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 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);
531 static struct attribute *sd_disk_attrs[] = {
532 &dev_attr_cache_type.attr,
534 &dev_attr_allow_restart.attr,
535 &dev_attr_manage_start_stop.attr,
536 &dev_attr_protection_type.attr,
537 &dev_attr_protection_mode.attr,
538 &dev_attr_app_tag_own.attr,
539 &dev_attr_thin_provisioning.attr,
540 &dev_attr_provisioning_mode.attr,
541 &dev_attr_zeroing_mode.attr,
542 &dev_attr_max_write_same_blocks.attr,
543 &dev_attr_max_medium_access_timeouts.attr,
546 ATTRIBUTE_GROUPS(sd_disk);
548 static struct class sd_disk_class = {
550 .owner = THIS_MODULE,
551 .dev_release = scsi_disk_release,
552 .dev_groups = sd_disk_groups,
555 static const struct dev_pm_ops sd_pm_ops = {
556 .suspend = sd_suspend_system,
558 .poweroff = sd_suspend_system,
559 .restore = sd_resume,
560 .runtime_suspend = sd_suspend_runtime,
561 .runtime_resume = sd_resume,
564 static struct scsi_driver sd_template = {
567 .owner = THIS_MODULE,
569 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
571 .shutdown = sd_shutdown,
575 .init_command = sd_init_command,
576 .uninit_command = sd_uninit_command,
578 .eh_action = sd_eh_action,
579 .eh_reset = sd_eh_reset,
583 * Dummy kobj_map->probe function.
584 * The default ->probe function will call modprobe, which is
585 * pointless as this module is already loaded.
587 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
593 * Device no to disk mapping:
595 * major disc2 disc p1
596 * |............|.............|....|....| <- dev_t
599 * Inside a major, we have 16k disks, however mapped non-
600 * contiguously. The first 16 disks are for major0, the next
601 * ones with major1, ... Disk 256 is for major0 again, disk 272
603 * As we stay compatible with our numbering scheme, we can reuse
604 * the well-know SCSI majors 8, 65--71, 136--143.
606 static int sd_major(int major_idx)
610 return SCSI_DISK0_MAJOR;
612 return SCSI_DISK1_MAJOR + major_idx - 1;
614 return SCSI_DISK8_MAJOR + major_idx - 8;
617 return 0; /* shut up gcc */
621 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
623 struct scsi_disk *sdkp = NULL;
625 mutex_lock(&sd_ref_mutex);
627 if (disk->private_data) {
628 sdkp = scsi_disk(disk);
629 if (scsi_device_get(sdkp->device) == 0)
630 get_device(&sdkp->dev);
634 mutex_unlock(&sd_ref_mutex);
638 static void scsi_disk_put(struct scsi_disk *sdkp)
640 struct scsi_device *sdev = sdkp->device;
642 mutex_lock(&sd_ref_mutex);
643 put_device(&sdkp->dev);
644 scsi_device_put(sdev);
645 mutex_unlock(&sd_ref_mutex);
648 #ifdef CONFIG_BLK_SED_OPAL
649 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
650 size_t len, bool send)
652 struct scsi_device *sdev = data;
656 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
658 put_unaligned_be16(spsp, &cdb[2]);
659 put_unaligned_be32(len, &cdb[6]);
661 ret = scsi_execute_req(sdev, cdb,
662 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
663 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
664 return ret <= 0 ? ret : -EIO;
666 #endif /* CONFIG_BLK_SED_OPAL */
669 * Look up the DIX operation based on whether the command is read or
670 * write and whether dix and dif are enabled.
672 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
674 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
675 static const unsigned int ops[] = { /* wrt dix dif */
676 SCSI_PROT_NORMAL, /* 0 0 0 */
677 SCSI_PROT_READ_STRIP, /* 0 0 1 */
678 SCSI_PROT_READ_INSERT, /* 0 1 0 */
679 SCSI_PROT_READ_PASS, /* 0 1 1 */
680 SCSI_PROT_NORMAL, /* 1 0 0 */
681 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
682 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
683 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
686 return ops[write << 2 | dix << 1 | dif];
690 * Returns a mask of the protection flags that are valid for a given DIX
693 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
695 static const unsigned int flag_mask[] = {
696 [SCSI_PROT_NORMAL] = 0,
698 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
699 SCSI_PROT_GUARD_CHECK |
700 SCSI_PROT_REF_CHECK |
701 SCSI_PROT_REF_INCREMENT,
703 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
704 SCSI_PROT_IP_CHECKSUM,
706 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
707 SCSI_PROT_GUARD_CHECK |
708 SCSI_PROT_REF_CHECK |
709 SCSI_PROT_REF_INCREMENT |
710 SCSI_PROT_IP_CHECKSUM,
712 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
713 SCSI_PROT_REF_INCREMENT,
715 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
716 SCSI_PROT_REF_CHECK |
717 SCSI_PROT_REF_INCREMENT |
718 SCSI_PROT_IP_CHECKSUM,
720 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
721 SCSI_PROT_GUARD_CHECK |
722 SCSI_PROT_REF_CHECK |
723 SCSI_PROT_REF_INCREMENT |
724 SCSI_PROT_IP_CHECKSUM,
727 return flag_mask[prot_op];
730 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
731 unsigned int dix, unsigned int dif)
733 struct bio *bio = scmd->request->bio;
734 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
735 unsigned int protect = 0;
737 if (dix) { /* DIX Type 0, 1, 2, 3 */
738 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
739 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
741 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
742 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
745 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
746 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
748 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
749 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
752 if (dif) { /* DIX/DIF Type 1, 2, 3 */
753 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
755 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
756 protect = 3 << 5; /* Disable target PI checking */
758 protect = 1 << 5; /* Enable target PI checking */
761 scsi_set_prot_op(scmd, prot_op);
762 scsi_set_prot_type(scmd, dif);
763 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
768 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
770 struct request_queue *q = sdkp->disk->queue;
771 unsigned int logical_block_size = sdkp->device->sector_size;
772 unsigned int max_blocks = 0;
774 q->limits.discard_alignment =
775 sdkp->unmap_alignment * logical_block_size;
776 q->limits.discard_granularity =
777 max(sdkp->physical_block_size,
778 sdkp->unmap_granularity * logical_block_size);
779 sdkp->provisioning_mode = mode;
785 blk_queue_max_discard_sectors(q, 0);
786 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
790 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
791 (u32)SD_MAX_WS16_BLOCKS);
795 if (sdkp->device->unmap_limit_for_ws)
796 max_blocks = sdkp->max_unmap_blocks;
798 max_blocks = sdkp->max_ws_blocks;
800 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
804 if (sdkp->device->unmap_limit_for_ws)
805 max_blocks = sdkp->max_unmap_blocks;
807 max_blocks = sdkp->max_ws_blocks;
809 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
813 max_blocks = min_not_zero(sdkp->max_ws_blocks,
814 (u32)SD_MAX_WS10_BLOCKS);
818 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
819 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
822 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
824 struct scsi_device *sdp = cmd->device;
825 struct request *rq = cmd->request;
826 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
827 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
828 unsigned int data_len = 24;
831 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
832 if (!rq->special_vec.bv_page)
833 return BLK_STS_RESOURCE;
834 clear_highpage(rq->special_vec.bv_page);
835 rq->special_vec.bv_offset = 0;
836 rq->special_vec.bv_len = data_len;
837 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
840 cmd->cmnd[0] = UNMAP;
843 buf = page_address(rq->special_vec.bv_page);
844 put_unaligned_be16(6 + 16, &buf[0]);
845 put_unaligned_be16(16, &buf[2]);
846 put_unaligned_be64(lba, &buf[8]);
847 put_unaligned_be32(nr_blocks, &buf[16]);
849 cmd->allowed = SD_MAX_RETRIES;
850 cmd->transfersize = data_len;
851 rq->timeout = SD_TIMEOUT;
853 return scsi_init_io(cmd);
856 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
859 struct scsi_device *sdp = cmd->device;
860 struct request *rq = cmd->request;
861 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
862 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
863 u32 data_len = sdp->sector_size;
865 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
866 if (!rq->special_vec.bv_page)
867 return BLK_STS_RESOURCE;
868 clear_highpage(rq->special_vec.bv_page);
869 rq->special_vec.bv_offset = 0;
870 rq->special_vec.bv_len = data_len;
871 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
874 cmd->cmnd[0] = WRITE_SAME_16;
876 cmd->cmnd[1] = 0x8; /* UNMAP */
877 put_unaligned_be64(lba, &cmd->cmnd[2]);
878 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
880 cmd->allowed = SD_MAX_RETRIES;
881 cmd->transfersize = data_len;
882 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
884 return scsi_init_io(cmd);
887 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
890 struct scsi_device *sdp = cmd->device;
891 struct request *rq = cmd->request;
892 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
893 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
894 u32 data_len = sdp->sector_size;
896 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
897 if (!rq->special_vec.bv_page)
898 return BLK_STS_RESOURCE;
899 clear_highpage(rq->special_vec.bv_page);
900 rq->special_vec.bv_offset = 0;
901 rq->special_vec.bv_len = data_len;
902 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905 cmd->cmnd[0] = WRITE_SAME;
907 cmd->cmnd[1] = 0x8; /* UNMAP */
908 put_unaligned_be32(lba, &cmd->cmnd[2]);
909 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
911 cmd->allowed = SD_MAX_RETRIES;
912 cmd->transfersize = data_len;
913 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
915 return scsi_init_io(cmd);
918 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
920 struct request *rq = cmd->request;
921 struct scsi_device *sdp = cmd->device;
922 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
923 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
924 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
926 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
927 switch (sdkp->zeroing_mode) {
928 case SD_ZERO_WS16_UNMAP:
929 return sd_setup_write_same16_cmnd(cmd, true);
930 case SD_ZERO_WS10_UNMAP:
931 return sd_setup_write_same10_cmnd(cmd, true);
935 if (sdp->no_write_same)
936 return BLK_STS_TARGET;
938 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
939 return sd_setup_write_same16_cmnd(cmd, false);
941 return sd_setup_write_same10_cmnd(cmd, false);
944 static void sd_config_write_same(struct scsi_disk *sdkp)
946 struct request_queue *q = sdkp->disk->queue;
947 unsigned int logical_block_size = sdkp->device->sector_size;
949 if (sdkp->device->no_write_same) {
950 sdkp->max_ws_blocks = 0;
954 /* Some devices can not handle block counts above 0xffff despite
955 * supporting WRITE SAME(16). Consequently we default to 64k
956 * blocks per I/O unless the device explicitly advertises a
959 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
960 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
961 (u32)SD_MAX_WS16_BLOCKS);
962 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
963 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
964 (u32)SD_MAX_WS10_BLOCKS);
966 sdkp->device->no_write_same = 1;
967 sdkp->max_ws_blocks = 0;
970 if (sdkp->lbprz && sdkp->lbpws)
971 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
972 else if (sdkp->lbprz && sdkp->lbpws10)
973 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
974 else if (sdkp->max_ws_blocks)
975 sdkp->zeroing_mode = SD_ZERO_WS;
977 sdkp->zeroing_mode = SD_ZERO_WRITE;
979 if (sdkp->max_ws_blocks &&
980 sdkp->physical_block_size > logical_block_size) {
982 * Reporting a maximum number of blocks that is not aligned
983 * on the device physical size would cause a large write same
984 * request to be split into physically unaligned chunks by
985 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
986 * even if the caller of these functions took care to align the
987 * large request. So make sure the maximum reported is aligned
988 * to the device physical block size. This is only an optional
989 * optimization for regular disks, but this is mandatory to
990 * avoid failure of large write same requests directed at
991 * sequential write required zones of host-managed ZBC disks.
993 sdkp->max_ws_blocks =
994 round_down(sdkp->max_ws_blocks,
995 bytes_to_logical(sdkp->device,
996 sdkp->physical_block_size));
1000 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1001 (logical_block_size >> 9));
1002 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1003 (logical_block_size >> 9));
1007 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1008 * @cmd: command to prepare
1010 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1011 * the preference indicated by the target device.
1013 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1015 struct request *rq = cmd->request;
1016 struct scsi_device *sdp = cmd->device;
1017 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1018 struct bio *bio = rq->bio;
1019 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1020 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023 if (sdkp->device->no_write_same)
1024 return BLK_STS_TARGET;
1026 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1028 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1030 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1032 cmd->cmnd[0] = WRITE_SAME_16;
1033 put_unaligned_be64(lba, &cmd->cmnd[2]);
1034 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037 cmd->cmnd[0] = WRITE_SAME;
1038 put_unaligned_be32(lba, &cmd->cmnd[2]);
1039 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042 cmd->transfersize = sdp->sector_size;
1043 cmd->allowed = SD_MAX_RETRIES;
1046 * For WRITE SAME the data transferred via the DATA OUT buffer is
1047 * different from the amount of data actually written to the target.
1049 * We set up __data_len to the amount of data transferred via the
1050 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1051 * to transfer a single sector of data first, but then reset it to
1052 * the amount of data to be written right after so that the I/O path
1053 * knows how much to actually write.
1055 rq->__data_len = sdp->sector_size;
1056 ret = scsi_init_io(cmd);
1057 rq->__data_len = blk_rq_bytes(rq);
1062 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1064 struct request *rq = cmd->request;
1066 /* flush requests don't perform I/O, zero the S/G table */
1067 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1069 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1071 cmd->transfersize = 0;
1072 cmd->allowed = SD_MAX_RETRIES;
1074 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1078 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1079 sector_t lba, unsigned int nr_blocks,
1080 unsigned char flags)
1082 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1083 if (unlikely(cmd->cmnd == NULL))
1084 return BLK_STS_RESOURCE;
1086 cmd->cmd_len = SD_EXT_CDB_SIZE;
1087 memset(cmd->cmnd, 0, cmd->cmd_len);
1089 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1090 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1091 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1092 cmd->cmnd[10] = flags;
1093 put_unaligned_be64(lba, &cmd->cmnd[12]);
1094 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1095 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1100 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1101 sector_t lba, unsigned int nr_blocks,
1102 unsigned char flags)
1105 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1106 cmd->cmnd[1] = flags;
1109 put_unaligned_be64(lba, &cmd->cmnd[2]);
1110 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1115 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1116 sector_t lba, unsigned int nr_blocks,
1117 unsigned char flags)
1120 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1121 cmd->cmnd[1] = flags;
1124 put_unaligned_be32(lba, &cmd->cmnd[2]);
1125 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1130 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1131 sector_t lba, unsigned int nr_blocks,
1132 unsigned char flags)
1134 /* Avoid that 0 blocks gets translated into 256 blocks. */
1135 if (WARN_ON_ONCE(nr_blocks == 0))
1136 return BLK_STS_IOERR;
1138 if (unlikely(flags & 0x8)) {
1140 * This happens only if this drive failed 10byte rw
1141 * command with ILLEGAL_REQUEST during operation and
1142 * thus turned off use_10_for_rw.
1144 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1145 return BLK_STS_IOERR;
1149 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1150 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1151 cmd->cmnd[2] = (lba >> 8) & 0xff;
1152 cmd->cmnd[3] = lba & 0xff;
1153 cmd->cmnd[4] = nr_blocks;
1159 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1161 struct request *rq = cmd->request;
1162 struct scsi_device *sdp = cmd->device;
1163 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1164 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1166 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1167 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1168 bool write = rq_data_dir(rq) == WRITE;
1169 unsigned char protect, fua;
1174 ret = scsi_init_io(cmd);
1175 if (ret != BLK_STS_OK)
1178 if (!scsi_device_online(sdp) || sdp->changed) {
1179 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1180 return BLK_STS_IOERR;
1183 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1184 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1185 return BLK_STS_IOERR;
1188 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1189 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1190 return BLK_STS_IOERR;
1194 * Some SD card readers can't handle accesses which touch the
1195 * last one or two logical blocks. Split accesses as needed.
1197 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1199 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1200 if (lba < threshold) {
1201 /* Access up to the threshold but not beyond */
1202 nr_blocks = threshold - lba;
1204 /* Access only a single logical block */
1209 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1210 dix = scsi_prot_sg_count(cmd);
1211 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1214 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1218 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1219 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1221 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1222 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1224 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1225 sdp->use_10_for_rw || protect) {
1226 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1229 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1233 if (unlikely(ret != BLK_STS_OK))
1237 * We shouldn't disconnect in the middle of a sector, so with a dumb
1238 * host adapter, it's safe to assume that we can at least transfer
1239 * this many bytes between each connect / disconnect.
1241 cmd->transfersize = sdp->sector_size;
1242 cmd->underflow = nr_blocks << 9;
1243 cmd->allowed = SD_MAX_RETRIES;
1244 cmd->sdb.length = nr_blocks * sdp->sector_size;
1247 scmd_printk(KERN_INFO, cmd,
1248 "%s: block=%llu, count=%d\n", __func__,
1249 (unsigned long long)blk_rq_pos(rq),
1250 blk_rq_sectors(rq)));
1252 scmd_printk(KERN_INFO, cmd,
1253 "%s %d/%u 512 byte blocks.\n",
1254 write ? "writing" : "reading", nr_blocks,
1255 blk_rq_sectors(rq)));
1258 * This indicates that the command is ready from our end to be
1264 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1266 struct request *rq = cmd->request;
1268 switch (req_op(rq)) {
1269 case REQ_OP_DISCARD:
1270 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1272 return sd_setup_unmap_cmnd(cmd);
1274 return sd_setup_write_same16_cmnd(cmd, true);
1276 return sd_setup_write_same10_cmnd(cmd, true);
1278 return sd_setup_write_same10_cmnd(cmd, false);
1280 return BLK_STS_TARGET;
1282 case REQ_OP_WRITE_ZEROES:
1283 return sd_setup_write_zeroes_cmnd(cmd);
1284 case REQ_OP_WRITE_SAME:
1285 return sd_setup_write_same_cmnd(cmd);
1287 return sd_setup_flush_cmnd(cmd);
1290 return sd_setup_read_write_cmnd(cmd);
1291 case REQ_OP_ZONE_RESET:
1292 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1294 case REQ_OP_ZONE_RESET_ALL:
1295 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1297 case REQ_OP_ZONE_OPEN:
1298 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1299 case REQ_OP_ZONE_CLOSE:
1300 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1301 case REQ_OP_ZONE_FINISH:
1302 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1305 return BLK_STS_NOTSUPP;
1309 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1311 struct request *rq = SCpnt->request;
1314 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1315 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1317 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1321 mempool_free(cmnd, sd_cdb_pool);
1326 * sd_open - open a scsi disk device
1327 * @bdev: Block device of the scsi disk to open
1328 * @mode: FMODE_* mask
1330 * Returns 0 if successful. Returns a negated errno value in case
1333 * Note: This can be called from a user context (e.g. fsck(1) )
1334 * or from within the kernel (e.g. as a result of a mount(1) ).
1335 * In the latter case @inode and @filp carry an abridged amount
1336 * of information as noted above.
1338 * Locking: called with bdev->bd_mutex held.
1340 static int sd_open(struct block_device *bdev, fmode_t mode)
1342 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1343 struct scsi_device *sdev;
1349 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1351 sdev = sdkp->device;
1354 * If the device is in error recovery, wait until it is done.
1355 * If the device is offline, then disallow any access to it.
1358 if (!scsi_block_when_processing_errors(sdev))
1361 if (sdev->removable || sdkp->write_prot)
1362 check_disk_change(bdev);
1365 * If the drive is empty, just let the open fail.
1367 retval = -ENOMEDIUM;
1368 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1372 * If the device has the write protect tab set, have the open fail
1373 * if the user expects to be able to write to the thing.
1376 if (sdkp->write_prot && (mode & FMODE_WRITE))
1380 * It is possible that the disk changing stuff resulted in
1381 * the device being taken offline. If this is the case,
1382 * report this to the user, and don't pretend that the
1383 * open actually succeeded.
1386 if (!scsi_device_online(sdev))
1389 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1390 if (scsi_block_when_processing_errors(sdev))
1391 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1397 scsi_disk_put(sdkp);
1402 * sd_release - invoked when the (last) close(2) is called on this
1404 * @disk: disk to release
1405 * @mode: FMODE_* mask
1409 * Note: may block (uninterruptible) if error recovery is underway
1412 * Locking: called with bdev->bd_mutex held.
1414 static void sd_release(struct gendisk *disk, fmode_t mode)
1416 struct scsi_disk *sdkp = scsi_disk(disk);
1417 struct scsi_device *sdev = sdkp->device;
1419 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1421 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1422 if (scsi_block_when_processing_errors(sdev))
1423 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1426 scsi_disk_put(sdkp);
1429 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1431 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1432 struct scsi_device *sdp = sdkp->device;
1433 struct Scsi_Host *host = sdp->host;
1434 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1437 /* default to most commonly used values */
1438 diskinfo[0] = 0x40; /* 1 << 6 */
1439 diskinfo[1] = 0x20; /* 1 << 5 */
1440 diskinfo[2] = capacity >> 11;
1442 /* override with calculated, extended default, or driver values */
1443 if (host->hostt->bios_param)
1444 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1446 scsicam_bios_param(bdev, capacity, diskinfo);
1448 geo->heads = diskinfo[0];
1449 geo->sectors = diskinfo[1];
1450 geo->cylinders = diskinfo[2];
1455 * sd_ioctl - process an ioctl
1456 * @bdev: target block device
1457 * @mode: FMODE_* mask
1458 * @cmd: ioctl command number
1459 * @arg: this is third argument given to ioctl(2) system call.
1460 * Often contains a pointer.
1462 * Returns 0 if successful (some ioctls return positive numbers on
1463 * success as well). Returns a negated errno value in case of error.
1465 * Note: most ioctls are forward onto the block subsystem or further
1466 * down in the scsi subsystem.
1468 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1469 unsigned int cmd, unsigned long arg)
1471 struct gendisk *disk = bdev->bd_disk;
1472 struct scsi_disk *sdkp = scsi_disk(disk);
1473 struct scsi_device *sdp = sdkp->device;
1474 void __user *p = (void __user *)arg;
1477 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1478 "cmd=0x%x\n", disk->disk_name, cmd));
1480 error = scsi_verify_blk_ioctl(bdev, cmd);
1485 * If we are in the middle of error recovery, don't let anyone
1486 * else try and use this device. Also, if error recovery fails, it
1487 * may try and take the device offline, in which case all further
1488 * access to the device is prohibited.
1490 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1491 (mode & FMODE_NDELAY) != 0);
1495 if (is_sed_ioctl(cmd))
1496 return sed_ioctl(sdkp->opal_dev, cmd, p);
1499 * Send SCSI addressing ioctls directly to mid level, send other
1500 * ioctls to block level and then onto mid level if they can't be
1504 case SCSI_IOCTL_GET_IDLUN:
1505 case SCSI_IOCTL_GET_BUS_NUMBER:
1506 error = scsi_ioctl(sdp, cmd, p);
1509 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1510 if (error != -ENOTTY)
1512 error = scsi_ioctl(sdp, cmd, p);
1519 static void set_media_not_present(struct scsi_disk *sdkp)
1521 if (sdkp->media_present)
1522 sdkp->device->changed = 1;
1524 if (sdkp->device->removable) {
1525 sdkp->media_present = 0;
1530 static int media_not_present(struct scsi_disk *sdkp,
1531 struct scsi_sense_hdr *sshdr)
1533 if (!scsi_sense_valid(sshdr))
1536 /* not invoked for commands that could return deferred errors */
1537 switch (sshdr->sense_key) {
1538 case UNIT_ATTENTION:
1540 /* medium not present */
1541 if (sshdr->asc == 0x3A) {
1542 set_media_not_present(sdkp);
1550 * sd_check_events - check media events
1551 * @disk: kernel device descriptor
1552 * @clearing: disk events currently being cleared
1554 * Returns mask of DISK_EVENT_*.
1556 * Note: this function is invoked from the block subsystem.
1558 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1560 struct scsi_disk *sdkp = scsi_disk_get(disk);
1561 struct scsi_device *sdp;
1568 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1571 * If the device is offline, don't send any commands - just pretend as
1572 * if the command failed. If the device ever comes back online, we
1573 * can deal with it then. It is only because of unrecoverable errors
1574 * that we would ever take a device offline in the first place.
1576 if (!scsi_device_online(sdp)) {
1577 set_media_not_present(sdkp);
1582 * Using TEST_UNIT_READY enables differentiation between drive with
1583 * no cartridge loaded - NOT READY, drive with changed cartridge -
1584 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1586 * Drives that auto spin down. eg iomega jaz 1G, will be started
1587 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1588 * sd_revalidate() is called.
1590 if (scsi_block_when_processing_errors(sdp)) {
1591 struct scsi_sense_hdr sshdr = { 0, };
1593 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1596 /* failed to execute TUR, assume media not present */
1597 if (host_byte(retval)) {
1598 set_media_not_present(sdkp);
1602 if (media_not_present(sdkp, &sshdr))
1607 * For removable scsi disk we have to recognise the presence
1608 * of a disk in the drive.
1610 if (!sdkp->media_present)
1612 sdkp->media_present = 1;
1615 * sdp->changed is set under the following conditions:
1617 * Medium present state has changed in either direction.
1618 * Device has indicated UNIT_ATTENTION.
1620 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1622 scsi_disk_put(sdkp);
1626 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1629 struct scsi_device *sdp = sdkp->device;
1630 const int timeout = sdp->request_queue->rq_timeout
1631 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1632 struct scsi_sense_hdr my_sshdr;
1634 if (!scsi_device_online(sdp))
1637 /* caller might not be interested in sense, but we need it */
1641 for (retries = 3; retries > 0; --retries) {
1642 unsigned char cmd[10] = { 0 };
1644 cmd[0] = SYNCHRONIZE_CACHE;
1646 * Leave the rest of the command zero to indicate
1649 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1650 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1656 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1658 if (driver_byte(res) == DRIVER_SENSE)
1659 sd_print_sense_hdr(sdkp, sshdr);
1661 /* we need to evaluate the error return */
1662 if (scsi_sense_valid(sshdr) &&
1663 (sshdr->asc == 0x3a || /* medium not present */
1664 sshdr->asc == 0x20 || /* invalid command */
1665 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1666 /* this is no error here */
1669 switch (host_byte(res)) {
1670 /* ignore errors due to racing a disconnection */
1671 case DID_BAD_TARGET:
1672 case DID_NO_CONNECT:
1674 /* signal the upper layer it might try again */
1678 case DID_SOFT_ERROR:
1687 static void sd_rescan(struct device *dev)
1689 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1691 revalidate_disk(sdkp->disk);
1695 #ifdef CONFIG_COMPAT
1697 * This gets directly called from VFS. When the ioctl
1698 * is not recognized we go back to the other translation paths.
1700 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1701 unsigned int cmd, unsigned long arg)
1703 struct gendisk *disk = bdev->bd_disk;
1704 struct scsi_disk *sdkp = scsi_disk(disk);
1705 struct scsi_device *sdev = sdkp->device;
1706 void __user *p = compat_ptr(arg);
1709 error = scsi_verify_blk_ioctl(bdev, cmd);
1713 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1714 (mode & FMODE_NDELAY) != 0);
1718 if (is_sed_ioctl(cmd))
1719 return sed_ioctl(sdkp->opal_dev, cmd, p);
1722 * Let the static ioctl translation table take care of it.
1724 if (!sdev->host->hostt->compat_ioctl)
1725 return -ENOIOCTLCMD;
1726 return sdev->host->hostt->compat_ioctl(sdev, cmd, p);
1730 static char sd_pr_type(enum pr_type type)
1733 case PR_WRITE_EXCLUSIVE:
1735 case PR_EXCLUSIVE_ACCESS:
1737 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1739 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1741 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1743 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1750 static int sd_pr_command(struct block_device *bdev, u8 sa,
1751 u64 key, u64 sa_key, u8 type, u8 flags)
1753 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1754 struct scsi_sense_hdr sshdr;
1756 u8 cmd[16] = { 0, };
1757 u8 data[24] = { 0, };
1759 cmd[0] = PERSISTENT_RESERVE_OUT;
1762 put_unaligned_be32(sizeof(data), &cmd[5]);
1764 put_unaligned_be64(key, &data[0]);
1765 put_unaligned_be64(sa_key, &data[8]);
1768 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1769 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1771 if (driver_byte(result) == DRIVER_SENSE &&
1772 scsi_sense_valid(&sshdr)) {
1773 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1774 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1780 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1783 if (flags & ~PR_FL_IGNORE_KEY)
1785 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1786 old_key, new_key, 0,
1787 (1 << 0) /* APTPL */);
1790 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1795 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1798 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1800 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1803 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1804 enum pr_type type, bool abort)
1806 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1807 sd_pr_type(type), 0);
1810 static int sd_pr_clear(struct block_device *bdev, u64 key)
1812 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1815 static const struct pr_ops sd_pr_ops = {
1816 .pr_register = sd_pr_register,
1817 .pr_reserve = sd_pr_reserve,
1818 .pr_release = sd_pr_release,
1819 .pr_preempt = sd_pr_preempt,
1820 .pr_clear = sd_pr_clear,
1823 static const struct block_device_operations sd_fops = {
1824 .owner = THIS_MODULE,
1826 .release = sd_release,
1828 .getgeo = sd_getgeo,
1829 #ifdef CONFIG_COMPAT
1830 .compat_ioctl = sd_compat_ioctl,
1832 .check_events = sd_check_events,
1833 .revalidate_disk = sd_revalidate_disk,
1834 .unlock_native_capacity = sd_unlock_native_capacity,
1835 .report_zones = sd_zbc_report_zones,
1836 .pr_ops = &sd_pr_ops,
1840 * sd_eh_reset - reset error handling callback
1841 * @scmd: sd-issued command that has failed
1843 * This function is called by the SCSI midlayer before starting
1844 * SCSI EH. When counting medium access failures we have to be
1845 * careful to register it only only once per device and SCSI EH run;
1846 * there might be several timed out commands which will cause the
1847 * 'max_medium_access_timeouts' counter to trigger after the first
1848 * SCSI EH run already and set the device to offline.
1849 * So this function resets the internal counter before starting SCSI EH.
1851 static void sd_eh_reset(struct scsi_cmnd *scmd)
1853 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1855 /* New SCSI EH run, reset gate variable */
1856 sdkp->ignore_medium_access_errors = false;
1860 * sd_eh_action - error handling callback
1861 * @scmd: sd-issued command that has failed
1862 * @eh_disp: The recovery disposition suggested by the midlayer
1864 * This function is called by the SCSI midlayer upon completion of an
1865 * error test command (currently TEST UNIT READY). The result of sending
1866 * the eh command is passed in eh_disp. We're looking for devices that
1867 * fail medium access commands but are OK with non access commands like
1868 * test unit ready (so wrongly see the device as having a successful
1871 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1873 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1874 struct scsi_device *sdev = scmd->device;
1876 if (!scsi_device_online(sdev) ||
1877 !scsi_medium_access_command(scmd) ||
1878 host_byte(scmd->result) != DID_TIME_OUT ||
1883 * The device has timed out executing a medium access command.
1884 * However, the TEST UNIT READY command sent during error
1885 * handling completed successfully. Either the device is in the
1886 * process of recovering or has it suffered an internal failure
1887 * that prevents access to the storage medium.
1889 if (!sdkp->ignore_medium_access_errors) {
1890 sdkp->medium_access_timed_out++;
1891 sdkp->ignore_medium_access_errors = true;
1895 * If the device keeps failing read/write commands but TEST UNIT
1896 * READY always completes successfully we assume that medium
1897 * access is no longer possible and take the device offline.
1899 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1900 scmd_printk(KERN_ERR, scmd,
1901 "Medium access timeout failure. Offlining disk!\n");
1902 mutex_lock(&sdev->state_mutex);
1903 scsi_device_set_state(sdev, SDEV_OFFLINE);
1904 mutex_unlock(&sdev->state_mutex);
1912 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1914 struct request *req = scmd->request;
1915 struct scsi_device *sdev = scmd->device;
1916 unsigned int transferred, good_bytes;
1917 u64 start_lba, end_lba, bad_lba;
1920 * Some commands have a payload smaller than the device logical
1921 * block size (e.g. INQUIRY on a 4K disk).
1923 if (scsi_bufflen(scmd) <= sdev->sector_size)
1926 /* Check if we have a 'bad_lba' information */
1927 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1928 SCSI_SENSE_BUFFERSIZE,
1933 * If the bad lba was reported incorrectly, we have no idea where
1936 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1937 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1938 if (bad_lba < start_lba || bad_lba >= end_lba)
1942 * resid is optional but mostly filled in. When it's unused,
1943 * its value is zero, so we assume the whole buffer transferred
1945 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1947 /* This computation should always be done in terms of the
1948 * resolution of the device's medium.
1950 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1952 return min(good_bytes, transferred);
1956 * sd_done - bottom half handler: called when the lower level
1957 * driver has completed (successfully or otherwise) a scsi command.
1958 * @SCpnt: mid-level's per command structure.
1960 * Note: potentially run from within an ISR. Must not block.
1962 static int sd_done(struct scsi_cmnd *SCpnt)
1964 int result = SCpnt->result;
1965 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1966 unsigned int sector_size = SCpnt->device->sector_size;
1968 struct scsi_sense_hdr sshdr;
1969 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1970 struct request *req = SCpnt->request;
1971 int sense_valid = 0;
1972 int sense_deferred = 0;
1974 switch (req_op(req)) {
1975 case REQ_OP_DISCARD:
1976 case REQ_OP_WRITE_ZEROES:
1977 case REQ_OP_WRITE_SAME:
1978 case REQ_OP_ZONE_RESET:
1979 case REQ_OP_ZONE_RESET_ALL:
1980 case REQ_OP_ZONE_OPEN:
1981 case REQ_OP_ZONE_CLOSE:
1982 case REQ_OP_ZONE_FINISH:
1984 good_bytes = blk_rq_bytes(req);
1985 scsi_set_resid(SCpnt, 0);
1988 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1993 * In case of bogus fw or device, we could end up having
1994 * an unaligned partial completion. Check this here and force
1997 resid = scsi_get_resid(SCpnt);
1998 if (resid & (sector_size - 1)) {
1999 sd_printk(KERN_INFO, sdkp,
2000 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2001 resid, sector_size);
2002 scsi_print_command(SCpnt);
2003 resid = min(scsi_bufflen(SCpnt),
2004 round_up(resid, sector_size));
2005 scsi_set_resid(SCpnt, resid);
2010 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2012 sense_deferred = scsi_sense_is_deferred(&sshdr);
2014 sdkp->medium_access_timed_out = 0;
2016 if (driver_byte(result) != DRIVER_SENSE &&
2017 (!sense_valid || sense_deferred))
2020 switch (sshdr.sense_key) {
2021 case HARDWARE_ERROR:
2023 good_bytes = sd_completed_bytes(SCpnt);
2025 case RECOVERED_ERROR:
2026 good_bytes = scsi_bufflen(SCpnt);
2029 /* This indicates a false check condition, so ignore it. An
2030 * unknown amount of data was transferred so treat it as an
2034 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2036 case ABORTED_COMMAND:
2037 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2038 good_bytes = sd_completed_bytes(SCpnt);
2040 case ILLEGAL_REQUEST:
2041 switch (sshdr.asc) {
2042 case 0x10: /* DIX: Host detected corruption */
2043 good_bytes = sd_completed_bytes(SCpnt);
2045 case 0x20: /* INVALID COMMAND OPCODE */
2046 case 0x24: /* INVALID FIELD IN CDB */
2047 switch (SCpnt->cmnd[0]) {
2049 sd_config_discard(sdkp, SD_LBP_DISABLE);
2053 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2054 sd_config_discard(sdkp, SD_LBP_DISABLE);
2056 sdkp->device->no_write_same = 1;
2057 sd_config_write_same(sdkp);
2058 req->rq_flags |= RQF_QUIET;
2069 if (sd_is_zoned(sdkp))
2070 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2072 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2073 "sd_done: completed %d of %d bytes\n",
2074 good_bytes, scsi_bufflen(SCpnt)));
2080 * spinup disk - called only in sd_revalidate_disk()
2083 sd_spinup_disk(struct scsi_disk *sdkp)
2085 unsigned char cmd[10];
2086 unsigned long spintime_expire = 0;
2087 int retries, spintime;
2088 unsigned int the_result;
2089 struct scsi_sense_hdr sshdr;
2090 int sense_valid = 0;
2094 /* Spin up drives, as required. Only do this at boot time */
2095 /* Spinup needs to be done for module loads too. */
2100 cmd[0] = TEST_UNIT_READY;
2101 memset((void *) &cmd[1], 0, 9);
2103 the_result = scsi_execute_req(sdkp->device, cmd,
2106 SD_MAX_RETRIES, NULL);
2109 * If the drive has indicated to us that it
2110 * doesn't have any media in it, don't bother
2111 * with any more polling.
2113 if (media_not_present(sdkp, &sshdr))
2117 sense_valid = scsi_sense_valid(&sshdr);
2119 } while (retries < 3 &&
2120 (!scsi_status_is_good(the_result) ||
2121 ((driver_byte(the_result) == DRIVER_SENSE) &&
2122 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2124 if (driver_byte(the_result) != DRIVER_SENSE) {
2125 /* no sense, TUR either succeeded or failed
2126 * with a status error */
2127 if(!spintime && !scsi_status_is_good(the_result)) {
2128 sd_print_result(sdkp, "Test Unit Ready failed",
2135 * The device does not want the automatic start to be issued.
2137 if (sdkp->device->no_start_on_add)
2140 if (sense_valid && sshdr.sense_key == NOT_READY) {
2141 if (sshdr.asc == 4 && sshdr.ascq == 3)
2142 break; /* manual intervention required */
2143 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2144 break; /* standby */
2145 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2146 break; /* unavailable */
2147 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2148 break; /* sanitize in progress */
2150 * Issue command to spin up drive when not ready
2153 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2154 cmd[0] = START_STOP;
2155 cmd[1] = 1; /* Return immediately */
2156 memset((void *) &cmd[2], 0, 8);
2157 cmd[4] = 1; /* Start spin cycle */
2158 if (sdkp->device->start_stop_pwr_cond)
2160 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2162 SD_TIMEOUT, SD_MAX_RETRIES,
2164 spintime_expire = jiffies + 100 * HZ;
2167 /* Wait 1 second for next try */
2169 printk(KERN_CONT ".");
2172 * Wait for USB flash devices with slow firmware.
2173 * Yes, this sense key/ASC combination shouldn't
2174 * occur here. It's characteristic of these devices.
2176 } else if (sense_valid &&
2177 sshdr.sense_key == UNIT_ATTENTION &&
2178 sshdr.asc == 0x28) {
2180 spintime_expire = jiffies + 5 * HZ;
2183 /* Wait 1 second for next try */
2186 /* we don't understand the sense code, so it's
2187 * probably pointless to loop */
2189 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2190 sd_print_sense_hdr(sdkp, &sshdr);
2195 } while (spintime && time_before_eq(jiffies, spintime_expire));
2198 if (scsi_status_is_good(the_result))
2199 printk(KERN_CONT "ready\n");
2201 printk(KERN_CONT "not responding...\n");
2206 * Determine whether disk supports Data Integrity Field.
2208 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2210 struct scsi_device *sdp = sdkp->device;
2214 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2217 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2219 if (type > T10_PI_TYPE3_PROTECTION)
2221 else if (scsi_host_dif_capable(sdp->host, type))
2224 if (sdkp->first_scan || type != sdkp->protection_type)
2227 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2228 " protection type %u. Disabling disk!\n",
2232 sd_printk(KERN_NOTICE, sdkp,
2233 "Enabling DIF Type %u protection\n", type);
2236 sd_printk(KERN_NOTICE, sdkp,
2237 "Disabling DIF Type %u protection\n", type);
2241 sdkp->protection_type = type;
2246 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2247 struct scsi_sense_hdr *sshdr, int sense_valid,
2250 if (driver_byte(the_result) == DRIVER_SENSE)
2251 sd_print_sense_hdr(sdkp, sshdr);
2253 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2256 * Set dirty bit for removable devices if not ready -
2257 * sometimes drives will not report this properly.
2259 if (sdp->removable &&
2260 sense_valid && sshdr->sense_key == NOT_READY)
2261 set_media_not_present(sdkp);
2264 * We used to set media_present to 0 here to indicate no media
2265 * in the drive, but some drives fail read capacity even with
2266 * media present, so we can't do that.
2268 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2272 #if RC16_LEN > SD_BUF_SIZE
2273 #error RC16_LEN must not be more than SD_BUF_SIZE
2276 #define READ_CAPACITY_RETRIES_ON_RESET 10
2278 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2279 unsigned char *buffer)
2281 unsigned char cmd[16];
2282 struct scsi_sense_hdr sshdr;
2283 int sense_valid = 0;
2285 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2286 unsigned int alignment;
2287 unsigned long long lba;
2288 unsigned sector_size;
2290 if (sdp->no_read_capacity_16)
2295 cmd[0] = SERVICE_ACTION_IN_16;
2296 cmd[1] = SAI_READ_CAPACITY_16;
2298 memset(buffer, 0, RC16_LEN);
2300 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2301 buffer, RC16_LEN, &sshdr,
2302 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2304 if (media_not_present(sdkp, &sshdr))
2308 sense_valid = scsi_sense_valid(&sshdr);
2310 sshdr.sense_key == ILLEGAL_REQUEST &&
2311 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2313 /* Invalid Command Operation Code or
2314 * Invalid Field in CDB, just retry
2315 * silently with RC10 */
2318 sshdr.sense_key == UNIT_ATTENTION &&
2319 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2320 /* Device reset might occur several times,
2321 * give it one more chance */
2322 if (--reset_retries > 0)
2327 } while (the_result && retries);
2330 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2331 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2335 sector_size = get_unaligned_be32(&buffer[8]);
2336 lba = get_unaligned_be64(&buffer[0]);
2338 if (sd_read_protection_type(sdkp, buffer) < 0) {
2343 /* Logical blocks per physical block exponent */
2344 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2347 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2349 /* Lowest aligned logical block */
2350 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2351 blk_queue_alignment_offset(sdp->request_queue, alignment);
2352 if (alignment && sdkp->first_scan)
2353 sd_printk(KERN_NOTICE, sdkp,
2354 "physical block alignment offset: %u\n", alignment);
2356 if (buffer[14] & 0x80) { /* LBPME */
2359 if (buffer[14] & 0x40) /* LBPRZ */
2362 sd_config_discard(sdkp, SD_LBP_WS16);
2365 sdkp->capacity = lba + 1;
2369 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2370 unsigned char *buffer)
2372 unsigned char cmd[16];
2373 struct scsi_sense_hdr sshdr;
2374 int sense_valid = 0;
2376 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2378 unsigned sector_size;
2381 cmd[0] = READ_CAPACITY;
2382 memset(&cmd[1], 0, 9);
2383 memset(buffer, 0, 8);
2385 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2387 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2389 if (media_not_present(sdkp, &sshdr))
2393 sense_valid = scsi_sense_valid(&sshdr);
2395 sshdr.sense_key == UNIT_ATTENTION &&
2396 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2397 /* Device reset might occur several times,
2398 * give it one more chance */
2399 if (--reset_retries > 0)
2404 } while (the_result && retries);
2407 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2408 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2412 sector_size = get_unaligned_be32(&buffer[4]);
2413 lba = get_unaligned_be32(&buffer[0]);
2415 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2416 /* Some buggy (usb cardreader) devices return an lba of
2417 0xffffffff when the want to report a size of 0 (with
2418 which they really mean no media is present) */
2420 sdkp->physical_block_size = sector_size;
2424 sdkp->capacity = lba + 1;
2425 sdkp->physical_block_size = sector_size;
2429 static int sd_try_rc16_first(struct scsi_device *sdp)
2431 if (sdp->host->max_cmd_len < 16)
2433 if (sdp->try_rc_10_first)
2435 if (sdp->scsi_level > SCSI_SPC_2)
2437 if (scsi_device_protection(sdp))
2443 * read disk capacity
2446 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2449 struct scsi_device *sdp = sdkp->device;
2451 if (sd_try_rc16_first(sdp)) {
2452 sector_size = read_capacity_16(sdkp, sdp, buffer);
2453 if (sector_size == -EOVERFLOW)
2455 if (sector_size == -ENODEV)
2457 if (sector_size < 0)
2458 sector_size = read_capacity_10(sdkp, sdp, buffer);
2459 if (sector_size < 0)
2462 sector_size = read_capacity_10(sdkp, sdp, buffer);
2463 if (sector_size == -EOVERFLOW)
2465 if (sector_size < 0)
2467 if ((sizeof(sdkp->capacity) > 4) &&
2468 (sdkp->capacity > 0xffffffffULL)) {
2469 int old_sector_size = sector_size;
2470 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2471 "Trying to use READ CAPACITY(16).\n");
2472 sector_size = read_capacity_16(sdkp, sdp, buffer);
2473 if (sector_size < 0) {
2474 sd_printk(KERN_NOTICE, sdkp,
2475 "Using 0xffffffff as device size\n");
2476 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2477 sector_size = old_sector_size;
2480 /* Remember that READ CAPACITY(16) succeeded */
2481 sdp->try_rc_10_first = 0;
2485 /* Some devices are known to return the total number of blocks,
2486 * not the highest block number. Some devices have versions
2487 * which do this and others which do not. Some devices we might
2488 * suspect of doing this but we don't know for certain.
2490 * If we know the reported capacity is wrong, decrement it. If
2491 * we can only guess, then assume the number of blocks is even
2492 * (usually true but not always) and err on the side of lowering
2495 if (sdp->fix_capacity ||
2496 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2497 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2498 "from its reported value: %llu\n",
2499 (unsigned long long) sdkp->capacity);
2504 if (sector_size == 0) {
2506 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2510 if (sector_size != 512 &&
2511 sector_size != 1024 &&
2512 sector_size != 2048 &&
2513 sector_size != 4096) {
2514 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2517 * The user might want to re-format the drive with
2518 * a supported sectorsize. Once this happens, it
2519 * would be relatively trivial to set the thing up.
2520 * For this reason, we leave the thing in the table.
2524 * set a bogus sector size so the normal read/write
2525 * logic in the block layer will eventually refuse any
2526 * request on this device without tripping over power
2527 * of two sector size assumptions
2531 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2532 blk_queue_physical_block_size(sdp->request_queue,
2533 sdkp->physical_block_size);
2534 sdkp->device->sector_size = sector_size;
2536 if (sdkp->capacity > 0xffffffff)
2537 sdp->use_16_for_rw = 1;
2542 * Print disk capacity
2545 sd_print_capacity(struct scsi_disk *sdkp,
2546 sector_t old_capacity)
2548 int sector_size = sdkp->device->sector_size;
2549 char cap_str_2[10], cap_str_10[10];
2551 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2554 string_get_size(sdkp->capacity, sector_size,
2555 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2556 string_get_size(sdkp->capacity, sector_size,
2557 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2559 sd_printk(KERN_NOTICE, sdkp,
2560 "%llu %d-byte logical blocks: (%s/%s)\n",
2561 (unsigned long long)sdkp->capacity,
2562 sector_size, cap_str_10, cap_str_2);
2564 if (sdkp->physical_block_size != sector_size)
2565 sd_printk(KERN_NOTICE, sdkp,
2566 "%u-byte physical blocks\n",
2567 sdkp->physical_block_size);
2569 sd_zbc_print_zones(sdkp);
2572 /* called with buffer of length 512 */
2574 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2575 unsigned char *buffer, int len, struct scsi_mode_data *data,
2576 struct scsi_sense_hdr *sshdr)
2578 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2579 SD_TIMEOUT, SD_MAX_RETRIES, data,
2584 * read write protect setting, if possible - called only in sd_revalidate_disk()
2585 * called with buffer of length SD_BUF_SIZE
2588 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2591 struct scsi_device *sdp = sdkp->device;
2592 struct scsi_mode_data data;
2593 int old_wp = sdkp->write_prot;
2595 set_disk_ro(sdkp->disk, 0);
2596 if (sdp->skip_ms_page_3f) {
2597 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2601 if (sdp->use_192_bytes_for_3f) {
2602 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2605 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2606 * We have to start carefully: some devices hang if we ask
2607 * for more than is available.
2609 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2612 * Second attempt: ask for page 0 When only page 0 is
2613 * implemented, a request for page 3F may return Sense Key
2614 * 5: Illegal Request, Sense Code 24: Invalid field in
2617 if (!scsi_status_is_good(res))
2618 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2621 * Third attempt: ask 255 bytes, as we did earlier.
2623 if (!scsi_status_is_good(res))
2624 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2628 if (!scsi_status_is_good(res)) {
2629 sd_first_printk(KERN_WARNING, sdkp,
2630 "Test WP failed, assume Write Enabled\n");
2632 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2633 set_disk_ro(sdkp->disk, sdkp->write_prot);
2634 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2635 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2636 sdkp->write_prot ? "on" : "off");
2637 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2643 * sd_read_cache_type - called only from sd_revalidate_disk()
2644 * called with buffer of length SD_BUF_SIZE
2647 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2650 struct scsi_device *sdp = sdkp->device;
2655 struct scsi_mode_data data;
2656 struct scsi_sense_hdr sshdr;
2657 int old_wce = sdkp->WCE;
2658 int old_rcd = sdkp->RCD;
2659 int old_dpofua = sdkp->DPOFUA;
2662 if (sdkp->cache_override)
2666 if (sdp->skip_ms_page_8) {
2667 if (sdp->type == TYPE_RBC)
2670 if (sdp->skip_ms_page_3f)
2673 if (sdp->use_192_bytes_for_3f)
2677 } else if (sdp->type == TYPE_RBC) {
2685 /* cautiously ask */
2686 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2689 if (!scsi_status_is_good(res))
2692 if (!data.header_length) {
2695 sd_first_printk(KERN_ERR, sdkp,
2696 "Missing header in MODE_SENSE response\n");
2699 /* that went OK, now ask for the proper length */
2703 * We're only interested in the first three bytes, actually.
2704 * But the data cache page is defined for the first 20.
2708 else if (len > SD_BUF_SIZE) {
2709 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2710 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2713 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2717 if (len > first_len)
2718 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2721 if (scsi_status_is_good(res)) {
2722 int offset = data.header_length + data.block_descriptor_length;
2724 while (offset < len) {
2725 u8 page_code = buffer[offset] & 0x3F;
2726 u8 spf = buffer[offset] & 0x40;
2728 if (page_code == 8 || page_code == 6) {
2729 /* We're interested only in the first 3 bytes.
2731 if (len - offset <= 2) {
2732 sd_first_printk(KERN_ERR, sdkp,
2733 "Incomplete mode parameter "
2737 modepage = page_code;
2741 /* Go to the next page */
2742 if (spf && len - offset > 3)
2743 offset += 4 + (buffer[offset+2] << 8) +
2745 else if (!spf && len - offset > 1)
2746 offset += 2 + buffer[offset+1];
2748 sd_first_printk(KERN_ERR, sdkp,
2750 "parameter data\n");
2756 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2760 if (modepage == 8) {
2761 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2762 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2764 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2768 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2769 if (sdp->broken_fua) {
2770 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2772 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2773 !sdkp->device->use_16_for_rw) {
2774 sd_first_printk(KERN_NOTICE, sdkp,
2775 "Uses READ/WRITE(6), disabling FUA\n");
2779 /* No cache flush allowed for write protected devices */
2780 if (sdkp->WCE && sdkp->write_prot)
2783 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2784 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2785 sd_printk(KERN_NOTICE, sdkp,
2786 "Write cache: %s, read cache: %s, %s\n",
2787 sdkp->WCE ? "enabled" : "disabled",
2788 sdkp->RCD ? "disabled" : "enabled",
2789 sdkp->DPOFUA ? "supports DPO and FUA"
2790 : "doesn't support DPO or FUA");
2796 if (scsi_sense_valid(&sshdr) &&
2797 sshdr.sense_key == ILLEGAL_REQUEST &&
2798 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2799 /* Invalid field in CDB */
2800 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2802 sd_first_printk(KERN_ERR, sdkp,
2803 "Asking for cache data failed\n");
2806 if (sdp->wce_default_on) {
2807 sd_first_printk(KERN_NOTICE, sdkp,
2808 "Assuming drive cache: write back\n");
2811 sd_first_printk(KERN_ERR, sdkp,
2812 "Assuming drive cache: write through\n");
2820 * The ATO bit indicates whether the DIF application tag is available
2821 * for use by the operating system.
2823 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2826 struct scsi_device *sdp = sdkp->device;
2827 struct scsi_mode_data data;
2828 struct scsi_sense_hdr sshdr;
2830 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2833 if (sdkp->protection_type == 0)
2836 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2837 SD_MAX_RETRIES, &data, &sshdr);
2839 if (!scsi_status_is_good(res) || !data.header_length ||
2841 sd_first_printk(KERN_WARNING, sdkp,
2842 "getting Control mode page failed, assume no ATO\n");
2844 if (scsi_sense_valid(&sshdr))
2845 sd_print_sense_hdr(sdkp, &sshdr);
2850 offset = data.header_length + data.block_descriptor_length;
2852 if ((buffer[offset] & 0x3f) != 0x0a) {
2853 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2857 if ((buffer[offset + 5] & 0x80) == 0)
2866 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2867 * @sdkp: disk to query
2869 static void sd_read_block_limits(struct scsi_disk *sdkp)
2871 unsigned int sector_sz = sdkp->device->sector_size;
2872 const int vpd_len = 64;
2873 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2876 /* Block Limits VPD */
2877 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2880 blk_queue_io_min(sdkp->disk->queue,
2881 get_unaligned_be16(&buffer[6]) * sector_sz);
2883 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2884 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2886 if (buffer[3] == 0x3c) {
2887 unsigned int lba_count, desc_count;
2889 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2894 lba_count = get_unaligned_be32(&buffer[20]);
2895 desc_count = get_unaligned_be32(&buffer[24]);
2897 if (lba_count && desc_count)
2898 sdkp->max_unmap_blocks = lba_count;
2900 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2902 if (buffer[32] & 0x80)
2903 sdkp->unmap_alignment =
2904 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2906 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2908 if (sdkp->max_unmap_blocks)
2909 sd_config_discard(sdkp, SD_LBP_UNMAP);
2911 sd_config_discard(sdkp, SD_LBP_WS16);
2913 } else { /* LBP VPD page tells us what to use */
2914 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2915 sd_config_discard(sdkp, SD_LBP_UNMAP);
2916 else if (sdkp->lbpws)
2917 sd_config_discard(sdkp, SD_LBP_WS16);
2918 else if (sdkp->lbpws10)
2919 sd_config_discard(sdkp, SD_LBP_WS10);
2921 sd_config_discard(sdkp, SD_LBP_DISABLE);
2930 * sd_read_block_characteristics - Query block dev. characteristics
2931 * @sdkp: disk to query
2933 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2935 struct request_queue *q = sdkp->disk->queue;
2936 unsigned char *buffer;
2938 const int vpd_len = 64;
2940 buffer = kmalloc(vpd_len, GFP_KERNEL);
2943 /* Block Device Characteristics VPD */
2944 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2947 rot = get_unaligned_be16(&buffer[4]);
2950 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2951 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2954 if (sdkp->device->type == TYPE_ZBC) {
2956 q->limits.zoned = BLK_ZONED_HM;
2958 sdkp->zoned = (buffer[8] >> 4) & 3;
2959 if (sdkp->zoned == 1 && !disk_has_partitions(sdkp->disk)) {
2961 q->limits.zoned = BLK_ZONED_HA;
2964 * Treat drive-managed devices and host-aware devices
2965 * with partitions as regular block devices.
2967 q->limits.zoned = BLK_ZONED_NONE;
2970 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2971 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2972 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2979 * sd_read_block_provisioning - Query provisioning VPD page
2980 * @sdkp: disk to query
2982 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2984 unsigned char *buffer;
2985 const int vpd_len = 8;
2987 if (sdkp->lbpme == 0)
2990 buffer = kmalloc(vpd_len, GFP_KERNEL);
2992 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2996 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2997 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2998 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3004 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3006 struct scsi_device *sdev = sdkp->device;
3008 if (sdev->host->no_write_same) {
3009 sdev->no_write_same = 1;
3014 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3015 /* too large values might cause issues with arcmsr */
3016 int vpd_buf_len = 64;
3018 sdev->no_report_opcodes = 1;
3020 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3021 * CODES is unsupported and the device has an ATA
3022 * Information VPD page (SAT).
3024 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3025 sdev->no_write_same = 1;
3028 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3031 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3035 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3037 struct scsi_device *sdev = sdkp->device;
3039 if (!sdev->security_supported)
3042 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3043 SECURITY_PROTOCOL_IN) == 1 &&
3044 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3045 SECURITY_PROTOCOL_OUT) == 1)
3050 * Determine the device's preferred I/O size for reads and writes
3051 * unless the reported value is unreasonably small, large, not a
3052 * multiple of the physical block size, or simply garbage.
3054 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3055 unsigned int dev_max)
3057 struct scsi_device *sdp = sdkp->device;
3058 unsigned int opt_xfer_bytes =
3059 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3061 if (sdkp->opt_xfer_blocks == 0)
3064 if (sdkp->opt_xfer_blocks > dev_max) {
3065 sd_first_printk(KERN_WARNING, sdkp,
3066 "Optimal transfer size %u logical blocks " \
3067 "> dev_max (%u logical blocks)\n",
3068 sdkp->opt_xfer_blocks, dev_max);
3072 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3073 sd_first_printk(KERN_WARNING, sdkp,
3074 "Optimal transfer size %u logical blocks " \
3075 "> sd driver limit (%u logical blocks)\n",
3076 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3080 if (opt_xfer_bytes < PAGE_SIZE) {
3081 sd_first_printk(KERN_WARNING, sdkp,
3082 "Optimal transfer size %u bytes < " \
3083 "PAGE_SIZE (%u bytes)\n",
3084 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3088 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3089 sd_first_printk(KERN_WARNING, sdkp,
3090 "Optimal transfer size %u bytes not a " \
3091 "multiple of physical block size (%u bytes)\n",
3092 opt_xfer_bytes, sdkp->physical_block_size);
3096 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3102 * sd_revalidate_disk - called the first time a new disk is seen,
3103 * performs disk spin up, read_capacity, etc.
3104 * @disk: struct gendisk we care about
3106 static int sd_revalidate_disk(struct gendisk *disk)
3108 struct scsi_disk *sdkp = scsi_disk(disk);
3109 struct scsi_device *sdp = sdkp->device;
3110 struct request_queue *q = sdkp->disk->queue;
3111 sector_t old_capacity = sdkp->capacity;
3112 unsigned char *buffer;
3113 unsigned int dev_max, rw_max;
3115 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3116 "sd_revalidate_disk\n"));
3119 * If the device is offline, don't try and read capacity or any
3120 * of the other niceties.
3122 if (!scsi_device_online(sdp))
3125 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3127 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3128 "allocation failure.\n");
3132 sd_spinup_disk(sdkp);
3135 * Without media there is no reason to ask; moreover, some devices
3136 * react badly if we do.
3138 if (sdkp->media_present) {
3139 sd_read_capacity(sdkp, buffer);
3142 * set the default to rotational. All non-rotational devices
3143 * support the block characteristics VPD page, which will
3144 * cause this to be updated correctly and any device which
3145 * doesn't support it should be treated as rotational.
3147 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3148 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3150 if (scsi_device_supports_vpd(sdp)) {
3151 sd_read_block_provisioning(sdkp);
3152 sd_read_block_limits(sdkp);
3153 sd_read_block_characteristics(sdkp);
3154 sd_zbc_read_zones(sdkp, buffer);
3157 sd_print_capacity(sdkp, old_capacity);
3159 sd_read_write_protect_flag(sdkp, buffer);
3160 sd_read_cache_type(sdkp, buffer);
3161 sd_read_app_tag_own(sdkp, buffer);
3162 sd_read_write_same(sdkp, buffer);
3163 sd_read_security(sdkp, buffer);
3167 * We now have all cache related info, determine how we deal
3168 * with flush requests.
3170 sd_set_flush_flag(sdkp);
3172 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3173 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3175 /* Some devices report a maximum block count for READ/WRITE requests. */
3176 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3177 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3179 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3180 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3181 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3183 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3184 (sector_t)BLK_DEF_MAX_SECTORS);
3186 /* Do not exceed controller limit */
3187 rw_max = min(rw_max, queue_max_hw_sectors(q));
3190 * Only update max_sectors if previously unset or if the current value
3191 * exceeds the capabilities of the hardware.
3193 if (sdkp->first_scan ||
3194 q->limits.max_sectors > q->limits.max_dev_sectors ||
3195 q->limits.max_sectors > q->limits.max_hw_sectors)
3196 q->limits.max_sectors = rw_max;
3198 sdkp->first_scan = 0;
3200 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3201 sd_config_write_same(sdkp);
3209 * sd_unlock_native_capacity - unlock native capacity
3210 * @disk: struct gendisk to set capacity for
3212 * Block layer calls this function if it detects that partitions
3213 * on @disk reach beyond the end of the device. If the SCSI host
3214 * implements ->unlock_native_capacity() method, it's invoked to
3215 * give it a chance to adjust the device capacity.
3218 * Defined by block layer. Might sleep.
3220 static void sd_unlock_native_capacity(struct gendisk *disk)
3222 struct scsi_device *sdev = scsi_disk(disk)->device;
3224 if (sdev->host->hostt->unlock_native_capacity)
3225 sdev->host->hostt->unlock_native_capacity(sdev);
3229 * sd_format_disk_name - format disk name
3230 * @prefix: name prefix - ie. "sd" for SCSI disks
3231 * @index: index of the disk to format name for
3232 * @buf: output buffer
3233 * @buflen: length of the output buffer
3235 * SCSI disk names starts at sda. The 26th device is sdz and the
3236 * 27th is sdaa. The last one for two lettered suffix is sdzz
3237 * which is followed by sdaaa.
3239 * This is basically 26 base counting with one extra 'nil' entry
3240 * at the beginning from the second digit on and can be
3241 * determined using similar method as 26 base conversion with the
3242 * index shifted -1 after each digit is computed.
3248 * 0 on success, -errno on failure.
3250 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3252 const int base = 'z' - 'a' + 1;
3253 char *begin = buf + strlen(prefix);
3254 char *end = buf + buflen;
3264 *--p = 'a' + (index % unit);
3265 index = (index / unit) - 1;
3266 } while (index >= 0);
3268 memmove(begin, p, end - p);
3269 memcpy(buf, prefix, strlen(prefix));
3275 * sd_probe - called during driver initialization and whenever a
3276 * new scsi device is attached to the system. It is called once
3277 * for each scsi device (not just disks) present.
3278 * @dev: pointer to device object
3280 * Returns 0 if successful (or not interested in this scsi device
3281 * (e.g. scanner)); 1 when there is an error.
3283 * Note: this function is invoked from the scsi mid-level.
3284 * This function sets up the mapping between a given
3285 * <host,channel,id,lun> (found in sdp) and new device name
3286 * (e.g. /dev/sda). More precisely it is the block device major
3287 * and minor number that is chosen here.
3289 * Assume sd_probe is not re-entrant (for time being)
3290 * Also think about sd_probe() and sd_remove() running coincidentally.
3292 static int sd_probe(struct device *dev)
3294 struct scsi_device *sdp = to_scsi_device(dev);
3295 struct scsi_disk *sdkp;
3300 scsi_autopm_get_device(sdp);
3302 if (sdp->type != TYPE_DISK &&
3303 sdp->type != TYPE_ZBC &&
3304 sdp->type != TYPE_MOD &&
3305 sdp->type != TYPE_RBC)
3308 #ifndef CONFIG_BLK_DEV_ZONED
3309 if (sdp->type == TYPE_ZBC)
3312 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3316 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3320 gd = alloc_disk(SD_MINORS);
3324 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3326 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3330 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3332 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3333 goto out_free_index;
3337 sdkp->driver = &sd_template;
3339 sdkp->index = index;
3340 atomic_set(&sdkp->openers, 0);
3341 atomic_set(&sdkp->device->ioerr_cnt, 0);
3343 if (!sdp->request_queue->rq_timeout) {
3344 if (sdp->type != TYPE_MOD)
3345 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3347 blk_queue_rq_timeout(sdp->request_queue,
3351 device_initialize(&sdkp->dev);
3352 sdkp->dev.parent = dev;
3353 sdkp->dev.class = &sd_disk_class;
3354 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3356 error = device_add(&sdkp->dev);
3358 goto out_free_index;
3361 dev_set_drvdata(dev, sdkp);
3363 gd->major = sd_major((index & 0xf0) >> 4);
3364 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3366 gd->fops = &sd_fops;
3367 gd->private_data = &sdkp->driver;
3368 gd->queue = sdkp->device->request_queue;
3370 /* defaults, until the device tells us otherwise */
3371 sdp->sector_size = 512;
3373 sdkp->media_present = 1;
3374 sdkp->write_prot = 0;
3375 sdkp->cache_override = 0;
3379 sdkp->first_scan = 1;
3380 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3382 sd_revalidate_disk(gd);
3384 gd->flags = GENHD_FL_EXT_DEVT;
3385 if (sdp->removable) {
3386 gd->flags |= GENHD_FL_REMOVABLE;
3387 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3388 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3391 blk_pm_runtime_init(sdp->request_queue, dev);
3392 if (sdp->rpm_autosuspend) {
3393 pm_runtime_set_autosuspend_delay(dev,
3394 sdp->host->hostt->rpm_autosuspend_delay);
3396 device_add_disk(dev, gd, NULL);
3398 sd_dif_config_host(sdkp);
3400 sd_revalidate_disk(gd);
3402 if (sdkp->security) {
3403 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3405 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3408 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3409 sdp->removable ? "removable " : "");
3410 scsi_autopm_put_device(sdp);
3415 ida_free(&sd_index_ida, index);
3421 scsi_autopm_put_device(sdp);
3426 * sd_remove - called whenever a scsi disk (previously recognized by
3427 * sd_probe) is detached from the system. It is called (potentially
3428 * multiple times) during sd module unload.
3429 * @dev: pointer to device object
3431 * Note: this function is invoked from the scsi mid-level.
3432 * This function potentially frees up a device name (e.g. /dev/sdc)
3433 * that could be re-used by a subsequent sd_probe().
3434 * This function is not called when the built-in sd driver is "exit-ed".
3436 static int sd_remove(struct device *dev)
3438 struct scsi_disk *sdkp;
3441 sdkp = dev_get_drvdata(dev);
3442 devt = disk_devt(sdkp->disk);
3443 scsi_autopm_get_device(sdkp->device);
3445 async_synchronize_full_domain(&scsi_sd_pm_domain);
3446 device_del(&sdkp->dev);
3447 del_gendisk(sdkp->disk);
3450 free_opal_dev(sdkp->opal_dev);
3452 blk_register_region(devt, SD_MINORS, NULL,
3453 sd_default_probe, NULL, NULL);
3455 mutex_lock(&sd_ref_mutex);
3456 dev_set_drvdata(dev, NULL);
3457 put_device(&sdkp->dev);
3458 mutex_unlock(&sd_ref_mutex);
3464 * scsi_disk_release - Called to free the scsi_disk structure
3465 * @dev: pointer to embedded class device
3467 * sd_ref_mutex must be held entering this routine. Because it is
3468 * called on last put, you should always use the scsi_disk_get()
3469 * scsi_disk_put() helpers which manipulate the semaphore directly
3470 * and never do a direct put_device.
3472 static void scsi_disk_release(struct device *dev)
3474 struct scsi_disk *sdkp = to_scsi_disk(dev);
3475 struct gendisk *disk = sdkp->disk;
3476 struct request_queue *q = disk->queue;
3478 ida_free(&sd_index_ida, sdkp->index);
3481 * Wait until all requests that are in progress have completed.
3482 * This is necessary to avoid that e.g. scsi_end_request() crashes
3483 * due to clearing the disk->private_data pointer. Wait from inside
3484 * scsi_disk_release() instead of from sd_release() to avoid that
3485 * freezing and unfreezing the request queue affects user space I/O
3486 * in case multiple processes open a /dev/sd... node concurrently.
3488 blk_mq_freeze_queue(q);
3489 blk_mq_unfreeze_queue(q);
3491 disk->private_data = NULL;
3493 put_device(&sdkp->device->sdev_gendev);
3498 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3500 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3501 struct scsi_sense_hdr sshdr;
3502 struct scsi_device *sdp = sdkp->device;
3506 cmd[4] |= 1; /* START */
3508 if (sdp->start_stop_pwr_cond)
3509 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3511 if (!scsi_device_online(sdp))
3514 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3515 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3517 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3518 if (driver_byte(res) == DRIVER_SENSE)
3519 sd_print_sense_hdr(sdkp, &sshdr);
3520 if (scsi_sense_valid(&sshdr) &&
3521 /* 0x3a is medium not present */
3526 /* SCSI error codes must not go to the generic layer */
3534 * Send a SYNCHRONIZE CACHE instruction down to the device through
3535 * the normal SCSI command structure. Wait for the command to
3538 static void sd_shutdown(struct device *dev)
3540 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3543 return; /* this can happen */
3545 if (pm_runtime_suspended(dev))
3548 if (sdkp->WCE && sdkp->media_present) {
3549 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3550 sd_sync_cache(sdkp, NULL);
3553 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3554 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3555 sd_start_stop_device(sdkp, 0);
3559 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3561 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3562 struct scsi_sense_hdr sshdr;
3565 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3568 if (sdkp->WCE && sdkp->media_present) {
3569 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3570 ret = sd_sync_cache(sdkp, &sshdr);
3573 /* ignore OFFLINE device */
3577 if (!scsi_sense_valid(&sshdr) ||
3578 sshdr.sense_key != ILLEGAL_REQUEST)
3582 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3583 * doesn't support sync. There's not much to do and
3584 * suspend shouldn't fail.
3590 if (sdkp->device->manage_start_stop) {
3591 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3592 /* an error is not worth aborting a system sleep */
3593 ret = sd_start_stop_device(sdkp, 0);
3594 if (ignore_stop_errors)
3601 static int sd_suspend_system(struct device *dev)
3603 return sd_suspend_common(dev, true);
3606 static int sd_suspend_runtime(struct device *dev)
3608 return sd_suspend_common(dev, false);
3611 static int sd_resume(struct device *dev)
3613 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3616 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3619 if (!sdkp->device->manage_start_stop)
3622 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3623 ret = sd_start_stop_device(sdkp, 1);
3625 opal_unlock_from_suspend(sdkp->opal_dev);
3630 * init_sd - entry point for this driver (both when built in or when
3633 * Note: this function registers this driver with the scsi mid-level.
3635 static int __init init_sd(void)
3637 int majors = 0, i, err;
3639 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3641 for (i = 0; i < SD_MAJORS; i++) {
3642 if (register_blkdev(sd_major(i), "sd") != 0)
3645 blk_register_region(sd_major(i), SD_MINORS, NULL,
3646 sd_default_probe, NULL, NULL);
3652 err = class_register(&sd_disk_class);
3656 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3658 if (!sd_cdb_cache) {
3659 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3664 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3666 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3671 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3672 if (!sd_page_pool) {
3673 printk(KERN_ERR "sd: can't init discard page pool\n");
3678 err = scsi_register_driver(&sd_template.gendrv);
3680 goto err_out_driver;
3685 mempool_destroy(sd_page_pool);
3688 mempool_destroy(sd_cdb_pool);
3691 kmem_cache_destroy(sd_cdb_cache);
3694 class_unregister(&sd_disk_class);
3696 for (i = 0; i < SD_MAJORS; i++)
3697 unregister_blkdev(sd_major(i), "sd");
3702 * exit_sd - exit point for this driver (when it is a module).
3704 * Note: this function unregisters this driver from the scsi mid-level.
3706 static void __exit exit_sd(void)
3710 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3712 scsi_unregister_driver(&sd_template.gendrv);
3713 mempool_destroy(sd_cdb_pool);
3714 mempool_destroy(sd_page_pool);
3715 kmem_cache_destroy(sd_cdb_cache);
3717 class_unregister(&sd_disk_class);
3719 for (i = 0; i < SD_MAJORS; i++) {
3720 blk_unregister_region(sd_major(i), SD_MINORS);
3721 unregister_blkdev(sd_major(i), "sd");
3725 module_init(init_sd);
3726 module_exit(exit_sd);
3728 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3730 scsi_print_sense_hdr(sdkp->device,
3731 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3734 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3736 const char *hb_string = scsi_hostbyte_string(result);
3737 const char *db_string = scsi_driverbyte_string(result);
3739 if (hb_string || db_string)
3740 sd_printk(KERN_INFO, sdkp,
3741 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3742 hb_string ? hb_string : "invalid",
3743 db_string ? db_string : "invalid");
3745 sd_printk(KERN_INFO, sdkp,
3746 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3747 msg, host_byte(result), driver_byte(result));