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));
1169 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170 bool write = rq_data_dir(rq) == WRITE;
1171 unsigned char protect, fua;
1176 ret = scsi_init_io(cmd);
1177 if (ret != BLK_STS_OK)
1180 if (!scsi_device_online(sdp) || sdp->changed) {
1181 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1182 return BLK_STS_IOERR;
1185 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1186 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1187 return BLK_STS_IOERR;
1190 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1191 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1192 return BLK_STS_IOERR;
1196 * Some SD card readers can't handle accesses which touch the
1197 * last one or two logical blocks. Split accesses as needed.
1199 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1201 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1202 if (lba < threshold) {
1203 /* Access up to the threshold but not beyond */
1204 nr_blocks = threshold - lba;
1206 /* Access only a single logical block */
1211 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1212 dix = scsi_prot_sg_count(cmd);
1213 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1216 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1220 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1221 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1223 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1224 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1226 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1227 sdp->use_10_for_rw || protect) {
1228 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1231 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1235 if (unlikely(ret != BLK_STS_OK))
1239 * We shouldn't disconnect in the middle of a sector, so with a dumb
1240 * host adapter, it's safe to assume that we can at least transfer
1241 * this many bytes between each connect / disconnect.
1243 cmd->transfersize = sdp->sector_size;
1244 cmd->underflow = nr_blocks << 9;
1245 cmd->allowed = SD_MAX_RETRIES;
1246 cmd->sdb.length = nr_blocks * sdp->sector_size;
1249 scmd_printk(KERN_INFO, cmd,
1250 "%s: block=%llu, count=%d\n", __func__,
1251 (unsigned long long)blk_rq_pos(rq),
1252 blk_rq_sectors(rq)));
1254 scmd_printk(KERN_INFO, cmd,
1255 "%s %d/%u 512 byte blocks.\n",
1256 write ? "writing" : "reading", nr_blocks,
1257 blk_rq_sectors(rq)));
1260 * This indicates that the command is ready from our end to be
1266 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1268 struct request *rq = cmd->request;
1270 switch (req_op(rq)) {
1271 case REQ_OP_DISCARD:
1272 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1274 return sd_setup_unmap_cmnd(cmd);
1276 return sd_setup_write_same16_cmnd(cmd, true);
1278 return sd_setup_write_same10_cmnd(cmd, true);
1280 return sd_setup_write_same10_cmnd(cmd, false);
1282 return BLK_STS_TARGET;
1284 case REQ_OP_WRITE_ZEROES:
1285 return sd_setup_write_zeroes_cmnd(cmd);
1286 case REQ_OP_WRITE_SAME:
1287 return sd_setup_write_same_cmnd(cmd);
1289 return sd_setup_flush_cmnd(cmd);
1292 return sd_setup_read_write_cmnd(cmd);
1293 case REQ_OP_ZONE_RESET:
1294 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1296 case REQ_OP_ZONE_RESET_ALL:
1297 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1299 case REQ_OP_ZONE_OPEN:
1300 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1301 case REQ_OP_ZONE_CLOSE:
1302 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1303 case REQ_OP_ZONE_FINISH:
1304 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1307 return BLK_STS_NOTSUPP;
1311 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1313 struct request *rq = SCpnt->request;
1316 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1317 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1319 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1323 mempool_free(cmnd, sd_cdb_pool);
1328 * sd_open - open a scsi disk device
1329 * @bdev: Block device of the scsi disk to open
1330 * @mode: FMODE_* mask
1332 * Returns 0 if successful. Returns a negated errno value in case
1335 * Note: This can be called from a user context (e.g. fsck(1) )
1336 * or from within the kernel (e.g. as a result of a mount(1) ).
1337 * In the latter case @inode and @filp carry an abridged amount
1338 * of information as noted above.
1340 * Locking: called with bdev->bd_mutex held.
1342 static int sd_open(struct block_device *bdev, fmode_t mode)
1344 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1345 struct scsi_device *sdev;
1351 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1353 sdev = sdkp->device;
1356 * If the device is in error recovery, wait until it is done.
1357 * If the device is offline, then disallow any access to it.
1360 if (!scsi_block_when_processing_errors(sdev))
1363 if (sdev->removable || sdkp->write_prot)
1364 check_disk_change(bdev);
1367 * If the drive is empty, just let the open fail.
1369 retval = -ENOMEDIUM;
1370 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1374 * If the device has the write protect tab set, have the open fail
1375 * if the user expects to be able to write to the thing.
1378 if (sdkp->write_prot && (mode & FMODE_WRITE))
1382 * It is possible that the disk changing stuff resulted in
1383 * the device being taken offline. If this is the case,
1384 * report this to the user, and don't pretend that the
1385 * open actually succeeded.
1388 if (!scsi_device_online(sdev))
1391 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1392 if (scsi_block_when_processing_errors(sdev))
1393 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1399 scsi_disk_put(sdkp);
1404 * sd_release - invoked when the (last) close(2) is called on this
1406 * @disk: disk to release
1407 * @mode: FMODE_* mask
1411 * Note: may block (uninterruptible) if error recovery is underway
1414 * Locking: called with bdev->bd_mutex held.
1416 static void sd_release(struct gendisk *disk, fmode_t mode)
1418 struct scsi_disk *sdkp = scsi_disk(disk);
1419 struct scsi_device *sdev = sdkp->device;
1421 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1423 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1424 if (scsi_block_when_processing_errors(sdev))
1425 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1428 scsi_disk_put(sdkp);
1431 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1433 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1434 struct scsi_device *sdp = sdkp->device;
1435 struct Scsi_Host *host = sdp->host;
1436 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1439 /* default to most commonly used values */
1440 diskinfo[0] = 0x40; /* 1 << 6 */
1441 diskinfo[1] = 0x20; /* 1 << 5 */
1442 diskinfo[2] = capacity >> 11;
1444 /* override with calculated, extended default, or driver values */
1445 if (host->hostt->bios_param)
1446 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1448 scsicam_bios_param(bdev, capacity, diskinfo);
1450 geo->heads = diskinfo[0];
1451 geo->sectors = diskinfo[1];
1452 geo->cylinders = diskinfo[2];
1457 * sd_ioctl - process an ioctl
1458 * @bdev: target block device
1459 * @mode: FMODE_* mask
1460 * @cmd: ioctl command number
1461 * @arg: this is third argument given to ioctl(2) system call.
1462 * Often contains a pointer.
1464 * Returns 0 if successful (some ioctls return positive numbers on
1465 * success as well). Returns a negated errno value in case of error.
1467 * Note: most ioctls are forward onto the block subsystem or further
1468 * down in the scsi subsystem.
1470 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1471 unsigned int cmd, unsigned long arg)
1473 struct gendisk *disk = bdev->bd_disk;
1474 struct scsi_disk *sdkp = scsi_disk(disk);
1475 struct scsi_device *sdp = sdkp->device;
1476 void __user *p = (void __user *)arg;
1479 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1480 "cmd=0x%x\n", disk->disk_name, cmd));
1482 error = scsi_verify_blk_ioctl(bdev, cmd);
1487 * If we are in the middle of error recovery, don't let anyone
1488 * else try and use this device. Also, if error recovery fails, it
1489 * may try and take the device offline, in which case all further
1490 * access to the device is prohibited.
1492 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1493 (mode & FMODE_NDELAY) != 0);
1497 if (is_sed_ioctl(cmd))
1498 return sed_ioctl(sdkp->opal_dev, cmd, p);
1501 * Send SCSI addressing ioctls directly to mid level, send other
1502 * ioctls to block level and then onto mid level if they can't be
1506 case SCSI_IOCTL_GET_IDLUN:
1507 case SCSI_IOCTL_GET_BUS_NUMBER:
1508 error = scsi_ioctl(sdp, cmd, p);
1511 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1512 if (error != -ENOTTY)
1514 error = scsi_ioctl(sdp, cmd, p);
1521 static void set_media_not_present(struct scsi_disk *sdkp)
1523 if (sdkp->media_present)
1524 sdkp->device->changed = 1;
1526 if (sdkp->device->removable) {
1527 sdkp->media_present = 0;
1532 static int media_not_present(struct scsi_disk *sdkp,
1533 struct scsi_sense_hdr *sshdr)
1535 if (!scsi_sense_valid(sshdr))
1538 /* not invoked for commands that could return deferred errors */
1539 switch (sshdr->sense_key) {
1540 case UNIT_ATTENTION:
1542 /* medium not present */
1543 if (sshdr->asc == 0x3A) {
1544 set_media_not_present(sdkp);
1552 * sd_check_events - check media events
1553 * @disk: kernel device descriptor
1554 * @clearing: disk events currently being cleared
1556 * Returns mask of DISK_EVENT_*.
1558 * Note: this function is invoked from the block subsystem.
1560 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1562 struct scsi_disk *sdkp = scsi_disk_get(disk);
1563 struct scsi_device *sdp;
1570 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1573 * If the device is offline, don't send any commands - just pretend as
1574 * if the command failed. If the device ever comes back online, we
1575 * can deal with it then. It is only because of unrecoverable errors
1576 * that we would ever take a device offline in the first place.
1578 if (!scsi_device_online(sdp)) {
1579 set_media_not_present(sdkp);
1584 * Using TEST_UNIT_READY enables differentiation between drive with
1585 * no cartridge loaded - NOT READY, drive with changed cartridge -
1586 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1588 * Drives that auto spin down. eg iomega jaz 1G, will be started
1589 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1590 * sd_revalidate() is called.
1592 if (scsi_block_when_processing_errors(sdp)) {
1593 struct scsi_sense_hdr sshdr = { 0, };
1595 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1598 /* failed to execute TUR, assume media not present */
1599 if (host_byte(retval)) {
1600 set_media_not_present(sdkp);
1604 if (media_not_present(sdkp, &sshdr))
1609 * For removable scsi disk we have to recognise the presence
1610 * of a disk in the drive.
1612 if (!sdkp->media_present)
1614 sdkp->media_present = 1;
1617 * sdp->changed is set under the following conditions:
1619 * Medium present state has changed in either direction.
1620 * Device has indicated UNIT_ATTENTION.
1622 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1624 scsi_disk_put(sdkp);
1628 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1631 struct scsi_device *sdp = sdkp->device;
1632 const int timeout = sdp->request_queue->rq_timeout
1633 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1634 struct scsi_sense_hdr my_sshdr;
1636 if (!scsi_device_online(sdp))
1639 /* caller might not be interested in sense, but we need it */
1643 for (retries = 3; retries > 0; --retries) {
1644 unsigned char cmd[10] = { 0 };
1646 cmd[0] = SYNCHRONIZE_CACHE;
1648 * Leave the rest of the command zero to indicate
1651 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1652 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1658 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1660 if (driver_byte(res) == DRIVER_SENSE)
1661 sd_print_sense_hdr(sdkp, sshdr);
1663 /* we need to evaluate the error return */
1664 if (scsi_sense_valid(sshdr) &&
1665 (sshdr->asc == 0x3a || /* medium not present */
1666 sshdr->asc == 0x20 || /* invalid command */
1667 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1668 /* this is no error here */
1671 switch (host_byte(res)) {
1672 /* ignore errors due to racing a disconnection */
1673 case DID_BAD_TARGET:
1674 case DID_NO_CONNECT:
1676 /* signal the upper layer it might try again */
1680 case DID_SOFT_ERROR:
1689 static void sd_rescan(struct device *dev)
1691 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1693 revalidate_disk(sdkp->disk);
1697 #ifdef CONFIG_COMPAT
1699 * This gets directly called from VFS. When the ioctl
1700 * is not recognized we go back to the other translation paths.
1702 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1703 unsigned int cmd, unsigned long arg)
1705 struct gendisk *disk = bdev->bd_disk;
1706 struct scsi_disk *sdkp = scsi_disk(disk);
1707 struct scsi_device *sdev = sdkp->device;
1708 void __user *p = compat_ptr(arg);
1711 error = scsi_verify_blk_ioctl(bdev, cmd);
1715 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1716 (mode & FMODE_NDELAY) != 0);
1720 if (is_sed_ioctl(cmd))
1721 return sed_ioctl(sdkp->opal_dev, cmd, p);
1724 * Let the static ioctl translation table take care of it.
1726 if (!sdev->host->hostt->compat_ioctl)
1727 return -ENOIOCTLCMD;
1728 return sdev->host->hostt->compat_ioctl(sdev, cmd, p);
1732 static char sd_pr_type(enum pr_type type)
1735 case PR_WRITE_EXCLUSIVE:
1737 case PR_EXCLUSIVE_ACCESS:
1739 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1741 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1743 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1745 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1752 static int sd_pr_command(struct block_device *bdev, u8 sa,
1753 u64 key, u64 sa_key, u8 type, u8 flags)
1755 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1756 struct scsi_sense_hdr sshdr;
1758 u8 cmd[16] = { 0, };
1759 u8 data[24] = { 0, };
1761 cmd[0] = PERSISTENT_RESERVE_OUT;
1764 put_unaligned_be32(sizeof(data), &cmd[5]);
1766 put_unaligned_be64(key, &data[0]);
1767 put_unaligned_be64(sa_key, &data[8]);
1770 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1771 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1773 if (driver_byte(result) == DRIVER_SENSE &&
1774 scsi_sense_valid(&sshdr)) {
1775 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1776 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1782 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1785 if (flags & ~PR_FL_IGNORE_KEY)
1787 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1788 old_key, new_key, 0,
1789 (1 << 0) /* APTPL */);
1792 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1797 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1800 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1802 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1805 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1806 enum pr_type type, bool abort)
1808 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1809 sd_pr_type(type), 0);
1812 static int sd_pr_clear(struct block_device *bdev, u64 key)
1814 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1817 static const struct pr_ops sd_pr_ops = {
1818 .pr_register = sd_pr_register,
1819 .pr_reserve = sd_pr_reserve,
1820 .pr_release = sd_pr_release,
1821 .pr_preempt = sd_pr_preempt,
1822 .pr_clear = sd_pr_clear,
1825 static const struct block_device_operations sd_fops = {
1826 .owner = THIS_MODULE,
1828 .release = sd_release,
1830 .getgeo = sd_getgeo,
1831 #ifdef CONFIG_COMPAT
1832 .compat_ioctl = sd_compat_ioctl,
1834 .check_events = sd_check_events,
1835 .revalidate_disk = sd_revalidate_disk,
1836 .unlock_native_capacity = sd_unlock_native_capacity,
1837 .report_zones = sd_zbc_report_zones,
1838 .pr_ops = &sd_pr_ops,
1842 * sd_eh_reset - reset error handling callback
1843 * @scmd: sd-issued command that has failed
1845 * This function is called by the SCSI midlayer before starting
1846 * SCSI EH. When counting medium access failures we have to be
1847 * careful to register it only only once per device and SCSI EH run;
1848 * there might be several timed out commands which will cause the
1849 * 'max_medium_access_timeouts' counter to trigger after the first
1850 * SCSI EH run already and set the device to offline.
1851 * So this function resets the internal counter before starting SCSI EH.
1853 static void sd_eh_reset(struct scsi_cmnd *scmd)
1855 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1857 /* New SCSI EH run, reset gate variable */
1858 sdkp->ignore_medium_access_errors = false;
1862 * sd_eh_action - error handling callback
1863 * @scmd: sd-issued command that has failed
1864 * @eh_disp: The recovery disposition suggested by the midlayer
1866 * This function is called by the SCSI midlayer upon completion of an
1867 * error test command (currently TEST UNIT READY). The result of sending
1868 * the eh command is passed in eh_disp. We're looking for devices that
1869 * fail medium access commands but are OK with non access commands like
1870 * test unit ready (so wrongly see the device as having a successful
1873 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1875 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1876 struct scsi_device *sdev = scmd->device;
1878 if (!scsi_device_online(sdev) ||
1879 !scsi_medium_access_command(scmd) ||
1880 host_byte(scmd->result) != DID_TIME_OUT ||
1885 * The device has timed out executing a medium access command.
1886 * However, the TEST UNIT READY command sent during error
1887 * handling completed successfully. Either the device is in the
1888 * process of recovering or has it suffered an internal failure
1889 * that prevents access to the storage medium.
1891 if (!sdkp->ignore_medium_access_errors) {
1892 sdkp->medium_access_timed_out++;
1893 sdkp->ignore_medium_access_errors = true;
1897 * If the device keeps failing read/write commands but TEST UNIT
1898 * READY always completes successfully we assume that medium
1899 * access is no longer possible and take the device offline.
1901 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1902 scmd_printk(KERN_ERR, scmd,
1903 "Medium access timeout failure. Offlining disk!\n");
1904 mutex_lock(&sdev->state_mutex);
1905 scsi_device_set_state(sdev, SDEV_OFFLINE);
1906 mutex_unlock(&sdev->state_mutex);
1914 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1916 struct request *req = scmd->request;
1917 struct scsi_device *sdev = scmd->device;
1918 unsigned int transferred, good_bytes;
1919 u64 start_lba, end_lba, bad_lba;
1922 * Some commands have a payload smaller than the device logical
1923 * block size (e.g. INQUIRY on a 4K disk).
1925 if (scsi_bufflen(scmd) <= sdev->sector_size)
1928 /* Check if we have a 'bad_lba' information */
1929 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1930 SCSI_SENSE_BUFFERSIZE,
1935 * If the bad lba was reported incorrectly, we have no idea where
1938 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1939 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1940 if (bad_lba < start_lba || bad_lba >= end_lba)
1944 * resid is optional but mostly filled in. When it's unused,
1945 * its value is zero, so we assume the whole buffer transferred
1947 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1949 /* This computation should always be done in terms of the
1950 * resolution of the device's medium.
1952 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1954 return min(good_bytes, transferred);
1958 * sd_done - bottom half handler: called when the lower level
1959 * driver has completed (successfully or otherwise) a scsi command.
1960 * @SCpnt: mid-level's per command structure.
1962 * Note: potentially run from within an ISR. Must not block.
1964 static int sd_done(struct scsi_cmnd *SCpnt)
1966 int result = SCpnt->result;
1967 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1968 unsigned int sector_size = SCpnt->device->sector_size;
1970 struct scsi_sense_hdr sshdr;
1971 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1972 struct request *req = SCpnt->request;
1973 int sense_valid = 0;
1974 int sense_deferred = 0;
1976 switch (req_op(req)) {
1977 case REQ_OP_DISCARD:
1978 case REQ_OP_WRITE_ZEROES:
1979 case REQ_OP_WRITE_SAME:
1980 case REQ_OP_ZONE_RESET:
1981 case REQ_OP_ZONE_RESET_ALL:
1982 case REQ_OP_ZONE_OPEN:
1983 case REQ_OP_ZONE_CLOSE:
1984 case REQ_OP_ZONE_FINISH:
1986 good_bytes = blk_rq_bytes(req);
1987 scsi_set_resid(SCpnt, 0);
1990 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1995 * In case of bogus fw or device, we could end up having
1996 * an unaligned partial completion. Check this here and force
1999 resid = scsi_get_resid(SCpnt);
2000 if (resid & (sector_size - 1)) {
2001 sd_printk(KERN_INFO, sdkp,
2002 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2003 resid, sector_size);
2004 scsi_print_command(SCpnt);
2005 resid = min(scsi_bufflen(SCpnt),
2006 round_up(resid, sector_size));
2007 scsi_set_resid(SCpnt, resid);
2012 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2014 sense_deferred = scsi_sense_is_deferred(&sshdr);
2016 sdkp->medium_access_timed_out = 0;
2018 if (driver_byte(result) != DRIVER_SENSE &&
2019 (!sense_valid || sense_deferred))
2022 switch (sshdr.sense_key) {
2023 case HARDWARE_ERROR:
2025 good_bytes = sd_completed_bytes(SCpnt);
2027 case RECOVERED_ERROR:
2028 good_bytes = scsi_bufflen(SCpnt);
2031 /* This indicates a false check condition, so ignore it. An
2032 * unknown amount of data was transferred so treat it as an
2036 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2038 case ABORTED_COMMAND:
2039 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2040 good_bytes = sd_completed_bytes(SCpnt);
2042 case ILLEGAL_REQUEST:
2043 switch (sshdr.asc) {
2044 case 0x10: /* DIX: Host detected corruption */
2045 good_bytes = sd_completed_bytes(SCpnt);
2047 case 0x20: /* INVALID COMMAND OPCODE */
2048 case 0x24: /* INVALID FIELD IN CDB */
2049 switch (SCpnt->cmnd[0]) {
2051 sd_config_discard(sdkp, SD_LBP_DISABLE);
2055 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2056 sd_config_discard(sdkp, SD_LBP_DISABLE);
2058 sdkp->device->no_write_same = 1;
2059 sd_config_write_same(sdkp);
2060 req->rq_flags |= RQF_QUIET;
2071 if (sd_is_zoned(sdkp))
2072 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2074 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2075 "sd_done: completed %d of %d bytes\n",
2076 good_bytes, scsi_bufflen(SCpnt)));
2082 * spinup disk - called only in sd_revalidate_disk()
2085 sd_spinup_disk(struct scsi_disk *sdkp)
2087 unsigned char cmd[10];
2088 unsigned long spintime_expire = 0;
2089 int retries, spintime;
2090 unsigned int the_result;
2091 struct scsi_sense_hdr sshdr;
2092 int sense_valid = 0;
2096 /* Spin up drives, as required. Only do this at boot time */
2097 /* Spinup needs to be done for module loads too. */
2102 cmd[0] = TEST_UNIT_READY;
2103 memset((void *) &cmd[1], 0, 9);
2105 the_result = scsi_execute_req(sdkp->device, cmd,
2108 SD_MAX_RETRIES, NULL);
2111 * If the drive has indicated to us that it
2112 * doesn't have any media in it, don't bother
2113 * with any more polling.
2115 if (media_not_present(sdkp, &sshdr))
2119 sense_valid = scsi_sense_valid(&sshdr);
2121 } while (retries < 3 &&
2122 (!scsi_status_is_good(the_result) ||
2123 ((driver_byte(the_result) == DRIVER_SENSE) &&
2124 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2126 if (driver_byte(the_result) != DRIVER_SENSE) {
2127 /* no sense, TUR either succeeded or failed
2128 * with a status error */
2129 if(!spintime && !scsi_status_is_good(the_result)) {
2130 sd_print_result(sdkp, "Test Unit Ready failed",
2137 * The device does not want the automatic start to be issued.
2139 if (sdkp->device->no_start_on_add)
2142 if (sense_valid && sshdr.sense_key == NOT_READY) {
2143 if (sshdr.asc == 4 && sshdr.ascq == 3)
2144 break; /* manual intervention required */
2145 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2146 break; /* standby */
2147 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2148 break; /* unavailable */
2149 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2150 break; /* sanitize in progress */
2152 * Issue command to spin up drive when not ready
2155 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2156 cmd[0] = START_STOP;
2157 cmd[1] = 1; /* Return immediately */
2158 memset((void *) &cmd[2], 0, 8);
2159 cmd[4] = 1; /* Start spin cycle */
2160 if (sdkp->device->start_stop_pwr_cond)
2162 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2164 SD_TIMEOUT, SD_MAX_RETRIES,
2166 spintime_expire = jiffies + 100 * HZ;
2169 /* Wait 1 second for next try */
2171 printk(KERN_CONT ".");
2174 * Wait for USB flash devices with slow firmware.
2175 * Yes, this sense key/ASC combination shouldn't
2176 * occur here. It's characteristic of these devices.
2178 } else if (sense_valid &&
2179 sshdr.sense_key == UNIT_ATTENTION &&
2180 sshdr.asc == 0x28) {
2182 spintime_expire = jiffies + 5 * HZ;
2185 /* Wait 1 second for next try */
2188 /* we don't understand the sense code, so it's
2189 * probably pointless to loop */
2191 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2192 sd_print_sense_hdr(sdkp, &sshdr);
2197 } while (spintime && time_before_eq(jiffies, spintime_expire));
2200 if (scsi_status_is_good(the_result))
2201 printk(KERN_CONT "ready\n");
2203 printk(KERN_CONT "not responding...\n");
2208 * Determine whether disk supports Data Integrity Field.
2210 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2212 struct scsi_device *sdp = sdkp->device;
2216 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2219 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2221 if (type > T10_PI_TYPE3_PROTECTION)
2223 else if (scsi_host_dif_capable(sdp->host, type))
2226 if (sdkp->first_scan || type != sdkp->protection_type)
2229 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2230 " protection type %u. Disabling disk!\n",
2234 sd_printk(KERN_NOTICE, sdkp,
2235 "Enabling DIF Type %u protection\n", type);
2238 sd_printk(KERN_NOTICE, sdkp,
2239 "Disabling DIF Type %u protection\n", type);
2243 sdkp->protection_type = type;
2248 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2249 struct scsi_sense_hdr *sshdr, int sense_valid,
2252 if (driver_byte(the_result) == DRIVER_SENSE)
2253 sd_print_sense_hdr(sdkp, sshdr);
2255 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2258 * Set dirty bit for removable devices if not ready -
2259 * sometimes drives will not report this properly.
2261 if (sdp->removable &&
2262 sense_valid && sshdr->sense_key == NOT_READY)
2263 set_media_not_present(sdkp);
2266 * We used to set media_present to 0 here to indicate no media
2267 * in the drive, but some drives fail read capacity even with
2268 * media present, so we can't do that.
2270 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2274 #if RC16_LEN > SD_BUF_SIZE
2275 #error RC16_LEN must not be more than SD_BUF_SIZE
2278 #define READ_CAPACITY_RETRIES_ON_RESET 10
2280 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2281 unsigned char *buffer)
2283 unsigned char cmd[16];
2284 struct scsi_sense_hdr sshdr;
2285 int sense_valid = 0;
2287 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2288 unsigned int alignment;
2289 unsigned long long lba;
2290 unsigned sector_size;
2292 if (sdp->no_read_capacity_16)
2297 cmd[0] = SERVICE_ACTION_IN_16;
2298 cmd[1] = SAI_READ_CAPACITY_16;
2300 memset(buffer, 0, RC16_LEN);
2302 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2303 buffer, RC16_LEN, &sshdr,
2304 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2306 if (media_not_present(sdkp, &sshdr))
2310 sense_valid = scsi_sense_valid(&sshdr);
2312 sshdr.sense_key == ILLEGAL_REQUEST &&
2313 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2315 /* Invalid Command Operation Code or
2316 * Invalid Field in CDB, just retry
2317 * silently with RC10 */
2320 sshdr.sense_key == UNIT_ATTENTION &&
2321 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2322 /* Device reset might occur several times,
2323 * give it one more chance */
2324 if (--reset_retries > 0)
2329 } while (the_result && retries);
2332 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2333 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2337 sector_size = get_unaligned_be32(&buffer[8]);
2338 lba = get_unaligned_be64(&buffer[0]);
2340 if (sd_read_protection_type(sdkp, buffer) < 0) {
2345 /* Logical blocks per physical block exponent */
2346 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2349 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2351 /* Lowest aligned logical block */
2352 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2353 blk_queue_alignment_offset(sdp->request_queue, alignment);
2354 if (alignment && sdkp->first_scan)
2355 sd_printk(KERN_NOTICE, sdkp,
2356 "physical block alignment offset: %u\n", alignment);
2358 if (buffer[14] & 0x80) { /* LBPME */
2361 if (buffer[14] & 0x40) /* LBPRZ */
2364 sd_config_discard(sdkp, SD_LBP_WS16);
2367 sdkp->capacity = lba + 1;
2371 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2372 unsigned char *buffer)
2374 unsigned char cmd[16];
2375 struct scsi_sense_hdr sshdr;
2376 int sense_valid = 0;
2378 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2380 unsigned sector_size;
2383 cmd[0] = READ_CAPACITY;
2384 memset(&cmd[1], 0, 9);
2385 memset(buffer, 0, 8);
2387 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2389 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2391 if (media_not_present(sdkp, &sshdr))
2395 sense_valid = scsi_sense_valid(&sshdr);
2397 sshdr.sense_key == UNIT_ATTENTION &&
2398 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2399 /* Device reset might occur several times,
2400 * give it one more chance */
2401 if (--reset_retries > 0)
2406 } while (the_result && retries);
2409 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2410 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2414 sector_size = get_unaligned_be32(&buffer[4]);
2415 lba = get_unaligned_be32(&buffer[0]);
2417 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2418 /* Some buggy (usb cardreader) devices return an lba of
2419 0xffffffff when the want to report a size of 0 (with
2420 which they really mean no media is present) */
2422 sdkp->physical_block_size = sector_size;
2426 sdkp->capacity = lba + 1;
2427 sdkp->physical_block_size = sector_size;
2431 static int sd_try_rc16_first(struct scsi_device *sdp)
2433 if (sdp->host->max_cmd_len < 16)
2435 if (sdp->try_rc_10_first)
2437 if (sdp->scsi_level > SCSI_SPC_2)
2439 if (scsi_device_protection(sdp))
2445 * read disk capacity
2448 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2451 struct scsi_device *sdp = sdkp->device;
2453 if (sd_try_rc16_first(sdp)) {
2454 sector_size = read_capacity_16(sdkp, sdp, buffer);
2455 if (sector_size == -EOVERFLOW)
2457 if (sector_size == -ENODEV)
2459 if (sector_size < 0)
2460 sector_size = read_capacity_10(sdkp, sdp, buffer);
2461 if (sector_size < 0)
2464 sector_size = read_capacity_10(sdkp, sdp, buffer);
2465 if (sector_size == -EOVERFLOW)
2467 if (sector_size < 0)
2469 if ((sizeof(sdkp->capacity) > 4) &&
2470 (sdkp->capacity > 0xffffffffULL)) {
2471 int old_sector_size = sector_size;
2472 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2473 "Trying to use READ CAPACITY(16).\n");
2474 sector_size = read_capacity_16(sdkp, sdp, buffer);
2475 if (sector_size < 0) {
2476 sd_printk(KERN_NOTICE, sdkp,
2477 "Using 0xffffffff as device size\n");
2478 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2479 sector_size = old_sector_size;
2482 /* Remember that READ CAPACITY(16) succeeded */
2483 sdp->try_rc_10_first = 0;
2487 /* Some devices are known to return the total number of blocks,
2488 * not the highest block number. Some devices have versions
2489 * which do this and others which do not. Some devices we might
2490 * suspect of doing this but we don't know for certain.
2492 * If we know the reported capacity is wrong, decrement it. If
2493 * we can only guess, then assume the number of blocks is even
2494 * (usually true but not always) and err on the side of lowering
2497 if (sdp->fix_capacity ||
2498 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2499 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2500 "from its reported value: %llu\n",
2501 (unsigned long long) sdkp->capacity);
2506 if (sector_size == 0) {
2508 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2512 if (sector_size != 512 &&
2513 sector_size != 1024 &&
2514 sector_size != 2048 &&
2515 sector_size != 4096) {
2516 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2519 * The user might want to re-format the drive with
2520 * a supported sectorsize. Once this happens, it
2521 * would be relatively trivial to set the thing up.
2522 * For this reason, we leave the thing in the table.
2526 * set a bogus sector size so the normal read/write
2527 * logic in the block layer will eventually refuse any
2528 * request on this device without tripping over power
2529 * of two sector size assumptions
2533 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2534 blk_queue_physical_block_size(sdp->request_queue,
2535 sdkp->physical_block_size);
2536 sdkp->device->sector_size = sector_size;
2538 if (sdkp->capacity > 0xffffffff)
2539 sdp->use_16_for_rw = 1;
2544 * Print disk capacity
2547 sd_print_capacity(struct scsi_disk *sdkp,
2548 sector_t old_capacity)
2550 int sector_size = sdkp->device->sector_size;
2551 char cap_str_2[10], cap_str_10[10];
2553 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2556 string_get_size(sdkp->capacity, sector_size,
2557 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2558 string_get_size(sdkp->capacity, sector_size,
2559 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2561 sd_printk(KERN_NOTICE, sdkp,
2562 "%llu %d-byte logical blocks: (%s/%s)\n",
2563 (unsigned long long)sdkp->capacity,
2564 sector_size, cap_str_10, cap_str_2);
2566 if (sdkp->physical_block_size != sector_size)
2567 sd_printk(KERN_NOTICE, sdkp,
2568 "%u-byte physical blocks\n",
2569 sdkp->physical_block_size);
2571 sd_zbc_print_zones(sdkp);
2574 /* called with buffer of length 512 */
2576 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2577 unsigned char *buffer, int len, struct scsi_mode_data *data,
2578 struct scsi_sense_hdr *sshdr)
2580 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2581 SD_TIMEOUT, SD_MAX_RETRIES, data,
2586 * read write protect setting, if possible - called only in sd_revalidate_disk()
2587 * called with buffer of length SD_BUF_SIZE
2590 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2593 struct scsi_device *sdp = sdkp->device;
2594 struct scsi_mode_data data;
2595 int old_wp = sdkp->write_prot;
2597 set_disk_ro(sdkp->disk, 0);
2598 if (sdp->skip_ms_page_3f) {
2599 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2603 if (sdp->use_192_bytes_for_3f) {
2604 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2607 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2608 * We have to start carefully: some devices hang if we ask
2609 * for more than is available.
2611 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2614 * Second attempt: ask for page 0 When only page 0 is
2615 * implemented, a request for page 3F may return Sense Key
2616 * 5: Illegal Request, Sense Code 24: Invalid field in
2619 if (!scsi_status_is_good(res))
2620 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2623 * Third attempt: ask 255 bytes, as we did earlier.
2625 if (!scsi_status_is_good(res))
2626 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2630 if (!scsi_status_is_good(res)) {
2631 sd_first_printk(KERN_WARNING, sdkp,
2632 "Test WP failed, assume Write Enabled\n");
2634 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2635 set_disk_ro(sdkp->disk, sdkp->write_prot);
2636 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2637 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2638 sdkp->write_prot ? "on" : "off");
2639 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2645 * sd_read_cache_type - called only from sd_revalidate_disk()
2646 * called with buffer of length SD_BUF_SIZE
2649 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2652 struct scsi_device *sdp = sdkp->device;
2657 struct scsi_mode_data data;
2658 struct scsi_sense_hdr sshdr;
2659 int old_wce = sdkp->WCE;
2660 int old_rcd = sdkp->RCD;
2661 int old_dpofua = sdkp->DPOFUA;
2664 if (sdkp->cache_override)
2668 if (sdp->skip_ms_page_8) {
2669 if (sdp->type == TYPE_RBC)
2672 if (sdp->skip_ms_page_3f)
2675 if (sdp->use_192_bytes_for_3f)
2679 } else if (sdp->type == TYPE_RBC) {
2687 /* cautiously ask */
2688 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2691 if (!scsi_status_is_good(res))
2694 if (!data.header_length) {
2697 sd_first_printk(KERN_ERR, sdkp,
2698 "Missing header in MODE_SENSE response\n");
2701 /* that went OK, now ask for the proper length */
2705 * We're only interested in the first three bytes, actually.
2706 * But the data cache page is defined for the first 20.
2710 else if (len > SD_BUF_SIZE) {
2711 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2712 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2715 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2719 if (len > first_len)
2720 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2723 if (scsi_status_is_good(res)) {
2724 int offset = data.header_length + data.block_descriptor_length;
2726 while (offset < len) {
2727 u8 page_code = buffer[offset] & 0x3F;
2728 u8 spf = buffer[offset] & 0x40;
2730 if (page_code == 8 || page_code == 6) {
2731 /* We're interested only in the first 3 bytes.
2733 if (len - offset <= 2) {
2734 sd_first_printk(KERN_ERR, sdkp,
2735 "Incomplete mode parameter "
2739 modepage = page_code;
2743 /* Go to the next page */
2744 if (spf && len - offset > 3)
2745 offset += 4 + (buffer[offset+2] << 8) +
2747 else if (!spf && len - offset > 1)
2748 offset += 2 + buffer[offset+1];
2750 sd_first_printk(KERN_ERR, sdkp,
2752 "parameter data\n");
2758 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2762 if (modepage == 8) {
2763 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2764 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2766 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2770 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2771 if (sdp->broken_fua) {
2772 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2774 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2775 !sdkp->device->use_16_for_rw) {
2776 sd_first_printk(KERN_NOTICE, sdkp,
2777 "Uses READ/WRITE(6), disabling FUA\n");
2781 /* No cache flush allowed for write protected devices */
2782 if (sdkp->WCE && sdkp->write_prot)
2785 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2786 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2787 sd_printk(KERN_NOTICE, sdkp,
2788 "Write cache: %s, read cache: %s, %s\n",
2789 sdkp->WCE ? "enabled" : "disabled",
2790 sdkp->RCD ? "disabled" : "enabled",
2791 sdkp->DPOFUA ? "supports DPO and FUA"
2792 : "doesn't support DPO or FUA");
2798 if (scsi_sense_valid(&sshdr) &&
2799 sshdr.sense_key == ILLEGAL_REQUEST &&
2800 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2801 /* Invalid field in CDB */
2802 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2804 sd_first_printk(KERN_ERR, sdkp,
2805 "Asking for cache data failed\n");
2808 if (sdp->wce_default_on) {
2809 sd_first_printk(KERN_NOTICE, sdkp,
2810 "Assuming drive cache: write back\n");
2813 sd_first_printk(KERN_ERR, sdkp,
2814 "Assuming drive cache: write through\n");
2822 * The ATO bit indicates whether the DIF application tag is available
2823 * for use by the operating system.
2825 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2828 struct scsi_device *sdp = sdkp->device;
2829 struct scsi_mode_data data;
2830 struct scsi_sense_hdr sshdr;
2832 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2835 if (sdkp->protection_type == 0)
2838 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2839 SD_MAX_RETRIES, &data, &sshdr);
2841 if (!scsi_status_is_good(res) || !data.header_length ||
2843 sd_first_printk(KERN_WARNING, sdkp,
2844 "getting Control mode page failed, assume no ATO\n");
2846 if (scsi_sense_valid(&sshdr))
2847 sd_print_sense_hdr(sdkp, &sshdr);
2852 offset = data.header_length + data.block_descriptor_length;
2854 if ((buffer[offset] & 0x3f) != 0x0a) {
2855 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2859 if ((buffer[offset + 5] & 0x80) == 0)
2868 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2869 * @sdkp: disk to query
2871 static void sd_read_block_limits(struct scsi_disk *sdkp)
2873 unsigned int sector_sz = sdkp->device->sector_size;
2874 const int vpd_len = 64;
2875 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2878 /* Block Limits VPD */
2879 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2882 blk_queue_io_min(sdkp->disk->queue,
2883 get_unaligned_be16(&buffer[6]) * sector_sz);
2885 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2886 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2888 if (buffer[3] == 0x3c) {
2889 unsigned int lba_count, desc_count;
2891 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2896 lba_count = get_unaligned_be32(&buffer[20]);
2897 desc_count = get_unaligned_be32(&buffer[24]);
2899 if (lba_count && desc_count)
2900 sdkp->max_unmap_blocks = lba_count;
2902 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2904 if (buffer[32] & 0x80)
2905 sdkp->unmap_alignment =
2906 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2908 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2910 if (sdkp->max_unmap_blocks)
2911 sd_config_discard(sdkp, SD_LBP_UNMAP);
2913 sd_config_discard(sdkp, SD_LBP_WS16);
2915 } else { /* LBP VPD page tells us what to use */
2916 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2917 sd_config_discard(sdkp, SD_LBP_UNMAP);
2918 else if (sdkp->lbpws)
2919 sd_config_discard(sdkp, SD_LBP_WS16);
2920 else if (sdkp->lbpws10)
2921 sd_config_discard(sdkp, SD_LBP_WS10);
2923 sd_config_discard(sdkp, SD_LBP_DISABLE);
2932 * sd_read_block_characteristics - Query block dev. characteristics
2933 * @sdkp: disk to query
2935 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2937 struct request_queue *q = sdkp->disk->queue;
2938 unsigned char *buffer;
2940 const int vpd_len = 64;
2942 buffer = kmalloc(vpd_len, GFP_KERNEL);
2945 /* Block Device Characteristics VPD */
2946 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2949 rot = get_unaligned_be16(&buffer[4]);
2952 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2953 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2956 if (sdkp->device->type == TYPE_ZBC) {
2958 q->limits.zoned = BLK_ZONED_HM;
2960 sdkp->zoned = (buffer[8] >> 4) & 3;
2961 if (sdkp->zoned == 1)
2963 q->limits.zoned = BLK_ZONED_HA;
2966 * Treat drive-managed devices as
2967 * regular block devices.
2969 q->limits.zoned = BLK_ZONED_NONE;
2971 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2972 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2973 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2980 * sd_read_block_provisioning - Query provisioning VPD page
2981 * @sdkp: disk to query
2983 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2985 unsigned char *buffer;
2986 const int vpd_len = 8;
2988 if (sdkp->lbpme == 0)
2991 buffer = kmalloc(vpd_len, GFP_KERNEL);
2993 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2997 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2998 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2999 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3005 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3007 struct scsi_device *sdev = sdkp->device;
3009 if (sdev->host->no_write_same) {
3010 sdev->no_write_same = 1;
3015 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3016 /* too large values might cause issues with arcmsr */
3017 int vpd_buf_len = 64;
3019 sdev->no_report_opcodes = 1;
3021 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3022 * CODES is unsupported and the device has an ATA
3023 * Information VPD page (SAT).
3025 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3026 sdev->no_write_same = 1;
3029 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3032 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3036 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3038 struct scsi_device *sdev = sdkp->device;
3040 if (!sdev->security_supported)
3043 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3044 SECURITY_PROTOCOL_IN) == 1 &&
3045 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3046 SECURITY_PROTOCOL_OUT) == 1)
3051 * Determine the device's preferred I/O size for reads and writes
3052 * unless the reported value is unreasonably small, large, not a
3053 * multiple of the physical block size, or simply garbage.
3055 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3056 unsigned int dev_max)
3058 struct scsi_device *sdp = sdkp->device;
3059 unsigned int opt_xfer_bytes =
3060 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3062 if (sdkp->opt_xfer_blocks == 0)
3065 if (sdkp->opt_xfer_blocks > dev_max) {
3066 sd_first_printk(KERN_WARNING, sdkp,
3067 "Optimal transfer size %u logical blocks " \
3068 "> dev_max (%u logical blocks)\n",
3069 sdkp->opt_xfer_blocks, dev_max);
3073 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3074 sd_first_printk(KERN_WARNING, sdkp,
3075 "Optimal transfer size %u logical blocks " \
3076 "> sd driver limit (%u logical blocks)\n",
3077 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3081 if (opt_xfer_bytes < PAGE_SIZE) {
3082 sd_first_printk(KERN_WARNING, sdkp,
3083 "Optimal transfer size %u bytes < " \
3084 "PAGE_SIZE (%u bytes)\n",
3085 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3089 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3090 sd_first_printk(KERN_WARNING, sdkp,
3091 "Optimal transfer size %u bytes not a " \
3092 "multiple of physical block size (%u bytes)\n",
3093 opt_xfer_bytes, sdkp->physical_block_size);
3097 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3103 * sd_revalidate_disk - called the first time a new disk is seen,
3104 * performs disk spin up, read_capacity, etc.
3105 * @disk: struct gendisk we care about
3107 static int sd_revalidate_disk(struct gendisk *disk)
3109 struct scsi_disk *sdkp = scsi_disk(disk);
3110 struct scsi_device *sdp = sdkp->device;
3111 struct request_queue *q = sdkp->disk->queue;
3112 sector_t old_capacity = sdkp->capacity;
3113 unsigned char *buffer;
3114 unsigned int dev_max, rw_max;
3116 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3117 "sd_revalidate_disk\n"));
3120 * If the device is offline, don't try and read capacity or any
3121 * of the other niceties.
3123 if (!scsi_device_online(sdp))
3126 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3128 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3129 "allocation failure.\n");
3133 sd_spinup_disk(sdkp);
3136 * Without media there is no reason to ask; moreover, some devices
3137 * react badly if we do.
3139 if (sdkp->media_present) {
3140 sd_read_capacity(sdkp, buffer);
3143 * set the default to rotational. All non-rotational devices
3144 * support the block characteristics VPD page, which will
3145 * cause this to be updated correctly and any device which
3146 * doesn't support it should be treated as rotational.
3148 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3149 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3151 if (scsi_device_supports_vpd(sdp)) {
3152 sd_read_block_provisioning(sdkp);
3153 sd_read_block_limits(sdkp);
3154 sd_read_block_characteristics(sdkp);
3155 sd_zbc_read_zones(sdkp, buffer);
3158 sd_print_capacity(sdkp, old_capacity);
3160 sd_read_write_protect_flag(sdkp, buffer);
3161 sd_read_cache_type(sdkp, buffer);
3162 sd_read_app_tag_own(sdkp, buffer);
3163 sd_read_write_same(sdkp, buffer);
3164 sd_read_security(sdkp, buffer);
3168 * We now have all cache related info, determine how we deal
3169 * with flush requests.
3171 sd_set_flush_flag(sdkp);
3173 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3174 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3176 /* Some devices report a maximum block count for READ/WRITE requests. */
3177 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3178 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3180 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3181 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3182 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3184 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3185 (sector_t)BLK_DEF_MAX_SECTORS);
3187 /* Do not exceed controller limit */
3188 rw_max = min(rw_max, queue_max_hw_sectors(q));
3191 * Only update max_sectors if previously unset or if the current value
3192 * exceeds the capabilities of the hardware.
3194 if (sdkp->first_scan ||
3195 q->limits.max_sectors > q->limits.max_dev_sectors ||
3196 q->limits.max_sectors > q->limits.max_hw_sectors)
3197 q->limits.max_sectors = rw_max;
3199 sdkp->first_scan = 0;
3201 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3202 sd_config_write_same(sdkp);
3210 * sd_unlock_native_capacity - unlock native capacity
3211 * @disk: struct gendisk to set capacity for
3213 * Block layer calls this function if it detects that partitions
3214 * on @disk reach beyond the end of the device. If the SCSI host
3215 * implements ->unlock_native_capacity() method, it's invoked to
3216 * give it a chance to adjust the device capacity.
3219 * Defined by block layer. Might sleep.
3221 static void sd_unlock_native_capacity(struct gendisk *disk)
3223 struct scsi_device *sdev = scsi_disk(disk)->device;
3225 if (sdev->host->hostt->unlock_native_capacity)
3226 sdev->host->hostt->unlock_native_capacity(sdev);
3230 * sd_format_disk_name - format disk name
3231 * @prefix: name prefix - ie. "sd" for SCSI disks
3232 * @index: index of the disk to format name for
3233 * @buf: output buffer
3234 * @buflen: length of the output buffer
3236 * SCSI disk names starts at sda. The 26th device is sdz and the
3237 * 27th is sdaa. The last one for two lettered suffix is sdzz
3238 * which is followed by sdaaa.
3240 * This is basically 26 base counting with one extra 'nil' entry
3241 * at the beginning from the second digit on and can be
3242 * determined using similar method as 26 base conversion with the
3243 * index shifted -1 after each digit is computed.
3249 * 0 on success, -errno on failure.
3251 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3253 const int base = 'z' - 'a' + 1;
3254 char *begin = buf + strlen(prefix);
3255 char *end = buf + buflen;
3265 *--p = 'a' + (index % unit);
3266 index = (index / unit) - 1;
3267 } while (index >= 0);
3269 memmove(begin, p, end - p);
3270 memcpy(buf, prefix, strlen(prefix));
3276 * sd_probe - called during driver initialization and whenever a
3277 * new scsi device is attached to the system. It is called once
3278 * for each scsi device (not just disks) present.
3279 * @dev: pointer to device object
3281 * Returns 0 if successful (or not interested in this scsi device
3282 * (e.g. scanner)); 1 when there is an error.
3284 * Note: this function is invoked from the scsi mid-level.
3285 * This function sets up the mapping between a given
3286 * <host,channel,id,lun> (found in sdp) and new device name
3287 * (e.g. /dev/sda). More precisely it is the block device major
3288 * and minor number that is chosen here.
3290 * Assume sd_probe is not re-entrant (for time being)
3291 * Also think about sd_probe() and sd_remove() running coincidentally.
3293 static int sd_probe(struct device *dev)
3295 struct scsi_device *sdp = to_scsi_device(dev);
3296 struct scsi_disk *sdkp;
3301 scsi_autopm_get_device(sdp);
3303 if (sdp->type != TYPE_DISK &&
3304 sdp->type != TYPE_ZBC &&
3305 sdp->type != TYPE_MOD &&
3306 sdp->type != TYPE_RBC)
3309 #ifndef CONFIG_BLK_DEV_ZONED
3310 if (sdp->type == TYPE_ZBC)
3313 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3317 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3321 gd = alloc_disk(SD_MINORS);
3325 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3327 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3331 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3333 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3334 goto out_free_index;
3338 sdkp->driver = &sd_template;
3340 sdkp->index = index;
3341 atomic_set(&sdkp->openers, 0);
3342 atomic_set(&sdkp->device->ioerr_cnt, 0);
3344 if (!sdp->request_queue->rq_timeout) {
3345 if (sdp->type != TYPE_MOD)
3346 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3348 blk_queue_rq_timeout(sdp->request_queue,
3352 device_initialize(&sdkp->dev);
3353 sdkp->dev.parent = dev;
3354 sdkp->dev.class = &sd_disk_class;
3355 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3357 error = device_add(&sdkp->dev);
3359 goto out_free_index;
3362 dev_set_drvdata(dev, sdkp);
3364 gd->major = sd_major((index & 0xf0) >> 4);
3365 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3367 gd->fops = &sd_fops;
3368 gd->private_data = &sdkp->driver;
3369 gd->queue = sdkp->device->request_queue;
3371 /* defaults, until the device tells us otherwise */
3372 sdp->sector_size = 512;
3374 sdkp->media_present = 1;
3375 sdkp->write_prot = 0;
3376 sdkp->cache_override = 0;
3380 sdkp->first_scan = 1;
3381 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3383 sd_revalidate_disk(gd);
3385 gd->flags = GENHD_FL_EXT_DEVT;
3386 if (sdp->removable) {
3387 gd->flags |= GENHD_FL_REMOVABLE;
3388 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3389 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3392 blk_pm_runtime_init(sdp->request_queue, dev);
3393 device_add_disk(dev, gd, NULL);
3395 sd_dif_config_host(sdkp);
3397 sd_revalidate_disk(gd);
3399 if (sdkp->security) {
3400 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3402 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3405 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3406 sdp->removable ? "removable " : "");
3407 scsi_autopm_put_device(sdp);
3412 ida_free(&sd_index_ida, index);
3418 scsi_autopm_put_device(sdp);
3423 * sd_remove - called whenever a scsi disk (previously recognized by
3424 * sd_probe) is detached from the system. It is called (potentially
3425 * multiple times) during sd module unload.
3426 * @dev: pointer to device object
3428 * Note: this function is invoked from the scsi mid-level.
3429 * This function potentially frees up a device name (e.g. /dev/sdc)
3430 * that could be re-used by a subsequent sd_probe().
3431 * This function is not called when the built-in sd driver is "exit-ed".
3433 static int sd_remove(struct device *dev)
3435 struct scsi_disk *sdkp;
3438 sdkp = dev_get_drvdata(dev);
3439 devt = disk_devt(sdkp->disk);
3440 scsi_autopm_get_device(sdkp->device);
3442 async_synchronize_full_domain(&scsi_sd_pm_domain);
3443 device_del(&sdkp->dev);
3444 del_gendisk(sdkp->disk);
3447 free_opal_dev(sdkp->opal_dev);
3449 blk_register_region(devt, SD_MINORS, NULL,
3450 sd_default_probe, NULL, NULL);
3452 mutex_lock(&sd_ref_mutex);
3453 dev_set_drvdata(dev, NULL);
3454 put_device(&sdkp->dev);
3455 mutex_unlock(&sd_ref_mutex);
3461 * scsi_disk_release - Called to free the scsi_disk structure
3462 * @dev: pointer to embedded class device
3464 * sd_ref_mutex must be held entering this routine. Because it is
3465 * called on last put, you should always use the scsi_disk_get()
3466 * scsi_disk_put() helpers which manipulate the semaphore directly
3467 * and never do a direct put_device.
3469 static void scsi_disk_release(struct device *dev)
3471 struct scsi_disk *sdkp = to_scsi_disk(dev);
3472 struct gendisk *disk = sdkp->disk;
3473 struct request_queue *q = disk->queue;
3475 ida_free(&sd_index_ida, sdkp->index);
3478 * Wait until all requests that are in progress have completed.
3479 * This is necessary to avoid that e.g. scsi_end_request() crashes
3480 * due to clearing the disk->private_data pointer. Wait from inside
3481 * scsi_disk_release() instead of from sd_release() to avoid that
3482 * freezing and unfreezing the request queue affects user space I/O
3483 * in case multiple processes open a /dev/sd... node concurrently.
3485 blk_mq_freeze_queue(q);
3486 blk_mq_unfreeze_queue(q);
3488 disk->private_data = NULL;
3490 put_device(&sdkp->device->sdev_gendev);
3495 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3497 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3498 struct scsi_sense_hdr sshdr;
3499 struct scsi_device *sdp = sdkp->device;
3503 cmd[4] |= 1; /* START */
3505 if (sdp->start_stop_pwr_cond)
3506 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3508 if (!scsi_device_online(sdp))
3511 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3512 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3514 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3515 if (driver_byte(res) == DRIVER_SENSE)
3516 sd_print_sense_hdr(sdkp, &sshdr);
3517 if (scsi_sense_valid(&sshdr) &&
3518 /* 0x3a is medium not present */
3523 /* SCSI error codes must not go to the generic layer */
3531 * Send a SYNCHRONIZE CACHE instruction down to the device through
3532 * the normal SCSI command structure. Wait for the command to
3535 static void sd_shutdown(struct device *dev)
3537 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3540 return; /* this can happen */
3542 if (pm_runtime_suspended(dev))
3545 if (sdkp->WCE && sdkp->media_present) {
3546 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3547 sd_sync_cache(sdkp, NULL);
3550 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3551 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3552 sd_start_stop_device(sdkp, 0);
3556 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3558 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3559 struct scsi_sense_hdr sshdr;
3562 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3565 if (sdkp->WCE && sdkp->media_present) {
3566 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3567 ret = sd_sync_cache(sdkp, &sshdr);
3570 /* ignore OFFLINE device */
3574 if (!scsi_sense_valid(&sshdr) ||
3575 sshdr.sense_key != ILLEGAL_REQUEST)
3579 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3580 * doesn't support sync. There's not much to do and
3581 * suspend shouldn't fail.
3587 if (sdkp->device->manage_start_stop) {
3588 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3589 /* an error is not worth aborting a system sleep */
3590 ret = sd_start_stop_device(sdkp, 0);
3591 if (ignore_stop_errors)
3598 static int sd_suspend_system(struct device *dev)
3600 return sd_suspend_common(dev, true);
3603 static int sd_suspend_runtime(struct device *dev)
3605 return sd_suspend_common(dev, false);
3608 static int sd_resume(struct device *dev)
3610 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3613 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3616 if (!sdkp->device->manage_start_stop)
3619 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3620 ret = sd_start_stop_device(sdkp, 1);
3622 opal_unlock_from_suspend(sdkp->opal_dev);
3627 * init_sd - entry point for this driver (both when built in or when
3630 * Note: this function registers this driver with the scsi mid-level.
3632 static int __init init_sd(void)
3634 int majors = 0, i, err;
3636 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3638 for (i = 0; i < SD_MAJORS; i++) {
3639 if (register_blkdev(sd_major(i), "sd") != 0)
3642 blk_register_region(sd_major(i), SD_MINORS, NULL,
3643 sd_default_probe, NULL, NULL);
3649 err = class_register(&sd_disk_class);
3653 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3655 if (!sd_cdb_cache) {
3656 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3661 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3663 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3668 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3669 if (!sd_page_pool) {
3670 printk(KERN_ERR "sd: can't init discard page pool\n");
3675 err = scsi_register_driver(&sd_template.gendrv);
3677 goto err_out_driver;
3682 mempool_destroy(sd_page_pool);
3685 mempool_destroy(sd_cdb_pool);
3688 kmem_cache_destroy(sd_cdb_cache);
3691 class_unregister(&sd_disk_class);
3693 for (i = 0; i < SD_MAJORS; i++)
3694 unregister_blkdev(sd_major(i), "sd");
3699 * exit_sd - exit point for this driver (when it is a module).
3701 * Note: this function unregisters this driver from the scsi mid-level.
3703 static void __exit exit_sd(void)
3707 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3709 scsi_unregister_driver(&sd_template.gendrv);
3710 mempool_destroy(sd_cdb_pool);
3711 mempool_destroy(sd_page_pool);
3712 kmem_cache_destroy(sd_cdb_cache);
3714 class_unregister(&sd_disk_class);
3716 for (i = 0; i < SD_MAJORS; i++) {
3717 blk_unregister_region(sd_major(i), SD_MINORS);
3718 unregister_blkdev(sd_major(i), "sd");
3722 module_init(init_sd);
3723 module_exit(exit_sd);
3725 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3726 struct scsi_sense_hdr *sshdr)
3728 scsi_print_sense_hdr(sdkp->device,
3729 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3732 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3735 const char *hb_string = scsi_hostbyte_string(result);
3736 const char *db_string = scsi_driverbyte_string(result);
3738 if (hb_string || db_string)
3739 sd_printk(KERN_INFO, sdkp,
3740 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3741 hb_string ? hb_string : "invalid",
3742 db_string ? db_string : "invalid");
3744 sd_printk(KERN_INFO, sdkp,
3745 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3746 msg, host_byte(result), driver_byte(result));