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/pm_runtime.h>
55 #include <linux/t10-pi.h>
56 #include <linux/uaccess.h>
57 #include <asm/unaligned.h>
59 #include <scsi/scsi.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_dbg.h>
62 #include <scsi/scsi_device.h>
63 #include <scsi/scsi_driver.h>
64 #include <scsi/scsi_eh.h>
65 #include <scsi/scsi_host.h>
66 #include <scsi/scsi_ioctl.h>
67 #include <scsi/scsicam.h>
70 #include "scsi_priv.h"
71 #include "scsi_logging.h"
73 MODULE_AUTHOR("Eric Youngdale");
74 MODULE_DESCRIPTION("SCSI disk (sd) driver");
75 MODULE_LICENSE("GPL");
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
98 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
104 static void sd_config_discard(struct scsi_disk *, unsigned int);
105 static void sd_config_write_same(struct scsi_disk *);
106 static int sd_revalidate_disk(struct gendisk *);
107 static void sd_unlock_native_capacity(struct gendisk *disk);
108 static int sd_probe(struct device *);
109 static int sd_remove(struct device *);
110 static void sd_shutdown(struct device *);
111 static int sd_suspend_system(struct device *);
112 static int sd_suspend_runtime(struct device *);
113 static int sd_resume(struct device *);
114 static void sd_rescan(struct device *);
115 static int sd_init_command(struct scsi_cmnd *SCpnt);
116 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117 static int sd_done(struct scsi_cmnd *);
118 static int sd_eh_action(struct scsi_cmnd *, int);
119 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
120 static void scsi_disk_release(struct device *cdev);
121 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
122 static void sd_print_result(const struct scsi_disk *, const char *, int);
124 static DEFINE_SPINLOCK(sd_index_lock);
125 static DEFINE_IDA(sd_index_ida);
127 /* This semaphore is used to mediate the 0->1 reference get in the
128 * face of object destruction (i.e. we can't allow a get on an
129 * object after last put) */
130 static DEFINE_MUTEX(sd_ref_mutex);
132 static struct kmem_cache *sd_cdb_cache;
133 static mempool_t *sd_cdb_pool;
135 static const char *sd_cache_types[] = {
136 "write through", "none", "write back",
137 "write back, no read (daft)"
140 static void sd_set_flush_flag(struct scsi_disk *sdkp)
142 bool wc = false, fua = false;
150 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
154 cache_type_store(struct device *dev, struct device_attribute *attr,
155 const char *buf, size_t count)
157 int i, ct = -1, rcd, wce, sp;
158 struct scsi_disk *sdkp = to_scsi_disk(dev);
159 struct scsi_device *sdp = sdkp->device;
162 struct scsi_mode_data data;
163 struct scsi_sense_hdr sshdr;
164 static const char temp[] = "temporary ";
167 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
168 /* no cache control on RBC devices; theoretically they
169 * can do it, but there's probably so many exceptions
170 * it's not worth the risk */
173 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
174 buf += sizeof(temp) - 1;
175 sdkp->cache_override = 1;
177 sdkp->cache_override = 0;
180 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
181 len = strlen(sd_cache_types[i]);
182 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
190 rcd = ct & 0x01 ? 1 : 0;
191 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
193 if (sdkp->cache_override) {
196 sd_set_flush_flag(sdkp);
200 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
201 SD_MAX_RETRIES, &data, NULL))
203 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
204 data.block_descriptor_length);
205 buffer_data = buffer + data.header_length +
206 data.block_descriptor_length;
207 buffer_data[2] &= ~0x05;
208 buffer_data[2] |= wce << 2 | rcd;
209 sp = buffer_data[0] & 0x80 ? 1 : 0;
210 buffer_data[0] &= ~0x80;
212 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
213 SD_MAX_RETRIES, &data, &sshdr)) {
214 if (scsi_sense_valid(&sshdr))
215 sd_print_sense_hdr(sdkp, &sshdr);
218 revalidate_disk(sdkp->disk);
223 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
226 struct scsi_disk *sdkp = to_scsi_disk(dev);
227 struct scsi_device *sdp = sdkp->device;
229 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
233 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
234 const char *buf, size_t count)
236 struct scsi_disk *sdkp = to_scsi_disk(dev);
237 struct scsi_device *sdp = sdkp->device;
239 if (!capable(CAP_SYS_ADMIN))
242 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
246 static DEVICE_ATTR_RW(manage_start_stop);
249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
251 struct scsi_disk *sdkp = to_scsi_disk(dev);
253 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
257 allow_restart_store(struct device *dev, struct device_attribute *attr,
258 const char *buf, size_t count)
260 struct scsi_disk *sdkp = to_scsi_disk(dev);
261 struct scsi_device *sdp = sdkp->device;
263 if (!capable(CAP_SYS_ADMIN))
266 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
269 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
273 static DEVICE_ATTR_RW(allow_restart);
276 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
278 struct scsi_disk *sdkp = to_scsi_disk(dev);
279 int ct = sdkp->RCD + 2*sdkp->WCE;
281 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
283 static DEVICE_ATTR_RW(cache_type);
286 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
288 struct scsi_disk *sdkp = to_scsi_disk(dev);
290 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
292 static DEVICE_ATTR_RO(FUA);
295 protection_type_show(struct device *dev, struct device_attribute *attr,
298 struct scsi_disk *sdkp = to_scsi_disk(dev);
300 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
304 protection_type_store(struct device *dev, struct device_attribute *attr,
305 const char *buf, size_t count)
307 struct scsi_disk *sdkp = to_scsi_disk(dev);
311 if (!capable(CAP_SYS_ADMIN))
314 err = kstrtouint(buf, 10, &val);
319 if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
320 sdkp->protection_type = val;
324 static DEVICE_ATTR_RW(protection_type);
327 protection_mode_show(struct device *dev, struct device_attribute *attr,
330 struct scsi_disk *sdkp = to_scsi_disk(dev);
331 struct scsi_device *sdp = sdkp->device;
332 unsigned int dif, dix;
334 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
335 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
337 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
343 return snprintf(buf, 20, "none\n");
345 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
347 static DEVICE_ATTR_RO(protection_mode);
350 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
352 struct scsi_disk *sdkp = to_scsi_disk(dev);
354 return snprintf(buf, 20, "%u\n", sdkp->ATO);
356 static DEVICE_ATTR_RO(app_tag_own);
359 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
362 struct scsi_disk *sdkp = to_scsi_disk(dev);
364 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
366 static DEVICE_ATTR_RO(thin_provisioning);
368 static const char *lbp_mode[] = {
369 [SD_LBP_FULL] = "full",
370 [SD_LBP_UNMAP] = "unmap",
371 [SD_LBP_WS16] = "writesame_16",
372 [SD_LBP_WS10] = "writesame_10",
373 [SD_LBP_ZERO] = "writesame_zero",
374 [SD_LBP_DISABLE] = "disabled",
378 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
381 struct scsi_disk *sdkp = to_scsi_disk(dev);
383 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
387 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
388 const char *buf, size_t count)
390 struct scsi_disk *sdkp = to_scsi_disk(dev);
391 struct scsi_device *sdp = sdkp->device;
393 if (!capable(CAP_SYS_ADMIN))
396 if (sd_is_zoned(sdkp)) {
397 sd_config_discard(sdkp, SD_LBP_DISABLE);
401 if (sdp->type != TYPE_DISK)
404 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
405 sd_config_discard(sdkp, SD_LBP_UNMAP);
406 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
407 sd_config_discard(sdkp, SD_LBP_WS16);
408 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
409 sd_config_discard(sdkp, SD_LBP_WS10);
410 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
411 sd_config_discard(sdkp, SD_LBP_ZERO);
412 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
413 sd_config_discard(sdkp, SD_LBP_DISABLE);
419 static DEVICE_ATTR_RW(provisioning_mode);
422 max_medium_access_timeouts_show(struct device *dev,
423 struct device_attribute *attr, char *buf)
425 struct scsi_disk *sdkp = to_scsi_disk(dev);
427 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
431 max_medium_access_timeouts_store(struct device *dev,
432 struct device_attribute *attr, const char *buf,
435 struct scsi_disk *sdkp = to_scsi_disk(dev);
438 if (!capable(CAP_SYS_ADMIN))
441 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
443 return err ? err : count;
445 static DEVICE_ATTR_RW(max_medium_access_timeouts);
448 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
451 struct scsi_disk *sdkp = to_scsi_disk(dev);
453 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
457 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
458 const char *buf, size_t count)
460 struct scsi_disk *sdkp = to_scsi_disk(dev);
461 struct scsi_device *sdp = sdkp->device;
465 if (!capable(CAP_SYS_ADMIN))
468 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
471 err = kstrtoul(buf, 10, &max);
477 sdp->no_write_same = 1;
478 else if (max <= SD_MAX_WS16_BLOCKS) {
479 sdp->no_write_same = 0;
480 sdkp->max_ws_blocks = max;
483 sd_config_write_same(sdkp);
487 static DEVICE_ATTR_RW(max_write_same_blocks);
489 static struct attribute *sd_disk_attrs[] = {
490 &dev_attr_cache_type.attr,
492 &dev_attr_allow_restart.attr,
493 &dev_attr_manage_start_stop.attr,
494 &dev_attr_protection_type.attr,
495 &dev_attr_protection_mode.attr,
496 &dev_attr_app_tag_own.attr,
497 &dev_attr_thin_provisioning.attr,
498 &dev_attr_provisioning_mode.attr,
499 &dev_attr_max_write_same_blocks.attr,
500 &dev_attr_max_medium_access_timeouts.attr,
503 ATTRIBUTE_GROUPS(sd_disk);
505 static struct class sd_disk_class = {
507 .owner = THIS_MODULE,
508 .dev_release = scsi_disk_release,
509 .dev_groups = sd_disk_groups,
512 static const struct dev_pm_ops sd_pm_ops = {
513 .suspend = sd_suspend_system,
515 .poweroff = sd_suspend_system,
516 .restore = sd_resume,
517 .runtime_suspend = sd_suspend_runtime,
518 .runtime_resume = sd_resume,
521 static struct scsi_driver sd_template = {
524 .owner = THIS_MODULE,
527 .shutdown = sd_shutdown,
531 .init_command = sd_init_command,
532 .uninit_command = sd_uninit_command,
534 .eh_action = sd_eh_action,
538 * Dummy kobj_map->probe function.
539 * The default ->probe function will call modprobe, which is
540 * pointless as this module is already loaded.
542 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
548 * Device no to disk mapping:
550 * major disc2 disc p1
551 * |............|.............|....|....| <- dev_t
554 * Inside a major, we have 16k disks, however mapped non-
555 * contiguously. The first 16 disks are for major0, the next
556 * ones with major1, ... Disk 256 is for major0 again, disk 272
558 * As we stay compatible with our numbering scheme, we can reuse
559 * the well-know SCSI majors 8, 65--71, 136--143.
561 static int sd_major(int major_idx)
565 return SCSI_DISK0_MAJOR;
567 return SCSI_DISK1_MAJOR + major_idx - 1;
569 return SCSI_DISK8_MAJOR + major_idx - 8;
572 return 0; /* shut up gcc */
576 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
578 struct scsi_disk *sdkp = NULL;
580 mutex_lock(&sd_ref_mutex);
582 if (disk->private_data) {
583 sdkp = scsi_disk(disk);
584 if (scsi_device_get(sdkp->device) == 0)
585 get_device(&sdkp->dev);
589 mutex_unlock(&sd_ref_mutex);
593 static void scsi_disk_put(struct scsi_disk *sdkp)
595 struct scsi_device *sdev = sdkp->device;
597 mutex_lock(&sd_ref_mutex);
598 put_device(&sdkp->dev);
599 scsi_device_put(sdev);
600 mutex_unlock(&sd_ref_mutex);
603 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
604 unsigned int dix, unsigned int dif)
606 struct bio *bio = scmd->request->bio;
607 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
608 unsigned int protect = 0;
610 if (dix) { /* DIX Type 0, 1, 2, 3 */
611 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
612 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
614 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
615 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
618 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
619 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
621 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
622 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
625 if (dif) { /* DIX/DIF Type 1, 2, 3 */
626 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
628 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
629 protect = 3 << 5; /* Disable target PI checking */
631 protect = 1 << 5; /* Enable target PI checking */
634 scsi_set_prot_op(scmd, prot_op);
635 scsi_set_prot_type(scmd, dif);
636 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
641 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
643 struct request_queue *q = sdkp->disk->queue;
644 unsigned int logical_block_size = sdkp->device->sector_size;
645 unsigned int max_blocks = 0;
647 q->limits.discard_zeroes_data = 0;
650 * When LBPRZ is reported, discard alignment and granularity
651 * must be fixed to the logical block size. Otherwise the block
652 * layer will drop misaligned portions of the request which can
653 * lead to data corruption. If LBPRZ is not set, we honor the
657 q->limits.discard_alignment = 0;
658 q->limits.discard_granularity = logical_block_size;
660 q->limits.discard_alignment = sdkp->unmap_alignment *
662 q->limits.discard_granularity =
663 max(sdkp->physical_block_size,
664 sdkp->unmap_granularity * logical_block_size);
667 sdkp->provisioning_mode = mode;
672 blk_queue_max_discard_sectors(q, 0);
673 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
677 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
678 (u32)SD_MAX_WS16_BLOCKS);
682 max_blocks = min_not_zero(sdkp->max_ws_blocks,
683 (u32)SD_MAX_WS16_BLOCKS);
684 q->limits.discard_zeroes_data = sdkp->lbprz;
688 max_blocks = min_not_zero(sdkp->max_ws_blocks,
689 (u32)SD_MAX_WS10_BLOCKS);
690 q->limits.discard_zeroes_data = sdkp->lbprz;
694 max_blocks = min_not_zero(sdkp->max_ws_blocks,
695 (u32)SD_MAX_WS10_BLOCKS);
696 q->limits.discard_zeroes_data = 1;
700 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
701 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
705 * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
706 * @sdp: scsi device to operate one
707 * @rq: Request to prepare
709 * Will issue either UNMAP or WRITE SAME(16) depending on preference
710 * indicated by target device.
712 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
714 struct request *rq = cmd->request;
715 struct scsi_device *sdp = cmd->device;
716 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
717 sector_t sector = blk_rq_pos(rq);
718 unsigned int nr_sectors = blk_rq_sectors(rq);
724 sector >>= ilog2(sdp->sector_size) - 9;
725 nr_sectors >>= ilog2(sdp->sector_size) - 9;
727 page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
729 return BLKPREP_DEFER;
731 switch (sdkp->provisioning_mode) {
733 buf = page_address(page);
736 cmd->cmnd[0] = UNMAP;
739 put_unaligned_be16(6 + 16, &buf[0]);
740 put_unaligned_be16(16, &buf[2]);
741 put_unaligned_be64(sector, &buf[8]);
742 put_unaligned_be32(nr_sectors, &buf[16]);
749 cmd->cmnd[0] = WRITE_SAME_16;
750 cmd->cmnd[1] = 0x8; /* UNMAP */
751 put_unaligned_be64(sector, &cmd->cmnd[2]);
752 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
754 len = sdkp->device->sector_size;
760 cmd->cmnd[0] = WRITE_SAME;
761 if (sdkp->provisioning_mode == SD_LBP_WS10)
762 cmd->cmnd[1] = 0x8; /* UNMAP */
763 put_unaligned_be32(sector, &cmd->cmnd[2]);
764 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
766 len = sdkp->device->sector_size;
770 ret = BLKPREP_INVALID;
774 rq->timeout = SD_TIMEOUT;
776 cmd->transfersize = len;
777 cmd->allowed = SD_MAX_RETRIES;
779 rq->special_vec.bv_page = page;
780 rq->special_vec.bv_offset = 0;
781 rq->special_vec.bv_len = len;
783 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
786 ret = scsi_init_io(cmd);
788 if (ret != BLKPREP_OK)
793 static void sd_config_write_same(struct scsi_disk *sdkp)
795 struct request_queue *q = sdkp->disk->queue;
796 unsigned int logical_block_size = sdkp->device->sector_size;
798 if (sdkp->device->no_write_same) {
799 sdkp->max_ws_blocks = 0;
803 /* Some devices can not handle block counts above 0xffff despite
804 * supporting WRITE SAME(16). Consequently we default to 64k
805 * blocks per I/O unless the device explicitly advertises a
808 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
809 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
810 (u32)SD_MAX_WS16_BLOCKS);
811 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
812 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
813 (u32)SD_MAX_WS10_BLOCKS);
815 sdkp->device->no_write_same = 1;
816 sdkp->max_ws_blocks = 0;
820 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
821 (logical_block_size >> 9));
825 * sd_setup_write_same_cmnd - write the same data to multiple blocks
826 * @cmd: command to prepare
828 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
829 * preference indicated by target device.
831 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
833 struct request *rq = cmd->request;
834 struct scsi_device *sdp = cmd->device;
835 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
836 struct bio *bio = rq->bio;
837 sector_t sector = blk_rq_pos(rq);
838 unsigned int nr_sectors = blk_rq_sectors(rq);
841 if (sdkp->device->no_write_same)
842 return BLKPREP_INVALID;
844 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
846 if (sd_is_zoned(sdkp)) {
847 ret = sd_zbc_setup_write_cmnd(cmd);
848 if (ret != BLKPREP_OK)
852 sector >>= ilog2(sdp->sector_size) - 9;
853 nr_sectors >>= ilog2(sdp->sector_size) - 9;
855 rq->timeout = SD_WRITE_SAME_TIMEOUT;
857 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
859 cmd->cmnd[0] = WRITE_SAME_16;
860 put_unaligned_be64(sector, &cmd->cmnd[2]);
861 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
864 cmd->cmnd[0] = WRITE_SAME;
865 put_unaligned_be32(sector, &cmd->cmnd[2]);
866 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
869 cmd->transfersize = sdp->sector_size;
870 cmd->allowed = SD_MAX_RETRIES;
871 return scsi_init_io(cmd);
874 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
876 struct request *rq = cmd->request;
878 /* flush requests don't perform I/O, zero the S/G table */
879 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
881 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
883 cmd->transfersize = 0;
884 cmd->allowed = SD_MAX_RETRIES;
886 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
890 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
892 struct request *rq = SCpnt->request;
893 struct scsi_device *sdp = SCpnt->device;
894 struct gendisk *disk = rq->rq_disk;
895 struct scsi_disk *sdkp = scsi_disk(disk);
896 sector_t block = blk_rq_pos(rq);
898 unsigned int this_count = blk_rq_sectors(rq);
899 unsigned int dif, dix;
900 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
902 unsigned char protect;
905 ret = sd_zbc_setup_write_cmnd(SCpnt);
906 if (ret != BLKPREP_OK)
910 ret = scsi_init_io(SCpnt);
911 if (ret != BLKPREP_OK)
915 /* from here on until we're complete, any goto out
916 * is used for a killable error condition */
920 scmd_printk(KERN_INFO, SCpnt,
921 "%s: block=%llu, count=%d\n",
922 __func__, (unsigned long long)block, this_count));
924 if (!sdp || !scsi_device_online(sdp) ||
925 block + blk_rq_sectors(rq) > get_capacity(disk)) {
926 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
927 "Finishing %u sectors\n",
928 blk_rq_sectors(rq)));
929 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
930 "Retry with 0x%p\n", SCpnt));
936 * quietly refuse to do anything to a changed disc until
937 * the changed bit has been reset
939 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
944 * Some SD card readers can't handle multi-sector accesses which touch
945 * the last one or two hardware sectors. Split accesses as needed.
947 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
948 (sdp->sector_size / 512);
950 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
951 if (block < threshold) {
952 /* Access up to the threshold but not beyond */
953 this_count = threshold - block;
955 /* Access only a single hardware sector */
956 this_count = sdp->sector_size / 512;
960 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
961 (unsigned long long)block));
964 * If we have a 1K hardware sectorsize, prevent access to single
965 * 512 byte sectors. In theory we could handle this - in fact
966 * the scsi cdrom driver must be able to handle this because
967 * we typically use 1K blocksizes, and cdroms typically have
968 * 2K hardware sectorsizes. Of course, things are simpler
969 * with the cdrom, since it is read-only. For performance
970 * reasons, the filesystems should be able to handle this
971 * and not force the scsi disk driver to use bounce buffers
974 if (sdp->sector_size == 1024) {
975 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
976 scmd_printk(KERN_ERR, SCpnt,
977 "Bad block number requested\n");
981 this_count = this_count >> 1;
984 if (sdp->sector_size == 2048) {
985 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
986 scmd_printk(KERN_ERR, SCpnt,
987 "Bad block number requested\n");
991 this_count = this_count >> 2;
994 if (sdp->sector_size == 4096) {
995 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
996 scmd_printk(KERN_ERR, SCpnt,
997 "Bad block number requested\n");
1001 this_count = this_count >> 3;
1004 if (rq_data_dir(rq) == WRITE) {
1005 SCpnt->cmnd[0] = WRITE_6;
1007 if (blk_integrity_rq(rq))
1008 sd_dif_prepare(SCpnt);
1010 } else if (rq_data_dir(rq) == READ) {
1011 SCpnt->cmnd[0] = READ_6;
1013 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1017 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1018 "%s %d/%u 512 byte blocks.\n",
1019 (rq_data_dir(rq) == WRITE) ?
1020 "writing" : "reading", this_count,
1021 blk_rq_sectors(rq)));
1023 dix = scsi_prot_sg_count(SCpnt);
1024 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1027 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1031 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1032 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1034 if (unlikely(SCpnt->cmnd == NULL)) {
1035 ret = BLKPREP_DEFER;
1039 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1040 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1041 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1042 SCpnt->cmnd[7] = 0x18;
1043 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1044 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1047 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1048 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1049 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1050 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1051 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1052 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1053 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1054 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1056 /* Expected Indirect LBA */
1057 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1058 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1059 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1060 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1062 /* Transfer length */
1063 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1064 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1065 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1066 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1067 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1068 SCpnt->cmnd[0] += READ_16 - READ_6;
1069 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1070 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1071 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1072 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1073 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1074 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1075 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1076 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1077 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1078 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1079 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1080 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1081 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1082 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1083 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1084 scsi_device_protection(SCpnt->device) ||
1085 SCpnt->device->use_10_for_rw) {
1086 SCpnt->cmnd[0] += READ_10 - READ_6;
1087 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1088 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1089 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1090 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1091 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1092 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1093 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1094 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1096 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1098 * This happens only if this drive failed
1099 * 10byte rw command with ILLEGAL_REQUEST
1100 * during operation and thus turned off
1103 scmd_printk(KERN_ERR, SCpnt,
1104 "FUA write on READ/WRITE(6) drive\n");
1108 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1109 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1110 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1111 SCpnt->cmnd[4] = (unsigned char) this_count;
1114 SCpnt->sdb.length = this_count * sdp->sector_size;
1117 * We shouldn't disconnect in the middle of a sector, so with a dumb
1118 * host adapter, it's safe to assume that we can at least transfer
1119 * this many bytes between each connect / disconnect.
1121 SCpnt->transfersize = sdp->sector_size;
1122 SCpnt->underflow = this_count << 9;
1123 SCpnt->allowed = SD_MAX_RETRIES;
1126 * This indicates that the command is ready from our end to be
1131 if (zoned_write && ret != BLKPREP_OK)
1132 sd_zbc_cancel_write_cmnd(SCpnt);
1137 static int sd_init_command(struct scsi_cmnd *cmd)
1139 struct request *rq = cmd->request;
1141 switch (req_op(rq)) {
1142 case REQ_OP_DISCARD:
1143 return sd_setup_discard_cmnd(cmd);
1144 case REQ_OP_WRITE_SAME:
1145 return sd_setup_write_same_cmnd(cmd);
1147 return sd_setup_flush_cmnd(cmd);
1150 return sd_setup_read_write_cmnd(cmd);
1151 case REQ_OP_ZONE_REPORT:
1152 return sd_zbc_setup_report_cmnd(cmd);
1153 case REQ_OP_ZONE_RESET:
1154 return sd_zbc_setup_reset_cmnd(cmd);
1160 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1162 struct request *rq = SCpnt->request;
1164 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1165 __free_page(rq->special_vec.bv_page);
1167 if (SCpnt->cmnd != rq->cmd) {
1168 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1175 * sd_open - open a scsi disk device
1176 * @inode: only i_rdev member may be used
1177 * @filp: only f_mode and f_flags may be used
1179 * Returns 0 if successful. Returns a negated errno value in case
1182 * Note: This can be called from a user context (e.g. fsck(1) )
1183 * or from within the kernel (e.g. as a result of a mount(1) ).
1184 * In the latter case @inode and @filp carry an abridged amount
1185 * of information as noted above.
1187 * Locking: called with bdev->bd_mutex held.
1189 static int sd_open(struct block_device *bdev, fmode_t mode)
1191 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1192 struct scsi_device *sdev;
1198 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1200 sdev = sdkp->device;
1203 * If the device is in error recovery, wait until it is done.
1204 * If the device is offline, then disallow any access to it.
1207 if (!scsi_block_when_processing_errors(sdev))
1210 if (sdev->removable || sdkp->write_prot)
1211 check_disk_change(bdev);
1214 * If the drive is empty, just let the open fail.
1216 retval = -ENOMEDIUM;
1217 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1221 * If the device has the write protect tab set, have the open fail
1222 * if the user expects to be able to write to the thing.
1225 if (sdkp->write_prot && (mode & FMODE_WRITE))
1229 * It is possible that the disk changing stuff resulted in
1230 * the device being taken offline. If this is the case,
1231 * report this to the user, and don't pretend that the
1232 * open actually succeeded.
1235 if (!scsi_device_online(sdev))
1238 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1239 if (scsi_block_when_processing_errors(sdev))
1240 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1246 scsi_disk_put(sdkp);
1251 * sd_release - invoked when the (last) close(2) is called on this
1253 * @inode: only i_rdev member may be used
1254 * @filp: only f_mode and f_flags may be used
1258 * Note: may block (uninterruptible) if error recovery is underway
1261 * Locking: called with bdev->bd_mutex held.
1263 static void sd_release(struct gendisk *disk, fmode_t mode)
1265 struct scsi_disk *sdkp = scsi_disk(disk);
1266 struct scsi_device *sdev = sdkp->device;
1268 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1270 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1271 if (scsi_block_when_processing_errors(sdev))
1272 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1276 * XXX and what if there are packets in flight and this close()
1277 * XXX is followed by a "rmmod sd_mod"?
1280 scsi_disk_put(sdkp);
1283 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1285 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1286 struct scsi_device *sdp = sdkp->device;
1287 struct Scsi_Host *host = sdp->host;
1288 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1291 /* default to most commonly used values */
1292 diskinfo[0] = 0x40; /* 1 << 6 */
1293 diskinfo[1] = 0x20; /* 1 << 5 */
1294 diskinfo[2] = capacity >> 11;
1296 /* override with calculated, extended default, or driver values */
1297 if (host->hostt->bios_param)
1298 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1300 scsicam_bios_param(bdev, capacity, diskinfo);
1302 geo->heads = diskinfo[0];
1303 geo->sectors = diskinfo[1];
1304 geo->cylinders = diskinfo[2];
1309 * sd_ioctl - process an ioctl
1310 * @inode: only i_rdev/i_bdev members may be used
1311 * @filp: only f_mode and f_flags may be used
1312 * @cmd: ioctl command number
1313 * @arg: this is third argument given to ioctl(2) system call.
1314 * Often contains a pointer.
1316 * Returns 0 if successful (some ioctls return positive numbers on
1317 * success as well). Returns a negated errno value in case of error.
1319 * Note: most ioctls are forward onto the block subsystem or further
1320 * down in the scsi subsystem.
1322 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1323 unsigned int cmd, unsigned long arg)
1325 struct gendisk *disk = bdev->bd_disk;
1326 struct scsi_disk *sdkp = scsi_disk(disk);
1327 struct scsi_device *sdp = sdkp->device;
1328 void __user *p = (void __user *)arg;
1331 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1332 "cmd=0x%x\n", disk->disk_name, cmd));
1334 error = scsi_verify_blk_ioctl(bdev, cmd);
1339 * If we are in the middle of error recovery, don't let anyone
1340 * else try and use this device. Also, if error recovery fails, it
1341 * may try and take the device offline, in which case all further
1342 * access to the device is prohibited.
1344 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1345 (mode & FMODE_NDELAY) != 0);
1350 * Send SCSI addressing ioctls directly to mid level, send other
1351 * ioctls to block level and then onto mid level if they can't be
1355 case SCSI_IOCTL_GET_IDLUN:
1356 case SCSI_IOCTL_GET_BUS_NUMBER:
1357 error = scsi_ioctl(sdp, cmd, p);
1360 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1361 if (error != -ENOTTY)
1363 error = scsi_ioctl(sdp, cmd, p);
1370 static void set_media_not_present(struct scsi_disk *sdkp)
1372 if (sdkp->media_present)
1373 sdkp->device->changed = 1;
1375 if (sdkp->device->removable) {
1376 sdkp->media_present = 0;
1381 static int media_not_present(struct scsi_disk *sdkp,
1382 struct scsi_sense_hdr *sshdr)
1384 if (!scsi_sense_valid(sshdr))
1387 /* not invoked for commands that could return deferred errors */
1388 switch (sshdr->sense_key) {
1389 case UNIT_ATTENTION:
1391 /* medium not present */
1392 if (sshdr->asc == 0x3A) {
1393 set_media_not_present(sdkp);
1401 * sd_check_events - check media events
1402 * @disk: kernel device descriptor
1403 * @clearing: disk events currently being cleared
1405 * Returns mask of DISK_EVENT_*.
1407 * Note: this function is invoked from the block subsystem.
1409 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1411 struct scsi_disk *sdkp = scsi_disk_get(disk);
1412 struct scsi_device *sdp;
1413 struct scsi_sense_hdr *sshdr = NULL;
1420 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1423 * If the device is offline, don't send any commands - just pretend as
1424 * if the command failed. If the device ever comes back online, we
1425 * can deal with it then. It is only because of unrecoverable errors
1426 * that we would ever take a device offline in the first place.
1428 if (!scsi_device_online(sdp)) {
1429 set_media_not_present(sdkp);
1434 * Using TEST_UNIT_READY enables differentiation between drive with
1435 * no cartridge loaded - NOT READY, drive with changed cartridge -
1436 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1438 * Drives that auto spin down. eg iomega jaz 1G, will be started
1439 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1440 * sd_revalidate() is called.
1444 if (scsi_block_when_processing_errors(sdp)) {
1445 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1446 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1450 /* failed to execute TUR, assume media not present */
1451 if (host_byte(retval)) {
1452 set_media_not_present(sdkp);
1456 if (media_not_present(sdkp, sshdr))
1460 * For removable scsi disk we have to recognise the presence
1461 * of a disk in the drive.
1463 if (!sdkp->media_present)
1465 sdkp->media_present = 1;
1468 * sdp->changed is set under the following conditions:
1470 * Medium present state has changed in either direction.
1471 * Device has indicated UNIT_ATTENTION.
1474 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1476 scsi_disk_put(sdkp);
1480 static int sd_sync_cache(struct scsi_disk *sdkp)
1483 struct scsi_device *sdp = sdkp->device;
1484 const int timeout = sdp->request_queue->rq_timeout
1485 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1486 struct scsi_sense_hdr sshdr;
1488 if (!scsi_device_online(sdp))
1491 for (retries = 3; retries > 0; --retries) {
1492 unsigned char cmd[10] = { 0 };
1494 cmd[0] = SYNCHRONIZE_CACHE;
1496 * Leave the rest of the command zero to indicate
1499 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1500 &sshdr, timeout, SD_MAX_RETRIES,
1507 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1509 if (driver_byte(res) & DRIVER_SENSE)
1510 sd_print_sense_hdr(sdkp, &sshdr);
1511 /* we need to evaluate the error return */
1512 if (scsi_sense_valid(&sshdr) &&
1513 (sshdr.asc == 0x3a || /* medium not present */
1514 sshdr.asc == 0x20)) /* invalid command */
1515 /* this is no error here */
1518 switch (host_byte(res)) {
1519 /* ignore errors due to racing a disconnection */
1520 case DID_BAD_TARGET:
1521 case DID_NO_CONNECT:
1523 /* signal the upper layer it might try again */
1527 case DID_SOFT_ERROR:
1536 static void sd_rescan(struct device *dev)
1538 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1540 revalidate_disk(sdkp->disk);
1544 #ifdef CONFIG_COMPAT
1546 * This gets directly called from VFS. When the ioctl
1547 * is not recognized we go back to the other translation paths.
1549 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1550 unsigned int cmd, unsigned long arg)
1552 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1555 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1556 (mode & FMODE_NDELAY) != 0);
1561 * Let the static ioctl translation table take care of it.
1563 if (!sdev->host->hostt->compat_ioctl)
1564 return -ENOIOCTLCMD;
1565 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1569 static char sd_pr_type(enum pr_type type)
1572 case PR_WRITE_EXCLUSIVE:
1574 case PR_EXCLUSIVE_ACCESS:
1576 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1578 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1580 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1582 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1589 static int sd_pr_command(struct block_device *bdev, u8 sa,
1590 u64 key, u64 sa_key, u8 type, u8 flags)
1592 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1593 struct scsi_sense_hdr sshdr;
1595 u8 cmd[16] = { 0, };
1596 u8 data[24] = { 0, };
1598 cmd[0] = PERSISTENT_RESERVE_OUT;
1601 put_unaligned_be32(sizeof(data), &cmd[5]);
1603 put_unaligned_be64(key, &data[0]);
1604 put_unaligned_be64(sa_key, &data[8]);
1607 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1608 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1610 if ((driver_byte(result) & DRIVER_SENSE) &&
1611 (scsi_sense_valid(&sshdr))) {
1612 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1613 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1619 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1622 if (flags & ~PR_FL_IGNORE_KEY)
1624 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1625 old_key, new_key, 0,
1626 (1 << 0) /* APTPL */);
1629 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1634 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1637 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1639 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1642 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1643 enum pr_type type, bool abort)
1645 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1646 sd_pr_type(type), 0);
1649 static int sd_pr_clear(struct block_device *bdev, u64 key)
1651 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1654 static const struct pr_ops sd_pr_ops = {
1655 .pr_register = sd_pr_register,
1656 .pr_reserve = sd_pr_reserve,
1657 .pr_release = sd_pr_release,
1658 .pr_preempt = sd_pr_preempt,
1659 .pr_clear = sd_pr_clear,
1662 static const struct block_device_operations sd_fops = {
1663 .owner = THIS_MODULE,
1665 .release = sd_release,
1667 .getgeo = sd_getgeo,
1668 #ifdef CONFIG_COMPAT
1669 .compat_ioctl = sd_compat_ioctl,
1671 .check_events = sd_check_events,
1672 .revalidate_disk = sd_revalidate_disk,
1673 .unlock_native_capacity = sd_unlock_native_capacity,
1674 .pr_ops = &sd_pr_ops,
1678 * sd_eh_action - error handling callback
1679 * @scmd: sd-issued command that has failed
1680 * @eh_disp: The recovery disposition suggested by the midlayer
1682 * This function is called by the SCSI midlayer upon completion of an
1683 * error test command (currently TEST UNIT READY). The result of sending
1684 * the eh command is passed in eh_disp. We're looking for devices that
1685 * fail medium access commands but are OK with non access commands like
1686 * test unit ready (so wrongly see the device as having a successful
1689 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1691 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1693 if (!scsi_device_online(scmd->device) ||
1694 !scsi_medium_access_command(scmd) ||
1695 host_byte(scmd->result) != DID_TIME_OUT ||
1700 * The device has timed out executing a medium access command.
1701 * However, the TEST UNIT READY command sent during error
1702 * handling completed successfully. Either the device is in the
1703 * process of recovering or has it suffered an internal failure
1704 * that prevents access to the storage medium.
1706 sdkp->medium_access_timed_out++;
1709 * If the device keeps failing read/write commands but TEST UNIT
1710 * READY always completes successfully we assume that medium
1711 * access is no longer possible and take the device offline.
1713 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1714 scmd_printk(KERN_ERR, scmd,
1715 "Medium access timeout failure. Offlining disk!\n");
1716 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1724 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1726 u64 start_lba = blk_rq_pos(scmd->request);
1727 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1728 u64 factor = scmd->device->sector_size / 512;
1732 * resid is optional but mostly filled in. When it's unused,
1733 * its value is zero, so we assume the whole buffer transferred
1735 unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1736 unsigned int good_bytes;
1738 if (scmd->request->cmd_type != REQ_TYPE_FS)
1741 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1742 SCSI_SENSE_BUFFERSIZE,
1747 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1750 /* be careful ... don't want any overflows */
1751 do_div(start_lba, factor);
1752 do_div(end_lba, factor);
1754 /* The bad lba was reported incorrectly, we have no idea where
1757 if (bad_lba < start_lba || bad_lba >= end_lba)
1760 /* This computation should always be done in terms of
1761 * the resolution of the device's medium.
1763 good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1764 return min(good_bytes, transferred);
1768 * sd_done - bottom half handler: called when the lower level
1769 * driver has completed (successfully or otherwise) a scsi command.
1770 * @SCpnt: mid-level's per command structure.
1772 * Note: potentially run from within an ISR. Must not block.
1774 static int sd_done(struct scsi_cmnd *SCpnt)
1776 int result = SCpnt->result;
1777 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1778 struct scsi_sense_hdr sshdr;
1779 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1780 struct request *req = SCpnt->request;
1781 int sense_valid = 0;
1782 int sense_deferred = 0;
1783 unsigned char op = SCpnt->cmnd[0];
1784 unsigned char unmap = SCpnt->cmnd[1] & 8;
1786 switch (req_op(req)) {
1787 case REQ_OP_DISCARD:
1788 case REQ_OP_WRITE_SAME:
1789 case REQ_OP_ZONE_RESET:
1791 good_bytes = blk_rq_bytes(req);
1792 scsi_set_resid(SCpnt, 0);
1795 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1798 case REQ_OP_ZONE_REPORT:
1800 good_bytes = scsi_bufflen(SCpnt)
1801 - scsi_get_resid(SCpnt);
1802 scsi_set_resid(SCpnt, 0);
1805 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1811 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1813 sense_deferred = scsi_sense_is_deferred(&sshdr);
1815 sdkp->medium_access_timed_out = 0;
1817 if (driver_byte(result) != DRIVER_SENSE &&
1818 (!sense_valid || sense_deferred))
1821 switch (sshdr.sense_key) {
1822 case HARDWARE_ERROR:
1824 good_bytes = sd_completed_bytes(SCpnt);
1826 case RECOVERED_ERROR:
1827 good_bytes = scsi_bufflen(SCpnt);
1830 /* This indicates a false check condition, so ignore it. An
1831 * unknown amount of data was transferred so treat it as an
1835 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1837 case ABORTED_COMMAND:
1838 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1839 good_bytes = sd_completed_bytes(SCpnt);
1841 case ILLEGAL_REQUEST:
1842 if (sshdr.asc == 0x10) /* DIX: Host detected corruption */
1843 good_bytes = sd_completed_bytes(SCpnt);
1844 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1845 if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1848 sd_config_discard(sdkp, SD_LBP_DISABLE);
1853 sd_config_discard(sdkp, SD_LBP_DISABLE);
1855 sdkp->device->no_write_same = 1;
1856 sd_config_write_same(sdkp);
1859 req->__data_len = blk_rq_bytes(req);
1860 req->rq_flags |= RQF_QUIET;
1870 if (sd_is_zoned(sdkp))
1871 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
1873 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1874 "sd_done: completed %d of %d bytes\n",
1875 good_bytes, scsi_bufflen(SCpnt)));
1877 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1878 sd_dif_complete(SCpnt, good_bytes);
1884 * spinup disk - called only in sd_revalidate_disk()
1887 sd_spinup_disk(struct scsi_disk *sdkp)
1889 unsigned char cmd[10];
1890 unsigned long spintime_expire = 0;
1891 int retries, spintime;
1892 unsigned int the_result;
1893 struct scsi_sense_hdr sshdr;
1894 int sense_valid = 0;
1898 /* Spin up drives, as required. Only do this at boot time */
1899 /* Spinup needs to be done for module loads too. */
1904 cmd[0] = TEST_UNIT_READY;
1905 memset((void *) &cmd[1], 0, 9);
1907 the_result = scsi_execute_req(sdkp->device, cmd,
1910 SD_MAX_RETRIES, NULL);
1913 * If the drive has indicated to us that it
1914 * doesn't have any media in it, don't bother
1915 * with any more polling.
1917 if (media_not_present(sdkp, &sshdr))
1921 sense_valid = scsi_sense_valid(&sshdr);
1923 } while (retries < 3 &&
1924 (!scsi_status_is_good(the_result) ||
1925 ((driver_byte(the_result) & DRIVER_SENSE) &&
1926 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1928 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1929 /* no sense, TUR either succeeded or failed
1930 * with a status error */
1931 if(!spintime && !scsi_status_is_good(the_result)) {
1932 sd_print_result(sdkp, "Test Unit Ready failed",
1939 * The device does not want the automatic start to be issued.
1941 if (sdkp->device->no_start_on_add)
1944 if (sense_valid && sshdr.sense_key == NOT_READY) {
1945 if (sshdr.asc == 4 && sshdr.ascq == 3)
1946 break; /* manual intervention required */
1947 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1948 break; /* standby */
1949 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1950 break; /* unavailable */
1952 * Issue command to spin up drive when not ready
1955 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1956 cmd[0] = START_STOP;
1957 cmd[1] = 1; /* Return immediately */
1958 memset((void *) &cmd[2], 0, 8);
1959 cmd[4] = 1; /* Start spin cycle */
1960 if (sdkp->device->start_stop_pwr_cond)
1962 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1964 SD_TIMEOUT, SD_MAX_RETRIES,
1966 spintime_expire = jiffies + 100 * HZ;
1969 /* Wait 1 second for next try */
1974 * Wait for USB flash devices with slow firmware.
1975 * Yes, this sense key/ASC combination shouldn't
1976 * occur here. It's characteristic of these devices.
1978 } else if (sense_valid &&
1979 sshdr.sense_key == UNIT_ATTENTION &&
1980 sshdr.asc == 0x28) {
1982 spintime_expire = jiffies + 5 * HZ;
1985 /* Wait 1 second for next try */
1988 /* we don't understand the sense code, so it's
1989 * probably pointless to loop */
1991 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1992 sd_print_sense_hdr(sdkp, &sshdr);
1997 } while (spintime && time_before_eq(jiffies, spintime_expire));
2000 if (scsi_status_is_good(the_result))
2003 printk("not responding...\n");
2008 * Determine whether disk supports Data Integrity Field.
2010 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2012 struct scsi_device *sdp = sdkp->device;
2016 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2019 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2021 if (type > T10_PI_TYPE3_PROTECTION)
2023 else if (scsi_host_dif_capable(sdp->host, type))
2026 if (sdkp->first_scan || type != sdkp->protection_type)
2029 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2030 " protection type %u. Disabling disk!\n",
2034 sd_printk(KERN_NOTICE, sdkp,
2035 "Enabling DIF Type %u protection\n", type);
2038 sd_printk(KERN_NOTICE, sdkp,
2039 "Disabling DIF Type %u protection\n", type);
2043 sdkp->protection_type = type;
2048 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2049 struct scsi_sense_hdr *sshdr, int sense_valid,
2052 if (driver_byte(the_result) & DRIVER_SENSE)
2053 sd_print_sense_hdr(sdkp, sshdr);
2055 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2058 * Set dirty bit for removable devices if not ready -
2059 * sometimes drives will not report this properly.
2061 if (sdp->removable &&
2062 sense_valid && sshdr->sense_key == NOT_READY)
2063 set_media_not_present(sdkp);
2066 * We used to set media_present to 0 here to indicate no media
2067 * in the drive, but some drives fail read capacity even with
2068 * media present, so we can't do that.
2070 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2074 #if RC16_LEN > SD_BUF_SIZE
2075 #error RC16_LEN must not be more than SD_BUF_SIZE
2078 #define READ_CAPACITY_RETRIES_ON_RESET 10
2080 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2081 unsigned char *buffer)
2083 unsigned char cmd[16];
2084 struct scsi_sense_hdr sshdr;
2085 int sense_valid = 0;
2087 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2088 unsigned int alignment;
2089 unsigned long long lba;
2090 unsigned sector_size;
2092 if (sdp->no_read_capacity_16)
2097 cmd[0] = SERVICE_ACTION_IN_16;
2098 cmd[1] = SAI_READ_CAPACITY_16;
2100 memset(buffer, 0, RC16_LEN);
2102 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2103 buffer, RC16_LEN, &sshdr,
2104 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2106 if (media_not_present(sdkp, &sshdr))
2110 sense_valid = scsi_sense_valid(&sshdr);
2112 sshdr.sense_key == ILLEGAL_REQUEST &&
2113 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2115 /* Invalid Command Operation Code or
2116 * Invalid Field in CDB, just retry
2117 * silently with RC10 */
2120 sshdr.sense_key == UNIT_ATTENTION &&
2121 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2122 /* Device reset might occur several times,
2123 * give it one more chance */
2124 if (--reset_retries > 0)
2129 } while (the_result && retries);
2132 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2133 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2137 sector_size = get_unaligned_be32(&buffer[8]);
2138 lba = get_unaligned_be64(&buffer[0]);
2140 if (sd_read_protection_type(sdkp, buffer) < 0) {
2145 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2146 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2147 "kernel compiled with support for large block "
2153 /* Logical blocks per physical block exponent */
2154 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2157 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2159 /* Lowest aligned logical block */
2160 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2161 blk_queue_alignment_offset(sdp->request_queue, alignment);
2162 if (alignment && sdkp->first_scan)
2163 sd_printk(KERN_NOTICE, sdkp,
2164 "physical block alignment offset: %u\n", alignment);
2166 if (buffer[14] & 0x80) { /* LBPME */
2169 if (buffer[14] & 0x40) /* LBPRZ */
2172 sd_config_discard(sdkp, SD_LBP_WS16);
2175 sdkp->capacity = lba + 1;
2179 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2180 unsigned char *buffer)
2182 unsigned char cmd[16];
2183 struct scsi_sense_hdr sshdr;
2184 int sense_valid = 0;
2186 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2188 unsigned sector_size;
2191 cmd[0] = READ_CAPACITY;
2192 memset(&cmd[1], 0, 9);
2193 memset(buffer, 0, 8);
2195 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2197 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2199 if (media_not_present(sdkp, &sshdr))
2203 sense_valid = scsi_sense_valid(&sshdr);
2205 sshdr.sense_key == UNIT_ATTENTION &&
2206 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2207 /* Device reset might occur several times,
2208 * give it one more chance */
2209 if (--reset_retries > 0)
2214 } while (the_result && retries);
2217 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2218 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2222 sector_size = get_unaligned_be32(&buffer[4]);
2223 lba = get_unaligned_be32(&buffer[0]);
2225 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2226 /* Some buggy (usb cardreader) devices return an lba of
2227 0xffffffff when the want to report a size of 0 (with
2228 which they really mean no media is present) */
2230 sdkp->physical_block_size = sector_size;
2234 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2235 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2236 "kernel compiled with support for large block "
2242 sdkp->capacity = lba + 1;
2243 sdkp->physical_block_size = sector_size;
2247 static int sd_try_rc16_first(struct scsi_device *sdp)
2249 if (sdp->host->max_cmd_len < 16)
2251 if (sdp->try_rc_10_first)
2253 if (sdp->scsi_level > SCSI_SPC_2)
2255 if (scsi_device_protection(sdp))
2261 * read disk capacity
2264 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2267 struct scsi_device *sdp = sdkp->device;
2269 if (sd_try_rc16_first(sdp)) {
2270 sector_size = read_capacity_16(sdkp, sdp, buffer);
2271 if (sector_size == -EOVERFLOW)
2273 if (sector_size == -ENODEV)
2275 if (sector_size < 0)
2276 sector_size = read_capacity_10(sdkp, sdp, buffer);
2277 if (sector_size < 0)
2280 sector_size = read_capacity_10(sdkp, sdp, buffer);
2281 if (sector_size == -EOVERFLOW)
2283 if (sector_size < 0)
2285 if ((sizeof(sdkp->capacity) > 4) &&
2286 (sdkp->capacity > 0xffffffffULL)) {
2287 int old_sector_size = sector_size;
2288 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2289 "Trying to use READ CAPACITY(16).\n");
2290 sector_size = read_capacity_16(sdkp, sdp, buffer);
2291 if (sector_size < 0) {
2292 sd_printk(KERN_NOTICE, sdkp,
2293 "Using 0xffffffff as device size\n");
2294 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2295 sector_size = old_sector_size;
2301 /* Some devices are known to return the total number of blocks,
2302 * not the highest block number. Some devices have versions
2303 * which do this and others which do not. Some devices we might
2304 * suspect of doing this but we don't know for certain.
2306 * If we know the reported capacity is wrong, decrement it. If
2307 * we can only guess, then assume the number of blocks is even
2308 * (usually true but not always) and err on the side of lowering
2311 if (sdp->fix_capacity ||
2312 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2313 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2314 "from its reported value: %llu\n",
2315 (unsigned long long) sdkp->capacity);
2320 if (sector_size == 0) {
2322 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2326 if (sector_size != 512 &&
2327 sector_size != 1024 &&
2328 sector_size != 2048 &&
2329 sector_size != 4096) {
2330 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2333 * The user might want to re-format the drive with
2334 * a supported sectorsize. Once this happens, it
2335 * would be relatively trivial to set the thing up.
2336 * For this reason, we leave the thing in the table.
2340 * set a bogus sector size so the normal read/write
2341 * logic in the block layer will eventually refuse any
2342 * request on this device without tripping over power
2343 * of two sector size assumptions
2347 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2348 blk_queue_physical_block_size(sdp->request_queue,
2349 sdkp->physical_block_size);
2350 sdkp->device->sector_size = sector_size;
2352 if (sdkp->capacity > 0xffffffff)
2353 sdp->use_16_for_rw = 1;
2358 * Print disk capacity
2361 sd_print_capacity(struct scsi_disk *sdkp,
2362 sector_t old_capacity)
2364 int sector_size = sdkp->device->sector_size;
2365 char cap_str_2[10], cap_str_10[10];
2367 string_get_size(sdkp->capacity, sector_size,
2368 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2369 string_get_size(sdkp->capacity, sector_size,
2370 STRING_UNITS_10, cap_str_10,
2371 sizeof(cap_str_10));
2373 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2374 sd_printk(KERN_NOTICE, sdkp,
2375 "%llu %d-byte logical blocks: (%s/%s)\n",
2376 (unsigned long long)sdkp->capacity,
2377 sector_size, cap_str_10, cap_str_2);
2379 if (sdkp->physical_block_size != sector_size)
2380 sd_printk(KERN_NOTICE, sdkp,
2381 "%u-byte physical blocks\n",
2382 sdkp->physical_block_size);
2384 sd_zbc_print_zones(sdkp);
2388 /* called with buffer of length 512 */
2390 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2391 unsigned char *buffer, int len, struct scsi_mode_data *data,
2392 struct scsi_sense_hdr *sshdr)
2394 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2395 SD_TIMEOUT, SD_MAX_RETRIES, data,
2400 * read write protect setting, if possible - called only in sd_revalidate_disk()
2401 * called with buffer of length SD_BUF_SIZE
2404 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2407 struct scsi_device *sdp = sdkp->device;
2408 struct scsi_mode_data data;
2409 int old_wp = sdkp->write_prot;
2411 set_disk_ro(sdkp->disk, 0);
2412 if (sdp->skip_ms_page_3f) {
2413 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2417 if (sdp->use_192_bytes_for_3f) {
2418 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2421 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2422 * We have to start carefully: some devices hang if we ask
2423 * for more than is available.
2425 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2428 * Second attempt: ask for page 0 When only page 0 is
2429 * implemented, a request for page 3F may return Sense Key
2430 * 5: Illegal Request, Sense Code 24: Invalid field in
2433 if (!scsi_status_is_good(res))
2434 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2437 * Third attempt: ask 255 bytes, as we did earlier.
2439 if (!scsi_status_is_good(res))
2440 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2444 if (!scsi_status_is_good(res)) {
2445 sd_first_printk(KERN_WARNING, sdkp,
2446 "Test WP failed, assume Write Enabled\n");
2448 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2449 set_disk_ro(sdkp->disk, sdkp->write_prot);
2450 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2451 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2452 sdkp->write_prot ? "on" : "off");
2453 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2459 * sd_read_cache_type - called only from sd_revalidate_disk()
2460 * called with buffer of length SD_BUF_SIZE
2463 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2466 struct scsi_device *sdp = sdkp->device;
2471 struct scsi_mode_data data;
2472 struct scsi_sense_hdr sshdr;
2473 int old_wce = sdkp->WCE;
2474 int old_rcd = sdkp->RCD;
2475 int old_dpofua = sdkp->DPOFUA;
2478 if (sdkp->cache_override)
2482 if (sdp->skip_ms_page_8) {
2483 if (sdp->type == TYPE_RBC)
2486 if (sdp->skip_ms_page_3f)
2489 if (sdp->use_192_bytes_for_3f)
2493 } else if (sdp->type == TYPE_RBC) {
2501 /* cautiously ask */
2502 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2505 if (!scsi_status_is_good(res))
2508 if (!data.header_length) {
2511 sd_first_printk(KERN_ERR, sdkp,
2512 "Missing header in MODE_SENSE response\n");
2515 /* that went OK, now ask for the proper length */
2519 * We're only interested in the first three bytes, actually.
2520 * But the data cache page is defined for the first 20.
2524 else if (len > SD_BUF_SIZE) {
2525 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2526 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2529 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2533 if (len > first_len)
2534 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2537 if (scsi_status_is_good(res)) {
2538 int offset = data.header_length + data.block_descriptor_length;
2540 while (offset < len) {
2541 u8 page_code = buffer[offset] & 0x3F;
2542 u8 spf = buffer[offset] & 0x40;
2544 if (page_code == 8 || page_code == 6) {
2545 /* We're interested only in the first 3 bytes.
2547 if (len - offset <= 2) {
2548 sd_first_printk(KERN_ERR, sdkp,
2549 "Incomplete mode parameter "
2553 modepage = page_code;
2557 /* Go to the next page */
2558 if (spf && len - offset > 3)
2559 offset += 4 + (buffer[offset+2] << 8) +
2561 else if (!spf && len - offset > 1)
2562 offset += 2 + buffer[offset+1];
2564 sd_first_printk(KERN_ERR, sdkp,
2566 "parameter data\n");
2572 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2576 if (modepage == 8) {
2577 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2578 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2580 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2584 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2585 if (sdp->broken_fua) {
2586 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2588 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2589 !sdkp->device->use_16_for_rw) {
2590 sd_first_printk(KERN_NOTICE, sdkp,
2591 "Uses READ/WRITE(6), disabling FUA\n");
2595 /* No cache flush allowed for write protected devices */
2596 if (sdkp->WCE && sdkp->write_prot)
2599 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2600 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2601 sd_printk(KERN_NOTICE, sdkp,
2602 "Write cache: %s, read cache: %s, %s\n",
2603 sdkp->WCE ? "enabled" : "disabled",
2604 sdkp->RCD ? "disabled" : "enabled",
2605 sdkp->DPOFUA ? "supports DPO and FUA"
2606 : "doesn't support DPO or FUA");
2612 if (scsi_sense_valid(&sshdr) &&
2613 sshdr.sense_key == ILLEGAL_REQUEST &&
2614 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2615 /* Invalid field in CDB */
2616 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2618 sd_first_printk(KERN_ERR, sdkp,
2619 "Asking for cache data failed\n");
2622 if (sdp->wce_default_on) {
2623 sd_first_printk(KERN_NOTICE, sdkp,
2624 "Assuming drive cache: write back\n");
2627 sd_first_printk(KERN_ERR, sdkp,
2628 "Assuming drive cache: write through\n");
2636 * The ATO bit indicates whether the DIF application tag is available
2637 * for use by the operating system.
2639 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2642 struct scsi_device *sdp = sdkp->device;
2643 struct scsi_mode_data data;
2644 struct scsi_sense_hdr sshdr;
2646 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2649 if (sdkp->protection_type == 0)
2652 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2653 SD_MAX_RETRIES, &data, &sshdr);
2655 if (!scsi_status_is_good(res) || !data.header_length ||
2657 sd_first_printk(KERN_WARNING, sdkp,
2658 "getting Control mode page failed, assume no ATO\n");
2660 if (scsi_sense_valid(&sshdr))
2661 sd_print_sense_hdr(sdkp, &sshdr);
2666 offset = data.header_length + data.block_descriptor_length;
2668 if ((buffer[offset] & 0x3f) != 0x0a) {
2669 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2673 if ((buffer[offset + 5] & 0x80) == 0)
2682 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2683 * @disk: disk to query
2685 static void sd_read_block_limits(struct scsi_disk *sdkp)
2687 unsigned int sector_sz = sdkp->device->sector_size;
2688 const int vpd_len = 64;
2689 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2692 /* Block Limits VPD */
2693 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2696 blk_queue_io_min(sdkp->disk->queue,
2697 get_unaligned_be16(&buffer[6]) * sector_sz);
2699 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2700 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2702 if (buffer[3] == 0x3c) {
2703 unsigned int lba_count, desc_count;
2705 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2710 lba_count = get_unaligned_be32(&buffer[20]);
2711 desc_count = get_unaligned_be32(&buffer[24]);
2713 if (lba_count && desc_count)
2714 sdkp->max_unmap_blocks = lba_count;
2716 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2718 if (buffer[32] & 0x80)
2719 sdkp->unmap_alignment =
2720 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2722 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2724 if (sdkp->max_unmap_blocks)
2725 sd_config_discard(sdkp, SD_LBP_UNMAP);
2727 sd_config_discard(sdkp, SD_LBP_WS16);
2729 } else { /* LBP VPD page tells us what to use */
2730 if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2731 sd_config_discard(sdkp, SD_LBP_UNMAP);
2732 else if (sdkp->lbpws)
2733 sd_config_discard(sdkp, SD_LBP_WS16);
2734 else if (sdkp->lbpws10)
2735 sd_config_discard(sdkp, SD_LBP_WS10);
2736 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2737 sd_config_discard(sdkp, SD_LBP_UNMAP);
2739 sd_config_discard(sdkp, SD_LBP_DISABLE);
2748 * sd_read_block_characteristics - Query block dev. characteristics
2749 * @disk: disk to query
2751 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2753 struct request_queue *q = sdkp->disk->queue;
2754 unsigned char *buffer;
2756 const int vpd_len = 64;
2758 buffer = kmalloc(vpd_len, GFP_KERNEL);
2761 /* Block Device Characteristics VPD */
2762 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2765 rot = get_unaligned_be16(&buffer[4]);
2768 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2769 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2772 if (sdkp->device->type == TYPE_ZBC) {
2774 q->limits.zoned = BLK_ZONED_HM;
2776 sdkp->zoned = (buffer[8] >> 4) & 3;
2777 if (sdkp->zoned == 1)
2779 q->limits.zoned = BLK_ZONED_HA;
2782 * Treat drive-managed devices as
2783 * regular block devices.
2785 q->limits.zoned = BLK_ZONED_NONE;
2787 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2788 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2789 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2796 * sd_read_block_provisioning - Query provisioning VPD page
2797 * @disk: disk to query
2799 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2801 unsigned char *buffer;
2802 const int vpd_len = 8;
2804 if (sdkp->lbpme == 0)
2807 buffer = kmalloc(vpd_len, GFP_KERNEL);
2809 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2813 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2814 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2815 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2821 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2823 struct scsi_device *sdev = sdkp->device;
2825 if (sdev->host->no_write_same) {
2826 sdev->no_write_same = 1;
2831 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2832 /* too large values might cause issues with arcmsr */
2833 int vpd_buf_len = 64;
2835 sdev->no_report_opcodes = 1;
2837 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2838 * CODES is unsupported and the device has an ATA
2839 * Information VPD page (SAT).
2841 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2842 sdev->no_write_same = 1;
2845 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2848 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2853 * sd_revalidate_disk - called the first time a new disk is seen,
2854 * performs disk spin up, read_capacity, etc.
2855 * @disk: struct gendisk we care about
2857 static int sd_revalidate_disk(struct gendisk *disk)
2859 struct scsi_disk *sdkp = scsi_disk(disk);
2860 struct scsi_device *sdp = sdkp->device;
2861 struct request_queue *q = sdkp->disk->queue;
2862 sector_t old_capacity = sdkp->capacity;
2863 unsigned char *buffer;
2864 unsigned int dev_max, rw_max;
2866 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2867 "sd_revalidate_disk\n"));
2870 * If the device is offline, don't try and read capacity or any
2871 * of the other niceties.
2873 if (!scsi_device_online(sdp))
2876 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2878 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2879 "allocation failure.\n");
2883 sd_spinup_disk(sdkp);
2886 * Without media there is no reason to ask; moreover, some devices
2887 * react badly if we do.
2889 if (sdkp->media_present) {
2890 sd_read_capacity(sdkp, buffer);
2892 if (scsi_device_supports_vpd(sdp)) {
2893 sd_read_block_provisioning(sdkp);
2894 sd_read_block_limits(sdkp);
2895 sd_read_block_characteristics(sdkp);
2896 sd_zbc_read_zones(sdkp, buffer);
2899 sd_print_capacity(sdkp, old_capacity);
2901 sd_read_write_protect_flag(sdkp, buffer);
2902 sd_read_cache_type(sdkp, buffer);
2903 sd_read_app_tag_own(sdkp, buffer);
2904 sd_read_write_same(sdkp, buffer);
2907 sdkp->first_scan = 0;
2910 * We now have all cache related info, determine how we deal
2911 * with flush requests.
2913 sd_set_flush_flag(sdkp);
2915 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2916 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2918 /* Some devices report a maximum block count for READ/WRITE requests. */
2919 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2920 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2923 * Use the device's preferred I/O size for reads and writes
2924 * unless the reported value is unreasonably small, large, or
2927 if (sdkp->opt_xfer_blocks &&
2928 sdkp->opt_xfer_blocks <= dev_max &&
2929 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2930 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2931 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2932 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2934 rw_max = BLK_DEF_MAX_SECTORS;
2936 /* Combine with controller limits */
2937 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2939 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2940 sd_config_write_same(sdkp);
2948 * sd_unlock_native_capacity - unlock native capacity
2949 * @disk: struct gendisk to set capacity for
2951 * Block layer calls this function if it detects that partitions
2952 * on @disk reach beyond the end of the device. If the SCSI host
2953 * implements ->unlock_native_capacity() method, it's invoked to
2954 * give it a chance to adjust the device capacity.
2957 * Defined by block layer. Might sleep.
2959 static void sd_unlock_native_capacity(struct gendisk *disk)
2961 struct scsi_device *sdev = scsi_disk(disk)->device;
2963 if (sdev->host->hostt->unlock_native_capacity)
2964 sdev->host->hostt->unlock_native_capacity(sdev);
2968 * sd_format_disk_name - format disk name
2969 * @prefix: name prefix - ie. "sd" for SCSI disks
2970 * @index: index of the disk to format name for
2971 * @buf: output buffer
2972 * @buflen: length of the output buffer
2974 * SCSI disk names starts at sda. The 26th device is sdz and the
2975 * 27th is sdaa. The last one for two lettered suffix is sdzz
2976 * which is followed by sdaaa.
2978 * This is basically 26 base counting with one extra 'nil' entry
2979 * at the beginning from the second digit on and can be
2980 * determined using similar method as 26 base conversion with the
2981 * index shifted -1 after each digit is computed.
2987 * 0 on success, -errno on failure.
2989 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2991 const int base = 'z' - 'a' + 1;
2992 char *begin = buf + strlen(prefix);
2993 char *end = buf + buflen;
3003 *--p = 'a' + (index % unit);
3004 index = (index / unit) - 1;
3005 } while (index >= 0);
3007 memmove(begin, p, end - p);
3008 memcpy(buf, prefix, strlen(prefix));
3014 * The asynchronous part of sd_probe
3016 static void sd_probe_async(void *data, async_cookie_t cookie)
3018 struct scsi_disk *sdkp = data;
3019 struct scsi_device *sdp;
3026 index = sdkp->index;
3027 dev = &sdp->sdev_gendev;
3029 gd->major = sd_major((index & 0xf0) >> 4);
3030 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3031 gd->minors = SD_MINORS;
3033 gd->fops = &sd_fops;
3034 gd->private_data = &sdkp->driver;
3035 gd->queue = sdkp->device->request_queue;
3037 /* defaults, until the device tells us otherwise */
3038 sdp->sector_size = 512;
3040 sdkp->media_present = 1;
3041 sdkp->write_prot = 0;
3042 sdkp->cache_override = 0;
3046 sdkp->first_scan = 1;
3047 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3049 sd_revalidate_disk(gd);
3051 gd->flags = GENHD_FL_EXT_DEVT;
3052 if (sdp->removable) {
3053 gd->flags |= GENHD_FL_REMOVABLE;
3054 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3057 blk_pm_runtime_init(sdp->request_queue, dev);
3058 device_add_disk(dev, gd);
3060 sd_dif_config_host(sdkp);
3062 sd_revalidate_disk(gd);
3064 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3065 sdp->removable ? "removable " : "");
3066 scsi_autopm_put_device(sdp);
3067 put_device(&sdkp->dev);
3071 * sd_probe - called during driver initialization and whenever a
3072 * new scsi device is attached to the system. It is called once
3073 * for each scsi device (not just disks) present.
3074 * @dev: pointer to device object
3076 * Returns 0 if successful (or not interested in this scsi device
3077 * (e.g. scanner)); 1 when there is an error.
3079 * Note: this function is invoked from the scsi mid-level.
3080 * This function sets up the mapping between a given
3081 * <host,channel,id,lun> (found in sdp) and new device name
3082 * (e.g. /dev/sda). More precisely it is the block device major
3083 * and minor number that is chosen here.
3085 * Assume sd_probe is not re-entrant (for time being)
3086 * Also think about sd_probe() and sd_remove() running coincidentally.
3088 static int sd_probe(struct device *dev)
3090 struct scsi_device *sdp = to_scsi_device(dev);
3091 struct scsi_disk *sdkp;
3096 scsi_autopm_get_device(sdp);
3098 if (sdp->type != TYPE_DISK &&
3099 sdp->type != TYPE_ZBC &&
3100 sdp->type != TYPE_MOD &&
3101 sdp->type != TYPE_RBC)
3104 #ifndef CONFIG_BLK_DEV_ZONED
3105 if (sdp->type == TYPE_ZBC)
3108 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3112 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3116 gd = alloc_disk(SD_MINORS);
3121 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3124 spin_lock(&sd_index_lock);
3125 error = ida_get_new(&sd_index_ida, &index);
3126 spin_unlock(&sd_index_lock);
3127 } while (error == -EAGAIN);
3130 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3134 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3136 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3137 goto out_free_index;
3141 sdkp->driver = &sd_template;
3143 sdkp->index = index;
3144 atomic_set(&sdkp->openers, 0);
3145 atomic_set(&sdkp->device->ioerr_cnt, 0);
3147 if (!sdp->request_queue->rq_timeout) {
3148 if (sdp->type != TYPE_MOD)
3149 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3151 blk_queue_rq_timeout(sdp->request_queue,
3155 device_initialize(&sdkp->dev);
3156 sdkp->dev.parent = dev;
3157 sdkp->dev.class = &sd_disk_class;
3158 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3160 error = device_add(&sdkp->dev);
3162 goto out_free_index;
3165 dev_set_drvdata(dev, sdkp);
3167 get_device(&sdkp->dev); /* prevent release before async_schedule */
3168 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3173 spin_lock(&sd_index_lock);
3174 ida_remove(&sd_index_ida, index);
3175 spin_unlock(&sd_index_lock);
3181 scsi_autopm_put_device(sdp);
3186 * sd_remove - called whenever a scsi disk (previously recognized by
3187 * sd_probe) is detached from the system. It is called (potentially
3188 * multiple times) during sd module unload.
3189 * @sdp: pointer to mid level scsi device object
3191 * Note: this function is invoked from the scsi mid-level.
3192 * This function potentially frees up a device name (e.g. /dev/sdc)
3193 * that could be re-used by a subsequent sd_probe().
3194 * This function is not called when the built-in sd driver is "exit-ed".
3196 static int sd_remove(struct device *dev)
3198 struct scsi_disk *sdkp;
3201 sdkp = dev_get_drvdata(dev);
3202 devt = disk_devt(sdkp->disk);
3203 scsi_autopm_get_device(sdkp->device);
3205 async_synchronize_full_domain(&scsi_sd_pm_domain);
3206 async_synchronize_full_domain(&scsi_sd_probe_domain);
3207 device_del(&sdkp->dev);
3208 del_gendisk(sdkp->disk);
3211 sd_zbc_remove(sdkp);
3213 blk_register_region(devt, SD_MINORS, NULL,
3214 sd_default_probe, NULL, NULL);
3216 mutex_lock(&sd_ref_mutex);
3217 dev_set_drvdata(dev, NULL);
3218 put_device(&sdkp->dev);
3219 mutex_unlock(&sd_ref_mutex);
3225 * scsi_disk_release - Called to free the scsi_disk structure
3226 * @dev: pointer to embedded class device
3228 * sd_ref_mutex must be held entering this routine. Because it is
3229 * called on last put, you should always use the scsi_disk_get()
3230 * scsi_disk_put() helpers which manipulate the semaphore directly
3231 * and never do a direct put_device.
3233 static void scsi_disk_release(struct device *dev)
3235 struct scsi_disk *sdkp = to_scsi_disk(dev);
3236 struct gendisk *disk = sdkp->disk;
3238 spin_lock(&sd_index_lock);
3239 ida_remove(&sd_index_ida, sdkp->index);
3240 spin_unlock(&sd_index_lock);
3242 disk->private_data = NULL;
3244 put_device(&sdkp->device->sdev_gendev);
3249 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3251 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3252 struct scsi_sense_hdr sshdr;
3253 struct scsi_device *sdp = sdkp->device;
3257 cmd[4] |= 1; /* START */
3259 if (sdp->start_stop_pwr_cond)
3260 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3262 if (!scsi_device_online(sdp))
3265 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3266 SD_TIMEOUT, SD_MAX_RETRIES, NULL, 0, RQF_PM);
3268 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3269 if (driver_byte(res) & DRIVER_SENSE)
3270 sd_print_sense_hdr(sdkp, &sshdr);
3271 if (scsi_sense_valid(&sshdr) &&
3272 /* 0x3a is medium not present */
3277 /* SCSI error codes must not go to the generic layer */
3285 * Send a SYNCHRONIZE CACHE instruction down to the device through
3286 * the normal SCSI command structure. Wait for the command to
3289 static void sd_shutdown(struct device *dev)
3291 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3294 return; /* this can happen */
3296 if (pm_runtime_suspended(dev))
3299 if (sdkp->WCE && sdkp->media_present) {
3300 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3301 sd_sync_cache(sdkp);
3304 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3305 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3306 sd_start_stop_device(sdkp, 0);
3310 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3312 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3315 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3318 if (sdkp->WCE && sdkp->media_present) {
3319 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3320 ret = sd_sync_cache(sdkp);
3322 /* ignore OFFLINE device */
3329 if (sdkp->device->manage_start_stop) {
3330 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3331 /* an error is not worth aborting a system sleep */
3332 ret = sd_start_stop_device(sdkp, 0);
3333 if (ignore_stop_errors)
3341 static int sd_suspend_system(struct device *dev)
3343 return sd_suspend_common(dev, true);
3346 static int sd_suspend_runtime(struct device *dev)
3348 return sd_suspend_common(dev, false);
3351 static int sd_resume(struct device *dev)
3353 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3355 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3358 if (!sdkp->device->manage_start_stop)
3361 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3362 return sd_start_stop_device(sdkp, 1);
3366 * init_sd - entry point for this driver (both when built in or when
3369 * Note: this function registers this driver with the scsi mid-level.
3371 static int __init init_sd(void)
3373 int majors = 0, i, err;
3375 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3377 for (i = 0; i < SD_MAJORS; i++) {
3378 if (register_blkdev(sd_major(i), "sd") != 0)
3381 blk_register_region(sd_major(i), SD_MINORS, NULL,
3382 sd_default_probe, NULL, NULL);
3388 err = class_register(&sd_disk_class);
3392 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3394 if (!sd_cdb_cache) {
3395 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3400 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3402 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3407 err = scsi_register_driver(&sd_template.gendrv);
3409 goto err_out_driver;
3414 mempool_destroy(sd_cdb_pool);
3417 kmem_cache_destroy(sd_cdb_cache);
3420 class_unregister(&sd_disk_class);
3422 for (i = 0; i < SD_MAJORS; i++)
3423 unregister_blkdev(sd_major(i), "sd");
3428 * exit_sd - exit point for this driver (when it is a module).
3430 * Note: this function unregisters this driver from the scsi mid-level.
3432 static void __exit exit_sd(void)
3436 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3438 scsi_unregister_driver(&sd_template.gendrv);
3439 mempool_destroy(sd_cdb_pool);
3440 kmem_cache_destroy(sd_cdb_cache);
3442 class_unregister(&sd_disk_class);
3444 for (i = 0; i < SD_MAJORS; i++) {
3445 blk_unregister_region(sd_major(i), SD_MINORS);
3446 unregister_blkdev(sd_major(i), "sd");
3450 module_init(init_sd);
3451 module_exit(exit_sd);
3453 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3454 struct scsi_sense_hdr *sshdr)
3456 scsi_print_sense_hdr(sdkp->device,
3457 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3460 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3463 const char *hb_string = scsi_hostbyte_string(result);
3464 const char *db_string = scsi_driverbyte_string(result);
3466 if (hb_string || db_string)
3467 sd_printk(KERN_INFO, sdkp,
3468 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3469 hb_string ? hb_string : "invalid",
3470 db_string ? db_string : "invalid");
3472 sd_printk(KERN_INFO, sdkp,
3473 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3474 msg, host_byte(result), driver_byte(result));