1 // SPDX-License-Identifier: GPL-2.0-only
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
15 * eight major numbers.
18 * sd_init and cleanups.
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
25 * Support 32k/1M disks.
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
36 #include <linux/module.h>
38 #include <linux/kernel.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
107 static void sd_config_discard(struct scsi_disk *, unsigned int);
108 static void sd_config_write_same(struct scsi_disk *);
109 static int sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static int sd_probe(struct device *);
112 static int sd_remove(struct device *);
113 static void sd_shutdown(struct device *);
114 static int sd_suspend_system(struct device *);
115 static int sd_suspend_runtime(struct device *);
116 static int sd_resume(struct device *);
117 static void sd_rescan(struct device *);
118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120 static int sd_done(struct scsi_cmnd *);
121 static void sd_eh_reset(struct scsi_cmnd *);
122 static int sd_eh_action(struct scsi_cmnd *, int);
123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124 static void scsi_disk_release(struct device *cdev);
125 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
126 static void sd_print_result(const struct scsi_disk *, const char *, int);
128 static DEFINE_IDA(sd_index_ida);
130 /* This semaphore is used to mediate the 0->1 reference get in the
131 * face of object destruction (i.e. we can't allow a get on an
132 * object after last put) */
133 static DEFINE_MUTEX(sd_ref_mutex);
135 static struct kmem_cache *sd_cdb_cache;
136 static mempool_t *sd_cdb_pool;
137 static mempool_t *sd_page_pool;
139 static const char *sd_cache_types[] = {
140 "write through", "none", "write back",
141 "write back, no read (daft)"
144 static void sd_set_flush_flag(struct scsi_disk *sdkp)
146 bool wc = false, fua = false;
154 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
158 cache_type_store(struct device *dev, struct device_attribute *attr,
159 const char *buf, size_t count)
161 int ct, rcd, wce, sp;
162 struct scsi_disk *sdkp = to_scsi_disk(dev);
163 struct scsi_device *sdp = sdkp->device;
166 struct scsi_mode_data data;
167 struct scsi_sense_hdr sshdr;
168 static const char temp[] = "temporary ";
171 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
172 /* no cache control on RBC devices; theoretically they
173 * can do it, but there's probably so many exceptions
174 * it's not worth the risk */
177 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
178 buf += sizeof(temp) - 1;
179 sdkp->cache_override = 1;
181 sdkp->cache_override = 0;
184 ct = sysfs_match_string(sd_cache_types, buf);
188 rcd = ct & 0x01 ? 1 : 0;
189 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
191 if (sdkp->cache_override) {
194 sd_set_flush_flag(sdkp);
198 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 SD_MAX_RETRIES, &data, NULL))
201 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 data.block_descriptor_length);
203 buffer_data = buffer + data.header_length +
204 data.block_descriptor_length;
205 buffer_data[2] &= ~0x05;
206 buffer_data[2] |= wce << 2 | rcd;
207 sp = buffer_data[0] & 0x80 ? 1 : 0;
208 buffer_data[0] &= ~0x80;
211 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
212 * received mode parameter buffer before doing MODE SELECT.
214 data.device_specific = 0;
216 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
217 SD_MAX_RETRIES, &data, &sshdr)) {
218 if (scsi_sense_valid(&sshdr))
219 sd_print_sense_hdr(sdkp, &sshdr);
222 revalidate_disk(sdkp->disk);
227 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
230 struct scsi_disk *sdkp = to_scsi_disk(dev);
231 struct scsi_device *sdp = sdkp->device;
233 return sprintf(buf, "%u\n", sdp->manage_start_stop);
237 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t count)
240 struct scsi_disk *sdkp = to_scsi_disk(dev);
241 struct scsi_device *sdp = sdkp->device;
244 if (!capable(CAP_SYS_ADMIN))
247 if (kstrtobool(buf, &v))
250 sdp->manage_start_stop = v;
254 static DEVICE_ATTR_RW(manage_start_stop);
257 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
259 struct scsi_disk *sdkp = to_scsi_disk(dev);
261 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
265 allow_restart_store(struct device *dev, struct device_attribute *attr,
266 const char *buf, size_t count)
269 struct scsi_disk *sdkp = to_scsi_disk(dev);
270 struct scsi_device *sdp = sdkp->device;
272 if (!capable(CAP_SYS_ADMIN))
275 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
278 if (kstrtobool(buf, &v))
281 sdp->allow_restart = v;
285 static DEVICE_ATTR_RW(allow_restart);
288 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
290 struct scsi_disk *sdkp = to_scsi_disk(dev);
291 int ct = sdkp->RCD + 2*sdkp->WCE;
293 return sprintf(buf, "%s\n", sd_cache_types[ct]);
295 static DEVICE_ATTR_RW(cache_type);
298 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
300 struct scsi_disk *sdkp = to_scsi_disk(dev);
302 return sprintf(buf, "%u\n", sdkp->DPOFUA);
304 static DEVICE_ATTR_RO(FUA);
307 protection_type_show(struct device *dev, struct device_attribute *attr,
310 struct scsi_disk *sdkp = to_scsi_disk(dev);
312 return sprintf(buf, "%u\n", sdkp->protection_type);
316 protection_type_store(struct device *dev, struct device_attribute *attr,
317 const char *buf, size_t count)
319 struct scsi_disk *sdkp = to_scsi_disk(dev);
323 if (!capable(CAP_SYS_ADMIN))
326 err = kstrtouint(buf, 10, &val);
331 if (val <= T10_PI_TYPE3_PROTECTION)
332 sdkp->protection_type = val;
336 static DEVICE_ATTR_RW(protection_type);
339 protection_mode_show(struct device *dev, struct device_attribute *attr,
342 struct scsi_disk *sdkp = to_scsi_disk(dev);
343 struct scsi_device *sdp = sdkp->device;
344 unsigned int dif, dix;
346 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
347 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
349 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
355 return sprintf(buf, "none\n");
357 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
359 static DEVICE_ATTR_RO(protection_mode);
362 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
366 return sprintf(buf, "%u\n", sdkp->ATO);
368 static DEVICE_ATTR_RO(app_tag_own);
371 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
374 struct scsi_disk *sdkp = to_scsi_disk(dev);
376 return sprintf(buf, "%u\n", sdkp->lbpme);
378 static DEVICE_ATTR_RO(thin_provisioning);
380 /* sysfs_match_string() requires dense arrays */
381 static const char *lbp_mode[] = {
382 [SD_LBP_FULL] = "full",
383 [SD_LBP_UNMAP] = "unmap",
384 [SD_LBP_WS16] = "writesame_16",
385 [SD_LBP_WS10] = "writesame_10",
386 [SD_LBP_ZERO] = "writesame_zero",
387 [SD_LBP_DISABLE] = "disabled",
391 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
394 struct scsi_disk *sdkp = to_scsi_disk(dev);
396 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
400 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
401 const char *buf, size_t count)
403 struct scsi_disk *sdkp = to_scsi_disk(dev);
404 struct scsi_device *sdp = sdkp->device;
407 if (!capable(CAP_SYS_ADMIN))
410 if (sd_is_zoned(sdkp)) {
411 sd_config_discard(sdkp, SD_LBP_DISABLE);
415 if (sdp->type != TYPE_DISK)
418 mode = sysfs_match_string(lbp_mode, buf);
422 sd_config_discard(sdkp, mode);
426 static DEVICE_ATTR_RW(provisioning_mode);
428 /* sysfs_match_string() requires dense arrays */
429 static const char *zeroing_mode[] = {
430 [SD_ZERO_WRITE] = "write",
431 [SD_ZERO_WS] = "writesame",
432 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
433 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
437 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
440 struct scsi_disk *sdkp = to_scsi_disk(dev);
442 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
446 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
447 const char *buf, size_t count)
449 struct scsi_disk *sdkp = to_scsi_disk(dev);
452 if (!capable(CAP_SYS_ADMIN))
455 mode = sysfs_match_string(zeroing_mode, buf);
459 sdkp->zeroing_mode = mode;
463 static DEVICE_ATTR_RW(zeroing_mode);
466 max_medium_access_timeouts_show(struct device *dev,
467 struct device_attribute *attr, char *buf)
469 struct scsi_disk *sdkp = to_scsi_disk(dev);
471 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
475 max_medium_access_timeouts_store(struct device *dev,
476 struct device_attribute *attr, const char *buf,
479 struct scsi_disk *sdkp = to_scsi_disk(dev);
482 if (!capable(CAP_SYS_ADMIN))
485 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
487 return err ? err : count;
489 static DEVICE_ATTR_RW(max_medium_access_timeouts);
492 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
495 struct scsi_disk *sdkp = to_scsi_disk(dev);
497 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
501 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
502 const char *buf, size_t count)
504 struct scsi_disk *sdkp = to_scsi_disk(dev);
505 struct scsi_device *sdp = sdkp->device;
509 if (!capable(CAP_SYS_ADMIN))
512 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
515 err = kstrtoul(buf, 10, &max);
521 sdp->no_write_same = 1;
522 else if (max <= SD_MAX_WS16_BLOCKS) {
523 sdp->no_write_same = 0;
524 sdkp->max_ws_blocks = max;
527 sd_config_write_same(sdkp);
531 static DEVICE_ATTR_RW(max_write_same_blocks);
533 static struct attribute *sd_disk_attrs[] = {
534 &dev_attr_cache_type.attr,
536 &dev_attr_allow_restart.attr,
537 &dev_attr_manage_start_stop.attr,
538 &dev_attr_protection_type.attr,
539 &dev_attr_protection_mode.attr,
540 &dev_attr_app_tag_own.attr,
541 &dev_attr_thin_provisioning.attr,
542 &dev_attr_provisioning_mode.attr,
543 &dev_attr_zeroing_mode.attr,
544 &dev_attr_max_write_same_blocks.attr,
545 &dev_attr_max_medium_access_timeouts.attr,
548 ATTRIBUTE_GROUPS(sd_disk);
550 static struct class sd_disk_class = {
552 .owner = THIS_MODULE,
553 .dev_release = scsi_disk_release,
554 .dev_groups = sd_disk_groups,
557 static const struct dev_pm_ops sd_pm_ops = {
558 .suspend = sd_suspend_system,
560 .poweroff = sd_suspend_system,
561 .restore = sd_resume,
562 .runtime_suspend = sd_suspend_runtime,
563 .runtime_resume = sd_resume,
566 static struct scsi_driver sd_template = {
569 .owner = THIS_MODULE,
572 .shutdown = sd_shutdown,
576 .init_command = sd_init_command,
577 .uninit_command = sd_uninit_command,
579 .eh_action = sd_eh_action,
580 .eh_reset = sd_eh_reset,
584 * Dummy kobj_map->probe function.
585 * The default ->probe function will call modprobe, which is
586 * pointless as this module is already loaded.
588 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
594 * Device no to disk mapping:
596 * major disc2 disc p1
597 * |............|.............|....|....| <- dev_t
600 * Inside a major, we have 16k disks, however mapped non-
601 * contiguously. The first 16 disks are for major0, the next
602 * ones with major1, ... Disk 256 is for major0 again, disk 272
604 * As we stay compatible with our numbering scheme, we can reuse
605 * the well-know SCSI majors 8, 65--71, 136--143.
607 static int sd_major(int major_idx)
611 return SCSI_DISK0_MAJOR;
613 return SCSI_DISK1_MAJOR + major_idx - 1;
615 return SCSI_DISK8_MAJOR + major_idx - 8;
618 return 0; /* shut up gcc */
622 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
624 struct scsi_disk *sdkp = NULL;
626 mutex_lock(&sd_ref_mutex);
628 if (disk->private_data) {
629 sdkp = scsi_disk(disk);
630 if (scsi_device_get(sdkp->device) == 0)
631 get_device(&sdkp->dev);
635 mutex_unlock(&sd_ref_mutex);
639 static void scsi_disk_put(struct scsi_disk *sdkp)
641 struct scsi_device *sdev = sdkp->device;
643 mutex_lock(&sd_ref_mutex);
644 put_device(&sdkp->dev);
645 scsi_device_put(sdev);
646 mutex_unlock(&sd_ref_mutex);
649 #ifdef CONFIG_BLK_SED_OPAL
650 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
651 size_t len, bool send)
653 struct scsi_device *sdev = data;
657 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
659 put_unaligned_be16(spsp, &cdb[2]);
660 put_unaligned_be32(len, &cdb[6]);
662 ret = scsi_execute_req(sdev, cdb,
663 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
664 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
665 return ret <= 0 ? ret : -EIO;
667 #endif /* CONFIG_BLK_SED_OPAL */
670 * Look up the DIX operation based on whether the command is read or
671 * write and whether dix and dif are enabled.
673 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
675 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
676 static const unsigned int ops[] = { /* wrt dix dif */
677 SCSI_PROT_NORMAL, /* 0 0 0 */
678 SCSI_PROT_READ_STRIP, /* 0 0 1 */
679 SCSI_PROT_READ_INSERT, /* 0 1 0 */
680 SCSI_PROT_READ_PASS, /* 0 1 1 */
681 SCSI_PROT_NORMAL, /* 1 0 0 */
682 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
683 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
684 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
687 return ops[write << 2 | dix << 1 | dif];
691 * Returns a mask of the protection flags that are valid for a given DIX
694 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
696 static const unsigned int flag_mask[] = {
697 [SCSI_PROT_NORMAL] = 0,
699 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
700 SCSI_PROT_GUARD_CHECK |
701 SCSI_PROT_REF_CHECK |
702 SCSI_PROT_REF_INCREMENT,
704 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
705 SCSI_PROT_IP_CHECKSUM,
707 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
708 SCSI_PROT_GUARD_CHECK |
709 SCSI_PROT_REF_CHECK |
710 SCSI_PROT_REF_INCREMENT |
711 SCSI_PROT_IP_CHECKSUM,
713 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
714 SCSI_PROT_REF_INCREMENT,
716 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
717 SCSI_PROT_REF_CHECK |
718 SCSI_PROT_REF_INCREMENT |
719 SCSI_PROT_IP_CHECKSUM,
721 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
722 SCSI_PROT_GUARD_CHECK |
723 SCSI_PROT_REF_CHECK |
724 SCSI_PROT_REF_INCREMENT |
725 SCSI_PROT_IP_CHECKSUM,
728 return flag_mask[prot_op];
731 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
732 unsigned int dix, unsigned int dif)
734 struct bio *bio = scmd->request->bio;
735 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
736 unsigned int protect = 0;
738 if (dix) { /* DIX Type 0, 1, 2, 3 */
739 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
740 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
742 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
743 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
746 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
747 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
749 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
750 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
753 if (dif) { /* DIX/DIF Type 1, 2, 3 */
754 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
756 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
757 protect = 3 << 5; /* Disable target PI checking */
759 protect = 1 << 5; /* Enable target PI checking */
762 scsi_set_prot_op(scmd, prot_op);
763 scsi_set_prot_type(scmd, dif);
764 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
769 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
771 struct request_queue *q = sdkp->disk->queue;
772 unsigned int logical_block_size = sdkp->device->sector_size;
773 unsigned int max_blocks = 0;
775 q->limits.discard_alignment =
776 sdkp->unmap_alignment * logical_block_size;
777 q->limits.discard_granularity =
778 max(sdkp->physical_block_size,
779 sdkp->unmap_granularity * logical_block_size);
780 sdkp->provisioning_mode = mode;
786 blk_queue_max_discard_sectors(q, 0);
787 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
791 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
792 (u32)SD_MAX_WS16_BLOCKS);
796 if (sdkp->device->unmap_limit_for_ws)
797 max_blocks = sdkp->max_unmap_blocks;
799 max_blocks = sdkp->max_ws_blocks;
801 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
805 if (sdkp->device->unmap_limit_for_ws)
806 max_blocks = sdkp->max_unmap_blocks;
808 max_blocks = sdkp->max_ws_blocks;
810 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
814 max_blocks = min_not_zero(sdkp->max_ws_blocks,
815 (u32)SD_MAX_WS10_BLOCKS);
819 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
820 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
823 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
825 struct scsi_device *sdp = cmd->device;
826 struct request *rq = cmd->request;
827 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
828 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
829 unsigned int data_len = 24;
832 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
833 if (!rq->special_vec.bv_page)
834 return BLK_STS_RESOURCE;
835 clear_highpage(rq->special_vec.bv_page);
836 rq->special_vec.bv_offset = 0;
837 rq->special_vec.bv_len = data_len;
838 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
841 cmd->cmnd[0] = UNMAP;
844 buf = page_address(rq->special_vec.bv_page);
845 put_unaligned_be16(6 + 16, &buf[0]);
846 put_unaligned_be16(16, &buf[2]);
847 put_unaligned_be64(lba, &buf[8]);
848 put_unaligned_be32(nr_blocks, &buf[16]);
850 cmd->allowed = SD_MAX_RETRIES;
851 cmd->transfersize = data_len;
852 rq->timeout = SD_TIMEOUT;
854 return scsi_init_io(cmd);
857 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
860 struct scsi_device *sdp = cmd->device;
861 struct request *rq = cmd->request;
862 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
863 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
864 u32 data_len = sdp->sector_size;
866 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
867 if (!rq->special_vec.bv_page)
868 return BLK_STS_RESOURCE;
869 clear_highpage(rq->special_vec.bv_page);
870 rq->special_vec.bv_offset = 0;
871 rq->special_vec.bv_len = data_len;
872 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
875 cmd->cmnd[0] = WRITE_SAME_16;
877 cmd->cmnd[1] = 0x8; /* UNMAP */
878 put_unaligned_be64(lba, &cmd->cmnd[2]);
879 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
881 cmd->allowed = SD_MAX_RETRIES;
882 cmd->transfersize = data_len;
883 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
885 return scsi_init_io(cmd);
888 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
891 struct scsi_device *sdp = cmd->device;
892 struct request *rq = cmd->request;
893 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
894 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
895 u32 data_len = sdp->sector_size;
897 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
898 if (!rq->special_vec.bv_page)
899 return BLK_STS_RESOURCE;
900 clear_highpage(rq->special_vec.bv_page);
901 rq->special_vec.bv_offset = 0;
902 rq->special_vec.bv_len = data_len;
903 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
906 cmd->cmnd[0] = WRITE_SAME;
908 cmd->cmnd[1] = 0x8; /* UNMAP */
909 put_unaligned_be32(lba, &cmd->cmnd[2]);
910 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
912 cmd->allowed = SD_MAX_RETRIES;
913 cmd->transfersize = data_len;
914 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
916 return scsi_init_io(cmd);
919 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
921 struct request *rq = cmd->request;
922 struct scsi_device *sdp = cmd->device;
923 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
924 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
925 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
927 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
928 switch (sdkp->zeroing_mode) {
929 case SD_ZERO_WS16_UNMAP:
930 return sd_setup_write_same16_cmnd(cmd, true);
931 case SD_ZERO_WS10_UNMAP:
932 return sd_setup_write_same10_cmnd(cmd, true);
936 if (sdp->no_write_same)
937 return BLK_STS_TARGET;
939 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
940 return sd_setup_write_same16_cmnd(cmd, false);
942 return sd_setup_write_same10_cmnd(cmd, false);
945 static void sd_config_write_same(struct scsi_disk *sdkp)
947 struct request_queue *q = sdkp->disk->queue;
948 unsigned int logical_block_size = sdkp->device->sector_size;
950 if (sdkp->device->no_write_same) {
951 sdkp->max_ws_blocks = 0;
955 /* Some devices can not handle block counts above 0xffff despite
956 * supporting WRITE SAME(16). Consequently we default to 64k
957 * blocks per I/O unless the device explicitly advertises a
960 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
961 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
962 (u32)SD_MAX_WS16_BLOCKS);
963 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
964 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
965 (u32)SD_MAX_WS10_BLOCKS);
967 sdkp->device->no_write_same = 1;
968 sdkp->max_ws_blocks = 0;
971 if (sdkp->lbprz && sdkp->lbpws)
972 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
973 else if (sdkp->lbprz && sdkp->lbpws10)
974 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
975 else if (sdkp->max_ws_blocks)
976 sdkp->zeroing_mode = SD_ZERO_WS;
978 sdkp->zeroing_mode = SD_ZERO_WRITE;
980 if (sdkp->max_ws_blocks &&
981 sdkp->physical_block_size > logical_block_size) {
983 * Reporting a maximum number of blocks that is not aligned
984 * on the device physical size would cause a large write same
985 * request to be split into physically unaligned chunks by
986 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
987 * even if the caller of these functions took care to align the
988 * large request. So make sure the maximum reported is aligned
989 * to the device physical block size. This is only an optional
990 * optimization for regular disks, but this is mandatory to
991 * avoid failure of large write same requests directed at
992 * sequential write required zones of host-managed ZBC disks.
994 sdkp->max_ws_blocks =
995 round_down(sdkp->max_ws_blocks,
996 bytes_to_logical(sdkp->device,
997 sdkp->physical_block_size));
1001 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1002 (logical_block_size >> 9));
1003 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1004 (logical_block_size >> 9));
1008 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1009 * @cmd: command to prepare
1011 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1012 * the preference indicated by the target device.
1014 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1016 struct request *rq = cmd->request;
1017 struct scsi_device *sdp = cmd->device;
1018 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1019 struct bio *bio = rq->bio;
1020 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1021 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1024 if (sdkp->device->no_write_same)
1025 return BLK_STS_TARGET;
1027 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1029 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1031 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1033 cmd->cmnd[0] = WRITE_SAME_16;
1034 put_unaligned_be64(lba, &cmd->cmnd[2]);
1035 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1038 cmd->cmnd[0] = WRITE_SAME;
1039 put_unaligned_be32(lba, &cmd->cmnd[2]);
1040 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1043 cmd->transfersize = sdp->sector_size;
1044 cmd->allowed = SD_MAX_RETRIES;
1047 * For WRITE SAME the data transferred via the DATA OUT buffer is
1048 * different from the amount of data actually written to the target.
1050 * We set up __data_len to the amount of data transferred via the
1051 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1052 * to transfer a single sector of data first, but then reset it to
1053 * the amount of data to be written right after so that the I/O path
1054 * knows how much to actually write.
1056 rq->__data_len = sdp->sector_size;
1057 ret = scsi_init_io(cmd);
1058 rq->__data_len = blk_rq_bytes(rq);
1063 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1065 struct request *rq = cmd->request;
1067 /* flush requests don't perform I/O, zero the S/G table */
1068 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1072 cmd->transfersize = 0;
1073 cmd->allowed = SD_MAX_RETRIES;
1075 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1079 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1080 sector_t lba, unsigned int nr_blocks,
1081 unsigned char flags)
1083 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1084 if (unlikely(cmd->cmnd == NULL))
1085 return BLK_STS_RESOURCE;
1087 cmd->cmd_len = SD_EXT_CDB_SIZE;
1088 memset(cmd->cmnd, 0, cmd->cmd_len);
1090 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1091 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1092 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1093 cmd->cmnd[10] = flags;
1094 put_unaligned_be64(lba, &cmd->cmnd[12]);
1095 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1096 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1101 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1102 sector_t lba, unsigned int nr_blocks,
1103 unsigned char flags)
1106 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1107 cmd->cmnd[1] = flags;
1110 put_unaligned_be64(lba, &cmd->cmnd[2]);
1111 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1116 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1117 sector_t lba, unsigned int nr_blocks,
1118 unsigned char flags)
1121 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1122 cmd->cmnd[1] = flags;
1125 put_unaligned_be32(lba, &cmd->cmnd[2]);
1126 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1131 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1132 sector_t lba, unsigned int nr_blocks,
1133 unsigned char flags)
1135 /* Avoid that 0 blocks gets translated into 256 blocks. */
1136 if (WARN_ON_ONCE(nr_blocks == 0))
1137 return BLK_STS_IOERR;
1139 if (unlikely(flags & 0x8)) {
1141 * This happens only if this drive failed 10byte rw
1142 * command with ILLEGAL_REQUEST during operation and
1143 * thus turned off use_10_for_rw.
1145 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1146 return BLK_STS_IOERR;
1150 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1151 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1152 cmd->cmnd[2] = (lba >> 8) & 0xff;
1153 cmd->cmnd[3] = lba & 0xff;
1154 cmd->cmnd[4] = nr_blocks;
1160 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1162 struct request *rq = cmd->request;
1163 struct scsi_device *sdp = cmd->device;
1164 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1165 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1167 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1169 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170 bool write = rq_data_dir(rq) == WRITE;
1171 unsigned char protect, fua;
1174 ret = scsi_init_io(cmd);
1175 if (ret != BLK_STS_OK)
1178 if (!scsi_device_online(sdp) || sdp->changed) {
1179 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1180 return BLK_STS_IOERR;
1183 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1184 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1185 return BLK_STS_IOERR;
1188 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1189 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1190 return BLK_STS_IOERR;
1194 * Some SD card readers can't handle accesses which touch the
1195 * last one or two logical blocks. Split accesses as needed.
1197 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1199 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1200 if (lba < threshold) {
1201 /* Access up to the threshold but not beyond */
1202 nr_blocks = threshold - lba;
1204 /* Access only a single logical block */
1209 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1210 dix = scsi_prot_sg_count(cmd);
1211 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1214 t10_pi_prepare(cmd->request, sdkp->protection_type);
1217 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1221 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1222 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1224 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1225 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1227 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1228 sdp->use_10_for_rw || protect) {
1229 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1232 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1236 if (unlikely(ret != BLK_STS_OK))
1240 * We shouldn't disconnect in the middle of a sector, so with a dumb
1241 * host adapter, it's safe to assume that we can at least transfer
1242 * this many bytes between each connect / disconnect.
1244 cmd->transfersize = sdp->sector_size;
1245 cmd->underflow = nr_blocks << 9;
1246 cmd->allowed = SD_MAX_RETRIES;
1247 cmd->sdb.length = nr_blocks * sdp->sector_size;
1250 scmd_printk(KERN_INFO, cmd,
1251 "%s: block=%llu, count=%d\n", __func__,
1252 (unsigned long long)blk_rq_pos(rq),
1253 blk_rq_sectors(rq)));
1255 scmd_printk(KERN_INFO, cmd,
1256 "%s %d/%u 512 byte blocks.\n",
1257 write ? "writing" : "reading", nr_blocks,
1258 blk_rq_sectors(rq)));
1261 * This indicates that the command is ready from our end to be
1267 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1269 struct request *rq = cmd->request;
1271 switch (req_op(rq)) {
1272 case REQ_OP_DISCARD:
1273 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1275 return sd_setup_unmap_cmnd(cmd);
1277 return sd_setup_write_same16_cmnd(cmd, true);
1279 return sd_setup_write_same10_cmnd(cmd, true);
1281 return sd_setup_write_same10_cmnd(cmd, false);
1283 return BLK_STS_TARGET;
1285 case REQ_OP_WRITE_ZEROES:
1286 return sd_setup_write_zeroes_cmnd(cmd);
1287 case REQ_OP_WRITE_SAME:
1288 return sd_setup_write_same_cmnd(cmd);
1290 return sd_setup_flush_cmnd(cmd);
1293 return sd_setup_read_write_cmnd(cmd);
1294 case REQ_OP_ZONE_RESET:
1295 return sd_zbc_setup_reset_cmnd(cmd);
1298 return BLK_STS_NOTSUPP;
1302 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1304 struct request *rq = SCpnt->request;
1307 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1308 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1310 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1314 mempool_free(cmnd, sd_cdb_pool);
1319 * sd_open - open a scsi disk device
1320 * @bdev: Block device of the scsi disk to open
1321 * @mode: FMODE_* mask
1323 * Returns 0 if successful. Returns a negated errno value in case
1326 * Note: This can be called from a user context (e.g. fsck(1) )
1327 * or from within the kernel (e.g. as a result of a mount(1) ).
1328 * In the latter case @inode and @filp carry an abridged amount
1329 * of information as noted above.
1331 * Locking: called with bdev->bd_mutex held.
1333 static int sd_open(struct block_device *bdev, fmode_t mode)
1335 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1336 struct scsi_device *sdev;
1342 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1344 sdev = sdkp->device;
1347 * If the device is in error recovery, wait until it is done.
1348 * If the device is offline, then disallow any access to it.
1351 if (!scsi_block_when_processing_errors(sdev))
1354 if (sdev->removable || sdkp->write_prot)
1355 check_disk_change(bdev);
1358 * If the drive is empty, just let the open fail.
1360 retval = -ENOMEDIUM;
1361 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1365 * If the device has the write protect tab set, have the open fail
1366 * if the user expects to be able to write to the thing.
1369 if (sdkp->write_prot && (mode & FMODE_WRITE))
1373 * It is possible that the disk changing stuff resulted in
1374 * the device being taken offline. If this is the case,
1375 * report this to the user, and don't pretend that the
1376 * open actually succeeded.
1379 if (!scsi_device_online(sdev))
1382 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1383 if (scsi_block_when_processing_errors(sdev))
1384 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1390 scsi_disk_put(sdkp);
1395 * sd_release - invoked when the (last) close(2) is called on this
1397 * @disk: disk to release
1398 * @mode: FMODE_* mask
1402 * Note: may block (uninterruptible) if error recovery is underway
1405 * Locking: called with bdev->bd_mutex held.
1407 static void sd_release(struct gendisk *disk, fmode_t mode)
1409 struct scsi_disk *sdkp = scsi_disk(disk);
1410 struct scsi_device *sdev = sdkp->device;
1412 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1414 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1415 if (scsi_block_when_processing_errors(sdev))
1416 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1419 scsi_disk_put(sdkp);
1422 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1424 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1425 struct scsi_device *sdp = sdkp->device;
1426 struct Scsi_Host *host = sdp->host;
1427 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1430 /* default to most commonly used values */
1431 diskinfo[0] = 0x40; /* 1 << 6 */
1432 diskinfo[1] = 0x20; /* 1 << 5 */
1433 diskinfo[2] = capacity >> 11;
1435 /* override with calculated, extended default, or driver values */
1436 if (host->hostt->bios_param)
1437 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1439 scsicam_bios_param(bdev, capacity, diskinfo);
1441 geo->heads = diskinfo[0];
1442 geo->sectors = diskinfo[1];
1443 geo->cylinders = diskinfo[2];
1448 * sd_ioctl - process an ioctl
1449 * @bdev: target block device
1450 * @mode: FMODE_* mask
1451 * @cmd: ioctl command number
1452 * @arg: this is third argument given to ioctl(2) system call.
1453 * Often contains a pointer.
1455 * Returns 0 if successful (some ioctls return positive numbers on
1456 * success as well). Returns a negated errno value in case of error.
1458 * Note: most ioctls are forward onto the block subsystem or further
1459 * down in the scsi subsystem.
1461 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1462 unsigned int cmd, unsigned long arg)
1464 struct gendisk *disk = bdev->bd_disk;
1465 struct scsi_disk *sdkp = scsi_disk(disk);
1466 struct scsi_device *sdp = sdkp->device;
1467 void __user *p = (void __user *)arg;
1470 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1471 "cmd=0x%x\n", disk->disk_name, cmd));
1473 error = scsi_verify_blk_ioctl(bdev, cmd);
1478 * If we are in the middle of error recovery, don't let anyone
1479 * else try and use this device. Also, if error recovery fails, it
1480 * may try and take the device offline, in which case all further
1481 * access to the device is prohibited.
1483 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1484 (mode & FMODE_NDELAY) != 0);
1488 if (is_sed_ioctl(cmd))
1489 return sed_ioctl(sdkp->opal_dev, cmd, p);
1492 * Send SCSI addressing ioctls directly to mid level, send other
1493 * ioctls to block level and then onto mid level if they can't be
1497 case SCSI_IOCTL_GET_IDLUN:
1498 case SCSI_IOCTL_GET_BUS_NUMBER:
1499 error = scsi_ioctl(sdp, cmd, p);
1502 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1503 if (error != -ENOTTY)
1505 error = scsi_ioctl(sdp, cmd, p);
1512 static void set_media_not_present(struct scsi_disk *sdkp)
1514 if (sdkp->media_present)
1515 sdkp->device->changed = 1;
1517 if (sdkp->device->removable) {
1518 sdkp->media_present = 0;
1523 static int media_not_present(struct scsi_disk *sdkp,
1524 struct scsi_sense_hdr *sshdr)
1526 if (!scsi_sense_valid(sshdr))
1529 /* not invoked for commands that could return deferred errors */
1530 switch (sshdr->sense_key) {
1531 case UNIT_ATTENTION:
1533 /* medium not present */
1534 if (sshdr->asc == 0x3A) {
1535 set_media_not_present(sdkp);
1543 * sd_check_events - check media events
1544 * @disk: kernel device descriptor
1545 * @clearing: disk events currently being cleared
1547 * Returns mask of DISK_EVENT_*.
1549 * Note: this function is invoked from the block subsystem.
1551 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1553 struct scsi_disk *sdkp = scsi_disk_get(disk);
1554 struct scsi_device *sdp;
1561 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1564 * If the device is offline, don't send any commands - just pretend as
1565 * if the command failed. If the device ever comes back online, we
1566 * can deal with it then. It is only because of unrecoverable errors
1567 * that we would ever take a device offline in the first place.
1569 if (!scsi_device_online(sdp)) {
1570 set_media_not_present(sdkp);
1575 * Using TEST_UNIT_READY enables differentiation between drive with
1576 * no cartridge loaded - NOT READY, drive with changed cartridge -
1577 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1579 * Drives that auto spin down. eg iomega jaz 1G, will be started
1580 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1581 * sd_revalidate() is called.
1583 if (scsi_block_when_processing_errors(sdp)) {
1584 struct scsi_sense_hdr sshdr = { 0, };
1586 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1589 /* failed to execute TUR, assume media not present */
1590 if (host_byte(retval)) {
1591 set_media_not_present(sdkp);
1595 if (media_not_present(sdkp, &sshdr))
1600 * For removable scsi disk we have to recognise the presence
1601 * of a disk in the drive.
1603 if (!sdkp->media_present)
1605 sdkp->media_present = 1;
1608 * sdp->changed is set under the following conditions:
1610 * Medium present state has changed in either direction.
1611 * Device has indicated UNIT_ATTENTION.
1613 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1615 scsi_disk_put(sdkp);
1619 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1622 struct scsi_device *sdp = sdkp->device;
1623 const int timeout = sdp->request_queue->rq_timeout
1624 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1625 struct scsi_sense_hdr my_sshdr;
1627 if (!scsi_device_online(sdp))
1630 /* caller might not be interested in sense, but we need it */
1634 for (retries = 3; retries > 0; --retries) {
1635 unsigned char cmd[10] = { 0 };
1637 cmd[0] = SYNCHRONIZE_CACHE;
1639 * Leave the rest of the command zero to indicate
1642 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1643 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1649 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1651 if (driver_byte(res) == DRIVER_SENSE)
1652 sd_print_sense_hdr(sdkp, sshdr);
1654 /* we need to evaluate the error return */
1655 if (scsi_sense_valid(sshdr) &&
1656 (sshdr->asc == 0x3a || /* medium not present */
1657 sshdr->asc == 0x20)) /* invalid command */
1658 /* this is no error here */
1661 switch (host_byte(res)) {
1662 /* ignore errors due to racing a disconnection */
1663 case DID_BAD_TARGET:
1664 case DID_NO_CONNECT:
1666 /* signal the upper layer it might try again */
1670 case DID_SOFT_ERROR:
1679 static void sd_rescan(struct device *dev)
1681 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1683 revalidate_disk(sdkp->disk);
1687 #ifdef CONFIG_COMPAT
1689 * This gets directly called from VFS. When the ioctl
1690 * is not recognized we go back to the other translation paths.
1692 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1693 unsigned int cmd, unsigned long arg)
1695 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1698 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1699 (mode & FMODE_NDELAY) != 0);
1704 * Let the static ioctl translation table take care of it.
1706 if (!sdev->host->hostt->compat_ioctl)
1707 return -ENOIOCTLCMD;
1708 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1712 static char sd_pr_type(enum pr_type type)
1715 case PR_WRITE_EXCLUSIVE:
1717 case PR_EXCLUSIVE_ACCESS:
1719 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1721 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1723 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1725 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1732 static int sd_pr_command(struct block_device *bdev, u8 sa,
1733 u64 key, u64 sa_key, u8 type, u8 flags)
1735 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1736 struct scsi_sense_hdr sshdr;
1738 u8 cmd[16] = { 0, };
1739 u8 data[24] = { 0, };
1741 cmd[0] = PERSISTENT_RESERVE_OUT;
1744 put_unaligned_be32(sizeof(data), &cmd[5]);
1746 put_unaligned_be64(key, &data[0]);
1747 put_unaligned_be64(sa_key, &data[8]);
1750 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1751 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1753 if (driver_byte(result) == DRIVER_SENSE &&
1754 scsi_sense_valid(&sshdr)) {
1755 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1756 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1762 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1765 if (flags & ~PR_FL_IGNORE_KEY)
1767 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1768 old_key, new_key, 0,
1769 (1 << 0) /* APTPL */);
1772 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1777 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1780 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1782 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1785 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1786 enum pr_type type, bool abort)
1788 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1789 sd_pr_type(type), 0);
1792 static int sd_pr_clear(struct block_device *bdev, u64 key)
1794 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1797 static const struct pr_ops sd_pr_ops = {
1798 .pr_register = sd_pr_register,
1799 .pr_reserve = sd_pr_reserve,
1800 .pr_release = sd_pr_release,
1801 .pr_preempt = sd_pr_preempt,
1802 .pr_clear = sd_pr_clear,
1805 static const struct block_device_operations sd_fops = {
1806 .owner = THIS_MODULE,
1808 .release = sd_release,
1810 .getgeo = sd_getgeo,
1811 #ifdef CONFIG_COMPAT
1812 .compat_ioctl = sd_compat_ioctl,
1814 .check_events = sd_check_events,
1815 .revalidate_disk = sd_revalidate_disk,
1816 .unlock_native_capacity = sd_unlock_native_capacity,
1817 .report_zones = sd_zbc_report_zones,
1818 .pr_ops = &sd_pr_ops,
1822 * sd_eh_reset - reset error handling callback
1823 * @scmd: sd-issued command that has failed
1825 * This function is called by the SCSI midlayer before starting
1826 * SCSI EH. When counting medium access failures we have to be
1827 * careful to register it only only once per device and SCSI EH run;
1828 * there might be several timed out commands which will cause the
1829 * 'max_medium_access_timeouts' counter to trigger after the first
1830 * SCSI EH run already and set the device to offline.
1831 * So this function resets the internal counter before starting SCSI EH.
1833 static void sd_eh_reset(struct scsi_cmnd *scmd)
1835 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1837 /* New SCSI EH run, reset gate variable */
1838 sdkp->ignore_medium_access_errors = false;
1842 * sd_eh_action - error handling callback
1843 * @scmd: sd-issued command that has failed
1844 * @eh_disp: The recovery disposition suggested by the midlayer
1846 * This function is called by the SCSI midlayer upon completion of an
1847 * error test command (currently TEST UNIT READY). The result of sending
1848 * the eh command is passed in eh_disp. We're looking for devices that
1849 * fail medium access commands but are OK with non access commands like
1850 * test unit ready (so wrongly see the device as having a successful
1853 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1855 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1856 struct scsi_device *sdev = scmd->device;
1858 if (!scsi_device_online(sdev) ||
1859 !scsi_medium_access_command(scmd) ||
1860 host_byte(scmd->result) != DID_TIME_OUT ||
1865 * The device has timed out executing a medium access command.
1866 * However, the TEST UNIT READY command sent during error
1867 * handling completed successfully. Either the device is in the
1868 * process of recovering or has it suffered an internal failure
1869 * that prevents access to the storage medium.
1871 if (!sdkp->ignore_medium_access_errors) {
1872 sdkp->medium_access_timed_out++;
1873 sdkp->ignore_medium_access_errors = true;
1877 * If the device keeps failing read/write commands but TEST UNIT
1878 * READY always completes successfully we assume that medium
1879 * access is no longer possible and take the device offline.
1881 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1882 scmd_printk(KERN_ERR, scmd,
1883 "Medium access timeout failure. Offlining disk!\n");
1884 mutex_lock(&sdev->state_mutex);
1885 scsi_device_set_state(sdev, SDEV_OFFLINE);
1886 mutex_unlock(&sdev->state_mutex);
1894 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1896 struct request *req = scmd->request;
1897 struct scsi_device *sdev = scmd->device;
1898 unsigned int transferred, good_bytes;
1899 u64 start_lba, end_lba, bad_lba;
1902 * Some commands have a payload smaller than the device logical
1903 * block size (e.g. INQUIRY on a 4K disk).
1905 if (scsi_bufflen(scmd) <= sdev->sector_size)
1908 /* Check if we have a 'bad_lba' information */
1909 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1910 SCSI_SENSE_BUFFERSIZE,
1915 * If the bad lba was reported incorrectly, we have no idea where
1918 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1919 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1920 if (bad_lba < start_lba || bad_lba >= end_lba)
1924 * resid is optional but mostly filled in. When it's unused,
1925 * its value is zero, so we assume the whole buffer transferred
1927 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1929 /* This computation should always be done in terms of the
1930 * resolution of the device's medium.
1932 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1934 return min(good_bytes, transferred);
1938 * sd_done - bottom half handler: called when the lower level
1939 * driver has completed (successfully or otherwise) a scsi command.
1940 * @SCpnt: mid-level's per command structure.
1942 * Note: potentially run from within an ISR. Must not block.
1944 static int sd_done(struct scsi_cmnd *SCpnt)
1946 int result = SCpnt->result;
1947 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1948 unsigned int sector_size = SCpnt->device->sector_size;
1950 struct scsi_sense_hdr sshdr;
1951 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1952 struct request *req = SCpnt->request;
1953 int sense_valid = 0;
1954 int sense_deferred = 0;
1956 switch (req_op(req)) {
1957 case REQ_OP_DISCARD:
1958 case REQ_OP_WRITE_ZEROES:
1959 case REQ_OP_WRITE_SAME:
1960 case REQ_OP_ZONE_RESET:
1962 good_bytes = blk_rq_bytes(req);
1963 scsi_set_resid(SCpnt, 0);
1966 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1971 * In case of bogus fw or device, we could end up having
1972 * an unaligned partial completion. Check this here and force
1975 resid = scsi_get_resid(SCpnt);
1976 if (resid & (sector_size - 1)) {
1977 sd_printk(KERN_INFO, sdkp,
1978 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1979 resid, sector_size);
1980 resid = min(scsi_bufflen(SCpnt),
1981 round_up(resid, sector_size));
1982 scsi_set_resid(SCpnt, resid);
1987 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1989 sense_deferred = scsi_sense_is_deferred(&sshdr);
1991 sdkp->medium_access_timed_out = 0;
1993 if (driver_byte(result) != DRIVER_SENSE &&
1994 (!sense_valid || sense_deferred))
1997 switch (sshdr.sense_key) {
1998 case HARDWARE_ERROR:
2000 good_bytes = sd_completed_bytes(SCpnt);
2002 case RECOVERED_ERROR:
2003 good_bytes = scsi_bufflen(SCpnt);
2006 /* This indicates a false check condition, so ignore it. An
2007 * unknown amount of data was transferred so treat it as an
2011 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2013 case ABORTED_COMMAND:
2014 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2015 good_bytes = sd_completed_bytes(SCpnt);
2017 case ILLEGAL_REQUEST:
2018 switch (sshdr.asc) {
2019 case 0x10: /* DIX: Host detected corruption */
2020 good_bytes = sd_completed_bytes(SCpnt);
2022 case 0x20: /* INVALID COMMAND OPCODE */
2023 case 0x24: /* INVALID FIELD IN CDB */
2024 switch (SCpnt->cmnd[0]) {
2026 sd_config_discard(sdkp, SD_LBP_DISABLE);
2030 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2031 sd_config_discard(sdkp, SD_LBP_DISABLE);
2033 sdkp->device->no_write_same = 1;
2034 sd_config_write_same(sdkp);
2035 req->rq_flags |= RQF_QUIET;
2046 if (sd_is_zoned(sdkp))
2047 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2049 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2050 "sd_done: completed %d of %d bytes\n",
2051 good_bytes, scsi_bufflen(SCpnt)));
2053 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2055 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2056 good_bytes / scsi_prot_interval(SCpnt));
2062 * spinup disk - called only in sd_revalidate_disk()
2065 sd_spinup_disk(struct scsi_disk *sdkp)
2067 unsigned char cmd[10];
2068 unsigned long spintime_expire = 0;
2069 int retries, spintime;
2070 unsigned int the_result;
2071 struct scsi_sense_hdr sshdr;
2072 int sense_valid = 0;
2076 /* Spin up drives, as required. Only do this at boot time */
2077 /* Spinup needs to be done for module loads too. */
2082 cmd[0] = TEST_UNIT_READY;
2083 memset((void *) &cmd[1], 0, 9);
2085 the_result = scsi_execute_req(sdkp->device, cmd,
2088 SD_MAX_RETRIES, NULL);
2091 * If the drive has indicated to us that it
2092 * doesn't have any media in it, don't bother
2093 * with any more polling.
2095 if (media_not_present(sdkp, &sshdr))
2099 sense_valid = scsi_sense_valid(&sshdr);
2101 } while (retries < 3 &&
2102 (!scsi_status_is_good(the_result) ||
2103 ((driver_byte(the_result) == DRIVER_SENSE) &&
2104 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2106 if (driver_byte(the_result) != DRIVER_SENSE) {
2107 /* no sense, TUR either succeeded or failed
2108 * with a status error */
2109 if(!spintime && !scsi_status_is_good(the_result)) {
2110 sd_print_result(sdkp, "Test Unit Ready failed",
2117 * The device does not want the automatic start to be issued.
2119 if (sdkp->device->no_start_on_add)
2122 if (sense_valid && sshdr.sense_key == NOT_READY) {
2123 if (sshdr.asc == 4 && sshdr.ascq == 3)
2124 break; /* manual intervention required */
2125 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2126 break; /* standby */
2127 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2128 break; /* unavailable */
2129 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2130 break; /* sanitize in progress */
2132 * Issue command to spin up drive when not ready
2135 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2136 cmd[0] = START_STOP;
2137 cmd[1] = 1; /* Return immediately */
2138 memset((void *) &cmd[2], 0, 8);
2139 cmd[4] = 1; /* Start spin cycle */
2140 if (sdkp->device->start_stop_pwr_cond)
2142 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2144 SD_TIMEOUT, SD_MAX_RETRIES,
2146 spintime_expire = jiffies + 100 * HZ;
2149 /* Wait 1 second for next try */
2151 printk(KERN_CONT ".");
2154 * Wait for USB flash devices with slow firmware.
2155 * Yes, this sense key/ASC combination shouldn't
2156 * occur here. It's characteristic of these devices.
2158 } else if (sense_valid &&
2159 sshdr.sense_key == UNIT_ATTENTION &&
2160 sshdr.asc == 0x28) {
2162 spintime_expire = jiffies + 5 * HZ;
2165 /* Wait 1 second for next try */
2168 /* we don't understand the sense code, so it's
2169 * probably pointless to loop */
2171 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2172 sd_print_sense_hdr(sdkp, &sshdr);
2177 } while (spintime && time_before_eq(jiffies, spintime_expire));
2180 if (scsi_status_is_good(the_result))
2181 printk(KERN_CONT "ready\n");
2183 printk(KERN_CONT "not responding...\n");
2188 * Determine whether disk supports Data Integrity Field.
2190 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2192 struct scsi_device *sdp = sdkp->device;
2196 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2199 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2201 if (type > T10_PI_TYPE3_PROTECTION)
2203 else if (scsi_host_dif_capable(sdp->host, type))
2206 if (sdkp->first_scan || type != sdkp->protection_type)
2209 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2210 " protection type %u. Disabling disk!\n",
2214 sd_printk(KERN_NOTICE, sdkp,
2215 "Enabling DIF Type %u protection\n", type);
2218 sd_printk(KERN_NOTICE, sdkp,
2219 "Disabling DIF Type %u protection\n", type);
2223 sdkp->protection_type = type;
2228 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2229 struct scsi_sense_hdr *sshdr, int sense_valid,
2232 if (driver_byte(the_result) == DRIVER_SENSE)
2233 sd_print_sense_hdr(sdkp, sshdr);
2235 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2238 * Set dirty bit for removable devices if not ready -
2239 * sometimes drives will not report this properly.
2241 if (sdp->removable &&
2242 sense_valid && sshdr->sense_key == NOT_READY)
2243 set_media_not_present(sdkp);
2246 * We used to set media_present to 0 here to indicate no media
2247 * in the drive, but some drives fail read capacity even with
2248 * media present, so we can't do that.
2250 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2254 #if RC16_LEN > SD_BUF_SIZE
2255 #error RC16_LEN must not be more than SD_BUF_SIZE
2258 #define READ_CAPACITY_RETRIES_ON_RESET 10
2260 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2261 unsigned char *buffer)
2263 unsigned char cmd[16];
2264 struct scsi_sense_hdr sshdr;
2265 int sense_valid = 0;
2267 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2268 unsigned int alignment;
2269 unsigned long long lba;
2270 unsigned sector_size;
2272 if (sdp->no_read_capacity_16)
2277 cmd[0] = SERVICE_ACTION_IN_16;
2278 cmd[1] = SAI_READ_CAPACITY_16;
2280 memset(buffer, 0, RC16_LEN);
2282 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2283 buffer, RC16_LEN, &sshdr,
2284 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2286 if (media_not_present(sdkp, &sshdr))
2290 sense_valid = scsi_sense_valid(&sshdr);
2292 sshdr.sense_key == ILLEGAL_REQUEST &&
2293 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2295 /* Invalid Command Operation Code or
2296 * Invalid Field in CDB, just retry
2297 * silently with RC10 */
2300 sshdr.sense_key == UNIT_ATTENTION &&
2301 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2302 /* Device reset might occur several times,
2303 * give it one more chance */
2304 if (--reset_retries > 0)
2309 } while (the_result && retries);
2312 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2313 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2317 sector_size = get_unaligned_be32(&buffer[8]);
2318 lba = get_unaligned_be64(&buffer[0]);
2320 if (sd_read_protection_type(sdkp, buffer) < 0) {
2325 /* Logical blocks per physical block exponent */
2326 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2329 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2331 /* Lowest aligned logical block */
2332 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2333 blk_queue_alignment_offset(sdp->request_queue, alignment);
2334 if (alignment && sdkp->first_scan)
2335 sd_printk(KERN_NOTICE, sdkp,
2336 "physical block alignment offset: %u\n", alignment);
2338 if (buffer[14] & 0x80) { /* LBPME */
2341 if (buffer[14] & 0x40) /* LBPRZ */
2344 sd_config_discard(sdkp, SD_LBP_WS16);
2347 sdkp->capacity = lba + 1;
2351 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2352 unsigned char *buffer)
2354 unsigned char cmd[16];
2355 struct scsi_sense_hdr sshdr;
2356 int sense_valid = 0;
2358 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2360 unsigned sector_size;
2363 cmd[0] = READ_CAPACITY;
2364 memset(&cmd[1], 0, 9);
2365 memset(buffer, 0, 8);
2367 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2369 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2371 if (media_not_present(sdkp, &sshdr))
2375 sense_valid = scsi_sense_valid(&sshdr);
2377 sshdr.sense_key == UNIT_ATTENTION &&
2378 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2379 /* Device reset might occur several times,
2380 * give it one more chance */
2381 if (--reset_retries > 0)
2386 } while (the_result && retries);
2389 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2390 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2394 sector_size = get_unaligned_be32(&buffer[4]);
2395 lba = get_unaligned_be32(&buffer[0]);
2397 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2398 /* Some buggy (usb cardreader) devices return an lba of
2399 0xffffffff when the want to report a size of 0 (with
2400 which they really mean no media is present) */
2402 sdkp->physical_block_size = sector_size;
2406 sdkp->capacity = lba + 1;
2407 sdkp->physical_block_size = sector_size;
2411 static int sd_try_rc16_first(struct scsi_device *sdp)
2413 if (sdp->host->max_cmd_len < 16)
2415 if (sdp->try_rc_10_first)
2417 if (sdp->scsi_level > SCSI_SPC_2)
2419 if (scsi_device_protection(sdp))
2425 * read disk capacity
2428 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2431 struct scsi_device *sdp = sdkp->device;
2433 if (sd_try_rc16_first(sdp)) {
2434 sector_size = read_capacity_16(sdkp, sdp, buffer);
2435 if (sector_size == -EOVERFLOW)
2437 if (sector_size == -ENODEV)
2439 if (sector_size < 0)
2440 sector_size = read_capacity_10(sdkp, sdp, buffer);
2441 if (sector_size < 0)
2444 sector_size = read_capacity_10(sdkp, sdp, buffer);
2445 if (sector_size == -EOVERFLOW)
2447 if (sector_size < 0)
2449 if ((sizeof(sdkp->capacity) > 4) &&
2450 (sdkp->capacity > 0xffffffffULL)) {
2451 int old_sector_size = sector_size;
2452 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2453 "Trying to use READ CAPACITY(16).\n");
2454 sector_size = read_capacity_16(sdkp, sdp, buffer);
2455 if (sector_size < 0) {
2456 sd_printk(KERN_NOTICE, sdkp,
2457 "Using 0xffffffff as device size\n");
2458 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2459 sector_size = old_sector_size;
2462 /* Remember that READ CAPACITY(16) succeeded */
2463 sdp->try_rc_10_first = 0;
2467 /* Some devices are known to return the total number of blocks,
2468 * not the highest block number. Some devices have versions
2469 * which do this and others which do not. Some devices we might
2470 * suspect of doing this but we don't know for certain.
2472 * If we know the reported capacity is wrong, decrement it. If
2473 * we can only guess, then assume the number of blocks is even
2474 * (usually true but not always) and err on the side of lowering
2477 if (sdp->fix_capacity ||
2478 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2479 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2480 "from its reported value: %llu\n",
2481 (unsigned long long) sdkp->capacity);
2486 if (sector_size == 0) {
2488 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2492 if (sector_size != 512 &&
2493 sector_size != 1024 &&
2494 sector_size != 2048 &&
2495 sector_size != 4096) {
2496 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2499 * The user might want to re-format the drive with
2500 * a supported sectorsize. Once this happens, it
2501 * would be relatively trivial to set the thing up.
2502 * For this reason, we leave the thing in the table.
2506 * set a bogus sector size so the normal read/write
2507 * logic in the block layer will eventually refuse any
2508 * request on this device without tripping over power
2509 * of two sector size assumptions
2513 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2514 blk_queue_physical_block_size(sdp->request_queue,
2515 sdkp->physical_block_size);
2516 sdkp->device->sector_size = sector_size;
2518 if (sdkp->capacity > 0xffffffff)
2519 sdp->use_16_for_rw = 1;
2524 * Print disk capacity
2527 sd_print_capacity(struct scsi_disk *sdkp,
2528 sector_t old_capacity)
2530 int sector_size = sdkp->device->sector_size;
2531 char cap_str_2[10], cap_str_10[10];
2533 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2536 string_get_size(sdkp->capacity, sector_size,
2537 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2538 string_get_size(sdkp->capacity, sector_size,
2539 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2541 sd_printk(KERN_NOTICE, sdkp,
2542 "%llu %d-byte logical blocks: (%s/%s)\n",
2543 (unsigned long long)sdkp->capacity,
2544 sector_size, cap_str_10, cap_str_2);
2546 if (sdkp->physical_block_size != sector_size)
2547 sd_printk(KERN_NOTICE, sdkp,
2548 "%u-byte physical blocks\n",
2549 sdkp->physical_block_size);
2551 sd_zbc_print_zones(sdkp);
2554 /* called with buffer of length 512 */
2556 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2557 unsigned char *buffer, int len, struct scsi_mode_data *data,
2558 struct scsi_sense_hdr *sshdr)
2560 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2561 SD_TIMEOUT, SD_MAX_RETRIES, data,
2566 * read write protect setting, if possible - called only in sd_revalidate_disk()
2567 * called with buffer of length SD_BUF_SIZE
2570 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2573 struct scsi_device *sdp = sdkp->device;
2574 struct scsi_mode_data data;
2575 int old_wp = sdkp->write_prot;
2577 set_disk_ro(sdkp->disk, 0);
2578 if (sdp->skip_ms_page_3f) {
2579 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2583 if (sdp->use_192_bytes_for_3f) {
2584 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2587 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2588 * We have to start carefully: some devices hang if we ask
2589 * for more than is available.
2591 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2594 * Second attempt: ask for page 0 When only page 0 is
2595 * implemented, a request for page 3F may return Sense Key
2596 * 5: Illegal Request, Sense Code 24: Invalid field in
2599 if (!scsi_status_is_good(res))
2600 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2603 * Third attempt: ask 255 bytes, as we did earlier.
2605 if (!scsi_status_is_good(res))
2606 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2610 if (!scsi_status_is_good(res)) {
2611 sd_first_printk(KERN_WARNING, sdkp,
2612 "Test WP failed, assume Write Enabled\n");
2614 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2615 set_disk_ro(sdkp->disk, sdkp->write_prot);
2616 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2617 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2618 sdkp->write_prot ? "on" : "off");
2619 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2625 * sd_read_cache_type - called only from sd_revalidate_disk()
2626 * called with buffer of length SD_BUF_SIZE
2629 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2632 struct scsi_device *sdp = sdkp->device;
2637 struct scsi_mode_data data;
2638 struct scsi_sense_hdr sshdr;
2639 int old_wce = sdkp->WCE;
2640 int old_rcd = sdkp->RCD;
2641 int old_dpofua = sdkp->DPOFUA;
2644 if (sdkp->cache_override)
2648 if (sdp->skip_ms_page_8) {
2649 if (sdp->type == TYPE_RBC)
2652 if (sdp->skip_ms_page_3f)
2655 if (sdp->use_192_bytes_for_3f)
2659 } else if (sdp->type == TYPE_RBC) {
2667 /* cautiously ask */
2668 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2671 if (!scsi_status_is_good(res))
2674 if (!data.header_length) {
2677 sd_first_printk(KERN_ERR, sdkp,
2678 "Missing header in MODE_SENSE response\n");
2681 /* that went OK, now ask for the proper length */
2685 * We're only interested in the first three bytes, actually.
2686 * But the data cache page is defined for the first 20.
2690 else if (len > SD_BUF_SIZE) {
2691 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2692 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2695 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2699 if (len > first_len)
2700 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2703 if (scsi_status_is_good(res)) {
2704 int offset = data.header_length + data.block_descriptor_length;
2706 while (offset < len) {
2707 u8 page_code = buffer[offset] & 0x3F;
2708 u8 spf = buffer[offset] & 0x40;
2710 if (page_code == 8 || page_code == 6) {
2711 /* We're interested only in the first 3 bytes.
2713 if (len - offset <= 2) {
2714 sd_first_printk(KERN_ERR, sdkp,
2715 "Incomplete mode parameter "
2719 modepage = page_code;
2723 /* Go to the next page */
2724 if (spf && len - offset > 3)
2725 offset += 4 + (buffer[offset+2] << 8) +
2727 else if (!spf && len - offset > 1)
2728 offset += 2 + buffer[offset+1];
2730 sd_first_printk(KERN_ERR, sdkp,
2732 "parameter data\n");
2738 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2742 if (modepage == 8) {
2743 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2744 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2746 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2750 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2751 if (sdp->broken_fua) {
2752 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2754 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2755 !sdkp->device->use_16_for_rw) {
2756 sd_first_printk(KERN_NOTICE, sdkp,
2757 "Uses READ/WRITE(6), disabling FUA\n");
2761 /* No cache flush allowed for write protected devices */
2762 if (sdkp->WCE && sdkp->write_prot)
2765 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2766 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2767 sd_printk(KERN_NOTICE, sdkp,
2768 "Write cache: %s, read cache: %s, %s\n",
2769 sdkp->WCE ? "enabled" : "disabled",
2770 sdkp->RCD ? "disabled" : "enabled",
2771 sdkp->DPOFUA ? "supports DPO and FUA"
2772 : "doesn't support DPO or FUA");
2778 if (scsi_sense_valid(&sshdr) &&
2779 sshdr.sense_key == ILLEGAL_REQUEST &&
2780 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2781 /* Invalid field in CDB */
2782 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2784 sd_first_printk(KERN_ERR, sdkp,
2785 "Asking for cache data failed\n");
2788 if (sdp->wce_default_on) {
2789 sd_first_printk(KERN_NOTICE, sdkp,
2790 "Assuming drive cache: write back\n");
2793 sd_first_printk(KERN_ERR, sdkp,
2794 "Assuming drive cache: write through\n");
2802 * The ATO bit indicates whether the DIF application tag is available
2803 * for use by the operating system.
2805 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2808 struct scsi_device *sdp = sdkp->device;
2809 struct scsi_mode_data data;
2810 struct scsi_sense_hdr sshdr;
2812 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2815 if (sdkp->protection_type == 0)
2818 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2819 SD_MAX_RETRIES, &data, &sshdr);
2821 if (!scsi_status_is_good(res) || !data.header_length ||
2823 sd_first_printk(KERN_WARNING, sdkp,
2824 "getting Control mode page failed, assume no ATO\n");
2826 if (scsi_sense_valid(&sshdr))
2827 sd_print_sense_hdr(sdkp, &sshdr);
2832 offset = data.header_length + data.block_descriptor_length;
2834 if ((buffer[offset] & 0x3f) != 0x0a) {
2835 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2839 if ((buffer[offset + 5] & 0x80) == 0)
2848 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2849 * @sdkp: disk to query
2851 static void sd_read_block_limits(struct scsi_disk *sdkp)
2853 unsigned int sector_sz = sdkp->device->sector_size;
2854 const int vpd_len = 64;
2855 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2858 /* Block Limits VPD */
2859 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2862 blk_queue_io_min(sdkp->disk->queue,
2863 get_unaligned_be16(&buffer[6]) * sector_sz);
2865 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2866 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2868 if (buffer[3] == 0x3c) {
2869 unsigned int lba_count, desc_count;
2871 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2876 lba_count = get_unaligned_be32(&buffer[20]);
2877 desc_count = get_unaligned_be32(&buffer[24]);
2879 if (lba_count && desc_count)
2880 sdkp->max_unmap_blocks = lba_count;
2882 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2884 if (buffer[32] & 0x80)
2885 sdkp->unmap_alignment =
2886 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2888 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2890 if (sdkp->max_unmap_blocks)
2891 sd_config_discard(sdkp, SD_LBP_UNMAP);
2893 sd_config_discard(sdkp, SD_LBP_WS16);
2895 } else { /* LBP VPD page tells us what to use */
2896 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2897 sd_config_discard(sdkp, SD_LBP_UNMAP);
2898 else if (sdkp->lbpws)
2899 sd_config_discard(sdkp, SD_LBP_WS16);
2900 else if (sdkp->lbpws10)
2901 sd_config_discard(sdkp, SD_LBP_WS10);
2903 sd_config_discard(sdkp, SD_LBP_DISABLE);
2912 * sd_read_block_characteristics - Query block dev. characteristics
2913 * @sdkp: disk to query
2915 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2917 struct request_queue *q = sdkp->disk->queue;
2918 unsigned char *buffer;
2920 const int vpd_len = 64;
2922 buffer = kmalloc(vpd_len, GFP_KERNEL);
2925 /* Block Device Characteristics VPD */
2926 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2929 rot = get_unaligned_be16(&buffer[4]);
2932 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2933 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2936 if (sdkp->device->type == TYPE_ZBC) {
2938 q->limits.zoned = BLK_ZONED_HM;
2940 sdkp->zoned = (buffer[8] >> 4) & 3;
2941 if (sdkp->zoned == 1)
2943 q->limits.zoned = BLK_ZONED_HA;
2946 * Treat drive-managed devices as
2947 * regular block devices.
2949 q->limits.zoned = BLK_ZONED_NONE;
2951 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2952 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2953 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2960 * sd_read_block_provisioning - Query provisioning VPD page
2961 * @sdkp: disk to query
2963 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2965 unsigned char *buffer;
2966 const int vpd_len = 8;
2968 if (sdkp->lbpme == 0)
2971 buffer = kmalloc(vpd_len, GFP_KERNEL);
2973 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2977 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2978 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2979 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2985 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2987 struct scsi_device *sdev = sdkp->device;
2989 if (sdev->host->no_write_same) {
2990 sdev->no_write_same = 1;
2995 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2996 /* too large values might cause issues with arcmsr */
2997 int vpd_buf_len = 64;
2999 sdev->no_report_opcodes = 1;
3001 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3002 * CODES is unsupported and the device has an ATA
3003 * Information VPD page (SAT).
3005 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3006 sdev->no_write_same = 1;
3009 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3012 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3016 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3018 struct scsi_device *sdev = sdkp->device;
3020 if (!sdev->security_supported)
3023 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3024 SECURITY_PROTOCOL_IN) == 1 &&
3025 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3026 SECURITY_PROTOCOL_OUT) == 1)
3031 * Determine the device's preferred I/O size for reads and writes
3032 * unless the reported value is unreasonably small, large, not a
3033 * multiple of the physical block size, or simply garbage.
3035 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3036 unsigned int dev_max)
3038 struct scsi_device *sdp = sdkp->device;
3039 unsigned int opt_xfer_bytes =
3040 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3042 if (sdkp->opt_xfer_blocks == 0)
3045 if (sdkp->opt_xfer_blocks > dev_max) {
3046 sd_first_printk(KERN_WARNING, sdkp,
3047 "Optimal transfer size %u logical blocks " \
3048 "> dev_max (%u logical blocks)\n",
3049 sdkp->opt_xfer_blocks, dev_max);
3053 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3054 sd_first_printk(KERN_WARNING, sdkp,
3055 "Optimal transfer size %u logical blocks " \
3056 "> sd driver limit (%u logical blocks)\n",
3057 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3061 if (opt_xfer_bytes < PAGE_SIZE) {
3062 sd_first_printk(KERN_WARNING, sdkp,
3063 "Optimal transfer size %u bytes < " \
3064 "PAGE_SIZE (%u bytes)\n",
3065 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3069 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3070 sd_first_printk(KERN_WARNING, sdkp,
3071 "Optimal transfer size %u bytes not a " \
3072 "multiple of physical block size (%u bytes)\n",
3073 opt_xfer_bytes, sdkp->physical_block_size);
3077 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3083 * sd_revalidate_disk - called the first time a new disk is seen,
3084 * performs disk spin up, read_capacity, etc.
3085 * @disk: struct gendisk we care about
3087 static int sd_revalidate_disk(struct gendisk *disk)
3089 struct scsi_disk *sdkp = scsi_disk(disk);
3090 struct scsi_device *sdp = sdkp->device;
3091 struct request_queue *q = sdkp->disk->queue;
3092 sector_t old_capacity = sdkp->capacity;
3093 unsigned char *buffer;
3094 unsigned int dev_max, rw_max;
3096 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3097 "sd_revalidate_disk\n"));
3100 * If the device is offline, don't try and read capacity or any
3101 * of the other niceties.
3103 if (!scsi_device_online(sdp))
3106 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3108 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3109 "allocation failure.\n");
3113 sd_spinup_disk(sdkp);
3116 * Without media there is no reason to ask; moreover, some devices
3117 * react badly if we do.
3119 if (sdkp->media_present) {
3120 sd_read_capacity(sdkp, buffer);
3123 * set the default to rotational. All non-rotational devices
3124 * support the block characteristics VPD page, which will
3125 * cause this to be updated correctly and any device which
3126 * doesn't support it should be treated as rotational.
3128 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3129 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3131 if (scsi_device_supports_vpd(sdp)) {
3132 sd_read_block_provisioning(sdkp);
3133 sd_read_block_limits(sdkp);
3134 sd_read_block_characteristics(sdkp);
3135 sd_zbc_read_zones(sdkp, buffer);
3138 sd_print_capacity(sdkp, old_capacity);
3140 sd_read_write_protect_flag(sdkp, buffer);
3141 sd_read_cache_type(sdkp, buffer);
3142 sd_read_app_tag_own(sdkp, buffer);
3143 sd_read_write_same(sdkp, buffer);
3144 sd_read_security(sdkp, buffer);
3148 * We now have all cache related info, determine how we deal
3149 * with flush requests.
3151 sd_set_flush_flag(sdkp);
3153 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3154 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3156 /* Some devices report a maximum block count for READ/WRITE requests. */
3157 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3158 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3160 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3161 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3162 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3164 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3165 (sector_t)BLK_DEF_MAX_SECTORS);
3167 /* Do not exceed controller limit */
3168 rw_max = min(rw_max, queue_max_hw_sectors(q));
3171 * Only update max_sectors if previously unset or if the current value
3172 * exceeds the capabilities of the hardware.
3174 if (sdkp->first_scan ||
3175 q->limits.max_sectors > q->limits.max_dev_sectors ||
3176 q->limits.max_sectors > q->limits.max_hw_sectors)
3177 q->limits.max_sectors = rw_max;
3179 sdkp->first_scan = 0;
3181 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3182 sd_config_write_same(sdkp);
3190 * sd_unlock_native_capacity - unlock native capacity
3191 * @disk: struct gendisk to set capacity for
3193 * Block layer calls this function if it detects that partitions
3194 * on @disk reach beyond the end of the device. If the SCSI host
3195 * implements ->unlock_native_capacity() method, it's invoked to
3196 * give it a chance to adjust the device capacity.
3199 * Defined by block layer. Might sleep.
3201 static void sd_unlock_native_capacity(struct gendisk *disk)
3203 struct scsi_device *sdev = scsi_disk(disk)->device;
3205 if (sdev->host->hostt->unlock_native_capacity)
3206 sdev->host->hostt->unlock_native_capacity(sdev);
3210 * sd_format_disk_name - format disk name
3211 * @prefix: name prefix - ie. "sd" for SCSI disks
3212 * @index: index of the disk to format name for
3213 * @buf: output buffer
3214 * @buflen: length of the output buffer
3216 * SCSI disk names starts at sda. The 26th device is sdz and the
3217 * 27th is sdaa. The last one for two lettered suffix is sdzz
3218 * which is followed by sdaaa.
3220 * This is basically 26 base counting with one extra 'nil' entry
3221 * at the beginning from the second digit on and can be
3222 * determined using similar method as 26 base conversion with the
3223 * index shifted -1 after each digit is computed.
3229 * 0 on success, -errno on failure.
3231 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3233 const int base = 'z' - 'a' + 1;
3234 char *begin = buf + strlen(prefix);
3235 char *end = buf + buflen;
3245 *--p = 'a' + (index % unit);
3246 index = (index / unit) - 1;
3247 } while (index >= 0);
3249 memmove(begin, p, end - p);
3250 memcpy(buf, prefix, strlen(prefix));
3256 * The asynchronous part of sd_probe
3258 static void sd_probe_async(void *data, async_cookie_t cookie)
3260 struct scsi_disk *sdkp = data;
3261 struct scsi_device *sdp;
3268 index = sdkp->index;
3269 dev = &sdp->sdev_gendev;
3271 gd->major = sd_major((index & 0xf0) >> 4);
3272 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3274 gd->fops = &sd_fops;
3275 gd->private_data = &sdkp->driver;
3276 gd->queue = sdkp->device->request_queue;
3278 /* defaults, until the device tells us otherwise */
3279 sdp->sector_size = 512;
3281 sdkp->media_present = 1;
3282 sdkp->write_prot = 0;
3283 sdkp->cache_override = 0;
3287 sdkp->first_scan = 1;
3288 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3290 sd_revalidate_disk(gd);
3292 gd->flags = GENHD_FL_EXT_DEVT;
3293 if (sdp->removable) {
3294 gd->flags |= GENHD_FL_REMOVABLE;
3295 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3296 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3299 blk_pm_runtime_init(sdp->request_queue, dev);
3300 device_add_disk(dev, gd, NULL);
3302 sd_dif_config_host(sdkp);
3304 sd_revalidate_disk(gd);
3306 if (sdkp->security) {
3307 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3309 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3312 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3313 sdp->removable ? "removable " : "");
3314 scsi_autopm_put_device(sdp);
3315 put_device(&sdkp->dev);
3319 * sd_probe - called during driver initialization and whenever a
3320 * new scsi device is attached to the system. It is called once
3321 * for each scsi device (not just disks) present.
3322 * @dev: pointer to device object
3324 * Returns 0 if successful (or not interested in this scsi device
3325 * (e.g. scanner)); 1 when there is an error.
3327 * Note: this function is invoked from the scsi mid-level.
3328 * This function sets up the mapping between a given
3329 * <host,channel,id,lun> (found in sdp) and new device name
3330 * (e.g. /dev/sda). More precisely it is the block device major
3331 * and minor number that is chosen here.
3333 * Assume sd_probe is not re-entrant (for time being)
3334 * Also think about sd_probe() and sd_remove() running coincidentally.
3336 static int sd_probe(struct device *dev)
3338 struct scsi_device *sdp = to_scsi_device(dev);
3339 struct scsi_disk *sdkp;
3344 scsi_autopm_get_device(sdp);
3346 if (sdp->type != TYPE_DISK &&
3347 sdp->type != TYPE_ZBC &&
3348 sdp->type != TYPE_MOD &&
3349 sdp->type != TYPE_RBC)
3352 #ifndef CONFIG_BLK_DEV_ZONED
3353 if (sdp->type == TYPE_ZBC)
3356 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3360 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3364 gd = alloc_disk(SD_MINORS);
3368 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3370 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3374 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3376 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3377 goto out_free_index;
3381 sdkp->driver = &sd_template;
3383 sdkp->index = index;
3384 atomic_set(&sdkp->openers, 0);
3385 atomic_set(&sdkp->device->ioerr_cnt, 0);
3387 if (!sdp->request_queue->rq_timeout) {
3388 if (sdp->type != TYPE_MOD)
3389 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3391 blk_queue_rq_timeout(sdp->request_queue,
3395 device_initialize(&sdkp->dev);
3396 sdkp->dev.parent = dev;
3397 sdkp->dev.class = &sd_disk_class;
3398 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3400 error = device_add(&sdkp->dev);
3402 goto out_free_index;
3405 dev_set_drvdata(dev, sdkp);
3407 get_device(&sdkp->dev); /* prevent release before async_schedule */
3408 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3413 ida_free(&sd_index_ida, index);
3419 scsi_autopm_put_device(sdp);
3424 * sd_remove - called whenever a scsi disk (previously recognized by
3425 * sd_probe) is detached from the system. It is called (potentially
3426 * multiple times) during sd module unload.
3427 * @dev: pointer to device object
3429 * Note: this function is invoked from the scsi mid-level.
3430 * This function potentially frees up a device name (e.g. /dev/sdc)
3431 * that could be re-used by a subsequent sd_probe().
3432 * This function is not called when the built-in sd driver is "exit-ed".
3434 static int sd_remove(struct device *dev)
3436 struct scsi_disk *sdkp;
3439 sdkp = dev_get_drvdata(dev);
3440 devt = disk_devt(sdkp->disk);
3441 scsi_autopm_get_device(sdkp->device);
3443 async_synchronize_full_domain(&scsi_sd_pm_domain);
3444 async_synchronize_full_domain(&scsi_sd_probe_domain);
3445 device_del(&sdkp->dev);
3446 del_gendisk(sdkp->disk);
3449 free_opal_dev(sdkp->opal_dev);
3451 blk_register_region(devt, SD_MINORS, NULL,
3452 sd_default_probe, NULL, NULL);
3454 mutex_lock(&sd_ref_mutex);
3455 dev_set_drvdata(dev, NULL);
3456 put_device(&sdkp->dev);
3457 mutex_unlock(&sd_ref_mutex);
3463 * scsi_disk_release - Called to free the scsi_disk structure
3464 * @dev: pointer to embedded class device
3466 * sd_ref_mutex must be held entering this routine. Because it is
3467 * called on last put, you should always use the scsi_disk_get()
3468 * scsi_disk_put() helpers which manipulate the semaphore directly
3469 * and never do a direct put_device.
3471 static void scsi_disk_release(struct device *dev)
3473 struct scsi_disk *sdkp = to_scsi_disk(dev);
3474 struct gendisk *disk = sdkp->disk;
3475 struct request_queue *q = disk->queue;
3477 ida_free(&sd_index_ida, sdkp->index);
3480 * Wait until all requests that are in progress have completed.
3481 * This is necessary to avoid that e.g. scsi_end_request() crashes
3482 * due to clearing the disk->private_data pointer. Wait from inside
3483 * scsi_disk_release() instead of from sd_release() to avoid that
3484 * freezing and unfreezing the request queue affects user space I/O
3485 * in case multiple processes open a /dev/sd... node concurrently.
3487 blk_mq_freeze_queue(q);
3488 blk_mq_unfreeze_queue(q);
3490 disk->private_data = NULL;
3492 put_device(&sdkp->device->sdev_gendev);
3497 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3499 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3500 struct scsi_sense_hdr sshdr;
3501 struct scsi_device *sdp = sdkp->device;
3505 cmd[4] |= 1; /* START */
3507 if (sdp->start_stop_pwr_cond)
3508 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3510 if (!scsi_device_online(sdp))
3513 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3514 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3516 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3517 if (driver_byte(res) == DRIVER_SENSE)
3518 sd_print_sense_hdr(sdkp, &sshdr);
3519 if (scsi_sense_valid(&sshdr) &&
3520 /* 0x3a is medium not present */
3525 /* SCSI error codes must not go to the generic layer */
3533 * Send a SYNCHRONIZE CACHE instruction down to the device through
3534 * the normal SCSI command structure. Wait for the command to
3537 static void sd_shutdown(struct device *dev)
3539 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3542 return; /* this can happen */
3544 if (pm_runtime_suspended(dev))
3547 if (sdkp->WCE && sdkp->media_present) {
3548 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3549 sd_sync_cache(sdkp, NULL);
3552 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3553 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3554 sd_start_stop_device(sdkp, 0);
3558 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3560 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3561 struct scsi_sense_hdr sshdr;
3564 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3567 if (sdkp->WCE && sdkp->media_present) {
3568 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3569 ret = sd_sync_cache(sdkp, &sshdr);
3572 /* ignore OFFLINE device */
3576 if (!scsi_sense_valid(&sshdr) ||
3577 sshdr.sense_key != ILLEGAL_REQUEST)
3581 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3582 * doesn't support sync. There's not much to do and
3583 * suspend shouldn't fail.
3589 if (sdkp->device->manage_start_stop) {
3590 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3591 /* an error is not worth aborting a system sleep */
3592 ret = sd_start_stop_device(sdkp, 0);
3593 if (ignore_stop_errors)
3600 static int sd_suspend_system(struct device *dev)
3602 return sd_suspend_common(dev, true);
3605 static int sd_suspend_runtime(struct device *dev)
3607 return sd_suspend_common(dev, false);
3610 static int sd_resume(struct device *dev)
3612 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3615 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3618 if (!sdkp->device->manage_start_stop)
3621 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3622 ret = sd_start_stop_device(sdkp, 1);
3624 opal_unlock_from_suspend(sdkp->opal_dev);
3629 * init_sd - entry point for this driver (both when built in or when
3632 * Note: this function registers this driver with the scsi mid-level.
3634 static int __init init_sd(void)
3636 int majors = 0, i, err;
3638 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3640 for (i = 0; i < SD_MAJORS; i++) {
3641 if (register_blkdev(sd_major(i), "sd") != 0)
3644 blk_register_region(sd_major(i), SD_MINORS, NULL,
3645 sd_default_probe, NULL, NULL);
3651 err = class_register(&sd_disk_class);
3655 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3657 if (!sd_cdb_cache) {
3658 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3663 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3665 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3670 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3671 if (!sd_page_pool) {
3672 printk(KERN_ERR "sd: can't init discard page pool\n");
3677 err = scsi_register_driver(&sd_template.gendrv);
3679 goto err_out_driver;
3684 mempool_destroy(sd_page_pool);
3687 mempool_destroy(sd_cdb_pool);
3690 kmem_cache_destroy(sd_cdb_cache);
3693 class_unregister(&sd_disk_class);
3695 for (i = 0; i < SD_MAJORS; i++)
3696 unregister_blkdev(sd_major(i), "sd");
3701 * exit_sd - exit point for this driver (when it is a module).
3703 * Note: this function unregisters this driver from the scsi mid-level.
3705 static void __exit exit_sd(void)
3709 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3711 scsi_unregister_driver(&sd_template.gendrv);
3712 mempool_destroy(sd_cdb_pool);
3713 mempool_destroy(sd_page_pool);
3714 kmem_cache_destroy(sd_cdb_cache);
3716 class_unregister(&sd_disk_class);
3718 for (i = 0; i < SD_MAJORS; i++) {
3719 blk_unregister_region(sd_major(i), SD_MINORS);
3720 unregister_blkdev(sd_major(i), "sd");
3724 module_init(init_sd);
3725 module_exit(exit_sd);
3727 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3728 struct scsi_sense_hdr *sshdr)
3730 scsi_print_sense_hdr(sdkp->device,
3731 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3734 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3737 const char *hb_string = scsi_hostbyte_string(result);
3738 const char *db_string = scsi_driverbyte_string(result);
3740 if (hb_string || db_string)
3741 sd_printk(KERN_INFO, sdkp,
3742 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3743 hb_string ? hb_string : "invalid",
3744 db_string ? db_string : "invalid");
3746 sd_printk(KERN_INFO, sdkp,
3747 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3748 msg, host_byte(result), driver_byte(result));