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/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/sed-opal.h>
54 #include <linux/pm_runtime.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int sd_probe(struct device *);
110 static int sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
126 static DEFINE_SPINLOCK(sd_index_lock);
127 static DEFINE_IDA(sd_index_ida);
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
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 queue_flag_clear_unlocked(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 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
760 static int 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 BLKPREP_DEFER;
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 int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
797 struct scsi_device *sdp = cmd->device;
798 struct request *rq = cmd->request;
799 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
800 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
801 u32 data_len = sdp->sector_size;
803 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
804 if (!rq->special_vec.bv_page)
805 return BLKPREP_DEFER;
806 clear_highpage(rq->special_vec.bv_page);
807 rq->special_vec.bv_offset = 0;
808 rq->special_vec.bv_len = data_len;
809 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
812 cmd->cmnd[0] = WRITE_SAME_16;
814 cmd->cmnd[1] = 0x8; /* UNMAP */
815 put_unaligned_be64(sector, &cmd->cmnd[2]);
816 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
818 cmd->allowed = SD_MAX_RETRIES;
819 cmd->transfersize = data_len;
820 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
821 scsi_req(rq)->resid_len = data_len;
823 return scsi_init_io(cmd);
826 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
828 struct scsi_device *sdp = cmd->device;
829 struct request *rq = cmd->request;
830 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
831 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
832 u32 data_len = sdp->sector_size;
834 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
835 if (!rq->special_vec.bv_page)
836 return BLKPREP_DEFER;
837 clear_highpage(rq->special_vec.bv_page);
838 rq->special_vec.bv_offset = 0;
839 rq->special_vec.bv_len = data_len;
840 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
843 cmd->cmnd[0] = WRITE_SAME;
845 cmd->cmnd[1] = 0x8; /* UNMAP */
846 put_unaligned_be32(sector, &cmd->cmnd[2]);
847 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
849 cmd->allowed = SD_MAX_RETRIES;
850 cmd->transfersize = data_len;
851 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
852 scsi_req(rq)->resid_len = data_len;
854 return scsi_init_io(cmd);
857 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
859 struct request *rq = cmd->request;
860 struct scsi_device *sdp = cmd->device;
861 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
862 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
863 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
866 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
867 switch (sdkp->zeroing_mode) {
868 case SD_ZERO_WS16_UNMAP:
869 ret = sd_setup_write_same16_cmnd(cmd, true);
871 case SD_ZERO_WS10_UNMAP:
872 ret = sd_setup_write_same10_cmnd(cmd, true);
877 if (sdp->no_write_same)
878 return BLKPREP_INVALID;
880 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
881 ret = sd_setup_write_same16_cmnd(cmd, false);
883 ret = sd_setup_write_same10_cmnd(cmd, false);
886 if (sd_is_zoned(sdkp) && ret == BLKPREP_OK)
887 return sd_zbc_write_lock_zone(cmd);
892 static void sd_config_write_same(struct scsi_disk *sdkp)
894 struct request_queue *q = sdkp->disk->queue;
895 unsigned int logical_block_size = sdkp->device->sector_size;
897 if (sdkp->device->no_write_same) {
898 sdkp->max_ws_blocks = 0;
902 /* Some devices can not handle block counts above 0xffff despite
903 * supporting WRITE SAME(16). Consequently we default to 64k
904 * blocks per I/O unless the device explicitly advertises a
907 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
908 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
909 (u32)SD_MAX_WS16_BLOCKS);
910 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
911 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
912 (u32)SD_MAX_WS10_BLOCKS);
914 sdkp->device->no_write_same = 1;
915 sdkp->max_ws_blocks = 0;
918 if (sdkp->lbprz && sdkp->lbpws)
919 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
920 else if (sdkp->lbprz && sdkp->lbpws10)
921 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
922 else if (sdkp->max_ws_blocks)
923 sdkp->zeroing_mode = SD_ZERO_WS;
925 sdkp->zeroing_mode = SD_ZERO_WRITE;
928 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
929 (logical_block_size >> 9));
930 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
931 (logical_block_size >> 9));
935 * sd_setup_write_same_cmnd - write the same data to multiple blocks
936 * @cmd: command to prepare
938 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
939 * the preference indicated by the target device.
941 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
943 struct request *rq = cmd->request;
944 struct scsi_device *sdp = cmd->device;
945 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
946 struct bio *bio = rq->bio;
947 sector_t sector = blk_rq_pos(rq);
948 unsigned int nr_sectors = blk_rq_sectors(rq);
949 unsigned int nr_bytes = blk_rq_bytes(rq);
952 if (sdkp->device->no_write_same)
953 return BLKPREP_INVALID;
955 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
957 if (sd_is_zoned(sdkp)) {
958 ret = sd_zbc_write_lock_zone(cmd);
959 if (ret != BLKPREP_OK)
963 sector >>= ilog2(sdp->sector_size) - 9;
964 nr_sectors >>= ilog2(sdp->sector_size) - 9;
966 rq->timeout = SD_WRITE_SAME_TIMEOUT;
968 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
970 cmd->cmnd[0] = WRITE_SAME_16;
971 put_unaligned_be64(sector, &cmd->cmnd[2]);
972 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
975 cmd->cmnd[0] = WRITE_SAME;
976 put_unaligned_be32(sector, &cmd->cmnd[2]);
977 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
980 cmd->transfersize = sdp->sector_size;
981 cmd->allowed = SD_MAX_RETRIES;
984 * For WRITE SAME the data transferred via the DATA OUT buffer is
985 * different from the amount of data actually written to the target.
987 * We set up __data_len to the amount of data transferred via the
988 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
989 * to transfer a single sector of data first, but then reset it to
990 * the amount of data to be written right after so that the I/O path
991 * knows how much to actually write.
993 rq->__data_len = sdp->sector_size;
994 ret = scsi_init_io(cmd);
995 rq->__data_len = nr_bytes;
997 if (sd_is_zoned(sdkp) && ret != BLKPREP_OK)
998 sd_zbc_write_unlock_zone(cmd);
1003 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1005 struct request *rq = cmd->request;
1007 /* flush requests don't perform I/O, zero the S/G table */
1008 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1010 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1012 cmd->transfersize = 0;
1013 cmd->allowed = SD_MAX_RETRIES;
1015 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1019 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1021 struct request *rq = SCpnt->request;
1022 struct scsi_device *sdp = SCpnt->device;
1023 struct gendisk *disk = rq->rq_disk;
1024 struct scsi_disk *sdkp = scsi_disk(disk);
1025 sector_t block = blk_rq_pos(rq);
1027 unsigned int this_count = blk_rq_sectors(rq);
1028 unsigned int dif, dix;
1029 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
1031 unsigned char protect;
1034 ret = sd_zbc_write_lock_zone(SCpnt);
1035 if (ret != BLKPREP_OK)
1039 ret = scsi_init_io(SCpnt);
1040 if (ret != BLKPREP_OK)
1042 WARN_ON_ONCE(SCpnt != rq->special);
1044 /* from here on until we're complete, any goto out
1045 * is used for a killable error condition */
1049 scmd_printk(KERN_INFO, SCpnt,
1050 "%s: block=%llu, count=%d\n",
1051 __func__, (unsigned long long)block, this_count));
1053 if (!sdp || !scsi_device_online(sdp) ||
1054 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1055 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1056 "Finishing %u sectors\n",
1057 blk_rq_sectors(rq)));
1058 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1059 "Retry with 0x%p\n", SCpnt));
1065 * quietly refuse to do anything to a changed disc until
1066 * the changed bit has been reset
1068 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1073 * Some SD card readers can't handle multi-sector accesses which touch
1074 * the last one or two hardware sectors. Split accesses as needed.
1076 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1077 (sdp->sector_size / 512);
1079 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1080 if (block < threshold) {
1081 /* Access up to the threshold but not beyond */
1082 this_count = threshold - block;
1084 /* Access only a single hardware sector */
1085 this_count = sdp->sector_size / 512;
1089 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1090 (unsigned long long)block));
1093 * If we have a 1K hardware sectorsize, prevent access to single
1094 * 512 byte sectors. In theory we could handle this - in fact
1095 * the scsi cdrom driver must be able to handle this because
1096 * we typically use 1K blocksizes, and cdroms typically have
1097 * 2K hardware sectorsizes. Of course, things are simpler
1098 * with the cdrom, since it is read-only. For performance
1099 * reasons, the filesystems should be able to handle this
1100 * and not force the scsi disk driver to use bounce buffers
1103 if (sdp->sector_size == 1024) {
1104 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1105 scmd_printk(KERN_ERR, SCpnt,
1106 "Bad block number requested\n");
1110 this_count = this_count >> 1;
1113 if (sdp->sector_size == 2048) {
1114 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1115 scmd_printk(KERN_ERR, SCpnt,
1116 "Bad block number requested\n");
1120 this_count = this_count >> 2;
1123 if (sdp->sector_size == 4096) {
1124 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1125 scmd_printk(KERN_ERR, SCpnt,
1126 "Bad block number requested\n");
1130 this_count = this_count >> 3;
1133 if (rq_data_dir(rq) == WRITE) {
1134 SCpnt->cmnd[0] = WRITE_6;
1136 if (blk_integrity_rq(rq))
1137 sd_dif_prepare(SCpnt);
1139 } else if (rq_data_dir(rq) == READ) {
1140 SCpnt->cmnd[0] = READ_6;
1142 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1146 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1147 "%s %d/%u 512 byte blocks.\n",
1148 (rq_data_dir(rq) == WRITE) ?
1149 "writing" : "reading", this_count,
1150 blk_rq_sectors(rq)));
1152 dix = scsi_prot_sg_count(SCpnt);
1153 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1156 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1160 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1161 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1163 if (unlikely(SCpnt->cmnd == NULL)) {
1164 ret = BLKPREP_DEFER;
1168 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1169 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1170 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1171 SCpnt->cmnd[7] = 0x18;
1172 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1173 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1176 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1177 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1178 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1179 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1180 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1181 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1182 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1183 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1185 /* Expected Indirect LBA */
1186 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1187 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1188 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1189 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1191 /* Transfer length */
1192 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1193 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1194 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1195 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1196 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1197 SCpnt->cmnd[0] += READ_16 - READ_6;
1198 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1199 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1200 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1201 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1202 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1203 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1204 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1205 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1206 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1207 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1208 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1209 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1210 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1211 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1212 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1213 scsi_device_protection(SCpnt->device) ||
1214 SCpnt->device->use_10_for_rw) {
1215 SCpnt->cmnd[0] += READ_10 - READ_6;
1216 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1217 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1218 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1219 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1220 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1221 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1222 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1223 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1225 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1227 * This happens only if this drive failed
1228 * 10byte rw command with ILLEGAL_REQUEST
1229 * during operation and thus turned off
1232 scmd_printk(KERN_ERR, SCpnt,
1233 "FUA write on READ/WRITE(6) drive\n");
1237 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1238 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1239 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1240 SCpnt->cmnd[4] = (unsigned char) this_count;
1243 SCpnt->sdb.length = this_count * sdp->sector_size;
1246 * We shouldn't disconnect in the middle of a sector, so with a dumb
1247 * host adapter, it's safe to assume that we can at least transfer
1248 * this many bytes between each connect / disconnect.
1250 SCpnt->transfersize = sdp->sector_size;
1251 SCpnt->underflow = this_count << 9;
1252 SCpnt->allowed = SD_MAX_RETRIES;
1255 * This indicates that the command is ready from our end to be
1260 if (zoned_write && ret != BLKPREP_OK)
1261 sd_zbc_write_unlock_zone(SCpnt);
1266 static int sd_init_command(struct scsi_cmnd *cmd)
1268 struct request *rq = cmd->request;
1270 switch (req_op(rq)) {
1271 case REQ_OP_DISCARD:
1272 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1274 return sd_setup_unmap_cmnd(cmd);
1276 return sd_setup_write_same16_cmnd(cmd, true);
1278 return sd_setup_write_same10_cmnd(cmd, true);
1280 return sd_setup_write_same10_cmnd(cmd, false);
1282 return BLKPREP_INVALID;
1284 case REQ_OP_WRITE_ZEROES:
1285 return sd_setup_write_zeroes_cmnd(cmd);
1286 case REQ_OP_WRITE_SAME:
1287 return sd_setup_write_same_cmnd(cmd);
1289 return sd_setup_flush_cmnd(cmd);
1292 return sd_setup_read_write_cmnd(cmd);
1293 case REQ_OP_ZONE_REPORT:
1294 return sd_zbc_setup_report_cmnd(cmd);
1295 case REQ_OP_ZONE_RESET:
1296 return sd_zbc_setup_reset_cmnd(cmd);
1299 return BLKPREP_KILL;
1303 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1305 struct request *rq = SCpnt->request;
1308 if (SCpnt->flags & SCMD_ZONE_WRITE_LOCK)
1309 sd_zbc_write_unlock_zone(SCpnt);
1311 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1312 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1314 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1318 mempool_free(cmnd, sd_cdb_pool);
1323 * sd_open - open a scsi disk device
1324 * @bdev: Block device of the scsi disk to open
1325 * @mode: FMODE_* mask
1327 * Returns 0 if successful. Returns a negated errno value in case
1330 * Note: This can be called from a user context (e.g. fsck(1) )
1331 * or from within the kernel (e.g. as a result of a mount(1) ).
1332 * In the latter case @inode and @filp carry an abridged amount
1333 * of information as noted above.
1335 * Locking: called with bdev->bd_mutex held.
1337 static int sd_open(struct block_device *bdev, fmode_t mode)
1339 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1340 struct scsi_device *sdev;
1346 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1348 sdev = sdkp->device;
1351 * If the device is in error recovery, wait until it is done.
1352 * If the device is offline, then disallow any access to it.
1355 if (!scsi_block_when_processing_errors(sdev))
1358 if (sdev->removable || sdkp->write_prot)
1359 check_disk_change(bdev);
1362 * If the drive is empty, just let the open fail.
1364 retval = -ENOMEDIUM;
1365 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1369 * If the device has the write protect tab set, have the open fail
1370 * if the user expects to be able to write to the thing.
1373 if (sdkp->write_prot && (mode & FMODE_WRITE))
1377 * It is possible that the disk changing stuff resulted in
1378 * the device being taken offline. If this is the case,
1379 * report this to the user, and don't pretend that the
1380 * open actually succeeded.
1383 if (!scsi_device_online(sdev))
1386 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1387 if (scsi_block_when_processing_errors(sdev))
1388 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1394 scsi_disk_put(sdkp);
1399 * sd_release - invoked when the (last) close(2) is called on this
1401 * @disk: disk to release
1402 * @mode: FMODE_* mask
1406 * Note: may block (uninterruptible) if error recovery is underway
1409 * Locking: called with bdev->bd_mutex held.
1411 static void sd_release(struct gendisk *disk, fmode_t mode)
1413 struct scsi_disk *sdkp = scsi_disk(disk);
1414 struct scsi_device *sdev = sdkp->device;
1416 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1418 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1419 if (scsi_block_when_processing_errors(sdev))
1420 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1423 scsi_disk_put(sdkp);
1426 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1428 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1429 struct scsi_device *sdp = sdkp->device;
1430 struct Scsi_Host *host = sdp->host;
1431 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1434 /* default to most commonly used values */
1435 diskinfo[0] = 0x40; /* 1 << 6 */
1436 diskinfo[1] = 0x20; /* 1 << 5 */
1437 diskinfo[2] = capacity >> 11;
1439 /* override with calculated, extended default, or driver values */
1440 if (host->hostt->bios_param)
1441 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1443 scsicam_bios_param(bdev, capacity, diskinfo);
1445 geo->heads = diskinfo[0];
1446 geo->sectors = diskinfo[1];
1447 geo->cylinders = diskinfo[2];
1452 * sd_ioctl - process an ioctl
1453 * @bdev: target block device
1454 * @mode: FMODE_* mask
1455 * @cmd: ioctl command number
1456 * @arg: this is third argument given to ioctl(2) system call.
1457 * Often contains a pointer.
1459 * Returns 0 if successful (some ioctls return positive numbers on
1460 * success as well). Returns a negated errno value in case of error.
1462 * Note: most ioctls are forward onto the block subsystem or further
1463 * down in the scsi subsystem.
1465 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1466 unsigned int cmd, unsigned long arg)
1468 struct gendisk *disk = bdev->bd_disk;
1469 struct scsi_disk *sdkp = scsi_disk(disk);
1470 struct scsi_device *sdp = sdkp->device;
1471 void __user *p = (void __user *)arg;
1474 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1475 "cmd=0x%x\n", disk->disk_name, cmd));
1477 error = scsi_verify_blk_ioctl(bdev, cmd);
1482 * If we are in the middle of error recovery, don't let anyone
1483 * else try and use this device. Also, if error recovery fails, it
1484 * may try and take the device offline, in which case all further
1485 * access to the device is prohibited.
1487 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1488 (mode & FMODE_NDELAY) != 0);
1492 if (is_sed_ioctl(cmd))
1493 return sed_ioctl(sdkp->opal_dev, cmd, p);
1496 * Send SCSI addressing ioctls directly to mid level, send other
1497 * ioctls to block level and then onto mid level if they can't be
1501 case SCSI_IOCTL_GET_IDLUN:
1502 case SCSI_IOCTL_GET_BUS_NUMBER:
1503 error = scsi_ioctl(sdp, cmd, p);
1506 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1507 if (error != -ENOTTY)
1509 error = scsi_ioctl(sdp, cmd, p);
1516 static void set_media_not_present(struct scsi_disk *sdkp)
1518 if (sdkp->media_present)
1519 sdkp->device->changed = 1;
1521 if (sdkp->device->removable) {
1522 sdkp->media_present = 0;
1527 static int media_not_present(struct scsi_disk *sdkp,
1528 struct scsi_sense_hdr *sshdr)
1530 if (!scsi_sense_valid(sshdr))
1533 /* not invoked for commands that could return deferred errors */
1534 switch (sshdr->sense_key) {
1535 case UNIT_ATTENTION:
1537 /* medium not present */
1538 if (sshdr->asc == 0x3A) {
1539 set_media_not_present(sdkp);
1547 * sd_check_events - check media events
1548 * @disk: kernel device descriptor
1549 * @clearing: disk events currently being cleared
1551 * Returns mask of DISK_EVENT_*.
1553 * Note: this function is invoked from the block subsystem.
1555 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1557 struct scsi_disk *sdkp = scsi_disk_get(disk);
1558 struct scsi_device *sdp;
1565 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1568 * If the device is offline, don't send any commands - just pretend as
1569 * if the command failed. If the device ever comes back online, we
1570 * can deal with it then. It is only because of unrecoverable errors
1571 * that we would ever take a device offline in the first place.
1573 if (!scsi_device_online(sdp)) {
1574 set_media_not_present(sdkp);
1579 * Using TEST_UNIT_READY enables differentiation between drive with
1580 * no cartridge loaded - NOT READY, drive with changed cartridge -
1581 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1583 * Drives that auto spin down. eg iomega jaz 1G, will be started
1584 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1585 * sd_revalidate() is called.
1587 if (scsi_block_when_processing_errors(sdp)) {
1588 struct scsi_sense_hdr sshdr = { 0, };
1590 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1593 /* failed to execute TUR, assume media not present */
1594 if (host_byte(retval)) {
1595 set_media_not_present(sdkp);
1599 if (media_not_present(sdkp, &sshdr))
1604 * For removable scsi disk we have to recognise the presence
1605 * of a disk in the drive.
1607 if (!sdkp->media_present)
1609 sdkp->media_present = 1;
1612 * sdp->changed is set under the following conditions:
1614 * Medium present state has changed in either direction.
1615 * Device has indicated UNIT_ATTENTION.
1617 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1619 scsi_disk_put(sdkp);
1623 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1626 struct scsi_device *sdp = sdkp->device;
1627 const int timeout = sdp->request_queue->rq_timeout
1628 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1629 struct scsi_sense_hdr my_sshdr;
1631 if (!scsi_device_online(sdp))
1634 /* caller might not be interested in sense, but we need it */
1638 for (retries = 3; retries > 0; --retries) {
1639 unsigned char cmd[10] = { 0 };
1641 cmd[0] = SYNCHRONIZE_CACHE;
1643 * Leave the rest of the command zero to indicate
1646 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1647 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1653 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1655 if (driver_byte(res) & DRIVER_SENSE)
1656 sd_print_sense_hdr(sdkp, sshdr);
1658 /* we need to evaluate the error return */
1659 if (scsi_sense_valid(sshdr) &&
1660 (sshdr->asc == 0x3a || /* medium not present */
1661 sshdr->asc == 0x20 || /* invalid command */
1662 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1663 /* this is no error here */
1666 switch (host_byte(res)) {
1667 /* ignore errors due to racing a disconnection */
1668 case DID_BAD_TARGET:
1669 case DID_NO_CONNECT:
1671 /* signal the upper layer it might try again */
1675 case DID_SOFT_ERROR:
1684 static void sd_rescan(struct device *dev)
1686 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1688 revalidate_disk(sdkp->disk);
1692 #ifdef CONFIG_COMPAT
1694 * This gets directly called from VFS. When the ioctl
1695 * is not recognized we go back to the other translation paths.
1697 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1698 unsigned int cmd, unsigned long arg)
1700 struct gendisk *disk = bdev->bd_disk;
1701 struct scsi_disk *sdkp = scsi_disk(disk);
1702 struct scsi_device *sdev = sdkp->device;
1703 void __user *p = compat_ptr(arg);
1706 error = scsi_verify_blk_ioctl(bdev, cmd);
1710 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1711 (mode & FMODE_NDELAY) != 0);
1715 if (is_sed_ioctl(cmd))
1716 return sed_ioctl(sdkp->opal_dev, cmd, p);
1719 * Let the static ioctl translation table take care of it.
1721 if (!sdev->host->hostt->compat_ioctl)
1722 return -ENOIOCTLCMD;
1723 return sdev->host->hostt->compat_ioctl(sdev, cmd, p);
1727 static char sd_pr_type(enum pr_type type)
1730 case PR_WRITE_EXCLUSIVE:
1732 case PR_EXCLUSIVE_ACCESS:
1734 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1736 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1738 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1740 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1747 static int sd_pr_command(struct block_device *bdev, u8 sa,
1748 u64 key, u64 sa_key, u8 type, u8 flags)
1750 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1751 struct scsi_sense_hdr sshdr;
1753 u8 cmd[16] = { 0, };
1754 u8 data[24] = { 0, };
1756 cmd[0] = PERSISTENT_RESERVE_OUT;
1759 put_unaligned_be32(sizeof(data), &cmd[5]);
1761 put_unaligned_be64(key, &data[0]);
1762 put_unaligned_be64(sa_key, &data[8]);
1765 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1766 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1768 if ((driver_byte(result) & DRIVER_SENSE) &&
1769 (scsi_sense_valid(&sshdr))) {
1770 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1771 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1777 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1780 if (flags & ~PR_FL_IGNORE_KEY)
1782 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1783 old_key, new_key, 0,
1784 (1 << 0) /* APTPL */);
1787 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1792 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1795 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1797 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1800 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1801 enum pr_type type, bool abort)
1803 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1804 sd_pr_type(type), 0);
1807 static int sd_pr_clear(struct block_device *bdev, u64 key)
1809 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1812 static const struct pr_ops sd_pr_ops = {
1813 .pr_register = sd_pr_register,
1814 .pr_reserve = sd_pr_reserve,
1815 .pr_release = sd_pr_release,
1816 .pr_preempt = sd_pr_preempt,
1817 .pr_clear = sd_pr_clear,
1820 static const struct block_device_operations sd_fops = {
1821 .owner = THIS_MODULE,
1823 .release = sd_release,
1825 .getgeo = sd_getgeo,
1826 #ifdef CONFIG_COMPAT
1827 .compat_ioctl = sd_compat_ioctl,
1829 .check_events = sd_check_events,
1830 .revalidate_disk = sd_revalidate_disk,
1831 .unlock_native_capacity = sd_unlock_native_capacity,
1832 .pr_ops = &sd_pr_ops,
1836 * sd_eh_reset - reset error handling callback
1837 * @scmd: sd-issued command that has failed
1839 * This function is called by the SCSI midlayer before starting
1840 * SCSI EH. When counting medium access failures we have to be
1841 * careful to register it only only once per device and SCSI EH run;
1842 * there might be several timed out commands which will cause the
1843 * 'max_medium_access_timeouts' counter to trigger after the first
1844 * SCSI EH run already and set the device to offline.
1845 * So this function resets the internal counter before starting SCSI EH.
1847 static void sd_eh_reset(struct scsi_cmnd *scmd)
1849 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1851 /* New SCSI EH run, reset gate variable */
1852 sdkp->ignore_medium_access_errors = false;
1856 * sd_eh_action - error handling callback
1857 * @scmd: sd-issued command that has failed
1858 * @eh_disp: The recovery disposition suggested by the midlayer
1860 * This function is called by the SCSI midlayer upon completion of an
1861 * error test command (currently TEST UNIT READY). The result of sending
1862 * the eh command is passed in eh_disp. We're looking for devices that
1863 * fail medium access commands but are OK with non access commands like
1864 * test unit ready (so wrongly see the device as having a successful
1867 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1869 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1870 struct scsi_device *sdev = scmd->device;
1872 if (!scsi_device_online(sdev) ||
1873 !scsi_medium_access_command(scmd) ||
1874 host_byte(scmd->result) != DID_TIME_OUT ||
1879 * The device has timed out executing a medium access command.
1880 * However, the TEST UNIT READY command sent during error
1881 * handling completed successfully. Either the device is in the
1882 * process of recovering or has it suffered an internal failure
1883 * that prevents access to the storage medium.
1885 if (!sdkp->ignore_medium_access_errors) {
1886 sdkp->medium_access_timed_out++;
1887 sdkp->ignore_medium_access_errors = true;
1891 * If the device keeps failing read/write commands but TEST UNIT
1892 * READY always completes successfully we assume that medium
1893 * access is no longer possible and take the device offline.
1895 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1896 scmd_printk(KERN_ERR, scmd,
1897 "Medium access timeout failure. Offlining disk!\n");
1898 mutex_lock(&sdev->state_mutex);
1899 scsi_device_set_state(sdev, SDEV_OFFLINE);
1900 mutex_unlock(&sdev->state_mutex);
1908 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1910 struct request *req = scmd->request;
1911 struct scsi_device *sdev = scmd->device;
1912 unsigned int transferred, good_bytes;
1913 u64 start_lba, end_lba, bad_lba;
1916 * Some commands have a payload smaller than the device logical
1917 * block size (e.g. INQUIRY on a 4K disk).
1919 if (scsi_bufflen(scmd) <= sdev->sector_size)
1922 /* Check if we have a 'bad_lba' information */
1923 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1924 SCSI_SENSE_BUFFERSIZE,
1929 * If the bad lba was reported incorrectly, we have no idea where
1932 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1933 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1934 if (bad_lba < start_lba || bad_lba >= end_lba)
1938 * resid is optional but mostly filled in. When it's unused,
1939 * its value is zero, so we assume the whole buffer transferred
1941 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1943 /* This computation should always be done in terms of the
1944 * resolution of the device's medium.
1946 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1948 return min(good_bytes, transferred);
1952 * sd_done - bottom half handler: called when the lower level
1953 * driver has completed (successfully or otherwise) a scsi command.
1954 * @SCpnt: mid-level's per command structure.
1956 * Note: potentially run from within an ISR. Must not block.
1958 static int sd_done(struct scsi_cmnd *SCpnt)
1960 int result = SCpnt->result;
1961 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1962 unsigned int sector_size = SCpnt->device->sector_size;
1964 struct scsi_sense_hdr sshdr;
1965 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1966 struct request *req = SCpnt->request;
1967 int sense_valid = 0;
1968 int sense_deferred = 0;
1970 switch (req_op(req)) {
1971 case REQ_OP_DISCARD:
1972 case REQ_OP_WRITE_ZEROES:
1973 case REQ_OP_WRITE_SAME:
1974 case REQ_OP_ZONE_RESET:
1976 good_bytes = blk_rq_bytes(req);
1977 scsi_set_resid(SCpnt, 0);
1980 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1983 case REQ_OP_ZONE_REPORT:
1984 /* To avoid that the block layer performs an incorrect
1985 * bio_advance() call and restart of the remainder of
1986 * incomplete report zone BIOs, always indicate a full
1987 * completion of REQ_OP_ZONE_REPORT.
1990 good_bytes = scsi_bufflen(SCpnt);
1991 scsi_set_resid(SCpnt, 0);
1994 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1999 * In case of bogus fw or device, we could end up having
2000 * an unaligned partial completion. Check this here and force
2003 resid = scsi_get_resid(SCpnt);
2004 if (resid & (sector_size - 1)) {
2005 sd_printk(KERN_INFO, sdkp,
2006 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2007 resid, sector_size);
2008 resid = min(scsi_bufflen(SCpnt),
2009 round_up(resid, sector_size));
2010 scsi_set_resid(SCpnt, resid);
2015 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2017 sense_deferred = scsi_sense_is_deferred(&sshdr);
2019 sdkp->medium_access_timed_out = 0;
2021 if (driver_byte(result) != DRIVER_SENSE &&
2022 (!sense_valid || sense_deferred))
2025 switch (sshdr.sense_key) {
2026 case HARDWARE_ERROR:
2028 good_bytes = sd_completed_bytes(SCpnt);
2030 case RECOVERED_ERROR:
2031 good_bytes = scsi_bufflen(SCpnt);
2034 /* This indicates a false check condition, so ignore it. An
2035 * unknown amount of data was transferred so treat it as an
2039 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2041 case ABORTED_COMMAND:
2042 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2043 good_bytes = sd_completed_bytes(SCpnt);
2045 case ILLEGAL_REQUEST:
2046 switch (sshdr.asc) {
2047 case 0x10: /* DIX: Host detected corruption */
2048 good_bytes = sd_completed_bytes(SCpnt);
2050 case 0x20: /* INVALID COMMAND OPCODE */
2051 case 0x24: /* INVALID FIELD IN CDB */
2052 switch (SCpnt->cmnd[0]) {
2054 sd_config_discard(sdkp, SD_LBP_DISABLE);
2058 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2059 sd_config_discard(sdkp, SD_LBP_DISABLE);
2061 sdkp->device->no_write_same = 1;
2062 sd_config_write_same(sdkp);
2063 req->__data_len = blk_rq_bytes(req);
2064 req->rq_flags |= RQF_QUIET;
2075 if (sd_is_zoned(sdkp))
2076 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2078 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2079 "sd_done: completed %d of %d bytes\n",
2080 good_bytes, scsi_bufflen(SCpnt)));
2082 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2083 sd_dif_complete(SCpnt, good_bytes);
2089 * spinup disk - called only in sd_revalidate_disk()
2092 sd_spinup_disk(struct scsi_disk *sdkp)
2094 unsigned char cmd[10];
2095 unsigned long spintime_expire = 0;
2096 int retries, spintime;
2097 unsigned int the_result;
2098 struct scsi_sense_hdr sshdr;
2099 int sense_valid = 0;
2103 /* Spin up drives, as required. Only do this at boot time */
2104 /* Spinup needs to be done for module loads too. */
2109 cmd[0] = TEST_UNIT_READY;
2110 memset((void *) &cmd[1], 0, 9);
2112 the_result = scsi_execute_req(sdkp->device, cmd,
2115 SD_MAX_RETRIES, NULL);
2118 * If the drive has indicated to us that it
2119 * doesn't have any media in it, don't bother
2120 * with any more polling.
2122 if (media_not_present(sdkp, &sshdr))
2126 sense_valid = scsi_sense_valid(&sshdr);
2128 } while (retries < 3 &&
2129 (!scsi_status_is_good(the_result) ||
2130 ((driver_byte(the_result) & DRIVER_SENSE) &&
2131 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2133 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2134 /* no sense, TUR either succeeded or failed
2135 * with a status error */
2136 if(!spintime && !scsi_status_is_good(the_result)) {
2137 sd_print_result(sdkp, "Test Unit Ready failed",
2144 * The device does not want the automatic start to be issued.
2146 if (sdkp->device->no_start_on_add)
2149 if (sense_valid && sshdr.sense_key == NOT_READY) {
2150 if (sshdr.asc == 4 && sshdr.ascq == 3)
2151 break; /* manual intervention required */
2152 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2153 break; /* standby */
2154 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2155 break; /* unavailable */
2156 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2157 break; /* sanitize in progress */
2159 * Issue command to spin up drive when not ready
2162 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2163 cmd[0] = START_STOP;
2164 cmd[1] = 1; /* Return immediately */
2165 memset((void *) &cmd[2], 0, 8);
2166 cmd[4] = 1; /* Start spin cycle */
2167 if (sdkp->device->start_stop_pwr_cond)
2169 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2171 SD_TIMEOUT, SD_MAX_RETRIES,
2173 spintime_expire = jiffies + 100 * HZ;
2176 /* Wait 1 second for next try */
2181 * Wait for USB flash devices with slow firmware.
2182 * Yes, this sense key/ASC combination shouldn't
2183 * occur here. It's characteristic of these devices.
2185 } else if (sense_valid &&
2186 sshdr.sense_key == UNIT_ATTENTION &&
2187 sshdr.asc == 0x28) {
2189 spintime_expire = jiffies + 5 * HZ;
2192 /* Wait 1 second for next try */
2195 /* we don't understand the sense code, so it's
2196 * probably pointless to loop */
2198 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2199 sd_print_sense_hdr(sdkp, &sshdr);
2204 } while (spintime && time_before_eq(jiffies, spintime_expire));
2207 if (scsi_status_is_good(the_result))
2210 printk("not responding...\n");
2215 * Determine whether disk supports Data Integrity Field.
2217 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2219 struct scsi_device *sdp = sdkp->device;
2223 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2224 sdkp->protection_type = 0;
2228 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2230 if (type > T10_PI_TYPE3_PROTECTION)
2232 else if (scsi_host_dif_capable(sdp->host, type))
2235 if (sdkp->first_scan || type != sdkp->protection_type)
2238 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2239 " protection type %u. Disabling disk!\n",
2243 sd_printk(KERN_NOTICE, sdkp,
2244 "Enabling DIF Type %u protection\n", type);
2247 sd_printk(KERN_NOTICE, sdkp,
2248 "Disabling DIF Type %u protection\n", type);
2252 sdkp->protection_type = type;
2257 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2258 struct scsi_sense_hdr *sshdr, int sense_valid,
2261 if (driver_byte(the_result) & DRIVER_SENSE)
2262 sd_print_sense_hdr(sdkp, sshdr);
2264 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2267 * Set dirty bit for removable devices if not ready -
2268 * sometimes drives will not report this properly.
2270 if (sdp->removable &&
2271 sense_valid && sshdr->sense_key == NOT_READY)
2272 set_media_not_present(sdkp);
2275 * We used to set media_present to 0 here to indicate no media
2276 * in the drive, but some drives fail read capacity even with
2277 * media present, so we can't do that.
2279 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2283 #if RC16_LEN > SD_BUF_SIZE
2284 #error RC16_LEN must not be more than SD_BUF_SIZE
2287 #define READ_CAPACITY_RETRIES_ON_RESET 10
2290 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2291 * and the reported logical block size is bigger than 512 bytes. Note
2292 * that last_sector is a u64 and therefore logical_to_sectors() is not
2295 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2297 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2299 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2305 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2306 unsigned char *buffer)
2308 unsigned char cmd[16];
2309 struct scsi_sense_hdr sshdr;
2310 int sense_valid = 0;
2312 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2313 unsigned int alignment;
2314 unsigned long long lba;
2315 unsigned sector_size;
2317 if (sdp->no_read_capacity_16)
2322 cmd[0] = SERVICE_ACTION_IN_16;
2323 cmd[1] = SAI_READ_CAPACITY_16;
2325 memset(buffer, 0, RC16_LEN);
2327 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2328 buffer, RC16_LEN, &sshdr,
2329 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2331 if (media_not_present(sdkp, &sshdr))
2335 sense_valid = scsi_sense_valid(&sshdr);
2337 sshdr.sense_key == ILLEGAL_REQUEST &&
2338 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2340 /* Invalid Command Operation Code or
2341 * Invalid Field in CDB, just retry
2342 * silently with RC10 */
2345 sshdr.sense_key == UNIT_ATTENTION &&
2346 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2347 /* Device reset might occur several times,
2348 * give it one more chance */
2349 if (--reset_retries > 0)
2354 } while (the_result && retries);
2357 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2358 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2362 sector_size = get_unaligned_be32(&buffer[8]);
2363 lba = get_unaligned_be64(&buffer[0]);
2365 if (sd_read_protection_type(sdkp, buffer) < 0) {
2370 if (!sd_addressable_capacity(lba, sector_size)) {
2371 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2372 "kernel compiled with support for large block "
2378 /* Logical blocks per physical block exponent */
2379 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2382 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2384 /* Lowest aligned logical block */
2385 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2386 blk_queue_alignment_offset(sdp->request_queue, alignment);
2387 if (alignment && sdkp->first_scan)
2388 sd_printk(KERN_NOTICE, sdkp,
2389 "physical block alignment offset: %u\n", alignment);
2391 if (buffer[14] & 0x80) { /* LBPME */
2394 if (buffer[14] & 0x40) /* LBPRZ */
2397 sd_config_discard(sdkp, SD_LBP_WS16);
2400 sdkp->capacity = lba + 1;
2404 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2405 unsigned char *buffer)
2407 unsigned char cmd[16];
2408 struct scsi_sense_hdr sshdr;
2409 int sense_valid = 0;
2411 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2413 unsigned sector_size;
2416 cmd[0] = READ_CAPACITY;
2417 memset(&cmd[1], 0, 9);
2418 memset(buffer, 0, 8);
2420 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2422 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2424 if (media_not_present(sdkp, &sshdr))
2428 sense_valid = scsi_sense_valid(&sshdr);
2430 sshdr.sense_key == UNIT_ATTENTION &&
2431 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2432 /* Device reset might occur several times,
2433 * give it one more chance */
2434 if (--reset_retries > 0)
2439 } while (the_result && retries);
2442 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2443 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2447 sector_size = get_unaligned_be32(&buffer[4]);
2448 lba = get_unaligned_be32(&buffer[0]);
2450 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2451 /* Some buggy (usb cardreader) devices return an lba of
2452 0xffffffff when the want to report a size of 0 (with
2453 which they really mean no media is present) */
2455 sdkp->physical_block_size = sector_size;
2459 if (!sd_addressable_capacity(lba, sector_size)) {
2460 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2461 "kernel compiled with support for large block "
2467 sdkp->capacity = lba + 1;
2468 sdkp->physical_block_size = sector_size;
2472 static int sd_try_rc16_first(struct scsi_device *sdp)
2474 if (sdp->host->max_cmd_len < 16)
2476 if (sdp->try_rc_10_first)
2478 if (sdp->scsi_level > SCSI_SPC_2)
2480 if (scsi_device_protection(sdp))
2486 * read disk capacity
2489 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2492 struct scsi_device *sdp = sdkp->device;
2494 if (sd_try_rc16_first(sdp)) {
2495 sector_size = read_capacity_16(sdkp, sdp, buffer);
2496 if (sector_size == -EOVERFLOW)
2498 if (sector_size == -ENODEV)
2500 if (sector_size < 0)
2501 sector_size = read_capacity_10(sdkp, sdp, buffer);
2502 if (sector_size < 0)
2505 sector_size = read_capacity_10(sdkp, sdp, buffer);
2506 if (sector_size == -EOVERFLOW)
2508 if (sector_size < 0)
2510 if ((sizeof(sdkp->capacity) > 4) &&
2511 (sdkp->capacity > 0xffffffffULL)) {
2512 int old_sector_size = sector_size;
2513 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2514 "Trying to use READ CAPACITY(16).\n");
2515 sector_size = read_capacity_16(sdkp, sdp, buffer);
2516 if (sector_size < 0) {
2517 sd_printk(KERN_NOTICE, sdkp,
2518 "Using 0xffffffff as device size\n");
2519 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2520 sector_size = old_sector_size;
2523 /* Remember that READ CAPACITY(16) succeeded */
2524 sdp->try_rc_10_first = 0;
2528 /* Some devices are known to return the total number of blocks,
2529 * not the highest block number. Some devices have versions
2530 * which do this and others which do not. Some devices we might
2531 * suspect of doing this but we don't know for certain.
2533 * If we know the reported capacity is wrong, decrement it. If
2534 * we can only guess, then assume the number of blocks is even
2535 * (usually true but not always) and err on the side of lowering
2538 if (sdp->fix_capacity ||
2539 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2540 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2541 "from its reported value: %llu\n",
2542 (unsigned long long) sdkp->capacity);
2547 if (sector_size == 0) {
2549 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2553 if (sector_size != 512 &&
2554 sector_size != 1024 &&
2555 sector_size != 2048 &&
2556 sector_size != 4096) {
2557 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2560 * The user might want to re-format the drive with
2561 * a supported sectorsize. Once this happens, it
2562 * would be relatively trivial to set the thing up.
2563 * For this reason, we leave the thing in the table.
2567 * set a bogus sector size so the normal read/write
2568 * logic in the block layer will eventually refuse any
2569 * request on this device without tripping over power
2570 * of two sector size assumptions
2574 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2575 blk_queue_physical_block_size(sdp->request_queue,
2576 sdkp->physical_block_size);
2577 sdkp->device->sector_size = sector_size;
2579 if (sdkp->capacity > 0xffffffff)
2580 sdp->use_16_for_rw = 1;
2585 * Print disk capacity
2588 sd_print_capacity(struct scsi_disk *sdkp,
2589 sector_t old_capacity)
2591 int sector_size = sdkp->device->sector_size;
2592 char cap_str_2[10], cap_str_10[10];
2594 string_get_size(sdkp->capacity, sector_size,
2595 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2596 string_get_size(sdkp->capacity, sector_size,
2597 STRING_UNITS_10, cap_str_10,
2598 sizeof(cap_str_10));
2600 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2601 sd_printk(KERN_NOTICE, sdkp,
2602 "%llu %d-byte logical blocks: (%s/%s)\n",
2603 (unsigned long long)sdkp->capacity,
2604 sector_size, cap_str_10, cap_str_2);
2606 if (sdkp->physical_block_size != sector_size)
2607 sd_printk(KERN_NOTICE, sdkp,
2608 "%u-byte physical blocks\n",
2609 sdkp->physical_block_size);
2611 sd_zbc_print_zones(sdkp);
2615 /* called with buffer of length 512 */
2617 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2618 unsigned char *buffer, int len, struct scsi_mode_data *data,
2619 struct scsi_sense_hdr *sshdr)
2621 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2622 SD_TIMEOUT, SD_MAX_RETRIES, data,
2627 * read write protect setting, if possible - called only in sd_revalidate_disk()
2628 * called with buffer of length SD_BUF_SIZE
2631 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2634 struct scsi_device *sdp = sdkp->device;
2635 struct scsi_mode_data data;
2636 int old_wp = sdkp->write_prot;
2638 set_disk_ro(sdkp->disk, 0);
2639 if (sdp->skip_ms_page_3f) {
2640 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2644 if (sdp->use_192_bytes_for_3f) {
2645 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2648 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2649 * We have to start carefully: some devices hang if we ask
2650 * for more than is available.
2652 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2655 * Second attempt: ask for page 0 When only page 0 is
2656 * implemented, a request for page 3F may return Sense Key
2657 * 5: Illegal Request, Sense Code 24: Invalid field in
2660 if (!scsi_status_is_good(res))
2661 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2664 * Third attempt: ask 255 bytes, as we did earlier.
2666 if (!scsi_status_is_good(res))
2667 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2671 if (!scsi_status_is_good(res)) {
2672 sd_first_printk(KERN_WARNING, sdkp,
2673 "Test WP failed, assume Write Enabled\n");
2675 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2676 set_disk_ro(sdkp->disk, sdkp->write_prot);
2677 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2678 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2679 sdkp->write_prot ? "on" : "off");
2680 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2686 * sd_read_cache_type - called only from sd_revalidate_disk()
2687 * called with buffer of length SD_BUF_SIZE
2690 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2693 struct scsi_device *sdp = sdkp->device;
2698 struct scsi_mode_data data;
2699 struct scsi_sense_hdr sshdr;
2700 int old_wce = sdkp->WCE;
2701 int old_rcd = sdkp->RCD;
2702 int old_dpofua = sdkp->DPOFUA;
2705 if (sdkp->cache_override)
2709 if (sdp->skip_ms_page_8) {
2710 if (sdp->type == TYPE_RBC)
2713 if (sdp->skip_ms_page_3f)
2716 if (sdp->use_192_bytes_for_3f)
2720 } else if (sdp->type == TYPE_RBC) {
2728 /* cautiously ask */
2729 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2732 if (!scsi_status_is_good(res))
2735 if (!data.header_length) {
2738 sd_first_printk(KERN_ERR, sdkp,
2739 "Missing header in MODE_SENSE response\n");
2742 /* that went OK, now ask for the proper length */
2746 * We're only interested in the first three bytes, actually.
2747 * But the data cache page is defined for the first 20.
2751 else if (len > SD_BUF_SIZE) {
2752 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2753 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2756 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2760 if (len > first_len)
2761 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2764 if (scsi_status_is_good(res)) {
2765 int offset = data.header_length + data.block_descriptor_length;
2767 while (offset < len) {
2768 u8 page_code = buffer[offset] & 0x3F;
2769 u8 spf = buffer[offset] & 0x40;
2771 if (page_code == 8 || page_code == 6) {
2772 /* We're interested only in the first 3 bytes.
2774 if (len - offset <= 2) {
2775 sd_first_printk(KERN_ERR, sdkp,
2776 "Incomplete mode parameter "
2780 modepage = page_code;
2784 /* Go to the next page */
2785 if (spf && len - offset > 3)
2786 offset += 4 + (buffer[offset+2] << 8) +
2788 else if (!spf && len - offset > 1)
2789 offset += 2 + buffer[offset+1];
2791 sd_first_printk(KERN_ERR, sdkp,
2793 "parameter data\n");
2799 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2803 if (modepage == 8) {
2804 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2805 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2807 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2811 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2812 if (sdp->broken_fua) {
2813 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2815 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2816 !sdkp->device->use_16_for_rw) {
2817 sd_first_printk(KERN_NOTICE, sdkp,
2818 "Uses READ/WRITE(6), disabling FUA\n");
2822 /* No cache flush allowed for write protected devices */
2823 if (sdkp->WCE && sdkp->write_prot)
2826 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2827 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2828 sd_printk(KERN_NOTICE, sdkp,
2829 "Write cache: %s, read cache: %s, %s\n",
2830 sdkp->WCE ? "enabled" : "disabled",
2831 sdkp->RCD ? "disabled" : "enabled",
2832 sdkp->DPOFUA ? "supports DPO and FUA"
2833 : "doesn't support DPO or FUA");
2839 if (scsi_sense_valid(&sshdr) &&
2840 sshdr.sense_key == ILLEGAL_REQUEST &&
2841 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2842 /* Invalid field in CDB */
2843 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2845 sd_first_printk(KERN_ERR, sdkp,
2846 "Asking for cache data failed\n");
2849 if (sdp->wce_default_on) {
2850 sd_first_printk(KERN_NOTICE, sdkp,
2851 "Assuming drive cache: write back\n");
2854 sd_first_printk(KERN_ERR, sdkp,
2855 "Assuming drive cache: write through\n");
2863 * The ATO bit indicates whether the DIF application tag is available
2864 * for use by the operating system.
2866 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2869 struct scsi_device *sdp = sdkp->device;
2870 struct scsi_mode_data data;
2871 struct scsi_sense_hdr sshdr;
2873 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2876 if (sdkp->protection_type == 0)
2879 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2880 SD_MAX_RETRIES, &data, &sshdr);
2882 if (!scsi_status_is_good(res) || !data.header_length ||
2884 sd_first_printk(KERN_WARNING, sdkp,
2885 "getting Control mode page failed, assume no ATO\n");
2887 if (scsi_sense_valid(&sshdr))
2888 sd_print_sense_hdr(sdkp, &sshdr);
2893 offset = data.header_length + data.block_descriptor_length;
2895 if ((buffer[offset] & 0x3f) != 0x0a) {
2896 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2900 if ((buffer[offset + 5] & 0x80) == 0)
2909 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2910 * @sdkp: disk to query
2912 static void sd_read_block_limits(struct scsi_disk *sdkp)
2914 unsigned int sector_sz = sdkp->device->sector_size;
2915 const int vpd_len = 64;
2916 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2919 /* Block Limits VPD */
2920 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2923 blk_queue_io_min(sdkp->disk->queue,
2924 get_unaligned_be16(&buffer[6]) * sector_sz);
2926 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2927 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2929 if (buffer[3] == 0x3c) {
2930 unsigned int lba_count, desc_count;
2932 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2937 lba_count = get_unaligned_be32(&buffer[20]);
2938 desc_count = get_unaligned_be32(&buffer[24]);
2940 if (lba_count && desc_count)
2941 sdkp->max_unmap_blocks = lba_count;
2943 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2945 if (buffer[32] & 0x80)
2946 sdkp->unmap_alignment =
2947 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2949 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2951 if (sdkp->max_unmap_blocks)
2952 sd_config_discard(sdkp, SD_LBP_UNMAP);
2954 sd_config_discard(sdkp, SD_LBP_WS16);
2956 } else { /* LBP VPD page tells us what to use */
2957 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2958 sd_config_discard(sdkp, SD_LBP_UNMAP);
2959 else if (sdkp->lbpws)
2960 sd_config_discard(sdkp, SD_LBP_WS16);
2961 else if (sdkp->lbpws10)
2962 sd_config_discard(sdkp, SD_LBP_WS10);
2964 sd_config_discard(sdkp, SD_LBP_DISABLE);
2973 * sd_read_block_characteristics - Query block dev. characteristics
2974 * @sdkp: disk to query
2976 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2978 struct request_queue *q = sdkp->disk->queue;
2979 unsigned char *buffer;
2981 const int vpd_len = 64;
2983 buffer = kmalloc(vpd_len, GFP_KERNEL);
2986 /* Block Device Characteristics VPD */
2987 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2990 rot = get_unaligned_be16(&buffer[4]);
2993 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2994 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2997 if (sdkp->device->type == TYPE_ZBC) {
2999 q->limits.zoned = BLK_ZONED_HM;
3001 sdkp->zoned = (buffer[8] >> 4) & 3;
3002 if (sdkp->zoned == 1)
3004 q->limits.zoned = BLK_ZONED_HA;
3007 * Treat drive-managed devices as
3008 * regular block devices.
3010 q->limits.zoned = BLK_ZONED_NONE;
3012 if (blk_queue_is_zoned(q) && sdkp->first_scan)
3013 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3014 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3021 * sd_read_block_provisioning - Query provisioning VPD page
3022 * @sdkp: disk to query
3024 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3026 unsigned char *buffer;
3027 const int vpd_len = 8;
3029 if (sdkp->lbpme == 0)
3032 buffer = kmalloc(vpd_len, GFP_KERNEL);
3034 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3038 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3039 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3040 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3046 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3048 struct scsi_device *sdev = sdkp->device;
3050 if (sdev->host->no_write_same) {
3051 sdev->no_write_same = 1;
3056 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3057 /* too large values might cause issues with arcmsr */
3058 int vpd_buf_len = 64;
3060 sdev->no_report_opcodes = 1;
3062 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3063 * CODES is unsupported and the device has an ATA
3064 * Information VPD page (SAT).
3066 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3067 sdev->no_write_same = 1;
3070 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3073 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3077 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3079 struct scsi_device *sdev = sdkp->device;
3081 if (!sdev->security_supported)
3084 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3085 SECURITY_PROTOCOL_IN) == 1 &&
3086 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3087 SECURITY_PROTOCOL_OUT) == 1)
3092 * Determine the device's preferred I/O size for reads and writes
3093 * unless the reported value is unreasonably small, large, not a
3094 * multiple of the physical block size, or simply garbage.
3096 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3097 unsigned int dev_max)
3099 struct scsi_device *sdp = sdkp->device;
3100 unsigned int opt_xfer_bytes =
3101 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3103 if (sdkp->opt_xfer_blocks == 0)
3106 if (sdkp->opt_xfer_blocks > dev_max) {
3107 sd_first_printk(KERN_WARNING, sdkp,
3108 "Optimal transfer size %u logical blocks " \
3109 "> dev_max (%u logical blocks)\n",
3110 sdkp->opt_xfer_blocks, dev_max);
3114 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3115 sd_first_printk(KERN_WARNING, sdkp,
3116 "Optimal transfer size %u logical blocks " \
3117 "> sd driver limit (%u logical blocks)\n",
3118 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3122 if (opt_xfer_bytes < PAGE_SIZE) {
3123 sd_first_printk(KERN_WARNING, sdkp,
3124 "Optimal transfer size %u bytes < " \
3125 "PAGE_SIZE (%u bytes)\n",
3126 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3130 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3131 sd_first_printk(KERN_WARNING, sdkp,
3132 "Optimal transfer size %u bytes not a " \
3133 "multiple of physical block size (%u bytes)\n",
3134 opt_xfer_bytes, sdkp->physical_block_size);
3138 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3144 * sd_revalidate_disk - called the first time a new disk is seen,
3145 * performs disk spin up, read_capacity, etc.
3146 * @disk: struct gendisk we care about
3148 static int sd_revalidate_disk(struct gendisk *disk)
3150 struct scsi_disk *sdkp = scsi_disk(disk);
3151 struct scsi_device *sdp = sdkp->device;
3152 struct request_queue *q = sdkp->disk->queue;
3153 sector_t old_capacity = sdkp->capacity;
3154 unsigned char *buffer;
3155 unsigned int dev_max, rw_max;
3157 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3158 "sd_revalidate_disk\n"));
3161 * If the device is offline, don't try and read capacity or any
3162 * of the other niceties.
3164 if (!scsi_device_online(sdp))
3167 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3169 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3170 "allocation failure.\n");
3174 sd_spinup_disk(sdkp);
3177 * Without media there is no reason to ask; moreover, some devices
3178 * react badly if we do.
3180 if (sdkp->media_present) {
3181 sd_read_capacity(sdkp, buffer);
3183 if (scsi_device_supports_vpd(sdp)) {
3184 sd_read_block_provisioning(sdkp);
3185 sd_read_block_limits(sdkp);
3186 sd_read_block_characteristics(sdkp);
3187 sd_zbc_read_zones(sdkp, buffer);
3190 sd_print_capacity(sdkp, old_capacity);
3192 sd_read_write_protect_flag(sdkp, buffer);
3193 sd_read_cache_type(sdkp, buffer);
3194 sd_read_app_tag_own(sdkp, buffer);
3195 sd_read_write_same(sdkp, buffer);
3196 sd_read_security(sdkp, buffer);
3200 * We now have all cache related info, determine how we deal
3201 * with flush requests.
3203 sd_set_flush_flag(sdkp);
3205 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3206 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3208 /* Some devices report a maximum block count for READ/WRITE requests. */
3209 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3210 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3212 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3213 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3214 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3216 q->limits.io_opt = 0;
3217 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3218 (sector_t)BLK_DEF_MAX_SECTORS);
3221 /* Do not exceed controller limit */
3222 rw_max = min(rw_max, queue_max_hw_sectors(q));
3225 * Only update max_sectors if previously unset or if the current value
3226 * exceeds the capabilities of the hardware.
3228 if (sdkp->first_scan ||
3229 q->limits.max_sectors > q->limits.max_dev_sectors ||
3230 q->limits.max_sectors > q->limits.max_hw_sectors)
3231 q->limits.max_sectors = rw_max;
3233 sdkp->first_scan = 0;
3235 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3236 sd_config_write_same(sdkp);
3244 * sd_unlock_native_capacity - unlock native capacity
3245 * @disk: struct gendisk to set capacity for
3247 * Block layer calls this function if it detects that partitions
3248 * on @disk reach beyond the end of the device. If the SCSI host
3249 * implements ->unlock_native_capacity() method, it's invoked to
3250 * give it a chance to adjust the device capacity.
3253 * Defined by block layer. Might sleep.
3255 static void sd_unlock_native_capacity(struct gendisk *disk)
3257 struct scsi_device *sdev = scsi_disk(disk)->device;
3259 if (sdev->host->hostt->unlock_native_capacity)
3260 sdev->host->hostt->unlock_native_capacity(sdev);
3264 * sd_format_disk_name - format disk name
3265 * @prefix: name prefix - ie. "sd" for SCSI disks
3266 * @index: index of the disk to format name for
3267 * @buf: output buffer
3268 * @buflen: length of the output buffer
3270 * SCSI disk names starts at sda. The 26th device is sdz and the
3271 * 27th is sdaa. The last one for two lettered suffix is sdzz
3272 * which is followed by sdaaa.
3274 * This is basically 26 base counting with one extra 'nil' entry
3275 * at the beginning from the second digit on and can be
3276 * determined using similar method as 26 base conversion with the
3277 * index shifted -1 after each digit is computed.
3283 * 0 on success, -errno on failure.
3285 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3287 const int base = 'z' - 'a' + 1;
3288 char *begin = buf + strlen(prefix);
3289 char *end = buf + buflen;
3299 *--p = 'a' + (index % unit);
3300 index = (index / unit) - 1;
3301 } while (index >= 0);
3303 memmove(begin, p, end - p);
3304 memcpy(buf, prefix, strlen(prefix));
3310 * The asynchronous part of sd_probe
3312 static void sd_probe_async(void *data, async_cookie_t cookie)
3314 struct scsi_disk *sdkp = data;
3315 struct scsi_device *sdp;
3322 index = sdkp->index;
3323 dev = &sdp->sdev_gendev;
3325 gd->major = sd_major((index & 0xf0) >> 4);
3326 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3328 gd->fops = &sd_fops;
3329 gd->private_data = &sdkp->driver;
3330 gd->queue = sdkp->device->request_queue;
3332 /* defaults, until the device tells us otherwise */
3333 sdp->sector_size = 512;
3335 sdkp->media_present = 1;
3336 sdkp->write_prot = 0;
3337 sdkp->cache_override = 0;
3341 sdkp->first_scan = 1;
3342 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3344 sd_revalidate_disk(gd);
3346 gd->flags = GENHD_FL_EXT_DEVT;
3347 if (sdp->removable) {
3348 gd->flags |= GENHD_FL_REMOVABLE;
3349 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3352 blk_pm_runtime_init(sdp->request_queue, dev);
3353 device_add_disk(dev, gd);
3355 sd_dif_config_host(sdkp);
3357 sd_revalidate_disk(gd);
3359 if (sdkp->security) {
3360 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3362 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3365 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3366 sdp->removable ? "removable " : "");
3367 scsi_autopm_put_device(sdp);
3368 put_device(&sdkp->dev);
3372 * sd_probe - called during driver initialization and whenever a
3373 * new scsi device is attached to the system. It is called once
3374 * for each scsi device (not just disks) present.
3375 * @dev: pointer to device object
3377 * Returns 0 if successful (or not interested in this scsi device
3378 * (e.g. scanner)); 1 when there is an error.
3380 * Note: this function is invoked from the scsi mid-level.
3381 * This function sets up the mapping between a given
3382 * <host,channel,id,lun> (found in sdp) and new device name
3383 * (e.g. /dev/sda). More precisely it is the block device major
3384 * and minor number that is chosen here.
3386 * Assume sd_probe is not re-entrant (for time being)
3387 * Also think about sd_probe() and sd_remove() running coincidentally.
3389 static int sd_probe(struct device *dev)
3391 struct scsi_device *sdp = to_scsi_device(dev);
3392 struct scsi_disk *sdkp;
3397 scsi_autopm_get_device(sdp);
3399 if (sdp->type != TYPE_DISK &&
3400 sdp->type != TYPE_ZBC &&
3401 sdp->type != TYPE_MOD &&
3402 sdp->type != TYPE_RBC)
3405 #ifndef CONFIG_BLK_DEV_ZONED
3406 if (sdp->type == TYPE_ZBC)
3409 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3413 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3417 gd = alloc_disk(SD_MINORS);
3422 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3425 spin_lock(&sd_index_lock);
3426 error = ida_get_new(&sd_index_ida, &index);
3427 spin_unlock(&sd_index_lock);
3428 } while (error == -EAGAIN);
3431 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3435 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3437 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3438 goto out_free_index;
3442 sdkp->driver = &sd_template;
3444 sdkp->index = index;
3445 atomic_set(&sdkp->openers, 0);
3446 atomic_set(&sdkp->device->ioerr_cnt, 0);
3448 if (!sdp->request_queue->rq_timeout) {
3449 if (sdp->type != TYPE_MOD)
3450 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3452 blk_queue_rq_timeout(sdp->request_queue,
3456 device_initialize(&sdkp->dev);
3457 sdkp->dev.parent = get_device(dev);
3458 sdkp->dev.class = &sd_disk_class;
3459 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3461 error = device_add(&sdkp->dev);
3463 put_device(&sdkp->dev);
3467 dev_set_drvdata(dev, sdkp);
3469 get_device(&sdkp->dev); /* prevent release before async_schedule */
3470 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3475 spin_lock(&sd_index_lock);
3476 ida_remove(&sd_index_ida, index);
3477 spin_unlock(&sd_index_lock);
3483 scsi_autopm_put_device(sdp);
3488 * sd_remove - called whenever a scsi disk (previously recognized by
3489 * sd_probe) is detached from the system. It is called (potentially
3490 * multiple times) during sd module unload.
3491 * @dev: pointer to device object
3493 * Note: this function is invoked from the scsi mid-level.
3494 * This function potentially frees up a device name (e.g. /dev/sdc)
3495 * that could be re-used by a subsequent sd_probe().
3496 * This function is not called when the built-in sd driver is "exit-ed".
3498 static int sd_remove(struct device *dev)
3500 struct scsi_disk *sdkp;
3503 sdkp = dev_get_drvdata(dev);
3504 devt = disk_devt(sdkp->disk);
3505 scsi_autopm_get_device(sdkp->device);
3507 async_synchronize_full_domain(&scsi_sd_pm_domain);
3508 async_synchronize_full_domain(&scsi_sd_probe_domain);
3509 device_del(&sdkp->dev);
3510 del_gendisk(sdkp->disk);
3513 sd_zbc_remove(sdkp);
3515 free_opal_dev(sdkp->opal_dev);
3517 blk_register_region(devt, SD_MINORS, NULL,
3518 sd_default_probe, NULL, NULL);
3520 mutex_lock(&sd_ref_mutex);
3521 dev_set_drvdata(dev, NULL);
3522 put_device(&sdkp->dev);
3523 mutex_unlock(&sd_ref_mutex);
3529 * scsi_disk_release - Called to free the scsi_disk structure
3530 * @dev: pointer to embedded class device
3532 * sd_ref_mutex must be held entering this routine. Because it is
3533 * called on last put, you should always use the scsi_disk_get()
3534 * scsi_disk_put() helpers which manipulate the semaphore directly
3535 * and never do a direct put_device.
3537 static void scsi_disk_release(struct device *dev)
3539 struct scsi_disk *sdkp = to_scsi_disk(dev);
3540 struct gendisk *disk = sdkp->disk;
3541 struct request_queue *q = disk->queue;
3543 spin_lock(&sd_index_lock);
3544 ida_remove(&sd_index_ida, sdkp->index);
3545 spin_unlock(&sd_index_lock);
3548 * Wait until all requests that are in progress have completed.
3549 * This is necessary to avoid that e.g. scsi_end_request() crashes
3550 * due to clearing the disk->private_data pointer. Wait from inside
3551 * scsi_disk_release() instead of from sd_release() to avoid that
3552 * freezing and unfreezing the request queue affects user space I/O
3553 * in case multiple processes open a /dev/sd... node concurrently.
3555 blk_mq_freeze_queue(q);
3556 blk_mq_unfreeze_queue(q);
3558 disk->private_data = NULL;
3560 put_device(&sdkp->device->sdev_gendev);
3565 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3567 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3568 struct scsi_sense_hdr sshdr;
3569 struct scsi_device *sdp = sdkp->device;
3573 cmd[4] |= 1; /* START */
3575 if (sdp->start_stop_pwr_cond)
3576 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3578 if (!scsi_device_online(sdp))
3581 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3582 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3584 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3585 if (driver_byte(res) & DRIVER_SENSE)
3586 sd_print_sense_hdr(sdkp, &sshdr);
3587 if (scsi_sense_valid(&sshdr) &&
3588 /* 0x3a is medium not present */
3593 /* SCSI error codes must not go to the generic layer */
3601 * Send a SYNCHRONIZE CACHE instruction down to the device through
3602 * the normal SCSI command structure. Wait for the command to
3605 static void sd_shutdown(struct device *dev)
3607 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3610 return; /* this can happen */
3612 if (pm_runtime_suspended(dev))
3615 if (sdkp->WCE && sdkp->media_present) {
3616 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3617 sd_sync_cache(sdkp, NULL);
3620 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3621 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3622 sd_start_stop_device(sdkp, 0);
3626 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3628 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3629 struct scsi_sense_hdr sshdr;
3632 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3635 if (sdkp->WCE && sdkp->media_present) {
3636 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3637 ret = sd_sync_cache(sdkp, &sshdr);
3640 /* ignore OFFLINE device */
3644 if (!scsi_sense_valid(&sshdr) ||
3645 sshdr.sense_key != ILLEGAL_REQUEST)
3649 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3650 * doesn't support sync. There's not much to do and
3651 * suspend shouldn't fail.
3657 if (sdkp->device->manage_start_stop) {
3658 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3659 /* an error is not worth aborting a system sleep */
3660 ret = sd_start_stop_device(sdkp, 0);
3661 if (ignore_stop_errors)
3668 static int sd_suspend_system(struct device *dev)
3670 return sd_suspend_common(dev, true);
3673 static int sd_suspend_runtime(struct device *dev)
3675 return sd_suspend_common(dev, false);
3678 static int sd_resume(struct device *dev)
3680 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3683 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3686 if (!sdkp->device->manage_start_stop)
3689 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3690 ret = sd_start_stop_device(sdkp, 1);
3692 opal_unlock_from_suspend(sdkp->opal_dev);
3697 * init_sd - entry point for this driver (both when built in or when
3700 * Note: this function registers this driver with the scsi mid-level.
3702 static int __init init_sd(void)
3704 int majors = 0, i, err;
3706 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3708 for (i = 0; i < SD_MAJORS; i++) {
3709 if (register_blkdev(sd_major(i), "sd") != 0)
3712 blk_register_region(sd_major(i), SD_MINORS, NULL,
3713 sd_default_probe, NULL, NULL);
3719 err = class_register(&sd_disk_class);
3723 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3725 if (!sd_cdb_cache) {
3726 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3731 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3733 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3738 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3739 if (!sd_page_pool) {
3740 printk(KERN_ERR "sd: can't init discard page pool\n");
3745 err = scsi_register_driver(&sd_template.gendrv);
3747 goto err_out_driver;
3752 mempool_destroy(sd_page_pool);
3755 mempool_destroy(sd_cdb_pool);
3758 kmem_cache_destroy(sd_cdb_cache);
3761 class_unregister(&sd_disk_class);
3763 for (i = 0; i < SD_MAJORS; i++)
3764 unregister_blkdev(sd_major(i), "sd");
3769 * exit_sd - exit point for this driver (when it is a module).
3771 * Note: this function unregisters this driver from the scsi mid-level.
3773 static void __exit exit_sd(void)
3777 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3779 scsi_unregister_driver(&sd_template.gendrv);
3780 mempool_destroy(sd_cdb_pool);
3781 mempool_destroy(sd_page_pool);
3782 kmem_cache_destroy(sd_cdb_cache);
3784 class_unregister(&sd_disk_class);
3786 for (i = 0; i < SD_MAJORS; i++) {
3787 blk_unregister_region(sd_major(i), SD_MINORS);
3788 unregister_blkdev(sd_major(i), "sd");
3792 module_init(init_sd);
3793 module_exit(exit_sd);
3795 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3796 struct scsi_sense_hdr *sshdr)
3798 scsi_print_sense_hdr(sdkp->device,
3799 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3802 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3805 const char *hb_string = scsi_hostbyte_string(result);
3806 const char *db_string = scsi_driverbyte_string(result);
3808 if (hb_string || db_string)
3809 sd_printk(KERN_INFO, sdkp,
3810 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3811 hb_string ? hb_string : "invalid",
3812 db_string ? db_string : "invalid");
3814 sd_printk(KERN_INFO, sdkp,
3815 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3816 msg, host_byte(result), driver_byte(result));