2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24 * Support 32k/1M disks.
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
35 #include <linux/module.h>
37 #include <linux/kernel.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
55 #include <linux/t10-pi.h>
56 #include <linux/uaccess.h>
57 #include <asm/unaligned.h>
59 #include <scsi/scsi.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_dbg.h>
62 #include <scsi/scsi_device.h>
63 #include <scsi/scsi_driver.h>
64 #include <scsi/scsi_eh.h>
65 #include <scsi/scsi_host.h>
66 #include <scsi/scsi_ioctl.h>
67 #include <scsi/scsicam.h>
70 #include "scsi_priv.h"
71 #include "scsi_logging.h"
73 MODULE_AUTHOR("Eric Youngdale");
74 MODULE_DESCRIPTION("SCSI disk (sd) driver");
75 MODULE_LICENSE("GPL");
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR
);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR
);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR
);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR
);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR
);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR
);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR
);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR
);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR
);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR
);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR
);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR
);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR
);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR
);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR
);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR
);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK
);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD
);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC
);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC
);
98 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
104 static void sd_config_discard(struct scsi_disk
*, unsigned int);
105 static void sd_config_write_same(struct scsi_disk
*);
106 static int sd_revalidate_disk(struct gendisk
*);
107 static void sd_unlock_native_capacity(struct gendisk
*disk
);
108 static int sd_probe(struct device
*);
109 static int sd_remove(struct device
*);
110 static void sd_shutdown(struct device
*);
111 static int sd_suspend_system(struct device
*);
112 static int sd_suspend_runtime(struct device
*);
113 static int sd_resume(struct device
*);
114 static void sd_rescan(struct device
*);
115 static int 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 int sd_eh_action(struct scsi_cmnd
*, int);
119 static void sd_read_capacity(struct scsi_disk
*sdkp
, unsigned char *buffer
);
120 static void scsi_disk_release(struct device
*cdev
);
121 static void sd_print_sense_hdr(struct scsi_disk
*, struct scsi_sense_hdr
*);
122 static void sd_print_result(const struct scsi_disk
*, const char *, int);
124 static DEFINE_SPINLOCK(sd_index_lock
);
125 static DEFINE_IDA(sd_index_ida
);
127 /* This semaphore is used to mediate the 0->1 reference get in the
128 * face of object destruction (i.e. we can't allow a get on an
129 * object after last put) */
130 static DEFINE_MUTEX(sd_ref_mutex
);
132 static struct kmem_cache
*sd_cdb_cache
;
133 static mempool_t
*sd_cdb_pool
;
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 i
, ct
= -1, 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 for (i
= 0; i
< ARRAY_SIZE(sd_cache_types
); i
++) {
181 len
= strlen(sd_cache_types
[i
]);
182 if (strncmp(sd_cache_types
[i
], buf
, len
) == 0 &&
190 rcd
= ct
& 0x01 ? 1 : 0;
191 wce
= (ct
& 0x02) && !sdkp
->write_prot
? 1 : 0;
193 if (sdkp
->cache_override
) {
196 sd_set_flush_flag(sdkp
);
200 if (scsi_mode_sense(sdp
, 0x08, 8, buffer
, sizeof(buffer
), SD_TIMEOUT
,
201 SD_MAX_RETRIES
, &data
, NULL
))
203 len
= min_t(size_t, sizeof(buffer
), data
.length
- data
.header_length
-
204 data
.block_descriptor_length
);
205 buffer_data
= buffer
+ data
.header_length
+
206 data
.block_descriptor_length
;
207 buffer_data
[2] &= ~0x05;
208 buffer_data
[2] |= wce
<< 2 | rcd
;
209 sp
= buffer_data
[0] & 0x80 ? 1 : 0;
210 buffer_data
[0] &= ~0x80;
212 if (scsi_mode_select(sdp
, 1, sp
, 8, buffer_data
, len
, SD_TIMEOUT
,
213 SD_MAX_RETRIES
, &data
, &sshdr
)) {
214 if (scsi_sense_valid(&sshdr
))
215 sd_print_sense_hdr(sdkp
, &sshdr
);
218 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 snprintf(buf
, 20, "%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
;
239 if (!capable(CAP_SYS_ADMIN
))
242 sdp
->manage_start_stop
= simple_strtoul(buf
, NULL
, 10);
246 static DEVICE_ATTR_RW(manage_start_stop
);
249 allow_restart_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
251 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
253 return snprintf(buf
, 40, "%d\n", sdkp
->device
->allow_restart
);
257 allow_restart_store(struct device
*dev
, struct device_attribute
*attr
,
258 const char *buf
, size_t count
)
260 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
261 struct scsi_device
*sdp
= sdkp
->device
;
263 if (!capable(CAP_SYS_ADMIN
))
266 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_ZBC
)
269 sdp
->allow_restart
= simple_strtoul(buf
, NULL
, 10);
273 static DEVICE_ATTR_RW(allow_restart
);
276 cache_type_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
278 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
279 int ct
= sdkp
->RCD
+ 2*sdkp
->WCE
;
281 return snprintf(buf
, 40, "%s\n", sd_cache_types
[ct
]);
283 static DEVICE_ATTR_RW(cache_type
);
286 FUA_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
288 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
290 return snprintf(buf
, 20, "%u\n", sdkp
->DPOFUA
);
292 static DEVICE_ATTR_RO(FUA
);
295 protection_type_show(struct device
*dev
, struct device_attribute
*attr
,
298 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
300 return snprintf(buf
, 20, "%u\n", sdkp
->protection_type
);
304 protection_type_store(struct device
*dev
, struct device_attribute
*attr
,
305 const char *buf
, size_t count
)
307 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
311 if (!capable(CAP_SYS_ADMIN
))
314 err
= kstrtouint(buf
, 10, &val
);
319 if (val
>= 0 && val
<= T10_PI_TYPE3_PROTECTION
)
320 sdkp
->protection_type
= val
;
324 static DEVICE_ATTR_RW(protection_type
);
327 protection_mode_show(struct device
*dev
, struct device_attribute
*attr
,
330 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
331 struct scsi_device
*sdp
= sdkp
->device
;
332 unsigned int dif
, dix
;
334 dif
= scsi_host_dif_capable(sdp
->host
, sdkp
->protection_type
);
335 dix
= scsi_host_dix_capable(sdp
->host
, sdkp
->protection_type
);
337 if (!dix
&& scsi_host_dix_capable(sdp
->host
, T10_PI_TYPE0_PROTECTION
)) {
343 return snprintf(buf
, 20, "none\n");
345 return snprintf(buf
, 20, "%s%u\n", dix
? "dix" : "dif", dif
);
347 static DEVICE_ATTR_RO(protection_mode
);
350 app_tag_own_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
352 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
354 return snprintf(buf
, 20, "%u\n", sdkp
->ATO
);
356 static DEVICE_ATTR_RO(app_tag_own
);
359 thin_provisioning_show(struct device
*dev
, struct device_attribute
*attr
,
362 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
364 return snprintf(buf
, 20, "%u\n", sdkp
->lbpme
);
366 static DEVICE_ATTR_RO(thin_provisioning
);
368 static const char *lbp_mode
[] = {
369 [SD_LBP_FULL
] = "full",
370 [SD_LBP_UNMAP
] = "unmap",
371 [SD_LBP_WS16
] = "writesame_16",
372 [SD_LBP_WS10
] = "writesame_10",
373 [SD_LBP_ZERO
] = "writesame_zero",
374 [SD_LBP_DISABLE
] = "disabled",
378 provisioning_mode_show(struct device
*dev
, struct device_attribute
*attr
,
381 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
383 return snprintf(buf
, 20, "%s\n", lbp_mode
[sdkp
->provisioning_mode
]);
387 provisioning_mode_store(struct device
*dev
, struct device_attribute
*attr
,
388 const char *buf
, size_t count
)
390 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
391 struct scsi_device
*sdp
= sdkp
->device
;
393 if (!capable(CAP_SYS_ADMIN
))
396 if (sd_is_zoned(sdkp
)) {
397 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
401 if (sdp
->type
!= TYPE_DISK
)
404 if (!strncmp(buf
, lbp_mode
[SD_LBP_UNMAP
], 20))
405 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
406 else if (!strncmp(buf
, lbp_mode
[SD_LBP_WS16
], 20))
407 sd_config_discard(sdkp
, SD_LBP_WS16
);
408 else if (!strncmp(buf
, lbp_mode
[SD_LBP_WS10
], 20))
409 sd_config_discard(sdkp
, SD_LBP_WS10
);
410 else if (!strncmp(buf
, lbp_mode
[SD_LBP_ZERO
], 20))
411 sd_config_discard(sdkp
, SD_LBP_ZERO
);
412 else if (!strncmp(buf
, lbp_mode
[SD_LBP_DISABLE
], 20))
413 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
419 static DEVICE_ATTR_RW(provisioning_mode
);
422 max_medium_access_timeouts_show(struct device
*dev
,
423 struct device_attribute
*attr
, char *buf
)
425 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
427 return snprintf(buf
, 20, "%u\n", sdkp
->max_medium_access_timeouts
);
431 max_medium_access_timeouts_store(struct device
*dev
,
432 struct device_attribute
*attr
, const char *buf
,
435 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
438 if (!capable(CAP_SYS_ADMIN
))
441 err
= kstrtouint(buf
, 10, &sdkp
->max_medium_access_timeouts
);
443 return err
? err
: count
;
445 static DEVICE_ATTR_RW(max_medium_access_timeouts
);
448 max_write_same_blocks_show(struct device
*dev
, struct device_attribute
*attr
,
451 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
453 return snprintf(buf
, 20, "%u\n", sdkp
->max_ws_blocks
);
457 max_write_same_blocks_store(struct device
*dev
, struct device_attribute
*attr
,
458 const char *buf
, size_t count
)
460 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
461 struct scsi_device
*sdp
= sdkp
->device
;
465 if (!capable(CAP_SYS_ADMIN
))
468 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_ZBC
)
471 err
= kstrtoul(buf
, 10, &max
);
477 sdp
->no_write_same
= 1;
478 else if (max
<= SD_MAX_WS16_BLOCKS
) {
479 sdp
->no_write_same
= 0;
480 sdkp
->max_ws_blocks
= max
;
483 sd_config_write_same(sdkp
);
487 static DEVICE_ATTR_RW(max_write_same_blocks
);
489 static struct attribute
*sd_disk_attrs
[] = {
490 &dev_attr_cache_type
.attr
,
492 &dev_attr_allow_restart
.attr
,
493 &dev_attr_manage_start_stop
.attr
,
494 &dev_attr_protection_type
.attr
,
495 &dev_attr_protection_mode
.attr
,
496 &dev_attr_app_tag_own
.attr
,
497 &dev_attr_thin_provisioning
.attr
,
498 &dev_attr_provisioning_mode
.attr
,
499 &dev_attr_max_write_same_blocks
.attr
,
500 &dev_attr_max_medium_access_timeouts
.attr
,
503 ATTRIBUTE_GROUPS(sd_disk
);
505 static struct class sd_disk_class
= {
507 .owner
= THIS_MODULE
,
508 .dev_release
= scsi_disk_release
,
509 .dev_groups
= sd_disk_groups
,
512 static const struct dev_pm_ops sd_pm_ops
= {
513 .suspend
= sd_suspend_system
,
515 .poweroff
= sd_suspend_system
,
516 .restore
= sd_resume
,
517 .runtime_suspend
= sd_suspend_runtime
,
518 .runtime_resume
= sd_resume
,
521 static struct scsi_driver sd_template
= {
524 .owner
= THIS_MODULE
,
527 .shutdown
= sd_shutdown
,
531 .init_command
= sd_init_command
,
532 .uninit_command
= sd_uninit_command
,
534 .eh_action
= sd_eh_action
,
538 * Dummy kobj_map->probe function.
539 * The default ->probe function will call modprobe, which is
540 * pointless as this module is already loaded.
542 static struct kobject
*sd_default_probe(dev_t devt
, int *partno
, void *data
)
548 * Device no to disk mapping:
550 * major disc2 disc p1
551 * |............|.............|....|....| <- dev_t
554 * Inside a major, we have 16k disks, however mapped non-
555 * contiguously. The first 16 disks are for major0, the next
556 * ones with major1, ... Disk 256 is for major0 again, disk 272
558 * As we stay compatible with our numbering scheme, we can reuse
559 * the well-know SCSI majors 8, 65--71, 136--143.
561 static int sd_major(int major_idx
)
565 return SCSI_DISK0_MAJOR
;
567 return SCSI_DISK1_MAJOR
+ major_idx
- 1;
569 return SCSI_DISK8_MAJOR
+ major_idx
- 8;
572 return 0; /* shut up gcc */
576 static struct scsi_disk
*scsi_disk_get(struct gendisk
*disk
)
578 struct scsi_disk
*sdkp
= NULL
;
580 mutex_lock(&sd_ref_mutex
);
582 if (disk
->private_data
) {
583 sdkp
= scsi_disk(disk
);
584 if (scsi_device_get(sdkp
->device
) == 0)
585 get_device(&sdkp
->dev
);
589 mutex_unlock(&sd_ref_mutex
);
593 static void scsi_disk_put(struct scsi_disk
*sdkp
)
595 struct scsi_device
*sdev
= sdkp
->device
;
597 mutex_lock(&sd_ref_mutex
);
598 put_device(&sdkp
->dev
);
599 scsi_device_put(sdev
);
600 mutex_unlock(&sd_ref_mutex
);
603 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd
*scmd
,
604 unsigned int dix
, unsigned int dif
)
606 struct bio
*bio
= scmd
->request
->bio
;
607 unsigned int prot_op
= sd_prot_op(rq_data_dir(scmd
->request
), dix
, dif
);
608 unsigned int protect
= 0;
610 if (dix
) { /* DIX Type 0, 1, 2, 3 */
611 if (bio_integrity_flagged(bio
, BIP_IP_CHECKSUM
))
612 scmd
->prot_flags
|= SCSI_PROT_IP_CHECKSUM
;
614 if (bio_integrity_flagged(bio
, BIP_CTRL_NOCHECK
) == false)
615 scmd
->prot_flags
|= SCSI_PROT_GUARD_CHECK
;
618 if (dif
!= T10_PI_TYPE3_PROTECTION
) { /* DIX/DIF Type 0, 1, 2 */
619 scmd
->prot_flags
|= SCSI_PROT_REF_INCREMENT
;
621 if (bio_integrity_flagged(bio
, BIP_CTRL_NOCHECK
) == false)
622 scmd
->prot_flags
|= SCSI_PROT_REF_CHECK
;
625 if (dif
) { /* DIX/DIF Type 1, 2, 3 */
626 scmd
->prot_flags
|= SCSI_PROT_TRANSFER_PI
;
628 if (bio_integrity_flagged(bio
, BIP_DISK_NOCHECK
))
629 protect
= 3 << 5; /* Disable target PI checking */
631 protect
= 1 << 5; /* Enable target PI checking */
634 scsi_set_prot_op(scmd
, prot_op
);
635 scsi_set_prot_type(scmd
, dif
);
636 scmd
->prot_flags
&= sd_prot_flag_mask(prot_op
);
641 static void sd_config_discard(struct scsi_disk
*sdkp
, unsigned int mode
)
643 struct request_queue
*q
= sdkp
->disk
->queue
;
644 unsigned int logical_block_size
= sdkp
->device
->sector_size
;
645 unsigned int max_blocks
= 0;
647 q
->limits
.discard_zeroes_data
= 0;
650 * When LBPRZ is reported, discard alignment and granularity
651 * must be fixed to the logical block size. Otherwise the block
652 * layer will drop misaligned portions of the request which can
653 * lead to data corruption. If LBPRZ is not set, we honor the
657 q
->limits
.discard_alignment
= 0;
658 q
->limits
.discard_granularity
= logical_block_size
;
660 q
->limits
.discard_alignment
= sdkp
->unmap_alignment
*
662 q
->limits
.discard_granularity
=
663 max(sdkp
->physical_block_size
,
664 sdkp
->unmap_granularity
* logical_block_size
);
667 sdkp
->provisioning_mode
= mode
;
672 blk_queue_max_discard_sectors(q
, 0);
673 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
677 max_blocks
= min_not_zero(sdkp
->max_unmap_blocks
,
678 (u32
)SD_MAX_WS16_BLOCKS
);
682 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
683 (u32
)SD_MAX_WS16_BLOCKS
);
684 q
->limits
.discard_zeroes_data
= sdkp
->lbprz
;
688 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
689 (u32
)SD_MAX_WS10_BLOCKS
);
690 q
->limits
.discard_zeroes_data
= sdkp
->lbprz
;
694 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
695 (u32
)SD_MAX_WS10_BLOCKS
);
696 q
->limits
.discard_zeroes_data
= 1;
700 blk_queue_max_discard_sectors(q
, max_blocks
* (logical_block_size
>> 9));
701 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
705 * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
706 * @sdp: scsi device to operate one
707 * @rq: Request to prepare
709 * Will issue either UNMAP or WRITE SAME(16) depending on preference
710 * indicated by target device.
712 static int sd_setup_discard_cmnd(struct scsi_cmnd
*cmd
)
714 struct request
*rq
= cmd
->request
;
715 struct scsi_device
*sdp
= cmd
->device
;
716 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
717 sector_t sector
= blk_rq_pos(rq
);
718 unsigned int nr_sectors
= blk_rq_sectors(rq
);
724 sector
>>= ilog2(sdp
->sector_size
) - 9;
725 nr_sectors
>>= ilog2(sdp
->sector_size
) - 9;
727 page
= alloc_page(GFP_ATOMIC
| __GFP_ZERO
);
729 return BLKPREP_DEFER
;
731 switch (sdkp
->provisioning_mode
) {
733 buf
= page_address(page
);
736 cmd
->cmnd
[0] = UNMAP
;
739 put_unaligned_be16(6 + 16, &buf
[0]);
740 put_unaligned_be16(16, &buf
[2]);
741 put_unaligned_be64(sector
, &buf
[8]);
742 put_unaligned_be32(nr_sectors
, &buf
[16]);
749 cmd
->cmnd
[0] = WRITE_SAME_16
;
750 cmd
->cmnd
[1] = 0x8; /* UNMAP */
751 put_unaligned_be64(sector
, &cmd
->cmnd
[2]);
752 put_unaligned_be32(nr_sectors
, &cmd
->cmnd
[10]);
754 len
= sdkp
->device
->sector_size
;
760 cmd
->cmnd
[0] = WRITE_SAME
;
761 if (sdkp
->provisioning_mode
== SD_LBP_WS10
)
762 cmd
->cmnd
[1] = 0x8; /* UNMAP */
763 put_unaligned_be32(sector
, &cmd
->cmnd
[2]);
764 put_unaligned_be16(nr_sectors
, &cmd
->cmnd
[7]);
766 len
= sdkp
->device
->sector_size
;
770 ret
= BLKPREP_INVALID
;
774 rq
->timeout
= SD_TIMEOUT
;
776 cmd
->transfersize
= len
;
777 cmd
->allowed
= SD_MAX_RETRIES
;
779 rq
->special_vec
.bv_page
= page
;
780 rq
->special_vec
.bv_offset
= 0;
781 rq
->special_vec
.bv_len
= len
;
783 rq
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
786 ret
= scsi_init_io(cmd
);
788 if (ret
!= BLKPREP_OK
)
793 static void sd_config_write_same(struct scsi_disk
*sdkp
)
795 struct request_queue
*q
= sdkp
->disk
->queue
;
796 unsigned int logical_block_size
= sdkp
->device
->sector_size
;
798 if (sdkp
->device
->no_write_same
) {
799 sdkp
->max_ws_blocks
= 0;
803 /* Some devices can not handle block counts above 0xffff despite
804 * supporting WRITE SAME(16). Consequently we default to 64k
805 * blocks per I/O unless the device explicitly advertises a
808 if (sdkp
->max_ws_blocks
> SD_MAX_WS10_BLOCKS
)
809 sdkp
->max_ws_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
810 (u32
)SD_MAX_WS16_BLOCKS
);
811 else if (sdkp
->ws16
|| sdkp
->ws10
|| sdkp
->device
->no_report_opcodes
)
812 sdkp
->max_ws_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
813 (u32
)SD_MAX_WS10_BLOCKS
);
815 sdkp
->device
->no_write_same
= 1;
816 sdkp
->max_ws_blocks
= 0;
820 blk_queue_max_write_same_sectors(q
, sdkp
->max_ws_blocks
*
821 (logical_block_size
>> 9));
825 * sd_setup_write_same_cmnd - write the same data to multiple blocks
826 * @cmd: command to prepare
828 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
829 * preference indicated by target device.
831 static int sd_setup_write_same_cmnd(struct scsi_cmnd
*cmd
)
833 struct request
*rq
= cmd
->request
;
834 struct scsi_device
*sdp
= cmd
->device
;
835 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
836 struct bio
*bio
= rq
->bio
;
837 sector_t sector
= blk_rq_pos(rq
);
838 unsigned int nr_sectors
= blk_rq_sectors(rq
);
839 unsigned int nr_bytes
= blk_rq_bytes(rq
);
842 if (sdkp
->device
->no_write_same
)
843 return BLKPREP_INVALID
;
845 BUG_ON(bio_offset(bio
) || bio_iovec(bio
).bv_len
!= sdp
->sector_size
);
847 if (sd_is_zoned(sdkp
)) {
848 ret
= sd_zbc_setup_write_cmnd(cmd
);
849 if (ret
!= BLKPREP_OK
)
853 sector
>>= ilog2(sdp
->sector_size
) - 9;
854 nr_sectors
>>= ilog2(sdp
->sector_size
) - 9;
856 rq
->timeout
= SD_WRITE_SAME_TIMEOUT
;
858 if (sdkp
->ws16
|| sector
> 0xffffffff || nr_sectors
> 0xffff) {
860 cmd
->cmnd
[0] = WRITE_SAME_16
;
861 put_unaligned_be64(sector
, &cmd
->cmnd
[2]);
862 put_unaligned_be32(nr_sectors
, &cmd
->cmnd
[10]);
865 cmd
->cmnd
[0] = WRITE_SAME
;
866 put_unaligned_be32(sector
, &cmd
->cmnd
[2]);
867 put_unaligned_be16(nr_sectors
, &cmd
->cmnd
[7]);
870 cmd
->transfersize
= sdp
->sector_size
;
871 cmd
->allowed
= SD_MAX_RETRIES
;
874 * For WRITE SAME the data transferred via the DATA OUT buffer is
875 * different from the amount of data actually written to the target.
877 * We set up __data_len to the amount of data transferred via the
878 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
879 * to transfer a single sector of data first, but then reset it to
880 * the amount of data to be written right after so that the I/O path
881 * knows how much to actually write.
883 rq
->__data_len
= sdp
->sector_size
;
884 ret
= scsi_init_io(cmd
);
885 rq
->__data_len
= nr_bytes
;
889 static int sd_setup_flush_cmnd(struct scsi_cmnd
*cmd
)
891 struct request
*rq
= cmd
->request
;
893 /* flush requests don't perform I/O, zero the S/G table */
894 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
896 cmd
->cmnd
[0] = SYNCHRONIZE_CACHE
;
898 cmd
->transfersize
= 0;
899 cmd
->allowed
= SD_MAX_RETRIES
;
901 rq
->timeout
= rq
->q
->rq_timeout
* SD_FLUSH_TIMEOUT_MULTIPLIER
;
905 static int sd_setup_read_write_cmnd(struct scsi_cmnd
*SCpnt
)
907 struct request
*rq
= SCpnt
->request
;
908 struct scsi_device
*sdp
= SCpnt
->device
;
909 struct gendisk
*disk
= rq
->rq_disk
;
910 struct scsi_disk
*sdkp
= scsi_disk(disk
);
911 sector_t block
= blk_rq_pos(rq
);
913 unsigned int this_count
= blk_rq_sectors(rq
);
914 unsigned int dif
, dix
;
915 bool zoned_write
= sd_is_zoned(sdkp
) && rq_data_dir(rq
) == WRITE
;
917 unsigned char protect
;
920 ret
= sd_zbc_setup_write_cmnd(SCpnt
);
921 if (ret
!= BLKPREP_OK
)
925 ret
= scsi_init_io(SCpnt
);
926 if (ret
!= BLKPREP_OK
)
930 /* from here on until we're complete, any goto out
931 * is used for a killable error condition */
935 scmd_printk(KERN_INFO
, SCpnt
,
936 "%s: block=%llu, count=%d\n",
937 __func__
, (unsigned long long)block
, this_count
));
939 if (!sdp
|| !scsi_device_online(sdp
) ||
940 block
+ blk_rq_sectors(rq
) > get_capacity(disk
)) {
941 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
,
942 "Finishing %u sectors\n",
943 blk_rq_sectors(rq
)));
944 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
,
945 "Retry with 0x%p\n", SCpnt
));
951 * quietly refuse to do anything to a changed disc until
952 * the changed bit has been reset
954 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
959 * Some SD card readers can't handle multi-sector accesses which touch
960 * the last one or two hardware sectors. Split accesses as needed.
962 threshold
= get_capacity(disk
) - SD_LAST_BUGGY_SECTORS
*
963 (sdp
->sector_size
/ 512);
965 if (unlikely(sdp
->last_sector_bug
&& block
+ this_count
> threshold
)) {
966 if (block
< threshold
) {
967 /* Access up to the threshold but not beyond */
968 this_count
= threshold
- block
;
970 /* Access only a single hardware sector */
971 this_count
= sdp
->sector_size
/ 512;
975 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
, "block=%llu\n",
976 (unsigned long long)block
));
979 * If we have a 1K hardware sectorsize, prevent access to single
980 * 512 byte sectors. In theory we could handle this - in fact
981 * the scsi cdrom driver must be able to handle this because
982 * we typically use 1K blocksizes, and cdroms typically have
983 * 2K hardware sectorsizes. Of course, things are simpler
984 * with the cdrom, since it is read-only. For performance
985 * reasons, the filesystems should be able to handle this
986 * and not force the scsi disk driver to use bounce buffers
989 if (sdp
->sector_size
== 1024) {
990 if ((block
& 1) || (blk_rq_sectors(rq
) & 1)) {
991 scmd_printk(KERN_ERR
, SCpnt
,
992 "Bad block number requested\n");
996 this_count
= this_count
>> 1;
999 if (sdp
->sector_size
== 2048) {
1000 if ((block
& 3) || (blk_rq_sectors(rq
) & 3)) {
1001 scmd_printk(KERN_ERR
, SCpnt
,
1002 "Bad block number requested\n");
1006 this_count
= this_count
>> 2;
1009 if (sdp
->sector_size
== 4096) {
1010 if ((block
& 7) || (blk_rq_sectors(rq
) & 7)) {
1011 scmd_printk(KERN_ERR
, SCpnt
,
1012 "Bad block number requested\n");
1016 this_count
= this_count
>> 3;
1019 if (rq_data_dir(rq
) == WRITE
) {
1020 SCpnt
->cmnd
[0] = WRITE_6
;
1022 if (blk_integrity_rq(rq
))
1023 sd_dif_prepare(SCpnt
);
1025 } else if (rq_data_dir(rq
) == READ
) {
1026 SCpnt
->cmnd
[0] = READ_6
;
1028 scmd_printk(KERN_ERR
, SCpnt
, "Unknown command %d\n", req_op(rq
));
1032 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
,
1033 "%s %d/%u 512 byte blocks.\n",
1034 (rq_data_dir(rq
) == WRITE
) ?
1035 "writing" : "reading", this_count
,
1036 blk_rq_sectors(rq
)));
1038 dix
= scsi_prot_sg_count(SCpnt
);
1039 dif
= scsi_host_dif_capable(SCpnt
->device
->host
, sdkp
->protection_type
);
1042 protect
= sd_setup_protect_cmnd(SCpnt
, dix
, dif
);
1046 if (protect
&& sdkp
->protection_type
== T10_PI_TYPE2_PROTECTION
) {
1047 SCpnt
->cmnd
= mempool_alloc(sd_cdb_pool
, GFP_ATOMIC
);
1049 if (unlikely(SCpnt
->cmnd
== NULL
)) {
1050 ret
= BLKPREP_DEFER
;
1054 SCpnt
->cmd_len
= SD_EXT_CDB_SIZE
;
1055 memset(SCpnt
->cmnd
, 0, SCpnt
->cmd_len
);
1056 SCpnt
->cmnd
[0] = VARIABLE_LENGTH_CMD
;
1057 SCpnt
->cmnd
[7] = 0x18;
1058 SCpnt
->cmnd
[9] = (rq_data_dir(rq
) == READ
) ? READ_32
: WRITE_32
;
1059 SCpnt
->cmnd
[10] = protect
| ((rq
->cmd_flags
& REQ_FUA
) ? 0x8 : 0);
1062 SCpnt
->cmnd
[12] = sizeof(block
) > 4 ? (unsigned char) (block
>> 56) & 0xff : 0;
1063 SCpnt
->cmnd
[13] = sizeof(block
) > 4 ? (unsigned char) (block
>> 48) & 0xff : 0;
1064 SCpnt
->cmnd
[14] = sizeof(block
) > 4 ? (unsigned char) (block
>> 40) & 0xff : 0;
1065 SCpnt
->cmnd
[15] = sizeof(block
) > 4 ? (unsigned char) (block
>> 32) & 0xff : 0;
1066 SCpnt
->cmnd
[16] = (unsigned char) (block
>> 24) & 0xff;
1067 SCpnt
->cmnd
[17] = (unsigned char) (block
>> 16) & 0xff;
1068 SCpnt
->cmnd
[18] = (unsigned char) (block
>> 8) & 0xff;
1069 SCpnt
->cmnd
[19] = (unsigned char) block
& 0xff;
1071 /* Expected Indirect LBA */
1072 SCpnt
->cmnd
[20] = (unsigned char) (block
>> 24) & 0xff;
1073 SCpnt
->cmnd
[21] = (unsigned char) (block
>> 16) & 0xff;
1074 SCpnt
->cmnd
[22] = (unsigned char) (block
>> 8) & 0xff;
1075 SCpnt
->cmnd
[23] = (unsigned char) block
& 0xff;
1077 /* Transfer length */
1078 SCpnt
->cmnd
[28] = (unsigned char) (this_count
>> 24) & 0xff;
1079 SCpnt
->cmnd
[29] = (unsigned char) (this_count
>> 16) & 0xff;
1080 SCpnt
->cmnd
[30] = (unsigned char) (this_count
>> 8) & 0xff;
1081 SCpnt
->cmnd
[31] = (unsigned char) this_count
& 0xff;
1082 } else if (sdp
->use_16_for_rw
|| (this_count
> 0xffff)) {
1083 SCpnt
->cmnd
[0] += READ_16
- READ_6
;
1084 SCpnt
->cmnd
[1] = protect
| ((rq
->cmd_flags
& REQ_FUA
) ? 0x8 : 0);
1085 SCpnt
->cmnd
[2] = sizeof(block
) > 4 ? (unsigned char) (block
>> 56) & 0xff : 0;
1086 SCpnt
->cmnd
[3] = sizeof(block
) > 4 ? (unsigned char) (block
>> 48) & 0xff : 0;
1087 SCpnt
->cmnd
[4] = sizeof(block
) > 4 ? (unsigned char) (block
>> 40) & 0xff : 0;
1088 SCpnt
->cmnd
[5] = sizeof(block
) > 4 ? (unsigned char) (block
>> 32) & 0xff : 0;
1089 SCpnt
->cmnd
[6] = (unsigned char) (block
>> 24) & 0xff;
1090 SCpnt
->cmnd
[7] = (unsigned char) (block
>> 16) & 0xff;
1091 SCpnt
->cmnd
[8] = (unsigned char) (block
>> 8) & 0xff;
1092 SCpnt
->cmnd
[9] = (unsigned char) block
& 0xff;
1093 SCpnt
->cmnd
[10] = (unsigned char) (this_count
>> 24) & 0xff;
1094 SCpnt
->cmnd
[11] = (unsigned char) (this_count
>> 16) & 0xff;
1095 SCpnt
->cmnd
[12] = (unsigned char) (this_count
>> 8) & 0xff;
1096 SCpnt
->cmnd
[13] = (unsigned char) this_count
& 0xff;
1097 SCpnt
->cmnd
[14] = SCpnt
->cmnd
[15] = 0;
1098 } else if ((this_count
> 0xff) || (block
> 0x1fffff) ||
1099 scsi_device_protection(SCpnt
->device
) ||
1100 SCpnt
->device
->use_10_for_rw
) {
1101 SCpnt
->cmnd
[0] += READ_10
- READ_6
;
1102 SCpnt
->cmnd
[1] = protect
| ((rq
->cmd_flags
& REQ_FUA
) ? 0x8 : 0);
1103 SCpnt
->cmnd
[2] = (unsigned char) (block
>> 24) & 0xff;
1104 SCpnt
->cmnd
[3] = (unsigned char) (block
>> 16) & 0xff;
1105 SCpnt
->cmnd
[4] = (unsigned char) (block
>> 8) & 0xff;
1106 SCpnt
->cmnd
[5] = (unsigned char) block
& 0xff;
1107 SCpnt
->cmnd
[6] = SCpnt
->cmnd
[9] = 0;
1108 SCpnt
->cmnd
[7] = (unsigned char) (this_count
>> 8) & 0xff;
1109 SCpnt
->cmnd
[8] = (unsigned char) this_count
& 0xff;
1111 if (unlikely(rq
->cmd_flags
& REQ_FUA
)) {
1113 * This happens only if this drive failed
1114 * 10byte rw command with ILLEGAL_REQUEST
1115 * during operation and thus turned off
1118 scmd_printk(KERN_ERR
, SCpnt
,
1119 "FUA write on READ/WRITE(6) drive\n");
1123 SCpnt
->cmnd
[1] |= (unsigned char) ((block
>> 16) & 0x1f);
1124 SCpnt
->cmnd
[2] = (unsigned char) ((block
>> 8) & 0xff);
1125 SCpnt
->cmnd
[3] = (unsigned char) block
& 0xff;
1126 SCpnt
->cmnd
[4] = (unsigned char) this_count
;
1129 SCpnt
->sdb
.length
= this_count
* sdp
->sector_size
;
1132 * We shouldn't disconnect in the middle of a sector, so with a dumb
1133 * host adapter, it's safe to assume that we can at least transfer
1134 * this many bytes between each connect / disconnect.
1136 SCpnt
->transfersize
= sdp
->sector_size
;
1137 SCpnt
->underflow
= this_count
<< 9;
1138 SCpnt
->allowed
= SD_MAX_RETRIES
;
1141 * This indicates that the command is ready from our end to be
1146 if (zoned_write
&& ret
!= BLKPREP_OK
)
1147 sd_zbc_cancel_write_cmnd(SCpnt
);
1152 static int sd_init_command(struct scsi_cmnd
*cmd
)
1154 struct request
*rq
= cmd
->request
;
1156 switch (req_op(rq
)) {
1157 case REQ_OP_DISCARD
:
1158 return sd_setup_discard_cmnd(cmd
);
1159 case REQ_OP_WRITE_SAME
:
1160 return sd_setup_write_same_cmnd(cmd
);
1162 return sd_setup_flush_cmnd(cmd
);
1165 return sd_setup_read_write_cmnd(cmd
);
1166 case REQ_OP_ZONE_REPORT
:
1167 return sd_zbc_setup_report_cmnd(cmd
);
1168 case REQ_OP_ZONE_RESET
:
1169 return sd_zbc_setup_reset_cmnd(cmd
);
1175 static void sd_uninit_command(struct scsi_cmnd
*SCpnt
)
1177 struct request
*rq
= SCpnt
->request
;
1179 if (rq
->rq_flags
& RQF_SPECIAL_PAYLOAD
)
1180 __free_page(rq
->special_vec
.bv_page
);
1182 if (SCpnt
->cmnd
!= rq
->cmd
) {
1183 mempool_free(SCpnt
->cmnd
, sd_cdb_pool
);
1190 * sd_open - open a scsi disk device
1191 * @inode: only i_rdev member may be used
1192 * @filp: only f_mode and f_flags may be used
1194 * Returns 0 if successful. Returns a negated errno value in case
1197 * Note: This can be called from a user context (e.g. fsck(1) )
1198 * or from within the kernel (e.g. as a result of a mount(1) ).
1199 * In the latter case @inode and @filp carry an abridged amount
1200 * of information as noted above.
1202 * Locking: called with bdev->bd_mutex held.
1204 static int sd_open(struct block_device
*bdev
, fmode_t mode
)
1206 struct scsi_disk
*sdkp
= scsi_disk_get(bdev
->bd_disk
);
1207 struct scsi_device
*sdev
;
1213 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_open\n"));
1215 sdev
= sdkp
->device
;
1218 * If the device is in error recovery, wait until it is done.
1219 * If the device is offline, then disallow any access to it.
1222 if (!scsi_block_when_processing_errors(sdev
))
1225 if (sdev
->removable
|| sdkp
->write_prot
)
1226 check_disk_change(bdev
);
1229 * If the drive is empty, just let the open fail.
1231 retval
= -ENOMEDIUM
;
1232 if (sdev
->removable
&& !sdkp
->media_present
&& !(mode
& FMODE_NDELAY
))
1236 * If the device has the write protect tab set, have the open fail
1237 * if the user expects to be able to write to the thing.
1240 if (sdkp
->write_prot
&& (mode
& FMODE_WRITE
))
1244 * It is possible that the disk changing stuff resulted in
1245 * the device being taken offline. If this is the case,
1246 * report this to the user, and don't pretend that the
1247 * open actually succeeded.
1250 if (!scsi_device_online(sdev
))
1253 if ((atomic_inc_return(&sdkp
->openers
) == 1) && sdev
->removable
) {
1254 if (scsi_block_when_processing_errors(sdev
))
1255 scsi_set_medium_removal(sdev
, SCSI_REMOVAL_PREVENT
);
1261 scsi_disk_put(sdkp
);
1266 * sd_release - invoked when the (last) close(2) is called on this
1268 * @inode: only i_rdev member may be used
1269 * @filp: only f_mode and f_flags may be used
1273 * Note: may block (uninterruptible) if error recovery is underway
1276 * Locking: called with bdev->bd_mutex held.
1278 static void sd_release(struct gendisk
*disk
, fmode_t mode
)
1280 struct scsi_disk
*sdkp
= scsi_disk(disk
);
1281 struct scsi_device
*sdev
= sdkp
->device
;
1283 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_release\n"));
1285 if (atomic_dec_return(&sdkp
->openers
) == 0 && sdev
->removable
) {
1286 if (scsi_block_when_processing_errors(sdev
))
1287 scsi_set_medium_removal(sdev
, SCSI_REMOVAL_ALLOW
);
1291 * XXX and what if there are packets in flight and this close()
1292 * XXX is followed by a "rmmod sd_mod"?
1295 scsi_disk_put(sdkp
);
1298 static int sd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1300 struct scsi_disk
*sdkp
= scsi_disk(bdev
->bd_disk
);
1301 struct scsi_device
*sdp
= sdkp
->device
;
1302 struct Scsi_Host
*host
= sdp
->host
;
1303 sector_t capacity
= logical_to_sectors(sdp
, sdkp
->capacity
);
1306 /* default to most commonly used values */
1307 diskinfo
[0] = 0x40; /* 1 << 6 */
1308 diskinfo
[1] = 0x20; /* 1 << 5 */
1309 diskinfo
[2] = capacity
>> 11;
1311 /* override with calculated, extended default, or driver values */
1312 if (host
->hostt
->bios_param
)
1313 host
->hostt
->bios_param(sdp
, bdev
, capacity
, diskinfo
);
1315 scsicam_bios_param(bdev
, capacity
, diskinfo
);
1317 geo
->heads
= diskinfo
[0];
1318 geo
->sectors
= diskinfo
[1];
1319 geo
->cylinders
= diskinfo
[2];
1324 * sd_ioctl - process an ioctl
1325 * @inode: only i_rdev/i_bdev members may be used
1326 * @filp: only f_mode and f_flags may be used
1327 * @cmd: ioctl command number
1328 * @arg: this is third argument given to ioctl(2) system call.
1329 * Often contains a pointer.
1331 * Returns 0 if successful (some ioctls return positive numbers on
1332 * success as well). Returns a negated errno value in case of error.
1334 * Note: most ioctls are forward onto the block subsystem or further
1335 * down in the scsi subsystem.
1337 static int sd_ioctl(struct block_device
*bdev
, fmode_t mode
,
1338 unsigned int cmd
, unsigned long arg
)
1340 struct gendisk
*disk
= bdev
->bd_disk
;
1341 struct scsi_disk
*sdkp
= scsi_disk(disk
);
1342 struct scsi_device
*sdp
= sdkp
->device
;
1343 void __user
*p
= (void __user
*)arg
;
1346 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO
, sdkp
, "sd_ioctl: disk=%s, "
1347 "cmd=0x%x\n", disk
->disk_name
, cmd
));
1349 error
= scsi_verify_blk_ioctl(bdev
, cmd
);
1354 * If we are in the middle of error recovery, don't let anyone
1355 * else try and use this device. Also, if error recovery fails, it
1356 * may try and take the device offline, in which case all further
1357 * access to the device is prohibited.
1359 error
= scsi_ioctl_block_when_processing_errors(sdp
, cmd
,
1360 (mode
& FMODE_NDELAY
) != 0);
1365 * Send SCSI addressing ioctls directly to mid level, send other
1366 * ioctls to block level and then onto mid level if they can't be
1370 case SCSI_IOCTL_GET_IDLUN
:
1371 case SCSI_IOCTL_GET_BUS_NUMBER
:
1372 error
= scsi_ioctl(sdp
, cmd
, p
);
1375 error
= scsi_cmd_blk_ioctl(bdev
, mode
, cmd
, p
);
1376 if (error
!= -ENOTTY
)
1378 error
= scsi_ioctl(sdp
, cmd
, p
);
1385 static void set_media_not_present(struct scsi_disk
*sdkp
)
1387 if (sdkp
->media_present
)
1388 sdkp
->device
->changed
= 1;
1390 if (sdkp
->device
->removable
) {
1391 sdkp
->media_present
= 0;
1396 static int media_not_present(struct scsi_disk
*sdkp
,
1397 struct scsi_sense_hdr
*sshdr
)
1399 if (!scsi_sense_valid(sshdr
))
1402 /* not invoked for commands that could return deferred errors */
1403 switch (sshdr
->sense_key
) {
1404 case UNIT_ATTENTION
:
1406 /* medium not present */
1407 if (sshdr
->asc
== 0x3A) {
1408 set_media_not_present(sdkp
);
1416 * sd_check_events - check media events
1417 * @disk: kernel device descriptor
1418 * @clearing: disk events currently being cleared
1420 * Returns mask of DISK_EVENT_*.
1422 * Note: this function is invoked from the block subsystem.
1424 static unsigned int sd_check_events(struct gendisk
*disk
, unsigned int clearing
)
1426 struct scsi_disk
*sdkp
= scsi_disk_get(disk
);
1427 struct scsi_device
*sdp
;
1428 struct scsi_sense_hdr
*sshdr
= NULL
;
1435 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_check_events\n"));
1438 * If the device is offline, don't send any commands - just pretend as
1439 * if the command failed. If the device ever comes back online, we
1440 * can deal with it then. It is only because of unrecoverable errors
1441 * that we would ever take a device offline in the first place.
1443 if (!scsi_device_online(sdp
)) {
1444 set_media_not_present(sdkp
);
1449 * Using TEST_UNIT_READY enables differentiation between drive with
1450 * no cartridge loaded - NOT READY, drive with changed cartridge -
1451 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1453 * Drives that auto spin down. eg iomega jaz 1G, will be started
1454 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1455 * sd_revalidate() is called.
1459 if (scsi_block_when_processing_errors(sdp
)) {
1460 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1461 retval
= scsi_test_unit_ready(sdp
, SD_TIMEOUT
, SD_MAX_RETRIES
,
1465 /* failed to execute TUR, assume media not present */
1466 if (host_byte(retval
)) {
1467 set_media_not_present(sdkp
);
1471 if (media_not_present(sdkp
, sshdr
))
1475 * For removable scsi disk we have to recognise the presence
1476 * of a disk in the drive.
1478 if (!sdkp
->media_present
)
1480 sdkp
->media_present
= 1;
1483 * sdp->changed is set under the following conditions:
1485 * Medium present state has changed in either direction.
1486 * Device has indicated UNIT_ATTENTION.
1489 retval
= sdp
->changed
? DISK_EVENT_MEDIA_CHANGE
: 0;
1491 scsi_disk_put(sdkp
);
1495 static int sd_sync_cache(struct scsi_disk
*sdkp
)
1498 struct scsi_device
*sdp
= sdkp
->device
;
1499 const int timeout
= sdp
->request_queue
->rq_timeout
1500 * SD_FLUSH_TIMEOUT_MULTIPLIER
;
1501 struct scsi_sense_hdr sshdr
;
1503 if (!scsi_device_online(sdp
))
1506 for (retries
= 3; retries
> 0; --retries
) {
1507 unsigned char cmd
[10] = { 0 };
1509 cmd
[0] = SYNCHRONIZE_CACHE
;
1511 * Leave the rest of the command zero to indicate
1514 res
= scsi_execute_req_flags(sdp
, cmd
, DMA_NONE
, NULL
, 0,
1515 &sshdr
, timeout
, SD_MAX_RETRIES
,
1522 sd_print_result(sdkp
, "Synchronize Cache(10) failed", res
);
1524 if (driver_byte(res
) & DRIVER_SENSE
)
1525 sd_print_sense_hdr(sdkp
, &sshdr
);
1526 /* we need to evaluate the error return */
1527 if (scsi_sense_valid(&sshdr
) &&
1528 (sshdr
.asc
== 0x3a || /* medium not present */
1529 sshdr
.asc
== 0x20)) /* invalid command */
1530 /* this is no error here */
1533 switch (host_byte(res
)) {
1534 /* ignore errors due to racing a disconnection */
1535 case DID_BAD_TARGET
:
1536 case DID_NO_CONNECT
:
1538 /* signal the upper layer it might try again */
1542 case DID_SOFT_ERROR
:
1551 static void sd_rescan(struct device
*dev
)
1553 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
1555 revalidate_disk(sdkp
->disk
);
1559 #ifdef CONFIG_COMPAT
1561 * This gets directly called from VFS. When the ioctl
1562 * is not recognized we go back to the other translation paths.
1564 static int sd_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1565 unsigned int cmd
, unsigned long arg
)
1567 struct scsi_device
*sdev
= scsi_disk(bdev
->bd_disk
)->device
;
1570 error
= scsi_ioctl_block_when_processing_errors(sdev
, cmd
,
1571 (mode
& FMODE_NDELAY
) != 0);
1576 * Let the static ioctl translation table take care of it.
1578 if (!sdev
->host
->hostt
->compat_ioctl
)
1579 return -ENOIOCTLCMD
;
1580 return sdev
->host
->hostt
->compat_ioctl(sdev
, cmd
, (void __user
*)arg
);
1584 static char sd_pr_type(enum pr_type type
)
1587 case PR_WRITE_EXCLUSIVE
:
1589 case PR_EXCLUSIVE_ACCESS
:
1591 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
1593 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
1595 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
1597 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
1604 static int sd_pr_command(struct block_device
*bdev
, u8 sa
,
1605 u64 key
, u64 sa_key
, u8 type
, u8 flags
)
1607 struct scsi_device
*sdev
= scsi_disk(bdev
->bd_disk
)->device
;
1608 struct scsi_sense_hdr sshdr
;
1610 u8 cmd
[16] = { 0, };
1611 u8 data
[24] = { 0, };
1613 cmd
[0] = PERSISTENT_RESERVE_OUT
;
1616 put_unaligned_be32(sizeof(data
), &cmd
[5]);
1618 put_unaligned_be64(key
, &data
[0]);
1619 put_unaligned_be64(sa_key
, &data
[8]);
1622 result
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, &data
, sizeof(data
),
1623 &sshdr
, SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
);
1625 if ((driver_byte(result
) & DRIVER_SENSE
) &&
1626 (scsi_sense_valid(&sshdr
))) {
1627 sdev_printk(KERN_INFO
, sdev
, "PR command failed: %d\n", result
);
1628 scsi_print_sense_hdr(sdev
, NULL
, &sshdr
);
1634 static int sd_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
1637 if (flags
& ~PR_FL_IGNORE_KEY
)
1639 return sd_pr_command(bdev
, (flags
& PR_FL_IGNORE_KEY
) ? 0x06 : 0x00,
1640 old_key
, new_key
, 0,
1641 (1 << 0) /* APTPL */);
1644 static int sd_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
1649 return sd_pr_command(bdev
, 0x01, key
, 0, sd_pr_type(type
), 0);
1652 static int sd_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
1654 return sd_pr_command(bdev
, 0x02, key
, 0, sd_pr_type(type
), 0);
1657 static int sd_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
1658 enum pr_type type
, bool abort
)
1660 return sd_pr_command(bdev
, abort
? 0x05 : 0x04, old_key
, new_key
,
1661 sd_pr_type(type
), 0);
1664 static int sd_pr_clear(struct block_device
*bdev
, u64 key
)
1666 return sd_pr_command(bdev
, 0x03, key
, 0, 0, 0);
1669 static const struct pr_ops sd_pr_ops
= {
1670 .pr_register
= sd_pr_register
,
1671 .pr_reserve
= sd_pr_reserve
,
1672 .pr_release
= sd_pr_release
,
1673 .pr_preempt
= sd_pr_preempt
,
1674 .pr_clear
= sd_pr_clear
,
1677 static const struct block_device_operations sd_fops
= {
1678 .owner
= THIS_MODULE
,
1680 .release
= sd_release
,
1682 .getgeo
= sd_getgeo
,
1683 #ifdef CONFIG_COMPAT
1684 .compat_ioctl
= sd_compat_ioctl
,
1686 .check_events
= sd_check_events
,
1687 .revalidate_disk
= sd_revalidate_disk
,
1688 .unlock_native_capacity
= sd_unlock_native_capacity
,
1689 .pr_ops
= &sd_pr_ops
,
1693 * sd_eh_action - error handling callback
1694 * @scmd: sd-issued command that has failed
1695 * @eh_disp: The recovery disposition suggested by the midlayer
1697 * This function is called by the SCSI midlayer upon completion of an
1698 * error test command (currently TEST UNIT READY). The result of sending
1699 * the eh command is passed in eh_disp. We're looking for devices that
1700 * fail medium access commands but are OK with non access commands like
1701 * test unit ready (so wrongly see the device as having a successful
1704 static int sd_eh_action(struct scsi_cmnd
*scmd
, int eh_disp
)
1706 struct scsi_disk
*sdkp
= scsi_disk(scmd
->request
->rq_disk
);
1708 if (!scsi_device_online(scmd
->device
) ||
1709 !scsi_medium_access_command(scmd
) ||
1710 host_byte(scmd
->result
) != DID_TIME_OUT
||
1715 * The device has timed out executing a medium access command.
1716 * However, the TEST UNIT READY command sent during error
1717 * handling completed successfully. Either the device is in the
1718 * process of recovering or has it suffered an internal failure
1719 * that prevents access to the storage medium.
1721 sdkp
->medium_access_timed_out
++;
1724 * If the device keeps failing read/write commands but TEST UNIT
1725 * READY always completes successfully we assume that medium
1726 * access is no longer possible and take the device offline.
1728 if (sdkp
->medium_access_timed_out
>= sdkp
->max_medium_access_timeouts
) {
1729 scmd_printk(KERN_ERR
, scmd
,
1730 "Medium access timeout failure. Offlining disk!\n");
1731 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1739 static unsigned int sd_completed_bytes(struct scsi_cmnd
*scmd
)
1741 u64 start_lba
= blk_rq_pos(scmd
->request
);
1742 u64 end_lba
= blk_rq_pos(scmd
->request
) + (scsi_bufflen(scmd
) / 512);
1743 u64 factor
= scmd
->device
->sector_size
/ 512;
1747 * resid is optional but mostly filled in. When it's unused,
1748 * its value is zero, so we assume the whole buffer transferred
1750 unsigned int transferred
= scsi_bufflen(scmd
) - scsi_get_resid(scmd
);
1751 unsigned int good_bytes
;
1753 if (scmd
->request
->cmd_type
!= REQ_TYPE_FS
)
1756 info_valid
= scsi_get_sense_info_fld(scmd
->sense_buffer
,
1757 SCSI_SENSE_BUFFERSIZE
,
1762 if (scsi_bufflen(scmd
) <= scmd
->device
->sector_size
)
1765 /* be careful ... don't want any overflows */
1766 do_div(start_lba
, factor
);
1767 do_div(end_lba
, factor
);
1769 /* The bad lba was reported incorrectly, we have no idea where
1772 if (bad_lba
< start_lba
|| bad_lba
>= end_lba
)
1775 /* This computation should always be done in terms of
1776 * the resolution of the device's medium.
1778 good_bytes
= (bad_lba
- start_lba
) * scmd
->device
->sector_size
;
1779 return min(good_bytes
, transferred
);
1783 * sd_done - bottom half handler: called when the lower level
1784 * driver has completed (successfully or otherwise) a scsi command.
1785 * @SCpnt: mid-level's per command structure.
1787 * Note: potentially run from within an ISR. Must not block.
1789 static int sd_done(struct scsi_cmnd
*SCpnt
)
1791 int result
= SCpnt
->result
;
1792 unsigned int good_bytes
= result
? 0 : scsi_bufflen(SCpnt
);
1793 unsigned int sector_size
= SCpnt
->device
->sector_size
;
1795 struct scsi_sense_hdr sshdr
;
1796 struct scsi_disk
*sdkp
= scsi_disk(SCpnt
->request
->rq_disk
);
1797 struct request
*req
= SCpnt
->request
;
1798 int sense_valid
= 0;
1799 int sense_deferred
= 0;
1800 unsigned char op
= SCpnt
->cmnd
[0];
1801 unsigned char unmap
= SCpnt
->cmnd
[1] & 8;
1803 switch (req_op(req
)) {
1804 case REQ_OP_DISCARD
:
1805 case REQ_OP_WRITE_SAME
:
1806 case REQ_OP_ZONE_RESET
:
1808 good_bytes
= blk_rq_bytes(req
);
1809 scsi_set_resid(SCpnt
, 0);
1812 scsi_set_resid(SCpnt
, blk_rq_bytes(req
));
1815 case REQ_OP_ZONE_REPORT
:
1817 good_bytes
= scsi_bufflen(SCpnt
)
1818 - scsi_get_resid(SCpnt
);
1819 scsi_set_resid(SCpnt
, 0);
1822 scsi_set_resid(SCpnt
, blk_rq_bytes(req
));
1827 * In case of bogus fw or device, we could end up having
1828 * an unaligned partial completion. Check this here and force
1831 resid
= scsi_get_resid(SCpnt
);
1832 if (resid
& (sector_size
- 1)) {
1833 sd_printk(KERN_INFO
, sdkp
,
1834 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1835 resid
, sector_size
);
1836 resid
= min(scsi_bufflen(SCpnt
),
1837 round_up(resid
, sector_size
));
1838 scsi_set_resid(SCpnt
, resid
);
1843 sense_valid
= scsi_command_normalize_sense(SCpnt
, &sshdr
);
1845 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
1847 sdkp
->medium_access_timed_out
= 0;
1849 if (driver_byte(result
) != DRIVER_SENSE
&&
1850 (!sense_valid
|| sense_deferred
))
1853 switch (sshdr
.sense_key
) {
1854 case HARDWARE_ERROR
:
1856 good_bytes
= sd_completed_bytes(SCpnt
);
1858 case RECOVERED_ERROR
:
1859 good_bytes
= scsi_bufflen(SCpnt
);
1862 /* This indicates a false check condition, so ignore it. An
1863 * unknown amount of data was transferred so treat it as an
1867 memset(SCpnt
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
1869 case ABORTED_COMMAND
:
1870 if (sshdr
.asc
== 0x10) /* DIF: Target detected corruption */
1871 good_bytes
= sd_completed_bytes(SCpnt
);
1873 case ILLEGAL_REQUEST
:
1874 if (sshdr
.asc
== 0x10) /* DIX: Host detected corruption */
1875 good_bytes
= sd_completed_bytes(SCpnt
);
1876 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1877 if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
1880 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
1885 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
1887 sdkp
->device
->no_write_same
= 1;
1888 sd_config_write_same(sdkp
);
1891 req
->__data_len
= blk_rq_bytes(req
);
1892 req
->rq_flags
|= RQF_QUIET
;
1902 if (sd_is_zoned(sdkp
))
1903 sd_zbc_complete(SCpnt
, good_bytes
, &sshdr
);
1905 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, SCpnt
,
1906 "sd_done: completed %d of %d bytes\n",
1907 good_bytes
, scsi_bufflen(SCpnt
)));
1909 if (rq_data_dir(SCpnt
->request
) == READ
&& scsi_prot_sg_count(SCpnt
))
1910 sd_dif_complete(SCpnt
, good_bytes
);
1916 * spinup disk - called only in sd_revalidate_disk()
1919 sd_spinup_disk(struct scsi_disk
*sdkp
)
1921 unsigned char cmd
[10];
1922 unsigned long spintime_expire
= 0;
1923 int retries
, spintime
;
1924 unsigned int the_result
;
1925 struct scsi_sense_hdr sshdr
;
1926 int sense_valid
= 0;
1930 /* Spin up drives, as required. Only do this at boot time */
1931 /* Spinup needs to be done for module loads too. */
1936 cmd
[0] = TEST_UNIT_READY
;
1937 memset((void *) &cmd
[1], 0, 9);
1939 the_result
= scsi_execute_req(sdkp
->device
, cmd
,
1942 SD_MAX_RETRIES
, NULL
);
1945 * If the drive has indicated to us that it
1946 * doesn't have any media in it, don't bother
1947 * with any more polling.
1949 if (media_not_present(sdkp
, &sshdr
))
1953 sense_valid
= scsi_sense_valid(&sshdr
);
1955 } while (retries
< 3 &&
1956 (!scsi_status_is_good(the_result
) ||
1957 ((driver_byte(the_result
) & DRIVER_SENSE
) &&
1958 sense_valid
&& sshdr
.sense_key
== UNIT_ATTENTION
)));
1960 if ((driver_byte(the_result
) & DRIVER_SENSE
) == 0) {
1961 /* no sense, TUR either succeeded or failed
1962 * with a status error */
1963 if(!spintime
&& !scsi_status_is_good(the_result
)) {
1964 sd_print_result(sdkp
, "Test Unit Ready failed",
1971 * The device does not want the automatic start to be issued.
1973 if (sdkp
->device
->no_start_on_add
)
1976 if (sense_valid
&& sshdr
.sense_key
== NOT_READY
) {
1977 if (sshdr
.asc
== 4 && sshdr
.ascq
== 3)
1978 break; /* manual intervention required */
1979 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0xb)
1980 break; /* standby */
1981 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0xc)
1982 break; /* unavailable */
1984 * Issue command to spin up drive when not ready
1987 sd_printk(KERN_NOTICE
, sdkp
, "Spinning up disk...");
1988 cmd
[0] = START_STOP
;
1989 cmd
[1] = 1; /* Return immediately */
1990 memset((void *) &cmd
[2], 0, 8);
1991 cmd
[4] = 1; /* Start spin cycle */
1992 if (sdkp
->device
->start_stop_pwr_cond
)
1994 scsi_execute_req(sdkp
->device
, cmd
, DMA_NONE
,
1996 SD_TIMEOUT
, SD_MAX_RETRIES
,
1998 spintime_expire
= jiffies
+ 100 * HZ
;
2001 /* Wait 1 second for next try */
2006 * Wait for USB flash devices with slow firmware.
2007 * Yes, this sense key/ASC combination shouldn't
2008 * occur here. It's characteristic of these devices.
2010 } else if (sense_valid
&&
2011 sshdr
.sense_key
== UNIT_ATTENTION
&&
2012 sshdr
.asc
== 0x28) {
2014 spintime_expire
= jiffies
+ 5 * HZ
;
2017 /* Wait 1 second for next try */
2020 /* we don't understand the sense code, so it's
2021 * probably pointless to loop */
2023 sd_printk(KERN_NOTICE
, sdkp
, "Unit Not Ready\n");
2024 sd_print_sense_hdr(sdkp
, &sshdr
);
2029 } while (spintime
&& time_before_eq(jiffies
, spintime_expire
));
2032 if (scsi_status_is_good(the_result
))
2035 printk("not responding...\n");
2040 * Determine whether disk supports Data Integrity Field.
2042 static int sd_read_protection_type(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2044 struct scsi_device
*sdp
= sdkp
->device
;
2048 if (scsi_device_protection(sdp
) == 0 || (buffer
[12] & 1) == 0)
2051 type
= ((buffer
[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2053 if (type
> T10_PI_TYPE3_PROTECTION
)
2055 else if (scsi_host_dif_capable(sdp
->host
, type
))
2058 if (sdkp
->first_scan
|| type
!= sdkp
->protection_type
)
2061 sd_printk(KERN_ERR
, sdkp
, "formatted with unsupported" \
2062 " protection type %u. Disabling disk!\n",
2066 sd_printk(KERN_NOTICE
, sdkp
,
2067 "Enabling DIF Type %u protection\n", type
);
2070 sd_printk(KERN_NOTICE
, sdkp
,
2071 "Disabling DIF Type %u protection\n", type
);
2075 sdkp
->protection_type
= type
;
2080 static void read_capacity_error(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2081 struct scsi_sense_hdr
*sshdr
, int sense_valid
,
2084 if (driver_byte(the_result
) & DRIVER_SENSE
)
2085 sd_print_sense_hdr(sdkp
, sshdr
);
2087 sd_printk(KERN_NOTICE
, sdkp
, "Sense not available.\n");
2090 * Set dirty bit for removable devices if not ready -
2091 * sometimes drives will not report this properly.
2093 if (sdp
->removable
&&
2094 sense_valid
&& sshdr
->sense_key
== NOT_READY
)
2095 set_media_not_present(sdkp
);
2098 * We used to set media_present to 0 here to indicate no media
2099 * in the drive, but some drives fail read capacity even with
2100 * media present, so we can't do that.
2102 sdkp
->capacity
= 0; /* unknown mapped to zero - as usual */
2106 #if RC16_LEN > SD_BUF_SIZE
2107 #error RC16_LEN must not be more than SD_BUF_SIZE
2110 #define READ_CAPACITY_RETRIES_ON_RESET 10
2113 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2114 * and the reported logical block size is bigger than 512 bytes. Note
2115 * that last_sector is a u64 and therefore logical_to_sectors() is not
2118 static bool sd_addressable_capacity(u64 lba
, unsigned int sector_size
)
2120 u64 last_sector
= (lba
+ 1ULL) << (ilog2(sector_size
) - 9);
2122 if (sizeof(sector_t
) == 4 && last_sector
> U32_MAX
)
2128 static int read_capacity_16(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2129 unsigned char *buffer
)
2131 unsigned char cmd
[16];
2132 struct scsi_sense_hdr sshdr
;
2133 int sense_valid
= 0;
2135 int retries
= 3, reset_retries
= READ_CAPACITY_RETRIES_ON_RESET
;
2136 unsigned int alignment
;
2137 unsigned long long lba
;
2138 unsigned sector_size
;
2140 if (sdp
->no_read_capacity_16
)
2145 cmd
[0] = SERVICE_ACTION_IN_16
;
2146 cmd
[1] = SAI_READ_CAPACITY_16
;
2148 memset(buffer
, 0, RC16_LEN
);
2150 the_result
= scsi_execute_req(sdp
, cmd
, DMA_FROM_DEVICE
,
2151 buffer
, RC16_LEN
, &sshdr
,
2152 SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
);
2154 if (media_not_present(sdkp
, &sshdr
))
2158 sense_valid
= scsi_sense_valid(&sshdr
);
2160 sshdr
.sense_key
== ILLEGAL_REQUEST
&&
2161 (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) &&
2163 /* Invalid Command Operation Code or
2164 * Invalid Field in CDB, just retry
2165 * silently with RC10 */
2168 sshdr
.sense_key
== UNIT_ATTENTION
&&
2169 sshdr
.asc
== 0x29 && sshdr
.ascq
== 0x00)
2170 /* Device reset might occur several times,
2171 * give it one more chance */
2172 if (--reset_retries
> 0)
2177 } while (the_result
&& retries
);
2180 sd_print_result(sdkp
, "Read Capacity(16) failed", the_result
);
2181 read_capacity_error(sdkp
, sdp
, &sshdr
, sense_valid
, the_result
);
2185 sector_size
= get_unaligned_be32(&buffer
[8]);
2186 lba
= get_unaligned_be64(&buffer
[0]);
2188 if (sd_read_protection_type(sdkp
, buffer
) < 0) {
2193 if (!sd_addressable_capacity(lba
, sector_size
)) {
2194 sd_printk(KERN_ERR
, sdkp
, "Too big for this kernel. Use a "
2195 "kernel compiled with support for large block "
2201 /* Logical blocks per physical block exponent */
2202 sdkp
->physical_block_size
= (1 << (buffer
[13] & 0xf)) * sector_size
;
2205 sdkp
->rc_basis
= (buffer
[12] >> 4) & 0x3;
2207 /* Lowest aligned logical block */
2208 alignment
= ((buffer
[14] & 0x3f) << 8 | buffer
[15]) * sector_size
;
2209 blk_queue_alignment_offset(sdp
->request_queue
, alignment
);
2210 if (alignment
&& sdkp
->first_scan
)
2211 sd_printk(KERN_NOTICE
, sdkp
,
2212 "physical block alignment offset: %u\n", alignment
);
2214 if (buffer
[14] & 0x80) { /* LBPME */
2217 if (buffer
[14] & 0x40) /* LBPRZ */
2220 sd_config_discard(sdkp
, SD_LBP_WS16
);
2223 sdkp
->capacity
= lba
+ 1;
2227 static int read_capacity_10(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2228 unsigned char *buffer
)
2230 unsigned char cmd
[16];
2231 struct scsi_sense_hdr sshdr
;
2232 int sense_valid
= 0;
2234 int retries
= 3, reset_retries
= READ_CAPACITY_RETRIES_ON_RESET
;
2236 unsigned sector_size
;
2239 cmd
[0] = READ_CAPACITY
;
2240 memset(&cmd
[1], 0, 9);
2241 memset(buffer
, 0, 8);
2243 the_result
= scsi_execute_req(sdp
, cmd
, DMA_FROM_DEVICE
,
2245 SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
);
2247 if (media_not_present(sdkp
, &sshdr
))
2251 sense_valid
= scsi_sense_valid(&sshdr
);
2253 sshdr
.sense_key
== UNIT_ATTENTION
&&
2254 sshdr
.asc
== 0x29 && sshdr
.ascq
== 0x00)
2255 /* Device reset might occur several times,
2256 * give it one more chance */
2257 if (--reset_retries
> 0)
2262 } while (the_result
&& retries
);
2265 sd_print_result(sdkp
, "Read Capacity(10) failed", the_result
);
2266 read_capacity_error(sdkp
, sdp
, &sshdr
, sense_valid
, the_result
);
2270 sector_size
= get_unaligned_be32(&buffer
[4]);
2271 lba
= get_unaligned_be32(&buffer
[0]);
2273 if (sdp
->no_read_capacity_16
&& (lba
== 0xffffffff)) {
2274 /* Some buggy (usb cardreader) devices return an lba of
2275 0xffffffff when the want to report a size of 0 (with
2276 which they really mean no media is present) */
2278 sdkp
->physical_block_size
= sector_size
;
2282 if (!sd_addressable_capacity(lba
, sector_size
)) {
2283 sd_printk(KERN_ERR
, sdkp
, "Too big for this kernel. Use a "
2284 "kernel compiled with support for large block "
2290 sdkp
->capacity
= lba
+ 1;
2291 sdkp
->physical_block_size
= sector_size
;
2295 static int sd_try_rc16_first(struct scsi_device
*sdp
)
2297 if (sdp
->host
->max_cmd_len
< 16)
2299 if (sdp
->try_rc_10_first
)
2301 if (sdp
->scsi_level
> SCSI_SPC_2
)
2303 if (scsi_device_protection(sdp
))
2309 * read disk capacity
2312 sd_read_capacity(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2315 struct scsi_device
*sdp
= sdkp
->device
;
2317 if (sd_try_rc16_first(sdp
)) {
2318 sector_size
= read_capacity_16(sdkp
, sdp
, buffer
);
2319 if (sector_size
== -EOVERFLOW
)
2321 if (sector_size
== -ENODEV
)
2323 if (sector_size
< 0)
2324 sector_size
= read_capacity_10(sdkp
, sdp
, buffer
);
2325 if (sector_size
< 0)
2328 sector_size
= read_capacity_10(sdkp
, sdp
, buffer
);
2329 if (sector_size
== -EOVERFLOW
)
2331 if (sector_size
< 0)
2333 if ((sizeof(sdkp
->capacity
) > 4) &&
2334 (sdkp
->capacity
> 0xffffffffULL
)) {
2335 int old_sector_size
= sector_size
;
2336 sd_printk(KERN_NOTICE
, sdkp
, "Very big device. "
2337 "Trying to use READ CAPACITY(16).\n");
2338 sector_size
= read_capacity_16(sdkp
, sdp
, buffer
);
2339 if (sector_size
< 0) {
2340 sd_printk(KERN_NOTICE
, sdkp
,
2341 "Using 0xffffffff as device size\n");
2342 sdkp
->capacity
= 1 + (sector_t
) 0xffffffff;
2343 sector_size
= old_sector_size
;
2349 /* Some devices are known to return the total number of blocks,
2350 * not the highest block number. Some devices have versions
2351 * which do this and others which do not. Some devices we might
2352 * suspect of doing this but we don't know for certain.
2354 * If we know the reported capacity is wrong, decrement it. If
2355 * we can only guess, then assume the number of blocks is even
2356 * (usually true but not always) and err on the side of lowering
2359 if (sdp
->fix_capacity
||
2360 (sdp
->guess_capacity
&& (sdkp
->capacity
& 0x01))) {
2361 sd_printk(KERN_INFO
, sdkp
, "Adjusting the sector count "
2362 "from its reported value: %llu\n",
2363 (unsigned long long) sdkp
->capacity
);
2368 if (sector_size
== 0) {
2370 sd_printk(KERN_NOTICE
, sdkp
, "Sector size 0 reported, "
2374 if (sector_size
!= 512 &&
2375 sector_size
!= 1024 &&
2376 sector_size
!= 2048 &&
2377 sector_size
!= 4096) {
2378 sd_printk(KERN_NOTICE
, sdkp
, "Unsupported sector size %d.\n",
2381 * The user might want to re-format the drive with
2382 * a supported sectorsize. Once this happens, it
2383 * would be relatively trivial to set the thing up.
2384 * For this reason, we leave the thing in the table.
2388 * set a bogus sector size so the normal read/write
2389 * logic in the block layer will eventually refuse any
2390 * request on this device without tripping over power
2391 * of two sector size assumptions
2395 blk_queue_logical_block_size(sdp
->request_queue
, sector_size
);
2396 blk_queue_physical_block_size(sdp
->request_queue
,
2397 sdkp
->physical_block_size
);
2398 sdkp
->device
->sector_size
= sector_size
;
2400 if (sdkp
->capacity
> 0xffffffff)
2401 sdp
->use_16_for_rw
= 1;
2406 * Print disk capacity
2409 sd_print_capacity(struct scsi_disk
*sdkp
,
2410 sector_t old_capacity
)
2412 int sector_size
= sdkp
->device
->sector_size
;
2413 char cap_str_2
[10], cap_str_10
[10];
2415 string_get_size(sdkp
->capacity
, sector_size
,
2416 STRING_UNITS_2
, cap_str_2
, sizeof(cap_str_2
));
2417 string_get_size(sdkp
->capacity
, sector_size
,
2418 STRING_UNITS_10
, cap_str_10
,
2419 sizeof(cap_str_10
));
2421 if (sdkp
->first_scan
|| old_capacity
!= sdkp
->capacity
) {
2422 sd_printk(KERN_NOTICE
, sdkp
,
2423 "%llu %d-byte logical blocks: (%s/%s)\n",
2424 (unsigned long long)sdkp
->capacity
,
2425 sector_size
, cap_str_10
, cap_str_2
);
2427 if (sdkp
->physical_block_size
!= sector_size
)
2428 sd_printk(KERN_NOTICE
, sdkp
,
2429 "%u-byte physical blocks\n",
2430 sdkp
->physical_block_size
);
2432 sd_zbc_print_zones(sdkp
);
2436 /* called with buffer of length 512 */
2438 sd_do_mode_sense(struct scsi_device
*sdp
, int dbd
, int modepage
,
2439 unsigned char *buffer
, int len
, struct scsi_mode_data
*data
,
2440 struct scsi_sense_hdr
*sshdr
)
2442 return scsi_mode_sense(sdp
, dbd
, modepage
, buffer
, len
,
2443 SD_TIMEOUT
, SD_MAX_RETRIES
, data
,
2448 * read write protect setting, if possible - called only in sd_revalidate_disk()
2449 * called with buffer of length SD_BUF_SIZE
2452 sd_read_write_protect_flag(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2455 struct scsi_device
*sdp
= sdkp
->device
;
2456 struct scsi_mode_data data
;
2457 int old_wp
= sdkp
->write_prot
;
2459 set_disk_ro(sdkp
->disk
, 0);
2460 if (sdp
->skip_ms_page_3f
) {
2461 sd_first_printk(KERN_NOTICE
, sdkp
, "Assuming Write Enabled\n");
2465 if (sdp
->use_192_bytes_for_3f
) {
2466 res
= sd_do_mode_sense(sdp
, 0, 0x3F, buffer
, 192, &data
, NULL
);
2469 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2470 * We have to start carefully: some devices hang if we ask
2471 * for more than is available.
2473 res
= sd_do_mode_sense(sdp
, 0, 0x3F, buffer
, 4, &data
, NULL
);
2476 * Second attempt: ask for page 0 When only page 0 is
2477 * implemented, a request for page 3F may return Sense Key
2478 * 5: Illegal Request, Sense Code 24: Invalid field in
2481 if (!scsi_status_is_good(res
))
2482 res
= sd_do_mode_sense(sdp
, 0, 0, buffer
, 4, &data
, NULL
);
2485 * Third attempt: ask 255 bytes, as we did earlier.
2487 if (!scsi_status_is_good(res
))
2488 res
= sd_do_mode_sense(sdp
, 0, 0x3F, buffer
, 255,
2492 if (!scsi_status_is_good(res
)) {
2493 sd_first_printk(KERN_WARNING
, sdkp
,
2494 "Test WP failed, assume Write Enabled\n");
2496 sdkp
->write_prot
= ((data
.device_specific
& 0x80) != 0);
2497 set_disk_ro(sdkp
->disk
, sdkp
->write_prot
);
2498 if (sdkp
->first_scan
|| old_wp
!= sdkp
->write_prot
) {
2499 sd_printk(KERN_NOTICE
, sdkp
, "Write Protect is %s\n",
2500 sdkp
->write_prot
? "on" : "off");
2501 sd_printk(KERN_DEBUG
, sdkp
, "Mode Sense: %4ph\n", buffer
);
2507 * sd_read_cache_type - called only from sd_revalidate_disk()
2508 * called with buffer of length SD_BUF_SIZE
2511 sd_read_cache_type(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2514 struct scsi_device
*sdp
= sdkp
->device
;
2519 struct scsi_mode_data data
;
2520 struct scsi_sense_hdr sshdr
;
2521 int old_wce
= sdkp
->WCE
;
2522 int old_rcd
= sdkp
->RCD
;
2523 int old_dpofua
= sdkp
->DPOFUA
;
2526 if (sdkp
->cache_override
)
2530 if (sdp
->skip_ms_page_8
) {
2531 if (sdp
->type
== TYPE_RBC
)
2534 if (sdp
->skip_ms_page_3f
)
2537 if (sdp
->use_192_bytes_for_3f
)
2541 } else if (sdp
->type
== TYPE_RBC
) {
2549 /* cautiously ask */
2550 res
= sd_do_mode_sense(sdp
, dbd
, modepage
, buffer
, first_len
,
2553 if (!scsi_status_is_good(res
))
2556 if (!data
.header_length
) {
2559 sd_first_printk(KERN_ERR
, sdkp
,
2560 "Missing header in MODE_SENSE response\n");
2563 /* that went OK, now ask for the proper length */
2567 * We're only interested in the first three bytes, actually.
2568 * But the data cache page is defined for the first 20.
2572 else if (len
> SD_BUF_SIZE
) {
2573 sd_first_printk(KERN_NOTICE
, sdkp
, "Truncating mode parameter "
2574 "data from %d to %d bytes\n", len
, SD_BUF_SIZE
);
2577 if (modepage
== 0x3F && sdp
->use_192_bytes_for_3f
)
2581 if (len
> first_len
)
2582 res
= sd_do_mode_sense(sdp
, dbd
, modepage
, buffer
, len
,
2585 if (scsi_status_is_good(res
)) {
2586 int offset
= data
.header_length
+ data
.block_descriptor_length
;
2588 while (offset
< len
) {
2589 u8 page_code
= buffer
[offset
] & 0x3F;
2590 u8 spf
= buffer
[offset
] & 0x40;
2592 if (page_code
== 8 || page_code
== 6) {
2593 /* We're interested only in the first 3 bytes.
2595 if (len
- offset
<= 2) {
2596 sd_first_printk(KERN_ERR
, sdkp
,
2597 "Incomplete mode parameter "
2601 modepage
= page_code
;
2605 /* Go to the next page */
2606 if (spf
&& len
- offset
> 3)
2607 offset
+= 4 + (buffer
[offset
+2] << 8) +
2609 else if (!spf
&& len
- offset
> 1)
2610 offset
+= 2 + buffer
[offset
+1];
2612 sd_first_printk(KERN_ERR
, sdkp
,
2614 "parameter data\n");
2620 sd_first_printk(KERN_ERR
, sdkp
, "No Caching mode page found\n");
2624 if (modepage
== 8) {
2625 sdkp
->WCE
= ((buffer
[offset
+ 2] & 0x04) != 0);
2626 sdkp
->RCD
= ((buffer
[offset
+ 2] & 0x01) != 0);
2628 sdkp
->WCE
= ((buffer
[offset
+ 2] & 0x01) == 0);
2632 sdkp
->DPOFUA
= (data
.device_specific
& 0x10) != 0;
2633 if (sdp
->broken_fua
) {
2634 sd_first_printk(KERN_NOTICE
, sdkp
, "Disabling FUA\n");
2636 } else if (sdkp
->DPOFUA
&& !sdkp
->device
->use_10_for_rw
&&
2637 !sdkp
->device
->use_16_for_rw
) {
2638 sd_first_printk(KERN_NOTICE
, sdkp
,
2639 "Uses READ/WRITE(6), disabling FUA\n");
2643 /* No cache flush allowed for write protected devices */
2644 if (sdkp
->WCE
&& sdkp
->write_prot
)
2647 if (sdkp
->first_scan
|| old_wce
!= sdkp
->WCE
||
2648 old_rcd
!= sdkp
->RCD
|| old_dpofua
!= sdkp
->DPOFUA
)
2649 sd_printk(KERN_NOTICE
, sdkp
,
2650 "Write cache: %s, read cache: %s, %s\n",
2651 sdkp
->WCE
? "enabled" : "disabled",
2652 sdkp
->RCD
? "disabled" : "enabled",
2653 sdkp
->DPOFUA
? "supports DPO and FUA"
2654 : "doesn't support DPO or FUA");
2660 if (scsi_sense_valid(&sshdr
) &&
2661 sshdr
.sense_key
== ILLEGAL_REQUEST
&&
2662 sshdr
.asc
== 0x24 && sshdr
.ascq
== 0x0)
2663 /* Invalid field in CDB */
2664 sd_first_printk(KERN_NOTICE
, sdkp
, "Cache data unavailable\n");
2666 sd_first_printk(KERN_ERR
, sdkp
,
2667 "Asking for cache data failed\n");
2670 if (sdp
->wce_default_on
) {
2671 sd_first_printk(KERN_NOTICE
, sdkp
,
2672 "Assuming drive cache: write back\n");
2675 sd_first_printk(KERN_ERR
, sdkp
,
2676 "Assuming drive cache: write through\n");
2684 * The ATO bit indicates whether the DIF application tag is available
2685 * for use by the operating system.
2687 static void sd_read_app_tag_own(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2690 struct scsi_device
*sdp
= sdkp
->device
;
2691 struct scsi_mode_data data
;
2692 struct scsi_sense_hdr sshdr
;
2694 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_ZBC
)
2697 if (sdkp
->protection_type
== 0)
2700 res
= scsi_mode_sense(sdp
, 1, 0x0a, buffer
, 36, SD_TIMEOUT
,
2701 SD_MAX_RETRIES
, &data
, &sshdr
);
2703 if (!scsi_status_is_good(res
) || !data
.header_length
||
2705 sd_first_printk(KERN_WARNING
, sdkp
,
2706 "getting Control mode page failed, assume no ATO\n");
2708 if (scsi_sense_valid(&sshdr
))
2709 sd_print_sense_hdr(sdkp
, &sshdr
);
2714 offset
= data
.header_length
+ data
.block_descriptor_length
;
2716 if ((buffer
[offset
] & 0x3f) != 0x0a) {
2717 sd_first_printk(KERN_ERR
, sdkp
, "ATO Got wrong page\n");
2721 if ((buffer
[offset
+ 5] & 0x80) == 0)
2730 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2731 * @disk: disk to query
2733 static void sd_read_block_limits(struct scsi_disk
*sdkp
)
2735 unsigned int sector_sz
= sdkp
->device
->sector_size
;
2736 const int vpd_len
= 64;
2737 unsigned char *buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2740 /* Block Limits VPD */
2741 scsi_get_vpd_page(sdkp
->device
, 0xb0, buffer
, vpd_len
))
2744 blk_queue_io_min(sdkp
->disk
->queue
,
2745 get_unaligned_be16(&buffer
[6]) * sector_sz
);
2747 sdkp
->max_xfer_blocks
= get_unaligned_be32(&buffer
[8]);
2748 sdkp
->opt_xfer_blocks
= get_unaligned_be32(&buffer
[12]);
2750 if (buffer
[3] == 0x3c) {
2751 unsigned int lba_count
, desc_count
;
2753 sdkp
->max_ws_blocks
= (u32
)get_unaligned_be64(&buffer
[36]);
2758 lba_count
= get_unaligned_be32(&buffer
[20]);
2759 desc_count
= get_unaligned_be32(&buffer
[24]);
2761 if (lba_count
&& desc_count
)
2762 sdkp
->max_unmap_blocks
= lba_count
;
2764 sdkp
->unmap_granularity
= get_unaligned_be32(&buffer
[28]);
2766 if (buffer
[32] & 0x80)
2767 sdkp
->unmap_alignment
=
2768 get_unaligned_be32(&buffer
[32]) & ~(1 << 31);
2770 if (!sdkp
->lbpvpd
) { /* LBP VPD page not provided */
2772 if (sdkp
->max_unmap_blocks
)
2773 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2775 sd_config_discard(sdkp
, SD_LBP_WS16
);
2777 } else { /* LBP VPD page tells us what to use */
2778 if (sdkp
->lbpu
&& sdkp
->max_unmap_blocks
&& !sdkp
->lbprz
)
2779 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2780 else if (sdkp
->lbpws
)
2781 sd_config_discard(sdkp
, SD_LBP_WS16
);
2782 else if (sdkp
->lbpws10
)
2783 sd_config_discard(sdkp
, SD_LBP_WS10
);
2784 else if (sdkp
->lbpu
&& sdkp
->max_unmap_blocks
)
2785 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2787 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
2796 * sd_read_block_characteristics - Query block dev. characteristics
2797 * @disk: disk to query
2799 static void sd_read_block_characteristics(struct scsi_disk
*sdkp
)
2801 struct request_queue
*q
= sdkp
->disk
->queue
;
2802 unsigned char *buffer
;
2804 const int vpd_len
= 64;
2806 buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2809 /* Block Device Characteristics VPD */
2810 scsi_get_vpd_page(sdkp
->device
, 0xb1, buffer
, vpd_len
))
2813 rot
= get_unaligned_be16(&buffer
[4]);
2816 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
2817 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, q
);
2820 if (sdkp
->device
->type
== TYPE_ZBC
) {
2822 q
->limits
.zoned
= BLK_ZONED_HM
;
2824 sdkp
->zoned
= (buffer
[8] >> 4) & 3;
2825 if (sdkp
->zoned
== 1)
2827 q
->limits
.zoned
= BLK_ZONED_HA
;
2830 * Treat drive-managed devices as
2831 * regular block devices.
2833 q
->limits
.zoned
= BLK_ZONED_NONE
;
2835 if (blk_queue_is_zoned(q
) && sdkp
->first_scan
)
2836 sd_printk(KERN_NOTICE
, sdkp
, "Host-%s zoned block device\n",
2837 q
->limits
.zoned
== BLK_ZONED_HM
? "managed" : "aware");
2844 * sd_read_block_provisioning - Query provisioning VPD page
2845 * @disk: disk to query
2847 static void sd_read_block_provisioning(struct scsi_disk
*sdkp
)
2849 unsigned char *buffer
;
2850 const int vpd_len
= 8;
2852 if (sdkp
->lbpme
== 0)
2855 buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2857 if (!buffer
|| scsi_get_vpd_page(sdkp
->device
, 0xb2, buffer
, vpd_len
))
2861 sdkp
->lbpu
= (buffer
[5] >> 7) & 1; /* UNMAP */
2862 sdkp
->lbpws
= (buffer
[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2863 sdkp
->lbpws10
= (buffer
[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2869 static void sd_read_write_same(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2871 struct scsi_device
*sdev
= sdkp
->device
;
2873 if (sdev
->host
->no_write_same
) {
2874 sdev
->no_write_same
= 1;
2879 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, INQUIRY
) < 0) {
2880 /* too large values might cause issues with arcmsr */
2881 int vpd_buf_len
= 64;
2883 sdev
->no_report_opcodes
= 1;
2885 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2886 * CODES is unsupported and the device has an ATA
2887 * Information VPD page (SAT).
2889 if (!scsi_get_vpd_page(sdev
, 0x89, buffer
, vpd_buf_len
))
2890 sdev
->no_write_same
= 1;
2893 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, WRITE_SAME_16
) == 1)
2896 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, WRITE_SAME
) == 1)
2901 * sd_revalidate_disk - called the first time a new disk is seen,
2902 * performs disk spin up, read_capacity, etc.
2903 * @disk: struct gendisk we care about
2905 static int sd_revalidate_disk(struct gendisk
*disk
)
2907 struct scsi_disk
*sdkp
= scsi_disk(disk
);
2908 struct scsi_device
*sdp
= sdkp
->device
;
2909 struct request_queue
*q
= sdkp
->disk
->queue
;
2910 sector_t old_capacity
= sdkp
->capacity
;
2911 unsigned char *buffer
;
2912 unsigned int dev_max
, rw_max
;
2914 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
,
2915 "sd_revalidate_disk\n"));
2918 * If the device is offline, don't try and read capacity or any
2919 * of the other niceties.
2921 if (!scsi_device_online(sdp
))
2924 buffer
= kmalloc(SD_BUF_SIZE
, GFP_KERNEL
);
2926 sd_printk(KERN_WARNING
, sdkp
, "sd_revalidate_disk: Memory "
2927 "allocation failure.\n");
2931 sd_spinup_disk(sdkp
);
2934 * Without media there is no reason to ask; moreover, some devices
2935 * react badly if we do.
2937 if (sdkp
->media_present
) {
2938 sd_read_capacity(sdkp
, buffer
);
2940 if (scsi_device_supports_vpd(sdp
)) {
2941 sd_read_block_provisioning(sdkp
);
2942 sd_read_block_limits(sdkp
);
2943 sd_read_block_characteristics(sdkp
);
2944 sd_zbc_read_zones(sdkp
, buffer
);
2947 sd_print_capacity(sdkp
, old_capacity
);
2949 sd_read_write_protect_flag(sdkp
, buffer
);
2950 sd_read_cache_type(sdkp
, buffer
);
2951 sd_read_app_tag_own(sdkp
, buffer
);
2952 sd_read_write_same(sdkp
, buffer
);
2955 sdkp
->first_scan
= 0;
2958 * We now have all cache related info, determine how we deal
2959 * with flush requests.
2961 sd_set_flush_flag(sdkp
);
2963 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2964 dev_max
= sdp
->use_16_for_rw
? SD_MAX_XFER_BLOCKS
: SD_DEF_XFER_BLOCKS
;
2966 /* Some devices report a maximum block count for READ/WRITE requests. */
2967 dev_max
= min_not_zero(dev_max
, sdkp
->max_xfer_blocks
);
2968 q
->limits
.max_dev_sectors
= logical_to_sectors(sdp
, dev_max
);
2971 * Use the device's preferred I/O size for reads and writes
2972 * unless the reported value is unreasonably small, large, or
2975 if (sdkp
->opt_xfer_blocks
&&
2976 sdkp
->opt_xfer_blocks
<= dev_max
&&
2977 sdkp
->opt_xfer_blocks
<= SD_DEF_XFER_BLOCKS
&&
2978 logical_to_bytes(sdp
, sdkp
->opt_xfer_blocks
) >= PAGE_SIZE
) {
2979 q
->limits
.io_opt
= logical_to_bytes(sdp
, sdkp
->opt_xfer_blocks
);
2980 rw_max
= logical_to_sectors(sdp
, sdkp
->opt_xfer_blocks
);
2982 rw_max
= min_not_zero(logical_to_sectors(sdp
, dev_max
),
2983 (sector_t
)BLK_DEF_MAX_SECTORS
);
2985 /* Combine with controller limits */
2986 q
->limits
.max_sectors
= min(rw_max
, queue_max_hw_sectors(q
));
2988 set_capacity(disk
, logical_to_sectors(sdp
, sdkp
->capacity
));
2989 sd_config_write_same(sdkp
);
2997 * sd_unlock_native_capacity - unlock native capacity
2998 * @disk: struct gendisk to set capacity for
3000 * Block layer calls this function if it detects that partitions
3001 * on @disk reach beyond the end of the device. If the SCSI host
3002 * implements ->unlock_native_capacity() method, it's invoked to
3003 * give it a chance to adjust the device capacity.
3006 * Defined by block layer. Might sleep.
3008 static void sd_unlock_native_capacity(struct gendisk
*disk
)
3010 struct scsi_device
*sdev
= scsi_disk(disk
)->device
;
3012 if (sdev
->host
->hostt
->unlock_native_capacity
)
3013 sdev
->host
->hostt
->unlock_native_capacity(sdev
);
3017 * sd_format_disk_name - format disk name
3018 * @prefix: name prefix - ie. "sd" for SCSI disks
3019 * @index: index of the disk to format name for
3020 * @buf: output buffer
3021 * @buflen: length of the output buffer
3023 * SCSI disk names starts at sda. The 26th device is sdz and the
3024 * 27th is sdaa. The last one for two lettered suffix is sdzz
3025 * which is followed by sdaaa.
3027 * This is basically 26 base counting with one extra 'nil' entry
3028 * at the beginning from the second digit on and can be
3029 * determined using similar method as 26 base conversion with the
3030 * index shifted -1 after each digit is computed.
3036 * 0 on success, -errno on failure.
3038 static int sd_format_disk_name(char *prefix
, int index
, char *buf
, int buflen
)
3040 const int base
= 'z' - 'a' + 1;
3041 char *begin
= buf
+ strlen(prefix
);
3042 char *end
= buf
+ buflen
;
3052 *--p
= 'a' + (index
% unit
);
3053 index
= (index
/ unit
) - 1;
3054 } while (index
>= 0);
3056 memmove(begin
, p
, end
- p
);
3057 memcpy(buf
, prefix
, strlen(prefix
));
3063 * The asynchronous part of sd_probe
3065 static void sd_probe_async(void *data
, async_cookie_t cookie
)
3067 struct scsi_disk
*sdkp
= data
;
3068 struct scsi_device
*sdp
;
3075 index
= sdkp
->index
;
3076 dev
= &sdp
->sdev_gendev
;
3078 gd
->major
= sd_major((index
& 0xf0) >> 4);
3079 gd
->first_minor
= ((index
& 0xf) << 4) | (index
& 0xfff00);
3080 gd
->minors
= SD_MINORS
;
3082 gd
->fops
= &sd_fops
;
3083 gd
->private_data
= &sdkp
->driver
;
3084 gd
->queue
= sdkp
->device
->request_queue
;
3086 /* defaults, until the device tells us otherwise */
3087 sdp
->sector_size
= 512;
3089 sdkp
->media_present
= 1;
3090 sdkp
->write_prot
= 0;
3091 sdkp
->cache_override
= 0;
3095 sdkp
->first_scan
= 1;
3096 sdkp
->max_medium_access_timeouts
= SD_MAX_MEDIUM_TIMEOUTS
;
3098 sd_revalidate_disk(gd
);
3100 gd
->flags
= GENHD_FL_EXT_DEVT
;
3101 if (sdp
->removable
) {
3102 gd
->flags
|= GENHD_FL_REMOVABLE
;
3103 gd
->events
|= DISK_EVENT_MEDIA_CHANGE
;
3106 blk_pm_runtime_init(sdp
->request_queue
, dev
);
3107 device_add_disk(dev
, gd
);
3109 sd_dif_config_host(sdkp
);
3111 sd_revalidate_disk(gd
);
3113 sd_printk(KERN_NOTICE
, sdkp
, "Attached SCSI %sdisk\n",
3114 sdp
->removable
? "removable " : "");
3115 scsi_autopm_put_device(sdp
);
3116 put_device(&sdkp
->dev
);
3120 * sd_probe - called during driver initialization and whenever a
3121 * new scsi device is attached to the system. It is called once
3122 * for each scsi device (not just disks) present.
3123 * @dev: pointer to device object
3125 * Returns 0 if successful (or not interested in this scsi device
3126 * (e.g. scanner)); 1 when there is an error.
3128 * Note: this function is invoked from the scsi mid-level.
3129 * This function sets up the mapping between a given
3130 * <host,channel,id,lun> (found in sdp) and new device name
3131 * (e.g. /dev/sda). More precisely it is the block device major
3132 * and minor number that is chosen here.
3134 * Assume sd_probe is not re-entrant (for time being)
3135 * Also think about sd_probe() and sd_remove() running coincidentally.
3137 static int sd_probe(struct device
*dev
)
3139 struct scsi_device
*sdp
= to_scsi_device(dev
);
3140 struct scsi_disk
*sdkp
;
3145 scsi_autopm_get_device(sdp
);
3147 if (sdp
->type
!= TYPE_DISK
&&
3148 sdp
->type
!= TYPE_ZBC
&&
3149 sdp
->type
!= TYPE_MOD
&&
3150 sdp
->type
!= TYPE_RBC
)
3153 #ifndef CONFIG_BLK_DEV_ZONED
3154 if (sdp
->type
== TYPE_ZBC
)
3157 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO
, sdp
,
3161 sdkp
= kzalloc(sizeof(*sdkp
), GFP_KERNEL
);
3165 gd
= alloc_disk(SD_MINORS
);
3170 if (!ida_pre_get(&sd_index_ida
, GFP_KERNEL
))
3173 spin_lock(&sd_index_lock
);
3174 error
= ida_get_new(&sd_index_ida
, &index
);
3175 spin_unlock(&sd_index_lock
);
3176 } while (error
== -EAGAIN
);
3179 sdev_printk(KERN_WARNING
, sdp
, "sd_probe: memory exhausted.\n");
3183 error
= sd_format_disk_name("sd", index
, gd
->disk_name
, DISK_NAME_LEN
);
3185 sdev_printk(KERN_WARNING
, sdp
, "SCSI disk (sd) name length exceeded.\n");
3186 goto out_free_index
;
3190 sdkp
->driver
= &sd_template
;
3192 sdkp
->index
= index
;
3193 atomic_set(&sdkp
->openers
, 0);
3194 atomic_set(&sdkp
->device
->ioerr_cnt
, 0);
3196 if (!sdp
->request_queue
->rq_timeout
) {
3197 if (sdp
->type
!= TYPE_MOD
)
3198 blk_queue_rq_timeout(sdp
->request_queue
, SD_TIMEOUT
);
3200 blk_queue_rq_timeout(sdp
->request_queue
,
3204 device_initialize(&sdkp
->dev
);
3205 sdkp
->dev
.parent
= dev
;
3206 sdkp
->dev
.class = &sd_disk_class
;
3207 dev_set_name(&sdkp
->dev
, "%s", dev_name(dev
));
3209 error
= device_add(&sdkp
->dev
);
3211 goto out_free_index
;
3214 dev_set_drvdata(dev
, sdkp
);
3216 get_device(&sdkp
->dev
); /* prevent release before async_schedule */
3217 async_schedule_domain(sd_probe_async
, sdkp
, &scsi_sd_probe_domain
);
3222 spin_lock(&sd_index_lock
);
3223 ida_remove(&sd_index_ida
, index
);
3224 spin_unlock(&sd_index_lock
);
3230 scsi_autopm_put_device(sdp
);
3235 * sd_remove - called whenever a scsi disk (previously recognized by
3236 * sd_probe) is detached from the system. It is called (potentially
3237 * multiple times) during sd module unload.
3238 * @sdp: pointer to mid level scsi device object
3240 * Note: this function is invoked from the scsi mid-level.
3241 * This function potentially frees up a device name (e.g. /dev/sdc)
3242 * that could be re-used by a subsequent sd_probe().
3243 * This function is not called when the built-in sd driver is "exit-ed".
3245 static int sd_remove(struct device
*dev
)
3247 struct scsi_disk
*sdkp
;
3250 sdkp
= dev_get_drvdata(dev
);
3251 devt
= disk_devt(sdkp
->disk
);
3252 scsi_autopm_get_device(sdkp
->device
);
3254 async_synchronize_full_domain(&scsi_sd_pm_domain
);
3255 async_synchronize_full_domain(&scsi_sd_probe_domain
);
3256 device_del(&sdkp
->dev
);
3257 del_gendisk(sdkp
->disk
);
3260 sd_zbc_remove(sdkp
);
3262 blk_register_region(devt
, SD_MINORS
, NULL
,
3263 sd_default_probe
, NULL
, NULL
);
3265 mutex_lock(&sd_ref_mutex
);
3266 dev_set_drvdata(dev
, NULL
);
3267 put_device(&sdkp
->dev
);
3268 mutex_unlock(&sd_ref_mutex
);
3274 * scsi_disk_release - Called to free the scsi_disk structure
3275 * @dev: pointer to embedded class device
3277 * sd_ref_mutex must be held entering this routine. Because it is
3278 * called on last put, you should always use the scsi_disk_get()
3279 * scsi_disk_put() helpers which manipulate the semaphore directly
3280 * and never do a direct put_device.
3282 static void scsi_disk_release(struct device
*dev
)
3284 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
3285 struct gendisk
*disk
= sdkp
->disk
;
3287 spin_lock(&sd_index_lock
);
3288 ida_remove(&sd_index_ida
, sdkp
->index
);
3289 spin_unlock(&sd_index_lock
);
3291 disk
->private_data
= NULL
;
3293 put_device(&sdkp
->device
->sdev_gendev
);
3298 static int sd_start_stop_device(struct scsi_disk
*sdkp
, int start
)
3300 unsigned char cmd
[6] = { START_STOP
}; /* START_VALID */
3301 struct scsi_sense_hdr sshdr
;
3302 struct scsi_device
*sdp
= sdkp
->device
;
3306 cmd
[4] |= 1; /* START */
3308 if (sdp
->start_stop_pwr_cond
)
3309 cmd
[4] |= start
? 1 << 4 : 3 << 4; /* Active or Standby */
3311 if (!scsi_device_online(sdp
))
3314 res
= scsi_execute_req_flags(sdp
, cmd
, DMA_NONE
, NULL
, 0, &sshdr
,
3315 SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
, 0, RQF_PM
);
3317 sd_print_result(sdkp
, "Start/Stop Unit failed", res
);
3318 if (driver_byte(res
) & DRIVER_SENSE
)
3319 sd_print_sense_hdr(sdkp
, &sshdr
);
3320 if (scsi_sense_valid(&sshdr
) &&
3321 /* 0x3a is medium not present */
3326 /* SCSI error codes must not go to the generic layer */
3334 * Send a SYNCHRONIZE CACHE instruction down to the device through
3335 * the normal SCSI command structure. Wait for the command to
3338 static void sd_shutdown(struct device
*dev
)
3340 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3343 return; /* this can happen */
3345 if (pm_runtime_suspended(dev
))
3348 if (sdkp
->WCE
&& sdkp
->media_present
) {
3349 sd_printk(KERN_NOTICE
, sdkp
, "Synchronizing SCSI cache\n");
3350 sd_sync_cache(sdkp
);
3353 if (system_state
!= SYSTEM_RESTART
&& sdkp
->device
->manage_start_stop
) {
3354 sd_printk(KERN_NOTICE
, sdkp
, "Stopping disk\n");
3355 sd_start_stop_device(sdkp
, 0);
3359 static int sd_suspend_common(struct device
*dev
, bool ignore_stop_errors
)
3361 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3364 if (!sdkp
) /* E.g.: runtime suspend following sd_remove() */
3367 if (sdkp
->WCE
&& sdkp
->media_present
) {
3368 sd_printk(KERN_NOTICE
, sdkp
, "Synchronizing SCSI cache\n");
3369 ret
= sd_sync_cache(sdkp
);
3371 /* ignore OFFLINE device */
3378 if (sdkp
->device
->manage_start_stop
) {
3379 sd_printk(KERN_NOTICE
, sdkp
, "Stopping disk\n");
3380 /* an error is not worth aborting a system sleep */
3381 ret
= sd_start_stop_device(sdkp
, 0);
3382 if (ignore_stop_errors
)
3390 static int sd_suspend_system(struct device
*dev
)
3392 return sd_suspend_common(dev
, true);
3395 static int sd_suspend_runtime(struct device
*dev
)
3397 return sd_suspend_common(dev
, false);
3400 static int sd_resume(struct device
*dev
)
3402 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3404 if (!sdkp
) /* E.g.: runtime resume at the start of sd_probe() */
3407 if (!sdkp
->device
->manage_start_stop
)
3410 sd_printk(KERN_NOTICE
, sdkp
, "Starting disk\n");
3411 return sd_start_stop_device(sdkp
, 1);
3415 * init_sd - entry point for this driver (both when built in or when
3418 * Note: this function registers this driver with the scsi mid-level.
3420 static int __init
init_sd(void)
3422 int majors
= 0, i
, err
;
3424 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3426 for (i
= 0; i
< SD_MAJORS
; i
++) {
3427 if (register_blkdev(sd_major(i
), "sd") != 0)
3430 blk_register_region(sd_major(i
), SD_MINORS
, NULL
,
3431 sd_default_probe
, NULL
, NULL
);
3437 err
= class_register(&sd_disk_class
);
3441 sd_cdb_cache
= kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE
,
3443 if (!sd_cdb_cache
) {
3444 printk(KERN_ERR
"sd: can't init extended cdb cache\n");
3449 sd_cdb_pool
= mempool_create_slab_pool(SD_MEMPOOL_SIZE
, sd_cdb_cache
);
3451 printk(KERN_ERR
"sd: can't init extended cdb pool\n");
3456 err
= scsi_register_driver(&sd_template
.gendrv
);
3458 goto err_out_driver
;
3463 mempool_destroy(sd_cdb_pool
);
3466 kmem_cache_destroy(sd_cdb_cache
);
3469 class_unregister(&sd_disk_class
);
3471 for (i
= 0; i
< SD_MAJORS
; i
++)
3472 unregister_blkdev(sd_major(i
), "sd");
3477 * exit_sd - exit point for this driver (when it is a module).
3479 * Note: this function unregisters this driver from the scsi mid-level.
3481 static void __exit
exit_sd(void)
3485 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3487 scsi_unregister_driver(&sd_template
.gendrv
);
3488 mempool_destroy(sd_cdb_pool
);
3489 kmem_cache_destroy(sd_cdb_cache
);
3491 class_unregister(&sd_disk_class
);
3493 for (i
= 0; i
< SD_MAJORS
; i
++) {
3494 blk_unregister_region(sd_major(i
), SD_MINORS
);
3495 unregister_blkdev(sd_major(i
), "sd");
3499 module_init(init_sd
);
3500 module_exit(exit_sd
);
3502 static void sd_print_sense_hdr(struct scsi_disk
*sdkp
,
3503 struct scsi_sense_hdr
*sshdr
)
3505 scsi_print_sense_hdr(sdkp
->device
,
3506 sdkp
->disk
? sdkp
->disk
->disk_name
: NULL
, sshdr
);
3509 static void sd_print_result(const struct scsi_disk
*sdkp
, const char *msg
,
3512 const char *hb_string
= scsi_hostbyte_string(result
);
3513 const char *db_string
= scsi_driverbyte_string(result
);
3515 if (hb_string
|| db_string
)
3516 sd_printk(KERN_INFO
, sdkp
,
3517 "%s: Result: hostbyte=%s driverbyte=%s\n", msg
,
3518 hb_string
? hb_string
: "invalid",
3519 db_string
? db_string
: "invalid");
3521 sd_printk(KERN_INFO
, sdkp
,
3522 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3523 msg
, host_byte(result
), driver_byte(result
));