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:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
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/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/mutex.h>
51 #include <linux/string_helpers.h>
52 #include <linux/async.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
106 static void sd_config_discard(struct scsi_disk *, unsigned int);
107 static void sd_config_write_same(struct scsi_disk *);
108 static int sd_revalidate_disk(struct gendisk *);
109 static void sd_unlock_native_capacity(struct gendisk *disk);
110 static int sd_probe(struct device *);
111 static int sd_remove(struct device *);
112 static void sd_shutdown(struct device *);
113 static int sd_suspend_system(struct device *);
114 static int sd_suspend_runtime(struct device *);
115 static int sd_resume(struct device *);
116 static void sd_rescan(struct device *);
117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
119 static int sd_done(struct scsi_cmnd *);
120 static void sd_eh_reset(struct scsi_cmnd *);
121 static int sd_eh_action(struct scsi_cmnd *, int);
122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
123 static void scsi_disk_release(struct device *cdev);
124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
125 static void sd_print_result(const struct scsi_disk *, const char *, int);
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;
136 static mempool_t *sd_page_pool;
138 static const char *sd_cache_types[] = {
139 "write through", "none", "write back",
140 "write back, no read (daft)"
143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
145 bool wc = false, fua = false;
153 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
157 cache_type_store(struct device *dev, struct device_attribute *attr,
158 const char *buf, size_t count)
160 int ct, rcd, wce, sp;
161 struct scsi_disk *sdkp = to_scsi_disk(dev);
162 struct scsi_device *sdp = sdkp->device;
165 struct scsi_mode_data data;
166 struct scsi_sense_hdr sshdr;
167 static const char temp[] = "temporary ";
170 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
171 /* no cache control on RBC devices; theoretically they
172 * can do it, but there's probably so many exceptions
173 * it's not worth the risk */
176 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
177 buf += sizeof(temp) - 1;
178 sdkp->cache_override = 1;
180 sdkp->cache_override = 0;
183 ct = sysfs_match_string(sd_cache_types, buf);
187 rcd = ct & 0x01 ? 1 : 0;
188 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 if (sdkp->cache_override) {
193 sd_set_flush_flag(sdkp);
197 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198 SD_MAX_RETRIES, &data, NULL))
200 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201 data.block_descriptor_length);
202 buffer_data = buffer + data.header_length +
203 data.block_descriptor_length;
204 buffer_data[2] &= ~0x05;
205 buffer_data[2] |= wce << 2 | rcd;
206 sp = buffer_data[0] & 0x80 ? 1 : 0;
207 buffer_data[0] &= ~0x80;
209 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
210 SD_MAX_RETRIES, &data, &sshdr)) {
211 if (scsi_sense_valid(&sshdr))
212 sd_print_sense_hdr(sdkp, &sshdr);
215 revalidate_disk(sdkp->disk);
220 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
223 struct scsi_disk *sdkp = to_scsi_disk(dev);
224 struct scsi_device *sdp = sdkp->device;
226 return sprintf(buf, "%u\n", sdp->manage_start_stop);
230 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
231 const char *buf, size_t count)
233 struct scsi_disk *sdkp = to_scsi_disk(dev);
234 struct scsi_device *sdp = sdkp->device;
237 if (!capable(CAP_SYS_ADMIN))
240 if (kstrtobool(buf, &v))
243 sdp->manage_start_stop = v;
247 static DEVICE_ATTR_RW(manage_start_stop);
250 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
252 struct scsi_disk *sdkp = to_scsi_disk(dev);
254 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
258 allow_restart_store(struct device *dev, struct device_attribute *attr,
259 const char *buf, size_t count)
262 struct scsi_disk *sdkp = to_scsi_disk(dev);
263 struct scsi_device *sdp = sdkp->device;
265 if (!capable(CAP_SYS_ADMIN))
268 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
271 if (kstrtobool(buf, &v))
274 sdp->allow_restart = v;
278 static DEVICE_ATTR_RW(allow_restart);
281 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
283 struct scsi_disk *sdkp = to_scsi_disk(dev);
284 int ct = sdkp->RCD + 2*sdkp->WCE;
286 return sprintf(buf, "%s\n", sd_cache_types[ct]);
288 static DEVICE_ATTR_RW(cache_type);
291 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
293 struct scsi_disk *sdkp = to_scsi_disk(dev);
295 return sprintf(buf, "%u\n", sdkp->DPOFUA);
297 static DEVICE_ATTR_RO(FUA);
300 protection_type_show(struct device *dev, struct device_attribute *attr,
303 struct scsi_disk *sdkp = to_scsi_disk(dev);
305 return sprintf(buf, "%u\n", sdkp->protection_type);
309 protection_type_store(struct device *dev, struct device_attribute *attr,
310 const char *buf, size_t count)
312 struct scsi_disk *sdkp = to_scsi_disk(dev);
316 if (!capable(CAP_SYS_ADMIN))
319 err = kstrtouint(buf, 10, &val);
324 if (val <= T10_PI_TYPE3_PROTECTION)
325 sdkp->protection_type = val;
329 static DEVICE_ATTR_RW(protection_type);
332 protection_mode_show(struct device *dev, struct device_attribute *attr,
335 struct scsi_disk *sdkp = to_scsi_disk(dev);
336 struct scsi_device *sdp = sdkp->device;
337 unsigned int dif, dix;
339 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
340 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
342 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
348 return sprintf(buf, "none\n");
350 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
352 static DEVICE_ATTR_RO(protection_mode);
355 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
357 struct scsi_disk *sdkp = to_scsi_disk(dev);
359 return sprintf(buf, "%u\n", sdkp->ATO);
361 static DEVICE_ATTR_RO(app_tag_own);
364 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
367 struct scsi_disk *sdkp = to_scsi_disk(dev);
369 return sprintf(buf, "%u\n", sdkp->lbpme);
371 static DEVICE_ATTR_RO(thin_provisioning);
373 /* sysfs_match_string() requires dense arrays */
374 static const char *lbp_mode[] = {
375 [SD_LBP_FULL] = "full",
376 [SD_LBP_UNMAP] = "unmap",
377 [SD_LBP_WS16] = "writesame_16",
378 [SD_LBP_WS10] = "writesame_10",
379 [SD_LBP_ZERO] = "writesame_zero",
380 [SD_LBP_DISABLE] = "disabled",
384 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
387 struct scsi_disk *sdkp = to_scsi_disk(dev);
389 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
393 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
394 const char *buf, size_t count)
396 struct scsi_disk *sdkp = to_scsi_disk(dev);
397 struct scsi_device *sdp = sdkp->device;
400 if (!capable(CAP_SYS_ADMIN))
403 if (sd_is_zoned(sdkp)) {
404 sd_config_discard(sdkp, SD_LBP_DISABLE);
408 if (sdp->type != TYPE_DISK)
411 mode = sysfs_match_string(lbp_mode, buf);
415 sd_config_discard(sdkp, mode);
419 static DEVICE_ATTR_RW(provisioning_mode);
421 /* sysfs_match_string() requires dense arrays */
422 static const char *zeroing_mode[] = {
423 [SD_ZERO_WRITE] = "write",
424 [SD_ZERO_WS] = "writesame",
425 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
426 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
430 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
433 struct scsi_disk *sdkp = to_scsi_disk(dev);
435 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
439 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
440 const char *buf, size_t count)
442 struct scsi_disk *sdkp = to_scsi_disk(dev);
445 if (!capable(CAP_SYS_ADMIN))
448 mode = sysfs_match_string(zeroing_mode, buf);
452 sdkp->zeroing_mode = mode;
456 static DEVICE_ATTR_RW(zeroing_mode);
459 max_medium_access_timeouts_show(struct device *dev,
460 struct device_attribute *attr, char *buf)
462 struct scsi_disk *sdkp = to_scsi_disk(dev);
464 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
468 max_medium_access_timeouts_store(struct device *dev,
469 struct device_attribute *attr, const char *buf,
472 struct scsi_disk *sdkp = to_scsi_disk(dev);
475 if (!capable(CAP_SYS_ADMIN))
478 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
480 return err ? err : count;
482 static DEVICE_ATTR_RW(max_medium_access_timeouts);
485 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
488 struct scsi_disk *sdkp = to_scsi_disk(dev);
490 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
494 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
495 const char *buf, size_t count)
497 struct scsi_disk *sdkp = to_scsi_disk(dev);
498 struct scsi_device *sdp = sdkp->device;
502 if (!capable(CAP_SYS_ADMIN))
505 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
508 err = kstrtoul(buf, 10, &max);
514 sdp->no_write_same = 1;
515 else if (max <= SD_MAX_WS16_BLOCKS) {
516 sdp->no_write_same = 0;
517 sdkp->max_ws_blocks = max;
520 sd_config_write_same(sdkp);
524 static DEVICE_ATTR_RW(max_write_same_blocks);
526 static struct attribute *sd_disk_attrs[] = {
527 &dev_attr_cache_type.attr,
529 &dev_attr_allow_restart.attr,
530 &dev_attr_manage_start_stop.attr,
531 &dev_attr_protection_type.attr,
532 &dev_attr_protection_mode.attr,
533 &dev_attr_app_tag_own.attr,
534 &dev_attr_thin_provisioning.attr,
535 &dev_attr_provisioning_mode.attr,
536 &dev_attr_zeroing_mode.attr,
537 &dev_attr_max_write_same_blocks.attr,
538 &dev_attr_max_medium_access_timeouts.attr,
541 ATTRIBUTE_GROUPS(sd_disk);
543 static struct class sd_disk_class = {
545 .owner = THIS_MODULE,
546 .dev_release = scsi_disk_release,
547 .dev_groups = sd_disk_groups,
550 static const struct dev_pm_ops sd_pm_ops = {
551 .suspend = sd_suspend_system,
553 .poweroff = sd_suspend_system,
554 .restore = sd_resume,
555 .runtime_suspend = sd_suspend_runtime,
556 .runtime_resume = sd_resume,
559 static struct scsi_driver sd_template = {
562 .owner = THIS_MODULE,
565 .shutdown = sd_shutdown,
569 .init_command = sd_init_command,
570 .uninit_command = sd_uninit_command,
572 .eh_action = sd_eh_action,
573 .eh_reset = sd_eh_reset,
577 * Dummy kobj_map->probe function.
578 * The default ->probe function will call modprobe, which is
579 * pointless as this module is already loaded.
581 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
587 * Device no to disk mapping:
589 * major disc2 disc p1
590 * |............|.............|....|....| <- dev_t
593 * Inside a major, we have 16k disks, however mapped non-
594 * contiguously. The first 16 disks are for major0, the next
595 * ones with major1, ... Disk 256 is for major0 again, disk 272
597 * As we stay compatible with our numbering scheme, we can reuse
598 * the well-know SCSI majors 8, 65--71, 136--143.
600 static int sd_major(int major_idx)
604 return SCSI_DISK0_MAJOR;
606 return SCSI_DISK1_MAJOR + major_idx - 1;
608 return SCSI_DISK8_MAJOR + major_idx - 8;
611 return 0; /* shut up gcc */
615 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
617 struct scsi_disk *sdkp = NULL;
619 mutex_lock(&sd_ref_mutex);
621 if (disk->private_data) {
622 sdkp = scsi_disk(disk);
623 if (scsi_device_get(sdkp->device) == 0)
624 get_device(&sdkp->dev);
628 mutex_unlock(&sd_ref_mutex);
632 static void scsi_disk_put(struct scsi_disk *sdkp)
634 struct scsi_device *sdev = sdkp->device;
636 mutex_lock(&sd_ref_mutex);
637 put_device(&sdkp->dev);
638 scsi_device_put(sdev);
639 mutex_unlock(&sd_ref_mutex);
642 #ifdef CONFIG_BLK_SED_OPAL
643 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
644 size_t len, bool send)
646 struct scsi_device *sdev = data;
650 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
652 put_unaligned_be16(spsp, &cdb[2]);
653 put_unaligned_be32(len, &cdb[6]);
655 ret = scsi_execute_req(sdev, cdb,
656 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
657 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
658 return ret <= 0 ? ret : -EIO;
660 #endif /* CONFIG_BLK_SED_OPAL */
662 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
663 unsigned int dix, unsigned int dif)
665 struct bio *bio = scmd->request->bio;
666 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
667 unsigned int protect = 0;
669 if (dix) { /* DIX Type 0, 1, 2, 3 */
670 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
671 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
673 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
674 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
677 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
678 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
680 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
681 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
684 if (dif) { /* DIX/DIF Type 1, 2, 3 */
685 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
687 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
688 protect = 3 << 5; /* Disable target PI checking */
690 protect = 1 << 5; /* Enable target PI checking */
693 scsi_set_prot_op(scmd, prot_op);
694 scsi_set_prot_type(scmd, dif);
695 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
700 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
702 struct request_queue *q = sdkp->disk->queue;
703 unsigned int logical_block_size = sdkp->device->sector_size;
704 unsigned int max_blocks = 0;
706 q->limits.discard_alignment =
707 sdkp->unmap_alignment * logical_block_size;
708 q->limits.discard_granularity =
709 max(sdkp->physical_block_size,
710 sdkp->unmap_granularity * logical_block_size);
711 sdkp->provisioning_mode = mode;
717 blk_queue_max_discard_sectors(q, 0);
718 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
722 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
723 (u32)SD_MAX_WS16_BLOCKS);
727 if (sdkp->device->unmap_limit_for_ws)
728 max_blocks = sdkp->max_unmap_blocks;
730 max_blocks = sdkp->max_ws_blocks;
732 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
736 if (sdkp->device->unmap_limit_for_ws)
737 max_blocks = sdkp->max_unmap_blocks;
739 max_blocks = sdkp->max_ws_blocks;
741 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
745 max_blocks = min_not_zero(sdkp->max_ws_blocks,
746 (u32)SD_MAX_WS10_BLOCKS);
750 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
751 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
754 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
756 struct scsi_device *sdp = cmd->device;
757 struct request *rq = cmd->request;
758 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
759 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
760 unsigned int data_len = 24;
763 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
764 if (!rq->special_vec.bv_page)
765 return BLK_STS_RESOURCE;
766 clear_highpage(rq->special_vec.bv_page);
767 rq->special_vec.bv_offset = 0;
768 rq->special_vec.bv_len = data_len;
769 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
772 cmd->cmnd[0] = UNMAP;
775 buf = page_address(rq->special_vec.bv_page);
776 put_unaligned_be16(6 + 16, &buf[0]);
777 put_unaligned_be16(16, &buf[2]);
778 put_unaligned_be64(sector, &buf[8]);
779 put_unaligned_be32(nr_sectors, &buf[16]);
781 cmd->allowed = SD_MAX_RETRIES;
782 cmd->transfersize = data_len;
783 rq->timeout = SD_TIMEOUT;
784 scsi_req(rq)->resid_len = data_len;
786 return scsi_init_io(cmd);
789 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
792 struct scsi_device *sdp = cmd->device;
793 struct request *rq = cmd->request;
794 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
795 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
796 u32 data_len = sdp->sector_size;
798 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
799 if (!rq->special_vec.bv_page)
800 return BLK_STS_RESOURCE;
801 clear_highpage(rq->special_vec.bv_page);
802 rq->special_vec.bv_offset = 0;
803 rq->special_vec.bv_len = data_len;
804 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
807 cmd->cmnd[0] = WRITE_SAME_16;
809 cmd->cmnd[1] = 0x8; /* UNMAP */
810 put_unaligned_be64(sector, &cmd->cmnd[2]);
811 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
813 cmd->allowed = SD_MAX_RETRIES;
814 cmd->transfersize = data_len;
815 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
816 scsi_req(rq)->resid_len = data_len;
818 return scsi_init_io(cmd);
821 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
824 struct scsi_device *sdp = cmd->device;
825 struct request *rq = cmd->request;
826 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
827 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
828 u32 data_len = sdp->sector_size;
830 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
831 if (!rq->special_vec.bv_page)
832 return BLK_STS_RESOURCE;
833 clear_highpage(rq->special_vec.bv_page);
834 rq->special_vec.bv_offset = 0;
835 rq->special_vec.bv_len = data_len;
836 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
839 cmd->cmnd[0] = WRITE_SAME;
841 cmd->cmnd[1] = 0x8; /* UNMAP */
842 put_unaligned_be32(sector, &cmd->cmnd[2]);
843 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
845 cmd->allowed = SD_MAX_RETRIES;
846 cmd->transfersize = data_len;
847 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
848 scsi_req(rq)->resid_len = data_len;
850 return scsi_init_io(cmd);
853 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
855 struct request *rq = cmd->request;
856 struct scsi_device *sdp = cmd->device;
857 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
858 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
859 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
861 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
862 switch (sdkp->zeroing_mode) {
863 case SD_ZERO_WS16_UNMAP:
864 return sd_setup_write_same16_cmnd(cmd, true);
865 case SD_ZERO_WS10_UNMAP:
866 return sd_setup_write_same10_cmnd(cmd, true);
870 if (sdp->no_write_same)
871 return BLK_STS_TARGET;
873 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
874 return sd_setup_write_same16_cmnd(cmd, false);
876 return sd_setup_write_same10_cmnd(cmd, false);
879 static void sd_config_write_same(struct scsi_disk *sdkp)
881 struct request_queue *q = sdkp->disk->queue;
882 unsigned int logical_block_size = sdkp->device->sector_size;
884 if (sdkp->device->no_write_same) {
885 sdkp->max_ws_blocks = 0;
889 /* Some devices can not handle block counts above 0xffff despite
890 * supporting WRITE SAME(16). Consequently we default to 64k
891 * blocks per I/O unless the device explicitly advertises a
894 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
895 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
896 (u32)SD_MAX_WS16_BLOCKS);
897 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
898 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
899 (u32)SD_MAX_WS10_BLOCKS);
901 sdkp->device->no_write_same = 1;
902 sdkp->max_ws_blocks = 0;
905 if (sdkp->lbprz && sdkp->lbpws)
906 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
907 else if (sdkp->lbprz && sdkp->lbpws10)
908 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
909 else if (sdkp->max_ws_blocks)
910 sdkp->zeroing_mode = SD_ZERO_WS;
912 sdkp->zeroing_mode = SD_ZERO_WRITE;
914 if (sdkp->max_ws_blocks &&
915 sdkp->physical_block_size > logical_block_size) {
917 * Reporting a maximum number of blocks that is not aligned
918 * on the device physical size would cause a large write same
919 * request to be split into physically unaligned chunks by
920 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
921 * even if the caller of these functions took care to align the
922 * large request. So make sure the maximum reported is aligned
923 * to the device physical block size. This is only an optional
924 * optimization for regular disks, but this is mandatory to
925 * avoid failure of large write same requests directed at
926 * sequential write required zones of host-managed ZBC disks.
928 sdkp->max_ws_blocks =
929 round_down(sdkp->max_ws_blocks,
930 bytes_to_logical(sdkp->device,
931 sdkp->physical_block_size));
935 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
936 (logical_block_size >> 9));
937 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
938 (logical_block_size >> 9));
942 * sd_setup_write_same_cmnd - write the same data to multiple blocks
943 * @cmd: command to prepare
945 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
946 * the preference indicated by the target device.
948 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
950 struct request *rq = cmd->request;
951 struct scsi_device *sdp = cmd->device;
952 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
953 struct bio *bio = rq->bio;
954 sector_t sector = blk_rq_pos(rq);
955 unsigned int nr_sectors = blk_rq_sectors(rq);
956 unsigned int nr_bytes = blk_rq_bytes(rq);
959 if (sdkp->device->no_write_same)
960 return BLK_STS_TARGET;
962 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
964 sector >>= ilog2(sdp->sector_size) - 9;
965 nr_sectors >>= ilog2(sdp->sector_size) - 9;
967 rq->timeout = SD_WRITE_SAME_TIMEOUT;
969 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
971 cmd->cmnd[0] = WRITE_SAME_16;
972 put_unaligned_be64(sector, &cmd->cmnd[2]);
973 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
976 cmd->cmnd[0] = WRITE_SAME;
977 put_unaligned_be32(sector, &cmd->cmnd[2]);
978 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
981 cmd->transfersize = sdp->sector_size;
982 cmd->allowed = SD_MAX_RETRIES;
985 * For WRITE SAME the data transferred via the DATA OUT buffer is
986 * different from the amount of data actually written to the target.
988 * We set up __data_len to the amount of data transferred via the
989 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
990 * to transfer a single sector of data first, but then reset it to
991 * the amount of data to be written right after so that the I/O path
992 * knows how much to actually write.
994 rq->__data_len = sdp->sector_size;
995 ret = scsi_init_io(cmd);
996 rq->__data_len = nr_bytes;
1001 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1003 struct request *rq = cmd->request;
1005 /* flush requests don't perform I/O, zero the S/G table */
1006 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1008 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1010 cmd->transfersize = 0;
1011 cmd->allowed = SD_MAX_RETRIES;
1013 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1017 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1019 struct request *rq = SCpnt->request;
1020 struct scsi_device *sdp = SCpnt->device;
1021 struct gendisk *disk = rq->rq_disk;
1022 struct scsi_disk *sdkp = scsi_disk(disk);
1023 sector_t block = blk_rq_pos(rq);
1025 unsigned int this_count = blk_rq_sectors(rq);
1026 unsigned int dif, dix;
1027 unsigned char protect;
1030 ret = scsi_init_io(SCpnt);
1031 if (ret != BLK_STS_OK)
1033 WARN_ON_ONCE(SCpnt != rq->special);
1036 scmd_printk(KERN_INFO, SCpnt,
1037 "%s: block=%llu, count=%d\n",
1038 __func__, (unsigned long long)block, this_count));
1040 if (!sdp || !scsi_device_online(sdp) ||
1041 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1042 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1043 "Finishing %u sectors\n",
1044 blk_rq_sectors(rq)));
1045 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1046 "Retry with 0x%p\n", SCpnt));
1047 return BLK_STS_IOERR;
1052 * quietly refuse to do anything to a changed disc until
1053 * the changed bit has been reset
1055 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1056 return BLK_STS_IOERR;
1060 * Some SD card readers can't handle multi-sector accesses which touch
1061 * the last one or two hardware sectors. Split accesses as needed.
1063 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1064 (sdp->sector_size / 512);
1066 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1067 if (block < threshold) {
1068 /* Access up to the threshold but not beyond */
1069 this_count = threshold - block;
1071 /* Access only a single hardware sector */
1072 this_count = sdp->sector_size / 512;
1076 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1077 (unsigned long long)block));
1080 * If we have a 1K hardware sectorsize, prevent access to single
1081 * 512 byte sectors. In theory we could handle this - in fact
1082 * the scsi cdrom driver must be able to handle this because
1083 * we typically use 1K blocksizes, and cdroms typically have
1084 * 2K hardware sectorsizes. Of course, things are simpler
1085 * with the cdrom, since it is read-only. For performance
1086 * reasons, the filesystems should be able to handle this
1087 * and not force the scsi disk driver to use bounce buffers
1090 if (sdp->sector_size == 1024) {
1091 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1092 scmd_printk(KERN_ERR, SCpnt,
1093 "Bad block number requested\n");
1094 return BLK_STS_IOERR;
1097 this_count = this_count >> 1;
1099 if (sdp->sector_size == 2048) {
1100 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1101 scmd_printk(KERN_ERR, SCpnt,
1102 "Bad block number requested\n");
1103 return BLK_STS_IOERR;
1106 this_count = this_count >> 2;
1108 if (sdp->sector_size == 4096) {
1109 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1110 scmd_printk(KERN_ERR, SCpnt,
1111 "Bad block number requested\n");
1112 return BLK_STS_IOERR;
1115 this_count = this_count >> 3;
1117 if (rq_data_dir(rq) == WRITE) {
1118 SCpnt->cmnd[0] = WRITE_6;
1120 if (blk_integrity_rq(rq))
1121 t10_pi_prepare(SCpnt->request, sdkp->protection_type);
1123 } else if (rq_data_dir(rq) == READ) {
1124 SCpnt->cmnd[0] = READ_6;
1126 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1127 return BLK_STS_IOERR;
1130 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1131 "%s %d/%u 512 byte blocks.\n",
1132 (rq_data_dir(rq) == WRITE) ?
1133 "writing" : "reading", this_count,
1134 blk_rq_sectors(rq)));
1136 dix = scsi_prot_sg_count(SCpnt);
1137 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1140 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1144 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1145 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1147 if (unlikely(!SCpnt->cmnd))
1148 return BLK_STS_RESOURCE;
1150 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1151 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1152 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1153 SCpnt->cmnd[7] = 0x18;
1154 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1155 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1158 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1159 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1160 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1161 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1162 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1163 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1164 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1165 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1167 /* Expected Indirect LBA */
1168 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1169 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1170 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1171 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1173 /* Transfer length */
1174 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1175 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1176 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1177 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1178 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1179 SCpnt->cmnd[0] += READ_16 - READ_6;
1180 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1181 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1182 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1183 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1184 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1185 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1186 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1187 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1188 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1189 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1190 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1191 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1192 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1193 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1194 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1195 scsi_device_protection(SCpnt->device) ||
1196 SCpnt->device->use_10_for_rw) {
1197 SCpnt->cmnd[0] += READ_10 - READ_6;
1198 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1199 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1200 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1201 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1202 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1203 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1204 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1205 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1207 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1209 * This happens only if this drive failed
1210 * 10byte rw command with ILLEGAL_REQUEST
1211 * during operation and thus turned off
1214 scmd_printk(KERN_ERR, SCpnt,
1215 "FUA write on READ/WRITE(6) drive\n");
1216 return BLK_STS_IOERR;
1219 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1220 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1221 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1222 SCpnt->cmnd[4] = (unsigned char) this_count;
1225 SCpnt->sdb.length = this_count * sdp->sector_size;
1228 * We shouldn't disconnect in the middle of a sector, so with a dumb
1229 * host adapter, it's safe to assume that we can at least transfer
1230 * this many bytes between each connect / disconnect.
1232 SCpnt->transfersize = sdp->sector_size;
1233 SCpnt->underflow = this_count << 9;
1234 SCpnt->allowed = SD_MAX_RETRIES;
1237 * This indicates that the command is ready from our end to be
1243 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1245 struct request *rq = cmd->request;
1247 switch (req_op(rq)) {
1248 case REQ_OP_DISCARD:
1249 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1251 return sd_setup_unmap_cmnd(cmd);
1253 return sd_setup_write_same16_cmnd(cmd, true);
1255 return sd_setup_write_same10_cmnd(cmd, true);
1257 return sd_setup_write_same10_cmnd(cmd, false);
1259 return BLK_STS_TARGET;
1261 case REQ_OP_WRITE_ZEROES:
1262 return sd_setup_write_zeroes_cmnd(cmd);
1263 case REQ_OP_WRITE_SAME:
1264 return sd_setup_write_same_cmnd(cmd);
1266 return sd_setup_flush_cmnd(cmd);
1269 return sd_setup_read_write_cmnd(cmd);
1270 case REQ_OP_ZONE_RESET:
1271 return sd_zbc_setup_reset_cmnd(cmd);
1274 return BLK_STS_NOTSUPP;
1278 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1280 struct request *rq = SCpnt->request;
1283 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1284 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1286 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1290 mempool_free(cmnd, sd_cdb_pool);
1295 * sd_open - open a scsi disk device
1296 * @bdev: Block device of the scsi disk to open
1297 * @mode: FMODE_* mask
1299 * Returns 0 if successful. Returns a negated errno value in case
1302 * Note: This can be called from a user context (e.g. fsck(1) )
1303 * or from within the kernel (e.g. as a result of a mount(1) ).
1304 * In the latter case @inode and @filp carry an abridged amount
1305 * of information as noted above.
1307 * Locking: called with bdev->bd_mutex held.
1309 static int sd_open(struct block_device *bdev, fmode_t mode)
1311 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1312 struct scsi_device *sdev;
1318 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1320 sdev = sdkp->device;
1323 * If the device is in error recovery, wait until it is done.
1324 * If the device is offline, then disallow any access to it.
1327 if (!scsi_block_when_processing_errors(sdev))
1330 if (sdev->removable || sdkp->write_prot)
1331 check_disk_change(bdev);
1334 * If the drive is empty, just let the open fail.
1336 retval = -ENOMEDIUM;
1337 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1341 * If the device has the write protect tab set, have the open fail
1342 * if the user expects to be able to write to the thing.
1345 if (sdkp->write_prot && (mode & FMODE_WRITE))
1349 * It is possible that the disk changing stuff resulted in
1350 * the device being taken offline. If this is the case,
1351 * report this to the user, and don't pretend that the
1352 * open actually succeeded.
1355 if (!scsi_device_online(sdev))
1358 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1359 if (scsi_block_when_processing_errors(sdev))
1360 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1366 scsi_disk_put(sdkp);
1371 * sd_release - invoked when the (last) close(2) is called on this
1373 * @disk: disk to release
1374 * @mode: FMODE_* mask
1378 * Note: may block (uninterruptible) if error recovery is underway
1381 * Locking: called with bdev->bd_mutex held.
1383 static void sd_release(struct gendisk *disk, fmode_t mode)
1385 struct scsi_disk *sdkp = scsi_disk(disk);
1386 struct scsi_device *sdev = sdkp->device;
1388 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1390 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1391 if (scsi_block_when_processing_errors(sdev))
1392 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1396 * XXX and what if there are packets in flight and this close()
1397 * XXX is followed by a "rmmod sd_mod"?
1400 scsi_disk_put(sdkp);
1403 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1405 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1406 struct scsi_device *sdp = sdkp->device;
1407 struct Scsi_Host *host = sdp->host;
1408 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1411 /* default to most commonly used values */
1412 diskinfo[0] = 0x40; /* 1 << 6 */
1413 diskinfo[1] = 0x20; /* 1 << 5 */
1414 diskinfo[2] = capacity >> 11;
1416 /* override with calculated, extended default, or driver values */
1417 if (host->hostt->bios_param)
1418 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1420 scsicam_bios_param(bdev, capacity, diskinfo);
1422 geo->heads = diskinfo[0];
1423 geo->sectors = diskinfo[1];
1424 geo->cylinders = diskinfo[2];
1429 * sd_ioctl - process an ioctl
1430 * @bdev: target block device
1431 * @mode: FMODE_* mask
1432 * @cmd: ioctl command number
1433 * @arg: this is third argument given to ioctl(2) system call.
1434 * Often contains a pointer.
1436 * Returns 0 if successful (some ioctls return positive numbers on
1437 * success as well). Returns a negated errno value in case of error.
1439 * Note: most ioctls are forward onto the block subsystem or further
1440 * down in the scsi subsystem.
1442 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1443 unsigned int cmd, unsigned long arg)
1445 struct gendisk *disk = bdev->bd_disk;
1446 struct scsi_disk *sdkp = scsi_disk(disk);
1447 struct scsi_device *sdp = sdkp->device;
1448 void __user *p = (void __user *)arg;
1451 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1452 "cmd=0x%x\n", disk->disk_name, cmd));
1454 error = scsi_verify_blk_ioctl(bdev, cmd);
1459 * If we are in the middle of error recovery, don't let anyone
1460 * else try and use this device. Also, if error recovery fails, it
1461 * may try and take the device offline, in which case all further
1462 * access to the device is prohibited.
1464 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1465 (mode & FMODE_NDELAY) != 0);
1469 if (is_sed_ioctl(cmd))
1470 return sed_ioctl(sdkp->opal_dev, cmd, p);
1473 * Send SCSI addressing ioctls directly to mid level, send other
1474 * ioctls to block level and then onto mid level if they can't be
1478 case SCSI_IOCTL_GET_IDLUN:
1479 case SCSI_IOCTL_GET_BUS_NUMBER:
1480 error = scsi_ioctl(sdp, cmd, p);
1483 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1484 if (error != -ENOTTY)
1486 error = scsi_ioctl(sdp, cmd, p);
1493 static void set_media_not_present(struct scsi_disk *sdkp)
1495 if (sdkp->media_present)
1496 sdkp->device->changed = 1;
1498 if (sdkp->device->removable) {
1499 sdkp->media_present = 0;
1504 static int media_not_present(struct scsi_disk *sdkp,
1505 struct scsi_sense_hdr *sshdr)
1507 if (!scsi_sense_valid(sshdr))
1510 /* not invoked for commands that could return deferred errors */
1511 switch (sshdr->sense_key) {
1512 case UNIT_ATTENTION:
1514 /* medium not present */
1515 if (sshdr->asc == 0x3A) {
1516 set_media_not_present(sdkp);
1524 * sd_check_events - check media events
1525 * @disk: kernel device descriptor
1526 * @clearing: disk events currently being cleared
1528 * Returns mask of DISK_EVENT_*.
1530 * Note: this function is invoked from the block subsystem.
1532 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1534 struct scsi_disk *sdkp = scsi_disk_get(disk);
1535 struct scsi_device *sdp;
1542 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1545 * If the device is offline, don't send any commands - just pretend as
1546 * if the command failed. If the device ever comes back online, we
1547 * can deal with it then. It is only because of unrecoverable errors
1548 * that we would ever take a device offline in the first place.
1550 if (!scsi_device_online(sdp)) {
1551 set_media_not_present(sdkp);
1556 * Using TEST_UNIT_READY enables differentiation between drive with
1557 * no cartridge loaded - NOT READY, drive with changed cartridge -
1558 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1560 * Drives that auto spin down. eg iomega jaz 1G, will be started
1561 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1562 * sd_revalidate() is called.
1564 if (scsi_block_when_processing_errors(sdp)) {
1565 struct scsi_sense_hdr sshdr = { 0, };
1567 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1570 /* failed to execute TUR, assume media not present */
1571 if (host_byte(retval)) {
1572 set_media_not_present(sdkp);
1576 if (media_not_present(sdkp, &sshdr))
1581 * For removable scsi disk we have to recognise the presence
1582 * of a disk in the drive.
1584 if (!sdkp->media_present)
1586 sdkp->media_present = 1;
1589 * sdp->changed is set under the following conditions:
1591 * Medium present state has changed in either direction.
1592 * Device has indicated UNIT_ATTENTION.
1594 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1596 scsi_disk_put(sdkp);
1600 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1603 struct scsi_device *sdp = sdkp->device;
1604 const int timeout = sdp->request_queue->rq_timeout
1605 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1606 struct scsi_sense_hdr my_sshdr;
1608 if (!scsi_device_online(sdp))
1611 /* caller might not be interested in sense, but we need it */
1615 for (retries = 3; retries > 0; --retries) {
1616 unsigned char cmd[10] = { 0 };
1618 cmd[0] = SYNCHRONIZE_CACHE;
1620 * Leave the rest of the command zero to indicate
1623 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1624 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1630 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1632 if (driver_byte(res) == DRIVER_SENSE)
1633 sd_print_sense_hdr(sdkp, sshdr);
1635 /* we need to evaluate the error return */
1636 if (scsi_sense_valid(sshdr) &&
1637 (sshdr->asc == 0x3a || /* medium not present */
1638 sshdr->asc == 0x20)) /* invalid command */
1639 /* this is no error here */
1642 switch (host_byte(res)) {
1643 /* ignore errors due to racing a disconnection */
1644 case DID_BAD_TARGET:
1645 case DID_NO_CONNECT:
1647 /* signal the upper layer it might try again */
1651 case DID_SOFT_ERROR:
1660 static void sd_rescan(struct device *dev)
1662 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1664 revalidate_disk(sdkp->disk);
1668 #ifdef CONFIG_COMPAT
1670 * This gets directly called from VFS. When the ioctl
1671 * is not recognized we go back to the other translation paths.
1673 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1674 unsigned int cmd, unsigned long arg)
1676 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1679 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1680 (mode & FMODE_NDELAY) != 0);
1685 * Let the static ioctl translation table take care of it.
1687 if (!sdev->host->hostt->compat_ioctl)
1688 return -ENOIOCTLCMD;
1689 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1693 static char sd_pr_type(enum pr_type type)
1696 case PR_WRITE_EXCLUSIVE:
1698 case PR_EXCLUSIVE_ACCESS:
1700 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1702 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1704 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1706 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1713 static int sd_pr_command(struct block_device *bdev, u8 sa,
1714 u64 key, u64 sa_key, u8 type, u8 flags)
1716 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1717 struct scsi_sense_hdr sshdr;
1719 u8 cmd[16] = { 0, };
1720 u8 data[24] = { 0, };
1722 cmd[0] = PERSISTENT_RESERVE_OUT;
1725 put_unaligned_be32(sizeof(data), &cmd[5]);
1727 put_unaligned_be64(key, &data[0]);
1728 put_unaligned_be64(sa_key, &data[8]);
1731 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1732 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1734 if (driver_byte(result) == DRIVER_SENSE &&
1735 scsi_sense_valid(&sshdr)) {
1736 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1737 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1743 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1746 if (flags & ~PR_FL_IGNORE_KEY)
1748 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1749 old_key, new_key, 0,
1750 (1 << 0) /* APTPL */);
1753 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1758 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1761 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1763 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1766 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1767 enum pr_type type, bool abort)
1769 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1770 sd_pr_type(type), 0);
1773 static int sd_pr_clear(struct block_device *bdev, u64 key)
1775 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1778 static const struct pr_ops sd_pr_ops = {
1779 .pr_register = sd_pr_register,
1780 .pr_reserve = sd_pr_reserve,
1781 .pr_release = sd_pr_release,
1782 .pr_preempt = sd_pr_preempt,
1783 .pr_clear = sd_pr_clear,
1786 static const struct block_device_operations sd_fops = {
1787 .owner = THIS_MODULE,
1789 .release = sd_release,
1791 .getgeo = sd_getgeo,
1792 #ifdef CONFIG_COMPAT
1793 .compat_ioctl = sd_compat_ioctl,
1795 .check_events = sd_check_events,
1796 .revalidate_disk = sd_revalidate_disk,
1797 .unlock_native_capacity = sd_unlock_native_capacity,
1798 .report_zones = sd_zbc_report_zones,
1799 .pr_ops = &sd_pr_ops,
1803 * sd_eh_reset - reset error handling callback
1804 * @scmd: sd-issued command that has failed
1806 * This function is called by the SCSI midlayer before starting
1807 * SCSI EH. When counting medium access failures we have to be
1808 * careful to register it only only once per device and SCSI EH run;
1809 * there might be several timed out commands which will cause the
1810 * 'max_medium_access_timeouts' counter to trigger after the first
1811 * SCSI EH run already and set the device to offline.
1812 * So this function resets the internal counter before starting SCSI EH.
1814 static void sd_eh_reset(struct scsi_cmnd *scmd)
1816 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1818 /* New SCSI EH run, reset gate variable */
1819 sdkp->ignore_medium_access_errors = false;
1823 * sd_eh_action - error handling callback
1824 * @scmd: sd-issued command that has failed
1825 * @eh_disp: The recovery disposition suggested by the midlayer
1827 * This function is called by the SCSI midlayer upon completion of an
1828 * error test command (currently TEST UNIT READY). The result of sending
1829 * the eh command is passed in eh_disp. We're looking for devices that
1830 * fail medium access commands but are OK with non access commands like
1831 * test unit ready (so wrongly see the device as having a successful
1834 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1836 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1837 struct scsi_device *sdev = scmd->device;
1839 if (!scsi_device_online(sdev) ||
1840 !scsi_medium_access_command(scmd) ||
1841 host_byte(scmd->result) != DID_TIME_OUT ||
1846 * The device has timed out executing a medium access command.
1847 * However, the TEST UNIT READY command sent during error
1848 * handling completed successfully. Either the device is in the
1849 * process of recovering or has it suffered an internal failure
1850 * that prevents access to the storage medium.
1852 if (!sdkp->ignore_medium_access_errors) {
1853 sdkp->medium_access_timed_out++;
1854 sdkp->ignore_medium_access_errors = true;
1858 * If the device keeps failing read/write commands but TEST UNIT
1859 * READY always completes successfully we assume that medium
1860 * access is no longer possible and take the device offline.
1862 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1863 scmd_printk(KERN_ERR, scmd,
1864 "Medium access timeout failure. Offlining disk!\n");
1865 mutex_lock(&sdev->state_mutex);
1866 scsi_device_set_state(sdev, SDEV_OFFLINE);
1867 mutex_unlock(&sdev->state_mutex);
1875 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1877 struct request *req = scmd->request;
1878 struct scsi_device *sdev = scmd->device;
1879 unsigned int transferred, good_bytes;
1880 u64 start_lba, end_lba, bad_lba;
1883 * Some commands have a payload smaller than the device logical
1884 * block size (e.g. INQUIRY on a 4K disk).
1886 if (scsi_bufflen(scmd) <= sdev->sector_size)
1889 /* Check if we have a 'bad_lba' information */
1890 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1891 SCSI_SENSE_BUFFERSIZE,
1896 * If the bad lba was reported incorrectly, we have no idea where
1899 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1900 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1901 if (bad_lba < start_lba || bad_lba >= end_lba)
1905 * resid is optional but mostly filled in. When it's unused,
1906 * its value is zero, so we assume the whole buffer transferred
1908 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1910 /* This computation should always be done in terms of the
1911 * resolution of the device's medium.
1913 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1915 return min(good_bytes, transferred);
1919 * sd_done - bottom half handler: called when the lower level
1920 * driver has completed (successfully or otherwise) a scsi command.
1921 * @SCpnt: mid-level's per command structure.
1923 * Note: potentially run from within an ISR. Must not block.
1925 static int sd_done(struct scsi_cmnd *SCpnt)
1927 int result = SCpnt->result;
1928 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1929 unsigned int sector_size = SCpnt->device->sector_size;
1931 struct scsi_sense_hdr sshdr;
1932 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1933 struct request *req = SCpnt->request;
1934 int sense_valid = 0;
1935 int sense_deferred = 0;
1937 switch (req_op(req)) {
1938 case REQ_OP_DISCARD:
1939 case REQ_OP_WRITE_ZEROES:
1940 case REQ_OP_WRITE_SAME:
1941 case REQ_OP_ZONE_RESET:
1943 good_bytes = blk_rq_bytes(req);
1944 scsi_set_resid(SCpnt, 0);
1947 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1952 * In case of bogus fw or device, we could end up having
1953 * an unaligned partial completion. Check this here and force
1956 resid = scsi_get_resid(SCpnt);
1957 if (resid & (sector_size - 1)) {
1958 sd_printk(KERN_INFO, sdkp,
1959 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1960 resid, sector_size);
1961 resid = min(scsi_bufflen(SCpnt),
1962 round_up(resid, sector_size));
1963 scsi_set_resid(SCpnt, resid);
1968 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1970 sense_deferred = scsi_sense_is_deferred(&sshdr);
1972 sdkp->medium_access_timed_out = 0;
1974 if (driver_byte(result) != DRIVER_SENSE &&
1975 (!sense_valid || sense_deferred))
1978 switch (sshdr.sense_key) {
1979 case HARDWARE_ERROR:
1981 good_bytes = sd_completed_bytes(SCpnt);
1983 case RECOVERED_ERROR:
1984 good_bytes = scsi_bufflen(SCpnt);
1987 /* This indicates a false check condition, so ignore it. An
1988 * unknown amount of data was transferred so treat it as an
1992 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1994 case ABORTED_COMMAND:
1995 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1996 good_bytes = sd_completed_bytes(SCpnt);
1998 case ILLEGAL_REQUEST:
1999 switch (sshdr.asc) {
2000 case 0x10: /* DIX: Host detected corruption */
2001 good_bytes = sd_completed_bytes(SCpnt);
2003 case 0x20: /* INVALID COMMAND OPCODE */
2004 case 0x24: /* INVALID FIELD IN CDB */
2005 switch (SCpnt->cmnd[0]) {
2007 sd_config_discard(sdkp, SD_LBP_DISABLE);
2011 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2012 sd_config_discard(sdkp, SD_LBP_DISABLE);
2014 sdkp->device->no_write_same = 1;
2015 sd_config_write_same(sdkp);
2016 req->rq_flags |= RQF_QUIET;
2027 if (sd_is_zoned(sdkp))
2028 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2030 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2031 "sd_done: completed %d of %d bytes\n",
2032 good_bytes, scsi_bufflen(SCpnt)));
2034 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2036 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2037 good_bytes / scsi_prot_interval(SCpnt));
2043 * spinup disk - called only in sd_revalidate_disk()
2046 sd_spinup_disk(struct scsi_disk *sdkp)
2048 unsigned char cmd[10];
2049 unsigned long spintime_expire = 0;
2050 int retries, spintime;
2051 unsigned int the_result;
2052 struct scsi_sense_hdr sshdr;
2053 int sense_valid = 0;
2057 /* Spin up drives, as required. Only do this at boot time */
2058 /* Spinup needs to be done for module loads too. */
2063 cmd[0] = TEST_UNIT_READY;
2064 memset((void *) &cmd[1], 0, 9);
2066 the_result = scsi_execute_req(sdkp->device, cmd,
2069 SD_MAX_RETRIES, NULL);
2072 * If the drive has indicated to us that it
2073 * doesn't have any media in it, don't bother
2074 * with any more polling.
2076 if (media_not_present(sdkp, &sshdr))
2080 sense_valid = scsi_sense_valid(&sshdr);
2082 } while (retries < 3 &&
2083 (!scsi_status_is_good(the_result) ||
2084 ((driver_byte(the_result) == DRIVER_SENSE) &&
2085 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2087 if (driver_byte(the_result) != DRIVER_SENSE) {
2088 /* no sense, TUR either succeeded or failed
2089 * with a status error */
2090 if(!spintime && !scsi_status_is_good(the_result)) {
2091 sd_print_result(sdkp, "Test Unit Ready failed",
2098 * The device does not want the automatic start to be issued.
2100 if (sdkp->device->no_start_on_add)
2103 if (sense_valid && sshdr.sense_key == NOT_READY) {
2104 if (sshdr.asc == 4 && sshdr.ascq == 3)
2105 break; /* manual intervention required */
2106 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2107 break; /* standby */
2108 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2109 break; /* unavailable */
2110 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2111 break; /* sanitize in progress */
2113 * Issue command to spin up drive when not ready
2116 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2117 cmd[0] = START_STOP;
2118 cmd[1] = 1; /* Return immediately */
2119 memset((void *) &cmd[2], 0, 8);
2120 cmd[4] = 1; /* Start spin cycle */
2121 if (sdkp->device->start_stop_pwr_cond)
2123 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2125 SD_TIMEOUT, SD_MAX_RETRIES,
2127 spintime_expire = jiffies + 100 * HZ;
2130 /* Wait 1 second for next try */
2132 printk(KERN_CONT ".");
2135 * Wait for USB flash devices with slow firmware.
2136 * Yes, this sense key/ASC combination shouldn't
2137 * occur here. It's characteristic of these devices.
2139 } else if (sense_valid &&
2140 sshdr.sense_key == UNIT_ATTENTION &&
2141 sshdr.asc == 0x28) {
2143 spintime_expire = jiffies + 5 * HZ;
2146 /* Wait 1 second for next try */
2149 /* we don't understand the sense code, so it's
2150 * probably pointless to loop */
2152 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2153 sd_print_sense_hdr(sdkp, &sshdr);
2158 } while (spintime && time_before_eq(jiffies, spintime_expire));
2161 if (scsi_status_is_good(the_result))
2162 printk(KERN_CONT "ready\n");
2164 printk(KERN_CONT "not responding...\n");
2169 * Determine whether disk supports Data Integrity Field.
2171 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2173 struct scsi_device *sdp = sdkp->device;
2177 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2180 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2182 if (type > T10_PI_TYPE3_PROTECTION)
2184 else if (scsi_host_dif_capable(sdp->host, type))
2187 if (sdkp->first_scan || type != sdkp->protection_type)
2190 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2191 " protection type %u. Disabling disk!\n",
2195 sd_printk(KERN_NOTICE, sdkp,
2196 "Enabling DIF Type %u protection\n", type);
2199 sd_printk(KERN_NOTICE, sdkp,
2200 "Disabling DIF Type %u protection\n", type);
2204 sdkp->protection_type = type;
2209 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2210 struct scsi_sense_hdr *sshdr, int sense_valid,
2213 if (driver_byte(the_result) == DRIVER_SENSE)
2214 sd_print_sense_hdr(sdkp, sshdr);
2216 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2219 * Set dirty bit for removable devices if not ready -
2220 * sometimes drives will not report this properly.
2222 if (sdp->removable &&
2223 sense_valid && sshdr->sense_key == NOT_READY)
2224 set_media_not_present(sdkp);
2227 * We used to set media_present to 0 here to indicate no media
2228 * in the drive, but some drives fail read capacity even with
2229 * media present, so we can't do that.
2231 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2235 #if RC16_LEN > SD_BUF_SIZE
2236 #error RC16_LEN must not be more than SD_BUF_SIZE
2239 #define READ_CAPACITY_RETRIES_ON_RESET 10
2242 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2243 * and the reported logical block size is bigger than 512 bytes. Note
2244 * that last_sector is a u64 and therefore logical_to_sectors() is not
2247 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2249 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2251 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2257 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2258 unsigned char *buffer)
2260 unsigned char cmd[16];
2261 struct scsi_sense_hdr sshdr;
2262 int sense_valid = 0;
2264 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2265 unsigned int alignment;
2266 unsigned long long lba;
2267 unsigned sector_size;
2269 if (sdp->no_read_capacity_16)
2274 cmd[0] = SERVICE_ACTION_IN_16;
2275 cmd[1] = SAI_READ_CAPACITY_16;
2277 memset(buffer, 0, RC16_LEN);
2279 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2280 buffer, RC16_LEN, &sshdr,
2281 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2283 if (media_not_present(sdkp, &sshdr))
2287 sense_valid = scsi_sense_valid(&sshdr);
2289 sshdr.sense_key == ILLEGAL_REQUEST &&
2290 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2292 /* Invalid Command Operation Code or
2293 * Invalid Field in CDB, just retry
2294 * silently with RC10 */
2297 sshdr.sense_key == UNIT_ATTENTION &&
2298 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2299 /* Device reset might occur several times,
2300 * give it one more chance */
2301 if (--reset_retries > 0)
2306 } while (the_result && retries);
2309 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2310 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2314 sector_size = get_unaligned_be32(&buffer[8]);
2315 lba = get_unaligned_be64(&buffer[0]);
2317 if (sd_read_protection_type(sdkp, buffer) < 0) {
2322 if (!sd_addressable_capacity(lba, sector_size)) {
2323 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2324 "kernel compiled with support for large block "
2330 /* Logical blocks per physical block exponent */
2331 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2334 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2336 /* Lowest aligned logical block */
2337 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2338 blk_queue_alignment_offset(sdp->request_queue, alignment);
2339 if (alignment && sdkp->first_scan)
2340 sd_printk(KERN_NOTICE, sdkp,
2341 "physical block alignment offset: %u\n", alignment);
2343 if (buffer[14] & 0x80) { /* LBPME */
2346 if (buffer[14] & 0x40) /* LBPRZ */
2349 sd_config_discard(sdkp, SD_LBP_WS16);
2352 sdkp->capacity = lba + 1;
2356 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2357 unsigned char *buffer)
2359 unsigned char cmd[16];
2360 struct scsi_sense_hdr sshdr;
2361 int sense_valid = 0;
2363 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2365 unsigned sector_size;
2368 cmd[0] = READ_CAPACITY;
2369 memset(&cmd[1], 0, 9);
2370 memset(buffer, 0, 8);
2372 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2374 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2376 if (media_not_present(sdkp, &sshdr))
2380 sense_valid = scsi_sense_valid(&sshdr);
2382 sshdr.sense_key == UNIT_ATTENTION &&
2383 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2384 /* Device reset might occur several times,
2385 * give it one more chance */
2386 if (--reset_retries > 0)
2391 } while (the_result && retries);
2394 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2395 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2399 sector_size = get_unaligned_be32(&buffer[4]);
2400 lba = get_unaligned_be32(&buffer[0]);
2402 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2403 /* Some buggy (usb cardreader) devices return an lba of
2404 0xffffffff when the want to report a size of 0 (with
2405 which they really mean no media is present) */
2407 sdkp->physical_block_size = sector_size;
2411 if (!sd_addressable_capacity(lba, sector_size)) {
2412 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2413 "kernel compiled with support for large block "
2419 sdkp->capacity = lba + 1;
2420 sdkp->physical_block_size = sector_size;
2424 static int sd_try_rc16_first(struct scsi_device *sdp)
2426 if (sdp->host->max_cmd_len < 16)
2428 if (sdp->try_rc_10_first)
2430 if (sdp->scsi_level > SCSI_SPC_2)
2432 if (scsi_device_protection(sdp))
2438 * read disk capacity
2441 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2444 struct scsi_device *sdp = sdkp->device;
2446 if (sd_try_rc16_first(sdp)) {
2447 sector_size = read_capacity_16(sdkp, sdp, buffer);
2448 if (sector_size == -EOVERFLOW)
2450 if (sector_size == -ENODEV)
2452 if (sector_size < 0)
2453 sector_size = read_capacity_10(sdkp, sdp, buffer);
2454 if (sector_size < 0)
2457 sector_size = read_capacity_10(sdkp, sdp, buffer);
2458 if (sector_size == -EOVERFLOW)
2460 if (sector_size < 0)
2462 if ((sizeof(sdkp->capacity) > 4) &&
2463 (sdkp->capacity > 0xffffffffULL)) {
2464 int old_sector_size = sector_size;
2465 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2466 "Trying to use READ CAPACITY(16).\n");
2467 sector_size = read_capacity_16(sdkp, sdp, buffer);
2468 if (sector_size < 0) {
2469 sd_printk(KERN_NOTICE, sdkp,
2470 "Using 0xffffffff as device size\n");
2471 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2472 sector_size = old_sector_size;
2475 /* Remember that READ CAPACITY(16) succeeded */
2476 sdp->try_rc_10_first = 0;
2480 /* Some devices are known to return the total number of blocks,
2481 * not the highest block number. Some devices have versions
2482 * which do this and others which do not. Some devices we might
2483 * suspect of doing this but we don't know for certain.
2485 * If we know the reported capacity is wrong, decrement it. If
2486 * we can only guess, then assume the number of blocks is even
2487 * (usually true but not always) and err on the side of lowering
2490 if (sdp->fix_capacity ||
2491 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2492 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2493 "from its reported value: %llu\n",
2494 (unsigned long long) sdkp->capacity);
2499 if (sector_size == 0) {
2501 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2505 if (sector_size != 512 &&
2506 sector_size != 1024 &&
2507 sector_size != 2048 &&
2508 sector_size != 4096) {
2509 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2512 * The user might want to re-format the drive with
2513 * a supported sectorsize. Once this happens, it
2514 * would be relatively trivial to set the thing up.
2515 * For this reason, we leave the thing in the table.
2519 * set a bogus sector size so the normal read/write
2520 * logic in the block layer will eventually refuse any
2521 * request on this device without tripping over power
2522 * of two sector size assumptions
2526 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2527 blk_queue_physical_block_size(sdp->request_queue,
2528 sdkp->physical_block_size);
2529 sdkp->device->sector_size = sector_size;
2531 if (sdkp->capacity > 0xffffffff)
2532 sdp->use_16_for_rw = 1;
2537 * Print disk capacity
2540 sd_print_capacity(struct scsi_disk *sdkp,
2541 sector_t old_capacity)
2543 int sector_size = sdkp->device->sector_size;
2544 char cap_str_2[10], cap_str_10[10];
2546 string_get_size(sdkp->capacity, sector_size,
2547 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2548 string_get_size(sdkp->capacity, sector_size,
2549 STRING_UNITS_10, cap_str_10,
2550 sizeof(cap_str_10));
2552 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2553 sd_printk(KERN_NOTICE, sdkp,
2554 "%llu %d-byte logical blocks: (%s/%s)\n",
2555 (unsigned long long)sdkp->capacity,
2556 sector_size, cap_str_10, cap_str_2);
2558 if (sdkp->physical_block_size != sector_size)
2559 sd_printk(KERN_NOTICE, sdkp,
2560 "%u-byte physical blocks\n",
2561 sdkp->physical_block_size);
2563 sd_zbc_print_zones(sdkp);
2567 /* called with buffer of length 512 */
2569 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2570 unsigned char *buffer, int len, struct scsi_mode_data *data,
2571 struct scsi_sense_hdr *sshdr)
2573 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2574 SD_TIMEOUT, SD_MAX_RETRIES, data,
2579 * read write protect setting, if possible - called only in sd_revalidate_disk()
2580 * called with buffer of length SD_BUF_SIZE
2583 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2586 struct scsi_device *sdp = sdkp->device;
2587 struct scsi_mode_data data;
2588 int disk_ro = get_disk_ro(sdkp->disk);
2589 int old_wp = sdkp->write_prot;
2591 set_disk_ro(sdkp->disk, 0);
2592 if (sdp->skip_ms_page_3f) {
2593 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2597 if (sdp->use_192_bytes_for_3f) {
2598 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2601 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2602 * We have to start carefully: some devices hang if we ask
2603 * for more than is available.
2605 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2608 * Second attempt: ask for page 0 When only page 0 is
2609 * implemented, a request for page 3F may return Sense Key
2610 * 5: Illegal Request, Sense Code 24: Invalid field in
2613 if (!scsi_status_is_good(res))
2614 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2617 * Third attempt: ask 255 bytes, as we did earlier.
2619 if (!scsi_status_is_good(res))
2620 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2624 if (!scsi_status_is_good(res)) {
2625 sd_first_printk(KERN_WARNING, sdkp,
2626 "Test WP failed, assume Write Enabled\n");
2628 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2629 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2630 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2631 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2632 sdkp->write_prot ? "on" : "off");
2633 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2639 * sd_read_cache_type - called only from sd_revalidate_disk()
2640 * called with buffer of length SD_BUF_SIZE
2643 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2646 struct scsi_device *sdp = sdkp->device;
2651 struct scsi_mode_data data;
2652 struct scsi_sense_hdr sshdr;
2653 int old_wce = sdkp->WCE;
2654 int old_rcd = sdkp->RCD;
2655 int old_dpofua = sdkp->DPOFUA;
2658 if (sdkp->cache_override)
2662 if (sdp->skip_ms_page_8) {
2663 if (sdp->type == TYPE_RBC)
2666 if (sdp->skip_ms_page_3f)
2669 if (sdp->use_192_bytes_for_3f)
2673 } else if (sdp->type == TYPE_RBC) {
2681 /* cautiously ask */
2682 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2685 if (!scsi_status_is_good(res))
2688 if (!data.header_length) {
2691 sd_first_printk(KERN_ERR, sdkp,
2692 "Missing header in MODE_SENSE response\n");
2695 /* that went OK, now ask for the proper length */
2699 * We're only interested in the first three bytes, actually.
2700 * But the data cache page is defined for the first 20.
2704 else if (len > SD_BUF_SIZE) {
2705 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2706 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2709 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2713 if (len > first_len)
2714 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2717 if (scsi_status_is_good(res)) {
2718 int offset = data.header_length + data.block_descriptor_length;
2720 while (offset < len) {
2721 u8 page_code = buffer[offset] & 0x3F;
2722 u8 spf = buffer[offset] & 0x40;
2724 if (page_code == 8 || page_code == 6) {
2725 /* We're interested only in the first 3 bytes.
2727 if (len - offset <= 2) {
2728 sd_first_printk(KERN_ERR, sdkp,
2729 "Incomplete mode parameter "
2733 modepage = page_code;
2737 /* Go to the next page */
2738 if (spf && len - offset > 3)
2739 offset += 4 + (buffer[offset+2] << 8) +
2741 else if (!spf && len - offset > 1)
2742 offset += 2 + buffer[offset+1];
2744 sd_first_printk(KERN_ERR, sdkp,
2746 "parameter data\n");
2752 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2756 if (modepage == 8) {
2757 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2758 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2760 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2764 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2765 if (sdp->broken_fua) {
2766 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2768 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2769 !sdkp->device->use_16_for_rw) {
2770 sd_first_printk(KERN_NOTICE, sdkp,
2771 "Uses READ/WRITE(6), disabling FUA\n");
2775 /* No cache flush allowed for write protected devices */
2776 if (sdkp->WCE && sdkp->write_prot)
2779 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2780 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2781 sd_printk(KERN_NOTICE, sdkp,
2782 "Write cache: %s, read cache: %s, %s\n",
2783 sdkp->WCE ? "enabled" : "disabled",
2784 sdkp->RCD ? "disabled" : "enabled",
2785 sdkp->DPOFUA ? "supports DPO and FUA"
2786 : "doesn't support DPO or FUA");
2792 if (scsi_sense_valid(&sshdr) &&
2793 sshdr.sense_key == ILLEGAL_REQUEST &&
2794 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2795 /* Invalid field in CDB */
2796 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2798 sd_first_printk(KERN_ERR, sdkp,
2799 "Asking for cache data failed\n");
2802 if (sdp->wce_default_on) {
2803 sd_first_printk(KERN_NOTICE, sdkp,
2804 "Assuming drive cache: write back\n");
2807 sd_first_printk(KERN_ERR, sdkp,
2808 "Assuming drive cache: write through\n");
2816 * The ATO bit indicates whether the DIF application tag is available
2817 * for use by the operating system.
2819 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2822 struct scsi_device *sdp = sdkp->device;
2823 struct scsi_mode_data data;
2824 struct scsi_sense_hdr sshdr;
2826 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2829 if (sdkp->protection_type == 0)
2832 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2833 SD_MAX_RETRIES, &data, &sshdr);
2835 if (!scsi_status_is_good(res) || !data.header_length ||
2837 sd_first_printk(KERN_WARNING, sdkp,
2838 "getting Control mode page failed, assume no ATO\n");
2840 if (scsi_sense_valid(&sshdr))
2841 sd_print_sense_hdr(sdkp, &sshdr);
2846 offset = data.header_length + data.block_descriptor_length;
2848 if ((buffer[offset] & 0x3f) != 0x0a) {
2849 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2853 if ((buffer[offset + 5] & 0x80) == 0)
2862 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2863 * @sdkp: disk to query
2865 static void sd_read_block_limits(struct scsi_disk *sdkp)
2867 unsigned int sector_sz = sdkp->device->sector_size;
2868 const int vpd_len = 64;
2869 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2872 /* Block Limits VPD */
2873 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2876 blk_queue_io_min(sdkp->disk->queue,
2877 get_unaligned_be16(&buffer[6]) * sector_sz);
2879 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2880 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2882 if (buffer[3] == 0x3c) {
2883 unsigned int lba_count, desc_count;
2885 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2890 lba_count = get_unaligned_be32(&buffer[20]);
2891 desc_count = get_unaligned_be32(&buffer[24]);
2893 if (lba_count && desc_count)
2894 sdkp->max_unmap_blocks = lba_count;
2896 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2898 if (buffer[32] & 0x80)
2899 sdkp->unmap_alignment =
2900 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2902 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2904 if (sdkp->max_unmap_blocks)
2905 sd_config_discard(sdkp, SD_LBP_UNMAP);
2907 sd_config_discard(sdkp, SD_LBP_WS16);
2909 } else { /* LBP VPD page tells us what to use */
2910 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2911 sd_config_discard(sdkp, SD_LBP_UNMAP);
2912 else if (sdkp->lbpws)
2913 sd_config_discard(sdkp, SD_LBP_WS16);
2914 else if (sdkp->lbpws10)
2915 sd_config_discard(sdkp, SD_LBP_WS10);
2917 sd_config_discard(sdkp, SD_LBP_DISABLE);
2926 * sd_read_block_characteristics - Query block dev. characteristics
2927 * @sdkp: disk to query
2929 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2931 struct request_queue *q = sdkp->disk->queue;
2932 unsigned char *buffer;
2934 const int vpd_len = 64;
2936 buffer = kmalloc(vpd_len, GFP_KERNEL);
2939 /* Block Device Characteristics VPD */
2940 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2943 rot = get_unaligned_be16(&buffer[4]);
2946 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2947 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2949 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2950 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
2953 if (sdkp->device->type == TYPE_ZBC) {
2955 q->limits.zoned = BLK_ZONED_HM;
2957 sdkp->zoned = (buffer[8] >> 4) & 3;
2958 if (sdkp->zoned == 1)
2960 q->limits.zoned = BLK_ZONED_HA;
2963 * Treat drive-managed devices as
2964 * regular block devices.
2966 q->limits.zoned = BLK_ZONED_NONE;
2968 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2969 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2970 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2977 * sd_read_block_provisioning - Query provisioning VPD page
2978 * @sdkp: disk to query
2980 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2982 unsigned char *buffer;
2983 const int vpd_len = 8;
2985 if (sdkp->lbpme == 0)
2988 buffer = kmalloc(vpd_len, GFP_KERNEL);
2990 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2994 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2995 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2996 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3002 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3004 struct scsi_device *sdev = sdkp->device;
3006 if (sdev->host->no_write_same) {
3007 sdev->no_write_same = 1;
3012 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3013 /* too large values might cause issues with arcmsr */
3014 int vpd_buf_len = 64;
3016 sdev->no_report_opcodes = 1;
3018 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3019 * CODES is unsupported and the device has an ATA
3020 * Information VPD page (SAT).
3022 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3023 sdev->no_write_same = 1;
3026 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3029 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3033 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3035 struct scsi_device *sdev = sdkp->device;
3037 if (!sdev->security_supported)
3040 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3041 SECURITY_PROTOCOL_IN) == 1 &&
3042 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3043 SECURITY_PROTOCOL_OUT) == 1)
3048 * sd_revalidate_disk - called the first time a new disk is seen,
3049 * performs disk spin up, read_capacity, etc.
3050 * @disk: struct gendisk we care about
3052 static int sd_revalidate_disk(struct gendisk *disk)
3054 struct scsi_disk *sdkp = scsi_disk(disk);
3055 struct scsi_device *sdp = sdkp->device;
3056 struct request_queue *q = sdkp->disk->queue;
3057 sector_t old_capacity = sdkp->capacity;
3058 unsigned char *buffer;
3059 unsigned int dev_max, rw_max;
3061 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3062 "sd_revalidate_disk\n"));
3065 * If the device is offline, don't try and read capacity or any
3066 * of the other niceties.
3068 if (!scsi_device_online(sdp))
3071 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3073 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3074 "allocation failure.\n");
3078 sd_spinup_disk(sdkp);
3081 * Without media there is no reason to ask; moreover, some devices
3082 * react badly if we do.
3084 if (sdkp->media_present) {
3085 sd_read_capacity(sdkp, buffer);
3087 if (scsi_device_supports_vpd(sdp)) {
3088 sd_read_block_provisioning(sdkp);
3089 sd_read_block_limits(sdkp);
3090 sd_read_block_characteristics(sdkp);
3091 sd_zbc_read_zones(sdkp, buffer);
3094 sd_print_capacity(sdkp, old_capacity);
3096 sd_read_write_protect_flag(sdkp, buffer);
3097 sd_read_cache_type(sdkp, buffer);
3098 sd_read_app_tag_own(sdkp, buffer);
3099 sd_read_write_same(sdkp, buffer);
3100 sd_read_security(sdkp, buffer);
3104 * We now have all cache related info, determine how we deal
3105 * with flush requests.
3107 sd_set_flush_flag(sdkp);
3109 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3110 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3112 /* Some devices report a maximum block count for READ/WRITE requests. */
3113 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3114 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3117 * Determine the device's preferred I/O size for reads and writes
3118 * unless the reported value is unreasonably small, large, or
3121 if (sdkp->opt_xfer_blocks &&
3122 sdkp->opt_xfer_blocks <= dev_max &&
3123 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3124 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3125 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3126 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3128 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3129 (sector_t)BLK_DEF_MAX_SECTORS);
3131 /* Do not exceed controller limit */
3132 rw_max = min(rw_max, queue_max_hw_sectors(q));
3135 * Only update max_sectors if previously unset or if the current value
3136 * exceeds the capabilities of the hardware.
3138 if (sdkp->first_scan ||
3139 q->limits.max_sectors > q->limits.max_dev_sectors ||
3140 q->limits.max_sectors > q->limits.max_hw_sectors)
3141 q->limits.max_sectors = rw_max;
3143 sdkp->first_scan = 0;
3145 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3146 sd_config_write_same(sdkp);
3154 * sd_unlock_native_capacity - unlock native capacity
3155 * @disk: struct gendisk to set capacity for
3157 * Block layer calls this function if it detects that partitions
3158 * on @disk reach beyond the end of the device. If the SCSI host
3159 * implements ->unlock_native_capacity() method, it's invoked to
3160 * give it a chance to adjust the device capacity.
3163 * Defined by block layer. Might sleep.
3165 static void sd_unlock_native_capacity(struct gendisk *disk)
3167 struct scsi_device *sdev = scsi_disk(disk)->device;
3169 if (sdev->host->hostt->unlock_native_capacity)
3170 sdev->host->hostt->unlock_native_capacity(sdev);
3174 * sd_format_disk_name - format disk name
3175 * @prefix: name prefix - ie. "sd" for SCSI disks
3176 * @index: index of the disk to format name for
3177 * @buf: output buffer
3178 * @buflen: length of the output buffer
3180 * SCSI disk names starts at sda. The 26th device is sdz and the
3181 * 27th is sdaa. The last one for two lettered suffix is sdzz
3182 * which is followed by sdaaa.
3184 * This is basically 26 base counting with one extra 'nil' entry
3185 * at the beginning from the second digit on and can be
3186 * determined using similar method as 26 base conversion with the
3187 * index shifted -1 after each digit is computed.
3193 * 0 on success, -errno on failure.
3195 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3197 const int base = 'z' - 'a' + 1;
3198 char *begin = buf + strlen(prefix);
3199 char *end = buf + buflen;
3209 *--p = 'a' + (index % unit);
3210 index = (index / unit) - 1;
3211 } while (index >= 0);
3213 memmove(begin, p, end - p);
3214 memcpy(buf, prefix, strlen(prefix));
3220 * The asynchronous part of sd_probe
3222 static void sd_probe_async(void *data, async_cookie_t cookie)
3224 struct scsi_disk *sdkp = data;
3225 struct scsi_device *sdp;
3232 index = sdkp->index;
3233 dev = &sdp->sdev_gendev;
3235 gd->major = sd_major((index & 0xf0) >> 4);
3236 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3238 gd->fops = &sd_fops;
3239 gd->private_data = &sdkp->driver;
3240 gd->queue = sdkp->device->request_queue;
3242 /* defaults, until the device tells us otherwise */
3243 sdp->sector_size = 512;
3245 sdkp->media_present = 1;
3246 sdkp->write_prot = 0;
3247 sdkp->cache_override = 0;
3251 sdkp->first_scan = 1;
3252 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3254 sd_revalidate_disk(gd);
3256 gd->flags = GENHD_FL_EXT_DEVT;
3257 if (sdp->removable) {
3258 gd->flags |= GENHD_FL_REMOVABLE;
3259 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3262 blk_pm_runtime_init(sdp->request_queue, dev);
3263 device_add_disk(dev, gd, NULL);
3265 sd_dif_config_host(sdkp);
3267 sd_revalidate_disk(gd);
3269 if (sdkp->security) {
3270 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3272 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3275 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3276 sdp->removable ? "removable " : "");
3277 scsi_autopm_put_device(sdp);
3278 put_device(&sdkp->dev);
3282 * sd_probe - called during driver initialization and whenever a
3283 * new scsi device is attached to the system. It is called once
3284 * for each scsi device (not just disks) present.
3285 * @dev: pointer to device object
3287 * Returns 0 if successful (or not interested in this scsi device
3288 * (e.g. scanner)); 1 when there is an error.
3290 * Note: this function is invoked from the scsi mid-level.
3291 * This function sets up the mapping between a given
3292 * <host,channel,id,lun> (found in sdp) and new device name
3293 * (e.g. /dev/sda). More precisely it is the block device major
3294 * and minor number that is chosen here.
3296 * Assume sd_probe is not re-entrant (for time being)
3297 * Also think about sd_probe() and sd_remove() running coincidentally.
3299 static int sd_probe(struct device *dev)
3301 struct scsi_device *sdp = to_scsi_device(dev);
3302 struct scsi_disk *sdkp;
3307 scsi_autopm_get_device(sdp);
3309 if (sdp->type != TYPE_DISK &&
3310 sdp->type != TYPE_ZBC &&
3311 sdp->type != TYPE_MOD &&
3312 sdp->type != TYPE_RBC)
3315 #ifndef CONFIG_BLK_DEV_ZONED
3316 if (sdp->type == TYPE_ZBC)
3319 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3323 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3327 gd = alloc_disk(SD_MINORS);
3331 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3333 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3337 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3339 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3340 goto out_free_index;
3344 sdkp->driver = &sd_template;
3346 sdkp->index = index;
3347 atomic_set(&sdkp->openers, 0);
3348 atomic_set(&sdkp->device->ioerr_cnt, 0);
3350 if (!sdp->request_queue->rq_timeout) {
3351 if (sdp->type != TYPE_MOD)
3352 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3354 blk_queue_rq_timeout(sdp->request_queue,
3358 device_initialize(&sdkp->dev);
3359 sdkp->dev.parent = dev;
3360 sdkp->dev.class = &sd_disk_class;
3361 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3363 error = device_add(&sdkp->dev);
3365 goto out_free_index;
3368 dev_set_drvdata(dev, sdkp);
3370 get_device(&sdkp->dev); /* prevent release before async_schedule */
3371 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3376 ida_free(&sd_index_ida, index);
3382 scsi_autopm_put_device(sdp);
3387 * sd_remove - called whenever a scsi disk (previously recognized by
3388 * sd_probe) is detached from the system. It is called (potentially
3389 * multiple times) during sd module unload.
3390 * @dev: pointer to device object
3392 * Note: this function is invoked from the scsi mid-level.
3393 * This function potentially frees up a device name (e.g. /dev/sdc)
3394 * that could be re-used by a subsequent sd_probe().
3395 * This function is not called when the built-in sd driver is "exit-ed".
3397 static int sd_remove(struct device *dev)
3399 struct scsi_disk *sdkp;
3402 sdkp = dev_get_drvdata(dev);
3403 devt = disk_devt(sdkp->disk);
3404 scsi_autopm_get_device(sdkp->device);
3406 async_synchronize_full_domain(&scsi_sd_pm_domain);
3407 async_synchronize_full_domain(&scsi_sd_probe_domain);
3408 device_del(&sdkp->dev);
3409 del_gendisk(sdkp->disk);
3412 free_opal_dev(sdkp->opal_dev);
3414 blk_register_region(devt, SD_MINORS, NULL,
3415 sd_default_probe, NULL, NULL);
3417 mutex_lock(&sd_ref_mutex);
3418 dev_set_drvdata(dev, NULL);
3419 put_device(&sdkp->dev);
3420 mutex_unlock(&sd_ref_mutex);
3426 * scsi_disk_release - Called to free the scsi_disk structure
3427 * @dev: pointer to embedded class device
3429 * sd_ref_mutex must be held entering this routine. Because it is
3430 * called on last put, you should always use the scsi_disk_get()
3431 * scsi_disk_put() helpers which manipulate the semaphore directly
3432 * and never do a direct put_device.
3434 static void scsi_disk_release(struct device *dev)
3436 struct scsi_disk *sdkp = to_scsi_disk(dev);
3437 struct gendisk *disk = sdkp->disk;
3439 ida_free(&sd_index_ida, sdkp->index);
3441 disk->private_data = NULL;
3443 put_device(&sdkp->device->sdev_gendev);
3448 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3450 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3451 struct scsi_sense_hdr sshdr;
3452 struct scsi_device *sdp = sdkp->device;
3456 cmd[4] |= 1; /* START */
3458 if (sdp->start_stop_pwr_cond)
3459 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3461 if (!scsi_device_online(sdp))
3464 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3465 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3467 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3468 if (driver_byte(res) == DRIVER_SENSE)
3469 sd_print_sense_hdr(sdkp, &sshdr);
3470 if (scsi_sense_valid(&sshdr) &&
3471 /* 0x3a is medium not present */
3476 /* SCSI error codes must not go to the generic layer */
3484 * Send a SYNCHRONIZE CACHE instruction down to the device through
3485 * the normal SCSI command structure. Wait for the command to
3488 static void sd_shutdown(struct device *dev)
3490 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3493 return; /* this can happen */
3495 if (pm_runtime_suspended(dev))
3498 if (sdkp->WCE && sdkp->media_present) {
3499 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3500 sd_sync_cache(sdkp, NULL);
3503 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3504 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3505 sd_start_stop_device(sdkp, 0);
3509 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3511 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3512 struct scsi_sense_hdr sshdr;
3515 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3518 if (sdkp->WCE && sdkp->media_present) {
3519 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3520 ret = sd_sync_cache(sdkp, &sshdr);
3523 /* ignore OFFLINE device */
3527 if (!scsi_sense_valid(&sshdr) ||
3528 sshdr.sense_key != ILLEGAL_REQUEST)
3532 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3533 * doesn't support sync. There's not much to do and
3534 * suspend shouldn't fail.
3540 if (sdkp->device->manage_start_stop) {
3541 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3542 /* an error is not worth aborting a system sleep */
3543 ret = sd_start_stop_device(sdkp, 0);
3544 if (ignore_stop_errors)
3551 static int sd_suspend_system(struct device *dev)
3553 return sd_suspend_common(dev, true);
3556 static int sd_suspend_runtime(struct device *dev)
3558 return sd_suspend_common(dev, false);
3561 static int sd_resume(struct device *dev)
3563 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3566 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3569 if (!sdkp->device->manage_start_stop)
3572 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3573 ret = sd_start_stop_device(sdkp, 1);
3575 opal_unlock_from_suspend(sdkp->opal_dev);
3580 * init_sd - entry point for this driver (both when built in or when
3583 * Note: this function registers this driver with the scsi mid-level.
3585 static int __init init_sd(void)
3587 int majors = 0, i, err;
3589 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3591 for (i = 0; i < SD_MAJORS; i++) {
3592 if (register_blkdev(sd_major(i), "sd") != 0)
3595 blk_register_region(sd_major(i), SD_MINORS, NULL,
3596 sd_default_probe, NULL, NULL);
3602 err = class_register(&sd_disk_class);
3606 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3608 if (!sd_cdb_cache) {
3609 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3614 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3616 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3621 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3622 if (!sd_page_pool) {
3623 printk(KERN_ERR "sd: can't init discard page pool\n");
3628 err = scsi_register_driver(&sd_template.gendrv);
3630 goto err_out_driver;
3635 mempool_destroy(sd_page_pool);
3638 mempool_destroy(sd_cdb_pool);
3641 kmem_cache_destroy(sd_cdb_cache);
3644 class_unregister(&sd_disk_class);
3646 for (i = 0; i < SD_MAJORS; i++)
3647 unregister_blkdev(sd_major(i), "sd");
3652 * exit_sd - exit point for this driver (when it is a module).
3654 * Note: this function unregisters this driver from the scsi mid-level.
3656 static void __exit exit_sd(void)
3660 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3662 scsi_unregister_driver(&sd_template.gendrv);
3663 mempool_destroy(sd_cdb_pool);
3664 mempool_destroy(sd_page_pool);
3665 kmem_cache_destroy(sd_cdb_cache);
3667 class_unregister(&sd_disk_class);
3669 for (i = 0; i < SD_MAJORS; i++) {
3670 blk_unregister_region(sd_major(i), SD_MINORS);
3671 unregister_blkdev(sd_major(i), "sd");
3675 module_init(init_sd);
3676 module_exit(exit_sd);
3678 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3679 struct scsi_sense_hdr *sshdr)
3681 scsi_print_sense_hdr(sdkp->device,
3682 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3685 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3688 const char *hb_string = scsi_hostbyte_string(result);
3689 const char *db_string = scsi_driverbyte_string(result);
3691 if (hb_string || db_string)
3692 sd_printk(KERN_INFO, sdkp,
3693 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3694 hb_string ? hb_string : "invalid",
3695 db_string ? db_string : "invalid");
3697 sd_printk(KERN_INFO, sdkp,
3698 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3699 msg, host_byte(result), driver_byte(result));