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/major.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
102 static void sd_config_discard(struct scsi_disk *, unsigned int);
103 static void sd_config_write_same(struct scsi_disk *);
104 static int sd_revalidate_disk(struct gendisk *);
105 static void sd_unlock_native_capacity(struct gendisk *disk);
106 static int sd_probe(struct device *);
107 static int sd_remove(struct device *);
108 static void sd_shutdown(struct device *);
109 static int sd_suspend_system(struct device *);
110 static int sd_suspend_runtime(struct device *);
111 static int sd_resume_system(struct device *);
112 static int sd_resume_runtime(struct device *);
113 static void sd_rescan(struct device *);
114 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
115 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
116 static int sd_done(struct scsi_cmnd *);
117 static void sd_eh_reset(struct scsi_cmnd *);
118 static int sd_eh_action(struct scsi_cmnd *, int);
119 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
120 static void scsi_disk_release(struct device *cdev);
122 static DEFINE_IDA(sd_index_ida);
124 static struct kmem_cache *sd_cdb_cache;
125 static mempool_t *sd_page_pool;
126 static struct lock_class_key sd_bio_compl_lkclass;
128 static const char *sd_cache_types[] = {
129 "write through", "none", "write back",
130 "write back, no read (daft)"
133 static void sd_set_flush_flag(struct scsi_disk *sdkp)
135 bool wc = false, fua = false;
143 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
147 cache_type_store(struct device *dev, struct device_attribute *attr,
148 const char *buf, size_t count)
150 int ct, rcd, wce, sp;
151 struct scsi_disk *sdkp = to_scsi_disk(dev);
152 struct scsi_device *sdp = sdkp->device;
155 struct scsi_mode_data data;
156 struct scsi_sense_hdr sshdr;
157 static const char temp[] = "temporary ";
160 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
161 /* no cache control on RBC devices; theoretically they
162 * can do it, but there's probably so many exceptions
163 * it's not worth the risk */
166 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
167 buf += sizeof(temp) - 1;
168 sdkp->cache_override = 1;
170 sdkp->cache_override = 0;
173 ct = sysfs_match_string(sd_cache_types, buf);
177 rcd = ct & 0x01 ? 1 : 0;
178 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
180 if (sdkp->cache_override) {
183 sd_set_flush_flag(sdkp);
187 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
188 sdkp->max_retries, &data, NULL))
190 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
191 data.block_descriptor_length);
192 buffer_data = buffer + data.header_length +
193 data.block_descriptor_length;
194 buffer_data[2] &= ~0x05;
195 buffer_data[2] |= wce << 2 | rcd;
196 sp = buffer_data[0] & 0x80 ? 1 : 0;
197 buffer_data[0] &= ~0x80;
200 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
201 * received mode parameter buffer before doing MODE SELECT.
203 data.device_specific = 0;
205 if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
206 sdkp->max_retries, &data, &sshdr)) {
207 if (scsi_sense_valid(&sshdr))
208 sd_print_sense_hdr(sdkp, &sshdr);
211 sd_revalidate_disk(sdkp->disk);
216 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
219 struct scsi_disk *sdkp = to_scsi_disk(dev);
220 struct scsi_device *sdp = sdkp->device;
222 return sprintf(buf, "%u\n", sdp->manage_start_stop);
226 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
227 const char *buf, size_t count)
229 struct scsi_disk *sdkp = to_scsi_disk(dev);
230 struct scsi_device *sdp = sdkp->device;
233 if (!capable(CAP_SYS_ADMIN))
236 if (kstrtobool(buf, &v))
239 sdp->manage_start_stop = v;
243 static DEVICE_ATTR_RW(manage_start_stop);
246 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
248 struct scsi_disk *sdkp = to_scsi_disk(dev);
250 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
254 allow_restart_store(struct device *dev, struct device_attribute *attr,
255 const char *buf, size_t count)
258 struct scsi_disk *sdkp = to_scsi_disk(dev);
259 struct scsi_device *sdp = sdkp->device;
261 if (!capable(CAP_SYS_ADMIN))
264 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
267 if (kstrtobool(buf, &v))
270 sdp->allow_restart = v;
274 static DEVICE_ATTR_RW(allow_restart);
277 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
279 struct scsi_disk *sdkp = to_scsi_disk(dev);
280 int ct = sdkp->RCD + 2*sdkp->WCE;
282 return sprintf(buf, "%s\n", sd_cache_types[ct]);
284 static DEVICE_ATTR_RW(cache_type);
287 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
289 struct scsi_disk *sdkp = to_scsi_disk(dev);
291 return sprintf(buf, "%u\n", sdkp->DPOFUA);
293 static DEVICE_ATTR_RO(FUA);
296 protection_type_show(struct device *dev, struct device_attribute *attr,
299 struct scsi_disk *sdkp = to_scsi_disk(dev);
301 return sprintf(buf, "%u\n", sdkp->protection_type);
305 protection_type_store(struct device *dev, struct device_attribute *attr,
306 const char *buf, size_t count)
308 struct scsi_disk *sdkp = to_scsi_disk(dev);
312 if (!capable(CAP_SYS_ADMIN))
315 err = kstrtouint(buf, 10, &val);
320 if (val <= T10_PI_TYPE3_PROTECTION)
321 sdkp->protection_type = val;
325 static DEVICE_ATTR_RW(protection_type);
328 protection_mode_show(struct device *dev, struct device_attribute *attr,
331 struct scsi_disk *sdkp = to_scsi_disk(dev);
332 struct scsi_device *sdp = sdkp->device;
333 unsigned int dif, dix;
335 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
336 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
338 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
344 return sprintf(buf, "none\n");
346 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
348 static DEVICE_ATTR_RO(protection_mode);
351 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
353 struct scsi_disk *sdkp = to_scsi_disk(dev);
355 return sprintf(buf, "%u\n", sdkp->ATO);
357 static DEVICE_ATTR_RO(app_tag_own);
360 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
363 struct scsi_disk *sdkp = to_scsi_disk(dev);
365 return sprintf(buf, "%u\n", sdkp->lbpme);
367 static DEVICE_ATTR_RO(thin_provisioning);
369 /* sysfs_match_string() requires dense arrays */
370 static const char *lbp_mode[] = {
371 [SD_LBP_FULL] = "full",
372 [SD_LBP_UNMAP] = "unmap",
373 [SD_LBP_WS16] = "writesame_16",
374 [SD_LBP_WS10] = "writesame_10",
375 [SD_LBP_ZERO] = "writesame_zero",
376 [SD_LBP_DISABLE] = "disabled",
380 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
383 struct scsi_disk *sdkp = to_scsi_disk(dev);
385 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
389 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
390 const char *buf, size_t count)
392 struct scsi_disk *sdkp = to_scsi_disk(dev);
393 struct scsi_device *sdp = sdkp->device;
396 if (!capable(CAP_SYS_ADMIN))
399 if (sd_is_zoned(sdkp)) {
400 sd_config_discard(sdkp, SD_LBP_DISABLE);
404 if (sdp->type != TYPE_DISK)
407 mode = sysfs_match_string(lbp_mode, buf);
411 sd_config_discard(sdkp, mode);
415 static DEVICE_ATTR_RW(provisioning_mode);
417 /* sysfs_match_string() requires dense arrays */
418 static const char *zeroing_mode[] = {
419 [SD_ZERO_WRITE] = "write",
420 [SD_ZERO_WS] = "writesame",
421 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
422 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
426 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
429 struct scsi_disk *sdkp = to_scsi_disk(dev);
431 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
435 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
436 const char *buf, size_t count)
438 struct scsi_disk *sdkp = to_scsi_disk(dev);
441 if (!capable(CAP_SYS_ADMIN))
444 mode = sysfs_match_string(zeroing_mode, buf);
448 sdkp->zeroing_mode = mode;
452 static DEVICE_ATTR_RW(zeroing_mode);
455 max_medium_access_timeouts_show(struct device *dev,
456 struct device_attribute *attr, char *buf)
458 struct scsi_disk *sdkp = to_scsi_disk(dev);
460 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
464 max_medium_access_timeouts_store(struct device *dev,
465 struct device_attribute *attr, const char *buf,
468 struct scsi_disk *sdkp = to_scsi_disk(dev);
471 if (!capable(CAP_SYS_ADMIN))
474 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
476 return err ? err : count;
478 static DEVICE_ATTR_RW(max_medium_access_timeouts);
481 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
484 struct scsi_disk *sdkp = to_scsi_disk(dev);
486 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
490 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
491 const char *buf, size_t count)
493 struct scsi_disk *sdkp = to_scsi_disk(dev);
494 struct scsi_device *sdp = sdkp->device;
498 if (!capable(CAP_SYS_ADMIN))
501 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
504 err = kstrtoul(buf, 10, &max);
510 sdp->no_write_same = 1;
511 else if (max <= SD_MAX_WS16_BLOCKS) {
512 sdp->no_write_same = 0;
513 sdkp->max_ws_blocks = max;
516 sd_config_write_same(sdkp);
520 static DEVICE_ATTR_RW(max_write_same_blocks);
523 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
525 struct scsi_disk *sdkp = to_scsi_disk(dev);
527 if (sdkp->device->type == TYPE_ZBC)
528 return sprintf(buf, "host-managed\n");
529 if (sdkp->zoned == 1)
530 return sprintf(buf, "host-aware\n");
531 if (sdkp->zoned == 2)
532 return sprintf(buf, "drive-managed\n");
533 return sprintf(buf, "none\n");
535 static DEVICE_ATTR_RO(zoned_cap);
538 max_retries_store(struct device *dev, struct device_attribute *attr,
539 const char *buf, size_t count)
541 struct scsi_disk *sdkp = to_scsi_disk(dev);
542 struct scsi_device *sdev = sdkp->device;
545 err = kstrtoint(buf, 10, &retries);
549 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
550 sdkp->max_retries = retries;
554 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
560 max_retries_show(struct device *dev, struct device_attribute *attr,
563 struct scsi_disk *sdkp = to_scsi_disk(dev);
565 return sprintf(buf, "%d\n", sdkp->max_retries);
568 static DEVICE_ATTR_RW(max_retries);
570 static struct attribute *sd_disk_attrs[] = {
571 &dev_attr_cache_type.attr,
573 &dev_attr_allow_restart.attr,
574 &dev_attr_manage_start_stop.attr,
575 &dev_attr_protection_type.attr,
576 &dev_attr_protection_mode.attr,
577 &dev_attr_app_tag_own.attr,
578 &dev_attr_thin_provisioning.attr,
579 &dev_attr_provisioning_mode.attr,
580 &dev_attr_zeroing_mode.attr,
581 &dev_attr_max_write_same_blocks.attr,
582 &dev_attr_max_medium_access_timeouts.attr,
583 &dev_attr_zoned_cap.attr,
584 &dev_attr_max_retries.attr,
587 ATTRIBUTE_GROUPS(sd_disk);
589 static struct class sd_disk_class = {
591 .owner = THIS_MODULE,
592 .dev_release = scsi_disk_release,
593 .dev_groups = sd_disk_groups,
596 static const struct dev_pm_ops sd_pm_ops = {
597 .suspend = sd_suspend_system,
598 .resume = sd_resume_system,
599 .poweroff = sd_suspend_system,
600 .restore = sd_resume_system,
601 .runtime_suspend = sd_suspend_runtime,
602 .runtime_resume = sd_resume_runtime,
605 static struct scsi_driver sd_template = {
608 .owner = THIS_MODULE,
610 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
612 .shutdown = sd_shutdown,
616 .init_command = sd_init_command,
617 .uninit_command = sd_uninit_command,
619 .eh_action = sd_eh_action,
620 .eh_reset = sd_eh_reset,
624 * Don't request a new module, as that could deadlock in multipath
627 static void sd_default_probe(dev_t devt)
632 * Device no to disk mapping:
634 * major disc2 disc p1
635 * |............|.............|....|....| <- dev_t
638 * Inside a major, we have 16k disks, however mapped non-
639 * contiguously. The first 16 disks are for major0, the next
640 * ones with major1, ... Disk 256 is for major0 again, disk 272
642 * As we stay compatible with our numbering scheme, we can reuse
643 * the well-know SCSI majors 8, 65--71, 136--143.
645 static int sd_major(int major_idx)
649 return SCSI_DISK0_MAJOR;
651 return SCSI_DISK1_MAJOR + major_idx - 1;
653 return SCSI_DISK8_MAJOR + major_idx - 8;
656 return 0; /* shut up gcc */
660 #ifdef CONFIG_BLK_SED_OPAL
661 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
662 size_t len, bool send)
664 struct scsi_disk *sdkp = data;
665 struct scsi_device *sdev = sdkp->device;
669 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
671 put_unaligned_be16(spsp, &cdb[2]);
672 put_unaligned_be32(len, &cdb[6]);
674 ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
675 buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
677 return ret <= 0 ? ret : -EIO;
679 #endif /* CONFIG_BLK_SED_OPAL */
682 * Look up the DIX operation based on whether the command is read or
683 * write and whether dix and dif are enabled.
685 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
687 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
688 static const unsigned int ops[] = { /* wrt dix dif */
689 SCSI_PROT_NORMAL, /* 0 0 0 */
690 SCSI_PROT_READ_STRIP, /* 0 0 1 */
691 SCSI_PROT_READ_INSERT, /* 0 1 0 */
692 SCSI_PROT_READ_PASS, /* 0 1 1 */
693 SCSI_PROT_NORMAL, /* 1 0 0 */
694 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
695 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
696 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
699 return ops[write << 2 | dix << 1 | dif];
703 * Returns a mask of the protection flags that are valid for a given DIX
706 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
708 static const unsigned int flag_mask[] = {
709 [SCSI_PROT_NORMAL] = 0,
711 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
712 SCSI_PROT_GUARD_CHECK |
713 SCSI_PROT_REF_CHECK |
714 SCSI_PROT_REF_INCREMENT,
716 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
717 SCSI_PROT_IP_CHECKSUM,
719 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
720 SCSI_PROT_GUARD_CHECK |
721 SCSI_PROT_REF_CHECK |
722 SCSI_PROT_REF_INCREMENT |
723 SCSI_PROT_IP_CHECKSUM,
725 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
726 SCSI_PROT_REF_INCREMENT,
728 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
729 SCSI_PROT_REF_CHECK |
730 SCSI_PROT_REF_INCREMENT |
731 SCSI_PROT_IP_CHECKSUM,
733 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
734 SCSI_PROT_GUARD_CHECK |
735 SCSI_PROT_REF_CHECK |
736 SCSI_PROT_REF_INCREMENT |
737 SCSI_PROT_IP_CHECKSUM,
740 return flag_mask[prot_op];
743 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
744 unsigned int dix, unsigned int dif)
746 struct request *rq = scsi_cmd_to_rq(scmd);
747 struct bio *bio = rq->bio;
748 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
749 unsigned int protect = 0;
751 if (dix) { /* DIX Type 0, 1, 2, 3 */
752 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
753 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
755 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
756 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
759 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
760 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
762 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
763 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
766 if (dif) { /* DIX/DIF Type 1, 2, 3 */
767 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
769 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
770 protect = 3 << 5; /* Disable target PI checking */
772 protect = 1 << 5; /* Enable target PI checking */
775 scsi_set_prot_op(scmd, prot_op);
776 scsi_set_prot_type(scmd, dif);
777 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
782 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
784 struct request_queue *q = sdkp->disk->queue;
785 unsigned int logical_block_size = sdkp->device->sector_size;
786 unsigned int max_blocks = 0;
788 q->limits.discard_alignment =
789 sdkp->unmap_alignment * logical_block_size;
790 q->limits.discard_granularity =
791 max(sdkp->physical_block_size,
792 sdkp->unmap_granularity * logical_block_size);
793 sdkp->provisioning_mode = mode;
799 blk_queue_max_discard_sectors(q, 0);
803 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
804 (u32)SD_MAX_WS16_BLOCKS);
808 if (sdkp->device->unmap_limit_for_ws)
809 max_blocks = sdkp->max_unmap_blocks;
811 max_blocks = sdkp->max_ws_blocks;
813 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
817 if (sdkp->device->unmap_limit_for_ws)
818 max_blocks = sdkp->max_unmap_blocks;
820 max_blocks = sdkp->max_ws_blocks;
822 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
826 max_blocks = min_not_zero(sdkp->max_ws_blocks,
827 (u32)SD_MAX_WS10_BLOCKS);
831 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
834 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
836 struct scsi_device *sdp = cmd->device;
837 struct request *rq = scsi_cmd_to_rq(cmd);
838 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
839 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
840 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
841 unsigned int data_len = 24;
844 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
845 if (!rq->special_vec.bv_page)
846 return BLK_STS_RESOURCE;
847 clear_highpage(rq->special_vec.bv_page);
848 rq->special_vec.bv_offset = 0;
849 rq->special_vec.bv_len = data_len;
850 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
853 cmd->cmnd[0] = UNMAP;
856 buf = bvec_virt(&rq->special_vec);
857 put_unaligned_be16(6 + 16, &buf[0]);
858 put_unaligned_be16(16, &buf[2]);
859 put_unaligned_be64(lba, &buf[8]);
860 put_unaligned_be32(nr_blocks, &buf[16]);
862 cmd->allowed = sdkp->max_retries;
863 cmd->transfersize = data_len;
864 rq->timeout = SD_TIMEOUT;
866 return scsi_alloc_sgtables(cmd);
869 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
872 struct scsi_device *sdp = cmd->device;
873 struct request *rq = scsi_cmd_to_rq(cmd);
874 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
875 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
876 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
877 u32 data_len = sdp->sector_size;
879 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
880 if (!rq->special_vec.bv_page)
881 return BLK_STS_RESOURCE;
882 clear_highpage(rq->special_vec.bv_page);
883 rq->special_vec.bv_offset = 0;
884 rq->special_vec.bv_len = data_len;
885 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
888 cmd->cmnd[0] = WRITE_SAME_16;
890 cmd->cmnd[1] = 0x8; /* UNMAP */
891 put_unaligned_be64(lba, &cmd->cmnd[2]);
892 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
894 cmd->allowed = sdkp->max_retries;
895 cmd->transfersize = data_len;
896 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
898 return scsi_alloc_sgtables(cmd);
901 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
904 struct scsi_device *sdp = cmd->device;
905 struct request *rq = scsi_cmd_to_rq(cmd);
906 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
907 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
908 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
909 u32 data_len = sdp->sector_size;
911 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
912 if (!rq->special_vec.bv_page)
913 return BLK_STS_RESOURCE;
914 clear_highpage(rq->special_vec.bv_page);
915 rq->special_vec.bv_offset = 0;
916 rq->special_vec.bv_len = data_len;
917 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
920 cmd->cmnd[0] = WRITE_SAME;
922 cmd->cmnd[1] = 0x8; /* UNMAP */
923 put_unaligned_be32(lba, &cmd->cmnd[2]);
924 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
926 cmd->allowed = sdkp->max_retries;
927 cmd->transfersize = data_len;
928 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
930 return scsi_alloc_sgtables(cmd);
933 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
935 struct request *rq = scsi_cmd_to_rq(cmd);
936 struct scsi_device *sdp = cmd->device;
937 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
938 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
939 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
941 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
942 switch (sdkp->zeroing_mode) {
943 case SD_ZERO_WS16_UNMAP:
944 return sd_setup_write_same16_cmnd(cmd, true);
945 case SD_ZERO_WS10_UNMAP:
946 return sd_setup_write_same10_cmnd(cmd, true);
950 if (sdp->no_write_same) {
951 rq->rq_flags |= RQF_QUIET;
952 return BLK_STS_TARGET;
955 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
956 return sd_setup_write_same16_cmnd(cmd, false);
958 return sd_setup_write_same10_cmnd(cmd, false);
961 static void sd_config_write_same(struct scsi_disk *sdkp)
963 struct request_queue *q = sdkp->disk->queue;
964 unsigned int logical_block_size = sdkp->device->sector_size;
966 if (sdkp->device->no_write_same) {
967 sdkp->max_ws_blocks = 0;
971 /* Some devices can not handle block counts above 0xffff despite
972 * supporting WRITE SAME(16). Consequently we default to 64k
973 * blocks per I/O unless the device explicitly advertises a
976 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
977 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
978 (u32)SD_MAX_WS16_BLOCKS);
979 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
980 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
981 (u32)SD_MAX_WS10_BLOCKS);
983 sdkp->device->no_write_same = 1;
984 sdkp->max_ws_blocks = 0;
987 if (sdkp->lbprz && sdkp->lbpws)
988 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
989 else if (sdkp->lbprz && sdkp->lbpws10)
990 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
991 else if (sdkp->max_ws_blocks)
992 sdkp->zeroing_mode = SD_ZERO_WS;
994 sdkp->zeroing_mode = SD_ZERO_WRITE;
996 if (sdkp->max_ws_blocks &&
997 sdkp->physical_block_size > logical_block_size) {
999 * Reporting a maximum number of blocks that is not aligned
1000 * on the device physical size would cause a large write same
1001 * request to be split into physically unaligned chunks by
1002 * __blkdev_issue_write_zeroes() even if the caller of this
1003 * functions took care to align the large request. So make sure
1004 * the maximum reported is aligned to the device physical block
1005 * size. This is only an optional optimization for regular
1006 * disks, but this is mandatory to avoid failure of large write
1007 * same requests directed at sequential write required zones of
1008 * host-managed ZBC disks.
1010 sdkp->max_ws_blocks =
1011 round_down(sdkp->max_ws_blocks,
1012 bytes_to_logical(sdkp->device,
1013 sdkp->physical_block_size));
1017 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1018 (logical_block_size >> 9));
1021 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1023 struct request *rq = scsi_cmd_to_rq(cmd);
1024 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1026 /* flush requests don't perform I/O, zero the S/G table */
1027 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1029 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1031 cmd->transfersize = 0;
1032 cmd->allowed = sdkp->max_retries;
1034 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1038 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1039 sector_t lba, unsigned int nr_blocks,
1040 unsigned char flags)
1042 cmd->cmd_len = SD_EXT_CDB_SIZE;
1043 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1044 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1045 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1046 cmd->cmnd[10] = flags;
1047 put_unaligned_be64(lba, &cmd->cmnd[12]);
1048 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1049 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1054 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1055 sector_t lba, unsigned int nr_blocks,
1056 unsigned char flags)
1059 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1060 cmd->cmnd[1] = flags;
1063 put_unaligned_be64(lba, &cmd->cmnd[2]);
1064 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1069 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1070 sector_t lba, unsigned int nr_blocks,
1071 unsigned char flags)
1074 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1075 cmd->cmnd[1] = flags;
1078 put_unaligned_be32(lba, &cmd->cmnd[2]);
1079 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1084 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1085 sector_t lba, unsigned int nr_blocks,
1086 unsigned char flags)
1088 /* Avoid that 0 blocks gets translated into 256 blocks. */
1089 if (WARN_ON_ONCE(nr_blocks == 0))
1090 return BLK_STS_IOERR;
1092 if (unlikely(flags & 0x8)) {
1094 * This happens only if this drive failed 10byte rw
1095 * command with ILLEGAL_REQUEST during operation and
1096 * thus turned off use_10_for_rw.
1098 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1099 return BLK_STS_IOERR;
1103 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1104 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1105 cmd->cmnd[2] = (lba >> 8) & 0xff;
1106 cmd->cmnd[3] = lba & 0xff;
1107 cmd->cmnd[4] = nr_blocks;
1113 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1115 struct request *rq = scsi_cmd_to_rq(cmd);
1116 struct scsi_device *sdp = cmd->device;
1117 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1118 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1120 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1121 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1122 bool write = rq_data_dir(rq) == WRITE;
1123 unsigned char protect, fua;
1128 ret = scsi_alloc_sgtables(cmd);
1129 if (ret != BLK_STS_OK)
1132 ret = BLK_STS_IOERR;
1133 if (!scsi_device_online(sdp) || sdp->changed) {
1134 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1138 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1139 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1143 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1144 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1149 * Some SD card readers can't handle accesses which touch the
1150 * last one or two logical blocks. Split accesses as needed.
1152 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1154 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1155 if (lba < threshold) {
1156 /* Access up to the threshold but not beyond */
1157 nr_blocks = threshold - lba;
1159 /* Access only a single logical block */
1164 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1165 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1170 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1171 dix = scsi_prot_sg_count(cmd);
1172 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1175 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1179 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1180 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1182 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1183 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1185 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1186 sdp->use_10_for_rw || protect) {
1187 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1190 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1194 if (unlikely(ret != BLK_STS_OK))
1198 * We shouldn't disconnect in the middle of a sector, so with a dumb
1199 * host adapter, it's safe to assume that we can at least transfer
1200 * this many bytes between each connect / disconnect.
1202 cmd->transfersize = sdp->sector_size;
1203 cmd->underflow = nr_blocks << 9;
1204 cmd->allowed = sdkp->max_retries;
1205 cmd->sdb.length = nr_blocks * sdp->sector_size;
1208 scmd_printk(KERN_INFO, cmd,
1209 "%s: block=%llu, count=%d\n", __func__,
1210 (unsigned long long)blk_rq_pos(rq),
1211 blk_rq_sectors(rq)));
1213 scmd_printk(KERN_INFO, cmd,
1214 "%s %d/%u 512 byte blocks.\n",
1215 write ? "writing" : "reading", nr_blocks,
1216 blk_rq_sectors(rq)));
1219 * This indicates that the command is ready from our end to be queued.
1223 scsi_free_sgtables(cmd);
1227 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1229 struct request *rq = scsi_cmd_to_rq(cmd);
1231 switch (req_op(rq)) {
1232 case REQ_OP_DISCARD:
1233 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1235 return sd_setup_unmap_cmnd(cmd);
1237 return sd_setup_write_same16_cmnd(cmd, true);
1239 return sd_setup_write_same10_cmnd(cmd, true);
1241 return sd_setup_write_same10_cmnd(cmd, false);
1243 return BLK_STS_TARGET;
1245 case REQ_OP_WRITE_ZEROES:
1246 return sd_setup_write_zeroes_cmnd(cmd);
1248 return sd_setup_flush_cmnd(cmd);
1251 case REQ_OP_ZONE_APPEND:
1252 return sd_setup_read_write_cmnd(cmd);
1253 case REQ_OP_ZONE_RESET:
1254 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1256 case REQ_OP_ZONE_RESET_ALL:
1257 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1259 case REQ_OP_ZONE_OPEN:
1260 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1261 case REQ_OP_ZONE_CLOSE:
1262 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1263 case REQ_OP_ZONE_FINISH:
1264 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1267 return BLK_STS_NOTSUPP;
1271 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1273 struct request *rq = scsi_cmd_to_rq(SCpnt);
1275 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1276 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1279 static bool sd_need_revalidate(struct block_device *bdev,
1280 struct scsi_disk *sdkp)
1282 if (sdkp->device->removable || sdkp->write_prot) {
1283 if (bdev_check_media_change(bdev))
1288 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1289 * nothing to do with partitions, BLKRRPART is used to force a full
1290 * revalidate after things like a format for historical reasons.
1292 return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1296 * sd_open - open a scsi disk device
1297 * @bdev: Block device of the scsi disk to open
1298 * @mode: FMODE_* mask
1300 * Returns 0 if successful. Returns a negated errno value in case
1303 * Note: This can be called from a user context (e.g. fsck(1) )
1304 * or from within the kernel (e.g. as a result of a mount(1) ).
1305 * In the latter case @inode and @filp carry an abridged amount
1306 * of information as noted above.
1308 * Locking: called with bdev->bd_disk->open_mutex held.
1310 static int sd_open(struct block_device *bdev, fmode_t mode)
1312 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1313 struct scsi_device *sdev = sdkp->device;
1316 if (scsi_device_get(sdev))
1319 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1322 * If the device is in error recovery, wait until it is done.
1323 * If the device is offline, then disallow any access to it.
1326 if (!scsi_block_when_processing_errors(sdev))
1329 if (sd_need_revalidate(bdev, sdkp))
1330 sd_revalidate_disk(bdev->bd_disk);
1333 * If the drive is empty, just let the open fail.
1335 retval = -ENOMEDIUM;
1336 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1340 * If the device has the write protect tab set, have the open fail
1341 * if the user expects to be able to write to the thing.
1344 if (sdkp->write_prot && (mode & FMODE_WRITE))
1348 * It is possible that the disk changing stuff resulted in
1349 * the device being taken offline. If this is the case,
1350 * report this to the user, and don't pretend that the
1351 * open actually succeeded.
1354 if (!scsi_device_online(sdev))
1357 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1358 if (scsi_block_when_processing_errors(sdev))
1359 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1365 scsi_device_put(sdev);
1370 * sd_release - invoked when the (last) close(2) is called on this
1372 * @disk: disk to release
1373 * @mode: FMODE_* mask
1377 * Note: may block (uninterruptible) if error recovery is underway
1380 * Locking: called with bdev->bd_disk->open_mutex held.
1382 static void sd_release(struct gendisk *disk, fmode_t mode)
1384 struct scsi_disk *sdkp = scsi_disk(disk);
1385 struct scsi_device *sdev = sdkp->device;
1387 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1389 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1390 if (scsi_block_when_processing_errors(sdev))
1391 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1394 scsi_device_put(sdev);
1397 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1399 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1400 struct scsi_device *sdp = sdkp->device;
1401 struct Scsi_Host *host = sdp->host;
1402 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1405 /* default to most commonly used values */
1406 diskinfo[0] = 0x40; /* 1 << 6 */
1407 diskinfo[1] = 0x20; /* 1 << 5 */
1408 diskinfo[2] = capacity >> 11;
1410 /* override with calculated, extended default, or driver values */
1411 if (host->hostt->bios_param)
1412 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1414 scsicam_bios_param(bdev, capacity, diskinfo);
1416 geo->heads = diskinfo[0];
1417 geo->sectors = diskinfo[1];
1418 geo->cylinders = diskinfo[2];
1423 * sd_ioctl - process an ioctl
1424 * @bdev: target block device
1425 * @mode: FMODE_* mask
1426 * @cmd: ioctl command number
1427 * @arg: this is third argument given to ioctl(2) system call.
1428 * Often contains a pointer.
1430 * Returns 0 if successful (some ioctls return positive numbers on
1431 * success as well). Returns a negated errno value in case of error.
1433 * Note: most ioctls are forward onto the block subsystem or further
1434 * down in the scsi subsystem.
1436 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1437 unsigned int cmd, unsigned long arg)
1439 struct gendisk *disk = bdev->bd_disk;
1440 struct scsi_disk *sdkp = scsi_disk(disk);
1441 struct scsi_device *sdp = sdkp->device;
1442 void __user *p = (void __user *)arg;
1445 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1446 "cmd=0x%x\n", disk->disk_name, cmd));
1448 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1449 return -ENOIOCTLCMD;
1452 * If we are in the middle of error recovery, don't let anyone
1453 * else try and use this device. Also, if error recovery fails, it
1454 * may try and take the device offline, in which case all further
1455 * access to the device is prohibited.
1457 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1458 (mode & FMODE_NDELAY) != 0);
1462 if (is_sed_ioctl(cmd))
1463 return sed_ioctl(sdkp->opal_dev, cmd, p);
1464 return scsi_ioctl(sdp, mode, cmd, p);
1467 static void set_media_not_present(struct scsi_disk *sdkp)
1469 if (sdkp->media_present)
1470 sdkp->device->changed = 1;
1472 if (sdkp->device->removable) {
1473 sdkp->media_present = 0;
1478 static int media_not_present(struct scsi_disk *sdkp,
1479 struct scsi_sense_hdr *sshdr)
1481 if (!scsi_sense_valid(sshdr))
1484 /* not invoked for commands that could return deferred errors */
1485 switch (sshdr->sense_key) {
1486 case UNIT_ATTENTION:
1488 /* medium not present */
1489 if (sshdr->asc == 0x3A) {
1490 set_media_not_present(sdkp);
1498 * sd_check_events - check media events
1499 * @disk: kernel device descriptor
1500 * @clearing: disk events currently being cleared
1502 * Returns mask of DISK_EVENT_*.
1504 * Note: this function is invoked from the block subsystem.
1506 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1508 struct scsi_disk *sdkp = disk->private_data;
1509 struct scsi_device *sdp;
1517 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1520 * If the device is offline, don't send any commands - just pretend as
1521 * if the command failed. If the device ever comes back online, we
1522 * can deal with it then. It is only because of unrecoverable errors
1523 * that we would ever take a device offline in the first place.
1525 if (!scsi_device_online(sdp)) {
1526 set_media_not_present(sdkp);
1531 * Using TEST_UNIT_READY enables differentiation between drive with
1532 * no cartridge loaded - NOT READY, drive with changed cartridge -
1533 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1535 * Drives that auto spin down. eg iomega jaz 1G, will be started
1536 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1537 * sd_revalidate() is called.
1539 if (scsi_block_when_processing_errors(sdp)) {
1540 struct scsi_sense_hdr sshdr = { 0, };
1542 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1545 /* failed to execute TUR, assume media not present */
1546 if (retval < 0 || host_byte(retval)) {
1547 set_media_not_present(sdkp);
1551 if (media_not_present(sdkp, &sshdr))
1556 * For removable scsi disk we have to recognise the presence
1557 * of a disk in the drive.
1559 if (!sdkp->media_present)
1561 sdkp->media_present = 1;
1564 * sdp->changed is set under the following conditions:
1566 * Medium present state has changed in either direction.
1567 * Device has indicated UNIT_ATTENTION.
1569 disk_changed = sdp->changed;
1571 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1574 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1577 struct scsi_device *sdp = sdkp->device;
1578 const int timeout = sdp->request_queue->rq_timeout
1579 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1580 struct scsi_sense_hdr my_sshdr;
1582 if (!scsi_device_online(sdp))
1585 /* caller might not be interested in sense, but we need it */
1589 for (retries = 3; retries > 0; --retries) {
1590 unsigned char cmd[10] = { 0 };
1592 cmd[0] = SYNCHRONIZE_CACHE;
1594 * Leave the rest of the command zero to indicate
1597 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1598 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1604 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1609 if (scsi_status_is_check_condition(res) &&
1610 scsi_sense_valid(sshdr)) {
1611 sd_print_sense_hdr(sdkp, sshdr);
1613 /* we need to evaluate the error return */
1614 if (sshdr->asc == 0x3a || /* medium not present */
1615 sshdr->asc == 0x20 || /* invalid command */
1616 (sshdr->asc == 0x74 && sshdr->ascq == 0x71)) /* drive is password locked */
1617 /* this is no error here */
1621 switch (host_byte(res)) {
1622 /* ignore errors due to racing a disconnection */
1623 case DID_BAD_TARGET:
1624 case DID_NO_CONNECT:
1626 /* signal the upper layer it might try again */
1630 case DID_SOFT_ERROR:
1639 static void sd_rescan(struct device *dev)
1641 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1643 sd_revalidate_disk(sdkp->disk);
1646 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1647 enum blk_unique_id type)
1649 struct scsi_device *sdev = scsi_disk(disk)->device;
1650 const struct scsi_vpd *vpd;
1651 const unsigned char *d;
1652 int ret = -ENXIO, len;
1655 vpd = rcu_dereference(sdev->vpd_pg83);
1660 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1661 /* we only care about designators with LU association */
1662 if (((d[1] >> 4) & 0x3) != 0x00)
1664 if ((d[1] & 0xf) != type)
1668 * Only exit early if a 16-byte descriptor was found. Otherwise
1669 * keep looking as one with more entropy might still show up.
1672 if (len != 8 && len != 12 && len != 16)
1675 memcpy(id, d + 4, len);
1684 static char sd_pr_type(enum pr_type type)
1687 case PR_WRITE_EXCLUSIVE:
1689 case PR_EXCLUSIVE_ACCESS:
1691 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1693 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1695 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1697 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1704 static int sd_pr_command(struct block_device *bdev, u8 sa,
1705 u64 key, u64 sa_key, u8 type, u8 flags)
1707 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1708 struct scsi_device *sdev = sdkp->device;
1709 struct scsi_sense_hdr sshdr;
1711 u8 cmd[16] = { 0, };
1712 u8 data[24] = { 0, };
1714 cmd[0] = PERSISTENT_RESERVE_OUT;
1717 put_unaligned_be32(sizeof(data), &cmd[5]);
1719 put_unaligned_be64(key, &data[0]);
1720 put_unaligned_be64(sa_key, &data[8]);
1723 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1724 &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1726 if (scsi_status_is_check_condition(result) &&
1727 scsi_sense_valid(&sshdr)) {
1728 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1729 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1735 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1738 if (flags & ~PR_FL_IGNORE_KEY)
1740 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1741 old_key, new_key, 0,
1742 (1 << 0) /* APTPL */);
1745 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1750 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1753 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1755 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1758 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1759 enum pr_type type, bool abort)
1761 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1762 sd_pr_type(type), 0);
1765 static int sd_pr_clear(struct block_device *bdev, u64 key)
1767 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1770 static const struct pr_ops sd_pr_ops = {
1771 .pr_register = sd_pr_register,
1772 .pr_reserve = sd_pr_reserve,
1773 .pr_release = sd_pr_release,
1774 .pr_preempt = sd_pr_preempt,
1775 .pr_clear = sd_pr_clear,
1778 static void scsi_disk_free_disk(struct gendisk *disk)
1780 struct scsi_disk *sdkp = scsi_disk(disk);
1782 put_device(&sdkp->disk_dev);
1785 static const struct block_device_operations sd_fops = {
1786 .owner = THIS_MODULE,
1788 .release = sd_release,
1790 .getgeo = sd_getgeo,
1791 .compat_ioctl = blkdev_compat_ptr_ioctl,
1792 .check_events = sd_check_events,
1793 .unlock_native_capacity = sd_unlock_native_capacity,
1794 .report_zones = sd_zbc_report_zones,
1795 .get_unique_id = sd_get_unique_id,
1796 .free_disk = scsi_disk_free_disk,
1797 .pr_ops = &sd_pr_ops,
1801 * sd_eh_reset - reset error handling callback
1802 * @scmd: sd-issued command that has failed
1804 * This function is called by the SCSI midlayer before starting
1805 * SCSI EH. When counting medium access failures we have to be
1806 * careful to register it only only once per device and SCSI EH run;
1807 * there might be several timed out commands which will cause the
1808 * 'max_medium_access_timeouts' counter to trigger after the first
1809 * SCSI EH run already and set the device to offline.
1810 * So this function resets the internal counter before starting SCSI EH.
1812 static void sd_eh_reset(struct scsi_cmnd *scmd)
1814 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1816 /* New SCSI EH run, reset gate variable */
1817 sdkp->ignore_medium_access_errors = false;
1821 * sd_eh_action - error handling callback
1822 * @scmd: sd-issued command that has failed
1823 * @eh_disp: The recovery disposition suggested by the midlayer
1825 * This function is called by the SCSI midlayer upon completion of an
1826 * error test command (currently TEST UNIT READY). The result of sending
1827 * the eh command is passed in eh_disp. We're looking for devices that
1828 * fail medium access commands but are OK with non access commands like
1829 * test unit ready (so wrongly see the device as having a successful
1832 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1834 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1835 struct scsi_device *sdev = scmd->device;
1837 if (!scsi_device_online(sdev) ||
1838 !scsi_medium_access_command(scmd) ||
1839 host_byte(scmd->result) != DID_TIME_OUT ||
1844 * The device has timed out executing a medium access command.
1845 * However, the TEST UNIT READY command sent during error
1846 * handling completed successfully. Either the device is in the
1847 * process of recovering or has it suffered an internal failure
1848 * that prevents access to the storage medium.
1850 if (!sdkp->ignore_medium_access_errors) {
1851 sdkp->medium_access_timed_out++;
1852 sdkp->ignore_medium_access_errors = true;
1856 * If the device keeps failing read/write commands but TEST UNIT
1857 * READY always completes successfully we assume that medium
1858 * access is no longer possible and take the device offline.
1860 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1861 scmd_printk(KERN_ERR, scmd,
1862 "Medium access timeout failure. Offlining disk!\n");
1863 mutex_lock(&sdev->state_mutex);
1864 scsi_device_set_state(sdev, SDEV_OFFLINE);
1865 mutex_unlock(&sdev->state_mutex);
1873 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1875 struct request *req = scsi_cmd_to_rq(scmd);
1876 struct scsi_device *sdev = scmd->device;
1877 unsigned int transferred, good_bytes;
1878 u64 start_lba, end_lba, bad_lba;
1881 * Some commands have a payload smaller than the device logical
1882 * block size (e.g. INQUIRY on a 4K disk).
1884 if (scsi_bufflen(scmd) <= sdev->sector_size)
1887 /* Check if we have a 'bad_lba' information */
1888 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1889 SCSI_SENSE_BUFFERSIZE,
1894 * If the bad lba was reported incorrectly, we have no idea where
1897 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1898 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1899 if (bad_lba < start_lba || bad_lba >= end_lba)
1903 * resid is optional but mostly filled in. When it's unused,
1904 * its value is zero, so we assume the whole buffer transferred
1906 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1908 /* This computation should always be done in terms of the
1909 * resolution of the device's medium.
1911 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1913 return min(good_bytes, transferred);
1917 * sd_done - bottom half handler: called when the lower level
1918 * driver has completed (successfully or otherwise) a scsi command.
1919 * @SCpnt: mid-level's per command structure.
1921 * Note: potentially run from within an ISR. Must not block.
1923 static int sd_done(struct scsi_cmnd *SCpnt)
1925 int result = SCpnt->result;
1926 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1927 unsigned int sector_size = SCpnt->device->sector_size;
1929 struct scsi_sense_hdr sshdr;
1930 struct request *req = scsi_cmd_to_rq(SCpnt);
1931 struct scsi_disk *sdkp = scsi_disk(req->q->disk);
1932 int sense_valid = 0;
1933 int sense_deferred = 0;
1935 switch (req_op(req)) {
1936 case REQ_OP_DISCARD:
1937 case REQ_OP_WRITE_ZEROES:
1938 case REQ_OP_ZONE_RESET:
1939 case REQ_OP_ZONE_RESET_ALL:
1940 case REQ_OP_ZONE_OPEN:
1941 case REQ_OP_ZONE_CLOSE:
1942 case REQ_OP_ZONE_FINISH:
1944 good_bytes = blk_rq_bytes(req);
1945 scsi_set_resid(SCpnt, 0);
1948 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1953 * In case of bogus fw or device, we could end up having
1954 * an unaligned partial completion. Check this here and force
1957 resid = scsi_get_resid(SCpnt);
1958 if (resid & (sector_size - 1)) {
1959 sd_printk(KERN_INFO, sdkp,
1960 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1961 resid, sector_size);
1962 scsi_print_command(SCpnt);
1963 resid = min(scsi_bufflen(SCpnt),
1964 round_up(resid, sector_size));
1965 scsi_set_resid(SCpnt, resid);
1970 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1972 sense_deferred = scsi_sense_is_deferred(&sshdr);
1974 sdkp->medium_access_timed_out = 0;
1976 if (!scsi_status_is_check_condition(result) &&
1977 (!sense_valid || sense_deferred))
1980 switch (sshdr.sense_key) {
1981 case HARDWARE_ERROR:
1983 good_bytes = sd_completed_bytes(SCpnt);
1985 case RECOVERED_ERROR:
1986 good_bytes = scsi_bufflen(SCpnt);
1989 /* This indicates a false check condition, so ignore it. An
1990 * unknown amount of data was transferred so treat it as an
1994 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1996 case ABORTED_COMMAND:
1997 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1998 good_bytes = sd_completed_bytes(SCpnt);
2000 case ILLEGAL_REQUEST:
2001 switch (sshdr.asc) {
2002 case 0x10: /* DIX: Host detected corruption */
2003 good_bytes = sd_completed_bytes(SCpnt);
2005 case 0x20: /* INVALID COMMAND OPCODE */
2006 case 0x24: /* INVALID FIELD IN CDB */
2007 switch (SCpnt->cmnd[0]) {
2009 sd_config_discard(sdkp, SD_LBP_DISABLE);
2013 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2014 sd_config_discard(sdkp, SD_LBP_DISABLE);
2016 sdkp->device->no_write_same = 1;
2017 sd_config_write_same(sdkp);
2018 req->rq_flags |= RQF_QUIET;
2029 if (sd_is_zoned(sdkp))
2030 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2032 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2033 "sd_done: completed %d of %d bytes\n",
2034 good_bytes, scsi_bufflen(SCpnt)));
2040 * spinup disk - called only in sd_revalidate_disk()
2043 sd_spinup_disk(struct scsi_disk *sdkp)
2045 unsigned char cmd[10];
2046 unsigned long spintime_expire = 0;
2047 int retries, spintime;
2048 unsigned int the_result;
2049 struct scsi_sense_hdr sshdr;
2050 int sense_valid = 0;
2054 /* Spin up drives, as required. Only do this at boot time */
2055 /* Spinup needs to be done for module loads too. */
2060 bool media_was_present = sdkp->media_present;
2062 cmd[0] = TEST_UNIT_READY;
2063 memset((void *) &cmd[1], 0, 9);
2065 the_result = scsi_execute_req(sdkp->device, cmd,
2068 sdkp->max_retries, NULL);
2071 * If the drive has indicated to us that it
2072 * doesn't have any media in it, don't bother
2073 * with any more polling.
2075 if (media_not_present(sdkp, &sshdr)) {
2076 if (media_was_present)
2077 sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2082 sense_valid = scsi_sense_valid(&sshdr);
2084 } while (retries < 3 &&
2085 (!scsi_status_is_good(the_result) ||
2086 (scsi_status_is_check_condition(the_result) &&
2087 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2089 if (!scsi_status_is_check_condition(the_result)) {
2090 /* no sense, TUR either succeeded or failed
2091 * with a status error */
2092 if(!spintime && !scsi_status_is_good(the_result)) {
2093 sd_print_result(sdkp, "Test Unit Ready failed",
2100 * The device does not want the automatic start to be issued.
2102 if (sdkp->device->no_start_on_add)
2105 if (sense_valid && sshdr.sense_key == NOT_READY) {
2106 if (sshdr.asc == 4 && sshdr.ascq == 3)
2107 break; /* manual intervention required */
2108 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2109 break; /* standby */
2110 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2111 break; /* unavailable */
2112 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2113 break; /* sanitize in progress */
2115 * Issue command to spin up drive when not ready
2118 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2119 cmd[0] = START_STOP;
2120 cmd[1] = 1; /* Return immediately */
2121 memset((void *) &cmd[2], 0, 8);
2122 cmd[4] = 1; /* Start spin cycle */
2123 if (sdkp->device->start_stop_pwr_cond)
2125 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2127 SD_TIMEOUT, sdkp->max_retries,
2129 spintime_expire = jiffies + 100 * HZ;
2132 /* Wait 1 second for next try */
2134 printk(KERN_CONT ".");
2137 * Wait for USB flash devices with slow firmware.
2138 * Yes, this sense key/ASC combination shouldn't
2139 * occur here. It's characteristic of these devices.
2141 } else if (sense_valid &&
2142 sshdr.sense_key == UNIT_ATTENTION &&
2143 sshdr.asc == 0x28) {
2145 spintime_expire = jiffies + 5 * HZ;
2148 /* Wait 1 second for next try */
2151 /* we don't understand the sense code, so it's
2152 * probably pointless to loop */
2154 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2155 sd_print_sense_hdr(sdkp, &sshdr);
2160 } while (spintime && time_before_eq(jiffies, spintime_expire));
2163 if (scsi_status_is_good(the_result))
2164 printk(KERN_CONT "ready\n");
2166 printk(KERN_CONT "not responding...\n");
2171 * Determine whether disk supports Data Integrity Field.
2173 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2175 struct scsi_device *sdp = sdkp->device;
2178 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2179 sdkp->protection_type = 0;
2183 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2185 if (type > T10_PI_TYPE3_PROTECTION) {
2186 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2187 " protection type %u. Disabling disk!\n",
2189 sdkp->protection_type = 0;
2193 sdkp->protection_type = type;
2198 static void sd_config_protection(struct scsi_disk *sdkp)
2200 struct scsi_device *sdp = sdkp->device;
2202 if (!sdkp->first_scan)
2205 sd_dif_config_host(sdkp);
2207 if (!sdkp->protection_type)
2210 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2211 sd_printk(KERN_NOTICE, sdkp,
2212 "Disabling DIF Type %u protection\n",
2213 sdkp->protection_type);
2214 sdkp->protection_type = 0;
2217 sd_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2218 sdkp->protection_type);
2221 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2222 struct scsi_sense_hdr *sshdr, int sense_valid,
2226 sd_print_sense_hdr(sdkp, sshdr);
2228 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2231 * Set dirty bit for removable devices if not ready -
2232 * sometimes drives will not report this properly.
2234 if (sdp->removable &&
2235 sense_valid && sshdr->sense_key == NOT_READY)
2236 set_media_not_present(sdkp);
2239 * We used to set media_present to 0 here to indicate no media
2240 * in the drive, but some drives fail read capacity even with
2241 * media present, so we can't do that.
2243 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2247 #if RC16_LEN > SD_BUF_SIZE
2248 #error RC16_LEN must not be more than SD_BUF_SIZE
2251 #define READ_CAPACITY_RETRIES_ON_RESET 10
2253 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2254 unsigned char *buffer)
2256 unsigned char cmd[16];
2257 struct scsi_sense_hdr sshdr;
2258 int sense_valid = 0;
2260 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2261 unsigned int alignment;
2262 unsigned long long lba;
2263 unsigned sector_size;
2265 if (sdp->no_read_capacity_16)
2270 cmd[0] = SERVICE_ACTION_IN_16;
2271 cmd[1] = SAI_READ_CAPACITY_16;
2273 memset(buffer, 0, RC16_LEN);
2275 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2276 buffer, RC16_LEN, &sshdr,
2277 SD_TIMEOUT, sdkp->max_retries, NULL);
2279 if (media_not_present(sdkp, &sshdr))
2282 if (the_result > 0) {
2283 sense_valid = scsi_sense_valid(&sshdr);
2285 sshdr.sense_key == ILLEGAL_REQUEST &&
2286 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2288 /* Invalid Command Operation Code or
2289 * Invalid Field in CDB, just retry
2290 * silently with RC10 */
2293 sshdr.sense_key == UNIT_ATTENTION &&
2294 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2295 /* Device reset might occur several times,
2296 * give it one more chance */
2297 if (--reset_retries > 0)
2302 } while (the_result && retries);
2305 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2306 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2310 sector_size = get_unaligned_be32(&buffer[8]);
2311 lba = get_unaligned_be64(&buffer[0]);
2313 if (sd_read_protection_type(sdkp, buffer) < 0) {
2318 /* Logical blocks per physical block exponent */
2319 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2322 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2324 /* Lowest aligned logical block */
2325 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2326 blk_queue_alignment_offset(sdp->request_queue, alignment);
2327 if (alignment && sdkp->first_scan)
2328 sd_printk(KERN_NOTICE, sdkp,
2329 "physical block alignment offset: %u\n", alignment);
2331 if (buffer[14] & 0x80) { /* LBPME */
2334 if (buffer[14] & 0x40) /* LBPRZ */
2337 sd_config_discard(sdkp, SD_LBP_WS16);
2340 sdkp->capacity = lba + 1;
2344 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2345 unsigned char *buffer)
2347 unsigned char cmd[16];
2348 struct scsi_sense_hdr sshdr;
2349 int sense_valid = 0;
2351 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2353 unsigned sector_size;
2356 cmd[0] = READ_CAPACITY;
2357 memset(&cmd[1], 0, 9);
2358 memset(buffer, 0, 8);
2360 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2362 SD_TIMEOUT, sdkp->max_retries, NULL);
2364 if (media_not_present(sdkp, &sshdr))
2367 if (the_result > 0) {
2368 sense_valid = scsi_sense_valid(&sshdr);
2370 sshdr.sense_key == UNIT_ATTENTION &&
2371 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2372 /* Device reset might occur several times,
2373 * give it one more chance */
2374 if (--reset_retries > 0)
2379 } while (the_result && retries);
2382 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2383 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2387 sector_size = get_unaligned_be32(&buffer[4]);
2388 lba = get_unaligned_be32(&buffer[0]);
2390 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2391 /* Some buggy (usb cardreader) devices return an lba of
2392 0xffffffff when the want to report a size of 0 (with
2393 which they really mean no media is present) */
2395 sdkp->physical_block_size = sector_size;
2399 sdkp->capacity = lba + 1;
2400 sdkp->physical_block_size = sector_size;
2404 static int sd_try_rc16_first(struct scsi_device *sdp)
2406 if (sdp->host->max_cmd_len < 16)
2408 if (sdp->try_rc_10_first)
2410 if (sdp->scsi_level > SCSI_SPC_2)
2412 if (scsi_device_protection(sdp))
2418 * read disk capacity
2421 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2424 struct scsi_device *sdp = sdkp->device;
2426 if (sd_try_rc16_first(sdp)) {
2427 sector_size = read_capacity_16(sdkp, sdp, buffer);
2428 if (sector_size == -EOVERFLOW)
2430 if (sector_size == -ENODEV)
2432 if (sector_size < 0)
2433 sector_size = read_capacity_10(sdkp, sdp, buffer);
2434 if (sector_size < 0)
2437 sector_size = read_capacity_10(sdkp, sdp, buffer);
2438 if (sector_size == -EOVERFLOW)
2440 if (sector_size < 0)
2442 if ((sizeof(sdkp->capacity) > 4) &&
2443 (sdkp->capacity > 0xffffffffULL)) {
2444 int old_sector_size = sector_size;
2445 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2446 "Trying to use READ CAPACITY(16).\n");
2447 sector_size = read_capacity_16(sdkp, sdp, buffer);
2448 if (sector_size < 0) {
2449 sd_printk(KERN_NOTICE, sdkp,
2450 "Using 0xffffffff as device size\n");
2451 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2452 sector_size = old_sector_size;
2455 /* Remember that READ CAPACITY(16) succeeded */
2456 sdp->try_rc_10_first = 0;
2460 /* Some devices are known to return the total number of blocks,
2461 * not the highest block number. Some devices have versions
2462 * which do this and others which do not. Some devices we might
2463 * suspect of doing this but we don't know for certain.
2465 * If we know the reported capacity is wrong, decrement it. If
2466 * we can only guess, then assume the number of blocks is even
2467 * (usually true but not always) and err on the side of lowering
2470 if (sdp->fix_capacity ||
2471 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2472 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2473 "from its reported value: %llu\n",
2474 (unsigned long long) sdkp->capacity);
2479 if (sector_size == 0) {
2481 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2485 if (sector_size != 512 &&
2486 sector_size != 1024 &&
2487 sector_size != 2048 &&
2488 sector_size != 4096) {
2489 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2492 * The user might want to re-format the drive with
2493 * a supported sectorsize. Once this happens, it
2494 * would be relatively trivial to set the thing up.
2495 * For this reason, we leave the thing in the table.
2499 * set a bogus sector size so the normal read/write
2500 * logic in the block layer will eventually refuse any
2501 * request on this device without tripping over power
2502 * of two sector size assumptions
2506 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2507 blk_queue_physical_block_size(sdp->request_queue,
2508 sdkp->physical_block_size);
2509 sdkp->device->sector_size = sector_size;
2511 if (sdkp->capacity > 0xffffffff)
2512 sdp->use_16_for_rw = 1;
2517 * Print disk capacity
2520 sd_print_capacity(struct scsi_disk *sdkp,
2521 sector_t old_capacity)
2523 int sector_size = sdkp->device->sector_size;
2524 char cap_str_2[10], cap_str_10[10];
2526 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2529 string_get_size(sdkp->capacity, sector_size,
2530 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2531 string_get_size(sdkp->capacity, sector_size,
2532 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2534 sd_printk(KERN_NOTICE, sdkp,
2535 "%llu %d-byte logical blocks: (%s/%s)\n",
2536 (unsigned long long)sdkp->capacity,
2537 sector_size, cap_str_10, cap_str_2);
2539 if (sdkp->physical_block_size != sector_size)
2540 sd_printk(KERN_NOTICE, sdkp,
2541 "%u-byte physical blocks\n",
2542 sdkp->physical_block_size);
2545 /* called with buffer of length 512 */
2547 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2548 unsigned char *buffer, int len, struct scsi_mode_data *data,
2549 struct scsi_sense_hdr *sshdr)
2552 * If we must use MODE SENSE(10), make sure that the buffer length
2553 * is at least 8 bytes so that the mode sense header fits.
2555 if (sdkp->device->use_10_for_ms && len < 8)
2558 return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2559 SD_TIMEOUT, sdkp->max_retries, data,
2564 * read write protect setting, if possible - called only in sd_revalidate_disk()
2565 * called with buffer of length SD_BUF_SIZE
2568 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2571 struct scsi_device *sdp = sdkp->device;
2572 struct scsi_mode_data data;
2573 int old_wp = sdkp->write_prot;
2575 set_disk_ro(sdkp->disk, 0);
2576 if (sdp->skip_ms_page_3f) {
2577 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2581 if (sdp->use_192_bytes_for_3f) {
2582 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2585 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2586 * We have to start carefully: some devices hang if we ask
2587 * for more than is available.
2589 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2592 * Second attempt: ask for page 0 When only page 0 is
2593 * implemented, a request for page 3F may return Sense Key
2594 * 5: Illegal Request, Sense Code 24: Invalid field in
2598 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2601 * Third attempt: ask 255 bytes, as we did earlier.
2604 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2609 sd_first_printk(KERN_WARNING, sdkp,
2610 "Test WP failed, assume Write Enabled\n");
2612 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2613 set_disk_ro(sdkp->disk, sdkp->write_prot);
2614 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2615 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2616 sdkp->write_prot ? "on" : "off");
2617 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2623 * sd_read_cache_type - called only from sd_revalidate_disk()
2624 * called with buffer of length SD_BUF_SIZE
2627 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2630 struct scsi_device *sdp = sdkp->device;
2635 struct scsi_mode_data data;
2636 struct scsi_sense_hdr sshdr;
2637 int old_wce = sdkp->WCE;
2638 int old_rcd = sdkp->RCD;
2639 int old_dpofua = sdkp->DPOFUA;
2642 if (sdkp->cache_override)
2646 if (sdp->skip_ms_page_8) {
2647 if (sdp->type == TYPE_RBC)
2650 if (sdp->skip_ms_page_3f)
2653 if (sdp->use_192_bytes_for_3f)
2657 } else if (sdp->type == TYPE_RBC) {
2665 /* cautiously ask */
2666 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2672 if (!data.header_length) {
2675 sd_first_printk(KERN_ERR, sdkp,
2676 "Missing header in MODE_SENSE response\n");
2679 /* that went OK, now ask for the proper length */
2683 * We're only interested in the first three bytes, actually.
2684 * But the data cache page is defined for the first 20.
2688 else if (len > SD_BUF_SIZE) {
2689 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2690 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2693 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2697 if (len > first_len)
2698 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2702 int offset = data.header_length + data.block_descriptor_length;
2704 while (offset < len) {
2705 u8 page_code = buffer[offset] & 0x3F;
2706 u8 spf = buffer[offset] & 0x40;
2708 if (page_code == 8 || page_code == 6) {
2709 /* We're interested only in the first 3 bytes.
2711 if (len - offset <= 2) {
2712 sd_first_printk(KERN_ERR, sdkp,
2713 "Incomplete mode parameter "
2717 modepage = page_code;
2721 /* Go to the next page */
2722 if (spf && len - offset > 3)
2723 offset += 4 + (buffer[offset+2] << 8) +
2725 else if (!spf && len - offset > 1)
2726 offset += 2 + buffer[offset+1];
2728 sd_first_printk(KERN_ERR, sdkp,
2730 "parameter data\n");
2736 sd_first_printk(KERN_WARNING, sdkp,
2737 "No Caching mode page found\n");
2741 if (modepage == 8) {
2742 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2743 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2745 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2749 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2750 if (sdp->broken_fua) {
2751 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2753 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2754 !sdkp->device->use_16_for_rw) {
2755 sd_first_printk(KERN_NOTICE, sdkp,
2756 "Uses READ/WRITE(6), disabling FUA\n");
2760 /* No cache flush allowed for write protected devices */
2761 if (sdkp->WCE && sdkp->write_prot)
2764 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2765 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2766 sd_printk(KERN_NOTICE, sdkp,
2767 "Write cache: %s, read cache: %s, %s\n",
2768 sdkp->WCE ? "enabled" : "disabled",
2769 sdkp->RCD ? "disabled" : "enabled",
2770 sdkp->DPOFUA ? "supports DPO and FUA"
2771 : "doesn't support DPO or FUA");
2777 if (scsi_sense_valid(&sshdr) &&
2778 sshdr.sense_key == ILLEGAL_REQUEST &&
2779 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2780 /* Invalid field in CDB */
2781 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2783 sd_first_printk(KERN_ERR, sdkp,
2784 "Asking for cache data failed\n");
2787 if (sdp->wce_default_on) {
2788 sd_first_printk(KERN_NOTICE, sdkp,
2789 "Assuming drive cache: write back\n");
2792 sd_first_printk(KERN_WARNING, sdkp,
2793 "Assuming drive cache: write through\n");
2801 * The ATO bit indicates whether the DIF application tag is available
2802 * for use by the operating system.
2804 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2807 struct scsi_device *sdp = sdkp->device;
2808 struct scsi_mode_data data;
2809 struct scsi_sense_hdr sshdr;
2811 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2814 if (sdkp->protection_type == 0)
2817 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2818 sdkp->max_retries, &data, &sshdr);
2820 if (res < 0 || !data.header_length ||
2822 sd_first_printk(KERN_WARNING, sdkp,
2823 "getting Control mode page failed, assume no ATO\n");
2825 if (scsi_sense_valid(&sshdr))
2826 sd_print_sense_hdr(sdkp, &sshdr);
2831 offset = data.header_length + data.block_descriptor_length;
2833 if ((buffer[offset] & 0x3f) != 0x0a) {
2834 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2838 if ((buffer[offset + 5] & 0x80) == 0)
2847 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2848 * @sdkp: disk to query
2850 static void sd_read_block_limits(struct scsi_disk *sdkp)
2852 struct scsi_vpd *vpd;
2856 vpd = rcu_dereference(sdkp->device->vpd_pgb0);
2857 if (!vpd || vpd->len < 16)
2860 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
2861 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
2862 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
2864 if (vpd->len >= 64) {
2865 unsigned int lba_count, desc_count;
2867 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
2872 lba_count = get_unaligned_be32(&vpd->data[20]);
2873 desc_count = get_unaligned_be32(&vpd->data[24]);
2875 if (lba_count && desc_count)
2876 sdkp->max_unmap_blocks = lba_count;
2878 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
2880 if (vpd->data[32] & 0x80)
2881 sdkp->unmap_alignment =
2882 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
2884 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2886 if (sdkp->max_unmap_blocks)
2887 sd_config_discard(sdkp, SD_LBP_UNMAP);
2889 sd_config_discard(sdkp, SD_LBP_WS16);
2891 } else { /* LBP VPD page tells us what to use */
2892 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2893 sd_config_discard(sdkp, SD_LBP_UNMAP);
2894 else if (sdkp->lbpws)
2895 sd_config_discard(sdkp, SD_LBP_WS16);
2896 else if (sdkp->lbpws10)
2897 sd_config_discard(sdkp, SD_LBP_WS10);
2899 sd_config_discard(sdkp, SD_LBP_DISABLE);
2908 * sd_read_block_characteristics - Query block dev. characteristics
2909 * @sdkp: disk to query
2911 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2913 struct request_queue *q = sdkp->disk->queue;
2914 struct scsi_vpd *vpd;
2919 vpd = rcu_dereference(sdkp->device->vpd_pgb1);
2921 if (!vpd || vpd->len < 8) {
2926 rot = get_unaligned_be16(&vpd->data[4]);
2927 zoned = (vpd->data[8] >> 4) & 3;
2931 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2932 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2935 if (sdkp->device->type == TYPE_ZBC) {
2937 * Host-managed: Per ZBC and ZAC specifications, writes in
2938 * sequential write required zones of host-managed devices must
2939 * be aligned to the device physical block size.
2941 disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
2942 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
2944 sdkp->zoned = zoned;
2945 if (sdkp->zoned == 1) {
2947 disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
2949 /* Regular disk or drive managed disk */
2950 disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
2954 if (!sdkp->first_scan)
2957 if (blk_queue_is_zoned(q)) {
2958 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2959 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2961 if (sdkp->zoned == 1)
2962 sd_printk(KERN_NOTICE, sdkp,
2963 "Host-aware SMR disk used as regular disk\n");
2964 else if (sdkp->zoned == 2)
2965 sd_printk(KERN_NOTICE, sdkp,
2966 "Drive-managed SMR disk\n");
2971 * sd_read_block_provisioning - Query provisioning VPD page
2972 * @sdkp: disk to query
2974 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2976 struct scsi_vpd *vpd;
2978 if (sdkp->lbpme == 0)
2982 vpd = rcu_dereference(sdkp->device->vpd_pgb2);
2984 if (!vpd || vpd->len < 8) {
2990 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */
2991 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
2992 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
2996 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2998 struct scsi_device *sdev = sdkp->device;
3000 if (sdev->host->no_write_same) {
3001 sdev->no_write_same = 1;
3006 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3007 struct scsi_vpd *vpd;
3009 sdev->no_report_opcodes = 1;
3011 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3012 * CODES is unsupported and the device has an ATA
3013 * Information VPD page (SAT).
3016 vpd = rcu_dereference(sdev->vpd_pg89);
3018 sdev->no_write_same = 1;
3022 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3025 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3029 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3031 struct scsi_device *sdev = sdkp->device;
3033 if (!sdev->security_supported)
3036 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3037 SECURITY_PROTOCOL_IN) == 1 &&
3038 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3039 SECURITY_PROTOCOL_OUT) == 1)
3043 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3045 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3049 * sd_read_cpr - Query concurrent positioning ranges
3050 * @sdkp: disk to query
3052 static void sd_read_cpr(struct scsi_disk *sdkp)
3054 struct blk_independent_access_ranges *iars = NULL;
3055 unsigned char *buffer = NULL;
3056 unsigned int nr_cpr = 0;
3057 int i, vpd_len, buf_len = SD_BUF_SIZE;
3061 * We need to have the capacity set first for the block layer to be
3062 * able to check the ranges.
3064 if (sdkp->first_scan)
3067 if (!sdkp->capacity)
3071 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3072 * leading to a maximum page size of 64 + 256*32 bytes.
3074 buf_len = 64 + 256*32;
3075 buffer = kmalloc(buf_len, GFP_KERNEL);
3076 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3079 /* We must have at least a 64B header and one 32B range descriptor */
3080 vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3081 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3082 sd_printk(KERN_ERR, sdkp,
3083 "Invalid Concurrent Positioning Ranges VPD page\n");
3087 nr_cpr = (vpd_len - 64) / 32;
3093 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3100 for (i = 0; i < nr_cpr; i++, desc += 32) {
3102 sd_printk(KERN_ERR, sdkp,
3103 "Invalid Concurrent Positioning Range number\n");
3108 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3109 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3113 disk_set_independent_access_ranges(sdkp->disk, iars);
3114 if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3115 sd_printk(KERN_NOTICE, sdkp,
3116 "%u concurrent positioning ranges\n", nr_cpr);
3117 sdkp->nr_actuators = nr_cpr;
3123 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3125 struct scsi_device *sdp = sdkp->device;
3126 unsigned int min_xfer_bytes =
3127 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3129 if (sdkp->min_xfer_blocks == 0)
3132 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3133 sd_first_printk(KERN_WARNING, sdkp,
3134 "Preferred minimum I/O size %u bytes not a " \
3135 "multiple of physical block size (%u bytes)\n",
3136 min_xfer_bytes, sdkp->physical_block_size);
3137 sdkp->min_xfer_blocks = 0;
3141 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3147 * Determine the device's preferred I/O size for reads and writes
3148 * unless the reported value is unreasonably small, large, not a
3149 * multiple of the physical block size, or simply garbage.
3151 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3152 unsigned int dev_max)
3154 struct scsi_device *sdp = sdkp->device;
3155 unsigned int opt_xfer_bytes =
3156 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3157 unsigned int min_xfer_bytes =
3158 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3160 if (sdkp->opt_xfer_blocks == 0)
3163 if (sdkp->opt_xfer_blocks > dev_max) {
3164 sd_first_printk(KERN_WARNING, sdkp,
3165 "Optimal transfer size %u logical blocks " \
3166 "> dev_max (%u logical blocks)\n",
3167 sdkp->opt_xfer_blocks, dev_max);
3171 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3172 sd_first_printk(KERN_WARNING, sdkp,
3173 "Optimal transfer size %u logical blocks " \
3174 "> sd driver limit (%u logical blocks)\n",
3175 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3179 if (opt_xfer_bytes < PAGE_SIZE) {
3180 sd_first_printk(KERN_WARNING, sdkp,
3181 "Optimal transfer size %u bytes < " \
3182 "PAGE_SIZE (%u bytes)\n",
3183 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3187 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3188 sd_first_printk(KERN_WARNING, sdkp,
3189 "Optimal transfer size %u bytes not a " \
3190 "multiple of preferred minimum block " \
3191 "size (%u bytes)\n",
3192 opt_xfer_bytes, min_xfer_bytes);
3196 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3197 sd_first_printk(KERN_WARNING, sdkp,
3198 "Optimal transfer size %u bytes not a " \
3199 "multiple of physical block size (%u bytes)\n",
3200 opt_xfer_bytes, sdkp->physical_block_size);
3204 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3210 * sd_revalidate_disk - called the first time a new disk is seen,
3211 * performs disk spin up, read_capacity, etc.
3212 * @disk: struct gendisk we care about
3214 static int sd_revalidate_disk(struct gendisk *disk)
3216 struct scsi_disk *sdkp = scsi_disk(disk);
3217 struct scsi_device *sdp = sdkp->device;
3218 struct request_queue *q = sdkp->disk->queue;
3219 sector_t old_capacity = sdkp->capacity;
3220 unsigned char *buffer;
3221 unsigned int dev_max, rw_max;
3223 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3224 "sd_revalidate_disk\n"));
3227 * If the device is offline, don't try and read capacity or any
3228 * of the other niceties.
3230 if (!scsi_device_online(sdp))
3233 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3235 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3236 "allocation failure.\n");
3240 sd_spinup_disk(sdkp);
3243 * Without media there is no reason to ask; moreover, some devices
3244 * react badly if we do.
3246 if (sdkp->media_present) {
3247 sd_read_capacity(sdkp, buffer);
3250 * set the default to rotational. All non-rotational devices
3251 * support the block characteristics VPD page, which will
3252 * cause this to be updated correctly and any device which
3253 * doesn't support it should be treated as rotational.
3255 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3256 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3258 if (scsi_device_supports_vpd(sdp)) {
3259 sd_read_block_provisioning(sdkp);
3260 sd_read_block_limits(sdkp);
3261 sd_read_block_characteristics(sdkp);
3262 sd_zbc_read_zones(sdkp, buffer);
3266 sd_print_capacity(sdkp, old_capacity);
3268 sd_read_write_protect_flag(sdkp, buffer);
3269 sd_read_cache_type(sdkp, buffer);
3270 sd_read_app_tag_own(sdkp, buffer);
3271 sd_read_write_same(sdkp, buffer);
3272 sd_read_security(sdkp, buffer);
3273 sd_config_protection(sdkp);
3277 * We now have all cache related info, determine how we deal
3278 * with flush requests.
3280 sd_set_flush_flag(sdkp);
3282 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3283 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3285 /* Some devices report a maximum block count for READ/WRITE requests. */
3286 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3287 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3289 if (sd_validate_min_xfer_size(sdkp))
3290 blk_queue_io_min(sdkp->disk->queue,
3291 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3293 blk_queue_io_min(sdkp->disk->queue, 0);
3295 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3296 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3297 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3299 q->limits.io_opt = 0;
3300 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3301 (sector_t)BLK_DEF_MAX_SECTORS);
3305 * Limit default to SCSI host optimal sector limit if set. There may be
3306 * an impact on performance for when the size of a request exceeds this
3309 rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3311 /* Do not exceed controller limit */
3312 rw_max = min(rw_max, queue_max_hw_sectors(q));
3315 * Only update max_sectors if previously unset or if the current value
3316 * exceeds the capabilities of the hardware.
3318 if (sdkp->first_scan ||
3319 q->limits.max_sectors > q->limits.max_dev_sectors ||
3320 q->limits.max_sectors > q->limits.max_hw_sectors)
3321 q->limits.max_sectors = rw_max;
3323 sdkp->first_scan = 0;
3325 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3326 sd_config_write_same(sdkp);
3330 * For a zoned drive, revalidating the zones can be done only once
3331 * the gendisk capacity is set. So if this fails, set back the gendisk
3334 if (sd_zbc_revalidate_zones(sdkp))
3335 set_capacity_and_notify(disk, 0);
3342 * sd_unlock_native_capacity - unlock native capacity
3343 * @disk: struct gendisk to set capacity for
3345 * Block layer calls this function if it detects that partitions
3346 * on @disk reach beyond the end of the device. If the SCSI host
3347 * implements ->unlock_native_capacity() method, it's invoked to
3348 * give it a chance to adjust the device capacity.
3351 * Defined by block layer. Might sleep.
3353 static void sd_unlock_native_capacity(struct gendisk *disk)
3355 struct scsi_device *sdev = scsi_disk(disk)->device;
3357 if (sdev->host->hostt->unlock_native_capacity)
3358 sdev->host->hostt->unlock_native_capacity(sdev);
3362 * sd_format_disk_name - format disk name
3363 * @prefix: name prefix - ie. "sd" for SCSI disks
3364 * @index: index of the disk to format name for
3365 * @buf: output buffer
3366 * @buflen: length of the output buffer
3368 * SCSI disk names starts at sda. The 26th device is sdz and the
3369 * 27th is sdaa. The last one for two lettered suffix is sdzz
3370 * which is followed by sdaaa.
3372 * This is basically 26 base counting with one extra 'nil' entry
3373 * at the beginning from the second digit on and can be
3374 * determined using similar method as 26 base conversion with the
3375 * index shifted -1 after each digit is computed.
3381 * 0 on success, -errno on failure.
3383 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3385 const int base = 'z' - 'a' + 1;
3386 char *begin = buf + strlen(prefix);
3387 char *end = buf + buflen;
3397 *--p = 'a' + (index % unit);
3398 index = (index / unit) - 1;
3399 } while (index >= 0);
3401 memmove(begin, p, end - p);
3402 memcpy(buf, prefix, strlen(prefix));
3408 * sd_probe - called during driver initialization and whenever a
3409 * new scsi device is attached to the system. It is called once
3410 * for each scsi device (not just disks) present.
3411 * @dev: pointer to device object
3413 * Returns 0 if successful (or not interested in this scsi device
3414 * (e.g. scanner)); 1 when there is an error.
3416 * Note: this function is invoked from the scsi mid-level.
3417 * This function sets up the mapping between a given
3418 * <host,channel,id,lun> (found in sdp) and new device name
3419 * (e.g. /dev/sda). More precisely it is the block device major
3420 * and minor number that is chosen here.
3422 * Assume sd_probe is not re-entrant (for time being)
3423 * Also think about sd_probe() and sd_remove() running coincidentally.
3425 static int sd_probe(struct device *dev)
3427 struct scsi_device *sdp = to_scsi_device(dev);
3428 struct scsi_disk *sdkp;
3433 scsi_autopm_get_device(sdp);
3435 if (sdp->type != TYPE_DISK &&
3436 sdp->type != TYPE_ZBC &&
3437 sdp->type != TYPE_MOD &&
3438 sdp->type != TYPE_RBC)
3441 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3442 sdev_printk(KERN_WARNING, sdp,
3443 "Unsupported ZBC host-managed device.\n");
3447 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3451 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3455 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3456 &sd_bio_compl_lkclass);
3460 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3462 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3466 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3468 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3469 goto out_free_index;
3474 sdkp->index = index;
3475 sdkp->max_retries = SD_MAX_RETRIES;
3476 atomic_set(&sdkp->openers, 0);
3477 atomic_set(&sdkp->device->ioerr_cnt, 0);
3479 if (!sdp->request_queue->rq_timeout) {
3480 if (sdp->type != TYPE_MOD)
3481 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3483 blk_queue_rq_timeout(sdp->request_queue,
3487 device_initialize(&sdkp->disk_dev);
3488 sdkp->disk_dev.parent = get_device(dev);
3489 sdkp->disk_dev.class = &sd_disk_class;
3490 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3492 error = device_add(&sdkp->disk_dev);
3494 put_device(&sdkp->disk_dev);
3498 dev_set_drvdata(dev, sdkp);
3500 gd->major = sd_major((index & 0xf0) >> 4);
3501 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3502 gd->minors = SD_MINORS;
3504 gd->fops = &sd_fops;
3505 gd->private_data = sdkp;
3507 /* defaults, until the device tells us otherwise */
3508 sdp->sector_size = 512;
3510 sdkp->media_present = 1;
3511 sdkp->write_prot = 0;
3512 sdkp->cache_override = 0;
3516 sdkp->first_scan = 1;
3517 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3519 sd_revalidate_disk(gd);
3521 if (sdp->removable) {
3522 gd->flags |= GENHD_FL_REMOVABLE;
3523 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3524 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3527 blk_pm_runtime_init(sdp->request_queue, dev);
3528 if (sdp->rpm_autosuspend) {
3529 pm_runtime_set_autosuspend_delay(dev,
3530 sdp->host->hostt->rpm_autosuspend_delay);
3533 error = device_add_disk(dev, gd, NULL);
3535 put_device(&sdkp->disk_dev);
3540 if (sdkp->security) {
3541 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3543 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3546 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3547 sdp->removable ? "removable " : "");
3548 scsi_autopm_put_device(sdp);
3553 ida_free(&sd_index_ida, index);
3559 scsi_autopm_put_device(sdp);
3564 * sd_remove - called whenever a scsi disk (previously recognized by
3565 * sd_probe) is detached from the system. It is called (potentially
3566 * multiple times) during sd module unload.
3567 * @dev: pointer to device object
3569 * Note: this function is invoked from the scsi mid-level.
3570 * This function potentially frees up a device name (e.g. /dev/sdc)
3571 * that could be re-used by a subsequent sd_probe().
3572 * This function is not called when the built-in sd driver is "exit-ed".
3574 static int sd_remove(struct device *dev)
3576 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3578 scsi_autopm_get_device(sdkp->device);
3580 device_del(&sdkp->disk_dev);
3581 del_gendisk(sdkp->disk);
3584 put_disk(sdkp->disk);
3588 static void scsi_disk_release(struct device *dev)
3590 struct scsi_disk *sdkp = to_scsi_disk(dev);
3592 ida_free(&sd_index_ida, sdkp->index);
3593 sd_zbc_free_zone_info(sdkp);
3594 put_device(&sdkp->device->sdev_gendev);
3595 free_opal_dev(sdkp->opal_dev);
3600 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3602 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3603 struct scsi_sense_hdr sshdr;
3604 struct scsi_device *sdp = sdkp->device;
3608 cmd[4] |= 1; /* START */
3610 if (sdp->start_stop_pwr_cond)
3611 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3613 if (!scsi_device_online(sdp))
3616 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3617 SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3619 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3620 if (res > 0 && scsi_sense_valid(&sshdr)) {
3621 sd_print_sense_hdr(sdkp, &sshdr);
3622 /* 0x3a is medium not present */
3623 if (sshdr.asc == 0x3a)
3628 /* SCSI error codes must not go to the generic layer */
3636 * Send a SYNCHRONIZE CACHE instruction down to the device through
3637 * the normal SCSI command structure. Wait for the command to
3640 static void sd_shutdown(struct device *dev)
3642 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3645 return; /* this can happen */
3647 if (pm_runtime_suspended(dev))
3650 if (sdkp->WCE && sdkp->media_present) {
3651 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3652 sd_sync_cache(sdkp, NULL);
3655 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3656 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3657 sd_start_stop_device(sdkp, 0);
3661 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3663 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3664 struct scsi_sense_hdr sshdr;
3667 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3670 if (sdkp->WCE && sdkp->media_present) {
3671 if (!sdkp->device->silence_suspend)
3672 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3673 ret = sd_sync_cache(sdkp, &sshdr);
3676 /* ignore OFFLINE device */
3680 if (!scsi_sense_valid(&sshdr) ||
3681 sshdr.sense_key != ILLEGAL_REQUEST)
3685 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3686 * doesn't support sync. There's not much to do and
3687 * suspend shouldn't fail.
3693 if (sdkp->device->manage_start_stop) {
3694 if (!sdkp->device->silence_suspend)
3695 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3696 /* an error is not worth aborting a system sleep */
3697 ret = sd_start_stop_device(sdkp, 0);
3698 if (ignore_stop_errors)
3705 static int sd_suspend_system(struct device *dev)
3707 if (pm_runtime_suspended(dev))
3710 return sd_suspend_common(dev, true);
3713 static int sd_suspend_runtime(struct device *dev)
3715 return sd_suspend_common(dev, false);
3718 static int sd_resume(struct device *dev)
3720 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3723 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3726 if (!sdkp->device->manage_start_stop)
3729 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3730 ret = sd_start_stop_device(sdkp, 1);
3732 opal_unlock_from_suspend(sdkp->opal_dev);
3736 static int sd_resume_system(struct device *dev)
3738 if (pm_runtime_suspended(dev))
3741 return sd_resume(dev);
3744 static int sd_resume_runtime(struct device *dev)
3746 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3747 struct scsi_device *sdp;
3749 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3754 if (sdp->ignore_media_change) {
3755 /* clear the device's sense data */
3756 static const u8 cmd[10] = { REQUEST_SENSE };
3758 if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3759 NULL, sdp->request_queue->rq_timeout, 1, 0,
3761 sd_printk(KERN_NOTICE, sdkp,
3762 "Failed to clear sense data\n");
3765 return sd_resume(dev);
3769 * init_sd - entry point for this driver (both when built in or when
3772 * Note: this function registers this driver with the scsi mid-level.
3774 static int __init init_sd(void)
3776 int majors = 0, i, err;
3778 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3780 for (i = 0; i < SD_MAJORS; i++) {
3781 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3789 err = class_register(&sd_disk_class);
3793 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3795 if (!sd_cdb_cache) {
3796 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3801 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3802 if (!sd_page_pool) {
3803 printk(KERN_ERR "sd: can't init discard page pool\n");
3808 err = scsi_register_driver(&sd_template.gendrv);
3810 goto err_out_driver;
3815 mempool_destroy(sd_page_pool);
3818 kmem_cache_destroy(sd_cdb_cache);
3821 class_unregister(&sd_disk_class);
3823 for (i = 0; i < SD_MAJORS; i++)
3824 unregister_blkdev(sd_major(i), "sd");
3829 * exit_sd - exit point for this driver (when it is a module).
3831 * Note: this function unregisters this driver from the scsi mid-level.
3833 static void __exit exit_sd(void)
3837 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3839 scsi_unregister_driver(&sd_template.gendrv);
3840 mempool_destroy(sd_page_pool);
3841 kmem_cache_destroy(sd_cdb_cache);
3843 class_unregister(&sd_disk_class);
3845 for (i = 0; i < SD_MAJORS; i++)
3846 unregister_blkdev(sd_major(i), "sd");
3849 module_init(init_sd);
3850 module_exit(exit_sd);
3852 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3854 scsi_print_sense_hdr(sdkp->device,
3855 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3858 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3860 const char *hb_string = scsi_hostbyte_string(result);
3863 sd_printk(KERN_INFO, sdkp,
3864 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3865 hb_string ? hb_string : "invalid",
3868 sd_printk(KERN_INFO, sdkp,
3869 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3870 msg, host_byte(result), "DRIVER_OK");