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/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/mutex.h>
51 #include <linux/string_helpers.h>
52 #include <linux/async.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
106 static void sd_config_discard(struct scsi_disk *, unsigned int);
107 static void sd_config_write_same(struct scsi_disk *);
108 static int sd_revalidate_disk(struct gendisk *);
109 static void sd_unlock_native_capacity(struct gendisk *disk);
110 static int sd_probe(struct device *);
111 static int sd_remove(struct device *);
112 static void sd_shutdown(struct device *);
113 static int sd_suspend_system(struct device *);
114 static int sd_suspend_runtime(struct device *);
115 static int sd_resume(struct device *);
116 static void sd_rescan(struct device *);
117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
119 static int sd_done(struct scsi_cmnd *);
120 static void sd_eh_reset(struct scsi_cmnd *);
121 static int sd_eh_action(struct scsi_cmnd *, int);
122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
123 static void scsi_disk_release(struct device *cdev);
124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
125 static void sd_print_result(const struct scsi_disk *, const char *, int);
127 static DEFINE_IDA(sd_index_ida);
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
136 static mempool_t *sd_page_pool;
138 static const char *sd_cache_types[] = {
139 "write through", "none", "write back",
140 "write back, no read (daft)"
143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
145 bool wc = false, fua = false;
153 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
157 cache_type_store(struct device *dev, struct device_attribute *attr,
158 const char *buf, size_t count)
160 int ct, rcd, wce, sp;
161 struct scsi_disk *sdkp = to_scsi_disk(dev);
162 struct scsi_device *sdp = sdkp->device;
165 struct scsi_mode_data data;
166 struct scsi_sense_hdr sshdr;
167 static const char temp[] = "temporary ";
170 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
171 /* no cache control on RBC devices; theoretically they
172 * can do it, but there's probably so many exceptions
173 * it's not worth the risk */
176 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
177 buf += sizeof(temp) - 1;
178 sdkp->cache_override = 1;
180 sdkp->cache_override = 0;
183 ct = sysfs_match_string(sd_cache_types, buf);
187 rcd = ct & 0x01 ? 1 : 0;
188 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 if (sdkp->cache_override) {
193 sd_set_flush_flag(sdkp);
197 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198 SD_MAX_RETRIES, &data, NULL))
200 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201 data.block_descriptor_length);
202 buffer_data = buffer + data.header_length +
203 data.block_descriptor_length;
204 buffer_data[2] &= ~0x05;
205 buffer_data[2] |= wce << 2 | rcd;
206 sp = buffer_data[0] & 0x80 ? 1 : 0;
207 buffer_data[0] &= ~0x80;
210 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
211 * received mode parameter buffer before doing MODE SELECT.
213 data.device_specific = 0;
215 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
216 SD_MAX_RETRIES, &data, &sshdr)) {
217 if (scsi_sense_valid(&sshdr))
218 sd_print_sense_hdr(sdkp, &sshdr);
221 revalidate_disk(sdkp->disk);
226 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
229 struct scsi_disk *sdkp = to_scsi_disk(dev);
230 struct scsi_device *sdp = sdkp->device;
232 return sprintf(buf, "%u\n", sdp->manage_start_stop);
236 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
237 const char *buf, size_t count)
239 struct scsi_disk *sdkp = to_scsi_disk(dev);
240 struct scsi_device *sdp = sdkp->device;
243 if (!capable(CAP_SYS_ADMIN))
246 if (kstrtobool(buf, &v))
249 sdp->manage_start_stop = v;
253 static DEVICE_ATTR_RW(manage_start_stop);
256 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
258 struct scsi_disk *sdkp = to_scsi_disk(dev);
260 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
264 allow_restart_store(struct device *dev, struct device_attribute *attr,
265 const char *buf, size_t count)
268 struct scsi_disk *sdkp = to_scsi_disk(dev);
269 struct scsi_device *sdp = sdkp->device;
271 if (!capable(CAP_SYS_ADMIN))
274 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
277 if (kstrtobool(buf, &v))
280 sdp->allow_restart = v;
284 static DEVICE_ATTR_RW(allow_restart);
287 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
289 struct scsi_disk *sdkp = to_scsi_disk(dev);
290 int ct = sdkp->RCD + 2*sdkp->WCE;
292 return sprintf(buf, "%s\n", sd_cache_types[ct]);
294 static DEVICE_ATTR_RW(cache_type);
297 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
299 struct scsi_disk *sdkp = to_scsi_disk(dev);
301 return sprintf(buf, "%u\n", sdkp->DPOFUA);
303 static DEVICE_ATTR_RO(FUA);
306 protection_type_show(struct device *dev, struct device_attribute *attr,
309 struct scsi_disk *sdkp = to_scsi_disk(dev);
311 return sprintf(buf, "%u\n", sdkp->protection_type);
315 protection_type_store(struct device *dev, struct device_attribute *attr,
316 const char *buf, size_t count)
318 struct scsi_disk *sdkp = to_scsi_disk(dev);
322 if (!capable(CAP_SYS_ADMIN))
325 err = kstrtouint(buf, 10, &val);
330 if (val <= T10_PI_TYPE3_PROTECTION)
331 sdkp->protection_type = val;
335 static DEVICE_ATTR_RW(protection_type);
338 protection_mode_show(struct device *dev, struct device_attribute *attr,
341 struct scsi_disk *sdkp = to_scsi_disk(dev);
342 struct scsi_device *sdp = sdkp->device;
343 unsigned int dif, dix;
345 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
346 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
348 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
354 return sprintf(buf, "none\n");
356 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
358 static DEVICE_ATTR_RO(protection_mode);
361 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
363 struct scsi_disk *sdkp = to_scsi_disk(dev);
365 return sprintf(buf, "%u\n", sdkp->ATO);
367 static DEVICE_ATTR_RO(app_tag_own);
370 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
373 struct scsi_disk *sdkp = to_scsi_disk(dev);
375 return sprintf(buf, "%u\n", sdkp->lbpme);
377 static DEVICE_ATTR_RO(thin_provisioning);
379 /* sysfs_match_string() requires dense arrays */
380 static const char *lbp_mode[] = {
381 [SD_LBP_FULL] = "full",
382 [SD_LBP_UNMAP] = "unmap",
383 [SD_LBP_WS16] = "writesame_16",
384 [SD_LBP_WS10] = "writesame_10",
385 [SD_LBP_ZERO] = "writesame_zero",
386 [SD_LBP_DISABLE] = "disabled",
390 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
393 struct scsi_disk *sdkp = to_scsi_disk(dev);
395 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
399 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
400 const char *buf, size_t count)
402 struct scsi_disk *sdkp = to_scsi_disk(dev);
403 struct scsi_device *sdp = sdkp->device;
406 if (!capable(CAP_SYS_ADMIN))
409 if (sd_is_zoned(sdkp)) {
410 sd_config_discard(sdkp, SD_LBP_DISABLE);
414 if (sdp->type != TYPE_DISK)
417 mode = sysfs_match_string(lbp_mode, buf);
421 sd_config_discard(sdkp, mode);
425 static DEVICE_ATTR_RW(provisioning_mode);
427 /* sysfs_match_string() requires dense arrays */
428 static const char *zeroing_mode[] = {
429 [SD_ZERO_WRITE] = "write",
430 [SD_ZERO_WS] = "writesame",
431 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
432 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
436 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
439 struct scsi_disk *sdkp = to_scsi_disk(dev);
441 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
445 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
446 const char *buf, size_t count)
448 struct scsi_disk *sdkp = to_scsi_disk(dev);
451 if (!capable(CAP_SYS_ADMIN))
454 mode = sysfs_match_string(zeroing_mode, buf);
458 sdkp->zeroing_mode = mode;
462 static DEVICE_ATTR_RW(zeroing_mode);
465 max_medium_access_timeouts_show(struct device *dev,
466 struct device_attribute *attr, char *buf)
468 struct scsi_disk *sdkp = to_scsi_disk(dev);
470 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
474 max_medium_access_timeouts_store(struct device *dev,
475 struct device_attribute *attr, const char *buf,
478 struct scsi_disk *sdkp = to_scsi_disk(dev);
481 if (!capable(CAP_SYS_ADMIN))
484 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
486 return err ? err : count;
488 static DEVICE_ATTR_RW(max_medium_access_timeouts);
491 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
494 struct scsi_disk *sdkp = to_scsi_disk(dev);
496 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
500 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
501 const char *buf, size_t count)
503 struct scsi_disk *sdkp = to_scsi_disk(dev);
504 struct scsi_device *sdp = sdkp->device;
508 if (!capable(CAP_SYS_ADMIN))
511 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
514 err = kstrtoul(buf, 10, &max);
520 sdp->no_write_same = 1;
521 else if (max <= SD_MAX_WS16_BLOCKS) {
522 sdp->no_write_same = 0;
523 sdkp->max_ws_blocks = max;
526 sd_config_write_same(sdkp);
530 static DEVICE_ATTR_RW(max_write_same_blocks);
532 static struct attribute *sd_disk_attrs[] = {
533 &dev_attr_cache_type.attr,
535 &dev_attr_allow_restart.attr,
536 &dev_attr_manage_start_stop.attr,
537 &dev_attr_protection_type.attr,
538 &dev_attr_protection_mode.attr,
539 &dev_attr_app_tag_own.attr,
540 &dev_attr_thin_provisioning.attr,
541 &dev_attr_provisioning_mode.attr,
542 &dev_attr_zeroing_mode.attr,
543 &dev_attr_max_write_same_blocks.attr,
544 &dev_attr_max_medium_access_timeouts.attr,
547 ATTRIBUTE_GROUPS(sd_disk);
549 static struct class sd_disk_class = {
551 .owner = THIS_MODULE,
552 .dev_release = scsi_disk_release,
553 .dev_groups = sd_disk_groups,
556 static const struct dev_pm_ops sd_pm_ops = {
557 .suspend = sd_suspend_system,
559 .poweroff = sd_suspend_system,
560 .restore = sd_resume,
561 .runtime_suspend = sd_suspend_runtime,
562 .runtime_resume = sd_resume,
565 static struct scsi_driver sd_template = {
568 .owner = THIS_MODULE,
571 .shutdown = sd_shutdown,
575 .init_command = sd_init_command,
576 .uninit_command = sd_uninit_command,
578 .eh_action = sd_eh_action,
579 .eh_reset = sd_eh_reset,
583 * Dummy kobj_map->probe function.
584 * The default ->probe function will call modprobe, which is
585 * pointless as this module is already loaded.
587 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
593 * Device no to disk mapping:
595 * major disc2 disc p1
596 * |............|.............|....|....| <- dev_t
599 * Inside a major, we have 16k disks, however mapped non-
600 * contiguously. The first 16 disks are for major0, the next
601 * ones with major1, ... Disk 256 is for major0 again, disk 272
603 * As we stay compatible with our numbering scheme, we can reuse
604 * the well-know SCSI majors 8, 65--71, 136--143.
606 static int sd_major(int major_idx)
610 return SCSI_DISK0_MAJOR;
612 return SCSI_DISK1_MAJOR + major_idx - 1;
614 return SCSI_DISK8_MAJOR + major_idx - 8;
617 return 0; /* shut up gcc */
621 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
623 struct scsi_disk *sdkp = NULL;
625 mutex_lock(&sd_ref_mutex);
627 if (disk->private_data) {
628 sdkp = scsi_disk(disk);
629 if (scsi_device_get(sdkp->device) == 0)
630 get_device(&sdkp->dev);
634 mutex_unlock(&sd_ref_mutex);
638 static void scsi_disk_put(struct scsi_disk *sdkp)
640 struct scsi_device *sdev = sdkp->device;
642 mutex_lock(&sd_ref_mutex);
643 put_device(&sdkp->dev);
644 scsi_device_put(sdev);
645 mutex_unlock(&sd_ref_mutex);
648 #ifdef CONFIG_BLK_SED_OPAL
649 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
650 size_t len, bool send)
652 struct scsi_device *sdev = data;
656 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
658 put_unaligned_be16(spsp, &cdb[2]);
659 put_unaligned_be32(len, &cdb[6]);
661 ret = scsi_execute_req(sdev, cdb,
662 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
663 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
664 return ret <= 0 ? ret : -EIO;
666 #endif /* CONFIG_BLK_SED_OPAL */
669 * Look up the DIX operation based on whether the command is read or
670 * write and whether dix and dif are enabled.
672 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
674 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
675 static const unsigned int ops[] = { /* wrt dix dif */
676 SCSI_PROT_NORMAL, /* 0 0 0 */
677 SCSI_PROT_READ_STRIP, /* 0 0 1 */
678 SCSI_PROT_READ_INSERT, /* 0 1 0 */
679 SCSI_PROT_READ_PASS, /* 0 1 1 */
680 SCSI_PROT_NORMAL, /* 1 0 0 */
681 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
682 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
683 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
686 return ops[write << 2 | dix << 1 | dif];
690 * Returns a mask of the protection flags that are valid for a given DIX
693 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
695 static const unsigned int flag_mask[] = {
696 [SCSI_PROT_NORMAL] = 0,
698 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
699 SCSI_PROT_GUARD_CHECK |
700 SCSI_PROT_REF_CHECK |
701 SCSI_PROT_REF_INCREMENT,
703 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
704 SCSI_PROT_IP_CHECKSUM,
706 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
707 SCSI_PROT_GUARD_CHECK |
708 SCSI_PROT_REF_CHECK |
709 SCSI_PROT_REF_INCREMENT |
710 SCSI_PROT_IP_CHECKSUM,
712 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
713 SCSI_PROT_REF_INCREMENT,
715 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
716 SCSI_PROT_REF_CHECK |
717 SCSI_PROT_REF_INCREMENT |
718 SCSI_PROT_IP_CHECKSUM,
720 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
721 SCSI_PROT_GUARD_CHECK |
722 SCSI_PROT_REF_CHECK |
723 SCSI_PROT_REF_INCREMENT |
724 SCSI_PROT_IP_CHECKSUM,
727 return flag_mask[prot_op];
730 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
731 unsigned int dix, unsigned int dif)
733 struct bio *bio = scmd->request->bio;
734 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
735 unsigned int protect = 0;
737 if (dix) { /* DIX Type 0, 1, 2, 3 */
738 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
739 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
741 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
742 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
745 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
746 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
748 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
749 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
752 if (dif) { /* DIX/DIF Type 1, 2, 3 */
753 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
755 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
756 protect = 3 << 5; /* Disable target PI checking */
758 protect = 1 << 5; /* Enable target PI checking */
761 scsi_set_prot_op(scmd, prot_op);
762 scsi_set_prot_type(scmd, dif);
763 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
768 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
770 struct request_queue *q = sdkp->disk->queue;
771 unsigned int logical_block_size = sdkp->device->sector_size;
772 unsigned int max_blocks = 0;
774 q->limits.discard_alignment =
775 sdkp->unmap_alignment * logical_block_size;
776 q->limits.discard_granularity =
777 max(sdkp->physical_block_size,
778 sdkp->unmap_granularity * logical_block_size);
779 sdkp->provisioning_mode = mode;
785 blk_queue_max_discard_sectors(q, 0);
786 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
790 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
791 (u32)SD_MAX_WS16_BLOCKS);
795 if (sdkp->device->unmap_limit_for_ws)
796 max_blocks = sdkp->max_unmap_blocks;
798 max_blocks = sdkp->max_ws_blocks;
800 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
804 if (sdkp->device->unmap_limit_for_ws)
805 max_blocks = sdkp->max_unmap_blocks;
807 max_blocks = sdkp->max_ws_blocks;
809 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
813 max_blocks = min_not_zero(sdkp->max_ws_blocks,
814 (u32)SD_MAX_WS10_BLOCKS);
818 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
819 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
822 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
824 struct scsi_device *sdp = cmd->device;
825 struct request *rq = cmd->request;
826 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
827 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
828 unsigned int data_len = 24;
831 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
832 if (!rq->special_vec.bv_page)
833 return BLK_STS_RESOURCE;
834 clear_highpage(rq->special_vec.bv_page);
835 rq->special_vec.bv_offset = 0;
836 rq->special_vec.bv_len = data_len;
837 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
840 cmd->cmnd[0] = UNMAP;
843 buf = page_address(rq->special_vec.bv_page);
844 put_unaligned_be16(6 + 16, &buf[0]);
845 put_unaligned_be16(16, &buf[2]);
846 put_unaligned_be64(lba, &buf[8]);
847 put_unaligned_be32(nr_blocks, &buf[16]);
849 cmd->allowed = SD_MAX_RETRIES;
850 cmd->transfersize = data_len;
851 rq->timeout = SD_TIMEOUT;
853 return scsi_init_io(cmd);
856 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
859 struct scsi_device *sdp = cmd->device;
860 struct request *rq = cmd->request;
861 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
862 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
863 u32 data_len = sdp->sector_size;
865 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
866 if (!rq->special_vec.bv_page)
867 return BLK_STS_RESOURCE;
868 clear_highpage(rq->special_vec.bv_page);
869 rq->special_vec.bv_offset = 0;
870 rq->special_vec.bv_len = data_len;
871 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
874 cmd->cmnd[0] = WRITE_SAME_16;
876 cmd->cmnd[1] = 0x8; /* UNMAP */
877 put_unaligned_be64(lba, &cmd->cmnd[2]);
878 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
880 cmd->allowed = SD_MAX_RETRIES;
881 cmd->transfersize = data_len;
882 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
884 return scsi_init_io(cmd);
887 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
890 struct scsi_device *sdp = cmd->device;
891 struct request *rq = cmd->request;
892 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
893 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
894 u32 data_len = sdp->sector_size;
896 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
897 if (!rq->special_vec.bv_page)
898 return BLK_STS_RESOURCE;
899 clear_highpage(rq->special_vec.bv_page);
900 rq->special_vec.bv_offset = 0;
901 rq->special_vec.bv_len = data_len;
902 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905 cmd->cmnd[0] = WRITE_SAME;
907 cmd->cmnd[1] = 0x8; /* UNMAP */
908 put_unaligned_be32(lba, &cmd->cmnd[2]);
909 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
911 cmd->allowed = SD_MAX_RETRIES;
912 cmd->transfersize = data_len;
913 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
915 return scsi_init_io(cmd);
918 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
920 struct request *rq = cmd->request;
921 struct scsi_device *sdp = cmd->device;
922 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
923 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
924 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
926 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
927 switch (sdkp->zeroing_mode) {
928 case SD_ZERO_WS16_UNMAP:
929 return sd_setup_write_same16_cmnd(cmd, true);
930 case SD_ZERO_WS10_UNMAP:
931 return sd_setup_write_same10_cmnd(cmd, true);
935 if (sdp->no_write_same)
936 return BLK_STS_TARGET;
938 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
939 return sd_setup_write_same16_cmnd(cmd, false);
941 return sd_setup_write_same10_cmnd(cmd, false);
944 static void sd_config_write_same(struct scsi_disk *sdkp)
946 struct request_queue *q = sdkp->disk->queue;
947 unsigned int logical_block_size = sdkp->device->sector_size;
949 if (sdkp->device->no_write_same) {
950 sdkp->max_ws_blocks = 0;
954 /* Some devices can not handle block counts above 0xffff despite
955 * supporting WRITE SAME(16). Consequently we default to 64k
956 * blocks per I/O unless the device explicitly advertises a
959 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
960 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
961 (u32)SD_MAX_WS16_BLOCKS);
962 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
963 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
964 (u32)SD_MAX_WS10_BLOCKS);
966 sdkp->device->no_write_same = 1;
967 sdkp->max_ws_blocks = 0;
970 if (sdkp->lbprz && sdkp->lbpws)
971 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
972 else if (sdkp->lbprz && sdkp->lbpws10)
973 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
974 else if (sdkp->max_ws_blocks)
975 sdkp->zeroing_mode = SD_ZERO_WS;
977 sdkp->zeroing_mode = SD_ZERO_WRITE;
979 if (sdkp->max_ws_blocks &&
980 sdkp->physical_block_size > logical_block_size) {
982 * Reporting a maximum number of blocks that is not aligned
983 * on the device physical size would cause a large write same
984 * request to be split into physically unaligned chunks by
985 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
986 * even if the caller of these functions took care to align the
987 * large request. So make sure the maximum reported is aligned
988 * to the device physical block size. This is only an optional
989 * optimization for regular disks, but this is mandatory to
990 * avoid failure of large write same requests directed at
991 * sequential write required zones of host-managed ZBC disks.
993 sdkp->max_ws_blocks =
994 round_down(sdkp->max_ws_blocks,
995 bytes_to_logical(sdkp->device,
996 sdkp->physical_block_size));
1000 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1001 (logical_block_size >> 9));
1002 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1003 (logical_block_size >> 9));
1007 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1008 * @cmd: command to prepare
1010 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1011 * the preference indicated by the target device.
1013 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1015 struct request *rq = cmd->request;
1016 struct scsi_device *sdp = cmd->device;
1017 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1018 struct bio *bio = rq->bio;
1019 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1020 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023 if (sdkp->device->no_write_same)
1024 return BLK_STS_TARGET;
1026 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1028 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1030 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1032 cmd->cmnd[0] = WRITE_SAME_16;
1033 put_unaligned_be64(lba, &cmd->cmnd[2]);
1034 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037 cmd->cmnd[0] = WRITE_SAME;
1038 put_unaligned_be32(lba, &cmd->cmnd[2]);
1039 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042 cmd->transfersize = sdp->sector_size;
1043 cmd->allowed = SD_MAX_RETRIES;
1046 * For WRITE SAME the data transferred via the DATA OUT buffer is
1047 * different from the amount of data actually written to the target.
1049 * We set up __data_len to the amount of data transferred via the
1050 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1051 * to transfer a single sector of data first, but then reset it to
1052 * the amount of data to be written right after so that the I/O path
1053 * knows how much to actually write.
1055 rq->__data_len = sdp->sector_size;
1056 ret = scsi_init_io(cmd);
1057 rq->__data_len = blk_rq_bytes(rq);
1062 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1064 struct request *rq = cmd->request;
1066 /* flush requests don't perform I/O, zero the S/G table */
1067 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1069 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1071 cmd->transfersize = 0;
1072 cmd->allowed = SD_MAX_RETRIES;
1074 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1078 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1079 sector_t lba, unsigned int nr_blocks,
1080 unsigned char flags)
1082 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1083 if (unlikely(cmd->cmnd == NULL))
1084 return BLK_STS_RESOURCE;
1086 cmd->cmd_len = SD_EXT_CDB_SIZE;
1087 memset(cmd->cmnd, 0, cmd->cmd_len);
1089 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1090 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1091 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1092 cmd->cmnd[10] = flags;
1093 put_unaligned_be64(lba, &cmd->cmnd[12]);
1094 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1095 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1100 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1101 sector_t lba, unsigned int nr_blocks,
1102 unsigned char flags)
1105 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1106 cmd->cmnd[1] = flags;
1109 put_unaligned_be64(lba, &cmd->cmnd[2]);
1110 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1115 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1116 sector_t lba, unsigned int nr_blocks,
1117 unsigned char flags)
1120 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1121 cmd->cmnd[1] = flags;
1124 put_unaligned_be32(lba, &cmd->cmnd[2]);
1125 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1130 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1131 sector_t lba, unsigned int nr_blocks,
1132 unsigned char flags)
1134 /* Avoid that 0 blocks gets translated into 256 blocks. */
1135 if (WARN_ON_ONCE(nr_blocks == 0))
1136 return BLK_STS_IOERR;
1138 if (unlikely(flags & 0x8)) {
1140 * This happens only if this drive failed 10byte rw
1141 * command with ILLEGAL_REQUEST during operation and
1142 * thus turned off use_10_for_rw.
1144 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1145 return BLK_STS_IOERR;
1149 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1150 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1151 cmd->cmnd[2] = (lba >> 8) & 0xff;
1152 cmd->cmnd[3] = lba & 0xff;
1153 cmd->cmnd[4] = nr_blocks;
1159 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1161 struct request *rq = cmd->request;
1162 struct scsi_device *sdp = cmd->device;
1163 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1164 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1166 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1168 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1169 bool write = rq_data_dir(rq) == WRITE;
1170 unsigned char protect, fua;
1173 ret = scsi_init_io(cmd);
1174 if (ret != BLK_STS_OK)
1177 if (!scsi_device_online(sdp) || sdp->changed) {
1178 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1179 return BLK_STS_IOERR;
1182 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1183 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1184 return BLK_STS_IOERR;
1187 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1188 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1189 return BLK_STS_IOERR;
1193 * Some SD card readers can't handle accesses which touch the
1194 * last one or two logical blocks. Split accesses as needed.
1196 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1198 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1199 if (lba < threshold) {
1200 /* Access up to the threshold but not beyond */
1201 nr_blocks = threshold - lba;
1203 /* Access only a single logical block */
1208 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1209 dix = scsi_prot_sg_count(cmd);
1210 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1213 t10_pi_prepare(cmd->request, sdkp->protection_type);
1216 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1220 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1221 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1223 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1224 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1226 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1227 sdp->use_10_for_rw || protect) {
1228 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1231 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1235 if (unlikely(ret != BLK_STS_OK))
1239 * We shouldn't disconnect in the middle of a sector, so with a dumb
1240 * host adapter, it's safe to assume that we can at least transfer
1241 * this many bytes between each connect / disconnect.
1243 cmd->transfersize = sdp->sector_size;
1244 cmd->underflow = nr_blocks << 9;
1245 cmd->allowed = SD_MAX_RETRIES;
1246 cmd->sdb.length = nr_blocks * sdp->sector_size;
1249 scmd_printk(KERN_INFO, cmd,
1250 "%s: block=%llu, count=%d\n", __func__,
1251 (unsigned long long)blk_rq_pos(rq),
1252 blk_rq_sectors(rq)));
1254 scmd_printk(KERN_INFO, cmd,
1255 "%s %d/%u 512 byte blocks.\n",
1256 write ? "writing" : "reading", nr_blocks,
1257 blk_rq_sectors(rq)));
1260 * This indicates that the command is ready from our end to be
1266 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1268 struct request *rq = cmd->request;
1270 switch (req_op(rq)) {
1271 case REQ_OP_DISCARD:
1272 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1274 return sd_setup_unmap_cmnd(cmd);
1276 return sd_setup_write_same16_cmnd(cmd, true);
1278 return sd_setup_write_same10_cmnd(cmd, true);
1280 return sd_setup_write_same10_cmnd(cmd, false);
1282 return BLK_STS_TARGET;
1284 case REQ_OP_WRITE_ZEROES:
1285 return sd_setup_write_zeroes_cmnd(cmd);
1286 case REQ_OP_WRITE_SAME:
1287 return sd_setup_write_same_cmnd(cmd);
1289 return sd_setup_flush_cmnd(cmd);
1292 return sd_setup_read_write_cmnd(cmd);
1293 case REQ_OP_ZONE_RESET:
1294 return sd_zbc_setup_reset_cmnd(cmd);
1297 return BLK_STS_NOTSUPP;
1301 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1303 struct request *rq = SCpnt->request;
1306 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1307 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1309 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1313 mempool_free(cmnd, sd_cdb_pool);
1318 * sd_open - open a scsi disk device
1319 * @bdev: Block device of the scsi disk to open
1320 * @mode: FMODE_* mask
1322 * Returns 0 if successful. Returns a negated errno value in case
1325 * Note: This can be called from a user context (e.g. fsck(1) )
1326 * or from within the kernel (e.g. as a result of a mount(1) ).
1327 * In the latter case @inode and @filp carry an abridged amount
1328 * of information as noted above.
1330 * Locking: called with bdev->bd_mutex held.
1332 static int sd_open(struct block_device *bdev, fmode_t mode)
1334 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1335 struct scsi_device *sdev;
1341 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1343 sdev = sdkp->device;
1346 * If the device is in error recovery, wait until it is done.
1347 * If the device is offline, then disallow any access to it.
1350 if (!scsi_block_when_processing_errors(sdev))
1353 if (sdev->removable || sdkp->write_prot)
1354 check_disk_change(bdev);
1357 * If the drive is empty, just let the open fail.
1359 retval = -ENOMEDIUM;
1360 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1364 * If the device has the write protect tab set, have the open fail
1365 * if the user expects to be able to write to the thing.
1368 if (sdkp->write_prot && (mode & FMODE_WRITE))
1372 * It is possible that the disk changing stuff resulted in
1373 * the device being taken offline. If this is the case,
1374 * report this to the user, and don't pretend that the
1375 * open actually succeeded.
1378 if (!scsi_device_online(sdev))
1381 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1382 if (scsi_block_when_processing_errors(sdev))
1383 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1389 scsi_disk_put(sdkp);
1394 * sd_release - invoked when the (last) close(2) is called on this
1396 * @disk: disk to release
1397 * @mode: FMODE_* mask
1401 * Note: may block (uninterruptible) if error recovery is underway
1404 * Locking: called with bdev->bd_mutex held.
1406 static void sd_release(struct gendisk *disk, fmode_t mode)
1408 struct scsi_disk *sdkp = scsi_disk(disk);
1409 struct scsi_device *sdev = sdkp->device;
1411 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1413 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1414 if (scsi_block_when_processing_errors(sdev))
1415 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1419 * XXX and what if there are packets in flight and this close()
1420 * XXX is followed by a "rmmod sd_mod"?
1423 scsi_disk_put(sdkp);
1426 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1428 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1429 struct scsi_device *sdp = sdkp->device;
1430 struct Scsi_Host *host = sdp->host;
1431 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1434 /* default to most commonly used values */
1435 diskinfo[0] = 0x40; /* 1 << 6 */
1436 diskinfo[1] = 0x20; /* 1 << 5 */
1437 diskinfo[2] = capacity >> 11;
1439 /* override with calculated, extended default, or driver values */
1440 if (host->hostt->bios_param)
1441 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1443 scsicam_bios_param(bdev, capacity, diskinfo);
1445 geo->heads = diskinfo[0];
1446 geo->sectors = diskinfo[1];
1447 geo->cylinders = diskinfo[2];
1452 * sd_ioctl - process an ioctl
1453 * @bdev: target block device
1454 * @mode: FMODE_* mask
1455 * @cmd: ioctl command number
1456 * @arg: this is third argument given to ioctl(2) system call.
1457 * Often contains a pointer.
1459 * Returns 0 if successful (some ioctls return positive numbers on
1460 * success as well). Returns a negated errno value in case of error.
1462 * Note: most ioctls are forward onto the block subsystem or further
1463 * down in the scsi subsystem.
1465 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1466 unsigned int cmd, unsigned long arg)
1468 struct gendisk *disk = bdev->bd_disk;
1469 struct scsi_disk *sdkp = scsi_disk(disk);
1470 struct scsi_device *sdp = sdkp->device;
1471 void __user *p = (void __user *)arg;
1474 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1475 "cmd=0x%x\n", disk->disk_name, cmd));
1477 error = scsi_verify_blk_ioctl(bdev, cmd);
1482 * If we are in the middle of error recovery, don't let anyone
1483 * else try and use this device. Also, if error recovery fails, it
1484 * may try and take the device offline, in which case all further
1485 * access to the device is prohibited.
1487 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1488 (mode & FMODE_NDELAY) != 0);
1492 if (is_sed_ioctl(cmd))
1493 return sed_ioctl(sdkp->opal_dev, cmd, p);
1496 * Send SCSI addressing ioctls directly to mid level, send other
1497 * ioctls to block level and then onto mid level if they can't be
1501 case SCSI_IOCTL_GET_IDLUN:
1502 case SCSI_IOCTL_GET_BUS_NUMBER:
1503 error = scsi_ioctl(sdp, cmd, p);
1506 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1507 if (error != -ENOTTY)
1509 error = scsi_ioctl(sdp, cmd, p);
1516 static void set_media_not_present(struct scsi_disk *sdkp)
1518 if (sdkp->media_present)
1519 sdkp->device->changed = 1;
1521 if (sdkp->device->removable) {
1522 sdkp->media_present = 0;
1527 static int media_not_present(struct scsi_disk *sdkp,
1528 struct scsi_sense_hdr *sshdr)
1530 if (!scsi_sense_valid(sshdr))
1533 /* not invoked for commands that could return deferred errors */
1534 switch (sshdr->sense_key) {
1535 case UNIT_ATTENTION:
1537 /* medium not present */
1538 if (sshdr->asc == 0x3A) {
1539 set_media_not_present(sdkp);
1547 * sd_check_events - check media events
1548 * @disk: kernel device descriptor
1549 * @clearing: disk events currently being cleared
1551 * Returns mask of DISK_EVENT_*.
1553 * Note: this function is invoked from the block subsystem.
1555 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1557 struct scsi_disk *sdkp = scsi_disk_get(disk);
1558 struct scsi_device *sdp;
1565 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1568 * If the device is offline, don't send any commands - just pretend as
1569 * if the command failed. If the device ever comes back online, we
1570 * can deal with it then. It is only because of unrecoverable errors
1571 * that we would ever take a device offline in the first place.
1573 if (!scsi_device_online(sdp)) {
1574 set_media_not_present(sdkp);
1579 * Using TEST_UNIT_READY enables differentiation between drive with
1580 * no cartridge loaded - NOT READY, drive with changed cartridge -
1581 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1583 * Drives that auto spin down. eg iomega jaz 1G, will be started
1584 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1585 * sd_revalidate() is called.
1587 if (scsi_block_when_processing_errors(sdp)) {
1588 struct scsi_sense_hdr sshdr = { 0, };
1590 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1593 /* failed to execute TUR, assume media not present */
1594 if (host_byte(retval)) {
1595 set_media_not_present(sdkp);
1599 if (media_not_present(sdkp, &sshdr))
1604 * For removable scsi disk we have to recognise the presence
1605 * of a disk in the drive.
1607 if (!sdkp->media_present)
1609 sdkp->media_present = 1;
1612 * sdp->changed is set under the following conditions:
1614 * Medium present state has changed in either direction.
1615 * Device has indicated UNIT_ATTENTION.
1617 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1619 scsi_disk_put(sdkp);
1623 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1626 struct scsi_device *sdp = sdkp->device;
1627 const int timeout = sdp->request_queue->rq_timeout
1628 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1629 struct scsi_sense_hdr my_sshdr;
1631 if (!scsi_device_online(sdp))
1634 /* caller might not be interested in sense, but we need it */
1638 for (retries = 3; retries > 0; --retries) {
1639 unsigned char cmd[10] = { 0 };
1641 cmd[0] = SYNCHRONIZE_CACHE;
1643 * Leave the rest of the command zero to indicate
1646 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1647 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1653 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1655 if (driver_byte(res) == DRIVER_SENSE)
1656 sd_print_sense_hdr(sdkp, sshdr);
1658 /* we need to evaluate the error return */
1659 if (scsi_sense_valid(sshdr) &&
1660 (sshdr->asc == 0x3a || /* medium not present */
1661 sshdr->asc == 0x20)) /* invalid command */
1662 /* this is no error here */
1665 switch (host_byte(res)) {
1666 /* ignore errors due to racing a disconnection */
1667 case DID_BAD_TARGET:
1668 case DID_NO_CONNECT:
1670 /* signal the upper layer it might try again */
1674 case DID_SOFT_ERROR:
1683 static void sd_rescan(struct device *dev)
1685 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1687 revalidate_disk(sdkp->disk);
1691 #ifdef CONFIG_COMPAT
1693 * This gets directly called from VFS. When the ioctl
1694 * is not recognized we go back to the other translation paths.
1696 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1697 unsigned int cmd, unsigned long arg)
1699 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1702 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1703 (mode & FMODE_NDELAY) != 0);
1708 * Let the static ioctl translation table take care of it.
1710 if (!sdev->host->hostt->compat_ioctl)
1711 return -ENOIOCTLCMD;
1712 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1716 static char sd_pr_type(enum pr_type type)
1719 case PR_WRITE_EXCLUSIVE:
1721 case PR_EXCLUSIVE_ACCESS:
1723 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1725 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1727 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1729 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1736 static int sd_pr_command(struct block_device *bdev, u8 sa,
1737 u64 key, u64 sa_key, u8 type, u8 flags)
1739 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1740 struct scsi_sense_hdr sshdr;
1742 u8 cmd[16] = { 0, };
1743 u8 data[24] = { 0, };
1745 cmd[0] = PERSISTENT_RESERVE_OUT;
1748 put_unaligned_be32(sizeof(data), &cmd[5]);
1750 put_unaligned_be64(key, &data[0]);
1751 put_unaligned_be64(sa_key, &data[8]);
1754 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1755 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1757 if (driver_byte(result) == DRIVER_SENSE &&
1758 scsi_sense_valid(&sshdr)) {
1759 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1760 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1766 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1769 if (flags & ~PR_FL_IGNORE_KEY)
1771 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1772 old_key, new_key, 0,
1773 (1 << 0) /* APTPL */);
1776 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1781 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1784 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1786 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1789 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1790 enum pr_type type, bool abort)
1792 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1793 sd_pr_type(type), 0);
1796 static int sd_pr_clear(struct block_device *bdev, u64 key)
1798 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1801 static const struct pr_ops sd_pr_ops = {
1802 .pr_register = sd_pr_register,
1803 .pr_reserve = sd_pr_reserve,
1804 .pr_release = sd_pr_release,
1805 .pr_preempt = sd_pr_preempt,
1806 .pr_clear = sd_pr_clear,
1809 static const struct block_device_operations sd_fops = {
1810 .owner = THIS_MODULE,
1812 .release = sd_release,
1814 .getgeo = sd_getgeo,
1815 #ifdef CONFIG_COMPAT
1816 .compat_ioctl = sd_compat_ioctl,
1818 .check_events = sd_check_events,
1819 .revalidate_disk = sd_revalidate_disk,
1820 .unlock_native_capacity = sd_unlock_native_capacity,
1821 .report_zones = sd_zbc_report_zones,
1822 .pr_ops = &sd_pr_ops,
1826 * sd_eh_reset - reset error handling callback
1827 * @scmd: sd-issued command that has failed
1829 * This function is called by the SCSI midlayer before starting
1830 * SCSI EH. When counting medium access failures we have to be
1831 * careful to register it only only once per device and SCSI EH run;
1832 * there might be several timed out commands which will cause the
1833 * 'max_medium_access_timeouts' counter to trigger after the first
1834 * SCSI EH run already and set the device to offline.
1835 * So this function resets the internal counter before starting SCSI EH.
1837 static void sd_eh_reset(struct scsi_cmnd *scmd)
1839 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1841 /* New SCSI EH run, reset gate variable */
1842 sdkp->ignore_medium_access_errors = false;
1846 * sd_eh_action - error handling callback
1847 * @scmd: sd-issued command that has failed
1848 * @eh_disp: The recovery disposition suggested by the midlayer
1850 * This function is called by the SCSI midlayer upon completion of an
1851 * error test command (currently TEST UNIT READY). The result of sending
1852 * the eh command is passed in eh_disp. We're looking for devices that
1853 * fail medium access commands but are OK with non access commands like
1854 * test unit ready (so wrongly see the device as having a successful
1857 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1859 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1860 struct scsi_device *sdev = scmd->device;
1862 if (!scsi_device_online(sdev) ||
1863 !scsi_medium_access_command(scmd) ||
1864 host_byte(scmd->result) != DID_TIME_OUT ||
1869 * The device has timed out executing a medium access command.
1870 * However, the TEST UNIT READY command sent during error
1871 * handling completed successfully. Either the device is in the
1872 * process of recovering or has it suffered an internal failure
1873 * that prevents access to the storage medium.
1875 if (!sdkp->ignore_medium_access_errors) {
1876 sdkp->medium_access_timed_out++;
1877 sdkp->ignore_medium_access_errors = true;
1881 * If the device keeps failing read/write commands but TEST UNIT
1882 * READY always completes successfully we assume that medium
1883 * access is no longer possible and take the device offline.
1885 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1886 scmd_printk(KERN_ERR, scmd,
1887 "Medium access timeout failure. Offlining disk!\n");
1888 mutex_lock(&sdev->state_mutex);
1889 scsi_device_set_state(sdev, SDEV_OFFLINE);
1890 mutex_unlock(&sdev->state_mutex);
1898 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1900 struct request *req = scmd->request;
1901 struct scsi_device *sdev = scmd->device;
1902 unsigned int transferred, good_bytes;
1903 u64 start_lba, end_lba, bad_lba;
1906 * Some commands have a payload smaller than the device logical
1907 * block size (e.g. INQUIRY on a 4K disk).
1909 if (scsi_bufflen(scmd) <= sdev->sector_size)
1912 /* Check if we have a 'bad_lba' information */
1913 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1914 SCSI_SENSE_BUFFERSIZE,
1919 * If the bad lba was reported incorrectly, we have no idea where
1922 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1923 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1924 if (bad_lba < start_lba || bad_lba >= end_lba)
1928 * resid is optional but mostly filled in. When it's unused,
1929 * its value is zero, so we assume the whole buffer transferred
1931 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1933 /* This computation should always be done in terms of the
1934 * resolution of the device's medium.
1936 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1938 return min(good_bytes, transferred);
1942 * sd_done - bottom half handler: called when the lower level
1943 * driver has completed (successfully or otherwise) a scsi command.
1944 * @SCpnt: mid-level's per command structure.
1946 * Note: potentially run from within an ISR. Must not block.
1948 static int sd_done(struct scsi_cmnd *SCpnt)
1950 int result = SCpnt->result;
1951 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1952 unsigned int sector_size = SCpnt->device->sector_size;
1954 struct scsi_sense_hdr sshdr;
1955 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1956 struct request *req = SCpnt->request;
1957 int sense_valid = 0;
1958 int sense_deferred = 0;
1960 switch (req_op(req)) {
1961 case REQ_OP_DISCARD:
1962 case REQ_OP_WRITE_ZEROES:
1963 case REQ_OP_WRITE_SAME:
1964 case REQ_OP_ZONE_RESET:
1966 good_bytes = blk_rq_bytes(req);
1967 scsi_set_resid(SCpnt, 0);
1970 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1975 * In case of bogus fw or device, we could end up having
1976 * an unaligned partial completion. Check this here and force
1979 resid = scsi_get_resid(SCpnt);
1980 if (resid & (sector_size - 1)) {
1981 sd_printk(KERN_INFO, sdkp,
1982 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1983 resid, sector_size);
1984 resid = min(scsi_bufflen(SCpnt),
1985 round_up(resid, sector_size));
1986 scsi_set_resid(SCpnt, resid);
1991 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1993 sense_deferred = scsi_sense_is_deferred(&sshdr);
1995 sdkp->medium_access_timed_out = 0;
1997 if (driver_byte(result) != DRIVER_SENSE &&
1998 (!sense_valid || sense_deferred))
2001 switch (sshdr.sense_key) {
2002 case HARDWARE_ERROR:
2004 good_bytes = sd_completed_bytes(SCpnt);
2006 case RECOVERED_ERROR:
2007 good_bytes = scsi_bufflen(SCpnt);
2010 /* This indicates a false check condition, so ignore it. An
2011 * unknown amount of data was transferred so treat it as an
2015 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2017 case ABORTED_COMMAND:
2018 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2019 good_bytes = sd_completed_bytes(SCpnt);
2021 case ILLEGAL_REQUEST:
2022 switch (sshdr.asc) {
2023 case 0x10: /* DIX: Host detected corruption */
2024 good_bytes = sd_completed_bytes(SCpnt);
2026 case 0x20: /* INVALID COMMAND OPCODE */
2027 case 0x24: /* INVALID FIELD IN CDB */
2028 switch (SCpnt->cmnd[0]) {
2030 sd_config_discard(sdkp, SD_LBP_DISABLE);
2034 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2035 sd_config_discard(sdkp, SD_LBP_DISABLE);
2037 sdkp->device->no_write_same = 1;
2038 sd_config_write_same(sdkp);
2039 req->rq_flags |= RQF_QUIET;
2050 if (sd_is_zoned(sdkp))
2051 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2053 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2054 "sd_done: completed %d of %d bytes\n",
2055 good_bytes, scsi_bufflen(SCpnt)));
2057 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2059 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2060 good_bytes / scsi_prot_interval(SCpnt));
2066 * spinup disk - called only in sd_revalidate_disk()
2069 sd_spinup_disk(struct scsi_disk *sdkp)
2071 unsigned char cmd[10];
2072 unsigned long spintime_expire = 0;
2073 int retries, spintime;
2074 unsigned int the_result;
2075 struct scsi_sense_hdr sshdr;
2076 int sense_valid = 0;
2080 /* Spin up drives, as required. Only do this at boot time */
2081 /* Spinup needs to be done for module loads too. */
2086 cmd[0] = TEST_UNIT_READY;
2087 memset((void *) &cmd[1], 0, 9);
2089 the_result = scsi_execute_req(sdkp->device, cmd,
2092 SD_MAX_RETRIES, NULL);
2095 * If the drive has indicated to us that it
2096 * doesn't have any media in it, don't bother
2097 * with any more polling.
2099 if (media_not_present(sdkp, &sshdr))
2103 sense_valid = scsi_sense_valid(&sshdr);
2105 } while (retries < 3 &&
2106 (!scsi_status_is_good(the_result) ||
2107 ((driver_byte(the_result) == DRIVER_SENSE) &&
2108 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2110 if (driver_byte(the_result) != DRIVER_SENSE) {
2111 /* no sense, TUR either succeeded or failed
2112 * with a status error */
2113 if(!spintime && !scsi_status_is_good(the_result)) {
2114 sd_print_result(sdkp, "Test Unit Ready failed",
2121 * The device does not want the automatic start to be issued.
2123 if (sdkp->device->no_start_on_add)
2126 if (sense_valid && sshdr.sense_key == NOT_READY) {
2127 if (sshdr.asc == 4 && sshdr.ascq == 3)
2128 break; /* manual intervention required */
2129 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2130 break; /* standby */
2131 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2132 break; /* unavailable */
2133 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2134 break; /* sanitize in progress */
2136 * Issue command to spin up drive when not ready
2139 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2140 cmd[0] = START_STOP;
2141 cmd[1] = 1; /* Return immediately */
2142 memset((void *) &cmd[2], 0, 8);
2143 cmd[4] = 1; /* Start spin cycle */
2144 if (sdkp->device->start_stop_pwr_cond)
2146 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2148 SD_TIMEOUT, SD_MAX_RETRIES,
2150 spintime_expire = jiffies + 100 * HZ;
2153 /* Wait 1 second for next try */
2155 printk(KERN_CONT ".");
2158 * Wait for USB flash devices with slow firmware.
2159 * Yes, this sense key/ASC combination shouldn't
2160 * occur here. It's characteristic of these devices.
2162 } else if (sense_valid &&
2163 sshdr.sense_key == UNIT_ATTENTION &&
2164 sshdr.asc == 0x28) {
2166 spintime_expire = jiffies + 5 * HZ;
2169 /* Wait 1 second for next try */
2172 /* we don't understand the sense code, so it's
2173 * probably pointless to loop */
2175 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2176 sd_print_sense_hdr(sdkp, &sshdr);
2181 } while (spintime && time_before_eq(jiffies, spintime_expire));
2184 if (scsi_status_is_good(the_result))
2185 printk(KERN_CONT "ready\n");
2187 printk(KERN_CONT "not responding...\n");
2192 * Determine whether disk supports Data Integrity Field.
2194 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2196 struct scsi_device *sdp = sdkp->device;
2200 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2203 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2205 if (type > T10_PI_TYPE3_PROTECTION)
2207 else if (scsi_host_dif_capable(sdp->host, type))
2210 if (sdkp->first_scan || type != sdkp->protection_type)
2213 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2214 " protection type %u. Disabling disk!\n",
2218 sd_printk(KERN_NOTICE, sdkp,
2219 "Enabling DIF Type %u protection\n", type);
2222 sd_printk(KERN_NOTICE, sdkp,
2223 "Disabling DIF Type %u protection\n", type);
2227 sdkp->protection_type = type;
2232 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2233 struct scsi_sense_hdr *sshdr, int sense_valid,
2236 if (driver_byte(the_result) == DRIVER_SENSE)
2237 sd_print_sense_hdr(sdkp, sshdr);
2239 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2242 * Set dirty bit for removable devices if not ready -
2243 * sometimes drives will not report this properly.
2245 if (sdp->removable &&
2246 sense_valid && sshdr->sense_key == NOT_READY)
2247 set_media_not_present(sdkp);
2250 * We used to set media_present to 0 here to indicate no media
2251 * in the drive, but some drives fail read capacity even with
2252 * media present, so we can't do that.
2254 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2258 #if RC16_LEN > SD_BUF_SIZE
2259 #error RC16_LEN must not be more than SD_BUF_SIZE
2262 #define READ_CAPACITY_RETRIES_ON_RESET 10
2265 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2266 * and the reported logical block size is bigger than 512 bytes. Note
2267 * that last_sector is a u64 and therefore logical_to_sectors() is not
2270 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2272 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2274 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2280 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2281 unsigned char *buffer)
2283 unsigned char cmd[16];
2284 struct scsi_sense_hdr sshdr;
2285 int sense_valid = 0;
2287 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2288 unsigned int alignment;
2289 unsigned long long lba;
2290 unsigned sector_size;
2292 if (sdp->no_read_capacity_16)
2297 cmd[0] = SERVICE_ACTION_IN_16;
2298 cmd[1] = SAI_READ_CAPACITY_16;
2300 memset(buffer, 0, RC16_LEN);
2302 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2303 buffer, RC16_LEN, &sshdr,
2304 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2306 if (media_not_present(sdkp, &sshdr))
2310 sense_valid = scsi_sense_valid(&sshdr);
2312 sshdr.sense_key == ILLEGAL_REQUEST &&
2313 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2315 /* Invalid Command Operation Code or
2316 * Invalid Field in CDB, just retry
2317 * silently with RC10 */
2320 sshdr.sense_key == UNIT_ATTENTION &&
2321 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2322 /* Device reset might occur several times,
2323 * give it one more chance */
2324 if (--reset_retries > 0)
2329 } while (the_result && retries);
2332 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2333 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2337 sector_size = get_unaligned_be32(&buffer[8]);
2338 lba = get_unaligned_be64(&buffer[0]);
2340 if (sd_read_protection_type(sdkp, buffer) < 0) {
2345 if (!sd_addressable_capacity(lba, sector_size)) {
2346 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2347 "kernel compiled with support for large block "
2353 /* Logical blocks per physical block exponent */
2354 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2357 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2359 /* Lowest aligned logical block */
2360 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2361 blk_queue_alignment_offset(sdp->request_queue, alignment);
2362 if (alignment && sdkp->first_scan)
2363 sd_printk(KERN_NOTICE, sdkp,
2364 "physical block alignment offset: %u\n", alignment);
2366 if (buffer[14] & 0x80) { /* LBPME */
2369 if (buffer[14] & 0x40) /* LBPRZ */
2372 sd_config_discard(sdkp, SD_LBP_WS16);
2375 sdkp->capacity = lba + 1;
2379 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2380 unsigned char *buffer)
2382 unsigned char cmd[16];
2383 struct scsi_sense_hdr sshdr;
2384 int sense_valid = 0;
2386 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2388 unsigned sector_size;
2391 cmd[0] = READ_CAPACITY;
2392 memset(&cmd[1], 0, 9);
2393 memset(buffer, 0, 8);
2395 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2397 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2399 if (media_not_present(sdkp, &sshdr))
2403 sense_valid = scsi_sense_valid(&sshdr);
2405 sshdr.sense_key == UNIT_ATTENTION &&
2406 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2407 /* Device reset might occur several times,
2408 * give it one more chance */
2409 if (--reset_retries > 0)
2414 } while (the_result && retries);
2417 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2418 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2422 sector_size = get_unaligned_be32(&buffer[4]);
2423 lba = get_unaligned_be32(&buffer[0]);
2425 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2426 /* Some buggy (usb cardreader) devices return an lba of
2427 0xffffffff when the want to report a size of 0 (with
2428 which they really mean no media is present) */
2430 sdkp->physical_block_size = sector_size;
2434 if (!sd_addressable_capacity(lba, sector_size)) {
2435 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2436 "kernel compiled with support for large block "
2442 sdkp->capacity = lba + 1;
2443 sdkp->physical_block_size = sector_size;
2447 static int sd_try_rc16_first(struct scsi_device *sdp)
2449 if (sdp->host->max_cmd_len < 16)
2451 if (sdp->try_rc_10_first)
2453 if (sdp->scsi_level > SCSI_SPC_2)
2455 if (scsi_device_protection(sdp))
2461 * read disk capacity
2464 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2467 struct scsi_device *sdp = sdkp->device;
2469 if (sd_try_rc16_first(sdp)) {
2470 sector_size = read_capacity_16(sdkp, sdp, buffer);
2471 if (sector_size == -EOVERFLOW)
2473 if (sector_size == -ENODEV)
2475 if (sector_size < 0)
2476 sector_size = read_capacity_10(sdkp, sdp, buffer);
2477 if (sector_size < 0)
2480 sector_size = read_capacity_10(sdkp, sdp, buffer);
2481 if (sector_size == -EOVERFLOW)
2483 if (sector_size < 0)
2485 if ((sizeof(sdkp->capacity) > 4) &&
2486 (sdkp->capacity > 0xffffffffULL)) {
2487 int old_sector_size = sector_size;
2488 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2489 "Trying to use READ CAPACITY(16).\n");
2490 sector_size = read_capacity_16(sdkp, sdp, buffer);
2491 if (sector_size < 0) {
2492 sd_printk(KERN_NOTICE, sdkp,
2493 "Using 0xffffffff as device size\n");
2494 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2495 sector_size = old_sector_size;
2498 /* Remember that READ CAPACITY(16) succeeded */
2499 sdp->try_rc_10_first = 0;
2503 /* Some devices are known to return the total number of blocks,
2504 * not the highest block number. Some devices have versions
2505 * which do this and others which do not. Some devices we might
2506 * suspect of doing this but we don't know for certain.
2508 * If we know the reported capacity is wrong, decrement it. If
2509 * we can only guess, then assume the number of blocks is even
2510 * (usually true but not always) and err on the side of lowering
2513 if (sdp->fix_capacity ||
2514 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2515 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2516 "from its reported value: %llu\n",
2517 (unsigned long long) sdkp->capacity);
2522 if (sector_size == 0) {
2524 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2528 if (sector_size != 512 &&
2529 sector_size != 1024 &&
2530 sector_size != 2048 &&
2531 sector_size != 4096) {
2532 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2535 * The user might want to re-format the drive with
2536 * a supported sectorsize. Once this happens, it
2537 * would be relatively trivial to set the thing up.
2538 * For this reason, we leave the thing in the table.
2542 * set a bogus sector size so the normal read/write
2543 * logic in the block layer will eventually refuse any
2544 * request on this device without tripping over power
2545 * of two sector size assumptions
2549 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2550 blk_queue_physical_block_size(sdp->request_queue,
2551 sdkp->physical_block_size);
2552 sdkp->device->sector_size = sector_size;
2554 if (sdkp->capacity > 0xffffffff)
2555 sdp->use_16_for_rw = 1;
2560 * Print disk capacity
2563 sd_print_capacity(struct scsi_disk *sdkp,
2564 sector_t old_capacity)
2566 int sector_size = sdkp->device->sector_size;
2567 char cap_str_2[10], cap_str_10[10];
2569 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2572 string_get_size(sdkp->capacity, sector_size,
2573 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2574 string_get_size(sdkp->capacity, sector_size,
2575 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2577 sd_printk(KERN_NOTICE, sdkp,
2578 "%llu %d-byte logical blocks: (%s/%s)\n",
2579 (unsigned long long)sdkp->capacity,
2580 sector_size, cap_str_10, cap_str_2);
2582 if (sdkp->physical_block_size != sector_size)
2583 sd_printk(KERN_NOTICE, sdkp,
2584 "%u-byte physical blocks\n",
2585 sdkp->physical_block_size);
2587 sd_zbc_print_zones(sdkp);
2590 /* called with buffer of length 512 */
2592 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2593 unsigned char *buffer, int len, struct scsi_mode_data *data,
2594 struct scsi_sense_hdr *sshdr)
2596 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2597 SD_TIMEOUT, SD_MAX_RETRIES, data,
2602 * read write protect setting, if possible - called only in sd_revalidate_disk()
2603 * called with buffer of length SD_BUF_SIZE
2606 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2609 struct scsi_device *sdp = sdkp->device;
2610 struct scsi_mode_data data;
2611 int disk_ro = get_disk_ro(sdkp->disk);
2612 int old_wp = sdkp->write_prot;
2614 set_disk_ro(sdkp->disk, 0);
2615 if (sdp->skip_ms_page_3f) {
2616 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2620 if (sdp->use_192_bytes_for_3f) {
2621 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2624 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2625 * We have to start carefully: some devices hang if we ask
2626 * for more than is available.
2628 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2631 * Second attempt: ask for page 0 When only page 0 is
2632 * implemented, a request for page 3F may return Sense Key
2633 * 5: Illegal Request, Sense Code 24: Invalid field in
2636 if (!scsi_status_is_good(res))
2637 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2640 * Third attempt: ask 255 bytes, as we did earlier.
2642 if (!scsi_status_is_good(res))
2643 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2647 if (!scsi_status_is_good(res)) {
2648 sd_first_printk(KERN_WARNING, sdkp,
2649 "Test WP failed, assume Write Enabled\n");
2651 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2652 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2653 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2654 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2655 sdkp->write_prot ? "on" : "off");
2656 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2662 * sd_read_cache_type - called only from sd_revalidate_disk()
2663 * called with buffer of length SD_BUF_SIZE
2666 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2669 struct scsi_device *sdp = sdkp->device;
2674 struct scsi_mode_data data;
2675 struct scsi_sense_hdr sshdr;
2676 int old_wce = sdkp->WCE;
2677 int old_rcd = sdkp->RCD;
2678 int old_dpofua = sdkp->DPOFUA;
2681 if (sdkp->cache_override)
2685 if (sdp->skip_ms_page_8) {
2686 if (sdp->type == TYPE_RBC)
2689 if (sdp->skip_ms_page_3f)
2692 if (sdp->use_192_bytes_for_3f)
2696 } else if (sdp->type == TYPE_RBC) {
2704 /* cautiously ask */
2705 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2708 if (!scsi_status_is_good(res))
2711 if (!data.header_length) {
2714 sd_first_printk(KERN_ERR, sdkp,
2715 "Missing header in MODE_SENSE response\n");
2718 /* that went OK, now ask for the proper length */
2722 * We're only interested in the first three bytes, actually.
2723 * But the data cache page is defined for the first 20.
2727 else if (len > SD_BUF_SIZE) {
2728 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2729 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2732 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2736 if (len > first_len)
2737 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2740 if (scsi_status_is_good(res)) {
2741 int offset = data.header_length + data.block_descriptor_length;
2743 while (offset < len) {
2744 u8 page_code = buffer[offset] & 0x3F;
2745 u8 spf = buffer[offset] & 0x40;
2747 if (page_code == 8 || page_code == 6) {
2748 /* We're interested only in the first 3 bytes.
2750 if (len - offset <= 2) {
2751 sd_first_printk(KERN_ERR, sdkp,
2752 "Incomplete mode parameter "
2756 modepage = page_code;
2760 /* Go to the next page */
2761 if (spf && len - offset > 3)
2762 offset += 4 + (buffer[offset+2] << 8) +
2764 else if (!spf && len - offset > 1)
2765 offset += 2 + buffer[offset+1];
2767 sd_first_printk(KERN_ERR, sdkp,
2769 "parameter data\n");
2775 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2779 if (modepage == 8) {
2780 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2781 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2783 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2787 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2788 if (sdp->broken_fua) {
2789 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2791 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2792 !sdkp->device->use_16_for_rw) {
2793 sd_first_printk(KERN_NOTICE, sdkp,
2794 "Uses READ/WRITE(6), disabling FUA\n");
2798 /* No cache flush allowed for write protected devices */
2799 if (sdkp->WCE && sdkp->write_prot)
2802 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2803 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2804 sd_printk(KERN_NOTICE, sdkp,
2805 "Write cache: %s, read cache: %s, %s\n",
2806 sdkp->WCE ? "enabled" : "disabled",
2807 sdkp->RCD ? "disabled" : "enabled",
2808 sdkp->DPOFUA ? "supports DPO and FUA"
2809 : "doesn't support DPO or FUA");
2815 if (scsi_sense_valid(&sshdr) &&
2816 sshdr.sense_key == ILLEGAL_REQUEST &&
2817 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2818 /* Invalid field in CDB */
2819 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2821 sd_first_printk(KERN_ERR, sdkp,
2822 "Asking for cache data failed\n");
2825 if (sdp->wce_default_on) {
2826 sd_first_printk(KERN_NOTICE, sdkp,
2827 "Assuming drive cache: write back\n");
2830 sd_first_printk(KERN_ERR, sdkp,
2831 "Assuming drive cache: write through\n");
2839 * The ATO bit indicates whether the DIF application tag is available
2840 * for use by the operating system.
2842 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2845 struct scsi_device *sdp = sdkp->device;
2846 struct scsi_mode_data data;
2847 struct scsi_sense_hdr sshdr;
2849 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2852 if (sdkp->protection_type == 0)
2855 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2856 SD_MAX_RETRIES, &data, &sshdr);
2858 if (!scsi_status_is_good(res) || !data.header_length ||
2860 sd_first_printk(KERN_WARNING, sdkp,
2861 "getting Control mode page failed, assume no ATO\n");
2863 if (scsi_sense_valid(&sshdr))
2864 sd_print_sense_hdr(sdkp, &sshdr);
2869 offset = data.header_length + data.block_descriptor_length;
2871 if ((buffer[offset] & 0x3f) != 0x0a) {
2872 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2876 if ((buffer[offset + 5] & 0x80) == 0)
2885 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2886 * @sdkp: disk to query
2888 static void sd_read_block_limits(struct scsi_disk *sdkp)
2890 unsigned int sector_sz = sdkp->device->sector_size;
2891 const int vpd_len = 64;
2892 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2895 /* Block Limits VPD */
2896 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2899 blk_queue_io_min(sdkp->disk->queue,
2900 get_unaligned_be16(&buffer[6]) * sector_sz);
2902 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2903 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2905 if (buffer[3] == 0x3c) {
2906 unsigned int lba_count, desc_count;
2908 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2913 lba_count = get_unaligned_be32(&buffer[20]);
2914 desc_count = get_unaligned_be32(&buffer[24]);
2916 if (lba_count && desc_count)
2917 sdkp->max_unmap_blocks = lba_count;
2919 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2921 if (buffer[32] & 0x80)
2922 sdkp->unmap_alignment =
2923 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2925 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2927 if (sdkp->max_unmap_blocks)
2928 sd_config_discard(sdkp, SD_LBP_UNMAP);
2930 sd_config_discard(sdkp, SD_LBP_WS16);
2932 } else { /* LBP VPD page tells us what to use */
2933 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2934 sd_config_discard(sdkp, SD_LBP_UNMAP);
2935 else if (sdkp->lbpws)
2936 sd_config_discard(sdkp, SD_LBP_WS16);
2937 else if (sdkp->lbpws10)
2938 sd_config_discard(sdkp, SD_LBP_WS10);
2940 sd_config_discard(sdkp, SD_LBP_DISABLE);
2949 * sd_read_block_characteristics - Query block dev. characteristics
2950 * @sdkp: disk to query
2952 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2954 struct request_queue *q = sdkp->disk->queue;
2955 unsigned char *buffer;
2957 const int vpd_len = 64;
2959 buffer = kmalloc(vpd_len, GFP_KERNEL);
2962 /* Block Device Characteristics VPD */
2963 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2966 rot = get_unaligned_be16(&buffer[4]);
2969 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2970 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2973 if (sdkp->device->type == TYPE_ZBC) {
2975 q->limits.zoned = BLK_ZONED_HM;
2977 sdkp->zoned = (buffer[8] >> 4) & 3;
2978 if (sdkp->zoned == 1)
2980 q->limits.zoned = BLK_ZONED_HA;
2983 * Treat drive-managed devices as
2984 * regular block devices.
2986 q->limits.zoned = BLK_ZONED_NONE;
2988 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2989 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2990 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2997 * sd_read_block_provisioning - Query provisioning VPD page
2998 * @sdkp: disk to query
3000 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3002 unsigned char *buffer;
3003 const int vpd_len = 8;
3005 if (sdkp->lbpme == 0)
3008 buffer = kmalloc(vpd_len, GFP_KERNEL);
3010 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3014 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3015 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3016 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3022 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3024 struct scsi_device *sdev = sdkp->device;
3026 if (sdev->host->no_write_same) {
3027 sdev->no_write_same = 1;
3032 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3033 /* too large values might cause issues with arcmsr */
3034 int vpd_buf_len = 64;
3036 sdev->no_report_opcodes = 1;
3038 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3039 * CODES is unsupported and the device has an ATA
3040 * Information VPD page (SAT).
3042 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3043 sdev->no_write_same = 1;
3046 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3049 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3053 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3055 struct scsi_device *sdev = sdkp->device;
3057 if (!sdev->security_supported)
3060 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3061 SECURITY_PROTOCOL_IN) == 1 &&
3062 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3063 SECURITY_PROTOCOL_OUT) == 1)
3068 * Determine the device's preferred I/O size for reads and writes
3069 * unless the reported value is unreasonably small, large, not a
3070 * multiple of the physical block size, or simply garbage.
3072 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3073 unsigned int dev_max)
3075 struct scsi_device *sdp = sdkp->device;
3076 unsigned int opt_xfer_bytes =
3077 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3079 if (sdkp->opt_xfer_blocks > dev_max) {
3080 sd_first_printk(KERN_WARNING, sdkp,
3081 "Optimal transfer size %u logical blocks " \
3082 "> dev_max (%u logical blocks)\n",
3083 sdkp->opt_xfer_blocks, dev_max);
3087 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3088 sd_first_printk(KERN_WARNING, sdkp,
3089 "Optimal transfer size %u logical blocks " \
3090 "> sd driver limit (%u logical blocks)\n",
3091 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3095 if (opt_xfer_bytes < PAGE_SIZE) {
3096 sd_first_printk(KERN_WARNING, sdkp,
3097 "Optimal transfer size %u bytes < " \
3098 "PAGE_SIZE (%u bytes)\n",
3099 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3103 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3104 sd_first_printk(KERN_WARNING, sdkp,
3105 "Optimal transfer size %u bytes not a " \
3106 "multiple of physical block size (%u bytes)\n",
3107 opt_xfer_bytes, sdkp->physical_block_size);
3111 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3117 * sd_revalidate_disk - called the first time a new disk is seen,
3118 * performs disk spin up, read_capacity, etc.
3119 * @disk: struct gendisk we care about
3121 static int sd_revalidate_disk(struct gendisk *disk)
3123 struct scsi_disk *sdkp = scsi_disk(disk);
3124 struct scsi_device *sdp = sdkp->device;
3125 struct request_queue *q = sdkp->disk->queue;
3126 sector_t old_capacity = sdkp->capacity;
3127 unsigned char *buffer;
3128 unsigned int dev_max, rw_max;
3130 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3131 "sd_revalidate_disk\n"));
3134 * If the device is offline, don't try and read capacity or any
3135 * of the other niceties.
3137 if (!scsi_device_online(sdp))
3140 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3142 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3143 "allocation failure.\n");
3147 sd_spinup_disk(sdkp);
3150 * Without media there is no reason to ask; moreover, some devices
3151 * react badly if we do.
3153 if (sdkp->media_present) {
3154 sd_read_capacity(sdkp, buffer);
3157 * set the default to rotational. All non-rotational devices
3158 * support the block characteristics VPD page, which will
3159 * cause this to be updated correctly and any device which
3160 * doesn't support it should be treated as rotational.
3162 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3163 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3165 if (scsi_device_supports_vpd(sdp)) {
3166 sd_read_block_provisioning(sdkp);
3167 sd_read_block_limits(sdkp);
3168 sd_read_block_characteristics(sdkp);
3169 sd_zbc_read_zones(sdkp, buffer);
3172 sd_print_capacity(sdkp, old_capacity);
3174 sd_read_write_protect_flag(sdkp, buffer);
3175 sd_read_cache_type(sdkp, buffer);
3176 sd_read_app_tag_own(sdkp, buffer);
3177 sd_read_write_same(sdkp, buffer);
3178 sd_read_security(sdkp, buffer);
3182 * We now have all cache related info, determine how we deal
3183 * with flush requests.
3185 sd_set_flush_flag(sdkp);
3187 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3188 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3190 /* Some devices report a maximum block count for READ/WRITE requests. */
3191 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3192 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3194 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3195 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3196 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3198 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3199 (sector_t)BLK_DEF_MAX_SECTORS);
3201 /* Do not exceed controller limit */
3202 rw_max = min(rw_max, queue_max_hw_sectors(q));
3205 * Only update max_sectors if previously unset or if the current value
3206 * exceeds the capabilities of the hardware.
3208 if (sdkp->first_scan ||
3209 q->limits.max_sectors > q->limits.max_dev_sectors ||
3210 q->limits.max_sectors > q->limits.max_hw_sectors)
3211 q->limits.max_sectors = rw_max;
3213 sdkp->first_scan = 0;
3215 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3216 sd_config_write_same(sdkp);
3224 * sd_unlock_native_capacity - unlock native capacity
3225 * @disk: struct gendisk to set capacity for
3227 * Block layer calls this function if it detects that partitions
3228 * on @disk reach beyond the end of the device. If the SCSI host
3229 * implements ->unlock_native_capacity() method, it's invoked to
3230 * give it a chance to adjust the device capacity.
3233 * Defined by block layer. Might sleep.
3235 static void sd_unlock_native_capacity(struct gendisk *disk)
3237 struct scsi_device *sdev = scsi_disk(disk)->device;
3239 if (sdev->host->hostt->unlock_native_capacity)
3240 sdev->host->hostt->unlock_native_capacity(sdev);
3244 * sd_format_disk_name - format disk name
3245 * @prefix: name prefix - ie. "sd" for SCSI disks
3246 * @index: index of the disk to format name for
3247 * @buf: output buffer
3248 * @buflen: length of the output buffer
3250 * SCSI disk names starts at sda. The 26th device is sdz and the
3251 * 27th is sdaa. The last one for two lettered suffix is sdzz
3252 * which is followed by sdaaa.
3254 * This is basically 26 base counting with one extra 'nil' entry
3255 * at the beginning from the second digit on and can be
3256 * determined using similar method as 26 base conversion with the
3257 * index shifted -1 after each digit is computed.
3263 * 0 on success, -errno on failure.
3265 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3267 const int base = 'z' - 'a' + 1;
3268 char *begin = buf + strlen(prefix);
3269 char *end = buf + buflen;
3279 *--p = 'a' + (index % unit);
3280 index = (index / unit) - 1;
3281 } while (index >= 0);
3283 memmove(begin, p, end - p);
3284 memcpy(buf, prefix, strlen(prefix));
3290 * The asynchronous part of sd_probe
3292 static void sd_probe_async(void *data, async_cookie_t cookie)
3294 struct scsi_disk *sdkp = data;
3295 struct scsi_device *sdp;
3302 index = sdkp->index;
3303 dev = &sdp->sdev_gendev;
3305 gd->major = sd_major((index & 0xf0) >> 4);
3306 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3308 gd->fops = &sd_fops;
3309 gd->private_data = &sdkp->driver;
3310 gd->queue = sdkp->device->request_queue;
3312 /* defaults, until the device tells us otherwise */
3313 sdp->sector_size = 512;
3315 sdkp->media_present = 1;
3316 sdkp->write_prot = 0;
3317 sdkp->cache_override = 0;
3321 sdkp->first_scan = 1;
3322 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3324 sd_revalidate_disk(gd);
3326 gd->flags = GENHD_FL_EXT_DEVT;
3327 if (sdp->removable) {
3328 gd->flags |= GENHD_FL_REMOVABLE;
3329 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3332 blk_pm_runtime_init(sdp->request_queue, dev);
3333 device_add_disk(dev, gd, NULL);
3335 sd_dif_config_host(sdkp);
3337 sd_revalidate_disk(gd);
3339 if (sdkp->security) {
3340 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3342 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3345 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3346 sdp->removable ? "removable " : "");
3347 scsi_autopm_put_device(sdp);
3348 put_device(&sdkp->dev);
3352 * sd_probe - called during driver initialization and whenever a
3353 * new scsi device is attached to the system. It is called once
3354 * for each scsi device (not just disks) present.
3355 * @dev: pointer to device object
3357 * Returns 0 if successful (or not interested in this scsi device
3358 * (e.g. scanner)); 1 when there is an error.
3360 * Note: this function is invoked from the scsi mid-level.
3361 * This function sets up the mapping between a given
3362 * <host,channel,id,lun> (found in sdp) and new device name
3363 * (e.g. /dev/sda). More precisely it is the block device major
3364 * and minor number that is chosen here.
3366 * Assume sd_probe is not re-entrant (for time being)
3367 * Also think about sd_probe() and sd_remove() running coincidentally.
3369 static int sd_probe(struct device *dev)
3371 struct scsi_device *sdp = to_scsi_device(dev);
3372 struct scsi_disk *sdkp;
3377 scsi_autopm_get_device(sdp);
3379 if (sdp->type != TYPE_DISK &&
3380 sdp->type != TYPE_ZBC &&
3381 sdp->type != TYPE_MOD &&
3382 sdp->type != TYPE_RBC)
3385 #ifndef CONFIG_BLK_DEV_ZONED
3386 if (sdp->type == TYPE_ZBC)
3389 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3393 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3397 gd = alloc_disk(SD_MINORS);
3401 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3403 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3407 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3409 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3410 goto out_free_index;
3414 sdkp->driver = &sd_template;
3416 sdkp->index = index;
3417 atomic_set(&sdkp->openers, 0);
3418 atomic_set(&sdkp->device->ioerr_cnt, 0);
3420 if (!sdp->request_queue->rq_timeout) {
3421 if (sdp->type != TYPE_MOD)
3422 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3424 blk_queue_rq_timeout(sdp->request_queue,
3428 device_initialize(&sdkp->dev);
3429 sdkp->dev.parent = dev;
3430 sdkp->dev.class = &sd_disk_class;
3431 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3433 error = device_add(&sdkp->dev);
3435 goto out_free_index;
3438 dev_set_drvdata(dev, sdkp);
3440 get_device(&sdkp->dev); /* prevent release before async_schedule */
3441 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3446 ida_free(&sd_index_ida, index);
3452 scsi_autopm_put_device(sdp);
3457 * sd_remove - called whenever a scsi disk (previously recognized by
3458 * sd_probe) is detached from the system. It is called (potentially
3459 * multiple times) during sd module unload.
3460 * @dev: pointer to device object
3462 * Note: this function is invoked from the scsi mid-level.
3463 * This function potentially frees up a device name (e.g. /dev/sdc)
3464 * that could be re-used by a subsequent sd_probe().
3465 * This function is not called when the built-in sd driver is "exit-ed".
3467 static int sd_remove(struct device *dev)
3469 struct scsi_disk *sdkp;
3472 sdkp = dev_get_drvdata(dev);
3473 devt = disk_devt(sdkp->disk);
3474 scsi_autopm_get_device(sdkp->device);
3476 async_synchronize_full_domain(&scsi_sd_pm_domain);
3477 async_synchronize_full_domain(&scsi_sd_probe_domain);
3478 device_del(&sdkp->dev);
3479 del_gendisk(sdkp->disk);
3482 free_opal_dev(sdkp->opal_dev);
3484 blk_register_region(devt, SD_MINORS, NULL,
3485 sd_default_probe, NULL, NULL);
3487 mutex_lock(&sd_ref_mutex);
3488 dev_set_drvdata(dev, NULL);
3489 put_device(&sdkp->dev);
3490 mutex_unlock(&sd_ref_mutex);
3496 * scsi_disk_release - Called to free the scsi_disk structure
3497 * @dev: pointer to embedded class device
3499 * sd_ref_mutex must be held entering this routine. Because it is
3500 * called on last put, you should always use the scsi_disk_get()
3501 * scsi_disk_put() helpers which manipulate the semaphore directly
3502 * and never do a direct put_device.
3504 static void scsi_disk_release(struct device *dev)
3506 struct scsi_disk *sdkp = to_scsi_disk(dev);
3507 struct gendisk *disk = sdkp->disk;
3509 ida_free(&sd_index_ida, sdkp->index);
3511 disk->private_data = NULL;
3513 put_device(&sdkp->device->sdev_gendev);
3518 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3520 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3521 struct scsi_sense_hdr sshdr;
3522 struct scsi_device *sdp = sdkp->device;
3526 cmd[4] |= 1; /* START */
3528 if (sdp->start_stop_pwr_cond)
3529 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3531 if (!scsi_device_online(sdp))
3534 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3535 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3537 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3538 if (driver_byte(res) == DRIVER_SENSE)
3539 sd_print_sense_hdr(sdkp, &sshdr);
3540 if (scsi_sense_valid(&sshdr) &&
3541 /* 0x3a is medium not present */
3546 /* SCSI error codes must not go to the generic layer */
3554 * Send a SYNCHRONIZE CACHE instruction down to the device through
3555 * the normal SCSI command structure. Wait for the command to
3558 static void sd_shutdown(struct device *dev)
3560 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3563 return; /* this can happen */
3565 if (pm_runtime_suspended(dev))
3568 if (sdkp->WCE && sdkp->media_present) {
3569 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3570 sd_sync_cache(sdkp, NULL);
3573 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3574 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3575 sd_start_stop_device(sdkp, 0);
3579 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3581 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3582 struct scsi_sense_hdr sshdr;
3585 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3588 if (sdkp->WCE && sdkp->media_present) {
3589 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3590 ret = sd_sync_cache(sdkp, &sshdr);
3593 /* ignore OFFLINE device */
3597 if (!scsi_sense_valid(&sshdr) ||
3598 sshdr.sense_key != ILLEGAL_REQUEST)
3602 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3603 * doesn't support sync. There's not much to do and
3604 * suspend shouldn't fail.
3610 if (sdkp->device->manage_start_stop) {
3611 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3612 /* an error is not worth aborting a system sleep */
3613 ret = sd_start_stop_device(sdkp, 0);
3614 if (ignore_stop_errors)
3621 static int sd_suspend_system(struct device *dev)
3623 return sd_suspend_common(dev, true);
3626 static int sd_suspend_runtime(struct device *dev)
3628 return sd_suspend_common(dev, false);
3631 static int sd_resume(struct device *dev)
3633 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3636 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3639 if (!sdkp->device->manage_start_stop)
3642 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3643 ret = sd_start_stop_device(sdkp, 1);
3645 opal_unlock_from_suspend(sdkp->opal_dev);
3650 * init_sd - entry point for this driver (both when built in or when
3653 * Note: this function registers this driver with the scsi mid-level.
3655 static int __init init_sd(void)
3657 int majors = 0, i, err;
3659 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3661 for (i = 0; i < SD_MAJORS; i++) {
3662 if (register_blkdev(sd_major(i), "sd") != 0)
3665 blk_register_region(sd_major(i), SD_MINORS, NULL,
3666 sd_default_probe, NULL, NULL);
3672 err = class_register(&sd_disk_class);
3676 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3678 if (!sd_cdb_cache) {
3679 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3684 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3686 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3691 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3692 if (!sd_page_pool) {
3693 printk(KERN_ERR "sd: can't init discard page pool\n");
3698 err = scsi_register_driver(&sd_template.gendrv);
3700 goto err_out_driver;
3705 mempool_destroy(sd_page_pool);
3708 mempool_destroy(sd_cdb_pool);
3711 kmem_cache_destroy(sd_cdb_cache);
3714 class_unregister(&sd_disk_class);
3716 for (i = 0; i < SD_MAJORS; i++)
3717 unregister_blkdev(sd_major(i), "sd");
3722 * exit_sd - exit point for this driver (when it is a module).
3724 * Note: this function unregisters this driver from the scsi mid-level.
3726 static void __exit exit_sd(void)
3730 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3732 scsi_unregister_driver(&sd_template.gendrv);
3733 mempool_destroy(sd_cdb_pool);
3734 mempool_destroy(sd_page_pool);
3735 kmem_cache_destroy(sd_cdb_cache);
3737 class_unregister(&sd_disk_class);
3739 for (i = 0; i < SD_MAJORS; i++) {
3740 blk_unregister_region(sd_major(i), SD_MINORS);
3741 unregister_blkdev(sd_major(i), "sd");
3745 module_init(init_sd);
3746 module_exit(exit_sd);
3748 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3749 struct scsi_sense_hdr *sshdr)
3751 scsi_print_sense_hdr(sdkp->device,
3752 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3755 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3758 const char *hb_string = scsi_hostbyte_string(result);
3759 const char *db_string = scsi_driverbyte_string(result);
3761 if (hb_string || db_string)
3762 sd_printk(KERN_INFO, sdkp,
3763 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3764 hb_string ? hb_string : "invalid",
3765 db_string ? db_string : "invalid");
3767 sd_printk(KERN_INFO, sdkp,
3768 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3769 msg, host_byte(result), driver_byte(result));