold
) {
1256 /* Access up to the threshold but not beyond */
1257 nr_blocks
= threshold
- lba
;
1259 /* Access only a single logical block */
1264 if (req_op(rq
) == REQ_OP_ZONE_APPEND
) {
1265 ret
= sd_zbc_prepare_zone_append(cmd
, &lba
, nr_blocks
);
1270 fua
= rq
->cmd_flags
& REQ_FUA
? 0x8 : 0;
1271 dix
= scsi_prot_sg_count(cmd
);
1272 dif
= scsi_host_dif_capable(cmd
->device
->host
, sdkp
->protection_type
);
1275 protect
= sd_setup_protect_cmnd(cmd
, dix
, dif
);
1279 if (protect
&& sdkp
->protection_type
== T10_PI_TYPE2_PROTECTION
) {
1280 ret
= sd_setup_rw32_cmnd(cmd
, write
, lba
, nr_blocks
,
1282 } else if (sdp
->use_16_for_rw
|| (nr_blocks
> 0xffff)) {
1283 ret
= sd_setup_rw16_cmnd(cmd
, write
, lba
, nr_blocks
,
1285 } else if ((nr_blocks
> 0xff) || (lba
> 0x1fffff) ||
1286 sdp
->use_10_for_rw
|| protect
) {
1287 ret
= sd_setup_rw10_cmnd(cmd
, write
, lba
, nr_blocks
,
1290 ret
= sd_setup_rw6_cmnd(cmd
, write
, lba
, nr_blocks
,
1294 if (unlikely(ret
!= BLK_STS_OK
))
1298 * We shouldn't disconnect in the middle of a sector, so with a dumb
1299 * host adapter, it's safe to assume that we can at least transfer
1300 * this many bytes between each connect / disconnect.
1302 cmd
->transfersize
= sdp
->sector_size
;
1303 cmd
->underflow
= nr_blocks
<< 9;
1304 cmd
->allowed
= sdkp
->max_retries
;
1305 cmd
->sdb
.length
= nr_blocks
* sdp
->sector_size
;
1308 scmd_printk(KERN_INFO
, cmd
,
1309 "%s: block=%llu, count=%d\n", __func__
,
1310 (unsigned long long)blk_rq_pos(rq
),
1311 blk_rq_sectors(rq
)));
1313 scmd_printk(KERN_INFO
, cmd
,
1314 "%s %d/%u 512 byte blocks.\n",
1315 write
? "writing" : "reading", nr_blocks
,
1316 blk_rq_sectors(rq
)));
1319 * This indicates that the command is ready from our end to be queued.
1323 scsi_free_sgtables(cmd
);
1327 static blk_status_t
sd_init_command(struct scsi_cmnd
*cmd
)
1329 struct request
*rq
= scsi_cmd_to_rq(cmd
);
1331 switch (req_op(rq
)) {
1332 case REQ_OP_DISCARD
:
1333 switch (scsi_disk(rq
->rq_disk
)->provisioning_mode
) {
1335 return sd_setup_unmap_cmnd(cmd
);
1337 return sd_setup_write_same16_cmnd(cmd
, true);
1339 return sd_setup_write_same10_cmnd(cmd
, true);
1341 return sd_setup_write_same10_cmnd(cmd
, false);
1343 return BLK_STS_TARGET
;
1345 case REQ_OP_WRITE_ZEROES
:
1346 return sd_setup_write_zeroes_cmnd(cmd
);
1347 case REQ_OP_WRITE_SAME
:
1348 return sd_setup_write_same_cmnd(cmd
);
1350 return sd_setup_flush_cmnd(cmd
);
1353 case REQ_OP_ZONE_APPEND
:
1354 return sd_setup_read_write_cmnd(cmd
);
1355 case REQ_OP_ZONE_RESET
:
1356 return sd_zbc_setup_zone_mgmt_cmnd(cmd
, ZO_RESET_WRITE_POINTER
,
1358 case REQ_OP_ZONE_RESET_ALL
:
1359 return sd_zbc_setup_zone_mgmt_cmnd(cmd
, ZO_RESET_WRITE_POINTER
,
1361 case REQ_OP_ZONE_OPEN
:
1362 return sd_zbc_setup_zone_mgmt_cmnd(cmd
, ZO_OPEN_ZONE
, false);
1363 case REQ_OP_ZONE_CLOSE
:
1364 return sd_zbc_setup_zone_mgmt_cmnd(cmd
, ZO_CLOSE_ZONE
, false);
1365 case REQ_OP_ZONE_FINISH
:
1366 return sd_zbc_setup_zone_mgmt_cmnd(cmd
, ZO_FINISH_ZONE
, false);
1369 return BLK_STS_NOTSUPP
;
1373 static void sd_uninit_command(struct scsi_cmnd
*SCpnt
)
1375 struct request
*rq
= scsi_cmd_to_rq(SCpnt
);
1378 if (rq
->rq_flags
& RQF_SPECIAL_PAYLOAD
)
1379 mempool_free(rq
->special_vec
.bv_page
, sd_page_pool
);
1381 if (SCpnt
->cmnd
!= scsi_req(rq
)->cmd
) {
1385 mempool_free(cmnd
, sd_cdb_pool
);
1389 static bool sd_need_revalidate(struct block_device
*bdev
,
1390 struct scsi_disk
*sdkp
)
1392 if (sdkp
->device
->removable
|| sdkp
->write_prot
) {
1393 if (bdev_check_media_change(bdev
))
1398 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1399 * nothing to do with partitions, BLKRRPART is used to force a full
1400 * revalidate after things like a format for historical reasons.
1402 return test_bit(GD_NEED_PART_SCAN
, &bdev
->bd_disk
->state
);
1406 * sd_open - open a scsi disk device
1407 * @bdev: Block device of the scsi disk to open
1408 * @mode: FMODE_* mask
1410 * Returns 0 if successful. Returns a negated errno value in case
1413 * Note: This can be called from a user context (e.g. fsck(1) )
1414 * or from within the kernel (e.g. as a result of a mount(1) ).
1415 * In the latter case @inode and @filp carry an abridged amount
1416 * of information as noted above.
1418 * Locking: called with bdev->bd_disk->open_mutex held.
1420 static int sd_open(struct block_device
*bdev
, fmode_t mode
)
1422 struct scsi_disk
*sdkp
= scsi_disk_get(bdev
->bd_disk
);
1423 struct scsi_device
*sdev
;
1429 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_open\n"));
1431 sdev
= sdkp
->device
;
1434 * If the device is in error recovery, wait until it is done.
1435 * If the device is offline, then disallow any access to it.
1438 if (!scsi_block_when_processing_errors(sdev
))
1441 if (sd_need_revalidate(bdev
, sdkp
))
1442 sd_revalidate_disk(bdev
->bd_disk
);
1445 * If the drive is empty, just let the open fail.
1447 retval
= -ENOMEDIUM
;
1448 if (sdev
->removable
&& !sdkp
->media_present
&& !(mode
& FMODE_NDELAY
))
1452 * If the device has the write protect tab set, have the open fail
1453 * if the user expects to be able to write to the thing.
1456 if (sdkp
->write_prot
&& (mode
& FMODE_WRITE
))
1460 * It is possible that the disk changing stuff resulted in
1461 * the device being taken offline. If this is the case,
1462 * report this to the user, and don't pretend that the
1463 * open actually succeeded.
1466 if (!scsi_device_online(sdev
))
1469 if ((atomic_inc_return(&sdkp
->openers
) == 1) && sdev
->removable
) {
1470 if (scsi_block_when_processing_errors(sdev
))
1471 scsi_set_medium_removal(sdev
, SCSI_REMOVAL_PREVENT
);
1477 scsi_disk_put(sdkp
);
1482 * sd_release - invoked when the (last) close(2) is called on this
1484 * @disk: disk to release
1485 * @mode: FMODE_* mask
1489 * Note: may block (uninterruptible) if error recovery is underway
1492 * Locking: called with bdev->bd_disk->open_mutex held.
1494 static void sd_release(struct gendisk
*disk
, fmode_t mode
)
1496 struct scsi_disk
*sdkp
= scsi_disk(disk
);
1497 struct scsi_device
*sdev
= sdkp
->device
;
1499 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_release\n"));
1501 if (atomic_dec_return(&sdkp
->openers
) == 0 && sdev
->removable
) {
1502 if (scsi_block_when_processing_errors(sdev
))
1503 scsi_set_medium_removal(sdev
, SCSI_REMOVAL_ALLOW
);
1506 scsi_disk_put(sdkp
);
1509 static int sd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1511 struct scsi_disk
*sdkp
= scsi_disk(bdev
->bd_disk
);
1512 struct scsi_device
*sdp
= sdkp
->device
;
1513 struct Scsi_Host
*host
= sdp
->host
;
1514 sector_t capacity
= logical_to_sectors(sdp
, sdkp
->capacity
);
1517 /* default to most commonly used values */
1518 diskinfo
[0] = 0x40; /* 1 << 6 */
1519 diskinfo
[1] = 0x20; /* 1 << 5 */
1520 diskinfo
[2] = capacity
>> 11;
1522 /* override with calculated, extended default, or driver values */
1523 if (host
->hostt
->bios_param
)
1524 host
->hostt
->bios_param(sdp
, bdev
, capacity
, diskinfo
);
1526 scsicam_bios_param(bdev
, capacity
, diskinfo
);
1528 geo
->heads
= diskinfo
[0];
1529 geo
->sectors
= diskinfo
[1];
1530 geo
->cylinders
= diskinfo
[2];
1535 * sd_ioctl - process an ioctl
1536 * @bdev: target block device
1537 * @mode: FMODE_* mask
1538 * @cmd: ioctl command number
1539 * @arg: this is third argument given to ioctl(2) system call.
1540 * Often contains a pointer.
1542 * Returns 0 if successful (some ioctls return positive numbers on
1543 * success as well). Returns a negated errno value in case of error.
1545 * Note: most ioctls are forward onto the block subsystem or further
1546 * down in the scsi subsystem.
1548 static int sd_ioctl(struct block_device
*bdev
, fmode_t mode
,
1549 unsigned int cmd
, unsigned long arg
)
1551 struct gendisk
*disk
= bdev
->bd_disk
;
1552 struct scsi_disk
*sdkp
= scsi_disk(disk
);
1553 struct scsi_device
*sdp
= sdkp
->device
;
1554 void __user
*p
= (void __user
*)arg
;
1557 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO
, sdkp
, "sd_ioctl: disk=%s, "
1558 "cmd=0x%x\n", disk
->disk_name
, cmd
));
1560 if (bdev_is_partition(bdev
) && !capable(CAP_SYS_RAWIO
))
1561 return -ENOIOCTLCMD
;
1564 * If we are in the middle of error recovery, don't let anyone
1565 * else try and use this device. Also, if error recovery fails, it
1566 * may try and take the device offline, in which case all further
1567 * access to the device is prohibited.
1569 error
= scsi_ioctl_block_when_processing_errors(sdp
, cmd
,
1570 (mode
& FMODE_NDELAY
) != 0);
1574 if (is_sed_ioctl(cmd
))
1575 return sed_ioctl(sdkp
->opal_dev
, cmd
, p
);
1576 return scsi_ioctl(sdp
, disk
, mode
, cmd
, p
);
1579 static void set_media_not_present(struct scsi_disk
*sdkp
)
1581 if (sdkp
->media_present
)
1582 sdkp
->device
->changed
= 1;
1584 if (sdkp
->device
->removable
) {
1585 sdkp
->media_present
= 0;
1590 static int media_not_present(struct scsi_disk
*sdkp
,
1591 struct scsi_sense_hdr
*sshdr
)
1593 if (!scsi_sense_valid(sshdr
))
1596 /* not invoked for commands that could return deferred errors */
1597 switch (sshdr
->sense_key
) {
1598 case UNIT_ATTENTION
:
1600 /* medium not present */
1601 if (sshdr
->asc
== 0x3A) {
1602 set_media_not_present(sdkp
);
1610 * sd_check_events - check media events
1611 * @disk: kernel device descriptor
1612 * @clearing: disk events currently being cleared
1614 * Returns mask of DISK_EVENT_*.
1616 * Note: this function is invoked from the block subsystem.
1618 static unsigned int sd_check_events(struct gendisk
*disk
, unsigned int clearing
)
1620 struct scsi_disk
*sdkp
= scsi_disk_get(disk
);
1621 struct scsi_device
*sdp
;
1629 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_check_events\n"));
1632 * If the device is offline, don't send any commands - just pretend as
1633 * if the command failed. If the device ever comes back online, we
1634 * can deal with it then. It is only because of unrecoverable errors
1635 * that we would ever take a device offline in the first place.
1637 if (!scsi_device_online(sdp
)) {
1638 set_media_not_present(sdkp
);
1643 * Using TEST_UNIT_READY enables differentiation between drive with
1644 * no cartridge loaded - NOT READY, drive with changed cartridge -
1645 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1647 * Drives that auto spin down. eg iomega jaz 1G, will be started
1648 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1649 * sd_revalidate() is called.
1651 if (scsi_block_when_processing_errors(sdp
)) {
1652 struct scsi_sense_hdr sshdr
= { 0, };
1654 retval
= scsi_test_unit_ready(sdp
, SD_TIMEOUT
, sdkp
->max_retries
,
1657 /* failed to execute TUR, assume media not present */
1658 if (retval
< 0 || host_byte(retval
)) {
1659 set_media_not_present(sdkp
);
1663 if (media_not_present(sdkp
, &sshdr
))
1668 * For removable scsi disk we have to recognise the presence
1669 * of a disk in the drive.
1671 if (!sdkp
->media_present
)
1673 sdkp
->media_present
= 1;
1676 * sdp->changed is set under the following conditions:
1678 * Medium present state has changed in either direction.
1679 * Device has indicated UNIT_ATTENTION.
1681 disk_changed
= sdp
->changed
;
1683 scsi_disk_put(sdkp
);
1684 return disk_changed
? DISK_EVENT_MEDIA_CHANGE
: 0;
1687 static int sd_sync_cache(struct scsi_disk
*sdkp
, struct scsi_sense_hdr
*sshdr
)
1690 struct scsi_device
*sdp
= sdkp
->device
;
1691 const int timeout
= sdp
->request_queue
->rq_timeout
1692 * SD_FLUSH_TIMEOUT_MULTIPLIER
;
1693 struct scsi_sense_hdr my_sshdr
;
1695 if (!scsi_device_online(sdp
))
1698 /* caller might not be interested in sense, but we need it */
1702 for (retries
= 3; retries
> 0; --retries
) {
1703 unsigned char cmd
[10] = { 0 };
1705 cmd
[0] = SYNCHRONIZE_CACHE
;
1707 * Leave the rest of the command zero to indicate
1710 res
= scsi_execute(sdp
, cmd
, DMA_NONE
, NULL
, 0, NULL
, sshdr
,
1711 timeout
, sdkp
->max_retries
, 0, RQF_PM
, NULL
);
1717 sd_print_result(sdkp
, "Synchronize Cache(10) failed", res
);
1722 if (scsi_status_is_check_condition(res
) &&
1723 scsi_sense_valid(sshdr
)) {
1724 sd_print_sense_hdr(sdkp
, sshdr
);
1726 /* we need to evaluate the error return */
1727 if (sshdr
->asc
== 0x3a || /* medium not present */
1728 sshdr
->asc
== 0x20 || /* invalid command */
1729 (sshdr
->asc
== 0x74 && sshdr
->ascq
== 0x71)) /* drive is password locked */
1730 /* this is no error here */
1734 switch (host_byte(res
)) {
1735 /* ignore errors due to racing a disconnection */
1736 case DID_BAD_TARGET
:
1737 case DID_NO_CONNECT
:
1739 /* signal the upper layer it might try again */
1743 case DID_SOFT_ERROR
:
1752 static void sd_rescan(struct device
*dev
)
1754 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
1756 sd_revalidate_disk(sdkp
->disk
);
1759 static char sd_pr_type(enum pr_type type
)
1762 case PR_WRITE_EXCLUSIVE
:
1764 case PR_EXCLUSIVE_ACCESS
:
1766 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
1768 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
1770 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
1772 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
1779 static int sd_pr_command(struct block_device
*bdev
, u8 sa
,
1780 u64 key
, u64 sa_key
, u8 type
, u8 flags
)
1782 struct scsi_disk
*sdkp
= scsi_disk(bdev
->bd_disk
);
1783 struct scsi_device
*sdev
= sdkp
->device
;
1784 struct scsi_sense_hdr sshdr
;
1786 u8 cmd
[16] = { 0, };
1787 u8 data
[24] = { 0, };
1789 cmd
[0] = PERSISTENT_RESERVE_OUT
;
1792 put_unaligned_be32(sizeof(data
), &cmd
[5]);
1794 put_unaligned_be64(key
, &data
[0]);
1795 put_unaligned_be64(sa_key
, &data
[8]);
1798 result
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, &data
, sizeof(data
),
1799 &sshdr
, SD_TIMEOUT
, sdkp
->max_retries
, NULL
);
1801 if (scsi_status_is_check_condition(result
) &&
1802 scsi_sense_valid(&sshdr
)) {
1803 sdev_printk(KERN_INFO
, sdev
, "PR command failed: %d\n", result
);
1804 scsi_print_sense_hdr(sdev
, NULL
, &sshdr
);
1810 static int sd_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
1813 if (flags
& ~PR_FL_IGNORE_KEY
)
1815 return sd_pr_command(bdev
, (flags
& PR_FL_IGNORE_KEY
) ? 0x06 : 0x00,
1816 old_key
, new_key
, 0,
1817 (1 << 0) /* APTPL */);
1820 static int sd_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
1825 return sd_pr_command(bdev
, 0x01, key
, 0, sd_pr_type(type
), 0);
1828 static int sd_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
1830 return sd_pr_command(bdev
, 0x02, key
, 0, sd_pr_type(type
), 0);
1833 static int sd_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
1834 enum pr_type type
, bool abort
)
1836 return sd_pr_command(bdev
, abort
? 0x05 : 0x04, old_key
, new_key
,
1837 sd_pr_type(type
), 0);
1840 static int sd_pr_clear(struct block_device
*bdev
, u64 key
)
1842 return sd_pr_command(bdev
, 0x03, key
, 0, 0, 0);
1845 static const struct pr_ops sd_pr_ops
= {
1846 .pr_register
= sd_pr_register
,
1847 .pr_reserve
= sd_pr_reserve
,
1848 .pr_release
= sd_pr_release
,
1849 .pr_preempt
= sd_pr_preempt
,
1850 .pr_clear
= sd_pr_clear
,
1853 static const struct block_device_operations sd_fops
= {
1854 .owner
= THIS_MODULE
,
1856 .release
= sd_release
,
1858 .getgeo
= sd_getgeo
,
1859 .compat_ioctl
= blkdev_compat_ptr_ioctl
,
1860 .check_events
= sd_check_events
,
1861 .unlock_native_capacity
= sd_unlock_native_capacity
,
1862 .report_zones
= sd_zbc_report_zones
,
1863 .pr_ops
= &sd_pr_ops
,
1867 * sd_eh_reset - reset error handling callback
1868 * @scmd: sd-issued command that has failed
1870 * This function is called by the SCSI midlayer before starting
1871 * SCSI EH. When counting medium access failures we have to be
1872 * careful to register it only only once per device and SCSI EH run;
1873 * there might be several timed out commands which will cause the
1874 * 'max_medium_access_timeouts' counter to trigger after the first
1875 * SCSI EH run already and set the device to offline.
1876 * So this function resets the internal counter before starting SCSI EH.
1878 static void sd_eh_reset(struct scsi_cmnd
*scmd
)
1880 struct scsi_disk
*sdkp
= scsi_disk(scsi_cmd_to_rq(scmd
)->rq_disk
);
1882 /* New SCSI EH run, reset gate variable */
1883 sdkp
->ignore_medium_access_errors
= false;
1887 * sd_eh_action - error handling callback
1888 * @scmd: sd-issued command that has failed
1889 * @eh_disp: The recovery disposition suggested by the midlayer
1891 * This function is called by the SCSI midlayer upon completion of an
1892 * error test command (currently TEST UNIT READY). The result of sending
1893 * the eh command is passed in eh_disp. We're looking for devices that
1894 * fail medium access commands but are OK with non access commands like
1895 * test unit ready (so wrongly see the device as having a successful
1898 static int sd_eh_action(struct scsi_cmnd
*scmd
, int eh_disp
)
1900 struct scsi_disk
*sdkp
= scsi_disk(scsi_cmd_to_rq(scmd
)->rq_disk
);
1901 struct scsi_device
*sdev
= scmd
->device
;
1903 if (!scsi_device_online(sdev
) ||
1904 !scsi_medium_access_command(scmd
) ||
1905 host_byte(scmd
->result
) != DID_TIME_OUT
||
1910 * The device has timed out executing a medium access command.
1911 * However, the TEST UNIT READY command sent during error
1912 * handling completed successfully. Either the device is in the
1913 * process of recovering or has it suffered an internal failure
1914 * that prevents access to the storage medium.
1916 if (!sdkp
->ignore_medium_access_errors
) {
1917 sdkp
->medium_access_timed_out
++;
1918 sdkp
->ignore_medium_access_errors
= true;
1922 * If the device keeps failing read/write commands but TEST UNIT
1923 * READY always completes successfully we assume that medium
1924 * access is no longer possible and take the device offline.
1926 if (sdkp
->medium_access_timed_out
>= sdkp
->max_medium_access_timeouts
) {
1927 scmd_printk(KERN_ERR
, scmd
,
1928 "Medium access timeout failure. Offlining disk!\n");
1929 mutex_lock(&sdev
->state_mutex
);
1930 scsi_device_set_state(sdev
, SDEV_OFFLINE
);
1931 mutex_unlock(&sdev
->state_mutex
);
1939 static unsigned int sd_completed_bytes(struct scsi_cmnd
*scmd
)
1941 struct request
*req
= scsi_cmd_to_rq(scmd
);
1942 struct scsi_device
*sdev
= scmd
->device
;
1943 unsigned int transferred
, good_bytes
;
1944 u64 start_lba
, end_lba
, bad_lba
;
1947 * Some commands have a payload smaller than the device logical
1948 * block size (e.g. INQUIRY on a 4K disk).
1950 if (scsi_bufflen(scmd
) <= sdev
->sector_size
)
1953 /* Check if we have a 'bad_lba' information */
1954 if (!scsi_get_sense_info_fld(scmd
->sense_buffer
,
1955 SCSI_SENSE_BUFFERSIZE
,
1960 * If the bad lba was reported incorrectly, we have no idea where
1963 start_lba
= sectors_to_logical(sdev
, blk_rq_pos(req
));
1964 end_lba
= start_lba
+ bytes_to_logical(sdev
, scsi_bufflen(scmd
));
1965 if (bad_lba
< start_lba
|| bad_lba
>= end_lba
)
1969 * resid is optional but mostly filled in. When it's unused,
1970 * its value is zero, so we assume the whole buffer transferred
1972 transferred
= scsi_bufflen(scmd
) - scsi_get_resid(scmd
);
1974 /* This computation should always be done in terms of the
1975 * resolution of the device's medium.
1977 good_bytes
= logical_to_bytes(sdev
, bad_lba
- start_lba
);
1979 return min(good_bytes
, transferred
);
1983 * sd_done - bottom half handler: called when the lower level
1984 * driver has completed (successfully or otherwise) a scsi command.
1985 * @SCpnt: mid-level's per command structure.
1987 * Note: potentially run from within an ISR. Must not block.
1989 static int sd_done(struct scsi_cmnd
*SCpnt
)
1991 int result
= SCpnt
->result
;
1992 unsigned int good_bytes
= result
? 0 : scsi_bufflen(SCpnt
);
1993 unsigned int sector_size
= SCpnt
->device
->sector_size
;
1995 struct scsi_sense_hdr sshdr
;
1996 struct request
*req
= scsi_cmd_to_rq(SCpnt
);
1997 struct scsi_disk
*sdkp
= scsi_disk(req
->rq_disk
);
1998 int sense_valid
= 0;
1999 int sense_deferred
= 0;
2001 switch (req_op(req
)) {
2002 case REQ_OP_DISCARD
:
2003 case REQ_OP_WRITE_ZEROES
:
2004 case REQ_OP_WRITE_SAME
:
2005 case REQ_OP_ZONE_RESET
:
2006 case REQ_OP_ZONE_RESET_ALL
:
2007 case REQ_OP_ZONE_OPEN
:
2008 case REQ_OP_ZONE_CLOSE
:
2009 case REQ_OP_ZONE_FINISH
:
2011 good_bytes
= blk_rq_bytes(req
);
2012 scsi_set_resid(SCpnt
, 0);
2015 scsi_set_resid(SCpnt
, blk_rq_bytes(req
));
2020 * In case of bogus fw or device, we could end up having
2021 * an unaligned partial completion. Check this here and force
2024 resid
= scsi_get_resid(SCpnt
);
2025 if (resid
& (sector_size
- 1)) {
2026 sd_printk(KERN_INFO
, sdkp
,
2027 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2028 resid
, sector_size
);
2029 scsi_print_command(SCpnt
);
2030 resid
= min(scsi_bufflen(SCpnt
),
2031 round_up(resid
, sector_size
));
2032 scsi_set_resid(SCpnt
, resid
);
2037 sense_valid
= scsi_command_normalize_sense(SCpnt
, &sshdr
);
2039 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
2041 sdkp
->medium_access_timed_out
= 0;
2043 if (!scsi_status_is_check_condition(result
) &&
2044 (!sense_valid
|| sense_deferred
))
2047 switch (sshdr
.sense_key
) {
2048 case HARDWARE_ERROR
:
2050 good_bytes
= sd_completed_bytes(SCpnt
);
2052 case RECOVERED_ERROR
:
2053 good_bytes
= scsi_bufflen(SCpnt
);
2056 /* This indicates a false check condition, so ignore it. An
2057 * unknown amount of data was transferred so treat it as an
2061 memset(SCpnt
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
2063 case ABORTED_COMMAND
:
2064 if (sshdr
.asc
== 0x10) /* DIF: Target detected corruption */
2065 good_bytes
= sd_completed_bytes(SCpnt
);
2067 case ILLEGAL_REQUEST
:
2068 switch (sshdr
.asc
) {
2069 case 0x10: /* DIX: Host detected corruption */
2070 good_bytes
= sd_completed_bytes(SCpnt
);
2072 case 0x20: /* INVALID COMMAND OPCODE */
2073 case 0x24: /* INVALID FIELD IN CDB */
2074 switch (SCpnt
->cmnd
[0]) {
2076 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
2080 if (SCpnt
->cmnd
[1] & 8) { /* UNMAP */
2081 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
2083 sdkp
->device
->no_write_same
= 1;
2084 sd_config_write_same(sdkp
);
2085 req
->rq_flags
|= RQF_QUIET
;
2096 if (sd_is_zoned(sdkp
))
2097 good_bytes
= sd_zbc_complete(SCpnt
, good_bytes
, &sshdr
);
2099 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, SCpnt
,
2100 "sd_done: completed %d of %d bytes\n",
2101 good_bytes
, scsi_bufflen(SCpnt
)));
2107 * spinup disk - called only in sd_revalidate_disk()
2110 sd_spinup_disk(struct scsi_disk
*sdkp
)
2112 unsigned char cmd
[10];
2113 unsigned long spintime_expire
= 0;
2114 int retries
, spintime
;
2115 unsigned int the_result
;
2116 struct scsi_sense_hdr sshdr
;
2117 int sense_valid
= 0;
2121 /* Spin up drives, as required. Only do this at boot time */
2122 /* Spinup needs to be done for module loads too. */
2127 bool media_was_present
= sdkp
->media_present
;
2129 cmd
[0] = TEST_UNIT_READY
;
2130 memset((void *) &cmd
[1], 0, 9);
2132 the_result
= scsi_execute_req(sdkp
->device
, cmd
,
2135 sdkp
->max_retries
, NULL
);
2138 * If the drive has indicated to us that it
2139 * doesn't have any media in it, don't bother
2140 * with any more polling.
2142 if (media_not_present(sdkp
, &sshdr
)) {
2143 if (media_was_present
)
2144 sd_printk(KERN_NOTICE
, sdkp
, "Media removed, stopped polling\n");
2149 sense_valid
= scsi_sense_valid(&sshdr
);
2151 } while (retries
< 3 &&
2152 (!scsi_status_is_good(the_result
) ||
2153 (scsi_status_is_check_condition(the_result
) &&
2154 sense_valid
&& sshdr
.sense_key
== UNIT_ATTENTION
)));
2156 if (!scsi_status_is_check_condition(the_result
)) {
2157 /* no sense, TUR either succeeded or failed
2158 * with a status error */
2159 if(!spintime
&& !scsi_status_is_good(the_result
)) {
2160 sd_print_result(sdkp
, "Test Unit Ready failed",
2167 * The device does not want the automatic start to be issued.
2169 if (sdkp
->device
->no_start_on_add
)
2172 if (sense_valid
&& sshdr
.sense_key
== NOT_READY
) {
2173 if (sshdr
.asc
== 4 && sshdr
.ascq
== 3)
2174 break; /* manual intervention required */
2175 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0xb)
2176 break; /* standby */
2177 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0xc)
2178 break; /* unavailable */
2179 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0x1b)
2180 break; /* sanitize in progress */
2182 * Issue command to spin up drive when not ready
2185 sd_printk(KERN_NOTICE
, sdkp
, "Spinning up disk...");
2186 cmd
[0] = START_STOP
;
2187 cmd
[1] = 1; /* Return immediately */
2188 memset((void *) &cmd
[2], 0, 8);
2189 cmd
[4] = 1; /* Start spin cycle */
2190 if (sdkp
->device
->start_stop_pwr_cond
)
2192 scsi_execute_req(sdkp
->device
, cmd
, DMA_NONE
,
2194 SD_TIMEOUT
, sdkp
->max_retries
,
2196 spintime_expire
= jiffies
+ 100 * HZ
;
2199 /* Wait 1 second for next try */
2201 printk(KERN_CONT
".");
2204 * Wait for USB flash devices with slow firmware.
2205 * Yes, this sense key/ASC combination shouldn't
2206 * occur here. It's characteristic of these devices.
2208 } else if (sense_valid
&&
2209 sshdr
.sense_key
== UNIT_ATTENTION
&&
2210 sshdr
.asc
== 0x28) {
2212 spintime_expire
= jiffies
+ 5 * HZ
;
2215 /* Wait 1 second for next try */
2218 /* we don't understand the sense code, so it's
2219 * probably pointless to loop */
2221 sd_printk(KERN_NOTICE
, sdkp
, "Unit Not Ready\n");
2222 sd_print_sense_hdr(sdkp
, &sshdr
);
2227 } while (spintime
&& time_before_eq(jiffies
, spintime_expire
));
2230 if (scsi_status_is_good(the_result
))
2231 printk(KERN_CONT
"ready\n");
2233 printk(KERN_CONT
"not responding...\n");
2238 * Determine whether disk supports Data Integrity Field.
2240 static int sd_read_protection_type(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2242 struct scsi_device
*sdp
= sdkp
->device
;
2246 if (scsi_device_protection(sdp
) == 0 || (buffer
[12] & 1) == 0) {
2247 sdkp
->protection_type
= 0;
2251 type
= ((buffer
[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2253 if (type
> T10_PI_TYPE3_PROTECTION
)
2255 else if (scsi_host_dif_capable(sdp
->host
, type
))
2258 if (sdkp
->first_scan
|| type
!= sdkp
->protection_type
)
2261 sd_printk(KERN_ERR
, sdkp
, "formatted with unsupported" \
2262 " protection type %u. Disabling disk!\n",
2266 sd_printk(KERN_NOTICE
, sdkp
,
2267 "Enabling DIF Type %u protection\n", type
);
2270 sd_printk(KERN_NOTICE
, sdkp
,
2271 "Disabling DIF Type %u protection\n", type
);
2275 sdkp
->protection_type
= type
;
2280 static void read_capacity_error(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2281 struct scsi_sense_hdr
*sshdr
, int sense_valid
,
2285 sd_print_sense_hdr(sdkp
, sshdr
);
2287 sd_printk(KERN_NOTICE
, sdkp
, "Sense not available.\n");
2290 * Set dirty bit for removable devices if not ready -
2291 * sometimes drives will not report this properly.
2293 if (sdp
->removable
&&
2294 sense_valid
&& sshdr
->sense_key
== NOT_READY
)
2295 set_media_not_present(sdkp
);
2298 * We used to set media_present to 0 here to indicate no media
2299 * in the drive, but some drives fail read capacity even with
2300 * media present, so we can't do that.
2302 sdkp
->capacity
= 0; /* unknown mapped to zero - as usual */
2306 #if RC16_LEN > SD_BUF_SIZE
2307 #error RC16_LEN must not be more than SD_BUF_SIZE
2310 #define READ_CAPACITY_RETRIES_ON_RESET 10
2312 static int read_capacity_16(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2313 unsigned char *buffer
)
2315 unsigned char cmd
[16];
2316 struct scsi_sense_hdr sshdr
;
2317 int sense_valid
= 0;
2319 int retries
= 3, reset_retries
= READ_CAPACITY_RETRIES_ON_RESET
;
2320 unsigned int alignment
;
2321 unsigned long long lba
;
2322 unsigned sector_size
;
2324 if (sdp
->no_read_capacity_16
)
2329 cmd
[0] = SERVICE_ACTION_IN_16
;
2330 cmd
[1] = SAI_READ_CAPACITY_16
;
2332 memset(buffer
, 0, RC16_LEN
);
2334 the_result
= scsi_execute_req(sdp
, cmd
, DMA_FROM_DEVICE
,
2335 buffer
, RC16_LEN
, &sshdr
,
2336 SD_TIMEOUT
, sdkp
->max_retries
, NULL
);
2338 if (media_not_present(sdkp
, &sshdr
))
2341 if (the_result
> 0) {
2342 sense_valid
= scsi_sense_valid(&sshdr
);
2344 sshdr
.sense_key
== ILLEGAL_REQUEST
&&
2345 (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) &&
2347 /* Invalid Command Operation Code or
2348 * Invalid Field in CDB, just retry
2349 * silently with RC10 */
2352 sshdr
.sense_key
== UNIT_ATTENTION
&&
2353 sshdr
.asc
== 0x29 && sshdr
.ascq
== 0x00)
2354 /* Device reset might occur several times,
2355 * give it one more chance */
2356 if (--reset_retries
> 0)
2361 } while (the_result
&& retries
);
2364 sd_print_result(sdkp
, "Read Capacity(16) failed", the_result
);
2365 read_capacity_error(sdkp
, sdp
, &sshdr
, sense_valid
, the_result
);
2369 sector_size
= get_unaligned_be32(&buffer
[8]);
2370 lba
= get_unaligned_be64(&buffer
[0]);
2372 if (sd_read_protection_type(sdkp
, buffer
) < 0) {
2377 /* Logical blocks per physical block exponent */
2378 sdkp
->physical_block_size
= (1 << (buffer
[13] & 0xf)) * sector_size
;
2381 sdkp
->rc_basis
= (buffer
[12] >> 4) & 0x3;
2383 /* Lowest aligned logical block */
2384 alignment
= ((buffer
[14] & 0x3f) << 8 | buffer
[15]) * sector_size
;
2385 blk_queue_alignment_offset(sdp
->request_queue
, alignment
);
2386 if (alignment
&& sdkp
->first_scan
)
2387 sd_printk(KERN_NOTICE
, sdkp
,
2388 "physical block alignment offset: %u\n", alignment
);
2390 if (buffer
[14] & 0x80) { /* LBPME */
2393 if (buffer
[14] & 0x40) /* LBPRZ */
2396 sd_config_discard(sdkp
, SD_LBP_WS16
);
2399 sdkp
->capacity
= lba
+ 1;
2403 static int read_capacity_10(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2404 unsigned char *buffer
)
2406 unsigned char cmd
[16];
2407 struct scsi_sense_hdr sshdr
;
2408 int sense_valid
= 0;
2410 int retries
= 3, reset_retries
= READ_CAPACITY_RETRIES_ON_RESET
;
2412 unsigned sector_size
;
2415 cmd
[0] = READ_CAPACITY
;
2416 memset(&cmd
[1], 0, 9);
2417 memset(buffer
, 0, 8);
2419 the_result
= scsi_execute_req(sdp
, cmd
, DMA_FROM_DEVICE
,
2421 SD_TIMEOUT
, sdkp
->max_retries
, NULL
);
2423 if (media_not_present(sdkp
, &sshdr
))
2426 if (the_result
> 0) {
2427 sense_valid
= scsi_sense_valid(&sshdr
);
2429 sshdr
.sense_key
== UNIT_ATTENTION
&&
2430 sshdr
.asc
== 0x29 && sshdr
.ascq
== 0x00)
2431 /* Device reset might occur several times,
2432 * give it one more chance */
2433 if (--reset_retries
> 0)
2438 } while (the_result
&& retries
);
2441 sd_print_result(sdkp
, "Read Capacity(10) failed", the_result
);
2442 read_capacity_error(sdkp
, sdp
, &sshdr
, sense_valid
, the_result
);
2446 sector_size
= get_unaligned_be32(&buffer
[4]);
2447 lba
= get_unaligned_be32(&buffer
[0]);
2449 if (sdp
->no_read_capacity_16
&& (lba
== 0xffffffff)) {
2450 /* Some buggy (usb cardreader) devices return an lba of
2451 0xffffffff when the want to report a size of 0 (with
2452 which they really mean no media is present) */
2454 sdkp
->physical_block_size
= sector_size
;
2458 sdkp
->capacity
= lba
+ 1;
2459 sdkp
->physical_block_size
= sector_size
;
2463 static int sd_try_rc16_first(struct scsi_device
*sdp
)
2465 if (sdp
->host
->max_cmd_len
< 16)
2467 if (sdp
->try_rc_10_first
)
2469 if (sdp
->scsi_level
> SCSI_SPC_2
)
2471 if (scsi_device_protection(sdp
))
2477 * read disk capacity
2480 sd_read_capacity(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2483 struct scsi_device
*sdp
= sdkp
->device
;
2485 if (sd_try_rc16_first(sdp
)) {
2486 sector_size
= read_capacity_16(sdkp
, sdp
, buffer
);
2487 if (sector_size
== -EOVERFLOW
)
2489 if (sector_size
== -ENODEV
)
2491 if (sector_size
< 0)
2492 sector_size
= read_capacity_10(sdkp
, sdp
, buffer
);
2493 if (sector_size
< 0)
2496 sector_size
= read_capacity_10(sdkp
, sdp
, buffer
);
2497 if (sector_size
== -EOVERFLOW
)
2499 if (sector_size
< 0)
2501 if ((sizeof(sdkp
->capacity
) > 4) &&
2502 (sdkp
->capacity
> 0xffffffffULL
)) {
2503 int old_sector_size
= sector_size
;
2504 sd_printk(KERN_NOTICE
, sdkp
, "Very big device. "
2505 "Trying to use READ CAPACITY(16).\n");
2506 sector_size
= read_capacity_16(sdkp
, sdp
, buffer
);
2507 if (sector_size
< 0) {
2508 sd_printk(KERN_NOTICE
, sdkp
,
2509 "Using 0xffffffff as device size\n");
2510 sdkp
->capacity
= 1 + (sector_t
) 0xffffffff;
2511 sector_size
= old_sector_size
;
2514 /* Remember that READ CAPACITY(16) succeeded */
2515 sdp
->try_rc_10_first
= 0;
2519 /* Some devices are known to return the total number of blocks,
2520 * not the highest block number. Some devices have versions
2521 * which do this and others which do not. Some devices we might
2522 * suspect of doing this but we don't know for certain.
2524 * If we know the reported capacity is wrong, decrement it. If
2525 * we can only guess, then assume the number of blocks is even
2526 * (usually true but not always) and err on the side of lowering
2529 if (sdp
->fix_capacity
||
2530 (sdp
->guess_capacity
&& (sdkp
->capacity
& 0x01))) {
2531 sd_printk(KERN_INFO
, sdkp
, "Adjusting the sector count "
2532 "from its reported value: %llu\n",
2533 (unsigned long long) sdkp
->capacity
);
2538 if (sector_size
== 0) {
2540 sd_printk(KERN_NOTICE
, sdkp
, "Sector size 0 reported, "
2544 if (sector_size
!= 512 &&
2545 sector_size
!= 1024 &&
2546 sector_size
!= 2048 &&
2547 sector_size
!= 4096) {
2548 sd_printk(KERN_NOTICE
, sdkp
, "Unsupported sector size %d.\n",
2551 * The user might want to re-format the drive with
2552 * a supported sectorsize. Once this happens, it
2553 * would be relatively trivial to set the thing up.
2554 * For this reason, we leave the thing in the table.
2558 * set a bogus sector size so the normal read/write
2559 * logic in the block layer will eventually refuse any
2560 * request on this device without tripping over power
2561 * of two sector size assumptions
2565 blk_queue_logical_block_size(sdp
->request_queue
, sector_size
);
2566 blk_queue_physical_block_size(sdp
->request_queue
,
2567 sdkp
->physical_block_size
);
2568 sdkp
->device
->sector_size
= sector_size
;
2570 if (sdkp
->capacity
> 0xffffffff)
2571 sdp
->use_16_for_rw
= 1;
2576 * Print disk capacity
2579 sd_print_capacity(struct scsi_disk
*sdkp
,
2580 sector_t old_capacity
)
2582 int sector_size
= sdkp
->device
->sector_size
;
2583 char cap_str_2
[10], cap_str_10
[10];
2585 if (!sdkp
->first_scan
&& old_capacity
== sdkp
->capacity
)
2588 string_get_size(sdkp
->capacity
, sector_size
,
2589 STRING_UNITS_2
, cap_str_2
, sizeof(cap_str_2
));
2590 string_get_size(sdkp
->capacity
, sector_size
,
2591 STRING_UNITS_10
, cap_str_10
, sizeof(cap_str_10
));
2593 sd_printk(KERN_NOTICE
, sdkp
,
2594 "%llu %d-byte logical blocks: (%s/%s)\n",
2595 (unsigned long long)sdkp
->capacity
,
2596 sector_size
, cap_str_10
, cap_str_2
);
2598 if (sdkp
->physical_block_size
!= sector_size
)
2599 sd_printk(KERN_NOTICE
, sdkp
,
2600 "%u-byte physical blocks\n",
2601 sdkp
->physical_block_size
);
2604 /* called with buffer of length 512 */
2606 sd_do_mode_sense(struct scsi_disk
*sdkp
, int dbd
, int modepage
,
2607 unsigned char *buffer
, int len
, struct scsi_mode_data
*data
,
2608 struct scsi_sense_hdr
*sshdr
)
2610 return scsi_mode_sense(sdkp
->device
, dbd
, modepage
, buffer
, len
,
2611 SD_TIMEOUT
, sdkp
->max_retries
, data
,
2616 * read write protect setting, if possible - called only in sd_revalidate_disk()
2617 * called with buffer of length SD_BUF_SIZE
2620 sd_read_write_protect_flag(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2623 struct scsi_device
*sdp
= sdkp
->device
;
2624 struct scsi_mode_data data
;
2625 int old_wp
= sdkp
->write_prot
;
2627 set_disk_ro(sdkp
->disk
, 0);
2628 if (sdp
->skip_ms_page_3f
) {
2629 sd_first_printk(KERN_NOTICE
, sdkp
, "Assuming Write Enabled\n");
2633 if (sdp
->use_192_bytes_for_3f
) {
2634 res
= sd_do_mode_sense(sdkp
, 0, 0x3F, buffer
, 192, &data
, NULL
);
2637 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2638 * We have to start carefully: some devices hang if we ask
2639 * for more than is available.
2641 res
= sd_do_mode_sense(sdkp
, 0, 0x3F, buffer
, 4, &data
, NULL
);
2644 * Second attempt: ask for page 0 When only page 0 is
2645 * implemented, a request for page 3F may return Sense Key
2646 * 5: Illegal Request, Sense Code 24: Invalid field in
2650 res
= sd_do_mode_sense(sdkp
, 0, 0, buffer
, 4, &data
, NULL
);
2653 * Third attempt: ask 255 bytes, as we did earlier.
2656 res
= sd_do_mode_sense(sdkp
, 0, 0x3F, buffer
, 255,
2661 sd_first_printk(KERN_WARNING
, sdkp
,
2662 "Test WP failed, assume Write Enabled\n");
2664 sdkp
->write_prot
= ((data
.device_specific
& 0x80) != 0);
2665 set_disk_ro(sdkp
->disk
, sdkp
->write_prot
);
2666 if (sdkp
->first_scan
|| old_wp
!= sdkp
->write_prot
) {
2667 sd_printk(KERN_NOTICE
, sdkp
, "Write Protect is %s\n",
2668 sdkp
->write_prot
? "on" : "off");
2669 sd_printk(KERN_DEBUG
, sdkp
, "Mode Sense: %4ph\n", buffer
);
2675 * sd_read_cache_type - called only from sd_revalidate_disk()
2676 * called with buffer of length SD_BUF_SIZE
2679 sd_read_cache_type(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2682 struct scsi_device
*sdp
= sdkp
->device
;
2687 struct scsi_mode_data data
;
2688 struct scsi_sense_hdr sshdr
;
2689 int old_wce
= sdkp
->WCE
;
2690 int old_rcd
= sdkp
->RCD
;
2691 int old_dpofua
= sdkp
->DPOFUA
;
2694 if (sdkp
->cache_override
)
2698 if (sdp
->skip_ms_page_8
) {
2699 if (sdp
->type
== TYPE_RBC
)
2702 if (sdp
->skip_ms_page_3f
)
2705 if (sdp
->use_192_bytes_for_3f
)
2709 } else if (sdp
->type
== TYPE_RBC
) {
2717 /* cautiously ask */
2718 res
= sd_do_mode_sense(sdkp
, dbd
, modepage
, buffer
, first_len
,
2724 if (!data
.header_length
) {
2727 sd_first_printk(KERN_ERR
, sdkp
,
2728 "Missing header in MODE_SENSE response\n");
2731 /* that went OK, now ask for the proper length */
2735 * We're only interested in the first three bytes, actually.
2736 * But the data cache page is defined for the first 20.
2740 else if (len
> SD_BUF_SIZE
) {
2741 sd_first_printk(KERN_NOTICE
, sdkp
, "Truncating mode parameter "
2742 "data from %d to %d bytes\n", len
, SD_BUF_SIZE
);
2745 if (modepage
== 0x3F && sdp
->use_192_bytes_for_3f
)
2749 if (len
> first_len
)
2750 res
= sd_do_mode_sense(sdkp
, dbd
, modepage
, buffer
, len
,
2754 int offset
= data
.header_length
+ data
.block_descriptor_length
;
2756 while (offset
< len
) {
2757 u8 page_code
= buffer
[offset
] & 0x3F;
2758 u8 spf
= buffer
[offset
] & 0x40;
2760 if (page_code
== 8 || page_code
== 6) {
2761 /* We're interested only in the first 3 bytes.
2763 if (len
- offset
<= 2) {
2764 sd_first_printk(KERN_ERR
, sdkp
,
2765 "Incomplete mode parameter "
2769 modepage
= page_code
;
2773 /* Go to the next page */
2774 if (spf
&& len
- offset
> 3)
2775 offset
+= 4 + (buffer
[offset
+2] << 8) +
2777 else if (!spf
&& len
- offset
> 1)
2778 offset
+= 2 + buffer
[offset
+1];
2780 sd_first_printk(KERN_ERR
, sdkp
,
2782 "parameter data\n");
2788 sd_first_printk(KERN_ERR
, sdkp
, "No Caching mode page found\n");
2792 if (modepage
== 8) {
2793 sdkp
->WCE
= ((buffer
[offset
+ 2] & 0x04) != 0);
2794 sdkp
->RCD
= ((buffer
[offset
+ 2] & 0x01) != 0);
2796 sdkp
->WCE
= ((buffer
[offset
+ 2] & 0x01) == 0);
2800 sdkp
->DPOFUA
= (data
.device_specific
& 0x10) != 0;
2801 if (sdp
->broken_fua
) {
2802 sd_first_printk(KERN_NOTICE
, sdkp
, "Disabling FUA\n");
2804 } else if (sdkp
->DPOFUA
&& !sdkp
->device
->use_10_for_rw
&&
2805 !sdkp
->device
->use_16_for_rw
) {
2806 sd_first_printk(KERN_NOTICE
, sdkp
,
2807 "Uses READ/WRITE(6), disabling FUA\n");
2811 /* No cache flush allowed for write protected devices */
2812 if (sdkp
->WCE
&& sdkp
->write_prot
)
2815 if (sdkp
->first_scan
|| old_wce
!= sdkp
->WCE
||
2816 old_rcd
!= sdkp
->RCD
|| old_dpofua
!= sdkp
->DPOFUA
)
2817 sd_printk(KERN_NOTICE
, sdkp
,
2818 "Write cache: %s, read cache: %s, %s\n",
2819 sdkp
->WCE
? "enabled" : "disabled",
2820 sdkp
->RCD
? "disabled" : "enabled",
2821 sdkp
->DPOFUA
? "supports DPO and FUA"
2822 : "doesn't support DPO or FUA");
2828 if (scsi_sense_valid(&sshdr
) &&
2829 sshdr
.sense_key
== ILLEGAL_REQUEST
&&
2830 sshdr
.asc
== 0x24 && sshdr
.ascq
== 0x0)
2831 /* Invalid field in CDB */
2832 sd_first_printk(KERN_NOTICE
, sdkp
, "Cache data unavailable\n");
2834 sd_first_printk(KERN_ERR
, sdkp
,
2835 "Asking for cache data failed\n");
2838 if (sdp
->wce_default_on
) {
2839 sd_first_printk(KERN_NOTICE
, sdkp
,
2840 "Assuming drive cache: write back\n");
2843 sd_first_printk(KERN_ERR
, sdkp
,
2844 "Assuming drive cache: write through\n");
2852 * The ATO bit indicates whether the DIF application tag is available
2853 * for use by the operating system.
2855 static void sd_read_app_tag_own(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2858 struct scsi_device
*sdp
= sdkp
->device
;
2859 struct scsi_mode_data data
;
2860 struct scsi_sense_hdr sshdr
;
2862 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_ZBC
)
2865 if (sdkp
->protection_type
== 0)
2868 res
= scsi_mode_sense(sdp
, 1, 0x0a, buffer
, 36, SD_TIMEOUT
,
2869 sdkp
->max_retries
, &data
, &sshdr
);
2871 if (res
< 0 || !data
.header_length
||
2873 sd_first_printk(KERN_WARNING
, sdkp
,
2874 "getting Control mode page failed, assume no ATO\n");
2876 if (scsi_sense_valid(&sshdr
))
2877 sd_print_sense_hdr(sdkp
, &sshdr
);
2882 offset
= data
.header_length
+ data
.block_descriptor_length
;
2884 if ((buffer
[offset
] & 0x3f) != 0x0a) {
2885 sd_first_printk(KERN_ERR
, sdkp
, "ATO Got wrong page\n");
2889 if ((buffer
[offset
+ 5] & 0x80) == 0)
2898 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2899 * @sdkp: disk to query
2901 static void sd_read_block_limits(struct scsi_disk
*sdkp
)
2903 unsigned int sector_sz
= sdkp
->device
->sector_size
;
2904 const int vpd_len
= 64;
2905 unsigned char *buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2908 /* Block Limits VPD */
2909 scsi_get_vpd_page(sdkp
->device
, 0xb0, buffer
, vpd_len
))
2912 blk_queue_io_min(sdkp
->disk
->queue
,
2913 get_unaligned_be16(&buffer
[6]) * sector_sz
);
2915 sdkp
->max_xfer_blocks
= get_unaligned_be32(&buffer
[8]);
2916 sdkp
->opt_xfer_blocks
= get_unaligned_be32(&buffer
[12]);
2918 if (buffer
[3] == 0x3c) {
2919 unsigned int lba_count
, desc_count
;
2921 sdkp
->max_ws_blocks
= (u32
)get_unaligned_be64(&buffer
[36]);
2926 lba_count
= get_unaligned_be32(&buffer
[20]);
2927 desc_count
= get_unaligned_be32(&buffer
[24]);
2929 if (lba_count
&& desc_count
)
2930 sdkp
->max_unmap_blocks
= lba_count
;
2932 sdkp
->unmap_granularity
= get_unaligned_be32(&buffer
[28]);
2934 if (buffer
[32] & 0x80)
2935 sdkp
->unmap_alignment
=
2936 get_unaligned_be32(&buffer
[32]) & ~(1 << 31);
2938 if (!sdkp
->lbpvpd
) { /* LBP VPD page not provided */
2940 if (sdkp
->max_unmap_blocks
)
2941 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2943 sd_config_discard(sdkp
, SD_LBP_WS16
);
2945 } else { /* LBP VPD page tells us what to use */
2946 if (sdkp
->lbpu
&& sdkp
->max_unmap_blocks
)
2947 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2948 else if (sdkp
->lbpws
)
2949 sd_config_discard(sdkp
, SD_LBP_WS16
);
2950 else if (sdkp
->lbpws10
)
2951 sd_config_discard(sdkp
, SD_LBP_WS10
);
2953 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
2962 * sd_read_block_characteristics - Query block dev. characteristics
2963 * @sdkp: disk to query
2965 static void sd_read_block_characteristics(struct scsi_disk
*sdkp
)
2967 struct request_queue
*q
= sdkp
->disk
->queue
;
2968 unsigned char *buffer
;
2970 const int vpd_len
= 64;
2972 buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2975 /* Block Device Characteristics VPD */
2976 scsi_get_vpd_page(sdkp
->device
, 0xb1, buffer
, vpd_len
))
2979 rot
= get_unaligned_be16(&buffer
[4]);
2982 blk_queue_flag_set(QUEUE_FLAG_NONROT
, q
);
2983 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM
, q
);
2986 if (sdkp
->device
->type
== TYPE_ZBC
) {
2988 blk_queue_set_zoned(sdkp
->disk
, BLK_ZONED_HM
);
2990 sdkp
->zoned
= (buffer
[8] >> 4) & 3;
2991 if (sdkp
->zoned
== 1) {
2993 blk_queue_set_zoned(sdkp
->disk
, BLK_ZONED_HA
);
2995 /* Regular disk or drive managed disk */
2996 blk_queue_set_zoned(sdkp
->disk
, BLK_ZONED_NONE
);
3000 if (!sdkp
->first_scan
)
3003 if (blk_queue_is_zoned(q
)) {
3004 sd_printk(KERN_NOTICE
, sdkp
, "Host-%s zoned block device\n",
3005 q
->limits
.zoned
== BLK_ZONED_HM
? "managed" : "aware");
3007 if (sdkp
->zoned
== 1)
3008 sd_printk(KERN_NOTICE
, sdkp
,
3009 "Host-aware SMR disk used as regular disk\n");
3010 else if (sdkp
->zoned
== 2)
3011 sd_printk(KERN_NOTICE
, sdkp
,
3012 "Drive-managed SMR disk\n");
3020 * sd_read_block_provisioning - Query provisioning VPD page
3021 * @sdkp: disk to query
3023 static void sd_read_block_provisioning(struct scsi_disk
*sdkp
)
3025 unsigned char *buffer
;
3026 const int vpd_len
= 8;
3028 if (sdkp
->lbpme
== 0)
3031 buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
3033 if (!buffer
|| scsi_get_vpd_page(sdkp
->device
, 0xb2, buffer
, vpd_len
))
3037 sdkp
->lbpu
= (buffer
[5] >> 7) & 1; /* UNMAP */
3038 sdkp
->lbpws
= (buffer
[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3039 sdkp
->lbpws10
= (buffer
[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3045 static void sd_read_write_same(struct scsi_disk
*sdkp
, unsigned char *buffer
)
3047 struct scsi_device
*sdev
= sdkp
->device
;
3049 if (sdev
->host
->no_write_same
) {
3050 sdev
->no_write_same
= 1;
3055 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, INQUIRY
) < 0) {
3056 /* too large values might cause issues with arcmsr */
3057 int vpd_buf_len
= 64;
3059 sdev
->no_report_opcodes
= 1;
3061 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3062 * CODES is unsupported and the device has an ATA
3063 * Information VPD page (SAT).
3065 if (!scsi_get_vpd_page(sdev
, 0x89, buffer
, vpd_buf_len
))
3066 sdev
->no_write_same
= 1;
3069 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, WRITE_SAME_16
) == 1)
3072 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, WRITE_SAME
) == 1)
3076 static void sd_read_security(struct scsi_disk
*sdkp
, unsigned char *buffer
)
3078 struct scsi_device
*sdev
= sdkp
->device
;
3080 if (!sdev
->security_supported
)
3083 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
,
3084 SECURITY_PROTOCOL_IN
) == 1 &&
3085 scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
,
3086 SECURITY_PROTOCOL_OUT
) == 1)
3091 * Determine the device's preferred I/O size for reads and writes
3092 * unless the reported value is unreasonably small, large, not a
3093 * multiple of the physical block size, or simply garbage.
3095 static bool sd_validate_opt_xfer_size(struct scsi_disk
*sdkp
,
3096 unsigned int dev_max
)
3098 struct scsi_device
*sdp
= sdkp
->device
;
3099 unsigned int opt_xfer_bytes
=
3100 logical_to_bytes(sdp
, sdkp
->opt_xfer_blocks
);
3102 if (sdkp
->opt_xfer_blocks
== 0)
3105 if (sdkp
->opt_xfer_blocks
> dev_max
) {
3106 sd_first_printk(KERN_WARNING
, sdkp
,
3107 "Optimal transfer size %u logical blocks " \
3108 "> dev_max (%u logical blocks)\n",
3109 sdkp
->opt_xfer_blocks
, dev_max
);
3113 if (sdkp
->opt_xfer_blocks
> SD_DEF_XFER_BLOCKS
) {
3114 sd_first_printk(KERN_WARNING
, sdkp
,
3115 "Optimal transfer size %u logical blocks " \
3116 "> sd driver limit (%u logical blocks)\n",
3117 sdkp
->opt_xfer_blocks
, SD_DEF_XFER_BLOCKS
);
3121 if (opt_xfer_bytes
< PAGE_SIZE
) {
3122 sd_first_printk(KERN_WARNING
, sdkp
,
3123 "Optimal transfer size %u bytes < " \
3124 "PAGE_SIZE (%u bytes)\n",
3125 opt_xfer_bytes
, (unsigned int)PAGE_SIZE
);
3129 if (opt_xfer_bytes
& (sdkp
->physical_block_size
- 1)) {
3130 sd_first_printk(KERN_WARNING
, sdkp
,
3131 "Optimal transfer size %u bytes not a " \
3132 "multiple of physical block size (%u bytes)\n",
3133 opt_xfer_bytes
, sdkp
->physical_block_size
);
3137 sd_first_printk(KERN_INFO
, sdkp
, "Optimal transfer size %u bytes\n",
3143 * sd_revalidate_disk - called the first time a new disk is seen,
3144 * performs disk spin up, read_capacity, etc.
3145 * @disk: struct gendisk we care about
3147 static int sd_revalidate_disk(struct gendisk
*disk
)
3149 struct scsi_disk
*sdkp
= scsi_disk(disk
);
3150 struct scsi_device
*sdp
= sdkp
->device
;
3151 struct request_queue
*q
= sdkp
->disk
->queue
;
3152 sector_t old_capacity
= sdkp
->capacity
;
3153 unsigned char *buffer
;
3154 unsigned int dev_max
, rw_max
;
3156 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
,
3157 "sd_revalidate_disk\n"));
3160 * If the device is offline, don't try and read capacity or any
3161 * of the other niceties.
3163 if (!scsi_device_online(sdp
))
3166 buffer
= kmalloc(SD_BUF_SIZE
, GFP_KERNEL
);
3168 sd_printk(KERN_WARNING
, sdkp
, "sd_revalidate_disk: Memory "
3169 "allocation failure.\n");
3173 sd_spinup_disk(sdkp
);
3176 * Without media there is no reason to ask; moreover, some devices
3177 * react badly if we do.
3179 if (sdkp
->media_present
) {
3180 sd_read_capacity(sdkp
, buffer
);
3183 * set the default to rotational. All non-rotational devices
3184 * support the block characteristics VPD page, which will
3185 * cause this to be updated correctly and any device which
3186 * doesn't support it should be treated as rotational.
3188 blk_queue_flag_clear(QUEUE_FLAG_NONROT
, q
);
3189 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM
, q
);
3191 if (scsi_device_supports_vpd(sdp
)) {
3192 sd_read_block_provisioning(sdkp
);
3193 sd_read_block_limits(sdkp
);
3194 sd_read_block_characteristics(sdkp
);
3195 sd_zbc_read_zones(sdkp
, buffer
);
3198 sd_print_capacity(sdkp
, old_capacity
);
3200 sd_read_write_protect_flag(sdkp
, buffer
);
3201 sd_read_cache_type(sdkp
, buffer
);
3202 sd_read_app_tag_own(sdkp
, buffer
);
3203 sd_read_write_same(sdkp
, buffer
);
3204 sd_read_security(sdkp
, buffer
);
3208 * We now have all cache related info, determine how we deal
3209 * with flush requests.
3211 sd_set_flush_flag(sdkp
);
3213 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3214 dev_max
= sdp
->use_16_for_rw
? SD_MAX_XFER_BLOCKS
: SD_DEF_XFER_BLOCKS
;
3216 /* Some devices report a maximum block count for READ/WRITE requests. */
3217 dev_max
= min_not_zero(dev_max
, sdkp
->max_xfer_blocks
);
3218 q
->limits
.max_dev_sectors
= logical_to_sectors(sdp
, dev_max
);
3220 if (sd_validate_opt_xfer_size(sdkp
, dev_max
)) {
3221 q
->limits
.io_opt
= logical_to_bytes(sdp
, sdkp
->opt_xfer_blocks
);
3222 rw_max
= logical_to_sectors(sdp
, sdkp
->opt_xfer_blocks
);
3224 q
->limits
.io_opt
= 0;
3225 rw_max
= min_not_zero(logical_to_sectors(sdp
, dev_max
),
3226 (sector_t
)BLK_DEF_MAX_SECTORS
);
3229 /* Do not exceed controller limit */
3230 rw_max
= min(rw_max
, queue_max_hw_sectors(q
));
3233 * Only update max_sectors if previously unset or if the current value
3234 * exceeds the capabilities of the hardware.
3236 if (sdkp
->first_scan
||
3237 q
->limits
.max_sectors
> q
->limits
.max_dev_sectors
||
3238 q
->limits
.max_sectors
> q
->limits
.max_hw_sectors
)
3239 q
->limits
.max_sectors
= rw_max
;
3241 sdkp
->first_scan
= 0;
3243 set_capacity_and_notify(disk
, logical_to_sectors(sdp
, sdkp
->capacity
));
3244 sd_config_write_same(sdkp
);
3248 * For a zoned drive, revalidating the zones can be done only once
3249 * the gendisk capacity is set. So if this fails, set back the gendisk
3252 if (sd_zbc_revalidate_zones(sdkp
))
3253 set_capacity_and_notify(disk
, 0);
3260 * sd_unlock_native_capacity - unlock native capacity
3261 * @disk: struct gendisk to set capacity for
3263 * Block layer calls this function if it detects that partitions
3264 * on @disk reach beyond the end of the device. If the SCSI host
3265 * implements ->unlock_native_capacity() method, it's invoked to
3266 * give it a chance to adjust the device capacity.
3269 * Defined by block layer. Might sleep.
3271 static void sd_unlock_native_capacity(struct gendisk
*disk
)
3273 struct scsi_device
*sdev
= scsi_disk(disk
)->device
;
3275 if (sdev
->host
->hostt
->unlock_native_capacity
)
3276 sdev
->host
->hostt
->unlock_native_capacity(sdev
);
3280 * sd_format_disk_name - format disk name
3281 * @prefix: name prefix - ie. "sd" for SCSI disks
3282 * @index: index of the disk to format name for
3283 * @buf: output buffer
3284 * @buflen: length of the output buffer
3286 * SCSI disk names starts at sda. The 26th device is sdz and the
3287 * 27th is sdaa. The last one for two lettered suffix is sdzz
3288 * which is followed by sdaaa.
3290 * This is basically 26 base counting with one extra 'nil' entry
3291 * at the beginning from the second digit on and can be
3292 * determined using similar method as 26 base conversion with the
3293 * index shifted -1 after each digit is computed.
3299 * 0 on success, -errno on failure.
3301 static int sd_format_disk_name(char *prefix
, int index
, char *buf
, int buflen
)
3303 const int base
= 'z' - 'a' + 1;
3304 char *begin
= buf
+ strlen(prefix
);
3305 char *end
= buf
+ buflen
;
3315 *--p
= 'a' + (index
% unit
);
3316 index
= (index
/ unit
) - 1;
3317 } while (index
>= 0);
3319 memmove(begin
, p
, end
- p
);
3320 memcpy(buf
, prefix
, strlen(prefix
));
3326 * sd_probe - called during driver initialization and whenever a
3327 * new scsi device is attached to the system. It is called once
3328 * for each scsi device (not just disks) present.
3329 * @dev: pointer to device object
3331 * Returns 0 if successful (or not interested in this scsi device
3332 * (e.g. scanner)); 1 when there is an error.
3334 * Note: this function is invoked from the scsi mid-level.
3335 * This function sets up the mapping between a given
3336 * <host,channel,id,lun> (found in sdp) and new device name
3337 * (e.g. /dev/sda). More precisely it is the block device major
3338 * and minor number that is chosen here.
3340 * Assume sd_probe is not re-entrant (for time being)
3341 * Also think about sd_probe() and sd_remove() running coincidentally.
3343 static int sd_probe(struct device
*dev
)
3345 struct scsi_device
*sdp
= to_scsi_device(dev
);
3346 struct scsi_disk
*sdkp
;
3351 scsi_autopm_get_device(sdp
);
3353 if (sdp
->type
!= TYPE_DISK
&&
3354 sdp
->type
!= TYPE_ZBC
&&
3355 sdp
->type
!= TYPE_MOD
&&
3356 sdp
->type
!= TYPE_RBC
)
3359 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED
) && sdp
->type
== TYPE_ZBC
) {
3360 sdev_printk(KERN_WARNING
, sdp
,
3361 "Unsupported ZBC host-managed device.\n");
3365 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO
, sdp
,
3369 sdkp
= kzalloc(sizeof(*sdkp
), GFP_KERNEL
);
3373 gd
= __alloc_disk_node(sdp
->request_queue
, NUMA_NO_NODE
,
3374 &sd_bio_compl_lkclass
);
3378 index
= ida_alloc(&sd_index_ida
, GFP_KERNEL
);
3380 sdev_printk(KERN_WARNING
, sdp
, "sd_probe: memory exhausted.\n");
3384 error
= sd_format_disk_name("sd", index
, gd
->disk_name
, DISK_NAME_LEN
);
3386 sdev_printk(KERN_WARNING
, sdp
, "SCSI disk (sd) name length exceeded.\n");
3387 goto out_free_index
;
3391 sdkp
->driver
= &sd_template
;
3393 sdkp
->index
= index
;
3394 sdkp
->max_retries
= SD_MAX_RETRIES
;
3395 atomic_set(&sdkp
->openers
, 0);
3396 atomic_set(&sdkp
->device
->ioerr_cnt
, 0);
3398 if (!sdp
->request_queue
->rq_timeout
) {
3399 if (sdp
->type
!= TYPE_MOD
)
3400 blk_queue_rq_timeout(sdp
->request_queue
, SD_TIMEOUT
);
3402 blk_queue_rq_timeout(sdp
->request_queue
,
3406 device_initialize(&sdkp
->dev
);
3407 sdkp
->dev
.parent
= get_device(dev
);
3408 sdkp
->dev
.class = &sd_disk_class
;
3409 dev_set_name(&sdkp
->dev
, "%s", dev_name(dev
));
3411 error
= device_add(&sdkp
->dev
);
3413 put_device(&sdkp
->dev
);
3417 dev_set_drvdata(dev
, sdkp
);
3419 gd
->major
= sd_major((index
& 0xf0) >> 4);
3420 gd
->first_minor
= ((index
& 0xf) << 4) | (index
& 0xfff00);
3421 gd
->minors
= SD_MINORS
;
3423 gd
->fops
= &sd_fops
;
3424 gd
->private_data
= &sdkp
->driver
;
3426 /* defaults, until the device tells us otherwise */
3427 sdp
->sector_size
= 512;
3429 sdkp
->media_present
= 1;
3430 sdkp
->write_prot
= 0;
3431 sdkp
->cache_override
= 0;
3435 sdkp
->first_scan
= 1;
3436 sdkp
->max_medium_access_timeouts
= SD_MAX_MEDIUM_TIMEOUTS
;
3438 sd_revalidate_disk(gd
);
3440 gd
->flags
= GENHD_FL_EXT_DEVT
;
3441 if (sdp
->removable
) {
3442 gd
->flags
|= GENHD_FL_REMOVABLE
;
3443 gd
->events
|= DISK_EVENT_MEDIA_CHANGE
;
3444 gd
->event_flags
= DISK_EVENT_FLAG_POLL
| DISK_EVENT_FLAG_UEVENT
;
3447 blk_pm_runtime_init(sdp
->request_queue
, dev
);
3448 if (sdp
->rpm_autosuspend
) {
3449 pm_runtime_set_autosuspend_delay(dev
,
3450 sdp
->host
->hostt
->rpm_autosuspend_delay
);
3452 device_add_disk(dev
, gd
, NULL
);
3454 sd_dif_config_host(sdkp
);
3456 sd_revalidate_disk(gd
);
3458 if (sdkp
->security
) {
3459 sdkp
->opal_dev
= init_opal_dev(sdkp
, &sd_sec_submit
);
3461 sd_printk(KERN_NOTICE
, sdkp
, "supports TCG Opal\n");
3464 sd_printk(KERN_NOTICE
, sdkp
, "Attached SCSI %sdisk\n",
3465 sdp
->removable
? "removable " : "");
3466 scsi_autopm_put_device(sdp
);
3471 ida_free(&sd_index_ida
, index
);
3475 sd_zbc_release_disk(sdkp
);
3478 scsi_autopm_put_device(sdp
);
3483 * sd_remove - called whenever a scsi disk (previously recognized by
3484 * sd_probe) is detached from the system. It is called (potentially
3485 * multiple times) during sd module unload.
3486 * @dev: pointer to device object
3488 * Note: this function is invoked from the scsi mid-level.
3489 * This function potentially frees up a device name (e.g. /dev/sdc)
3490 * that could be re-used by a subsequent sd_probe().
3491 * This function is not called when the built-in sd driver is "exit-ed".
3493 static int sd_remove(struct device
*dev
)
3495 struct scsi_disk
*sdkp
;
3497 sdkp
= dev_get_drvdata(dev
);
3498 scsi_autopm_get_device(sdkp
->device
);
3500 async_synchronize_full_domain(&scsi_sd_pm_domain
);
3501 device_del(&sdkp
->dev
);
3502 del_gendisk(sdkp
->disk
);
3505 free_opal_dev(sdkp
->opal_dev
);
3507 mutex_lock(&sd_ref_mutex
);
3508 dev_set_drvdata(dev
, NULL
);
3509 put_device(&sdkp
->dev
);
3510 mutex_unlock(&sd_ref_mutex
);
3516 * scsi_disk_release - Called to free the scsi_disk structure
3517 * @dev: pointer to embedded class device
3519 * sd_ref_mutex must be held entering this routine. Because it is
3520 * called on last put, you should always use the scsi_disk_get()
3521 * scsi_disk_put() helpers which manipulate the semaphore directly
3522 * and never do a direct put_device.
3524 static void scsi_disk_release(struct device
*dev
)
3526 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
3527 struct gendisk
*disk
= sdkp
->disk
;
3528 struct request_queue
*q
= disk
->queue
;
3530 ida_free(&sd_index_ida
, sdkp
->index
);
3533 * Wait until all requests that are in progress have completed.
3534 * This is necessary to avoid that e.g. scsi_end_request() crashes
3535 * due to clearing the disk->private_data pointer. Wait from inside
3536 * scsi_disk_release() instead of from sd_release() to avoid that
3537 * freezing and unfreezing the request queue affects user space I/O
3538 * in case multiple processes open a /dev/sd... node concurrently.
3540 blk_mq_freeze_queue(q
);
3541 blk_mq_unfreeze_queue(q
);
3543 disk
->private_data
= NULL
;
3545 put_device(&sdkp
->device
->sdev_gendev
);
3547 sd_zbc_release_disk(sdkp
);
3552 static int sd_start_stop_device(struct scsi_disk
*sdkp
, int start
)
3554 unsigned char cmd
[6] = { START_STOP
}; /* START_VALID */
3555 struct scsi_sense_hdr sshdr
;
3556 struct scsi_device
*sdp
= sdkp
->device
;
3560 cmd
[4] |= 1; /* START */
3562 if (sdp
->start_stop_pwr_cond
)
3563 cmd
[4] |= start
? 1 << 4 : 3 << 4; /* Active or Standby */
3565 if (!scsi_device_online(sdp
))
3568 res
= scsi_execute(sdp
, cmd
, DMA_NONE
, NULL
, 0, NULL
, &sshdr
,
3569 SD_TIMEOUT
, sdkp
->max_retries
, 0, RQF_PM
, NULL
);
3571 sd_print_result(sdkp
, "Start/Stop Unit failed", res
);
3572 if (res
> 0 && scsi_sense_valid(&sshdr
)) {
3573 sd_print_sense_hdr(sdkp
, &sshdr
);
3574 /* 0x3a is medium not present */
3575 if (sshdr
.asc
== 0x3a)
3580 /* SCSI error codes must not go to the generic layer */
3588 * Send a SYNCHRONIZE CACHE instruction down to the device through
3589 * the normal SCSI command structure. Wait for the command to
3592 static void sd_shutdown(struct device
*dev
)
3594 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3597 return; /* this can happen */
3599 if (pm_runtime_suspended(dev
))
3602 if (sdkp
->WCE
&& sdkp
->media_present
) {
3603 sd_printk(KERN_NOTICE
, sdkp
, "Synchronizing SCSI cache\n");
3604 sd_sync_cache(sdkp
, NULL
);
3607 if (system_state
!= SYSTEM_RESTART
&& sdkp
->device
->manage_start_stop
) {
3608 sd_printk(KERN_NOTICE
, sdkp
, "Stopping disk\n");
3609 sd_start_stop_device(sdkp
, 0);
3613 static int sd_suspend_common(struct device
*dev
, bool ignore_stop_errors
)
3615 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3616 struct scsi_sense_hdr sshdr
;
3619 if (!sdkp
) /* E.g.: runtime suspend following sd_remove() */
3622 if (sdkp
->WCE
&& sdkp
->media_present
) {
3623 sd_printk(KERN_NOTICE
, sdkp
, "Synchronizing SCSI cache\n");
3624 ret
= sd_sync_cache(sdkp
, &sshdr
);
3627 /* ignore OFFLINE device */
3631 if (!scsi_sense_valid(&sshdr
) ||
3632 sshdr
.sense_key
!= ILLEGAL_REQUEST
)
3636 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3637 * doesn't support sync. There's not much to do and
3638 * suspend shouldn't fail.
3644 if (sdkp
->device
->manage_start_stop
) {
3645 sd_printk(KERN_NOTICE
, sdkp
, "Stopping disk\n");
3646 /* an error is not worth aborting a system sleep */
3647 ret
= sd_start_stop_device(sdkp
, 0);
3648 if (ignore_stop_errors
)
3655 static int sd_suspend_system(struct device
*dev
)
3657 return sd_suspend_common(dev
, true);
3660 static int sd_suspend_runtime(struct device
*dev
)
3662 return sd_suspend_common(dev
, false);
3665 static int sd_resume(struct device
*dev
)
3667 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3670 if (!sdkp
) /* E.g.: runtime resume at the start of sd_probe() */
3673 if (!sdkp
->device
->manage_start_stop
)
3676 sd_printk(KERN_NOTICE
, sdkp
, "Starting disk\n");
3677 ret
= sd_start_stop_device(sdkp
, 1);
3679 opal_unlock_from_suspend(sdkp
->opal_dev
);
3683 static int sd_resume_runtime(struct device
*dev
)
3685 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3686 struct scsi_device
*sdp
;
3688 if (!sdkp
) /* E.g.: runtime resume at the start of sd_probe() */
3693 if (sdp
->ignore_media_change
) {
3694 /* clear the device's sense data */
3695 static const u8 cmd
[10] = { REQUEST_SENSE
};
3697 if (scsi_execute(sdp
, cmd
, DMA_NONE
, NULL
, 0, NULL
,
3698 NULL
, sdp
->request_queue
->rq_timeout
, 1, 0,
3700 sd_printk(KERN_NOTICE
, sdkp
,
3701 "Failed to clear sense data\n");
3704 return sd_resume(dev
);
3708 * init_sd - entry point for this driver (both when built in or when
3711 * Note: this function registers this driver with the scsi mid-level.
3713 static int __init
init_sd(void)
3715 int majors
= 0, i
, err
;
3717 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3719 for (i
= 0; i
< SD_MAJORS
; i
++) {
3720 if (__register_blkdev(sd_major(i
), "sd", sd_default_probe
))
3728 err
= class_register(&sd_disk_class
);
3732 sd_cdb_cache
= kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE
,
3734 if (!sd_cdb_cache
) {
3735 printk(KERN_ERR
"sd: can't init extended cdb cache\n");
3740 sd_cdb_pool
= mempool_create_slab_pool(SD_MEMPOOL_SIZE
, sd_cdb_cache
);
3742 printk(KERN_ERR
"sd: can't init extended cdb pool\n");
3747 sd_page_pool
= mempool_create_page_pool(SD_MEMPOOL_SIZE
, 0);
3748 if (!sd_page_pool
) {
3749 printk(KERN_ERR
"sd: can't init discard page pool\n");
3754 err
= scsi_register_driver(&sd_template
.gendrv
);
3756 goto err_out_driver
;
3761 mempool_destroy(sd_page_pool
);
3764 mempool_destroy(sd_cdb_pool
);
3767 kmem_cache_destroy(sd_cdb_cache
);
3770 class_unregister(&sd_disk_class
);
3772 for (i
= 0; i
< SD_MAJORS
; i
++)
3773 unregister_blkdev(sd_major(i
), "sd");
3778 * exit_sd - exit point for this driver (when it is a module).
3780 * Note: this function unregisters this driver from the scsi mid-level.
3782 static void __exit
exit_sd(void)
3786 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3788 scsi_unregister_driver(&sd_template
.gendrv
);
3789 mempool_destroy(sd_cdb_pool
);
3790 mempool_destroy(sd_page_pool
);
3791 kmem_cache_destroy(sd_cdb_cache
);
3793 class_unregister(&sd_disk_class
);
3795 for (i
= 0; i
< SD_MAJORS
; i
++)
3796 unregister_blkdev(sd_major(i
), "sd");
3799 module_init(init_sd
);
3800 module_exit(exit_sd
);
3802 void sd_print_sense_hdr(struct scsi_disk
*sdkp
, struct scsi_sense_hdr
*sshdr
)
3804 scsi_print_sense_hdr(sdkp
->device
,
3805 sdkp
->disk
? sdkp
->disk
->disk_name
: NULL
, sshdr
);
3808 void sd_print_result(const struct scsi_disk
*sdkp
, const char *msg
, int result
)
3810 const char *hb_string
= scsi_hostbyte_string(result
);
3813 sd_printk(KERN_INFO
, sdkp
,
3814 "%s: Result: hostbyte=%s driverbyte=%s\n", msg
,
3815 hb_string
? hb_string
: "invalid",
3818 sd_printk(KERN_INFO
, sdkp
,
3819 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3820 msg
, host_byte(result
), "DRIVER_OK");