2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24 * Support 32k/1M disks.
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
35 #include <linux/module.h>
37 #include <linux/kernel.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/blk-pm.h>
49 #include <linux/delay.h>
50 #include <linux/mutex.h>
51 #include <linux/string_helpers.h>
52 #include <linux/async.h>
53 #include <linux/slab.h>
54 #include <linux/sed-opal.h>
55 #include <linux/pm_runtime.h>
57 #include <linux/t10-pi.h>
58 #include <linux/uaccess.h>
59 #include <asm/unaligned.h>
61 #include <scsi/scsi.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_dbg.h>
64 #include <scsi/scsi_device.h>
65 #include <scsi/scsi_driver.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_host.h>
68 #include <scsi/scsi_ioctl.h>
69 #include <scsi/scsicam.h>
72 #include "scsi_priv.h"
73 #include "scsi_logging.h"
75 MODULE_AUTHOR("Eric Youngdale");
76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
77 MODULE_LICENSE("GPL");
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
106 static void sd_config_discard(struct scsi_disk *, unsigned int);
107 static void sd_config_write_same(struct scsi_disk *);
108 static int sd_revalidate_disk(struct gendisk *);
109 static void sd_unlock_native_capacity(struct gendisk *disk);
110 static int sd_probe(struct device *);
111 static int sd_remove(struct device *);
112 static void sd_shutdown(struct device *);
113 static int sd_suspend_system(struct device *);
114 static int sd_suspend_runtime(struct device *);
115 static int sd_resume(struct device *);
116 static void sd_rescan(struct device *);
117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
119 static int sd_done(struct scsi_cmnd *);
120 static void sd_eh_reset(struct scsi_cmnd *);
121 static int sd_eh_action(struct scsi_cmnd *, int);
122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
123 static void scsi_disk_release(struct device *cdev);
124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
125 static void sd_print_result(const struct scsi_disk *, const char *, int);
127 static DEFINE_IDA(sd_index_ida);
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
136 static mempool_t *sd_page_pool;
138 static const char *sd_cache_types[] = {
139 "write through", "none", "write back",
140 "write back, no read (daft)"
143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
145 bool wc = false, fua = false;
153 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
157 cache_type_store(struct device *dev, struct device_attribute *attr,
158 const char *buf, size_t count)
160 int ct, rcd, wce, sp;
161 struct scsi_disk *sdkp = to_scsi_disk(dev);
162 struct scsi_device *sdp = sdkp->device;
165 struct scsi_mode_data data;
166 struct scsi_sense_hdr sshdr;
167 static const char temp[] = "temporary ";
170 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
171 /* no cache control on RBC devices; theoretically they
172 * can do it, but there's probably so many exceptions
173 * it's not worth the risk */
176 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
177 buf += sizeof(temp) - 1;
178 sdkp->cache_override = 1;
180 sdkp->cache_override = 0;
183 ct = sysfs_match_string(sd_cache_types, buf);
187 rcd = ct & 0x01 ? 1 : 0;
188 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 if (sdkp->cache_override) {
193 sd_set_flush_flag(sdkp);
197 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198 SD_MAX_RETRIES, &data, NULL))
200 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201 data.block_descriptor_length);
202 buffer_data = buffer + data.header_length +
203 data.block_descriptor_length;
204 buffer_data[2] &= ~0x05;
205 buffer_data[2] |= wce << 2 | rcd;
206 sp = buffer_data[0] & 0x80 ? 1 : 0;
207 buffer_data[0] &= ~0x80;
210 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
211 * received mode parameter buffer before doing MODE SELECT.
213 data.device_specific = 0;
215 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
216 SD_MAX_RETRIES, &data, &sshdr)) {
217 if (scsi_sense_valid(&sshdr))
218 sd_print_sense_hdr(sdkp, &sshdr);
221 revalidate_disk(sdkp->disk);
226 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
229 struct scsi_disk *sdkp = to_scsi_disk(dev);
230 struct scsi_device *sdp = sdkp->device;
232 return sprintf(buf, "%u\n", sdp->manage_start_stop);
236 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
237 const char *buf, size_t count)
239 struct scsi_disk *sdkp = to_scsi_disk(dev);
240 struct scsi_device *sdp = sdkp->device;
243 if (!capable(CAP_SYS_ADMIN))
246 if (kstrtobool(buf, &v))
249 sdp->manage_start_stop = v;
253 static DEVICE_ATTR_RW(manage_start_stop);
256 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
258 struct scsi_disk *sdkp = to_scsi_disk(dev);
260 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
264 allow_restart_store(struct device *dev, struct device_attribute *attr,
265 const char *buf, size_t count)
268 struct scsi_disk *sdkp = to_scsi_disk(dev);
269 struct scsi_device *sdp = sdkp->device;
271 if (!capable(CAP_SYS_ADMIN))
274 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
277 if (kstrtobool(buf, &v))
280 sdp->allow_restart = v;
284 static DEVICE_ATTR_RW(allow_restart);
287 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
289 struct scsi_disk *sdkp = to_scsi_disk(dev);
290 int ct = sdkp->RCD + 2*sdkp->WCE;
292 return sprintf(buf, "%s\n", sd_cache_types[ct]);
294 static DEVICE_ATTR_RW(cache_type);
297 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
299 struct scsi_disk *sdkp = to_scsi_disk(dev);
301 return sprintf(buf, "%u\n", sdkp->DPOFUA);
303 static DEVICE_ATTR_RO(FUA);
306 protection_type_show(struct device *dev, struct device_attribute *attr,
309 struct scsi_disk *sdkp = to_scsi_disk(dev);
311 return sprintf(buf, "%u\n", sdkp->protection_type);
315 protection_type_store(struct device *dev, struct device_attribute *attr,
316 const char *buf, size_t count)
318 struct scsi_disk *sdkp = to_scsi_disk(dev);
322 if (!capable(CAP_SYS_ADMIN))
325 err = kstrtouint(buf, 10, &val);
330 if (val <= T10_PI_TYPE3_PROTECTION)
331 sdkp->protection_type = val;
335 static DEVICE_ATTR_RW(protection_type);
338 protection_mode_show(struct device *dev, struct device_attribute *attr,
341 struct scsi_disk *sdkp = to_scsi_disk(dev);
342 struct scsi_device *sdp = sdkp->device;
343 unsigned int dif, dix;
345 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
346 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
348 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
354 return sprintf(buf, "none\n");
356 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
358 static DEVICE_ATTR_RO(protection_mode);
361 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
363 struct scsi_disk *sdkp = to_scsi_disk(dev);
365 return sprintf(buf, "%u\n", sdkp->ATO);
367 static DEVICE_ATTR_RO(app_tag_own);
370 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
373 struct scsi_disk *sdkp = to_scsi_disk(dev);
375 return sprintf(buf, "%u\n", sdkp->lbpme);
377 static DEVICE_ATTR_RO(thin_provisioning);
379 /* sysfs_match_string() requires dense arrays */
380 static const char *lbp_mode[] = {
381 [SD_LBP_FULL] = "full",
382 [SD_LBP_UNMAP] = "unmap",
383 [SD_LBP_WS16] = "writesame_16",
384 [SD_LBP_WS10] = "writesame_10",
385 [SD_LBP_ZERO] = "writesame_zero",
386 [SD_LBP_DISABLE] = "disabled",
390 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
393 struct scsi_disk *sdkp = to_scsi_disk(dev);
395 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
399 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
400 const char *buf, size_t count)
402 struct scsi_disk *sdkp = to_scsi_disk(dev);
403 struct scsi_device *sdp = sdkp->device;
406 if (!capable(CAP_SYS_ADMIN))
409 if (sd_is_zoned(sdkp)) {
410 sd_config_discard(sdkp, SD_LBP_DISABLE);
414 if (sdp->type != TYPE_DISK)
417 mode = sysfs_match_string(lbp_mode, buf);
421 sd_config_discard(sdkp, mode);
425 static DEVICE_ATTR_RW(provisioning_mode);
427 /* sysfs_match_string() requires dense arrays */
428 static const char *zeroing_mode[] = {
429 [SD_ZERO_WRITE] = "write",
430 [SD_ZERO_WS] = "writesame",
431 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
432 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
436 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
439 struct scsi_disk *sdkp = to_scsi_disk(dev);
441 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
445 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
446 const char *buf, size_t count)
448 struct scsi_disk *sdkp = to_scsi_disk(dev);
451 if (!capable(CAP_SYS_ADMIN))
454 mode = sysfs_match_string(zeroing_mode, buf);
458 sdkp->zeroing_mode = mode;
462 static DEVICE_ATTR_RW(zeroing_mode);
465 max_medium_access_timeouts_show(struct device *dev,
466 struct device_attribute *attr, char *buf)
468 struct scsi_disk *sdkp = to_scsi_disk(dev);
470 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
474 max_medium_access_timeouts_store(struct device *dev,
475 struct device_attribute *attr, const char *buf,
478 struct scsi_disk *sdkp = to_scsi_disk(dev);
481 if (!capable(CAP_SYS_ADMIN))
484 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
486 return err ? err : count;
488 static DEVICE_ATTR_RW(max_medium_access_timeouts);
491 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
494 struct scsi_disk *sdkp = to_scsi_disk(dev);
496 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
500 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
501 const char *buf, size_t count)
503 struct scsi_disk *sdkp = to_scsi_disk(dev);
504 struct scsi_device *sdp = sdkp->device;
508 if (!capable(CAP_SYS_ADMIN))
511 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
514 err = kstrtoul(buf, 10, &max);
520 sdp->no_write_same = 1;
521 else if (max <= SD_MAX_WS16_BLOCKS) {
522 sdp->no_write_same = 0;
523 sdkp->max_ws_blocks = max;
526 sd_config_write_same(sdkp);
530 static DEVICE_ATTR_RW(max_write_same_blocks);
532 static struct attribute *sd_disk_attrs[] = {
533 &dev_attr_cache_type.attr,
535 &dev_attr_allow_restart.attr,
536 &dev_attr_manage_start_stop.attr,
537 &dev_attr_protection_type.attr,
538 &dev_attr_protection_mode.attr,
539 &dev_attr_app_tag_own.attr,
540 &dev_attr_thin_provisioning.attr,
541 &dev_attr_provisioning_mode.attr,
542 &dev_attr_zeroing_mode.attr,
543 &dev_attr_max_write_same_blocks.attr,
544 &dev_attr_max_medium_access_timeouts.attr,
547 ATTRIBUTE_GROUPS(sd_disk);
549 static struct class sd_disk_class = {
551 .owner = THIS_MODULE,
552 .dev_release = scsi_disk_release,
553 .dev_groups = sd_disk_groups,
556 static const struct dev_pm_ops sd_pm_ops = {
557 .suspend = sd_suspend_system,
559 .poweroff = sd_suspend_system,
560 .restore = sd_resume,
561 .runtime_suspend = sd_suspend_runtime,
562 .runtime_resume = sd_resume,
565 static struct scsi_driver sd_template = {
568 .owner = THIS_MODULE,
571 .shutdown = sd_shutdown,
575 .init_command = sd_init_command,
576 .uninit_command = sd_uninit_command,
578 .eh_action = sd_eh_action,
579 .eh_reset = sd_eh_reset,
583 * Dummy kobj_map->probe function.
584 * The default ->probe function will call modprobe, which is
585 * pointless as this module is already loaded.
587 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
593 * Device no to disk mapping:
595 * major disc2 disc p1
596 * |............|.............|....|....| <- dev_t
599 * Inside a major, we have 16k disks, however mapped non-
600 * contiguously. The first 16 disks are for major0, the next
601 * ones with major1, ... Disk 256 is for major0 again, disk 272
603 * As we stay compatible with our numbering scheme, we can reuse
604 * the well-know SCSI majors 8, 65--71, 136--143.
606 static int sd_major(int major_idx)
610 return SCSI_DISK0_MAJOR;
612 return SCSI_DISK1_MAJOR + major_idx - 1;
614 return SCSI_DISK8_MAJOR + major_idx - 8;
617 return 0; /* shut up gcc */
621 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
623 struct scsi_disk *sdkp = NULL;
625 mutex_lock(&sd_ref_mutex);
627 if (disk->private_data) {
628 sdkp = scsi_disk(disk);
629 if (scsi_device_get(sdkp->device) == 0)
630 get_device(&sdkp->dev);
634 mutex_unlock(&sd_ref_mutex);
638 static void scsi_disk_put(struct scsi_disk *sdkp)
640 struct scsi_device *sdev = sdkp->device;
642 mutex_lock(&sd_ref_mutex);
643 put_device(&sdkp->dev);
644 scsi_device_put(sdev);
645 mutex_unlock(&sd_ref_mutex);
648 #ifdef CONFIG_BLK_SED_OPAL
649 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
650 size_t len, bool send)
652 struct scsi_device *sdev = data;
656 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
658 put_unaligned_be16(spsp, &cdb[2]);
659 put_unaligned_be32(len, &cdb[6]);
661 ret = scsi_execute_req(sdev, cdb,
662 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
663 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
664 return ret <= 0 ? ret : -EIO;
666 #endif /* CONFIG_BLK_SED_OPAL */
668 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
669 unsigned int dix, unsigned int dif)
671 struct bio *bio = scmd->request->bio;
672 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
673 unsigned int protect = 0;
675 if (dix) { /* DIX Type 0, 1, 2, 3 */
676 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
677 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
679 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
680 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
683 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
684 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
686 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
687 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
690 if (dif) { /* DIX/DIF Type 1, 2, 3 */
691 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
693 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
694 protect = 3 << 5; /* Disable target PI checking */
696 protect = 1 << 5; /* Enable target PI checking */
699 scsi_set_prot_op(scmd, prot_op);
700 scsi_set_prot_type(scmd, dif);
701 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
706 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
708 struct request_queue *q = sdkp->disk->queue;
709 unsigned int logical_block_size = sdkp->device->sector_size;
710 unsigned int max_blocks = 0;
712 q->limits.discard_alignment =
713 sdkp->unmap_alignment * logical_block_size;
714 q->limits.discard_granularity =
715 max(sdkp->physical_block_size,
716 sdkp->unmap_granularity * logical_block_size);
717 sdkp->provisioning_mode = mode;
723 blk_queue_max_discard_sectors(q, 0);
724 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
728 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
729 (u32)SD_MAX_WS16_BLOCKS);
733 if (sdkp->device->unmap_limit_for_ws)
734 max_blocks = sdkp->max_unmap_blocks;
736 max_blocks = sdkp->max_ws_blocks;
738 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
742 if (sdkp->device->unmap_limit_for_ws)
743 max_blocks = sdkp->max_unmap_blocks;
745 max_blocks = sdkp->max_ws_blocks;
747 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
751 max_blocks = min_not_zero(sdkp->max_ws_blocks,
752 (u32)SD_MAX_WS10_BLOCKS);
756 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
757 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
760 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
762 struct scsi_device *sdp = cmd->device;
763 struct request *rq = cmd->request;
764 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
765 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
766 unsigned int data_len = 24;
769 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
770 if (!rq->special_vec.bv_page)
771 return BLK_STS_RESOURCE;
772 clear_highpage(rq->special_vec.bv_page);
773 rq->special_vec.bv_offset = 0;
774 rq->special_vec.bv_len = data_len;
775 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
778 cmd->cmnd[0] = UNMAP;
781 buf = page_address(rq->special_vec.bv_page);
782 put_unaligned_be16(6 + 16, &buf[0]);
783 put_unaligned_be16(16, &buf[2]);
784 put_unaligned_be64(sector, &buf[8]);
785 put_unaligned_be32(nr_sectors, &buf[16]);
787 cmd->allowed = SD_MAX_RETRIES;
788 cmd->transfersize = data_len;
789 rq->timeout = SD_TIMEOUT;
790 scsi_req(rq)->resid_len = data_len;
792 return scsi_init_io(cmd);
795 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
798 struct scsi_device *sdp = cmd->device;
799 struct request *rq = cmd->request;
800 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
801 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
802 u32 data_len = sdp->sector_size;
804 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
805 if (!rq->special_vec.bv_page)
806 return BLK_STS_RESOURCE;
807 clear_highpage(rq->special_vec.bv_page);
808 rq->special_vec.bv_offset = 0;
809 rq->special_vec.bv_len = data_len;
810 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
813 cmd->cmnd[0] = WRITE_SAME_16;
815 cmd->cmnd[1] = 0x8; /* UNMAP */
816 put_unaligned_be64(sector, &cmd->cmnd[2]);
817 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
819 cmd->allowed = SD_MAX_RETRIES;
820 cmd->transfersize = data_len;
821 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
822 scsi_req(rq)->resid_len = data_len;
824 return scsi_init_io(cmd);
827 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
830 struct scsi_device *sdp = cmd->device;
831 struct request *rq = cmd->request;
832 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
833 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
834 u32 data_len = sdp->sector_size;
836 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
837 if (!rq->special_vec.bv_page)
838 return BLK_STS_RESOURCE;
839 clear_highpage(rq->special_vec.bv_page);
840 rq->special_vec.bv_offset = 0;
841 rq->special_vec.bv_len = data_len;
842 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
845 cmd->cmnd[0] = WRITE_SAME;
847 cmd->cmnd[1] = 0x8; /* UNMAP */
848 put_unaligned_be32(sector, &cmd->cmnd[2]);
849 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
851 cmd->allowed = SD_MAX_RETRIES;
852 cmd->transfersize = data_len;
853 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
854 scsi_req(rq)->resid_len = data_len;
856 return scsi_init_io(cmd);
859 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
861 struct request *rq = cmd->request;
862 struct scsi_device *sdp = cmd->device;
863 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
864 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
865 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
867 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
868 switch (sdkp->zeroing_mode) {
869 case SD_ZERO_WS16_UNMAP:
870 return sd_setup_write_same16_cmnd(cmd, true);
871 case SD_ZERO_WS10_UNMAP:
872 return sd_setup_write_same10_cmnd(cmd, true);
876 if (sdp->no_write_same)
877 return BLK_STS_TARGET;
879 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
880 return sd_setup_write_same16_cmnd(cmd, false);
882 return sd_setup_write_same10_cmnd(cmd, false);
885 static void sd_config_write_same(struct scsi_disk *sdkp)
887 struct request_queue *q = sdkp->disk->queue;
888 unsigned int logical_block_size = sdkp->device->sector_size;
890 if (sdkp->device->no_write_same) {
891 sdkp->max_ws_blocks = 0;
895 /* Some devices can not handle block counts above 0xffff despite
896 * supporting WRITE SAME(16). Consequently we default to 64k
897 * blocks per I/O unless the device explicitly advertises a
900 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
901 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
902 (u32)SD_MAX_WS16_BLOCKS);
903 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
904 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
905 (u32)SD_MAX_WS10_BLOCKS);
907 sdkp->device->no_write_same = 1;
908 sdkp->max_ws_blocks = 0;
911 if (sdkp->lbprz && sdkp->lbpws)
912 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
913 else if (sdkp->lbprz && sdkp->lbpws10)
914 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
915 else if (sdkp->max_ws_blocks)
916 sdkp->zeroing_mode = SD_ZERO_WS;
918 sdkp->zeroing_mode = SD_ZERO_WRITE;
920 if (sdkp->max_ws_blocks &&
921 sdkp->physical_block_size > logical_block_size) {
923 * Reporting a maximum number of blocks that is not aligned
924 * on the device physical size would cause a large write same
925 * request to be split into physically unaligned chunks by
926 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
927 * even if the caller of these functions took care to align the
928 * large request. So make sure the maximum reported is aligned
929 * to the device physical block size. This is only an optional
930 * optimization for regular disks, but this is mandatory to
931 * avoid failure of large write same requests directed at
932 * sequential write required zones of host-managed ZBC disks.
934 sdkp->max_ws_blocks =
935 round_down(sdkp->max_ws_blocks,
936 bytes_to_logical(sdkp->device,
937 sdkp->physical_block_size));
941 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
942 (logical_block_size >> 9));
943 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
944 (logical_block_size >> 9));
948 * sd_setup_write_same_cmnd - write the same data to multiple blocks
949 * @cmd: command to prepare
951 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
952 * the preference indicated by the target device.
954 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
956 struct request *rq = cmd->request;
957 struct scsi_device *sdp = cmd->device;
958 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
959 struct bio *bio = rq->bio;
960 sector_t sector = blk_rq_pos(rq);
961 unsigned int nr_sectors = blk_rq_sectors(rq);
962 unsigned int nr_bytes = blk_rq_bytes(rq);
965 if (sdkp->device->no_write_same)
966 return BLK_STS_TARGET;
968 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
970 sector >>= ilog2(sdp->sector_size) - 9;
971 nr_sectors >>= ilog2(sdp->sector_size) - 9;
973 rq->timeout = SD_WRITE_SAME_TIMEOUT;
975 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
977 cmd->cmnd[0] = WRITE_SAME_16;
978 put_unaligned_be64(sector, &cmd->cmnd[2]);
979 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
982 cmd->cmnd[0] = WRITE_SAME;
983 put_unaligned_be32(sector, &cmd->cmnd[2]);
984 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
987 cmd->transfersize = sdp->sector_size;
988 cmd->allowed = SD_MAX_RETRIES;
991 * For WRITE SAME the data transferred via the DATA OUT buffer is
992 * different from the amount of data actually written to the target.
994 * We set up __data_len to the amount of data transferred via the
995 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
996 * to transfer a single sector of data first, but then reset it to
997 * the amount of data to be written right after so that the I/O path
998 * knows how much to actually write.
1000 rq->__data_len = sdp->sector_size;
1001 ret = scsi_init_io(cmd);
1002 rq->__data_len = nr_bytes;
1007 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1009 struct request *rq = cmd->request;
1011 /* flush requests don't perform I/O, zero the S/G table */
1012 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1014 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1016 cmd->transfersize = 0;
1017 cmd->allowed = SD_MAX_RETRIES;
1019 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1023 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1025 struct request *rq = SCpnt->request;
1026 struct scsi_device *sdp = SCpnt->device;
1027 struct gendisk *disk = rq->rq_disk;
1028 struct scsi_disk *sdkp = scsi_disk(disk);
1029 sector_t block = blk_rq_pos(rq);
1031 unsigned int this_count = blk_rq_sectors(rq);
1032 unsigned int dif, dix;
1033 unsigned char protect;
1036 ret = scsi_init_io(SCpnt);
1037 if (ret != BLK_STS_OK)
1039 WARN_ON_ONCE(SCpnt != rq->special);
1042 scmd_printk(KERN_INFO, SCpnt,
1043 "%s: block=%llu, count=%d\n",
1044 __func__, (unsigned long long)block, this_count));
1046 if (!sdp || !scsi_device_online(sdp) ||
1047 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1048 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1049 "Finishing %u sectors\n",
1050 blk_rq_sectors(rq)));
1051 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1052 "Retry with 0x%p\n", SCpnt));
1053 return BLK_STS_IOERR;
1058 * quietly refuse to do anything to a changed disc until
1059 * the changed bit has been reset
1061 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1062 return BLK_STS_IOERR;
1066 * Some SD card readers can't handle multi-sector accesses which touch
1067 * the last one or two hardware sectors. Split accesses as needed.
1069 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1070 (sdp->sector_size / 512);
1072 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1073 if (block < threshold) {
1074 /* Access up to the threshold but not beyond */
1075 this_count = threshold - block;
1077 /* Access only a single hardware sector */
1078 this_count = sdp->sector_size / 512;
1082 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1083 (unsigned long long)block));
1086 * If we have a 1K hardware sectorsize, prevent access to single
1087 * 512 byte sectors. In theory we could handle this - in fact
1088 * the scsi cdrom driver must be able to handle this because
1089 * we typically use 1K blocksizes, and cdroms typically have
1090 * 2K hardware sectorsizes. Of course, things are simpler
1091 * with the cdrom, since it is read-only. For performance
1092 * reasons, the filesystems should be able to handle this
1093 * and not force the scsi disk driver to use bounce buffers
1096 if (sdp->sector_size == 1024) {
1097 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1098 scmd_printk(KERN_ERR, SCpnt,
1099 "Bad block number requested\n");
1100 return BLK_STS_IOERR;
1103 this_count = this_count >> 1;
1105 if (sdp->sector_size == 2048) {
1106 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1107 scmd_printk(KERN_ERR, SCpnt,
1108 "Bad block number requested\n");
1109 return BLK_STS_IOERR;
1112 this_count = this_count >> 2;
1114 if (sdp->sector_size == 4096) {
1115 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1116 scmd_printk(KERN_ERR, SCpnt,
1117 "Bad block number requested\n");
1118 return BLK_STS_IOERR;
1121 this_count = this_count >> 3;
1123 if (rq_data_dir(rq) == WRITE) {
1124 SCpnt->cmnd[0] = WRITE_6;
1126 if (blk_integrity_rq(rq))
1127 t10_pi_prepare(SCpnt->request, sdkp->protection_type);
1129 } else if (rq_data_dir(rq) == READ) {
1130 SCpnt->cmnd[0] = READ_6;
1132 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1133 return BLK_STS_IOERR;
1136 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1137 "%s %d/%u 512 byte blocks.\n",
1138 (rq_data_dir(rq) == WRITE) ?
1139 "writing" : "reading", this_count,
1140 blk_rq_sectors(rq)));
1142 dix = scsi_prot_sg_count(SCpnt);
1143 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1146 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1150 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1151 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1153 if (unlikely(!SCpnt->cmnd))
1154 return BLK_STS_RESOURCE;
1156 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1157 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1158 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1159 SCpnt->cmnd[7] = 0x18;
1160 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1161 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1164 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1165 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1166 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1167 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1168 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1169 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1170 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1171 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1173 /* Expected Indirect LBA */
1174 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1175 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1176 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1177 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1179 /* Transfer length */
1180 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1181 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1182 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1183 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1184 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1185 SCpnt->cmnd[0] += READ_16 - READ_6;
1186 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1187 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1188 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1189 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1190 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1191 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1192 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1193 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1194 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1195 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1196 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1197 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1198 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1199 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1200 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1201 scsi_device_protection(SCpnt->device) ||
1202 SCpnt->device->use_10_for_rw) {
1203 SCpnt->cmnd[0] += READ_10 - READ_6;
1204 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1205 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1206 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1207 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1208 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1209 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1210 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1211 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1213 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1215 * This happens only if this drive failed
1216 * 10byte rw command with ILLEGAL_REQUEST
1217 * during operation and thus turned off
1220 scmd_printk(KERN_ERR, SCpnt,
1221 "FUA write on READ/WRITE(6) drive\n");
1222 return BLK_STS_IOERR;
1225 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1226 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1227 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1228 SCpnt->cmnd[4] = (unsigned char) this_count;
1231 SCpnt->sdb.length = this_count * sdp->sector_size;
1234 * We shouldn't disconnect in the middle of a sector, so with a dumb
1235 * host adapter, it's safe to assume that we can at least transfer
1236 * this many bytes between each connect / disconnect.
1238 SCpnt->transfersize = sdp->sector_size;
1239 SCpnt->underflow = this_count << 9;
1240 SCpnt->allowed = SD_MAX_RETRIES;
1243 * This indicates that the command is ready from our end to be
1249 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1251 struct request *rq = cmd->request;
1253 switch (req_op(rq)) {
1254 case REQ_OP_DISCARD:
1255 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1257 return sd_setup_unmap_cmnd(cmd);
1259 return sd_setup_write_same16_cmnd(cmd, true);
1261 return sd_setup_write_same10_cmnd(cmd, true);
1263 return sd_setup_write_same10_cmnd(cmd, false);
1265 return BLK_STS_TARGET;
1267 case REQ_OP_WRITE_ZEROES:
1268 return sd_setup_write_zeroes_cmnd(cmd);
1269 case REQ_OP_WRITE_SAME:
1270 return sd_setup_write_same_cmnd(cmd);
1272 return sd_setup_flush_cmnd(cmd);
1275 return sd_setup_read_write_cmnd(cmd);
1276 case REQ_OP_ZONE_RESET:
1277 return sd_zbc_setup_reset_cmnd(cmd);
1280 return BLK_STS_NOTSUPP;
1284 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1286 struct request *rq = SCpnt->request;
1289 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1290 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1292 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1296 mempool_free(cmnd, sd_cdb_pool);
1301 * sd_open - open a scsi disk device
1302 * @bdev: Block device of the scsi disk to open
1303 * @mode: FMODE_* mask
1305 * Returns 0 if successful. Returns a negated errno value in case
1308 * Note: This can be called from a user context (e.g. fsck(1) )
1309 * or from within the kernel (e.g. as a result of a mount(1) ).
1310 * In the latter case @inode and @filp carry an abridged amount
1311 * of information as noted above.
1313 * Locking: called with bdev->bd_mutex held.
1315 static int sd_open(struct block_device *bdev, fmode_t mode)
1317 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1318 struct scsi_device *sdev;
1324 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1326 sdev = sdkp->device;
1329 * If the device is in error recovery, wait until it is done.
1330 * If the device is offline, then disallow any access to it.
1333 if (!scsi_block_when_processing_errors(sdev))
1336 if (sdev->removable || sdkp->write_prot)
1337 check_disk_change(bdev);
1340 * If the drive is empty, just let the open fail.
1342 retval = -ENOMEDIUM;
1343 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1347 * If the device has the write protect tab set, have the open fail
1348 * if the user expects to be able to write to the thing.
1351 if (sdkp->write_prot && (mode & FMODE_WRITE))
1355 * It is possible that the disk changing stuff resulted in
1356 * the device being taken offline. If this is the case,
1357 * report this to the user, and don't pretend that the
1358 * open actually succeeded.
1361 if (!scsi_device_online(sdev))
1364 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1365 if (scsi_block_when_processing_errors(sdev))
1366 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1372 scsi_disk_put(sdkp);
1377 * sd_release - invoked when the (last) close(2) is called on this
1379 * @disk: disk to release
1380 * @mode: FMODE_* mask
1384 * Note: may block (uninterruptible) if error recovery is underway
1387 * Locking: called with bdev->bd_mutex held.
1389 static void sd_release(struct gendisk *disk, fmode_t mode)
1391 struct scsi_disk *sdkp = scsi_disk(disk);
1392 struct scsi_device *sdev = sdkp->device;
1394 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1396 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1397 if (scsi_block_when_processing_errors(sdev))
1398 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1402 * XXX and what if there are packets in flight and this close()
1403 * XXX is followed by a "rmmod sd_mod"?
1406 scsi_disk_put(sdkp);
1409 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1411 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1412 struct scsi_device *sdp = sdkp->device;
1413 struct Scsi_Host *host = sdp->host;
1414 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1417 /* default to most commonly used values */
1418 diskinfo[0] = 0x40; /* 1 << 6 */
1419 diskinfo[1] = 0x20; /* 1 << 5 */
1420 diskinfo[2] = capacity >> 11;
1422 /* override with calculated, extended default, or driver values */
1423 if (host->hostt->bios_param)
1424 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1426 scsicam_bios_param(bdev, capacity, diskinfo);
1428 geo->heads = diskinfo[0];
1429 geo->sectors = diskinfo[1];
1430 geo->cylinders = diskinfo[2];
1435 * sd_ioctl - process an ioctl
1436 * @bdev: target block device
1437 * @mode: FMODE_* mask
1438 * @cmd: ioctl command number
1439 * @arg: this is third argument given to ioctl(2) system call.
1440 * Often contains a pointer.
1442 * Returns 0 if successful (some ioctls return positive numbers on
1443 * success as well). Returns a negated errno value in case of error.
1445 * Note: most ioctls are forward onto the block subsystem or further
1446 * down in the scsi subsystem.
1448 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1449 unsigned int cmd, unsigned long arg)
1451 struct gendisk *disk = bdev->bd_disk;
1452 struct scsi_disk *sdkp = scsi_disk(disk);
1453 struct scsi_device *sdp = sdkp->device;
1454 void __user *p = (void __user *)arg;
1457 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1458 "cmd=0x%x\n", disk->disk_name, cmd));
1460 error = scsi_verify_blk_ioctl(bdev, cmd);
1465 * If we are in the middle of error recovery, don't let anyone
1466 * else try and use this device. Also, if error recovery fails, it
1467 * may try and take the device offline, in which case all further
1468 * access to the device is prohibited.
1470 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1471 (mode & FMODE_NDELAY) != 0);
1475 if (is_sed_ioctl(cmd))
1476 return sed_ioctl(sdkp->opal_dev, cmd, p);
1479 * Send SCSI addressing ioctls directly to mid level, send other
1480 * ioctls to block level and then onto mid level if they can't be
1484 case SCSI_IOCTL_GET_IDLUN:
1485 case SCSI_IOCTL_GET_BUS_NUMBER:
1486 error = scsi_ioctl(sdp, cmd, p);
1489 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1490 if (error != -ENOTTY)
1492 error = scsi_ioctl(sdp, cmd, p);
1499 static void set_media_not_present(struct scsi_disk *sdkp)
1501 if (sdkp->media_present)
1502 sdkp->device->changed = 1;
1504 if (sdkp->device->removable) {
1505 sdkp->media_present = 0;
1510 static int media_not_present(struct scsi_disk *sdkp,
1511 struct scsi_sense_hdr *sshdr)
1513 if (!scsi_sense_valid(sshdr))
1516 /* not invoked for commands that could return deferred errors */
1517 switch (sshdr->sense_key) {
1518 case UNIT_ATTENTION:
1520 /* medium not present */
1521 if (sshdr->asc == 0x3A) {
1522 set_media_not_present(sdkp);
1530 * sd_check_events - check media events
1531 * @disk: kernel device descriptor
1532 * @clearing: disk events currently being cleared
1534 * Returns mask of DISK_EVENT_*.
1536 * Note: this function is invoked from the block subsystem.
1538 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1540 struct scsi_disk *sdkp = scsi_disk_get(disk);
1541 struct scsi_device *sdp;
1548 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1551 * If the device is offline, don't send any commands - just pretend as
1552 * if the command failed. If the device ever comes back online, we
1553 * can deal with it then. It is only because of unrecoverable errors
1554 * that we would ever take a device offline in the first place.
1556 if (!scsi_device_online(sdp)) {
1557 set_media_not_present(sdkp);
1562 * Using TEST_UNIT_READY enables differentiation between drive with
1563 * no cartridge loaded - NOT READY, drive with changed cartridge -
1564 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1566 * Drives that auto spin down. eg iomega jaz 1G, will be started
1567 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1568 * sd_revalidate() is called.
1570 if (scsi_block_when_processing_errors(sdp)) {
1571 struct scsi_sense_hdr sshdr = { 0, };
1573 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1576 /* failed to execute TUR, assume media not present */
1577 if (host_byte(retval)) {
1578 set_media_not_present(sdkp);
1582 if (media_not_present(sdkp, &sshdr))
1587 * For removable scsi disk we have to recognise the presence
1588 * of a disk in the drive.
1590 if (!sdkp->media_present)
1592 sdkp->media_present = 1;
1595 * sdp->changed is set under the following conditions:
1597 * Medium present state has changed in either direction.
1598 * Device has indicated UNIT_ATTENTION.
1600 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1602 scsi_disk_put(sdkp);
1606 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1609 struct scsi_device *sdp = sdkp->device;
1610 const int timeout = sdp->request_queue->rq_timeout
1611 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1612 struct scsi_sense_hdr my_sshdr;
1614 if (!scsi_device_online(sdp))
1617 /* caller might not be interested in sense, but we need it */
1621 for (retries = 3; retries > 0; --retries) {
1622 unsigned char cmd[10] = { 0 };
1624 cmd[0] = SYNCHRONIZE_CACHE;
1626 * Leave the rest of the command zero to indicate
1629 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1630 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1636 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1638 if (driver_byte(res) == DRIVER_SENSE)
1639 sd_print_sense_hdr(sdkp, sshdr);
1641 /* we need to evaluate the error return */
1642 if (scsi_sense_valid(sshdr) &&
1643 (sshdr->asc == 0x3a || /* medium not present */
1644 sshdr->asc == 0x20)) /* invalid command */
1645 /* this is no error here */
1648 switch (host_byte(res)) {
1649 /* ignore errors due to racing a disconnection */
1650 case DID_BAD_TARGET:
1651 case DID_NO_CONNECT:
1653 /* signal the upper layer it might try again */
1657 case DID_SOFT_ERROR:
1666 static void sd_rescan(struct device *dev)
1668 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1670 revalidate_disk(sdkp->disk);
1674 #ifdef CONFIG_COMPAT
1676 * This gets directly called from VFS. When the ioctl
1677 * is not recognized we go back to the other translation paths.
1679 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1680 unsigned int cmd, unsigned long arg)
1682 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1685 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1686 (mode & FMODE_NDELAY) != 0);
1691 * Let the static ioctl translation table take care of it.
1693 if (!sdev->host->hostt->compat_ioctl)
1694 return -ENOIOCTLCMD;
1695 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1699 static char sd_pr_type(enum pr_type type)
1702 case PR_WRITE_EXCLUSIVE:
1704 case PR_EXCLUSIVE_ACCESS:
1706 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1708 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1710 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1712 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1719 static int sd_pr_command(struct block_device *bdev, u8 sa,
1720 u64 key, u64 sa_key, u8 type, u8 flags)
1722 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1723 struct scsi_sense_hdr sshdr;
1725 u8 cmd[16] = { 0, };
1726 u8 data[24] = { 0, };
1728 cmd[0] = PERSISTENT_RESERVE_OUT;
1731 put_unaligned_be32(sizeof(data), &cmd[5]);
1733 put_unaligned_be64(key, &data[0]);
1734 put_unaligned_be64(sa_key, &data[8]);
1737 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1738 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1740 if (driver_byte(result) == DRIVER_SENSE &&
1741 scsi_sense_valid(&sshdr)) {
1742 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1743 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1749 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1752 if (flags & ~PR_FL_IGNORE_KEY)
1754 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1755 old_key, new_key, 0,
1756 (1 << 0) /* APTPL */);
1759 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1764 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1767 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1769 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1772 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1773 enum pr_type type, bool abort)
1775 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1776 sd_pr_type(type), 0);
1779 static int sd_pr_clear(struct block_device *bdev, u64 key)
1781 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1784 static const struct pr_ops sd_pr_ops = {
1785 .pr_register = sd_pr_register,
1786 .pr_reserve = sd_pr_reserve,
1787 .pr_release = sd_pr_release,
1788 .pr_preempt = sd_pr_preempt,
1789 .pr_clear = sd_pr_clear,
1792 static const struct block_device_operations sd_fops = {
1793 .owner = THIS_MODULE,
1795 .release = sd_release,
1797 .getgeo = sd_getgeo,
1798 #ifdef CONFIG_COMPAT
1799 .compat_ioctl = sd_compat_ioctl,
1801 .check_events = sd_check_events,
1802 .revalidate_disk = sd_revalidate_disk,
1803 .unlock_native_capacity = sd_unlock_native_capacity,
1804 .report_zones = sd_zbc_report_zones,
1805 .pr_ops = &sd_pr_ops,
1809 * sd_eh_reset - reset error handling callback
1810 * @scmd: sd-issued command that has failed
1812 * This function is called by the SCSI midlayer before starting
1813 * SCSI EH. When counting medium access failures we have to be
1814 * careful to register it only only once per device and SCSI EH run;
1815 * there might be several timed out commands which will cause the
1816 * 'max_medium_access_timeouts' counter to trigger after the first
1817 * SCSI EH run already and set the device to offline.
1818 * So this function resets the internal counter before starting SCSI EH.
1820 static void sd_eh_reset(struct scsi_cmnd *scmd)
1822 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1824 /* New SCSI EH run, reset gate variable */
1825 sdkp->ignore_medium_access_errors = false;
1829 * sd_eh_action - error handling callback
1830 * @scmd: sd-issued command that has failed
1831 * @eh_disp: The recovery disposition suggested by the midlayer
1833 * This function is called by the SCSI midlayer upon completion of an
1834 * error test command (currently TEST UNIT READY). The result of sending
1835 * the eh command is passed in eh_disp. We're looking for devices that
1836 * fail medium access commands but are OK with non access commands like
1837 * test unit ready (so wrongly see the device as having a successful
1840 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1842 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1843 struct scsi_device *sdev = scmd->device;
1845 if (!scsi_device_online(sdev) ||
1846 !scsi_medium_access_command(scmd) ||
1847 host_byte(scmd->result) != DID_TIME_OUT ||
1852 * The device has timed out executing a medium access command.
1853 * However, the TEST UNIT READY command sent during error
1854 * handling completed successfully. Either the device is in the
1855 * process of recovering or has it suffered an internal failure
1856 * that prevents access to the storage medium.
1858 if (!sdkp->ignore_medium_access_errors) {
1859 sdkp->medium_access_timed_out++;
1860 sdkp->ignore_medium_access_errors = true;
1864 * If the device keeps failing read/write commands but TEST UNIT
1865 * READY always completes successfully we assume that medium
1866 * access is no longer possible and take the device offline.
1868 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1869 scmd_printk(KERN_ERR, scmd,
1870 "Medium access timeout failure. Offlining disk!\n");
1871 mutex_lock(&sdev->state_mutex);
1872 scsi_device_set_state(sdev, SDEV_OFFLINE);
1873 mutex_unlock(&sdev->state_mutex);
1881 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1883 struct request *req = scmd->request;
1884 struct scsi_device *sdev = scmd->device;
1885 unsigned int transferred, good_bytes;
1886 u64 start_lba, end_lba, bad_lba;
1889 * Some commands have a payload smaller than the device logical
1890 * block size (e.g. INQUIRY on a 4K disk).
1892 if (scsi_bufflen(scmd) <= sdev->sector_size)
1895 /* Check if we have a 'bad_lba' information */
1896 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1897 SCSI_SENSE_BUFFERSIZE,
1902 * If the bad lba was reported incorrectly, we have no idea where
1905 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1906 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1907 if (bad_lba < start_lba || bad_lba >= end_lba)
1911 * resid is optional but mostly filled in. When it's unused,
1912 * its value is zero, so we assume the whole buffer transferred
1914 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1916 /* This computation should always be done in terms of the
1917 * resolution of the device's medium.
1919 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1921 return min(good_bytes, transferred);
1925 * sd_done - bottom half handler: called when the lower level
1926 * driver has completed (successfully or otherwise) a scsi command.
1927 * @SCpnt: mid-level's per command structure.
1929 * Note: potentially run from within an ISR. Must not block.
1931 static int sd_done(struct scsi_cmnd *SCpnt)
1933 int result = SCpnt->result;
1934 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1935 unsigned int sector_size = SCpnt->device->sector_size;
1937 struct scsi_sense_hdr sshdr;
1938 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1939 struct request *req = SCpnt->request;
1940 int sense_valid = 0;
1941 int sense_deferred = 0;
1943 switch (req_op(req)) {
1944 case REQ_OP_DISCARD:
1945 case REQ_OP_WRITE_ZEROES:
1946 case REQ_OP_WRITE_SAME:
1947 case REQ_OP_ZONE_RESET:
1949 good_bytes = blk_rq_bytes(req);
1950 scsi_set_resid(SCpnt, 0);
1953 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1958 * In case of bogus fw or device, we could end up having
1959 * an unaligned partial completion. Check this here and force
1962 resid = scsi_get_resid(SCpnt);
1963 if (resid & (sector_size - 1)) {
1964 sd_printk(KERN_INFO, sdkp,
1965 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1966 resid, sector_size);
1967 resid = min(scsi_bufflen(SCpnt),
1968 round_up(resid, sector_size));
1969 scsi_set_resid(SCpnt, resid);
1974 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1976 sense_deferred = scsi_sense_is_deferred(&sshdr);
1978 sdkp->medium_access_timed_out = 0;
1980 if (driver_byte(result) != DRIVER_SENSE &&
1981 (!sense_valid || sense_deferred))
1984 switch (sshdr.sense_key) {
1985 case HARDWARE_ERROR:
1987 good_bytes = sd_completed_bytes(SCpnt);
1989 case RECOVERED_ERROR:
1990 good_bytes = scsi_bufflen(SCpnt);
1993 /* This indicates a false check condition, so ignore it. An
1994 * unknown amount of data was transferred so treat it as an
1998 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2000 case ABORTED_COMMAND:
2001 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2002 good_bytes = sd_completed_bytes(SCpnt);
2004 case ILLEGAL_REQUEST:
2005 switch (sshdr.asc) {
2006 case 0x10: /* DIX: Host detected corruption */
2007 good_bytes = sd_completed_bytes(SCpnt);
2009 case 0x20: /* INVALID COMMAND OPCODE */
2010 case 0x24: /* INVALID FIELD IN CDB */
2011 switch (SCpnt->cmnd[0]) {
2013 sd_config_discard(sdkp, SD_LBP_DISABLE);
2017 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2018 sd_config_discard(sdkp, SD_LBP_DISABLE);
2020 sdkp->device->no_write_same = 1;
2021 sd_config_write_same(sdkp);
2022 req->rq_flags |= RQF_QUIET;
2033 if (sd_is_zoned(sdkp))
2034 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2036 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2037 "sd_done: completed %d of %d bytes\n",
2038 good_bytes, scsi_bufflen(SCpnt)));
2040 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2042 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2043 good_bytes / scsi_prot_interval(SCpnt));
2049 * spinup disk - called only in sd_revalidate_disk()
2052 sd_spinup_disk(struct scsi_disk *sdkp)
2054 unsigned char cmd[10];
2055 unsigned long spintime_expire = 0;
2056 int retries, spintime;
2057 unsigned int the_result;
2058 struct scsi_sense_hdr sshdr;
2059 int sense_valid = 0;
2063 /* Spin up drives, as required. Only do this at boot time */
2064 /* Spinup needs to be done for module loads too. */
2069 cmd[0] = TEST_UNIT_READY;
2070 memset((void *) &cmd[1], 0, 9);
2072 the_result = scsi_execute_req(sdkp->device, cmd,
2075 SD_MAX_RETRIES, NULL);
2078 * If the drive has indicated to us that it
2079 * doesn't have any media in it, don't bother
2080 * with any more polling.
2082 if (media_not_present(sdkp, &sshdr))
2086 sense_valid = scsi_sense_valid(&sshdr);
2088 } while (retries < 3 &&
2089 (!scsi_status_is_good(the_result) ||
2090 ((driver_byte(the_result) == DRIVER_SENSE) &&
2091 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2093 if (driver_byte(the_result) != DRIVER_SENSE) {
2094 /* no sense, TUR either succeeded or failed
2095 * with a status error */
2096 if(!spintime && !scsi_status_is_good(the_result)) {
2097 sd_print_result(sdkp, "Test Unit Ready failed",
2104 * The device does not want the automatic start to be issued.
2106 if (sdkp->device->no_start_on_add)
2109 if (sense_valid && sshdr.sense_key == NOT_READY) {
2110 if (sshdr.asc == 4 && sshdr.ascq == 3)
2111 break; /* manual intervention required */
2112 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2113 break; /* standby */
2114 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2115 break; /* unavailable */
2116 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2117 break; /* sanitize in progress */
2119 * Issue command to spin up drive when not ready
2122 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2123 cmd[0] = START_STOP;
2124 cmd[1] = 1; /* Return immediately */
2125 memset((void *) &cmd[2], 0, 8);
2126 cmd[4] = 1; /* Start spin cycle */
2127 if (sdkp->device->start_stop_pwr_cond)
2129 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2131 SD_TIMEOUT, SD_MAX_RETRIES,
2133 spintime_expire = jiffies + 100 * HZ;
2136 /* Wait 1 second for next try */
2138 printk(KERN_CONT ".");
2141 * Wait for USB flash devices with slow firmware.
2142 * Yes, this sense key/ASC combination shouldn't
2143 * occur here. It's characteristic of these devices.
2145 } else if (sense_valid &&
2146 sshdr.sense_key == UNIT_ATTENTION &&
2147 sshdr.asc == 0x28) {
2149 spintime_expire = jiffies + 5 * HZ;
2152 /* Wait 1 second for next try */
2155 /* we don't understand the sense code, so it's
2156 * probably pointless to loop */
2158 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2159 sd_print_sense_hdr(sdkp, &sshdr);
2164 } while (spintime && time_before_eq(jiffies, spintime_expire));
2167 if (scsi_status_is_good(the_result))
2168 printk(KERN_CONT "ready\n");
2170 printk(KERN_CONT "not responding...\n");
2175 * Determine whether disk supports Data Integrity Field.
2177 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2179 struct scsi_device *sdp = sdkp->device;
2183 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2186 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2188 if (type > T10_PI_TYPE3_PROTECTION)
2190 else if (scsi_host_dif_capable(sdp->host, type))
2193 if (sdkp->first_scan || type != sdkp->protection_type)
2196 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2197 " protection type %u. Disabling disk!\n",
2201 sd_printk(KERN_NOTICE, sdkp,
2202 "Enabling DIF Type %u protection\n", type);
2205 sd_printk(KERN_NOTICE, sdkp,
2206 "Disabling DIF Type %u protection\n", type);
2210 sdkp->protection_type = type;
2215 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2216 struct scsi_sense_hdr *sshdr, int sense_valid,
2219 if (driver_byte(the_result) == DRIVER_SENSE)
2220 sd_print_sense_hdr(sdkp, sshdr);
2222 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2225 * Set dirty bit for removable devices if not ready -
2226 * sometimes drives will not report this properly.
2228 if (sdp->removable &&
2229 sense_valid && sshdr->sense_key == NOT_READY)
2230 set_media_not_present(sdkp);
2233 * We used to set media_present to 0 here to indicate no media
2234 * in the drive, but some drives fail read capacity even with
2235 * media present, so we can't do that.
2237 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2241 #if RC16_LEN > SD_BUF_SIZE
2242 #error RC16_LEN must not be more than SD_BUF_SIZE
2245 #define READ_CAPACITY_RETRIES_ON_RESET 10
2248 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2249 * and the reported logical block size is bigger than 512 bytes. Note
2250 * that last_sector is a u64 and therefore logical_to_sectors() is not
2253 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2255 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2257 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2263 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2264 unsigned char *buffer)
2266 unsigned char cmd[16];
2267 struct scsi_sense_hdr sshdr;
2268 int sense_valid = 0;
2270 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2271 unsigned int alignment;
2272 unsigned long long lba;
2273 unsigned sector_size;
2275 if (sdp->no_read_capacity_16)
2280 cmd[0] = SERVICE_ACTION_IN_16;
2281 cmd[1] = SAI_READ_CAPACITY_16;
2283 memset(buffer, 0, RC16_LEN);
2285 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2286 buffer, RC16_LEN, &sshdr,
2287 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2289 if (media_not_present(sdkp, &sshdr))
2293 sense_valid = scsi_sense_valid(&sshdr);
2295 sshdr.sense_key == ILLEGAL_REQUEST &&
2296 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2298 /* Invalid Command Operation Code or
2299 * Invalid Field in CDB, just retry
2300 * silently with RC10 */
2303 sshdr.sense_key == UNIT_ATTENTION &&
2304 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2305 /* Device reset might occur several times,
2306 * give it one more chance */
2307 if (--reset_retries > 0)
2312 } while (the_result && retries);
2315 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2316 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2320 sector_size = get_unaligned_be32(&buffer[8]);
2321 lba = get_unaligned_be64(&buffer[0]);
2323 if (sd_read_protection_type(sdkp, buffer) < 0) {
2328 if (!sd_addressable_capacity(lba, sector_size)) {
2329 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2330 "kernel compiled with support for large block "
2336 /* Logical blocks per physical block exponent */
2337 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2340 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2342 /* Lowest aligned logical block */
2343 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2344 blk_queue_alignment_offset(sdp->request_queue, alignment);
2345 if (alignment && sdkp->first_scan)
2346 sd_printk(KERN_NOTICE, sdkp,
2347 "physical block alignment offset: %u\n", alignment);
2349 if (buffer[14] & 0x80) { /* LBPME */
2352 if (buffer[14] & 0x40) /* LBPRZ */
2355 sd_config_discard(sdkp, SD_LBP_WS16);
2358 sdkp->capacity = lba + 1;
2362 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2363 unsigned char *buffer)
2365 unsigned char cmd[16];
2366 struct scsi_sense_hdr sshdr;
2367 int sense_valid = 0;
2369 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2371 unsigned sector_size;
2374 cmd[0] = READ_CAPACITY;
2375 memset(&cmd[1], 0, 9);
2376 memset(buffer, 0, 8);
2378 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2380 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2382 if (media_not_present(sdkp, &sshdr))
2386 sense_valid = scsi_sense_valid(&sshdr);
2388 sshdr.sense_key == UNIT_ATTENTION &&
2389 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2390 /* Device reset might occur several times,
2391 * give it one more chance */
2392 if (--reset_retries > 0)
2397 } while (the_result && retries);
2400 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2401 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2405 sector_size = get_unaligned_be32(&buffer[4]);
2406 lba = get_unaligned_be32(&buffer[0]);
2408 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2409 /* Some buggy (usb cardreader) devices return an lba of
2410 0xffffffff when the want to report a size of 0 (with
2411 which they really mean no media is present) */
2413 sdkp->physical_block_size = sector_size;
2417 if (!sd_addressable_capacity(lba, sector_size)) {
2418 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2419 "kernel compiled with support for large block "
2425 sdkp->capacity = lba + 1;
2426 sdkp->physical_block_size = sector_size;
2430 static int sd_try_rc16_first(struct scsi_device *sdp)
2432 if (sdp->host->max_cmd_len < 16)
2434 if (sdp->try_rc_10_first)
2436 if (sdp->scsi_level > SCSI_SPC_2)
2438 if (scsi_device_protection(sdp))
2444 * read disk capacity
2447 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2450 struct scsi_device *sdp = sdkp->device;
2452 if (sd_try_rc16_first(sdp)) {
2453 sector_size = read_capacity_16(sdkp, sdp, buffer);
2454 if (sector_size == -EOVERFLOW)
2456 if (sector_size == -ENODEV)
2458 if (sector_size < 0)
2459 sector_size = read_capacity_10(sdkp, sdp, buffer);
2460 if (sector_size < 0)
2463 sector_size = read_capacity_10(sdkp, sdp, buffer);
2464 if (sector_size == -EOVERFLOW)
2466 if (sector_size < 0)
2468 if ((sizeof(sdkp->capacity) > 4) &&
2469 (sdkp->capacity > 0xffffffffULL)) {
2470 int old_sector_size = sector_size;
2471 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2472 "Trying to use READ CAPACITY(16).\n");
2473 sector_size = read_capacity_16(sdkp, sdp, buffer);
2474 if (sector_size < 0) {
2475 sd_printk(KERN_NOTICE, sdkp,
2476 "Using 0xffffffff as device size\n");
2477 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2478 sector_size = old_sector_size;
2481 /* Remember that READ CAPACITY(16) succeeded */
2482 sdp->try_rc_10_first = 0;
2486 /* Some devices are known to return the total number of blocks,
2487 * not the highest block number. Some devices have versions
2488 * which do this and others which do not. Some devices we might
2489 * suspect of doing this but we don't know for certain.
2491 * If we know the reported capacity is wrong, decrement it. If
2492 * we can only guess, then assume the number of blocks is even
2493 * (usually true but not always) and err on the side of lowering
2496 if (sdp->fix_capacity ||
2497 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2498 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2499 "from its reported value: %llu\n",
2500 (unsigned long long) sdkp->capacity);
2505 if (sector_size == 0) {
2507 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2511 if (sector_size != 512 &&
2512 sector_size != 1024 &&
2513 sector_size != 2048 &&
2514 sector_size != 4096) {
2515 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2518 * The user might want to re-format the drive with
2519 * a supported sectorsize. Once this happens, it
2520 * would be relatively trivial to set the thing up.
2521 * For this reason, we leave the thing in the table.
2525 * set a bogus sector size so the normal read/write
2526 * logic in the block layer will eventually refuse any
2527 * request on this device without tripping over power
2528 * of two sector size assumptions
2532 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2533 blk_queue_physical_block_size(sdp->request_queue,
2534 sdkp->physical_block_size);
2535 sdkp->device->sector_size = sector_size;
2537 if (sdkp->capacity > 0xffffffff)
2538 sdp->use_16_for_rw = 1;
2543 * Print disk capacity
2546 sd_print_capacity(struct scsi_disk *sdkp,
2547 sector_t old_capacity)
2549 int sector_size = sdkp->device->sector_size;
2550 char cap_str_2[10], cap_str_10[10];
2552 string_get_size(sdkp->capacity, sector_size,
2553 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2554 string_get_size(sdkp->capacity, sector_size,
2555 STRING_UNITS_10, cap_str_10,
2556 sizeof(cap_str_10));
2558 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2559 sd_printk(KERN_NOTICE, sdkp,
2560 "%llu %d-byte logical blocks: (%s/%s)\n",
2561 (unsigned long long)sdkp->capacity,
2562 sector_size, cap_str_10, cap_str_2);
2564 if (sdkp->physical_block_size != sector_size)
2565 sd_printk(KERN_NOTICE, sdkp,
2566 "%u-byte physical blocks\n",
2567 sdkp->physical_block_size);
2569 sd_zbc_print_zones(sdkp);
2573 /* called with buffer of length 512 */
2575 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2576 unsigned char *buffer, int len, struct scsi_mode_data *data,
2577 struct scsi_sense_hdr *sshdr)
2579 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2580 SD_TIMEOUT, SD_MAX_RETRIES, data,
2585 * read write protect setting, if possible - called only in sd_revalidate_disk()
2586 * called with buffer of length SD_BUF_SIZE
2589 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2592 struct scsi_device *sdp = sdkp->device;
2593 struct scsi_mode_data data;
2594 int disk_ro = get_disk_ro(sdkp->disk);
2595 int old_wp = sdkp->write_prot;
2597 set_disk_ro(sdkp->disk, 0);
2598 if (sdp->skip_ms_page_3f) {
2599 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2603 if (sdp->use_192_bytes_for_3f) {
2604 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2607 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2608 * We have to start carefully: some devices hang if we ask
2609 * for more than is available.
2611 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2614 * Second attempt: ask for page 0 When only page 0 is
2615 * implemented, a request for page 3F may return Sense Key
2616 * 5: Illegal Request, Sense Code 24: Invalid field in
2619 if (!scsi_status_is_good(res))
2620 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2623 * Third attempt: ask 255 bytes, as we did earlier.
2625 if (!scsi_status_is_good(res))
2626 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2630 if (!scsi_status_is_good(res)) {
2631 sd_first_printk(KERN_WARNING, sdkp,
2632 "Test WP failed, assume Write Enabled\n");
2634 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2635 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2636 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2637 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2638 sdkp->write_prot ? "on" : "off");
2639 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2645 * sd_read_cache_type - called only from sd_revalidate_disk()
2646 * called with buffer of length SD_BUF_SIZE
2649 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2652 struct scsi_device *sdp = sdkp->device;
2657 struct scsi_mode_data data;
2658 struct scsi_sense_hdr sshdr;
2659 int old_wce = sdkp->WCE;
2660 int old_rcd = sdkp->RCD;
2661 int old_dpofua = sdkp->DPOFUA;
2664 if (sdkp->cache_override)
2668 if (sdp->skip_ms_page_8) {
2669 if (sdp->type == TYPE_RBC)
2672 if (sdp->skip_ms_page_3f)
2675 if (sdp->use_192_bytes_for_3f)
2679 } else if (sdp->type == TYPE_RBC) {
2687 /* cautiously ask */
2688 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2691 if (!scsi_status_is_good(res))
2694 if (!data.header_length) {
2697 sd_first_printk(KERN_ERR, sdkp,
2698 "Missing header in MODE_SENSE response\n");
2701 /* that went OK, now ask for the proper length */
2705 * We're only interested in the first three bytes, actually.
2706 * But the data cache page is defined for the first 20.
2710 else if (len > SD_BUF_SIZE) {
2711 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2712 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2715 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2719 if (len > first_len)
2720 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2723 if (scsi_status_is_good(res)) {
2724 int offset = data.header_length + data.block_descriptor_length;
2726 while (offset < len) {
2727 u8 page_code = buffer[offset] & 0x3F;
2728 u8 spf = buffer[offset] & 0x40;
2730 if (page_code == 8 || page_code == 6) {
2731 /* We're interested only in the first 3 bytes.
2733 if (len - offset <= 2) {
2734 sd_first_printk(KERN_ERR, sdkp,
2735 "Incomplete mode parameter "
2739 modepage = page_code;
2743 /* Go to the next page */
2744 if (spf && len - offset > 3)
2745 offset += 4 + (buffer[offset+2] << 8) +
2747 else if (!spf && len - offset > 1)
2748 offset += 2 + buffer[offset+1];
2750 sd_first_printk(KERN_ERR, sdkp,
2752 "parameter data\n");
2758 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2762 if (modepage == 8) {
2763 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2764 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2766 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2770 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2771 if (sdp->broken_fua) {
2772 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2774 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2775 !sdkp->device->use_16_for_rw) {
2776 sd_first_printk(KERN_NOTICE, sdkp,
2777 "Uses READ/WRITE(6), disabling FUA\n");
2781 /* No cache flush allowed for write protected devices */
2782 if (sdkp->WCE && sdkp->write_prot)
2785 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2786 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2787 sd_printk(KERN_NOTICE, sdkp,
2788 "Write cache: %s, read cache: %s, %s\n",
2789 sdkp->WCE ? "enabled" : "disabled",
2790 sdkp->RCD ? "disabled" : "enabled",
2791 sdkp->DPOFUA ? "supports DPO and FUA"
2792 : "doesn't support DPO or FUA");
2798 if (scsi_sense_valid(&sshdr) &&
2799 sshdr.sense_key == ILLEGAL_REQUEST &&
2800 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2801 /* Invalid field in CDB */
2802 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2804 sd_first_printk(KERN_ERR, sdkp,
2805 "Asking for cache data failed\n");
2808 if (sdp->wce_default_on) {
2809 sd_first_printk(KERN_NOTICE, sdkp,
2810 "Assuming drive cache: write back\n");
2813 sd_first_printk(KERN_ERR, sdkp,
2814 "Assuming drive cache: write through\n");
2822 * The ATO bit indicates whether the DIF application tag is available
2823 * for use by the operating system.
2825 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2828 struct scsi_device *sdp = sdkp->device;
2829 struct scsi_mode_data data;
2830 struct scsi_sense_hdr sshdr;
2832 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2835 if (sdkp->protection_type == 0)
2838 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2839 SD_MAX_RETRIES, &data, &sshdr);
2841 if (!scsi_status_is_good(res) || !data.header_length ||
2843 sd_first_printk(KERN_WARNING, sdkp,
2844 "getting Control mode page failed, assume no ATO\n");
2846 if (scsi_sense_valid(&sshdr))
2847 sd_print_sense_hdr(sdkp, &sshdr);
2852 offset = data.header_length + data.block_descriptor_length;
2854 if ((buffer[offset] & 0x3f) != 0x0a) {
2855 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2859 if ((buffer[offset + 5] & 0x80) == 0)
2868 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2869 * @sdkp: disk to query
2871 static void sd_read_block_limits(struct scsi_disk *sdkp)
2873 unsigned int sector_sz = sdkp->device->sector_size;
2874 const int vpd_len = 64;
2875 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2878 /* Block Limits VPD */
2879 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2882 blk_queue_io_min(sdkp->disk->queue,
2883 get_unaligned_be16(&buffer[6]) * sector_sz);
2885 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2886 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2888 if (buffer[3] == 0x3c) {
2889 unsigned int lba_count, desc_count;
2891 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2896 lba_count = get_unaligned_be32(&buffer[20]);
2897 desc_count = get_unaligned_be32(&buffer[24]);
2899 if (lba_count && desc_count)
2900 sdkp->max_unmap_blocks = lba_count;
2902 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2904 if (buffer[32] & 0x80)
2905 sdkp->unmap_alignment =
2906 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2908 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2910 if (sdkp->max_unmap_blocks)
2911 sd_config_discard(sdkp, SD_LBP_UNMAP);
2913 sd_config_discard(sdkp, SD_LBP_WS16);
2915 } else { /* LBP VPD page tells us what to use */
2916 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2917 sd_config_discard(sdkp, SD_LBP_UNMAP);
2918 else if (sdkp->lbpws)
2919 sd_config_discard(sdkp, SD_LBP_WS16);
2920 else if (sdkp->lbpws10)
2921 sd_config_discard(sdkp, SD_LBP_WS10);
2923 sd_config_discard(sdkp, SD_LBP_DISABLE);
2932 * sd_read_block_characteristics - Query block dev. characteristics
2933 * @sdkp: disk to query
2935 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2937 struct request_queue *q = sdkp->disk->queue;
2938 unsigned char *buffer;
2940 const int vpd_len = 64;
2942 buffer = kmalloc(vpd_len, GFP_KERNEL);
2945 /* Block Device Characteristics VPD */
2946 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2949 rot = get_unaligned_be16(&buffer[4]);
2952 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2953 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2956 if (sdkp->device->type == TYPE_ZBC) {
2958 q->limits.zoned = BLK_ZONED_HM;
2960 sdkp->zoned = (buffer[8] >> 4) & 3;
2961 if (sdkp->zoned == 1)
2963 q->limits.zoned = BLK_ZONED_HA;
2966 * Treat drive-managed devices as
2967 * regular block devices.
2969 q->limits.zoned = BLK_ZONED_NONE;
2971 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2972 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2973 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2980 * sd_read_block_provisioning - Query provisioning VPD page
2981 * @sdkp: disk to query
2983 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2985 unsigned char *buffer;
2986 const int vpd_len = 8;
2988 if (sdkp->lbpme == 0)
2991 buffer = kmalloc(vpd_len, GFP_KERNEL);
2993 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2997 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2998 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2999 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3005 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3007 struct scsi_device *sdev = sdkp->device;
3009 if (sdev->host->no_write_same) {
3010 sdev->no_write_same = 1;
3015 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3016 /* too large values might cause issues with arcmsr */
3017 int vpd_buf_len = 64;
3019 sdev->no_report_opcodes = 1;
3021 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3022 * CODES is unsupported and the device has an ATA
3023 * Information VPD page (SAT).
3025 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3026 sdev->no_write_same = 1;
3029 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3032 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3036 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3038 struct scsi_device *sdev = sdkp->device;
3040 if (!sdev->security_supported)
3043 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3044 SECURITY_PROTOCOL_IN) == 1 &&
3045 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3046 SECURITY_PROTOCOL_OUT) == 1)
3051 * sd_revalidate_disk - called the first time a new disk is seen,
3052 * performs disk spin up, read_capacity, etc.
3053 * @disk: struct gendisk we care about
3055 static int sd_revalidate_disk(struct gendisk *disk)
3057 struct scsi_disk *sdkp = scsi_disk(disk);
3058 struct scsi_device *sdp = sdkp->device;
3059 struct request_queue *q = sdkp->disk->queue;
3060 sector_t old_capacity = sdkp->capacity;
3061 unsigned char *buffer;
3062 unsigned int dev_max, rw_max;
3064 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3065 "sd_revalidate_disk\n"));
3068 * If the device is offline, don't try and read capacity or any
3069 * of the other niceties.
3071 if (!scsi_device_online(sdp))
3074 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3076 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3077 "allocation failure.\n");
3081 sd_spinup_disk(sdkp);
3084 * Without media there is no reason to ask; moreover, some devices
3085 * react badly if we do.
3087 if (sdkp->media_present) {
3088 sd_read_capacity(sdkp, buffer);
3091 * set the default to rotational. All non-rotational devices
3092 * support the block characteristics VPD page, which will
3093 * cause this to be updated correctly and any device which
3094 * doesn't support it should be treated as rotational.
3096 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3097 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3099 if (scsi_device_supports_vpd(sdp)) {
3100 sd_read_block_provisioning(sdkp);
3101 sd_read_block_limits(sdkp);
3102 sd_read_block_characteristics(sdkp);
3103 sd_zbc_read_zones(sdkp, buffer);
3106 sd_print_capacity(sdkp, old_capacity);
3108 sd_read_write_protect_flag(sdkp, buffer);
3109 sd_read_cache_type(sdkp, buffer);
3110 sd_read_app_tag_own(sdkp, buffer);
3111 sd_read_write_same(sdkp, buffer);
3112 sd_read_security(sdkp, buffer);
3116 * We now have all cache related info, determine how we deal
3117 * with flush requests.
3119 sd_set_flush_flag(sdkp);
3121 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3122 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3124 /* Some devices report a maximum block count for READ/WRITE requests. */
3125 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3126 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3129 * Determine the device's preferred I/O size for reads and writes
3130 * unless the reported value is unreasonably small, large, or
3133 if (sdkp->opt_xfer_blocks &&
3134 sdkp->opt_xfer_blocks <= dev_max &&
3135 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3136 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3137 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3138 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3140 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3141 (sector_t)BLK_DEF_MAX_SECTORS);
3143 /* Do not exceed controller limit */
3144 rw_max = min(rw_max, queue_max_hw_sectors(q));
3147 * Only update max_sectors if previously unset or if the current value
3148 * exceeds the capabilities of the hardware.
3150 if (sdkp->first_scan ||
3151 q->limits.max_sectors > q->limits.max_dev_sectors ||
3152 q->limits.max_sectors > q->limits.max_hw_sectors)
3153 q->limits.max_sectors = rw_max;
3155 sdkp->first_scan = 0;
3157 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3158 sd_config_write_same(sdkp);
3166 * sd_unlock_native_capacity - unlock native capacity
3167 * @disk: struct gendisk to set capacity for
3169 * Block layer calls this function if it detects that partitions
3170 * on @disk reach beyond the end of the device. If the SCSI host
3171 * implements ->unlock_native_capacity() method, it's invoked to
3172 * give it a chance to adjust the device capacity.
3175 * Defined by block layer. Might sleep.
3177 static void sd_unlock_native_capacity(struct gendisk *disk)
3179 struct scsi_device *sdev = scsi_disk(disk)->device;
3181 if (sdev->host->hostt->unlock_native_capacity)
3182 sdev->host->hostt->unlock_native_capacity(sdev);
3186 * sd_format_disk_name - format disk name
3187 * @prefix: name prefix - ie. "sd" for SCSI disks
3188 * @index: index of the disk to format name for
3189 * @buf: output buffer
3190 * @buflen: length of the output buffer
3192 * SCSI disk names starts at sda. The 26th device is sdz and the
3193 * 27th is sdaa. The last one for two lettered suffix is sdzz
3194 * which is followed by sdaaa.
3196 * This is basically 26 base counting with one extra 'nil' entry
3197 * at the beginning from the second digit on and can be
3198 * determined using similar method as 26 base conversion with the
3199 * index shifted -1 after each digit is computed.
3205 * 0 on success, -errno on failure.
3207 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3209 const int base = 'z' - 'a' + 1;
3210 char *begin = buf + strlen(prefix);
3211 char *end = buf + buflen;
3221 *--p = 'a' + (index % unit);
3222 index = (index / unit) - 1;
3223 } while (index >= 0);
3225 memmove(begin, p, end - p);
3226 memcpy(buf, prefix, strlen(prefix));
3232 * The asynchronous part of sd_probe
3234 static void sd_probe_async(void *data, async_cookie_t cookie)
3236 struct scsi_disk *sdkp = data;
3237 struct scsi_device *sdp;
3244 index = sdkp->index;
3245 dev = &sdp->sdev_gendev;
3247 gd->major = sd_major((index & 0xf0) >> 4);
3248 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3250 gd->fops = &sd_fops;
3251 gd->private_data = &sdkp->driver;
3252 gd->queue = sdkp->device->request_queue;
3254 /* defaults, until the device tells us otherwise */
3255 sdp->sector_size = 512;
3257 sdkp->media_present = 1;
3258 sdkp->write_prot = 0;
3259 sdkp->cache_override = 0;
3263 sdkp->first_scan = 1;
3264 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3266 sd_revalidate_disk(gd);
3268 gd->flags = GENHD_FL_EXT_DEVT;
3269 if (sdp->removable) {
3270 gd->flags |= GENHD_FL_REMOVABLE;
3271 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3274 blk_pm_runtime_init(sdp->request_queue, dev);
3275 device_add_disk(dev, gd, NULL);
3277 sd_dif_config_host(sdkp);
3279 sd_revalidate_disk(gd);
3281 if (sdkp->security) {
3282 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3284 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3287 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3288 sdp->removable ? "removable " : "");
3289 scsi_autopm_put_device(sdp);
3290 put_device(&sdkp->dev);
3294 * sd_probe - called during driver initialization and whenever a
3295 * new scsi device is attached to the system. It is called once
3296 * for each scsi device (not just disks) present.
3297 * @dev: pointer to device object
3299 * Returns 0 if successful (or not interested in this scsi device
3300 * (e.g. scanner)); 1 when there is an error.
3302 * Note: this function is invoked from the scsi mid-level.
3303 * This function sets up the mapping between a given
3304 * <host,channel,id,lun> (found in sdp) and new device name
3305 * (e.g. /dev/sda). More precisely it is the block device major
3306 * and minor number that is chosen here.
3308 * Assume sd_probe is not re-entrant (for time being)
3309 * Also think about sd_probe() and sd_remove() running coincidentally.
3311 static int sd_probe(struct device *dev)
3313 struct scsi_device *sdp = to_scsi_device(dev);
3314 struct scsi_disk *sdkp;
3319 scsi_autopm_get_device(sdp);
3321 if (sdp->type != TYPE_DISK &&
3322 sdp->type != TYPE_ZBC &&
3323 sdp->type != TYPE_MOD &&
3324 sdp->type != TYPE_RBC)
3327 #ifndef CONFIG_BLK_DEV_ZONED
3328 if (sdp->type == TYPE_ZBC)
3331 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3335 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3339 gd = alloc_disk(SD_MINORS);
3343 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3345 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3349 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3351 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3352 goto out_free_index;
3356 sdkp->driver = &sd_template;
3358 sdkp->index = index;
3359 atomic_set(&sdkp->openers, 0);
3360 atomic_set(&sdkp->device->ioerr_cnt, 0);
3362 if (!sdp->request_queue->rq_timeout) {
3363 if (sdp->type != TYPE_MOD)
3364 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3366 blk_queue_rq_timeout(sdp->request_queue,
3370 device_initialize(&sdkp->dev);
3371 sdkp->dev.parent = dev;
3372 sdkp->dev.class = &sd_disk_class;
3373 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3375 error = device_add(&sdkp->dev);
3377 goto out_free_index;
3380 dev_set_drvdata(dev, sdkp);
3382 get_device(&sdkp->dev); /* prevent release before async_schedule */
3383 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3388 ida_free(&sd_index_ida, index);
3394 scsi_autopm_put_device(sdp);
3399 * sd_remove - called whenever a scsi disk (previously recognized by
3400 * sd_probe) is detached from the system. It is called (potentially
3401 * multiple times) during sd module unload.
3402 * @dev: pointer to device object
3404 * Note: this function is invoked from the scsi mid-level.
3405 * This function potentially frees up a device name (e.g. /dev/sdc)
3406 * that could be re-used by a subsequent sd_probe().
3407 * This function is not called when the built-in sd driver is "exit-ed".
3409 static int sd_remove(struct device *dev)
3411 struct scsi_disk *sdkp;
3414 sdkp = dev_get_drvdata(dev);
3415 devt = disk_devt(sdkp->disk);
3416 scsi_autopm_get_device(sdkp->device);
3418 async_synchronize_full_domain(&scsi_sd_pm_domain);
3419 async_synchronize_full_domain(&scsi_sd_probe_domain);
3420 device_del(&sdkp->dev);
3421 del_gendisk(sdkp->disk);
3424 free_opal_dev(sdkp->opal_dev);
3426 blk_register_region(devt, SD_MINORS, NULL,
3427 sd_default_probe, NULL, NULL);
3429 mutex_lock(&sd_ref_mutex);
3430 dev_set_drvdata(dev, NULL);
3431 put_device(&sdkp->dev);
3432 mutex_unlock(&sd_ref_mutex);
3438 * scsi_disk_release - Called to free the scsi_disk structure
3439 * @dev: pointer to embedded class device
3441 * sd_ref_mutex must be held entering this routine. Because it is
3442 * called on last put, you should always use the scsi_disk_get()
3443 * scsi_disk_put() helpers which manipulate the semaphore directly
3444 * and never do a direct put_device.
3446 static void scsi_disk_release(struct device *dev)
3448 struct scsi_disk *sdkp = to_scsi_disk(dev);
3449 struct gendisk *disk = sdkp->disk;
3451 ida_free(&sd_index_ida, sdkp->index);
3453 disk->private_data = NULL;
3455 put_device(&sdkp->device->sdev_gendev);
3460 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3462 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3463 struct scsi_sense_hdr sshdr;
3464 struct scsi_device *sdp = sdkp->device;
3468 cmd[4] |= 1; /* START */
3470 if (sdp->start_stop_pwr_cond)
3471 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3473 if (!scsi_device_online(sdp))
3476 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3477 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3479 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3480 if (driver_byte(res) == DRIVER_SENSE)
3481 sd_print_sense_hdr(sdkp, &sshdr);
3482 if (scsi_sense_valid(&sshdr) &&
3483 /* 0x3a is medium not present */
3488 /* SCSI error codes must not go to the generic layer */
3496 * Send a SYNCHRONIZE CACHE instruction down to the device through
3497 * the normal SCSI command structure. Wait for the command to
3500 static void sd_shutdown(struct device *dev)
3502 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3505 return; /* this can happen */
3507 if (pm_runtime_suspended(dev))
3510 if (sdkp->WCE && sdkp->media_present) {
3511 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3512 sd_sync_cache(sdkp, NULL);
3515 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3516 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3517 sd_start_stop_device(sdkp, 0);
3521 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3523 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3524 struct scsi_sense_hdr sshdr;
3527 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3530 if (sdkp->WCE && sdkp->media_present) {
3531 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3532 ret = sd_sync_cache(sdkp, &sshdr);
3535 /* ignore OFFLINE device */
3539 if (!scsi_sense_valid(&sshdr) ||
3540 sshdr.sense_key != ILLEGAL_REQUEST)
3544 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3545 * doesn't support sync. There's not much to do and
3546 * suspend shouldn't fail.
3552 if (sdkp->device->manage_start_stop) {
3553 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3554 /* an error is not worth aborting a system sleep */
3555 ret = sd_start_stop_device(sdkp, 0);
3556 if (ignore_stop_errors)
3563 static int sd_suspend_system(struct device *dev)
3565 return sd_suspend_common(dev, true);
3568 static int sd_suspend_runtime(struct device *dev)
3570 return sd_suspend_common(dev, false);
3573 static int sd_resume(struct device *dev)
3575 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3578 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3581 if (!sdkp->device->manage_start_stop)
3584 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3585 ret = sd_start_stop_device(sdkp, 1);
3587 opal_unlock_from_suspend(sdkp->opal_dev);
3592 * init_sd - entry point for this driver (both when built in or when
3595 * Note: this function registers this driver with the scsi mid-level.
3597 static int __init init_sd(void)
3599 int majors = 0, i, err;
3601 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3603 for (i = 0; i < SD_MAJORS; i++) {
3604 if (register_blkdev(sd_major(i), "sd") != 0)
3607 blk_register_region(sd_major(i), SD_MINORS, NULL,
3608 sd_default_probe, NULL, NULL);
3614 err = class_register(&sd_disk_class);
3618 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3620 if (!sd_cdb_cache) {
3621 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3626 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3628 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3633 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3634 if (!sd_page_pool) {
3635 printk(KERN_ERR "sd: can't init discard page pool\n");
3640 err = scsi_register_driver(&sd_template.gendrv);
3642 goto err_out_driver;
3647 mempool_destroy(sd_page_pool);
3650 mempool_destroy(sd_cdb_pool);
3653 kmem_cache_destroy(sd_cdb_cache);
3656 class_unregister(&sd_disk_class);
3658 for (i = 0; i < SD_MAJORS; i++)
3659 unregister_blkdev(sd_major(i), "sd");
3664 * exit_sd - exit point for this driver (when it is a module).
3666 * Note: this function unregisters this driver from the scsi mid-level.
3668 static void __exit exit_sd(void)
3672 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3674 scsi_unregister_driver(&sd_template.gendrv);
3675 mempool_destroy(sd_cdb_pool);
3676 mempool_destroy(sd_page_pool);
3677 kmem_cache_destroy(sd_cdb_cache);
3679 class_unregister(&sd_disk_class);
3681 for (i = 0; i < SD_MAJORS; i++) {
3682 blk_unregister_region(sd_major(i), SD_MINORS);
3683 unregister_blkdev(sd_major(i), "sd");
3687 module_init(init_sd);
3688 module_exit(exit_sd);
3690 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3691 struct scsi_sense_hdr *sshdr)
3693 scsi_print_sense_hdr(sdkp->device,
3694 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3697 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3700 const char *hb_string = scsi_hostbyte_string(result);
3701 const char *db_string = scsi_driverbyte_string(result);
3703 if (hb_string || db_string)
3704 sd_printk(KERN_INFO, sdkp,
3705 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3706 hb_string ? hb_string : "invalid",
3707 db_string ? db_string : "invalid");
3709 sd_printk(KERN_INFO, sdkp,
3710 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3711 msg, host_byte(result), driver_byte(result));