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);
126 static DEFINE_IDA(sd_index_ida);
128 /* This semaphore is used to mediate the 0->1 reference get in the
129 * face of object destruction (i.e. we can't allow a get on an
130 * object after last put) */
131 static DEFINE_MUTEX(sd_ref_mutex);
133 static struct kmem_cache *sd_cdb_cache;
134 static mempool_t *sd_cdb_pool;
135 static mempool_t *sd_page_pool;
137 static const char *sd_cache_types[] = {
138 "write through", "none", "write back",
139 "write back, no read (daft)"
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 bool wc = false, fua = false;
152 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 const char *buf, size_t count)
159 int ct, rcd, wce, sp;
160 struct scsi_disk *sdkp = to_scsi_disk(dev);
161 struct scsi_device *sdp = sdkp->device;
164 struct scsi_mode_data data;
165 struct scsi_sense_hdr sshdr;
166 static const char temp[] = "temporary ";
169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 /* no cache control on RBC devices; theoretically they
171 * can do it, but there's probably so many exceptions
172 * it's not worth the risk */
175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 buf += sizeof(temp) - 1;
177 sdkp->cache_override = 1;
179 sdkp->cache_override = 0;
182 ct = sysfs_match_string(sd_cache_types, buf);
186 rcd = ct & 0x01 ? 1 : 0;
187 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
189 if (sdkp->cache_override) {
192 sd_set_flush_flag(sdkp);
196 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 SD_MAX_RETRIES, &data, NULL))
199 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 data.block_descriptor_length);
201 buffer_data = buffer + data.header_length +
202 data.block_descriptor_length;
203 buffer_data[2] &= ~0x05;
204 buffer_data[2] |= wce << 2 | rcd;
205 sp = buffer_data[0] & 0x80 ? 1 : 0;
206 buffer_data[0] &= ~0x80;
209 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
210 * received mode parameter buffer before doing MODE SELECT.
212 data.device_specific = 0;
214 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
215 SD_MAX_RETRIES, &data, &sshdr)) {
216 if (scsi_sense_valid(&sshdr))
217 sd_print_sense_hdr(sdkp, &sshdr);
220 sd_revalidate_disk(sdkp->disk);
225 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
228 struct scsi_disk *sdkp = to_scsi_disk(dev);
229 struct scsi_device *sdp = sdkp->device;
231 return sprintf(buf, "%u\n", sdp->manage_start_stop);
235 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
236 const char *buf, size_t count)
238 struct scsi_disk *sdkp = to_scsi_disk(dev);
239 struct scsi_device *sdp = sdkp->device;
242 if (!capable(CAP_SYS_ADMIN))
245 if (kstrtobool(buf, &v))
248 sdp->manage_start_stop = v;
252 static DEVICE_ATTR_RW(manage_start_stop);
255 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
257 struct scsi_disk *sdkp = to_scsi_disk(dev);
259 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
263 allow_restart_store(struct device *dev, struct device_attribute *attr,
264 const char *buf, size_t count)
267 struct scsi_disk *sdkp = to_scsi_disk(dev);
268 struct scsi_device *sdp = sdkp->device;
270 if (!capable(CAP_SYS_ADMIN))
273 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
276 if (kstrtobool(buf, &v))
279 sdp->allow_restart = v;
283 static DEVICE_ATTR_RW(allow_restart);
286 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
288 struct scsi_disk *sdkp = to_scsi_disk(dev);
289 int ct = sdkp->RCD + 2*sdkp->WCE;
291 return sprintf(buf, "%s\n", sd_cache_types[ct]);
293 static DEVICE_ATTR_RW(cache_type);
296 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
298 struct scsi_disk *sdkp = to_scsi_disk(dev);
300 return sprintf(buf, "%u\n", sdkp->DPOFUA);
302 static DEVICE_ATTR_RO(FUA);
305 protection_type_show(struct device *dev, struct device_attribute *attr,
308 struct scsi_disk *sdkp = to_scsi_disk(dev);
310 return sprintf(buf, "%u\n", sdkp->protection_type);
314 protection_type_store(struct device *dev, struct device_attribute *attr,
315 const char *buf, size_t count)
317 struct scsi_disk *sdkp = to_scsi_disk(dev);
321 if (!capable(CAP_SYS_ADMIN))
324 err = kstrtouint(buf, 10, &val);
329 if (val <= T10_PI_TYPE3_PROTECTION)
330 sdkp->protection_type = val;
334 static DEVICE_ATTR_RW(protection_type);
337 protection_mode_show(struct device *dev, struct device_attribute *attr,
340 struct scsi_disk *sdkp = to_scsi_disk(dev);
341 struct scsi_device *sdp = sdkp->device;
342 unsigned int dif, dix;
344 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
345 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
347 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
353 return sprintf(buf, "none\n");
355 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
357 static DEVICE_ATTR_RO(protection_mode);
360 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
362 struct scsi_disk *sdkp = to_scsi_disk(dev);
364 return sprintf(buf, "%u\n", sdkp->ATO);
366 static DEVICE_ATTR_RO(app_tag_own);
369 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
372 struct scsi_disk *sdkp = to_scsi_disk(dev);
374 return sprintf(buf, "%u\n", sdkp->lbpme);
376 static DEVICE_ATTR_RO(thin_provisioning);
378 /* sysfs_match_string() requires dense arrays */
379 static const char *lbp_mode[] = {
380 [SD_LBP_FULL] = "full",
381 [SD_LBP_UNMAP] = "unmap",
382 [SD_LBP_WS16] = "writesame_16",
383 [SD_LBP_WS10] = "writesame_10",
384 [SD_LBP_ZERO] = "writesame_zero",
385 [SD_LBP_DISABLE] = "disabled",
389 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
392 struct scsi_disk *sdkp = to_scsi_disk(dev);
394 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
398 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
399 const char *buf, size_t count)
401 struct scsi_disk *sdkp = to_scsi_disk(dev);
402 struct scsi_device *sdp = sdkp->device;
405 if (!capable(CAP_SYS_ADMIN))
408 if (sd_is_zoned(sdkp)) {
409 sd_config_discard(sdkp, SD_LBP_DISABLE);
413 if (sdp->type != TYPE_DISK)
416 mode = sysfs_match_string(lbp_mode, buf);
420 sd_config_discard(sdkp, mode);
424 static DEVICE_ATTR_RW(provisioning_mode);
426 /* sysfs_match_string() requires dense arrays */
427 static const char *zeroing_mode[] = {
428 [SD_ZERO_WRITE] = "write",
429 [SD_ZERO_WS] = "writesame",
430 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
431 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
435 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
438 struct scsi_disk *sdkp = to_scsi_disk(dev);
440 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
444 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
445 const char *buf, size_t count)
447 struct scsi_disk *sdkp = to_scsi_disk(dev);
450 if (!capable(CAP_SYS_ADMIN))
453 mode = sysfs_match_string(zeroing_mode, buf);
457 sdkp->zeroing_mode = mode;
461 static DEVICE_ATTR_RW(zeroing_mode);
464 max_medium_access_timeouts_show(struct device *dev,
465 struct device_attribute *attr, char *buf)
467 struct scsi_disk *sdkp = to_scsi_disk(dev);
469 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
473 max_medium_access_timeouts_store(struct device *dev,
474 struct device_attribute *attr, const char *buf,
477 struct scsi_disk *sdkp = to_scsi_disk(dev);
480 if (!capable(CAP_SYS_ADMIN))
483 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
485 return err ? err : count;
487 static DEVICE_ATTR_RW(max_medium_access_timeouts);
490 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
493 struct scsi_disk *sdkp = to_scsi_disk(dev);
495 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
499 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
500 const char *buf, size_t count)
502 struct scsi_disk *sdkp = to_scsi_disk(dev);
503 struct scsi_device *sdp = sdkp->device;
507 if (!capable(CAP_SYS_ADMIN))
510 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
513 err = kstrtoul(buf, 10, &max);
519 sdp->no_write_same = 1;
520 else if (max <= SD_MAX_WS16_BLOCKS) {
521 sdp->no_write_same = 0;
522 sdkp->max_ws_blocks = max;
525 sd_config_write_same(sdkp);
529 static DEVICE_ATTR_RW(max_write_same_blocks);
532 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
534 struct scsi_disk *sdkp = to_scsi_disk(dev);
536 if (sdkp->device->type == TYPE_ZBC)
537 return sprintf(buf, "host-managed\n");
538 if (sdkp->zoned == 1)
539 return sprintf(buf, "host-aware\n");
540 if (sdkp->zoned == 2)
541 return sprintf(buf, "drive-managed\n");
542 return sprintf(buf, "none\n");
544 static DEVICE_ATTR_RO(zoned_cap);
546 static struct attribute *sd_disk_attrs[] = {
547 &dev_attr_cache_type.attr,
549 &dev_attr_allow_restart.attr,
550 &dev_attr_manage_start_stop.attr,
551 &dev_attr_protection_type.attr,
552 &dev_attr_protection_mode.attr,
553 &dev_attr_app_tag_own.attr,
554 &dev_attr_thin_provisioning.attr,
555 &dev_attr_provisioning_mode.attr,
556 &dev_attr_zeroing_mode.attr,
557 &dev_attr_max_write_same_blocks.attr,
558 &dev_attr_max_medium_access_timeouts.attr,
559 &dev_attr_zoned_cap.attr,
562 ATTRIBUTE_GROUPS(sd_disk);
564 static struct class sd_disk_class = {
566 .owner = THIS_MODULE,
567 .dev_release = scsi_disk_release,
568 .dev_groups = sd_disk_groups,
571 static const struct dev_pm_ops sd_pm_ops = {
572 .suspend = sd_suspend_system,
574 .poweroff = sd_suspend_system,
575 .restore = sd_resume,
576 .runtime_suspend = sd_suspend_runtime,
577 .runtime_resume = sd_resume,
580 static struct scsi_driver sd_template = {
583 .owner = THIS_MODULE,
585 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
587 .shutdown = sd_shutdown,
591 .init_command = sd_init_command,
592 .uninit_command = sd_uninit_command,
594 .eh_action = sd_eh_action,
595 .eh_reset = sd_eh_reset,
599 * Dummy kobj_map->probe function.
600 * The default ->probe function will call modprobe, which is
601 * pointless as this module is already loaded.
603 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
609 * Device no to disk mapping:
611 * major disc2 disc p1
612 * |............|.............|....|....| <- dev_t
615 * Inside a major, we have 16k disks, however mapped non-
616 * contiguously. The first 16 disks are for major0, the next
617 * ones with major1, ... Disk 256 is for major0 again, disk 272
619 * As we stay compatible with our numbering scheme, we can reuse
620 * the well-know SCSI majors 8, 65--71, 136--143.
622 static int sd_major(int major_idx)
626 return SCSI_DISK0_MAJOR;
628 return SCSI_DISK1_MAJOR + major_idx - 1;
630 return SCSI_DISK8_MAJOR + major_idx - 8;
633 return 0; /* shut up gcc */
637 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
639 struct scsi_disk *sdkp = NULL;
641 mutex_lock(&sd_ref_mutex);
643 if (disk->private_data) {
644 sdkp = scsi_disk(disk);
645 if (scsi_device_get(sdkp->device) == 0)
646 get_device(&sdkp->dev);
650 mutex_unlock(&sd_ref_mutex);
654 static void scsi_disk_put(struct scsi_disk *sdkp)
656 struct scsi_device *sdev = sdkp->device;
658 mutex_lock(&sd_ref_mutex);
659 put_device(&sdkp->dev);
660 scsi_device_put(sdev);
661 mutex_unlock(&sd_ref_mutex);
664 #ifdef CONFIG_BLK_SED_OPAL
665 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
666 size_t len, bool send)
668 struct scsi_device *sdev = data;
672 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
674 put_unaligned_be16(spsp, &cdb[2]);
675 put_unaligned_be32(len, &cdb[6]);
677 ret = scsi_execute_req(sdev, cdb,
678 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
679 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
680 return ret <= 0 ? ret : -EIO;
682 #endif /* CONFIG_BLK_SED_OPAL */
685 * Look up the DIX operation based on whether the command is read or
686 * write and whether dix and dif are enabled.
688 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
690 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
691 static const unsigned int ops[] = { /* wrt dix dif */
692 SCSI_PROT_NORMAL, /* 0 0 0 */
693 SCSI_PROT_READ_STRIP, /* 0 0 1 */
694 SCSI_PROT_READ_INSERT, /* 0 1 0 */
695 SCSI_PROT_READ_PASS, /* 0 1 1 */
696 SCSI_PROT_NORMAL, /* 1 0 0 */
697 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
698 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
699 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
702 return ops[write << 2 | dix << 1 | dif];
706 * Returns a mask of the protection flags that are valid for a given DIX
709 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
711 static const unsigned int flag_mask[] = {
712 [SCSI_PROT_NORMAL] = 0,
714 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
715 SCSI_PROT_GUARD_CHECK |
716 SCSI_PROT_REF_CHECK |
717 SCSI_PROT_REF_INCREMENT,
719 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
720 SCSI_PROT_IP_CHECKSUM,
722 [SCSI_PROT_READ_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,
728 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
729 SCSI_PROT_REF_INCREMENT,
731 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
732 SCSI_PROT_REF_CHECK |
733 SCSI_PROT_REF_INCREMENT |
734 SCSI_PROT_IP_CHECKSUM,
736 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
737 SCSI_PROT_GUARD_CHECK |
738 SCSI_PROT_REF_CHECK |
739 SCSI_PROT_REF_INCREMENT |
740 SCSI_PROT_IP_CHECKSUM,
743 return flag_mask[prot_op];
746 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
747 unsigned int dix, unsigned int dif)
749 struct bio *bio = scmd->request->bio;
750 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
751 unsigned int protect = 0;
753 if (dix) { /* DIX Type 0, 1, 2, 3 */
754 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
755 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
757 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
758 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
761 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
762 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
764 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
765 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
768 if (dif) { /* DIX/DIF Type 1, 2, 3 */
769 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
771 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
772 protect = 3 << 5; /* Disable target PI checking */
774 protect = 1 << 5; /* Enable target PI checking */
777 scsi_set_prot_op(scmd, prot_op);
778 scsi_set_prot_type(scmd, dif);
779 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
784 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
786 struct request_queue *q = sdkp->disk->queue;
787 unsigned int logical_block_size = sdkp->device->sector_size;
788 unsigned int max_blocks = 0;
790 q->limits.discard_alignment =
791 sdkp->unmap_alignment * logical_block_size;
792 q->limits.discard_granularity =
793 max(sdkp->physical_block_size,
794 sdkp->unmap_granularity * logical_block_size);
795 sdkp->provisioning_mode = mode;
801 blk_queue_max_discard_sectors(q, 0);
802 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
806 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
807 (u32)SD_MAX_WS16_BLOCKS);
811 if (sdkp->device->unmap_limit_for_ws)
812 max_blocks = sdkp->max_unmap_blocks;
814 max_blocks = sdkp->max_ws_blocks;
816 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
820 if (sdkp->device->unmap_limit_for_ws)
821 max_blocks = sdkp->max_unmap_blocks;
823 max_blocks = sdkp->max_ws_blocks;
825 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
829 max_blocks = min_not_zero(sdkp->max_ws_blocks,
830 (u32)SD_MAX_WS10_BLOCKS);
834 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
835 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
838 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
840 struct scsi_device *sdp = cmd->device;
841 struct request *rq = cmd->request;
842 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
843 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
844 unsigned int data_len = 24;
847 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
848 if (!rq->special_vec.bv_page)
849 return BLK_STS_RESOURCE;
850 clear_highpage(rq->special_vec.bv_page);
851 rq->special_vec.bv_offset = 0;
852 rq->special_vec.bv_len = data_len;
853 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
856 cmd->cmnd[0] = UNMAP;
859 buf = page_address(rq->special_vec.bv_page);
860 put_unaligned_be16(6 + 16, &buf[0]);
861 put_unaligned_be16(16, &buf[2]);
862 put_unaligned_be64(lba, &buf[8]);
863 put_unaligned_be32(nr_blocks, &buf[16]);
865 cmd->allowed = SD_MAX_RETRIES;
866 cmd->transfersize = data_len;
867 rq->timeout = SD_TIMEOUT;
869 return scsi_init_io(cmd);
872 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
875 struct scsi_device *sdp = cmd->device;
876 struct request *rq = cmd->request;
877 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
878 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
879 u32 data_len = sdp->sector_size;
881 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 if (!rq->special_vec.bv_page)
883 return BLK_STS_RESOURCE;
884 clear_highpage(rq->special_vec.bv_page);
885 rq->special_vec.bv_offset = 0;
886 rq->special_vec.bv_len = data_len;
887 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
890 cmd->cmnd[0] = WRITE_SAME_16;
892 cmd->cmnd[1] = 0x8; /* UNMAP */
893 put_unaligned_be64(lba, &cmd->cmnd[2]);
894 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
896 cmd->allowed = SD_MAX_RETRIES;
897 cmd->transfersize = data_len;
898 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
900 return scsi_init_io(cmd);
903 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
906 struct scsi_device *sdp = cmd->device;
907 struct request *rq = cmd->request;
908 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
909 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
910 u32 data_len = sdp->sector_size;
912 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
913 if (!rq->special_vec.bv_page)
914 return BLK_STS_RESOURCE;
915 clear_highpage(rq->special_vec.bv_page);
916 rq->special_vec.bv_offset = 0;
917 rq->special_vec.bv_len = data_len;
918 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
921 cmd->cmnd[0] = WRITE_SAME;
923 cmd->cmnd[1] = 0x8; /* UNMAP */
924 put_unaligned_be32(lba, &cmd->cmnd[2]);
925 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
927 cmd->allowed = SD_MAX_RETRIES;
928 cmd->transfersize = data_len;
929 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
931 return scsi_init_io(cmd);
934 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
936 struct request *rq = cmd->request;
937 struct scsi_device *sdp = cmd->device;
938 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
939 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
940 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
942 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
943 switch (sdkp->zeroing_mode) {
944 case SD_ZERO_WS16_UNMAP:
945 return sd_setup_write_same16_cmnd(cmd, true);
946 case SD_ZERO_WS10_UNMAP:
947 return sd_setup_write_same10_cmnd(cmd, true);
951 if (sdp->no_write_same)
952 return BLK_STS_TARGET;
954 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
955 return sd_setup_write_same16_cmnd(cmd, false);
957 return sd_setup_write_same10_cmnd(cmd, false);
960 static void sd_config_write_same(struct scsi_disk *sdkp)
962 struct request_queue *q = sdkp->disk->queue;
963 unsigned int logical_block_size = sdkp->device->sector_size;
965 if (sdkp->device->no_write_same) {
966 sdkp->max_ws_blocks = 0;
970 /* Some devices can not handle block counts above 0xffff despite
971 * supporting WRITE SAME(16). Consequently we default to 64k
972 * blocks per I/O unless the device explicitly advertises a
975 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
976 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
977 (u32)SD_MAX_WS16_BLOCKS);
978 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
979 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
980 (u32)SD_MAX_WS10_BLOCKS);
982 sdkp->device->no_write_same = 1;
983 sdkp->max_ws_blocks = 0;
986 if (sdkp->lbprz && sdkp->lbpws)
987 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
988 else if (sdkp->lbprz && sdkp->lbpws10)
989 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
990 else if (sdkp->max_ws_blocks)
991 sdkp->zeroing_mode = SD_ZERO_WS;
993 sdkp->zeroing_mode = SD_ZERO_WRITE;
995 if (sdkp->max_ws_blocks &&
996 sdkp->physical_block_size > logical_block_size) {
998 * Reporting a maximum number of blocks that is not aligned
999 * on the device physical size would cause a large write same
1000 * request to be split into physically unaligned chunks by
1001 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1002 * even if the caller of these functions took care to align the
1003 * large request. So make sure the maximum reported is aligned
1004 * to the device physical block size. This is only an optional
1005 * optimization for regular disks, but this is mandatory to
1006 * avoid failure of large write same requests directed at
1007 * sequential write required zones of host-managed ZBC disks.
1009 sdkp->max_ws_blocks =
1010 round_down(sdkp->max_ws_blocks,
1011 bytes_to_logical(sdkp->device,
1012 sdkp->physical_block_size));
1016 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1017 (logical_block_size >> 9));
1018 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1019 (logical_block_size >> 9));
1023 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1024 * @cmd: command to prepare
1026 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1027 * the preference indicated by the target device.
1029 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1031 struct request *rq = cmd->request;
1032 struct scsi_device *sdp = cmd->device;
1033 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1034 struct bio *bio = rq->bio;
1035 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1036 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1039 if (sdkp->device->no_write_same)
1040 return BLK_STS_TARGET;
1042 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1044 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1046 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1048 cmd->cmnd[0] = WRITE_SAME_16;
1049 put_unaligned_be64(lba, &cmd->cmnd[2]);
1050 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1053 cmd->cmnd[0] = WRITE_SAME;
1054 put_unaligned_be32(lba, &cmd->cmnd[2]);
1055 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1058 cmd->transfersize = sdp->sector_size;
1059 cmd->allowed = SD_MAX_RETRIES;
1062 * For WRITE SAME the data transferred via the DATA OUT buffer is
1063 * different from the amount of data actually written to the target.
1065 * We set up __data_len to the amount of data transferred via the
1066 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1067 * to transfer a single sector of data first, but then reset it to
1068 * the amount of data to be written right after so that the I/O path
1069 * knows how much to actually write.
1071 rq->__data_len = sdp->sector_size;
1072 ret = scsi_init_io(cmd);
1073 rq->__data_len = blk_rq_bytes(rq);
1078 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1080 struct request *rq = cmd->request;
1082 /* flush requests don't perform I/O, zero the S/G table */
1083 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1085 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1087 cmd->transfersize = 0;
1088 cmd->allowed = SD_MAX_RETRIES;
1090 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1094 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1095 sector_t lba, unsigned int nr_blocks,
1096 unsigned char flags)
1098 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1099 if (unlikely(cmd->cmnd == NULL))
1100 return BLK_STS_RESOURCE;
1102 cmd->cmd_len = SD_EXT_CDB_SIZE;
1103 memset(cmd->cmnd, 0, cmd->cmd_len);
1105 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1106 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1107 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1108 cmd->cmnd[10] = flags;
1109 put_unaligned_be64(lba, &cmd->cmnd[12]);
1110 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1111 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1116 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1117 sector_t lba, unsigned int nr_blocks,
1118 unsigned char flags)
1121 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1122 cmd->cmnd[1] = flags;
1125 put_unaligned_be64(lba, &cmd->cmnd[2]);
1126 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1131 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1132 sector_t lba, unsigned int nr_blocks,
1133 unsigned char flags)
1136 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1137 cmd->cmnd[1] = flags;
1140 put_unaligned_be32(lba, &cmd->cmnd[2]);
1141 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1146 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1147 sector_t lba, unsigned int nr_blocks,
1148 unsigned char flags)
1150 /* Avoid that 0 blocks gets translated into 256 blocks. */
1151 if (WARN_ON_ONCE(nr_blocks == 0))
1152 return BLK_STS_IOERR;
1154 if (unlikely(flags & 0x8)) {
1156 * This happens only if this drive failed 10byte rw
1157 * command with ILLEGAL_REQUEST during operation and
1158 * thus turned off use_10_for_rw.
1160 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1161 return BLK_STS_IOERR;
1165 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1166 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1167 cmd->cmnd[2] = (lba >> 8) & 0xff;
1168 cmd->cmnd[3] = lba & 0xff;
1169 cmd->cmnd[4] = nr_blocks;
1175 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1177 struct request *rq = cmd->request;
1178 struct scsi_device *sdp = cmd->device;
1179 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1180 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1182 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1183 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1184 bool write = rq_data_dir(rq) == WRITE;
1185 unsigned char protect, fua;
1190 ret = scsi_init_io(cmd);
1191 if (ret != BLK_STS_OK)
1194 if (!scsi_device_online(sdp) || sdp->changed) {
1195 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1196 return BLK_STS_IOERR;
1199 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1200 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1201 return BLK_STS_IOERR;
1204 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1205 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1206 return BLK_STS_IOERR;
1210 * Some SD card readers can't handle accesses which touch the
1211 * last one or two logical blocks. Split accesses as needed.
1213 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1215 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1216 if (lba < threshold) {
1217 /* Access up to the threshold but not beyond */
1218 nr_blocks = threshold - lba;
1220 /* Access only a single logical block */
1225 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1226 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1231 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1232 dix = scsi_prot_sg_count(cmd);
1233 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1236 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1240 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1241 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1243 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1244 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1246 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1247 sdp->use_10_for_rw || protect) {
1248 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1251 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1255 if (unlikely(ret != BLK_STS_OK))
1259 * We shouldn't disconnect in the middle of a sector, so with a dumb
1260 * host adapter, it's safe to assume that we can at least transfer
1261 * this many bytes between each connect / disconnect.
1263 cmd->transfersize = sdp->sector_size;
1264 cmd->underflow = nr_blocks << 9;
1265 cmd->allowed = SD_MAX_RETRIES;
1266 cmd->sdb.length = nr_blocks * sdp->sector_size;
1269 scmd_printk(KERN_INFO, cmd,
1270 "%s: block=%llu, count=%d\n", __func__,
1271 (unsigned long long)blk_rq_pos(rq),
1272 blk_rq_sectors(rq)));
1274 scmd_printk(KERN_INFO, cmd,
1275 "%s %d/%u 512 byte blocks.\n",
1276 write ? "writing" : "reading", nr_blocks,
1277 blk_rq_sectors(rq)));
1280 * This indicates that the command is ready from our end to be
1286 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1288 struct request *rq = cmd->request;
1290 switch (req_op(rq)) {
1291 case REQ_OP_DISCARD:
1292 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1294 return sd_setup_unmap_cmnd(cmd);
1296 return sd_setup_write_same16_cmnd(cmd, true);
1298 return sd_setup_write_same10_cmnd(cmd, true);
1300 return sd_setup_write_same10_cmnd(cmd, false);
1302 return BLK_STS_TARGET;
1304 case REQ_OP_WRITE_ZEROES:
1305 return sd_setup_write_zeroes_cmnd(cmd);
1306 case REQ_OP_WRITE_SAME:
1307 return sd_setup_write_same_cmnd(cmd);
1309 return sd_setup_flush_cmnd(cmd);
1312 case REQ_OP_ZONE_APPEND:
1313 return sd_setup_read_write_cmnd(cmd);
1314 case REQ_OP_ZONE_RESET:
1315 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1317 case REQ_OP_ZONE_RESET_ALL:
1318 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1320 case REQ_OP_ZONE_OPEN:
1321 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1322 case REQ_OP_ZONE_CLOSE:
1323 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1324 case REQ_OP_ZONE_FINISH:
1325 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1328 return BLK_STS_NOTSUPP;
1332 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1334 struct request *rq = SCpnt->request;
1337 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1338 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1340 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1344 mempool_free(cmnd, sd_cdb_pool);
1349 * sd_open - open a scsi disk device
1350 * @bdev: Block device of the scsi disk to open
1351 * @mode: FMODE_* mask
1353 * Returns 0 if successful. Returns a negated errno value in case
1356 * Note: This can be called from a user context (e.g. fsck(1) )
1357 * or from within the kernel (e.g. as a result of a mount(1) ).
1358 * In the latter case @inode and @filp carry an abridged amount
1359 * of information as noted above.
1361 * Locking: called with bdev->bd_mutex held.
1363 static int sd_open(struct block_device *bdev, fmode_t mode)
1365 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1366 struct scsi_device *sdev;
1372 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1374 sdev = sdkp->device;
1377 * If the device is in error recovery, wait until it is done.
1378 * If the device is offline, then disallow any access to it.
1381 if (!scsi_block_when_processing_errors(sdev))
1384 if (sdev->removable || sdkp->write_prot) {
1385 if (bdev_check_media_change(bdev))
1386 sd_revalidate_disk(bdev->bd_disk);
1390 * If the drive is empty, just let the open fail.
1392 retval = -ENOMEDIUM;
1393 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1397 * If the device has the write protect tab set, have the open fail
1398 * if the user expects to be able to write to the thing.
1401 if (sdkp->write_prot && (mode & FMODE_WRITE))
1405 * It is possible that the disk changing stuff resulted in
1406 * the device being taken offline. If this is the case,
1407 * report this to the user, and don't pretend that the
1408 * open actually succeeded.
1411 if (!scsi_device_online(sdev))
1414 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1415 if (scsi_block_when_processing_errors(sdev))
1416 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1422 scsi_disk_put(sdkp);
1427 * sd_release - invoked when the (last) close(2) is called on this
1429 * @disk: disk to release
1430 * @mode: FMODE_* mask
1434 * Note: may block (uninterruptible) if error recovery is underway
1437 * Locking: called with bdev->bd_mutex held.
1439 static void sd_release(struct gendisk *disk, fmode_t mode)
1441 struct scsi_disk *sdkp = scsi_disk(disk);
1442 struct scsi_device *sdev = sdkp->device;
1444 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1446 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1447 if (scsi_block_when_processing_errors(sdev))
1448 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1451 scsi_disk_put(sdkp);
1454 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1456 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1457 struct scsi_device *sdp = sdkp->device;
1458 struct Scsi_Host *host = sdp->host;
1459 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1462 /* default to most commonly used values */
1463 diskinfo[0] = 0x40; /* 1 << 6 */
1464 diskinfo[1] = 0x20; /* 1 << 5 */
1465 diskinfo[2] = capacity >> 11;
1467 /* override with calculated, extended default, or driver values */
1468 if (host->hostt->bios_param)
1469 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1471 scsicam_bios_param(bdev, capacity, diskinfo);
1473 geo->heads = diskinfo[0];
1474 geo->sectors = diskinfo[1];
1475 geo->cylinders = diskinfo[2];
1480 * sd_ioctl - process an ioctl
1481 * @bdev: target block device
1482 * @mode: FMODE_* mask
1483 * @cmd: ioctl command number
1484 * @p: this is third argument given to ioctl(2) system call.
1485 * Often contains a pointer.
1487 * Returns 0 if successful (some ioctls return positive numbers on
1488 * success as well). Returns a negated errno value in case of error.
1490 * Note: most ioctls are forward onto the block subsystem or further
1491 * down in the scsi subsystem.
1493 static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1494 unsigned int cmd, void __user *p)
1496 struct gendisk *disk = bdev->bd_disk;
1497 struct scsi_disk *sdkp = scsi_disk(disk);
1498 struct scsi_device *sdp = sdkp->device;
1501 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1502 "cmd=0x%x\n", disk->disk_name, cmd));
1504 error = scsi_verify_blk_ioctl(bdev, cmd);
1509 * If we are in the middle of error recovery, don't let anyone
1510 * else try and use this device. Also, if error recovery fails, it
1511 * may try and take the device offline, in which case all further
1512 * access to the device is prohibited.
1514 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1515 (mode & FMODE_NDELAY) != 0);
1519 if (is_sed_ioctl(cmd))
1520 return sed_ioctl(sdkp->opal_dev, cmd, p);
1523 * Send SCSI addressing ioctls directly to mid level, send other
1524 * ioctls to block level and then onto mid level if they can't be
1528 case SCSI_IOCTL_GET_IDLUN:
1529 case SCSI_IOCTL_GET_BUS_NUMBER:
1530 error = scsi_ioctl(sdp, cmd, p);
1533 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1540 static void set_media_not_present(struct scsi_disk *sdkp)
1542 if (sdkp->media_present)
1543 sdkp->device->changed = 1;
1545 if (sdkp->device->removable) {
1546 sdkp->media_present = 0;
1551 static int media_not_present(struct scsi_disk *sdkp,
1552 struct scsi_sense_hdr *sshdr)
1554 if (!scsi_sense_valid(sshdr))
1557 /* not invoked for commands that could return deferred errors */
1558 switch (sshdr->sense_key) {
1559 case UNIT_ATTENTION:
1561 /* medium not present */
1562 if (sshdr->asc == 0x3A) {
1563 set_media_not_present(sdkp);
1571 * sd_check_events - check media events
1572 * @disk: kernel device descriptor
1573 * @clearing: disk events currently being cleared
1575 * Returns mask of DISK_EVENT_*.
1577 * Note: this function is invoked from the block subsystem.
1579 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1581 struct scsi_disk *sdkp = scsi_disk_get(disk);
1582 struct scsi_device *sdp;
1589 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1592 * If the device is offline, don't send any commands - just pretend as
1593 * if the command failed. If the device ever comes back online, we
1594 * can deal with it then. It is only because of unrecoverable errors
1595 * that we would ever take a device offline in the first place.
1597 if (!scsi_device_online(sdp)) {
1598 set_media_not_present(sdkp);
1603 * Using TEST_UNIT_READY enables differentiation between drive with
1604 * no cartridge loaded - NOT READY, drive with changed cartridge -
1605 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1607 * Drives that auto spin down. eg iomega jaz 1G, will be started
1608 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1609 * sd_revalidate() is called.
1611 if (scsi_block_when_processing_errors(sdp)) {
1612 struct scsi_sense_hdr sshdr = { 0, };
1614 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1617 /* failed to execute TUR, assume media not present */
1618 if (host_byte(retval)) {
1619 set_media_not_present(sdkp);
1623 if (media_not_present(sdkp, &sshdr))
1628 * For removable scsi disk we have to recognise the presence
1629 * of a disk in the drive.
1631 if (!sdkp->media_present)
1633 sdkp->media_present = 1;
1636 * sdp->changed is set under the following conditions:
1638 * Medium present state has changed in either direction.
1639 * Device has indicated UNIT_ATTENTION.
1641 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1643 scsi_disk_put(sdkp);
1647 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1650 struct scsi_device *sdp = sdkp->device;
1651 const int timeout = sdp->request_queue->rq_timeout
1652 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1653 struct scsi_sense_hdr my_sshdr;
1655 if (!scsi_device_online(sdp))
1658 /* caller might not be interested in sense, but we need it */
1662 for (retries = 3; retries > 0; --retries) {
1663 unsigned char cmd[10] = { 0 };
1665 cmd[0] = SYNCHRONIZE_CACHE;
1667 * Leave the rest of the command zero to indicate
1670 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1671 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1677 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1679 if (driver_byte(res) == DRIVER_SENSE)
1680 sd_print_sense_hdr(sdkp, sshdr);
1682 /* we need to evaluate the error return */
1683 if (scsi_sense_valid(sshdr) &&
1684 (sshdr->asc == 0x3a || /* medium not present */
1685 sshdr->asc == 0x20 || /* invalid command */
1686 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1687 /* this is no error here */
1690 switch (host_byte(res)) {
1691 /* ignore errors due to racing a disconnection */
1692 case DID_BAD_TARGET:
1693 case DID_NO_CONNECT:
1695 /* signal the upper layer it might try again */
1699 case DID_SOFT_ERROR:
1708 static void sd_rescan(struct device *dev)
1710 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1713 ret = sd_revalidate_disk(sdkp->disk);
1714 revalidate_disk_size(sdkp->disk, ret == 0);
1717 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1718 unsigned int cmd, unsigned long arg)
1720 void __user *p = (void __user *)arg;
1723 ret = sd_ioctl_common(bdev, mode, cmd, p);
1727 return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1730 #ifdef CONFIG_COMPAT
1731 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1732 unsigned int cmd, unsigned long arg)
1734 void __user *p = compat_ptr(arg);
1737 ret = sd_ioctl_common(bdev, mode, cmd, p);
1741 return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1745 static char sd_pr_type(enum pr_type type)
1748 case PR_WRITE_EXCLUSIVE:
1750 case PR_EXCLUSIVE_ACCESS:
1752 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1754 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1756 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1758 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1765 static int sd_pr_command(struct block_device *bdev, u8 sa,
1766 u64 key, u64 sa_key, u8 type, u8 flags)
1768 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1769 struct scsi_sense_hdr sshdr;
1771 u8 cmd[16] = { 0, };
1772 u8 data[24] = { 0, };
1774 cmd[0] = PERSISTENT_RESERVE_OUT;
1777 put_unaligned_be32(sizeof(data), &cmd[5]);
1779 put_unaligned_be64(key, &data[0]);
1780 put_unaligned_be64(sa_key, &data[8]);
1783 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1784 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1786 if (driver_byte(result) == DRIVER_SENSE &&
1787 scsi_sense_valid(&sshdr)) {
1788 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1789 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1795 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1798 if (flags & ~PR_FL_IGNORE_KEY)
1800 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1801 old_key, new_key, 0,
1802 (1 << 0) /* APTPL */);
1805 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1810 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1813 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1815 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1818 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1819 enum pr_type type, bool abort)
1821 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1822 sd_pr_type(type), 0);
1825 static int sd_pr_clear(struct block_device *bdev, u64 key)
1827 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1830 static const struct pr_ops sd_pr_ops = {
1831 .pr_register = sd_pr_register,
1832 .pr_reserve = sd_pr_reserve,
1833 .pr_release = sd_pr_release,
1834 .pr_preempt = sd_pr_preempt,
1835 .pr_clear = sd_pr_clear,
1838 static const struct block_device_operations sd_fops = {
1839 .owner = THIS_MODULE,
1841 .release = sd_release,
1843 .getgeo = sd_getgeo,
1844 #ifdef CONFIG_COMPAT
1845 .compat_ioctl = sd_compat_ioctl,
1847 .check_events = sd_check_events,
1848 .unlock_native_capacity = sd_unlock_native_capacity,
1849 .report_zones = sd_zbc_report_zones,
1850 .pr_ops = &sd_pr_ops,
1854 * sd_eh_reset - reset error handling callback
1855 * @scmd: sd-issued command that has failed
1857 * This function is called by the SCSI midlayer before starting
1858 * SCSI EH. When counting medium access failures we have to be
1859 * careful to register it only only once per device and SCSI EH run;
1860 * there might be several timed out commands which will cause the
1861 * 'max_medium_access_timeouts' counter to trigger after the first
1862 * SCSI EH run already and set the device to offline.
1863 * So this function resets the internal counter before starting SCSI EH.
1865 static void sd_eh_reset(struct scsi_cmnd *scmd)
1867 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1869 /* New SCSI EH run, reset gate variable */
1870 sdkp->ignore_medium_access_errors = false;
1874 * sd_eh_action - error handling callback
1875 * @scmd: sd-issued command that has failed
1876 * @eh_disp: The recovery disposition suggested by the midlayer
1878 * This function is called by the SCSI midlayer upon completion of an
1879 * error test command (currently TEST UNIT READY). The result of sending
1880 * the eh command is passed in eh_disp. We're looking for devices that
1881 * fail medium access commands but are OK with non access commands like
1882 * test unit ready (so wrongly see the device as having a successful
1885 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1887 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1888 struct scsi_device *sdev = scmd->device;
1890 if (!scsi_device_online(sdev) ||
1891 !scsi_medium_access_command(scmd) ||
1892 host_byte(scmd->result) != DID_TIME_OUT ||
1897 * The device has timed out executing a medium access command.
1898 * However, the TEST UNIT READY command sent during error
1899 * handling completed successfully. Either the device is in the
1900 * process of recovering or has it suffered an internal failure
1901 * that prevents access to the storage medium.
1903 if (!sdkp->ignore_medium_access_errors) {
1904 sdkp->medium_access_timed_out++;
1905 sdkp->ignore_medium_access_errors = true;
1909 * If the device keeps failing read/write commands but TEST UNIT
1910 * READY always completes successfully we assume that medium
1911 * access is no longer possible and take the device offline.
1913 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1914 scmd_printk(KERN_ERR, scmd,
1915 "Medium access timeout failure. Offlining disk!\n");
1916 mutex_lock(&sdev->state_mutex);
1917 scsi_device_set_state(sdev, SDEV_OFFLINE);
1918 mutex_unlock(&sdev->state_mutex);
1926 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1928 struct request *req = scmd->request;
1929 struct scsi_device *sdev = scmd->device;
1930 unsigned int transferred, good_bytes;
1931 u64 start_lba, end_lba, bad_lba;
1934 * Some commands have a payload smaller than the device logical
1935 * block size (e.g. INQUIRY on a 4K disk).
1937 if (scsi_bufflen(scmd) <= sdev->sector_size)
1940 /* Check if we have a 'bad_lba' information */
1941 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1942 SCSI_SENSE_BUFFERSIZE,
1947 * If the bad lba was reported incorrectly, we have no idea where
1950 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1951 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1952 if (bad_lba < start_lba || bad_lba >= end_lba)
1956 * resid is optional but mostly filled in. When it's unused,
1957 * its value is zero, so we assume the whole buffer transferred
1959 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1961 /* This computation should always be done in terms of the
1962 * resolution of the device's medium.
1964 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1966 return min(good_bytes, transferred);
1970 * sd_done - bottom half handler: called when the lower level
1971 * driver has completed (successfully or otherwise) a scsi command.
1972 * @SCpnt: mid-level's per command structure.
1974 * Note: potentially run from within an ISR. Must not block.
1976 static int sd_done(struct scsi_cmnd *SCpnt)
1978 int result = SCpnt->result;
1979 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1980 unsigned int sector_size = SCpnt->device->sector_size;
1982 struct scsi_sense_hdr sshdr;
1983 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1984 struct request *req = SCpnt->request;
1985 int sense_valid = 0;
1986 int sense_deferred = 0;
1988 switch (req_op(req)) {
1989 case REQ_OP_DISCARD:
1990 case REQ_OP_WRITE_ZEROES:
1991 case REQ_OP_WRITE_SAME:
1992 case REQ_OP_ZONE_RESET:
1993 case REQ_OP_ZONE_RESET_ALL:
1994 case REQ_OP_ZONE_OPEN:
1995 case REQ_OP_ZONE_CLOSE:
1996 case REQ_OP_ZONE_FINISH:
1998 good_bytes = blk_rq_bytes(req);
1999 scsi_set_resid(SCpnt, 0);
2002 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2007 * In case of bogus fw or device, we could end up having
2008 * an unaligned partial completion. Check this here and force
2011 resid = scsi_get_resid(SCpnt);
2012 if (resid & (sector_size - 1)) {
2013 sd_printk(KERN_INFO, sdkp,
2014 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2015 resid, sector_size);
2016 scsi_print_command(SCpnt);
2017 resid = min(scsi_bufflen(SCpnt),
2018 round_up(resid, sector_size));
2019 scsi_set_resid(SCpnt, resid);
2024 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2026 sense_deferred = scsi_sense_is_deferred(&sshdr);
2028 sdkp->medium_access_timed_out = 0;
2030 if (driver_byte(result) != DRIVER_SENSE &&
2031 (!sense_valid || sense_deferred))
2034 switch (sshdr.sense_key) {
2035 case HARDWARE_ERROR:
2037 good_bytes = sd_completed_bytes(SCpnt);
2039 case RECOVERED_ERROR:
2040 good_bytes = scsi_bufflen(SCpnt);
2043 /* This indicates a false check condition, so ignore it. An
2044 * unknown amount of data was transferred so treat it as an
2048 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2050 case ABORTED_COMMAND:
2051 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2052 good_bytes = sd_completed_bytes(SCpnt);
2054 case ILLEGAL_REQUEST:
2055 switch (sshdr.asc) {
2056 case 0x10: /* DIX: Host detected corruption */
2057 good_bytes = sd_completed_bytes(SCpnt);
2059 case 0x20: /* INVALID COMMAND OPCODE */
2060 case 0x24: /* INVALID FIELD IN CDB */
2061 switch (SCpnt->cmnd[0]) {
2063 sd_config_discard(sdkp, SD_LBP_DISABLE);
2067 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2068 sd_config_discard(sdkp, SD_LBP_DISABLE);
2070 sdkp->device->no_write_same = 1;
2071 sd_config_write_same(sdkp);
2072 req->rq_flags |= RQF_QUIET;
2083 if (sd_is_zoned(sdkp))
2084 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2086 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2087 "sd_done: completed %d of %d bytes\n",
2088 good_bytes, scsi_bufflen(SCpnt)));
2094 * spinup disk - called only in sd_revalidate_disk()
2097 sd_spinup_disk(struct scsi_disk *sdkp)
2099 unsigned char cmd[10];
2100 unsigned long spintime_expire = 0;
2101 int retries, spintime;
2102 unsigned int the_result;
2103 struct scsi_sense_hdr sshdr;
2104 int sense_valid = 0;
2108 /* Spin up drives, as required. Only do this at boot time */
2109 /* Spinup needs to be done for module loads too. */
2114 cmd[0] = TEST_UNIT_READY;
2115 memset((void *) &cmd[1], 0, 9);
2117 the_result = scsi_execute_req(sdkp->device, cmd,
2120 SD_MAX_RETRIES, NULL);
2123 * If the drive has indicated to us that it
2124 * doesn't have any media in it, don't bother
2125 * with any more polling.
2127 if (media_not_present(sdkp, &sshdr))
2131 sense_valid = scsi_sense_valid(&sshdr);
2133 } while (retries < 3 &&
2134 (!scsi_status_is_good(the_result) ||
2135 ((driver_byte(the_result) == DRIVER_SENSE) &&
2136 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2138 if (driver_byte(the_result) != DRIVER_SENSE) {
2139 /* no sense, TUR either succeeded or failed
2140 * with a status error */
2141 if(!spintime && !scsi_status_is_good(the_result)) {
2142 sd_print_result(sdkp, "Test Unit Ready failed",
2149 * The device does not want the automatic start to be issued.
2151 if (sdkp->device->no_start_on_add)
2154 if (sense_valid && sshdr.sense_key == NOT_READY) {
2155 if (sshdr.asc == 4 && sshdr.ascq == 3)
2156 break; /* manual intervention required */
2157 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2158 break; /* standby */
2159 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2160 break; /* unavailable */
2161 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2162 break; /* sanitize in progress */
2164 * Issue command to spin up drive when not ready
2167 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2168 cmd[0] = START_STOP;
2169 cmd[1] = 1; /* Return immediately */
2170 memset((void *) &cmd[2], 0, 8);
2171 cmd[4] = 1; /* Start spin cycle */
2172 if (sdkp->device->start_stop_pwr_cond)
2174 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2176 SD_TIMEOUT, SD_MAX_RETRIES,
2178 spintime_expire = jiffies + 100 * HZ;
2181 /* Wait 1 second for next try */
2183 printk(KERN_CONT ".");
2186 * Wait for USB flash devices with slow firmware.
2187 * Yes, this sense key/ASC combination shouldn't
2188 * occur here. It's characteristic of these devices.
2190 } else if (sense_valid &&
2191 sshdr.sense_key == UNIT_ATTENTION &&
2192 sshdr.asc == 0x28) {
2194 spintime_expire = jiffies + 5 * HZ;
2197 /* Wait 1 second for next try */
2200 /* we don't understand the sense code, so it's
2201 * probably pointless to loop */
2203 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2204 sd_print_sense_hdr(sdkp, &sshdr);
2209 } while (spintime && time_before_eq(jiffies, spintime_expire));
2212 if (scsi_status_is_good(the_result))
2213 printk(KERN_CONT "ready\n");
2215 printk(KERN_CONT "not responding...\n");
2220 * Determine whether disk supports Data Integrity Field.
2222 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2224 struct scsi_device *sdp = sdkp->device;
2228 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2229 sdkp->protection_type = 0;
2233 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2235 if (type > T10_PI_TYPE3_PROTECTION)
2237 else if (scsi_host_dif_capable(sdp->host, type))
2240 if (sdkp->first_scan || type != sdkp->protection_type)
2243 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2244 " protection type %u. Disabling disk!\n",
2248 sd_printk(KERN_NOTICE, sdkp,
2249 "Enabling DIF Type %u protection\n", type);
2252 sd_printk(KERN_NOTICE, sdkp,
2253 "Disabling DIF Type %u protection\n", type);
2257 sdkp->protection_type = type;
2262 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2263 struct scsi_sense_hdr *sshdr, int sense_valid,
2266 if (driver_byte(the_result) == DRIVER_SENSE)
2267 sd_print_sense_hdr(sdkp, sshdr);
2269 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2272 * Set dirty bit for removable devices if not ready -
2273 * sometimes drives will not report this properly.
2275 if (sdp->removable &&
2276 sense_valid && sshdr->sense_key == NOT_READY)
2277 set_media_not_present(sdkp);
2280 * We used to set media_present to 0 here to indicate no media
2281 * in the drive, but some drives fail read capacity even with
2282 * media present, so we can't do that.
2284 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2288 #if RC16_LEN > SD_BUF_SIZE
2289 #error RC16_LEN must not be more than SD_BUF_SIZE
2292 #define READ_CAPACITY_RETRIES_ON_RESET 10
2294 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2295 unsigned char *buffer)
2297 unsigned char cmd[16];
2298 struct scsi_sense_hdr sshdr;
2299 int sense_valid = 0;
2301 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2302 unsigned int alignment;
2303 unsigned long long lba;
2304 unsigned sector_size;
2306 if (sdp->no_read_capacity_16)
2311 cmd[0] = SERVICE_ACTION_IN_16;
2312 cmd[1] = SAI_READ_CAPACITY_16;
2314 memset(buffer, 0, RC16_LEN);
2316 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2317 buffer, RC16_LEN, &sshdr,
2318 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2320 if (media_not_present(sdkp, &sshdr))
2324 sense_valid = scsi_sense_valid(&sshdr);
2326 sshdr.sense_key == ILLEGAL_REQUEST &&
2327 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2329 /* Invalid Command Operation Code or
2330 * Invalid Field in CDB, just retry
2331 * silently with RC10 */
2334 sshdr.sense_key == UNIT_ATTENTION &&
2335 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2336 /* Device reset might occur several times,
2337 * give it one more chance */
2338 if (--reset_retries > 0)
2343 } while (the_result && retries);
2346 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2347 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2351 sector_size = get_unaligned_be32(&buffer[8]);
2352 lba = get_unaligned_be64(&buffer[0]);
2354 if (sd_read_protection_type(sdkp, buffer) < 0) {
2359 /* Logical blocks per physical block exponent */
2360 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2363 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2365 /* Lowest aligned logical block */
2366 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2367 blk_queue_alignment_offset(sdp->request_queue, alignment);
2368 if (alignment && sdkp->first_scan)
2369 sd_printk(KERN_NOTICE, sdkp,
2370 "physical block alignment offset: %u\n", alignment);
2372 if (buffer[14] & 0x80) { /* LBPME */
2375 if (buffer[14] & 0x40) /* LBPRZ */
2378 sd_config_discard(sdkp, SD_LBP_WS16);
2381 sdkp->capacity = lba + 1;
2385 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2386 unsigned char *buffer)
2388 unsigned char cmd[16];
2389 struct scsi_sense_hdr sshdr;
2390 int sense_valid = 0;
2392 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2394 unsigned sector_size;
2397 cmd[0] = READ_CAPACITY;
2398 memset(&cmd[1], 0, 9);
2399 memset(buffer, 0, 8);
2401 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2403 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2405 if (media_not_present(sdkp, &sshdr))
2409 sense_valid = scsi_sense_valid(&sshdr);
2411 sshdr.sense_key == UNIT_ATTENTION &&
2412 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2413 /* Device reset might occur several times,
2414 * give it one more chance */
2415 if (--reset_retries > 0)
2420 } while (the_result && retries);
2423 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2424 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2428 sector_size = get_unaligned_be32(&buffer[4]);
2429 lba = get_unaligned_be32(&buffer[0]);
2431 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2432 /* Some buggy (usb cardreader) devices return an lba of
2433 0xffffffff when the want to report a size of 0 (with
2434 which they really mean no media is present) */
2436 sdkp->physical_block_size = sector_size;
2440 sdkp->capacity = lba + 1;
2441 sdkp->physical_block_size = sector_size;
2445 static int sd_try_rc16_first(struct scsi_device *sdp)
2447 if (sdp->host->max_cmd_len < 16)
2449 if (sdp->try_rc_10_first)
2451 if (sdp->scsi_level > SCSI_SPC_2)
2453 if (scsi_device_protection(sdp))
2459 * read disk capacity
2462 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2465 struct scsi_device *sdp = sdkp->device;
2467 if (sd_try_rc16_first(sdp)) {
2468 sector_size = read_capacity_16(sdkp, sdp, buffer);
2469 if (sector_size == -EOVERFLOW)
2471 if (sector_size == -ENODEV)
2473 if (sector_size < 0)
2474 sector_size = read_capacity_10(sdkp, sdp, buffer);
2475 if (sector_size < 0)
2478 sector_size = read_capacity_10(sdkp, sdp, buffer);
2479 if (sector_size == -EOVERFLOW)
2481 if (sector_size < 0)
2483 if ((sizeof(sdkp->capacity) > 4) &&
2484 (sdkp->capacity > 0xffffffffULL)) {
2485 int old_sector_size = sector_size;
2486 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2487 "Trying to use READ CAPACITY(16).\n");
2488 sector_size = read_capacity_16(sdkp, sdp, buffer);
2489 if (sector_size < 0) {
2490 sd_printk(KERN_NOTICE, sdkp,
2491 "Using 0xffffffff as device size\n");
2492 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2493 sector_size = old_sector_size;
2496 /* Remember that READ CAPACITY(16) succeeded */
2497 sdp->try_rc_10_first = 0;
2501 /* Some devices are known to return the total number of blocks,
2502 * not the highest block number. Some devices have versions
2503 * which do this and others which do not. Some devices we might
2504 * suspect of doing this but we don't know for certain.
2506 * If we know the reported capacity is wrong, decrement it. If
2507 * we can only guess, then assume the number of blocks is even
2508 * (usually true but not always) and err on the side of lowering
2511 if (sdp->fix_capacity ||
2512 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2513 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2514 "from its reported value: %llu\n",
2515 (unsigned long long) sdkp->capacity);
2520 if (sector_size == 0) {
2522 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2526 if (sector_size != 512 &&
2527 sector_size != 1024 &&
2528 sector_size != 2048 &&
2529 sector_size != 4096) {
2530 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2533 * The user might want to re-format the drive with
2534 * a supported sectorsize. Once this happens, it
2535 * would be relatively trivial to set the thing up.
2536 * For this reason, we leave the thing in the table.
2540 * set a bogus sector size so the normal read/write
2541 * logic in the block layer will eventually refuse any
2542 * request on this device without tripping over power
2543 * of two sector size assumptions
2547 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2548 blk_queue_physical_block_size(sdp->request_queue,
2549 sdkp->physical_block_size);
2550 sdkp->device->sector_size = sector_size;
2552 if (sdkp->capacity > 0xffffffff)
2553 sdp->use_16_for_rw = 1;
2558 * Print disk capacity
2561 sd_print_capacity(struct scsi_disk *sdkp,
2562 sector_t old_capacity)
2564 int sector_size = sdkp->device->sector_size;
2565 char cap_str_2[10], cap_str_10[10];
2567 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2570 string_get_size(sdkp->capacity, sector_size,
2571 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2572 string_get_size(sdkp->capacity, sector_size,
2573 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2575 sd_printk(KERN_NOTICE, sdkp,
2576 "%llu %d-byte logical blocks: (%s/%s)\n",
2577 (unsigned long long)sdkp->capacity,
2578 sector_size, cap_str_10, cap_str_2);
2580 if (sdkp->physical_block_size != sector_size)
2581 sd_printk(KERN_NOTICE, sdkp,
2582 "%u-byte physical blocks\n",
2583 sdkp->physical_block_size);
2586 /* called with buffer of length 512 */
2588 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2589 unsigned char *buffer, int len, struct scsi_mode_data *data,
2590 struct scsi_sense_hdr *sshdr)
2592 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2593 SD_TIMEOUT, SD_MAX_RETRIES, data,
2598 * read write protect setting, if possible - called only in sd_revalidate_disk()
2599 * called with buffer of length SD_BUF_SIZE
2602 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2605 struct scsi_device *sdp = sdkp->device;
2606 struct scsi_mode_data data;
2607 int old_wp = sdkp->write_prot;
2609 set_disk_ro(sdkp->disk, 0);
2610 if (sdp->skip_ms_page_3f) {
2611 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2615 if (sdp->use_192_bytes_for_3f) {
2616 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2619 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2620 * We have to start carefully: some devices hang if we ask
2621 * for more than is available.
2623 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2626 * Second attempt: ask for page 0 When only page 0 is
2627 * implemented, a request for page 3F may return Sense Key
2628 * 5: Illegal Request, Sense Code 24: Invalid field in
2631 if (!scsi_status_is_good(res))
2632 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2635 * Third attempt: ask 255 bytes, as we did earlier.
2637 if (!scsi_status_is_good(res))
2638 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2642 if (!scsi_status_is_good(res)) {
2643 sd_first_printk(KERN_WARNING, sdkp,
2644 "Test WP failed, assume Write Enabled\n");
2646 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2647 set_disk_ro(sdkp->disk, sdkp->write_prot);
2648 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2649 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2650 sdkp->write_prot ? "on" : "off");
2651 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2657 * sd_read_cache_type - called only from sd_revalidate_disk()
2658 * called with buffer of length SD_BUF_SIZE
2661 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2664 struct scsi_device *sdp = sdkp->device;
2669 struct scsi_mode_data data;
2670 struct scsi_sense_hdr sshdr;
2671 int old_wce = sdkp->WCE;
2672 int old_rcd = sdkp->RCD;
2673 int old_dpofua = sdkp->DPOFUA;
2676 if (sdkp->cache_override)
2680 if (sdp->skip_ms_page_8) {
2681 if (sdp->type == TYPE_RBC)
2684 if (sdp->skip_ms_page_3f)
2687 if (sdp->use_192_bytes_for_3f)
2691 } else if (sdp->type == TYPE_RBC) {
2699 /* cautiously ask */
2700 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2703 if (!scsi_status_is_good(res))
2706 if (!data.header_length) {
2709 sd_first_printk(KERN_ERR, sdkp,
2710 "Missing header in MODE_SENSE response\n");
2713 /* that went OK, now ask for the proper length */
2717 * We're only interested in the first three bytes, actually.
2718 * But the data cache page is defined for the first 20.
2722 else if (len > SD_BUF_SIZE) {
2723 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2724 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2727 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2731 if (len > first_len)
2732 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2735 if (scsi_status_is_good(res)) {
2736 int offset = data.header_length + data.block_descriptor_length;
2738 while (offset < len) {
2739 u8 page_code = buffer[offset] & 0x3F;
2740 u8 spf = buffer[offset] & 0x40;
2742 if (page_code == 8 || page_code == 6) {
2743 /* We're interested only in the first 3 bytes.
2745 if (len - offset <= 2) {
2746 sd_first_printk(KERN_ERR, sdkp,
2747 "Incomplete mode parameter "
2751 modepage = page_code;
2755 /* Go to the next page */
2756 if (spf && len - offset > 3)
2757 offset += 4 + (buffer[offset+2] << 8) +
2759 else if (!spf && len - offset > 1)
2760 offset += 2 + buffer[offset+1];
2762 sd_first_printk(KERN_ERR, sdkp,
2764 "parameter data\n");
2770 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2774 if (modepage == 8) {
2775 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2776 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2778 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2782 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2783 if (sdp->broken_fua) {
2784 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2786 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2787 !sdkp->device->use_16_for_rw) {
2788 sd_first_printk(KERN_NOTICE, sdkp,
2789 "Uses READ/WRITE(6), disabling FUA\n");
2793 /* No cache flush allowed for write protected devices */
2794 if (sdkp->WCE && sdkp->write_prot)
2797 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2798 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2799 sd_printk(KERN_NOTICE, sdkp,
2800 "Write cache: %s, read cache: %s, %s\n",
2801 sdkp->WCE ? "enabled" : "disabled",
2802 sdkp->RCD ? "disabled" : "enabled",
2803 sdkp->DPOFUA ? "supports DPO and FUA"
2804 : "doesn't support DPO or FUA");
2810 if (scsi_sense_valid(&sshdr) &&
2811 sshdr.sense_key == ILLEGAL_REQUEST &&
2812 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2813 /* Invalid field in CDB */
2814 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2816 sd_first_printk(KERN_ERR, sdkp,
2817 "Asking for cache data failed\n");
2820 if (sdp->wce_default_on) {
2821 sd_first_printk(KERN_NOTICE, sdkp,
2822 "Assuming drive cache: write back\n");
2825 sd_first_printk(KERN_ERR, sdkp,
2826 "Assuming drive cache: write through\n");
2834 * The ATO bit indicates whether the DIF application tag is available
2835 * for use by the operating system.
2837 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2840 struct scsi_device *sdp = sdkp->device;
2841 struct scsi_mode_data data;
2842 struct scsi_sense_hdr sshdr;
2844 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2847 if (sdkp->protection_type == 0)
2850 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2851 SD_MAX_RETRIES, &data, &sshdr);
2853 if (!scsi_status_is_good(res) || !data.header_length ||
2855 sd_first_printk(KERN_WARNING, sdkp,
2856 "getting Control mode page failed, assume no ATO\n");
2858 if (scsi_sense_valid(&sshdr))
2859 sd_print_sense_hdr(sdkp, &sshdr);
2864 offset = data.header_length + data.block_descriptor_length;
2866 if ((buffer[offset] & 0x3f) != 0x0a) {
2867 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2871 if ((buffer[offset + 5] & 0x80) == 0)
2880 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2881 * @sdkp: disk to query
2883 static void sd_read_block_limits(struct scsi_disk *sdkp)
2885 unsigned int sector_sz = sdkp->device->sector_size;
2886 const int vpd_len = 64;
2887 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2890 /* Block Limits VPD */
2891 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2894 blk_queue_io_min(sdkp->disk->queue,
2895 get_unaligned_be16(&buffer[6]) * sector_sz);
2897 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2898 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2900 if (buffer[3] == 0x3c) {
2901 unsigned int lba_count, desc_count;
2903 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2908 lba_count = get_unaligned_be32(&buffer[20]);
2909 desc_count = get_unaligned_be32(&buffer[24]);
2911 if (lba_count && desc_count)
2912 sdkp->max_unmap_blocks = lba_count;
2914 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2916 if (buffer[32] & 0x80)
2917 sdkp->unmap_alignment =
2918 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2920 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2922 if (sdkp->max_unmap_blocks)
2923 sd_config_discard(sdkp, SD_LBP_UNMAP);
2925 sd_config_discard(sdkp, SD_LBP_WS16);
2927 } else { /* LBP VPD page tells us what to use */
2928 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2929 sd_config_discard(sdkp, SD_LBP_UNMAP);
2930 else if (sdkp->lbpws)
2931 sd_config_discard(sdkp, SD_LBP_WS16);
2932 else if (sdkp->lbpws10)
2933 sd_config_discard(sdkp, SD_LBP_WS10);
2935 sd_config_discard(sdkp, SD_LBP_DISABLE);
2944 * sd_read_block_characteristics - Query block dev. characteristics
2945 * @sdkp: disk to query
2947 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2949 struct request_queue *q = sdkp->disk->queue;
2950 unsigned char *buffer;
2952 const int vpd_len = 64;
2954 buffer = kmalloc(vpd_len, GFP_KERNEL);
2957 /* Block Device Characteristics VPD */
2958 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2961 rot = get_unaligned_be16(&buffer[4]);
2964 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2965 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2968 if (sdkp->device->type == TYPE_ZBC) {
2970 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
2972 sdkp->zoned = (buffer[8] >> 4) & 3;
2973 if (sdkp->zoned == 1) {
2975 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
2977 /* Regular disk or drive managed disk */
2978 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
2982 if (!sdkp->first_scan)
2985 if (blk_queue_is_zoned(q)) {
2986 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2987 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2989 if (sdkp->zoned == 1)
2990 sd_printk(KERN_NOTICE, sdkp,
2991 "Host-aware SMR disk used as regular disk\n");
2992 else if (sdkp->zoned == 2)
2993 sd_printk(KERN_NOTICE, sdkp,
2994 "Drive-managed SMR disk\n");
3002 * sd_read_block_provisioning - Query provisioning VPD page
3003 * @sdkp: disk to query
3005 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3007 unsigned char *buffer;
3008 const int vpd_len = 8;
3010 if (sdkp->lbpme == 0)
3013 buffer = kmalloc(vpd_len, GFP_KERNEL);
3015 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3019 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3020 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3021 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3027 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3029 struct scsi_device *sdev = sdkp->device;
3031 if (sdev->host->no_write_same) {
3032 sdev->no_write_same = 1;
3037 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3038 /* too large values might cause issues with arcmsr */
3039 int vpd_buf_len = 64;
3041 sdev->no_report_opcodes = 1;
3043 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3044 * CODES is unsupported and the device has an ATA
3045 * Information VPD page (SAT).
3047 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3048 sdev->no_write_same = 1;
3051 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3054 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3058 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3060 struct scsi_device *sdev = sdkp->device;
3062 if (!sdev->security_supported)
3065 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3066 SECURITY_PROTOCOL_IN) == 1 &&
3067 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3068 SECURITY_PROTOCOL_OUT) == 1)
3073 * Determine the device's preferred I/O size for reads and writes
3074 * unless the reported value is unreasonably small, large, not a
3075 * multiple of the physical block size, or simply garbage.
3077 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3078 unsigned int dev_max)
3080 struct scsi_device *sdp = sdkp->device;
3081 unsigned int opt_xfer_bytes =
3082 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3084 if (sdkp->opt_xfer_blocks == 0)
3087 if (sdkp->opt_xfer_blocks > dev_max) {
3088 sd_first_printk(KERN_WARNING, sdkp,
3089 "Optimal transfer size %u logical blocks " \
3090 "> dev_max (%u logical blocks)\n",
3091 sdkp->opt_xfer_blocks, dev_max);
3095 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3096 sd_first_printk(KERN_WARNING, sdkp,
3097 "Optimal transfer size %u logical blocks " \
3098 "> sd driver limit (%u logical blocks)\n",
3099 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3103 if (opt_xfer_bytes < PAGE_SIZE) {
3104 sd_first_printk(KERN_WARNING, sdkp,
3105 "Optimal transfer size %u bytes < " \
3106 "PAGE_SIZE (%u bytes)\n",
3107 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3111 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3112 sd_first_printk(KERN_WARNING, sdkp,
3113 "Optimal transfer size %u bytes not a " \
3114 "multiple of physical block size (%u bytes)\n",
3115 opt_xfer_bytes, sdkp->physical_block_size);
3119 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3125 * sd_revalidate_disk - called the first time a new disk is seen,
3126 * performs disk spin up, read_capacity, etc.
3127 * @disk: struct gendisk we care about
3129 static int sd_revalidate_disk(struct gendisk *disk)
3131 struct scsi_disk *sdkp = scsi_disk(disk);
3132 struct scsi_device *sdp = sdkp->device;
3133 struct request_queue *q = sdkp->disk->queue;
3134 sector_t old_capacity = sdkp->capacity;
3135 unsigned char *buffer;
3136 unsigned int dev_max, rw_max;
3138 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3139 "sd_revalidate_disk\n"));
3142 * If the device is offline, don't try and read capacity or any
3143 * of the other niceties.
3145 if (!scsi_device_online(sdp))
3148 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3150 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3151 "allocation failure.\n");
3155 sd_spinup_disk(sdkp);
3158 * Without media there is no reason to ask; moreover, some devices
3159 * react badly if we do.
3161 if (sdkp->media_present) {
3162 sd_read_capacity(sdkp, buffer);
3165 * set the default to rotational. All non-rotational devices
3166 * support the block characteristics VPD page, which will
3167 * cause this to be updated correctly and any device which
3168 * doesn't support it should be treated as rotational.
3170 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3171 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3173 if (scsi_device_supports_vpd(sdp)) {
3174 sd_read_block_provisioning(sdkp);
3175 sd_read_block_limits(sdkp);
3176 sd_read_block_characteristics(sdkp);
3177 sd_zbc_read_zones(sdkp, buffer);
3180 sd_print_capacity(sdkp, old_capacity);
3182 sd_read_write_protect_flag(sdkp, buffer);
3183 sd_read_cache_type(sdkp, buffer);
3184 sd_read_app_tag_own(sdkp, buffer);
3185 sd_read_write_same(sdkp, buffer);
3186 sd_read_security(sdkp, buffer);
3190 * We now have all cache related info, determine how we deal
3191 * with flush requests.
3193 sd_set_flush_flag(sdkp);
3195 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3196 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3198 /* Some devices report a maximum block count for READ/WRITE requests. */
3199 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3200 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3202 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3203 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3204 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3206 q->limits.io_opt = 0;
3207 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3208 (sector_t)BLK_DEF_MAX_SECTORS);
3211 /* Do not exceed controller limit */
3212 rw_max = min(rw_max, queue_max_hw_sectors(q));
3215 * Only update max_sectors if previously unset or if the current value
3216 * exceeds the capabilities of the hardware.
3218 if (sdkp->first_scan ||
3219 q->limits.max_sectors > q->limits.max_dev_sectors ||
3220 q->limits.max_sectors > q->limits.max_hw_sectors)
3221 q->limits.max_sectors = rw_max;
3223 sdkp->first_scan = 0;
3225 set_capacity_revalidate_and_notify(disk,
3226 logical_to_sectors(sdp, sdkp->capacity), false);
3227 sd_config_write_same(sdkp);
3231 * For a zoned drive, revalidating the zones can be done only once
3232 * the gendisk capacity is set. So if this fails, set back the gendisk
3235 if (sd_zbc_revalidate_zones(sdkp))
3236 set_capacity_revalidate_and_notify(disk, 0, false);
3243 * sd_unlock_native_capacity - unlock native capacity
3244 * @disk: struct gendisk to set capacity for
3246 * Block layer calls this function if it detects that partitions
3247 * on @disk reach beyond the end of the device. If the SCSI host
3248 * implements ->unlock_native_capacity() method, it's invoked to
3249 * give it a chance to adjust the device capacity.
3252 * Defined by block layer. Might sleep.
3254 static void sd_unlock_native_capacity(struct gendisk *disk)
3256 struct scsi_device *sdev = scsi_disk(disk)->device;
3258 if (sdev->host->hostt->unlock_native_capacity)
3259 sdev->host->hostt->unlock_native_capacity(sdev);
3263 * sd_format_disk_name - format disk name
3264 * @prefix: name prefix - ie. "sd" for SCSI disks
3265 * @index: index of the disk to format name for
3266 * @buf: output buffer
3267 * @buflen: length of the output buffer
3269 * SCSI disk names starts at sda. The 26th device is sdz and the
3270 * 27th is sdaa. The last one for two lettered suffix is sdzz
3271 * which is followed by sdaaa.
3273 * This is basically 26 base counting with one extra 'nil' entry
3274 * at the beginning from the second digit on and can be
3275 * determined using similar method as 26 base conversion with the
3276 * index shifted -1 after each digit is computed.
3282 * 0 on success, -errno on failure.
3284 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3286 const int base = 'z' - 'a' + 1;
3287 char *begin = buf + strlen(prefix);
3288 char *end = buf + buflen;
3298 *--p = 'a' + (index % unit);
3299 index = (index / unit) - 1;
3300 } while (index >= 0);
3302 memmove(begin, p, end - p);
3303 memcpy(buf, prefix, strlen(prefix));
3309 * sd_probe - called during driver initialization and whenever a
3310 * new scsi device is attached to the system. It is called once
3311 * for each scsi device (not just disks) present.
3312 * @dev: pointer to device object
3314 * Returns 0 if successful (or not interested in this scsi device
3315 * (e.g. scanner)); 1 when there is an error.
3317 * Note: this function is invoked from the scsi mid-level.
3318 * This function sets up the mapping between a given
3319 * <host,channel,id,lun> (found in sdp) and new device name
3320 * (e.g. /dev/sda). More precisely it is the block device major
3321 * and minor number that is chosen here.
3323 * Assume sd_probe is not re-entrant (for time being)
3324 * Also think about sd_probe() and sd_remove() running coincidentally.
3326 static int sd_probe(struct device *dev)
3328 struct scsi_device *sdp = to_scsi_device(dev);
3329 struct scsi_disk *sdkp;
3334 scsi_autopm_get_device(sdp);
3336 if (sdp->type != TYPE_DISK &&
3337 sdp->type != TYPE_ZBC &&
3338 sdp->type != TYPE_MOD &&
3339 sdp->type != TYPE_RBC)
3342 #ifndef CONFIG_BLK_DEV_ZONED
3343 if (sdp->type == TYPE_ZBC)
3346 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3350 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3354 gd = alloc_disk(SD_MINORS);
3358 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3360 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3364 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3366 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3367 goto out_free_index;
3371 sdkp->driver = &sd_template;
3373 sdkp->index = index;
3374 atomic_set(&sdkp->openers, 0);
3375 atomic_set(&sdkp->device->ioerr_cnt, 0);
3377 if (!sdp->request_queue->rq_timeout) {
3378 if (sdp->type != TYPE_MOD)
3379 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3381 blk_queue_rq_timeout(sdp->request_queue,
3385 device_initialize(&sdkp->dev);
3386 sdkp->dev.parent = dev;
3387 sdkp->dev.class = &sd_disk_class;
3388 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3390 error = device_add(&sdkp->dev);
3392 goto out_free_index;
3395 dev_set_drvdata(dev, sdkp);
3397 gd->major = sd_major((index & 0xf0) >> 4);
3398 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3400 gd->fops = &sd_fops;
3401 gd->private_data = &sdkp->driver;
3402 gd->queue = sdkp->device->request_queue;
3404 /* defaults, until the device tells us otherwise */
3405 sdp->sector_size = 512;
3407 sdkp->media_present = 1;
3408 sdkp->write_prot = 0;
3409 sdkp->cache_override = 0;
3413 sdkp->first_scan = 1;
3414 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3416 sd_revalidate_disk(gd);
3418 gd->flags = GENHD_FL_EXT_DEVT;
3419 if (sdp->removable) {
3420 gd->flags |= GENHD_FL_REMOVABLE;
3421 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3422 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3425 blk_pm_runtime_init(sdp->request_queue, dev);
3426 if (sdp->rpm_autosuspend) {
3427 pm_runtime_set_autosuspend_delay(dev,
3428 sdp->host->hostt->rpm_autosuspend_delay);
3430 device_add_disk(dev, gd, NULL);
3432 sd_dif_config_host(sdkp);
3434 sd_revalidate_disk(gd);
3436 if (sdkp->security) {
3437 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3439 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3442 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3443 sdp->removable ? "removable " : "");
3444 scsi_autopm_put_device(sdp);
3449 ida_free(&sd_index_ida, index);
3453 sd_zbc_release_disk(sdkp);
3456 scsi_autopm_put_device(sdp);
3461 * sd_remove - called whenever a scsi disk (previously recognized by
3462 * sd_probe) is detached from the system. It is called (potentially
3463 * multiple times) during sd module unload.
3464 * @dev: pointer to device object
3466 * Note: this function is invoked from the scsi mid-level.
3467 * This function potentially frees up a device name (e.g. /dev/sdc)
3468 * that could be re-used by a subsequent sd_probe().
3469 * This function is not called when the built-in sd driver is "exit-ed".
3471 static int sd_remove(struct device *dev)
3473 struct scsi_disk *sdkp;
3476 sdkp = dev_get_drvdata(dev);
3477 devt = disk_devt(sdkp->disk);
3478 scsi_autopm_get_device(sdkp->device);
3480 async_synchronize_full_domain(&scsi_sd_pm_domain);
3481 device_del(&sdkp->dev);
3482 del_gendisk(sdkp->disk);
3485 free_opal_dev(sdkp->opal_dev);
3487 blk_register_region(devt, SD_MINORS, NULL,
3488 sd_default_probe, NULL, NULL);
3490 mutex_lock(&sd_ref_mutex);
3491 dev_set_drvdata(dev, NULL);
3492 put_device(&sdkp->dev);
3493 mutex_unlock(&sd_ref_mutex);
3499 * scsi_disk_release - Called to free the scsi_disk structure
3500 * @dev: pointer to embedded class device
3502 * sd_ref_mutex must be held entering this routine. Because it is
3503 * called on last put, you should always use the scsi_disk_get()
3504 * scsi_disk_put() helpers which manipulate the semaphore directly
3505 * and never do a direct put_device.
3507 static void scsi_disk_release(struct device *dev)
3509 struct scsi_disk *sdkp = to_scsi_disk(dev);
3510 struct gendisk *disk = sdkp->disk;
3511 struct request_queue *q = disk->queue;
3513 ida_free(&sd_index_ida, sdkp->index);
3516 * Wait until all requests that are in progress have completed.
3517 * This is necessary to avoid that e.g. scsi_end_request() crashes
3518 * due to clearing the disk->private_data pointer. Wait from inside
3519 * scsi_disk_release() instead of from sd_release() to avoid that
3520 * freezing and unfreezing the request queue affects user space I/O
3521 * in case multiple processes open a /dev/sd... node concurrently.
3523 blk_mq_freeze_queue(q);
3524 blk_mq_unfreeze_queue(q);
3526 disk->private_data = NULL;
3528 put_device(&sdkp->device->sdev_gendev);
3530 sd_zbc_release_disk(sdkp);
3535 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3537 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3538 struct scsi_sense_hdr sshdr;
3539 struct scsi_device *sdp = sdkp->device;
3543 cmd[4] |= 1; /* START */
3545 if (sdp->start_stop_pwr_cond)
3546 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3548 if (!scsi_device_online(sdp))
3551 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3552 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3554 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3555 if (driver_byte(res) == DRIVER_SENSE)
3556 sd_print_sense_hdr(sdkp, &sshdr);
3557 if (scsi_sense_valid(&sshdr) &&
3558 /* 0x3a is medium not present */
3563 /* SCSI error codes must not go to the generic layer */
3571 * Send a SYNCHRONIZE CACHE instruction down to the device through
3572 * the normal SCSI command structure. Wait for the command to
3575 static void sd_shutdown(struct device *dev)
3577 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3580 return; /* this can happen */
3582 if (pm_runtime_suspended(dev))
3585 if (sdkp->WCE && sdkp->media_present) {
3586 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3587 sd_sync_cache(sdkp, NULL);
3590 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3591 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3592 sd_start_stop_device(sdkp, 0);
3596 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3598 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3599 struct scsi_sense_hdr sshdr;
3602 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3605 if (sdkp->WCE && sdkp->media_present) {
3606 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3607 ret = sd_sync_cache(sdkp, &sshdr);
3610 /* ignore OFFLINE device */
3614 if (!scsi_sense_valid(&sshdr) ||
3615 sshdr.sense_key != ILLEGAL_REQUEST)
3619 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3620 * doesn't support sync. There's not much to do and
3621 * suspend shouldn't fail.
3627 if (sdkp->device->manage_start_stop) {
3628 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3629 /* an error is not worth aborting a system sleep */
3630 ret = sd_start_stop_device(sdkp, 0);
3631 if (ignore_stop_errors)
3638 static int sd_suspend_system(struct device *dev)
3640 return sd_suspend_common(dev, true);
3643 static int sd_suspend_runtime(struct device *dev)
3645 return sd_suspend_common(dev, false);
3648 static int sd_resume(struct device *dev)
3650 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3653 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3656 if (!sdkp->device->manage_start_stop)
3659 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3660 ret = sd_start_stop_device(sdkp, 1);
3662 opal_unlock_from_suspend(sdkp->opal_dev);
3667 * init_sd - entry point for this driver (both when built in or when
3670 * Note: this function registers this driver with the scsi mid-level.
3672 static int __init init_sd(void)
3674 int majors = 0, i, err;
3676 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3678 for (i = 0; i < SD_MAJORS; i++) {
3679 if (register_blkdev(sd_major(i), "sd") != 0)
3682 blk_register_region(sd_major(i), SD_MINORS, NULL,
3683 sd_default_probe, NULL, NULL);
3689 err = class_register(&sd_disk_class);
3693 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3695 if (!sd_cdb_cache) {
3696 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3701 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3703 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3708 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3709 if (!sd_page_pool) {
3710 printk(KERN_ERR "sd: can't init discard page pool\n");
3715 err = scsi_register_driver(&sd_template.gendrv);
3717 goto err_out_driver;
3722 mempool_destroy(sd_page_pool);
3725 mempool_destroy(sd_cdb_pool);
3728 kmem_cache_destroy(sd_cdb_cache);
3731 class_unregister(&sd_disk_class);
3733 for (i = 0; i < SD_MAJORS; i++)
3734 unregister_blkdev(sd_major(i), "sd");
3739 * exit_sd - exit point for this driver (when it is a module).
3741 * Note: this function unregisters this driver from the scsi mid-level.
3743 static void __exit exit_sd(void)
3747 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3749 scsi_unregister_driver(&sd_template.gendrv);
3750 mempool_destroy(sd_cdb_pool);
3751 mempool_destroy(sd_page_pool);
3752 kmem_cache_destroy(sd_cdb_cache);
3754 class_unregister(&sd_disk_class);
3756 for (i = 0; i < SD_MAJORS; i++) {
3757 blk_unregister_region(sd_major(i), SD_MINORS);
3758 unregister_blkdev(sd_major(i), "sd");
3762 module_init(init_sd);
3763 module_exit(exit_sd);
3765 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3767 scsi_print_sense_hdr(sdkp->device,
3768 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3771 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3773 const char *hb_string = scsi_hostbyte_string(result);
3774 const char *db_string = scsi_driverbyte_string(result);
3776 if (hb_string || db_string)
3777 sd_printk(KERN_INFO, sdkp,
3778 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3779 hb_string ? hb_string : "invalid",
3780 db_string ? db_string : "invalid");
3782 sd_printk(KERN_INFO, sdkp,
3783 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3784 msg, host_byte(result), driver_byte(result));