2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
14 * eight major numbers.
17 * sd_init and cleanups.
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
24 * Support 32k/1M disks.
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
35 #include <linux/module.h>
37 #include <linux/kernel.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/sed-opal.h>
54 #include <linux/pm_runtime.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int sd_probe(struct device *);
110 static int sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
126 static DEFINE_SPINLOCK(sd_index_lock);
127 static DEFINE_IDA(sd_index_ida);
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_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;
208 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209 SD_MAX_RETRIES, &data, &sshdr)) {
210 if (scsi_sense_valid(&sshdr))
211 sd_print_sense_hdr(sdkp, &sshdr);
214 revalidate_disk(sdkp->disk);
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
222 struct scsi_disk *sdkp = to_scsi_disk(dev);
223 struct scsi_device *sdp = sdkp->device;
225 return sprintf(buf, "%u\n", sdp->manage_start_stop);
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230 const char *buf, size_t count)
232 struct scsi_disk *sdkp = to_scsi_disk(dev);
233 struct scsi_device *sdp = sdkp->device;
236 if (!capable(CAP_SYS_ADMIN))
239 if (kstrtobool(buf, &v))
242 sdp->manage_start_stop = v;
246 static DEVICE_ATTR_RW(manage_start_stop);
249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
251 struct scsi_disk *sdkp = to_scsi_disk(dev);
253 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
257 allow_restart_store(struct device *dev, struct device_attribute *attr,
258 const char *buf, size_t count)
261 struct scsi_disk *sdkp = to_scsi_disk(dev);
262 struct scsi_device *sdp = sdkp->device;
264 if (!capable(CAP_SYS_ADMIN))
267 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
270 if (kstrtobool(buf, &v))
273 sdp->allow_restart = v;
277 static DEVICE_ATTR_RW(allow_restart);
280 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
282 struct scsi_disk *sdkp = to_scsi_disk(dev);
283 int ct = sdkp->RCD + 2*sdkp->WCE;
285 return sprintf(buf, "%s\n", sd_cache_types[ct]);
287 static DEVICE_ATTR_RW(cache_type);
290 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
292 struct scsi_disk *sdkp = to_scsi_disk(dev);
294 return sprintf(buf, "%u\n", sdkp->DPOFUA);
296 static DEVICE_ATTR_RO(FUA);
299 protection_type_show(struct device *dev, struct device_attribute *attr,
302 struct scsi_disk *sdkp = to_scsi_disk(dev);
304 return sprintf(buf, "%u\n", sdkp->protection_type);
308 protection_type_store(struct device *dev, struct device_attribute *attr,
309 const char *buf, size_t count)
311 struct scsi_disk *sdkp = to_scsi_disk(dev);
315 if (!capable(CAP_SYS_ADMIN))
318 err = kstrtouint(buf, 10, &val);
323 if (val <= T10_PI_TYPE3_PROTECTION)
324 sdkp->protection_type = val;
328 static DEVICE_ATTR_RW(protection_type);
331 protection_mode_show(struct device *dev, struct device_attribute *attr,
334 struct scsi_disk *sdkp = to_scsi_disk(dev);
335 struct scsi_device *sdp = sdkp->device;
336 unsigned int dif, dix;
338 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
339 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
341 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
347 return sprintf(buf, "none\n");
349 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
351 static DEVICE_ATTR_RO(protection_mode);
354 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
356 struct scsi_disk *sdkp = to_scsi_disk(dev);
358 return sprintf(buf, "%u\n", sdkp->ATO);
360 static DEVICE_ATTR_RO(app_tag_own);
363 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
366 struct scsi_disk *sdkp = to_scsi_disk(dev);
368 return sprintf(buf, "%u\n", sdkp->lbpme);
370 static DEVICE_ATTR_RO(thin_provisioning);
372 /* sysfs_match_string() requires dense arrays */
373 static const char *lbp_mode[] = {
374 [SD_LBP_FULL] = "full",
375 [SD_LBP_UNMAP] = "unmap",
376 [SD_LBP_WS16] = "writesame_16",
377 [SD_LBP_WS10] = "writesame_10",
378 [SD_LBP_ZERO] = "writesame_zero",
379 [SD_LBP_DISABLE] = "disabled",
383 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
386 struct scsi_disk *sdkp = to_scsi_disk(dev);
388 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
392 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
393 const char *buf, size_t count)
395 struct scsi_disk *sdkp = to_scsi_disk(dev);
396 struct scsi_device *sdp = sdkp->device;
399 if (!capable(CAP_SYS_ADMIN))
402 if (sd_is_zoned(sdkp)) {
403 sd_config_discard(sdkp, SD_LBP_DISABLE);
407 if (sdp->type != TYPE_DISK)
410 mode = sysfs_match_string(lbp_mode, buf);
414 sd_config_discard(sdkp, mode);
418 static DEVICE_ATTR_RW(provisioning_mode);
420 /* sysfs_match_string() requires dense arrays */
421 static const char *zeroing_mode[] = {
422 [SD_ZERO_WRITE] = "write",
423 [SD_ZERO_WS] = "writesame",
424 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
425 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
429 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
432 struct scsi_disk *sdkp = to_scsi_disk(dev);
434 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
438 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
439 const char *buf, size_t count)
441 struct scsi_disk *sdkp = to_scsi_disk(dev);
444 if (!capable(CAP_SYS_ADMIN))
447 mode = sysfs_match_string(zeroing_mode, buf);
451 sdkp->zeroing_mode = mode;
455 static DEVICE_ATTR_RW(zeroing_mode);
458 max_medium_access_timeouts_show(struct device *dev,
459 struct device_attribute *attr, char *buf)
461 struct scsi_disk *sdkp = to_scsi_disk(dev);
463 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
467 max_medium_access_timeouts_store(struct device *dev,
468 struct device_attribute *attr, const char *buf,
471 struct scsi_disk *sdkp = to_scsi_disk(dev);
474 if (!capable(CAP_SYS_ADMIN))
477 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
479 return err ? err : count;
481 static DEVICE_ATTR_RW(max_medium_access_timeouts);
484 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
487 struct scsi_disk *sdkp = to_scsi_disk(dev);
489 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
493 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
494 const char *buf, size_t count)
496 struct scsi_disk *sdkp = to_scsi_disk(dev);
497 struct scsi_device *sdp = sdkp->device;
501 if (!capable(CAP_SYS_ADMIN))
504 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
507 err = kstrtoul(buf, 10, &max);
513 sdp->no_write_same = 1;
514 else if (max <= SD_MAX_WS16_BLOCKS) {
515 sdp->no_write_same = 0;
516 sdkp->max_ws_blocks = max;
519 sd_config_write_same(sdkp);
523 static DEVICE_ATTR_RW(max_write_same_blocks);
525 static struct attribute *sd_disk_attrs[] = {
526 &dev_attr_cache_type.attr,
528 &dev_attr_allow_restart.attr,
529 &dev_attr_manage_start_stop.attr,
530 &dev_attr_protection_type.attr,
531 &dev_attr_protection_mode.attr,
532 &dev_attr_app_tag_own.attr,
533 &dev_attr_thin_provisioning.attr,
534 &dev_attr_provisioning_mode.attr,
535 &dev_attr_zeroing_mode.attr,
536 &dev_attr_max_write_same_blocks.attr,
537 &dev_attr_max_medium_access_timeouts.attr,
540 ATTRIBUTE_GROUPS(sd_disk);
542 static struct class sd_disk_class = {
544 .owner = THIS_MODULE,
545 .dev_release = scsi_disk_release,
546 .dev_groups = sd_disk_groups,
549 static const struct dev_pm_ops sd_pm_ops = {
550 .suspend = sd_suspend_system,
552 .poweroff = sd_suspend_system,
553 .restore = sd_resume,
554 .runtime_suspend = sd_suspend_runtime,
555 .runtime_resume = sd_resume,
558 static struct scsi_driver sd_template = {
561 .owner = THIS_MODULE,
564 .shutdown = sd_shutdown,
568 .init_command = sd_init_command,
569 .uninit_command = sd_uninit_command,
571 .eh_action = sd_eh_action,
572 .eh_reset = sd_eh_reset,
576 * Dummy kobj_map->probe function.
577 * The default ->probe function will call modprobe, which is
578 * pointless as this module is already loaded.
580 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
586 * Device no to disk mapping:
588 * major disc2 disc p1
589 * |............|.............|....|....| <- dev_t
592 * Inside a major, we have 16k disks, however mapped non-
593 * contiguously. The first 16 disks are for major0, the next
594 * ones with major1, ... Disk 256 is for major0 again, disk 272
596 * As we stay compatible with our numbering scheme, we can reuse
597 * the well-know SCSI majors 8, 65--71, 136--143.
599 static int sd_major(int major_idx)
603 return SCSI_DISK0_MAJOR;
605 return SCSI_DISK1_MAJOR + major_idx - 1;
607 return SCSI_DISK8_MAJOR + major_idx - 8;
610 return 0; /* shut up gcc */
614 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
616 struct scsi_disk *sdkp = NULL;
618 mutex_lock(&sd_ref_mutex);
620 if (disk->private_data) {
621 sdkp = scsi_disk(disk);
622 if (scsi_device_get(sdkp->device) == 0)
623 get_device(&sdkp->dev);
627 mutex_unlock(&sd_ref_mutex);
631 static void scsi_disk_put(struct scsi_disk *sdkp)
633 struct scsi_device *sdev = sdkp->device;
635 mutex_lock(&sd_ref_mutex);
636 put_device(&sdkp->dev);
637 scsi_device_put(sdev);
638 mutex_unlock(&sd_ref_mutex);
641 #ifdef CONFIG_BLK_SED_OPAL
642 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
643 size_t len, bool send)
645 struct scsi_device *sdev = data;
649 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
651 put_unaligned_be16(spsp, &cdb[2]);
652 put_unaligned_be32(len, &cdb[6]);
654 ret = scsi_execute_req(sdev, cdb,
655 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
656 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
657 return ret <= 0 ? ret : -EIO;
659 #endif /* CONFIG_BLK_SED_OPAL */
661 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
662 unsigned int dix, unsigned int dif)
664 struct bio *bio = scmd->request->bio;
665 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
666 unsigned int protect = 0;
668 if (dix) { /* DIX Type 0, 1, 2, 3 */
669 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
670 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
672 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
673 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
676 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
677 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
679 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
680 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
683 if (dif) { /* DIX/DIF Type 1, 2, 3 */
684 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
686 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
687 protect = 3 << 5; /* Disable target PI checking */
689 protect = 1 << 5; /* Enable target PI checking */
692 scsi_set_prot_op(scmd, prot_op);
693 scsi_set_prot_type(scmd, dif);
694 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
699 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
701 struct request_queue *q = sdkp->disk->queue;
702 unsigned int logical_block_size = sdkp->device->sector_size;
703 unsigned int max_blocks = 0;
705 q->limits.discard_alignment =
706 sdkp->unmap_alignment * logical_block_size;
707 q->limits.discard_granularity =
708 max(sdkp->physical_block_size,
709 sdkp->unmap_granularity * logical_block_size);
710 sdkp->provisioning_mode = mode;
716 blk_queue_max_discard_sectors(q, 0);
717 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
721 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
722 (u32)SD_MAX_WS16_BLOCKS);
726 if (sdkp->device->unmap_limit_for_ws)
727 max_blocks = sdkp->max_unmap_blocks;
729 max_blocks = sdkp->max_ws_blocks;
731 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
735 if (sdkp->device->unmap_limit_for_ws)
736 max_blocks = sdkp->max_unmap_blocks;
738 max_blocks = sdkp->max_ws_blocks;
740 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
744 max_blocks = min_not_zero(sdkp->max_ws_blocks,
745 (u32)SD_MAX_WS10_BLOCKS);
749 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
750 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
753 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
755 struct scsi_device *sdp = cmd->device;
756 struct request *rq = cmd->request;
757 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
758 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
759 unsigned int data_len = 24;
762 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
763 if (!rq->special_vec.bv_page)
764 return BLKPREP_DEFER;
765 rq->special_vec.bv_offset = 0;
766 rq->special_vec.bv_len = data_len;
767 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
770 cmd->cmnd[0] = UNMAP;
773 buf = page_address(rq->special_vec.bv_page);
774 put_unaligned_be16(6 + 16, &buf[0]);
775 put_unaligned_be16(16, &buf[2]);
776 put_unaligned_be64(sector, &buf[8]);
777 put_unaligned_be32(nr_sectors, &buf[16]);
779 cmd->allowed = SD_MAX_RETRIES;
780 cmd->transfersize = data_len;
781 rq->timeout = SD_TIMEOUT;
782 scsi_req(rq)->resid_len = data_len;
784 return scsi_init_io(cmd);
787 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
789 struct scsi_device *sdp = cmd->device;
790 struct request *rq = cmd->request;
791 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
792 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
793 u32 data_len = sdp->sector_size;
795 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
796 if (!rq->special_vec.bv_page)
797 return BLKPREP_DEFER;
798 rq->special_vec.bv_offset = 0;
799 rq->special_vec.bv_len = data_len;
800 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
803 cmd->cmnd[0] = WRITE_SAME_16;
805 cmd->cmnd[1] = 0x8; /* UNMAP */
806 put_unaligned_be64(sector, &cmd->cmnd[2]);
807 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
809 cmd->allowed = SD_MAX_RETRIES;
810 cmd->transfersize = data_len;
811 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
812 scsi_req(rq)->resid_len = data_len;
814 return scsi_init_io(cmd);
817 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
819 struct scsi_device *sdp = cmd->device;
820 struct request *rq = cmd->request;
821 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
822 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
823 u32 data_len = sdp->sector_size;
825 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
826 if (!rq->special_vec.bv_page)
827 return BLKPREP_DEFER;
828 rq->special_vec.bv_offset = 0;
829 rq->special_vec.bv_len = data_len;
830 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
833 cmd->cmnd[0] = WRITE_SAME;
835 cmd->cmnd[1] = 0x8; /* UNMAP */
836 put_unaligned_be32(sector, &cmd->cmnd[2]);
837 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
839 cmd->allowed = SD_MAX_RETRIES;
840 cmd->transfersize = data_len;
841 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
842 scsi_req(rq)->resid_len = data_len;
844 return scsi_init_io(cmd);
847 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
849 struct request *rq = cmd->request;
850 struct scsi_device *sdp = cmd->device;
851 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
852 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
853 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
856 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
857 switch (sdkp->zeroing_mode) {
858 case SD_ZERO_WS16_UNMAP:
859 ret = sd_setup_write_same16_cmnd(cmd, true);
861 case SD_ZERO_WS10_UNMAP:
862 ret = sd_setup_write_same10_cmnd(cmd, true);
867 if (sdp->no_write_same)
868 return BLKPREP_INVALID;
870 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
871 ret = sd_setup_write_same16_cmnd(cmd, false);
873 ret = sd_setup_write_same10_cmnd(cmd, false);
876 if (sd_is_zoned(sdkp) && ret == BLKPREP_OK)
877 return sd_zbc_write_lock_zone(cmd);
882 static void sd_config_write_same(struct scsi_disk *sdkp)
884 struct request_queue *q = sdkp->disk->queue;
885 unsigned int logical_block_size = sdkp->device->sector_size;
887 if (sdkp->device->no_write_same) {
888 sdkp->max_ws_blocks = 0;
892 /* Some devices can not handle block counts above 0xffff despite
893 * supporting WRITE SAME(16). Consequently we default to 64k
894 * blocks per I/O unless the device explicitly advertises a
897 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
898 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
899 (u32)SD_MAX_WS16_BLOCKS);
900 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
901 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
902 (u32)SD_MAX_WS10_BLOCKS);
904 sdkp->device->no_write_same = 1;
905 sdkp->max_ws_blocks = 0;
908 if (sdkp->lbprz && sdkp->lbpws)
909 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
910 else if (sdkp->lbprz && sdkp->lbpws10)
911 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
912 else if (sdkp->max_ws_blocks)
913 sdkp->zeroing_mode = SD_ZERO_WS;
915 sdkp->zeroing_mode = SD_ZERO_WRITE;
917 if (sdkp->max_ws_blocks &&
918 sdkp->physical_block_size > logical_block_size) {
920 * Reporting a maximum number of blocks that is not aligned
921 * on the device physical size would cause a large write same
922 * request to be split into physically unaligned chunks by
923 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
924 * even if the caller of these functions took care to align the
925 * large request. So make sure the maximum reported is aligned
926 * to the device physical block size. This is only an optional
927 * optimization for regular disks, but this is mandatory to
928 * avoid failure of large write same requests directed at
929 * sequential write required zones of host-managed ZBC disks.
931 sdkp->max_ws_blocks =
932 round_down(sdkp->max_ws_blocks,
933 bytes_to_logical(sdkp->device,
934 sdkp->physical_block_size));
938 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
939 (logical_block_size >> 9));
940 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
941 (logical_block_size >> 9));
945 * sd_setup_write_same_cmnd - write the same data to multiple blocks
946 * @cmd: command to prepare
948 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
949 * the preference indicated by the target device.
951 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
953 struct request *rq = cmd->request;
954 struct scsi_device *sdp = cmd->device;
955 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
956 struct bio *bio = rq->bio;
957 sector_t sector = blk_rq_pos(rq);
958 unsigned int nr_sectors = blk_rq_sectors(rq);
959 unsigned int nr_bytes = blk_rq_bytes(rq);
962 if (sdkp->device->no_write_same)
963 return BLKPREP_INVALID;
965 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
967 if (sd_is_zoned(sdkp)) {
968 ret = sd_zbc_write_lock_zone(cmd);
969 if (ret != BLKPREP_OK)
973 sector >>= ilog2(sdp->sector_size) - 9;
974 nr_sectors >>= ilog2(sdp->sector_size) - 9;
976 rq->timeout = SD_WRITE_SAME_TIMEOUT;
978 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
980 cmd->cmnd[0] = WRITE_SAME_16;
981 put_unaligned_be64(sector, &cmd->cmnd[2]);
982 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
985 cmd->cmnd[0] = WRITE_SAME;
986 put_unaligned_be32(sector, &cmd->cmnd[2]);
987 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
990 cmd->transfersize = sdp->sector_size;
991 cmd->allowed = SD_MAX_RETRIES;
994 * For WRITE SAME the data transferred via the DATA OUT buffer is
995 * different from the amount of data actually written to the target.
997 * We set up __data_len to the amount of data transferred via the
998 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
999 * to transfer a single sector of data first, but then reset it to
1000 * the amount of data to be written right after so that the I/O path
1001 * knows how much to actually write.
1003 rq->__data_len = sdp->sector_size;
1004 ret = scsi_init_io(cmd);
1005 rq->__data_len = nr_bytes;
1007 if (sd_is_zoned(sdkp) && ret != BLKPREP_OK)
1008 sd_zbc_write_unlock_zone(cmd);
1013 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1015 struct request *rq = cmd->request;
1017 /* flush requests don't perform I/O, zero the S/G table */
1018 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1020 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1022 cmd->transfersize = 0;
1023 cmd->allowed = SD_MAX_RETRIES;
1025 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1029 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1031 struct request *rq = SCpnt->request;
1032 struct scsi_device *sdp = SCpnt->device;
1033 struct gendisk *disk = rq->rq_disk;
1034 struct scsi_disk *sdkp = scsi_disk(disk);
1035 sector_t block = blk_rq_pos(rq);
1037 unsigned int this_count = blk_rq_sectors(rq);
1038 unsigned int dif, dix;
1039 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
1041 unsigned char protect;
1044 ret = sd_zbc_write_lock_zone(SCpnt);
1045 if (ret != BLKPREP_OK)
1049 ret = scsi_init_io(SCpnt);
1050 if (ret != BLKPREP_OK)
1052 WARN_ON_ONCE(SCpnt != rq->special);
1054 /* from here on until we're complete, any goto out
1055 * is used for a killable error condition */
1059 scmd_printk(KERN_INFO, SCpnt,
1060 "%s: block=%llu, count=%d\n",
1061 __func__, (unsigned long long)block, this_count));
1063 if (!sdp || !scsi_device_online(sdp) ||
1064 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1065 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1066 "Finishing %u sectors\n",
1067 blk_rq_sectors(rq)));
1068 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1069 "Retry with 0x%p\n", SCpnt));
1075 * quietly refuse to do anything to a changed disc until
1076 * the changed bit has been reset
1078 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1083 * Some SD card readers can't handle multi-sector accesses which touch
1084 * the last one or two hardware sectors. Split accesses as needed.
1086 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1087 (sdp->sector_size / 512);
1089 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1090 if (block < threshold) {
1091 /* Access up to the threshold but not beyond */
1092 this_count = threshold - block;
1094 /* Access only a single hardware sector */
1095 this_count = sdp->sector_size / 512;
1099 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1100 (unsigned long long)block));
1103 * If we have a 1K hardware sectorsize, prevent access to single
1104 * 512 byte sectors. In theory we could handle this - in fact
1105 * the scsi cdrom driver must be able to handle this because
1106 * we typically use 1K blocksizes, and cdroms typically have
1107 * 2K hardware sectorsizes. Of course, things are simpler
1108 * with the cdrom, since it is read-only. For performance
1109 * reasons, the filesystems should be able to handle this
1110 * and not force the scsi disk driver to use bounce buffers
1113 if (sdp->sector_size == 1024) {
1114 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1115 scmd_printk(KERN_ERR, SCpnt,
1116 "Bad block number requested\n");
1120 this_count = this_count >> 1;
1123 if (sdp->sector_size == 2048) {
1124 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1125 scmd_printk(KERN_ERR, SCpnt,
1126 "Bad block number requested\n");
1130 this_count = this_count >> 2;
1133 if (sdp->sector_size == 4096) {
1134 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1135 scmd_printk(KERN_ERR, SCpnt,
1136 "Bad block number requested\n");
1140 this_count = this_count >> 3;
1143 if (rq_data_dir(rq) == WRITE) {
1144 SCpnt->cmnd[0] = WRITE_6;
1146 if (blk_integrity_rq(rq))
1147 sd_dif_prepare(SCpnt);
1149 } else if (rq_data_dir(rq) == READ) {
1150 SCpnt->cmnd[0] = READ_6;
1152 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1156 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1157 "%s %d/%u 512 byte blocks.\n",
1158 (rq_data_dir(rq) == WRITE) ?
1159 "writing" : "reading", this_count,
1160 blk_rq_sectors(rq)));
1162 dix = scsi_prot_sg_count(SCpnt);
1163 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1166 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1170 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1171 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1173 if (unlikely(SCpnt->cmnd == NULL)) {
1174 ret = BLKPREP_DEFER;
1178 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1179 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1180 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1181 SCpnt->cmnd[7] = 0x18;
1182 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1183 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1186 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1187 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1188 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1189 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1190 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1191 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1192 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1193 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1195 /* Expected Indirect LBA */
1196 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1197 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1198 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1199 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1201 /* Transfer length */
1202 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1203 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1204 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1205 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1206 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1207 SCpnt->cmnd[0] += READ_16 - READ_6;
1208 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1209 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1210 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1211 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1212 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1213 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1214 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1215 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1216 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1217 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1218 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1219 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1220 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1221 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1222 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1223 scsi_device_protection(SCpnt->device) ||
1224 SCpnt->device->use_10_for_rw) {
1225 SCpnt->cmnd[0] += READ_10 - READ_6;
1226 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1227 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1228 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1229 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1230 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1231 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1232 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1233 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1235 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1237 * This happens only if this drive failed
1238 * 10byte rw command with ILLEGAL_REQUEST
1239 * during operation and thus turned off
1242 scmd_printk(KERN_ERR, SCpnt,
1243 "FUA write on READ/WRITE(6) drive\n");
1247 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1248 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1249 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1250 SCpnt->cmnd[4] = (unsigned char) this_count;
1253 SCpnt->sdb.length = this_count * sdp->sector_size;
1256 * We shouldn't disconnect in the middle of a sector, so with a dumb
1257 * host adapter, it's safe to assume that we can at least transfer
1258 * this many bytes between each connect / disconnect.
1260 SCpnt->transfersize = sdp->sector_size;
1261 SCpnt->underflow = this_count << 9;
1262 SCpnt->allowed = SD_MAX_RETRIES;
1265 * This indicates that the command is ready from our end to be
1270 if (zoned_write && ret != BLKPREP_OK)
1271 sd_zbc_write_unlock_zone(SCpnt);
1276 static int sd_init_command(struct scsi_cmnd *cmd)
1278 struct request *rq = cmd->request;
1280 switch (req_op(rq)) {
1281 case REQ_OP_DISCARD:
1282 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1284 return sd_setup_unmap_cmnd(cmd);
1286 return sd_setup_write_same16_cmnd(cmd, true);
1288 return sd_setup_write_same10_cmnd(cmd, true);
1290 return sd_setup_write_same10_cmnd(cmd, false);
1292 return BLKPREP_INVALID;
1294 case REQ_OP_WRITE_ZEROES:
1295 return sd_setup_write_zeroes_cmnd(cmd);
1296 case REQ_OP_WRITE_SAME:
1297 return sd_setup_write_same_cmnd(cmd);
1299 return sd_setup_flush_cmnd(cmd);
1302 return sd_setup_read_write_cmnd(cmd);
1303 case REQ_OP_ZONE_REPORT:
1304 return sd_zbc_setup_report_cmnd(cmd);
1305 case REQ_OP_ZONE_RESET:
1306 return sd_zbc_setup_reset_cmnd(cmd);
1312 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1314 struct request *rq = SCpnt->request;
1316 if (SCpnt->flags & SCMD_ZONE_WRITE_LOCK)
1317 sd_zbc_write_unlock_zone(SCpnt);
1319 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1320 __free_page(rq->special_vec.bv_page);
1322 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1323 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1330 * sd_open - open a scsi disk device
1331 * @bdev: Block device of the scsi disk to open
1332 * @mode: FMODE_* mask
1334 * Returns 0 if successful. Returns a negated errno value in case
1337 * Note: This can be called from a user context (e.g. fsck(1) )
1338 * or from within the kernel (e.g. as a result of a mount(1) ).
1339 * In the latter case @inode and @filp carry an abridged amount
1340 * of information as noted above.
1342 * Locking: called with bdev->bd_mutex held.
1344 static int sd_open(struct block_device *bdev, fmode_t mode)
1346 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1347 struct scsi_device *sdev;
1353 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1355 sdev = sdkp->device;
1358 * If the device is in error recovery, wait until it is done.
1359 * If the device is offline, then disallow any access to it.
1362 if (!scsi_block_when_processing_errors(sdev))
1365 if (sdev->removable || sdkp->write_prot)
1366 check_disk_change(bdev);
1369 * If the drive is empty, just let the open fail.
1371 retval = -ENOMEDIUM;
1372 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1376 * If the device has the write protect tab set, have the open fail
1377 * if the user expects to be able to write to the thing.
1380 if (sdkp->write_prot && (mode & FMODE_WRITE))
1384 * It is possible that the disk changing stuff resulted in
1385 * the device being taken offline. If this is the case,
1386 * report this to the user, and don't pretend that the
1387 * open actually succeeded.
1390 if (!scsi_device_online(sdev))
1393 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1394 if (scsi_block_when_processing_errors(sdev))
1395 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1401 scsi_disk_put(sdkp);
1406 * sd_release - invoked when the (last) close(2) is called on this
1408 * @disk: disk to release
1409 * @mode: FMODE_* mask
1413 * Note: may block (uninterruptible) if error recovery is underway
1416 * Locking: called with bdev->bd_mutex held.
1418 static void sd_release(struct gendisk *disk, fmode_t mode)
1420 struct scsi_disk *sdkp = scsi_disk(disk);
1421 struct scsi_device *sdev = sdkp->device;
1423 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1425 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1426 if (scsi_block_when_processing_errors(sdev))
1427 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1431 * XXX and what if there are packets in flight and this close()
1432 * XXX is followed by a "rmmod sd_mod"?
1435 scsi_disk_put(sdkp);
1438 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1440 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1441 struct scsi_device *sdp = sdkp->device;
1442 struct Scsi_Host *host = sdp->host;
1443 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1446 /* default to most commonly used values */
1447 diskinfo[0] = 0x40; /* 1 << 6 */
1448 diskinfo[1] = 0x20; /* 1 << 5 */
1449 diskinfo[2] = capacity >> 11;
1451 /* override with calculated, extended default, or driver values */
1452 if (host->hostt->bios_param)
1453 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1455 scsicam_bios_param(bdev, capacity, diskinfo);
1457 geo->heads = diskinfo[0];
1458 geo->sectors = diskinfo[1];
1459 geo->cylinders = diskinfo[2];
1464 * sd_ioctl - process an ioctl
1465 * @bdev: target block device
1466 * @mode: FMODE_* mask
1467 * @cmd: ioctl command number
1468 * @arg: this is third argument given to ioctl(2) system call.
1469 * Often contains a pointer.
1471 * Returns 0 if successful (some ioctls return positive numbers on
1472 * success as well). Returns a negated errno value in case of error.
1474 * Note: most ioctls are forward onto the block subsystem or further
1475 * down in the scsi subsystem.
1477 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1478 unsigned int cmd, unsigned long arg)
1480 struct gendisk *disk = bdev->bd_disk;
1481 struct scsi_disk *sdkp = scsi_disk(disk);
1482 struct scsi_device *sdp = sdkp->device;
1483 void __user *p = (void __user *)arg;
1486 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1487 "cmd=0x%x\n", disk->disk_name, cmd));
1489 error = scsi_verify_blk_ioctl(bdev, cmd);
1494 * If we are in the middle of error recovery, don't let anyone
1495 * else try and use this device. Also, if error recovery fails, it
1496 * may try and take the device offline, in which case all further
1497 * access to the device is prohibited.
1499 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1500 (mode & FMODE_NDELAY) != 0);
1504 if (is_sed_ioctl(cmd))
1505 return sed_ioctl(sdkp->opal_dev, cmd, p);
1508 * Send SCSI addressing ioctls directly to mid level, send other
1509 * ioctls to block level and then onto mid level if they can't be
1513 case SCSI_IOCTL_GET_IDLUN:
1514 case SCSI_IOCTL_GET_BUS_NUMBER:
1515 error = scsi_ioctl(sdp, cmd, p);
1518 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1519 if (error != -ENOTTY)
1521 error = scsi_ioctl(sdp, cmd, p);
1528 static void set_media_not_present(struct scsi_disk *sdkp)
1530 if (sdkp->media_present)
1531 sdkp->device->changed = 1;
1533 if (sdkp->device->removable) {
1534 sdkp->media_present = 0;
1539 static int media_not_present(struct scsi_disk *sdkp,
1540 struct scsi_sense_hdr *sshdr)
1542 if (!scsi_sense_valid(sshdr))
1545 /* not invoked for commands that could return deferred errors */
1546 switch (sshdr->sense_key) {
1547 case UNIT_ATTENTION:
1549 /* medium not present */
1550 if (sshdr->asc == 0x3A) {
1551 set_media_not_present(sdkp);
1559 * sd_check_events - check media events
1560 * @disk: kernel device descriptor
1561 * @clearing: disk events currently being cleared
1563 * Returns mask of DISK_EVENT_*.
1565 * Note: this function is invoked from the block subsystem.
1567 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1569 struct scsi_disk *sdkp = scsi_disk_get(disk);
1570 struct scsi_device *sdp;
1577 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1580 * If the device is offline, don't send any commands - just pretend as
1581 * if the command failed. If the device ever comes back online, we
1582 * can deal with it then. It is only because of unrecoverable errors
1583 * that we would ever take a device offline in the first place.
1585 if (!scsi_device_online(sdp)) {
1586 set_media_not_present(sdkp);
1591 * Using TEST_UNIT_READY enables differentiation between drive with
1592 * no cartridge loaded - NOT READY, drive with changed cartridge -
1593 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1595 * Drives that auto spin down. eg iomega jaz 1G, will be started
1596 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1597 * sd_revalidate() is called.
1599 if (scsi_block_when_processing_errors(sdp)) {
1600 struct scsi_sense_hdr sshdr = { 0, };
1602 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1605 /* failed to execute TUR, assume media not present */
1606 if (host_byte(retval)) {
1607 set_media_not_present(sdkp);
1611 if (media_not_present(sdkp, &sshdr))
1616 * For removable scsi disk we have to recognise the presence
1617 * of a disk in the drive.
1619 if (!sdkp->media_present)
1621 sdkp->media_present = 1;
1624 * sdp->changed is set under the following conditions:
1626 * Medium present state has changed in either direction.
1627 * Device has indicated UNIT_ATTENTION.
1629 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1631 scsi_disk_put(sdkp);
1635 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1638 struct scsi_device *sdp = sdkp->device;
1639 const int timeout = sdp->request_queue->rq_timeout
1640 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1641 struct scsi_sense_hdr my_sshdr;
1643 if (!scsi_device_online(sdp))
1646 /* caller might not be interested in sense, but we need it */
1650 for (retries = 3; retries > 0; --retries) {
1651 unsigned char cmd[10] = { 0 };
1653 cmd[0] = SYNCHRONIZE_CACHE;
1655 * Leave the rest of the command zero to indicate
1658 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1659 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1665 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1667 if (driver_byte(res) & DRIVER_SENSE)
1668 sd_print_sense_hdr(sdkp, sshdr);
1670 /* we need to evaluate the error return */
1671 if (scsi_sense_valid(sshdr) &&
1672 (sshdr->asc == 0x3a || /* medium not present */
1673 sshdr->asc == 0x20)) /* invalid command */
1674 /* this is no error here */
1677 switch (host_byte(res)) {
1678 /* ignore errors due to racing a disconnection */
1679 case DID_BAD_TARGET:
1680 case DID_NO_CONNECT:
1682 /* signal the upper layer it might try again */
1686 case DID_SOFT_ERROR:
1695 static void sd_rescan(struct device *dev)
1697 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1699 revalidate_disk(sdkp->disk);
1703 #ifdef CONFIG_COMPAT
1705 * This gets directly called from VFS. When the ioctl
1706 * is not recognized we go back to the other translation paths.
1708 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1709 unsigned int cmd, unsigned long arg)
1711 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1714 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1715 (mode & FMODE_NDELAY) != 0);
1720 * Let the static ioctl translation table take care of it.
1722 if (!sdev->host->hostt->compat_ioctl)
1723 return -ENOIOCTLCMD;
1724 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1728 static char sd_pr_type(enum pr_type type)
1731 case PR_WRITE_EXCLUSIVE:
1733 case PR_EXCLUSIVE_ACCESS:
1735 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1737 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1739 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1741 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1748 static int sd_pr_command(struct block_device *bdev, u8 sa,
1749 u64 key, u64 sa_key, u8 type, u8 flags)
1751 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1752 struct scsi_sense_hdr sshdr;
1754 u8 cmd[16] = { 0, };
1755 u8 data[24] = { 0, };
1757 cmd[0] = PERSISTENT_RESERVE_OUT;
1760 put_unaligned_be32(sizeof(data), &cmd[5]);
1762 put_unaligned_be64(key, &data[0]);
1763 put_unaligned_be64(sa_key, &data[8]);
1766 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1767 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1769 if ((driver_byte(result) & DRIVER_SENSE) &&
1770 (scsi_sense_valid(&sshdr))) {
1771 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1772 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1778 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1781 if (flags & ~PR_FL_IGNORE_KEY)
1783 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1784 old_key, new_key, 0,
1785 (1 << 0) /* APTPL */);
1788 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1793 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1796 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1798 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1801 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1802 enum pr_type type, bool abort)
1804 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1805 sd_pr_type(type), 0);
1808 static int sd_pr_clear(struct block_device *bdev, u64 key)
1810 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1813 static const struct pr_ops sd_pr_ops = {
1814 .pr_register = sd_pr_register,
1815 .pr_reserve = sd_pr_reserve,
1816 .pr_release = sd_pr_release,
1817 .pr_preempt = sd_pr_preempt,
1818 .pr_clear = sd_pr_clear,
1821 static const struct block_device_operations sd_fops = {
1822 .owner = THIS_MODULE,
1824 .release = sd_release,
1826 .getgeo = sd_getgeo,
1827 #ifdef CONFIG_COMPAT
1828 .compat_ioctl = sd_compat_ioctl,
1830 .check_events = sd_check_events,
1831 .revalidate_disk = sd_revalidate_disk,
1832 .unlock_native_capacity = sd_unlock_native_capacity,
1833 .pr_ops = &sd_pr_ops,
1837 * sd_eh_reset - reset error handling callback
1838 * @scmd: sd-issued command that has failed
1840 * This function is called by the SCSI midlayer before starting
1841 * SCSI EH. When counting medium access failures we have to be
1842 * careful to register it only only once per device and SCSI EH run;
1843 * there might be several timed out commands which will cause the
1844 * 'max_medium_access_timeouts' counter to trigger after the first
1845 * SCSI EH run already and set the device to offline.
1846 * So this function resets the internal counter before starting SCSI EH.
1848 static void sd_eh_reset(struct scsi_cmnd *scmd)
1850 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1852 /* New SCSI EH run, reset gate variable */
1853 sdkp->ignore_medium_access_errors = false;
1857 * sd_eh_action - error handling callback
1858 * @scmd: sd-issued command that has failed
1859 * @eh_disp: The recovery disposition suggested by the midlayer
1861 * This function is called by the SCSI midlayer upon completion of an
1862 * error test command (currently TEST UNIT READY). The result of sending
1863 * the eh command is passed in eh_disp. We're looking for devices that
1864 * fail medium access commands but are OK with non access commands like
1865 * test unit ready (so wrongly see the device as having a successful
1868 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1870 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1871 struct scsi_device *sdev = scmd->device;
1873 if (!scsi_device_online(sdev) ||
1874 !scsi_medium_access_command(scmd) ||
1875 host_byte(scmd->result) != DID_TIME_OUT ||
1880 * The device has timed out executing a medium access command.
1881 * However, the TEST UNIT READY command sent during error
1882 * handling completed successfully. Either the device is in the
1883 * process of recovering or has it suffered an internal failure
1884 * that prevents access to the storage medium.
1886 if (!sdkp->ignore_medium_access_errors) {
1887 sdkp->medium_access_timed_out++;
1888 sdkp->ignore_medium_access_errors = true;
1892 * If the device keeps failing read/write commands but TEST UNIT
1893 * READY always completes successfully we assume that medium
1894 * access is no longer possible and take the device offline.
1896 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1897 scmd_printk(KERN_ERR, scmd,
1898 "Medium access timeout failure. Offlining disk!\n");
1899 mutex_lock(&sdev->state_mutex);
1900 scsi_device_set_state(sdev, SDEV_OFFLINE);
1901 mutex_unlock(&sdev->state_mutex);
1909 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1911 struct request *req = scmd->request;
1912 struct scsi_device *sdev = scmd->device;
1913 unsigned int transferred, good_bytes;
1914 u64 start_lba, end_lba, bad_lba;
1917 * Some commands have a payload smaller than the device logical
1918 * block size (e.g. INQUIRY on a 4K disk).
1920 if (scsi_bufflen(scmd) <= sdev->sector_size)
1923 /* Check if we have a 'bad_lba' information */
1924 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1925 SCSI_SENSE_BUFFERSIZE,
1930 * If the bad lba was reported incorrectly, we have no idea where
1933 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1934 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1935 if (bad_lba < start_lba || bad_lba >= end_lba)
1939 * resid is optional but mostly filled in. When it's unused,
1940 * its value is zero, so we assume the whole buffer transferred
1942 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1944 /* This computation should always be done in terms of the
1945 * resolution of the device's medium.
1947 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1949 return min(good_bytes, transferred);
1953 * sd_done - bottom half handler: called when the lower level
1954 * driver has completed (successfully or otherwise) a scsi command.
1955 * @SCpnt: mid-level's per command structure.
1957 * Note: potentially run from within an ISR. Must not block.
1959 static int sd_done(struct scsi_cmnd *SCpnt)
1961 int result = SCpnt->result;
1962 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1963 unsigned int sector_size = SCpnt->device->sector_size;
1965 struct scsi_sense_hdr sshdr;
1966 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1967 struct request *req = SCpnt->request;
1968 int sense_valid = 0;
1969 int sense_deferred = 0;
1971 switch (req_op(req)) {
1972 case REQ_OP_DISCARD:
1973 case REQ_OP_WRITE_ZEROES:
1974 case REQ_OP_WRITE_SAME:
1975 case REQ_OP_ZONE_RESET:
1977 good_bytes = blk_rq_bytes(req);
1978 scsi_set_resid(SCpnt, 0);
1981 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1984 case REQ_OP_ZONE_REPORT:
1986 good_bytes = scsi_bufflen(SCpnt)
1987 - scsi_get_resid(SCpnt);
1988 scsi_set_resid(SCpnt, 0);
1991 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1996 * In case of bogus fw or device, we could end up having
1997 * an unaligned partial completion. Check this here and force
2000 resid = scsi_get_resid(SCpnt);
2001 if (resid & (sector_size - 1)) {
2002 sd_printk(KERN_INFO, sdkp,
2003 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2004 resid, sector_size);
2005 resid = min(scsi_bufflen(SCpnt),
2006 round_up(resid, sector_size));
2007 scsi_set_resid(SCpnt, resid);
2012 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2014 sense_deferred = scsi_sense_is_deferred(&sshdr);
2016 sdkp->medium_access_timed_out = 0;
2018 if (driver_byte(result) != DRIVER_SENSE &&
2019 (!sense_valid || sense_deferred))
2022 switch (sshdr.sense_key) {
2023 case HARDWARE_ERROR:
2025 good_bytes = sd_completed_bytes(SCpnt);
2027 case RECOVERED_ERROR:
2028 good_bytes = scsi_bufflen(SCpnt);
2031 /* This indicates a false check condition, so ignore it. An
2032 * unknown amount of data was transferred so treat it as an
2036 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2038 case ABORTED_COMMAND:
2039 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2040 good_bytes = sd_completed_bytes(SCpnt);
2042 case ILLEGAL_REQUEST:
2043 switch (sshdr.asc) {
2044 case 0x10: /* DIX: Host detected corruption */
2045 good_bytes = sd_completed_bytes(SCpnt);
2047 case 0x20: /* INVALID COMMAND OPCODE */
2048 case 0x24: /* INVALID FIELD IN CDB */
2049 switch (SCpnt->cmnd[0]) {
2051 sd_config_discard(sdkp, SD_LBP_DISABLE);
2055 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2056 sd_config_discard(sdkp, SD_LBP_DISABLE);
2058 sdkp->device->no_write_same = 1;
2059 sd_config_write_same(sdkp);
2060 req->__data_len = blk_rq_bytes(req);
2061 req->rq_flags |= RQF_QUIET;
2072 if (sd_is_zoned(sdkp))
2073 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2075 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2076 "sd_done: completed %d of %d bytes\n",
2077 good_bytes, scsi_bufflen(SCpnt)));
2079 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2080 sd_dif_complete(SCpnt, good_bytes);
2086 * spinup disk - called only in sd_revalidate_disk()
2089 sd_spinup_disk(struct scsi_disk *sdkp)
2091 unsigned char cmd[10];
2092 unsigned long spintime_expire = 0;
2093 int retries, spintime;
2094 unsigned int the_result;
2095 struct scsi_sense_hdr sshdr;
2096 int sense_valid = 0;
2100 /* Spin up drives, as required. Only do this at boot time */
2101 /* Spinup needs to be done for module loads too. */
2106 cmd[0] = TEST_UNIT_READY;
2107 memset((void *) &cmd[1], 0, 9);
2109 the_result = scsi_execute_req(sdkp->device, cmd,
2112 SD_MAX_RETRIES, NULL);
2115 * If the drive has indicated to us that it
2116 * doesn't have any media in it, don't bother
2117 * with any more polling.
2119 if (media_not_present(sdkp, &sshdr))
2123 sense_valid = scsi_sense_valid(&sshdr);
2125 } while (retries < 3 &&
2126 (!scsi_status_is_good(the_result) ||
2127 ((driver_byte(the_result) & DRIVER_SENSE) &&
2128 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2130 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2131 /* no sense, TUR either succeeded or failed
2132 * with a status error */
2133 if(!spintime && !scsi_status_is_good(the_result)) {
2134 sd_print_result(sdkp, "Test Unit Ready failed",
2141 * The device does not want the automatic start to be issued.
2143 if (sdkp->device->no_start_on_add)
2146 if (sense_valid && sshdr.sense_key == NOT_READY) {
2147 if (sshdr.asc == 4 && sshdr.ascq == 3)
2148 break; /* manual intervention required */
2149 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2150 break; /* standby */
2151 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2152 break; /* unavailable */
2154 * Issue command to spin up drive when not ready
2157 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2158 cmd[0] = START_STOP;
2159 cmd[1] = 1; /* Return immediately */
2160 memset((void *) &cmd[2], 0, 8);
2161 cmd[4] = 1; /* Start spin cycle */
2162 if (sdkp->device->start_stop_pwr_cond)
2164 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2166 SD_TIMEOUT, SD_MAX_RETRIES,
2168 spintime_expire = jiffies + 100 * HZ;
2171 /* Wait 1 second for next try */
2176 * Wait for USB flash devices with slow firmware.
2177 * Yes, this sense key/ASC combination shouldn't
2178 * occur here. It's characteristic of these devices.
2180 } else if (sense_valid &&
2181 sshdr.sense_key == UNIT_ATTENTION &&
2182 sshdr.asc == 0x28) {
2184 spintime_expire = jiffies + 5 * HZ;
2187 /* Wait 1 second for next try */
2190 /* we don't understand the sense code, so it's
2191 * probably pointless to loop */
2193 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2194 sd_print_sense_hdr(sdkp, &sshdr);
2199 } while (spintime && time_before_eq(jiffies, spintime_expire));
2202 if (scsi_status_is_good(the_result))
2205 printk("not responding...\n");
2210 * Determine whether disk supports Data Integrity Field.
2212 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2214 struct scsi_device *sdp = sdkp->device;
2218 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2221 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2223 if (type > T10_PI_TYPE3_PROTECTION)
2225 else if (scsi_host_dif_capable(sdp->host, type))
2228 if (sdkp->first_scan || type != sdkp->protection_type)
2231 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2232 " protection type %u. Disabling disk!\n",
2236 sd_printk(KERN_NOTICE, sdkp,
2237 "Enabling DIF Type %u protection\n", type);
2240 sd_printk(KERN_NOTICE, sdkp,
2241 "Disabling DIF Type %u protection\n", type);
2245 sdkp->protection_type = type;
2250 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2251 struct scsi_sense_hdr *sshdr, int sense_valid,
2254 if (driver_byte(the_result) & DRIVER_SENSE)
2255 sd_print_sense_hdr(sdkp, sshdr);
2257 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2260 * Set dirty bit for removable devices if not ready -
2261 * sometimes drives will not report this properly.
2263 if (sdp->removable &&
2264 sense_valid && sshdr->sense_key == NOT_READY)
2265 set_media_not_present(sdkp);
2268 * We used to set media_present to 0 here to indicate no media
2269 * in the drive, but some drives fail read capacity even with
2270 * media present, so we can't do that.
2272 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2276 #if RC16_LEN > SD_BUF_SIZE
2277 #error RC16_LEN must not be more than SD_BUF_SIZE
2280 #define READ_CAPACITY_RETRIES_ON_RESET 10
2283 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2284 * and the reported logical block size is bigger than 512 bytes. Note
2285 * that last_sector is a u64 and therefore logical_to_sectors() is not
2288 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2290 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2292 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2298 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2299 unsigned char *buffer)
2301 unsigned char cmd[16];
2302 struct scsi_sense_hdr sshdr;
2303 int sense_valid = 0;
2305 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2306 unsigned int alignment;
2307 unsigned long long lba;
2308 unsigned sector_size;
2310 if (sdp->no_read_capacity_16)
2315 cmd[0] = SERVICE_ACTION_IN_16;
2316 cmd[1] = SAI_READ_CAPACITY_16;
2318 memset(buffer, 0, RC16_LEN);
2320 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2321 buffer, RC16_LEN, &sshdr,
2322 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2324 if (media_not_present(sdkp, &sshdr))
2328 sense_valid = scsi_sense_valid(&sshdr);
2330 sshdr.sense_key == ILLEGAL_REQUEST &&
2331 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2333 /* Invalid Command Operation Code or
2334 * Invalid Field in CDB, just retry
2335 * silently with RC10 */
2338 sshdr.sense_key == UNIT_ATTENTION &&
2339 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2340 /* Device reset might occur several times,
2341 * give it one more chance */
2342 if (--reset_retries > 0)
2347 } while (the_result && retries);
2350 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2351 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2355 sector_size = get_unaligned_be32(&buffer[8]);
2356 lba = get_unaligned_be64(&buffer[0]);
2358 if (sd_read_protection_type(sdkp, buffer) < 0) {
2363 if (!sd_addressable_capacity(lba, sector_size)) {
2364 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2365 "kernel compiled with support for large block "
2371 /* Logical blocks per physical block exponent */
2372 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2375 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2377 /* Lowest aligned logical block */
2378 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2379 blk_queue_alignment_offset(sdp->request_queue, alignment);
2380 if (alignment && sdkp->first_scan)
2381 sd_printk(KERN_NOTICE, sdkp,
2382 "physical block alignment offset: %u\n", alignment);
2384 if (buffer[14] & 0x80) { /* LBPME */
2387 if (buffer[14] & 0x40) /* LBPRZ */
2390 sd_config_discard(sdkp, SD_LBP_WS16);
2393 sdkp->capacity = lba + 1;
2397 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2398 unsigned char *buffer)
2400 unsigned char cmd[16];
2401 struct scsi_sense_hdr sshdr;
2402 int sense_valid = 0;
2404 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2406 unsigned sector_size;
2409 cmd[0] = READ_CAPACITY;
2410 memset(&cmd[1], 0, 9);
2411 memset(buffer, 0, 8);
2413 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2415 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2417 if (media_not_present(sdkp, &sshdr))
2421 sense_valid = scsi_sense_valid(&sshdr);
2423 sshdr.sense_key == UNIT_ATTENTION &&
2424 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2425 /* Device reset might occur several times,
2426 * give it one more chance */
2427 if (--reset_retries > 0)
2432 } while (the_result && retries);
2435 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2436 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2440 sector_size = get_unaligned_be32(&buffer[4]);
2441 lba = get_unaligned_be32(&buffer[0]);
2443 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2444 /* Some buggy (usb cardreader) devices return an lba of
2445 0xffffffff when the want to report a size of 0 (with
2446 which they really mean no media is present) */
2448 sdkp->physical_block_size = sector_size;
2452 if (!sd_addressable_capacity(lba, sector_size)) {
2453 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2454 "kernel compiled with support for large block "
2460 sdkp->capacity = lba + 1;
2461 sdkp->physical_block_size = sector_size;
2465 static int sd_try_rc16_first(struct scsi_device *sdp)
2467 if (sdp->host->max_cmd_len < 16)
2469 if (sdp->try_rc_10_first)
2471 if (sdp->scsi_level > SCSI_SPC_2)
2473 if (scsi_device_protection(sdp))
2479 * read disk capacity
2482 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2485 struct scsi_device *sdp = sdkp->device;
2487 if (sd_try_rc16_first(sdp)) {
2488 sector_size = read_capacity_16(sdkp, sdp, buffer);
2489 if (sector_size == -EOVERFLOW)
2491 if (sector_size == -ENODEV)
2493 if (sector_size < 0)
2494 sector_size = read_capacity_10(sdkp, sdp, buffer);
2495 if (sector_size < 0)
2498 sector_size = read_capacity_10(sdkp, sdp, buffer);
2499 if (sector_size == -EOVERFLOW)
2501 if (sector_size < 0)
2503 if ((sizeof(sdkp->capacity) > 4) &&
2504 (sdkp->capacity > 0xffffffffULL)) {
2505 int old_sector_size = sector_size;
2506 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2507 "Trying to use READ CAPACITY(16).\n");
2508 sector_size = read_capacity_16(sdkp, sdp, buffer);
2509 if (sector_size < 0) {
2510 sd_printk(KERN_NOTICE, sdkp,
2511 "Using 0xffffffff as device size\n");
2512 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2513 sector_size = old_sector_size;
2519 /* Some devices are known to return the total number of blocks,
2520 * not the highest block number. Some devices have versions
2521 * which do this and others which do not. Some devices we might
2522 * suspect of doing this but we don't know for certain.
2524 * If we know the reported capacity is wrong, decrement it. If
2525 * we can only guess, then assume the number of blocks is even
2526 * (usually true but not always) and err on the side of lowering
2529 if (sdp->fix_capacity ||
2530 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2531 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2532 "from its reported value: %llu\n",
2533 (unsigned long long) sdkp->capacity);
2538 if (sector_size == 0) {
2540 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2544 if (sector_size != 512 &&
2545 sector_size != 1024 &&
2546 sector_size != 2048 &&
2547 sector_size != 4096) {
2548 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2551 * The user might want to re-format the drive with
2552 * a supported sectorsize. Once this happens, it
2553 * would be relatively trivial to set the thing up.
2554 * For this reason, we leave the thing in the table.
2558 * set a bogus sector size so the normal read/write
2559 * logic in the block layer will eventually refuse any
2560 * request on this device without tripping over power
2561 * of two sector size assumptions
2565 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2566 blk_queue_physical_block_size(sdp->request_queue,
2567 sdkp->physical_block_size);
2568 sdkp->device->sector_size = sector_size;
2570 if (sdkp->capacity > 0xffffffff)
2571 sdp->use_16_for_rw = 1;
2576 * Print disk capacity
2579 sd_print_capacity(struct scsi_disk *sdkp,
2580 sector_t old_capacity)
2582 int sector_size = sdkp->device->sector_size;
2583 char cap_str_2[10], cap_str_10[10];
2585 string_get_size(sdkp->capacity, sector_size,
2586 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2587 string_get_size(sdkp->capacity, sector_size,
2588 STRING_UNITS_10, cap_str_10,
2589 sizeof(cap_str_10));
2591 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2592 sd_printk(KERN_NOTICE, sdkp,
2593 "%llu %d-byte logical blocks: (%s/%s)\n",
2594 (unsigned long long)sdkp->capacity,
2595 sector_size, cap_str_10, cap_str_2);
2597 if (sdkp->physical_block_size != sector_size)
2598 sd_printk(KERN_NOTICE, sdkp,
2599 "%u-byte physical blocks\n",
2600 sdkp->physical_block_size);
2602 sd_zbc_print_zones(sdkp);
2606 /* called with buffer of length 512 */
2608 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2609 unsigned char *buffer, int len, struct scsi_mode_data *data,
2610 struct scsi_sense_hdr *sshdr)
2612 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2613 SD_TIMEOUT, SD_MAX_RETRIES, data,
2618 * read write protect setting, if possible - called only in sd_revalidate_disk()
2619 * called with buffer of length SD_BUF_SIZE
2622 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2625 struct scsi_device *sdp = sdkp->device;
2626 struct scsi_mode_data data;
2627 int old_wp = sdkp->write_prot;
2629 set_disk_ro(sdkp->disk, 0);
2630 if (sdp->skip_ms_page_3f) {
2631 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2635 if (sdp->use_192_bytes_for_3f) {
2636 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2639 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2640 * We have to start carefully: some devices hang if we ask
2641 * for more than is available.
2643 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2646 * Second attempt: ask for page 0 When only page 0 is
2647 * implemented, a request for page 3F may return Sense Key
2648 * 5: Illegal Request, Sense Code 24: Invalid field in
2651 if (!scsi_status_is_good(res))
2652 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2655 * Third attempt: ask 255 bytes, as we did earlier.
2657 if (!scsi_status_is_good(res))
2658 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2662 if (!scsi_status_is_good(res)) {
2663 sd_first_printk(KERN_WARNING, sdkp,
2664 "Test WP failed, assume Write Enabled\n");
2666 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2667 set_disk_ro(sdkp->disk, sdkp->write_prot);
2668 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2669 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2670 sdkp->write_prot ? "on" : "off");
2671 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2677 * sd_read_cache_type - called only from sd_revalidate_disk()
2678 * called with buffer of length SD_BUF_SIZE
2681 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2684 struct scsi_device *sdp = sdkp->device;
2689 struct scsi_mode_data data;
2690 struct scsi_sense_hdr sshdr;
2691 int old_wce = sdkp->WCE;
2692 int old_rcd = sdkp->RCD;
2693 int old_dpofua = sdkp->DPOFUA;
2696 if (sdkp->cache_override)
2700 if (sdp->skip_ms_page_8) {
2701 if (sdp->type == TYPE_RBC)
2704 if (sdp->skip_ms_page_3f)
2707 if (sdp->use_192_bytes_for_3f)
2711 } else if (sdp->type == TYPE_RBC) {
2719 /* cautiously ask */
2720 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2723 if (!scsi_status_is_good(res))
2726 if (!data.header_length) {
2729 sd_first_printk(KERN_ERR, sdkp,
2730 "Missing header in MODE_SENSE response\n");
2733 /* that went OK, now ask for the proper length */
2737 * We're only interested in the first three bytes, actually.
2738 * But the data cache page is defined for the first 20.
2742 else if (len > SD_BUF_SIZE) {
2743 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2744 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2747 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2751 if (len > first_len)
2752 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2755 if (scsi_status_is_good(res)) {
2756 int offset = data.header_length + data.block_descriptor_length;
2758 while (offset < len) {
2759 u8 page_code = buffer[offset] & 0x3F;
2760 u8 spf = buffer[offset] & 0x40;
2762 if (page_code == 8 || page_code == 6) {
2763 /* We're interested only in the first 3 bytes.
2765 if (len - offset <= 2) {
2766 sd_first_printk(KERN_ERR, sdkp,
2767 "Incomplete mode parameter "
2771 modepage = page_code;
2775 /* Go to the next page */
2776 if (spf && len - offset > 3)
2777 offset += 4 + (buffer[offset+2] << 8) +
2779 else if (!spf && len - offset > 1)
2780 offset += 2 + buffer[offset+1];
2782 sd_first_printk(KERN_ERR, sdkp,
2784 "parameter data\n");
2790 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2794 if (modepage == 8) {
2795 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2796 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2798 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2802 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2803 if (sdp->broken_fua) {
2804 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2806 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2807 !sdkp->device->use_16_for_rw) {
2808 sd_first_printk(KERN_NOTICE, sdkp,
2809 "Uses READ/WRITE(6), disabling FUA\n");
2813 /* No cache flush allowed for write protected devices */
2814 if (sdkp->WCE && sdkp->write_prot)
2817 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2818 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2819 sd_printk(KERN_NOTICE, sdkp,
2820 "Write cache: %s, read cache: %s, %s\n",
2821 sdkp->WCE ? "enabled" : "disabled",
2822 sdkp->RCD ? "disabled" : "enabled",
2823 sdkp->DPOFUA ? "supports DPO and FUA"
2824 : "doesn't support DPO or FUA");
2830 if (scsi_sense_valid(&sshdr) &&
2831 sshdr.sense_key == ILLEGAL_REQUEST &&
2832 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2833 /* Invalid field in CDB */
2834 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2836 sd_first_printk(KERN_ERR, sdkp,
2837 "Asking for cache data failed\n");
2840 if (sdp->wce_default_on) {
2841 sd_first_printk(KERN_NOTICE, sdkp,
2842 "Assuming drive cache: write back\n");
2845 sd_first_printk(KERN_ERR, sdkp,
2846 "Assuming drive cache: write through\n");
2854 * The ATO bit indicates whether the DIF application tag is available
2855 * for use by the operating system.
2857 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2860 struct scsi_device *sdp = sdkp->device;
2861 struct scsi_mode_data data;
2862 struct scsi_sense_hdr sshdr;
2864 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2867 if (sdkp->protection_type == 0)
2870 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2871 SD_MAX_RETRIES, &data, &sshdr);
2873 if (!scsi_status_is_good(res) || !data.header_length ||
2875 sd_first_printk(KERN_WARNING, sdkp,
2876 "getting Control mode page failed, assume no ATO\n");
2878 if (scsi_sense_valid(&sshdr))
2879 sd_print_sense_hdr(sdkp, &sshdr);
2884 offset = data.header_length + data.block_descriptor_length;
2886 if ((buffer[offset] & 0x3f) != 0x0a) {
2887 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2891 if ((buffer[offset + 5] & 0x80) == 0)
2900 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2901 * @sdkp: disk to query
2903 static void sd_read_block_limits(struct scsi_disk *sdkp)
2905 unsigned int sector_sz = sdkp->device->sector_size;
2906 const int vpd_len = 64;
2907 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2910 /* Block Limits VPD */
2911 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2914 blk_queue_io_min(sdkp->disk->queue,
2915 get_unaligned_be16(&buffer[6]) * sector_sz);
2917 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2918 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2920 if (buffer[3] == 0x3c) {
2921 unsigned int lba_count, desc_count;
2923 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2928 lba_count = get_unaligned_be32(&buffer[20]);
2929 desc_count = get_unaligned_be32(&buffer[24]);
2931 if (lba_count && desc_count)
2932 sdkp->max_unmap_blocks = lba_count;
2934 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2936 if (buffer[32] & 0x80)
2937 sdkp->unmap_alignment =
2938 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2940 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2942 if (sdkp->max_unmap_blocks)
2943 sd_config_discard(sdkp, SD_LBP_UNMAP);
2945 sd_config_discard(sdkp, SD_LBP_WS16);
2947 } else { /* LBP VPD page tells us what to use */
2948 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2949 sd_config_discard(sdkp, SD_LBP_UNMAP);
2950 else if (sdkp->lbpws)
2951 sd_config_discard(sdkp, SD_LBP_WS16);
2952 else if (sdkp->lbpws10)
2953 sd_config_discard(sdkp, SD_LBP_WS10);
2955 sd_config_discard(sdkp, SD_LBP_DISABLE);
2964 * sd_read_block_characteristics - Query block dev. characteristics
2965 * @sdkp: disk to query
2967 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2969 struct request_queue *q = sdkp->disk->queue;
2970 unsigned char *buffer;
2972 const int vpd_len = 64;
2974 buffer = kmalloc(vpd_len, GFP_KERNEL);
2977 /* Block Device Characteristics VPD */
2978 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2981 rot = get_unaligned_be16(&buffer[4]);
2984 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2985 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2988 if (sdkp->device->type == TYPE_ZBC) {
2990 q->limits.zoned = BLK_ZONED_HM;
2992 sdkp->zoned = (buffer[8] >> 4) & 3;
2993 if (sdkp->zoned == 1)
2995 q->limits.zoned = BLK_ZONED_HA;
2998 * Treat drive-managed devices as
2999 * regular block devices.
3001 q->limits.zoned = BLK_ZONED_NONE;
3003 if (blk_queue_is_zoned(q) && sdkp->first_scan)
3004 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3005 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3012 * sd_read_block_provisioning - Query provisioning VPD page
3013 * @sdkp: disk to query
3015 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3017 unsigned char *buffer;
3018 const int vpd_len = 8;
3020 if (sdkp->lbpme == 0)
3023 buffer = kmalloc(vpd_len, GFP_KERNEL);
3025 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3029 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3030 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3031 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3037 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3039 struct scsi_device *sdev = sdkp->device;
3041 if (sdev->host->no_write_same) {
3042 sdev->no_write_same = 1;
3047 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3048 /* too large values might cause issues with arcmsr */
3049 int vpd_buf_len = 64;
3051 sdev->no_report_opcodes = 1;
3053 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3054 * CODES is unsupported and the device has an ATA
3055 * Information VPD page (SAT).
3057 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3058 sdev->no_write_same = 1;
3061 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3064 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3068 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3070 struct scsi_device *sdev = sdkp->device;
3072 if (!sdev->security_supported)
3075 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3076 SECURITY_PROTOCOL_IN) == 1 &&
3077 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3078 SECURITY_PROTOCOL_OUT) == 1)
3083 * sd_revalidate_disk - called the first time a new disk is seen,
3084 * performs disk spin up, read_capacity, etc.
3085 * @disk: struct gendisk we care about
3087 static int sd_revalidate_disk(struct gendisk *disk)
3089 struct scsi_disk *sdkp = scsi_disk(disk);
3090 struct scsi_device *sdp = sdkp->device;
3091 struct request_queue *q = sdkp->disk->queue;
3092 sector_t old_capacity = sdkp->capacity;
3093 unsigned char *buffer;
3094 unsigned int dev_max, rw_max;
3096 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3097 "sd_revalidate_disk\n"));
3100 * If the device is offline, don't try and read capacity or any
3101 * of the other niceties.
3103 if (!scsi_device_online(sdp))
3106 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3108 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3109 "allocation failure.\n");
3113 sd_spinup_disk(sdkp);
3116 * Without media there is no reason to ask; moreover, some devices
3117 * react badly if we do.
3119 if (sdkp->media_present) {
3120 sd_read_capacity(sdkp, buffer);
3122 if (scsi_device_supports_vpd(sdp)) {
3123 sd_read_block_provisioning(sdkp);
3124 sd_read_block_limits(sdkp);
3125 sd_read_block_characteristics(sdkp);
3126 sd_zbc_read_zones(sdkp, buffer);
3129 sd_print_capacity(sdkp, old_capacity);
3131 sd_read_write_protect_flag(sdkp, buffer);
3132 sd_read_cache_type(sdkp, buffer);
3133 sd_read_app_tag_own(sdkp, buffer);
3134 sd_read_write_same(sdkp, buffer);
3135 sd_read_security(sdkp, buffer);
3139 * We now have all cache related info, determine how we deal
3140 * with flush requests.
3142 sd_set_flush_flag(sdkp);
3144 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3145 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3147 /* Some devices report a maximum block count for READ/WRITE requests. */
3148 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3149 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3152 * Determine the device's preferred I/O size for reads and writes
3153 * unless the reported value is unreasonably small, large, or
3156 if (sdkp->opt_xfer_blocks &&
3157 sdkp->opt_xfer_blocks <= dev_max &&
3158 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3159 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3160 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3161 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3163 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3164 (sector_t)BLK_DEF_MAX_SECTORS);
3166 /* Do not exceed controller limit */
3167 rw_max = min(rw_max, queue_max_hw_sectors(q));
3170 * Only update max_sectors if previously unset or if the current value
3171 * exceeds the capabilities of the hardware.
3173 if (sdkp->first_scan ||
3174 q->limits.max_sectors > q->limits.max_dev_sectors ||
3175 q->limits.max_sectors > q->limits.max_hw_sectors)
3176 q->limits.max_sectors = rw_max;
3178 sdkp->first_scan = 0;
3180 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3181 sd_config_write_same(sdkp);
3189 * sd_unlock_native_capacity - unlock native capacity
3190 * @disk: struct gendisk to set capacity for
3192 * Block layer calls this function if it detects that partitions
3193 * on @disk reach beyond the end of the device. If the SCSI host
3194 * implements ->unlock_native_capacity() method, it's invoked to
3195 * give it a chance to adjust the device capacity.
3198 * Defined by block layer. Might sleep.
3200 static void sd_unlock_native_capacity(struct gendisk *disk)
3202 struct scsi_device *sdev = scsi_disk(disk)->device;
3204 if (sdev->host->hostt->unlock_native_capacity)
3205 sdev->host->hostt->unlock_native_capacity(sdev);
3209 * sd_format_disk_name - format disk name
3210 * @prefix: name prefix - ie. "sd" for SCSI disks
3211 * @index: index of the disk to format name for
3212 * @buf: output buffer
3213 * @buflen: length of the output buffer
3215 * SCSI disk names starts at sda. The 26th device is sdz and the
3216 * 27th is sdaa. The last one for two lettered suffix is sdzz
3217 * which is followed by sdaaa.
3219 * This is basically 26 base counting with one extra 'nil' entry
3220 * at the beginning from the second digit on and can be
3221 * determined using similar method as 26 base conversion with the
3222 * index shifted -1 after each digit is computed.
3228 * 0 on success, -errno on failure.
3230 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3232 const int base = 'z' - 'a' + 1;
3233 char *begin = buf + strlen(prefix);
3234 char *end = buf + buflen;
3244 *--p = 'a' + (index % unit);
3245 index = (index / unit) - 1;
3246 } while (index >= 0);
3248 memmove(begin, p, end - p);
3249 memcpy(buf, prefix, strlen(prefix));
3255 * The asynchronous part of sd_probe
3257 static void sd_probe_async(void *data, async_cookie_t cookie)
3259 struct scsi_disk *sdkp = data;
3260 struct scsi_device *sdp;
3267 index = sdkp->index;
3268 dev = &sdp->sdev_gendev;
3270 gd->major = sd_major((index & 0xf0) >> 4);
3271 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3273 gd->fops = &sd_fops;
3274 gd->private_data = &sdkp->driver;
3275 gd->queue = sdkp->device->request_queue;
3277 /* defaults, until the device tells us otherwise */
3278 sdp->sector_size = 512;
3280 sdkp->media_present = 1;
3281 sdkp->write_prot = 0;
3282 sdkp->cache_override = 0;
3286 sdkp->first_scan = 1;
3287 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3289 sd_revalidate_disk(gd);
3291 gd->flags = GENHD_FL_EXT_DEVT;
3292 if (sdp->removable) {
3293 gd->flags |= GENHD_FL_REMOVABLE;
3294 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3297 blk_pm_runtime_init(sdp->request_queue, dev);
3298 device_add_disk(dev, gd);
3300 sd_dif_config_host(sdkp);
3302 sd_revalidate_disk(gd);
3304 if (sdkp->security) {
3305 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3307 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3310 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3311 sdp->removable ? "removable " : "");
3312 scsi_autopm_put_device(sdp);
3313 put_device(&sdkp->dev);
3317 * sd_probe - called during driver initialization and whenever a
3318 * new scsi device is attached to the system. It is called once
3319 * for each scsi device (not just disks) present.
3320 * @dev: pointer to device object
3322 * Returns 0 if successful (or not interested in this scsi device
3323 * (e.g. scanner)); 1 when there is an error.
3325 * Note: this function is invoked from the scsi mid-level.
3326 * This function sets up the mapping between a given
3327 * <host,channel,id,lun> (found in sdp) and new device name
3328 * (e.g. /dev/sda). More precisely it is the block device major
3329 * and minor number that is chosen here.
3331 * Assume sd_probe is not re-entrant (for time being)
3332 * Also think about sd_probe() and sd_remove() running coincidentally.
3334 static int sd_probe(struct device *dev)
3336 struct scsi_device *sdp = to_scsi_device(dev);
3337 struct scsi_disk *sdkp;
3342 scsi_autopm_get_device(sdp);
3344 if (sdp->type != TYPE_DISK &&
3345 sdp->type != TYPE_ZBC &&
3346 sdp->type != TYPE_MOD &&
3347 sdp->type != TYPE_RBC)
3350 #ifndef CONFIG_BLK_DEV_ZONED
3351 if (sdp->type == TYPE_ZBC)
3354 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3358 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3362 gd = alloc_disk(SD_MINORS);
3367 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3370 spin_lock(&sd_index_lock);
3371 error = ida_get_new(&sd_index_ida, &index);
3372 spin_unlock(&sd_index_lock);
3373 } while (error == -EAGAIN);
3376 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3380 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3382 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3383 goto out_free_index;
3387 sdkp->driver = &sd_template;
3389 sdkp->index = index;
3390 atomic_set(&sdkp->openers, 0);
3391 atomic_set(&sdkp->device->ioerr_cnt, 0);
3393 if (!sdp->request_queue->rq_timeout) {
3394 if (sdp->type != TYPE_MOD)
3395 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3397 blk_queue_rq_timeout(sdp->request_queue,
3401 device_initialize(&sdkp->dev);
3402 sdkp->dev.parent = dev;
3403 sdkp->dev.class = &sd_disk_class;
3404 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3406 error = device_add(&sdkp->dev);
3408 goto out_free_index;
3411 dev_set_drvdata(dev, sdkp);
3413 get_device(&sdkp->dev); /* prevent release before async_schedule */
3414 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3419 spin_lock(&sd_index_lock);
3420 ida_remove(&sd_index_ida, index);
3421 spin_unlock(&sd_index_lock);
3427 scsi_autopm_put_device(sdp);
3432 * sd_remove - called whenever a scsi disk (previously recognized by
3433 * sd_probe) is detached from the system. It is called (potentially
3434 * multiple times) during sd module unload.
3435 * @dev: pointer to device object
3437 * Note: this function is invoked from the scsi mid-level.
3438 * This function potentially frees up a device name (e.g. /dev/sdc)
3439 * that could be re-used by a subsequent sd_probe().
3440 * This function is not called when the built-in sd driver is "exit-ed".
3442 static int sd_remove(struct device *dev)
3444 struct scsi_disk *sdkp;
3447 sdkp = dev_get_drvdata(dev);
3448 devt = disk_devt(sdkp->disk);
3449 scsi_autopm_get_device(sdkp->device);
3451 async_synchronize_full_domain(&scsi_sd_pm_domain);
3452 async_synchronize_full_domain(&scsi_sd_probe_domain);
3453 device_del(&sdkp->dev);
3454 del_gendisk(sdkp->disk);
3457 sd_zbc_remove(sdkp);
3459 free_opal_dev(sdkp->opal_dev);
3461 blk_register_region(devt, SD_MINORS, NULL,
3462 sd_default_probe, NULL, NULL);
3464 mutex_lock(&sd_ref_mutex);
3465 dev_set_drvdata(dev, NULL);
3466 put_device(&sdkp->dev);
3467 mutex_unlock(&sd_ref_mutex);
3473 * scsi_disk_release - Called to free the scsi_disk structure
3474 * @dev: pointer to embedded class device
3476 * sd_ref_mutex must be held entering this routine. Because it is
3477 * called on last put, you should always use the scsi_disk_get()
3478 * scsi_disk_put() helpers which manipulate the semaphore directly
3479 * and never do a direct put_device.
3481 static void scsi_disk_release(struct device *dev)
3483 struct scsi_disk *sdkp = to_scsi_disk(dev);
3484 struct gendisk *disk = sdkp->disk;
3486 spin_lock(&sd_index_lock);
3487 ida_remove(&sd_index_ida, sdkp->index);
3488 spin_unlock(&sd_index_lock);
3490 disk->private_data = NULL;
3492 put_device(&sdkp->device->sdev_gendev);
3497 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3499 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3500 struct scsi_sense_hdr sshdr;
3501 struct scsi_device *sdp = sdkp->device;
3505 cmd[4] |= 1; /* START */
3507 if (sdp->start_stop_pwr_cond)
3508 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3510 if (!scsi_device_online(sdp))
3513 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3514 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3516 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3517 if (driver_byte(res) & DRIVER_SENSE)
3518 sd_print_sense_hdr(sdkp, &sshdr);
3519 if (scsi_sense_valid(&sshdr) &&
3520 /* 0x3a is medium not present */
3525 /* SCSI error codes must not go to the generic layer */
3533 * Send a SYNCHRONIZE CACHE instruction down to the device through
3534 * the normal SCSI command structure. Wait for the command to
3537 static void sd_shutdown(struct device *dev)
3539 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3542 return; /* this can happen */
3544 if (pm_runtime_suspended(dev))
3547 if (sdkp->WCE && sdkp->media_present) {
3548 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3549 sd_sync_cache(sdkp, NULL);
3552 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3553 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3554 sd_start_stop_device(sdkp, 0);
3558 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3560 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3561 struct scsi_sense_hdr sshdr;
3564 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3567 if (sdkp->WCE && sdkp->media_present) {
3568 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3569 ret = sd_sync_cache(sdkp, &sshdr);
3572 /* ignore OFFLINE device */
3576 if (!scsi_sense_valid(&sshdr) ||
3577 sshdr.sense_key != ILLEGAL_REQUEST)
3581 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3582 * doesn't support sync. There's not much to do and
3583 * suspend shouldn't fail.
3589 if (sdkp->device->manage_start_stop) {
3590 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3591 /* an error is not worth aborting a system sleep */
3592 ret = sd_start_stop_device(sdkp, 0);
3593 if (ignore_stop_errors)
3600 static int sd_suspend_system(struct device *dev)
3602 return sd_suspend_common(dev, true);
3605 static int sd_suspend_runtime(struct device *dev)
3607 return sd_suspend_common(dev, false);
3610 static int sd_resume(struct device *dev)
3612 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3615 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3618 if (!sdkp->device->manage_start_stop)
3621 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3622 ret = sd_start_stop_device(sdkp, 1);
3624 opal_unlock_from_suspend(sdkp->opal_dev);
3629 * init_sd - entry point for this driver (both when built in or when
3632 * Note: this function registers this driver with the scsi mid-level.
3634 static int __init init_sd(void)
3636 int majors = 0, i, err;
3638 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3640 for (i = 0; i < SD_MAJORS; i++) {
3641 if (register_blkdev(sd_major(i), "sd") != 0)
3644 blk_register_region(sd_major(i), SD_MINORS, NULL,
3645 sd_default_probe, NULL, NULL);
3651 err = class_register(&sd_disk_class);
3655 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3657 if (!sd_cdb_cache) {
3658 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3663 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3665 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3670 err = scsi_register_driver(&sd_template.gendrv);
3672 goto err_out_driver;
3677 mempool_destroy(sd_cdb_pool);
3680 kmem_cache_destroy(sd_cdb_cache);
3683 class_unregister(&sd_disk_class);
3685 for (i = 0; i < SD_MAJORS; i++)
3686 unregister_blkdev(sd_major(i), "sd");
3691 * exit_sd - exit point for this driver (when it is a module).
3693 * Note: this function unregisters this driver from the scsi mid-level.
3695 static void __exit exit_sd(void)
3699 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3701 scsi_unregister_driver(&sd_template.gendrv);
3702 mempool_destroy(sd_cdb_pool);
3703 kmem_cache_destroy(sd_cdb_cache);
3705 class_unregister(&sd_disk_class);
3707 for (i = 0; i < SD_MAJORS; i++) {
3708 blk_unregister_region(sd_major(i), SD_MINORS);
3709 unregister_blkdev(sd_major(i), "sd");
3713 module_init(init_sd);
3714 module_exit(exit_sd);
3716 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3717 struct scsi_sense_hdr *sshdr)
3719 scsi_print_sense_hdr(sdkp->device,
3720 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3723 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3726 const char *hb_string = scsi_hostbyte_string(result);
3727 const char *db_string = scsi_driverbyte_string(result);
3729 if (hb_string || db_string)
3730 sd_printk(KERN_INFO, sdkp,
3731 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3732 hb_string ? hb_string : "invalid",
3733 db_string ? db_string : "invalid");
3735 sd_printk(KERN_INFO, sdkp,
3736 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3737 msg, host_byte(result), driver_byte(result));