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;
137 static const char *sd_cache_types[] = {
138 "write through", "none", "write back",
139 "write back, no read (daft)"
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 bool wc = false, fua = false;
152 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 const char *buf, size_t count)
159 int ct, rcd, wce, sp;
160 struct scsi_disk *sdkp = to_scsi_disk(dev);
161 struct scsi_device *sdp = sdkp->device;
164 struct scsi_mode_data data;
165 struct scsi_sense_hdr sshdr;
166 static const char temp[] = "temporary ";
169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 /* no cache control on RBC devices; theoretically they
171 * can do it, but there's probably so many exceptions
172 * it's not worth the risk */
175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 buf += sizeof(temp) - 1;
177 sdkp->cache_override = 1;
179 sdkp->cache_override = 0;
182 ct = sysfs_match_string(sd_cache_types, buf);
186 rcd = ct & 0x01 ? 1 : 0;
187 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
189 if (sdkp->cache_override) {
192 sd_set_flush_flag(sdkp);
196 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 SD_MAX_RETRIES, &data, NULL))
199 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 data.block_descriptor_length);
201 buffer_data = buffer + data.header_length +
202 data.block_descriptor_length;
203 buffer_data[2] &= ~0x05;
204 buffer_data[2] |= wce << 2 | rcd;
205 sp = buffer_data[0] & 0x80 ? 1 : 0;
206 buffer_data[0] &= ~0x80;
208 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209 SD_MAX_RETRIES, &data, &sshdr)) {
210 if (scsi_sense_valid(&sshdr))
211 sd_print_sense_hdr(sdkp, &sshdr);
214 revalidate_disk(sdkp->disk);
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
222 struct scsi_disk *sdkp = to_scsi_disk(dev);
223 struct scsi_device *sdp = sdkp->device;
225 return sprintf(buf, "%u\n", sdp->manage_start_stop);
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230 const char *buf, size_t count)
232 struct scsi_disk *sdkp = to_scsi_disk(dev);
233 struct scsi_device *sdp = sdkp->device;
236 if (!capable(CAP_SYS_ADMIN))
239 if (kstrtobool(buf, &v))
242 sdp->manage_start_stop = v;
246 static DEVICE_ATTR_RW(manage_start_stop);
249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
251 struct scsi_disk *sdkp = to_scsi_disk(dev);
253 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
257 allow_restart_store(struct device *dev, struct device_attribute *attr,
258 const char *buf, size_t count)
261 struct scsi_disk *sdkp = to_scsi_disk(dev);
262 struct scsi_device *sdp = sdkp->device;
264 if (!capable(CAP_SYS_ADMIN))
267 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
270 if (kstrtobool(buf, &v))
273 sdp->allow_restart = v;
277 static DEVICE_ATTR_RW(allow_restart);
280 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
282 struct scsi_disk *sdkp = to_scsi_disk(dev);
283 int ct = sdkp->RCD + 2*sdkp->WCE;
285 return sprintf(buf, "%s\n", sd_cache_types[ct]);
287 static DEVICE_ATTR_RW(cache_type);
290 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
292 struct scsi_disk *sdkp = to_scsi_disk(dev);
294 return sprintf(buf, "%u\n", sdkp->DPOFUA);
296 static DEVICE_ATTR_RO(FUA);
299 protection_type_show(struct device *dev, struct device_attribute *attr,
302 struct scsi_disk *sdkp = to_scsi_disk(dev);
304 return sprintf(buf, "%u\n", sdkp->protection_type);
308 protection_type_store(struct device *dev, struct device_attribute *attr,
309 const char *buf, size_t count)
311 struct scsi_disk *sdkp = to_scsi_disk(dev);
315 if (!capable(CAP_SYS_ADMIN))
318 err = kstrtouint(buf, 10, &val);
323 if (val <= T10_PI_TYPE3_PROTECTION)
324 sdkp->protection_type = val;
328 static DEVICE_ATTR_RW(protection_type);
331 protection_mode_show(struct device *dev, struct device_attribute *attr,
334 struct scsi_disk *sdkp = to_scsi_disk(dev);
335 struct scsi_device *sdp = sdkp->device;
336 unsigned int dif, dix;
338 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
339 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
341 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
347 return sprintf(buf, "none\n");
349 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
351 static DEVICE_ATTR_RO(protection_mode);
354 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
356 struct scsi_disk *sdkp = to_scsi_disk(dev);
358 return sprintf(buf, "%u\n", sdkp->ATO);
360 static DEVICE_ATTR_RO(app_tag_own);
363 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
366 struct scsi_disk *sdkp = to_scsi_disk(dev);
368 return sprintf(buf, "%u\n", sdkp->lbpme);
370 static DEVICE_ATTR_RO(thin_provisioning);
372 /* sysfs_match_string() requires dense arrays */
373 static const char *lbp_mode[] = {
374 [SD_LBP_FULL] = "full",
375 [SD_LBP_UNMAP] = "unmap",
376 [SD_LBP_WS16] = "writesame_16",
377 [SD_LBP_WS10] = "writesame_10",
378 [SD_LBP_ZERO] = "writesame_zero",
379 [SD_LBP_DISABLE] = "disabled",
383 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
386 struct scsi_disk *sdkp = to_scsi_disk(dev);
388 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
392 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
393 const char *buf, size_t count)
395 struct scsi_disk *sdkp = to_scsi_disk(dev);
396 struct scsi_device *sdp = sdkp->device;
399 if (!capable(CAP_SYS_ADMIN))
402 if (sd_is_zoned(sdkp)) {
403 sd_config_discard(sdkp, SD_LBP_DISABLE);
407 if (sdp->type != TYPE_DISK)
410 mode = sysfs_match_string(lbp_mode, buf);
414 sd_config_discard(sdkp, mode);
418 static DEVICE_ATTR_RW(provisioning_mode);
420 /* sysfs_match_string() requires dense arrays */
421 static const char *zeroing_mode[] = {
422 [SD_ZERO_WRITE] = "write",
423 [SD_ZERO_WS] = "writesame",
424 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
425 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
429 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
432 struct scsi_disk *sdkp = to_scsi_disk(dev);
434 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
438 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
439 const char *buf, size_t count)
441 struct scsi_disk *sdkp = to_scsi_disk(dev);
444 if (!capable(CAP_SYS_ADMIN))
447 mode = sysfs_match_string(zeroing_mode, buf);
451 sdkp->zeroing_mode = mode;
455 static DEVICE_ATTR_RW(zeroing_mode);
458 max_medium_access_timeouts_show(struct device *dev,
459 struct device_attribute *attr, char *buf)
461 struct scsi_disk *sdkp = to_scsi_disk(dev);
463 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
467 max_medium_access_timeouts_store(struct device *dev,
468 struct device_attribute *attr, const char *buf,
471 struct scsi_disk *sdkp = to_scsi_disk(dev);
474 if (!capable(CAP_SYS_ADMIN))
477 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
479 return err ? err : count;
481 static DEVICE_ATTR_RW(max_medium_access_timeouts);
484 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
487 struct scsi_disk *sdkp = to_scsi_disk(dev);
489 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
493 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
494 const char *buf, size_t count)
496 struct scsi_disk *sdkp = to_scsi_disk(dev);
497 struct scsi_device *sdp = sdkp->device;
501 if (!capable(CAP_SYS_ADMIN))
504 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
507 err = kstrtoul(buf, 10, &max);
513 sdp->no_write_same = 1;
514 else if (max <= SD_MAX_WS16_BLOCKS) {
515 sdp->no_write_same = 0;
516 sdkp->max_ws_blocks = max;
519 sd_config_write_same(sdkp);
523 static DEVICE_ATTR_RW(max_write_same_blocks);
525 static struct attribute *sd_disk_attrs[] = {
526 &dev_attr_cache_type.attr,
528 &dev_attr_allow_restart.attr,
529 &dev_attr_manage_start_stop.attr,
530 &dev_attr_protection_type.attr,
531 &dev_attr_protection_mode.attr,
532 &dev_attr_app_tag_own.attr,
533 &dev_attr_thin_provisioning.attr,
534 &dev_attr_provisioning_mode.attr,
535 &dev_attr_zeroing_mode.attr,
536 &dev_attr_max_write_same_blocks.attr,
537 &dev_attr_max_medium_access_timeouts.attr,
540 ATTRIBUTE_GROUPS(sd_disk);
542 static struct class sd_disk_class = {
544 .owner = THIS_MODULE,
545 .dev_release = scsi_disk_release,
546 .dev_groups = sd_disk_groups,
549 static const struct dev_pm_ops sd_pm_ops = {
550 .suspend = sd_suspend_system,
552 .poweroff = sd_suspend_system,
553 .restore = sd_resume,
554 .runtime_suspend = sd_suspend_runtime,
555 .runtime_resume = sd_resume,
558 static struct scsi_driver sd_template = {
561 .owner = THIS_MODULE,
564 .shutdown = sd_shutdown,
568 .init_command = sd_init_command,
569 .uninit_command = sd_uninit_command,
571 .eh_action = sd_eh_action,
572 .eh_reset = sd_eh_reset,
576 * Dummy kobj_map->probe function.
577 * The default ->probe function will call modprobe, which is
578 * pointless as this module is already loaded.
580 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
586 * Device no to disk mapping:
588 * major disc2 disc p1
589 * |............|.............|....|....| <- dev_t
592 * Inside a major, we have 16k disks, however mapped non-
593 * contiguously. The first 16 disks are for major0, the next
594 * ones with major1, ... Disk 256 is for major0 again, disk 272
596 * As we stay compatible with our numbering scheme, we can reuse
597 * the well-know SCSI majors 8, 65--71, 136--143.
599 static int sd_major(int major_idx)
603 return SCSI_DISK0_MAJOR;
605 return SCSI_DISK1_MAJOR + major_idx - 1;
607 return SCSI_DISK8_MAJOR + major_idx - 8;
610 return 0; /* shut up gcc */
614 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
616 struct scsi_disk *sdkp = NULL;
618 mutex_lock(&sd_ref_mutex);
620 if (disk->private_data) {
621 sdkp = scsi_disk(disk);
622 if (scsi_device_get(sdkp->device) == 0)
623 get_device(&sdkp->dev);
627 mutex_unlock(&sd_ref_mutex);
631 static void scsi_disk_put(struct scsi_disk *sdkp)
633 struct scsi_device *sdev = sdkp->device;
635 mutex_lock(&sd_ref_mutex);
636 put_device(&sdkp->dev);
637 scsi_device_put(sdev);
638 mutex_unlock(&sd_ref_mutex);
641 #ifdef CONFIG_BLK_SED_OPAL
642 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
643 size_t len, bool send)
645 struct scsi_device *sdev = data;
649 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
651 put_unaligned_be16(spsp, &cdb[2]);
652 put_unaligned_be32(len, &cdb[6]);
654 ret = scsi_execute_req(sdev, cdb,
655 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
656 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
657 return ret <= 0 ? ret : -EIO;
659 #endif /* CONFIG_BLK_SED_OPAL */
661 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
662 unsigned int dix, unsigned int dif)
664 struct bio *bio = scmd->request->bio;
665 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
666 unsigned int protect = 0;
668 if (dix) { /* DIX Type 0, 1, 2, 3 */
669 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
670 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
672 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
673 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
676 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
677 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
679 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
680 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
683 if (dif) { /* DIX/DIF Type 1, 2, 3 */
684 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
686 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
687 protect = 3 << 5; /* Disable target PI checking */
689 protect = 1 << 5; /* Enable target PI checking */
692 scsi_set_prot_op(scmd, prot_op);
693 scsi_set_prot_type(scmd, dif);
694 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
699 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
701 struct request_queue *q = sdkp->disk->queue;
702 unsigned int logical_block_size = sdkp->device->sector_size;
703 unsigned int max_blocks = 0;
705 q->limits.discard_alignment =
706 sdkp->unmap_alignment * logical_block_size;
707 q->limits.discard_granularity =
708 max(sdkp->physical_block_size,
709 sdkp->unmap_granularity * logical_block_size);
710 sdkp->provisioning_mode = mode;
716 blk_queue_max_discard_sectors(q, 0);
717 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
721 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
722 (u32)SD_MAX_WS16_BLOCKS);
726 if (sdkp->device->unmap_limit_for_ws)
727 max_blocks = sdkp->max_unmap_blocks;
729 max_blocks = sdkp->max_ws_blocks;
731 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
735 if (sdkp->device->unmap_limit_for_ws)
736 max_blocks = sdkp->max_unmap_blocks;
738 max_blocks = sdkp->max_ws_blocks;
740 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
744 max_blocks = min_not_zero(sdkp->max_ws_blocks,
745 (u32)SD_MAX_WS10_BLOCKS);
749 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
750 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
753 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
755 struct scsi_device *sdp = cmd->device;
756 struct request *rq = cmd->request;
757 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
758 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
759 unsigned int data_len = 24;
762 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
763 if (!rq->special_vec.bv_page)
764 return BLK_STS_RESOURCE;
765 rq->special_vec.bv_offset = 0;
766 rq->special_vec.bv_len = data_len;
767 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
770 cmd->cmnd[0] = UNMAP;
773 buf = page_address(rq->special_vec.bv_page);
774 put_unaligned_be16(6 + 16, &buf[0]);
775 put_unaligned_be16(16, &buf[2]);
776 put_unaligned_be64(sector, &buf[8]);
777 put_unaligned_be32(nr_sectors, &buf[16]);
779 cmd->allowed = SD_MAX_RETRIES;
780 cmd->transfersize = data_len;
781 rq->timeout = SD_TIMEOUT;
782 scsi_req(rq)->resid_len = data_len;
784 return scsi_init_io(cmd);
787 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
790 struct scsi_device *sdp = cmd->device;
791 struct request *rq = cmd->request;
792 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
793 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
794 u32 data_len = sdp->sector_size;
796 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
797 if (!rq->special_vec.bv_page)
798 return BLK_STS_RESOURCE;
799 rq->special_vec.bv_offset = 0;
800 rq->special_vec.bv_len = data_len;
801 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
804 cmd->cmnd[0] = WRITE_SAME_16;
806 cmd->cmnd[1] = 0x8; /* UNMAP */
807 put_unaligned_be64(sector, &cmd->cmnd[2]);
808 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
810 cmd->allowed = SD_MAX_RETRIES;
811 cmd->transfersize = data_len;
812 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
813 scsi_req(rq)->resid_len = data_len;
815 return scsi_init_io(cmd);
818 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
821 struct scsi_device *sdp = cmd->device;
822 struct request *rq = cmd->request;
823 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
824 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
825 u32 data_len = sdp->sector_size;
827 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
828 if (!rq->special_vec.bv_page)
829 return BLK_STS_RESOURCE;
830 rq->special_vec.bv_offset = 0;
831 rq->special_vec.bv_len = data_len;
832 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
835 cmd->cmnd[0] = WRITE_SAME;
837 cmd->cmnd[1] = 0x8; /* UNMAP */
838 put_unaligned_be32(sector, &cmd->cmnd[2]);
839 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
841 cmd->allowed = SD_MAX_RETRIES;
842 cmd->transfersize = data_len;
843 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
844 scsi_req(rq)->resid_len = data_len;
846 return scsi_init_io(cmd);
849 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
851 struct request *rq = cmd->request;
852 struct scsi_device *sdp = cmd->device;
853 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
854 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
855 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
857 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
858 switch (sdkp->zeroing_mode) {
859 case SD_ZERO_WS16_UNMAP:
860 return sd_setup_write_same16_cmnd(cmd, true);
861 case SD_ZERO_WS10_UNMAP:
862 return sd_setup_write_same10_cmnd(cmd, true);
866 if (sdp->no_write_same)
867 return BLK_STS_TARGET;
869 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
870 return sd_setup_write_same16_cmnd(cmd, false);
872 return sd_setup_write_same10_cmnd(cmd, false);
875 static void sd_config_write_same(struct scsi_disk *sdkp)
877 struct request_queue *q = sdkp->disk->queue;
878 unsigned int logical_block_size = sdkp->device->sector_size;
880 if (sdkp->device->no_write_same) {
881 sdkp->max_ws_blocks = 0;
885 /* Some devices can not handle block counts above 0xffff despite
886 * supporting WRITE SAME(16). Consequently we default to 64k
887 * blocks per I/O unless the device explicitly advertises a
890 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
891 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
892 (u32)SD_MAX_WS16_BLOCKS);
893 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
894 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
895 (u32)SD_MAX_WS10_BLOCKS);
897 sdkp->device->no_write_same = 1;
898 sdkp->max_ws_blocks = 0;
901 if (sdkp->lbprz && sdkp->lbpws)
902 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
903 else if (sdkp->lbprz && sdkp->lbpws10)
904 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
905 else if (sdkp->max_ws_blocks)
906 sdkp->zeroing_mode = SD_ZERO_WS;
908 sdkp->zeroing_mode = SD_ZERO_WRITE;
910 if (sdkp->max_ws_blocks &&
911 sdkp->physical_block_size > logical_block_size) {
913 * Reporting a maximum number of blocks that is not aligned
914 * on the device physical size would cause a large write same
915 * request to be split into physically unaligned chunks by
916 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
917 * even if the caller of these functions took care to align the
918 * large request. So make sure the maximum reported is aligned
919 * to the device physical block size. This is only an optional
920 * optimization for regular disks, but this is mandatory to
921 * avoid failure of large write same requests directed at
922 * sequential write required zones of host-managed ZBC disks.
924 sdkp->max_ws_blocks =
925 round_down(sdkp->max_ws_blocks,
926 bytes_to_logical(sdkp->device,
927 sdkp->physical_block_size));
931 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
932 (logical_block_size >> 9));
933 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
934 (logical_block_size >> 9));
938 * sd_setup_write_same_cmnd - write the same data to multiple blocks
939 * @cmd: command to prepare
941 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
942 * the preference indicated by the target device.
944 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
946 struct request *rq = cmd->request;
947 struct scsi_device *sdp = cmd->device;
948 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
949 struct bio *bio = rq->bio;
950 sector_t sector = blk_rq_pos(rq);
951 unsigned int nr_sectors = blk_rq_sectors(rq);
952 unsigned int nr_bytes = blk_rq_bytes(rq);
955 if (sdkp->device->no_write_same)
956 return BLK_STS_TARGET;
958 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
960 sector >>= ilog2(sdp->sector_size) - 9;
961 nr_sectors >>= ilog2(sdp->sector_size) - 9;
963 rq->timeout = SD_WRITE_SAME_TIMEOUT;
965 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
967 cmd->cmnd[0] = WRITE_SAME_16;
968 put_unaligned_be64(sector, &cmd->cmnd[2]);
969 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
972 cmd->cmnd[0] = WRITE_SAME;
973 put_unaligned_be32(sector, &cmd->cmnd[2]);
974 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
977 cmd->transfersize = sdp->sector_size;
978 cmd->allowed = SD_MAX_RETRIES;
981 * For WRITE SAME the data transferred via the DATA OUT buffer is
982 * different from the amount of data actually written to the target.
984 * We set up __data_len to the amount of data transferred via the
985 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
986 * to transfer a single sector of data first, but then reset it to
987 * the amount of data to be written right after so that the I/O path
988 * knows how much to actually write.
990 rq->__data_len = sdp->sector_size;
991 ret = scsi_init_io(cmd);
992 rq->__data_len = nr_bytes;
997 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
999 struct request *rq = cmd->request;
1001 /* flush requests don't perform I/O, zero the S/G table */
1002 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1004 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1006 cmd->transfersize = 0;
1007 cmd->allowed = SD_MAX_RETRIES;
1009 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1013 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1015 struct request *rq = SCpnt->request;
1016 struct scsi_device *sdp = SCpnt->device;
1017 struct gendisk *disk = rq->rq_disk;
1018 struct scsi_disk *sdkp = scsi_disk(disk);
1019 sector_t block = blk_rq_pos(rq);
1021 unsigned int this_count = blk_rq_sectors(rq);
1022 unsigned int dif, dix;
1023 unsigned char protect;
1026 ret = scsi_init_io(SCpnt);
1027 if (ret != BLK_STS_OK)
1029 WARN_ON_ONCE(SCpnt != rq->special);
1032 scmd_printk(KERN_INFO, SCpnt,
1033 "%s: block=%llu, count=%d\n",
1034 __func__, (unsigned long long)block, this_count));
1036 if (!sdp || !scsi_device_online(sdp) ||
1037 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1038 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1039 "Finishing %u sectors\n",
1040 blk_rq_sectors(rq)));
1041 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1042 "Retry with 0x%p\n", SCpnt));
1043 return BLK_STS_IOERR;
1048 * quietly refuse to do anything to a changed disc until
1049 * the changed bit has been reset
1051 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1052 return BLK_STS_IOERR;
1056 * Some SD card readers can't handle multi-sector accesses which touch
1057 * the last one or two hardware sectors. Split accesses as needed.
1059 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1060 (sdp->sector_size / 512);
1062 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1063 if (block < threshold) {
1064 /* Access up to the threshold but not beyond */
1065 this_count = threshold - block;
1067 /* Access only a single hardware sector */
1068 this_count = sdp->sector_size / 512;
1072 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1073 (unsigned long long)block));
1076 * If we have a 1K hardware sectorsize, prevent access to single
1077 * 512 byte sectors. In theory we could handle this - in fact
1078 * the scsi cdrom driver must be able to handle this because
1079 * we typically use 1K blocksizes, and cdroms typically have
1080 * 2K hardware sectorsizes. Of course, things are simpler
1081 * with the cdrom, since it is read-only. For performance
1082 * reasons, the filesystems should be able to handle this
1083 * and not force the scsi disk driver to use bounce buffers
1086 if (sdp->sector_size == 1024) {
1087 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1088 scmd_printk(KERN_ERR, SCpnt,
1089 "Bad block number requested\n");
1090 return BLK_STS_IOERR;
1093 this_count = this_count >> 1;
1095 if (sdp->sector_size == 2048) {
1096 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1097 scmd_printk(KERN_ERR, SCpnt,
1098 "Bad block number requested\n");
1099 return BLK_STS_IOERR;
1102 this_count = this_count >> 2;
1104 if (sdp->sector_size == 4096) {
1105 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1106 scmd_printk(KERN_ERR, SCpnt,
1107 "Bad block number requested\n");
1108 return BLK_STS_IOERR;
1111 this_count = this_count >> 3;
1113 if (rq_data_dir(rq) == WRITE) {
1114 SCpnt->cmnd[0] = WRITE_6;
1116 if (blk_integrity_rq(rq))
1117 t10_pi_prepare(SCpnt->request, sdkp->protection_type);
1119 } else if (rq_data_dir(rq) == READ) {
1120 SCpnt->cmnd[0] = READ_6;
1122 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1123 return BLK_STS_IOERR;
1126 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1127 "%s %d/%u 512 byte blocks.\n",
1128 (rq_data_dir(rq) == WRITE) ?
1129 "writing" : "reading", this_count,
1130 blk_rq_sectors(rq)));
1132 dix = scsi_prot_sg_count(SCpnt);
1133 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1136 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1140 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1141 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1143 if (unlikely(!SCpnt->cmnd))
1144 return BLK_STS_RESOURCE;
1146 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1147 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1148 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1149 SCpnt->cmnd[7] = 0x18;
1150 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1151 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1154 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1155 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1156 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1157 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1158 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1159 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1160 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1161 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1163 /* Expected Indirect LBA */
1164 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1165 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1166 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1167 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1169 /* Transfer length */
1170 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1171 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1172 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1173 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1174 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1175 SCpnt->cmnd[0] += READ_16 - READ_6;
1176 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1177 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1178 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1179 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1180 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1181 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1182 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1183 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1184 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1185 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1186 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1187 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1188 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1189 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1190 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1191 scsi_device_protection(SCpnt->device) ||
1192 SCpnt->device->use_10_for_rw) {
1193 SCpnt->cmnd[0] += READ_10 - READ_6;
1194 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1195 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1196 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1197 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1198 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1199 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1200 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1201 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1203 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1205 * This happens only if this drive failed
1206 * 10byte rw command with ILLEGAL_REQUEST
1207 * during operation and thus turned off
1210 scmd_printk(KERN_ERR, SCpnt,
1211 "FUA write on READ/WRITE(6) drive\n");
1212 return BLK_STS_IOERR;
1215 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1216 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1217 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1218 SCpnt->cmnd[4] = (unsigned char) this_count;
1221 SCpnt->sdb.length = this_count * sdp->sector_size;
1224 * We shouldn't disconnect in the middle of a sector, so with a dumb
1225 * host adapter, it's safe to assume that we can at least transfer
1226 * this many bytes between each connect / disconnect.
1228 SCpnt->transfersize = sdp->sector_size;
1229 SCpnt->underflow = this_count << 9;
1230 SCpnt->allowed = SD_MAX_RETRIES;
1233 * This indicates that the command is ready from our end to be
1239 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1241 struct request *rq = cmd->request;
1243 switch (req_op(rq)) {
1244 case REQ_OP_DISCARD:
1245 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1247 return sd_setup_unmap_cmnd(cmd);
1249 return sd_setup_write_same16_cmnd(cmd, true);
1251 return sd_setup_write_same10_cmnd(cmd, true);
1253 return sd_setup_write_same10_cmnd(cmd, false);
1255 return BLK_STS_TARGET;
1257 case REQ_OP_WRITE_ZEROES:
1258 return sd_setup_write_zeroes_cmnd(cmd);
1259 case REQ_OP_WRITE_SAME:
1260 return sd_setup_write_same_cmnd(cmd);
1262 return sd_setup_flush_cmnd(cmd);
1265 return sd_setup_read_write_cmnd(cmd);
1266 case REQ_OP_ZONE_RESET:
1267 return sd_zbc_setup_reset_cmnd(cmd);
1270 return BLK_STS_NOTSUPP;
1274 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1276 struct request *rq = SCpnt->request;
1279 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1280 __free_page(rq->special_vec.bv_page);
1282 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1286 mempool_free(cmnd, sd_cdb_pool);
1291 * sd_open - open a scsi disk device
1292 * @bdev: Block device of the scsi disk to open
1293 * @mode: FMODE_* mask
1295 * Returns 0 if successful. Returns a negated errno value in case
1298 * Note: This can be called from a user context (e.g. fsck(1) )
1299 * or from within the kernel (e.g. as a result of a mount(1) ).
1300 * In the latter case @inode and @filp carry an abridged amount
1301 * of information as noted above.
1303 * Locking: called with bdev->bd_mutex held.
1305 static int sd_open(struct block_device *bdev, fmode_t mode)
1307 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1308 struct scsi_device *sdev;
1314 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1316 sdev = sdkp->device;
1319 * If the device is in error recovery, wait until it is done.
1320 * If the device is offline, then disallow any access to it.
1323 if (!scsi_block_when_processing_errors(sdev))
1326 if (sdev->removable || sdkp->write_prot)
1327 check_disk_change(bdev);
1330 * If the drive is empty, just let the open fail.
1332 retval = -ENOMEDIUM;
1333 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1337 * If the device has the write protect tab set, have the open fail
1338 * if the user expects to be able to write to the thing.
1341 if (sdkp->write_prot && (mode & FMODE_WRITE))
1345 * It is possible that the disk changing stuff resulted in
1346 * the device being taken offline. If this is the case,
1347 * report this to the user, and don't pretend that the
1348 * open actually succeeded.
1351 if (!scsi_device_online(sdev))
1354 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1355 if (scsi_block_when_processing_errors(sdev))
1356 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1362 scsi_disk_put(sdkp);
1367 * sd_release - invoked when the (last) close(2) is called on this
1369 * @disk: disk to release
1370 * @mode: FMODE_* mask
1374 * Note: may block (uninterruptible) if error recovery is underway
1377 * Locking: called with bdev->bd_mutex held.
1379 static void sd_release(struct gendisk *disk, fmode_t mode)
1381 struct scsi_disk *sdkp = scsi_disk(disk);
1382 struct scsi_device *sdev = sdkp->device;
1384 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1386 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1387 if (scsi_block_when_processing_errors(sdev))
1388 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1392 * XXX and what if there are packets in flight and this close()
1393 * XXX is followed by a "rmmod sd_mod"?
1396 scsi_disk_put(sdkp);
1399 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1401 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1402 struct scsi_device *sdp = sdkp->device;
1403 struct Scsi_Host *host = sdp->host;
1404 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1407 /* default to most commonly used values */
1408 diskinfo[0] = 0x40; /* 1 << 6 */
1409 diskinfo[1] = 0x20; /* 1 << 5 */
1410 diskinfo[2] = capacity >> 11;
1412 /* override with calculated, extended default, or driver values */
1413 if (host->hostt->bios_param)
1414 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1416 scsicam_bios_param(bdev, capacity, diskinfo);
1418 geo->heads = diskinfo[0];
1419 geo->sectors = diskinfo[1];
1420 geo->cylinders = diskinfo[2];
1425 * sd_ioctl - process an ioctl
1426 * @bdev: target block device
1427 * @mode: FMODE_* mask
1428 * @cmd: ioctl command number
1429 * @arg: this is third argument given to ioctl(2) system call.
1430 * Often contains a pointer.
1432 * Returns 0 if successful (some ioctls return positive numbers on
1433 * success as well). Returns a negated errno value in case of error.
1435 * Note: most ioctls are forward onto the block subsystem or further
1436 * down in the scsi subsystem.
1438 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1439 unsigned int cmd, unsigned long arg)
1441 struct gendisk *disk = bdev->bd_disk;
1442 struct scsi_disk *sdkp = scsi_disk(disk);
1443 struct scsi_device *sdp = sdkp->device;
1444 void __user *p = (void __user *)arg;
1447 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1448 "cmd=0x%x\n", disk->disk_name, cmd));
1450 error = scsi_verify_blk_ioctl(bdev, cmd);
1455 * If we are in the middle of error recovery, don't let anyone
1456 * else try and use this device. Also, if error recovery fails, it
1457 * may try and take the device offline, in which case all further
1458 * access to the device is prohibited.
1460 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1461 (mode & FMODE_NDELAY) != 0);
1465 if (is_sed_ioctl(cmd))
1466 return sed_ioctl(sdkp->opal_dev, cmd, p);
1469 * Send SCSI addressing ioctls directly to mid level, send other
1470 * ioctls to block level and then onto mid level if they can't be
1474 case SCSI_IOCTL_GET_IDLUN:
1475 case SCSI_IOCTL_GET_BUS_NUMBER:
1476 error = scsi_ioctl(sdp, cmd, p);
1479 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1480 if (error != -ENOTTY)
1482 error = scsi_ioctl(sdp, cmd, p);
1489 static void set_media_not_present(struct scsi_disk *sdkp)
1491 if (sdkp->media_present)
1492 sdkp->device->changed = 1;
1494 if (sdkp->device->removable) {
1495 sdkp->media_present = 0;
1500 static int media_not_present(struct scsi_disk *sdkp,
1501 struct scsi_sense_hdr *sshdr)
1503 if (!scsi_sense_valid(sshdr))
1506 /* not invoked for commands that could return deferred errors */
1507 switch (sshdr->sense_key) {
1508 case UNIT_ATTENTION:
1510 /* medium not present */
1511 if (sshdr->asc == 0x3A) {
1512 set_media_not_present(sdkp);
1520 * sd_check_events - check media events
1521 * @disk: kernel device descriptor
1522 * @clearing: disk events currently being cleared
1524 * Returns mask of DISK_EVENT_*.
1526 * Note: this function is invoked from the block subsystem.
1528 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1530 struct scsi_disk *sdkp = scsi_disk_get(disk);
1531 struct scsi_device *sdp;
1538 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1541 * If the device is offline, don't send any commands - just pretend as
1542 * if the command failed. If the device ever comes back online, we
1543 * can deal with it then. It is only because of unrecoverable errors
1544 * that we would ever take a device offline in the first place.
1546 if (!scsi_device_online(sdp)) {
1547 set_media_not_present(sdkp);
1552 * Using TEST_UNIT_READY enables differentiation between drive with
1553 * no cartridge loaded - NOT READY, drive with changed cartridge -
1554 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1556 * Drives that auto spin down. eg iomega jaz 1G, will be started
1557 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1558 * sd_revalidate() is called.
1560 if (scsi_block_when_processing_errors(sdp)) {
1561 struct scsi_sense_hdr sshdr = { 0, };
1563 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1566 /* failed to execute TUR, assume media not present */
1567 if (host_byte(retval)) {
1568 set_media_not_present(sdkp);
1572 if (media_not_present(sdkp, &sshdr))
1577 * For removable scsi disk we have to recognise the presence
1578 * of a disk in the drive.
1580 if (!sdkp->media_present)
1582 sdkp->media_present = 1;
1585 * sdp->changed is set under the following conditions:
1587 * Medium present state has changed in either direction.
1588 * Device has indicated UNIT_ATTENTION.
1590 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1592 scsi_disk_put(sdkp);
1596 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1599 struct scsi_device *sdp = sdkp->device;
1600 const int timeout = sdp->request_queue->rq_timeout
1601 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1602 struct scsi_sense_hdr my_sshdr;
1604 if (!scsi_device_online(sdp))
1607 /* caller might not be interested in sense, but we need it */
1611 for (retries = 3; retries > 0; --retries) {
1612 unsigned char cmd[10] = { 0 };
1614 cmd[0] = SYNCHRONIZE_CACHE;
1616 * Leave the rest of the command zero to indicate
1619 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1620 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1626 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1628 if (driver_byte(res) == DRIVER_SENSE)
1629 sd_print_sense_hdr(sdkp, sshdr);
1631 /* we need to evaluate the error return */
1632 if (scsi_sense_valid(sshdr) &&
1633 (sshdr->asc == 0x3a || /* medium not present */
1634 sshdr->asc == 0x20)) /* invalid command */
1635 /* this is no error here */
1638 switch (host_byte(res)) {
1639 /* ignore errors due to racing a disconnection */
1640 case DID_BAD_TARGET:
1641 case DID_NO_CONNECT:
1643 /* signal the upper layer it might try again */
1647 case DID_SOFT_ERROR:
1656 static void sd_rescan(struct device *dev)
1658 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1660 revalidate_disk(sdkp->disk);
1664 #ifdef CONFIG_COMPAT
1666 * This gets directly called from VFS. When the ioctl
1667 * is not recognized we go back to the other translation paths.
1669 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1670 unsigned int cmd, unsigned long arg)
1672 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1675 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1676 (mode & FMODE_NDELAY) != 0);
1681 * Let the static ioctl translation table take care of it.
1683 if (!sdev->host->hostt->compat_ioctl)
1684 return -ENOIOCTLCMD;
1685 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1689 static char sd_pr_type(enum pr_type type)
1692 case PR_WRITE_EXCLUSIVE:
1694 case PR_EXCLUSIVE_ACCESS:
1696 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1698 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1700 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1702 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1709 static int sd_pr_command(struct block_device *bdev, u8 sa,
1710 u64 key, u64 sa_key, u8 type, u8 flags)
1712 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1713 struct scsi_sense_hdr sshdr;
1715 u8 cmd[16] = { 0, };
1716 u8 data[24] = { 0, };
1718 cmd[0] = PERSISTENT_RESERVE_OUT;
1721 put_unaligned_be32(sizeof(data), &cmd[5]);
1723 put_unaligned_be64(key, &data[0]);
1724 put_unaligned_be64(sa_key, &data[8]);
1727 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1728 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1730 if (driver_byte(result) == DRIVER_SENSE &&
1731 scsi_sense_valid(&sshdr)) {
1732 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1733 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1739 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1742 if (flags & ~PR_FL_IGNORE_KEY)
1744 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1745 old_key, new_key, 0,
1746 (1 << 0) /* APTPL */);
1749 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1754 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1757 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1759 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1762 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1763 enum pr_type type, bool abort)
1765 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1766 sd_pr_type(type), 0);
1769 static int sd_pr_clear(struct block_device *bdev, u64 key)
1771 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1774 static const struct pr_ops sd_pr_ops = {
1775 .pr_register = sd_pr_register,
1776 .pr_reserve = sd_pr_reserve,
1777 .pr_release = sd_pr_release,
1778 .pr_preempt = sd_pr_preempt,
1779 .pr_clear = sd_pr_clear,
1782 static const struct block_device_operations sd_fops = {
1783 .owner = THIS_MODULE,
1785 .release = sd_release,
1787 .getgeo = sd_getgeo,
1788 #ifdef CONFIG_COMPAT
1789 .compat_ioctl = sd_compat_ioctl,
1791 .check_events = sd_check_events,
1792 .revalidate_disk = sd_revalidate_disk,
1793 .unlock_native_capacity = sd_unlock_native_capacity,
1794 .report_zones = sd_zbc_report_zones,
1795 .pr_ops = &sd_pr_ops,
1799 * sd_eh_reset - reset error handling callback
1800 * @scmd: sd-issued command that has failed
1802 * This function is called by the SCSI midlayer before starting
1803 * SCSI EH. When counting medium access failures we have to be
1804 * careful to register it only only once per device and SCSI EH run;
1805 * there might be several timed out commands which will cause the
1806 * 'max_medium_access_timeouts' counter to trigger after the first
1807 * SCSI EH run already and set the device to offline.
1808 * So this function resets the internal counter before starting SCSI EH.
1810 static void sd_eh_reset(struct scsi_cmnd *scmd)
1812 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1814 /* New SCSI EH run, reset gate variable */
1815 sdkp->ignore_medium_access_errors = false;
1819 * sd_eh_action - error handling callback
1820 * @scmd: sd-issued command that has failed
1821 * @eh_disp: The recovery disposition suggested by the midlayer
1823 * This function is called by the SCSI midlayer upon completion of an
1824 * error test command (currently TEST UNIT READY). The result of sending
1825 * the eh command is passed in eh_disp. We're looking for devices that
1826 * fail medium access commands but are OK with non access commands like
1827 * test unit ready (so wrongly see the device as having a successful
1830 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1832 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1833 struct scsi_device *sdev = scmd->device;
1835 if (!scsi_device_online(sdev) ||
1836 !scsi_medium_access_command(scmd) ||
1837 host_byte(scmd->result) != DID_TIME_OUT ||
1842 * The device has timed out executing a medium access command.
1843 * However, the TEST UNIT READY command sent during error
1844 * handling completed successfully. Either the device is in the
1845 * process of recovering or has it suffered an internal failure
1846 * that prevents access to the storage medium.
1848 if (!sdkp->ignore_medium_access_errors) {
1849 sdkp->medium_access_timed_out++;
1850 sdkp->ignore_medium_access_errors = true;
1854 * If the device keeps failing read/write commands but TEST UNIT
1855 * READY always completes successfully we assume that medium
1856 * access is no longer possible and take the device offline.
1858 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1859 scmd_printk(KERN_ERR, scmd,
1860 "Medium access timeout failure. Offlining disk!\n");
1861 mutex_lock(&sdev->state_mutex);
1862 scsi_device_set_state(sdev, SDEV_OFFLINE);
1863 mutex_unlock(&sdev->state_mutex);
1871 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1873 struct request *req = scmd->request;
1874 struct scsi_device *sdev = scmd->device;
1875 unsigned int transferred, good_bytes;
1876 u64 start_lba, end_lba, bad_lba;
1879 * Some commands have a payload smaller than the device logical
1880 * block size (e.g. INQUIRY on a 4K disk).
1882 if (scsi_bufflen(scmd) <= sdev->sector_size)
1885 /* Check if we have a 'bad_lba' information */
1886 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1887 SCSI_SENSE_BUFFERSIZE,
1892 * If the bad lba was reported incorrectly, we have no idea where
1895 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1896 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1897 if (bad_lba < start_lba || bad_lba >= end_lba)
1901 * resid is optional but mostly filled in. When it's unused,
1902 * its value is zero, so we assume the whole buffer transferred
1904 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1906 /* This computation should always be done in terms of the
1907 * resolution of the device's medium.
1909 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1911 return min(good_bytes, transferred);
1915 * sd_done - bottom half handler: called when the lower level
1916 * driver has completed (successfully or otherwise) a scsi command.
1917 * @SCpnt: mid-level's per command structure.
1919 * Note: potentially run from within an ISR. Must not block.
1921 static int sd_done(struct scsi_cmnd *SCpnt)
1923 int result = SCpnt->result;
1924 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1925 unsigned int sector_size = SCpnt->device->sector_size;
1927 struct scsi_sense_hdr sshdr;
1928 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1929 struct request *req = SCpnt->request;
1930 int sense_valid = 0;
1931 int sense_deferred = 0;
1933 switch (req_op(req)) {
1934 case REQ_OP_DISCARD:
1935 case REQ_OP_WRITE_ZEROES:
1936 case REQ_OP_WRITE_SAME:
1937 case REQ_OP_ZONE_RESET:
1939 good_bytes = blk_rq_bytes(req);
1940 scsi_set_resid(SCpnt, 0);
1943 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1948 * In case of bogus fw or device, we could end up having
1949 * an unaligned partial completion. Check this here and force
1952 resid = scsi_get_resid(SCpnt);
1953 if (resid & (sector_size - 1)) {
1954 sd_printk(KERN_INFO, sdkp,
1955 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1956 resid, sector_size);
1957 resid = min(scsi_bufflen(SCpnt),
1958 round_up(resid, sector_size));
1959 scsi_set_resid(SCpnt, resid);
1964 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1966 sense_deferred = scsi_sense_is_deferred(&sshdr);
1968 sdkp->medium_access_timed_out = 0;
1970 if (driver_byte(result) != DRIVER_SENSE &&
1971 (!sense_valid || sense_deferred))
1974 switch (sshdr.sense_key) {
1975 case HARDWARE_ERROR:
1977 good_bytes = sd_completed_bytes(SCpnt);
1979 case RECOVERED_ERROR:
1980 good_bytes = scsi_bufflen(SCpnt);
1983 /* This indicates a false check condition, so ignore it. An
1984 * unknown amount of data was transferred so treat it as an
1988 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1990 case ABORTED_COMMAND:
1991 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1992 good_bytes = sd_completed_bytes(SCpnt);
1994 case ILLEGAL_REQUEST:
1995 switch (sshdr.asc) {
1996 case 0x10: /* DIX: Host detected corruption */
1997 good_bytes = sd_completed_bytes(SCpnt);
1999 case 0x20: /* INVALID COMMAND OPCODE */
2000 case 0x24: /* INVALID FIELD IN CDB */
2001 switch (SCpnt->cmnd[0]) {
2003 sd_config_discard(sdkp, SD_LBP_DISABLE);
2007 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2008 sd_config_discard(sdkp, SD_LBP_DISABLE);
2010 sdkp->device->no_write_same = 1;
2011 sd_config_write_same(sdkp);
2012 req->rq_flags |= RQF_QUIET;
2023 if (sd_is_zoned(sdkp))
2024 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2026 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2027 "sd_done: completed %d of %d bytes\n",
2028 good_bytes, scsi_bufflen(SCpnt)));
2030 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2032 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2033 good_bytes / scsi_prot_interval(SCpnt));
2039 * spinup disk - called only in sd_revalidate_disk()
2042 sd_spinup_disk(struct scsi_disk *sdkp)
2044 unsigned char cmd[10];
2045 unsigned long spintime_expire = 0;
2046 int retries, spintime;
2047 unsigned int the_result;
2048 struct scsi_sense_hdr sshdr;
2049 int sense_valid = 0;
2053 /* Spin up drives, as required. Only do this at boot time */
2054 /* Spinup needs to be done for module loads too. */
2059 cmd[0] = TEST_UNIT_READY;
2060 memset((void *) &cmd[1], 0, 9);
2062 the_result = scsi_execute_req(sdkp->device, cmd,
2065 SD_MAX_RETRIES, NULL);
2068 * If the drive has indicated to us that it
2069 * doesn't have any media in it, don't bother
2070 * with any more polling.
2072 if (media_not_present(sdkp, &sshdr))
2076 sense_valid = scsi_sense_valid(&sshdr);
2078 } while (retries < 3 &&
2079 (!scsi_status_is_good(the_result) ||
2080 ((driver_byte(the_result) == DRIVER_SENSE) &&
2081 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2083 if (driver_byte(the_result) != DRIVER_SENSE) {
2084 /* no sense, TUR either succeeded or failed
2085 * with a status error */
2086 if(!spintime && !scsi_status_is_good(the_result)) {
2087 sd_print_result(sdkp, "Test Unit Ready failed",
2094 * The device does not want the automatic start to be issued.
2096 if (sdkp->device->no_start_on_add)
2099 if (sense_valid && sshdr.sense_key == NOT_READY) {
2100 if (sshdr.asc == 4 && sshdr.ascq == 3)
2101 break; /* manual intervention required */
2102 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2103 break; /* standby */
2104 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2105 break; /* unavailable */
2106 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2107 break; /* sanitize in progress */
2109 * Issue command to spin up drive when not ready
2112 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2113 cmd[0] = START_STOP;
2114 cmd[1] = 1; /* Return immediately */
2115 memset((void *) &cmd[2], 0, 8);
2116 cmd[4] = 1; /* Start spin cycle */
2117 if (sdkp->device->start_stop_pwr_cond)
2119 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2121 SD_TIMEOUT, SD_MAX_RETRIES,
2123 spintime_expire = jiffies + 100 * HZ;
2126 /* Wait 1 second for next try */
2128 printk(KERN_CONT ".");
2131 * Wait for USB flash devices with slow firmware.
2132 * Yes, this sense key/ASC combination shouldn't
2133 * occur here. It's characteristic of these devices.
2135 } else if (sense_valid &&
2136 sshdr.sense_key == UNIT_ATTENTION &&
2137 sshdr.asc == 0x28) {
2139 spintime_expire = jiffies + 5 * HZ;
2142 /* Wait 1 second for next try */
2145 /* we don't understand the sense code, so it's
2146 * probably pointless to loop */
2148 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2149 sd_print_sense_hdr(sdkp, &sshdr);
2154 } while (spintime && time_before_eq(jiffies, spintime_expire));
2157 if (scsi_status_is_good(the_result))
2158 printk(KERN_CONT "ready\n");
2160 printk(KERN_CONT "not responding...\n");
2165 * Determine whether disk supports Data Integrity Field.
2167 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2169 struct scsi_device *sdp = sdkp->device;
2173 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2176 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2178 if (type > T10_PI_TYPE3_PROTECTION)
2180 else if (scsi_host_dif_capable(sdp->host, type))
2183 if (sdkp->first_scan || type != sdkp->protection_type)
2186 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2187 " protection type %u. Disabling disk!\n",
2191 sd_printk(KERN_NOTICE, sdkp,
2192 "Enabling DIF Type %u protection\n", type);
2195 sd_printk(KERN_NOTICE, sdkp,
2196 "Disabling DIF Type %u protection\n", type);
2200 sdkp->protection_type = type;
2205 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2206 struct scsi_sense_hdr *sshdr, int sense_valid,
2209 if (driver_byte(the_result) == DRIVER_SENSE)
2210 sd_print_sense_hdr(sdkp, sshdr);
2212 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2215 * Set dirty bit for removable devices if not ready -
2216 * sometimes drives will not report this properly.
2218 if (sdp->removable &&
2219 sense_valid && sshdr->sense_key == NOT_READY)
2220 set_media_not_present(sdkp);
2223 * We used to set media_present to 0 here to indicate no media
2224 * in the drive, but some drives fail read capacity even with
2225 * media present, so we can't do that.
2227 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2231 #if RC16_LEN > SD_BUF_SIZE
2232 #error RC16_LEN must not be more than SD_BUF_SIZE
2235 #define READ_CAPACITY_RETRIES_ON_RESET 10
2238 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2239 * and the reported logical block size is bigger than 512 bytes. Note
2240 * that last_sector is a u64 and therefore logical_to_sectors() is not
2243 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2245 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2247 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2253 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2254 unsigned char *buffer)
2256 unsigned char cmd[16];
2257 struct scsi_sense_hdr sshdr;
2258 int sense_valid = 0;
2260 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2261 unsigned int alignment;
2262 unsigned long long lba;
2263 unsigned sector_size;
2265 if (sdp->no_read_capacity_16)
2270 cmd[0] = SERVICE_ACTION_IN_16;
2271 cmd[1] = SAI_READ_CAPACITY_16;
2273 memset(buffer, 0, RC16_LEN);
2275 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2276 buffer, RC16_LEN, &sshdr,
2277 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2279 if (media_not_present(sdkp, &sshdr))
2283 sense_valid = scsi_sense_valid(&sshdr);
2285 sshdr.sense_key == ILLEGAL_REQUEST &&
2286 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2288 /* Invalid Command Operation Code or
2289 * Invalid Field in CDB, just retry
2290 * silently with RC10 */
2293 sshdr.sense_key == UNIT_ATTENTION &&
2294 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2295 /* Device reset might occur several times,
2296 * give it one more chance */
2297 if (--reset_retries > 0)
2302 } while (the_result && retries);
2305 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2306 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2310 sector_size = get_unaligned_be32(&buffer[8]);
2311 lba = get_unaligned_be64(&buffer[0]);
2313 if (sd_read_protection_type(sdkp, buffer) < 0) {
2318 if (!sd_addressable_capacity(lba, sector_size)) {
2319 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2320 "kernel compiled with support for large block "
2326 /* Logical blocks per physical block exponent */
2327 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2330 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2332 /* Lowest aligned logical block */
2333 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2334 blk_queue_alignment_offset(sdp->request_queue, alignment);
2335 if (alignment && sdkp->first_scan)
2336 sd_printk(KERN_NOTICE, sdkp,
2337 "physical block alignment offset: %u\n", alignment);
2339 if (buffer[14] & 0x80) { /* LBPME */
2342 if (buffer[14] & 0x40) /* LBPRZ */
2345 sd_config_discard(sdkp, SD_LBP_WS16);
2348 sdkp->capacity = lba + 1;
2352 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2353 unsigned char *buffer)
2355 unsigned char cmd[16];
2356 struct scsi_sense_hdr sshdr;
2357 int sense_valid = 0;
2359 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2361 unsigned sector_size;
2364 cmd[0] = READ_CAPACITY;
2365 memset(&cmd[1], 0, 9);
2366 memset(buffer, 0, 8);
2368 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2370 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2372 if (media_not_present(sdkp, &sshdr))
2376 sense_valid = scsi_sense_valid(&sshdr);
2378 sshdr.sense_key == UNIT_ATTENTION &&
2379 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2380 /* Device reset might occur several times,
2381 * give it one more chance */
2382 if (--reset_retries > 0)
2387 } while (the_result && retries);
2390 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2391 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2395 sector_size = get_unaligned_be32(&buffer[4]);
2396 lba = get_unaligned_be32(&buffer[0]);
2398 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2399 /* Some buggy (usb cardreader) devices return an lba of
2400 0xffffffff when the want to report a size of 0 (with
2401 which they really mean no media is present) */
2403 sdkp->physical_block_size = sector_size;
2407 if (!sd_addressable_capacity(lba, sector_size)) {
2408 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2409 "kernel compiled with support for large block "
2415 sdkp->capacity = lba + 1;
2416 sdkp->physical_block_size = sector_size;
2420 static int sd_try_rc16_first(struct scsi_device *sdp)
2422 if (sdp->host->max_cmd_len < 16)
2424 if (sdp->try_rc_10_first)
2426 if (sdp->scsi_level > SCSI_SPC_2)
2428 if (scsi_device_protection(sdp))
2434 * read disk capacity
2437 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2440 struct scsi_device *sdp = sdkp->device;
2442 if (sd_try_rc16_first(sdp)) {
2443 sector_size = read_capacity_16(sdkp, sdp, buffer);
2444 if (sector_size == -EOVERFLOW)
2446 if (sector_size == -ENODEV)
2448 if (sector_size < 0)
2449 sector_size = read_capacity_10(sdkp, sdp, buffer);
2450 if (sector_size < 0)
2453 sector_size = read_capacity_10(sdkp, sdp, buffer);
2454 if (sector_size == -EOVERFLOW)
2456 if (sector_size < 0)
2458 if ((sizeof(sdkp->capacity) > 4) &&
2459 (sdkp->capacity > 0xffffffffULL)) {
2460 int old_sector_size = sector_size;
2461 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2462 "Trying to use READ CAPACITY(16).\n");
2463 sector_size = read_capacity_16(sdkp, sdp, buffer);
2464 if (sector_size < 0) {
2465 sd_printk(KERN_NOTICE, sdkp,
2466 "Using 0xffffffff as device size\n");
2467 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2468 sector_size = old_sector_size;
2471 /* Remember that READ CAPACITY(16) succeeded */
2472 sdp->try_rc_10_first = 0;
2476 /* Some devices are known to return the total number of blocks,
2477 * not the highest block number. Some devices have versions
2478 * which do this and others which do not. Some devices we might
2479 * suspect of doing this but we don't know for certain.
2481 * If we know the reported capacity is wrong, decrement it. If
2482 * we can only guess, then assume the number of blocks is even
2483 * (usually true but not always) and err on the side of lowering
2486 if (sdp->fix_capacity ||
2487 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2488 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2489 "from its reported value: %llu\n",
2490 (unsigned long long) sdkp->capacity);
2495 if (sector_size == 0) {
2497 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2501 if (sector_size != 512 &&
2502 sector_size != 1024 &&
2503 sector_size != 2048 &&
2504 sector_size != 4096) {
2505 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2508 * The user might want to re-format the drive with
2509 * a supported sectorsize. Once this happens, it
2510 * would be relatively trivial to set the thing up.
2511 * For this reason, we leave the thing in the table.
2515 * set a bogus sector size so the normal read/write
2516 * logic in the block layer will eventually refuse any
2517 * request on this device without tripping over power
2518 * of two sector size assumptions
2522 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2523 blk_queue_physical_block_size(sdp->request_queue,
2524 sdkp->physical_block_size);
2525 sdkp->device->sector_size = sector_size;
2527 if (sdkp->capacity > 0xffffffff)
2528 sdp->use_16_for_rw = 1;
2533 * Print disk capacity
2536 sd_print_capacity(struct scsi_disk *sdkp,
2537 sector_t old_capacity)
2539 int sector_size = sdkp->device->sector_size;
2540 char cap_str_2[10], cap_str_10[10];
2542 string_get_size(sdkp->capacity, sector_size,
2543 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2544 string_get_size(sdkp->capacity, sector_size,
2545 STRING_UNITS_10, cap_str_10,
2546 sizeof(cap_str_10));
2548 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2549 sd_printk(KERN_NOTICE, sdkp,
2550 "%llu %d-byte logical blocks: (%s/%s)\n",
2551 (unsigned long long)sdkp->capacity,
2552 sector_size, cap_str_10, cap_str_2);
2554 if (sdkp->physical_block_size != sector_size)
2555 sd_printk(KERN_NOTICE, sdkp,
2556 "%u-byte physical blocks\n",
2557 sdkp->physical_block_size);
2559 sd_zbc_print_zones(sdkp);
2563 /* called with buffer of length 512 */
2565 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2566 unsigned char *buffer, int len, struct scsi_mode_data *data,
2567 struct scsi_sense_hdr *sshdr)
2569 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2570 SD_TIMEOUT, SD_MAX_RETRIES, data,
2575 * read write protect setting, if possible - called only in sd_revalidate_disk()
2576 * called with buffer of length SD_BUF_SIZE
2579 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2582 struct scsi_device *sdp = sdkp->device;
2583 struct scsi_mode_data data;
2584 int disk_ro = get_disk_ro(sdkp->disk);
2585 int old_wp = sdkp->write_prot;
2587 set_disk_ro(sdkp->disk, 0);
2588 if (sdp->skip_ms_page_3f) {
2589 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2593 if (sdp->use_192_bytes_for_3f) {
2594 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2597 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2598 * We have to start carefully: some devices hang if we ask
2599 * for more than is available.
2601 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2604 * Second attempt: ask for page 0 When only page 0 is
2605 * implemented, a request for page 3F may return Sense Key
2606 * 5: Illegal Request, Sense Code 24: Invalid field in
2609 if (!scsi_status_is_good(res))
2610 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2613 * Third attempt: ask 255 bytes, as we did earlier.
2615 if (!scsi_status_is_good(res))
2616 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2620 if (!scsi_status_is_good(res)) {
2621 sd_first_printk(KERN_WARNING, sdkp,
2622 "Test WP failed, assume Write Enabled\n");
2624 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2625 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2626 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2627 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2628 sdkp->write_prot ? "on" : "off");
2629 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2635 * sd_read_cache_type - called only from sd_revalidate_disk()
2636 * called with buffer of length SD_BUF_SIZE
2639 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2642 struct scsi_device *sdp = sdkp->device;
2647 struct scsi_mode_data data;
2648 struct scsi_sense_hdr sshdr;
2649 int old_wce = sdkp->WCE;
2650 int old_rcd = sdkp->RCD;
2651 int old_dpofua = sdkp->DPOFUA;
2654 if (sdkp->cache_override)
2658 if (sdp->skip_ms_page_8) {
2659 if (sdp->type == TYPE_RBC)
2662 if (sdp->skip_ms_page_3f)
2665 if (sdp->use_192_bytes_for_3f)
2669 } else if (sdp->type == TYPE_RBC) {
2677 /* cautiously ask */
2678 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2681 if (!scsi_status_is_good(res))
2684 if (!data.header_length) {
2687 sd_first_printk(KERN_ERR, sdkp,
2688 "Missing header in MODE_SENSE response\n");
2691 /* that went OK, now ask for the proper length */
2695 * We're only interested in the first three bytes, actually.
2696 * But the data cache page is defined for the first 20.
2700 else if (len > SD_BUF_SIZE) {
2701 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2702 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2705 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2709 if (len > first_len)
2710 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2713 if (scsi_status_is_good(res)) {
2714 int offset = data.header_length + data.block_descriptor_length;
2716 while (offset < len) {
2717 u8 page_code = buffer[offset] & 0x3F;
2718 u8 spf = buffer[offset] & 0x40;
2720 if (page_code == 8 || page_code == 6) {
2721 /* We're interested only in the first 3 bytes.
2723 if (len - offset <= 2) {
2724 sd_first_printk(KERN_ERR, sdkp,
2725 "Incomplete mode parameter "
2729 modepage = page_code;
2733 /* Go to the next page */
2734 if (spf && len - offset > 3)
2735 offset += 4 + (buffer[offset+2] << 8) +
2737 else if (!spf && len - offset > 1)
2738 offset += 2 + buffer[offset+1];
2740 sd_first_printk(KERN_ERR, sdkp,
2742 "parameter data\n");
2748 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2752 if (modepage == 8) {
2753 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2754 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2756 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2760 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2761 if (sdp->broken_fua) {
2762 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2764 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2765 !sdkp->device->use_16_for_rw) {
2766 sd_first_printk(KERN_NOTICE, sdkp,
2767 "Uses READ/WRITE(6), disabling FUA\n");
2771 /* No cache flush allowed for write protected devices */
2772 if (sdkp->WCE && sdkp->write_prot)
2775 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2776 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2777 sd_printk(KERN_NOTICE, sdkp,
2778 "Write cache: %s, read cache: %s, %s\n",
2779 sdkp->WCE ? "enabled" : "disabled",
2780 sdkp->RCD ? "disabled" : "enabled",
2781 sdkp->DPOFUA ? "supports DPO and FUA"
2782 : "doesn't support DPO or FUA");
2788 if (scsi_sense_valid(&sshdr) &&
2789 sshdr.sense_key == ILLEGAL_REQUEST &&
2790 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2791 /* Invalid field in CDB */
2792 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2794 sd_first_printk(KERN_ERR, sdkp,
2795 "Asking for cache data failed\n");
2798 if (sdp->wce_default_on) {
2799 sd_first_printk(KERN_NOTICE, sdkp,
2800 "Assuming drive cache: write back\n");
2803 sd_first_printk(KERN_ERR, sdkp,
2804 "Assuming drive cache: write through\n");
2812 * The ATO bit indicates whether the DIF application tag is available
2813 * for use by the operating system.
2815 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2818 struct scsi_device *sdp = sdkp->device;
2819 struct scsi_mode_data data;
2820 struct scsi_sense_hdr sshdr;
2822 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2825 if (sdkp->protection_type == 0)
2828 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2829 SD_MAX_RETRIES, &data, &sshdr);
2831 if (!scsi_status_is_good(res) || !data.header_length ||
2833 sd_first_printk(KERN_WARNING, sdkp,
2834 "getting Control mode page failed, assume no ATO\n");
2836 if (scsi_sense_valid(&sshdr))
2837 sd_print_sense_hdr(sdkp, &sshdr);
2842 offset = data.header_length + data.block_descriptor_length;
2844 if ((buffer[offset] & 0x3f) != 0x0a) {
2845 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2849 if ((buffer[offset + 5] & 0x80) == 0)
2858 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2859 * @sdkp: disk to query
2861 static void sd_read_block_limits(struct scsi_disk *sdkp)
2863 unsigned int sector_sz = sdkp->device->sector_size;
2864 const int vpd_len = 64;
2865 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2868 /* Block Limits VPD */
2869 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2872 blk_queue_io_min(sdkp->disk->queue,
2873 get_unaligned_be16(&buffer[6]) * sector_sz);
2875 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2876 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2878 if (buffer[3] == 0x3c) {
2879 unsigned int lba_count, desc_count;
2881 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2886 lba_count = get_unaligned_be32(&buffer[20]);
2887 desc_count = get_unaligned_be32(&buffer[24]);
2889 if (lba_count && desc_count)
2890 sdkp->max_unmap_blocks = lba_count;
2892 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2894 if (buffer[32] & 0x80)
2895 sdkp->unmap_alignment =
2896 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2898 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2900 if (sdkp->max_unmap_blocks)
2901 sd_config_discard(sdkp, SD_LBP_UNMAP);
2903 sd_config_discard(sdkp, SD_LBP_WS16);
2905 } else { /* LBP VPD page tells us what to use */
2906 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2907 sd_config_discard(sdkp, SD_LBP_UNMAP);
2908 else if (sdkp->lbpws)
2909 sd_config_discard(sdkp, SD_LBP_WS16);
2910 else if (sdkp->lbpws10)
2911 sd_config_discard(sdkp, SD_LBP_WS10);
2913 sd_config_discard(sdkp, SD_LBP_DISABLE);
2922 * sd_read_block_characteristics - Query block dev. characteristics
2923 * @sdkp: disk to query
2925 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2927 struct request_queue *q = sdkp->disk->queue;
2928 unsigned char *buffer;
2930 const int vpd_len = 64;
2932 buffer = kmalloc(vpd_len, GFP_KERNEL);
2935 /* Block Device Characteristics VPD */
2936 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2939 rot = get_unaligned_be16(&buffer[4]);
2942 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2943 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2945 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2946 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
2949 if (sdkp->device->type == TYPE_ZBC) {
2951 q->limits.zoned = BLK_ZONED_HM;
2953 sdkp->zoned = (buffer[8] >> 4) & 3;
2954 if (sdkp->zoned == 1)
2956 q->limits.zoned = BLK_ZONED_HA;
2959 * Treat drive-managed devices as
2960 * regular block devices.
2962 q->limits.zoned = BLK_ZONED_NONE;
2964 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2965 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2966 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2973 * sd_read_block_provisioning - Query provisioning VPD page
2974 * @sdkp: disk to query
2976 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2978 unsigned char *buffer;
2979 const int vpd_len = 8;
2981 if (sdkp->lbpme == 0)
2984 buffer = kmalloc(vpd_len, GFP_KERNEL);
2986 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2990 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2991 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2992 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2998 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3000 struct scsi_device *sdev = sdkp->device;
3002 if (sdev->host->no_write_same) {
3003 sdev->no_write_same = 1;
3008 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3009 /* too large values might cause issues with arcmsr */
3010 int vpd_buf_len = 64;
3012 sdev->no_report_opcodes = 1;
3014 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3015 * CODES is unsupported and the device has an ATA
3016 * Information VPD page (SAT).
3018 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3019 sdev->no_write_same = 1;
3022 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3025 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3029 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3031 struct scsi_device *sdev = sdkp->device;
3033 if (!sdev->security_supported)
3036 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3037 SECURITY_PROTOCOL_IN) == 1 &&
3038 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3039 SECURITY_PROTOCOL_OUT) == 1)
3044 * sd_revalidate_disk - called the first time a new disk is seen,
3045 * performs disk spin up, read_capacity, etc.
3046 * @disk: struct gendisk we care about
3048 static int sd_revalidate_disk(struct gendisk *disk)
3050 struct scsi_disk *sdkp = scsi_disk(disk);
3051 struct scsi_device *sdp = sdkp->device;
3052 struct request_queue *q = sdkp->disk->queue;
3053 sector_t old_capacity = sdkp->capacity;
3054 unsigned char *buffer;
3055 unsigned int dev_max, rw_max;
3057 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3058 "sd_revalidate_disk\n"));
3061 * If the device is offline, don't try and read capacity or any
3062 * of the other niceties.
3064 if (!scsi_device_online(sdp))
3067 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3069 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3070 "allocation failure.\n");
3074 sd_spinup_disk(sdkp);
3077 * Without media there is no reason to ask; moreover, some devices
3078 * react badly if we do.
3080 if (sdkp->media_present) {
3081 sd_read_capacity(sdkp, buffer);
3083 if (scsi_device_supports_vpd(sdp)) {
3084 sd_read_block_provisioning(sdkp);
3085 sd_read_block_limits(sdkp);
3086 sd_read_block_characteristics(sdkp);
3087 sd_zbc_read_zones(sdkp, buffer);
3090 sd_print_capacity(sdkp, old_capacity);
3092 sd_read_write_protect_flag(sdkp, buffer);
3093 sd_read_cache_type(sdkp, buffer);
3094 sd_read_app_tag_own(sdkp, buffer);
3095 sd_read_write_same(sdkp, buffer);
3096 sd_read_security(sdkp, buffer);
3100 * We now have all cache related info, determine how we deal
3101 * with flush requests.
3103 sd_set_flush_flag(sdkp);
3105 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3106 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3108 /* Some devices report a maximum block count for READ/WRITE requests. */
3109 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3110 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3113 * Determine the device's preferred I/O size for reads and writes
3114 * unless the reported value is unreasonably small, large, or
3117 if (sdkp->opt_xfer_blocks &&
3118 sdkp->opt_xfer_blocks <= dev_max &&
3119 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3120 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3121 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3122 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3124 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3125 (sector_t)BLK_DEF_MAX_SECTORS);
3127 /* Do not exceed controller limit */
3128 rw_max = min(rw_max, queue_max_hw_sectors(q));
3131 * Only update max_sectors if previously unset or if the current value
3132 * exceeds the capabilities of the hardware.
3134 if (sdkp->first_scan ||
3135 q->limits.max_sectors > q->limits.max_dev_sectors ||
3136 q->limits.max_sectors > q->limits.max_hw_sectors)
3137 q->limits.max_sectors = rw_max;
3139 sdkp->first_scan = 0;
3141 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3142 sd_config_write_same(sdkp);
3150 * sd_unlock_native_capacity - unlock native capacity
3151 * @disk: struct gendisk to set capacity for
3153 * Block layer calls this function if it detects that partitions
3154 * on @disk reach beyond the end of the device. If the SCSI host
3155 * implements ->unlock_native_capacity() method, it's invoked to
3156 * give it a chance to adjust the device capacity.
3159 * Defined by block layer. Might sleep.
3161 static void sd_unlock_native_capacity(struct gendisk *disk)
3163 struct scsi_device *sdev = scsi_disk(disk)->device;
3165 if (sdev->host->hostt->unlock_native_capacity)
3166 sdev->host->hostt->unlock_native_capacity(sdev);
3170 * sd_format_disk_name - format disk name
3171 * @prefix: name prefix - ie. "sd" for SCSI disks
3172 * @index: index of the disk to format name for
3173 * @buf: output buffer
3174 * @buflen: length of the output buffer
3176 * SCSI disk names starts at sda. The 26th device is sdz and the
3177 * 27th is sdaa. The last one for two lettered suffix is sdzz
3178 * which is followed by sdaaa.
3180 * This is basically 26 base counting with one extra 'nil' entry
3181 * at the beginning from the second digit on and can be
3182 * determined using similar method as 26 base conversion with the
3183 * index shifted -1 after each digit is computed.
3189 * 0 on success, -errno on failure.
3191 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3193 const int base = 'z' - 'a' + 1;
3194 char *begin = buf + strlen(prefix);
3195 char *end = buf + buflen;
3205 *--p = 'a' + (index % unit);
3206 index = (index / unit) - 1;
3207 } while (index >= 0);
3209 memmove(begin, p, end - p);
3210 memcpy(buf, prefix, strlen(prefix));
3216 * The asynchronous part of sd_probe
3218 static void sd_probe_async(void *data, async_cookie_t cookie)
3220 struct scsi_disk *sdkp = data;
3221 struct scsi_device *sdp;
3228 index = sdkp->index;
3229 dev = &sdp->sdev_gendev;
3231 gd->major = sd_major((index & 0xf0) >> 4);
3232 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3234 gd->fops = &sd_fops;
3235 gd->private_data = &sdkp->driver;
3236 gd->queue = sdkp->device->request_queue;
3238 /* defaults, until the device tells us otherwise */
3239 sdp->sector_size = 512;
3241 sdkp->media_present = 1;
3242 sdkp->write_prot = 0;
3243 sdkp->cache_override = 0;
3247 sdkp->first_scan = 1;
3248 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3250 sd_revalidate_disk(gd);
3252 gd->flags = GENHD_FL_EXT_DEVT;
3253 if (sdp->removable) {
3254 gd->flags |= GENHD_FL_REMOVABLE;
3255 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3258 blk_pm_runtime_init(sdp->request_queue, dev);
3259 device_add_disk(dev, gd, NULL);
3261 sd_dif_config_host(sdkp);
3263 sd_revalidate_disk(gd);
3265 if (sdkp->security) {
3266 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3268 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3271 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3272 sdp->removable ? "removable " : "");
3273 scsi_autopm_put_device(sdp);
3274 put_device(&sdkp->dev);
3278 * sd_probe - called during driver initialization and whenever a
3279 * new scsi device is attached to the system. It is called once
3280 * for each scsi device (not just disks) present.
3281 * @dev: pointer to device object
3283 * Returns 0 if successful (or not interested in this scsi device
3284 * (e.g. scanner)); 1 when there is an error.
3286 * Note: this function is invoked from the scsi mid-level.
3287 * This function sets up the mapping between a given
3288 * <host,channel,id,lun> (found in sdp) and new device name
3289 * (e.g. /dev/sda). More precisely it is the block device major
3290 * and minor number that is chosen here.
3292 * Assume sd_probe is not re-entrant (for time being)
3293 * Also think about sd_probe() and sd_remove() running coincidentally.
3295 static int sd_probe(struct device *dev)
3297 struct scsi_device *sdp = to_scsi_device(dev);
3298 struct scsi_disk *sdkp;
3303 scsi_autopm_get_device(sdp);
3305 if (sdp->type != TYPE_DISK &&
3306 sdp->type != TYPE_ZBC &&
3307 sdp->type != TYPE_MOD &&
3308 sdp->type != TYPE_RBC)
3311 #ifndef CONFIG_BLK_DEV_ZONED
3312 if (sdp->type == TYPE_ZBC)
3315 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3319 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3323 gd = alloc_disk(SD_MINORS);
3327 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3329 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3333 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3335 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3336 goto out_free_index;
3340 sdkp->driver = &sd_template;
3342 sdkp->index = index;
3343 atomic_set(&sdkp->openers, 0);
3344 atomic_set(&sdkp->device->ioerr_cnt, 0);
3346 if (!sdp->request_queue->rq_timeout) {
3347 if (sdp->type != TYPE_MOD)
3348 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3350 blk_queue_rq_timeout(sdp->request_queue,
3354 device_initialize(&sdkp->dev);
3355 sdkp->dev.parent = dev;
3356 sdkp->dev.class = &sd_disk_class;
3357 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3359 error = device_add(&sdkp->dev);
3361 goto out_free_index;
3364 dev_set_drvdata(dev, sdkp);
3366 get_device(&sdkp->dev); /* prevent release before async_schedule */
3367 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3372 ida_free(&sd_index_ida, index);
3378 scsi_autopm_put_device(sdp);
3383 * sd_remove - called whenever a scsi disk (previously recognized by
3384 * sd_probe) is detached from the system. It is called (potentially
3385 * multiple times) during sd module unload.
3386 * @dev: pointer to device object
3388 * Note: this function is invoked from the scsi mid-level.
3389 * This function potentially frees up a device name (e.g. /dev/sdc)
3390 * that could be re-used by a subsequent sd_probe().
3391 * This function is not called when the built-in sd driver is "exit-ed".
3393 static int sd_remove(struct device *dev)
3395 struct scsi_disk *sdkp;
3398 sdkp = dev_get_drvdata(dev);
3399 devt = disk_devt(sdkp->disk);
3400 scsi_autopm_get_device(sdkp->device);
3402 async_synchronize_full_domain(&scsi_sd_pm_domain);
3403 async_synchronize_full_domain(&scsi_sd_probe_domain);
3404 device_del(&sdkp->dev);
3405 del_gendisk(sdkp->disk);
3408 free_opal_dev(sdkp->opal_dev);
3410 blk_register_region(devt, SD_MINORS, NULL,
3411 sd_default_probe, NULL, NULL);
3413 mutex_lock(&sd_ref_mutex);
3414 dev_set_drvdata(dev, NULL);
3415 put_device(&sdkp->dev);
3416 mutex_unlock(&sd_ref_mutex);
3422 * scsi_disk_release - Called to free the scsi_disk structure
3423 * @dev: pointer to embedded class device
3425 * sd_ref_mutex must be held entering this routine. Because it is
3426 * called on last put, you should always use the scsi_disk_get()
3427 * scsi_disk_put() helpers which manipulate the semaphore directly
3428 * and never do a direct put_device.
3430 static void scsi_disk_release(struct device *dev)
3432 struct scsi_disk *sdkp = to_scsi_disk(dev);
3433 struct gendisk *disk = sdkp->disk;
3435 ida_free(&sd_index_ida, sdkp->index);
3437 disk->private_data = NULL;
3439 put_device(&sdkp->device->sdev_gendev);
3444 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3446 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3447 struct scsi_sense_hdr sshdr;
3448 struct scsi_device *sdp = sdkp->device;
3452 cmd[4] |= 1; /* START */
3454 if (sdp->start_stop_pwr_cond)
3455 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3457 if (!scsi_device_online(sdp))
3460 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3461 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3463 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3464 if (driver_byte(res) == DRIVER_SENSE)
3465 sd_print_sense_hdr(sdkp, &sshdr);
3466 if (scsi_sense_valid(&sshdr) &&
3467 /* 0x3a is medium not present */
3472 /* SCSI error codes must not go to the generic layer */
3480 * Send a SYNCHRONIZE CACHE instruction down to the device through
3481 * the normal SCSI command structure. Wait for the command to
3484 static void sd_shutdown(struct device *dev)
3486 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3489 return; /* this can happen */
3491 if (pm_runtime_suspended(dev))
3494 if (sdkp->WCE && sdkp->media_present) {
3495 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3496 sd_sync_cache(sdkp, NULL);
3499 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3500 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3501 sd_start_stop_device(sdkp, 0);
3505 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3507 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3508 struct scsi_sense_hdr sshdr;
3511 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3514 if (sdkp->WCE && sdkp->media_present) {
3515 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3516 ret = sd_sync_cache(sdkp, &sshdr);
3519 /* ignore OFFLINE device */
3523 if (!scsi_sense_valid(&sshdr) ||
3524 sshdr.sense_key != ILLEGAL_REQUEST)
3528 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3529 * doesn't support sync. There's not much to do and
3530 * suspend shouldn't fail.
3536 if (sdkp->device->manage_start_stop) {
3537 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3538 /* an error is not worth aborting a system sleep */
3539 ret = sd_start_stop_device(sdkp, 0);
3540 if (ignore_stop_errors)
3547 static int sd_suspend_system(struct device *dev)
3549 return sd_suspend_common(dev, true);
3552 static int sd_suspend_runtime(struct device *dev)
3554 return sd_suspend_common(dev, false);
3557 static int sd_resume(struct device *dev)
3559 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3562 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3565 if (!sdkp->device->manage_start_stop)
3568 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3569 ret = sd_start_stop_device(sdkp, 1);
3571 opal_unlock_from_suspend(sdkp->opal_dev);
3576 * init_sd - entry point for this driver (both when built in or when
3579 * Note: this function registers this driver with the scsi mid-level.
3581 static int __init init_sd(void)
3583 int majors = 0, i, err;
3585 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3587 for (i = 0; i < SD_MAJORS; i++) {
3588 if (register_blkdev(sd_major(i), "sd") != 0)
3591 blk_register_region(sd_major(i), SD_MINORS, NULL,
3592 sd_default_probe, NULL, NULL);
3598 err = class_register(&sd_disk_class);
3602 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3604 if (!sd_cdb_cache) {
3605 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3610 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3612 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3617 err = scsi_register_driver(&sd_template.gendrv);
3619 goto err_out_driver;
3624 mempool_destroy(sd_cdb_pool);
3627 kmem_cache_destroy(sd_cdb_cache);
3630 class_unregister(&sd_disk_class);
3632 for (i = 0; i < SD_MAJORS; i++)
3633 unregister_blkdev(sd_major(i), "sd");
3638 * exit_sd - exit point for this driver (when it is a module).
3640 * Note: this function unregisters this driver from the scsi mid-level.
3642 static void __exit exit_sd(void)
3646 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3648 scsi_unregister_driver(&sd_template.gendrv);
3649 mempool_destroy(sd_cdb_pool);
3650 kmem_cache_destroy(sd_cdb_cache);
3652 class_unregister(&sd_disk_class);
3654 for (i = 0; i < SD_MAJORS; i++) {
3655 blk_unregister_region(sd_major(i), SD_MINORS);
3656 unregister_blkdev(sd_major(i), "sd");
3660 module_init(init_sd);
3661 module_exit(exit_sd);
3663 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3664 struct scsi_sense_hdr *sshdr)
3666 scsi_print_sense_hdr(sdkp->device,
3667 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3670 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3673 const char *hb_string = scsi_hostbyte_string(result);
3674 const char *db_string = scsi_driverbyte_string(result);
3676 if (hb_string || db_string)
3677 sd_printk(KERN_INFO, sdkp,
3678 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3679 hb_string ? hb_string : "invalid",
3680 db_string ? db_string : "invalid");
3682 sd_printk(KERN_INFO, sdkp,
3683 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3684 msg, host_byte(result), driver_byte(result));