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/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/rw_hint.h>
51 #include <linux/major.h>
52 #include <linux/mutex.h>
53 #include <linux/string_helpers.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>
71 #include <scsi/scsi_common.h>
74 #include "scsi_priv.h"
75 #include "scsi_logging.h"
77 MODULE_AUTHOR("Eric Youngdale");
78 MODULE_DESCRIPTION("SCSI disk (sd) driver");
79 MODULE_LICENSE("GPL");
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
96 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
100 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
104 static void sd_config_discard(struct scsi_disk *, unsigned int);
105 static void sd_config_write_same(struct scsi_disk *);
106 static int sd_revalidate_disk(struct gendisk *);
107 static void sd_unlock_native_capacity(struct gendisk *disk);
108 static void sd_shutdown(struct device *);
109 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
110 static void scsi_disk_release(struct device *cdev);
112 static DEFINE_IDA(sd_index_ida);
114 static mempool_t *sd_page_pool;
115 static struct lock_class_key sd_bio_compl_lkclass;
117 static const char *sd_cache_types[] = {
118 "write through", "none", "write back",
119 "write back, no read (daft)"
122 static void sd_set_flush_flag(struct scsi_disk *sdkp)
124 bool wc = false, fua = false;
132 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
136 cache_type_store(struct device *dev, struct device_attribute *attr,
137 const char *buf, size_t count)
139 int ct, rcd, wce, sp;
140 struct scsi_disk *sdkp = to_scsi_disk(dev);
141 struct scsi_device *sdp = sdkp->device;
144 struct scsi_mode_data data;
145 struct scsi_sense_hdr sshdr;
146 static const char temp[] = "temporary ";
149 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
150 /* no cache control on RBC devices; theoretically they
151 * can do it, but there's probably so many exceptions
152 * it's not worth the risk */
155 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
156 buf += sizeof(temp) - 1;
157 sdkp->cache_override = 1;
159 sdkp->cache_override = 0;
162 ct = sysfs_match_string(sd_cache_types, buf);
166 rcd = ct & 0x01 ? 1 : 0;
167 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
169 if (sdkp->cache_override) {
172 sd_set_flush_flag(sdkp);
176 if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
177 sdkp->max_retries, &data, NULL))
179 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
180 data.block_descriptor_length);
181 buffer_data = buffer + data.header_length +
182 data.block_descriptor_length;
183 buffer_data[2] &= ~0x05;
184 buffer_data[2] |= wce << 2 | rcd;
185 sp = buffer_data[0] & 0x80 ? 1 : 0;
186 buffer_data[0] &= ~0x80;
189 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
190 * received mode parameter buffer before doing MODE SELECT.
192 data.device_specific = 0;
194 ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
195 sdkp->max_retries, &data, &sshdr);
197 if (ret > 0 && scsi_sense_valid(&sshdr))
198 sd_print_sense_hdr(sdkp, &sshdr);
201 sd_revalidate_disk(sdkp->disk);
206 manage_start_stop_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
209 struct scsi_disk *sdkp = to_scsi_disk(dev);
210 struct scsi_device *sdp = sdkp->device;
212 return sysfs_emit(buf, "%u\n",
213 sdp->manage_system_start_stop &&
214 sdp->manage_runtime_start_stop &&
215 sdp->manage_shutdown);
217 static DEVICE_ATTR_RO(manage_start_stop);
220 manage_system_start_stop_show(struct device *dev,
221 struct device_attribute *attr, char *buf)
223 struct scsi_disk *sdkp = to_scsi_disk(dev);
224 struct scsi_device *sdp = sdkp->device;
226 return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
230 manage_system_start_stop_store(struct device *dev,
231 struct device_attribute *attr,
232 const char *buf, size_t count)
234 struct scsi_disk *sdkp = to_scsi_disk(dev);
235 struct scsi_device *sdp = sdkp->device;
238 if (!capable(CAP_SYS_ADMIN))
241 if (kstrtobool(buf, &v))
244 sdp->manage_system_start_stop = v;
248 static DEVICE_ATTR_RW(manage_system_start_stop);
251 manage_runtime_start_stop_show(struct device *dev,
252 struct device_attribute *attr, char *buf)
254 struct scsi_disk *sdkp = to_scsi_disk(dev);
255 struct scsi_device *sdp = sdkp->device;
257 return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
261 manage_runtime_start_stop_store(struct device *dev,
262 struct device_attribute *attr,
263 const char *buf, size_t count)
265 struct scsi_disk *sdkp = to_scsi_disk(dev);
266 struct scsi_device *sdp = sdkp->device;
269 if (!capable(CAP_SYS_ADMIN))
272 if (kstrtobool(buf, &v))
275 sdp->manage_runtime_start_stop = v;
279 static DEVICE_ATTR_RW(manage_runtime_start_stop);
281 static ssize_t manage_shutdown_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
284 struct scsi_disk *sdkp = to_scsi_disk(dev);
285 struct scsi_device *sdp = sdkp->device;
287 return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
290 static ssize_t manage_shutdown_store(struct device *dev,
291 struct device_attribute *attr,
292 const char *buf, size_t count)
294 struct scsi_disk *sdkp = to_scsi_disk(dev);
295 struct scsi_device *sdp = sdkp->device;
298 if (!capable(CAP_SYS_ADMIN))
301 if (kstrtobool(buf, &v))
304 sdp->manage_shutdown = v;
308 static DEVICE_ATTR_RW(manage_shutdown);
311 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
313 struct scsi_disk *sdkp = to_scsi_disk(dev);
315 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
319 allow_restart_store(struct device *dev, struct device_attribute *attr,
320 const char *buf, size_t count)
323 struct scsi_disk *sdkp = to_scsi_disk(dev);
324 struct scsi_device *sdp = sdkp->device;
326 if (!capable(CAP_SYS_ADMIN))
329 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
332 if (kstrtobool(buf, &v))
335 sdp->allow_restart = v;
339 static DEVICE_ATTR_RW(allow_restart);
342 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
344 struct scsi_disk *sdkp = to_scsi_disk(dev);
345 int ct = sdkp->RCD + 2*sdkp->WCE;
347 return sprintf(buf, "%s\n", sd_cache_types[ct]);
349 static DEVICE_ATTR_RW(cache_type);
352 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
354 struct scsi_disk *sdkp = to_scsi_disk(dev);
356 return sprintf(buf, "%u\n", sdkp->DPOFUA);
358 static DEVICE_ATTR_RO(FUA);
361 protection_type_show(struct device *dev, struct device_attribute *attr,
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
366 return sprintf(buf, "%u\n", sdkp->protection_type);
370 protection_type_store(struct device *dev, struct device_attribute *attr,
371 const char *buf, size_t count)
373 struct scsi_disk *sdkp = to_scsi_disk(dev);
377 if (!capable(CAP_SYS_ADMIN))
380 err = kstrtouint(buf, 10, &val);
385 if (val <= T10_PI_TYPE3_PROTECTION)
386 sdkp->protection_type = val;
390 static DEVICE_ATTR_RW(protection_type);
393 protection_mode_show(struct device *dev, struct device_attribute *attr,
396 struct scsi_disk *sdkp = to_scsi_disk(dev);
397 struct scsi_device *sdp = sdkp->device;
398 unsigned int dif, dix;
400 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
401 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
403 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
409 return sprintf(buf, "none\n");
411 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
413 static DEVICE_ATTR_RO(protection_mode);
416 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
418 struct scsi_disk *sdkp = to_scsi_disk(dev);
420 return sprintf(buf, "%u\n", sdkp->ATO);
422 static DEVICE_ATTR_RO(app_tag_own);
425 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
428 struct scsi_disk *sdkp = to_scsi_disk(dev);
430 return sprintf(buf, "%u\n", sdkp->lbpme);
432 static DEVICE_ATTR_RO(thin_provisioning);
434 /* sysfs_match_string() requires dense arrays */
435 static const char *lbp_mode[] = {
436 [SD_LBP_FULL] = "full",
437 [SD_LBP_UNMAP] = "unmap",
438 [SD_LBP_WS16] = "writesame_16",
439 [SD_LBP_WS10] = "writesame_10",
440 [SD_LBP_ZERO] = "writesame_zero",
441 [SD_LBP_DISABLE] = "disabled",
445 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
448 struct scsi_disk *sdkp = to_scsi_disk(dev);
450 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
454 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
455 const char *buf, size_t count)
457 struct scsi_disk *sdkp = to_scsi_disk(dev);
458 struct scsi_device *sdp = sdkp->device;
461 if (!capable(CAP_SYS_ADMIN))
464 if (sd_is_zoned(sdkp)) {
465 sd_config_discard(sdkp, SD_LBP_DISABLE);
469 if (sdp->type != TYPE_DISK)
472 mode = sysfs_match_string(lbp_mode, buf);
476 sd_config_discard(sdkp, mode);
480 static DEVICE_ATTR_RW(provisioning_mode);
482 /* sysfs_match_string() requires dense arrays */
483 static const char *zeroing_mode[] = {
484 [SD_ZERO_WRITE] = "write",
485 [SD_ZERO_WS] = "writesame",
486 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
487 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
491 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
494 struct scsi_disk *sdkp = to_scsi_disk(dev);
496 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
500 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
501 const char *buf, size_t count)
503 struct scsi_disk *sdkp = to_scsi_disk(dev);
506 if (!capable(CAP_SYS_ADMIN))
509 mode = sysfs_match_string(zeroing_mode, buf);
513 sdkp->zeroing_mode = mode;
517 static DEVICE_ATTR_RW(zeroing_mode);
520 max_medium_access_timeouts_show(struct device *dev,
521 struct device_attribute *attr, char *buf)
523 struct scsi_disk *sdkp = to_scsi_disk(dev);
525 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
529 max_medium_access_timeouts_store(struct device *dev,
530 struct device_attribute *attr, const char *buf,
533 struct scsi_disk *sdkp = to_scsi_disk(dev);
536 if (!capable(CAP_SYS_ADMIN))
539 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
541 return err ? err : count;
543 static DEVICE_ATTR_RW(max_medium_access_timeouts);
546 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
549 struct scsi_disk *sdkp = to_scsi_disk(dev);
551 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
555 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
556 const char *buf, size_t count)
558 struct scsi_disk *sdkp = to_scsi_disk(dev);
559 struct scsi_device *sdp = sdkp->device;
563 if (!capable(CAP_SYS_ADMIN))
566 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
569 err = kstrtoul(buf, 10, &max);
575 sdp->no_write_same = 1;
576 else if (max <= SD_MAX_WS16_BLOCKS) {
577 sdp->no_write_same = 0;
578 sdkp->max_ws_blocks = max;
581 sd_config_write_same(sdkp);
585 static DEVICE_ATTR_RW(max_write_same_blocks);
588 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
590 struct scsi_disk *sdkp = to_scsi_disk(dev);
592 if (sdkp->device->type == TYPE_ZBC)
593 return sprintf(buf, "host-managed\n");
594 if (sdkp->zoned == 1)
595 return sprintf(buf, "host-aware\n");
596 if (sdkp->zoned == 2)
597 return sprintf(buf, "drive-managed\n");
598 return sprintf(buf, "none\n");
600 static DEVICE_ATTR_RO(zoned_cap);
603 max_retries_store(struct device *dev, struct device_attribute *attr,
604 const char *buf, size_t count)
606 struct scsi_disk *sdkp = to_scsi_disk(dev);
607 struct scsi_device *sdev = sdkp->device;
610 err = kstrtoint(buf, 10, &retries);
614 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
615 sdkp->max_retries = retries;
619 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
625 max_retries_show(struct device *dev, struct device_attribute *attr,
628 struct scsi_disk *sdkp = to_scsi_disk(dev);
630 return sprintf(buf, "%d\n", sdkp->max_retries);
633 static DEVICE_ATTR_RW(max_retries);
635 static struct attribute *sd_disk_attrs[] = {
636 &dev_attr_cache_type.attr,
638 &dev_attr_allow_restart.attr,
639 &dev_attr_manage_start_stop.attr,
640 &dev_attr_manage_system_start_stop.attr,
641 &dev_attr_manage_runtime_start_stop.attr,
642 &dev_attr_manage_shutdown.attr,
643 &dev_attr_protection_type.attr,
644 &dev_attr_protection_mode.attr,
645 &dev_attr_app_tag_own.attr,
646 &dev_attr_thin_provisioning.attr,
647 &dev_attr_provisioning_mode.attr,
648 &dev_attr_zeroing_mode.attr,
649 &dev_attr_max_write_same_blocks.attr,
650 &dev_attr_max_medium_access_timeouts.attr,
651 &dev_attr_zoned_cap.attr,
652 &dev_attr_max_retries.attr,
655 ATTRIBUTE_GROUPS(sd_disk);
657 static struct class sd_disk_class = {
659 .dev_release = scsi_disk_release,
660 .dev_groups = sd_disk_groups,
664 * Don't request a new module, as that could deadlock in multipath
667 static void sd_default_probe(dev_t devt)
672 * Device no to disk mapping:
674 * major disc2 disc p1
675 * |............|.............|....|....| <- dev_t
678 * Inside a major, we have 16k disks, however mapped non-
679 * contiguously. The first 16 disks are for major0, the next
680 * ones with major1, ... Disk 256 is for major0 again, disk 272
682 * As we stay compatible with our numbering scheme, we can reuse
683 * the well-know SCSI majors 8, 65--71, 136--143.
685 static int sd_major(int major_idx)
689 return SCSI_DISK0_MAJOR;
691 return SCSI_DISK1_MAJOR + major_idx - 1;
693 return SCSI_DISK8_MAJOR + major_idx - 8;
696 return 0; /* shut up gcc */
700 #ifdef CONFIG_BLK_SED_OPAL
701 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
702 size_t len, bool send)
704 struct scsi_disk *sdkp = data;
705 struct scsi_device *sdev = sdkp->device;
707 const struct scsi_exec_args exec_args = {
708 .req_flags = BLK_MQ_REQ_PM,
712 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
714 put_unaligned_be16(spsp, &cdb[2]);
715 put_unaligned_be32(len, &cdb[6]);
717 ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
718 buffer, len, SD_TIMEOUT, sdkp->max_retries,
720 return ret <= 0 ? ret : -EIO;
722 #endif /* CONFIG_BLK_SED_OPAL */
725 * Look up the DIX operation based on whether the command is read or
726 * write and whether dix and dif are enabled.
728 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
730 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
731 static const unsigned int ops[] = { /* wrt dix dif */
732 SCSI_PROT_NORMAL, /* 0 0 0 */
733 SCSI_PROT_READ_STRIP, /* 0 0 1 */
734 SCSI_PROT_READ_INSERT, /* 0 1 0 */
735 SCSI_PROT_READ_PASS, /* 0 1 1 */
736 SCSI_PROT_NORMAL, /* 1 0 0 */
737 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
738 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
739 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
742 return ops[write << 2 | dix << 1 | dif];
746 * Returns a mask of the protection flags that are valid for a given DIX
749 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
751 static const unsigned int flag_mask[] = {
752 [SCSI_PROT_NORMAL] = 0,
754 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
755 SCSI_PROT_GUARD_CHECK |
756 SCSI_PROT_REF_CHECK |
757 SCSI_PROT_REF_INCREMENT,
759 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
760 SCSI_PROT_IP_CHECKSUM,
762 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
763 SCSI_PROT_GUARD_CHECK |
764 SCSI_PROT_REF_CHECK |
765 SCSI_PROT_REF_INCREMENT |
766 SCSI_PROT_IP_CHECKSUM,
768 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
769 SCSI_PROT_REF_INCREMENT,
771 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
772 SCSI_PROT_REF_CHECK |
773 SCSI_PROT_REF_INCREMENT |
774 SCSI_PROT_IP_CHECKSUM,
776 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
777 SCSI_PROT_GUARD_CHECK |
778 SCSI_PROT_REF_CHECK |
779 SCSI_PROT_REF_INCREMENT |
780 SCSI_PROT_IP_CHECKSUM,
783 return flag_mask[prot_op];
786 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
787 unsigned int dix, unsigned int dif)
789 struct request *rq = scsi_cmd_to_rq(scmd);
790 struct bio *bio = rq->bio;
791 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
792 unsigned int protect = 0;
794 if (dix) { /* DIX Type 0, 1, 2, 3 */
795 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
796 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
798 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
799 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
802 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
803 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
805 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
806 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
809 if (dif) { /* DIX/DIF Type 1, 2, 3 */
810 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
812 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
813 protect = 3 << 5; /* Disable target PI checking */
815 protect = 1 << 5; /* Enable target PI checking */
818 scsi_set_prot_op(scmd, prot_op);
819 scsi_set_prot_type(scmd, dif);
820 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
825 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
827 struct request_queue *q = sdkp->disk->queue;
828 unsigned int logical_block_size = sdkp->device->sector_size;
829 unsigned int max_blocks = 0;
831 q->limits.discard_alignment =
832 sdkp->unmap_alignment * logical_block_size;
833 q->limits.discard_granularity =
834 max(sdkp->physical_block_size,
835 sdkp->unmap_granularity * logical_block_size);
836 sdkp->provisioning_mode = mode;
842 blk_queue_max_discard_sectors(q, 0);
846 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
847 (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_WS16_BLOCKS);
860 if (sdkp->device->unmap_limit_for_ws)
861 max_blocks = sdkp->max_unmap_blocks;
863 max_blocks = sdkp->max_ws_blocks;
865 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
869 max_blocks = min_not_zero(sdkp->max_ws_blocks,
870 (u32)SD_MAX_WS10_BLOCKS);
874 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
877 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
881 page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
884 clear_highpage(page);
885 bvec_set_page(&rq->special_vec, page, data_len, 0);
886 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
887 return bvec_virt(&rq->special_vec);
890 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
892 struct scsi_device *sdp = cmd->device;
893 struct request *rq = scsi_cmd_to_rq(cmd);
894 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
895 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
896 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
897 unsigned int data_len = 24;
900 buf = sd_set_special_bvec(rq, data_len);
902 return BLK_STS_RESOURCE;
905 cmd->cmnd[0] = UNMAP;
908 put_unaligned_be16(6 + 16, &buf[0]);
909 put_unaligned_be16(16, &buf[2]);
910 put_unaligned_be64(lba, &buf[8]);
911 put_unaligned_be32(nr_blocks, &buf[16]);
913 cmd->allowed = sdkp->max_retries;
914 cmd->transfersize = data_len;
915 rq->timeout = SD_TIMEOUT;
917 return scsi_alloc_sgtables(cmd);
920 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
923 struct scsi_device *sdp = cmd->device;
924 struct request *rq = scsi_cmd_to_rq(cmd);
925 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
926 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
927 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
928 u32 data_len = sdp->sector_size;
930 if (!sd_set_special_bvec(rq, data_len))
931 return BLK_STS_RESOURCE;
934 cmd->cmnd[0] = WRITE_SAME_16;
936 cmd->cmnd[1] = 0x8; /* UNMAP */
937 put_unaligned_be64(lba, &cmd->cmnd[2]);
938 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
940 cmd->allowed = sdkp->max_retries;
941 cmd->transfersize = data_len;
942 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
944 return scsi_alloc_sgtables(cmd);
947 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
950 struct scsi_device *sdp = cmd->device;
951 struct request *rq = scsi_cmd_to_rq(cmd);
952 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
953 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
954 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
955 u32 data_len = sdp->sector_size;
957 if (!sd_set_special_bvec(rq, data_len))
958 return BLK_STS_RESOURCE;
961 cmd->cmnd[0] = WRITE_SAME;
963 cmd->cmnd[1] = 0x8; /* UNMAP */
964 put_unaligned_be32(lba, &cmd->cmnd[2]);
965 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
967 cmd->allowed = sdkp->max_retries;
968 cmd->transfersize = data_len;
969 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
971 return scsi_alloc_sgtables(cmd);
974 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
976 struct request *rq = scsi_cmd_to_rq(cmd);
977 struct scsi_device *sdp = cmd->device;
978 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
979 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
980 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
982 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
983 switch (sdkp->zeroing_mode) {
984 case SD_ZERO_WS16_UNMAP:
985 return sd_setup_write_same16_cmnd(cmd, true);
986 case SD_ZERO_WS10_UNMAP:
987 return sd_setup_write_same10_cmnd(cmd, true);
991 if (sdp->no_write_same) {
992 rq->rq_flags |= RQF_QUIET;
993 return BLK_STS_TARGET;
996 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
997 return sd_setup_write_same16_cmnd(cmd, false);
999 return sd_setup_write_same10_cmnd(cmd, false);
1002 static void sd_config_write_same(struct scsi_disk *sdkp)
1004 struct request_queue *q = sdkp->disk->queue;
1005 unsigned int logical_block_size = sdkp->device->sector_size;
1007 if (sdkp->device->no_write_same) {
1008 sdkp->max_ws_blocks = 0;
1012 /* Some devices can not handle block counts above 0xffff despite
1013 * supporting WRITE SAME(16). Consequently we default to 64k
1014 * blocks per I/O unless the device explicitly advertises a
1017 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1018 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1019 (u32)SD_MAX_WS16_BLOCKS);
1020 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1021 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1022 (u32)SD_MAX_WS10_BLOCKS);
1024 sdkp->device->no_write_same = 1;
1025 sdkp->max_ws_blocks = 0;
1028 if (sdkp->lbprz && sdkp->lbpws)
1029 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1030 else if (sdkp->lbprz && sdkp->lbpws10)
1031 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1032 else if (sdkp->max_ws_blocks)
1033 sdkp->zeroing_mode = SD_ZERO_WS;
1035 sdkp->zeroing_mode = SD_ZERO_WRITE;
1037 if (sdkp->max_ws_blocks &&
1038 sdkp->physical_block_size > logical_block_size) {
1040 * Reporting a maximum number of blocks that is not aligned
1041 * on the device physical size would cause a large write same
1042 * request to be split into physically unaligned chunks by
1043 * __blkdev_issue_write_zeroes() even if the caller of this
1044 * functions took care to align the large request. So make sure
1045 * the maximum reported is aligned to the device physical block
1046 * size. This is only an optional optimization for regular
1047 * disks, but this is mandatory to avoid failure of large write
1048 * same requests directed at sequential write required zones of
1049 * host-managed ZBC disks.
1051 sdkp->max_ws_blocks =
1052 round_down(sdkp->max_ws_blocks,
1053 bytes_to_logical(sdkp->device,
1054 sdkp->physical_block_size));
1058 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1059 (logical_block_size >> 9));
1062 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1064 struct request *rq = scsi_cmd_to_rq(cmd);
1065 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1067 /* flush requests don't perform I/O, zero the S/G table */
1068 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070 if (cmd->device->use_16_for_sync) {
1071 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1074 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1077 cmd->transfersize = 0;
1078 cmd->allowed = sdkp->max_retries;
1080 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1085 * sd_group_number() - Compute the GROUP NUMBER field
1086 * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1089 * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1090 * 0: no relative lifetime.
1091 * 1: shortest relative lifetime.
1092 * 2: second shortest relative lifetime.
1093 * 3 - 0x3d: intermediate relative lifetimes.
1094 * 0x3e: second longest relative lifetime.
1095 * 0x3f: longest relative lifetime.
1097 static u8 sd_group_number(struct scsi_cmnd *cmd)
1099 const struct request *rq = scsi_cmd_to_rq(cmd);
1100 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1105 return min3((u32)rq->write_hint, (u32)sdkp->permanent_stream_count,
1109 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1110 sector_t lba, unsigned int nr_blocks,
1111 unsigned char flags, unsigned int dld)
1113 cmd->cmd_len = SD_EXT_CDB_SIZE;
1114 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1115 cmd->cmnd[6] = sd_group_number(cmd);
1116 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1117 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1118 cmd->cmnd[10] = flags;
1119 cmd->cmnd[11] = dld & 0x07;
1120 put_unaligned_be64(lba, &cmd->cmnd[12]);
1121 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1122 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1127 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1128 sector_t lba, unsigned int nr_blocks,
1129 unsigned char flags, unsigned int dld)
1132 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1133 cmd->cmnd[1] = flags | ((dld >> 2) & 0x01);
1134 cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1136 put_unaligned_be64(lba, &cmd->cmnd[2]);
1137 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1142 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1143 sector_t lba, unsigned int nr_blocks,
1144 unsigned char flags)
1147 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1148 cmd->cmnd[1] = flags;
1149 cmd->cmnd[6] = sd_group_number(cmd);
1151 put_unaligned_be32(lba, &cmd->cmnd[2]);
1152 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1157 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1158 sector_t lba, unsigned int nr_blocks,
1159 unsigned char flags)
1161 /* Avoid that 0 blocks gets translated into 256 blocks. */
1162 if (WARN_ON_ONCE(nr_blocks == 0))
1163 return BLK_STS_IOERR;
1165 if (unlikely(flags & 0x8)) {
1167 * This happens only if this drive failed 10byte rw
1168 * command with ILLEGAL_REQUEST during operation and
1169 * thus turned off use_10_for_rw.
1171 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1172 return BLK_STS_IOERR;
1176 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1177 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1178 cmd->cmnd[2] = (lba >> 8) & 0xff;
1179 cmd->cmnd[3] = lba & 0xff;
1180 cmd->cmnd[4] = nr_blocks;
1187 * Check if a command has a duration limit set. If it does, and the target
1188 * device supports CDL and the feature is enabled, return the limit
1189 * descriptor index to use. Return 0 (no limit) otherwise.
1191 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1193 struct scsi_device *sdp = sdkp->device;
1196 if (!sdp->cdl_supported || !sdp->cdl_enable)
1200 * Use "no limit" if the request ioprio does not specify a duration
1203 hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1204 if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1205 hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1208 return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1211 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1213 struct request *rq = scsi_cmd_to_rq(cmd);
1214 struct scsi_device *sdp = cmd->device;
1215 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1216 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1218 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1219 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1220 bool write = rq_data_dir(rq) == WRITE;
1221 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->q->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 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1264 dix = scsi_prot_sg_count(cmd);
1265 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1266 dld = sd_cdl_dld(sdkp, cmd);
1269 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1273 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1274 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1275 protect | fua, dld);
1276 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1277 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1278 protect | fua, dld);
1279 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1280 sdp->use_10_for_rw || protect || rq->write_hint) {
1281 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1284 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1288 if (unlikely(ret != BLK_STS_OK))
1292 * We shouldn't disconnect in the middle of a sector, so with a dumb
1293 * host adapter, it's safe to assume that we can at least transfer
1294 * this many bytes between each connect / disconnect.
1296 cmd->transfersize = sdp->sector_size;
1297 cmd->underflow = nr_blocks << 9;
1298 cmd->allowed = sdkp->max_retries;
1299 cmd->sdb.length = nr_blocks * sdp->sector_size;
1302 scmd_printk(KERN_INFO, cmd,
1303 "%s: block=%llu, count=%d\n", __func__,
1304 (unsigned long long)blk_rq_pos(rq),
1305 blk_rq_sectors(rq)));
1307 scmd_printk(KERN_INFO, cmd,
1308 "%s %d/%u 512 byte blocks.\n",
1309 write ? "writing" : "reading", nr_blocks,
1310 blk_rq_sectors(rq)));
1313 * This indicates that the command is ready from our end to be queued.
1317 scsi_free_sgtables(cmd);
1321 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1323 struct request *rq = scsi_cmd_to_rq(cmd);
1325 switch (req_op(rq)) {
1326 case REQ_OP_DISCARD:
1327 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1329 return sd_setup_unmap_cmnd(cmd);
1331 return sd_setup_write_same16_cmnd(cmd, true);
1333 return sd_setup_write_same10_cmnd(cmd, true);
1335 return sd_setup_write_same10_cmnd(cmd, false);
1337 return BLK_STS_TARGET;
1339 case REQ_OP_WRITE_ZEROES:
1340 return sd_setup_write_zeroes_cmnd(cmd);
1342 return sd_setup_flush_cmnd(cmd);
1345 return sd_setup_read_write_cmnd(cmd);
1346 case REQ_OP_ZONE_RESET:
1347 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1349 case REQ_OP_ZONE_RESET_ALL:
1350 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1352 case REQ_OP_ZONE_OPEN:
1353 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1354 case REQ_OP_ZONE_CLOSE:
1355 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1356 case REQ_OP_ZONE_FINISH:
1357 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1360 return BLK_STS_NOTSUPP;
1364 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1366 struct request *rq = scsi_cmd_to_rq(SCpnt);
1368 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1369 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1372 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1374 if (sdkp->device->removable || sdkp->write_prot) {
1375 if (disk_check_media_change(disk))
1380 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1381 * nothing to do with partitions, BLKRRPART is used to force a full
1382 * revalidate after things like a format for historical reasons.
1384 return test_bit(GD_NEED_PART_SCAN, &disk->state);
1388 * sd_open - open a scsi disk device
1389 * @disk: disk to open
1392 * Returns 0 if successful. Returns a negated errno value in case
1395 * Note: This can be called from a user context (e.g. fsck(1) )
1396 * or from within the kernel (e.g. as a result of a mount(1) ).
1397 * In the latter case @inode and @filp carry an abridged amount
1398 * of information as noted above.
1400 * Locking: called with disk->open_mutex held.
1402 static int sd_open(struct gendisk *disk, blk_mode_t mode)
1404 struct scsi_disk *sdkp = scsi_disk(disk);
1405 struct scsi_device *sdev = sdkp->device;
1408 if (scsi_device_get(sdev))
1411 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1414 * If the device is in error recovery, wait until it is done.
1415 * If the device is offline, then disallow any access to it.
1418 if (!scsi_block_when_processing_errors(sdev))
1421 if (sd_need_revalidate(disk, sdkp))
1422 sd_revalidate_disk(disk);
1425 * If the drive is empty, just let the open fail.
1427 retval = -ENOMEDIUM;
1428 if (sdev->removable && !sdkp->media_present &&
1429 !(mode & BLK_OPEN_NDELAY))
1433 * If the device has the write protect tab set, have the open fail
1434 * if the user expects to be able to write to the thing.
1437 if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1441 * It is possible that the disk changing stuff resulted in
1442 * the device being taken offline. If this is the case,
1443 * report this to the user, and don't pretend that the
1444 * open actually succeeded.
1447 if (!scsi_device_online(sdev))
1450 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1451 if (scsi_block_when_processing_errors(sdev))
1452 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1458 scsi_device_put(sdev);
1463 * sd_release - invoked when the (last) close(2) is called on this
1465 * @disk: disk to release
1469 * Note: may block (uninterruptible) if error recovery is underway
1472 * Locking: called with disk->open_mutex held.
1474 static void sd_release(struct gendisk *disk)
1476 struct scsi_disk *sdkp = scsi_disk(disk);
1477 struct scsi_device *sdev = sdkp->device;
1479 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1481 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1482 if (scsi_block_when_processing_errors(sdev))
1483 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1486 scsi_device_put(sdev);
1489 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1491 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1492 struct scsi_device *sdp = sdkp->device;
1493 struct Scsi_Host *host = sdp->host;
1494 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1497 /* default to most commonly used values */
1498 diskinfo[0] = 0x40; /* 1 << 6 */
1499 diskinfo[1] = 0x20; /* 1 << 5 */
1500 diskinfo[2] = capacity >> 11;
1502 /* override with calculated, extended default, or driver values */
1503 if (host->hostt->bios_param)
1504 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1506 scsicam_bios_param(bdev, capacity, diskinfo);
1508 geo->heads = diskinfo[0];
1509 geo->sectors = diskinfo[1];
1510 geo->cylinders = diskinfo[2];
1515 * sd_ioctl - process an ioctl
1516 * @bdev: target block device
1518 * @cmd: ioctl command number
1519 * @arg: this is third argument given to ioctl(2) system call.
1520 * Often contains a pointer.
1522 * Returns 0 if successful (some ioctls return positive numbers on
1523 * success as well). Returns a negated errno value in case of error.
1525 * Note: most ioctls are forward onto the block subsystem or further
1526 * down in the scsi subsystem.
1528 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1529 unsigned int cmd, unsigned long arg)
1531 struct gendisk *disk = bdev->bd_disk;
1532 struct scsi_disk *sdkp = scsi_disk(disk);
1533 struct scsi_device *sdp = sdkp->device;
1534 void __user *p = (void __user *)arg;
1537 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1538 "cmd=0x%x\n", disk->disk_name, cmd));
1540 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1541 return -ENOIOCTLCMD;
1544 * If we are in the middle of error recovery, don't let anyone
1545 * else try and use this device. Also, if error recovery fails, it
1546 * may try and take the device offline, in which case all further
1547 * access to the device is prohibited.
1549 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1550 (mode & BLK_OPEN_NDELAY));
1554 if (is_sed_ioctl(cmd))
1555 return sed_ioctl(sdkp->opal_dev, cmd, p);
1556 return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1559 static void set_media_not_present(struct scsi_disk *sdkp)
1561 if (sdkp->media_present)
1562 sdkp->device->changed = 1;
1564 if (sdkp->device->removable) {
1565 sdkp->media_present = 0;
1570 static int media_not_present(struct scsi_disk *sdkp,
1571 struct scsi_sense_hdr *sshdr)
1573 if (!scsi_sense_valid(sshdr))
1576 /* not invoked for commands that could return deferred errors */
1577 switch (sshdr->sense_key) {
1578 case UNIT_ATTENTION:
1580 /* medium not present */
1581 if (sshdr->asc == 0x3A) {
1582 set_media_not_present(sdkp);
1590 * sd_check_events - check media events
1591 * @disk: kernel device descriptor
1592 * @clearing: disk events currently being cleared
1594 * Returns mask of DISK_EVENT_*.
1596 * Note: this function is invoked from the block subsystem.
1598 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1600 struct scsi_disk *sdkp = disk->private_data;
1601 struct scsi_device *sdp;
1609 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1612 * If the device is offline, don't send any commands - just pretend as
1613 * if the command failed. If the device ever comes back online, we
1614 * can deal with it then. It is only because of unrecoverable errors
1615 * that we would ever take a device offline in the first place.
1617 if (!scsi_device_online(sdp)) {
1618 set_media_not_present(sdkp);
1623 * Using TEST_UNIT_READY enables differentiation between drive with
1624 * no cartridge loaded - NOT READY, drive with changed cartridge -
1625 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1627 * Drives that auto spin down. eg iomega jaz 1G, will be started
1628 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1629 * sd_revalidate() is called.
1631 if (scsi_block_when_processing_errors(sdp)) {
1632 struct scsi_sense_hdr sshdr = { 0, };
1634 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1637 /* failed to execute TUR, assume media not present */
1638 if (retval < 0 || host_byte(retval)) {
1639 set_media_not_present(sdkp);
1643 if (media_not_present(sdkp, &sshdr))
1648 * For removable scsi disk we have to recognise the presence
1649 * of a disk in the drive.
1651 if (!sdkp->media_present)
1653 sdkp->media_present = 1;
1656 * sdp->changed is set under the following conditions:
1658 * Medium present state has changed in either direction.
1659 * Device has indicated UNIT_ATTENTION.
1661 disk_changed = sdp->changed;
1663 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1666 static int sd_sync_cache(struct scsi_disk *sdkp)
1669 struct scsi_device *sdp = sdkp->device;
1670 const int timeout = sdp->request_queue->rq_timeout
1671 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1672 /* Leave the rest of the command zero to indicate flush everything. */
1673 const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1674 SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1675 struct scsi_sense_hdr sshdr;
1676 struct scsi_failure failure_defs[] = {
1679 .result = SCMD_FAILURE_RESULT_ANY,
1683 struct scsi_failures failures = {
1684 .failure_definitions = failure_defs,
1686 const struct scsi_exec_args exec_args = {
1687 .req_flags = BLK_MQ_REQ_PM,
1689 .failures = &failures,
1692 if (!scsi_device_online(sdp))
1695 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1696 sdkp->max_retries, &exec_args);
1698 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1703 if (scsi_status_is_check_condition(res) &&
1704 scsi_sense_valid(&sshdr)) {
1705 sd_print_sense_hdr(sdkp, &sshdr);
1707 /* we need to evaluate the error return */
1708 if (sshdr.asc == 0x3a || /* medium not present */
1709 sshdr.asc == 0x20 || /* invalid command */
1710 (sshdr.asc == 0x74 && sshdr.ascq == 0x71)) /* drive is password locked */
1711 /* this is no error here */
1714 * This drive doesn't support sync and there's not much
1715 * we can do because this is called during shutdown
1716 * or suspend so just return success so those operations
1719 if (sshdr.sense_key == ILLEGAL_REQUEST)
1723 switch (host_byte(res)) {
1724 /* ignore errors due to racing a disconnection */
1725 case DID_BAD_TARGET:
1726 case DID_NO_CONNECT:
1728 /* signal the upper layer it might try again */
1732 case DID_SOFT_ERROR:
1741 static void sd_rescan(struct device *dev)
1743 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1745 sd_revalidate_disk(sdkp->disk);
1748 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1749 enum blk_unique_id type)
1751 struct scsi_device *sdev = scsi_disk(disk)->device;
1752 const struct scsi_vpd *vpd;
1753 const unsigned char *d;
1754 int ret = -ENXIO, len;
1757 vpd = rcu_dereference(sdev->vpd_pg83);
1762 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1763 /* we only care about designators with LU association */
1764 if (((d[1] >> 4) & 0x3) != 0x00)
1766 if ((d[1] & 0xf) != type)
1770 * Only exit early if a 16-byte descriptor was found. Otherwise
1771 * keep looking as one with more entropy might still show up.
1774 if (len != 8 && len != 12 && len != 16)
1777 memcpy(id, d + 4, len);
1786 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1788 switch (host_byte(result)) {
1789 case DID_TRANSPORT_MARGINAL:
1790 case DID_TRANSPORT_DISRUPTED:
1792 return PR_STS_RETRY_PATH_FAILURE;
1793 case DID_NO_CONNECT:
1794 return PR_STS_PATH_FAILED;
1795 case DID_TRANSPORT_FAILFAST:
1796 return PR_STS_PATH_FAST_FAILED;
1799 switch (status_byte(result)) {
1800 case SAM_STAT_RESERVATION_CONFLICT:
1801 return PR_STS_RESERVATION_CONFLICT;
1802 case SAM_STAT_CHECK_CONDITION:
1803 if (!scsi_sense_valid(sshdr))
1804 return PR_STS_IOERR;
1806 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1807 (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1812 return PR_STS_IOERR;
1816 static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1817 unsigned char *data, int data_len)
1819 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1820 struct scsi_device *sdev = sdkp->device;
1821 struct scsi_sense_hdr sshdr;
1822 u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1823 struct scsi_failure failure_defs[] = {
1825 .sense = UNIT_ATTENTION,
1826 .asc = SCMD_FAILURE_ASC_ANY,
1827 .ascq = SCMD_FAILURE_ASCQ_ANY,
1829 .result = SAM_STAT_CHECK_CONDITION,
1833 struct scsi_failures failures = {
1834 .failure_definitions = failure_defs,
1836 const struct scsi_exec_args exec_args = {
1838 .failures = &failures,
1842 put_unaligned_be16(data_len, &cmd[7]);
1844 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1845 SD_TIMEOUT, sdkp->max_retries, &exec_args);
1846 if (scsi_status_is_check_condition(result) &&
1847 scsi_sense_valid(&sshdr)) {
1848 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1849 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1855 return sd_scsi_to_pr_err(&sshdr, result);
1858 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1860 int result, i, data_offset, num_copy_keys;
1861 u32 num_keys = keys_info->num_keys;
1862 int data_len = num_keys * 8 + 8;
1865 data = kzalloc(data_len, GFP_KERNEL);
1869 result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1873 keys_info->generation = get_unaligned_be32(&data[0]);
1874 keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1877 num_copy_keys = min(num_keys, keys_info->num_keys);
1879 for (i = 0; i < num_copy_keys; i++) {
1880 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1889 static int sd_pr_read_reservation(struct block_device *bdev,
1890 struct pr_held_reservation *rsv)
1892 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1893 struct scsi_device *sdev = sdkp->device;
1897 result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1901 len = get_unaligned_be32(&data[4]);
1905 /* Make sure we have at least the key and type */
1907 sdev_printk(KERN_INFO, sdev,
1908 "READ RESERVATION failed due to short return buffer of %d bytes\n",
1913 rsv->generation = get_unaligned_be32(&data[0]);
1914 rsv->key = get_unaligned_be64(&data[8]);
1915 rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1919 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1920 u64 sa_key, enum scsi_pr_type type, u8 flags)
1922 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1923 struct scsi_device *sdev = sdkp->device;
1924 struct scsi_sense_hdr sshdr;
1925 struct scsi_failure failure_defs[] = {
1927 .sense = UNIT_ATTENTION,
1928 .asc = SCMD_FAILURE_ASC_ANY,
1929 .ascq = SCMD_FAILURE_ASCQ_ANY,
1931 .result = SAM_STAT_CHECK_CONDITION,
1935 struct scsi_failures failures = {
1936 .failure_definitions = failure_defs,
1938 const struct scsi_exec_args exec_args = {
1940 .failures = &failures,
1943 u8 cmd[16] = { 0, };
1944 u8 data[24] = { 0, };
1946 cmd[0] = PERSISTENT_RESERVE_OUT;
1949 put_unaligned_be32(sizeof(data), &cmd[5]);
1951 put_unaligned_be64(key, &data[0]);
1952 put_unaligned_be64(sa_key, &data[8]);
1955 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1956 sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1959 if (scsi_status_is_check_condition(result) &&
1960 scsi_sense_valid(&sshdr)) {
1961 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1962 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1968 return sd_scsi_to_pr_err(&sshdr, result);
1971 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1974 if (flags & ~PR_FL_IGNORE_KEY)
1976 return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1977 old_key, new_key, 0,
1978 (1 << 0) /* APTPL */);
1981 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1986 return sd_pr_out_command(bdev, 0x01, key, 0,
1987 block_pr_type_to_scsi(type), 0);
1990 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1992 return sd_pr_out_command(bdev, 0x02, key, 0,
1993 block_pr_type_to_scsi(type), 0);
1996 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1997 enum pr_type type, bool abort)
1999 return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2000 block_pr_type_to_scsi(type), 0);
2003 static int sd_pr_clear(struct block_device *bdev, u64 key)
2005 return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2008 static const struct pr_ops sd_pr_ops = {
2009 .pr_register = sd_pr_register,
2010 .pr_reserve = sd_pr_reserve,
2011 .pr_release = sd_pr_release,
2012 .pr_preempt = sd_pr_preempt,
2013 .pr_clear = sd_pr_clear,
2014 .pr_read_keys = sd_pr_read_keys,
2015 .pr_read_reservation = sd_pr_read_reservation,
2018 static void scsi_disk_free_disk(struct gendisk *disk)
2020 struct scsi_disk *sdkp = scsi_disk(disk);
2022 put_device(&sdkp->disk_dev);
2025 static const struct block_device_operations sd_fops = {
2026 .owner = THIS_MODULE,
2028 .release = sd_release,
2030 .getgeo = sd_getgeo,
2031 .compat_ioctl = blkdev_compat_ptr_ioctl,
2032 .check_events = sd_check_events,
2033 .unlock_native_capacity = sd_unlock_native_capacity,
2034 .report_zones = sd_zbc_report_zones,
2035 .get_unique_id = sd_get_unique_id,
2036 .free_disk = scsi_disk_free_disk,
2037 .pr_ops = &sd_pr_ops,
2041 * sd_eh_reset - reset error handling callback
2042 * @scmd: sd-issued command that has failed
2044 * This function is called by the SCSI midlayer before starting
2045 * SCSI EH. When counting medium access failures we have to be
2046 * careful to register it only only once per device and SCSI EH run;
2047 * there might be several timed out commands which will cause the
2048 * 'max_medium_access_timeouts' counter to trigger after the first
2049 * SCSI EH run already and set the device to offline.
2050 * So this function resets the internal counter before starting SCSI EH.
2052 static void sd_eh_reset(struct scsi_cmnd *scmd)
2054 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2056 /* New SCSI EH run, reset gate variable */
2057 sdkp->ignore_medium_access_errors = false;
2061 * sd_eh_action - error handling callback
2062 * @scmd: sd-issued command that has failed
2063 * @eh_disp: The recovery disposition suggested by the midlayer
2065 * This function is called by the SCSI midlayer upon completion of an
2066 * error test command (currently TEST UNIT READY). The result of sending
2067 * the eh command is passed in eh_disp. We're looking for devices that
2068 * fail medium access commands but are OK with non access commands like
2069 * test unit ready (so wrongly see the device as having a successful
2072 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2074 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2075 struct scsi_device *sdev = scmd->device;
2077 if (!scsi_device_online(sdev) ||
2078 !scsi_medium_access_command(scmd) ||
2079 host_byte(scmd->result) != DID_TIME_OUT ||
2084 * The device has timed out executing a medium access command.
2085 * However, the TEST UNIT READY command sent during error
2086 * handling completed successfully. Either the device is in the
2087 * process of recovering or has it suffered an internal failure
2088 * that prevents access to the storage medium.
2090 if (!sdkp->ignore_medium_access_errors) {
2091 sdkp->medium_access_timed_out++;
2092 sdkp->ignore_medium_access_errors = true;
2096 * If the device keeps failing read/write commands but TEST UNIT
2097 * READY always completes successfully we assume that medium
2098 * access is no longer possible and take the device offline.
2100 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2101 scmd_printk(KERN_ERR, scmd,
2102 "Medium access timeout failure. Offlining disk!\n");
2103 mutex_lock(&sdev->state_mutex);
2104 scsi_device_set_state(sdev, SDEV_OFFLINE);
2105 mutex_unlock(&sdev->state_mutex);
2113 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2115 struct request *req = scsi_cmd_to_rq(scmd);
2116 struct scsi_device *sdev = scmd->device;
2117 unsigned int transferred, good_bytes;
2118 u64 start_lba, end_lba, bad_lba;
2121 * Some commands have a payload smaller than the device logical
2122 * block size (e.g. INQUIRY on a 4K disk).
2124 if (scsi_bufflen(scmd) <= sdev->sector_size)
2127 /* Check if we have a 'bad_lba' information */
2128 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2129 SCSI_SENSE_BUFFERSIZE,
2134 * If the bad lba was reported incorrectly, we have no idea where
2137 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2138 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2139 if (bad_lba < start_lba || bad_lba >= end_lba)
2143 * resid is optional but mostly filled in. When it's unused,
2144 * its value is zero, so we assume the whole buffer transferred
2146 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2148 /* This computation should always be done in terms of the
2149 * resolution of the device's medium.
2151 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2153 return min(good_bytes, transferred);
2157 * sd_done - bottom half handler: called when the lower level
2158 * driver has completed (successfully or otherwise) a scsi command.
2159 * @SCpnt: mid-level's per command structure.
2161 * Note: potentially run from within an ISR. Must not block.
2163 static int sd_done(struct scsi_cmnd *SCpnt)
2165 int result = SCpnt->result;
2166 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2167 unsigned int sector_size = SCpnt->device->sector_size;
2169 struct scsi_sense_hdr sshdr;
2170 struct request *req = scsi_cmd_to_rq(SCpnt);
2171 struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2172 int sense_valid = 0;
2173 int sense_deferred = 0;
2175 switch (req_op(req)) {
2176 case REQ_OP_DISCARD:
2177 case REQ_OP_WRITE_ZEROES:
2178 case REQ_OP_ZONE_RESET:
2179 case REQ_OP_ZONE_RESET_ALL:
2180 case REQ_OP_ZONE_OPEN:
2181 case REQ_OP_ZONE_CLOSE:
2182 case REQ_OP_ZONE_FINISH:
2184 good_bytes = blk_rq_bytes(req);
2185 scsi_set_resid(SCpnt, 0);
2188 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2193 * In case of bogus fw or device, we could end up having
2194 * an unaligned partial completion. Check this here and force
2197 resid = scsi_get_resid(SCpnt);
2198 if (resid & (sector_size - 1)) {
2199 sd_printk(KERN_INFO, sdkp,
2200 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2201 resid, sector_size);
2202 scsi_print_command(SCpnt);
2203 resid = min(scsi_bufflen(SCpnt),
2204 round_up(resid, sector_size));
2205 scsi_set_resid(SCpnt, resid);
2210 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2212 sense_deferred = scsi_sense_is_deferred(&sshdr);
2214 sdkp->medium_access_timed_out = 0;
2216 if (!scsi_status_is_check_condition(result) &&
2217 (!sense_valid || sense_deferred))
2220 switch (sshdr.sense_key) {
2221 case HARDWARE_ERROR:
2223 good_bytes = sd_completed_bytes(SCpnt);
2225 case RECOVERED_ERROR:
2226 good_bytes = scsi_bufflen(SCpnt);
2229 /* This indicates a false check condition, so ignore it. An
2230 * unknown amount of data was transferred so treat it as an
2234 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2236 case ABORTED_COMMAND:
2237 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2238 good_bytes = sd_completed_bytes(SCpnt);
2240 case ILLEGAL_REQUEST:
2241 switch (sshdr.asc) {
2242 case 0x10: /* DIX: Host detected corruption */
2243 good_bytes = sd_completed_bytes(SCpnt);
2245 case 0x20: /* INVALID COMMAND OPCODE */
2246 case 0x24: /* INVALID FIELD IN CDB */
2247 switch (SCpnt->cmnd[0]) {
2249 sd_config_discard(sdkp, SD_LBP_DISABLE);
2253 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2254 sd_config_discard(sdkp, SD_LBP_DISABLE);
2256 sdkp->device->no_write_same = 1;
2257 sd_config_write_same(sdkp);
2258 req->rq_flags |= RQF_QUIET;
2269 if (sd_is_zoned(sdkp))
2270 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2272 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2273 "sd_done: completed %d of %d bytes\n",
2274 good_bytes, scsi_bufflen(SCpnt)));
2280 * spinup disk - called only in sd_revalidate_disk()
2283 sd_spinup_disk(struct scsi_disk *sdkp)
2285 static const u8 cmd[10] = { TEST_UNIT_READY };
2286 unsigned long spintime_expire = 0;
2287 int spintime, sense_valid = 0;
2288 unsigned int the_result;
2289 struct scsi_sense_hdr sshdr;
2290 struct scsi_failure failure_defs[] = {
2291 /* Do not retry Medium Not Present */
2293 .sense = UNIT_ATTENTION,
2295 .ascq = SCMD_FAILURE_ASCQ_ANY,
2296 .result = SAM_STAT_CHECK_CONDITION,
2301 .ascq = SCMD_FAILURE_ASCQ_ANY,
2302 .result = SAM_STAT_CHECK_CONDITION,
2304 /* Retry when scsi_status_is_good would return false 3 times */
2306 .result = SCMD_FAILURE_STAT_ANY,
2311 struct scsi_failures failures = {
2312 .failure_definitions = failure_defs,
2314 const struct scsi_exec_args exec_args = {
2316 .failures = &failures,
2321 /* Spin up drives, as required. Only do this at boot time */
2322 /* Spinup needs to be done for module loads too. */
2324 bool media_was_present = sdkp->media_present;
2326 scsi_failures_reset_retries(&failures);
2328 the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2329 NULL, 0, SD_TIMEOUT,
2330 sdkp->max_retries, &exec_args);
2333 if (the_result > 0) {
2335 * If the drive has indicated to us that it doesn't
2336 * have any media in it, don't bother with any more
2339 if (media_not_present(sdkp, &sshdr)) {
2340 if (media_was_present)
2341 sd_printk(KERN_NOTICE, sdkp,
2342 "Media removed, stopped polling\n");
2345 sense_valid = scsi_sense_valid(&sshdr);
2348 if (!scsi_status_is_check_condition(the_result)) {
2349 /* no sense, TUR either succeeded or failed
2350 * with a status error */
2351 if(!spintime && !scsi_status_is_good(the_result)) {
2352 sd_print_result(sdkp, "Test Unit Ready failed",
2359 * The device does not want the automatic start to be issued.
2361 if (sdkp->device->no_start_on_add)
2364 if (sense_valid && sshdr.sense_key == NOT_READY) {
2365 if (sshdr.asc == 4 && sshdr.ascq == 3)
2366 break; /* manual intervention required */
2367 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2368 break; /* standby */
2369 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2370 break; /* unavailable */
2371 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2372 break; /* sanitize in progress */
2373 if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2374 break; /* depopulation in progress */
2375 if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2376 break; /* depopulation restoration in progress */
2378 * Issue command to spin up drive when not ready
2381 /* Return immediately and start spin cycle */
2382 const u8 start_cmd[10] = {
2385 [4] = sdkp->device->start_stop_pwr_cond ?
2389 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2390 scsi_execute_cmd(sdkp->device, start_cmd,
2391 REQ_OP_DRV_IN, NULL, 0,
2392 SD_TIMEOUT, sdkp->max_retries,
2394 spintime_expire = jiffies + 100 * HZ;
2397 /* Wait 1 second for next try */
2399 printk(KERN_CONT ".");
2402 * Wait for USB flash devices with slow firmware.
2403 * Yes, this sense key/ASC combination shouldn't
2404 * occur here. It's characteristic of these devices.
2406 } else if (sense_valid &&
2407 sshdr.sense_key == UNIT_ATTENTION &&
2408 sshdr.asc == 0x28) {
2410 spintime_expire = jiffies + 5 * HZ;
2413 /* Wait 1 second for next try */
2416 /* we don't understand the sense code, so it's
2417 * probably pointless to loop */
2419 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2420 sd_print_sense_hdr(sdkp, &sshdr);
2425 } while (spintime && time_before_eq(jiffies, spintime_expire));
2428 if (scsi_status_is_good(the_result))
2429 printk(KERN_CONT "ready\n");
2431 printk(KERN_CONT "not responding...\n");
2436 * Determine whether disk supports Data Integrity Field.
2438 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2440 struct scsi_device *sdp = sdkp->device;
2443 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2444 sdkp->protection_type = 0;
2448 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2450 if (type > T10_PI_TYPE3_PROTECTION) {
2451 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2452 " protection type %u. Disabling disk!\n",
2454 sdkp->protection_type = 0;
2458 sdkp->protection_type = type;
2463 static void sd_config_protection(struct scsi_disk *sdkp)
2465 struct scsi_device *sdp = sdkp->device;
2467 sd_dif_config_host(sdkp);
2469 if (!sdkp->protection_type)
2472 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2473 sd_first_printk(KERN_NOTICE, sdkp,
2474 "Disabling DIF Type %u protection\n",
2475 sdkp->protection_type);
2476 sdkp->protection_type = 0;
2479 sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2480 sdkp->protection_type);
2483 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2484 struct scsi_sense_hdr *sshdr, int sense_valid,
2488 sd_print_sense_hdr(sdkp, sshdr);
2490 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2493 * Set dirty bit for removable devices if not ready -
2494 * sometimes drives will not report this properly.
2496 if (sdp->removable &&
2497 sense_valid && sshdr->sense_key == NOT_READY)
2498 set_media_not_present(sdkp);
2501 * We used to set media_present to 0 here to indicate no media
2502 * in the drive, but some drives fail read capacity even with
2503 * media present, so we can't do that.
2505 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2509 #if RC16_LEN > SD_BUF_SIZE
2510 #error RC16_LEN must not be more than SD_BUF_SIZE
2513 #define READ_CAPACITY_RETRIES_ON_RESET 10
2515 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2516 unsigned char *buffer)
2518 unsigned char cmd[16];
2519 struct scsi_sense_hdr sshdr;
2520 const struct scsi_exec_args exec_args = {
2523 int sense_valid = 0;
2525 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2526 unsigned int alignment;
2527 unsigned long long lba;
2528 unsigned sector_size;
2530 if (sdp->no_read_capacity_16)
2535 cmd[0] = SERVICE_ACTION_IN_16;
2536 cmd[1] = SAI_READ_CAPACITY_16;
2538 memset(buffer, 0, RC16_LEN);
2540 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2541 buffer, RC16_LEN, SD_TIMEOUT,
2542 sdkp->max_retries, &exec_args);
2543 if (the_result > 0) {
2544 if (media_not_present(sdkp, &sshdr))
2547 sense_valid = scsi_sense_valid(&sshdr);
2549 sshdr.sense_key == ILLEGAL_REQUEST &&
2550 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2552 /* Invalid Command Operation Code or
2553 * Invalid Field in CDB, just retry
2554 * silently with RC10 */
2557 sshdr.sense_key == UNIT_ATTENTION &&
2558 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2559 /* Device reset might occur several times,
2560 * give it one more chance */
2561 if (--reset_retries > 0)
2566 } while (the_result && retries);
2569 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2570 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2574 sector_size = get_unaligned_be32(&buffer[8]);
2575 lba = get_unaligned_be64(&buffer[0]);
2577 if (sd_read_protection_type(sdkp, buffer) < 0) {
2582 /* Logical blocks per physical block exponent */
2583 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2586 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2588 /* Lowest aligned logical block */
2589 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2590 blk_queue_alignment_offset(sdp->request_queue, alignment);
2591 if (alignment && sdkp->first_scan)
2592 sd_printk(KERN_NOTICE, sdkp,
2593 "physical block alignment offset: %u\n", alignment);
2595 if (buffer[14] & 0x80) { /* LBPME */
2598 if (buffer[14] & 0x40) /* LBPRZ */
2601 sd_config_discard(sdkp, SD_LBP_WS16);
2604 sdkp->capacity = lba + 1;
2608 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2609 unsigned char *buffer)
2611 static const u8 cmd[10] = { READ_CAPACITY };
2612 struct scsi_sense_hdr sshdr;
2613 struct scsi_failure failure_defs[] = {
2614 /* Do not retry Medium Not Present */
2616 .sense = UNIT_ATTENTION,
2618 .result = SAM_STAT_CHECK_CONDITION,
2623 .result = SAM_STAT_CHECK_CONDITION,
2625 /* Device reset might occur several times so retry a lot */
2627 .sense = UNIT_ATTENTION,
2629 .allowed = READ_CAPACITY_RETRIES_ON_RESET,
2630 .result = SAM_STAT_CHECK_CONDITION,
2632 /* Any other error not listed above retry 3 times */
2634 .result = SCMD_FAILURE_RESULT_ANY,
2639 struct scsi_failures failures = {
2640 .failure_definitions = failure_defs,
2642 const struct scsi_exec_args exec_args = {
2644 .failures = &failures,
2646 int sense_valid = 0;
2649 unsigned sector_size;
2651 memset(buffer, 0, 8);
2653 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2654 8, SD_TIMEOUT, sdkp->max_retries,
2657 if (the_result > 0) {
2658 sense_valid = scsi_sense_valid(&sshdr);
2660 if (media_not_present(sdkp, &sshdr))
2665 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2666 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2670 sector_size = get_unaligned_be32(&buffer[4]);
2671 lba = get_unaligned_be32(&buffer[0]);
2673 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2674 /* Some buggy (usb cardreader) devices return an lba of
2675 0xffffffff when the want to report a size of 0 (with
2676 which they really mean no media is present) */
2678 sdkp->physical_block_size = sector_size;
2682 sdkp->capacity = lba + 1;
2683 sdkp->physical_block_size = sector_size;
2687 static int sd_try_rc16_first(struct scsi_device *sdp)
2689 if (sdp->host->max_cmd_len < 16)
2691 if (sdp->try_rc_10_first)
2693 if (sdp->scsi_level > SCSI_SPC_2)
2695 if (scsi_device_protection(sdp))
2701 * read disk capacity
2704 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2707 struct scsi_device *sdp = sdkp->device;
2709 if (sd_try_rc16_first(sdp)) {
2710 sector_size = read_capacity_16(sdkp, sdp, buffer);
2711 if (sector_size == -EOVERFLOW)
2713 if (sector_size == -ENODEV)
2715 if (sector_size < 0)
2716 sector_size = read_capacity_10(sdkp, sdp, buffer);
2717 if (sector_size < 0)
2720 sector_size = read_capacity_10(sdkp, sdp, buffer);
2721 if (sector_size == -EOVERFLOW)
2723 if (sector_size < 0)
2725 if ((sizeof(sdkp->capacity) > 4) &&
2726 (sdkp->capacity > 0xffffffffULL)) {
2727 int old_sector_size = sector_size;
2728 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2729 "Trying to use READ CAPACITY(16).\n");
2730 sector_size = read_capacity_16(sdkp, sdp, buffer);
2731 if (sector_size < 0) {
2732 sd_printk(KERN_NOTICE, sdkp,
2733 "Using 0xffffffff as device size\n");
2734 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2735 sector_size = old_sector_size;
2738 /* Remember that READ CAPACITY(16) succeeded */
2739 sdp->try_rc_10_first = 0;
2743 /* Some devices are known to return the total number of blocks,
2744 * not the highest block number. Some devices have versions
2745 * which do this and others which do not. Some devices we might
2746 * suspect of doing this but we don't know for certain.
2748 * If we know the reported capacity is wrong, decrement it. If
2749 * we can only guess, then assume the number of blocks is even
2750 * (usually true but not always) and err on the side of lowering
2753 if (sdp->fix_capacity ||
2754 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2755 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2756 "from its reported value: %llu\n",
2757 (unsigned long long) sdkp->capacity);
2762 if (sector_size == 0) {
2764 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2768 if (sector_size != 512 &&
2769 sector_size != 1024 &&
2770 sector_size != 2048 &&
2771 sector_size != 4096) {
2772 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2775 * The user might want to re-format the drive with
2776 * a supported sectorsize. Once this happens, it
2777 * would be relatively trivial to set the thing up.
2778 * For this reason, we leave the thing in the table.
2782 * set a bogus sector size so the normal read/write
2783 * logic in the block layer will eventually refuse any
2784 * request on this device without tripping over power
2785 * of two sector size assumptions
2789 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2790 blk_queue_physical_block_size(sdp->request_queue,
2791 sdkp->physical_block_size);
2792 sdkp->device->sector_size = sector_size;
2794 if (sdkp->capacity > 0xffffffff)
2795 sdp->use_16_for_rw = 1;
2800 * Print disk capacity
2803 sd_print_capacity(struct scsi_disk *sdkp,
2804 sector_t old_capacity)
2806 int sector_size = sdkp->device->sector_size;
2807 char cap_str_2[10], cap_str_10[10];
2809 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2812 string_get_size(sdkp->capacity, sector_size,
2813 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2814 string_get_size(sdkp->capacity, sector_size,
2815 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2817 sd_printk(KERN_NOTICE, sdkp,
2818 "%llu %d-byte logical blocks: (%s/%s)\n",
2819 (unsigned long long)sdkp->capacity,
2820 sector_size, cap_str_10, cap_str_2);
2822 if (sdkp->physical_block_size != sector_size)
2823 sd_printk(KERN_NOTICE, sdkp,
2824 "%u-byte physical blocks\n",
2825 sdkp->physical_block_size);
2828 /* called with buffer of length 512 */
2830 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2831 unsigned char *buffer, int len, struct scsi_mode_data *data,
2832 struct scsi_sense_hdr *sshdr)
2835 * If we must use MODE SENSE(10), make sure that the buffer length
2836 * is at least 8 bytes so that the mode sense header fits.
2838 if (sdkp->device->use_10_for_ms && len < 8)
2841 return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2842 SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2846 * read write protect setting, if possible - called only in sd_revalidate_disk()
2847 * called with buffer of length SD_BUF_SIZE
2850 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2853 struct scsi_device *sdp = sdkp->device;
2854 struct scsi_mode_data data;
2855 int old_wp = sdkp->write_prot;
2857 set_disk_ro(sdkp->disk, 0);
2858 if (sdp->skip_ms_page_3f) {
2859 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2863 if (sdp->use_192_bytes_for_3f) {
2864 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2867 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2868 * We have to start carefully: some devices hang if we ask
2869 * for more than is available.
2871 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2874 * Second attempt: ask for page 0 When only page 0 is
2875 * implemented, a request for page 3F may return Sense Key
2876 * 5: Illegal Request, Sense Code 24: Invalid field in
2880 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2883 * Third attempt: ask 255 bytes, as we did earlier.
2886 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2891 sd_first_printk(KERN_WARNING, sdkp,
2892 "Test WP failed, assume Write Enabled\n");
2894 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2895 set_disk_ro(sdkp->disk, sdkp->write_prot);
2896 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2897 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2898 sdkp->write_prot ? "on" : "off");
2899 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2905 * sd_read_cache_type - called only from sd_revalidate_disk()
2906 * called with buffer of length SD_BUF_SIZE
2909 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2912 struct scsi_device *sdp = sdkp->device;
2917 struct scsi_mode_data data;
2918 struct scsi_sense_hdr sshdr;
2919 int old_wce = sdkp->WCE;
2920 int old_rcd = sdkp->RCD;
2921 int old_dpofua = sdkp->DPOFUA;
2924 if (sdkp->cache_override)
2928 if (sdp->skip_ms_page_8) {
2929 if (sdp->type == TYPE_RBC)
2932 if (sdp->skip_ms_page_3f)
2935 if (sdp->use_192_bytes_for_3f)
2939 } else if (sdp->type == TYPE_RBC) {
2947 /* cautiously ask */
2948 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2954 if (!data.header_length) {
2957 sd_first_printk(KERN_ERR, sdkp,
2958 "Missing header in MODE_SENSE response\n");
2961 /* that went OK, now ask for the proper length */
2965 * We're only interested in the first three bytes, actually.
2966 * But the data cache page is defined for the first 20.
2970 else if (len > SD_BUF_SIZE) {
2971 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2972 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2975 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2979 if (len > first_len)
2980 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2984 int offset = data.header_length + data.block_descriptor_length;
2986 while (offset < len) {
2987 u8 page_code = buffer[offset] & 0x3F;
2988 u8 spf = buffer[offset] & 0x40;
2990 if (page_code == 8 || page_code == 6) {
2991 /* We're interested only in the first 3 bytes.
2993 if (len - offset <= 2) {
2994 sd_first_printk(KERN_ERR, sdkp,
2995 "Incomplete mode parameter "
2999 modepage = page_code;
3003 /* Go to the next page */
3004 if (spf && len - offset > 3)
3005 offset += 4 + (buffer[offset+2] << 8) +
3007 else if (!spf && len - offset > 1)
3008 offset += 2 + buffer[offset+1];
3010 sd_first_printk(KERN_ERR, sdkp,
3012 "parameter data\n");
3018 sd_first_printk(KERN_WARNING, sdkp,
3019 "No Caching mode page found\n");
3023 if (modepage == 8) {
3024 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3025 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3027 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3031 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3032 if (sdp->broken_fua) {
3033 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3035 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3036 !sdkp->device->use_16_for_rw) {
3037 sd_first_printk(KERN_NOTICE, sdkp,
3038 "Uses READ/WRITE(6), disabling FUA\n");
3042 /* No cache flush allowed for write protected devices */
3043 if (sdkp->WCE && sdkp->write_prot)
3046 if (sdkp->first_scan || old_wce != sdkp->WCE ||
3047 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3048 sd_printk(KERN_NOTICE, sdkp,
3049 "Write cache: %s, read cache: %s, %s\n",
3050 sdkp->WCE ? "enabled" : "disabled",
3051 sdkp->RCD ? "disabled" : "enabled",
3052 sdkp->DPOFUA ? "supports DPO and FUA"
3053 : "doesn't support DPO or FUA");
3059 if (res == -EIO && scsi_sense_valid(&sshdr) &&
3060 sshdr.sense_key == ILLEGAL_REQUEST &&
3061 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3062 /* Invalid field in CDB */
3063 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3065 sd_first_printk(KERN_ERR, sdkp,
3066 "Asking for cache data failed\n");
3069 if (sdp->wce_default_on) {
3070 sd_first_printk(KERN_NOTICE, sdkp,
3071 "Assuming drive cache: write back\n");
3074 sd_first_printk(KERN_WARNING, sdkp,
3075 "Assuming drive cache: write through\n");
3082 static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3084 u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3086 struct scsi_stream_status_header h;
3087 struct scsi_stream_status s;
3089 struct scsi_device *sdev = sdkp->device;
3090 struct scsi_sense_hdr sshdr;
3091 const struct scsi_exec_args exec_args = {
3096 put_unaligned_be16(stream_id, &cdb[4]);
3097 put_unaligned_be32(sizeof(buf), &cdb[10]);
3099 res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3100 SD_TIMEOUT, sdkp->max_retries, &exec_args);
3103 if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3104 sd_print_sense_hdr(sdkp, &sshdr);
3107 if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3109 return buf.h.stream_status[0].perm;
3112 static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3114 struct scsi_device *sdp = sdkp->device;
3115 const struct scsi_io_group_descriptor *desc, *start, *end;
3116 u16 permanent_stream_count_old;
3117 struct scsi_sense_hdr sshdr;
3118 struct scsi_mode_data data;
3121 res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3122 /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3123 sdkp->max_retries, &data, &sshdr);
3126 start = (void *)buffer + data.header_length + 16;
3127 end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3130 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3131 * should assign the lowest numbered stream identifiers to permanent
3134 for (desc = start; desc < end; desc++)
3135 if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3137 permanent_stream_count_old = sdkp->permanent_stream_count;
3138 sdkp->permanent_stream_count = desc - start;
3139 if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3140 sd_printk(KERN_INFO, sdkp,
3141 "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3142 sdkp->permanent_stream_count);
3143 else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3144 sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3145 sdkp->permanent_stream_count);
3149 * The ATO bit indicates whether the DIF application tag is available
3150 * for use by the operating system.
3152 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3155 struct scsi_device *sdp = sdkp->device;
3156 struct scsi_mode_data data;
3157 struct scsi_sense_hdr sshdr;
3159 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3162 if (sdkp->protection_type == 0)
3165 res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3166 sdkp->max_retries, &data, &sshdr);
3168 if (res < 0 || !data.header_length ||
3170 sd_first_printk(KERN_WARNING, sdkp,
3171 "getting Control mode page failed, assume no ATO\n");
3173 if (res == -EIO && scsi_sense_valid(&sshdr))
3174 sd_print_sense_hdr(sdkp, &sshdr);
3179 offset = data.header_length + data.block_descriptor_length;
3181 if ((buffer[offset] & 0x3f) != 0x0a) {
3182 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3186 if ((buffer[offset + 5] & 0x80) == 0)
3195 * sd_read_block_limits - Query disk device for preferred I/O sizes.
3196 * @sdkp: disk to query
3198 static void sd_read_block_limits(struct scsi_disk *sdkp)
3200 struct scsi_vpd *vpd;
3204 vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3205 if (!vpd || vpd->len < 16)
3208 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3209 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3210 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3212 if (vpd->len >= 64) {
3213 unsigned int lba_count, desc_count;
3215 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3220 lba_count = get_unaligned_be32(&vpd->data[20]);
3221 desc_count = get_unaligned_be32(&vpd->data[24]);
3223 if (lba_count && desc_count)
3224 sdkp->max_unmap_blocks = lba_count;
3226 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3228 if (vpd->data[32] & 0x80)
3229 sdkp->unmap_alignment =
3230 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3232 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3234 if (sdkp->max_unmap_blocks)
3235 sd_config_discard(sdkp, SD_LBP_UNMAP);
3237 sd_config_discard(sdkp, SD_LBP_WS16);
3239 } else { /* LBP VPD page tells us what to use */
3240 if (sdkp->lbpu && sdkp->max_unmap_blocks)
3241 sd_config_discard(sdkp, SD_LBP_UNMAP);
3242 else if (sdkp->lbpws)
3243 sd_config_discard(sdkp, SD_LBP_WS16);
3244 else if (sdkp->lbpws10)
3245 sd_config_discard(sdkp, SD_LBP_WS10);
3247 sd_config_discard(sdkp, SD_LBP_DISABLE);
3255 /* Parse the Block Limits Extension VPD page (0xb7) */
3256 static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3258 struct scsi_vpd *vpd;
3261 vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3262 if (vpd && vpd->len >= 2)
3263 sdkp->rscs = vpd->data[5] & 1;
3268 * sd_read_block_characteristics - Query block dev. characteristics
3269 * @sdkp: disk to query
3271 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3273 struct request_queue *q = sdkp->disk->queue;
3274 struct scsi_vpd *vpd;
3278 vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3280 if (!vpd || vpd->len < 8) {
3285 rot = get_unaligned_be16(&vpd->data[4]);
3286 sdkp->zoned = (vpd->data[8] >> 4) & 3;
3290 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3291 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3295 #ifdef CONFIG_BLK_DEV_ZONED /* sd_probe rejects ZBD devices early otherwise */
3296 if (sdkp->device->type == TYPE_ZBC) {
3300 disk_set_zoned(sdkp->disk);
3303 * Per ZBC and ZAC specifications, writes in sequential write
3304 * required zones of host-managed devices must be aligned to
3305 * the device physical block size.
3307 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3310 * Host-aware devices are treated as conventional.
3312 WARN_ON_ONCE(blk_queue_is_zoned(q));
3314 #endif /* CONFIG_BLK_DEV_ZONED */
3316 if (!sdkp->first_scan)
3319 if (blk_queue_is_zoned(q))
3320 sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3321 else if (sdkp->zoned == 1)
3322 sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3323 else if (sdkp->zoned == 2)
3324 sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3328 * sd_read_block_provisioning - Query provisioning VPD page
3329 * @sdkp: disk to query
3331 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3333 struct scsi_vpd *vpd;
3335 if (sdkp->lbpme == 0)
3339 vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3341 if (!vpd || vpd->len < 8) {
3347 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */
3348 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3349 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3353 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3355 struct scsi_device *sdev = sdkp->device;
3357 if (sdev->host->no_write_same) {
3358 sdev->no_write_same = 1;
3363 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3364 struct scsi_vpd *vpd;
3366 sdev->no_report_opcodes = 1;
3368 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3369 * CODES is unsupported and the device has an ATA
3370 * Information VPD page (SAT).
3373 vpd = rcu_dereference(sdev->vpd_pg89);
3375 sdev->no_write_same = 1;
3379 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3382 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3386 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3388 struct scsi_device *sdev = sdkp->device;
3390 if (!sdev->security_supported)
3393 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3394 SECURITY_PROTOCOL_IN, 0) == 1 &&
3395 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3396 SECURITY_PROTOCOL_OUT, 0) == 1)
3400 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3402 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3406 * sd_read_cpr - Query concurrent positioning ranges
3407 * @sdkp: disk to query
3409 static void sd_read_cpr(struct scsi_disk *sdkp)
3411 struct blk_independent_access_ranges *iars = NULL;
3412 unsigned char *buffer = NULL;
3413 unsigned int nr_cpr = 0;
3414 int i, vpd_len, buf_len = SD_BUF_SIZE;
3418 * We need to have the capacity set first for the block layer to be
3419 * able to check the ranges.
3421 if (sdkp->first_scan)
3424 if (!sdkp->capacity)
3428 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3429 * leading to a maximum page size of 64 + 256*32 bytes.
3431 buf_len = 64 + 256*32;
3432 buffer = kmalloc(buf_len, GFP_KERNEL);
3433 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3436 /* We must have at least a 64B header and one 32B range descriptor */
3437 vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3438 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3439 sd_printk(KERN_ERR, sdkp,
3440 "Invalid Concurrent Positioning Ranges VPD page\n");
3444 nr_cpr = (vpd_len - 64) / 32;
3450 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3457 for (i = 0; i < nr_cpr; i++, desc += 32) {
3459 sd_printk(KERN_ERR, sdkp,
3460 "Invalid Concurrent Positioning Range number\n");
3465 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3466 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3470 disk_set_independent_access_ranges(sdkp->disk, iars);
3471 if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3472 sd_printk(KERN_NOTICE, sdkp,
3473 "%u concurrent positioning ranges\n", nr_cpr);
3474 sdkp->nr_actuators = nr_cpr;
3480 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3482 struct scsi_device *sdp = sdkp->device;
3483 unsigned int min_xfer_bytes =
3484 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3486 if (sdkp->min_xfer_blocks == 0)
3489 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3490 sd_first_printk(KERN_WARNING, sdkp,
3491 "Preferred minimum I/O size %u bytes not a " \
3492 "multiple of physical block size (%u bytes)\n",
3493 min_xfer_bytes, sdkp->physical_block_size);
3494 sdkp->min_xfer_blocks = 0;
3498 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3504 * Determine the device's preferred I/O size for reads and writes
3505 * unless the reported value is unreasonably small, large, not a
3506 * multiple of the physical block size, or simply garbage.
3508 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3509 unsigned int dev_max)
3511 struct scsi_device *sdp = sdkp->device;
3512 unsigned int opt_xfer_bytes =
3513 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3514 unsigned int min_xfer_bytes =
3515 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3517 if (sdkp->opt_xfer_blocks == 0)
3520 if (sdkp->opt_xfer_blocks > dev_max) {
3521 sd_first_printk(KERN_WARNING, sdkp,
3522 "Optimal transfer size %u logical blocks " \
3523 "> dev_max (%u logical blocks)\n",
3524 sdkp->opt_xfer_blocks, dev_max);
3528 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3529 sd_first_printk(KERN_WARNING, sdkp,
3530 "Optimal transfer size %u logical blocks " \
3531 "> sd driver limit (%u logical blocks)\n",
3532 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3536 if (opt_xfer_bytes < PAGE_SIZE) {
3537 sd_first_printk(KERN_WARNING, sdkp,
3538 "Optimal transfer size %u bytes < " \
3539 "PAGE_SIZE (%u bytes)\n",
3540 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3544 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3545 sd_first_printk(KERN_WARNING, sdkp,
3546 "Optimal transfer size %u bytes not a " \
3547 "multiple of preferred minimum block " \
3548 "size (%u bytes)\n",
3549 opt_xfer_bytes, min_xfer_bytes);
3553 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3554 sd_first_printk(KERN_WARNING, sdkp,
3555 "Optimal transfer size %u bytes not a " \
3556 "multiple of physical block size (%u bytes)\n",
3557 opt_xfer_bytes, sdkp->physical_block_size);
3561 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3566 static void sd_read_block_zero(struct scsi_disk *sdkp)
3568 unsigned int buf_len = sdkp->device->sector_size;
3569 char *buffer, cmd[10] = { };
3571 buffer = kmalloc(buf_len, GFP_KERNEL);
3576 put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3577 put_unaligned_be16(1, &cmd[7]); /* Transfer 1 logical block */
3579 scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3580 SD_TIMEOUT, sdkp->max_retries, NULL);
3585 * sd_revalidate_disk - called the first time a new disk is seen,
3586 * performs disk spin up, read_capacity, etc.
3587 * @disk: struct gendisk we care about
3589 static int sd_revalidate_disk(struct gendisk *disk)
3591 struct scsi_disk *sdkp = scsi_disk(disk);
3592 struct scsi_device *sdp = sdkp->device;
3593 struct request_queue *q = sdkp->disk->queue;
3594 sector_t old_capacity = sdkp->capacity;
3595 unsigned char *buffer;
3596 unsigned int dev_max, rw_max;
3598 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3599 "sd_revalidate_disk\n"));
3602 * If the device is offline, don't try and read capacity or any
3603 * of the other niceties.
3605 if (!scsi_device_online(sdp))
3608 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3610 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3611 "allocation failure.\n");
3615 sd_spinup_disk(sdkp);
3618 * Without media there is no reason to ask; moreover, some devices
3619 * react badly if we do.
3621 if (sdkp->media_present) {
3622 sd_read_capacity(sdkp, buffer);
3624 * Some USB/UAS devices return generic values for mode pages
3625 * until the media has been accessed. Trigger a READ operation
3626 * to force the device to populate mode pages.
3628 if (sdp->read_before_ms)
3629 sd_read_block_zero(sdkp);
3631 * set the default to rotational. All non-rotational devices
3632 * support the block characteristics VPD page, which will
3633 * cause this to be updated correctly and any device which
3634 * doesn't support it should be treated as rotational.
3636 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3637 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3639 if (scsi_device_supports_vpd(sdp)) {
3640 sd_read_block_provisioning(sdkp);
3641 sd_read_block_limits(sdkp);
3642 sd_read_block_limits_ext(sdkp);
3643 sd_read_block_characteristics(sdkp);
3644 sd_zbc_read_zones(sdkp, buffer);
3648 sd_print_capacity(sdkp, old_capacity);
3650 sd_read_write_protect_flag(sdkp, buffer);
3651 sd_read_cache_type(sdkp, buffer);
3652 sd_read_io_hints(sdkp, buffer);
3653 sd_read_app_tag_own(sdkp, buffer);
3654 sd_read_write_same(sdkp, buffer);
3655 sd_read_security(sdkp, buffer);
3656 sd_config_protection(sdkp);
3660 * We now have all cache related info, determine how we deal
3661 * with flush requests.
3663 sd_set_flush_flag(sdkp);
3665 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3666 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3668 /* Some devices report a maximum block count for READ/WRITE requests. */
3669 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3670 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3672 if (sd_validate_min_xfer_size(sdkp))
3673 blk_queue_io_min(sdkp->disk->queue,
3674 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3676 blk_queue_io_min(sdkp->disk->queue, 0);
3678 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3679 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3680 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3682 q->limits.io_opt = 0;
3683 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3684 (sector_t)BLK_DEF_MAX_SECTORS_CAP);
3688 * Limit default to SCSI host optimal sector limit if set. There may be
3689 * an impact on performance for when the size of a request exceeds this
3692 rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3694 /* Do not exceed controller limit */
3695 rw_max = min(rw_max, queue_max_hw_sectors(q));
3698 * Only update max_sectors if previously unset or if the current value
3699 * exceeds the capabilities of the hardware.
3701 if (sdkp->first_scan ||
3702 q->limits.max_sectors > q->limits.max_dev_sectors ||
3703 q->limits.max_sectors > q->limits.max_hw_sectors)
3704 q->limits.max_sectors = rw_max;
3706 sdkp->first_scan = 0;
3708 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3709 sd_config_write_same(sdkp);
3713 * For a zoned drive, revalidating the zones can be done only once
3714 * the gendisk capacity is set. So if this fails, set back the gendisk
3717 if (sd_zbc_revalidate_zones(sdkp))
3718 set_capacity_and_notify(disk, 0);
3725 * sd_unlock_native_capacity - unlock native capacity
3726 * @disk: struct gendisk to set capacity for
3728 * Block layer calls this function if it detects that partitions
3729 * on @disk reach beyond the end of the device. If the SCSI host
3730 * implements ->unlock_native_capacity() method, it's invoked to
3731 * give it a chance to adjust the device capacity.
3734 * Defined by block layer. Might sleep.
3736 static void sd_unlock_native_capacity(struct gendisk *disk)
3738 struct scsi_device *sdev = scsi_disk(disk)->device;
3740 if (sdev->host->hostt->unlock_native_capacity)
3741 sdev->host->hostt->unlock_native_capacity(sdev);
3745 * sd_format_disk_name - format disk name
3746 * @prefix: name prefix - ie. "sd" for SCSI disks
3747 * @index: index of the disk to format name for
3748 * @buf: output buffer
3749 * @buflen: length of the output buffer
3751 * SCSI disk names starts at sda. The 26th device is sdz and the
3752 * 27th is sdaa. The last one for two lettered suffix is sdzz
3753 * which is followed by sdaaa.
3755 * This is basically 26 base counting with one extra 'nil' entry
3756 * at the beginning from the second digit on and can be
3757 * determined using similar method as 26 base conversion with the
3758 * index shifted -1 after each digit is computed.
3764 * 0 on success, -errno on failure.
3766 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3768 const int base = 'z' - 'a' + 1;
3769 char *begin = buf + strlen(prefix);
3770 char *end = buf + buflen;
3780 *--p = 'a' + (index % unit);
3781 index = (index / unit) - 1;
3782 } while (index >= 0);
3784 memmove(begin, p, end - p);
3785 memcpy(buf, prefix, strlen(prefix));
3791 * sd_probe - called during driver initialization and whenever a
3792 * new scsi device is attached to the system. It is called once
3793 * for each scsi device (not just disks) present.
3794 * @dev: pointer to device object
3796 * Returns 0 if successful (or not interested in this scsi device
3797 * (e.g. scanner)); 1 when there is an error.
3799 * Note: this function is invoked from the scsi mid-level.
3800 * This function sets up the mapping between a given
3801 * <host,channel,id,lun> (found in sdp) and new device name
3802 * (e.g. /dev/sda). More precisely it is the block device major
3803 * and minor number that is chosen here.
3805 * Assume sd_probe is not re-entrant (for time being)
3806 * Also think about sd_probe() and sd_remove() running coincidentally.
3808 static int sd_probe(struct device *dev)
3810 struct scsi_device *sdp = to_scsi_device(dev);
3811 struct scsi_disk *sdkp;
3816 scsi_autopm_get_device(sdp);
3818 if (sdp->type != TYPE_DISK &&
3819 sdp->type != TYPE_ZBC &&
3820 sdp->type != TYPE_MOD &&
3821 sdp->type != TYPE_RBC)
3824 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3825 sdev_printk(KERN_WARNING, sdp,
3826 "Unsupported ZBC host-managed device.\n");
3830 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3834 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3838 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3839 &sd_bio_compl_lkclass);
3843 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3845 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3849 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3851 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3852 goto out_free_index;
3857 sdkp->index = index;
3858 sdkp->max_retries = SD_MAX_RETRIES;
3859 atomic_set(&sdkp->openers, 0);
3860 atomic_set(&sdkp->device->ioerr_cnt, 0);
3862 if (!sdp->request_queue->rq_timeout) {
3863 if (sdp->type != TYPE_MOD)
3864 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3866 blk_queue_rq_timeout(sdp->request_queue,
3870 device_initialize(&sdkp->disk_dev);
3871 sdkp->disk_dev.parent = get_device(dev);
3872 sdkp->disk_dev.class = &sd_disk_class;
3873 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3875 error = device_add(&sdkp->disk_dev);
3877 put_device(&sdkp->disk_dev);
3881 dev_set_drvdata(dev, sdkp);
3883 gd->major = sd_major((index & 0xf0) >> 4);
3884 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3885 gd->minors = SD_MINORS;
3887 gd->fops = &sd_fops;
3888 gd->private_data = sdkp;
3890 /* defaults, until the device tells us otherwise */
3891 sdp->sector_size = 512;
3893 sdkp->media_present = 1;
3894 sdkp->write_prot = 0;
3895 sdkp->cache_override = 0;
3899 sdkp->first_scan = 1;
3900 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3902 sd_revalidate_disk(gd);
3904 if (sdp->removable) {
3905 gd->flags |= GENHD_FL_REMOVABLE;
3906 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3907 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3910 blk_pm_runtime_init(sdp->request_queue, dev);
3911 if (sdp->rpm_autosuspend) {
3912 pm_runtime_set_autosuspend_delay(dev,
3913 sdp->host->rpm_autosuspend_delay);
3916 error = device_add_disk(dev, gd, NULL);
3918 device_unregister(&sdkp->disk_dev);
3923 if (sdkp->security) {
3924 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3926 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3929 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3930 sdp->removable ? "removable " : "");
3931 scsi_autopm_put_device(sdp);
3936 ida_free(&sd_index_ida, index);
3942 scsi_autopm_put_device(sdp);
3947 * sd_remove - called whenever a scsi disk (previously recognized by
3948 * sd_probe) is detached from the system. It is called (potentially
3949 * multiple times) during sd module unload.
3950 * @dev: pointer to device object
3952 * Note: this function is invoked from the scsi mid-level.
3953 * This function potentially frees up a device name (e.g. /dev/sdc)
3954 * that could be re-used by a subsequent sd_probe().
3955 * This function is not called when the built-in sd driver is "exit-ed".
3957 static int sd_remove(struct device *dev)
3959 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3961 scsi_autopm_get_device(sdkp->device);
3963 device_del(&sdkp->disk_dev);
3964 del_gendisk(sdkp->disk);
3965 if (!sdkp->suspended)
3968 put_disk(sdkp->disk);
3972 static void scsi_disk_release(struct device *dev)
3974 struct scsi_disk *sdkp = to_scsi_disk(dev);
3976 ida_free(&sd_index_ida, sdkp->index);
3977 put_device(&sdkp->device->sdev_gendev);
3978 free_opal_dev(sdkp->opal_dev);
3983 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3985 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3986 struct scsi_sense_hdr sshdr;
3987 const struct scsi_exec_args exec_args = {
3989 .req_flags = BLK_MQ_REQ_PM,
3991 struct scsi_device *sdp = sdkp->device;
3995 cmd[4] |= 1; /* START */
3997 if (sdp->start_stop_pwr_cond)
3998 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
4000 if (!scsi_device_online(sdp))
4003 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4004 sdkp->max_retries, &exec_args);
4006 sd_print_result(sdkp, "Start/Stop Unit failed", res);
4007 if (res > 0 && scsi_sense_valid(&sshdr)) {
4008 sd_print_sense_hdr(sdkp, &sshdr);
4009 /* 0x3a is medium not present */
4010 if (sshdr.asc == 0x3a)
4015 /* SCSI error codes must not go to the generic layer */
4023 * Send a SYNCHRONIZE CACHE instruction down to the device through
4024 * the normal SCSI command structure. Wait for the command to
4027 static void sd_shutdown(struct device *dev)
4029 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4032 return; /* this can happen */
4034 if (pm_runtime_suspended(dev))
4037 if (sdkp->WCE && sdkp->media_present) {
4038 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4039 sd_sync_cache(sdkp);
4042 if ((system_state != SYSTEM_RESTART &&
4043 sdkp->device->manage_system_start_stop) ||
4044 (system_state == SYSTEM_POWER_OFF &&
4045 sdkp->device->manage_shutdown)) {
4046 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4047 sd_start_stop_device(sdkp, 0);
4051 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4053 return (sdev->manage_system_start_stop && !runtime) ||
4054 (sdev->manage_runtime_start_stop && runtime);
4057 static int sd_suspend_common(struct device *dev, bool runtime)
4059 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4062 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
4065 if (sdkp->WCE && sdkp->media_present) {
4066 if (!sdkp->device->silence_suspend)
4067 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4068 ret = sd_sync_cache(sdkp);
4069 /* ignore OFFLINE device */
4077 if (sd_do_start_stop(sdkp->device, runtime)) {
4078 if (!sdkp->device->silence_suspend)
4079 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4080 /* an error is not worth aborting a system sleep */
4081 ret = sd_start_stop_device(sdkp, 0);
4087 sdkp->suspended = true;
4092 static int sd_suspend_system(struct device *dev)
4094 if (pm_runtime_suspended(dev))
4097 return sd_suspend_common(dev, false);
4100 static int sd_suspend_runtime(struct device *dev)
4102 return sd_suspend_common(dev, true);
4105 static int sd_resume(struct device *dev)
4107 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4109 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4111 if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4112 sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4119 static int sd_resume_common(struct device *dev, bool runtime)
4121 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4124 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4127 if (!sd_do_start_stop(sdkp->device, runtime)) {
4128 sdkp->suspended = false;
4132 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4133 ret = sd_start_stop_device(sdkp, 1);
4136 sdkp->suspended = false;
4142 static int sd_resume_system(struct device *dev)
4144 if (pm_runtime_suspended(dev)) {
4145 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4146 struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4148 if (sdp && sdp->force_runtime_start_on_system_start)
4149 pm_request_resume(dev);
4154 return sd_resume_common(dev, false);
4157 static int sd_resume_runtime(struct device *dev)
4159 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4160 struct scsi_device *sdp;
4162 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4167 if (sdp->ignore_media_change) {
4168 /* clear the device's sense data */
4169 static const u8 cmd[10] = { REQUEST_SENSE };
4170 const struct scsi_exec_args exec_args = {
4171 .req_flags = BLK_MQ_REQ_PM,
4174 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4175 sdp->request_queue->rq_timeout, 1,
4177 sd_printk(KERN_NOTICE, sdkp,
4178 "Failed to clear sense data\n");
4181 return sd_resume_common(dev, true);
4184 static const struct dev_pm_ops sd_pm_ops = {
4185 .suspend = sd_suspend_system,
4186 .resume = sd_resume_system,
4187 .poweroff = sd_suspend_system,
4188 .restore = sd_resume_system,
4189 .runtime_suspend = sd_suspend_runtime,
4190 .runtime_resume = sd_resume_runtime,
4193 static struct scsi_driver sd_template = {
4197 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
4198 .remove = sd_remove,
4199 .shutdown = sd_shutdown,
4202 .rescan = sd_rescan,
4203 .resume = sd_resume,
4204 .init_command = sd_init_command,
4205 .uninit_command = sd_uninit_command,
4207 .eh_action = sd_eh_action,
4208 .eh_reset = sd_eh_reset,
4212 * init_sd - entry point for this driver (both when built in or when
4215 * Note: this function registers this driver with the scsi mid-level.
4217 static int __init init_sd(void)
4219 int majors = 0, i, err;
4221 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4223 for (i = 0; i < SD_MAJORS; i++) {
4224 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4232 err = class_register(&sd_disk_class);
4236 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4237 if (!sd_page_pool) {
4238 printk(KERN_ERR "sd: can't init discard page pool\n");
4243 err = scsi_register_driver(&sd_template.gendrv);
4245 goto err_out_driver;
4250 mempool_destroy(sd_page_pool);
4252 class_unregister(&sd_disk_class);
4254 for (i = 0; i < SD_MAJORS; i++)
4255 unregister_blkdev(sd_major(i), "sd");
4260 * exit_sd - exit point for this driver (when it is a module).
4262 * Note: this function unregisters this driver from the scsi mid-level.
4264 static void __exit exit_sd(void)
4268 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4270 scsi_unregister_driver(&sd_template.gendrv);
4271 mempool_destroy(sd_page_pool);
4273 class_unregister(&sd_disk_class);
4275 for (i = 0; i < SD_MAJORS; i++)
4276 unregister_blkdev(sd_major(i), "sd");
4279 module_init(init_sd);
4280 module_exit(exit_sd);
4282 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4284 scsi_print_sense_hdr(sdkp->device,
4285 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4288 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4290 const char *hb_string = scsi_hostbyte_string(result);
4293 sd_printk(KERN_INFO, sdkp,
4294 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4295 hb_string ? hb_string : "invalid",
4298 sd_printk(KERN_INFO, sdkp,
4299 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4300 msg, host_byte(result), "DRIVER_OK");