1 // SPDX-License-Identifier: GPL-2.0-only
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
36 #include <linux/module.h>
38 #include <linux/kernel.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
107 static void sd_config_discard(struct scsi_disk *, unsigned int);
108 static void sd_config_write_same(struct scsi_disk *);
109 static int sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static int sd_probe(struct device *);
112 static int sd_remove(struct device *);
113 static void sd_shutdown(struct device *);
114 static int sd_suspend_system(struct device *);
115 static int sd_suspend_runtime(struct device *);
116 static int sd_resume(struct device *);
117 static void sd_rescan(struct device *);
118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120 static int sd_done(struct scsi_cmnd *);
121 static void sd_eh_reset(struct scsi_cmnd *);
122 static int sd_eh_action(struct scsi_cmnd *, int);
123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124 static void scsi_disk_release(struct device *cdev);
125 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
126 static void sd_print_result(const struct scsi_disk *, const char *, int);
128 static DEFINE_IDA(sd_index_ida);
130 /* This semaphore is used to mediate the 0->1 reference get in the
131 * face of object destruction (i.e. we can't allow a get on an
132 * object after last put) */
133 static DEFINE_MUTEX(sd_ref_mutex);
135 static struct kmem_cache *sd_cdb_cache;
136 static mempool_t *sd_cdb_pool;
137 static mempool_t *sd_page_pool;
139 static const char *sd_cache_types[] = {
140 "write through", "none", "write back",
141 "write back, no read (daft)"
144 static void sd_set_flush_flag(struct scsi_disk *sdkp)
146 bool wc = false, fua = false;
154 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
158 cache_type_store(struct device *dev, struct device_attribute *attr,
159 const char *buf, size_t count)
161 int ct, rcd, wce, sp;
162 struct scsi_disk *sdkp = to_scsi_disk(dev);
163 struct scsi_device *sdp = sdkp->device;
166 struct scsi_mode_data data;
167 struct scsi_sense_hdr sshdr;
168 static const char temp[] = "temporary ";
171 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
172 /* no cache control on RBC devices; theoretically they
173 * can do it, but there's probably so many exceptions
174 * it's not worth the risk */
177 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
178 buf += sizeof(temp) - 1;
179 sdkp->cache_override = 1;
181 sdkp->cache_override = 0;
184 ct = sysfs_match_string(sd_cache_types, buf);
188 rcd = ct & 0x01 ? 1 : 0;
189 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
191 if (sdkp->cache_override) {
194 sd_set_flush_flag(sdkp);
198 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 SD_MAX_RETRIES, &data, NULL))
201 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 data.block_descriptor_length);
203 buffer_data = buffer + data.header_length +
204 data.block_descriptor_length;
205 buffer_data[2] &= ~0x05;
206 buffer_data[2] |= wce << 2 | rcd;
207 sp = buffer_data[0] & 0x80 ? 1 : 0;
208 buffer_data[0] &= ~0x80;
211 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
212 * received mode parameter buffer before doing MODE SELECT.
214 data.device_specific = 0;
216 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
217 SD_MAX_RETRIES, &data, &sshdr)) {
218 if (scsi_sense_valid(&sshdr))
219 sd_print_sense_hdr(sdkp, &sshdr);
222 revalidate_disk(sdkp->disk);
227 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
230 struct scsi_disk *sdkp = to_scsi_disk(dev);
231 struct scsi_device *sdp = sdkp->device;
233 return sprintf(buf, "%u\n", sdp->manage_start_stop);
237 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t count)
240 struct scsi_disk *sdkp = to_scsi_disk(dev);
241 struct scsi_device *sdp = sdkp->device;
244 if (!capable(CAP_SYS_ADMIN))
247 if (kstrtobool(buf, &v))
250 sdp->manage_start_stop = v;
254 static DEVICE_ATTR_RW(manage_start_stop);
257 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
259 struct scsi_disk *sdkp = to_scsi_disk(dev);
261 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
265 allow_restart_store(struct device *dev, struct device_attribute *attr,
266 const char *buf, size_t count)
269 struct scsi_disk *sdkp = to_scsi_disk(dev);
270 struct scsi_device *sdp = sdkp->device;
272 if (!capable(CAP_SYS_ADMIN))
275 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
278 if (kstrtobool(buf, &v))
281 sdp->allow_restart = v;
285 static DEVICE_ATTR_RW(allow_restart);
288 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
290 struct scsi_disk *sdkp = to_scsi_disk(dev);
291 int ct = sdkp->RCD + 2*sdkp->WCE;
293 return sprintf(buf, "%s\n", sd_cache_types[ct]);
295 static DEVICE_ATTR_RW(cache_type);
298 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
300 struct scsi_disk *sdkp = to_scsi_disk(dev);
302 return sprintf(buf, "%u\n", sdkp->DPOFUA);
304 static DEVICE_ATTR_RO(FUA);
307 protection_type_show(struct device *dev, struct device_attribute *attr,
310 struct scsi_disk *sdkp = to_scsi_disk(dev);
312 return sprintf(buf, "%u\n", sdkp->protection_type);
316 protection_type_store(struct device *dev, struct device_attribute *attr,
317 const char *buf, size_t count)
319 struct scsi_disk *sdkp = to_scsi_disk(dev);
323 if (!capable(CAP_SYS_ADMIN))
326 err = kstrtouint(buf, 10, &val);
331 if (val <= T10_PI_TYPE3_PROTECTION)
332 sdkp->protection_type = val;
336 static DEVICE_ATTR_RW(protection_type);
339 protection_mode_show(struct device *dev, struct device_attribute *attr,
342 struct scsi_disk *sdkp = to_scsi_disk(dev);
343 struct scsi_device *sdp = sdkp->device;
344 unsigned int dif, dix;
346 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
347 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
349 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
355 return sprintf(buf, "none\n");
357 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
359 static DEVICE_ATTR_RO(protection_mode);
362 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
366 return sprintf(buf, "%u\n", sdkp->ATO);
368 static DEVICE_ATTR_RO(app_tag_own);
371 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
374 struct scsi_disk *sdkp = to_scsi_disk(dev);
376 return sprintf(buf, "%u\n", sdkp->lbpme);
378 static DEVICE_ATTR_RO(thin_provisioning);
380 /* sysfs_match_string() requires dense arrays */
381 static const char *lbp_mode[] = {
382 [SD_LBP_FULL] = "full",
383 [SD_LBP_UNMAP] = "unmap",
384 [SD_LBP_WS16] = "writesame_16",
385 [SD_LBP_WS10] = "writesame_10",
386 [SD_LBP_ZERO] = "writesame_zero",
387 [SD_LBP_DISABLE] = "disabled",
391 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
394 struct scsi_disk *sdkp = to_scsi_disk(dev);
396 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
400 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
401 const char *buf, size_t count)
403 struct scsi_disk *sdkp = to_scsi_disk(dev);
404 struct scsi_device *sdp = sdkp->device;
407 if (!capable(CAP_SYS_ADMIN))
410 if (sd_is_zoned(sdkp)) {
411 sd_config_discard(sdkp, SD_LBP_DISABLE);
415 if (sdp->type != TYPE_DISK)
418 mode = sysfs_match_string(lbp_mode, buf);
422 sd_config_discard(sdkp, mode);
426 static DEVICE_ATTR_RW(provisioning_mode);
428 /* sysfs_match_string() requires dense arrays */
429 static const char *zeroing_mode[] = {
430 [SD_ZERO_WRITE] = "write",
431 [SD_ZERO_WS] = "writesame",
432 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
433 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
437 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
440 struct scsi_disk *sdkp = to_scsi_disk(dev);
442 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
446 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
447 const char *buf, size_t count)
449 struct scsi_disk *sdkp = to_scsi_disk(dev);
452 if (!capable(CAP_SYS_ADMIN))
455 mode = sysfs_match_string(zeroing_mode, buf);
459 sdkp->zeroing_mode = mode;
463 static DEVICE_ATTR_RW(zeroing_mode);
466 max_medium_access_timeouts_show(struct device *dev,
467 struct device_attribute *attr, char *buf)
469 struct scsi_disk *sdkp = to_scsi_disk(dev);
471 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
475 max_medium_access_timeouts_store(struct device *dev,
476 struct device_attribute *attr, const char *buf,
479 struct scsi_disk *sdkp = to_scsi_disk(dev);
482 if (!capable(CAP_SYS_ADMIN))
485 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
487 return err ? err : count;
489 static DEVICE_ATTR_RW(max_medium_access_timeouts);
492 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
495 struct scsi_disk *sdkp = to_scsi_disk(dev);
497 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
501 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
502 const char *buf, size_t count)
504 struct scsi_disk *sdkp = to_scsi_disk(dev);
505 struct scsi_device *sdp = sdkp->device;
509 if (!capable(CAP_SYS_ADMIN))
512 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
515 err = kstrtoul(buf, 10, &max);
521 sdp->no_write_same = 1;
522 else if (max <= SD_MAX_WS16_BLOCKS) {
523 sdp->no_write_same = 0;
524 sdkp->max_ws_blocks = max;
527 sd_config_write_same(sdkp);
531 static DEVICE_ATTR_RW(max_write_same_blocks);
533 static struct attribute *sd_disk_attrs[] = {
534 &dev_attr_cache_type.attr,
536 &dev_attr_allow_restart.attr,
537 &dev_attr_manage_start_stop.attr,
538 &dev_attr_protection_type.attr,
539 &dev_attr_protection_mode.attr,
540 &dev_attr_app_tag_own.attr,
541 &dev_attr_thin_provisioning.attr,
542 &dev_attr_provisioning_mode.attr,
543 &dev_attr_zeroing_mode.attr,
544 &dev_attr_max_write_same_blocks.attr,
545 &dev_attr_max_medium_access_timeouts.attr,
548 ATTRIBUTE_GROUPS(sd_disk);
550 static struct class sd_disk_class = {
552 .owner = THIS_MODULE,
553 .dev_release = scsi_disk_release,
554 .dev_groups = sd_disk_groups,
557 static const struct dev_pm_ops sd_pm_ops = {
558 .suspend = sd_suspend_system,
560 .poweroff = sd_suspend_system,
561 .restore = sd_resume,
562 .runtime_suspend = sd_suspend_runtime,
563 .runtime_resume = sd_resume,
566 static struct scsi_driver sd_template = {
569 .owner = THIS_MODULE,
571 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
573 .shutdown = sd_shutdown,
577 .init_command = sd_init_command,
578 .uninit_command = sd_uninit_command,
580 .eh_action = sd_eh_action,
581 .eh_reset = sd_eh_reset,
585 * Dummy kobj_map->probe function.
586 * The default ->probe function will call modprobe, which is
587 * pointless as this module is already loaded.
589 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
595 * Device no to disk mapping:
597 * major disc2 disc p1
598 * |............|.............|....|....| <- dev_t
601 * Inside a major, we have 16k disks, however mapped non-
602 * contiguously. The first 16 disks are for major0, the next
603 * ones with major1, ... Disk 256 is for major0 again, disk 272
605 * As we stay compatible with our numbering scheme, we can reuse
606 * the well-know SCSI majors 8, 65--71, 136--143.
608 static int sd_major(int major_idx)
612 return SCSI_DISK0_MAJOR;
614 return SCSI_DISK1_MAJOR + major_idx - 1;
616 return SCSI_DISK8_MAJOR + major_idx - 8;
619 return 0; /* shut up gcc */
623 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
625 struct scsi_disk *sdkp = NULL;
627 mutex_lock(&sd_ref_mutex);
629 if (disk->private_data) {
630 sdkp = scsi_disk(disk);
631 if (scsi_device_get(sdkp->device) == 0)
632 get_device(&sdkp->dev);
636 mutex_unlock(&sd_ref_mutex);
640 static void scsi_disk_put(struct scsi_disk *sdkp)
642 struct scsi_device *sdev = sdkp->device;
644 mutex_lock(&sd_ref_mutex);
645 put_device(&sdkp->dev);
646 scsi_device_put(sdev);
647 mutex_unlock(&sd_ref_mutex);
650 #ifdef CONFIG_BLK_SED_OPAL
651 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
652 size_t len, bool send)
654 struct scsi_device *sdev = data;
658 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
660 put_unaligned_be16(spsp, &cdb[2]);
661 put_unaligned_be32(len, &cdb[6]);
663 ret = scsi_execute_req(sdev, cdb,
664 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
665 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
666 return ret <= 0 ? ret : -EIO;
668 #endif /* CONFIG_BLK_SED_OPAL */
671 * Look up the DIX operation based on whether the command is read or
672 * write and whether dix and dif are enabled.
674 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
676 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
677 static const unsigned int ops[] = { /* wrt dix dif */
678 SCSI_PROT_NORMAL, /* 0 0 0 */
679 SCSI_PROT_READ_STRIP, /* 0 0 1 */
680 SCSI_PROT_READ_INSERT, /* 0 1 0 */
681 SCSI_PROT_READ_PASS, /* 0 1 1 */
682 SCSI_PROT_NORMAL, /* 1 0 0 */
683 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
684 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
685 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
688 return ops[write << 2 | dix << 1 | dif];
692 * Returns a mask of the protection flags that are valid for a given DIX
695 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
697 static const unsigned int flag_mask[] = {
698 [SCSI_PROT_NORMAL] = 0,
700 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
701 SCSI_PROT_GUARD_CHECK |
702 SCSI_PROT_REF_CHECK |
703 SCSI_PROT_REF_INCREMENT,
705 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
706 SCSI_PROT_IP_CHECKSUM,
708 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
709 SCSI_PROT_GUARD_CHECK |
710 SCSI_PROT_REF_CHECK |
711 SCSI_PROT_REF_INCREMENT |
712 SCSI_PROT_IP_CHECKSUM,
714 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
715 SCSI_PROT_REF_INCREMENT,
717 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
718 SCSI_PROT_REF_CHECK |
719 SCSI_PROT_REF_INCREMENT |
720 SCSI_PROT_IP_CHECKSUM,
722 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
723 SCSI_PROT_GUARD_CHECK |
724 SCSI_PROT_REF_CHECK |
725 SCSI_PROT_REF_INCREMENT |
726 SCSI_PROT_IP_CHECKSUM,
729 return flag_mask[prot_op];
732 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
733 unsigned int dix, unsigned int dif)
735 struct bio *bio = scmd->request->bio;
736 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
737 unsigned int protect = 0;
739 if (dix) { /* DIX Type 0, 1, 2, 3 */
740 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
741 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
743 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
744 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
747 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
748 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
750 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
751 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
754 if (dif) { /* DIX/DIF Type 1, 2, 3 */
755 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
757 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
758 protect = 3 << 5; /* Disable target PI checking */
760 protect = 1 << 5; /* Enable target PI checking */
763 scsi_set_prot_op(scmd, prot_op);
764 scsi_set_prot_type(scmd, dif);
765 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
770 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
772 struct request_queue *q = sdkp->disk->queue;
773 unsigned int logical_block_size = sdkp->device->sector_size;
774 unsigned int max_blocks = 0;
776 q->limits.discard_alignment =
777 sdkp->unmap_alignment * logical_block_size;
778 q->limits.discard_granularity =
779 max(sdkp->physical_block_size,
780 sdkp->unmap_granularity * logical_block_size);
781 sdkp->provisioning_mode = mode;
787 blk_queue_max_discard_sectors(q, 0);
788 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
792 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
793 (u32)SD_MAX_WS16_BLOCKS);
797 if (sdkp->device->unmap_limit_for_ws)
798 max_blocks = sdkp->max_unmap_blocks;
800 max_blocks = sdkp->max_ws_blocks;
802 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
806 if (sdkp->device->unmap_limit_for_ws)
807 max_blocks = sdkp->max_unmap_blocks;
809 max_blocks = sdkp->max_ws_blocks;
811 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
815 max_blocks = min_not_zero(sdkp->max_ws_blocks,
816 (u32)SD_MAX_WS10_BLOCKS);
820 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
821 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
824 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
826 struct scsi_device *sdp = cmd->device;
827 struct request *rq = cmd->request;
828 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
829 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
830 unsigned int data_len = 24;
833 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
834 if (!rq->special_vec.bv_page)
835 return BLK_STS_RESOURCE;
836 clear_highpage(rq->special_vec.bv_page);
837 rq->special_vec.bv_offset = 0;
838 rq->special_vec.bv_len = data_len;
839 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
842 cmd->cmnd[0] = UNMAP;
845 buf = page_address(rq->special_vec.bv_page);
846 put_unaligned_be16(6 + 16, &buf[0]);
847 put_unaligned_be16(16, &buf[2]);
848 put_unaligned_be64(lba, &buf[8]);
849 put_unaligned_be32(nr_blocks, &buf[16]);
851 cmd->allowed = SD_MAX_RETRIES;
852 cmd->transfersize = data_len;
853 rq->timeout = SD_TIMEOUT;
855 return scsi_init_io(cmd);
858 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
861 struct scsi_device *sdp = cmd->device;
862 struct request *rq = cmd->request;
863 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
864 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
865 u32 data_len = sdp->sector_size;
867 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
868 if (!rq->special_vec.bv_page)
869 return BLK_STS_RESOURCE;
870 clear_highpage(rq->special_vec.bv_page);
871 rq->special_vec.bv_offset = 0;
872 rq->special_vec.bv_len = data_len;
873 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
876 cmd->cmnd[0] = WRITE_SAME_16;
878 cmd->cmnd[1] = 0x8; /* UNMAP */
879 put_unaligned_be64(lba, &cmd->cmnd[2]);
880 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
882 cmd->allowed = SD_MAX_RETRIES;
883 cmd->transfersize = data_len;
884 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
886 return scsi_init_io(cmd);
889 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
892 struct scsi_device *sdp = cmd->device;
893 struct request *rq = cmd->request;
894 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
895 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
896 u32 data_len = sdp->sector_size;
898 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
899 if (!rq->special_vec.bv_page)
900 return BLK_STS_RESOURCE;
901 clear_highpage(rq->special_vec.bv_page);
902 rq->special_vec.bv_offset = 0;
903 rq->special_vec.bv_len = data_len;
904 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
907 cmd->cmnd[0] = WRITE_SAME;
909 cmd->cmnd[1] = 0x8; /* UNMAP */
910 put_unaligned_be32(lba, &cmd->cmnd[2]);
911 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
913 cmd->allowed = SD_MAX_RETRIES;
914 cmd->transfersize = data_len;
915 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
917 return scsi_init_io(cmd);
920 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
922 struct request *rq = cmd->request;
923 struct scsi_device *sdp = cmd->device;
924 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
925 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
926 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
928 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
929 switch (sdkp->zeroing_mode) {
930 case SD_ZERO_WS16_UNMAP:
931 return sd_setup_write_same16_cmnd(cmd, true);
932 case SD_ZERO_WS10_UNMAP:
933 return sd_setup_write_same10_cmnd(cmd, true);
937 if (sdp->no_write_same)
938 return BLK_STS_TARGET;
940 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
941 return sd_setup_write_same16_cmnd(cmd, false);
943 return sd_setup_write_same10_cmnd(cmd, false);
946 static void sd_config_write_same(struct scsi_disk *sdkp)
948 struct request_queue *q = sdkp->disk->queue;
949 unsigned int logical_block_size = sdkp->device->sector_size;
951 if (sdkp->device->no_write_same) {
952 sdkp->max_ws_blocks = 0;
956 /* Some devices can not handle block counts above 0xffff despite
957 * supporting WRITE SAME(16). Consequently we default to 64k
958 * blocks per I/O unless the device explicitly advertises a
961 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
962 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
963 (u32)SD_MAX_WS16_BLOCKS);
964 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
965 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
966 (u32)SD_MAX_WS10_BLOCKS);
968 sdkp->device->no_write_same = 1;
969 sdkp->max_ws_blocks = 0;
972 if (sdkp->lbprz && sdkp->lbpws)
973 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
974 else if (sdkp->lbprz && sdkp->lbpws10)
975 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
976 else if (sdkp->max_ws_blocks)
977 sdkp->zeroing_mode = SD_ZERO_WS;
979 sdkp->zeroing_mode = SD_ZERO_WRITE;
981 if (sdkp->max_ws_blocks &&
982 sdkp->physical_block_size > logical_block_size) {
984 * Reporting a maximum number of blocks that is not aligned
985 * on the device physical size would cause a large write same
986 * request to be split into physically unaligned chunks by
987 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
988 * even if the caller of these functions took care to align the
989 * large request. So make sure the maximum reported is aligned
990 * to the device physical block size. This is only an optional
991 * optimization for regular disks, but this is mandatory to
992 * avoid failure of large write same requests directed at
993 * sequential write required zones of host-managed ZBC disks.
995 sdkp->max_ws_blocks =
996 round_down(sdkp->max_ws_blocks,
997 bytes_to_logical(sdkp->device,
998 sdkp->physical_block_size));
1002 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1003 (logical_block_size >> 9));
1004 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1005 (logical_block_size >> 9));
1009 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1010 * @cmd: command to prepare
1012 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1013 * the preference indicated by the target device.
1015 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1017 struct request *rq = cmd->request;
1018 struct scsi_device *sdp = cmd->device;
1019 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1020 struct bio *bio = rq->bio;
1021 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1022 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1025 if (sdkp->device->no_write_same)
1026 return BLK_STS_TARGET;
1028 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1030 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1032 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1034 cmd->cmnd[0] = WRITE_SAME_16;
1035 put_unaligned_be64(lba, &cmd->cmnd[2]);
1036 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1039 cmd->cmnd[0] = WRITE_SAME;
1040 put_unaligned_be32(lba, &cmd->cmnd[2]);
1041 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1044 cmd->transfersize = sdp->sector_size;
1045 cmd->allowed = SD_MAX_RETRIES;
1048 * For WRITE SAME the data transferred via the DATA OUT buffer is
1049 * different from the amount of data actually written to the target.
1051 * We set up __data_len to the amount of data transferred via the
1052 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1053 * to transfer a single sector of data first, but then reset it to
1054 * the amount of data to be written right after so that the I/O path
1055 * knows how much to actually write.
1057 rq->__data_len = sdp->sector_size;
1058 ret = scsi_init_io(cmd);
1059 rq->__data_len = blk_rq_bytes(rq);
1064 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1066 struct request *rq = cmd->request;
1068 /* flush requests don't perform I/O, zero the S/G table */
1069 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1071 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1073 cmd->transfersize = 0;
1074 cmd->allowed = SD_MAX_RETRIES;
1076 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1080 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1081 sector_t lba, unsigned int nr_blocks,
1082 unsigned char flags)
1084 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1085 if (unlikely(cmd->cmnd == NULL))
1086 return BLK_STS_RESOURCE;
1088 cmd->cmd_len = SD_EXT_CDB_SIZE;
1089 memset(cmd->cmnd, 0, cmd->cmd_len);
1091 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1092 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1093 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1094 cmd->cmnd[10] = flags;
1095 put_unaligned_be64(lba, &cmd->cmnd[12]);
1096 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1097 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1102 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1103 sector_t lba, unsigned int nr_blocks,
1104 unsigned char flags)
1107 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1108 cmd->cmnd[1] = flags;
1111 put_unaligned_be64(lba, &cmd->cmnd[2]);
1112 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1117 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1118 sector_t lba, unsigned int nr_blocks,
1119 unsigned char flags)
1122 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1123 cmd->cmnd[1] = flags;
1126 put_unaligned_be32(lba, &cmd->cmnd[2]);
1127 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1132 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1133 sector_t lba, unsigned int nr_blocks,
1134 unsigned char flags)
1136 /* Avoid that 0 blocks gets translated into 256 blocks. */
1137 if (WARN_ON_ONCE(nr_blocks == 0))
1138 return BLK_STS_IOERR;
1140 if (unlikely(flags & 0x8)) {
1142 * This happens only if this drive failed 10byte rw
1143 * command with ILLEGAL_REQUEST during operation and
1144 * thus turned off use_10_for_rw.
1146 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1147 return BLK_STS_IOERR;
1151 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1152 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1153 cmd->cmnd[2] = (lba >> 8) & 0xff;
1154 cmd->cmnd[3] = lba & 0xff;
1155 cmd->cmnd[4] = nr_blocks;
1161 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1163 struct request *rq = cmd->request;
1164 struct scsi_device *sdp = cmd->device;
1165 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1166 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1168 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1170 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1171 bool write = rq_data_dir(rq) == WRITE;
1172 unsigned char protect, fua;
1175 ret = scsi_init_io(cmd);
1176 if (ret != BLK_STS_OK)
1179 if (!scsi_device_online(sdp) || sdp->changed) {
1180 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1181 return BLK_STS_IOERR;
1184 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1185 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1186 return BLK_STS_IOERR;
1189 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1190 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1191 return BLK_STS_IOERR;
1195 * Some SD card readers can't handle accesses which touch the
1196 * last one or two logical blocks. Split accesses as needed.
1198 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1200 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1201 if (lba < threshold) {
1202 /* Access up to the threshold but not beyond */
1203 nr_blocks = threshold - lba;
1205 /* Access only a single logical block */
1210 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1211 dix = scsi_prot_sg_count(cmd);
1212 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1215 t10_pi_prepare(cmd->request, sdkp->protection_type);
1218 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1222 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1223 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1225 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1226 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1228 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1229 sdp->use_10_for_rw || protect) {
1230 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1233 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1237 if (unlikely(ret != BLK_STS_OK))
1241 * We shouldn't disconnect in the middle of a sector, so with a dumb
1242 * host adapter, it's safe to assume that we can at least transfer
1243 * this many bytes between each connect / disconnect.
1245 cmd->transfersize = sdp->sector_size;
1246 cmd->underflow = nr_blocks << 9;
1247 cmd->allowed = SD_MAX_RETRIES;
1248 cmd->sdb.length = nr_blocks * sdp->sector_size;
1251 scmd_printk(KERN_INFO, cmd,
1252 "%s: block=%llu, count=%d\n", __func__,
1253 (unsigned long long)blk_rq_pos(rq),
1254 blk_rq_sectors(rq)));
1256 scmd_printk(KERN_INFO, cmd,
1257 "%s %d/%u 512 byte blocks.\n",
1258 write ? "writing" : "reading", nr_blocks,
1259 blk_rq_sectors(rq)));
1262 * This indicates that the command is ready from our end to be
1268 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1270 struct request *rq = cmd->request;
1272 switch (req_op(rq)) {
1273 case REQ_OP_DISCARD:
1274 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1276 return sd_setup_unmap_cmnd(cmd);
1278 return sd_setup_write_same16_cmnd(cmd, true);
1280 return sd_setup_write_same10_cmnd(cmd, true);
1282 return sd_setup_write_same10_cmnd(cmd, false);
1284 return BLK_STS_TARGET;
1286 case REQ_OP_WRITE_ZEROES:
1287 return sd_setup_write_zeroes_cmnd(cmd);
1288 case REQ_OP_WRITE_SAME:
1289 return sd_setup_write_same_cmnd(cmd);
1291 return sd_setup_flush_cmnd(cmd);
1294 return sd_setup_read_write_cmnd(cmd);
1295 case REQ_OP_ZONE_RESET:
1296 return sd_zbc_setup_reset_cmnd(cmd, false);
1297 case REQ_OP_ZONE_RESET_ALL:
1298 return sd_zbc_setup_reset_cmnd(cmd, true);
1301 return BLK_STS_NOTSUPP;
1305 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1307 struct request *rq = SCpnt->request;
1310 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1311 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1313 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1317 mempool_free(cmnd, sd_cdb_pool);
1322 * sd_open - open a scsi disk device
1323 * @bdev: Block device of the scsi disk to open
1324 * @mode: FMODE_* mask
1326 * Returns 0 if successful. Returns a negated errno value in case
1329 * Note: This can be called from a user context (e.g. fsck(1) )
1330 * or from within the kernel (e.g. as a result of a mount(1) ).
1331 * In the latter case @inode and @filp carry an abridged amount
1332 * of information as noted above.
1334 * Locking: called with bdev->bd_mutex held.
1336 static int sd_open(struct block_device *bdev, fmode_t mode)
1338 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1339 struct scsi_device *sdev;
1345 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1347 sdev = sdkp->device;
1350 * If the device is in error recovery, wait until it is done.
1351 * If the device is offline, then disallow any access to it.
1354 if (!scsi_block_when_processing_errors(sdev))
1357 if (sdev->removable || sdkp->write_prot)
1358 check_disk_change(bdev);
1361 * If the drive is empty, just let the open fail.
1363 retval = -ENOMEDIUM;
1364 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1368 * If the device has the write protect tab set, have the open fail
1369 * if the user expects to be able to write to the thing.
1372 if (sdkp->write_prot && (mode & FMODE_WRITE))
1376 * It is possible that the disk changing stuff resulted in
1377 * the device being taken offline. If this is the case,
1378 * report this to the user, and don't pretend that the
1379 * open actually succeeded.
1382 if (!scsi_device_online(sdev))
1385 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1386 if (scsi_block_when_processing_errors(sdev))
1387 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1393 scsi_disk_put(sdkp);
1398 * sd_release - invoked when the (last) close(2) is called on this
1400 * @disk: disk to release
1401 * @mode: FMODE_* mask
1405 * Note: may block (uninterruptible) if error recovery is underway
1408 * Locking: called with bdev->bd_mutex held.
1410 static void sd_release(struct gendisk *disk, fmode_t mode)
1412 struct scsi_disk *sdkp = scsi_disk(disk);
1413 struct scsi_device *sdev = sdkp->device;
1415 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1417 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1418 if (scsi_block_when_processing_errors(sdev))
1419 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1422 scsi_disk_put(sdkp);
1425 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1427 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1428 struct scsi_device *sdp = sdkp->device;
1429 struct Scsi_Host *host = sdp->host;
1430 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1433 /* default to most commonly used values */
1434 diskinfo[0] = 0x40; /* 1 << 6 */
1435 diskinfo[1] = 0x20; /* 1 << 5 */
1436 diskinfo[2] = capacity >> 11;
1438 /* override with calculated, extended default, or driver values */
1439 if (host->hostt->bios_param)
1440 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1442 scsicam_bios_param(bdev, capacity, diskinfo);
1444 geo->heads = diskinfo[0];
1445 geo->sectors = diskinfo[1];
1446 geo->cylinders = diskinfo[2];
1451 * sd_ioctl - process an ioctl
1452 * @bdev: target block device
1453 * @mode: FMODE_* mask
1454 * @cmd: ioctl command number
1455 * @arg: this is third argument given to ioctl(2) system call.
1456 * Often contains a pointer.
1458 * Returns 0 if successful (some ioctls return positive numbers on
1459 * success as well). Returns a negated errno value in case of error.
1461 * Note: most ioctls are forward onto the block subsystem or further
1462 * down in the scsi subsystem.
1464 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1465 unsigned int cmd, unsigned long arg)
1467 struct gendisk *disk = bdev->bd_disk;
1468 struct scsi_disk *sdkp = scsi_disk(disk);
1469 struct scsi_device *sdp = sdkp->device;
1470 void __user *p = (void __user *)arg;
1473 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1474 "cmd=0x%x\n", disk->disk_name, cmd));
1476 error = scsi_verify_blk_ioctl(bdev, cmd);
1481 * If we are in the middle of error recovery, don't let anyone
1482 * else try and use this device. Also, if error recovery fails, it
1483 * may try and take the device offline, in which case all further
1484 * access to the device is prohibited.
1486 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1487 (mode & FMODE_NDELAY) != 0);
1491 if (is_sed_ioctl(cmd))
1492 return sed_ioctl(sdkp->opal_dev, cmd, p);
1495 * Send SCSI addressing ioctls directly to mid level, send other
1496 * ioctls to block level and then onto mid level if they can't be
1500 case SCSI_IOCTL_GET_IDLUN:
1501 case SCSI_IOCTL_GET_BUS_NUMBER:
1502 error = scsi_ioctl(sdp, cmd, p);
1505 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1506 if (error != -ENOTTY)
1508 error = scsi_ioctl(sdp, cmd, p);
1515 static void set_media_not_present(struct scsi_disk *sdkp)
1517 if (sdkp->media_present)
1518 sdkp->device->changed = 1;
1520 if (sdkp->device->removable) {
1521 sdkp->media_present = 0;
1526 static int media_not_present(struct scsi_disk *sdkp,
1527 struct scsi_sense_hdr *sshdr)
1529 if (!scsi_sense_valid(sshdr))
1532 /* not invoked for commands that could return deferred errors */
1533 switch (sshdr->sense_key) {
1534 case UNIT_ATTENTION:
1536 /* medium not present */
1537 if (sshdr->asc == 0x3A) {
1538 set_media_not_present(sdkp);
1546 * sd_check_events - check media events
1547 * @disk: kernel device descriptor
1548 * @clearing: disk events currently being cleared
1550 * Returns mask of DISK_EVENT_*.
1552 * Note: this function is invoked from the block subsystem.
1554 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1556 struct scsi_disk *sdkp = scsi_disk_get(disk);
1557 struct scsi_device *sdp;
1564 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1567 * If the device is offline, don't send any commands - just pretend as
1568 * if the command failed. If the device ever comes back online, we
1569 * can deal with it then. It is only because of unrecoverable errors
1570 * that we would ever take a device offline in the first place.
1572 if (!scsi_device_online(sdp)) {
1573 set_media_not_present(sdkp);
1578 * Using TEST_UNIT_READY enables differentiation between drive with
1579 * no cartridge loaded - NOT READY, drive with changed cartridge -
1580 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1582 * Drives that auto spin down. eg iomega jaz 1G, will be started
1583 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1584 * sd_revalidate() is called.
1586 if (scsi_block_when_processing_errors(sdp)) {
1587 struct scsi_sense_hdr sshdr = { 0, };
1589 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1592 /* failed to execute TUR, assume media not present */
1593 if (host_byte(retval)) {
1594 set_media_not_present(sdkp);
1598 if (media_not_present(sdkp, &sshdr))
1603 * For removable scsi disk we have to recognise the presence
1604 * of a disk in the drive.
1606 if (!sdkp->media_present)
1608 sdkp->media_present = 1;
1611 * sdp->changed is set under the following conditions:
1613 * Medium present state has changed in either direction.
1614 * Device has indicated UNIT_ATTENTION.
1616 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1618 scsi_disk_put(sdkp);
1622 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1625 struct scsi_device *sdp = sdkp->device;
1626 const int timeout = sdp->request_queue->rq_timeout
1627 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1628 struct scsi_sense_hdr my_sshdr;
1630 if (!scsi_device_online(sdp))
1633 /* caller might not be interested in sense, but we need it */
1637 for (retries = 3; retries > 0; --retries) {
1638 unsigned char cmd[10] = { 0 };
1640 cmd[0] = SYNCHRONIZE_CACHE;
1642 * Leave the rest of the command zero to indicate
1645 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1646 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1652 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1654 if (driver_byte(res) == DRIVER_SENSE)
1655 sd_print_sense_hdr(sdkp, sshdr);
1657 /* we need to evaluate the error return */
1658 if (scsi_sense_valid(sshdr) &&
1659 (sshdr->asc == 0x3a || /* medium not present */
1660 sshdr->asc == 0x20)) /* invalid command */
1661 /* this is no error here */
1664 switch (host_byte(res)) {
1665 /* ignore errors due to racing a disconnection */
1666 case DID_BAD_TARGET:
1667 case DID_NO_CONNECT:
1669 /* signal the upper layer it might try again */
1673 case DID_SOFT_ERROR:
1682 static void sd_rescan(struct device *dev)
1684 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1686 revalidate_disk(sdkp->disk);
1690 #ifdef CONFIG_COMPAT
1692 * This gets directly called from VFS. When the ioctl
1693 * is not recognized we go back to the other translation paths.
1695 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1696 unsigned int cmd, unsigned long arg)
1698 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1701 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1702 (mode & FMODE_NDELAY) != 0);
1707 * Let the static ioctl translation table take care of it.
1709 if (!sdev->host->hostt->compat_ioctl)
1710 return -ENOIOCTLCMD;
1711 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1715 static char sd_pr_type(enum pr_type type)
1718 case PR_WRITE_EXCLUSIVE:
1720 case PR_EXCLUSIVE_ACCESS:
1722 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1724 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1726 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1728 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1735 static int sd_pr_command(struct block_device *bdev, u8 sa,
1736 u64 key, u64 sa_key, u8 type, u8 flags)
1738 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1739 struct scsi_sense_hdr sshdr;
1741 u8 cmd[16] = { 0, };
1742 u8 data[24] = { 0, };
1744 cmd[0] = PERSISTENT_RESERVE_OUT;
1747 put_unaligned_be32(sizeof(data), &cmd[5]);
1749 put_unaligned_be64(key, &data[0]);
1750 put_unaligned_be64(sa_key, &data[8]);
1753 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1754 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1756 if (driver_byte(result) == DRIVER_SENSE &&
1757 scsi_sense_valid(&sshdr)) {
1758 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1759 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1765 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1768 if (flags & ~PR_FL_IGNORE_KEY)
1770 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1771 old_key, new_key, 0,
1772 (1 << 0) /* APTPL */);
1775 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1780 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1783 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1785 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1788 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1789 enum pr_type type, bool abort)
1791 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1792 sd_pr_type(type), 0);
1795 static int sd_pr_clear(struct block_device *bdev, u64 key)
1797 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1800 static const struct pr_ops sd_pr_ops = {
1801 .pr_register = sd_pr_register,
1802 .pr_reserve = sd_pr_reserve,
1803 .pr_release = sd_pr_release,
1804 .pr_preempt = sd_pr_preempt,
1805 .pr_clear = sd_pr_clear,
1808 static const struct block_device_operations sd_fops = {
1809 .owner = THIS_MODULE,
1811 .release = sd_release,
1813 .getgeo = sd_getgeo,
1814 #ifdef CONFIG_COMPAT
1815 .compat_ioctl = sd_compat_ioctl,
1817 .check_events = sd_check_events,
1818 .revalidate_disk = sd_revalidate_disk,
1819 .unlock_native_capacity = sd_unlock_native_capacity,
1820 .report_zones = sd_zbc_report_zones,
1821 .pr_ops = &sd_pr_ops,
1825 * sd_eh_reset - reset error handling callback
1826 * @scmd: sd-issued command that has failed
1828 * This function is called by the SCSI midlayer before starting
1829 * SCSI EH. When counting medium access failures we have to be
1830 * careful to register it only only once per device and SCSI EH run;
1831 * there might be several timed out commands which will cause the
1832 * 'max_medium_access_timeouts' counter to trigger after the first
1833 * SCSI EH run already and set the device to offline.
1834 * So this function resets the internal counter before starting SCSI EH.
1836 static void sd_eh_reset(struct scsi_cmnd *scmd)
1838 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1840 /* New SCSI EH run, reset gate variable */
1841 sdkp->ignore_medium_access_errors = false;
1845 * sd_eh_action - error handling callback
1846 * @scmd: sd-issued command that has failed
1847 * @eh_disp: The recovery disposition suggested by the midlayer
1849 * This function is called by the SCSI midlayer upon completion of an
1850 * error test command (currently TEST UNIT READY). The result of sending
1851 * the eh command is passed in eh_disp. We're looking for devices that
1852 * fail medium access commands but are OK with non access commands like
1853 * test unit ready (so wrongly see the device as having a successful
1856 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1858 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1859 struct scsi_device *sdev = scmd->device;
1861 if (!scsi_device_online(sdev) ||
1862 !scsi_medium_access_command(scmd) ||
1863 host_byte(scmd->result) != DID_TIME_OUT ||
1868 * The device has timed out executing a medium access command.
1869 * However, the TEST UNIT READY command sent during error
1870 * handling completed successfully. Either the device is in the
1871 * process of recovering or has it suffered an internal failure
1872 * that prevents access to the storage medium.
1874 if (!sdkp->ignore_medium_access_errors) {
1875 sdkp->medium_access_timed_out++;
1876 sdkp->ignore_medium_access_errors = true;
1880 * If the device keeps failing read/write commands but TEST UNIT
1881 * READY always completes successfully we assume that medium
1882 * access is no longer possible and take the device offline.
1884 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1885 scmd_printk(KERN_ERR, scmd,
1886 "Medium access timeout failure. Offlining disk!\n");
1887 mutex_lock(&sdev->state_mutex);
1888 scsi_device_set_state(sdev, SDEV_OFFLINE);
1889 mutex_unlock(&sdev->state_mutex);
1897 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1899 struct request *req = scmd->request;
1900 struct scsi_device *sdev = scmd->device;
1901 unsigned int transferred, good_bytes;
1902 u64 start_lba, end_lba, bad_lba;
1905 * Some commands have a payload smaller than the device logical
1906 * block size (e.g. INQUIRY on a 4K disk).
1908 if (scsi_bufflen(scmd) <= sdev->sector_size)
1911 /* Check if we have a 'bad_lba' information */
1912 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1913 SCSI_SENSE_BUFFERSIZE,
1918 * If the bad lba was reported incorrectly, we have no idea where
1921 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1922 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1923 if (bad_lba < start_lba || bad_lba >= end_lba)
1927 * resid is optional but mostly filled in. When it's unused,
1928 * its value is zero, so we assume the whole buffer transferred
1930 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1932 /* This computation should always be done in terms of the
1933 * resolution of the device's medium.
1935 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1937 return min(good_bytes, transferred);
1941 * sd_done - bottom half handler: called when the lower level
1942 * driver has completed (successfully or otherwise) a scsi command.
1943 * @SCpnt: mid-level's per command structure.
1945 * Note: potentially run from within an ISR. Must not block.
1947 static int sd_done(struct scsi_cmnd *SCpnt)
1949 int result = SCpnt->result;
1950 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1951 unsigned int sector_size = SCpnt->device->sector_size;
1953 struct scsi_sense_hdr sshdr;
1954 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1955 struct request *req = SCpnt->request;
1956 int sense_valid = 0;
1957 int sense_deferred = 0;
1959 switch (req_op(req)) {
1960 case REQ_OP_DISCARD:
1961 case REQ_OP_WRITE_ZEROES:
1962 case REQ_OP_WRITE_SAME:
1963 case REQ_OP_ZONE_RESET:
1964 case REQ_OP_ZONE_RESET_ALL:
1966 good_bytes = blk_rq_bytes(req);
1967 scsi_set_resid(SCpnt, 0);
1970 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1975 * In case of bogus fw or device, we could end up having
1976 * an unaligned partial completion. Check this here and force
1979 resid = scsi_get_resid(SCpnt);
1980 if (resid & (sector_size - 1)) {
1981 sd_printk(KERN_INFO, sdkp,
1982 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1983 resid, sector_size);
1984 resid = min(scsi_bufflen(SCpnt),
1985 round_up(resid, sector_size));
1986 scsi_set_resid(SCpnt, resid);
1991 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1993 sense_deferred = scsi_sense_is_deferred(&sshdr);
1995 sdkp->medium_access_timed_out = 0;
1997 if (driver_byte(result) != DRIVER_SENSE &&
1998 (!sense_valid || sense_deferred))
2001 switch (sshdr.sense_key) {
2002 case HARDWARE_ERROR:
2004 good_bytes = sd_completed_bytes(SCpnt);
2006 case RECOVERED_ERROR:
2007 good_bytes = scsi_bufflen(SCpnt);
2010 /* This indicates a false check condition, so ignore it. An
2011 * unknown amount of data was transferred so treat it as an
2015 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2017 case ABORTED_COMMAND:
2018 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2019 good_bytes = sd_completed_bytes(SCpnt);
2021 case ILLEGAL_REQUEST:
2022 switch (sshdr.asc) {
2023 case 0x10: /* DIX: Host detected corruption */
2024 good_bytes = sd_completed_bytes(SCpnt);
2026 case 0x20: /* INVALID COMMAND OPCODE */
2027 case 0x24: /* INVALID FIELD IN CDB */
2028 switch (SCpnt->cmnd[0]) {
2030 sd_config_discard(sdkp, SD_LBP_DISABLE);
2034 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2035 sd_config_discard(sdkp, SD_LBP_DISABLE);
2037 sdkp->device->no_write_same = 1;
2038 sd_config_write_same(sdkp);
2039 req->rq_flags |= RQF_QUIET;
2050 if (sd_is_zoned(sdkp))
2051 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2053 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2054 "sd_done: completed %d of %d bytes\n",
2055 good_bytes, scsi_bufflen(SCpnt)));
2057 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2059 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2060 good_bytes / scsi_prot_interval(SCpnt));
2066 * spinup disk - called only in sd_revalidate_disk()
2069 sd_spinup_disk(struct scsi_disk *sdkp)
2071 unsigned char cmd[10];
2072 unsigned long spintime_expire = 0;
2073 int retries, spintime;
2074 unsigned int the_result;
2075 struct scsi_sense_hdr sshdr;
2076 int sense_valid = 0;
2080 /* Spin up drives, as required. Only do this at boot time */
2081 /* Spinup needs to be done for module loads too. */
2086 cmd[0] = TEST_UNIT_READY;
2087 memset((void *) &cmd[1], 0, 9);
2089 the_result = scsi_execute_req(sdkp->device, cmd,
2092 SD_MAX_RETRIES, NULL);
2095 * If the drive has indicated to us that it
2096 * doesn't have any media in it, don't bother
2097 * with any more polling.
2099 if (media_not_present(sdkp, &sshdr))
2103 sense_valid = scsi_sense_valid(&sshdr);
2105 } while (retries < 3 &&
2106 (!scsi_status_is_good(the_result) ||
2107 ((driver_byte(the_result) == DRIVER_SENSE) &&
2108 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2110 if (driver_byte(the_result) != DRIVER_SENSE) {
2111 /* no sense, TUR either succeeded or failed
2112 * with a status error */
2113 if(!spintime && !scsi_status_is_good(the_result)) {
2114 sd_print_result(sdkp, "Test Unit Ready failed",
2121 * The device does not want the automatic start to be issued.
2123 if (sdkp->device->no_start_on_add)
2126 if (sense_valid && sshdr.sense_key == NOT_READY) {
2127 if (sshdr.asc == 4 && sshdr.ascq == 3)
2128 break; /* manual intervention required */
2129 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2130 break; /* standby */
2131 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2132 break; /* unavailable */
2133 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2134 break; /* sanitize in progress */
2136 * Issue command to spin up drive when not ready
2139 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2140 cmd[0] = START_STOP;
2141 cmd[1] = 1; /* Return immediately */
2142 memset((void *) &cmd[2], 0, 8);
2143 cmd[4] = 1; /* Start spin cycle */
2144 if (sdkp->device->start_stop_pwr_cond)
2146 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2148 SD_TIMEOUT, SD_MAX_RETRIES,
2150 spintime_expire = jiffies + 100 * HZ;
2153 /* Wait 1 second for next try */
2155 printk(KERN_CONT ".");
2158 * Wait for USB flash devices with slow firmware.
2159 * Yes, this sense key/ASC combination shouldn't
2160 * occur here. It's characteristic of these devices.
2162 } else if (sense_valid &&
2163 sshdr.sense_key == UNIT_ATTENTION &&
2164 sshdr.asc == 0x28) {
2166 spintime_expire = jiffies + 5 * HZ;
2169 /* Wait 1 second for next try */
2172 /* we don't understand the sense code, so it's
2173 * probably pointless to loop */
2175 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2176 sd_print_sense_hdr(sdkp, &sshdr);
2181 } while (spintime && time_before_eq(jiffies, spintime_expire));
2184 if (scsi_status_is_good(the_result))
2185 printk(KERN_CONT "ready\n");
2187 printk(KERN_CONT "not responding...\n");
2192 * Determine whether disk supports Data Integrity Field.
2194 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2196 struct scsi_device *sdp = sdkp->device;
2200 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2203 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2205 if (type > T10_PI_TYPE3_PROTECTION)
2207 else if (scsi_host_dif_capable(sdp->host, type))
2210 if (sdkp->first_scan || type != sdkp->protection_type)
2213 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2214 " protection type %u. Disabling disk!\n",
2218 sd_printk(KERN_NOTICE, sdkp,
2219 "Enabling DIF Type %u protection\n", type);
2222 sd_printk(KERN_NOTICE, sdkp,
2223 "Disabling DIF Type %u protection\n", type);
2227 sdkp->protection_type = type;
2232 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2233 struct scsi_sense_hdr *sshdr, int sense_valid,
2236 if (driver_byte(the_result) == DRIVER_SENSE)
2237 sd_print_sense_hdr(sdkp, sshdr);
2239 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2242 * Set dirty bit for removable devices if not ready -
2243 * sometimes drives will not report this properly.
2245 if (sdp->removable &&
2246 sense_valid && sshdr->sense_key == NOT_READY)
2247 set_media_not_present(sdkp);
2250 * We used to set media_present to 0 here to indicate no media
2251 * in the drive, but some drives fail read capacity even with
2252 * media present, so we can't do that.
2254 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2258 #if RC16_LEN > SD_BUF_SIZE
2259 #error RC16_LEN must not be more than SD_BUF_SIZE
2262 #define READ_CAPACITY_RETRIES_ON_RESET 10
2264 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2265 unsigned char *buffer)
2267 unsigned char cmd[16];
2268 struct scsi_sense_hdr sshdr;
2269 int sense_valid = 0;
2271 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2272 unsigned int alignment;
2273 unsigned long long lba;
2274 unsigned sector_size;
2276 if (sdp->no_read_capacity_16)
2281 cmd[0] = SERVICE_ACTION_IN_16;
2282 cmd[1] = SAI_READ_CAPACITY_16;
2284 memset(buffer, 0, RC16_LEN);
2286 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2287 buffer, RC16_LEN, &sshdr,
2288 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2290 if (media_not_present(sdkp, &sshdr))
2294 sense_valid = scsi_sense_valid(&sshdr);
2296 sshdr.sense_key == ILLEGAL_REQUEST &&
2297 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2299 /* Invalid Command Operation Code or
2300 * Invalid Field in CDB, just retry
2301 * silently with RC10 */
2304 sshdr.sense_key == UNIT_ATTENTION &&
2305 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2306 /* Device reset might occur several times,
2307 * give it one more chance */
2308 if (--reset_retries > 0)
2313 } while (the_result && retries);
2316 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2317 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2321 sector_size = get_unaligned_be32(&buffer[8]);
2322 lba = get_unaligned_be64(&buffer[0]);
2324 if (sd_read_protection_type(sdkp, buffer) < 0) {
2329 /* Logical blocks per physical block exponent */
2330 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2333 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2335 /* Lowest aligned logical block */
2336 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2337 blk_queue_alignment_offset(sdp->request_queue, alignment);
2338 if (alignment && sdkp->first_scan)
2339 sd_printk(KERN_NOTICE, sdkp,
2340 "physical block alignment offset: %u\n", alignment);
2342 if (buffer[14] & 0x80) { /* LBPME */
2345 if (buffer[14] & 0x40) /* LBPRZ */
2348 sd_config_discard(sdkp, SD_LBP_WS16);
2351 sdkp->capacity = lba + 1;
2355 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2356 unsigned char *buffer)
2358 unsigned char cmd[16];
2359 struct scsi_sense_hdr sshdr;
2360 int sense_valid = 0;
2362 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2364 unsigned sector_size;
2367 cmd[0] = READ_CAPACITY;
2368 memset(&cmd[1], 0, 9);
2369 memset(buffer, 0, 8);
2371 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2373 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2375 if (media_not_present(sdkp, &sshdr))
2379 sense_valid = scsi_sense_valid(&sshdr);
2381 sshdr.sense_key == UNIT_ATTENTION &&
2382 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2383 /* Device reset might occur several times,
2384 * give it one more chance */
2385 if (--reset_retries > 0)
2390 } while (the_result && retries);
2393 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2394 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2398 sector_size = get_unaligned_be32(&buffer[4]);
2399 lba = get_unaligned_be32(&buffer[0]);
2401 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2402 /* Some buggy (usb cardreader) devices return an lba of
2403 0xffffffff when the want to report a size of 0 (with
2404 which they really mean no media is present) */
2406 sdkp->physical_block_size = sector_size;
2410 sdkp->capacity = lba + 1;
2411 sdkp->physical_block_size = sector_size;
2415 static int sd_try_rc16_first(struct scsi_device *sdp)
2417 if (sdp->host->max_cmd_len < 16)
2419 if (sdp->try_rc_10_first)
2421 if (sdp->scsi_level > SCSI_SPC_2)
2423 if (scsi_device_protection(sdp))
2429 * read disk capacity
2432 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2435 struct scsi_device *sdp = sdkp->device;
2437 if (sd_try_rc16_first(sdp)) {
2438 sector_size = read_capacity_16(sdkp, sdp, buffer);
2439 if (sector_size == -EOVERFLOW)
2441 if (sector_size == -ENODEV)
2443 if (sector_size < 0)
2444 sector_size = read_capacity_10(sdkp, sdp, buffer);
2445 if (sector_size < 0)
2448 sector_size = read_capacity_10(sdkp, sdp, buffer);
2449 if (sector_size == -EOVERFLOW)
2451 if (sector_size < 0)
2453 if ((sizeof(sdkp->capacity) > 4) &&
2454 (sdkp->capacity > 0xffffffffULL)) {
2455 int old_sector_size = sector_size;
2456 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2457 "Trying to use READ CAPACITY(16).\n");
2458 sector_size = read_capacity_16(sdkp, sdp, buffer);
2459 if (sector_size < 0) {
2460 sd_printk(KERN_NOTICE, sdkp,
2461 "Using 0xffffffff as device size\n");
2462 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2463 sector_size = old_sector_size;
2466 /* Remember that READ CAPACITY(16) succeeded */
2467 sdp->try_rc_10_first = 0;
2471 /* Some devices are known to return the total number of blocks,
2472 * not the highest block number. Some devices have versions
2473 * which do this and others which do not. Some devices we might
2474 * suspect of doing this but we don't know for certain.
2476 * If we know the reported capacity is wrong, decrement it. If
2477 * we can only guess, then assume the number of blocks is even
2478 * (usually true but not always) and err on the side of lowering
2481 if (sdp->fix_capacity ||
2482 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2483 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2484 "from its reported value: %llu\n",
2485 (unsigned long long) sdkp->capacity);
2490 if (sector_size == 0) {
2492 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2496 if (sector_size != 512 &&
2497 sector_size != 1024 &&
2498 sector_size != 2048 &&
2499 sector_size != 4096) {
2500 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2503 * The user might want to re-format the drive with
2504 * a supported sectorsize. Once this happens, it
2505 * would be relatively trivial to set the thing up.
2506 * For this reason, we leave the thing in the table.
2510 * set a bogus sector size so the normal read/write
2511 * logic in the block layer will eventually refuse any
2512 * request on this device without tripping over power
2513 * of two sector size assumptions
2517 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2518 blk_queue_physical_block_size(sdp->request_queue,
2519 sdkp->physical_block_size);
2520 sdkp->device->sector_size = sector_size;
2522 if (sdkp->capacity > 0xffffffff)
2523 sdp->use_16_for_rw = 1;
2528 * Print disk capacity
2531 sd_print_capacity(struct scsi_disk *sdkp,
2532 sector_t old_capacity)
2534 int sector_size = sdkp->device->sector_size;
2535 char cap_str_2[10], cap_str_10[10];
2537 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2540 string_get_size(sdkp->capacity, sector_size,
2541 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2542 string_get_size(sdkp->capacity, sector_size,
2543 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2545 sd_printk(KERN_NOTICE, sdkp,
2546 "%llu %d-byte logical blocks: (%s/%s)\n",
2547 (unsigned long long)sdkp->capacity,
2548 sector_size, cap_str_10, cap_str_2);
2550 if (sdkp->physical_block_size != sector_size)
2551 sd_printk(KERN_NOTICE, sdkp,
2552 "%u-byte physical blocks\n",
2553 sdkp->physical_block_size);
2555 sd_zbc_print_zones(sdkp);
2558 /* called with buffer of length 512 */
2560 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2561 unsigned char *buffer, int len, struct scsi_mode_data *data,
2562 struct scsi_sense_hdr *sshdr)
2564 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2565 SD_TIMEOUT, SD_MAX_RETRIES, data,
2570 * read write protect setting, if possible - called only in sd_revalidate_disk()
2571 * called with buffer of length SD_BUF_SIZE
2574 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2577 struct scsi_device *sdp = sdkp->device;
2578 struct scsi_mode_data data;
2579 int old_wp = sdkp->write_prot;
2581 set_disk_ro(sdkp->disk, 0);
2582 if (sdp->skip_ms_page_3f) {
2583 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2587 if (sdp->use_192_bytes_for_3f) {
2588 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2591 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2592 * We have to start carefully: some devices hang if we ask
2593 * for more than is available.
2595 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2598 * Second attempt: ask for page 0 When only page 0 is
2599 * implemented, a request for page 3F may return Sense Key
2600 * 5: Illegal Request, Sense Code 24: Invalid field in
2603 if (!scsi_status_is_good(res))
2604 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2607 * Third attempt: ask 255 bytes, as we did earlier.
2609 if (!scsi_status_is_good(res))
2610 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2614 if (!scsi_status_is_good(res)) {
2615 sd_first_printk(KERN_WARNING, sdkp,
2616 "Test WP failed, assume Write Enabled\n");
2618 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2619 set_disk_ro(sdkp->disk, sdkp->write_prot);
2620 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2621 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2622 sdkp->write_prot ? "on" : "off");
2623 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2629 * sd_read_cache_type - called only from sd_revalidate_disk()
2630 * called with buffer of length SD_BUF_SIZE
2633 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2636 struct scsi_device *sdp = sdkp->device;
2641 struct scsi_mode_data data;
2642 struct scsi_sense_hdr sshdr;
2643 int old_wce = sdkp->WCE;
2644 int old_rcd = sdkp->RCD;
2645 int old_dpofua = sdkp->DPOFUA;
2648 if (sdkp->cache_override)
2652 if (sdp->skip_ms_page_8) {
2653 if (sdp->type == TYPE_RBC)
2656 if (sdp->skip_ms_page_3f)
2659 if (sdp->use_192_bytes_for_3f)
2663 } else if (sdp->type == TYPE_RBC) {
2671 /* cautiously ask */
2672 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2675 if (!scsi_status_is_good(res))
2678 if (!data.header_length) {
2681 sd_first_printk(KERN_ERR, sdkp,
2682 "Missing header in MODE_SENSE response\n");
2685 /* that went OK, now ask for the proper length */
2689 * We're only interested in the first three bytes, actually.
2690 * But the data cache page is defined for the first 20.
2694 else if (len > SD_BUF_SIZE) {
2695 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2696 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2699 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2703 if (len > first_len)
2704 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2707 if (scsi_status_is_good(res)) {
2708 int offset = data.header_length + data.block_descriptor_length;
2710 while (offset < len) {
2711 u8 page_code = buffer[offset] & 0x3F;
2712 u8 spf = buffer[offset] & 0x40;
2714 if (page_code == 8 || page_code == 6) {
2715 /* We're interested only in the first 3 bytes.
2717 if (len - offset <= 2) {
2718 sd_first_printk(KERN_ERR, sdkp,
2719 "Incomplete mode parameter "
2723 modepage = page_code;
2727 /* Go to the next page */
2728 if (spf && len - offset > 3)
2729 offset += 4 + (buffer[offset+2] << 8) +
2731 else if (!spf && len - offset > 1)
2732 offset += 2 + buffer[offset+1];
2734 sd_first_printk(KERN_ERR, sdkp,
2736 "parameter data\n");
2742 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2746 if (modepage == 8) {
2747 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2748 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2750 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2754 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2755 if (sdp->broken_fua) {
2756 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2758 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2759 !sdkp->device->use_16_for_rw) {
2760 sd_first_printk(KERN_NOTICE, sdkp,
2761 "Uses READ/WRITE(6), disabling FUA\n");
2765 /* No cache flush allowed for write protected devices */
2766 if (sdkp->WCE && sdkp->write_prot)
2769 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2770 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2771 sd_printk(KERN_NOTICE, sdkp,
2772 "Write cache: %s, read cache: %s, %s\n",
2773 sdkp->WCE ? "enabled" : "disabled",
2774 sdkp->RCD ? "disabled" : "enabled",
2775 sdkp->DPOFUA ? "supports DPO and FUA"
2776 : "doesn't support DPO or FUA");
2782 if (scsi_sense_valid(&sshdr) &&
2783 sshdr.sense_key == ILLEGAL_REQUEST &&
2784 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2785 /* Invalid field in CDB */
2786 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2788 sd_first_printk(KERN_ERR, sdkp,
2789 "Asking for cache data failed\n");
2792 if (sdp->wce_default_on) {
2793 sd_first_printk(KERN_NOTICE, sdkp,
2794 "Assuming drive cache: write back\n");
2797 sd_first_printk(KERN_ERR, sdkp,
2798 "Assuming drive cache: write through\n");
2806 * The ATO bit indicates whether the DIF application tag is available
2807 * for use by the operating system.
2809 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2812 struct scsi_device *sdp = sdkp->device;
2813 struct scsi_mode_data data;
2814 struct scsi_sense_hdr sshdr;
2816 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2819 if (sdkp->protection_type == 0)
2822 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2823 SD_MAX_RETRIES, &data, &sshdr);
2825 if (!scsi_status_is_good(res) || !data.header_length ||
2827 sd_first_printk(KERN_WARNING, sdkp,
2828 "getting Control mode page failed, assume no ATO\n");
2830 if (scsi_sense_valid(&sshdr))
2831 sd_print_sense_hdr(sdkp, &sshdr);
2836 offset = data.header_length + data.block_descriptor_length;
2838 if ((buffer[offset] & 0x3f) != 0x0a) {
2839 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2843 if ((buffer[offset + 5] & 0x80) == 0)
2852 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2853 * @sdkp: disk to query
2855 static void sd_read_block_limits(struct scsi_disk *sdkp)
2857 unsigned int sector_sz = sdkp->device->sector_size;
2858 const int vpd_len = 64;
2859 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2862 /* Block Limits VPD */
2863 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2866 blk_queue_io_min(sdkp->disk->queue,
2867 get_unaligned_be16(&buffer[6]) * sector_sz);
2869 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2870 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2872 if (buffer[3] == 0x3c) {
2873 unsigned int lba_count, desc_count;
2875 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2880 lba_count = get_unaligned_be32(&buffer[20]);
2881 desc_count = get_unaligned_be32(&buffer[24]);
2883 if (lba_count && desc_count)
2884 sdkp->max_unmap_blocks = lba_count;
2886 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2888 if (buffer[32] & 0x80)
2889 sdkp->unmap_alignment =
2890 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2892 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2894 if (sdkp->max_unmap_blocks)
2895 sd_config_discard(sdkp, SD_LBP_UNMAP);
2897 sd_config_discard(sdkp, SD_LBP_WS16);
2899 } else { /* LBP VPD page tells us what to use */
2900 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2901 sd_config_discard(sdkp, SD_LBP_UNMAP);
2902 else if (sdkp->lbpws)
2903 sd_config_discard(sdkp, SD_LBP_WS16);
2904 else if (sdkp->lbpws10)
2905 sd_config_discard(sdkp, SD_LBP_WS10);
2907 sd_config_discard(sdkp, SD_LBP_DISABLE);
2916 * sd_read_block_characteristics - Query block dev. characteristics
2917 * @sdkp: disk to query
2919 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2921 struct request_queue *q = sdkp->disk->queue;
2922 unsigned char *buffer;
2924 const int vpd_len = 64;
2926 buffer = kmalloc(vpd_len, GFP_KERNEL);
2929 /* Block Device Characteristics VPD */
2930 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2933 rot = get_unaligned_be16(&buffer[4]);
2936 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2937 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2940 if (sdkp->device->type == TYPE_ZBC) {
2942 q->limits.zoned = BLK_ZONED_HM;
2944 sdkp->zoned = (buffer[8] >> 4) & 3;
2945 if (sdkp->zoned == 1)
2947 q->limits.zoned = BLK_ZONED_HA;
2950 * Treat drive-managed devices as
2951 * regular block devices.
2953 q->limits.zoned = BLK_ZONED_NONE;
2955 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2956 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2957 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2964 * sd_read_block_provisioning - Query provisioning VPD page
2965 * @sdkp: disk to query
2967 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2969 unsigned char *buffer;
2970 const int vpd_len = 8;
2972 if (sdkp->lbpme == 0)
2975 buffer = kmalloc(vpd_len, GFP_KERNEL);
2977 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2981 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2982 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2983 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2989 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2991 struct scsi_device *sdev = sdkp->device;
2993 if (sdev->host->no_write_same) {
2994 sdev->no_write_same = 1;
2999 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3000 /* too large values might cause issues with arcmsr */
3001 int vpd_buf_len = 64;
3003 sdev->no_report_opcodes = 1;
3005 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3006 * CODES is unsupported and the device has an ATA
3007 * Information VPD page (SAT).
3009 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3010 sdev->no_write_same = 1;
3013 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3016 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3020 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3022 struct scsi_device *sdev = sdkp->device;
3024 if (!sdev->security_supported)
3027 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3028 SECURITY_PROTOCOL_IN) == 1 &&
3029 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3030 SECURITY_PROTOCOL_OUT) == 1)
3035 * Determine the device's preferred I/O size for reads and writes
3036 * unless the reported value is unreasonably small, large, not a
3037 * multiple of the physical block size, or simply garbage.
3039 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3040 unsigned int dev_max)
3042 struct scsi_device *sdp = sdkp->device;
3043 unsigned int opt_xfer_bytes =
3044 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3046 if (sdkp->opt_xfer_blocks == 0)
3049 if (sdkp->opt_xfer_blocks > dev_max) {
3050 sd_first_printk(KERN_WARNING, sdkp,
3051 "Optimal transfer size %u logical blocks " \
3052 "> dev_max (%u logical blocks)\n",
3053 sdkp->opt_xfer_blocks, dev_max);
3057 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3058 sd_first_printk(KERN_WARNING, sdkp,
3059 "Optimal transfer size %u logical blocks " \
3060 "> sd driver limit (%u logical blocks)\n",
3061 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3065 if (opt_xfer_bytes < PAGE_SIZE) {
3066 sd_first_printk(KERN_WARNING, sdkp,
3067 "Optimal transfer size %u bytes < " \
3068 "PAGE_SIZE (%u bytes)\n",
3069 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3073 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3074 sd_first_printk(KERN_WARNING, sdkp,
3075 "Optimal transfer size %u bytes not a " \
3076 "multiple of physical block size (%u bytes)\n",
3077 opt_xfer_bytes, sdkp->physical_block_size);
3081 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3087 * sd_revalidate_disk - called the first time a new disk is seen,
3088 * performs disk spin up, read_capacity, etc.
3089 * @disk: struct gendisk we care about
3091 static int sd_revalidate_disk(struct gendisk *disk)
3093 struct scsi_disk *sdkp = scsi_disk(disk);
3094 struct scsi_device *sdp = sdkp->device;
3095 struct request_queue *q = sdkp->disk->queue;
3096 sector_t old_capacity = sdkp->capacity;
3097 unsigned char *buffer;
3098 unsigned int dev_max, rw_max;
3100 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3101 "sd_revalidate_disk\n"));
3104 * If the device is offline, don't try and read capacity or any
3105 * of the other niceties.
3107 if (!scsi_device_online(sdp))
3110 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3112 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3113 "allocation failure.\n");
3117 sd_spinup_disk(sdkp);
3120 * Without media there is no reason to ask; moreover, some devices
3121 * react badly if we do.
3123 if (sdkp->media_present) {
3124 sd_read_capacity(sdkp, buffer);
3127 * set the default to rotational. All non-rotational devices
3128 * support the block characteristics VPD page, which will
3129 * cause this to be updated correctly and any device which
3130 * doesn't support it should be treated as rotational.
3132 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3133 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3135 if (scsi_device_supports_vpd(sdp)) {
3136 sd_read_block_provisioning(sdkp);
3137 sd_read_block_limits(sdkp);
3138 sd_read_block_characteristics(sdkp);
3139 sd_zbc_read_zones(sdkp, buffer);
3142 sd_print_capacity(sdkp, old_capacity);
3144 sd_read_write_protect_flag(sdkp, buffer);
3145 sd_read_cache_type(sdkp, buffer);
3146 sd_read_app_tag_own(sdkp, buffer);
3147 sd_read_write_same(sdkp, buffer);
3148 sd_read_security(sdkp, buffer);
3152 * We now have all cache related info, determine how we deal
3153 * with flush requests.
3155 sd_set_flush_flag(sdkp);
3157 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3158 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3160 /* Some devices report a maximum block count for READ/WRITE requests. */
3161 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3162 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3164 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3165 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3166 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3168 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3169 (sector_t)BLK_DEF_MAX_SECTORS);
3171 /* Do not exceed controller limit */
3172 rw_max = min(rw_max, queue_max_hw_sectors(q));
3175 * Only update max_sectors if previously unset or if the current value
3176 * exceeds the capabilities of the hardware.
3178 if (sdkp->first_scan ||
3179 q->limits.max_sectors > q->limits.max_dev_sectors ||
3180 q->limits.max_sectors > q->limits.max_hw_sectors)
3181 q->limits.max_sectors = rw_max;
3183 sdkp->first_scan = 0;
3185 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3186 sd_config_write_same(sdkp);
3194 * sd_unlock_native_capacity - unlock native capacity
3195 * @disk: struct gendisk to set capacity for
3197 * Block layer calls this function if it detects that partitions
3198 * on @disk reach beyond the end of the device. If the SCSI host
3199 * implements ->unlock_native_capacity() method, it's invoked to
3200 * give it a chance to adjust the device capacity.
3203 * Defined by block layer. Might sleep.
3205 static void sd_unlock_native_capacity(struct gendisk *disk)
3207 struct scsi_device *sdev = scsi_disk(disk)->device;
3209 if (sdev->host->hostt->unlock_native_capacity)
3210 sdev->host->hostt->unlock_native_capacity(sdev);
3214 * sd_format_disk_name - format disk name
3215 * @prefix: name prefix - ie. "sd" for SCSI disks
3216 * @index: index of the disk to format name for
3217 * @buf: output buffer
3218 * @buflen: length of the output buffer
3220 * SCSI disk names starts at sda. The 26th device is sdz and the
3221 * 27th is sdaa. The last one for two lettered suffix is sdzz
3222 * which is followed by sdaaa.
3224 * This is basically 26 base counting with one extra 'nil' entry
3225 * at the beginning from the second digit on and can be
3226 * determined using similar method as 26 base conversion with the
3227 * index shifted -1 after each digit is computed.
3233 * 0 on success, -errno on failure.
3235 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3237 const int base = 'z' - 'a' + 1;
3238 char *begin = buf + strlen(prefix);
3239 char *end = buf + buflen;
3249 *--p = 'a' + (index % unit);
3250 index = (index / unit) - 1;
3251 } while (index >= 0);
3253 memmove(begin, p, end - p);
3254 memcpy(buf, prefix, strlen(prefix));
3260 * sd_probe - called during driver initialization and whenever a
3261 * new scsi device is attached to the system. It is called once
3262 * for each scsi device (not just disks) present.
3263 * @dev: pointer to device object
3265 * Returns 0 if successful (or not interested in this scsi device
3266 * (e.g. scanner)); 1 when there is an error.
3268 * Note: this function is invoked from the scsi mid-level.
3269 * This function sets up the mapping between a given
3270 * <host,channel,id,lun> (found in sdp) and new device name
3271 * (e.g. /dev/sda). More precisely it is the block device major
3272 * and minor number that is chosen here.
3274 * Assume sd_probe is not re-entrant (for time being)
3275 * Also think about sd_probe() and sd_remove() running coincidentally.
3277 static int sd_probe(struct device *dev)
3279 struct scsi_device *sdp = to_scsi_device(dev);
3280 struct scsi_disk *sdkp;
3285 scsi_autopm_get_device(sdp);
3287 if (sdp->type != TYPE_DISK &&
3288 sdp->type != TYPE_ZBC &&
3289 sdp->type != TYPE_MOD &&
3290 sdp->type != TYPE_RBC)
3293 #ifndef CONFIG_BLK_DEV_ZONED
3294 if (sdp->type == TYPE_ZBC)
3297 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3301 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3305 gd = alloc_disk(SD_MINORS);
3309 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3311 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3315 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3317 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3318 goto out_free_index;
3322 sdkp->driver = &sd_template;
3324 sdkp->index = index;
3325 atomic_set(&sdkp->openers, 0);
3326 atomic_set(&sdkp->device->ioerr_cnt, 0);
3328 if (!sdp->request_queue->rq_timeout) {
3329 if (sdp->type != TYPE_MOD)
3330 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3332 blk_queue_rq_timeout(sdp->request_queue,
3336 device_initialize(&sdkp->dev);
3337 sdkp->dev.parent = dev;
3338 sdkp->dev.class = &sd_disk_class;
3339 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3341 error = device_add(&sdkp->dev);
3343 goto out_free_index;
3346 dev_set_drvdata(dev, sdkp);
3348 gd->major = sd_major((index & 0xf0) >> 4);
3349 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3351 gd->fops = &sd_fops;
3352 gd->private_data = &sdkp->driver;
3353 gd->queue = sdkp->device->request_queue;
3355 /* defaults, until the device tells us otherwise */
3356 sdp->sector_size = 512;
3358 sdkp->media_present = 1;
3359 sdkp->write_prot = 0;
3360 sdkp->cache_override = 0;
3364 sdkp->first_scan = 1;
3365 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3367 sd_revalidate_disk(gd);
3369 gd->flags = GENHD_FL_EXT_DEVT;
3370 if (sdp->removable) {
3371 gd->flags |= GENHD_FL_REMOVABLE;
3372 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3373 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3376 blk_pm_runtime_init(sdp->request_queue, dev);
3377 device_add_disk(dev, gd, NULL);
3379 sd_dif_config_host(sdkp);
3381 sd_revalidate_disk(gd);
3383 if (sdkp->security) {
3384 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3386 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3389 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3390 sdp->removable ? "removable " : "");
3391 scsi_autopm_put_device(sdp);
3396 ida_free(&sd_index_ida, index);
3402 scsi_autopm_put_device(sdp);
3407 * sd_remove - called whenever a scsi disk (previously recognized by
3408 * sd_probe) is detached from the system. It is called (potentially
3409 * multiple times) during sd module unload.
3410 * @dev: pointer to device object
3412 * Note: this function is invoked from the scsi mid-level.
3413 * This function potentially frees up a device name (e.g. /dev/sdc)
3414 * that could be re-used by a subsequent sd_probe().
3415 * This function is not called when the built-in sd driver is "exit-ed".
3417 static int sd_remove(struct device *dev)
3419 struct scsi_disk *sdkp;
3422 sdkp = dev_get_drvdata(dev);
3423 devt = disk_devt(sdkp->disk);
3424 scsi_autopm_get_device(sdkp->device);
3426 async_synchronize_full_domain(&scsi_sd_pm_domain);
3427 device_del(&sdkp->dev);
3428 del_gendisk(sdkp->disk);
3431 free_opal_dev(sdkp->opal_dev);
3433 blk_register_region(devt, SD_MINORS, NULL,
3434 sd_default_probe, NULL, NULL);
3436 mutex_lock(&sd_ref_mutex);
3437 dev_set_drvdata(dev, NULL);
3438 put_device(&sdkp->dev);
3439 mutex_unlock(&sd_ref_mutex);
3445 * scsi_disk_release - Called to free the scsi_disk structure
3446 * @dev: pointer to embedded class device
3448 * sd_ref_mutex must be held entering this routine. Because it is
3449 * called on last put, you should always use the scsi_disk_get()
3450 * scsi_disk_put() helpers which manipulate the semaphore directly
3451 * and never do a direct put_device.
3453 static void scsi_disk_release(struct device *dev)
3455 struct scsi_disk *sdkp = to_scsi_disk(dev);
3456 struct gendisk *disk = sdkp->disk;
3457 struct request_queue *q = disk->queue;
3459 ida_free(&sd_index_ida, sdkp->index);
3462 * Wait until all requests that are in progress have completed.
3463 * This is necessary to avoid that e.g. scsi_end_request() crashes
3464 * due to clearing the disk->private_data pointer. Wait from inside
3465 * scsi_disk_release() instead of from sd_release() to avoid that
3466 * freezing and unfreezing the request queue affects user space I/O
3467 * in case multiple processes open a /dev/sd... node concurrently.
3469 blk_mq_freeze_queue(q);
3470 blk_mq_unfreeze_queue(q);
3472 disk->private_data = NULL;
3474 put_device(&sdkp->device->sdev_gendev);
3479 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3481 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3482 struct scsi_sense_hdr sshdr;
3483 struct scsi_device *sdp = sdkp->device;
3487 cmd[4] |= 1; /* START */
3489 if (sdp->start_stop_pwr_cond)
3490 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3492 if (!scsi_device_online(sdp))
3495 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3496 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3498 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3499 if (driver_byte(res) == DRIVER_SENSE)
3500 sd_print_sense_hdr(sdkp, &sshdr);
3501 if (scsi_sense_valid(&sshdr) &&
3502 /* 0x3a is medium not present */
3507 /* SCSI error codes must not go to the generic layer */
3515 * Send a SYNCHRONIZE CACHE instruction down to the device through
3516 * the normal SCSI command structure. Wait for the command to
3519 static void sd_shutdown(struct device *dev)
3521 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3524 return; /* this can happen */
3526 if (pm_runtime_suspended(dev))
3529 if (sdkp->WCE && sdkp->media_present) {
3530 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3531 sd_sync_cache(sdkp, NULL);
3534 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3535 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3536 sd_start_stop_device(sdkp, 0);
3540 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3542 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3543 struct scsi_sense_hdr sshdr;
3546 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3549 if (sdkp->WCE && sdkp->media_present) {
3550 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3551 ret = sd_sync_cache(sdkp, &sshdr);
3554 /* ignore OFFLINE device */
3558 if (!scsi_sense_valid(&sshdr) ||
3559 sshdr.sense_key != ILLEGAL_REQUEST)
3563 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3564 * doesn't support sync. There's not much to do and
3565 * suspend shouldn't fail.
3571 if (sdkp->device->manage_start_stop) {
3572 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3573 /* an error is not worth aborting a system sleep */
3574 ret = sd_start_stop_device(sdkp, 0);
3575 if (ignore_stop_errors)
3582 static int sd_suspend_system(struct device *dev)
3584 return sd_suspend_common(dev, true);
3587 static int sd_suspend_runtime(struct device *dev)
3589 return sd_suspend_common(dev, false);
3592 static int sd_resume(struct device *dev)
3594 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3597 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3600 if (!sdkp->device->manage_start_stop)
3603 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3604 ret = sd_start_stop_device(sdkp, 1);
3606 opal_unlock_from_suspend(sdkp->opal_dev);
3611 * init_sd - entry point for this driver (both when built in or when
3614 * Note: this function registers this driver with the scsi mid-level.
3616 static int __init init_sd(void)
3618 int majors = 0, i, err;
3620 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3622 for (i = 0; i < SD_MAJORS; i++) {
3623 if (register_blkdev(sd_major(i), "sd") != 0)
3626 blk_register_region(sd_major(i), SD_MINORS, NULL,
3627 sd_default_probe, NULL, NULL);
3633 err = class_register(&sd_disk_class);
3637 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3639 if (!sd_cdb_cache) {
3640 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3645 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3647 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3652 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3653 if (!sd_page_pool) {
3654 printk(KERN_ERR "sd: can't init discard page pool\n");
3659 err = scsi_register_driver(&sd_template.gendrv);
3661 goto err_out_driver;
3666 mempool_destroy(sd_page_pool);
3669 mempool_destroy(sd_cdb_pool);
3672 kmem_cache_destroy(sd_cdb_cache);
3675 class_unregister(&sd_disk_class);
3677 for (i = 0; i < SD_MAJORS; i++)
3678 unregister_blkdev(sd_major(i), "sd");
3683 * exit_sd - exit point for this driver (when it is a module).
3685 * Note: this function unregisters this driver from the scsi mid-level.
3687 static void __exit exit_sd(void)
3691 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3693 scsi_unregister_driver(&sd_template.gendrv);
3694 mempool_destroy(sd_cdb_pool);
3695 mempool_destroy(sd_page_pool);
3696 kmem_cache_destroy(sd_cdb_cache);
3698 class_unregister(&sd_disk_class);
3700 for (i = 0; i < SD_MAJORS; i++) {
3701 blk_unregister_region(sd_major(i), SD_MINORS);
3702 unregister_blkdev(sd_major(i), "sd");
3706 module_init(init_sd);
3707 module_exit(exit_sd);
3709 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3710 struct scsi_sense_hdr *sshdr)
3712 scsi_print_sense_hdr(sdkp->device,
3713 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3716 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3719 const char *hb_string = scsi_hostbyte_string(result);
3720 const char *db_string = scsi_driverbyte_string(result);
3722 if (hb_string || db_string)
3723 sd_printk(KERN_INFO, sdkp,
3724 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3725 hb_string ? hb_string : "invalid",
3726 db_string ? db_string : "invalid");
3728 sd_printk(KERN_INFO, sdkp,
3729 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3730 msg, host_byte(result), driver_byte(result));