1 // SPDX-License-Identifier: GPL-2.0-only
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
36 #include <linux/module.h>
38 #include <linux/kernel.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
103 static void sd_config_discard(struct scsi_disk *, unsigned int);
104 static void sd_config_write_same(struct scsi_disk *);
105 static int sd_revalidate_disk(struct gendisk *);
106 static void sd_unlock_native_capacity(struct gendisk *disk);
107 static int sd_probe(struct device *);
108 static int sd_remove(struct device *);
109 static void sd_shutdown(struct device *);
110 static int sd_suspend_system(struct device *);
111 static int sd_suspend_runtime(struct device *);
112 static int sd_resume(struct device *);
113 static int sd_resume_runtime(struct device *);
114 static void sd_rescan(struct device *);
115 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
116 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117 static int sd_done(struct scsi_cmnd *);
118 static void sd_eh_reset(struct scsi_cmnd *);
119 static int sd_eh_action(struct scsi_cmnd *, int);
120 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
121 static void scsi_disk_release(struct device *cdev);
123 static DEFINE_IDA(sd_index_ida);
125 /* This semaphore is used to mediate the 0->1 reference get in the
126 * face of object destruction (i.e. we can't allow a get on an
127 * object after last put) */
128 static DEFINE_MUTEX(sd_ref_mutex);
130 static struct kmem_cache *sd_cdb_cache;
131 static mempool_t *sd_cdb_pool;
132 static mempool_t *sd_page_pool;
134 static const char *sd_cache_types[] = {
135 "write through", "none", "write back",
136 "write back, no read (daft)"
139 static void sd_set_flush_flag(struct scsi_disk *sdkp)
141 bool wc = false, fua = false;
149 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
153 cache_type_store(struct device *dev, struct device_attribute *attr,
154 const char *buf, size_t count)
156 int ct, rcd, wce, sp;
157 struct scsi_disk *sdkp = to_scsi_disk(dev);
158 struct scsi_device *sdp = sdkp->device;
161 struct scsi_mode_data data;
162 struct scsi_sense_hdr sshdr;
163 static const char temp[] = "temporary ";
166 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
167 /* no cache control on RBC devices; theoretically they
168 * can do it, but there's probably so many exceptions
169 * it's not worth the risk */
172 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
173 buf += sizeof(temp) - 1;
174 sdkp->cache_override = 1;
176 sdkp->cache_override = 0;
179 ct = sysfs_match_string(sd_cache_types, buf);
183 rcd = ct & 0x01 ? 1 : 0;
184 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
186 if (sdkp->cache_override) {
189 sd_set_flush_flag(sdkp);
193 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
194 sdkp->max_retries, &data, NULL))
196 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
197 data.block_descriptor_length);
198 buffer_data = buffer + data.header_length +
199 data.block_descriptor_length;
200 buffer_data[2] &= ~0x05;
201 buffer_data[2] |= wce << 2 | rcd;
202 sp = buffer_data[0] & 0x80 ? 1 : 0;
203 buffer_data[0] &= ~0x80;
206 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
207 * received mode parameter buffer before doing MODE SELECT.
209 data.device_specific = 0;
211 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
212 sdkp->max_retries, &data, &sshdr)) {
213 if (scsi_sense_valid(&sshdr))
214 sd_print_sense_hdr(sdkp, &sshdr);
217 sd_revalidate_disk(sdkp->disk);
222 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
225 struct scsi_disk *sdkp = to_scsi_disk(dev);
226 struct scsi_device *sdp = sdkp->device;
228 return sprintf(buf, "%u\n", sdp->manage_start_stop);
232 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
233 const char *buf, size_t count)
235 struct scsi_disk *sdkp = to_scsi_disk(dev);
236 struct scsi_device *sdp = sdkp->device;
239 if (!capable(CAP_SYS_ADMIN))
242 if (kstrtobool(buf, &v))
245 sdp->manage_start_stop = v;
249 static DEVICE_ATTR_RW(manage_start_stop);
252 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
254 struct scsi_disk *sdkp = to_scsi_disk(dev);
256 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
260 allow_restart_store(struct device *dev, struct device_attribute *attr,
261 const char *buf, size_t count)
264 struct scsi_disk *sdkp = to_scsi_disk(dev);
265 struct scsi_device *sdp = sdkp->device;
267 if (!capable(CAP_SYS_ADMIN))
270 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
273 if (kstrtobool(buf, &v))
276 sdp->allow_restart = v;
280 static DEVICE_ATTR_RW(allow_restart);
283 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
285 struct scsi_disk *sdkp = to_scsi_disk(dev);
286 int ct = sdkp->RCD + 2*sdkp->WCE;
288 return sprintf(buf, "%s\n", sd_cache_types[ct]);
290 static DEVICE_ATTR_RW(cache_type);
293 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
295 struct scsi_disk *sdkp = to_scsi_disk(dev);
297 return sprintf(buf, "%u\n", sdkp->DPOFUA);
299 static DEVICE_ATTR_RO(FUA);
302 protection_type_show(struct device *dev, struct device_attribute *attr,
305 struct scsi_disk *sdkp = to_scsi_disk(dev);
307 return sprintf(buf, "%u\n", sdkp->protection_type);
311 protection_type_store(struct device *dev, struct device_attribute *attr,
312 const char *buf, size_t count)
314 struct scsi_disk *sdkp = to_scsi_disk(dev);
318 if (!capable(CAP_SYS_ADMIN))
321 err = kstrtouint(buf, 10, &val);
326 if (val <= T10_PI_TYPE3_PROTECTION)
327 sdkp->protection_type = val;
331 static DEVICE_ATTR_RW(protection_type);
334 protection_mode_show(struct device *dev, struct device_attribute *attr,
337 struct scsi_disk *sdkp = to_scsi_disk(dev);
338 struct scsi_device *sdp = sdkp->device;
339 unsigned int dif, dix;
341 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
342 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
344 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
350 return sprintf(buf, "none\n");
352 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
354 static DEVICE_ATTR_RO(protection_mode);
357 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
359 struct scsi_disk *sdkp = to_scsi_disk(dev);
361 return sprintf(buf, "%u\n", sdkp->ATO);
363 static DEVICE_ATTR_RO(app_tag_own);
366 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
369 struct scsi_disk *sdkp = to_scsi_disk(dev);
371 return sprintf(buf, "%u\n", sdkp->lbpme);
373 static DEVICE_ATTR_RO(thin_provisioning);
375 /* sysfs_match_string() requires dense arrays */
376 static const char *lbp_mode[] = {
377 [SD_LBP_FULL] = "full",
378 [SD_LBP_UNMAP] = "unmap",
379 [SD_LBP_WS16] = "writesame_16",
380 [SD_LBP_WS10] = "writesame_10",
381 [SD_LBP_ZERO] = "writesame_zero",
382 [SD_LBP_DISABLE] = "disabled",
386 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
389 struct scsi_disk *sdkp = to_scsi_disk(dev);
391 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
395 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
396 const char *buf, size_t count)
398 struct scsi_disk *sdkp = to_scsi_disk(dev);
399 struct scsi_device *sdp = sdkp->device;
402 if (!capable(CAP_SYS_ADMIN))
405 if (sd_is_zoned(sdkp)) {
406 sd_config_discard(sdkp, SD_LBP_DISABLE);
410 if (sdp->type != TYPE_DISK)
413 mode = sysfs_match_string(lbp_mode, buf);
417 sd_config_discard(sdkp, mode);
421 static DEVICE_ATTR_RW(provisioning_mode);
423 /* sysfs_match_string() requires dense arrays */
424 static const char *zeroing_mode[] = {
425 [SD_ZERO_WRITE] = "write",
426 [SD_ZERO_WS] = "writesame",
427 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
428 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
432 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
435 struct scsi_disk *sdkp = to_scsi_disk(dev);
437 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
441 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
442 const char *buf, size_t count)
444 struct scsi_disk *sdkp = to_scsi_disk(dev);
447 if (!capable(CAP_SYS_ADMIN))
450 mode = sysfs_match_string(zeroing_mode, buf);
454 sdkp->zeroing_mode = mode;
458 static DEVICE_ATTR_RW(zeroing_mode);
461 max_medium_access_timeouts_show(struct device *dev,
462 struct device_attribute *attr, char *buf)
464 struct scsi_disk *sdkp = to_scsi_disk(dev);
466 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
470 max_medium_access_timeouts_store(struct device *dev,
471 struct device_attribute *attr, const char *buf,
474 struct scsi_disk *sdkp = to_scsi_disk(dev);
477 if (!capable(CAP_SYS_ADMIN))
480 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
482 return err ? err : count;
484 static DEVICE_ATTR_RW(max_medium_access_timeouts);
487 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
490 struct scsi_disk *sdkp = to_scsi_disk(dev);
492 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
496 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
497 const char *buf, size_t count)
499 struct scsi_disk *sdkp = to_scsi_disk(dev);
500 struct scsi_device *sdp = sdkp->device;
504 if (!capable(CAP_SYS_ADMIN))
507 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
510 err = kstrtoul(buf, 10, &max);
516 sdp->no_write_same = 1;
517 else if (max <= SD_MAX_WS16_BLOCKS) {
518 sdp->no_write_same = 0;
519 sdkp->max_ws_blocks = max;
522 sd_config_write_same(sdkp);
526 static DEVICE_ATTR_RW(max_write_same_blocks);
529 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
531 struct scsi_disk *sdkp = to_scsi_disk(dev);
533 if (sdkp->device->type == TYPE_ZBC)
534 return sprintf(buf, "host-managed\n");
535 if (sdkp->zoned == 1)
536 return sprintf(buf, "host-aware\n");
537 if (sdkp->zoned == 2)
538 return sprintf(buf, "drive-managed\n");
539 return sprintf(buf, "none\n");
541 static DEVICE_ATTR_RO(zoned_cap);
544 max_retries_store(struct device *dev, struct device_attribute *attr,
545 const char *buf, size_t count)
547 struct scsi_disk *sdkp = to_scsi_disk(dev);
548 struct scsi_device *sdev = sdkp->device;
551 err = kstrtoint(buf, 10, &retries);
555 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
556 sdkp->max_retries = retries;
560 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
566 max_retries_show(struct device *dev, struct device_attribute *attr,
569 struct scsi_disk *sdkp = to_scsi_disk(dev);
571 return sprintf(buf, "%d\n", sdkp->max_retries);
574 static DEVICE_ATTR_RW(max_retries);
576 static struct attribute *sd_disk_attrs[] = {
577 &dev_attr_cache_type.attr,
579 &dev_attr_allow_restart.attr,
580 &dev_attr_manage_start_stop.attr,
581 &dev_attr_protection_type.attr,
582 &dev_attr_protection_mode.attr,
583 &dev_attr_app_tag_own.attr,
584 &dev_attr_thin_provisioning.attr,
585 &dev_attr_provisioning_mode.attr,
586 &dev_attr_zeroing_mode.attr,
587 &dev_attr_max_write_same_blocks.attr,
588 &dev_attr_max_medium_access_timeouts.attr,
589 &dev_attr_zoned_cap.attr,
590 &dev_attr_max_retries.attr,
593 ATTRIBUTE_GROUPS(sd_disk);
595 static struct class sd_disk_class = {
597 .owner = THIS_MODULE,
598 .dev_release = scsi_disk_release,
599 .dev_groups = sd_disk_groups,
602 static const struct dev_pm_ops sd_pm_ops = {
603 .suspend = sd_suspend_system,
605 .poweroff = sd_suspend_system,
606 .restore = sd_resume,
607 .runtime_suspend = sd_suspend_runtime,
608 .runtime_resume = sd_resume_runtime,
611 static struct scsi_driver sd_template = {
614 .owner = THIS_MODULE,
616 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
618 .shutdown = sd_shutdown,
622 .init_command = sd_init_command,
623 .uninit_command = sd_uninit_command,
625 .eh_action = sd_eh_action,
626 .eh_reset = sd_eh_reset,
630 * Don't request a new module, as that could deadlock in multipath
633 static void sd_default_probe(dev_t devt)
638 * Device no to disk mapping:
640 * major disc2 disc p1
641 * |............|.............|....|....| <- dev_t
644 * Inside a major, we have 16k disks, however mapped non-
645 * contiguously. The first 16 disks are for major0, the next
646 * ones with major1, ... Disk 256 is for major0 again, disk 272
648 * As we stay compatible with our numbering scheme, we can reuse
649 * the well-know SCSI majors 8, 65--71, 136--143.
651 static int sd_major(int major_idx)
655 return SCSI_DISK0_MAJOR;
657 return SCSI_DISK1_MAJOR + major_idx - 1;
659 return SCSI_DISK8_MAJOR + major_idx - 8;
662 return 0; /* shut up gcc */
666 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
668 struct scsi_disk *sdkp = NULL;
670 mutex_lock(&sd_ref_mutex);
672 if (disk->private_data) {
673 sdkp = scsi_disk(disk);
674 if (scsi_device_get(sdkp->device) == 0)
675 get_device(&sdkp->dev);
679 mutex_unlock(&sd_ref_mutex);
683 static void scsi_disk_put(struct scsi_disk *sdkp)
685 struct scsi_device *sdev = sdkp->device;
687 mutex_lock(&sd_ref_mutex);
688 put_device(&sdkp->dev);
689 scsi_device_put(sdev);
690 mutex_unlock(&sd_ref_mutex);
693 #ifdef CONFIG_BLK_SED_OPAL
694 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
695 size_t len, bool send)
697 struct scsi_disk *sdkp = data;
698 struct scsi_device *sdev = sdkp->device;
702 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
704 put_unaligned_be16(spsp, &cdb[2]);
705 put_unaligned_be32(len, &cdb[6]);
707 ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
708 buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
710 return ret <= 0 ? ret : -EIO;
712 #endif /* CONFIG_BLK_SED_OPAL */
715 * Look up the DIX operation based on whether the command is read or
716 * write and whether dix and dif are enabled.
718 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
720 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
721 static const unsigned int ops[] = { /* wrt dix dif */
722 SCSI_PROT_NORMAL, /* 0 0 0 */
723 SCSI_PROT_READ_STRIP, /* 0 0 1 */
724 SCSI_PROT_READ_INSERT, /* 0 1 0 */
725 SCSI_PROT_READ_PASS, /* 0 1 1 */
726 SCSI_PROT_NORMAL, /* 1 0 0 */
727 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
728 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
729 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
732 return ops[write << 2 | dix << 1 | dif];
736 * Returns a mask of the protection flags that are valid for a given DIX
739 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
741 static const unsigned int flag_mask[] = {
742 [SCSI_PROT_NORMAL] = 0,
744 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
745 SCSI_PROT_GUARD_CHECK |
746 SCSI_PROT_REF_CHECK |
747 SCSI_PROT_REF_INCREMENT,
749 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
750 SCSI_PROT_IP_CHECKSUM,
752 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
753 SCSI_PROT_GUARD_CHECK |
754 SCSI_PROT_REF_CHECK |
755 SCSI_PROT_REF_INCREMENT |
756 SCSI_PROT_IP_CHECKSUM,
758 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
759 SCSI_PROT_REF_INCREMENT,
761 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
762 SCSI_PROT_REF_CHECK |
763 SCSI_PROT_REF_INCREMENT |
764 SCSI_PROT_IP_CHECKSUM,
766 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
767 SCSI_PROT_GUARD_CHECK |
768 SCSI_PROT_REF_CHECK |
769 SCSI_PROT_REF_INCREMENT |
770 SCSI_PROT_IP_CHECKSUM,
773 return flag_mask[prot_op];
776 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
777 unsigned int dix, unsigned int dif)
779 struct request *rq = scsi_cmd_to_rq(scmd);
780 struct bio *bio = rq->bio;
781 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
782 unsigned int protect = 0;
784 if (dix) { /* DIX Type 0, 1, 2, 3 */
785 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
786 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
788 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
789 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
792 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
793 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
795 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
796 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
799 if (dif) { /* DIX/DIF Type 1, 2, 3 */
800 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
802 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
803 protect = 3 << 5; /* Disable target PI checking */
805 protect = 1 << 5; /* Enable target PI checking */
808 scsi_set_prot_op(scmd, prot_op);
809 scsi_set_prot_type(scmd, dif);
810 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
815 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
817 struct request_queue *q = sdkp->disk->queue;
818 unsigned int logical_block_size = sdkp->device->sector_size;
819 unsigned int max_blocks = 0;
821 q->limits.discard_alignment =
822 sdkp->unmap_alignment * logical_block_size;
823 q->limits.discard_granularity =
824 max(sdkp->physical_block_size,
825 sdkp->unmap_granularity * logical_block_size);
826 sdkp->provisioning_mode = mode;
832 blk_queue_max_discard_sectors(q, 0);
833 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
837 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
838 (u32)SD_MAX_WS16_BLOCKS);
842 if (sdkp->device->unmap_limit_for_ws)
843 max_blocks = sdkp->max_unmap_blocks;
845 max_blocks = sdkp->max_ws_blocks;
847 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
851 if (sdkp->device->unmap_limit_for_ws)
852 max_blocks = sdkp->max_unmap_blocks;
854 max_blocks = sdkp->max_ws_blocks;
856 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
860 max_blocks = min_not_zero(sdkp->max_ws_blocks,
861 (u32)SD_MAX_WS10_BLOCKS);
865 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
866 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
869 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
871 struct scsi_device *sdp = cmd->device;
872 struct request *rq = scsi_cmd_to_rq(cmd);
873 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
874 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
875 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
876 unsigned int data_len = 24;
879 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
880 if (!rq->special_vec.bv_page)
881 return BLK_STS_RESOURCE;
882 clear_highpage(rq->special_vec.bv_page);
883 rq->special_vec.bv_offset = 0;
884 rq->special_vec.bv_len = data_len;
885 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
888 cmd->cmnd[0] = UNMAP;
891 buf = page_address(rq->special_vec.bv_page);
892 put_unaligned_be16(6 + 16, &buf[0]);
893 put_unaligned_be16(16, &buf[2]);
894 put_unaligned_be64(lba, &buf[8]);
895 put_unaligned_be32(nr_blocks, &buf[16]);
897 cmd->allowed = sdkp->max_retries;
898 cmd->transfersize = data_len;
899 rq->timeout = SD_TIMEOUT;
901 return scsi_alloc_sgtables(cmd);
904 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
907 struct scsi_device *sdp = cmd->device;
908 struct request *rq = scsi_cmd_to_rq(cmd);
909 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
910 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
911 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
912 u32 data_len = sdp->sector_size;
914 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
915 if (!rq->special_vec.bv_page)
916 return BLK_STS_RESOURCE;
917 clear_highpage(rq->special_vec.bv_page);
918 rq->special_vec.bv_offset = 0;
919 rq->special_vec.bv_len = data_len;
920 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
923 cmd->cmnd[0] = WRITE_SAME_16;
925 cmd->cmnd[1] = 0x8; /* UNMAP */
926 put_unaligned_be64(lba, &cmd->cmnd[2]);
927 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
929 cmd->allowed = sdkp->max_retries;
930 cmd->transfersize = data_len;
931 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
933 return scsi_alloc_sgtables(cmd);
936 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
939 struct scsi_device *sdp = cmd->device;
940 struct request *rq = scsi_cmd_to_rq(cmd);
941 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
942 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
943 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
944 u32 data_len = sdp->sector_size;
946 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
947 if (!rq->special_vec.bv_page)
948 return BLK_STS_RESOURCE;
949 clear_highpage(rq->special_vec.bv_page);
950 rq->special_vec.bv_offset = 0;
951 rq->special_vec.bv_len = data_len;
952 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
955 cmd->cmnd[0] = WRITE_SAME;
957 cmd->cmnd[1] = 0x8; /* UNMAP */
958 put_unaligned_be32(lba, &cmd->cmnd[2]);
959 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
961 cmd->allowed = sdkp->max_retries;
962 cmd->transfersize = data_len;
963 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
965 return scsi_alloc_sgtables(cmd);
968 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
970 struct request *rq = scsi_cmd_to_rq(cmd);
971 struct scsi_device *sdp = cmd->device;
972 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
973 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
974 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
976 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
977 switch (sdkp->zeroing_mode) {
978 case SD_ZERO_WS16_UNMAP:
979 return sd_setup_write_same16_cmnd(cmd, true);
980 case SD_ZERO_WS10_UNMAP:
981 return sd_setup_write_same10_cmnd(cmd, true);
985 if (sdp->no_write_same) {
986 rq->rq_flags |= RQF_QUIET;
987 return BLK_STS_TARGET;
990 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
991 return sd_setup_write_same16_cmnd(cmd, false);
993 return sd_setup_write_same10_cmnd(cmd, false);
996 static void sd_config_write_same(struct scsi_disk *sdkp)
998 struct request_queue *q = sdkp->disk->queue;
999 unsigned int logical_block_size = sdkp->device->sector_size;
1001 if (sdkp->device->no_write_same) {
1002 sdkp->max_ws_blocks = 0;
1006 /* Some devices can not handle block counts above 0xffff despite
1007 * supporting WRITE SAME(16). Consequently we default to 64k
1008 * blocks per I/O unless the device explicitly advertises a
1011 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1012 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1013 (u32)SD_MAX_WS16_BLOCKS);
1014 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1015 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1016 (u32)SD_MAX_WS10_BLOCKS);
1018 sdkp->device->no_write_same = 1;
1019 sdkp->max_ws_blocks = 0;
1022 if (sdkp->lbprz && sdkp->lbpws)
1023 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1024 else if (sdkp->lbprz && sdkp->lbpws10)
1025 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1026 else if (sdkp->max_ws_blocks)
1027 sdkp->zeroing_mode = SD_ZERO_WS;
1029 sdkp->zeroing_mode = SD_ZERO_WRITE;
1031 if (sdkp->max_ws_blocks &&
1032 sdkp->physical_block_size > logical_block_size) {
1034 * Reporting a maximum number of blocks that is not aligned
1035 * on the device physical size would cause a large write same
1036 * request to be split into physically unaligned chunks by
1037 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1038 * even if the caller of these functions took care to align the
1039 * large request. So make sure the maximum reported is aligned
1040 * to the device physical block size. This is only an optional
1041 * optimization for regular disks, but this is mandatory to
1042 * avoid failure of large write same requests directed at
1043 * sequential write required zones of host-managed ZBC disks.
1045 sdkp->max_ws_blocks =
1046 round_down(sdkp->max_ws_blocks,
1047 bytes_to_logical(sdkp->device,
1048 sdkp->physical_block_size));
1052 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1053 (logical_block_size >> 9));
1054 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1055 (logical_block_size >> 9));
1059 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1060 * @cmd: command to prepare
1062 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1063 * the preference indicated by the target device.
1065 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1067 struct request *rq = scsi_cmd_to_rq(cmd);
1068 struct scsi_device *sdp = cmd->device;
1069 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1070 struct bio *bio = rq->bio;
1071 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1072 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1075 if (sdkp->device->no_write_same)
1076 return BLK_STS_TARGET;
1078 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1080 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1082 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1084 cmd->cmnd[0] = WRITE_SAME_16;
1085 put_unaligned_be64(lba, &cmd->cmnd[2]);
1086 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1089 cmd->cmnd[0] = WRITE_SAME;
1090 put_unaligned_be32(lba, &cmd->cmnd[2]);
1091 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1094 cmd->transfersize = sdp->sector_size;
1095 cmd->allowed = sdkp->max_retries;
1098 * For WRITE SAME the data transferred via the DATA OUT buffer is
1099 * different from the amount of data actually written to the target.
1101 * We set up __data_len to the amount of data transferred via the
1102 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1103 * to transfer a single sector of data first, but then reset it to
1104 * the amount of data to be written right after so that the I/O path
1105 * knows how much to actually write.
1107 rq->__data_len = sdp->sector_size;
1108 ret = scsi_alloc_sgtables(cmd);
1109 rq->__data_len = blk_rq_bytes(rq);
1114 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1116 struct request *rq = scsi_cmd_to_rq(cmd);
1117 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1119 /* flush requests don't perform I/O, zero the S/G table */
1120 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1122 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1124 cmd->transfersize = 0;
1125 cmd->allowed = sdkp->max_retries;
1127 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1131 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1132 sector_t lba, unsigned int nr_blocks,
1133 unsigned char flags)
1135 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1136 if (unlikely(cmd->cmnd == NULL))
1137 return BLK_STS_RESOURCE;
1139 cmd->cmd_len = SD_EXT_CDB_SIZE;
1140 memset(cmd->cmnd, 0, cmd->cmd_len);
1142 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1143 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1144 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1145 cmd->cmnd[10] = flags;
1146 put_unaligned_be64(lba, &cmd->cmnd[12]);
1147 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1148 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1153 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1154 sector_t lba, unsigned int nr_blocks,
1155 unsigned char flags)
1158 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1159 cmd->cmnd[1] = flags;
1162 put_unaligned_be64(lba, &cmd->cmnd[2]);
1163 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1168 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1169 sector_t lba, unsigned int nr_blocks,
1170 unsigned char flags)
1173 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1174 cmd->cmnd[1] = flags;
1177 put_unaligned_be32(lba, &cmd->cmnd[2]);
1178 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1183 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1184 sector_t lba, unsigned int nr_blocks,
1185 unsigned char flags)
1187 /* Avoid that 0 blocks gets translated into 256 blocks. */
1188 if (WARN_ON_ONCE(nr_blocks == 0))
1189 return BLK_STS_IOERR;
1191 if (unlikely(flags & 0x8)) {
1193 * This happens only if this drive failed 10byte rw
1194 * command with ILLEGAL_REQUEST during operation and
1195 * thus turned off use_10_for_rw.
1197 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1198 return BLK_STS_IOERR;
1202 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1203 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1204 cmd->cmnd[2] = (lba >> 8) & 0xff;
1205 cmd->cmnd[3] = lba & 0xff;
1206 cmd->cmnd[4] = nr_blocks;
1212 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1214 struct request *rq = scsi_cmd_to_rq(cmd);
1215 struct scsi_device *sdp = cmd->device;
1216 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1217 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1219 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1220 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1221 bool write = rq_data_dir(rq) == WRITE;
1222 unsigned char protect, fua;
1227 ret = scsi_alloc_sgtables(cmd);
1228 if (ret != BLK_STS_OK)
1231 ret = BLK_STS_IOERR;
1232 if (!scsi_device_online(sdp) || sdp->changed) {
1233 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1237 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1238 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1242 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1243 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1248 * Some SD card readers can't handle accesses which touch the
1249 * last one or two logical blocks. Split accesses as needed.
1251 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1253 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1254 if (lba < threshold) {
1255 /* Access up to the threshold but not beyond */
1256 nr_blocks = threshold - lba;
1258 /* Access only a single logical block */
1263 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1264 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1269 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1270 dix = scsi_prot_sg_count(cmd);
1271 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1274 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1278 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1279 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1281 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1282 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1284 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1285 sdp->use_10_for_rw || protect) {
1286 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1289 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1293 if (unlikely(ret != BLK_STS_OK))
1297 * We shouldn't disconnect in the middle of a sector, so with a dumb
1298 * host adapter, it's safe to assume that we can at least transfer
1299 * this many bytes between each connect / disconnect.
1301 cmd->transfersize = sdp->sector_size;
1302 cmd->underflow = nr_blocks << 9;
1303 cmd->allowed = sdkp->max_retries;
1304 cmd->sdb.length = nr_blocks * sdp->sector_size;
1307 scmd_printk(KERN_INFO, cmd,
1308 "%s: block=%llu, count=%d\n", __func__,
1309 (unsigned long long)blk_rq_pos(rq),
1310 blk_rq_sectors(rq)));
1312 scmd_printk(KERN_INFO, cmd,
1313 "%s %d/%u 512 byte blocks.\n",
1314 write ? "writing" : "reading", nr_blocks,
1315 blk_rq_sectors(rq)));
1318 * This indicates that the command is ready from our end to be queued.
1322 scsi_free_sgtables(cmd);
1326 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1328 struct request *rq = scsi_cmd_to_rq(cmd);
1330 switch (req_op(rq)) {
1331 case REQ_OP_DISCARD:
1332 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1334 return sd_setup_unmap_cmnd(cmd);
1336 return sd_setup_write_same16_cmnd(cmd, true);
1338 return sd_setup_write_same10_cmnd(cmd, true);
1340 return sd_setup_write_same10_cmnd(cmd, false);
1342 return BLK_STS_TARGET;
1344 case REQ_OP_WRITE_ZEROES:
1345 return sd_setup_write_zeroes_cmnd(cmd);
1346 case REQ_OP_WRITE_SAME:
1347 return sd_setup_write_same_cmnd(cmd);
1349 return sd_setup_flush_cmnd(cmd);
1352 case REQ_OP_ZONE_APPEND:
1353 return sd_setup_read_write_cmnd(cmd);
1354 case REQ_OP_ZONE_RESET:
1355 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1357 case REQ_OP_ZONE_RESET_ALL:
1358 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1360 case REQ_OP_ZONE_OPEN:
1361 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1362 case REQ_OP_ZONE_CLOSE:
1363 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1364 case REQ_OP_ZONE_FINISH:
1365 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1368 return BLK_STS_NOTSUPP;
1372 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1374 struct request *rq = scsi_cmd_to_rq(SCpnt);
1377 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1378 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1380 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1384 mempool_free(cmnd, sd_cdb_pool);
1388 static bool sd_need_revalidate(struct block_device *bdev,
1389 struct scsi_disk *sdkp)
1391 if (sdkp->device->removable || sdkp->write_prot) {
1392 if (bdev_check_media_change(bdev))
1397 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1398 * nothing to do with partitions, BLKRRPART is used to force a full
1399 * revalidate after things like a format for historical reasons.
1401 return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1405 * sd_open - open a scsi disk device
1406 * @bdev: Block device of the scsi disk to open
1407 * @mode: FMODE_* mask
1409 * Returns 0 if successful. Returns a negated errno value in case
1412 * Note: This can be called from a user context (e.g. fsck(1) )
1413 * or from within the kernel (e.g. as a result of a mount(1) ).
1414 * In the latter case @inode and @filp carry an abridged amount
1415 * of information as noted above.
1417 * Locking: called with bdev->bd_disk->open_mutex held.
1419 static int sd_open(struct block_device *bdev, fmode_t mode)
1421 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1422 struct scsi_device *sdev;
1428 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1430 sdev = sdkp->device;
1433 * If the device is in error recovery, wait until it is done.
1434 * If the device is offline, then disallow any access to it.
1437 if (!scsi_block_when_processing_errors(sdev))
1440 if (sd_need_revalidate(bdev, sdkp))
1441 sd_revalidate_disk(bdev->bd_disk);
1444 * If the drive is empty, just let the open fail.
1446 retval = -ENOMEDIUM;
1447 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1451 * If the device has the write protect tab set, have the open fail
1452 * if the user expects to be able to write to the thing.
1455 if (sdkp->write_prot && (mode & FMODE_WRITE))
1459 * It is possible that the disk changing stuff resulted in
1460 * the device being taken offline. If this is the case,
1461 * report this to the user, and don't pretend that the
1462 * open actually succeeded.
1465 if (!scsi_device_online(sdev))
1468 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1469 if (scsi_block_when_processing_errors(sdev))
1470 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1476 scsi_disk_put(sdkp);
1481 * sd_release - invoked when the (last) close(2) is called on this
1483 * @disk: disk to release
1484 * @mode: FMODE_* mask
1488 * Note: may block (uninterruptible) if error recovery is underway
1491 * Locking: called with bdev->bd_disk->open_mutex held.
1493 static void sd_release(struct gendisk *disk, fmode_t mode)
1495 struct scsi_disk *sdkp = scsi_disk(disk);
1496 struct scsi_device *sdev = sdkp->device;
1498 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1500 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1501 if (scsi_block_when_processing_errors(sdev))
1502 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1505 scsi_disk_put(sdkp);
1508 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1510 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1511 struct scsi_device *sdp = sdkp->device;
1512 struct Scsi_Host *host = sdp->host;
1513 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1516 /* default to most commonly used values */
1517 diskinfo[0] = 0x40; /* 1 << 6 */
1518 diskinfo[1] = 0x20; /* 1 << 5 */
1519 diskinfo[2] = capacity >> 11;
1521 /* override with calculated, extended default, or driver values */
1522 if (host->hostt->bios_param)
1523 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1525 scsicam_bios_param(bdev, capacity, diskinfo);
1527 geo->heads = diskinfo[0];
1528 geo->sectors = diskinfo[1];
1529 geo->cylinders = diskinfo[2];
1534 * sd_ioctl - process an ioctl
1535 * @bdev: target block device
1536 * @mode: FMODE_* mask
1537 * @cmd: ioctl command number
1538 * @arg: this is third argument given to ioctl(2) system call.
1539 * Often contains a pointer.
1541 * Returns 0 if successful (some ioctls return positive numbers on
1542 * success as well). Returns a negated errno value in case of error.
1544 * Note: most ioctls are forward onto the block subsystem or further
1545 * down in the scsi subsystem.
1547 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1548 unsigned int cmd, unsigned long arg)
1550 struct gendisk *disk = bdev->bd_disk;
1551 struct scsi_disk *sdkp = scsi_disk(disk);
1552 struct scsi_device *sdp = sdkp->device;
1553 void __user *p = (void __user *)arg;
1556 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1557 "cmd=0x%x\n", disk->disk_name, cmd));
1559 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1560 return -ENOIOCTLCMD;
1563 * If we are in the middle of error recovery, don't let anyone
1564 * else try and use this device. Also, if error recovery fails, it
1565 * may try and take the device offline, in which case all further
1566 * access to the device is prohibited.
1568 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1569 (mode & FMODE_NDELAY) != 0);
1573 if (is_sed_ioctl(cmd))
1574 return sed_ioctl(sdkp->opal_dev, cmd, p);
1575 return scsi_ioctl(sdp, disk, mode, cmd, p);
1578 static void set_media_not_present(struct scsi_disk *sdkp)
1580 if (sdkp->media_present)
1581 sdkp->device->changed = 1;
1583 if (sdkp->device->removable) {
1584 sdkp->media_present = 0;
1589 static int media_not_present(struct scsi_disk *sdkp,
1590 struct scsi_sense_hdr *sshdr)
1592 if (!scsi_sense_valid(sshdr))
1595 /* not invoked for commands that could return deferred errors */
1596 switch (sshdr->sense_key) {
1597 case UNIT_ATTENTION:
1599 /* medium not present */
1600 if (sshdr->asc == 0x3A) {
1601 set_media_not_present(sdkp);
1609 * sd_check_events - check media events
1610 * @disk: kernel device descriptor
1611 * @clearing: disk events currently being cleared
1613 * Returns mask of DISK_EVENT_*.
1615 * Note: this function is invoked from the block subsystem.
1617 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1619 struct scsi_disk *sdkp = scsi_disk_get(disk);
1620 struct scsi_device *sdp;
1628 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1631 * If the device is offline, don't send any commands - just pretend as
1632 * if the command failed. If the device ever comes back online, we
1633 * can deal with it then. It is only because of unrecoverable errors
1634 * that we would ever take a device offline in the first place.
1636 if (!scsi_device_online(sdp)) {
1637 set_media_not_present(sdkp);
1642 * Using TEST_UNIT_READY enables differentiation between drive with
1643 * no cartridge loaded - NOT READY, drive with changed cartridge -
1644 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1646 * Drives that auto spin down. eg iomega jaz 1G, will be started
1647 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1648 * sd_revalidate() is called.
1650 if (scsi_block_when_processing_errors(sdp)) {
1651 struct scsi_sense_hdr sshdr = { 0, };
1653 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1656 /* failed to execute TUR, assume media not present */
1657 if (retval < 0 || host_byte(retval)) {
1658 set_media_not_present(sdkp);
1662 if (media_not_present(sdkp, &sshdr))
1667 * For removable scsi disk we have to recognise the presence
1668 * of a disk in the drive.
1670 if (!sdkp->media_present)
1672 sdkp->media_present = 1;
1675 * sdp->changed is set under the following conditions:
1677 * Medium present state has changed in either direction.
1678 * Device has indicated UNIT_ATTENTION.
1680 disk_changed = sdp->changed;
1682 scsi_disk_put(sdkp);
1683 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1686 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1689 struct scsi_device *sdp = sdkp->device;
1690 const int timeout = sdp->request_queue->rq_timeout
1691 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1692 struct scsi_sense_hdr my_sshdr;
1694 if (!scsi_device_online(sdp))
1697 /* caller might not be interested in sense, but we need it */
1701 for (retries = 3; retries > 0; --retries) {
1702 unsigned char cmd[10] = { 0 };
1704 cmd[0] = SYNCHRONIZE_CACHE;
1706 * Leave the rest of the command zero to indicate
1709 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1710 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1716 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1721 if (scsi_status_is_check_condition(res) &&
1722 scsi_sense_valid(sshdr)) {
1723 sd_print_sense_hdr(sdkp, sshdr);
1725 /* we need to evaluate the error return */
1726 if (sshdr->asc == 0x3a || /* medium not present */
1727 sshdr->asc == 0x20 || /* invalid command */
1728 (sshdr->asc == 0x74 && sshdr->ascq == 0x71)) /* drive is password locked */
1729 /* this is no error here */
1733 switch (host_byte(res)) {
1734 /* ignore errors due to racing a disconnection */
1735 case DID_BAD_TARGET:
1736 case DID_NO_CONNECT:
1738 /* signal the upper layer it might try again */
1742 case DID_SOFT_ERROR:
1751 static void sd_rescan(struct device *dev)
1753 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1755 sd_revalidate_disk(sdkp->disk);
1758 static char sd_pr_type(enum pr_type type)
1761 case PR_WRITE_EXCLUSIVE:
1763 case PR_EXCLUSIVE_ACCESS:
1765 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1767 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1769 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1771 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1778 static int sd_pr_command(struct block_device *bdev, u8 sa,
1779 u64 key, u64 sa_key, u8 type, u8 flags)
1781 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1782 struct scsi_device *sdev = sdkp->device;
1783 struct scsi_sense_hdr sshdr;
1785 u8 cmd[16] = { 0, };
1786 u8 data[24] = { 0, };
1788 cmd[0] = PERSISTENT_RESERVE_OUT;
1791 put_unaligned_be32(sizeof(data), &cmd[5]);
1793 put_unaligned_be64(key, &data[0]);
1794 put_unaligned_be64(sa_key, &data[8]);
1797 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1798 &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1800 if (scsi_status_is_check_condition(result) &&
1801 scsi_sense_valid(&sshdr)) {
1802 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1803 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1809 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1812 if (flags & ~PR_FL_IGNORE_KEY)
1814 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1815 old_key, new_key, 0,
1816 (1 << 0) /* APTPL */);
1819 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1824 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1827 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1829 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1832 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1833 enum pr_type type, bool abort)
1835 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1836 sd_pr_type(type), 0);
1839 static int sd_pr_clear(struct block_device *bdev, u64 key)
1841 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1844 static const struct pr_ops sd_pr_ops = {
1845 .pr_register = sd_pr_register,
1846 .pr_reserve = sd_pr_reserve,
1847 .pr_release = sd_pr_release,
1848 .pr_preempt = sd_pr_preempt,
1849 .pr_clear = sd_pr_clear,
1852 static const struct block_device_operations sd_fops = {
1853 .owner = THIS_MODULE,
1855 .release = sd_release,
1857 .getgeo = sd_getgeo,
1858 .compat_ioctl = blkdev_compat_ptr_ioctl,
1859 .check_events = sd_check_events,
1860 .unlock_native_capacity = sd_unlock_native_capacity,
1861 .report_zones = sd_zbc_report_zones,
1862 .pr_ops = &sd_pr_ops,
1866 * sd_eh_reset - reset error handling callback
1867 * @scmd: sd-issued command that has failed
1869 * This function is called by the SCSI midlayer before starting
1870 * SCSI EH. When counting medium access failures we have to be
1871 * careful to register it only only once per device and SCSI EH run;
1872 * there might be several timed out commands which will cause the
1873 * 'max_medium_access_timeouts' counter to trigger after the first
1874 * SCSI EH run already and set the device to offline.
1875 * So this function resets the internal counter before starting SCSI EH.
1877 static void sd_eh_reset(struct scsi_cmnd *scmd)
1879 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1881 /* New SCSI EH run, reset gate variable */
1882 sdkp->ignore_medium_access_errors = false;
1886 * sd_eh_action - error handling callback
1887 * @scmd: sd-issued command that has failed
1888 * @eh_disp: The recovery disposition suggested by the midlayer
1890 * This function is called by the SCSI midlayer upon completion of an
1891 * error test command (currently TEST UNIT READY). The result of sending
1892 * the eh command is passed in eh_disp. We're looking for devices that
1893 * fail medium access commands but are OK with non access commands like
1894 * test unit ready (so wrongly see the device as having a successful
1897 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1899 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1900 struct scsi_device *sdev = scmd->device;
1902 if (!scsi_device_online(sdev) ||
1903 !scsi_medium_access_command(scmd) ||
1904 host_byte(scmd->result) != DID_TIME_OUT ||
1909 * The device has timed out executing a medium access command.
1910 * However, the TEST UNIT READY command sent during error
1911 * handling completed successfully. Either the device is in the
1912 * process of recovering or has it suffered an internal failure
1913 * that prevents access to the storage medium.
1915 if (!sdkp->ignore_medium_access_errors) {
1916 sdkp->medium_access_timed_out++;
1917 sdkp->ignore_medium_access_errors = true;
1921 * If the device keeps failing read/write commands but TEST UNIT
1922 * READY always completes successfully we assume that medium
1923 * access is no longer possible and take the device offline.
1925 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1926 scmd_printk(KERN_ERR, scmd,
1927 "Medium access timeout failure. Offlining disk!\n");
1928 mutex_lock(&sdev->state_mutex);
1929 scsi_device_set_state(sdev, SDEV_OFFLINE);
1930 mutex_unlock(&sdev->state_mutex);
1938 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1940 struct request *req = scsi_cmd_to_rq(scmd);
1941 struct scsi_device *sdev = scmd->device;
1942 unsigned int transferred, good_bytes;
1943 u64 start_lba, end_lba, bad_lba;
1946 * Some commands have a payload smaller than the device logical
1947 * block size (e.g. INQUIRY on a 4K disk).
1949 if (scsi_bufflen(scmd) <= sdev->sector_size)
1952 /* Check if we have a 'bad_lba' information */
1953 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1954 SCSI_SENSE_BUFFERSIZE,
1959 * If the bad lba was reported incorrectly, we have no idea where
1962 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1963 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1964 if (bad_lba < start_lba || bad_lba >= end_lba)
1968 * resid is optional but mostly filled in. When it's unused,
1969 * its value is zero, so we assume the whole buffer transferred
1971 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1973 /* This computation should always be done in terms of the
1974 * resolution of the device's medium.
1976 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1978 return min(good_bytes, transferred);
1982 * sd_done - bottom half handler: called when the lower level
1983 * driver has completed (successfully or otherwise) a scsi command.
1984 * @SCpnt: mid-level's per command structure.
1986 * Note: potentially run from within an ISR. Must not block.
1988 static int sd_done(struct scsi_cmnd *SCpnt)
1990 int result = SCpnt->result;
1991 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1992 unsigned int sector_size = SCpnt->device->sector_size;
1994 struct scsi_sense_hdr sshdr;
1995 struct request *req = scsi_cmd_to_rq(SCpnt);
1996 struct scsi_disk *sdkp = scsi_disk(req->rq_disk);
1997 int sense_valid = 0;
1998 int sense_deferred = 0;
2000 switch (req_op(req)) {
2001 case REQ_OP_DISCARD:
2002 case REQ_OP_WRITE_ZEROES:
2003 case REQ_OP_WRITE_SAME:
2004 case REQ_OP_ZONE_RESET:
2005 case REQ_OP_ZONE_RESET_ALL:
2006 case REQ_OP_ZONE_OPEN:
2007 case REQ_OP_ZONE_CLOSE:
2008 case REQ_OP_ZONE_FINISH:
2010 good_bytes = blk_rq_bytes(req);
2011 scsi_set_resid(SCpnt, 0);
2014 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2019 * In case of bogus fw or device, we could end up having
2020 * an unaligned partial completion. Check this here and force
2023 resid = scsi_get_resid(SCpnt);
2024 if (resid & (sector_size - 1)) {
2025 sd_printk(KERN_INFO, sdkp,
2026 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2027 resid, sector_size);
2028 scsi_print_command(SCpnt);
2029 resid = min(scsi_bufflen(SCpnt),
2030 round_up(resid, sector_size));
2031 scsi_set_resid(SCpnt, resid);
2036 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2038 sense_deferred = scsi_sense_is_deferred(&sshdr);
2040 sdkp->medium_access_timed_out = 0;
2042 if (!scsi_status_is_check_condition(result) &&
2043 (!sense_valid || sense_deferred))
2046 switch (sshdr.sense_key) {
2047 case HARDWARE_ERROR:
2049 good_bytes = sd_completed_bytes(SCpnt);
2051 case RECOVERED_ERROR:
2052 good_bytes = scsi_bufflen(SCpnt);
2055 /* This indicates a false check condition, so ignore it. An
2056 * unknown amount of data was transferred so treat it as an
2060 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2062 case ABORTED_COMMAND:
2063 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2064 good_bytes = sd_completed_bytes(SCpnt);
2066 case ILLEGAL_REQUEST:
2067 switch (sshdr.asc) {
2068 case 0x10: /* DIX: Host detected corruption */
2069 good_bytes = sd_completed_bytes(SCpnt);
2071 case 0x20: /* INVALID COMMAND OPCODE */
2072 case 0x24: /* INVALID FIELD IN CDB */
2073 switch (SCpnt->cmnd[0]) {
2075 sd_config_discard(sdkp, SD_LBP_DISABLE);
2079 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2080 sd_config_discard(sdkp, SD_LBP_DISABLE);
2082 sdkp->device->no_write_same = 1;
2083 sd_config_write_same(sdkp);
2084 req->rq_flags |= RQF_QUIET;
2095 if (sd_is_zoned(sdkp))
2096 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2098 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2099 "sd_done: completed %d of %d bytes\n",
2100 good_bytes, scsi_bufflen(SCpnt)));
2106 * spinup disk - called only in sd_revalidate_disk()
2109 sd_spinup_disk(struct scsi_disk *sdkp)
2111 unsigned char cmd[10];
2112 unsigned long spintime_expire = 0;
2113 int retries, spintime;
2114 unsigned int the_result;
2115 struct scsi_sense_hdr sshdr;
2116 int sense_valid = 0;
2120 /* Spin up drives, as required. Only do this at boot time */
2121 /* Spinup needs to be done for module loads too. */
2126 cmd[0] = TEST_UNIT_READY;
2127 memset((void *) &cmd[1], 0, 9);
2129 the_result = scsi_execute_req(sdkp->device, cmd,
2132 sdkp->max_retries, NULL);
2135 * If the drive has indicated to us that it
2136 * doesn't have any media in it, don't bother
2137 * with any more polling.
2139 if (media_not_present(sdkp, &sshdr)) {
2140 sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2145 sense_valid = scsi_sense_valid(&sshdr);
2147 } while (retries < 3 &&
2148 (!scsi_status_is_good(the_result) ||
2149 (scsi_status_is_check_condition(the_result) &&
2150 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2152 if (!scsi_status_is_check_condition(the_result)) {
2153 /* no sense, TUR either succeeded or failed
2154 * with a status error */
2155 if(!spintime && !scsi_status_is_good(the_result)) {
2156 sd_print_result(sdkp, "Test Unit Ready failed",
2163 * The device does not want the automatic start to be issued.
2165 if (sdkp->device->no_start_on_add)
2168 if (sense_valid && sshdr.sense_key == NOT_READY) {
2169 if (sshdr.asc == 4 && sshdr.ascq == 3)
2170 break; /* manual intervention required */
2171 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2172 break; /* standby */
2173 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2174 break; /* unavailable */
2175 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2176 break; /* sanitize in progress */
2178 * Issue command to spin up drive when not ready
2181 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2182 cmd[0] = START_STOP;
2183 cmd[1] = 1; /* Return immediately */
2184 memset((void *) &cmd[2], 0, 8);
2185 cmd[4] = 1; /* Start spin cycle */
2186 if (sdkp->device->start_stop_pwr_cond)
2188 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2190 SD_TIMEOUT, sdkp->max_retries,
2192 spintime_expire = jiffies + 100 * HZ;
2195 /* Wait 1 second for next try */
2197 printk(KERN_CONT ".");
2200 * Wait for USB flash devices with slow firmware.
2201 * Yes, this sense key/ASC combination shouldn't
2202 * occur here. It's characteristic of these devices.
2204 } else if (sense_valid &&
2205 sshdr.sense_key == UNIT_ATTENTION &&
2206 sshdr.asc == 0x28) {
2208 spintime_expire = jiffies + 5 * HZ;
2211 /* Wait 1 second for next try */
2214 /* we don't understand the sense code, so it's
2215 * probably pointless to loop */
2217 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2218 sd_print_sense_hdr(sdkp, &sshdr);
2223 } while (spintime && time_before_eq(jiffies, spintime_expire));
2226 if (scsi_status_is_good(the_result))
2227 printk(KERN_CONT "ready\n");
2229 printk(KERN_CONT "not responding...\n");
2234 * Determine whether disk supports Data Integrity Field.
2236 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2238 struct scsi_device *sdp = sdkp->device;
2242 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2243 sdkp->protection_type = 0;
2247 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2249 if (type > T10_PI_TYPE3_PROTECTION)
2251 else if (scsi_host_dif_capable(sdp->host, type))
2254 if (sdkp->first_scan || type != sdkp->protection_type)
2257 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2258 " protection type %u. Disabling disk!\n",
2262 sd_printk(KERN_NOTICE, sdkp,
2263 "Enabling DIF Type %u protection\n", type);
2266 sd_printk(KERN_NOTICE, sdkp,
2267 "Disabling DIF Type %u protection\n", type);
2271 sdkp->protection_type = type;
2276 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2277 struct scsi_sense_hdr *sshdr, int sense_valid,
2281 sd_print_sense_hdr(sdkp, sshdr);
2283 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2286 * Set dirty bit for removable devices if not ready -
2287 * sometimes drives will not report this properly.
2289 if (sdp->removable &&
2290 sense_valid && sshdr->sense_key == NOT_READY)
2291 set_media_not_present(sdkp);
2294 * We used to set media_present to 0 here to indicate no media
2295 * in the drive, but some drives fail read capacity even with
2296 * media present, so we can't do that.
2298 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2302 #if RC16_LEN > SD_BUF_SIZE
2303 #error RC16_LEN must not be more than SD_BUF_SIZE
2306 #define READ_CAPACITY_RETRIES_ON_RESET 10
2308 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2309 unsigned char *buffer)
2311 unsigned char cmd[16];
2312 struct scsi_sense_hdr sshdr;
2313 int sense_valid = 0;
2315 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2316 unsigned int alignment;
2317 unsigned long long lba;
2318 unsigned sector_size;
2320 if (sdp->no_read_capacity_16)
2325 cmd[0] = SERVICE_ACTION_IN_16;
2326 cmd[1] = SAI_READ_CAPACITY_16;
2328 memset(buffer, 0, RC16_LEN);
2330 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2331 buffer, RC16_LEN, &sshdr,
2332 SD_TIMEOUT, sdkp->max_retries, NULL);
2334 if (media_not_present(sdkp, &sshdr))
2337 if (the_result > 0) {
2338 sense_valid = scsi_sense_valid(&sshdr);
2340 sshdr.sense_key == ILLEGAL_REQUEST &&
2341 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2343 /* Invalid Command Operation Code or
2344 * Invalid Field in CDB, just retry
2345 * silently with RC10 */
2348 sshdr.sense_key == UNIT_ATTENTION &&
2349 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2350 /* Device reset might occur several times,
2351 * give it one more chance */
2352 if (--reset_retries > 0)
2357 } while (the_result && retries);
2360 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2361 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2365 sector_size = get_unaligned_be32(&buffer[8]);
2366 lba = get_unaligned_be64(&buffer[0]);
2368 if (sd_read_protection_type(sdkp, buffer) < 0) {
2373 /* Logical blocks per physical block exponent */
2374 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2377 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2379 /* Lowest aligned logical block */
2380 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2381 blk_queue_alignment_offset(sdp->request_queue, alignment);
2382 if (alignment && sdkp->first_scan)
2383 sd_printk(KERN_NOTICE, sdkp,
2384 "physical block alignment offset: %u\n", alignment);
2386 if (buffer[14] & 0x80) { /* LBPME */
2389 if (buffer[14] & 0x40) /* LBPRZ */
2392 sd_config_discard(sdkp, SD_LBP_WS16);
2395 sdkp->capacity = lba + 1;
2399 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2400 unsigned char *buffer)
2402 unsigned char cmd[16];
2403 struct scsi_sense_hdr sshdr;
2404 int sense_valid = 0;
2406 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2408 unsigned sector_size;
2411 cmd[0] = READ_CAPACITY;
2412 memset(&cmd[1], 0, 9);
2413 memset(buffer, 0, 8);
2415 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2417 SD_TIMEOUT, sdkp->max_retries, NULL);
2419 if (media_not_present(sdkp, &sshdr))
2422 if (the_result > 0) {
2423 sense_valid = scsi_sense_valid(&sshdr);
2425 sshdr.sense_key == UNIT_ATTENTION &&
2426 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2427 /* Device reset might occur several times,
2428 * give it one more chance */
2429 if (--reset_retries > 0)
2434 } while (the_result && retries);
2437 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2438 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2442 sector_size = get_unaligned_be32(&buffer[4]);
2443 lba = get_unaligned_be32(&buffer[0]);
2445 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2446 /* Some buggy (usb cardreader) devices return an lba of
2447 0xffffffff when the want to report a size of 0 (with
2448 which they really mean no media is present) */
2450 sdkp->physical_block_size = sector_size;
2454 sdkp->capacity = lba + 1;
2455 sdkp->physical_block_size = sector_size;
2459 static int sd_try_rc16_first(struct scsi_device *sdp)
2461 if (sdp->host->max_cmd_len < 16)
2463 if (sdp->try_rc_10_first)
2465 if (sdp->scsi_level > SCSI_SPC_2)
2467 if (scsi_device_protection(sdp))
2473 * read disk capacity
2476 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2479 struct scsi_device *sdp = sdkp->device;
2481 if (sd_try_rc16_first(sdp)) {
2482 sector_size = read_capacity_16(sdkp, sdp, buffer);
2483 if (sector_size == -EOVERFLOW)
2485 if (sector_size == -ENODEV)
2487 if (sector_size < 0)
2488 sector_size = read_capacity_10(sdkp, sdp, buffer);
2489 if (sector_size < 0)
2492 sector_size = read_capacity_10(sdkp, sdp, buffer);
2493 if (sector_size == -EOVERFLOW)
2495 if (sector_size < 0)
2497 if ((sizeof(sdkp->capacity) > 4) &&
2498 (sdkp->capacity > 0xffffffffULL)) {
2499 int old_sector_size = sector_size;
2500 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2501 "Trying to use READ CAPACITY(16).\n");
2502 sector_size = read_capacity_16(sdkp, sdp, buffer);
2503 if (sector_size < 0) {
2504 sd_printk(KERN_NOTICE, sdkp,
2505 "Using 0xffffffff as device size\n");
2506 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2507 sector_size = old_sector_size;
2510 /* Remember that READ CAPACITY(16) succeeded */
2511 sdp->try_rc_10_first = 0;
2515 /* Some devices are known to return the total number of blocks,
2516 * not the highest block number. Some devices have versions
2517 * which do this and others which do not. Some devices we might
2518 * suspect of doing this but we don't know for certain.
2520 * If we know the reported capacity is wrong, decrement it. If
2521 * we can only guess, then assume the number of blocks is even
2522 * (usually true but not always) and err on the side of lowering
2525 if (sdp->fix_capacity ||
2526 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2527 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2528 "from its reported value: %llu\n",
2529 (unsigned long long) sdkp->capacity);
2534 if (sector_size == 0) {
2536 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2540 if (sector_size != 512 &&
2541 sector_size != 1024 &&
2542 sector_size != 2048 &&
2543 sector_size != 4096) {
2544 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2547 * The user might want to re-format the drive with
2548 * a supported sectorsize. Once this happens, it
2549 * would be relatively trivial to set the thing up.
2550 * For this reason, we leave the thing in the table.
2554 * set a bogus sector size so the normal read/write
2555 * logic in the block layer will eventually refuse any
2556 * request on this device without tripping over power
2557 * of two sector size assumptions
2561 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2562 blk_queue_physical_block_size(sdp->request_queue,
2563 sdkp->physical_block_size);
2564 sdkp->device->sector_size = sector_size;
2566 if (sdkp->capacity > 0xffffffff)
2567 sdp->use_16_for_rw = 1;
2572 * Print disk capacity
2575 sd_print_capacity(struct scsi_disk *sdkp,
2576 sector_t old_capacity)
2578 int sector_size = sdkp->device->sector_size;
2579 char cap_str_2[10], cap_str_10[10];
2581 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2584 string_get_size(sdkp->capacity, sector_size,
2585 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2586 string_get_size(sdkp->capacity, sector_size,
2587 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2589 sd_printk(KERN_NOTICE, sdkp,
2590 "%llu %d-byte logical blocks: (%s/%s)\n",
2591 (unsigned long long)sdkp->capacity,
2592 sector_size, cap_str_10, cap_str_2);
2594 if (sdkp->physical_block_size != sector_size)
2595 sd_printk(KERN_NOTICE, sdkp,
2596 "%u-byte physical blocks\n",
2597 sdkp->physical_block_size);
2600 /* called with buffer of length 512 */
2602 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2603 unsigned char *buffer, int len, struct scsi_mode_data *data,
2604 struct scsi_sense_hdr *sshdr)
2606 return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2607 SD_TIMEOUT, sdkp->max_retries, data,
2612 * read write protect setting, if possible - called only in sd_revalidate_disk()
2613 * called with buffer of length SD_BUF_SIZE
2616 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2619 struct scsi_device *sdp = sdkp->device;
2620 struct scsi_mode_data data;
2621 int old_wp = sdkp->write_prot;
2623 set_disk_ro(sdkp->disk, 0);
2624 if (sdp->skip_ms_page_3f) {
2625 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2629 if (sdp->use_192_bytes_for_3f) {
2630 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2633 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2634 * We have to start carefully: some devices hang if we ask
2635 * for more than is available.
2637 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2640 * Second attempt: ask for page 0 When only page 0 is
2641 * implemented, a request for page 3F may return Sense Key
2642 * 5: Illegal Request, Sense Code 24: Invalid field in
2646 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2649 * Third attempt: ask 255 bytes, as we did earlier.
2652 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2657 sd_first_printk(KERN_WARNING, sdkp,
2658 "Test WP failed, assume Write Enabled\n");
2660 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2661 set_disk_ro(sdkp->disk, sdkp->write_prot);
2662 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2663 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2664 sdkp->write_prot ? "on" : "off");
2665 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2671 * sd_read_cache_type - called only from sd_revalidate_disk()
2672 * called with buffer of length SD_BUF_SIZE
2675 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2678 struct scsi_device *sdp = sdkp->device;
2683 struct scsi_mode_data data;
2684 struct scsi_sense_hdr sshdr;
2685 int old_wce = sdkp->WCE;
2686 int old_rcd = sdkp->RCD;
2687 int old_dpofua = sdkp->DPOFUA;
2690 if (sdkp->cache_override)
2694 if (sdp->skip_ms_page_8) {
2695 if (sdp->type == TYPE_RBC)
2698 if (sdp->skip_ms_page_3f)
2701 if (sdp->use_192_bytes_for_3f)
2705 } else if (sdp->type == TYPE_RBC) {
2713 /* cautiously ask */
2714 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2720 if (!data.header_length) {
2723 sd_first_printk(KERN_ERR, sdkp,
2724 "Missing header in MODE_SENSE response\n");
2727 /* that went OK, now ask for the proper length */
2731 * We're only interested in the first three bytes, actually.
2732 * But the data cache page is defined for the first 20.
2736 else if (len > SD_BUF_SIZE) {
2737 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2738 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2741 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2745 if (len > first_len)
2746 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2750 int offset = data.header_length + data.block_descriptor_length;
2752 while (offset < len) {
2753 u8 page_code = buffer[offset] & 0x3F;
2754 u8 spf = buffer[offset] & 0x40;
2756 if (page_code == 8 || page_code == 6) {
2757 /* We're interested only in the first 3 bytes.
2759 if (len - offset <= 2) {
2760 sd_first_printk(KERN_ERR, sdkp,
2761 "Incomplete mode parameter "
2765 modepage = page_code;
2769 /* Go to the next page */
2770 if (spf && len - offset > 3)
2771 offset += 4 + (buffer[offset+2] << 8) +
2773 else if (!spf && len - offset > 1)
2774 offset += 2 + buffer[offset+1];
2776 sd_first_printk(KERN_ERR, sdkp,
2778 "parameter data\n");
2784 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2788 if (modepage == 8) {
2789 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2790 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2792 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2796 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2797 if (sdp->broken_fua) {
2798 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2800 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2801 !sdkp->device->use_16_for_rw) {
2802 sd_first_printk(KERN_NOTICE, sdkp,
2803 "Uses READ/WRITE(6), disabling FUA\n");
2807 /* No cache flush allowed for write protected devices */
2808 if (sdkp->WCE && sdkp->write_prot)
2811 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2812 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2813 sd_printk(KERN_NOTICE, sdkp,
2814 "Write cache: %s, read cache: %s, %s\n",
2815 sdkp->WCE ? "enabled" : "disabled",
2816 sdkp->RCD ? "disabled" : "enabled",
2817 sdkp->DPOFUA ? "supports DPO and FUA"
2818 : "doesn't support DPO or FUA");
2824 if (scsi_sense_valid(&sshdr) &&
2825 sshdr.sense_key == ILLEGAL_REQUEST &&
2826 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2827 /* Invalid field in CDB */
2828 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2830 sd_first_printk(KERN_ERR, sdkp,
2831 "Asking for cache data failed\n");
2834 if (sdp->wce_default_on) {
2835 sd_first_printk(KERN_NOTICE, sdkp,
2836 "Assuming drive cache: write back\n");
2839 sd_first_printk(KERN_ERR, sdkp,
2840 "Assuming drive cache: write through\n");
2848 * The ATO bit indicates whether the DIF application tag is available
2849 * for use by the operating system.
2851 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2854 struct scsi_device *sdp = sdkp->device;
2855 struct scsi_mode_data data;
2856 struct scsi_sense_hdr sshdr;
2858 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2861 if (sdkp->protection_type == 0)
2864 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2865 sdkp->max_retries, &data, &sshdr);
2867 if (res < 0 || !data.header_length ||
2869 sd_first_printk(KERN_WARNING, sdkp,
2870 "getting Control mode page failed, assume no ATO\n");
2872 if (scsi_sense_valid(&sshdr))
2873 sd_print_sense_hdr(sdkp, &sshdr);
2878 offset = data.header_length + data.block_descriptor_length;
2880 if ((buffer[offset] & 0x3f) != 0x0a) {
2881 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2885 if ((buffer[offset + 5] & 0x80) == 0)
2894 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2895 * @sdkp: disk to query
2897 static void sd_read_block_limits(struct scsi_disk *sdkp)
2899 unsigned int sector_sz = sdkp->device->sector_size;
2900 const int vpd_len = 64;
2901 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2904 /* Block Limits VPD */
2905 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2908 blk_queue_io_min(sdkp->disk->queue,
2909 get_unaligned_be16(&buffer[6]) * sector_sz);
2911 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2912 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2914 if (buffer[3] == 0x3c) {
2915 unsigned int lba_count, desc_count;
2917 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2922 lba_count = get_unaligned_be32(&buffer[20]);
2923 desc_count = get_unaligned_be32(&buffer[24]);
2925 if (lba_count && desc_count)
2926 sdkp->max_unmap_blocks = lba_count;
2928 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2930 if (buffer[32] & 0x80)
2931 sdkp->unmap_alignment =
2932 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2934 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2936 if (sdkp->max_unmap_blocks)
2937 sd_config_discard(sdkp, SD_LBP_UNMAP);
2939 sd_config_discard(sdkp, SD_LBP_WS16);
2941 } else { /* LBP VPD page tells us what to use */
2942 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2943 sd_config_discard(sdkp, SD_LBP_UNMAP);
2944 else if (sdkp->lbpws)
2945 sd_config_discard(sdkp, SD_LBP_WS16);
2946 else if (sdkp->lbpws10)
2947 sd_config_discard(sdkp, SD_LBP_WS10);
2949 sd_config_discard(sdkp, SD_LBP_DISABLE);
2958 * sd_read_block_characteristics - Query block dev. characteristics
2959 * @sdkp: disk to query
2961 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2963 struct request_queue *q = sdkp->disk->queue;
2964 unsigned char *buffer;
2966 const int vpd_len = 64;
2968 buffer = kmalloc(vpd_len, GFP_KERNEL);
2971 /* Block Device Characteristics VPD */
2972 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2975 rot = get_unaligned_be16(&buffer[4]);
2978 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2979 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2982 if (sdkp->device->type == TYPE_ZBC) {
2984 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
2986 sdkp->zoned = (buffer[8] >> 4) & 3;
2987 if (sdkp->zoned == 1) {
2989 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
2991 /* Regular disk or drive managed disk */
2992 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
2996 if (!sdkp->first_scan)
2999 if (blk_queue_is_zoned(q)) {
3000 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3001 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3003 if (sdkp->zoned == 1)
3004 sd_printk(KERN_NOTICE, sdkp,
3005 "Host-aware SMR disk used as regular disk\n");
3006 else if (sdkp->zoned == 2)
3007 sd_printk(KERN_NOTICE, sdkp,
3008 "Drive-managed SMR disk\n");
3016 * sd_read_block_provisioning - Query provisioning VPD page
3017 * @sdkp: disk to query
3019 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3021 unsigned char *buffer;
3022 const int vpd_len = 8;
3024 if (sdkp->lbpme == 0)
3027 buffer = kmalloc(vpd_len, GFP_KERNEL);
3029 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3033 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3034 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3035 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3041 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3043 struct scsi_device *sdev = sdkp->device;
3045 if (sdev->host->no_write_same) {
3046 sdev->no_write_same = 1;
3051 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3052 /* too large values might cause issues with arcmsr */
3053 int vpd_buf_len = 64;
3055 sdev->no_report_opcodes = 1;
3057 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3058 * CODES is unsupported and the device has an ATA
3059 * Information VPD page (SAT).
3061 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3062 sdev->no_write_same = 1;
3065 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3068 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3072 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3074 struct scsi_device *sdev = sdkp->device;
3076 if (!sdev->security_supported)
3079 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3080 SECURITY_PROTOCOL_IN) == 1 &&
3081 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3082 SECURITY_PROTOCOL_OUT) == 1)
3087 * Determine the device's preferred I/O size for reads and writes
3088 * unless the reported value is unreasonably small, large, not a
3089 * multiple of the physical block size, or simply garbage.
3091 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3092 unsigned int dev_max)
3094 struct scsi_device *sdp = sdkp->device;
3095 unsigned int opt_xfer_bytes =
3096 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3098 if (sdkp->opt_xfer_blocks == 0)
3101 if (sdkp->opt_xfer_blocks > dev_max) {
3102 sd_first_printk(KERN_WARNING, sdkp,
3103 "Optimal transfer size %u logical blocks " \
3104 "> dev_max (%u logical blocks)\n",
3105 sdkp->opt_xfer_blocks, dev_max);
3109 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3110 sd_first_printk(KERN_WARNING, sdkp,
3111 "Optimal transfer size %u logical blocks " \
3112 "> sd driver limit (%u logical blocks)\n",
3113 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3117 if (opt_xfer_bytes < PAGE_SIZE) {
3118 sd_first_printk(KERN_WARNING, sdkp,
3119 "Optimal transfer size %u bytes < " \
3120 "PAGE_SIZE (%u bytes)\n",
3121 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3125 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3126 sd_first_printk(KERN_WARNING, sdkp,
3127 "Optimal transfer size %u bytes not a " \
3128 "multiple of physical block size (%u bytes)\n",
3129 opt_xfer_bytes, sdkp->physical_block_size);
3133 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3139 * sd_revalidate_disk - called the first time a new disk is seen,
3140 * performs disk spin up, read_capacity, etc.
3141 * @disk: struct gendisk we care about
3143 static int sd_revalidate_disk(struct gendisk *disk)
3145 struct scsi_disk *sdkp = scsi_disk(disk);
3146 struct scsi_device *sdp = sdkp->device;
3147 struct request_queue *q = sdkp->disk->queue;
3148 sector_t old_capacity = sdkp->capacity;
3149 unsigned char *buffer;
3150 unsigned int dev_max, rw_max;
3152 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3153 "sd_revalidate_disk\n"));
3156 * If the device is offline, don't try and read capacity or any
3157 * of the other niceties.
3159 if (!scsi_device_online(sdp))
3162 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3164 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3165 "allocation failure.\n");
3169 sd_spinup_disk(sdkp);
3172 * Without media there is no reason to ask; moreover, some devices
3173 * react badly if we do.
3175 if (sdkp->media_present) {
3176 sd_read_capacity(sdkp, buffer);
3179 * set the default to rotational. All non-rotational devices
3180 * support the block characteristics VPD page, which will
3181 * cause this to be updated correctly and any device which
3182 * doesn't support it should be treated as rotational.
3184 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3185 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3187 if (scsi_device_supports_vpd(sdp)) {
3188 sd_read_block_provisioning(sdkp);
3189 sd_read_block_limits(sdkp);
3190 sd_read_block_characteristics(sdkp);
3191 sd_zbc_read_zones(sdkp, buffer);
3194 sd_print_capacity(sdkp, old_capacity);
3196 sd_read_write_protect_flag(sdkp, buffer);
3197 sd_read_cache_type(sdkp, buffer);
3198 sd_read_app_tag_own(sdkp, buffer);
3199 sd_read_write_same(sdkp, buffer);
3200 sd_read_security(sdkp, buffer);
3204 * We now have all cache related info, determine how we deal
3205 * with flush requests.
3207 sd_set_flush_flag(sdkp);
3209 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3210 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3212 /* Some devices report a maximum block count for READ/WRITE requests. */
3213 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3214 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3216 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3217 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3218 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3220 q->limits.io_opt = 0;
3221 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3222 (sector_t)BLK_DEF_MAX_SECTORS);
3225 /* Do not exceed controller limit */
3226 rw_max = min(rw_max, queue_max_hw_sectors(q));
3229 * Only update max_sectors if previously unset or if the current value
3230 * exceeds the capabilities of the hardware.
3232 if (sdkp->first_scan ||
3233 q->limits.max_sectors > q->limits.max_dev_sectors ||
3234 q->limits.max_sectors > q->limits.max_hw_sectors)
3235 q->limits.max_sectors = rw_max;
3237 sdkp->first_scan = 0;
3239 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3240 sd_config_write_same(sdkp);
3244 * For a zoned drive, revalidating the zones can be done only once
3245 * the gendisk capacity is set. So if this fails, set back the gendisk
3248 if (sd_zbc_revalidate_zones(sdkp))
3249 set_capacity_and_notify(disk, 0);
3256 * sd_unlock_native_capacity - unlock native capacity
3257 * @disk: struct gendisk to set capacity for
3259 * Block layer calls this function if it detects that partitions
3260 * on @disk reach beyond the end of the device. If the SCSI host
3261 * implements ->unlock_native_capacity() method, it's invoked to
3262 * give it a chance to adjust the device capacity.
3265 * Defined by block layer. Might sleep.
3267 static void sd_unlock_native_capacity(struct gendisk *disk)
3269 struct scsi_device *sdev = scsi_disk(disk)->device;
3271 if (sdev->host->hostt->unlock_native_capacity)
3272 sdev->host->hostt->unlock_native_capacity(sdev);
3276 * sd_format_disk_name - format disk name
3277 * @prefix: name prefix - ie. "sd" for SCSI disks
3278 * @index: index of the disk to format name for
3279 * @buf: output buffer
3280 * @buflen: length of the output buffer
3282 * SCSI disk names starts at sda. The 26th device is sdz and the
3283 * 27th is sdaa. The last one for two lettered suffix is sdzz
3284 * which is followed by sdaaa.
3286 * This is basically 26 base counting with one extra 'nil' entry
3287 * at the beginning from the second digit on and can be
3288 * determined using similar method as 26 base conversion with the
3289 * index shifted -1 after each digit is computed.
3295 * 0 on success, -errno on failure.
3297 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3299 const int base = 'z' - 'a' + 1;
3300 char *begin = buf + strlen(prefix);
3301 char *end = buf + buflen;
3311 *--p = 'a' + (index % unit);
3312 index = (index / unit) - 1;
3313 } while (index >= 0);
3315 memmove(begin, p, end - p);
3316 memcpy(buf, prefix, strlen(prefix));
3322 * sd_probe - called during driver initialization and whenever a
3323 * new scsi device is attached to the system. It is called once
3324 * for each scsi device (not just disks) present.
3325 * @dev: pointer to device object
3327 * Returns 0 if successful (or not interested in this scsi device
3328 * (e.g. scanner)); 1 when there is an error.
3330 * Note: this function is invoked from the scsi mid-level.
3331 * This function sets up the mapping between a given
3332 * <host,channel,id,lun> (found in sdp) and new device name
3333 * (e.g. /dev/sda). More precisely it is the block device major
3334 * and minor number that is chosen here.
3336 * Assume sd_probe is not re-entrant (for time being)
3337 * Also think about sd_probe() and sd_remove() running coincidentally.
3339 static int sd_probe(struct device *dev)
3341 struct scsi_device *sdp = to_scsi_device(dev);
3342 struct scsi_disk *sdkp;
3347 scsi_autopm_get_device(sdp);
3349 if (sdp->type != TYPE_DISK &&
3350 sdp->type != TYPE_ZBC &&
3351 sdp->type != TYPE_MOD &&
3352 sdp->type != TYPE_RBC)
3355 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3356 sdev_printk(KERN_WARNING, sdp,
3357 "Unsupported ZBC host-managed device.\n");
3361 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3365 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3369 gd = alloc_disk(SD_MINORS);
3373 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3375 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3379 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3381 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3382 goto out_free_index;
3386 sdkp->driver = &sd_template;
3388 sdkp->index = index;
3389 sdkp->max_retries = SD_MAX_RETRIES;
3390 atomic_set(&sdkp->openers, 0);
3391 atomic_set(&sdkp->device->ioerr_cnt, 0);
3393 if (!sdp->request_queue->rq_timeout) {
3394 if (sdp->type != TYPE_MOD)
3395 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3397 blk_queue_rq_timeout(sdp->request_queue,
3401 device_initialize(&sdkp->dev);
3402 sdkp->dev.parent = dev;
3403 sdkp->dev.class = &sd_disk_class;
3404 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3406 error = device_add(&sdkp->dev);
3408 goto out_free_index;
3411 dev_set_drvdata(dev, sdkp);
3413 gd->major = sd_major((index & 0xf0) >> 4);
3414 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3416 gd->fops = &sd_fops;
3417 gd->private_data = &sdkp->driver;
3418 gd->queue = sdkp->device->request_queue;
3420 /* defaults, until the device tells us otherwise */
3421 sdp->sector_size = 512;
3423 sdkp->media_present = 1;
3424 sdkp->write_prot = 0;
3425 sdkp->cache_override = 0;
3429 sdkp->first_scan = 1;
3430 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3432 sd_revalidate_disk(gd);
3434 gd->flags = GENHD_FL_EXT_DEVT;
3435 if (sdp->removable) {
3436 gd->flags |= GENHD_FL_REMOVABLE;
3437 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3438 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3441 blk_pm_runtime_init(sdp->request_queue, dev);
3442 if (sdp->rpm_autosuspend) {
3443 pm_runtime_set_autosuspend_delay(dev,
3444 sdp->host->hostt->rpm_autosuspend_delay);
3446 device_add_disk(dev, gd, NULL);
3448 sd_dif_config_host(sdkp);
3450 sd_revalidate_disk(gd);
3452 if (sdkp->security) {
3453 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3455 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3458 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3459 sdp->removable ? "removable " : "");
3460 scsi_autopm_put_device(sdp);
3465 ida_free(&sd_index_ida, index);
3469 sd_zbc_release_disk(sdkp);
3472 scsi_autopm_put_device(sdp);
3477 * sd_remove - called whenever a scsi disk (previously recognized by
3478 * sd_probe) is detached from the system. It is called (potentially
3479 * multiple times) during sd module unload.
3480 * @dev: pointer to device object
3482 * Note: this function is invoked from the scsi mid-level.
3483 * This function potentially frees up a device name (e.g. /dev/sdc)
3484 * that could be re-used by a subsequent sd_probe().
3485 * This function is not called when the built-in sd driver is "exit-ed".
3487 static int sd_remove(struct device *dev)
3489 struct scsi_disk *sdkp;
3491 sdkp = dev_get_drvdata(dev);
3492 scsi_autopm_get_device(sdkp->device);
3494 async_synchronize_full_domain(&scsi_sd_pm_domain);
3495 device_del(&sdkp->dev);
3496 del_gendisk(sdkp->disk);
3499 free_opal_dev(sdkp->opal_dev);
3501 mutex_lock(&sd_ref_mutex);
3502 dev_set_drvdata(dev, NULL);
3503 put_device(&sdkp->dev);
3504 mutex_unlock(&sd_ref_mutex);
3510 * scsi_disk_release - Called to free the scsi_disk structure
3511 * @dev: pointer to embedded class device
3513 * sd_ref_mutex must be held entering this routine. Because it is
3514 * called on last put, you should always use the scsi_disk_get()
3515 * scsi_disk_put() helpers which manipulate the semaphore directly
3516 * and never do a direct put_device.
3518 static void scsi_disk_release(struct device *dev)
3520 struct scsi_disk *sdkp = to_scsi_disk(dev);
3521 struct gendisk *disk = sdkp->disk;
3522 struct request_queue *q = disk->queue;
3524 ida_free(&sd_index_ida, sdkp->index);
3527 * Wait until all requests that are in progress have completed.
3528 * This is necessary to avoid that e.g. scsi_end_request() crashes
3529 * due to clearing the disk->private_data pointer. Wait from inside
3530 * scsi_disk_release() instead of from sd_release() to avoid that
3531 * freezing and unfreezing the request queue affects user space I/O
3532 * in case multiple processes open a /dev/sd... node concurrently.
3534 blk_mq_freeze_queue(q);
3535 blk_mq_unfreeze_queue(q);
3537 disk->private_data = NULL;
3539 put_device(&sdkp->device->sdev_gendev);
3541 sd_zbc_release_disk(sdkp);
3546 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3548 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3549 struct scsi_sense_hdr sshdr;
3550 struct scsi_device *sdp = sdkp->device;
3554 cmd[4] |= 1; /* START */
3556 if (sdp->start_stop_pwr_cond)
3557 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3559 if (!scsi_device_online(sdp))
3562 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3563 SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3565 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3566 if (res > 0 && scsi_sense_valid(&sshdr)) {
3567 sd_print_sense_hdr(sdkp, &sshdr);
3568 /* 0x3a is medium not present */
3569 if (sshdr.asc == 0x3a)
3574 /* SCSI error codes must not go to the generic layer */
3582 * Send a SYNCHRONIZE CACHE instruction down to the device through
3583 * the normal SCSI command structure. Wait for the command to
3586 static void sd_shutdown(struct device *dev)
3588 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3591 return; /* this can happen */
3593 if (pm_runtime_suspended(dev))
3596 if (sdkp->WCE && sdkp->media_present) {
3597 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3598 sd_sync_cache(sdkp, NULL);
3601 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3602 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3603 sd_start_stop_device(sdkp, 0);
3607 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3609 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3610 struct scsi_sense_hdr sshdr;
3613 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3616 if (sdkp->WCE && sdkp->media_present) {
3617 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3618 ret = sd_sync_cache(sdkp, &sshdr);
3621 /* ignore OFFLINE device */
3625 if (!scsi_sense_valid(&sshdr) ||
3626 sshdr.sense_key != ILLEGAL_REQUEST)
3630 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3631 * doesn't support sync. There's not much to do and
3632 * suspend shouldn't fail.
3638 if (sdkp->device->manage_start_stop) {
3639 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3640 /* an error is not worth aborting a system sleep */
3641 ret = sd_start_stop_device(sdkp, 0);
3642 if (ignore_stop_errors)
3649 static int sd_suspend_system(struct device *dev)
3651 return sd_suspend_common(dev, true);
3654 static int sd_suspend_runtime(struct device *dev)
3656 return sd_suspend_common(dev, false);
3659 static int sd_resume(struct device *dev)
3661 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3664 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3667 if (!sdkp->device->manage_start_stop)
3670 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3671 ret = sd_start_stop_device(sdkp, 1);
3673 opal_unlock_from_suspend(sdkp->opal_dev);
3677 static int sd_resume_runtime(struct device *dev)
3679 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3680 struct scsi_device *sdp = sdkp->device;
3682 if (sdp->ignore_media_change) {
3683 /* clear the device's sense data */
3684 static const u8 cmd[10] = { REQUEST_SENSE };
3686 if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3687 NULL, sdp->request_queue->rq_timeout, 1, 0,
3689 sd_printk(KERN_NOTICE, sdkp,
3690 "Failed to clear sense data\n");
3693 return sd_resume(dev);
3697 * init_sd - entry point for this driver (both when built in or when
3700 * Note: this function registers this driver with the scsi mid-level.
3702 static int __init init_sd(void)
3704 int majors = 0, i, err;
3706 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3708 for (i = 0; i < SD_MAJORS; i++) {
3709 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3717 err = class_register(&sd_disk_class);
3721 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3723 if (!sd_cdb_cache) {
3724 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3729 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3731 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3736 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3737 if (!sd_page_pool) {
3738 printk(KERN_ERR "sd: can't init discard page pool\n");
3743 err = scsi_register_driver(&sd_template.gendrv);
3745 goto err_out_driver;
3750 mempool_destroy(sd_page_pool);
3753 mempool_destroy(sd_cdb_pool);
3756 kmem_cache_destroy(sd_cdb_cache);
3759 class_unregister(&sd_disk_class);
3761 for (i = 0; i < SD_MAJORS; i++)
3762 unregister_blkdev(sd_major(i), "sd");
3767 * exit_sd - exit point for this driver (when it is a module).
3769 * Note: this function unregisters this driver from the scsi mid-level.
3771 static void __exit exit_sd(void)
3775 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3777 scsi_unregister_driver(&sd_template.gendrv);
3778 mempool_destroy(sd_cdb_pool);
3779 mempool_destroy(sd_page_pool);
3780 kmem_cache_destroy(sd_cdb_cache);
3782 class_unregister(&sd_disk_class);
3784 for (i = 0; i < SD_MAJORS; i++)
3785 unregister_blkdev(sd_major(i), "sd");
3788 module_init(init_sd);
3789 module_exit(exit_sd);
3791 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3793 scsi_print_sense_hdr(sdkp->device,
3794 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3797 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3799 const char *hb_string = scsi_hostbyte_string(result);
3802 sd_printk(KERN_INFO, sdkp,
3803 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3804 hb_string ? hb_string : "invalid",
3807 sd_printk(KERN_INFO, sdkp,
3808 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3809 msg, host_byte(result), "DRIVER_OK");