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 <asm/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
);
97 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
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 void sd_rescan(struct device
*);
114 static int sd_init_command(struct scsi_cmnd
*SCpnt
);
115 static void sd_uninit_command(struct scsi_cmnd
*SCpnt
);
116 static int sd_done(struct scsi_cmnd
*);
117 static int sd_eh_action(struct scsi_cmnd
*, int);
118 static void sd_read_capacity(struct scsi_disk
*sdkp
, unsigned char *buffer
);
119 static void scsi_disk_release(struct device
*cdev
);
120 static void sd_print_sense_hdr(struct scsi_disk
*, struct scsi_sense_hdr
*);
121 static void sd_print_result(const struct scsi_disk
*, const char *, int);
123 static DEFINE_SPINLOCK(sd_index_lock
);
124 static DEFINE_IDA(sd_index_ida
);
126 /* This semaphore is used to mediate the 0->1 reference get in the
127 * face of object destruction (i.e. we can't allow a get on an
128 * object after last put) */
129 static DEFINE_MUTEX(sd_ref_mutex
);
131 static struct kmem_cache
*sd_cdb_cache
;
132 static mempool_t
*sd_cdb_pool
;
134 static const char *sd_cache_types
[] = {
135 "write through", "none", "write back",
136 "write back, no read (daft)"
139 static void sd_set_flush_flag(struct scsi_disk
*sdkp
)
141 bool wc
= false, fua
= false;
149 blk_queue_write_cache(sdkp
->disk
->queue
, wc
, fua
);
153 cache_type_store(struct device
*dev
, struct device_attribute
*attr
,
154 const char *buf
, size_t count
)
156 int i
, ct
= -1, rcd
, wce
, sp
;
157 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
158 struct scsi_device
*sdp
= sdkp
->device
;
161 struct scsi_mode_data data
;
162 struct scsi_sense_hdr sshdr
;
163 static const char temp
[] = "temporary ";
166 if (sdp
->type
!= TYPE_DISK
)
167 /* no cache control on RBC devices; theoretically they
168 * can do it, but there's probably so many exceptions
169 * it's not worth the risk */
172 if (strncmp(buf
, temp
, sizeof(temp
) - 1) == 0) {
173 buf
+= sizeof(temp
) - 1;
174 sdkp
->cache_override
= 1;
176 sdkp
->cache_override
= 0;
179 for (i
= 0; i
< ARRAY_SIZE(sd_cache_types
); i
++) {
180 len
= strlen(sd_cache_types
[i
]);
181 if (strncmp(sd_cache_types
[i
], buf
, len
) == 0 &&
189 rcd
= ct
& 0x01 ? 1 : 0;
190 wce
= (ct
& 0x02) && !sdkp
->write_prot
? 1 : 0;
192 if (sdkp
->cache_override
) {
195 sd_set_flush_flag(sdkp
);
199 if (scsi_mode_sense(sdp
, 0x08, 8, buffer
, sizeof(buffer
), SD_TIMEOUT
,
200 SD_MAX_RETRIES
, &data
, NULL
))
202 len
= min_t(size_t, sizeof(buffer
), data
.length
- data
.header_length
-
203 data
.block_descriptor_length
);
204 buffer_data
= buffer
+ data
.header_length
+
205 data
.block_descriptor_length
;
206 buffer_data
[2] &= ~0x05;
207 buffer_data
[2] |= wce
<< 2 | rcd
;
208 sp
= buffer_data
[0] & 0x80 ? 1 : 0;
209 buffer_data
[0] &= ~0x80;
211 if (scsi_mode_select(sdp
, 1, sp
, 8, buffer_data
, len
, SD_TIMEOUT
,
212 SD_MAX_RETRIES
, &data
, &sshdr
)) {
213 if (scsi_sense_valid(&sshdr
))
214 sd_print_sense_hdr(sdkp
, &sshdr
);
217 revalidate_disk(sdkp
->disk
);
222 manage_start_stop_show(struct device
*dev
, struct device_attribute
*attr
,
225 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
226 struct scsi_device
*sdp
= sdkp
->device
;
228 return snprintf(buf
, 20, "%u\n", sdp
->manage_start_stop
);
232 manage_start_stop_store(struct device
*dev
, struct device_attribute
*attr
,
233 const char *buf
, size_t count
)
235 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
236 struct scsi_device
*sdp
= sdkp
->device
;
239 if (!capable(CAP_SYS_ADMIN
))
242 if (kstrtobool(buf
, &v
))
245 sdp
->manage_start_stop
= v
;
249 static DEVICE_ATTR_RW(manage_start_stop
);
252 allow_restart_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
254 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
256 return snprintf(buf
, 40, "%d\n", sdkp
->device
->allow_restart
);
260 allow_restart_store(struct device
*dev
, struct device_attribute
*attr
,
261 const char *buf
, size_t count
)
264 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
265 struct scsi_device
*sdp
= sdkp
->device
;
267 if (!capable(CAP_SYS_ADMIN
))
270 if (sdp
->type
!= TYPE_DISK
)
273 if (kstrtobool(buf
, &v
))
276 sdp
->allow_restart
= v
;
280 static DEVICE_ATTR_RW(allow_restart
);
283 cache_type_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
285 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
286 int ct
= sdkp
->RCD
+ 2*sdkp
->WCE
;
288 return snprintf(buf
, 40, "%s\n", sd_cache_types
[ct
]);
290 static DEVICE_ATTR_RW(cache_type
);
293 FUA_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
295 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
297 return snprintf(buf
, 20, "%u\n", sdkp
->DPOFUA
);
299 static DEVICE_ATTR_RO(FUA
);
302 protection_type_show(struct device
*dev
, struct device_attribute
*attr
,
305 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
307 return snprintf(buf
, 20, "%u\n", sdkp
->protection_type
);
311 protection_type_store(struct device
*dev
, struct device_attribute
*attr
,
312 const char *buf
, size_t count
)
314 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
318 if (!capable(CAP_SYS_ADMIN
))
321 err
= kstrtouint(buf
, 10, &val
);
326 if (val
>= 0 && val
<= T10_PI_TYPE3_PROTECTION
)
327 sdkp
->protection_type
= val
;
331 static DEVICE_ATTR_RW(protection_type
);
334 protection_mode_show(struct device
*dev
, struct device_attribute
*attr
,
337 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
338 struct scsi_device
*sdp
= sdkp
->device
;
339 unsigned int dif
, dix
;
341 dif
= scsi_host_dif_capable(sdp
->host
, sdkp
->protection_type
);
342 dix
= scsi_host_dix_capable(sdp
->host
, sdkp
->protection_type
);
344 if (!dix
&& scsi_host_dix_capable(sdp
->host
, T10_PI_TYPE0_PROTECTION
)) {
350 return snprintf(buf
, 20, "none\n");
352 return snprintf(buf
, 20, "%s%u\n", dix
? "dix" : "dif", dif
);
354 static DEVICE_ATTR_RO(protection_mode
);
357 app_tag_own_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
359 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
361 return snprintf(buf
, 20, "%u\n", sdkp
->ATO
);
363 static DEVICE_ATTR_RO(app_tag_own
);
366 thin_provisioning_show(struct device
*dev
, struct device_attribute
*attr
,
369 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
371 return snprintf(buf
, 20, "%u\n", sdkp
->lbpme
);
373 static DEVICE_ATTR_RO(thin_provisioning
);
375 static const char *lbp_mode
[] = {
376 [SD_LBP_FULL
] = "full",
377 [SD_LBP_UNMAP
] = "unmap",
378 [SD_LBP_WS16
] = "writesame_16",
379 [SD_LBP_WS10
] = "writesame_10",
380 [SD_LBP_ZERO
] = "writesame_zero",
381 [SD_LBP_DISABLE
] = "disabled",
385 provisioning_mode_show(struct device
*dev
, struct device_attribute
*attr
,
388 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
390 return snprintf(buf
, 20, "%s\n", lbp_mode
[sdkp
->provisioning_mode
]);
394 provisioning_mode_store(struct device
*dev
, struct device_attribute
*attr
,
395 const char *buf
, size_t count
)
397 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
398 struct scsi_device
*sdp
= sdkp
->device
;
400 if (!capable(CAP_SYS_ADMIN
))
403 if (sdp
->type
!= TYPE_DISK
)
406 if (!strncmp(buf
, lbp_mode
[SD_LBP_UNMAP
], 20))
407 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
408 else if (!strncmp(buf
, lbp_mode
[SD_LBP_WS16
], 20))
409 sd_config_discard(sdkp
, SD_LBP_WS16
);
410 else if (!strncmp(buf
, lbp_mode
[SD_LBP_WS10
], 20))
411 sd_config_discard(sdkp
, SD_LBP_WS10
);
412 else if (!strncmp(buf
, lbp_mode
[SD_LBP_ZERO
], 20))
413 sd_config_discard(sdkp
, SD_LBP_ZERO
);
414 else if (!strncmp(buf
, lbp_mode
[SD_LBP_DISABLE
], 20))
415 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
421 static DEVICE_ATTR_RW(provisioning_mode
);
424 max_medium_access_timeouts_show(struct device
*dev
,
425 struct device_attribute
*attr
, char *buf
)
427 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
429 return snprintf(buf
, 20, "%u\n", sdkp
->max_medium_access_timeouts
);
433 max_medium_access_timeouts_store(struct device
*dev
,
434 struct device_attribute
*attr
, const char *buf
,
437 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
440 if (!capable(CAP_SYS_ADMIN
))
443 err
= kstrtouint(buf
, 10, &sdkp
->max_medium_access_timeouts
);
445 return err
? err
: count
;
447 static DEVICE_ATTR_RW(max_medium_access_timeouts
);
450 max_write_same_blocks_show(struct device
*dev
, struct device_attribute
*attr
,
453 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
455 return snprintf(buf
, 20, "%u\n", sdkp
->max_ws_blocks
);
459 max_write_same_blocks_store(struct device
*dev
, struct device_attribute
*attr
,
460 const char *buf
, size_t count
)
462 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
463 struct scsi_device
*sdp
= sdkp
->device
;
467 if (!capable(CAP_SYS_ADMIN
))
470 if (sdp
->type
!= TYPE_DISK
)
473 err
= kstrtoul(buf
, 10, &max
);
479 sdp
->no_write_same
= 1;
480 else if (max
<= SD_MAX_WS16_BLOCKS
) {
481 sdp
->no_write_same
= 0;
482 sdkp
->max_ws_blocks
= max
;
485 sd_config_write_same(sdkp
);
489 static DEVICE_ATTR_RW(max_write_same_blocks
);
491 static struct attribute
*sd_disk_attrs
[] = {
492 &dev_attr_cache_type
.attr
,
494 &dev_attr_allow_restart
.attr
,
495 &dev_attr_manage_start_stop
.attr
,
496 &dev_attr_protection_type
.attr
,
497 &dev_attr_protection_mode
.attr
,
498 &dev_attr_app_tag_own
.attr
,
499 &dev_attr_thin_provisioning
.attr
,
500 &dev_attr_provisioning_mode
.attr
,
501 &dev_attr_max_write_same_blocks
.attr
,
502 &dev_attr_max_medium_access_timeouts
.attr
,
505 ATTRIBUTE_GROUPS(sd_disk
);
507 static struct class sd_disk_class
= {
509 .owner
= THIS_MODULE
,
510 .dev_release
= scsi_disk_release
,
511 .dev_groups
= sd_disk_groups
,
514 static const struct dev_pm_ops sd_pm_ops
= {
515 .suspend
= sd_suspend_system
,
517 .poweroff
= sd_suspend_system
,
518 .restore
= sd_resume
,
519 .runtime_suspend
= sd_suspend_runtime
,
520 .runtime_resume
= sd_resume
,
523 static struct scsi_driver sd_template
= {
526 .owner
= THIS_MODULE
,
529 .shutdown
= sd_shutdown
,
533 .init_command
= sd_init_command
,
534 .uninit_command
= sd_uninit_command
,
536 .eh_action
= sd_eh_action
,
540 * Dummy kobj_map->probe function.
541 * The default ->probe function will call modprobe, which is
542 * pointless as this module is already loaded.
544 static struct kobject
*sd_default_probe(dev_t devt
, int *partno
, void *data
)
550 * Device no to disk mapping:
552 * major disc2 disc p1
553 * |............|.............|....|....| <- dev_t
556 * Inside a major, we have 16k disks, however mapped non-
557 * contiguously. The first 16 disks are for major0, the next
558 * ones with major1, ... Disk 256 is for major0 again, disk 272
560 * As we stay compatible with our numbering scheme, we can reuse
561 * the well-know SCSI majors 8, 65--71, 136--143.
563 static int sd_major(int major_idx
)
567 return SCSI_DISK0_MAJOR
;
569 return SCSI_DISK1_MAJOR
+ major_idx
- 1;
571 return SCSI_DISK8_MAJOR
+ major_idx
- 8;
574 return 0; /* shut up gcc */
578 static struct scsi_disk
*scsi_disk_get(struct gendisk
*disk
)
580 struct scsi_disk
*sdkp
= NULL
;
582 mutex_lock(&sd_ref_mutex
);
584 if (disk
->private_data
) {
585 sdkp
= scsi_disk(disk
);
586 if (scsi_device_get(sdkp
->device
) == 0)
587 get_device(&sdkp
->dev
);
591 mutex_unlock(&sd_ref_mutex
);
595 static void scsi_disk_put(struct scsi_disk
*sdkp
)
597 struct scsi_device
*sdev
= sdkp
->device
;
599 mutex_lock(&sd_ref_mutex
);
600 put_device(&sdkp
->dev
);
601 scsi_device_put(sdev
);
602 mutex_unlock(&sd_ref_mutex
);
605 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd
*scmd
,
606 unsigned int dix
, unsigned int dif
)
608 struct bio
*bio
= scmd
->request
->bio
;
609 unsigned int prot_op
= sd_prot_op(rq_data_dir(scmd
->request
), dix
, dif
);
610 unsigned int protect
= 0;
612 if (dix
) { /* DIX Type 0, 1, 2, 3 */
613 if (bio_integrity_flagged(bio
, BIP_IP_CHECKSUM
))
614 scmd
->prot_flags
|= SCSI_PROT_IP_CHECKSUM
;
616 if (bio_integrity_flagged(bio
, BIP_CTRL_NOCHECK
) == false)
617 scmd
->prot_flags
|= SCSI_PROT_GUARD_CHECK
;
620 if (dif
!= T10_PI_TYPE3_PROTECTION
) { /* DIX/DIF Type 0, 1, 2 */
621 scmd
->prot_flags
|= SCSI_PROT_REF_INCREMENT
;
623 if (bio_integrity_flagged(bio
, BIP_CTRL_NOCHECK
) == false)
624 scmd
->prot_flags
|= SCSI_PROT_REF_CHECK
;
627 if (dif
) { /* DIX/DIF Type 1, 2, 3 */
628 scmd
->prot_flags
|= SCSI_PROT_TRANSFER_PI
;
630 if (bio_integrity_flagged(bio
, BIP_DISK_NOCHECK
))
631 protect
= 3 << 5; /* Disable target PI checking */
633 protect
= 1 << 5; /* Enable target PI checking */
636 scsi_set_prot_op(scmd
, prot_op
);
637 scsi_set_prot_type(scmd
, dif
);
638 scmd
->prot_flags
&= sd_prot_flag_mask(prot_op
);
643 static void sd_config_discard(struct scsi_disk
*sdkp
, unsigned int mode
)
645 struct request_queue
*q
= sdkp
->disk
->queue
;
646 unsigned int logical_block_size
= sdkp
->device
->sector_size
;
647 unsigned int max_blocks
= 0;
649 q
->limits
.discard_zeroes_data
= 0;
652 * When LBPRZ is reported, discard alignment and granularity
653 * must be fixed to the logical block size. Otherwise the block
654 * layer will drop misaligned portions of the request which can
655 * lead to data corruption. If LBPRZ is not set, we honor the
659 q
->limits
.discard_alignment
= 0;
660 q
->limits
.discard_granularity
= logical_block_size
;
662 q
->limits
.discard_alignment
= sdkp
->unmap_alignment
*
664 q
->limits
.discard_granularity
=
665 max(sdkp
->physical_block_size
,
666 sdkp
->unmap_granularity
* logical_block_size
);
669 sdkp
->provisioning_mode
= mode
;
674 blk_queue_max_discard_sectors(q
, 0);
675 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
679 max_blocks
= min_not_zero(sdkp
->max_unmap_blocks
,
680 (u32
)SD_MAX_WS16_BLOCKS
);
684 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
685 (u32
)SD_MAX_WS16_BLOCKS
);
686 q
->limits
.discard_zeroes_data
= sdkp
->lbprz
;
690 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
691 (u32
)SD_MAX_WS10_BLOCKS
);
692 q
->limits
.discard_zeroes_data
= sdkp
->lbprz
;
696 max_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
697 (u32
)SD_MAX_WS10_BLOCKS
);
698 q
->limits
.discard_zeroes_data
= 1;
702 blk_queue_max_discard_sectors(q
, max_blocks
* (logical_block_size
>> 9));
703 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
707 * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
708 * @sdp: scsi device to operate one
709 * @rq: Request to prepare
711 * Will issue either UNMAP or WRITE SAME(16) depending on preference
712 * indicated by target device.
714 static int sd_setup_discard_cmnd(struct scsi_cmnd
*cmd
)
716 struct request
*rq
= cmd
->request
;
717 struct scsi_device
*sdp
= cmd
->device
;
718 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
719 sector_t sector
= blk_rq_pos(rq
);
720 unsigned int nr_sectors
= blk_rq_sectors(rq
);
721 unsigned int nr_bytes
= blk_rq_bytes(rq
);
727 sector
>>= ilog2(sdp
->sector_size
) - 9;
728 nr_sectors
>>= ilog2(sdp
->sector_size
) - 9;
730 page
= alloc_page(GFP_ATOMIC
| __GFP_ZERO
);
732 return BLKPREP_DEFER
;
734 switch (sdkp
->provisioning_mode
) {
736 buf
= page_address(page
);
739 cmd
->cmnd
[0] = UNMAP
;
742 put_unaligned_be16(6 + 16, &buf
[0]);
743 put_unaligned_be16(16, &buf
[2]);
744 put_unaligned_be64(sector
, &buf
[8]);
745 put_unaligned_be32(nr_sectors
, &buf
[16]);
752 cmd
->cmnd
[0] = WRITE_SAME_16
;
753 cmd
->cmnd
[1] = 0x8; /* UNMAP */
754 put_unaligned_be64(sector
, &cmd
->cmnd
[2]);
755 put_unaligned_be32(nr_sectors
, &cmd
->cmnd
[10]);
757 len
= sdkp
->device
->sector_size
;
763 cmd
->cmnd
[0] = WRITE_SAME
;
764 if (sdkp
->provisioning_mode
== SD_LBP_WS10
)
765 cmd
->cmnd
[1] = 0x8; /* UNMAP */
766 put_unaligned_be32(sector
, &cmd
->cmnd
[2]);
767 put_unaligned_be16(nr_sectors
, &cmd
->cmnd
[7]);
769 len
= sdkp
->device
->sector_size
;
773 ret
= BLKPREP_INVALID
;
777 rq
->completion_data
= page
;
778 rq
->timeout
= SD_TIMEOUT
;
780 cmd
->transfersize
= len
;
781 cmd
->allowed
= SD_MAX_RETRIES
;
784 * Initially __data_len is set to the amount of data that needs to be
785 * transferred to the target. This amount depends on whether WRITE SAME
786 * or UNMAP is being used. After the scatterlist has been mapped by
787 * scsi_init_io() we set __data_len to the size of the area to be
788 * discarded on disk. This allows us to report completion on the full
789 * amount of blocks described by the request.
791 blk_add_request_payload(rq
, page
, 0, len
);
792 ret
= scsi_init_io(cmd
);
793 rq
->__data_len
= nr_bytes
;
796 if (ret
!= BLKPREP_OK
)
801 static void sd_config_write_same(struct scsi_disk
*sdkp
)
803 struct request_queue
*q
= sdkp
->disk
->queue
;
804 unsigned int logical_block_size
= sdkp
->device
->sector_size
;
806 if (sdkp
->device
->no_write_same
) {
807 sdkp
->max_ws_blocks
= 0;
811 /* Some devices can not handle block counts above 0xffff despite
812 * supporting WRITE SAME(16). Consequently we default to 64k
813 * blocks per I/O unless the device explicitly advertises a
816 if (sdkp
->max_ws_blocks
> SD_MAX_WS10_BLOCKS
)
817 sdkp
->max_ws_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
818 (u32
)SD_MAX_WS16_BLOCKS
);
819 else if (sdkp
->ws16
|| sdkp
->ws10
|| sdkp
->device
->no_report_opcodes
)
820 sdkp
->max_ws_blocks
= min_not_zero(sdkp
->max_ws_blocks
,
821 (u32
)SD_MAX_WS10_BLOCKS
);
823 sdkp
->device
->no_write_same
= 1;
824 sdkp
->max_ws_blocks
= 0;
828 blk_queue_max_write_same_sectors(q
, sdkp
->max_ws_blocks
*
829 (logical_block_size
>> 9));
833 * sd_setup_write_same_cmnd - write the same data to multiple blocks
834 * @cmd: command to prepare
836 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
837 * preference indicated by target device.
839 static int sd_setup_write_same_cmnd(struct scsi_cmnd
*cmd
)
841 struct request
*rq
= cmd
->request
;
842 struct scsi_device
*sdp
= cmd
->device
;
843 struct scsi_disk
*sdkp
= scsi_disk(rq
->rq_disk
);
844 struct bio
*bio
= rq
->bio
;
845 sector_t sector
= blk_rq_pos(rq
);
846 unsigned int nr_sectors
= blk_rq_sectors(rq
);
847 unsigned int nr_bytes
= blk_rq_bytes(rq
);
850 if (sdkp
->device
->no_write_same
)
851 return BLKPREP_INVALID
;
853 BUG_ON(bio_offset(bio
) || bio_iovec(bio
).bv_len
!= sdp
->sector_size
);
855 sector
>>= ilog2(sdp
->sector_size
) - 9;
856 nr_sectors
>>= ilog2(sdp
->sector_size
) - 9;
858 rq
->timeout
= SD_WRITE_SAME_TIMEOUT
;
860 if (sdkp
->ws16
|| sector
> 0xffffffff || nr_sectors
> 0xffff) {
862 cmd
->cmnd
[0] = WRITE_SAME_16
;
863 put_unaligned_be64(sector
, &cmd
->cmnd
[2]);
864 put_unaligned_be32(nr_sectors
, &cmd
->cmnd
[10]);
867 cmd
->cmnd
[0] = WRITE_SAME
;
868 put_unaligned_be32(sector
, &cmd
->cmnd
[2]);
869 put_unaligned_be16(nr_sectors
, &cmd
->cmnd
[7]);
872 cmd
->transfersize
= sdp
->sector_size
;
873 cmd
->allowed
= SD_MAX_RETRIES
;
876 * For WRITE_SAME the data transferred in the DATA IN buffer is
877 * different from the amount of data actually written to the target.
879 * We set up __data_len to the amount of data transferred from the
880 * DATA IN buffer so that blk_rq_map_sg set up the proper S/G list
881 * to transfer a single sector of data first, but then reset it to
882 * the amount of data to be written right after so that the I/O path
883 * knows how much to actually write.
885 rq
->__data_len
= sdp
->sector_size
;
886 ret
= scsi_init_io(cmd
);
887 rq
->__data_len
= nr_bytes
;
891 static int sd_setup_flush_cmnd(struct scsi_cmnd
*cmd
)
893 struct request
*rq
= cmd
->request
;
895 /* flush requests don't perform I/O, zero the S/G table */
896 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
898 cmd
->cmnd
[0] = SYNCHRONIZE_CACHE
;
900 cmd
->transfersize
= 0;
901 cmd
->allowed
= SD_MAX_RETRIES
;
903 rq
->timeout
= rq
->q
->rq_timeout
* SD_FLUSH_TIMEOUT_MULTIPLIER
;
907 static int sd_setup_read_write_cmnd(struct scsi_cmnd
*SCpnt
)
909 struct request
*rq
= SCpnt
->request
;
910 struct scsi_device
*sdp
= SCpnt
->device
;
911 struct gendisk
*disk
= rq
->rq_disk
;
912 struct scsi_disk
*sdkp
;
913 sector_t block
= blk_rq_pos(rq
);
915 unsigned int this_count
= blk_rq_sectors(rq
);
916 unsigned int dif
, dix
;
918 unsigned char protect
;
920 ret
= scsi_init_io(SCpnt
);
921 if (ret
!= BLKPREP_OK
)
924 sdkp
= scsi_disk(disk
);
926 /* from here on until we're complete, any goto out
927 * is used for a killable error condition */
931 scmd_printk(KERN_INFO
, SCpnt
,
932 "%s: block=%llu, count=%d\n",
933 __func__
, (unsigned long long)block
, this_count
));
935 if (!sdp
|| !scsi_device_online(sdp
) ||
936 block
+ blk_rq_sectors(rq
) > get_capacity(disk
)) {
937 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
,
938 "Finishing %u sectors\n",
939 blk_rq_sectors(rq
)));
940 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
,
941 "Retry with 0x%p\n", SCpnt
));
947 * quietly refuse to do anything to a changed disc until
948 * the changed bit has been reset
950 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
955 * Some SD card readers can't handle multi-sector accesses which touch
956 * the last one or two hardware sectors. Split accesses as needed.
958 threshold
= get_capacity(disk
) - SD_LAST_BUGGY_SECTORS
*
959 (sdp
->sector_size
/ 512);
961 if (unlikely(sdp
->last_sector_bug
&& block
+ this_count
> threshold
)) {
962 if (block
< threshold
) {
963 /* Access up to the threshold but not beyond */
964 this_count
= threshold
- block
;
966 /* Access only a single hardware sector */
967 this_count
= sdp
->sector_size
/ 512;
971 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
, "block=%llu\n",
972 (unsigned long long)block
));
975 * If we have a 1K hardware sectorsize, prevent access to single
976 * 512 byte sectors. In theory we could handle this - in fact
977 * the scsi cdrom driver must be able to handle this because
978 * we typically use 1K blocksizes, and cdroms typically have
979 * 2K hardware sectorsizes. Of course, things are simpler
980 * with the cdrom, since it is read-only. For performance
981 * reasons, the filesystems should be able to handle this
982 * and not force the scsi disk driver to use bounce buffers
985 if (sdp
->sector_size
== 1024) {
986 if ((block
& 1) || (blk_rq_sectors(rq
) & 1)) {
987 scmd_printk(KERN_ERR
, SCpnt
,
988 "Bad block number requested\n");
992 this_count
= this_count
>> 1;
995 if (sdp
->sector_size
== 2048) {
996 if ((block
& 3) || (blk_rq_sectors(rq
) & 3)) {
997 scmd_printk(KERN_ERR
, SCpnt
,
998 "Bad block number requested\n");
1002 this_count
= this_count
>> 2;
1005 if (sdp
->sector_size
== 4096) {
1006 if ((block
& 7) || (blk_rq_sectors(rq
) & 7)) {
1007 scmd_printk(KERN_ERR
, SCpnt
,
1008 "Bad block number requested\n");
1012 this_count
= this_count
>> 3;
1015 if (rq_data_dir(rq
) == WRITE
) {
1016 SCpnt
->cmnd
[0] = WRITE_6
;
1018 if (blk_integrity_rq(rq
))
1019 sd_dif_prepare(SCpnt
);
1021 } else if (rq_data_dir(rq
) == READ
) {
1022 SCpnt
->cmnd
[0] = READ_6
;
1024 scmd_printk(KERN_ERR
, SCpnt
, "Unknown command %llu,%llx\n",
1025 req_op(rq
), (unsigned long long) rq
->cmd_flags
);
1029 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO
, SCpnt
,
1030 "%s %d/%u 512 byte blocks.\n",
1031 (rq_data_dir(rq
) == WRITE
) ?
1032 "writing" : "reading", this_count
,
1033 blk_rq_sectors(rq
)));
1035 dix
= scsi_prot_sg_count(SCpnt
);
1036 dif
= scsi_host_dif_capable(SCpnt
->device
->host
, sdkp
->protection_type
);
1039 protect
= sd_setup_protect_cmnd(SCpnt
, dix
, dif
);
1043 if (protect
&& sdkp
->protection_type
== T10_PI_TYPE2_PROTECTION
) {
1044 SCpnt
->cmnd
= mempool_alloc(sd_cdb_pool
, GFP_ATOMIC
);
1046 if (unlikely(SCpnt
->cmnd
== NULL
)) {
1047 ret
= BLKPREP_DEFER
;
1051 SCpnt
->cmd_len
= SD_EXT_CDB_SIZE
;
1052 memset(SCpnt
->cmnd
, 0, SCpnt
->cmd_len
);
1053 SCpnt
->cmnd
[0] = VARIABLE_LENGTH_CMD
;
1054 SCpnt
->cmnd
[7] = 0x18;
1055 SCpnt
->cmnd
[9] = (rq_data_dir(rq
) == READ
) ? READ_32
: WRITE_32
;
1056 SCpnt
->cmnd
[10] = protect
| ((rq
->cmd_flags
& REQ_FUA
) ? 0x8 : 0);
1059 SCpnt
->cmnd
[12] = sizeof(block
) > 4 ? (unsigned char) (block
>> 56) & 0xff : 0;
1060 SCpnt
->cmnd
[13] = sizeof(block
) > 4 ? (unsigned char) (block
>> 48) & 0xff : 0;
1061 SCpnt
->cmnd
[14] = sizeof(block
) > 4 ? (unsigned char) (block
>> 40) & 0xff : 0;
1062 SCpnt
->cmnd
[15] = sizeof(block
) > 4 ? (unsigned char) (block
>> 32) & 0xff : 0;
1063 SCpnt
->cmnd
[16] = (unsigned char) (block
>> 24) & 0xff;
1064 SCpnt
->cmnd
[17] = (unsigned char) (block
>> 16) & 0xff;
1065 SCpnt
->cmnd
[18] = (unsigned char) (block
>> 8) & 0xff;
1066 SCpnt
->cmnd
[19] = (unsigned char) block
& 0xff;
1068 /* Expected Indirect LBA */
1069 SCpnt
->cmnd
[20] = (unsigned char) (block
>> 24) & 0xff;
1070 SCpnt
->cmnd
[21] = (unsigned char) (block
>> 16) & 0xff;
1071 SCpnt
->cmnd
[22] = (unsigned char) (block
>> 8) & 0xff;
1072 SCpnt
->cmnd
[23] = (unsigned char) block
& 0xff;
1074 /* Transfer length */
1075 SCpnt
->cmnd
[28] = (unsigned char) (this_count
>> 24) & 0xff;
1076 SCpnt
->cmnd
[29] = (unsigned char) (this_count
>> 16) & 0xff;
1077 SCpnt
->cmnd
[30] = (unsigned char) (this_count
>> 8) & 0xff;
1078 SCpnt
->cmnd
[31] = (unsigned char) this_count
& 0xff;
1079 } else if (sdp
->use_16_for_rw
|| (this_count
> 0xffff)) {
1080 SCpnt
->cmnd
[0] += READ_16
- READ_6
;
1081 SCpnt
->cmnd
[1] = protect
| ((rq
->cmd_flags
& REQ_FUA
) ? 0x8 : 0);
1082 SCpnt
->cmnd
[2] = sizeof(block
) > 4 ? (unsigned char) (block
>> 56) & 0xff : 0;
1083 SCpnt
->cmnd
[3] = sizeof(block
) > 4 ? (unsigned char) (block
>> 48) & 0xff : 0;
1084 SCpnt
->cmnd
[4] = sizeof(block
) > 4 ? (unsigned char) (block
>> 40) & 0xff : 0;
1085 SCpnt
->cmnd
[5] = sizeof(block
) > 4 ? (unsigned char) (block
>> 32) & 0xff : 0;
1086 SCpnt
->cmnd
[6] = (unsigned char) (block
>> 24) & 0xff;
1087 SCpnt
->cmnd
[7] = (unsigned char) (block
>> 16) & 0xff;
1088 SCpnt
->cmnd
[8] = (unsigned char) (block
>> 8) & 0xff;
1089 SCpnt
->cmnd
[9] = (unsigned char) block
& 0xff;
1090 SCpnt
->cmnd
[10] = (unsigned char) (this_count
>> 24) & 0xff;
1091 SCpnt
->cmnd
[11] = (unsigned char) (this_count
>> 16) & 0xff;
1092 SCpnt
->cmnd
[12] = (unsigned char) (this_count
>> 8) & 0xff;
1093 SCpnt
->cmnd
[13] = (unsigned char) this_count
& 0xff;
1094 SCpnt
->cmnd
[14] = SCpnt
->cmnd
[15] = 0;
1095 } else if ((this_count
> 0xff) || (block
> 0x1fffff) ||
1096 scsi_device_protection(SCpnt
->device
) ||
1097 SCpnt
->device
->use_10_for_rw
) {
1098 SCpnt
->cmnd
[0] += READ_10
- READ_6
;
1099 SCpnt
->cmnd
[1] = protect
| ((rq
->cmd_flags
& REQ_FUA
) ? 0x8 : 0);
1100 SCpnt
->cmnd
[2] = (unsigned char) (block
>> 24) & 0xff;
1101 SCpnt
->cmnd
[3] = (unsigned char) (block
>> 16) & 0xff;
1102 SCpnt
->cmnd
[4] = (unsigned char) (block
>> 8) & 0xff;
1103 SCpnt
->cmnd
[5] = (unsigned char) block
& 0xff;
1104 SCpnt
->cmnd
[6] = SCpnt
->cmnd
[9] = 0;
1105 SCpnt
->cmnd
[7] = (unsigned char) (this_count
>> 8) & 0xff;
1106 SCpnt
->cmnd
[8] = (unsigned char) this_count
& 0xff;
1108 if (unlikely(rq
->cmd_flags
& REQ_FUA
)) {
1110 * This happens only if this drive failed
1111 * 10byte rw command with ILLEGAL_REQUEST
1112 * during operation and thus turned off
1115 scmd_printk(KERN_ERR
, SCpnt
,
1116 "FUA write on READ/WRITE(6) drive\n");
1120 SCpnt
->cmnd
[1] |= (unsigned char) ((block
>> 16) & 0x1f);
1121 SCpnt
->cmnd
[2] = (unsigned char) ((block
>> 8) & 0xff);
1122 SCpnt
->cmnd
[3] = (unsigned char) block
& 0xff;
1123 SCpnt
->cmnd
[4] = (unsigned char) this_count
;
1126 SCpnt
->sdb
.length
= this_count
* sdp
->sector_size
;
1129 * We shouldn't disconnect in the middle of a sector, so with a dumb
1130 * host adapter, it's safe to assume that we can at least transfer
1131 * this many bytes between each connect / disconnect.
1133 SCpnt
->transfersize
= sdp
->sector_size
;
1134 SCpnt
->underflow
= this_count
<< 9;
1135 SCpnt
->allowed
= SD_MAX_RETRIES
;
1138 * This indicates that the command is ready from our end to be
1146 static int sd_init_command(struct scsi_cmnd
*cmd
)
1148 struct request
*rq
= cmd
->request
;
1150 switch (req_op(rq
)) {
1151 case REQ_OP_DISCARD
:
1152 return sd_setup_discard_cmnd(cmd
);
1153 case REQ_OP_WRITE_SAME
:
1154 return sd_setup_write_same_cmnd(cmd
);
1156 return sd_setup_flush_cmnd(cmd
);
1159 return sd_setup_read_write_cmnd(cmd
);
1165 static void sd_uninit_command(struct scsi_cmnd
*SCpnt
)
1167 struct request
*rq
= SCpnt
->request
;
1169 if (req_op(rq
) == REQ_OP_DISCARD
)
1170 __free_page(rq
->completion_data
);
1172 if (SCpnt
->cmnd
!= rq
->cmd
) {
1173 mempool_free(SCpnt
->cmnd
, sd_cdb_pool
);
1180 * sd_open - open a scsi disk device
1181 * @inode: only i_rdev member may be used
1182 * @filp: only f_mode and f_flags may be used
1184 * Returns 0 if successful. Returns a negated errno value in case
1187 * Note: This can be called from a user context (e.g. fsck(1) )
1188 * or from within the kernel (e.g. as a result of a mount(1) ).
1189 * In the latter case @inode and @filp carry an abridged amount
1190 * of information as noted above.
1192 * Locking: called with bdev->bd_mutex held.
1194 static int sd_open(struct block_device
*bdev
, fmode_t mode
)
1196 struct scsi_disk
*sdkp
= scsi_disk_get(bdev
->bd_disk
);
1197 struct scsi_device
*sdev
;
1203 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_open\n"));
1205 sdev
= sdkp
->device
;
1208 * If the device is in error recovery, wait until it is done.
1209 * If the device is offline, then disallow any access to it.
1212 if (!scsi_block_when_processing_errors(sdev
))
1215 if (sdev
->removable
|| sdkp
->write_prot
)
1216 check_disk_change(bdev
);
1219 * If the drive is empty, just let the open fail.
1221 retval
= -ENOMEDIUM
;
1222 if (sdev
->removable
&& !sdkp
->media_present
&& !(mode
& FMODE_NDELAY
))
1226 * If the device has the write protect tab set, have the open fail
1227 * if the user expects to be able to write to the thing.
1230 if (sdkp
->write_prot
&& (mode
& FMODE_WRITE
))
1234 * It is possible that the disk changing stuff resulted in
1235 * the device being taken offline. If this is the case,
1236 * report this to the user, and don't pretend that the
1237 * open actually succeeded.
1240 if (!scsi_device_online(sdev
))
1243 if ((atomic_inc_return(&sdkp
->openers
) == 1) && sdev
->removable
) {
1244 if (scsi_block_when_processing_errors(sdev
))
1245 scsi_set_medium_removal(sdev
, SCSI_REMOVAL_PREVENT
);
1251 scsi_disk_put(sdkp
);
1256 * sd_release - invoked when the (last) close(2) is called on this
1258 * @inode: only i_rdev member may be used
1259 * @filp: only f_mode and f_flags may be used
1263 * Note: may block (uninterruptible) if error recovery is underway
1266 * Locking: called with bdev->bd_mutex held.
1268 static void sd_release(struct gendisk
*disk
, fmode_t mode
)
1270 struct scsi_disk
*sdkp
= scsi_disk(disk
);
1271 struct scsi_device
*sdev
= sdkp
->device
;
1273 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_release\n"));
1275 if (atomic_dec_return(&sdkp
->openers
) == 0 && sdev
->removable
) {
1276 if (scsi_block_when_processing_errors(sdev
))
1277 scsi_set_medium_removal(sdev
, SCSI_REMOVAL_ALLOW
);
1281 * XXX and what if there are packets in flight and this close()
1282 * XXX is followed by a "rmmod sd_mod"?
1285 scsi_disk_put(sdkp
);
1288 static int sd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1290 struct scsi_disk
*sdkp
= scsi_disk(bdev
->bd_disk
);
1291 struct scsi_device
*sdp
= sdkp
->device
;
1292 struct Scsi_Host
*host
= sdp
->host
;
1293 sector_t capacity
= logical_to_sectors(sdp
, sdkp
->capacity
);
1296 /* default to most commonly used values */
1297 diskinfo
[0] = 0x40; /* 1 << 6 */
1298 diskinfo
[1] = 0x20; /* 1 << 5 */
1299 diskinfo
[2] = capacity
>> 11;
1301 /* override with calculated, extended default, or driver values */
1302 if (host
->hostt
->bios_param
)
1303 host
->hostt
->bios_param(sdp
, bdev
, capacity
, diskinfo
);
1305 scsicam_bios_param(bdev
, capacity
, diskinfo
);
1307 geo
->heads
= diskinfo
[0];
1308 geo
->sectors
= diskinfo
[1];
1309 geo
->cylinders
= diskinfo
[2];
1314 * sd_ioctl - process an ioctl
1315 * @inode: only i_rdev/i_bdev members may be used
1316 * @filp: only f_mode and f_flags may be used
1317 * @cmd: ioctl command number
1318 * @arg: this is third argument given to ioctl(2) system call.
1319 * Often contains a pointer.
1321 * Returns 0 if successful (some ioctls return positive numbers on
1322 * success as well). Returns a negated errno value in case of error.
1324 * Note: most ioctls are forward onto the block subsystem or further
1325 * down in the scsi subsystem.
1327 static int sd_ioctl(struct block_device
*bdev
, fmode_t mode
,
1328 unsigned int cmd
, unsigned long arg
)
1330 struct gendisk
*disk
= bdev
->bd_disk
;
1331 struct scsi_disk
*sdkp
= scsi_disk(disk
);
1332 struct scsi_device
*sdp
= sdkp
->device
;
1333 void __user
*p
= (void __user
*)arg
;
1336 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO
, sdkp
, "sd_ioctl: disk=%s, "
1337 "cmd=0x%x\n", disk
->disk_name
, cmd
));
1339 error
= scsi_verify_blk_ioctl(bdev
, cmd
);
1344 * If we are in the middle of error recovery, don't let anyone
1345 * else try and use this device. Also, if error recovery fails, it
1346 * may try and take the device offline, in which case all further
1347 * access to the device is prohibited.
1349 error
= scsi_ioctl_block_when_processing_errors(sdp
, cmd
,
1350 (mode
& FMODE_NDELAY
) != 0);
1355 * Send SCSI addressing ioctls directly to mid level, send other
1356 * ioctls to block level and then onto mid level if they can't be
1360 case SCSI_IOCTL_GET_IDLUN
:
1361 case SCSI_IOCTL_GET_BUS_NUMBER
:
1362 error
= scsi_ioctl(sdp
, cmd
, p
);
1365 error
= scsi_cmd_blk_ioctl(bdev
, mode
, cmd
, p
);
1366 if (error
!= -ENOTTY
)
1368 error
= scsi_ioctl(sdp
, cmd
, p
);
1375 static void set_media_not_present(struct scsi_disk
*sdkp
)
1377 if (sdkp
->media_present
)
1378 sdkp
->device
->changed
= 1;
1380 if (sdkp
->device
->removable
) {
1381 sdkp
->media_present
= 0;
1386 static int media_not_present(struct scsi_disk
*sdkp
,
1387 struct scsi_sense_hdr
*sshdr
)
1389 if (!scsi_sense_valid(sshdr
))
1392 /* not invoked for commands that could return deferred errors */
1393 switch (sshdr
->sense_key
) {
1394 case UNIT_ATTENTION
:
1396 /* medium not present */
1397 if (sshdr
->asc
== 0x3A) {
1398 set_media_not_present(sdkp
);
1406 * sd_check_events - check media events
1407 * @disk: kernel device descriptor
1408 * @clearing: disk events currently being cleared
1410 * Returns mask of DISK_EVENT_*.
1412 * Note: this function is invoked from the block subsystem.
1414 static unsigned int sd_check_events(struct gendisk
*disk
, unsigned int clearing
)
1416 struct scsi_disk
*sdkp
= scsi_disk_get(disk
);
1417 struct scsi_device
*sdp
;
1418 struct scsi_sense_hdr
*sshdr
= NULL
;
1425 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
, "sd_check_events\n"));
1428 * If the device is offline, don't send any commands - just pretend as
1429 * if the command failed. If the device ever comes back online, we
1430 * can deal with it then. It is only because of unrecoverable errors
1431 * that we would ever take a device offline in the first place.
1433 if (!scsi_device_online(sdp
)) {
1434 set_media_not_present(sdkp
);
1439 * Using TEST_UNIT_READY enables differentiation between drive with
1440 * no cartridge loaded - NOT READY, drive with changed cartridge -
1441 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1443 * Drives that auto spin down. eg iomega jaz 1G, will be started
1444 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1445 * sd_revalidate() is called.
1449 if (scsi_block_when_processing_errors(sdp
)) {
1450 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1451 retval
= scsi_test_unit_ready(sdp
, SD_TIMEOUT
, SD_MAX_RETRIES
,
1455 /* failed to execute TUR, assume media not present */
1456 if (host_byte(retval
)) {
1457 set_media_not_present(sdkp
);
1461 if (media_not_present(sdkp
, sshdr
))
1465 * For removable scsi disk we have to recognise the presence
1466 * of a disk in the drive.
1468 if (!sdkp
->media_present
)
1470 sdkp
->media_present
= 1;
1473 * sdp->changed is set under the following conditions:
1475 * Medium present state has changed in either direction.
1476 * Device has indicated UNIT_ATTENTION.
1479 retval
= sdp
->changed
? DISK_EVENT_MEDIA_CHANGE
: 0;
1481 scsi_disk_put(sdkp
);
1485 static int sd_sync_cache(struct scsi_disk
*sdkp
)
1488 struct scsi_device
*sdp
= sdkp
->device
;
1489 const int timeout
= sdp
->request_queue
->rq_timeout
1490 * SD_FLUSH_TIMEOUT_MULTIPLIER
;
1491 struct scsi_sense_hdr sshdr
;
1493 if (!scsi_device_online(sdp
))
1496 for (retries
= 3; retries
> 0; --retries
) {
1497 unsigned char cmd
[10] = { 0 };
1499 cmd
[0] = SYNCHRONIZE_CACHE
;
1501 * Leave the rest of the command zero to indicate
1504 res
= scsi_execute_req_flags(sdp
, cmd
, DMA_NONE
, NULL
, 0,
1505 &sshdr
, timeout
, SD_MAX_RETRIES
,
1512 sd_print_result(sdkp
, "Synchronize Cache(10) failed", res
);
1514 if (driver_byte(res
) & DRIVER_SENSE
)
1515 sd_print_sense_hdr(sdkp
, &sshdr
);
1516 /* we need to evaluate the error return */
1517 if (scsi_sense_valid(&sshdr
) &&
1518 (sshdr
.asc
== 0x3a || /* medium not present */
1519 sshdr
.asc
== 0x20)) /* invalid command */
1520 /* this is no error here */
1523 switch (host_byte(res
)) {
1524 /* ignore errors due to racing a disconnection */
1525 case DID_BAD_TARGET
:
1526 case DID_NO_CONNECT
:
1528 /* signal the upper layer it might try again */
1532 case DID_SOFT_ERROR
:
1541 static void sd_rescan(struct device
*dev
)
1543 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
1545 revalidate_disk(sdkp
->disk
);
1549 #ifdef CONFIG_COMPAT
1551 * This gets directly called from VFS. When the ioctl
1552 * is not recognized we go back to the other translation paths.
1554 static int sd_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1555 unsigned int cmd
, unsigned long arg
)
1557 struct scsi_device
*sdev
= scsi_disk(bdev
->bd_disk
)->device
;
1560 error
= scsi_ioctl_block_when_processing_errors(sdev
, cmd
,
1561 (mode
& FMODE_NDELAY
) != 0);
1566 * Let the static ioctl translation table take care of it.
1568 if (!sdev
->host
->hostt
->compat_ioctl
)
1569 return -ENOIOCTLCMD
;
1570 return sdev
->host
->hostt
->compat_ioctl(sdev
, cmd
, (void __user
*)arg
);
1574 static char sd_pr_type(enum pr_type type
)
1577 case PR_WRITE_EXCLUSIVE
:
1579 case PR_EXCLUSIVE_ACCESS
:
1581 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
1583 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
1585 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
1587 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
1594 static int sd_pr_command(struct block_device
*bdev
, u8 sa
,
1595 u64 key
, u64 sa_key
, u8 type
, u8 flags
)
1597 struct scsi_device
*sdev
= scsi_disk(bdev
->bd_disk
)->device
;
1598 struct scsi_sense_hdr sshdr
;
1600 u8 cmd
[16] = { 0, };
1601 u8 data
[24] = { 0, };
1603 cmd
[0] = PERSISTENT_RESERVE_OUT
;
1606 put_unaligned_be32(sizeof(data
), &cmd
[5]);
1608 put_unaligned_be64(key
, &data
[0]);
1609 put_unaligned_be64(sa_key
, &data
[8]);
1612 result
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, &data
, sizeof(data
),
1613 &sshdr
, SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
);
1615 if ((driver_byte(result
) & DRIVER_SENSE
) &&
1616 (scsi_sense_valid(&sshdr
))) {
1617 sdev_printk(KERN_INFO
, sdev
, "PR command failed: %d\n", result
);
1618 scsi_print_sense_hdr(sdev
, NULL
, &sshdr
);
1624 static int sd_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
1627 if (flags
& ~PR_FL_IGNORE_KEY
)
1629 return sd_pr_command(bdev
, (flags
& PR_FL_IGNORE_KEY
) ? 0x06 : 0x00,
1630 old_key
, new_key
, 0,
1631 (1 << 0) /* APTPL */);
1634 static int sd_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
1639 return sd_pr_command(bdev
, 0x01, key
, 0, sd_pr_type(type
), 0);
1642 static int sd_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
1644 return sd_pr_command(bdev
, 0x02, key
, 0, sd_pr_type(type
), 0);
1647 static int sd_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
1648 enum pr_type type
, bool abort
)
1650 return sd_pr_command(bdev
, abort
? 0x05 : 0x04, old_key
, new_key
,
1651 sd_pr_type(type
), 0);
1654 static int sd_pr_clear(struct block_device
*bdev
, u64 key
)
1656 return sd_pr_command(bdev
, 0x03, key
, 0, 0, 0);
1659 static const struct pr_ops sd_pr_ops
= {
1660 .pr_register
= sd_pr_register
,
1661 .pr_reserve
= sd_pr_reserve
,
1662 .pr_release
= sd_pr_release
,
1663 .pr_preempt
= sd_pr_preempt
,
1664 .pr_clear
= sd_pr_clear
,
1667 static const struct block_device_operations sd_fops
= {
1668 .owner
= THIS_MODULE
,
1670 .release
= sd_release
,
1672 .getgeo
= sd_getgeo
,
1673 #ifdef CONFIG_COMPAT
1674 .compat_ioctl
= sd_compat_ioctl
,
1676 .check_events
= sd_check_events
,
1677 .revalidate_disk
= sd_revalidate_disk
,
1678 .unlock_native_capacity
= sd_unlock_native_capacity
,
1679 .pr_ops
= &sd_pr_ops
,
1683 * sd_eh_action - error handling callback
1684 * @scmd: sd-issued command that has failed
1685 * @eh_disp: The recovery disposition suggested by the midlayer
1687 * This function is called by the SCSI midlayer upon completion of an
1688 * error test command (currently TEST UNIT READY). The result of sending
1689 * the eh command is passed in eh_disp. We're looking for devices that
1690 * fail medium access commands but are OK with non access commands like
1691 * test unit ready (so wrongly see the device as having a successful
1694 static int sd_eh_action(struct scsi_cmnd
*scmd
, int eh_disp
)
1696 struct scsi_disk
*sdkp
= scsi_disk(scmd
->request
->rq_disk
);
1698 if (!scsi_device_online(scmd
->device
) ||
1699 !scsi_medium_access_command(scmd
) ||
1700 host_byte(scmd
->result
) != DID_TIME_OUT
||
1705 * The device has timed out executing a medium access command.
1706 * However, the TEST UNIT READY command sent during error
1707 * handling completed successfully. Either the device is in the
1708 * process of recovering or has it suffered an internal failure
1709 * that prevents access to the storage medium.
1711 sdkp
->medium_access_timed_out
++;
1714 * If the device keeps failing read/write commands but TEST UNIT
1715 * READY always completes successfully we assume that medium
1716 * access is no longer possible and take the device offline.
1718 if (sdkp
->medium_access_timed_out
>= sdkp
->max_medium_access_timeouts
) {
1719 scmd_printk(KERN_ERR
, scmd
,
1720 "Medium access timeout failure. Offlining disk!\n");
1721 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1729 static unsigned int sd_completed_bytes(struct scsi_cmnd
*scmd
)
1731 u64 start_lba
= blk_rq_pos(scmd
->request
);
1732 u64 end_lba
= blk_rq_pos(scmd
->request
) + (scsi_bufflen(scmd
) / 512);
1733 u64 factor
= scmd
->device
->sector_size
/ 512;
1737 * resid is optional but mostly filled in. When it's unused,
1738 * its value is zero, so we assume the whole buffer transferred
1740 unsigned int transferred
= scsi_bufflen(scmd
) - scsi_get_resid(scmd
);
1741 unsigned int good_bytes
;
1743 if (scmd
->request
->cmd_type
!= REQ_TYPE_FS
)
1746 info_valid
= scsi_get_sense_info_fld(scmd
->sense_buffer
,
1747 SCSI_SENSE_BUFFERSIZE
,
1752 if (scsi_bufflen(scmd
) <= scmd
->device
->sector_size
)
1755 /* be careful ... don't want any overflows */
1756 do_div(start_lba
, factor
);
1757 do_div(end_lba
, factor
);
1759 /* The bad lba was reported incorrectly, we have no idea where
1762 if (bad_lba
< start_lba
|| bad_lba
>= end_lba
)
1765 /* This computation should always be done in terms of
1766 * the resolution of the device's medium.
1768 good_bytes
= (bad_lba
- start_lba
) * scmd
->device
->sector_size
;
1769 return min(good_bytes
, transferred
);
1773 * sd_done - bottom half handler: called when the lower level
1774 * driver has completed (successfully or otherwise) a scsi command.
1775 * @SCpnt: mid-level's per command structure.
1777 * Note: potentially run from within an ISR. Must not block.
1779 static int sd_done(struct scsi_cmnd
*SCpnt
)
1781 int result
= SCpnt
->result
;
1782 unsigned int good_bytes
= result
? 0 : scsi_bufflen(SCpnt
);
1783 struct scsi_sense_hdr sshdr
;
1784 struct scsi_disk
*sdkp
= scsi_disk(SCpnt
->request
->rq_disk
);
1785 struct request
*req
= SCpnt
->request
;
1786 int sense_valid
= 0;
1787 int sense_deferred
= 0;
1788 unsigned char op
= SCpnt
->cmnd
[0];
1789 unsigned char unmap
= SCpnt
->cmnd
[1] & 8;
1791 if (req_op(req
) == REQ_OP_DISCARD
|| req_op(req
) == REQ_OP_WRITE_SAME
) {
1793 good_bytes
= blk_rq_bytes(req
);
1794 scsi_set_resid(SCpnt
, 0);
1797 scsi_set_resid(SCpnt
, blk_rq_bytes(req
));
1802 sense_valid
= scsi_command_normalize_sense(SCpnt
, &sshdr
);
1804 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
1806 sdkp
->medium_access_timed_out
= 0;
1808 if (driver_byte(result
) != DRIVER_SENSE
&&
1809 (!sense_valid
|| sense_deferred
))
1812 switch (sshdr
.sense_key
) {
1813 case HARDWARE_ERROR
:
1815 good_bytes
= sd_completed_bytes(SCpnt
);
1817 case RECOVERED_ERROR
:
1818 good_bytes
= scsi_bufflen(SCpnt
);
1821 /* This indicates a false check condition, so ignore it. An
1822 * unknown amount of data was transferred so treat it as an
1826 memset(SCpnt
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
1828 case ABORTED_COMMAND
:
1829 if (sshdr
.asc
== 0x10) /* DIF: Target detected corruption */
1830 good_bytes
= sd_completed_bytes(SCpnt
);
1832 case ILLEGAL_REQUEST
:
1833 if (sshdr
.asc
== 0x10) /* DIX: Host detected corruption */
1834 good_bytes
= sd_completed_bytes(SCpnt
);
1835 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1836 if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
1839 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
1844 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
1846 sdkp
->device
->no_write_same
= 1;
1847 sd_config_write_same(sdkp
);
1850 req
->__data_len
= blk_rq_bytes(req
);
1851 req
->cmd_flags
|= REQ_QUIET
;
1860 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, SCpnt
,
1861 "sd_done: completed %d of %d bytes\n",
1862 good_bytes
, scsi_bufflen(SCpnt
)));
1864 if (rq_data_dir(SCpnt
->request
) == READ
&& scsi_prot_sg_count(SCpnt
))
1865 sd_dif_complete(SCpnt
, good_bytes
);
1871 * spinup disk - called only in sd_revalidate_disk()
1874 sd_spinup_disk(struct scsi_disk
*sdkp
)
1876 unsigned char cmd
[10];
1877 unsigned long spintime_expire
= 0;
1878 int retries
, spintime
;
1879 unsigned int the_result
;
1880 struct scsi_sense_hdr sshdr
;
1881 int sense_valid
= 0;
1885 /* Spin up drives, as required. Only do this at boot time */
1886 /* Spinup needs to be done for module loads too. */
1891 cmd
[0] = TEST_UNIT_READY
;
1892 memset((void *) &cmd
[1], 0, 9);
1894 the_result
= scsi_execute_req(sdkp
->device
, cmd
,
1897 SD_MAX_RETRIES
, NULL
);
1900 * If the drive has indicated to us that it
1901 * doesn't have any media in it, don't bother
1902 * with any more polling.
1904 if (media_not_present(sdkp
, &sshdr
))
1908 sense_valid
= scsi_sense_valid(&sshdr
);
1910 } while (retries
< 3 &&
1911 (!scsi_status_is_good(the_result
) ||
1912 ((driver_byte(the_result
) & DRIVER_SENSE
) &&
1913 sense_valid
&& sshdr
.sense_key
== UNIT_ATTENTION
)));
1915 if ((driver_byte(the_result
) & DRIVER_SENSE
) == 0) {
1916 /* no sense, TUR either succeeded or failed
1917 * with a status error */
1918 if(!spintime
&& !scsi_status_is_good(the_result
)) {
1919 sd_print_result(sdkp
, "Test Unit Ready failed",
1926 * The device does not want the automatic start to be issued.
1928 if (sdkp
->device
->no_start_on_add
)
1931 if (sense_valid
&& sshdr
.sense_key
== NOT_READY
) {
1932 if (sshdr
.asc
== 4 && sshdr
.ascq
== 3)
1933 break; /* manual intervention required */
1934 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0xb)
1935 break; /* standby */
1936 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0xc)
1937 break; /* unavailable */
1938 if (sshdr
.asc
== 4 && sshdr
.ascq
== 0x1b)
1939 break; /* sanitize in progress */
1941 * Issue command to spin up drive when not ready
1944 sd_printk(KERN_NOTICE
, sdkp
, "Spinning up disk...");
1945 cmd
[0] = START_STOP
;
1946 cmd
[1] = 1; /* Return immediately */
1947 memset((void *) &cmd
[2], 0, 8);
1948 cmd
[4] = 1; /* Start spin cycle */
1949 if (sdkp
->device
->start_stop_pwr_cond
)
1951 scsi_execute_req(sdkp
->device
, cmd
, DMA_NONE
,
1953 SD_TIMEOUT
, SD_MAX_RETRIES
,
1955 spintime_expire
= jiffies
+ 100 * HZ
;
1958 /* Wait 1 second for next try */
1963 * Wait for USB flash devices with slow firmware.
1964 * Yes, this sense key/ASC combination shouldn't
1965 * occur here. It's characteristic of these devices.
1967 } else if (sense_valid
&&
1968 sshdr
.sense_key
== UNIT_ATTENTION
&&
1969 sshdr
.asc
== 0x28) {
1971 spintime_expire
= jiffies
+ 5 * HZ
;
1974 /* Wait 1 second for next try */
1977 /* we don't understand the sense code, so it's
1978 * probably pointless to loop */
1980 sd_printk(KERN_NOTICE
, sdkp
, "Unit Not Ready\n");
1981 sd_print_sense_hdr(sdkp
, &sshdr
);
1986 } while (spintime
&& time_before_eq(jiffies
, spintime_expire
));
1989 if (scsi_status_is_good(the_result
))
1992 printk("not responding...\n");
1998 * Determine whether disk supports Data Integrity Field.
2000 static int sd_read_protection_type(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2002 struct scsi_device
*sdp
= sdkp
->device
;
2006 if (scsi_device_protection(sdp
) == 0 || (buffer
[12] & 1) == 0)
2009 type
= ((buffer
[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2011 if (type
> T10_PI_TYPE3_PROTECTION
)
2013 else if (scsi_host_dif_capable(sdp
->host
, type
))
2016 if (sdkp
->first_scan
|| type
!= sdkp
->protection_type
)
2019 sd_printk(KERN_ERR
, sdkp
, "formatted with unsupported" \
2020 " protection type %u. Disabling disk!\n",
2024 sd_printk(KERN_NOTICE
, sdkp
,
2025 "Enabling DIF Type %u protection\n", type
);
2028 sd_printk(KERN_NOTICE
, sdkp
,
2029 "Disabling DIF Type %u protection\n", type
);
2033 sdkp
->protection_type
= type
;
2038 static void read_capacity_error(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2039 struct scsi_sense_hdr
*sshdr
, int sense_valid
,
2042 if (driver_byte(the_result
) & DRIVER_SENSE
)
2043 sd_print_sense_hdr(sdkp
, sshdr
);
2045 sd_printk(KERN_NOTICE
, sdkp
, "Sense not available.\n");
2048 * Set dirty bit for removable devices if not ready -
2049 * sometimes drives will not report this properly.
2051 if (sdp
->removable
&&
2052 sense_valid
&& sshdr
->sense_key
== NOT_READY
)
2053 set_media_not_present(sdkp
);
2056 * We used to set media_present to 0 here to indicate no media
2057 * in the drive, but some drives fail read capacity even with
2058 * media present, so we can't do that.
2060 sdkp
->capacity
= 0; /* unknown mapped to zero - as usual */
2064 #if RC16_LEN > SD_BUF_SIZE
2065 #error RC16_LEN must not be more than SD_BUF_SIZE
2068 #define READ_CAPACITY_RETRIES_ON_RESET 10
2071 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2072 * and the reported logical block size is bigger than 512 bytes. Note
2073 * that last_sector is a u64 and therefore logical_to_sectors() is not
2076 static bool sd_addressable_capacity(u64 lba
, unsigned int sector_size
)
2078 u64 last_sector
= (lba
+ 1ULL) << (ilog2(sector_size
) - 9);
2080 if (sizeof(sector_t
) == 4 && last_sector
> U32_MAX
)
2086 static int read_capacity_16(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2087 unsigned char *buffer
)
2089 unsigned char cmd
[16];
2090 struct scsi_sense_hdr sshdr
;
2091 int sense_valid
= 0;
2093 int retries
= 3, reset_retries
= READ_CAPACITY_RETRIES_ON_RESET
;
2094 unsigned int alignment
;
2095 unsigned long long lba
;
2096 unsigned sector_size
;
2098 if (sdp
->no_read_capacity_16
)
2103 cmd
[0] = SERVICE_ACTION_IN_16
;
2104 cmd
[1] = SAI_READ_CAPACITY_16
;
2106 memset(buffer
, 0, RC16_LEN
);
2108 the_result
= scsi_execute_req(sdp
, cmd
, DMA_FROM_DEVICE
,
2109 buffer
, RC16_LEN
, &sshdr
,
2110 SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
);
2112 if (media_not_present(sdkp
, &sshdr
))
2116 sense_valid
= scsi_sense_valid(&sshdr
);
2118 sshdr
.sense_key
== ILLEGAL_REQUEST
&&
2119 (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) &&
2121 /* Invalid Command Operation Code or
2122 * Invalid Field in CDB, just retry
2123 * silently with RC10 */
2126 sshdr
.sense_key
== UNIT_ATTENTION
&&
2127 sshdr
.asc
== 0x29 && sshdr
.ascq
== 0x00)
2128 /* Device reset might occur several times,
2129 * give it one more chance */
2130 if (--reset_retries
> 0)
2135 } while (the_result
&& retries
);
2138 sd_print_result(sdkp
, "Read Capacity(16) failed", the_result
);
2139 read_capacity_error(sdkp
, sdp
, &sshdr
, sense_valid
, the_result
);
2143 sector_size
= get_unaligned_be32(&buffer
[8]);
2144 lba
= get_unaligned_be64(&buffer
[0]);
2146 if (sd_read_protection_type(sdkp
, buffer
) < 0) {
2151 if (!sd_addressable_capacity(lba
, sector_size
)) {
2152 sd_printk(KERN_ERR
, sdkp
, "Too big for this kernel. Use a "
2153 "kernel compiled with support for large block "
2159 /* Logical blocks per physical block exponent */
2160 sdkp
->physical_block_size
= (1 << (buffer
[13] & 0xf)) * sector_size
;
2162 /* Lowest aligned logical block */
2163 alignment
= ((buffer
[14] & 0x3f) << 8 | buffer
[15]) * sector_size
;
2164 blk_queue_alignment_offset(sdp
->request_queue
, alignment
);
2165 if (alignment
&& sdkp
->first_scan
)
2166 sd_printk(KERN_NOTICE
, sdkp
,
2167 "physical block alignment offset: %u\n", alignment
);
2169 if (buffer
[14] & 0x80) { /* LBPME */
2172 if (buffer
[14] & 0x40) /* LBPRZ */
2175 sd_config_discard(sdkp
, SD_LBP_WS16
);
2178 sdkp
->capacity
= lba
+ 1;
2182 static int read_capacity_10(struct scsi_disk
*sdkp
, struct scsi_device
*sdp
,
2183 unsigned char *buffer
)
2185 unsigned char cmd
[16];
2186 struct scsi_sense_hdr sshdr
;
2187 int sense_valid
= 0;
2189 int retries
= 3, reset_retries
= READ_CAPACITY_RETRIES_ON_RESET
;
2191 unsigned sector_size
;
2194 cmd
[0] = READ_CAPACITY
;
2195 memset(&cmd
[1], 0, 9);
2196 memset(buffer
, 0, 8);
2198 the_result
= scsi_execute_req(sdp
, cmd
, DMA_FROM_DEVICE
,
2200 SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
);
2202 if (media_not_present(sdkp
, &sshdr
))
2206 sense_valid
= scsi_sense_valid(&sshdr
);
2208 sshdr
.sense_key
== UNIT_ATTENTION
&&
2209 sshdr
.asc
== 0x29 && sshdr
.ascq
== 0x00)
2210 /* Device reset might occur several times,
2211 * give it one more chance */
2212 if (--reset_retries
> 0)
2217 } while (the_result
&& retries
);
2220 sd_print_result(sdkp
, "Read Capacity(10) failed", the_result
);
2221 read_capacity_error(sdkp
, sdp
, &sshdr
, sense_valid
, the_result
);
2225 sector_size
= get_unaligned_be32(&buffer
[4]);
2226 lba
= get_unaligned_be32(&buffer
[0]);
2228 if (sdp
->no_read_capacity_16
&& (lba
== 0xffffffff)) {
2229 /* Some buggy (usb cardreader) devices return an lba of
2230 0xffffffff when the want to report a size of 0 (with
2231 which they really mean no media is present) */
2233 sdkp
->physical_block_size
= sector_size
;
2237 if (!sd_addressable_capacity(lba
, sector_size
)) {
2238 sd_printk(KERN_ERR
, sdkp
, "Too big for this kernel. Use a "
2239 "kernel compiled with support for large block "
2245 sdkp
->capacity
= lba
+ 1;
2246 sdkp
->physical_block_size
= sector_size
;
2250 static int sd_try_rc16_first(struct scsi_device
*sdp
)
2252 if (sdp
->host
->max_cmd_len
< 16)
2254 if (sdp
->try_rc_10_first
)
2256 if (sdp
->scsi_level
> SCSI_SPC_2
)
2258 if (scsi_device_protection(sdp
))
2264 * read disk capacity
2267 sd_read_capacity(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2270 struct scsi_device
*sdp
= sdkp
->device
;
2271 sector_t old_capacity
= sdkp
->capacity
;
2273 if (sd_try_rc16_first(sdp
)) {
2274 sector_size
= read_capacity_16(sdkp
, sdp
, buffer
);
2275 if (sector_size
== -EOVERFLOW
)
2277 if (sector_size
== -ENODEV
)
2279 if (sector_size
< 0)
2280 sector_size
= read_capacity_10(sdkp
, sdp
, buffer
);
2281 if (sector_size
< 0)
2284 sector_size
= read_capacity_10(sdkp
, sdp
, buffer
);
2285 if (sector_size
== -EOVERFLOW
)
2287 if (sector_size
< 0)
2289 if ((sizeof(sdkp
->capacity
) > 4) &&
2290 (sdkp
->capacity
> 0xffffffffULL
)) {
2291 int old_sector_size
= sector_size
;
2292 sd_printk(KERN_NOTICE
, sdkp
, "Very big device. "
2293 "Trying to use READ CAPACITY(16).\n");
2294 sector_size
= read_capacity_16(sdkp
, sdp
, buffer
);
2295 if (sector_size
< 0) {
2296 sd_printk(KERN_NOTICE
, sdkp
,
2297 "Using 0xffffffff as device size\n");
2298 sdkp
->capacity
= 1 + (sector_t
) 0xffffffff;
2299 sector_size
= old_sector_size
;
2305 /* Some devices are known to return the total number of blocks,
2306 * not the highest block number. Some devices have versions
2307 * which do this and others which do not. Some devices we might
2308 * suspect of doing this but we don't know for certain.
2310 * If we know the reported capacity is wrong, decrement it. If
2311 * we can only guess, then assume the number of blocks is even
2312 * (usually true but not always) and err on the side of lowering
2315 if (sdp
->fix_capacity
||
2316 (sdp
->guess_capacity
&& (sdkp
->capacity
& 0x01))) {
2317 sd_printk(KERN_INFO
, sdkp
, "Adjusting the sector count "
2318 "from its reported value: %llu\n",
2319 (unsigned long long) sdkp
->capacity
);
2324 if (sector_size
== 0) {
2326 sd_printk(KERN_NOTICE
, sdkp
, "Sector size 0 reported, "
2330 if (sector_size
!= 512 &&
2331 sector_size
!= 1024 &&
2332 sector_size
!= 2048 &&
2333 sector_size
!= 4096) {
2334 sd_printk(KERN_NOTICE
, sdkp
, "Unsupported sector size %d.\n",
2337 * The user might want to re-format the drive with
2338 * a supported sectorsize. Once this happens, it
2339 * would be relatively trivial to set the thing up.
2340 * For this reason, we leave the thing in the table.
2344 * set a bogus sector size so the normal read/write
2345 * logic in the block layer will eventually refuse any
2346 * request on this device without tripping over power
2347 * of two sector size assumptions
2351 blk_queue_logical_block_size(sdp
->request_queue
, sector_size
);
2354 char cap_str_2
[10], cap_str_10
[10];
2356 string_get_size(sdkp
->capacity
, sector_size
,
2357 STRING_UNITS_2
, cap_str_2
, sizeof(cap_str_2
));
2358 string_get_size(sdkp
->capacity
, sector_size
,
2359 STRING_UNITS_10
, cap_str_10
,
2360 sizeof(cap_str_10
));
2362 if (sdkp
->first_scan
|| old_capacity
!= sdkp
->capacity
) {
2363 sd_printk(KERN_NOTICE
, sdkp
,
2364 "%llu %d-byte logical blocks: (%s/%s)\n",
2365 (unsigned long long)sdkp
->capacity
,
2366 sector_size
, cap_str_10
, cap_str_2
);
2368 if (sdkp
->physical_block_size
!= sector_size
)
2369 sd_printk(KERN_NOTICE
, sdkp
,
2370 "%u-byte physical blocks\n",
2371 sdkp
->physical_block_size
);
2375 if (sdkp
->capacity
> 0xffffffff)
2376 sdp
->use_16_for_rw
= 1;
2378 blk_queue_physical_block_size(sdp
->request_queue
,
2379 sdkp
->physical_block_size
);
2380 sdkp
->device
->sector_size
= sector_size
;
2383 /* called with buffer of length 512 */
2385 sd_do_mode_sense(struct scsi_device
*sdp
, int dbd
, int modepage
,
2386 unsigned char *buffer
, int len
, struct scsi_mode_data
*data
,
2387 struct scsi_sense_hdr
*sshdr
)
2389 return scsi_mode_sense(sdp
, dbd
, modepage
, buffer
, len
,
2390 SD_TIMEOUT
, SD_MAX_RETRIES
, data
,
2395 * read write protect setting, if possible - called only in sd_revalidate_disk()
2396 * called with buffer of length SD_BUF_SIZE
2399 sd_read_write_protect_flag(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2402 struct scsi_device
*sdp
= sdkp
->device
;
2403 struct scsi_mode_data data
;
2404 int disk_ro
= get_disk_ro(sdkp
->disk
);
2405 int old_wp
= sdkp
->write_prot
;
2407 set_disk_ro(sdkp
->disk
, 0);
2408 if (sdp
->skip_ms_page_3f
) {
2409 sd_first_printk(KERN_NOTICE
, sdkp
, "Assuming Write Enabled\n");
2413 if (sdp
->use_192_bytes_for_3f
) {
2414 res
= sd_do_mode_sense(sdp
, 0, 0x3F, buffer
, 192, &data
, NULL
);
2417 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2418 * We have to start carefully: some devices hang if we ask
2419 * for more than is available.
2421 res
= sd_do_mode_sense(sdp
, 0, 0x3F, buffer
, 4, &data
, NULL
);
2424 * Second attempt: ask for page 0 When only page 0 is
2425 * implemented, a request for page 3F may return Sense Key
2426 * 5: Illegal Request, Sense Code 24: Invalid field in
2429 if (!scsi_status_is_good(res
))
2430 res
= sd_do_mode_sense(sdp
, 0, 0, buffer
, 4, &data
, NULL
);
2433 * Third attempt: ask 255 bytes, as we did earlier.
2435 if (!scsi_status_is_good(res
))
2436 res
= sd_do_mode_sense(sdp
, 0, 0x3F, buffer
, 255,
2440 if (!scsi_status_is_good(res
)) {
2441 sd_first_printk(KERN_WARNING
, sdkp
,
2442 "Test WP failed, assume Write Enabled\n");
2444 sdkp
->write_prot
= ((data
.device_specific
& 0x80) != 0);
2445 set_disk_ro(sdkp
->disk
, sdkp
->write_prot
|| disk_ro
);
2446 if (sdkp
->first_scan
|| old_wp
!= sdkp
->write_prot
) {
2447 sd_printk(KERN_NOTICE
, sdkp
, "Write Protect is %s\n",
2448 sdkp
->write_prot
? "on" : "off");
2449 sd_printk(KERN_DEBUG
, sdkp
,
2450 "Mode Sense: %02x %02x %02x %02x\n",
2451 buffer
[0], buffer
[1], buffer
[2], buffer
[3]);
2457 * sd_read_cache_type - called only from sd_revalidate_disk()
2458 * called with buffer of length SD_BUF_SIZE
2461 sd_read_cache_type(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2464 struct scsi_device
*sdp
= sdkp
->device
;
2469 struct scsi_mode_data data
;
2470 struct scsi_sense_hdr sshdr
;
2471 int old_wce
= sdkp
->WCE
;
2472 int old_rcd
= sdkp
->RCD
;
2473 int old_dpofua
= sdkp
->DPOFUA
;
2476 if (sdkp
->cache_override
)
2480 if (sdp
->skip_ms_page_8
) {
2481 if (sdp
->type
== TYPE_RBC
)
2484 if (sdp
->skip_ms_page_3f
)
2487 if (sdp
->use_192_bytes_for_3f
)
2491 } else if (sdp
->type
== TYPE_RBC
) {
2499 /* cautiously ask */
2500 res
= sd_do_mode_sense(sdp
, dbd
, modepage
, buffer
, first_len
,
2503 if (!scsi_status_is_good(res
))
2506 if (!data
.header_length
) {
2509 sd_first_printk(KERN_ERR
, sdkp
,
2510 "Missing header in MODE_SENSE response\n");
2513 /* that went OK, now ask for the proper length */
2517 * We're only interested in the first three bytes, actually.
2518 * But the data cache page is defined for the first 20.
2522 else if (len
> SD_BUF_SIZE
) {
2523 sd_first_printk(KERN_NOTICE
, sdkp
, "Truncating mode parameter "
2524 "data from %d to %d bytes\n", len
, SD_BUF_SIZE
);
2527 if (modepage
== 0x3F && sdp
->use_192_bytes_for_3f
)
2531 if (len
> first_len
)
2532 res
= sd_do_mode_sense(sdp
, dbd
, modepage
, buffer
, len
,
2535 if (scsi_status_is_good(res
)) {
2536 int offset
= data
.header_length
+ data
.block_descriptor_length
;
2538 while (offset
< len
) {
2539 u8 page_code
= buffer
[offset
] & 0x3F;
2540 u8 spf
= buffer
[offset
] & 0x40;
2542 if (page_code
== 8 || page_code
== 6) {
2543 /* We're interested only in the first 3 bytes.
2545 if (len
- offset
<= 2) {
2546 sd_first_printk(KERN_ERR
, sdkp
,
2547 "Incomplete mode parameter "
2551 modepage
= page_code
;
2555 /* Go to the next page */
2556 if (spf
&& len
- offset
> 3)
2557 offset
+= 4 + (buffer
[offset
+2] << 8) +
2559 else if (!spf
&& len
- offset
> 1)
2560 offset
+= 2 + buffer
[offset
+1];
2562 sd_first_printk(KERN_ERR
, sdkp
,
2564 "parameter data\n");
2570 sd_first_printk(KERN_ERR
, sdkp
, "No Caching mode page found\n");
2574 if (modepage
== 8) {
2575 sdkp
->WCE
= ((buffer
[offset
+ 2] & 0x04) != 0);
2576 sdkp
->RCD
= ((buffer
[offset
+ 2] & 0x01) != 0);
2578 sdkp
->WCE
= ((buffer
[offset
+ 2] & 0x01) == 0);
2582 sdkp
->DPOFUA
= (data
.device_specific
& 0x10) != 0;
2583 if (sdp
->broken_fua
) {
2584 sd_first_printk(KERN_NOTICE
, sdkp
, "Disabling FUA\n");
2586 } else if (sdkp
->DPOFUA
&& !sdkp
->device
->use_10_for_rw
&&
2587 !sdkp
->device
->use_16_for_rw
) {
2588 sd_first_printk(KERN_NOTICE
, sdkp
,
2589 "Uses READ/WRITE(6), disabling FUA\n");
2593 /* No cache flush allowed for write protected devices */
2594 if (sdkp
->WCE
&& sdkp
->write_prot
)
2597 if (sdkp
->first_scan
|| old_wce
!= sdkp
->WCE
||
2598 old_rcd
!= sdkp
->RCD
|| old_dpofua
!= sdkp
->DPOFUA
)
2599 sd_printk(KERN_NOTICE
, sdkp
,
2600 "Write cache: %s, read cache: %s, %s\n",
2601 sdkp
->WCE
? "enabled" : "disabled",
2602 sdkp
->RCD
? "disabled" : "enabled",
2603 sdkp
->DPOFUA
? "supports DPO and FUA"
2604 : "doesn't support DPO or FUA");
2610 if (scsi_sense_valid(&sshdr
) &&
2611 sshdr
.sense_key
== ILLEGAL_REQUEST
&&
2612 sshdr
.asc
== 0x24 && sshdr
.ascq
== 0x0)
2613 /* Invalid field in CDB */
2614 sd_first_printk(KERN_NOTICE
, sdkp
, "Cache data unavailable\n");
2616 sd_first_printk(KERN_ERR
, sdkp
,
2617 "Asking for cache data failed\n");
2620 if (sdp
->wce_default_on
) {
2621 sd_first_printk(KERN_NOTICE
, sdkp
,
2622 "Assuming drive cache: write back\n");
2625 sd_first_printk(KERN_ERR
, sdkp
,
2626 "Assuming drive cache: write through\n");
2634 * The ATO bit indicates whether the DIF application tag is available
2635 * for use by the operating system.
2637 static void sd_read_app_tag_own(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2640 struct scsi_device
*sdp
= sdkp
->device
;
2641 struct scsi_mode_data data
;
2642 struct scsi_sense_hdr sshdr
;
2644 if (sdp
->type
!= TYPE_DISK
)
2647 if (sdkp
->protection_type
== 0)
2650 res
= scsi_mode_sense(sdp
, 1, 0x0a, buffer
, 36, SD_TIMEOUT
,
2651 SD_MAX_RETRIES
, &data
, &sshdr
);
2653 if (!scsi_status_is_good(res
) || !data
.header_length
||
2655 sd_first_printk(KERN_WARNING
, sdkp
,
2656 "getting Control mode page failed, assume no ATO\n");
2658 if (scsi_sense_valid(&sshdr
))
2659 sd_print_sense_hdr(sdkp
, &sshdr
);
2664 offset
= data
.header_length
+ data
.block_descriptor_length
;
2666 if ((buffer
[offset
] & 0x3f) != 0x0a) {
2667 sd_first_printk(KERN_ERR
, sdkp
, "ATO Got wrong page\n");
2671 if ((buffer
[offset
+ 5] & 0x80) == 0)
2680 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2681 * @disk: disk to query
2683 static void sd_read_block_limits(struct scsi_disk
*sdkp
)
2685 unsigned int sector_sz
= sdkp
->device
->sector_size
;
2686 const int vpd_len
= 64;
2687 unsigned char *buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2690 /* Block Limits VPD */
2691 scsi_get_vpd_page(sdkp
->device
, 0xb0, buffer
, vpd_len
))
2694 blk_queue_io_min(sdkp
->disk
->queue
,
2695 get_unaligned_be16(&buffer
[6]) * sector_sz
);
2697 sdkp
->max_xfer_blocks
= get_unaligned_be32(&buffer
[8]);
2698 sdkp
->opt_xfer_blocks
= get_unaligned_be32(&buffer
[12]);
2700 if (buffer
[3] == 0x3c) {
2701 unsigned int lba_count
, desc_count
;
2703 sdkp
->max_ws_blocks
= (u32
)get_unaligned_be64(&buffer
[36]);
2708 lba_count
= get_unaligned_be32(&buffer
[20]);
2709 desc_count
= get_unaligned_be32(&buffer
[24]);
2711 if (lba_count
&& desc_count
)
2712 sdkp
->max_unmap_blocks
= lba_count
;
2714 sdkp
->unmap_granularity
= get_unaligned_be32(&buffer
[28]);
2716 if (buffer
[32] & 0x80)
2717 sdkp
->unmap_alignment
=
2718 get_unaligned_be32(&buffer
[32]) & ~(1 << 31);
2720 if (!sdkp
->lbpvpd
) { /* LBP VPD page not provided */
2722 if (sdkp
->max_unmap_blocks
)
2723 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2725 sd_config_discard(sdkp
, SD_LBP_WS16
);
2727 } else { /* LBP VPD page tells us what to use */
2728 if (sdkp
->lbpu
&& sdkp
->max_unmap_blocks
&& !sdkp
->lbprz
)
2729 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2730 else if (sdkp
->lbpws
)
2731 sd_config_discard(sdkp
, SD_LBP_WS16
);
2732 else if (sdkp
->lbpws10
)
2733 sd_config_discard(sdkp
, SD_LBP_WS10
);
2734 else if (sdkp
->lbpu
&& sdkp
->max_unmap_blocks
)
2735 sd_config_discard(sdkp
, SD_LBP_UNMAP
);
2737 sd_config_discard(sdkp
, SD_LBP_DISABLE
);
2746 * sd_read_block_characteristics - Query block dev. characteristics
2747 * @disk: disk to query
2749 static void sd_read_block_characteristics(struct scsi_disk
*sdkp
)
2751 unsigned char *buffer
;
2753 const int vpd_len
= 64;
2755 buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2758 /* Block Device Characteristics VPD */
2759 scsi_get_vpd_page(sdkp
->device
, 0xb1, buffer
, vpd_len
))
2762 rot
= get_unaligned_be16(&buffer
[4]);
2765 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, sdkp
->disk
->queue
);
2766 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, sdkp
->disk
->queue
);
2774 * sd_read_block_provisioning - Query provisioning VPD page
2775 * @disk: disk to query
2777 static void sd_read_block_provisioning(struct scsi_disk
*sdkp
)
2779 unsigned char *buffer
;
2780 const int vpd_len
= 8;
2782 if (sdkp
->lbpme
== 0)
2785 buffer
= kmalloc(vpd_len
, GFP_KERNEL
);
2787 if (!buffer
|| scsi_get_vpd_page(sdkp
->device
, 0xb2, buffer
, vpd_len
))
2791 sdkp
->lbpu
= (buffer
[5] >> 7) & 1; /* UNMAP */
2792 sdkp
->lbpws
= (buffer
[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2793 sdkp
->lbpws10
= (buffer
[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2799 static void sd_read_write_same(struct scsi_disk
*sdkp
, unsigned char *buffer
)
2801 struct scsi_device
*sdev
= sdkp
->device
;
2803 if (sdev
->host
->no_write_same
) {
2804 sdev
->no_write_same
= 1;
2809 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, INQUIRY
) < 0) {
2810 /* too large values might cause issues with arcmsr */
2811 int vpd_buf_len
= 64;
2813 sdev
->no_report_opcodes
= 1;
2815 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2816 * CODES is unsupported and the device has an ATA
2817 * Information VPD page (SAT).
2819 if (!scsi_get_vpd_page(sdev
, 0x89, buffer
, vpd_buf_len
))
2820 sdev
->no_write_same
= 1;
2823 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, WRITE_SAME_16
) == 1)
2826 if (scsi_report_opcode(sdev
, buffer
, SD_BUF_SIZE
, WRITE_SAME
) == 1)
2831 * sd_revalidate_disk - called the first time a new disk is seen,
2832 * performs disk spin up, read_capacity, etc.
2833 * @disk: struct gendisk we care about
2835 static int sd_revalidate_disk(struct gendisk
*disk
)
2837 struct scsi_disk
*sdkp
= scsi_disk(disk
);
2838 struct scsi_device
*sdp
= sdkp
->device
;
2839 struct request_queue
*q
= sdkp
->disk
->queue
;
2840 unsigned char *buffer
;
2841 unsigned int dev_max
, rw_max
;
2843 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO
, sdkp
,
2844 "sd_revalidate_disk\n"));
2847 * If the device is offline, don't try and read capacity or any
2848 * of the other niceties.
2850 if (!scsi_device_online(sdp
))
2853 buffer
= kmalloc(SD_BUF_SIZE
, GFP_KERNEL
);
2855 sd_printk(KERN_WARNING
, sdkp
, "sd_revalidate_disk: Memory "
2856 "allocation failure.\n");
2860 sd_spinup_disk(sdkp
);
2863 * Without media there is no reason to ask; moreover, some devices
2864 * react badly if we do.
2866 if (sdkp
->media_present
) {
2867 sd_read_capacity(sdkp
, buffer
);
2869 if (scsi_device_supports_vpd(sdp
)) {
2870 sd_read_block_provisioning(sdkp
);
2871 sd_read_block_limits(sdkp
);
2872 sd_read_block_characteristics(sdkp
);
2875 sd_read_write_protect_flag(sdkp
, buffer
);
2876 sd_read_cache_type(sdkp
, buffer
);
2877 sd_read_app_tag_own(sdkp
, buffer
);
2878 sd_read_write_same(sdkp
, buffer
);
2882 * We now have all cache related info, determine how we deal
2883 * with flush requests.
2885 sd_set_flush_flag(sdkp
);
2887 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2888 dev_max
= sdp
->use_16_for_rw
? SD_MAX_XFER_BLOCKS
: SD_DEF_XFER_BLOCKS
;
2890 /* Some devices report a maximum block count for READ/WRITE requests. */
2891 dev_max
= min_not_zero(dev_max
, sdkp
->max_xfer_blocks
);
2892 q
->limits
.max_dev_sectors
= logical_to_sectors(sdp
, dev_max
);
2895 * Determine the device's preferred I/O size for reads and writes
2896 * unless the reported value is unreasonably small, large, or
2899 if (sdkp
->opt_xfer_blocks
&&
2900 sdkp
->opt_xfer_blocks
<= dev_max
&&
2901 sdkp
->opt_xfer_blocks
<= SD_DEF_XFER_BLOCKS
&&
2902 logical_to_bytes(sdp
, sdkp
->opt_xfer_blocks
) >= PAGE_SIZE
) {
2903 q
->limits
.io_opt
= logical_to_bytes(sdp
, sdkp
->opt_xfer_blocks
);
2904 rw_max
= logical_to_sectors(sdp
, sdkp
->opt_xfer_blocks
);
2906 rw_max
= min_not_zero(logical_to_sectors(sdp
, dev_max
),
2907 (sector_t
)BLK_DEF_MAX_SECTORS
);
2909 /* Do not exceed controller limit */
2910 rw_max
= min(rw_max
, queue_max_hw_sectors(q
));
2913 * Only update max_sectors if previously unset or if the current value
2914 * exceeds the capabilities of the hardware.
2916 if (sdkp
->first_scan
||
2917 q
->limits
.max_sectors
> q
->limits
.max_dev_sectors
||
2918 q
->limits
.max_sectors
> q
->limits
.max_hw_sectors
)
2919 q
->limits
.max_sectors
= rw_max
;
2921 sdkp
->first_scan
= 0;
2923 set_capacity(disk
, logical_to_sectors(sdp
, sdkp
->capacity
));
2924 sd_config_write_same(sdkp
);
2932 * sd_unlock_native_capacity - unlock native capacity
2933 * @disk: struct gendisk to set capacity for
2935 * Block layer calls this function if it detects that partitions
2936 * on @disk reach beyond the end of the device. If the SCSI host
2937 * implements ->unlock_native_capacity() method, it's invoked to
2938 * give it a chance to adjust the device capacity.
2941 * Defined by block layer. Might sleep.
2943 static void sd_unlock_native_capacity(struct gendisk
*disk
)
2945 struct scsi_device
*sdev
= scsi_disk(disk
)->device
;
2947 if (sdev
->host
->hostt
->unlock_native_capacity
)
2948 sdev
->host
->hostt
->unlock_native_capacity(sdev
);
2952 * sd_format_disk_name - format disk name
2953 * @prefix: name prefix - ie. "sd" for SCSI disks
2954 * @index: index of the disk to format name for
2955 * @buf: output buffer
2956 * @buflen: length of the output buffer
2958 * SCSI disk names starts at sda. The 26th device is sdz and the
2959 * 27th is sdaa. The last one for two lettered suffix is sdzz
2960 * which is followed by sdaaa.
2962 * This is basically 26 base counting with one extra 'nil' entry
2963 * at the beginning from the second digit on and can be
2964 * determined using similar method as 26 base conversion with the
2965 * index shifted -1 after each digit is computed.
2971 * 0 on success, -errno on failure.
2973 static int sd_format_disk_name(char *prefix
, int index
, char *buf
, int buflen
)
2975 const int base
= 'z' - 'a' + 1;
2976 char *begin
= buf
+ strlen(prefix
);
2977 char *end
= buf
+ buflen
;
2987 *--p
= 'a' + (index
% unit
);
2988 index
= (index
/ unit
) - 1;
2989 } while (index
>= 0);
2991 memmove(begin
, p
, end
- p
);
2992 memcpy(buf
, prefix
, strlen(prefix
));
2998 * The asynchronous part of sd_probe
3000 static void sd_probe_async(void *data
, async_cookie_t cookie
)
3002 struct scsi_disk
*sdkp
= data
;
3003 struct scsi_device
*sdp
;
3010 index
= sdkp
->index
;
3011 dev
= &sdp
->sdev_gendev
;
3013 gd
->major
= sd_major((index
& 0xf0) >> 4);
3014 gd
->first_minor
= ((index
& 0xf) << 4) | (index
& 0xfff00);
3015 gd
->minors
= SD_MINORS
;
3017 gd
->fops
= &sd_fops
;
3018 gd
->private_data
= &sdkp
->driver
;
3019 gd
->queue
= sdkp
->device
->request_queue
;
3021 /* defaults, until the device tells us otherwise */
3022 sdp
->sector_size
= 512;
3024 sdkp
->media_present
= 1;
3025 sdkp
->write_prot
= 0;
3026 sdkp
->cache_override
= 0;
3030 sdkp
->first_scan
= 1;
3031 sdkp
->max_medium_access_timeouts
= SD_MAX_MEDIUM_TIMEOUTS
;
3033 sd_revalidate_disk(gd
);
3035 gd
->flags
= GENHD_FL_EXT_DEVT
;
3036 if (sdp
->removable
) {
3037 gd
->flags
|= GENHD_FL_REMOVABLE
;
3038 gd
->events
|= DISK_EVENT_MEDIA_CHANGE
;
3041 blk_pm_runtime_init(sdp
->request_queue
, dev
);
3042 device_add_disk(dev
, gd
);
3044 sd_dif_config_host(sdkp
);
3046 sd_revalidate_disk(gd
);
3048 sd_printk(KERN_NOTICE
, sdkp
, "Attached SCSI %sdisk\n",
3049 sdp
->removable
? "removable " : "");
3050 scsi_autopm_put_device(sdp
);
3051 put_device(&sdkp
->dev
);
3055 * sd_probe - called during driver initialization and whenever a
3056 * new scsi device is attached to the system. It is called once
3057 * for each scsi device (not just disks) present.
3058 * @dev: pointer to device object
3060 * Returns 0 if successful (or not interested in this scsi device
3061 * (e.g. scanner)); 1 when there is an error.
3063 * Note: this function is invoked from the scsi mid-level.
3064 * This function sets up the mapping between a given
3065 * <host,channel,id,lun> (found in sdp) and new device name
3066 * (e.g. /dev/sda). More precisely it is the block device major
3067 * and minor number that is chosen here.
3069 * Assume sd_probe is not re-entrant (for time being)
3070 * Also think about sd_probe() and sd_remove() running coincidentally.
3072 static int sd_probe(struct device
*dev
)
3074 struct scsi_device
*sdp
= to_scsi_device(dev
);
3075 struct scsi_disk
*sdkp
;
3080 scsi_autopm_get_device(sdp
);
3082 if (sdp
->type
!= TYPE_DISK
&& sdp
->type
!= TYPE_MOD
&& sdp
->type
!= TYPE_RBC
)
3085 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO
, sdp
,
3089 sdkp
= kzalloc(sizeof(*sdkp
), GFP_KERNEL
);
3093 gd
= alloc_disk(SD_MINORS
);
3098 if (!ida_pre_get(&sd_index_ida
, GFP_KERNEL
))
3101 spin_lock(&sd_index_lock
);
3102 error
= ida_get_new(&sd_index_ida
, &index
);
3103 spin_unlock(&sd_index_lock
);
3104 } while (error
== -EAGAIN
);
3107 sdev_printk(KERN_WARNING
, sdp
, "sd_probe: memory exhausted.\n");
3111 error
= sd_format_disk_name("sd", index
, gd
->disk_name
, DISK_NAME_LEN
);
3113 sdev_printk(KERN_WARNING
, sdp
, "SCSI disk (sd) name length exceeded.\n");
3114 goto out_free_index
;
3118 sdkp
->driver
= &sd_template
;
3120 sdkp
->index
= index
;
3121 atomic_set(&sdkp
->openers
, 0);
3122 atomic_set(&sdkp
->device
->ioerr_cnt
, 0);
3124 if (!sdp
->request_queue
->rq_timeout
) {
3125 if (sdp
->type
!= TYPE_MOD
)
3126 blk_queue_rq_timeout(sdp
->request_queue
, SD_TIMEOUT
);
3128 blk_queue_rq_timeout(sdp
->request_queue
,
3132 device_initialize(&sdkp
->dev
);
3133 sdkp
->dev
.parent
= dev
;
3134 sdkp
->dev
.class = &sd_disk_class
;
3135 dev_set_name(&sdkp
->dev
, "%s", dev_name(dev
));
3137 error
= device_add(&sdkp
->dev
);
3139 goto out_free_index
;
3142 dev_set_drvdata(dev
, sdkp
);
3144 get_device(&sdkp
->dev
); /* prevent release before async_schedule */
3145 async_schedule_domain(sd_probe_async
, sdkp
, &scsi_sd_probe_domain
);
3150 spin_lock(&sd_index_lock
);
3151 ida_remove(&sd_index_ida
, index
);
3152 spin_unlock(&sd_index_lock
);
3158 scsi_autopm_put_device(sdp
);
3163 * sd_remove - called whenever a scsi disk (previously recognized by
3164 * sd_probe) is detached from the system. It is called (potentially
3165 * multiple times) during sd module unload.
3166 * @sdp: pointer to mid level scsi device object
3168 * Note: this function is invoked from the scsi mid-level.
3169 * This function potentially frees up a device name (e.g. /dev/sdc)
3170 * that could be re-used by a subsequent sd_probe().
3171 * This function is not called when the built-in sd driver is "exit-ed".
3173 static int sd_remove(struct device
*dev
)
3175 struct scsi_disk
*sdkp
;
3178 sdkp
= dev_get_drvdata(dev
);
3179 devt
= disk_devt(sdkp
->disk
);
3180 scsi_autopm_get_device(sdkp
->device
);
3182 async_synchronize_full_domain(&scsi_sd_pm_domain
);
3183 async_synchronize_full_domain(&scsi_sd_probe_domain
);
3184 device_del(&sdkp
->dev
);
3185 del_gendisk(sdkp
->disk
);
3188 blk_register_region(devt
, SD_MINORS
, NULL
,
3189 sd_default_probe
, NULL
, NULL
);
3191 mutex_lock(&sd_ref_mutex
);
3192 dev_set_drvdata(dev
, NULL
);
3193 put_device(&sdkp
->dev
);
3194 mutex_unlock(&sd_ref_mutex
);
3200 * scsi_disk_release - Called to free the scsi_disk structure
3201 * @dev: pointer to embedded class device
3203 * sd_ref_mutex must be held entering this routine. Because it is
3204 * called on last put, you should always use the scsi_disk_get()
3205 * scsi_disk_put() helpers which manipulate the semaphore directly
3206 * and never do a direct put_device.
3208 static void scsi_disk_release(struct device
*dev
)
3210 struct scsi_disk
*sdkp
= to_scsi_disk(dev
);
3211 struct gendisk
*disk
= sdkp
->disk
;
3213 spin_lock(&sd_index_lock
);
3214 ida_remove(&sd_index_ida
, sdkp
->index
);
3215 spin_unlock(&sd_index_lock
);
3217 disk
->private_data
= NULL
;
3219 put_device(&sdkp
->device
->sdev_gendev
);
3224 static int sd_start_stop_device(struct scsi_disk
*sdkp
, int start
)
3226 unsigned char cmd
[6] = { START_STOP
}; /* START_VALID */
3227 struct scsi_sense_hdr sshdr
;
3228 struct scsi_device
*sdp
= sdkp
->device
;
3232 cmd
[4] |= 1; /* START */
3234 if (sdp
->start_stop_pwr_cond
)
3235 cmd
[4] |= start
? 1 << 4 : 3 << 4; /* Active or Standby */
3237 if (!scsi_device_online(sdp
))
3240 res
= scsi_execute_req_flags(sdp
, cmd
, DMA_NONE
, NULL
, 0, &sshdr
,
3241 SD_TIMEOUT
, SD_MAX_RETRIES
, NULL
, REQ_PM
);
3243 sd_print_result(sdkp
, "Start/Stop Unit failed", res
);
3244 if (driver_byte(res
) & DRIVER_SENSE
)
3245 sd_print_sense_hdr(sdkp
, &sshdr
);
3246 if (scsi_sense_valid(&sshdr
) &&
3247 /* 0x3a is medium not present */
3252 /* SCSI error codes must not go to the generic layer */
3260 * Send a SYNCHRONIZE CACHE instruction down to the device through
3261 * the normal SCSI command structure. Wait for the command to
3264 static void sd_shutdown(struct device
*dev
)
3266 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3269 return; /* this can happen */
3271 if (pm_runtime_suspended(dev
))
3274 if (sdkp
->WCE
&& sdkp
->media_present
) {
3275 sd_printk(KERN_NOTICE
, sdkp
, "Synchronizing SCSI cache\n");
3276 sd_sync_cache(sdkp
);
3279 if (system_state
!= SYSTEM_RESTART
&& sdkp
->device
->manage_start_stop
) {
3280 sd_printk(KERN_NOTICE
, sdkp
, "Stopping disk\n");
3281 sd_start_stop_device(sdkp
, 0);
3285 static int sd_suspend_common(struct device
*dev
, bool ignore_stop_errors
)
3287 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3290 if (!sdkp
) /* E.g.: runtime suspend following sd_remove() */
3293 if (sdkp
->WCE
&& sdkp
->media_present
) {
3294 sd_printk(KERN_NOTICE
, sdkp
, "Synchronizing SCSI cache\n");
3295 ret
= sd_sync_cache(sdkp
);
3297 /* ignore OFFLINE device */
3304 if (sdkp
->device
->manage_start_stop
) {
3305 sd_printk(KERN_NOTICE
, sdkp
, "Stopping disk\n");
3306 /* an error is not worth aborting a system sleep */
3307 ret
= sd_start_stop_device(sdkp
, 0);
3308 if (ignore_stop_errors
)
3316 static int sd_suspend_system(struct device
*dev
)
3318 return sd_suspend_common(dev
, true);
3321 static int sd_suspend_runtime(struct device
*dev
)
3323 return sd_suspend_common(dev
, false);
3326 static int sd_resume(struct device
*dev
)
3328 struct scsi_disk
*sdkp
= dev_get_drvdata(dev
);
3330 if (!sdkp
) /* E.g.: runtime resume at the start of sd_probe() */
3333 if (!sdkp
->device
->manage_start_stop
)
3336 sd_printk(KERN_NOTICE
, sdkp
, "Starting disk\n");
3337 return sd_start_stop_device(sdkp
, 1);
3341 * init_sd - entry point for this driver (both when built in or when
3344 * Note: this function registers this driver with the scsi mid-level.
3346 static int __init
init_sd(void)
3348 int majors
= 0, i
, err
;
3350 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3352 for (i
= 0; i
< SD_MAJORS
; i
++) {
3353 if (register_blkdev(sd_major(i
), "sd") != 0)
3356 blk_register_region(sd_major(i
), SD_MINORS
, NULL
,
3357 sd_default_probe
, NULL
, NULL
);
3363 err
= class_register(&sd_disk_class
);
3367 sd_cdb_cache
= kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE
,
3369 if (!sd_cdb_cache
) {
3370 printk(KERN_ERR
"sd: can't init extended cdb cache\n");
3375 sd_cdb_pool
= mempool_create_slab_pool(SD_MEMPOOL_SIZE
, sd_cdb_cache
);
3377 printk(KERN_ERR
"sd: can't init extended cdb pool\n");
3382 err
= scsi_register_driver(&sd_template
.gendrv
);
3384 goto err_out_driver
;
3389 mempool_destroy(sd_cdb_pool
);
3392 kmem_cache_destroy(sd_cdb_cache
);
3395 class_unregister(&sd_disk_class
);
3397 for (i
= 0; i
< SD_MAJORS
; i
++)
3398 unregister_blkdev(sd_major(i
), "sd");
3403 * exit_sd - exit point for this driver (when it is a module).
3405 * Note: this function unregisters this driver from the scsi mid-level.
3407 static void __exit
exit_sd(void)
3411 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3413 scsi_unregister_driver(&sd_template
.gendrv
);
3414 mempool_destroy(sd_cdb_pool
);
3415 kmem_cache_destroy(sd_cdb_cache
);
3417 class_unregister(&sd_disk_class
);
3419 for (i
= 0; i
< SD_MAJORS
; i
++) {
3420 blk_unregister_region(sd_major(i
), SD_MINORS
);
3421 unregister_blkdev(sd_major(i
), "sd");
3425 module_init(init_sd
);
3426 module_exit(exit_sd
);
3428 static void sd_print_sense_hdr(struct scsi_disk
*sdkp
,
3429 struct scsi_sense_hdr
*sshdr
)
3431 scsi_print_sense_hdr(sdkp
->device
,
3432 sdkp
->disk
? sdkp
->disk
->disk_name
: NULL
, sshdr
);
3435 static void sd_print_result(const struct scsi_disk
*sdkp
, const char *msg
,
3438 const char *hb_string
= scsi_hostbyte_string(result
);
3439 const char *db_string
= scsi_driverbyte_string(result
);
3441 if (hb_string
|| db_string
)
3442 sd_printk(KERN_INFO
, sdkp
,
3443 "%s: Result: hostbyte=%s driverbyte=%s\n", msg
,
3444 hb_string
? hb_string
: "invalid",
3445 db_string
? db_string
: "invalid");
3447 sd_printk(KERN_INFO
, sdkp
,
3448 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3449 msg
, host_byte(result
), driver_byte(result
));